Commit | Line | Data |
---|---|---|
785eecbb RS |
1 | ;;; cc-engine.el --- core syntax guessing engine for CC mode |
2 | ||
92ab3834 | 3 | ;; Copyright (C) 1985, 1987, 1992, 1993, 1994, 1995, 1996, 1997, 1998, |
114f9c96 | 4 | ;; 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010 |
d7a0267c | 5 | ;; Free Software Foundation, Inc. |
785eecbb | 6 | |
e309f66c AM |
7 | ;; Authors: 2001- Alan Mackenzie |
8 | ;; 1998- Martin Stjernholm | |
d9e94c22 | 9 | ;; 1992-1999 Barry A. Warsaw |
5858f68c GM |
10 | ;; 1987 Dave Detlefs |
11 | ;; 1987 Stewart Clamen | |
785eecbb | 12 | ;; 1985 Richard M. Stallman |
0ec8351b | 13 | ;; Maintainer: bug-cc-mode@gnu.org |
785eecbb | 14 | ;; Created: 22-Apr-1997 (split from cc-mode.el) |
6430c434 | 15 | ;; Version: See cc-mode.el |
785eecbb RS |
16 | ;; Keywords: c languages oop |
17 | ||
18 | ;; This file is part of GNU Emacs. | |
19 | ||
b1fc2b50 | 20 | ;; GNU Emacs is free software: you can redistribute it and/or modify |
785eecbb | 21 | ;; it under the terms of the GNU General Public License as published by |
b1fc2b50 GM |
22 | ;; the Free Software Foundation, either version 3 of the License, or |
23 | ;; (at your option) any later version. | |
785eecbb RS |
24 | |
25 | ;; GNU Emacs is distributed in the hope that it will be useful, | |
26 | ;; but WITHOUT ANY WARRANTY; without even the implied warranty of | |
27 | ;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
28 | ;; GNU General Public License for more details. | |
29 | ||
30 | ;; You should have received a copy of the GNU General Public License | |
b1fc2b50 | 31 | ;; along with GNU Emacs. If not, see <http://www.gnu.org/licenses/>. |
785eecbb | 32 | |
3afbc435 PJ |
33 | ;;; Commentary: |
34 | ||
a66cd3ee MS |
35 | ;; The functions which have docstring documentation can be considered |
36 | ;; part of an API which other packages can use in CC Mode buffers. | |
37 | ;; Otoh, undocumented functions and functions with the documentation | |
38 | ;; in comments are considered purely internal and can change semantics | |
39 | ;; or even disappear in the future. | |
40 | ;; | |
41 | ;; (This policy applies to CC Mode as a whole, not just this file. It | |
42 | ;; probably also applies to many other Emacs packages, but here it's | |
43 | ;; clearly spelled out.) | |
44 | ||
d9e94c22 MS |
45 | ;; Hidden buffer changes |
46 | ;; | |
47 | ;; Various functions in CC Mode use text properties for caching and | |
48 | ;; syntactic markup purposes, and those of them that might modify such | |
0386b551 AM |
49 | ;; properties but still don't modify the buffer in a visible way are |
50 | ;; said to do "hidden buffer changes". They should be used within | |
51 | ;; `c-save-buffer-state' or a similar function that saves and restores | |
52 | ;; buffer modifiedness, disables buffer change hooks, etc. | |
d9e94c22 | 53 | ;; |
0386b551 AM |
54 | ;; Interactive functions are assumed to not do hidden buffer changes, |
55 | ;; except in the specific parts of them that do real changes. | |
d9e94c22 | 56 | ;; |
0386b551 AM |
57 | ;; Lineup functions are assumed to do hidden buffer changes. They |
58 | ;; must not do real changes, though. | |
d9e94c22 | 59 | ;; |
0386b551 AM |
60 | ;; All other functions that do hidden buffer changes have that noted |
61 | ;; in their doc string or comment. | |
62 | ;; | |
63 | ;; The intention with this system is to avoid wrapping every leaf | |
64 | ;; function that do hidden buffer changes inside | |
65 | ;; `c-save-buffer-state'. It should be used as near the top of the | |
66 | ;; interactive functions as possible. | |
67 | ;; | |
68 | ;; Functions called during font locking are allowed to do hidden | |
69 | ;; buffer changes since the font-lock package run them in a context | |
70 | ;; similar to `c-save-buffer-state' (in fact, that function is heavily | |
71 | ;; inspired by `save-buffer-state' in the font-lock package). | |
d9e94c22 MS |
72 | |
73 | ;; Use of text properties | |
74 | ;; | |
75 | ;; CC Mode uses several text properties internally to mark up various | |
76 | ;; positions, e.g. to improve speed and to eliminate glitches in | |
77 | ;; interactive refontification. | |
78 | ;; | |
7bfc3fdb MS |
79 | ;; Note: This doc is for internal use only. Other packages should not |
80 | ;; assume that these text properties are used as described here. | |
81 | ;; | |
0ec1d2c5 AM |
82 | ;; 'category |
83 | ;; Used for "indirection". With its help, some other property can | |
84 | ;; be cheaply and easily switched on or off everywhere it occurs. | |
85 | ;; | |
d9e94c22 | 86 | ;; 'syntax-table |
c8018ede AM |
87 | ;; Used to modify the syntax of some characters. It is used to |
88 | ;; mark the "<" and ">" of angle bracket parens with paren syntax, and | |
89 | ;; to "hide" obtrusive characters in preprocessor lines. | |
d9e94c22 MS |
90 | ;; |
91 | ;; This property is used on single characters and is therefore | |
92 | ;; always treated as front and rear nonsticky (or start and end open | |
93 | ;; in XEmacs vocabulary). It's therefore installed on | |
94 | ;; `text-property-default-nonsticky' if that variable exists (Emacs | |
95 | ;; >= 21). | |
96 | ;; | |
97 | ;; 'c-is-sws and 'c-in-sws | |
98 | ;; Used by `c-forward-syntactic-ws' and `c-backward-syntactic-ws' to | |
99 | ;; speed them up. See the comment blurb before `c-put-is-sws' | |
100 | ;; below for further details. | |
101 | ;; | |
102 | ;; 'c-type | |
103 | ;; This property is used on single characters to mark positions with | |
0386b551 AM |
104 | ;; special syntactic relevance of various sorts. Its primary use is |
105 | ;; to avoid glitches when multiline constructs are refontified | |
d9e94c22 MS |
106 | ;; interactively (on font lock decoration level 3). It's cleared in |
107 | ;; a region before it's fontified and is then put on relevant chars | |
108 | ;; in that region as they are encountered during the fontification. | |
109 | ;; The value specifies the kind of position: | |
110 | ;; | |
111 | ;; 'c-decl-arg-start | |
112 | ;; Put on the last char of the token preceding each declaration | |
113 | ;; inside a declaration style arglist (typically in a function | |
114 | ;; prototype). | |
115 | ;; | |
116 | ;; 'c-decl-end | |
117 | ;; Put on the last char of the token preceding a declaration. | |
118 | ;; This is used in cases where declaration boundaries can't be | |
119 | ;; recognized simply by looking for a token like ";" or "}". | |
120 | ;; `c-type-decl-end-used' must be set if this is used (see also | |
121 | ;; `c-find-decl-spots'). | |
122 | ;; | |
123 | ;; 'c-<>-arg-sep | |
124 | ;; Put on the commas that separate arguments in angle bracket | |
125 | ;; arglists like C++ template arglists. | |
126 | ;; | |
127 | ;; 'c-decl-id-start and 'c-decl-type-start | |
128 | ;; Put on the last char of the token preceding each declarator | |
129 | ;; in the declarator list of a declaration. They are also used | |
130 | ;; between the identifiers cases like enum declarations. | |
131 | ;; 'c-decl-type-start is used when the declarators are types, | |
132 | ;; 'c-decl-id-start otherwise. | |
133 | ;; | |
134 | ;; 'c-awk-NL-prop | |
135 | ;; Used in AWK mode to mark the various kinds of newlines. See | |
136 | ;; cc-awk.el. | |
137 | ||
3afbc435 PJ |
138 | ;;; Code: |
139 | ||
0ec8351b | 140 | (eval-when-compile |
51f606de | 141 | (let ((load-path |
130c507e GM |
142 | (if (and (boundp 'byte-compile-dest-file) |
143 | (stringp byte-compile-dest-file)) | |
144 | (cons (file-name-directory byte-compile-dest-file) load-path) | |
51f606de | 145 | load-path))) |
d9e94c22 | 146 | (load "cc-bytecomp" nil t))) |
130c507e GM |
147 | |
148 | (cc-require 'cc-defs) | |
d9e94c22 | 149 | (cc-require-when-compile 'cc-langs) |
130c507e | 150 | (cc-require 'cc-vars) |
d9e94c22 | 151 | |
130c507e GM |
152 | ;; Silence the compiler. |
153 | (cc-bytecomp-defun buffer-syntactic-context) ; XEmacs | |
0ec8351b | 154 | |
51f606de | 155 | \f |
d9e94c22 MS |
156 | ;; Make declarations for all the `c-lang-defvar' variables in cc-langs. |
157 | ||
158 | (defmacro c-declare-lang-variables () | |
159 | `(progn | |
485fe977 RS |
160 | ,@(apply 'nconc |
161 | (mapcar (lambda (init) | |
162 | `(,(if (elt init 2) | |
163 | `(defvar ,(car init) nil ,(elt init 2)) | |
164 | `(defvar ,(car init) nil)) | |
165 | (make-variable-buffer-local ',(car init)))) | |
166 | (cdr c-lang-variable-inits))))) | |
d9e94c22 MS |
167 | (c-declare-lang-variables) |
168 | ||
169 | \f | |
170 | ;;; Internal state variables. | |
171 | ||
172 | ;; Internal state of hungry delete key feature | |
173 | (defvar c-hungry-delete-key nil) | |
174 | (make-variable-buffer-local 'c-hungry-delete-key) | |
175 | ||
0386b551 AM |
176 | ;; The electric flag (toggled by `c-toggle-electric-state'). |
177 | ;; If t, electric actions (like automatic reindentation, and (if | |
178 | ;; c-auto-newline is also set) auto newlining) will happen when an electric | |
179 | ;; key like `{' is pressed (or an electric keyword like `else'). | |
180 | (defvar c-electric-flag t) | |
181 | (make-variable-buffer-local 'c-electric-flag) | |
182 | ||
d9e94c22 MS |
183 | ;; Internal state of auto newline feature. |
184 | (defvar c-auto-newline nil) | |
185 | (make-variable-buffer-local 'c-auto-newline) | |
186 | ||
0386b551 | 187 | ;; Included in the mode line to indicate the active submodes. |
cb694ab7 AM |
188 | ;; (defvar c-submode-indicators nil) |
189 | ;; (make-variable-buffer-local 'c-submode-indicators) | |
d9e94c22 | 190 | |
a66cd3ee MS |
191 | (defun c-calculate-state (arg prevstate) |
192 | ;; Calculate the new state of PREVSTATE, t or nil, based on arg. If | |
193 | ;; arg is nil or zero, toggle the state. If arg is negative, turn | |
194 | ;; the state off, and if arg is positive, turn the state on | |
195 | (if (or (not arg) | |
196 | (zerop (setq arg (prefix-numeric-value arg)))) | |
197 | (not prevstate) | |
198 | (> arg 0))) | |
199 | ||
d9e94c22 | 200 | ;; Dynamically bound cache for `c-in-literal'. |
130c507e | 201 | (defvar c-in-literal-cache t) |
d9e94c22 | 202 | |
d9e94c22 | 203 | \f |
037558bf MS |
204 | ;; Basic handling of preprocessor directives. |
205 | ||
206 | ;; This is a dynamically bound cache used together with | |
207 | ;; `c-query-macro-start' and `c-query-and-set-macro-start'. It only | |
208 | ;; works as long as point doesn't cross a macro boundary. | |
209 | (defvar c-macro-start 'unknown) | |
210 | ||
211 | (defsubst c-query-and-set-macro-start () | |
037558bf MS |
212 | (if (symbolp c-macro-start) |
213 | (setq c-macro-start (save-excursion | |
0386b551 AM |
214 | (c-save-buffer-state () |
215 | (and (c-beginning-of-macro) | |
216 | (point))))) | |
037558bf MS |
217 | c-macro-start)) |
218 | ||
219 | (defsubst c-query-macro-start () | |
037558bf MS |
220 | (if (symbolp c-macro-start) |
221 | (save-excursion | |
0386b551 AM |
222 | (c-save-buffer-state () |
223 | (and (c-beginning-of-macro) | |
224 | (point)))) | |
037558bf MS |
225 | c-macro-start)) |
226 | ||
227 | (defun c-beginning-of-macro (&optional lim) | |
228 | "Go to the beginning of a preprocessor directive. | |
229 | Leave point at the beginning of the directive and return t if in one, | |
230 | otherwise return nil and leave point unchanged. | |
231 | ||
0386b551 AM |
232 | Note that this function might do hidden buffer changes. See the |
233 | comment at the start of cc-engine.el for more info." | |
037558bf MS |
234 | (when c-opt-cpp-prefix |
235 | (let ((here (point))) | |
236 | (save-restriction | |
237 | (if lim (narrow-to-region lim (point-max))) | |
238 | (beginning-of-line) | |
239 | (while (eq (char-before (1- (point))) ?\\) | |
240 | (forward-line -1)) | |
241 | (back-to-indentation) | |
242 | (if (and (<= (point) here) | |
243 | (looking-at c-opt-cpp-start)) | |
244 | t | |
245 | (goto-char here) | |
246 | nil))))) | |
247 | ||
248 | (defun c-end-of-macro () | |
249 | "Go to the end of a preprocessor directive. | |
0386b551 AM |
250 | More accurately, move the point to the end of the closest following |
251 | line that doesn't end with a line continuation backslash - no check is | |
252 | done that the point is inside a cpp directive to begin with. | |
037558bf | 253 | |
0386b551 AM |
254 | Note that this function might do hidden buffer changes. See the |
255 | comment at the start of cc-engine.el for more info." | |
037558bf MS |
256 | (while (progn |
257 | (end-of-line) | |
258 | (when (and (eq (char-before) ?\\) | |
259 | (not (eobp))) | |
260 | (forward-char) | |
261 | t)))) | |
262 | ||
0ec1d2c5 AM |
263 | (defun c-syntactic-end-of-macro () |
264 | ;; Go to the end of a CPP directive, or a "safe" pos just before. | |
265 | ;; | |
266 | ;; This is normally the end of the next non-escaped line. A "safe" | |
267 | ;; position is one not within a string or comment. (The EOL on a line | |
268 | ;; comment is NOT "safe"). | |
269 | ;; | |
270 | ;; This function must only be called from the beginning of a CPP construct. | |
271 | ;; | |
272 | ;; Note that this function might do hidden buffer changes. See the comment | |
273 | ;; at the start of cc-engine.el for more info. | |
274 | (let* ((here (point)) | |
275 | (there (progn (c-end-of-macro) (point))) | |
276 | (s (parse-partial-sexp here there))) | |
277 | (while (and (or (nth 3 s) ; in a string | |
278 | (nth 4 s)) ; in a comment (maybe at end of line comment) | |
279 | (> there here)) ; No infinite loops, please. | |
280 | (setq there (1- (nth 8 s))) | |
281 | (setq s (parse-partial-sexp here there))) | |
282 | (point))) | |
283 | ||
51c9af45 AM |
284 | (defun c-forward-over-cpp-define-id () |
285 | ;; Assuming point is at the "#" that introduces a preprocessor | |
286 | ;; directive, it's moved forward to the end of the identifier which is | |
287 | ;; "#define"d (or whatever c-opt-cpp-macro-define specifies). Non-nil | |
288 | ;; is returned in this case, in all other cases nil is returned and | |
289 | ;; point isn't moved. | |
290 | ;; | |
291 | ;; This function might do hidden buffer changes. | |
292 | (when (and c-opt-cpp-macro-define-id | |
293 | (looking-at c-opt-cpp-macro-define-id)) | |
294 | (goto-char (match-end 0)))) | |
295 | ||
037558bf MS |
296 | (defun c-forward-to-cpp-define-body () |
297 | ;; Assuming point is at the "#" that introduces a preprocessor | |
298 | ;; directive, it's moved forward to the start of the definition body | |
0386b551 AM |
299 | ;; if it's a "#define" (or whatever c-opt-cpp-macro-define |
300 | ;; specifies). Non-nil is returned in this case, in all other cases | |
301 | ;; nil is returned and point isn't moved. | |
302 | ;; | |
303 | ;; This function might do hidden buffer changes. | |
304 | (when (and c-opt-cpp-macro-define-start | |
305 | (looking-at c-opt-cpp-macro-define-start) | |
037558bf MS |
306 | (not (= (match-end 0) (c-point 'eol)))) |
307 | (goto-char (match-end 0)))) | |
308 | ||
309 | \f | |
d9e94c22 MS |
310 | ;;; Basic utility functions. |
311 | ||
0386b551 | 312 | (defun c-syntactic-content (from to paren-level) |
d9e94c22 MS |
313 | ;; Return the given region as a string where all syntactic |
314 | ;; whitespace is removed or, where necessary, replaced with a single | |
0386b551 AM |
315 | ;; space. If PAREN-LEVEL is given then all parens in the region are |
316 | ;; collapsed to "()", "[]" etc. | |
317 | ;; | |
318 | ;; This function might do hidden buffer changes. | |
319 | ||
d9e94c22 | 320 | (save-excursion |
0386b551 AM |
321 | (save-restriction |
322 | (narrow-to-region from to) | |
323 | (goto-char from) | |
324 | (let* ((parts (list nil)) (tail parts) pos in-paren) | |
325 | ||
326 | (while (re-search-forward c-syntactic-ws-start to t) | |
327 | (goto-char (setq pos (match-beginning 0))) | |
328 | (c-forward-syntactic-ws) | |
329 | (if (= (point) pos) | |
330 | (forward-char) | |
331 | ||
332 | (when paren-level | |
333 | (save-excursion | |
334 | (setq in-paren (= (car (parse-partial-sexp from pos 1)) 1) | |
335 | pos (point)))) | |
336 | ||
337 | (if (and (> pos from) | |
338 | (< (point) to) | |
339 | (looking-at "\\w\\|\\s_") | |
340 | (save-excursion | |
341 | (goto-char (1- pos)) | |
342 | (looking-at "\\w\\|\\s_"))) | |
343 | (progn | |
344 | (setcdr tail (list (buffer-substring-no-properties from pos) | |
345 | " ")) | |
346 | (setq tail (cddr tail))) | |
347 | (setcdr tail (list (buffer-substring-no-properties from pos))) | |
348 | (setq tail (cdr tail))) | |
349 | ||
350 | (when in-paren | |
351 | (when (= (car (parse-partial-sexp pos to -1)) -1) | |
352 | (setcdr tail (list (buffer-substring-no-properties | |
353 | (1- (point)) (point)))) | |
354 | (setq tail (cdr tail)))) | |
355 | ||
356 | (setq from (point)))) | |
357 | ||
358 | (setcdr tail (list (buffer-substring-no-properties from to))) | |
359 | (apply 'concat (cdr parts)))))) | |
360 | ||
361 | (defun c-shift-line-indentation (shift-amt) | |
362 | ;; Shift the indentation of the current line with the specified | |
363 | ;; amount (positive inwards). The buffer is modified only if | |
364 | ;; SHIFT-AMT isn't equal to zero. | |
365 | (let ((pos (- (point-max) (point))) | |
366 | (c-macro-start c-macro-start) | |
367 | tmp-char-inserted) | |
368 | (if (zerop shift-amt) | |
369 | nil | |
370 | ;; If we're on an empty line inside a macro, we take the point | |
371 | ;; to be at the current indentation and shift it to the | |
372 | ;; appropriate column. This way we don't treat the extra | |
373 | ;; whitespace out to the line continuation as indentation. | |
374 | (when (and (c-query-and-set-macro-start) | |
375 | (looking-at "[ \t]*\\\\$") | |
376 | (save-excursion | |
377 | (skip-chars-backward " \t") | |
378 | (bolp))) | |
379 | (insert ?x) | |
380 | (backward-char) | |
381 | (setq tmp-char-inserted t)) | |
382 | (unwind-protect | |
383 | (let ((col (current-indentation))) | |
384 | (delete-region (c-point 'bol) (c-point 'boi)) | |
385 | (beginning-of-line) | |
386 | (indent-to (+ col shift-amt))) | |
387 | (when tmp-char-inserted | |
388 | (delete-char 1)))) | |
389 | ;; If initial point was within line's indentation and we're not on | |
390 | ;; a line with a line continuation in a macro, position after the | |
391 | ;; indentation. Else stay at same point in text. | |
392 | (if (and (< (point) (c-point 'boi)) | |
393 | (not tmp-char-inserted)) | |
394 | (back-to-indentation) | |
395 | (if (> (- (point-max) pos) (point)) | |
396 | (goto-char (- (point-max) pos)))))) | |
d9e94c22 MS |
397 | |
398 | (defsubst c-keyword-sym (keyword) | |
399 | ;; Return non-nil if the string KEYWORD is a known keyword. More | |
400 | ;; precisely, the value is the symbol for the keyword in | |
401 | ;; `c-keywords-obarray'. | |
402 | (intern-soft keyword c-keywords-obarray)) | |
403 | ||
404 | (defsubst c-keyword-member (keyword-sym lang-constant) | |
405 | ;; Return non-nil if the symbol KEYWORD-SYM, as returned by | |
406 | ;; `c-keyword-sym', is a member of LANG-CONSTANT, which is the name | |
407 | ;; of a language constant that ends with "-kwds". If KEYWORD-SYM is | |
408 | ;; nil then the result is nil. | |
409 | (get keyword-sym lang-constant)) | |
410 | ||
411 | ;; String syntax chars, suitable for skip-syntax-(forward|backward). | |
412 | (defconst c-string-syntax (if (memq 'gen-string-delim c-emacs-features) | |
413 | "\"|" | |
414 | "\"")) | |
415 | ||
0386b551 | 416 | ;; Regexp matching string limit syntax. |
d9e94c22 MS |
417 | (defconst c-string-limit-regexp (if (memq 'gen-string-delim c-emacs-features) |
418 | "\\s\"\\|\\s|" | |
419 | "\\s\"")) | |
420 | ||
0386b551 AM |
421 | ;; Regexp matching WS followed by string limit syntax. |
422 | (defconst c-ws*-string-limit-regexp | |
423 | (concat "[ \t]*\\(" c-string-limit-regexp "\\)")) | |
424 | ||
d9e94c22 MS |
425 | ;; Holds formatted error strings for the few cases where parse errors |
426 | ;; are reported. | |
a66cd3ee | 427 | (defvar c-parsing-error nil) |
d9e94c22 MS |
428 | (make-variable-buffer-local 'c-parsing-error) |
429 | ||
430 | (defun c-echo-parsing-error (&optional quiet) | |
d9e94c22 MS |
431 | (when (and c-report-syntactic-errors c-parsing-error (not quiet)) |
432 | (c-benign-error "%s" c-parsing-error)) | |
433 | c-parsing-error) | |
434 | ||
435 | ;; Faces given to comments and string literals. This is used in some | |
436 | ;; situations to speed up recognition; it isn't mandatory that font | |
437 | ;; locking is in use. This variable is extended with the face in | |
438 | ;; `c-doc-face-name' when fontification is activated in cc-fonts.el. | |
1e330469 | 439 | (defvar c-literal-faces |
0386b551 AM |
440 | (append '(font-lock-comment-face font-lock-string-face) |
441 | (when (facep 'font-lock-comment-delimiter-face) | |
442 | ;; New in Emacs 22. | |
443 | '(font-lock-comment-delimiter-face)))) | |
444 | ||
445 | (defsubst c-put-c-type-property (pos value) | |
446 | ;; Put a c-type property with the given value at POS. | |
447 | (c-put-char-property pos 'c-type value)) | |
448 | ||
449 | (defun c-clear-c-type-property (from to value) | |
450 | ;; Remove all occurences of the c-type property that has the given | |
451 | ;; value in the region between FROM and TO. VALUE is assumed to not | |
452 | ;; be nil. | |
453 | ;; | |
454 | ;; Note: This assumes that c-type is put on single chars only; it's | |
455 | ;; very inefficient if matching properties cover large regions. | |
456 | (save-excursion | |
457 | (goto-char from) | |
458 | (while (progn | |
459 | (when (eq (get-text-property (point) 'c-type) value) | |
460 | (c-clear-char-property (point) 'c-type)) | |
461 | (goto-char (next-single-property-change (point) 'c-type nil to)) | |
462 | (< (point) to))))) | |
037558bf | 463 | |
d9e94c22 MS |
464 | \f |
465 | ;; Some debug tools to visualize various special positions. This | |
466 | ;; debug code isn't as portable as the rest of CC Mode. | |
467 | ||
468 | (cc-bytecomp-defun overlays-in) | |
469 | (cc-bytecomp-defun overlay-get) | |
470 | (cc-bytecomp-defun overlay-start) | |
471 | (cc-bytecomp-defun overlay-end) | |
472 | (cc-bytecomp-defun delete-overlay) | |
473 | (cc-bytecomp-defun overlay-put) | |
474 | (cc-bytecomp-defun make-overlay) | |
475 | ||
476 | (defun c-debug-add-face (beg end face) | |
477 | (c-save-buffer-state ((overlays (overlays-in beg end)) overlay) | |
478 | (while overlays | |
479 | (setq overlay (car overlays) | |
480 | overlays (cdr overlays)) | |
481 | (when (eq (overlay-get overlay 'face) face) | |
482 | (setq beg (min beg (overlay-start overlay)) | |
483 | end (max end (overlay-end overlay))) | |
484 | (delete-overlay overlay))) | |
485 | (overlay-put (make-overlay beg end) 'face face))) | |
486 | ||
487 | (defun c-debug-remove-face (beg end face) | |
488 | (c-save-buffer-state ((overlays (overlays-in beg end)) overlay | |
489 | (ol-beg beg) (ol-end end)) | |
490 | (while overlays | |
491 | (setq overlay (car overlays) | |
492 | overlays (cdr overlays)) | |
493 | (when (eq (overlay-get overlay 'face) face) | |
494 | (setq ol-beg (min ol-beg (overlay-start overlay)) | |
495 | ol-end (max ol-end (overlay-end overlay))) | |
496 | (delete-overlay overlay))) | |
497 | (when (< ol-beg beg) | |
498 | (overlay-put (make-overlay ol-beg beg) 'face face)) | |
499 | (when (> ol-end end) | |
500 | (overlay-put (make-overlay end ol-end) 'face face)))) | |
501 | ||
502 | \f | |
503 | ;; `c-beginning-of-statement-1' and accompanying stuff. | |
130c507e | 504 | |
64001211 RS |
505 | ;; KLUDGE ALERT: c-maybe-labelp is used to pass information between |
506 | ;; c-crosses-statement-barrier-p and c-beginning-of-statement-1. A | |
507 | ;; better way should be implemented, but this will at least shut up | |
508 | ;; the byte compiler. | |
0386b551 | 509 | (defvar c-maybe-labelp) |
64001211 | 510 | |
d9e94c22 MS |
511 | ;; New awk-compatible version of c-beginning-of-statement-1, ACM 2002/6/22 |
512 | ||
a66cd3ee MS |
513 | ;; Macros used internally in c-beginning-of-statement-1 for the |
514 | ;; automaton actions. | |
515 | (defmacro c-bos-push-state () | |
516 | '(setq stack (cons (cons state saved-pos) | |
517 | stack))) | |
518 | (defmacro c-bos-pop-state (&optional do-if-done) | |
519 | `(if (setq state (car (car stack)) | |
520 | saved-pos (cdr (car stack)) | |
521 | stack (cdr stack)) | |
522 | t | |
523 | ,do-if-done | |
524 | (throw 'loop nil))) | |
525 | (defmacro c-bos-pop-state-and-retry () | |
526 | '(throw 'loop (setq state (car (car stack)) | |
527 | saved-pos (cdr (car stack)) | |
528 | ;; Throw nil if stack is empty, else throw non-nil. | |
529 | stack (cdr stack)))) | |
530 | (defmacro c-bos-save-pos () | |
531 | '(setq saved-pos (vector pos tok ptok pptok))) | |
532 | (defmacro c-bos-restore-pos () | |
533 | '(unless (eq (elt saved-pos 0) start) | |
534 | (setq pos (elt saved-pos 0) | |
535 | tok (elt saved-pos 1) | |
536 | ptok (elt saved-pos 2) | |
537 | pptok (elt saved-pos 3)) | |
538 | (goto-char pos) | |
539 | (setq sym nil))) | |
540 | (defmacro c-bos-save-error-info (missing got) | |
541 | `(setq saved-pos (vector pos ,missing ,got))) | |
542 | (defmacro c-bos-report-error () | |
543 | '(unless noerror | |
544 | (setq c-parsing-error | |
545 | (format "No matching `%s' found for `%s' on line %d" | |
546 | (elt saved-pos 1) | |
547 | (elt saved-pos 2) | |
548 | (1+ (count-lines (point-min) | |
549 | (c-point 'bol (elt saved-pos 0)))))))) | |
550 | ||
551 | (defun c-beginning-of-statement-1 (&optional lim ignore-labels | |
552 | noerror comma-delim) | |
553 | "Move to the start of the current statement or declaration, or to | |
554 | the previous one if already at the beginning of one. Only | |
555 | statements/declarations on the same level are considered, i.e. don't | |
556 | move into or out of sexps (not even normal expression parentheses). | |
557 | ||
a85fd6da AM |
558 | If point is already at the earliest statment within braces or parens, |
559 | this function doesn't move back into any whitespace preceding it; it | |
560 | returns 'same in this case. | |
561 | ||
d9e94c22 MS |
562 | Stop at statement continuation tokens like \"else\", \"catch\", |
563 | \"finally\" and the \"while\" in \"do ... while\" if the start point | |
564 | is within the continuation. If starting at such a token, move to the | |
565 | corresponding statement start. If at the beginning of a statement, | |
566 | move to the closest containing statement if there is any. This might | |
567 | also stop at a continuation clause. | |
a66cd3ee | 568 | |
0386b551 AM |
569 | Labels are treated as part of the following statements if |
570 | IGNORE-LABELS is non-nil. (FIXME: Doesn't work if we stop at a known | |
cb694ab7 AM |
571 | statement start keyword.) Otherwise, each label is treated as a |
572 | separate statement. | |
a66cd3ee | 573 | |
cb694ab7 AM |
574 | Macros are ignored \(i.e. skipped over) unless point is within one, in |
575 | which case the content of the macro is treated as normal code. Aside | |
576 | from any normal statement starts found in it, stop at the first token | |
577 | of the content in the macro, i.e. the expression of an \"#if\" or the | |
578 | start of the definition in a \"#define\". Also stop at start of | |
579 | macros before leaving them. | |
a66cd3ee | 580 | |
a85fd6da | 581 | Return: |
d28e7f28 | 582 | 'label if stopped at a label or \"case...:\" or \"default:\"; |
a85fd6da AM |
583 | 'same if stopped at the beginning of the current statement; |
584 | 'up if stepped to a containing statement; | |
585 | 'previous if stepped to a preceding statement; | |
586 | 'beginning if stepped from a statement continuation clause to | |
587 | its start clause; or | |
588 | 'macro if stepped to a macro start. | |
589 | Note that 'same and not 'label is returned if stopped at the same | |
590 | label without crossing the colon character. | |
a66cd3ee MS |
591 | |
592 | LIM may be given to limit the search. If the search hits the limit, | |
593 | point will be left at the closest following token, or at the start | |
594 | position if that is less ('same is returned in this case). | |
595 | ||
596 | NOERROR turns off error logging to `c-parsing-error'. | |
597 | ||
cb694ab7 AM |
598 | Normally only ';' and virtual semicolons are considered to delimit |
599 | statements, but if COMMA-DELIM is non-nil then ',' is treated | |
600 | as a delimiter too. | |
0386b551 AM |
601 | |
602 | Note that this function might do hidden buffer changes. See the | |
603 | comment at the start of cc-engine.el for more info." | |
a66cd3ee | 604 | |
d9e94c22 MS |
605 | ;; The bulk of this function is a pushdown automaton that looks at statement |
606 | ;; boundaries and the tokens (such as "while") in c-opt-block-stmt-key. Its | |
607 | ;; purpose is to keep track of nested statements, ensuring that such | |
608 | ;; statments are skipped over in their entirety (somewhat akin to what C-M-p | |
609 | ;; does with nested braces/brackets/parentheses). | |
a66cd3ee MS |
610 | ;; |
611 | ;; Note: The position of a boundary is the following token. | |
612 | ;; | |
d9e94c22 MS |
613 | ;; Beginning with the current token (the one following point), move back one |
614 | ;; sexp at a time (where a sexp is, more or less, either a token or the | |
615 | ;; entire contents of a brace/bracket/paren pair). Each time a statement | |
616 | ;; boundary is crossed or a "while"-like token is found, update the state of | |
617 | ;; the PDA. Stop at the beginning of a statement when the stack (holding | |
618 | ;; nested statement info) is empty and the position has been moved. | |
619 | ;; | |
620 | ;; The following variables constitute the PDA: | |
621 | ;; | |
622 | ;; sym: This is either the "while"-like token (e.g. 'for) we've just | |
623 | ;; scanned back over, 'boundary if we've just gone back over a | |
624 | ;; statement boundary, or nil otherwise. | |
625 | ;; state: takes one of the values (nil else else-boundary while | |
626 | ;; while-boundary catch catch-boundary). | |
627 | ;; nil means "no "while"-like token yet scanned". | |
628 | ;; 'else, for example, means "just gone back over an else". | |
629 | ;; 'else-boundary means "just gone back over a statement boundary | |
630 | ;; immediately after having gone back over an else". | |
631 | ;; saved-pos: A vector of either saved positions (tok ptok pptok, etc.) or | |
632 | ;; of error reporting information. | |
633 | ;; stack: The stack onto which the PDA pushes its state. Each entry | |
634 | ;; consists of a saved value of state and saved-pos. An entry is | |
635 | ;; pushed when we move back over a "continuation" token (e.g. else) | |
636 | ;; and popped when we encounter the corresponding opening token | |
637 | ;; (e.g. if). | |
638 | ;; | |
639 | ;; | |
b414f371 | 640 | ;; The following diagram briefly outlines the PDA. |
a66cd3ee MS |
641 | ;; |
642 | ;; Common state: | |
d9e94c22 MS |
643 | ;; "else": Push state, goto state `else'. |
644 | ;; "while": Push state, goto state `while'. | |
645 | ;; "catch" or "finally": Push state, goto state `catch'. | |
646 | ;; boundary: Pop state. | |
a66cd3ee MS |
647 | ;; other: Do nothing special. |
648 | ;; | |
d9e94c22 MS |
649 | ;; State `else': |
650 | ;; boundary: Goto state `else-boundary'. | |
651 | ;; other: Error, pop state, retry token. | |
652 | ;; | |
653 | ;; State `else-boundary': | |
654 | ;; "if": Pop state. | |
655 | ;; boundary: Error, pop state. | |
656 | ;; other: See common state. | |
657 | ;; | |
658 | ;; State `while': | |
659 | ;; boundary: Save position, goto state `while-boundary'. | |
660 | ;; other: Pop state, retry token. | |
661 | ;; | |
662 | ;; State `while-boundary': | |
663 | ;; "do": Pop state. | |
664 | ;; boundary: Restore position if it's not at start, pop state. [*see below] | |
665 | ;; other: See common state. | |
666 | ;; | |
667 | ;; State `catch': | |
668 | ;; boundary: Goto state `catch-boundary'. | |
669 | ;; other: Error, pop state, retry token. | |
670 | ;; | |
671 | ;; State `catch-boundary': | |
672 | ;; "try": Pop state. | |
673 | ;; "catch": Goto state `catch'. | |
674 | ;; boundary: Error, pop state. | |
675 | ;; other: See common state. | |
676 | ;; | |
677 | ;; [*] In the `while-boundary' state, we had pushed a 'while state, and were | |
678 | ;; searching for a "do" which would have opened a do-while. If we didn't | |
679 | ;; find it, we discard the analysis done since the "while", go back to this | |
680 | ;; token in the buffer and restart the scanning there, this time WITHOUT | |
681 | ;; pushing the 'while state onto the stack. | |
682 | ;; | |
a66cd3ee MS |
683 | ;; In addition to the above there is some special handling of labels |
684 | ;; and macros. | |
685 | ||
686 | (let ((case-fold-search nil) | |
687 | (start (point)) | |
688 | macro-start | |
689 | (delims (if comma-delim '(?\; ?,) '(?\;))) | |
690 | (c-stmt-delim-chars (if comma-delim | |
691 | c-stmt-delim-chars-with-comma | |
692 | c-stmt-delim-chars)) | |
d28e7f28 | 693 | c-in-literal-cache c-maybe-labelp after-case:-pos saved |
0386b551 AM |
694 | ;; Current position. |
695 | pos | |
696 | ;; Position of last stmt boundary character (e.g. ;). | |
697 | boundary-pos | |
698 | ;; The position of the last sexp or bound that follows the | |
699 | ;; first found colon, i.e. the start of the nonlabel part of | |
700 | ;; the statement. It's `start' if a colon is found just after | |
701 | ;; the start. | |
702 | after-labels-pos | |
703 | ;; Like `after-labels-pos', but the first such position inside | |
704 | ;; a label, i.e. the start of the last label before the start | |
705 | ;; of the nonlabel part of the statement. | |
706 | last-label-pos | |
707 | ;; The last position where a label is possible provided the | |
708 | ;; statement started there. It's nil as long as no invalid | |
709 | ;; label content has been found (according to | |
710 | ;; `c-nonlabel-token-key'. It's `start' if no valid label | |
711 | ;; content was found in the label. Note that we might still | |
712 | ;; regard it a label if it starts with `c-label-kwds'. | |
713 | label-good-pos | |
714 | ;; Symbol just scanned back over (e.g. 'while or 'boundary). | |
715 | ;; See above. | |
716 | sym | |
717 | ;; Current state in the automaton. See above. | |
718 | state | |
719 | ;; Current saved positions. See above. | |
720 | saved-pos | |
721 | ;; Stack of conses (state . saved-pos). | |
722 | stack | |
723 | ;; Regexp which matches "for", "if", etc. | |
724 | (cond-key (or c-opt-block-stmt-key | |
a66cd3ee | 725 | "\\<\\>")) ; Matches nothing. |
0386b551 AM |
726 | ;; Return value. |
727 | (ret 'same) | |
728 | ;; Positions of the last three sexps or bounds we've stopped at. | |
729 | tok ptok pptok) | |
a66cd3ee MS |
730 | |
731 | (save-restriction | |
732 | (if lim (narrow-to-region lim (point-max))) | |
733 | ||
734 | (if (save-excursion | |
735 | (and (c-beginning-of-macro) | |
736 | (/= (point) start))) | |
737 | (setq macro-start (point))) | |
738 | ||
d9e94c22 | 739 | ;; Try to skip back over unary operator characters, to register |
a66cd3ee MS |
740 | ;; that we've moved. |
741 | (while (progn | |
742 | (setq pos (point)) | |
0386b551 AM |
743 | (c-backward-syntactic-ws) |
744 | ;; Protect post-++/-- operators just before a virtual semicolon. | |
745 | (and (not (c-at-vsemi-p)) | |
746 | (/= (skip-chars-backward "-+!*&~@`#") 0)))) | |
d9e94c22 MS |
747 | |
748 | ;; Skip back over any semicolon here. If it was a bare semicolon, we're | |
0386b551 | 749 | ;; done. Later on we ignore the boundaries for statements that don't |
d9e94c22 MS |
750 | ;; contain any sexp. The only thing that is affected is that the error |
751 | ;; checking is a little less strict, and we really don't bother. | |
a66cd3ee MS |
752 | (if (and (memq (char-before) delims) |
753 | (progn (forward-char -1) | |
754 | (setq saved (point)) | |
0386b551 | 755 | (c-backward-syntactic-ws) |
a66cd3ee MS |
756 | (or (memq (char-before) delims) |
757 | (memq (char-before) '(?: nil)) | |
d9e94c22 | 758 | (eq (char-syntax (char-before)) ?\() |
0386b551 | 759 | (c-at-vsemi-p)))) |
a66cd3ee MS |
760 | (setq ret 'previous |
761 | pos saved) | |
762 | ||
763 | ;; Begin at start and not pos to detect macros if we stand | |
764 | ;; directly after the #. | |
765 | (goto-char start) | |
766 | (if (looking-at "\\<\\|\\W") | |
767 | ;; Record this as the first token if not starting inside it. | |
768 | (setq tok start)) | |
769 | ||
d9e94c22 MS |
770 | ;; The following while loop goes back one sexp (balanced parens, |
771 | ;; etc. with contents, or symbol or suchlike) each iteration. This | |
772 | ;; movement is accomplished with a call to scan-sexps approx 130 lines | |
773 | ;; below. | |
a66cd3ee MS |
774 | (while |
775 | (catch 'loop ;; Throw nil to break, non-nil to continue. | |
776 | (cond | |
a66cd3ee | 777 | ((save-excursion |
0386b551 | 778 | (and macro-start ; Always NIL for AWK. |
a66cd3ee MS |
779 | (progn (skip-chars-backward " \t") |
780 | (eq (char-before) ?#)) | |
781 | (progn (setq saved (1- (point))) | |
782 | (beginning-of-line) | |
783 | (not (eq (char-before (1- (point))) ?\\))) | |
d9e94c22 | 784 | (looking-at c-opt-cpp-start) |
a66cd3ee MS |
785 | (progn (skip-chars-forward " \t") |
786 | (eq (point) saved)))) | |
787 | (goto-char saved) | |
788 | (if (and (c-forward-to-cpp-define-body) | |
789 | (progn (c-forward-syntactic-ws start) | |
790 | (< (point) start))) | |
791 | ;; Stop at the first token in the content of the macro. | |
792 | (setq pos (point) | |
793 | ignore-labels t) ; Avoid the label check on exit. | |
794 | (setq pos saved | |
795 | ret 'macro | |
796 | ignore-labels t)) | |
797 | (throw 'loop nil)) | |
798 | ||
d9e94c22 MS |
799 | ;; Do a round through the automaton if we've just passed a |
800 | ;; statement boundary or passed a "while"-like token. | |
a66cd3ee MS |
801 | ((or sym |
802 | (and (looking-at cond-key) | |
803 | (setq sym (intern (match-string 1))))) | |
804 | ||
805 | (when (and (< pos start) (null stack)) | |
806 | (throw 'loop nil)) | |
807 | ||
d9e94c22 MS |
808 | ;; The PDA state handling. |
809 | ;; | |
037558bf | 810 | ;; Refer to the description of the PDA in the opening |
d9e94c22 MS |
811 | ;; comments. In the following OR form, the first leaf |
812 | ;; attempts to handles one of the specific actions detailed | |
813 | ;; (e.g., finding token "if" whilst in state `else-boundary'). | |
814 | ;; We drop through to the second leaf (which handles common | |
815 | ;; state) if no specific handler is found in the first cond. | |
816 | ;; If a parsing error is detected (e.g. an "else" with no | |
817 | ;; preceding "if"), we throw to the enclosing catch. | |
818 | ;; | |
819 | ;; Note that the (eq state 'else) means | |
820 | ;; "we've just passed an else", NOT "we're looking for an | |
821 | ;; else". | |
a66cd3ee MS |
822 | (or (cond |
823 | ((eq state 'else) | |
824 | (if (eq sym 'boundary) | |
825 | (setq state 'else-boundary) | |
826 | (c-bos-report-error) | |
827 | (c-bos-pop-state-and-retry))) | |
828 | ||
829 | ((eq state 'else-boundary) | |
830 | (cond ((eq sym 'if) | |
831 | (c-bos-pop-state (setq ret 'beginning))) | |
832 | ((eq sym 'boundary) | |
833 | (c-bos-report-error) | |
834 | (c-bos-pop-state)))) | |
835 | ||
836 | ((eq state 'while) | |
837 | (if (and (eq sym 'boundary) | |
838 | ;; Since this can cause backtracking we do a | |
839 | ;; little more careful analysis to avoid it: | |
840 | ;; If there's a label in front of the while | |
841 | ;; it can't be part of a do-while. | |
842 | (not after-labels-pos)) | |
843 | (progn (c-bos-save-pos) | |
844 | (setq state 'while-boundary)) | |
d9e94c22 | 845 | (c-bos-pop-state-and-retry))) ; Can't be a do-while |
a66cd3ee MS |
846 | |
847 | ((eq state 'while-boundary) | |
848 | (cond ((eq sym 'do) | |
849 | (c-bos-pop-state (setq ret 'beginning))) | |
d9e94c22 MS |
850 | ((eq sym 'boundary) ; isn't a do-while |
851 | (c-bos-restore-pos) ; the position of the while | |
852 | (c-bos-pop-state)))) ; no longer searching for do. | |
a66cd3ee MS |
853 | |
854 | ((eq state 'catch) | |
855 | (if (eq sym 'boundary) | |
856 | (setq state 'catch-boundary) | |
857 | (c-bos-report-error) | |
858 | (c-bos-pop-state-and-retry))) | |
859 | ||
860 | ((eq state 'catch-boundary) | |
861 | (cond | |
862 | ((eq sym 'try) | |
863 | (c-bos-pop-state (setq ret 'beginning))) | |
864 | ((eq sym 'catch) | |
865 | (setq state 'catch)) | |
866 | ((eq sym 'boundary) | |
867 | (c-bos-report-error) | |
868 | (c-bos-pop-state))))) | |
869 | ||
d9e94c22 MS |
870 | ;; This is state common. We get here when the previous |
871 | ;; cond statement found no particular state handler. | |
a66cd3ee | 872 | (cond ((eq sym 'boundary) |
d9e94c22 MS |
873 | ;; If we have a boundary at the start |
874 | ;; position we push a frame to go to the | |
875 | ;; previous statement. | |
876 | (if (>= pos start) | |
877 | (c-bos-push-state) | |
878 | (c-bos-pop-state))) | |
a66cd3ee MS |
879 | ((eq sym 'else) |
880 | (c-bos-push-state) | |
881 | (c-bos-save-error-info 'if 'else) | |
882 | (setq state 'else)) | |
883 | ((eq sym 'while) | |
0386b551 AM |
884 | ;; Is this a real while, or a do-while? |
885 | ;; The next `when' triggers unless we are SURE that | |
886 | ;; the `while' is not the tailend of a `do-while'. | |
a66cd3ee | 887 | (when (or (not pptok) |
d9e94c22 | 888 | (memq (char-after pptok) delims) |
0386b551 AM |
889 | ;; The following kludge is to prevent |
890 | ;; infinite recursion when called from | |
891 | ;; c-awk-after-if-for-while-condition-p, | |
892 | ;; or the like. | |
893 | (and (eq (point) start) | |
894 | (c-vsemi-status-unknown-p)) | |
895 | (c-at-vsemi-p pptok)) | |
a66cd3ee MS |
896 | ;; Since this can cause backtracking we do a |
897 | ;; little more careful analysis to avoid it: If | |
0386b551 AM |
898 | ;; the while isn't followed by a (possibly |
899 | ;; virtual) semicolon it can't be a do-while. | |
a66cd3ee MS |
900 | (c-bos-push-state) |
901 | (setq state 'while))) | |
902 | ((memq sym '(catch finally)) | |
903 | (c-bos-push-state) | |
904 | (c-bos-save-error-info 'try sym) | |
905 | (setq state 'catch)))) | |
906 | ||
907 | (when c-maybe-labelp | |
908 | ;; We're either past a statement boundary or at the | |
909 | ;; start of a statement, so throw away any label data | |
910 | ;; for the previous one. | |
911 | (setq after-labels-pos nil | |
912 | last-label-pos nil | |
913 | c-maybe-labelp nil)))) | |
914 | ||
d9e94c22 MS |
915 | ;; Step to the previous sexp, but not if we crossed a |
916 | ;; boundary, since that doesn't consume an sexp. | |
a66cd3ee MS |
917 | (if (eq sym 'boundary) |
918 | (setq ret 'previous) | |
d9e94c22 MS |
919 | |
920 | ;; HERE IS THE SINGLE PLACE INSIDE THE PDA LOOP WHERE WE MOVE | |
0386b551 AM |
921 | ;; BACKWARDS THROUGH THE SOURCE. |
922 | ||
923 | ;; This is typically fast with the caching done by | |
924 | ;; c-(backward|forward)-sws. | |
925 | (c-backward-syntactic-ws) | |
926 | ||
927 | (let ((before-sws-pos (point)) | |
928 | ;; Set as long as we have to continue jumping by sexps. | |
929 | ;; It's the position to use as end in the next round. | |
930 | sexp-loop-continue-pos | |
931 | ;; The end position of the area to search for statement | |
932 | ;; barriers in this round. | |
933 | (sexp-loop-end-pos pos)) | |
934 | ||
cb694ab7 | 935 | ;; The following while goes back one sexp per iteration. |
0386b551 AM |
936 | (while |
937 | (progn | |
938 | (unless (c-safe (c-backward-sexp) t) | |
939 | ;; Give up if we hit an unbalanced block. Since the | |
940 | ;; stack won't be empty the code below will report a | |
941 | ;; suitable error. | |
a66cd3ee | 942 | (throw 'loop nil)) |
0386b551 AM |
943 | |
944 | ;; Check if the sexp movement crossed a statement or | |
945 | ;; declaration boundary. But first modify the point | |
946 | ;; so that `c-crosses-statement-barrier-p' only looks | |
947 | ;; at the non-sexp chars following the sexp. | |
948 | (save-excursion | |
949 | (when (setq | |
950 | boundary-pos | |
951 | (cond | |
952 | ((if macro-start | |
953 | nil | |
954 | (save-excursion | |
955 | (when (c-beginning-of-macro) | |
956 | ;; Set continuation position in case | |
957 | ;; `c-crosses-statement-barrier-p' | |
958 | ;; doesn't detect anything below. | |
959 | (setq sexp-loop-continue-pos (point))))) | |
960 | ;; If the sexp movement took us into a | |
961 | ;; macro then there were only some non-sexp | |
962 | ;; chars after it. Skip out of the macro | |
963 | ;; to analyze them but not the non-sexp | |
964 | ;; chars that might be inside the macro. | |
965 | (c-end-of-macro) | |
966 | (c-crosses-statement-barrier-p | |
967 | (point) sexp-loop-end-pos)) | |
968 | ||
969 | ((and | |
970 | (eq (char-after) ?{) | |
971 | (not (c-looking-at-inexpr-block lim nil t))) | |
972 | ;; Passed a block sexp. That's a boundary | |
973 | ;; alright. | |
974 | (point)) | |
975 | ||
976 | ((looking-at "\\s\(") | |
977 | ;; Passed some other paren. Only analyze | |
978 | ;; the non-sexp chars after it. | |
979 | (goto-char (1+ (c-down-list-backward | |
980 | before-sws-pos))) | |
981 | ;; We're at a valid token start position | |
982 | ;; (outside the `save-excursion') if | |
983 | ;; `c-crosses-statement-barrier-p' failed. | |
984 | (c-crosses-statement-barrier-p | |
985 | (point) sexp-loop-end-pos)) | |
986 | ||
987 | (t | |
988 | ;; Passed a symbol sexp or line | |
989 | ;; continuation. It doesn't matter that | |
990 | ;; it's included in the analyzed region. | |
991 | (if (c-crosses-statement-barrier-p | |
992 | (point) sexp-loop-end-pos) | |
993 | t | |
994 | ;; If it was a line continuation then we | |
995 | ;; have to continue looping. | |
996 | (if (looking-at "\\\\$") | |
997 | (setq sexp-loop-continue-pos (point))) | |
998 | nil)))) | |
999 | ||
1000 | (setq pptok ptok | |
1001 | ptok tok | |
1002 | tok boundary-pos | |
1003 | sym 'boundary) | |
1004 | ;; Like a C "continue". Analyze the next sexp. | |
1005 | (throw 'loop t))) | |
1006 | ||
d28e7f28 | 1007 | sexp-loop-continue-pos) ; End of "go back a sexp" loop condition. |
0386b551 AM |
1008 | (goto-char sexp-loop-continue-pos) |
1009 | (setq sexp-loop-end-pos sexp-loop-continue-pos | |
1010 | sexp-loop-continue-pos nil)))) | |
a66cd3ee MS |
1011 | |
1012 | ;; ObjC method def? | |
1013 | (when (and c-opt-method-key | |
1014 | (setq saved (c-in-method-def-p))) | |
1015 | (setq pos saved | |
1016 | ignore-labels t) ; Avoid the label check on exit. | |
1017 | (throw 'loop nil)) | |
1018 | ||
0386b551 AM |
1019 | ;; Handle labels. |
1020 | (unless (eq ignore-labels t) | |
1021 | (when (numberp c-maybe-labelp) | |
cb694ab7 AM |
1022 | ;; `c-crosses-statement-barrier-p' has found a colon, so we |
1023 | ;; might be in a label now. Have we got a real label | |
1024 | ;; (including a case label) or something like C++'s "public:"? | |
d28e7f28 AM |
1025 | ;; A case label might use an expression rather than a token. |
1026 | (setq after-case:-pos (or tok start)) | |
1027 | (if (looking-at c-nonlabel-token-key) ; e.g. "while" or "'a'" | |
1028 | (setq c-maybe-labelp nil) | |
1029 | (if after-labels-pos ; Have we already encountered a label? | |
1030 | (if (not last-label-pos) | |
1031 | (setq last-label-pos (or tok start))) | |
1032 | (setq after-labels-pos (or tok start))) | |
1033 | (setq c-maybe-labelp t | |
1034 | label-good-pos nil))) ; bogus "label" | |
cb694ab7 AM |
1035 | |
1036 | (when (and (not label-good-pos) ; i.e. no invalid "label"'s yet | |
1037 | ; been found. | |
1038 | (looking-at c-nonlabel-token-key)) ; e.g. "while :" | |
0386b551 AM |
1039 | ;; We're in a potential label and it's the first |
1040 | ;; time we've found something that isn't allowed in | |
1041 | ;; one. | |
1042 | (setq label-good-pos (or tok start)))) | |
1043 | ||
1044 | ;; We've moved back by a sexp, so update the token positions. | |
a66cd3ee MS |
1045 | (setq sym nil |
1046 | pptok ptok | |
1047 | ptok tok | |
1048 | tok (point) | |
d9e94c22 | 1049 | pos tok))) ; Not nil (for the while loop). |
a66cd3ee MS |
1050 | |
1051 | ;; If the stack isn't empty there might be errors to report. | |
1052 | (while stack | |
1053 | (if (and (vectorp saved-pos) (eq (length saved-pos) 3)) | |
1054 | (c-bos-report-error)) | |
1055 | (setq saved-pos (cdr (car stack)) | |
1056 | stack (cdr stack))) | |
1057 | ||
1058 | (when (and (eq ret 'same) | |
1059 | (not (memq sym '(boundary ignore nil)))) | |
1060 | ;; Need to investigate closer whether we've crossed | |
1061 | ;; between a substatement and its containing statement. | |
1062 | (if (setq saved (if (looking-at c-block-stmt-1-key) | |
1063 | ptok | |
1064 | pptok)) | |
1065 | (cond ((> start saved) (setq pos saved)) | |
1066 | ((= start saved) (setq ret 'up))))) | |
1067 | ||
0386b551 AM |
1068 | (when (and (not ignore-labels) |
1069 | (eq c-maybe-labelp t) | |
d9e94c22 | 1070 | (not (eq ret 'beginning)) |
0386b551 AM |
1071 | after-labels-pos |
1072 | (or (not label-good-pos) | |
1073 | (<= label-good-pos pos) | |
1074 | (progn | |
1075 | (goto-char (if (and last-label-pos | |
1076 | (< last-label-pos start)) | |
1077 | last-label-pos | |
1078 | pos)) | |
1079 | (looking-at c-label-kwds-regexp)))) | |
a66cd3ee MS |
1080 | ;; We're in a label. Maybe we should step to the statement |
1081 | ;; after it. | |
1082 | (if (< after-labels-pos start) | |
1083 | (setq pos after-labels-pos) | |
1084 | (setq ret 'label) | |
0386b551 AM |
1085 | (if (and last-label-pos (< last-label-pos start)) |
1086 | ;; Might have jumped over several labels. Go to the last one. | |
a66cd3ee MS |
1087 | (setq pos last-label-pos))))) |
1088 | ||
d28e7f28 | 1089 | ;; Have we got "case <expression>:"? |
a66cd3ee | 1090 | (goto-char pos) |
d28e7f28 AM |
1091 | (when (and after-case:-pos |
1092 | (not (eq ret 'beginning)) | |
1093 | (looking-at c-case-kwds-regexp)) | |
1094 | (if (< after-case:-pos start) | |
0000ee90 AM |
1095 | (setq pos after-case:-pos)) |
1096 | (if (eq ret 'same) | |
1097 | (setq ret 'label))) | |
d28e7f28 AM |
1098 | |
1099 | ;; Skip over the unary operators that can start the statement. | |
a66cd3ee | 1100 | (while (progn |
0386b551 AM |
1101 | (c-backward-syntactic-ws) |
1102 | ;; protect AWK post-inc/decrement operators, etc. | |
1103 | (and (not (c-at-vsemi-p (point))) | |
1104 | (/= (skip-chars-backward "-+!*&~@`#") 0))) | |
a66cd3ee MS |
1105 | (setq pos (point))) |
1106 | (goto-char pos) | |
1107 | ret))) | |
785eecbb | 1108 | |
785eecbb | 1109 | (defun c-crosses-statement-barrier-p (from to) |
a66cd3ee MS |
1110 | "Return non-nil if buffer positions FROM to TO cross one or more |
1111 | statement or declaration boundaries. The returned value is actually | |
d9e94c22 MS |
1112 | the position of the earliest boundary char. FROM must not be within |
1113 | a string or comment. | |
a66cd3ee MS |
1114 | |
1115 | The variable `c-maybe-labelp' is set to the position of the first `:' that | |
1116 | might start a label (i.e. not part of `::' and not preceded by `?'). If a | |
0386b551 AM |
1117 | single `?' is found, then `c-maybe-labelp' is cleared. |
1118 | ||
1119 | For AWK, a statement which is terminated by an EOL (not a \; or a }) is | |
1120 | regarded as having a \"virtual semicolon\" immediately after the last token on | |
b414f371 | 1121 | the line. If this virtual semicolon is _at_ from, the function recognizes it. |
0386b551 AM |
1122 | |
1123 | Note that this function might do hidden buffer changes. See the | |
1124 | comment at the start of cc-engine.el for more info." | |
a66cd3ee MS |
1125 | (let ((skip-chars c-stmt-delim-chars) |
1126 | lit-range) | |
1127 | (save-excursion | |
1128 | (catch 'done | |
1129 | (goto-char from) | |
1130 | (while (progn (skip-chars-forward skip-chars to) | |
785eecbb | 1131 | (< (point) to)) |
0386b551 AM |
1132 | (cond |
1133 | ((setq lit-range (c-literal-limits from)) ; Have we landed in a string/comment? | |
1134 | (goto-char (cdr lit-range))) | |
1135 | ((eq (char-after) ?:) | |
1136 | (forward-char) | |
1137 | (if (and (eq (char-after) ?:) | |
1138 | (< (point) to)) | |
1139 | ;; Ignore scope operators. | |
1140 | (forward-char) | |
1141 | (setq c-maybe-labelp (1- (point))))) | |
1142 | ((eq (char-after) ??) | |
1143 | ;; A question mark. Can't be a label, so stop | |
1144 | ;; looking for more : and ?. | |
1145 | (setq c-maybe-labelp nil | |
1146 | skip-chars (substring c-stmt-delim-chars 0 -2))) | |
1147 | ((memq (char-after) '(?# ?\n ?\r)) ; A virtual semicolon? | |
1148 | (if (and (eq (char-before) ?\\) (memq (char-after) '(?\n ?\r))) | |
1149 | (backward-char)) | |
1150 | (skip-chars-backward " \t" from) | |
1151 | (if (c-at-vsemi-p) | |
1152 | (throw 'done (point)) | |
1153 | (forward-line))) | |
1154 | (t (throw 'done (point))))) | |
1155 | ;; In trailing space after an as yet undetected virtual semicolon? | |
1156 | (c-backward-syntactic-ws from) | |
1157 | (if (and (< (point) to) | |
1158 | (c-at-vsemi-p)) | |
1159 | (point) | |
1160 | nil))))) | |
1161 | ||
1162 | (defun c-at-statement-start-p () | |
1163 | "Return non-nil if the point is at the first token in a statement | |
1164 | or somewhere in the syntactic whitespace before it. | |
1165 | ||
1166 | A \"statement\" here is not restricted to those inside code blocks. | |
1167 | Any kind of declaration-like construct that occur outside function | |
1168 | bodies is also considered a \"statement\". | |
1169 | ||
1170 | Note that this function might do hidden buffer changes. See the | |
1171 | comment at the start of cc-engine.el for more info." | |
1172 | ||
1173 | (save-excursion | |
1174 | (let ((end (point)) | |
1175 | c-maybe-labelp) | |
1176 | (c-syntactic-skip-backward (substring c-stmt-delim-chars 1) nil t) | |
1177 | (or (bobp) | |
1178 | (eq (char-before) ?}) | |
1179 | (and (eq (char-before) ?{) | |
1180 | (not (and c-special-brace-lists | |
1181 | (progn (backward-char) | |
1182 | (c-looking-at-special-brace-list))))) | |
1183 | (c-crosses-statement-barrier-p (point) end))))) | |
1184 | ||
1185 | (defun c-at-expression-start-p () | |
1186 | "Return non-nil if the point is at the first token in an expression or | |
1187 | statement, or somewhere in the syntactic whitespace before it. | |
1188 | ||
1189 | An \"expression\" here is a bit different from the normal language | |
1190 | grammar sense: It's any sequence of expression tokens except commas, | |
1191 | unless they are enclosed inside parentheses of some kind. Also, an | |
1192 | expression never continues past an enclosing parenthesis, but it might | |
1193 | contain parenthesis pairs of any sort except braces. | |
1194 | ||
1195 | Since expressions never cross statement boundaries, this function also | |
1196 | recognizes statement beginnings, just like `c-at-statement-start-p'. | |
1197 | ||
1198 | Note that this function might do hidden buffer changes. See the | |
1199 | comment at the start of cc-engine.el for more info." | |
1200 | ||
1201 | (save-excursion | |
1202 | (let ((end (point)) | |
1203 | (c-stmt-delim-chars c-stmt-delim-chars-with-comma) | |
1204 | c-maybe-labelp) | |
1205 | (c-syntactic-skip-backward (substring c-stmt-delim-chars 1) nil t) | |
1206 | (or (bobp) | |
1207 | (memq (char-before) '(?{ ?})) | |
1208 | (save-excursion (backward-char) | |
1209 | (looking-at "\\s(")) | |
1210 | (c-crosses-statement-barrier-p (point) end))))) | |
785eecbb RS |
1211 | |
1212 | \f | |
d9e94c22 MS |
1213 | ;; A set of functions that covers various idiosyncrasies in |
1214 | ;; implementations of `forward-comment'. | |
1215 | ||
1216 | ;; Note: Some emacsen considers incorrectly that any line comment | |
1217 | ;; ending with a backslash continues to the next line. I can't think | |
1218 | ;; of any way to work around that in a reliable way without changing | |
1219 | ;; the buffer, though. Suggestions welcome. ;) (No, temporarily | |
1220 | ;; changing the syntax for backslash doesn't work since we must treat | |
1221 | ;; escapes in string literals correctly.) | |
1222 | ||
1223 | (defun c-forward-single-comment () | |
1224 | "Move forward past whitespace and the closest following comment, if any. | |
1225 | Return t if a comment was found, nil otherwise. In either case, the | |
1226 | point is moved past the following whitespace. Line continuations, | |
1227 | i.e. a backslashes followed by line breaks, are treated as whitespace. | |
1228 | The line breaks that end line comments are considered to be the | |
1229 | comment enders, so the point will be put on the beginning of the next | |
1230 | line if it moved past a line comment. | |
1231 | ||
1232 | This function does not do any hidden buffer changes." | |
1233 | ||
1234 | (let ((start (point))) | |
1235 | (when (looking-at "\\([ \t\n\r\f\v]\\|\\\\[\n\r]\\)+") | |
1236 | (goto-char (match-end 0))) | |
1237 | ||
1238 | (when (forward-comment 1) | |
1239 | (if (eobp) | |
1240 | ;; Some emacsen (e.g. XEmacs 21) return t when moving | |
1241 | ;; forwards at eob. | |
1242 | nil | |
1243 | ||
1244 | ;; Emacs includes the ending newline in a b-style (c++) | |
1245 | ;; comment, but XEmacs doesn't. We depend on the Emacs | |
1246 | ;; behavior (which also is symmetric). | |
1247 | (if (and (eolp) (elt (parse-partial-sexp start (point)) 7)) | |
1248 | (condition-case nil (forward-char 1))) | |
1249 | ||
1250 | t)))) | |
1251 | ||
1252 | (defsubst c-forward-comments () | |
1253 | "Move forward past all following whitespace and comments. | |
1254 | Line continuations, i.e. a backslashes followed by line breaks, are | |
1255 | treated as whitespace. | |
1256 | ||
0386b551 AM |
1257 | Note that this function might do hidden buffer changes. See the |
1258 | comment at the start of cc-engine.el for more info." | |
d9e94c22 MS |
1259 | |
1260 | (while (or | |
1261 | ;; If forward-comment in at least XEmacs 21 is given a large | |
1262 | ;; positive value, it'll loop all the way through if it hits | |
1263 | ;; eob. | |
1264 | (and (forward-comment 5) | |
1265 | ;; Some emacsen (e.g. XEmacs 21) return t when moving | |
1266 | ;; forwards at eob. | |
1267 | (not (eobp))) | |
1268 | ||
1269 | (when (looking-at "\\\\[\n\r]") | |
1270 | (forward-char 2) | |
1271 | t)))) | |
1272 | ||
1273 | (defun c-backward-single-comment () | |
1274 | "Move backward past whitespace and the closest preceding comment, if any. | |
1275 | Return t if a comment was found, nil otherwise. In either case, the | |
1276 | point is moved past the preceding whitespace. Line continuations, | |
1277 | i.e. a backslashes followed by line breaks, are treated as whitespace. | |
1278 | The line breaks that end line comments are considered to be the | |
1279 | comment enders, so the point cannot be at the end of the same line to | |
1280 | move over a line comment. | |
1281 | ||
1282 | This function does not do any hidden buffer changes." | |
1283 | ||
1284 | (let ((start (point))) | |
1285 | ;; When we got newline terminated comments, forward-comment in all | |
1286 | ;; supported emacsen so far will stop at eol of each line not | |
1287 | ;; ending with a comment when moving backwards. This corrects for | |
1288 | ;; that, and at the same time handles line continuations. | |
1289 | (while (progn | |
1290 | (skip-chars-backward " \t\n\r\f\v") | |
1291 | (and (looking-at "[\n\r]") | |
0386b551 | 1292 | (eq (char-before) ?\\))) |
d9e94c22 MS |
1293 | (backward-char)) |
1294 | ||
1295 | (if (bobp) | |
1296 | ;; Some emacsen (e.g. Emacs 19.34) return t when moving | |
1297 | ;; backwards at bob. | |
1298 | nil | |
1299 | ||
1300 | ;; Leave point after the closest following newline if we've | |
1301 | ;; backed up over any above, since forward-comment won't move | |
1302 | ;; backward over a line comment if point is at the end of the | |
1303 | ;; same line. | |
1304 | (re-search-forward "\\=\\s *[\n\r]" start t) | |
1305 | ||
1306 | (if (if (forward-comment -1) | |
1307 | (if (eolp) | |
1308 | ;; If forward-comment above succeeded and we're at eol | |
1309 | ;; then the newline we moved over above didn't end a | |
1310 | ;; line comment, so we give it another go. | |
1311 | (forward-comment -1) | |
1312 | t)) | |
1313 | ||
1314 | ;; Emacs <= 20 and XEmacs move back over the closer of a | |
1315 | ;; block comment that lacks an opener. | |
1316 | (if (looking-at "\\*/") | |
1317 | (progn (forward-char 2) nil) | |
1318 | t))))) | |
1319 | ||
1320 | (defsubst c-backward-comments () | |
1321 | "Move backward past all preceding whitespace and comments. | |
1322 | Line continuations, i.e. a backslashes followed by line breaks, are | |
1323 | treated as whitespace. The line breaks that end line comments are | |
1324 | considered to be the comment enders, so the point cannot be at the end | |
0386b551 AM |
1325 | of the same line to move over a line comment. Unlike |
1326 | c-backward-syntactic-ws, this function doesn't move back over | |
1327 | preprocessor directives. | |
d9e94c22 | 1328 | |
0386b551 AM |
1329 | Note that this function might do hidden buffer changes. See the |
1330 | comment at the start of cc-engine.el for more info." | |
d9e94c22 MS |
1331 | |
1332 | (let ((start (point))) | |
1333 | (while (and | |
0386b551 | 1334 | ;; `forward-comment' in some emacsen (e.g. XEmacs 21.4) |
d9e94c22 MS |
1335 | ;; return t when moving backwards at bob. |
1336 | (not (bobp)) | |
1337 | ||
1338 | (if (forward-comment -1) | |
1339 | (if (looking-at "\\*/") | |
1340 | ;; Emacs <= 20 and XEmacs move back over the | |
1341 | ;; closer of a block comment that lacks an opener. | |
1342 | (progn (forward-char 2) nil) | |
1343 | t) | |
1344 | ||
1345 | ;; XEmacs treats line continuations as whitespace but | |
1346 | ;; only in the backward direction, which seems a bit | |
1347 | ;; odd. Anyway, this is necessary for Emacs. | |
1348 | (when (and (looking-at "[\n\r]") | |
1349 | (eq (char-before) ?\\) | |
1350 | (< (point) start)) | |
1351 | (backward-char) | |
1352 | t)))))) | |
1353 | ||
1354 | \f | |
d9e94c22 | 1355 | ;; Tools for skipping over syntactic whitespace. |
a66cd3ee | 1356 | |
d9e94c22 MS |
1357 | ;; The following functions use text properties to cache searches over |
1358 | ;; large regions of syntactic whitespace. It works as follows: | |
1359 | ;; | |
1360 | ;; o If a syntactic whitespace region contains anything but simple | |
1361 | ;; whitespace (i.e. space, tab and line breaks), the text property | |
1362 | ;; `c-in-sws' is put over it. At places where we have stopped | |
1363 | ;; within that region there's also a `c-is-sws' text property. | |
1364 | ;; That since there typically are nested whitespace inside that | |
1365 | ;; must be handled separately, e.g. whitespace inside a comment or | |
1366 | ;; cpp directive. Thus, from one point with `c-is-sws' it's safe | |
1367 | ;; to jump to another point with that property within the same | |
1368 | ;; `c-in-sws' region. It can be likened to a ladder where | |
1369 | ;; `c-in-sws' marks the bars and `c-is-sws' the rungs. | |
1370 | ;; | |
1371 | ;; o The `c-is-sws' property is put on the simple whitespace chars at | |
1372 | ;; a "rung position" and also maybe on the first following char. | |
1373 | ;; As many characters as can be conveniently found in this range | |
1374 | ;; are marked, but no assumption can be made that the whole range | |
1375 | ;; is marked (it could be clobbered by later changes, for | |
1376 | ;; instance). | |
1377 | ;; | |
1378 | ;; Note that some part of the beginning of a sequence of simple | |
1379 | ;; whitespace might be part of the end of a preceding line comment | |
1380 | ;; or cpp directive and must not be considered part of the "rung". | |
1381 | ;; Such whitespace is some amount of horizontal whitespace followed | |
1382 | ;; by a newline. In the case of cpp directives it could also be | |
1383 | ;; two newlines with horizontal whitespace between them. | |
1384 | ;; | |
1385 | ;; The reason to include the first following char is to cope with | |
1386 | ;; "rung positions" that doesn't have any ordinary whitespace. If | |
1387 | ;; `c-is-sws' is put on a token character it does not have | |
1388 | ;; `c-in-sws' set simultaneously. That's the only case when that | |
1389 | ;; can occur, and the reason for not extending the `c-in-sws' | |
1390 | ;; region to cover it is that the `c-in-sws' region could then be | |
1391 | ;; accidentally merged with a following one if the token is only | |
1392 | ;; one character long. | |
1393 | ;; | |
1394 | ;; o On buffer changes the `c-in-sws' and `c-is-sws' properties are | |
1395 | ;; removed in the changed region. If the change was inside | |
1396 | ;; syntactic whitespace that means that the "ladder" is broken, but | |
1397 | ;; a later call to `c-forward-sws' or `c-backward-sws' will use the | |
1398 | ;; parts on either side and use an ordinary search only to "repair" | |
1399 | ;; the gap. | |
1400 | ;; | |
1401 | ;; Special care needs to be taken if a region is removed: If there | |
1402 | ;; are `c-in-sws' on both sides of it which do not connect inside | |
1403 | ;; the region then they can't be joined. If e.g. a marked macro is | |
1404 | ;; broken, syntactic whitespace inside the new text might be | |
1405 | ;; marked. If those marks would become connected with the old | |
1406 | ;; `c-in-sws' range around the macro then we could get a ladder | |
1407 | ;; with one end outside the macro and the other at some whitespace | |
1408 | ;; within it. | |
1409 | ;; | |
1410 | ;; The main motivation for this system is to increase the speed in | |
1411 | ;; skipping over the large whitespace regions that can occur at the | |
1412 | ;; top level in e.g. header files that contain a lot of comments and | |
1413 | ;; cpp directives. For small comments inside code it's probably | |
1414 | ;; slower than using `forward-comment' straightforwardly, but speed is | |
1415 | ;; not a significant factor there anyway. | |
1416 | ||
1417 | ; (defface c-debug-is-sws-face | |
1418 | ; '((t (:background "GreenYellow"))) | |
1419 | ; "Debug face to mark the `c-is-sws' property.") | |
1420 | ; (defface c-debug-in-sws-face | |
1421 | ; '((t (:underline t))) | |
1422 | ; "Debug face to mark the `c-in-sws' property.") | |
1423 | ||
1424 | ; (defun c-debug-put-sws-faces () | |
1425 | ; ;; Put the sws debug faces on all the `c-is-sws' and `c-in-sws' | |
1426 | ; ;; properties in the buffer. | |
1427 | ; (interactive) | |
1428 | ; (save-excursion | |
0386b551 | 1429 | ; (c-save-buffer-state (in-face) |
d9e94c22 MS |
1430 | ; (goto-char (point-min)) |
1431 | ; (setq in-face (if (get-text-property (point) 'c-is-sws) | |
1432 | ; (point))) | |
1433 | ; (while (progn | |
1434 | ; (goto-char (next-single-property-change | |
1435 | ; (point) 'c-is-sws nil (point-max))) | |
1436 | ; (if in-face | |
1437 | ; (progn | |
1438 | ; (c-debug-add-face in-face (point) 'c-debug-is-sws-face) | |
1439 | ; (setq in-face nil)) | |
1440 | ; (setq in-face (point))) | |
1441 | ; (not (eobp)))) | |
1442 | ; (goto-char (point-min)) | |
1443 | ; (setq in-face (if (get-text-property (point) 'c-in-sws) | |
1444 | ; (point))) | |
1445 | ; (while (progn | |
1446 | ; (goto-char (next-single-property-change | |
1447 | ; (point) 'c-in-sws nil (point-max))) | |
1448 | ; (if in-face | |
1449 | ; (progn | |
1450 | ; (c-debug-add-face in-face (point) 'c-debug-in-sws-face) | |
1451 | ; (setq in-face nil)) | |
1452 | ; (setq in-face (point))) | |
1453 | ; (not (eobp))))))) | |
1454 | ||
1455 | (defmacro c-debug-sws-msg (&rest args) | |
1456 | ;;`(message ,@args) | |
1457 | ) | |
1458 | ||
1459 | (defmacro c-put-is-sws (beg end) | |
0386b551 | 1460 | ;; This macro does a hidden buffer change. |
d9e94c22 MS |
1461 | `(let ((beg ,beg) (end ,end)) |
1462 | (put-text-property beg end 'c-is-sws t) | |
1463 | ,@(when (facep 'c-debug-is-sws-face) | |
1464 | `((c-debug-add-face beg end 'c-debug-is-sws-face))))) | |
1465 | ||
1466 | (defmacro c-put-in-sws (beg end) | |
0386b551 | 1467 | ;; This macro does a hidden buffer change. |
d9e94c22 MS |
1468 | `(let ((beg ,beg) (end ,end)) |
1469 | (put-text-property beg end 'c-in-sws t) | |
1470 | ,@(when (facep 'c-debug-is-sws-face) | |
1471 | `((c-debug-add-face beg end 'c-debug-in-sws-face))))) | |
1472 | ||
1473 | (defmacro c-remove-is-sws (beg end) | |
0386b551 | 1474 | ;; This macro does a hidden buffer change. |
d9e94c22 MS |
1475 | `(let ((beg ,beg) (end ,end)) |
1476 | (remove-text-properties beg end '(c-is-sws nil)) | |
1477 | ,@(when (facep 'c-debug-is-sws-face) | |
1478 | `((c-debug-remove-face beg end 'c-debug-is-sws-face))))) | |
1479 | ||
1480 | (defmacro c-remove-in-sws (beg end) | |
0386b551 | 1481 | ;; This macro does a hidden buffer change. |
d9e94c22 MS |
1482 | `(let ((beg ,beg) (end ,end)) |
1483 | (remove-text-properties beg end '(c-in-sws nil)) | |
1484 | ,@(when (facep 'c-debug-is-sws-face) | |
1485 | `((c-debug-remove-face beg end 'c-debug-in-sws-face))))) | |
1486 | ||
1487 | (defmacro c-remove-is-and-in-sws (beg end) | |
0386b551 | 1488 | ;; This macro does a hidden buffer change. |
d9e94c22 MS |
1489 | `(let ((beg ,beg) (end ,end)) |
1490 | (remove-text-properties beg end '(c-is-sws nil c-in-sws nil)) | |
1491 | ,@(when (facep 'c-debug-is-sws-face) | |
1492 | `((c-debug-remove-face beg end 'c-debug-is-sws-face) | |
1493 | (c-debug-remove-face beg end 'c-debug-in-sws-face))))) | |
1494 | ||
1495 | (defsubst c-invalidate-sws-region-after (beg end) | |
1496 | ;; Called from `after-change-functions'. Note that if | |
1497 | ;; `c-forward-sws' or `c-backward-sws' are used outside | |
1498 | ;; `c-save-buffer-state' or similar then this will remove the cache | |
1499 | ;; properties right after they're added. | |
0386b551 AM |
1500 | ;; |
1501 | ;; This function does hidden buffer changes. | |
d9e94c22 MS |
1502 | |
1503 | (save-excursion | |
1504 | ;; Adjust the end to remove the properties in any following simple | |
1505 | ;; ws up to and including the next line break, if there is any | |
1506 | ;; after the changed region. This is necessary e.g. when a rung | |
1507 | ;; marked empty line is converted to a line comment by inserting | |
1508 | ;; "//" before the line break. In that case the line break would | |
1509 | ;; keep the rung mark which could make a later `c-backward-sws' | |
1510 | ;; move into the line comment instead of over it. | |
1511 | (goto-char end) | |
1512 | (skip-chars-forward " \t\f\v") | |
1513 | (when (and (eolp) (not (eobp))) | |
1514 | (setq end (1+ (point))))) | |
1515 | ||
1516 | (when (and (= beg end) | |
1517 | (get-text-property beg 'c-in-sws) | |
2a15eb73 | 1518 | (> beg (point-min)) |
d9e94c22 MS |
1519 | (get-text-property (1- beg) 'c-in-sws)) |
1520 | ;; Ensure that an `c-in-sws' range gets broken. Note that it isn't | |
1521 | ;; safe to keep a range that was continuous before the change. E.g: | |
1522 | ;; | |
1523 | ;; #define foo | |
1524 | ;; \ | |
1525 | ;; bar | |
1526 | ;; | |
1527 | ;; There can be a "ladder" between "#" and "b". Now, if the newline | |
1528 | ;; after "foo" is removed then "bar" will become part of the cpp | |
1529 | ;; directive instead of a syntactically relevant token. In that | |
1530 | ;; case there's no longer syntactic ws from "#" to "b". | |
1531 | (setq beg (1- beg))) | |
1532 | ||
1533 | (c-debug-sws-msg "c-invalidate-sws-region-after [%s..%s]" beg end) | |
1534 | (c-remove-is-and-in-sws beg end)) | |
1535 | ||
1536 | (defun c-forward-sws () | |
1537 | ;; Used by `c-forward-syntactic-ws' to implement the unbounded search. | |
0386b551 AM |
1538 | ;; |
1539 | ;; This function might do hidden buffer changes. | |
d9e94c22 MS |
1540 | |
1541 | (let (;; `rung-pos' is set to a position as early as possible in the | |
1542 | ;; unmarked part of the simple ws region. | |
1543 | (rung-pos (point)) next-rung-pos rung-end-pos last-put-in-sws-pos | |
1544 | rung-is-marked next-rung-is-marked simple-ws-end | |
1545 | ;; `safe-start' is set when it's safe to cache the start position. | |
1546 | ;; It's not set if we've initially skipped over comments and line | |
1547 | ;; continuations since we might have gone out through the end of a | |
1548 | ;; macro then. This provision makes `c-forward-sws' not populate the | |
1549 | ;; cache in the majority of cases, but otoh is `c-backward-sws' by far | |
1550 | ;; more common. | |
1551 | safe-start) | |
1552 | ||
1553 | ;; Skip simple ws and do a quick check on the following character to see | |
1554 | ;; if it's anything that can't start syntactic ws, so we can bail out | |
1555 | ;; early in the majority of cases when there just are a few ws chars. | |
1556 | (skip-chars-forward " \t\n\r\f\v") | |
1557 | (when (looking-at c-syntactic-ws-start) | |
1558 | ||
1559 | (setq rung-end-pos (min (1+ (point)) (point-max))) | |
1560 | (if (setq rung-is-marked (text-property-any rung-pos rung-end-pos | |
1561 | 'c-is-sws t)) | |
1562 | ;; Find the last rung position to avoid setting properties in all | |
1563 | ;; the cases when the marked rung is complete. | |
1564 | ;; (`next-single-property-change' is certain to move at least one | |
1565 | ;; step forward.) | |
1566 | (setq rung-pos (1- (next-single-property-change | |
1567 | rung-is-marked 'c-is-sws nil rung-end-pos))) | |
1568 | ;; Got no marked rung here. Since the simple ws might have started | |
1569 | ;; inside a line comment or cpp directive we must set `rung-pos' as | |
1570 | ;; high as possible. | |
1571 | (setq rung-pos (point))) | |
1572 | ||
1573 | (while | |
1574 | (progn | |
1575 | (while | |
1576 | (when (and rung-is-marked | |
1577 | (get-text-property (point) 'c-in-sws)) | |
1578 | ||
1579 | ;; The following search is the main reason that `c-in-sws' | |
1580 | ;; and `c-is-sws' aren't combined to one property. | |
1581 | (goto-char (next-single-property-change | |
1582 | (point) 'c-in-sws nil (point-max))) | |
1583 | (unless (get-text-property (point) 'c-is-sws) | |
1584 | ;; If the `c-in-sws' region extended past the last | |
1585 | ;; `c-is-sws' char we have to go back a bit. | |
1586 | (or (get-text-property (1- (point)) 'c-is-sws) | |
1587 | (goto-char (previous-single-property-change | |
1588 | (point) 'c-is-sws))) | |
1589 | (backward-char)) | |
1590 | ||
1591 | (c-debug-sws-msg | |
1592 | "c-forward-sws cached move %s -> %s (max %s)" | |
1593 | rung-pos (point) (point-max)) | |
1594 | ||
1595 | (setq rung-pos (point)) | |
1596 | (and (> (skip-chars-forward " \t\n\r\f\v") 0) | |
1597 | (not (eobp)))) | |
1598 | ||
1599 | ;; We'll loop here if there is simple ws after the last rung. | |
1600 | ;; That means that there's been some change in it and it's | |
1601 | ;; possible that we've stepped into another ladder, so extend | |
1602 | ;; the previous one to join with it if there is one, and try to | |
1603 | ;; use the cache again. | |
1604 | (c-debug-sws-msg | |
1605 | "c-forward-sws extending rung with [%s..%s] (max %s)" | |
1606 | (1+ rung-pos) (1+ (point)) (point-max)) | |
1607 | (unless (get-text-property (point) 'c-is-sws) | |
1608 | ;; Remove any `c-in-sws' property from the last char of | |
1609 | ;; the rung before we mark it with `c-is-sws', so that we | |
1610 | ;; won't connect with the remains of a broken "ladder". | |
1611 | (c-remove-in-sws (point) (1+ (point)))) | |
1612 | (c-put-is-sws (1+ rung-pos) | |
1613 | (1+ (point))) | |
1614 | (c-put-in-sws rung-pos | |
1615 | (setq rung-pos (point) | |
1616 | last-put-in-sws-pos rung-pos))) | |
1617 | ||
1618 | (setq simple-ws-end (point)) | |
1619 | (c-forward-comments) | |
1620 | ||
1621 | (cond | |
1622 | ((/= (point) simple-ws-end) | |
1623 | ;; Skipped over comments. Don't cache at eob in case the buffer | |
1624 | ;; is narrowed. | |
1625 | (not (eobp))) | |
1626 | ||
1627 | ((save-excursion | |
1628 | (and c-opt-cpp-prefix | |
1629 | (looking-at c-opt-cpp-start) | |
1630 | (progn (skip-chars-backward " \t") | |
1631 | (bolp)) | |
1632 | (or (bobp) | |
1633 | (progn (backward-char) | |
1634 | (not (eq (char-before) ?\\)))))) | |
1635 | ;; Skip a preprocessor directive. | |
1636 | (end-of-line) | |
1637 | (while (and (eq (char-before) ?\\) | |
1638 | (= (forward-line 1) 0)) | |
1639 | (end-of-line)) | |
1640 | (forward-line 1) | |
1641 | (setq safe-start t) | |
1642 | ;; Don't cache at eob in case the buffer is narrowed. | |
1643 | (not (eobp))))) | |
1644 | ||
1645 | ;; We've searched over a piece of non-white syntactic ws. See if this | |
1646 | ;; can be cached. | |
1647 | (setq next-rung-pos (point)) | |
1648 | (skip-chars-forward " \t\n\r\f\v") | |
1649 | (setq rung-end-pos (min (1+ (point)) (point-max))) | |
1650 | ||
1651 | (if (or | |
1652 | ;; Cache if we haven't skipped comments only, and if we started | |
1653 | ;; either from a marked rung or from a completely uncached | |
1654 | ;; position. | |
1655 | (and safe-start | |
1656 | (or rung-is-marked | |
1657 | (not (get-text-property simple-ws-end 'c-in-sws)))) | |
1658 | ||
1659 | ;; See if there's a marked rung in the encountered simple ws. If | |
1660 | ;; so then we can cache, unless `safe-start' is nil. Even then | |
1661 | ;; we need to do this to check if the cache can be used for the | |
1662 | ;; next step. | |
1663 | (and (setq next-rung-is-marked | |
1664 | (text-property-any next-rung-pos rung-end-pos | |
1665 | 'c-is-sws t)) | |
1666 | safe-start)) | |
b2acd789 | 1667 | |
0ec8351b | 1668 | (progn |
d9e94c22 MS |
1669 | (c-debug-sws-msg |
1670 | "c-forward-sws caching [%s..%s] - [%s..%s] (max %s)" | |
1671 | rung-pos (1+ simple-ws-end) next-rung-pos rung-end-pos | |
1672 | (point-max)) | |
1673 | ||
1674 | ;; Remove the properties for any nested ws that might be cached. | |
1675 | ;; Only necessary for `c-is-sws' since `c-in-sws' will be set | |
1676 | ;; anyway. | |
1677 | (c-remove-is-sws (1+ simple-ws-end) next-rung-pos) | |
1678 | (unless (and rung-is-marked (= rung-pos simple-ws-end)) | |
1679 | (c-put-is-sws rung-pos | |
1680 | (1+ simple-ws-end)) | |
1681 | (setq rung-is-marked t)) | |
1682 | (c-put-in-sws rung-pos | |
1683 | (setq rung-pos (point) | |
1684 | last-put-in-sws-pos rung-pos)) | |
1685 | (unless (get-text-property (1- rung-end-pos) 'c-is-sws) | |
1686 | ;; Remove any `c-in-sws' property from the last char of | |
1687 | ;; the rung before we mark it with `c-is-sws', so that we | |
1688 | ;; won't connect with the remains of a broken "ladder". | |
1689 | (c-remove-in-sws (1- rung-end-pos) rung-end-pos)) | |
1690 | (c-put-is-sws next-rung-pos | |
1691 | rung-end-pos)) | |
1692 | ||
1693 | (c-debug-sws-msg | |
1694 | "c-forward-sws not caching [%s..%s] - [%s..%s] (max %s)" | |
1695 | rung-pos (1+ simple-ws-end) next-rung-pos rung-end-pos | |
1696 | (point-max)) | |
1697 | ||
1698 | ;; Set `rung-pos' for the next rung. It's the same thing here as | |
1699 | ;; initially, except that the rung position is set as early as | |
1700 | ;; possible since we can't be in the ending ws of a line comment or | |
1701 | ;; cpp directive now. | |
1702 | (if (setq rung-is-marked next-rung-is-marked) | |
1703 | (setq rung-pos (1- (next-single-property-change | |
1704 | rung-is-marked 'c-is-sws nil rung-end-pos))) | |
1705 | (setq rung-pos next-rung-pos)) | |
1706 | (setq safe-start t))) | |
1707 | ||
1708 | ;; Make sure that the newly marked `c-in-sws' region doesn't connect to | |
1709 | ;; another one after the point (which might occur when editing inside a | |
1710 | ;; comment or macro). | |
1711 | (when (eq last-put-in-sws-pos (point)) | |
1712 | (cond ((< last-put-in-sws-pos (point-max)) | |
1713 | (c-debug-sws-msg | |
1714 | "c-forward-sws clearing at %s for cache separation" | |
1715 | last-put-in-sws-pos) | |
1716 | (c-remove-in-sws last-put-in-sws-pos | |
1717 | (1+ last-put-in-sws-pos))) | |
1718 | (t | |
1719 | ;; If at eob we have to clear the last character before the end | |
1720 | ;; instead since the buffer might be narrowed and there might | |
1721 | ;; be a `c-in-sws' after (point-max). In this case it's | |
1722 | ;; necessary to clear both properties. | |
1723 | (c-debug-sws-msg | |
1724 | "c-forward-sws clearing thoroughly at %s for cache separation" | |
1725 | (1- last-put-in-sws-pos)) | |
1726 | (c-remove-is-and-in-sws (1- last-put-in-sws-pos) | |
1727 | last-put-in-sws-pos)))) | |
1728 | ))) | |
b2acd789 | 1729 | |
d9e94c22 MS |
1730 | (defun c-backward-sws () |
1731 | ;; Used by `c-backward-syntactic-ws' to implement the unbounded search. | |
0386b551 AM |
1732 | ;; |
1733 | ;; This function might do hidden buffer changes. | |
d9e94c22 MS |
1734 | |
1735 | (let (;; `rung-pos' is set to a position as late as possible in the unmarked | |
1736 | ;; part of the simple ws region. | |
1737 | (rung-pos (point)) next-rung-pos last-put-in-sws-pos | |
1738 | rung-is-marked simple-ws-beg cmt-skip-pos) | |
1739 | ||
1740 | ;; Skip simple horizontal ws and do a quick check on the preceding | |
1741 | ;; character to see if it's anying that can't end syntactic ws, so we can | |
1742 | ;; bail out early in the majority of cases when there just are a few ws | |
1743 | ;; chars. Newlines are complicated in the backward direction, so we can't | |
1744 | ;; skip over them. | |
1745 | (skip-chars-backward " \t\f") | |
1746 | (when (and (not (bobp)) | |
1747 | (save-excursion | |
1748 | (backward-char) | |
1749 | (looking-at c-syntactic-ws-end))) | |
1750 | ||
1751 | ;; Try to find a rung position in the simple ws preceding point, so that | |
1752 | ;; we can get a cache hit even if the last bit of the simple ws has | |
1753 | ;; changed recently. | |
1754 | (setq simple-ws-beg (point)) | |
1755 | (skip-chars-backward " \t\n\r\f\v") | |
1756 | (if (setq rung-is-marked (text-property-any | |
1757 | (point) (min (1+ rung-pos) (point-max)) | |
1758 | 'c-is-sws t)) | |
1759 | ;; `rung-pos' will be the earliest marked position, which means that | |
1760 | ;; there might be later unmarked parts in the simple ws region. | |
1761 | ;; It's not worth the effort to fix that; the last part of the | |
1762 | ;; simple ws is also typically edited often, so it could be wasted. | |
1763 | (goto-char (setq rung-pos rung-is-marked)) | |
1764 | (goto-char simple-ws-beg)) | |
1765 | ||
1766 | (while | |
1767 | (progn | |
1768 | (while | |
1769 | (when (and rung-is-marked | |
1770 | (not (bobp)) | |
1771 | (get-text-property (1- (point)) 'c-in-sws)) | |
1772 | ||
1773 | ;; The following search is the main reason that `c-in-sws' | |
1774 | ;; and `c-is-sws' aren't combined to one property. | |
1775 | (goto-char (previous-single-property-change | |
1776 | (point) 'c-in-sws nil (point-min))) | |
1777 | (unless (get-text-property (point) 'c-is-sws) | |
1778 | ;; If the `c-in-sws' region extended past the first | |
1779 | ;; `c-is-sws' char we have to go forward a bit. | |
1780 | (goto-char (next-single-property-change | |
1781 | (point) 'c-is-sws))) | |
1782 | ||
1783 | (c-debug-sws-msg | |
1784 | "c-backward-sws cached move %s <- %s (min %s)" | |
1785 | (point) rung-pos (point-min)) | |
1786 | ||
1787 | (setq rung-pos (point)) | |
1788 | (if (and (< (min (skip-chars-backward " \t\f\v") | |
1789 | (progn | |
1790 | (setq simple-ws-beg (point)) | |
1791 | (skip-chars-backward " \t\n\r\f\v"))) | |
1792 | 0) | |
1793 | (setq rung-is-marked | |
1794 | (text-property-any (point) rung-pos | |
1795 | 'c-is-sws t))) | |
1796 | t | |
1797 | (goto-char simple-ws-beg) | |
1798 | nil)) | |
1799 | ||
1800 | ;; We'll loop here if there is simple ws before the first rung. | |
1801 | ;; That means that there's been some change in it and it's | |
1802 | ;; possible that we've stepped into another ladder, so extend | |
1803 | ;; the previous one to join with it if there is one, and try to | |
1804 | ;; use the cache again. | |
1805 | (c-debug-sws-msg | |
1806 | "c-backward-sws extending rung with [%s..%s] (min %s)" | |
1807 | rung-is-marked rung-pos (point-min)) | |
1808 | (unless (get-text-property (1- rung-pos) 'c-is-sws) | |
1809 | ;; Remove any `c-in-sws' property from the last char of | |
1810 | ;; the rung before we mark it with `c-is-sws', so that we | |
1811 | ;; won't connect with the remains of a broken "ladder". | |
1812 | (c-remove-in-sws (1- rung-pos) rung-pos)) | |
1813 | (c-put-is-sws rung-is-marked | |
1814 | rung-pos) | |
1815 | (c-put-in-sws rung-is-marked | |
1816 | (1- rung-pos)) | |
1817 | (setq rung-pos rung-is-marked | |
1818 | last-put-in-sws-pos rung-pos)) | |
1819 | ||
1820 | (c-backward-comments) | |
1821 | (setq cmt-skip-pos (point)) | |
a66cd3ee | 1822 | |
d9e94c22 MS |
1823 | (cond |
1824 | ((and c-opt-cpp-prefix | |
1825 | (/= cmt-skip-pos simple-ws-beg) | |
1826 | (c-beginning-of-macro)) | |
1827 | ;; Inside a cpp directive. See if it should be skipped over. | |
1828 | (let ((cpp-beg (point))) | |
1829 | ||
1830 | ;; Move back over all line continuations in the region skipped | |
1831 | ;; over by `c-backward-comments'. If we go past it then we | |
1832 | ;; started inside the cpp directive. | |
1833 | (goto-char simple-ws-beg) | |
1834 | (beginning-of-line) | |
1835 | (while (and (> (point) cmt-skip-pos) | |
1836 | (progn (backward-char) | |
1837 | (eq (char-before) ?\\))) | |
1838 | (beginning-of-line)) | |
1839 | ||
1840 | (if (< (point) cmt-skip-pos) | |
1841 | ;; Don't move past the cpp directive if we began inside | |
1842 | ;; it. Note that the position at the end of the last line | |
1843 | ;; of the macro is also considered to be within it. | |
1844 | (progn (goto-char cmt-skip-pos) | |
1845 | nil) | |
1846 | ||
1847 | ;; It's worthwhile to spend a little bit of effort on finding | |
1848 | ;; the end of the macro, to get a good `simple-ws-beg' | |
1849 | ;; position for the cache. Note that `c-backward-comments' | |
1850 | ;; could have stepped over some comments before going into | |
1851 | ;; the macro, and then `simple-ws-beg' must be kept on the | |
1852 | ;; same side of those comments. | |
1853 | (goto-char simple-ws-beg) | |
1854 | (skip-chars-backward " \t\n\r\f\v") | |
1855 | (if (eq (char-before) ?\\) | |
1856 | (forward-char)) | |
1857 | (forward-line 1) | |
1858 | (if (< (point) simple-ws-beg) | |
1859 | ;; Might happen if comments after the macro were skipped | |
1860 | ;; over. | |
1861 | (setq simple-ws-beg (point))) | |
1862 | ||
1863 | (goto-char cpp-beg) | |
1864 | t))) | |
1865 | ||
1866 | ((/= (save-excursion | |
1867 | (skip-chars-forward " \t\n\r\f\v" simple-ws-beg) | |
1868 | (setq next-rung-pos (point))) | |
1869 | simple-ws-beg) | |
1870 | ;; Skipped over comments. Must put point at the end of | |
1871 | ;; the simple ws at point since we might be after a line | |
1872 | ;; comment or cpp directive that's been partially | |
1873 | ;; narrowed out, and we can't risk marking the simple ws | |
1874 | ;; at the end of it. | |
1875 | (goto-char next-rung-pos) | |
1876 | t))) | |
1877 | ||
1878 | ;; We've searched over a piece of non-white syntactic ws. See if this | |
1879 | ;; can be cached. | |
1880 | (setq next-rung-pos (point)) | |
1881 | (skip-chars-backward " \t\f\v") | |
1882 | ||
1883 | (if (or | |
1884 | ;; Cache if we started either from a marked rung or from a | |
1885 | ;; completely uncached position. | |
1886 | rung-is-marked | |
1887 | (not (get-text-property (1- simple-ws-beg) 'c-in-sws)) | |
1888 | ||
1889 | ;; Cache if there's a marked rung in the encountered simple ws. | |
1890 | (save-excursion | |
1891 | (skip-chars-backward " \t\n\r\f\v") | |
1892 | (text-property-any (point) (min (1+ next-rung-pos) (point-max)) | |
1893 | 'c-is-sws t))) | |
a66cd3ee | 1894 | |
d9e94c22 MS |
1895 | (progn |
1896 | (c-debug-sws-msg | |
1897 | "c-backward-sws caching [%s..%s] - [%s..%s] (min %s)" | |
1898 | (point) (1+ next-rung-pos) | |
1899 | simple-ws-beg (min (1+ rung-pos) (point-max)) | |
1900 | (point-min)) | |
1901 | ||
1902 | ;; Remove the properties for any nested ws that might be cached. | |
1903 | ;; Only necessary for `c-is-sws' since `c-in-sws' will be set | |
1904 | ;; anyway. | |
1905 | (c-remove-is-sws (1+ next-rung-pos) simple-ws-beg) | |
1906 | (unless (and rung-is-marked (= simple-ws-beg rung-pos)) | |
1907 | (let ((rung-end-pos (min (1+ rung-pos) (point-max)))) | |
1908 | (unless (get-text-property (1- rung-end-pos) 'c-is-sws) | |
1909 | ;; Remove any `c-in-sws' property from the last char of | |
1910 | ;; the rung before we mark it with `c-is-sws', so that we | |
1911 | ;; won't connect with the remains of a broken "ladder". | |
1912 | (c-remove-in-sws (1- rung-end-pos) rung-end-pos)) | |
1913 | (c-put-is-sws simple-ws-beg | |
1914 | rung-end-pos) | |
1915 | (setq rung-is-marked t))) | |
1916 | (c-put-in-sws (setq simple-ws-beg (point) | |
1917 | last-put-in-sws-pos simple-ws-beg) | |
1918 | rung-pos) | |
1919 | (c-put-is-sws (setq rung-pos simple-ws-beg) | |
1920 | (1+ next-rung-pos))) | |
1921 | ||
1922 | (c-debug-sws-msg | |
1923 | "c-backward-sws not caching [%s..%s] - [%s..%s] (min %s)" | |
1924 | (point) (1+ next-rung-pos) | |
1925 | simple-ws-beg (min (1+ rung-pos) (point-max)) | |
1926 | (point-min)) | |
1927 | (setq rung-pos next-rung-pos | |
1928 | simple-ws-beg (point)) | |
1929 | )) | |
1930 | ||
1931 | ;; Make sure that the newly marked `c-in-sws' region doesn't connect to | |
1932 | ;; another one before the point (which might occur when editing inside a | |
1933 | ;; comment or macro). | |
1934 | (when (eq last-put-in-sws-pos (point)) | |
1935 | (cond ((< (point-min) last-put-in-sws-pos) | |
1936 | (c-debug-sws-msg | |
1937 | "c-backward-sws clearing at %s for cache separation" | |
1938 | (1- last-put-in-sws-pos)) | |
1939 | (c-remove-in-sws (1- last-put-in-sws-pos) | |
1940 | last-put-in-sws-pos)) | |
1941 | ((> (point-min) 1) | |
1942 | ;; If at bob and the buffer is narrowed, we have to clear the | |
1943 | ;; character we're standing on instead since there might be a | |
1944 | ;; `c-in-sws' before (point-min). In this case it's necessary | |
1945 | ;; to clear both properties. | |
1946 | (c-debug-sws-msg | |
1947 | "c-backward-sws clearing thoroughly at %s for cache separation" | |
1948 | last-put-in-sws-pos) | |
1949 | (c-remove-is-and-in-sws last-put-in-sws-pos | |
1950 | (1+ last-put-in-sws-pos))))) | |
1951 | ))) | |
785eecbb | 1952 | |
d9e94c22 | 1953 | \f |
580fba94 AM |
1954 | ;; Other whitespace tools |
1955 | (defun c-partial-ws-p (beg end) | |
1956 | ;; Is the region (beg end) WS, and is there WS (or BOB/EOB) next to the | |
1957 | ;; region? This is a "heuristic" function. ..... | |
b414f371 | 1958 | ;; |
c421028a AM |
1959 | ;; The motivation for the second bit is to check whether removing this |
1960 | ;; region would coalesce two symbols. | |
580fba94 AM |
1961 | ;; |
1962 | ;; FIXME!!! This function doesn't check virtual semicolons in any way. Be | |
1963 | ;; careful about using this function for, e.g. AWK. (2007/3/7) | |
1964 | (save-excursion | |
1965 | (let ((end+1 (min (1+ end) (point-max)))) | |
1966 | (or (progn (goto-char (max (point-min) (1- beg))) | |
1967 | (c-skip-ws-forward end) | |
1968 | (eq (point) end)) | |
1969 | (progn (goto-char beg) | |
1970 | (c-skip-ws-forward end+1) | |
1971 | (eq (point) end+1)))))) | |
1972 | \f | |
0386b551 | 1973 | ;; A system for finding noteworthy parens before the point. |
e1c458ae | 1974 | |
0ec1d2c5 AM |
1975 | (defconst c-state-cache-too-far 5000) |
1976 | ;; A maximum comfortable scanning distance, e.g. between | |
1977 | ;; `c-state-cache-good-pos' and "HERE" (where we call c-parse-state). When | |
1978 | ;; this distance is exceeded, we take "emergency meausures", e.g. by clearing | |
1979 | ;; the cache and starting again from point-min or a beginning of defun. This | |
1980 | ;; value can be tuned for efficiency or set to a lower value for testing. | |
1981 | ||
d9e94c22 MS |
1982 | (defvar c-state-cache nil) |
1983 | (make-variable-buffer-local 'c-state-cache) | |
1984 | ;; The state cache used by `c-parse-state' to cut down the amount of | |
0ec1d2c5 AM |
1985 | ;; searching. It's the result from some earlier `c-parse-state' call. See |
1986 | ;; `c-parse-state''s doc string for details of its structure. | |
0386b551 | 1987 | ;; |
d9e94c22 MS |
1988 | ;; The use of the cached info is more effective if the next |
1989 | ;; `c-parse-state' call is on a line close by the one the cached state | |
1990 | ;; was made at; the cache can actually slow down a little if the | |
1991 | ;; cached state was made very far back in the buffer. The cache is | |
1992 | ;; most effective if `c-parse-state' is used on each line while moving | |
1993 | ;; forward. | |
e1c458ae | 1994 | |
0386b551 AM |
1995 | (defvar c-state-cache-good-pos 1) |
1996 | (make-variable-buffer-local 'c-state-cache-good-pos) | |
0ec1d2c5 AM |
1997 | ;; This is a position where `c-state-cache' is known to be correct, or |
1998 | ;; nil (see below). It's a position inside one of the recorded unclosed | |
1999 | ;; parens or the top level, but not further nested inside any literal or | |
2000 | ;; subparen that is closed before the last recorded position. | |
0386b551 AM |
2001 | ;; |
2002 | ;; The exact position is chosen to try to be close to yet earlier than | |
2003 | ;; the position where `c-state-cache' will be called next. Right now | |
2004 | ;; the heuristic is to set it to the position after the last found | |
2005 | ;; closing paren (of any type) before the line on which | |
2006 | ;; `c-parse-state' was called. That is chosen primarily to work well | |
2007 | ;; with refontification of the current line. | |
0ec1d2c5 AM |
2008 | ;; |
2009 | ;; 2009-07-28: When `c-state-point-min' and the last position where | |
2010 | ;; `c-parse-state' or for which `c-invalidate-state-cache' was called, are | |
2011 | ;; both in the same literal, there is no such "good position", and | |
2012 | ;; c-state-cache-good-pos is then nil. This is the ONLY circumstance in which | |
2013 | ;; it can be nil. In this case, `c-state-point-min-literal' will be non-nil. | |
2014 | ;; | |
2015 | ;; 2009-06-12: In a brace desert, c-state-cache-good-pos may also be in | |
2016 | ;; the middle of the desert, as long as it is not within a brace pair | |
2017 | ;; recorded in `c-state-cache' or a paren/bracket pair. | |
2018 | ||
2019 | ||
2020 | ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; | |
2021 | ;; We maintain a simple cache of positions which aren't in a literal, so as to | |
2022 | ;; speed up testing for non-literality. | |
2023 | (defconst c-state-nonlit-pos-interval 10000) | |
2024 | ;; The approximate interval between entries in `c-state-nonlit-pos-cache'. | |
2025 | ||
2026 | (defvar c-state-nonlit-pos-cache nil) | |
2027 | (make-variable-buffer-local 'c-state-nonlit-pos-cache) | |
2028 | ;; A list of buffer positions which are known not to be in a literal. This is | |
2029 | ;; ordered with higher positions at the front of the list. Only those which | |
2030 | ;; are less than `c-state-nonlit-pos-cache-limit' are valid. | |
2031 | ||
2032 | (defvar c-state-nonlit-pos-cache-limit 1) | |
2033 | (make-variable-buffer-local 'c-state-nonlit-pos-cache-limit) | |
2034 | ;; An upper limit on valid entries in `c-state-nonlit-pos-cache'. This is | |
2035 | ;; reduced by buffer changes, and increased by invocations of | |
2036 | ;; `c-state-literal-at'. | |
2037 | ||
2038 | (defsubst c-state-pp-to-literal (from to) | |
2039 | ;; Do a parse-partial-sexp from FROM to TO, returning the bounds of any | |
2040 | ;; literal at TO as a cons, otherwise NIL. | |
2041 | ;; FROM must not be in a literal, and the buffer should already be wide | |
2042 | ;; enough. | |
2043 | (save-excursion | |
2044 | (let ((s (parse-partial-sexp from to))) | |
2045 | (when (or (nth 3 s) (nth 4 s)) ; in a string or comment | |
2046 | (parse-partial-sexp (point) (point-max) | |
2047 | nil ; TARGETDEPTH | |
2048 | nil ; STOPBEFORE | |
2049 | s ; OLDSTATE | |
2050 | 'syntax-table) ; stop at end of literal | |
2051 | (cons (nth 8 s) (point)))))) | |
2052 | ||
2053 | (defun c-state-literal-at (here) | |
2054 | ;; If position HERE is inside a literal, return (START . END), the | |
2055 | ;; boundaries of the literal (which may be outside the accessible bit of the | |
2056 | ;; buffer). Otherwise, return nil. | |
2057 | ;; | |
2058 | ;; This function is almost the same as `c-literal-limits'. It differs in | |
2059 | ;; that it is a lower level function, and that it rigourously follows the | |
2060 | ;; syntax from BOB, whereas `c-literal-limits' uses a "local" safe position. | |
2061 | (save-restriction | |
2062 | (widen) | |
2063 | (save-excursion | |
2064 | (let ((c c-state-nonlit-pos-cache) | |
2065 | pos npos lit) | |
2066 | ;; Trim the cache to take account of buffer changes. | |
2067 | (while (and c (> (car c) c-state-nonlit-pos-cache-limit)) | |
2068 | (setq c (cdr c))) | |
2069 | (setq c-state-nonlit-pos-cache c) | |
2070 | ||
2071 | (while (and c (> (car c) here)) | |
2072 | (setq c (cdr c))) | |
2073 | (setq pos (or (car c) (point-min))) | |
2074 | ||
2075 | (while (<= (setq npos (+ pos c-state-nonlit-pos-interval)) | |
2076 | here) | |
2077 | (setq lit (c-state-pp-to-literal pos npos)) | |
2078 | (setq pos (or (cdr lit) npos)) ; end of literal containing npos. | |
2079 | (setq c-state-nonlit-pos-cache (cons pos c-state-nonlit-pos-cache))) | |
2080 | ||
2081 | (if (> pos c-state-nonlit-pos-cache-limit) | |
2082 | (setq c-state-nonlit-pos-cache-limit pos)) | |
2083 | (if (< pos here) | |
2084 | (setq lit (c-state-pp-to-literal pos here))) | |
2085 | lit)))) | |
2086 | ||
2087 | (defsubst c-state-lit-beg (pos) | |
2088 | ;; Return the start of the literal containing POS, or POS itself. | |
2089 | (or (car (c-state-literal-at pos)) | |
2090 | pos)) | |
2091 | ||
2092 | (defsubst c-state-cache-non-literal-place (pos state) | |
2093 | ;; Return a position outside of a string/comment at or before POS. | |
2094 | ;; STATE is the parse-partial-sexp state at POS. | |
2095 | (if (or (nth 3 state) ; in a string? | |
2096 | (nth 4 state)) ; in a comment? | |
2097 | (nth 8 state) | |
2098 | pos)) | |
2099 | ||
2100 | ||
2101 | ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; | |
2102 | ;; Stuff to do with point-min, and coping with any literal there. | |
2103 | (defvar c-state-point-min 1) | |
2104 | (make-variable-buffer-local 'c-state-point-min) | |
2105 | ;; This is (point-min) when `c-state-cache' was last calculated. A change of | |
2106 | ;; narrowing is likely to affect the parens that are visible before the point. | |
2107 | ||
2108 | (defvar c-state-point-min-lit-type nil) | |
2109 | (make-variable-buffer-local 'c-state-point-min-lit-type) | |
2110 | (defvar c-state-point-min-lit-start nil) | |
2111 | (make-variable-buffer-local 'c-state-point-min-lit-start) | |
2112 | ;; These two variables define the literal, if any, containing point-min. | |
2113 | ;; Their values are, respectively, 'string, c, or c++, and the start of the | |
2114 | ;; literal. If there's no literal there, they're both nil. | |
2115 | ||
2116 | (defvar c-state-min-scan-pos 1) | |
2117 | (make-variable-buffer-local 'c-state-min-scan-pos) | |
2118 | ;; This is the earliest buffer-pos from which scanning can be done. It is | |
2119 | ;; either the end of the literal containing point-min, or point-min itself. | |
2120 | ;; It becomes nil if the buffer is changed earlier than this point. | |
2121 | (defun c-state-get-min-scan-pos () | |
2122 | ;; Return the lowest valid scanning pos. This will be the end of the | |
2123 | ;; literal enclosing point-min, or point-min itself. | |
2124 | (or c-state-min-scan-pos | |
2125 | (save-restriction | |
2126 | (save-excursion | |
2127 | (widen) | |
2128 | (goto-char c-state-point-min-lit-start) | |
2129 | (if (eq c-state-point-min-lit-type 'string) | |
2130 | (forward-sexp) | |
2131 | (forward-comment 1)) | |
2132 | (setq c-state-min-scan-pos (point)))))) | |
2133 | ||
2134 | (defun c-state-mark-point-min-literal () | |
2135 | ;; Determine the properties of any literal containing POINT-MIN, setting the | |
2136 | ;; variables `c-state-point-min-lit-type', `c-state-point-min-lit-start', | |
2137 | ;; and `c-state-min-scan-pos' accordingly. The return value is meaningless. | |
2138 | (let ((p-min (point-min)) | |
2139 | lit) | |
2140 | (save-restriction | |
2141 | (widen) | |
2142 | (setq lit (c-state-literal-at p-min)) | |
2143 | (if lit | |
2144 | (setq c-state-point-min-lit-type | |
2145 | (save-excursion | |
2146 | (goto-char (car lit)) | |
2147 | (cond | |
2148 | ((looking-at c-block-comment-start-regexp) 'c) | |
2149 | ((looking-at c-line-comment-starter) 'c++) | |
2150 | (t 'string))) | |
2151 | c-state-point-min-lit-start (car lit) | |
2152 | c-state-min-scan-pos (cdr lit)) | |
2153 | (setq c-state-point-min-lit-type nil | |
2154 | c-state-point-min-lit-start nil | |
2155 | c-state-min-scan-pos p-min))))) | |
2156 | ||
2157 | ||
2158 | ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; | |
2159 | ;; A variable which signals a brace dessert - helpful for reducing the number | |
2160 | ;; of fruitless backward scans. | |
2161 | (defvar c-state-brace-pair-desert nil) | |
2162 | (make-variable-buffer-local 'c-state-brace-pair-desert) | |
2163 | ;; Used only in `c-append-lower-brace-pair-to-state-cache'. It is set when an | |
2164 | ;; that defun has searched backwards for a brace pair and not found one. Its | |
2165 | ;; value is either nil or a cons (PA . FROM), where PA is the position of the | |
2166 | ;; enclosing opening paren/brace/bracket which bounds the backwards search (or | |
2167 | ;; nil when at top level) and FROM is where the backward search started. It | |
2168 | ;; is reset to nil in `c-invalidate-state-cache'. | |
2169 | ||
2170 | ||
2171 | ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; | |
2172 | ;; Lowish level functions/macros which work directly on `c-state-cache', or a | |
2173 | ;; list of like structure. | |
2174 | (defmacro c-state-cache-top-lparen (&optional cache) | |
2175 | ;; Return the address of the top left brace/bracket/paren recorded in CACHE | |
2176 | ;; (default `c-state-cache') (or nil). | |
2177 | (let ((cash (or cache 'c-state-cache))) | |
2178 | `(if (consp (car ,cash)) | |
2179 | (caar ,cash) | |
2180 | (car ,cash)))) | |
2181 | ||
2182 | (defmacro c-state-cache-top-paren (&optional cache) | |
2183 | ;; Return the address of the latest brace/bracket/paren (whether left or | |
2184 | ;; right) recorded in CACHE (default `c-state-cache') or nil. | |
2185 | (let ((cash (or cache 'c-state-cache))) | |
2186 | `(if (consp (car ,cash)) | |
2187 | (cdar ,cash) | |
2188 | (car ,cash)))) | |
2189 | ||
2190 | (defmacro c-state-cache-after-top-paren (&optional cache) | |
2191 | ;; Return the position just after the latest brace/bracket/paren (whether | |
2192 | ;; left or right) recorded in CACHE (default `c-state-cache') or nil. | |
2193 | (let ((cash (or cache 'c-state-cache))) | |
2194 | `(if (consp (car ,cash)) | |
2195 | (cdar ,cash) | |
2196 | (and (car ,cash) | |
2197 | (1+ (car ,cash)))))) | |
2198 | ||
2199 | (defun c-get-cache-scan-pos (here) | |
2200 | ;; From the state-cache, determine the buffer position from which we might | |
2201 | ;; scan forward to HERE to update this cache. This position will be just | |
2202 | ;; after a paren/brace/bracket recorded in the cache, if possible, otherwise | |
2203 | ;; return the earliest position in the accessible region which isn't within | |
2204 | ;; a literal. If the visible portion of the buffer is entirely within a | |
2205 | ;; literal, return NIL. | |
2206 | (let ((c c-state-cache) elt) | |
2207 | ;(while (>= (or (c-state-cache-top-lparen c) 1) here) | |
2208 | (while (and c | |
2209 | (>= (c-state-cache-top-lparen c) here)) | |
2210 | (setq c (cdr c))) | |
2211 | ||
2212 | (setq elt (car c)) | |
2213 | (cond | |
2214 | ((consp elt) | |
2215 | (if (> (cdr elt) here) | |
2216 | (1+ (car elt)) | |
2217 | (cdr elt))) | |
2218 | (elt (1+ elt)) | |
2219 | ((<= (c-state-get-min-scan-pos) here) | |
2220 | (c-state-get-min-scan-pos)) | |
2221 | (t nil)))) | |
2222 | ||
2223 | ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; | |
2224 | ;; Variables which keep track of preprocessor constructs. | |
2225 | (defvar c-state-old-cpp-beg nil) | |
2226 | (make-variable-buffer-local 'c-state-old-cpp-beg) | |
2227 | (defvar c-state-old-cpp-end nil) | |
2228 | (make-variable-buffer-local 'c-state-old-cpp-end) | |
2229 | ;; These are the limits of the macro containing point at the previous call of | |
2230 | ;; `c-parse-state', or nil. | |
2231 | ||
2232 | ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; | |
2233 | ;; Defuns which analyse the buffer, yet don't change `c-state-cache'. | |
2234 | (defun c-get-fallback-scan-pos (here) | |
2235 | ;; Return a start position for building `c-state-cache' from | |
2236 | ;; scratch. This will be at the top level, 2 defuns back. | |
0386b551 AM |
2237 | (save-excursion |
2238 | ;; Go back 2 bods, but ignore any bogus positions returned by | |
2239 | ;; beginning-of-defun (i.e. open paren in column zero). | |
2240 | (goto-char here) | |
2241 | (let ((cnt 2)) | |
2242 | (while (not (or (bobp) (zerop cnt))) | |
0ec1d2c5 | 2243 | (c-beginning-of-defun-1) ; Pure elisp BOD. |
0386b551 AM |
2244 | (if (eq (char-after) ?\{) |
2245 | (setq cnt (1- cnt))))) | |
2246 | (point))) | |
2247 | ||
0ec1d2c5 AM |
2248 | (defun c-state-balance-parens-backwards (here top) |
2249 | ;; Return the position of the opening paren/brace/bracket before HERE which | |
2250 | ;; matches the outermost close p/b/b between HERE and TOP, like this: | |
2251 | ;; | |
2252 | ;; ...................................... | |
2253 | ;; | | | |
2254 | ;; ( [ ( ........... ) ( ) ] ) | |
2255 | ;; ^ ^ ^ | |
2256 | ;; | | | | |
2257 | ;; return HERE TOP | |
2258 | ;; | |
2259 | ;; If there aren't enough opening paren/brace/brackets, return the position | |
2260 | ;; of the outermost one found, or HERE it there are none. If there are no | |
2261 | ;; closeing p/b/bs between HERE and TOP, return HERE. HERE and TOP must not | |
2262 | ;; be inside literals. Only the accessible portion of the buffer will be | |
2263 | ;; scanned. | |
2264 | ||
2265 | ;; PART 1: scan from `here' up to `top', accumulating ")"s which enclose | |
2266 | ;; `here'. Go round the next loop each time we pass over such a ")". These | |
2267 | ;; probably match "("s before `here'. | |
2268 | (let (pos pa ren+1 lonely-rens) | |
2269 | (save-excursion | |
2270 | (save-restriction | |
2271 | (narrow-to-region (point-min) top) ; This can move point, sometimes. | |
2272 | (setq pos here) | |
2273 | (c-safe | |
2274 | (while | |
2275 | (setq ren+1 (scan-lists pos 1 1)) ; might signal | |
2276 | (setq lonely-rens (cons ren+1 lonely-rens) | |
2277 | pos ren+1))))) | |
2278 | ||
2279 | ;; PART 2: Scan back before `here' searching for the "("s | |
2280 | ;; matching/mismatching the ")"s found above. We only need to direct the | |
2281 | ;; caller to scan when we've encountered unmatched right parens. | |
2282 | (when lonely-rens | |
2283 | (setq pos here) | |
2284 | (c-safe | |
2285 | (while | |
2286 | (and lonely-rens ; actual values aren't used. | |
2287 | (setq pa (scan-lists pos -1 1))) | |
2288 | (setq pos pa) | |
2289 | (setq lonely-rens (cdr lonely-rens)))) ;) | |
2290 | ) | |
2291 | pos)) | |
2292 | ||
2293 | (defun c-parse-state-get-strategy (here good-pos) | |
2294 | ;; Determine the scanning strategy for adjusting `c-parse-state', attempting | |
2295 | ;; to minimise the amount of scanning. HERE is the pertinent position in | |
2296 | ;; the buffer, GOOD-POS is a position where `c-state-cache' (possibly with | |
2297 | ;; its head trimmed) is known to be good, or nil if there is no such | |
2298 | ;; position. | |
2299 | ;; | |
2300 | ;; The return value is a list, one of the following: | |
2301 | ;; | |
2302 | ;; o - ('forward CACHE-POS START-POINT) - scan forward from START-POINT, | |
2303 | ;; which is not less than CACHE-POS. | |
2304 | ;; o - ('backward CACHE-POS nil) - scan backwards (from HERE). | |
2305 | ;; o - ('BOD nil START-POINT) - scan forwards from START-POINT, which is at the | |
2306 | ;; top level. | |
2307 | ;; o - ('IN-LIT nil nil) - point is inside the literal containing point-min. | |
2308 | ;; , where CACHE-POS is the highest position recorded in `c-state-cache' at | |
2309 | ;; or below HERE. | |
2310 | (let ((cache-pos (c-get-cache-scan-pos here)) ; highest position below HERE in cache (or 1) | |
2311 | BOD-pos ; position of 2nd BOD before HERE. | |
2312 | strategy ; 'forward, 'backward, 'BOD, or 'IN-LIT. | |
2313 | start-point | |
2314 | how-far) ; putative scanning distance. | |
2315 | (setq good-pos (or good-pos (c-state-get-min-scan-pos))) | |
2316 | (cond | |
2317 | ((< here (c-state-get-min-scan-pos)) | |
2318 | (setq strategy 'IN-LIT | |
2319 | start-point nil | |
2320 | cache-pos nil | |
2321 | how-far 0)) | |
2322 | ((<= good-pos here) | |
2323 | (setq strategy 'forward | |
2324 | start-point (max good-pos cache-pos) | |
2325 | how-far (- here start-point))) | |
2326 | ((< (- good-pos here) (- here cache-pos)) ; FIXME!!! ; apply some sort of weighting. | |
2327 | (setq strategy 'backward | |
2328 | how-far (- good-pos here))) | |
2329 | (t | |
2330 | (setq strategy 'forward | |
2331 | how-far (- here cache-pos) | |
2332 | start-point cache-pos))) | |
2333 | ||
2334 | ;; Might we be better off starting from the top level, two defuns back, | |
2335 | ;; instead? | |
2336 | (when (> how-far c-state-cache-too-far) | |
2337 | (setq BOD-pos (c-get-fallback-scan-pos here)) ; somewhat EXPENSIVE!!! | |
2338 | (if (< (- here BOD-pos) how-far) | |
2339 | (setq strategy 'BOD | |
2340 | start-point BOD-pos))) | |
2341 | ||
2342 | (list | |
2343 | strategy | |
2344 | (and (memq strategy '(forward backward)) cache-pos) | |
2345 | (and (memq strategy '(forward BOD)) start-point)))) | |
2346 | ||
2347 | ||
2348 | ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; | |
2349 | ;; Routines which change `c-state-cache' and associated values. | |
2350 | (defun c-renarrow-state-cache () | |
2351 | ;; The region (more precisely, point-min) has changed since we | |
2352 | ;; calculated `c-state-cache'. Amend `c-state-cache' accordingly. | |
2353 | (if (< (point-min) c-state-point-min) | |
2354 | ;; If point-min has MOVED BACKWARDS then we drop the state completely. | |
2355 | ;; It would be possible to do a better job here and recalculate the top | |
2356 | ;; only. | |
2357 | (progn | |
2358 | (c-state-mark-point-min-literal) | |
2359 | (setq c-state-cache nil | |
2360 | c-state-cache-good-pos c-state-min-scan-pos | |
2361 | c-state-brace-pair-desert nil)) | |
2362 | ||
2363 | ;; point-min has MOVED FORWARD. | |
2364 | ||
2365 | ;; Is the new point-min inside a (different) literal? | |
2366 | (unless (and c-state-point-min-lit-start ; at prev. point-min | |
2367 | (< (point-min) (c-state-get-min-scan-pos))) | |
2368 | (c-state-mark-point-min-literal)) | |
2369 | ||
2370 | ;; Cut off a bit of the tail from `c-state-cache'. | |
2371 | (let ((ptr (cons nil c-state-cache)) | |
2372 | pa) | |
2373 | (while (and (setq pa (c-state-cache-top-lparen (cdr ptr))) | |
2374 | (>= pa (point-min))) | |
2375 | (setq ptr (cdr ptr))) | |
2376 | ||
2377 | (when (consp ptr) | |
2378 | (if (eq (cdr ptr) c-state-cache) | |
2379 | (setq c-state-cache nil | |
2380 | c-state-cache-good-pos c-state-min-scan-pos) | |
2381 | (setcdr ptr nil) | |
2382 | (setq c-state-cache-good-pos (1+ (c-state-cache-top-lparen)))) | |
2383 | ))) | |
2384 | ||
2385 | (setq c-state-point-min (point-min))) | |
2386 | ||
2387 | (defun c-append-lower-brace-pair-to-state-cache (from &optional upper-lim) | |
2388 | ;; If there is a brace pair preceding FROM in the buffer (not necessarily | |
2389 | ;; immediately preceding), push a cons onto `c-state-cache' to represent it. | |
2390 | ;; FROM must not be inside a literal. If UPPER-LIM is non-nil, we append | |
2391 | ;; the highest brace pair whose "}" is below UPPER-LIM. | |
2392 | ;; | |
2393 | ;; Return non-nil when this has been done. | |
2394 | ;; | |
2395 | ;; This routine should be fast. Since it can get called a LOT, we maintain | |
2396 | ;; `c-state-brace-pair-desert', a small cache of "failures", such that we | |
2397 | ;; reduce the time wasted in repeated fruitless searches in brace deserts. | |
2398 | (save-excursion | |
2399 | (save-restriction | |
2400 | (let ((bra from) ce ; Positions of "{" and "}". | |
2401 | new-cons | |
2402 | (cache-pos (c-state-cache-top-lparen)) ; might be nil. | |
2403 | (macro-start-or-from | |
2404 | (progn (goto-char from) | |
2405 | (c-beginning-of-macro) | |
2406 | (point)))) | |
2407 | (or upper-lim (setq upper-lim from)) | |
2408 | ||
2409 | ;; If we're essentially repeating a fruitless search, just give up. | |
2410 | (unless (and c-state-brace-pair-desert | |
2411 | (eq cache-pos (car c-state-brace-pair-desert)) | |
2412 | (<= from (cdr c-state-brace-pair-desert))) | |
2413 | ;; Only search what we absolutely need to: | |
2414 | (if (and c-state-brace-pair-desert | |
2415 | (> from (cdr c-state-brace-pair-desert))) | |
2416 | (narrow-to-region (cdr c-state-brace-pair-desert) (point-max))) | |
2417 | ||
2418 | ;; In the next pair of nested loops, the inner one moves back past a | |
2419 | ;; pair of (mis-)matching parens or brackets; the outer one moves | |
2420 | ;; back over a sequence of unmatched close brace/paren/bracket each | |
2421 | ;; time round. | |
2422 | (while | |
2423 | (progn | |
2424 | (c-safe | |
2425 | (while | |
2426 | (and (setq ce (scan-lists bra -1 -1)) ; back past )/]/}; might signal | |
2427 | (setq bra (scan-lists ce -1 1)) ; back past (/[/{; might signal | |
2428 | (or (> ce upper-lim) | |
2429 | (not (eq (char-after bra) ?\{)) | |
2430 | (and (goto-char bra) | |
2431 | (c-beginning-of-macro) | |
2432 | (< (point) macro-start-or-from)))))) | |
2433 | (and ce (< ce bra))) | |
2434 | (setq bra ce)) ; If we just backed over an unbalanced closing | |
2435 | ; brace, ignore it. | |
2436 | ||
2437 | (if (and ce (< bra ce) (eq (char-after bra) ?\{)) | |
2438 | ;; We've found the desired brace-pair. | |
2439 | (progn | |
2440 | (setq new-cons (cons bra (1+ ce))) | |
2441 | (cond | |
2442 | ((consp (car c-state-cache)) | |
2443 | (setcar c-state-cache new-cons)) | |
2444 | ((and (numberp (car c-state-cache)) ; probably never happens | |
2445 | (< ce (car c-state-cache))) | |
2446 | (setcdr c-state-cache | |
2447 | (cons new-cons (cdr c-state-cache)))) | |
2448 | (t (setq c-state-cache (cons new-cons c-state-cache))))) | |
2449 | ||
2450 | ;; We haven't found a brace pair. Record this. | |
2451 | (setq c-state-brace-pair-desert (cons cache-pos from)))))))) | |
2452 | ||
2453 | (defsubst c-state-push-any-brace-pair (bra+1 macro-start-or-here) | |
2454 | ;; If BRA+1 is nil, do nothing. Otherwise, BRA+1 is the buffer position | |
2455 | ;; following a {, and that brace has a (mis-)matching } (or ]), and we | |
2456 | ;; "push" "a" brace pair onto `c-state-cache'. | |
2457 | ;; | |
2458 | ;; Here "push" means overwrite the top element if it's itself a brace-pair, | |
2459 | ;; otherwise push it normally. | |
2460 | ;; | |
2461 | ;; The brace pair we push is normally the one surrounding BRA+1, but if the | |
2462 | ;; latter is inside a macro, not being a macro containing | |
2463 | ;; MACRO-START-OR-HERE, we scan backwards through the buffer for a non-macro | |
2464 | ;; base pair. This latter case is assumed to be rare. | |
2465 | ;; | |
2466 | ;; Note: POINT is not preserved in this routine. | |
2467 | (if bra+1 | |
2468 | (if (or (> bra+1 macro-start-or-here) | |
2469 | (progn (goto-char bra+1) | |
2470 | (not (c-beginning-of-macro)))) | |
2471 | (setq c-state-cache | |
2472 | (cons (cons (1- bra+1) | |
2473 | (scan-lists bra+1 1 1)) | |
2474 | (if (consp (car c-state-cache)) | |
2475 | (cdr c-state-cache) | |
2476 | c-state-cache))) | |
2477 | ;; N.B. This defsubst codes one method for the simple, normal case, | |
2478 | ;; and a more sophisticated, slower way for the general case. Don't | |
2479 | ;; eliminate this defsubst - it's a speed optimisation. | |
2480 | (c-append-lower-brace-pair-to-state-cache (1- bra+1))))) | |
2481 | ||
2482 | (defun c-append-to-state-cache (from) | |
2483 | ;; Scan the buffer from FROM to (point-max), adding elements into | |
2484 | ;; `c-state-cache' for braces etc. Return a candidate for | |
2485 | ;; `c-state-cache-good-pos'. | |
2486 | ;; | |
2487 | ;; FROM must be after the latest brace/paren/bracket in `c-state-cache', if | |
2488 | ;; any. Typically, it is immediately after it. It must not be inside a | |
2489 | ;; literal. | |
2490 | (let ((here-bol (c-point 'bol (point-max))) | |
2491 | (macro-start-or-here | |
2492 | (save-excursion (goto-char (point-max)) | |
2493 | (if (c-beginning-of-macro) | |
2494 | (point) | |
2495 | (point-max)))) | |
2496 | pa+1 ; pos just after an opening PAren (or brace). | |
2497 | (ren+1 from) ; usually a pos just after an closing paREN etc. | |
2498 | ; Is actually the pos. to scan for a (/{/[ from, | |
2499 | ; which sometimes is after a silly )/}/]. | |
2500 | paren+1 ; Pos after some opening or closing paren. | |
2501 | paren+1s ; A list of `paren+1's; used to determine a | |
2502 | ; good-pos. | |
2503 | bra+1 ce+1 ; just after L/R bra-ces. | |
2504 | bra+1s ; list of OLD values of bra+1. | |
2505 | mstart) ; start of a macro. | |
2506 | ||
2507 | (save-excursion | |
2508 | ;; Each time round the following loop, we enter a succesively deeper | |
2509 | ;; level of brace/paren nesting. (Except sometimes we "continue at | |
2510 | ;; the existing level".) `pa+1' is a pos inside an opening | |
2511 | ;; brace/paren/bracket, usually just after it. | |
2512 | (while | |
2513 | (progn | |
2514 | ;; Each time round the next loop moves forward over an opening then | |
2515 | ;; a closing brace/bracket/paren. This loop is white hot, so it | |
2516 | ;; plays ugly tricks to go fast. DON'T PUT ANYTHING INTO THIS | |
2517 | ;; LOOP WHICH ISN'T ABSOLUTELY NECESSARY!!! It terminates when a | |
2518 | ;; call of `scan-lists' signals an error, which happens when there | |
2519 | ;; are no more b/b/p's to scan. | |
2520 | (c-safe | |
2521 | (while t | |
2522 | (setq pa+1 (scan-lists ren+1 1 -1) ; Into (/{/[; might signal | |
2523 | paren+1s (cons pa+1 paren+1s)) | |
2524 | (setq ren+1 (scan-lists pa+1 1 1)) ; Out of )/}/]; might signal | |
2525 | (if (and (eq (char-before pa+1) ?{)) ; Check for a macro later. | |
2526 | (setq bra+1 pa+1)) | |
2527 | (setcar paren+1s ren+1))) | |
2528 | ||
2529 | (if (and pa+1 (> pa+1 ren+1)) | |
2530 | ;; We've just entered a deeper nesting level. | |
2531 | (progn | |
2532 | ;; Insert the brace pair (if present) and the single open | |
2533 | ;; paren/brace/bracket into `c-state-cache' It cannot be | |
2534 | ;; inside a macro, except one around point, because of what | |
2535 | ;; `c-neutralize-syntax-in-CPP' has done. | |
2536 | (c-state-push-any-brace-pair bra+1 macro-start-or-here) | |
2537 | ;; Insert the opening brace/bracket/paren position. | |
2538 | (setq c-state-cache (cons (1- pa+1) c-state-cache)) | |
2539 | ;; Clear admin stuff for the next more nested part of the scan. | |
2540 | (setq ren+1 pa+1 pa+1 nil bra+1 nil bra+1s nil) | |
2541 | t) ; Carry on the loop | |
2542 | ||
2543 | ;; All open p/b/b's at this nesting level, if any, have probably | |
2544 | ;; been closed by matching/mismatching ones. We're probably | |
2545 | ;; finished - we just need to check for having found an | |
2546 | ;; unmatched )/}/], which we ignore. Such a )/}/] can't be in a | |
2547 | ;; macro, due the action of `c-neutralize-syntax-in-CPP'. | |
2548 | (c-safe (setq ren+1 (scan-lists ren+1 1 1)))))) ; acts as loop control. | |
2549 | ||
2550 | ;; Record the final, innermost, brace-pair if there is one. | |
2551 | (c-state-push-any-brace-pair bra+1 macro-start-or-here) | |
2552 | ||
2553 | ;; Determine a good pos | |
2554 | (while (and (setq paren+1 (car paren+1s)) | |
2555 | (> (if (> paren+1 macro-start-or-here) | |
2556 | paren+1 | |
2557 | (goto-char paren+1) | |
2558 | (setq mstart (and (c-beginning-of-macro) | |
2559 | (point))) | |
2560 | (or mstart paren+1)) | |
2561 | here-bol)) | |
2562 | (setq paren+1s (cdr paren+1s))) | |
2563 | (cond | |
2564 | ((and paren+1 mstart) | |
2565 | (min paren+1 mstart)) | |
2566 | (paren+1) | |
2567 | (t from))))) | |
2568 | ||
2569 | (defun c-remove-stale-state-cache (good-pos pps-point) | |
2570 | ;; Remove stale entries from the `c-cache-state', i.e. those which will | |
2571 | ;; not be in it when it is amended for position (point-max). | |
2572 | ;; Additionally, the "outermost" open-brace entry before (point-max) | |
2573 | ;; will be converted to a cons if the matching close-brace is scanned. | |
2574 | ;; | |
2575 | ;; GOOD-POS is a "maximal" "safe position" - there must be no open | |
2576 | ;; parens/braces/brackets between GOOD-POS and (point-max). | |
2577 | ;; | |
2578 | ;; As a second thing, calculate the result of parse-partial-sexp at | |
2579 | ;; PPS-POINT, w.r.t. GOOD-POS. The motivation here is that | |
2580 | ;; `c-state-cache-good-pos' may become PPS-POINT, but the caller may need to | |
2581 | ;; adjust it to get outside a string/comment. (Sorry about this! The code | |
2582 | ;; needs to be FAST). | |
2583 | ;; | |
2584 | ;; Return a list (GOOD-POS SCAN-BACK-POS PPS-STATE), where | |
2585 | ;; o - GOOD-POS is a position where the new value `c-state-cache' is known | |
2586 | ;; to be good (we aim for this to be as high as possible); | |
2587 | ;; o - SCAN-BACK-POS, if not nil, indicates there may be a brace pair | |
2588 | ;; preceding POS which needs to be recorded in `c-state-cache'. It is a | |
2589 | ;; position to scan backwards from. | |
2590 | ;; o - PPS-STATE is the parse-partial-sexp state at PPS-POINT. | |
2591 | (save-restriction | |
2592 | (narrow-to-region 1 (point-max)) | |
2593 | (save-excursion | |
2594 | (let* ((in-macro-start ; point-max or beginning of macro containing it | |
2595 | (save-excursion | |
2596 | (goto-char (point-max)) | |
2597 | (and (c-beginning-of-macro) | |
2598 | (point)))) | |
2599 | (good-pos-actual-macro-start ; Start of macro containing good-pos | |
2600 | ; or nil | |
2601 | (and (< good-pos (point-max)) | |
2602 | (save-excursion | |
2603 | (goto-char good-pos) | |
2604 | (and (c-beginning-of-macro) | |
2605 | (point))))) | |
2606 | (good-pos-actual-macro-end ; End of this macro, (maybe | |
2607 | ; (point-max)), or nil. | |
2608 | (and good-pos-actual-macro-start | |
2609 | (save-excursion | |
2610 | (goto-char good-pos-actual-macro-start) | |
2611 | (c-end-of-macro) | |
2612 | (point)))) | |
2613 | pps-state ; Will be 9 or 10 elements long. | |
2614 | pos | |
2615 | upper-lim ; ,beyond which `c-state-cache' entries are removed | |
2616 | scan-back-pos | |
2617 | pair-beg pps-point-state target-depth) | |
2618 | ||
2619 | ;; Remove entries beyond (point-max). Also remove any entries inside | |
2620 | ;; a macro, unless (point-max) is in the same macro. | |
2621 | (setq upper-lim | |
2622 | (if (or (null c-state-old-cpp-beg) | |
2623 | (and (> (point-max) c-state-old-cpp-beg) | |
2624 | (< (point-max) c-state-old-cpp-end))) | |
2625 | (point-max) | |
2626 | (min (point-max) c-state-old-cpp-beg))) | |
2627 | (while (and c-state-cache (> (c-state-cache-top-lparen) upper-lim)) | |
2628 | (setq c-state-cache (cdr c-state-cache))) | |
2629 | ;; If `upper-lim' is inside the last recorded brace pair, remove its | |
2630 | ;; RBrace and indicate we'll need to search backwards for a previous | |
2631 | ;; brace pair. | |
2632 | (when (and c-state-cache | |
2633 | (consp (car c-state-cache)) | |
2634 | (> (cdar c-state-cache) upper-lim)) | |
2635 | (setcar c-state-cache (caar c-state-cache)) | |
2636 | (setq scan-back-pos (car c-state-cache))) | |
2637 | ||
2638 | ;; The next loop jumps forward out of a nested level of parens each | |
2639 | ;; time round; the corresponding elements in `c-state-cache' are | |
2640 | ;; removed. `pos' is just after the brace-pair or the open paren at | |
2641 | ;; (car c-state-cache). There can be no open parens/braces/brackets | |
2642 | ;; between `good-pos'/`good-pos-actual-macro-start' and (point-max), | |
2643 | ;; due to the interface spec to this function. | |
2644 | (setq pos (if good-pos-actual-macro-end | |
2645 | (1+ good-pos-actual-macro-end) ; get outside the macro as | |
2646 | ; marked by a `category' text property. | |
2647 | good-pos)) | |
2648 | (goto-char pos) | |
2649 | (while (and c-state-cache | |
2650 | (< (point) (point-max))) | |
2651 | (cond | |
2652 | ((null pps-state) ; first time through | |
2653 | (setq target-depth -1)) | |
2654 | ((eq (car pps-state) target-depth) ; found closing ),},] | |
2655 | (setq target-depth (1- (car pps-state)))) | |
2656 | ;; Do nothing when we've merely reached pps-point. | |
2657 | ) | |
2658 | ||
2659 | ;; Scan! | |
2660 | (setq pps-state | |
2661 | (parse-partial-sexp | |
2662 | (point) (if (< (point) pps-point) pps-point (point-max)) | |
2663 | target-depth | |
2664 | nil pps-state)) | |
2665 | ||
2666 | (if (= (point) pps-point) | |
2667 | (setq pps-point-state pps-state)) | |
2668 | ||
2669 | (when (eq (car pps-state) target-depth) | |
2670 | (setq pos (point)) ; POS is now just after an R-paren/brace. | |
2671 | (cond | |
2672 | ((and (consp (car c-state-cache)) | |
2673 | (eq (point) (cdar c-state-cache))) | |
2674 | ;; We've just moved out of the paren pair containing the brace-pair | |
2675 | ;; at (car c-state-cache). `pair-beg' is where the open paren is, | |
2676 | ;; and is potentially where the open brace of a cons in | |
2677 | ;; c-state-cache will be. | |
2678 | (setq pair-beg (car-safe (cdr c-state-cache)) | |
2679 | c-state-cache (cdr-safe (cdr c-state-cache)))) ; remove {}pair + containing Lparen. | |
2680 | ((numberp (car c-state-cache)) | |
2681 | (setq pair-beg (car c-state-cache) | |
2682 | c-state-cache (cdr c-state-cache))) ; remove this | |
2683 | ; containing Lparen | |
2684 | ((numberp (cadr c-state-cache)) | |
2685 | (setq pair-beg (cadr c-state-cache) | |
2686 | c-state-cache (cddr c-state-cache))) ; Remove a paren pair | |
2687 | ; together with enclosed brace pair. | |
2688 | ;; (t nil) ; Ignore an unmated Rparen. | |
2689 | ))) | |
2690 | ||
2691 | (if (< (point) pps-point) | |
2692 | (setq pps-state (parse-partial-sexp (point) pps-point | |
2693 | nil nil ; TARGETDEPTH, STOPBEFORE | |
2694 | pps-state))) | |
2695 | ||
2696 | ;; If the last paren pair we moved out of was actually a brace pair, | |
2697 | ;; insert it into `c-state-cache'. | |
2698 | (when (and pair-beg (eq (char-after pair-beg) ?{)) | |
2699 | (if (consp (car-safe c-state-cache)) | |
2700 | (setq c-state-cache (cdr c-state-cache))) | |
2701 | (setq c-state-cache (cons (cons pair-beg pos) | |
2702 | c-state-cache))) | |
2703 | ||
2704 | (list pos scan-back-pos pps-state))))) | |
2705 | ||
2706 | (defun c-remove-stale-state-cache-backwards (here cache-pos) | |
2707 | ;; Strip stale elements of `c-state-cache' by moving backwards through the | |
2708 | ;; buffer, and inform the caller of the scenario detected. | |
2709 | ;; | |
2710 | ;; HERE is the position we're setting `c-state-cache' for. | |
2711 | ;; CACHE-POS is just after the latest recorded position in `c-state-cache' | |
2712 | ;; before HERE, or a position at or near point-min which isn't in a | |
2713 | ;; literal. | |
2714 | ;; | |
2715 | ;; This function must only be called only when (> `c-state-cache-good-pos' | |
2716 | ;; HERE). Usually the gap between CACHE-POS and HERE is large. It is thus | |
2717 | ;; optimised to eliminate (or minimise) scanning between these two | |
2718 | ;; positions. | |
2719 | ;; | |
2720 | ;; Return a three element list (GOOD-POS SCAN-BACK-POS FWD-FLAG), where: | |
2721 | ;; o - GOOD-POS is a "good position", where `c-state-cache' is valid, or | |
2722 | ;; could become so after missing elements are inserted into | |
2723 | ;; `c-state-cache'. This is JUST AFTER an opening or closing | |
2724 | ;; brace/paren/bracket which is already in `c-state-cache' or just before | |
2725 | ;; one otherwise. exceptionally (when there's no such b/p/b handy) the BOL | |
2726 | ;; before `here''s line, or the start of the literal containing it. | |
2727 | ;; o - SCAN-BACK-POS, if non-nil, indicates there may be a brace pair | |
2728 | ;; preceding POS which isn't recorded in `c-state-cache'. It is a position | |
2729 | ;; to scan backwards from. | |
2730 | ;; o - FWD-FLAG, if non-nil, indicates there may be parens/braces between | |
2731 | ;; POS and HERE which aren't recorded in `c-state-cache'. | |
2732 | ;; | |
2733 | ;; The comments in this defun use "paren" to mean parenthesis or square | |
2734 | ;; bracket (as contrasted with a brace), and "(" and ")" likewise. | |
2735 | ;; | |
2736 | ;; . {..} (..) (..) ( .. { } ) (...) ( .... . ..) | |
2737 | ;; | | | | | | | |
2738 | ;; CP E here D C good | |
2739 | (let ((pos c-state-cache-good-pos) | |
2740 | pa ren ; positions of "(" and ")" | |
2741 | dropped-cons ; whether the last element dropped from `c-state-cache' | |
2742 | ; was a cons (representing a brace-pair) | |
2743 | good-pos ; see above. | |
2744 | lit ; (START . END) of a literal containing some point. | |
2745 | here-lit-start here-lit-end ; bounds of literal containing `here' | |
2746 | ; or `here' itself. | |
2747 | (here-bol (c-point 'bol here)) | |
2748 | (too-far-back (max (- here c-state-cache-too-far) 1))) | |
2749 | ||
2750 | ;; Remove completely irrelevant entries from `c-state-cache'. | |
2751 | (while (and c-state-cache | |
2752 | (>= (setq pa (c-state-cache-top-lparen)) here)) | |
2753 | (setq dropped-cons (consp (car c-state-cache))) | |
2754 | (setq c-state-cache (cdr c-state-cache)) | |
2755 | (setq pos pa)) | |
2756 | ;; At this stage, (> pos here); | |
2757 | ;; (< (c-state-cache-top-lparen) here) (or is nil). | |
2758 | ||
2759 | ;; CASE 1: The top of the cache is a brace pair which now encloses `here'. | |
2760 | ;; As good-pos, return the address. of the "{". | |
2761 | (if (and (consp (car c-state-cache)) | |
2762 | (> (cdar c-state-cache) here)) | |
2763 | ;; Since we've no knowledge of what's inside these braces, we have no | |
2764 | ;; alternative but to direct the caller to scan the buffer from the | |
2765 | ;; opening brace. | |
2766 | (progn | |
2767 | (setq pos (caar c-state-cache)) | |
2768 | (setcar c-state-cache pos) | |
2769 | (list (1+ pos) pos t)) ; return value. We've just converted a brace | |
2770 | ; pair entry into a { entry, so the caller | |
2771 | ; needs to search for a brace pair before the | |
2772 | ; {. | |
2773 | ||
2774 | ;; ;; `here' might be inside a literal. Check for this. | |
2775 | (setq lit (c-state-literal-at here) | |
2776 | here-lit-start (or (car lit) here) | |
2777 | here-lit-end (or (cdr lit) here)) | |
2778 | ||
2779 | ;; `here' might be nested inside any depth of parens (or brackets but | |
2780 | ;; not braces). Scan backwards to find the outermost such opening | |
2781 | ;; paren, if there is one. This will be the scan position to return. | |
2782 | (save-restriction | |
2783 | (narrow-to-region cache-pos (point-max)) | |
2784 | (setq pos (c-state-balance-parens-backwards here-lit-end pos))) | |
2785 | ||
2786 | (if (< pos here-lit-start) | |
2787 | ;; CASE 2: Address of outermost ( or [ which now encloses `here', | |
2788 | ;; but didn't enclose the (previous) `c-state-cache-good-pos'. If | |
2789 | ;; there is a brace pair preceding this, it will already be in | |
2790 | ;; `c-state-cache', unless there was a brace pair after it, | |
2791 | ;; i.e. there'll only be one to scan for if we've just deleted one. | |
2792 | (list pos (and dropped-cons pos) t) ; Return value. | |
2793 | ||
2794 | ;; `here' isn't enclosed in a (previously unrecorded) bracket/paren. | |
2795 | ;; Further forward scanning isn't needed, but we still need to find a | |
2796 | ;; GOOD-POS. Step out of all enclosing "("s on HERE's line. | |
2797 | (save-restriction | |
2798 | (narrow-to-region here-bol (point-max)) | |
2799 | (setq pos here-lit-start) | |
2800 | (c-safe (while (setq pa (scan-lists pos -1 1)) | |
2801 | (setq pos pa)))) ; might signal | |
2802 | (if (setq ren (c-safe-scan-lists pos -1 -1 too-far-back)) | |
2803 | ;; CASE 3: After a }/)/] before `here''s BOL. | |
2804 | (list (1+ ren) (and dropped-cons pos) nil) ; Return value | |
2805 | ||
2806 | ;; CASE 4; Best of a bad job: BOL before `here-bol', or beginning of | |
2807 | ;; literal containing it. | |
2808 | (setq good-pos (c-state-lit-beg (c-point 'bopl here-bol))) | |
2809 | (list good-pos (and dropped-cons good-pos) nil)))))) | |
2810 | ||
2811 | ||
2812 | ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; | |
2813 | ;; Externally visible routines. | |
2814 | ||
2815 | (defun c-state-cache-init () | |
2816 | (setq c-state-cache nil | |
2817 | c-state-cache-good-pos 1 | |
2818 | c-state-nonlit-pos-cache nil | |
2819 | c-state-nonlit-pos-cache-limit 1 | |
2820 | c-state-brace-pair-desert nil | |
2821 | c-state-point-min 1 | |
2822 | c-state-point-min-lit-type nil | |
2823 | c-state-point-min-lit-start nil | |
2824 | c-state-min-scan-pos 1 | |
2825 | c-state-old-cpp-beg nil | |
2826 | c-state-old-cpp-end nil) | |
2827 | (c-state-mark-point-min-literal)) | |
2828 | ||
2829 | (defun c-invalidate-state-cache-1 (here) | |
2830 | ;; Invalidate all info on `c-state-cache' that applies to the buffer at HERE | |
2831 | ;; or higher and set `c-state-cache-good-pos' accordingly. The cache is | |
2832 | ;; left in a consistent state. | |
2833 | ;; | |
2834 | ;; This is much like `c-whack-state-after', but it never changes a paren | |
2835 | ;; pair element into an open paren element. Doing that would mean that the | |
2836 | ;; new open paren wouldn't have the required preceding paren pair element. | |
785eecbb | 2837 | ;; |
0ec1d2c5 AM |
2838 | ;; This function is called from c-after-change. |
2839 | ||
2840 | ;; The cache of non-literals: | |
2841 | (if (< here c-state-nonlit-pos-cache-limit) | |
2842 | (setq c-state-nonlit-pos-cache-limit here)) | |
2843 | ||
2844 | ;; `c-state-cache': | |
2845 | ;; Case 1: if `here' is in a literal containing point-min, everything | |
2846 | ;; becomes (or is already) nil. | |
2847 | (if (or (null c-state-cache-good-pos) | |
2848 | (< here (c-state-get-min-scan-pos))) | |
2849 | (setq c-state-cache nil | |
2850 | c-state-cache-good-pos nil | |
2851 | c-state-min-scan-pos nil) | |
2852 | ||
2853 | ;;; Truncate `c-state-cache' and set `c-state-cache-good-pos' to a value below | |
2854 | ;;; `here'. To maintain its consistency, we may need to insert a new brace | |
2855 | ;;; pair. | |
2856 | (let ((here-bol (c-point 'bol here)) | |
2857 | too-high-pa ; recorded {/(/[ next above here, or nil. | |
2858 | dropped-cons ; was the last removed element a brace pair? | |
2859 | pa) | |
2860 | ;; The easy bit - knock over-the-top bits off `c-state-cache'. | |
2861 | (while (and c-state-cache | |
2862 | (>= (setq pa (c-state-cache-top-paren)) here)) | |
2863 | (setq dropped-cons (consp (car c-state-cache)) | |
2864 | too-high-pa (c-state-cache-top-lparen) | |
2865 | c-state-cache (cdr c-state-cache))) | |
2866 | ||
2867 | ;; Do we need to add in an earlier brace pair, having lopped one off? | |
2868 | (if (and dropped-cons | |
2869 | (< too-high-pa (+ here c-state-cache-too-far))) | |
2870 | (c-append-lower-brace-pair-to-state-cache too-high-pa here-bol)) | |
2871 | (setq c-state-cache-good-pos (or (c-state-cache-after-top-paren) | |
2872 | (c-state-get-min-scan-pos))))) | |
2873 | ||
2874 | ;; The brace-pair desert marker: | |
2875 | (when (car c-state-brace-pair-desert) | |
2876 | (if (< here (car c-state-brace-pair-desert)) | |
2877 | (setq c-state-brace-pair-desert nil) | |
2878 | (if (< here (cdr c-state-brace-pair-desert)) | |
2879 | (setcdr c-state-brace-pair-desert here))))) | |
2880 | ||
2881 | (defun c-parse-state-1 () | |
2882 | ;; Find and record all noteworthy parens between some good point earlier in | |
2883 | ;; the file and point. That good point is at least the beginning of the | |
2884 | ;; top-level construct we are in, or the beginning of the preceding | |
2885 | ;; top-level construct if we aren't in one. | |
2886 | ;; | |
2887 | ;; The returned value is a list of the noteworthy parens with the last one | |
2888 | ;; first. If an element in the list is an integer, it's the position of an | |
2889 | ;; open paren (of any type) which has not been closed before the point. If | |
2890 | ;; an element is a cons, it gives the position of a closed BRACE paren | |
2891 | ;; pair[*]; the car is the start brace position and the cdr is the position | |
2892 | ;; following the closing brace. Only the last closed brace paren pair | |
2893 | ;; before each open paren and before the point is recorded, and thus the | |
2894 | ;; state never contains two cons elements in succession. When a close brace | |
2895 | ;; has no matching open brace (e.g., the matching brace is outside the | |
2896 | ;; visible region), it is not represented in the returned value. | |
2897 | ;; | |
2898 | ;; [*] N.B. The close "brace" might be a mismatching close bracket or paren. | |
2899 | ;; This defun explicitly treats mismatching parens/braces/brackets as | |
2900 | ;; matching. It is the open brace which makes it a "brace" pair. | |
2901 | ;; | |
2902 | ;; If POINT is within a macro, open parens and brace pairs within | |
2903 | ;; THIS macro MIGHT be recorded. This depends on whether their | |
2904 | ;; syntactic properties have been suppressed by | |
2905 | ;; `c-neutralize-syntax-in-CPP'. This might need fixing (2008-12-11). | |
d9e94c22 MS |
2906 | ;; |
2907 | ;; Currently no characters which are given paren syntax with the | |
2908 | ;; syntax-table property are recorded, i.e. angle bracket arglist | |
2909 | ;; parens are never present here. Note that this might change. | |
2910 | ;; | |
0386b551 | 2911 | ;; BUG: This function doesn't cope entirely well with unbalanced |
0ec1d2c5 AM |
2912 | ;; parens in macros. (2008-12-11: this has probably been resolved |
2913 | ;; by the function `c-neutralize-syntax-in-CPP'.) E.g. in the | |
2914 | ;; following case the brace before the macro isn't balanced with the | |
2915 | ;; one after it: | |
0386b551 AM |
2916 | ;; |
2917 | ;; { | |
2918 | ;; #define X { | |
2919 | ;; } | |
2920 | ;; | |
0ec1d2c5 AM |
2921 | ;; Note to maintainers: this function DOES get called with point |
2922 | ;; within comments and strings, so don't assume it doesn't! | |
2923 | ;; | |
0386b551 | 2924 | ;; This function might do hidden buffer changes. |
0ec1d2c5 AM |
2925 | (let* ((here (point)) |
2926 | (here-bopl (c-point 'bopl)) | |
2927 | strategy ; 'forward, 'backward etc.. | |
2928 | ;; Candidate positions to start scanning from: | |
2929 | cache-pos ; highest position below HERE already existing in | |
2930 | ; cache (or 1). | |
2931 | good-pos | |
2932 | start-point | |
2933 | bopl-state | |
2934 | res | |
2935 | scan-backward-pos scan-forward-p) ; used for 'backward. | |
2936 | ;; If POINT-MIN has changed, adjust the cache | |
2937 | (unless (= (point-min) c-state-point-min) | |
2938 | (c-renarrow-state-cache)) | |
2939 | ||
2940 | ;; Strategy? | |
2941 | (setq res (c-parse-state-get-strategy here c-state-cache-good-pos) | |
2942 | strategy (car res) | |
2943 | cache-pos (cadr res) | |
2944 | start-point (nth 2 res)) | |
2945 | ||
2946 | (when (eq strategy 'BOD) | |
2947 | (setq c-state-cache nil | |
2948 | c-state-cache-good-pos start-point)) | |
2949 | ||
2950 | ;; SCAN! | |
2951 | (save-restriction | |
2952 | (cond | |
2953 | ((memq strategy '(forward BOD)) | |
0386b551 | 2954 | (narrow-to-region (point-min) here) |
0ec1d2c5 AM |
2955 | (setq res (c-remove-stale-state-cache start-point here-bopl)) |
2956 | (setq cache-pos (car res) | |
2957 | scan-backward-pos (cadr res) | |
2958 | bopl-state (car (cddr res))) ; will be nil if (< here-bopl | |
2959 | ; start-point) | |
2960 | (if scan-backward-pos | |
2961 | (c-append-lower-brace-pair-to-state-cache scan-backward-pos)) | |
2962 | (setq good-pos | |
2963 | (c-append-to-state-cache cache-pos)) | |
2964 | (setq c-state-cache-good-pos | |
2965 | (if (and bopl-state | |
2966 | (< good-pos (- here c-state-cache-too-far))) | |
2967 | (c-state-cache-non-literal-place here-bopl bopl-state) | |
2968 | good-pos))) | |
2969 | ||
2970 | ((eq strategy 'backward) | |
2971 | (setq res (c-remove-stale-state-cache-backwards here cache-pos) | |
2972 | good-pos (car res) | |
2973 | scan-backward-pos (cadr res) | |
2974 | scan-forward-p (car (cddr res))) | |
2975 | (if scan-backward-pos | |
2976 | (c-append-lower-brace-pair-to-state-cache | |
2977 | scan-backward-pos)) | |
2978 | (setq c-state-cache-good-pos | |
2979 | (if scan-forward-p | |
2980 | (progn (narrow-to-region (point-min) here) | |
2981 | (c-append-to-state-cache good-pos)) | |
2982 | ||
2983 | (c-get-cache-scan-pos good-pos)))) | |
2984 | ||
2985 | (t ; (eq strategy 'IN-LIT) | |
2986 | (setq c-state-cache nil | |
2987 | c-state-cache-good-pos nil))))) | |
2988 | ||
2989 | c-state-cache) | |
2990 | ||
2991 | (defun c-invalidate-state-cache (here) | |
2992 | ;; This is a wrapper over `c-invalidate-state-cache-1'. | |
2993 | ;; | |
2994 | ;; It suppresses the syntactic effect of the < and > (template) brackets and | |
2995 | ;; of all parens in preprocessor constructs, except for any such construct | |
2996 | ;; containing point. We can then call `c-invalidate-state-cache-1' without | |
2997 | ;; worrying further about macros and template delimiters. | |
2998 | (c-with-<->-as-parens-suppressed | |
2999 | (if c-state-old-cpp-beg | |
3000 | (c-with-all-but-one-cpps-commented-out | |
3001 | c-state-old-cpp-beg c-state-old-cpp-end | |
3002 | (c-invalidate-state-cache-1 here)) | |
3003 | (c-with-cpps-commented-out | |
3004 | (c-invalidate-state-cache-1 here))))) | |
3005 | ||
3006 | (defun c-parse-state () | |
3007 | ;; This is a wrapper over `c-parse-state-1'. See that function for a | |
3008 | ;; description of the functionality and return value. | |
3009 | ;; | |
3010 | ;; It suppresses the syntactic effect of the < and > (template) brackets and | |
3011 | ;; of all parens in preprocessor constructs, except for any such construct | |
3012 | ;; containing point. We can then call `c-parse-state-1' without worrying | |
3013 | ;; further about macros and template delimiters. | |
3014 | (let (here-cpp-beg here-cpp-end) | |
3015 | (save-excursion | |
3016 | (when (c-beginning-of-macro) | |
3017 | (setq here-cpp-beg (point)) | |
3018 | (unless | |
3019 | (> (setq here-cpp-end (c-syntactic-end-of-macro)) | |
3020 | here-cpp-beg) | |
3021 | (setq here-cpp-beg nil here-cpp-end nil)))) | |
3022 | ;; FIXME!!! Put in a `condition-case' here to protect the integrity of the | |
3023 | ;; subsystem. | |
3024 | (prog1 | |
3025 | (c-with-<->-as-parens-suppressed | |
3026 | (if (and here-cpp-beg (> here-cpp-end here-cpp-beg)) | |
3027 | (c-with-all-but-one-cpps-commented-out | |
3028 | here-cpp-beg here-cpp-end | |
3029 | (c-parse-state-1)) | |
3030 | (c-with-cpps-commented-out | |
3031 | (c-parse-state-1)))) | |
3032 | (setq c-state-old-cpp-beg here-cpp-beg | |
3033 | c-state-old-cpp-end here-cpp-end)))) | |
3034 | ||
3035 | ;; Debug tool to catch cache inconsistencies. This is called from | |
3036 | ;; 000tests.el. | |
a66cd3ee MS |
3037 | (defvar c-debug-parse-state nil) |
3038 | (unless (fboundp 'c-real-parse-state) | |
3039 | (fset 'c-real-parse-state (symbol-function 'c-parse-state))) | |
3040 | (cc-bytecomp-defun c-real-parse-state) | |
3041 | (defun c-debug-parse-state () | |
0ec1d2c5 | 3042 | (let ((here (point)) (res1 (c-real-parse-state)) res2) |
0386b551 | 3043 | (let ((c-state-cache nil) |
0ec1d2c5 AM |
3044 | (c-state-cache-good-pos 1) |
3045 | (c-state-nonlit-pos-cache nil) | |
3046 | (c-state-nonlit-pos-cache-limit 1) | |
3047 | (c-state-brace-pair-desert nil) | |
3048 | (c-state-point-min 1) | |
3049 | (c-state-point-min-lit-type nil) | |
3050 | (c-state-point-min-lit-start nil) | |
3051 | (c-state-min-scan-pos 1) | |
3052 | (c-state-old-cpp-beg nil) | |
3053 | (c-state-old-cpp-end nil)) | |
a66cd3ee MS |
3054 | (setq res2 (c-real-parse-state))) |
3055 | (unless (equal res1 res2) | |
0386b551 AM |
3056 | ;; The cache can actually go further back due to the ad-hoc way |
3057 | ;; the first paren is found, so try to whack off a bit of its | |
3058 | ;; start before complaining. | |
3059 | (save-excursion | |
3060 | (goto-char (or (c-least-enclosing-brace res2) (point))) | |
3061 | (c-beginning-of-defun-1) | |
3062 | (while (not (or (bobp) (eq (char-after) ?{))) | |
3063 | (c-beginning-of-defun-1)) | |
3064 | (unless (equal (c-whack-state-before (point) res1) res2) | |
0ec1d2c5 | 3065 | (message (concat "c-parse-state inconsistency at %s: " |
0386b551 | 3066 | "using cache: %s, from scratch: %s") |
0ec1d2c5 | 3067 | here res1 res2)))) |
a66cd3ee | 3068 | res1)) |
0ec1d2c5 | 3069 | |
a66cd3ee MS |
3070 | (defun c-toggle-parse-state-debug (&optional arg) |
3071 | (interactive "P") | |
3072 | (setq c-debug-parse-state (c-calculate-state arg c-debug-parse-state)) | |
3073 | (fset 'c-parse-state (symbol-function (if c-debug-parse-state | |
3074 | 'c-debug-parse-state | |
3075 | 'c-real-parse-state))) | |
3076 | (c-keep-region-active)) | |
0386b551 AM |
3077 | (when c-debug-parse-state |
3078 | (c-toggle-parse-state-debug 1)) | |
a66cd3ee | 3079 | |
0ec1d2c5 | 3080 | \f |
d9e94c22 MS |
3081 | (defun c-whack-state-before (bufpos paren-state) |
3082 | ;; Whack off any state information from PAREN-STATE which lies | |
3083 | ;; before BUFPOS. Not destructive on PAREN-STATE. | |
d9e94c22 MS |
3084 | (let* ((newstate (list nil)) |
3085 | (ptr newstate) | |
3086 | car) | |
3087 | (while paren-state | |
3088 | (setq car (car paren-state) | |
3089 | paren-state (cdr paren-state)) | |
3090 | (if (< (if (consp car) (car car) car) bufpos) | |
3091 | (setq paren-state nil) | |
3092 | (setcdr ptr (list car)) | |
3093 | (setq ptr (cdr ptr)))) | |
3094 | (cdr newstate))) | |
3095 | ||
3096 | (defun c-whack-state-after (bufpos paren-state) | |
3097 | ;; Whack off any state information from PAREN-STATE which lies at or | |
3098 | ;; after BUFPOS. Not destructive on PAREN-STATE. | |
d9e94c22 MS |
3099 | (catch 'done |
3100 | (while paren-state | |
3101 | (let ((car (car paren-state))) | |
3102 | (if (consp car) | |
3103 | ;; just check the car, because in a balanced brace | |
3104 | ;; expression, it must be impossible for the corresponding | |
3105 | ;; close brace to be before point, but the open brace to | |
3106 | ;; be after. | |
3107 | (if (<= bufpos (car car)) | |
3108 | nil ; whack it off | |
3109 | (if (< bufpos (cdr car)) | |
3110 | ;; its possible that the open brace is before | |
3111 | ;; bufpos, but the close brace is after. In that | |
3112 | ;; case, convert this to a non-cons element. The | |
3113 | ;; rest of the state is before bufpos, so we're | |
3114 | ;; done. | |
3115 | (throw 'done (cons (car car) (cdr paren-state))) | |
3116 | ;; we know that both the open and close braces are | |
3117 | ;; before bufpos, so we also know that everything else | |
3118 | ;; on state is before bufpos. | |
3119 | (throw 'done paren-state))) | |
3120 | (if (<= bufpos car) | |
3121 | nil ; whack it off | |
3122 | ;; it's before bufpos, so everything else should too. | |
3123 | (throw 'done paren-state))) | |
3124 | (setq paren-state (cdr paren-state))) | |
3125 | nil))) | |
3126 | ||
3127 | (defun c-most-enclosing-brace (paren-state &optional bufpos) | |
3128 | ;; Return the bufpos of the innermost enclosing open paren before | |
0386b551 | 3129 | ;; bufpos, or nil if none was found. |
d9e94c22 MS |
3130 | (let (enclosingp) |
3131 | (or bufpos (setq bufpos 134217727)) | |
3132 | (while paren-state | |
3133 | (setq enclosingp (car paren-state) | |
3134 | paren-state (cdr paren-state)) | |
3135 | (if (or (consp enclosingp) | |
3136 | (>= enclosingp bufpos)) | |
3137 | (setq enclosingp nil) | |
d9e94c22 MS |
3138 | (setq paren-state nil))) |
3139 | enclosingp)) | |
3140 | ||
0386b551 AM |
3141 | (defun c-least-enclosing-brace (paren-state) |
3142 | ;; Return the bufpos of the outermost enclosing open paren, or nil | |
3143 | ;; if none was found. | |
d9e94c22 | 3144 | (let (pos elem) |
d9e94c22 MS |
3145 | (while paren-state |
3146 | (setq elem (car paren-state) | |
3147 | paren-state (cdr paren-state)) | |
0386b551 AM |
3148 | (if (integerp elem) |
3149 | (setq pos elem))) | |
d9e94c22 MS |
3150 | pos)) |
3151 | ||
3152 | (defun c-safe-position (bufpos paren-state) | |
0386b551 AM |
3153 | ;; Return the closest "safe" position recorded on PAREN-STATE that |
3154 | ;; is higher up than BUFPOS. Return nil if PAREN-STATE doesn't | |
3155 | ;; contain any. Return nil if BUFPOS is nil, which is useful to | |
3156 | ;; find the closest limit before a given limit that might be nil. | |
d9e94c22 | 3157 | ;; |
0386b551 AM |
3158 | ;; A "safe" position is a position at or after a recorded open |
3159 | ;; paren, or after a recorded close paren. The returned position is | |
3160 | ;; thus either the first position after a close brace, or the first | |
3161 | ;; position after an enclosing paren, or at the enclosing paren in | |
3162 | ;; case BUFPOS is immediately after it. | |
d9e94c22 MS |
3163 | (when bufpos |
3164 | (let (elem) | |
3165 | (catch 'done | |
3166 | (while paren-state | |
3167 | (setq elem (car paren-state)) | |
3168 | (if (consp elem) | |
3169 | (cond ((< (cdr elem) bufpos) | |
3170 | (throw 'done (cdr elem))) | |
3171 | ((< (car elem) bufpos) | |
3172 | ;; See below. | |
3173 | (throw 'done (min (1+ (car elem)) bufpos)))) | |
3174 | (if (< elem bufpos) | |
3175 | ;; elem is the position at and not after the opening paren, so | |
3176 | ;; we can go forward one more step unless it's equal to | |
3177 | ;; bufpos. This is useful in some cases avoid an extra paren | |
3178 | ;; level between the safe position and bufpos. | |
3179 | (throw 'done (min (1+ elem) bufpos)))) | |
3180 | (setq paren-state (cdr paren-state))))))) | |
3181 | ||
3182 | (defun c-beginning-of-syntax () | |
3183 | ;; This is used for `font-lock-beginning-of-syntax-function'. It | |
3184 | ;; goes to the closest previous point that is known to be outside | |
3185 | ;; any string literal or comment. `c-state-cache' is used if it has | |
3186 | ;; a position in the vicinity. | |
3187 | (let* ((paren-state c-state-cache) | |
3188 | elem | |
3189 | ||
3190 | (pos (catch 'done | |
3191 | ;; Note: Similar code in `c-safe-position'. The | |
3192 | ;; difference is that we accept a safe position at | |
3193 | ;; the point and don't bother to go forward past open | |
3194 | ;; parens. | |
3195 | (while paren-state | |
3196 | (setq elem (car paren-state)) | |
3197 | (if (consp elem) | |
3198 | (cond ((<= (cdr elem) (point)) | |
3199 | (throw 'done (cdr elem))) | |
3200 | ((<= (car elem) (point)) | |
3201 | (throw 'done (car elem)))) | |
3202 | (if (<= elem (point)) | |
3203 | (throw 'done elem))) | |
3204 | (setq paren-state (cdr paren-state))) | |
3205 | (point-min)))) | |
3206 | ||
3207 | (if (> pos (- (point) 4000)) | |
3208 | (goto-char pos) | |
3209 | ;; The position is far back. Try `c-beginning-of-defun-1' | |
3210 | ;; (although we can't be entirely sure it will go to a position | |
3211 | ;; outside a comment or string in current emacsen). FIXME: | |
3212 | ;; Consult `syntax-ppss' here. | |
3213 | (c-beginning-of-defun-1) | |
3214 | (if (< (point) pos) | |
3215 | (goto-char pos))))) | |
3216 | ||
3217 | \f | |
3218 | ;; Tools for scanning identifiers and other tokens. | |
3219 | ||
3220 | (defun c-on-identifier () | |
3221 | "Return non-nil if the point is on or directly after an identifier. | |
3222 | Keywords are recognized and not considered identifiers. If an | |
3223 | identifier is detected, the returned value is its starting position. | |
0386b551 AM |
3224 | If an identifier ends at the point and another begins at it \(can only |
3225 | happen in Pike) then the point for the preceding one is returned. | |
d9e94c22 | 3226 | |
0386b551 AM |
3227 | Note that this function might do hidden buffer changes. See the |
3228 | comment at the start of cc-engine.el for more info." | |
3229 | ||
3230 | ;; FIXME: Shouldn't this function handle "operator" in C++? | |
d9e94c22 MS |
3231 | |
3232 | (save-excursion | |
0386b551 AM |
3233 | (skip-syntax-backward "w_") |
3234 | ||
3235 | (or | |
3236 | ||
3237 | ;; Check for a normal (non-keyword) identifier. | |
3238 | (and (looking-at c-symbol-start) | |
3239 | (not (looking-at c-keywords-regexp)) | |
3240 | (point)) | |
3241 | ||
3242 | (when (c-major-mode-is 'pike-mode) | |
3243 | ;; Handle the `<operator> syntax in Pike. | |
3244 | (let ((pos (point))) | |
3245 | (skip-chars-backward "-!%&*+/<=>^|~[]()") | |
3246 | (and (if (< (skip-chars-backward "`") 0) | |
3247 | t | |
3248 | (goto-char pos) | |
3249 | (eq (char-after) ?\`)) | |
3250 | (looking-at c-symbol-key) | |
3251 | (>= (match-end 0) pos) | |
3252 | (point)))) | |
3253 | ||
3254 | ;; Handle the "operator +" syntax in C++. | |
3255 | (when (and c-overloadable-operators-regexp | |
3256 | (= (c-backward-token-2 0) 0)) | |
3257 | ||
3258 | (cond ((and (looking-at c-overloadable-operators-regexp) | |
51c9af45 | 3259 | (or (not c-opt-op-identifier-prefix) |
0386b551 | 3260 | (and (= (c-backward-token-2 1) 0) |
51c9af45 | 3261 | (looking-at c-opt-op-identifier-prefix)))) |
0386b551 AM |
3262 | (point)) |
3263 | ||
3264 | ((save-excursion | |
51c9af45 AM |
3265 | (and c-opt-op-identifier-prefix |
3266 | (looking-at c-opt-op-identifier-prefix) | |
0386b551 AM |
3267 | (= (c-forward-token-2 1) 0) |
3268 | (looking-at c-overloadable-operators-regexp))) | |
3269 | (point)))) | |
3270 | ||
3271 | ))) | |
d9e94c22 MS |
3272 | |
3273 | (defsubst c-simple-skip-symbol-backward () | |
3274 | ;; If the point is at the end of a symbol then skip backward to the | |
3275 | ;; beginning of it. Don't move otherwise. Return non-nil if point | |
3276 | ;; moved. | |
0386b551 AM |
3277 | ;; |
3278 | ;; This function might do hidden buffer changes. | |
d9e94c22 MS |
3279 | (or (< (skip-syntax-backward "w_") 0) |
3280 | (and (c-major-mode-is 'pike-mode) | |
3281 | ;; Handle the `<operator> syntax in Pike. | |
3282 | (let ((pos (point))) | |
2a15eb73 | 3283 | (if (and (< (skip-chars-backward "-!%&*+/<=>^|~[]()") 0) |
d9e94c22 MS |
3284 | (< (skip-chars-backward "`") 0) |
3285 | (looking-at c-symbol-key) | |
3286 | (>= (match-end 0) pos)) | |
3287 | t | |
3288 | (goto-char pos) | |
3289 | nil))))) | |
3290 | ||
0386b551 | 3291 | (defun c-beginning-of-current-token (&optional back-limit) |
d9e94c22 MS |
3292 | ;; Move to the beginning of the current token. Do not move if not |
3293 | ;; in the middle of one. BACK-LIMIT may be used to bound the | |
3294 | ;; backward search; if given it's assumed to be at the boundary | |
a85fd6da | 3295 | ;; between two tokens. Return non-nil if the point is moved, nil |
580fba94 | 3296 | ;; otherwise. |
0386b551 AM |
3297 | ;; |
3298 | ;; This function might do hidden buffer changes. | |
d9e94c22 | 3299 | (let ((start (point))) |
580fba94 AM |
3300 | (if (looking-at "\\w\\|\\s_") |
3301 | (skip-syntax-backward "w_" back-limit) | |
3302 | (when (< (skip-syntax-backward ".()" back-limit) 0) | |
3303 | (while (let ((pos (or (and (looking-at c-nonsymbol-token-regexp) | |
3304 | (match-end 0)) | |
3305 | ;; `c-nonsymbol-token-regexp' should always match | |
3306 | ;; since we've skipped backward over punctuator | |
3307 | ;; or paren syntax, but consume one char in case | |
3308 | ;; it doesn't so that we don't leave point before | |
3309 | ;; some earlier incorrect token. | |
3310 | (1+ (point))))) | |
3311 | (if (<= pos start) | |
3312 | (goto-char pos)))))) | |
3313 | (< (point) start))) | |
d9e94c22 | 3314 | |
ff959bab | 3315 | (defun c-end-of-current-token (&optional back-limit) |
d9e94c22 MS |
3316 | ;; Move to the end of the current token. Do not move if not in the |
3317 | ;; middle of one. BACK-LIMIT may be used to bound the backward | |
3318 | ;; search; if given it's assumed to be at the boundary between two | |
ff959bab | 3319 | ;; tokens. Return non-nil if the point is moved, nil otherwise. |
0386b551 AM |
3320 | ;; |
3321 | ;; This function might do hidden buffer changes. | |
d9e94c22 MS |
3322 | (let ((start (point))) |
3323 | (cond ((< (skip-syntax-backward "w_" (1- start)) 0) | |
3324 | (skip-syntax-forward "w_")) | |
3325 | ((< (skip-syntax-backward ".()" back-limit) 0) | |
3326 | (while (progn | |
3327 | (if (looking-at c-nonsymbol-token-regexp) | |
3328 | (goto-char (match-end 0)) | |
3329 | ;; `c-nonsymbol-token-regexp' should always match since | |
3330 | ;; we've skipped backward over punctuator or paren | |
3331 | ;; syntax, but move forward in case it doesn't so that | |
3332 | ;; we don't leave point earlier than we started with. | |
3333 | (forward-char)) | |
ff959bab MS |
3334 | (< (point) start))))) |
3335 | (> (point) start))) | |
d9e94c22 MS |
3336 | |
3337 | (defconst c-jump-syntax-balanced | |
3338 | (if (memq 'gen-string-delim c-emacs-features) | |
3339 | "\\w\\|\\s_\\|\\s\(\\|\\s\)\\|\\s\"\\|\\s|" | |
3340 | "\\w\\|\\s_\\|\\s\(\\|\\s\)\\|\\s\"")) | |
3341 | ||
3342 | (defconst c-jump-syntax-unbalanced | |
3343 | (if (memq 'gen-string-delim c-emacs-features) | |
3344 | "\\w\\|\\s_\\|\\s\"\\|\\s|" | |
3345 | "\\w\\|\\s_\\|\\s\"")) | |
3346 | ||
3347 | (defun c-forward-token-2 (&optional count balanced limit) | |
3348 | "Move forward by tokens. | |
3349 | A token is defined as all symbols and identifiers which aren't | |
3350 | syntactic whitespace \(note that multicharacter tokens like \"==\" are | |
3351 | treated properly). Point is always either left at the beginning of a | |
3352 | token or not moved at all. COUNT specifies the number of tokens to | |
3353 | move; a negative COUNT moves in the opposite direction. A COUNT of 0 | |
3354 | moves to the next token beginning only if not already at one. If | |
3355 | BALANCED is true, move over balanced parens, otherwise move into them. | |
3356 | Also, if BALANCED is true, never move out of an enclosing paren. | |
3357 | ||
3358 | LIMIT sets the limit for the movement and defaults to the point limit. | |
3359 | The case when LIMIT is set in the middle of a token, comment or macro | |
3360 | is handled correctly, i.e. the point won't be left there. | |
3361 | ||
3362 | Return the number of tokens left to move \(positive or negative). If | |
3363 | BALANCED is true, a move over a balanced paren counts as one. Note | |
3364 | that if COUNT is 0 and no appropriate token beginning is found, 1 will | |
3365 | be returned. Thus, a return value of 0 guarantees that point is at | |
3366 | the requested position and a return value less \(without signs) than | |
0386b551 AM |
3367 | COUNT guarantees that point is at the beginning of some token. |
3368 | ||
3369 | Note that this function might do hidden buffer changes. See the | |
3370 | comment at the start of cc-engine.el for more info." | |
d9e94c22 MS |
3371 | |
3372 | (or count (setq count 1)) | |
3373 | (if (< count 0) | |
3374 | (- (c-backward-token-2 (- count) balanced limit)) | |
3375 | ||
3376 | (let ((jump-syntax (if balanced | |
3377 | c-jump-syntax-balanced | |
3378 | c-jump-syntax-unbalanced)) | |
3379 | (last (point)) | |
3380 | (prev (point))) | |
3381 | ||
3382 | (if (zerop count) | |
3383 | ;; If count is zero we should jump if in the middle of a token. | |
3384 | (c-end-of-current-token)) | |
3385 | ||
3386 | (save-restriction | |
3387 | (if limit (narrow-to-region (point-min) limit)) | |
3388 | (if (/= (point) | |
3389 | (progn (c-forward-syntactic-ws) (point))) | |
3390 | ;; Skip whitespace. Count this as a move if we did in | |
3391 | ;; fact move. | |
3392 | (setq count (max (1- count) 0))) | |
3393 | ||
3394 | (if (eobp) | |
3395 | ;; Moved out of bounds. Make sure the returned count isn't zero. | |
3396 | (progn | |
3397 | (if (zerop count) (setq count 1)) | |
3398 | (goto-char last)) | |
3399 | ||
3400 | ;; Use `condition-case' to avoid having the limit tests | |
3401 | ;; inside the loop. | |
3402 | (condition-case nil | |
3403 | (while (and | |
3404 | (> count 0) | |
3405 | (progn | |
3406 | (setq last (point)) | |
3407 | (cond ((looking-at jump-syntax) | |
3408 | (goto-char (scan-sexps (point) 1)) | |
3409 | t) | |
3410 | ((looking-at c-nonsymbol-token-regexp) | |
3411 | (goto-char (match-end 0)) | |
3412 | t) | |
3413 | ;; `c-nonsymbol-token-regexp' above should always | |
3414 | ;; match if there are correct tokens. Try to | |
3415 | ;; widen to see if the limit was set in the | |
3416 | ;; middle of one, else fall back to treating | |
3417 | ;; the offending thing as a one character token. | |
3418 | ((and limit | |
3419 | (save-restriction | |
3420 | (widen) | |
3421 | (looking-at c-nonsymbol-token-regexp))) | |
3422 | nil) | |
3423 | (t | |
3424 | (forward-char) | |
3425 | t)))) | |
3426 | (c-forward-syntactic-ws) | |
3427 | (setq prev last | |
3428 | count (1- count))) | |
3429 | (error (goto-char last))) | |
3430 | ||
3431 | (when (eobp) | |
3432 | (goto-char prev) | |
3433 | (setq count (1+ count))))) | |
3434 | ||
3435 | count))) | |
3436 | ||
3437 | (defun c-backward-token-2 (&optional count balanced limit) | |
3438 | "Move backward by tokens. | |
3439 | See `c-forward-token-2' for details." | |
3440 | ||
3441 | (or count (setq count 1)) | |
3442 | (if (< count 0) | |
3443 | (- (c-forward-token-2 (- count) balanced limit)) | |
3444 | ||
3445 | (or limit (setq limit (point-min))) | |
3446 | (let ((jump-syntax (if balanced | |
3447 | c-jump-syntax-balanced | |
3448 | c-jump-syntax-unbalanced)) | |
3449 | (last (point))) | |
3450 | ||
3451 | (if (zerop count) | |
3452 | ;; The count is zero so try to skip to the beginning of the | |
3453 | ;; current token. | |
3454 | (if (> (point) | |
3455 | (progn (c-beginning-of-current-token) (point))) | |
3456 | (if (< (point) limit) | |
3457 | ;; The limit is inside the same token, so return 1. | |
3458 | (setq count 1)) | |
3459 | ||
3460 | ;; We're not in the middle of a token. If there's | |
3461 | ;; whitespace after the point then we must move backward, | |
3462 | ;; so set count to 1 in that case. | |
3463 | (and (looking-at c-syntactic-ws-start) | |
3464 | ;; If we're looking at a '#' that might start a cpp | |
3465 | ;; directive then we have to do a more elaborate check. | |
3466 | (or (/= (char-after) ?#) | |
3467 | (not c-opt-cpp-prefix) | |
3468 | (save-excursion | |
3469 | (and (= (point) | |
3470 | (progn (beginning-of-line) | |
3471 | (looking-at "[ \t]*") | |
3472 | (match-end 0))) | |
3473 | (or (bobp) | |
3474 | (progn (backward-char) | |
3475 | (not (eq (char-before) ?\\))))))) | |
3476 | (setq count 1)))) | |
3477 | ||
3478 | ;; Use `condition-case' to avoid having to check for buffer | |
3479 | ;; limits in `backward-char', `scan-sexps' and `goto-char' below. | |
3480 | (condition-case nil | |
3481 | (while (and | |
3482 | (> count 0) | |
3483 | (progn | |
3484 | (c-backward-syntactic-ws) | |
3485 | (backward-char) | |
3486 | (if (looking-at jump-syntax) | |
3487 | (goto-char (scan-sexps (1+ (point)) -1)) | |
3488 | ;; This can be very inefficient if there's a long | |
3489 | ;; sequence of operator tokens without any separation. | |
3490 | ;; That doesn't happen in practice, anyway. | |
3491 | (c-beginning-of-current-token)) | |
3492 | (>= (point) limit))) | |
3493 | (setq last (point) | |
3494 | count (1- count))) | |
3495 | (error (goto-char last))) | |
3496 | ||
3497 | (if (< (point) limit) | |
3498 | (goto-char last)) | |
3499 | ||
3500 | count))) | |
3501 | ||
3502 | (defun c-forward-token-1 (&optional count balanced limit) | |
3503 | "Like `c-forward-token-2' but doesn't treat multicharacter operator | |
3504 | tokens like \"==\" as single tokens, i.e. all sequences of symbol | |
3505 | characters are jumped over character by character. This function is | |
3506 | for compatibility only; it's only a wrapper over `c-forward-token-2'." | |
3507 | (let ((c-nonsymbol-token-regexp "\\s.\\|\\s\(\\|\\s\)")) | |
3508 | (c-forward-token-2 count balanced limit))) | |
3509 | ||
3510 | (defun c-backward-token-1 (&optional count balanced limit) | |
3511 | "Like `c-backward-token-2' but doesn't treat multicharacter operator | |
3512 | tokens like \"==\" as single tokens, i.e. all sequences of symbol | |
3513 | characters are jumped over character by character. This function is | |
3514 | for compatibility only; it's only a wrapper over `c-backward-token-2'." | |
3515 | (let ((c-nonsymbol-token-regexp "\\s.\\|\\s\(\\|\\s\)")) | |
3516 | (c-backward-token-2 count balanced limit))) | |
3517 | ||
3518 | \f | |
3519 | ;; Tools for doing searches restricted to syntactically relevant text. | |
3520 | ||
3521 | (defun c-syntactic-re-search-forward (regexp &optional bound noerror | |
3522 | paren-level not-inside-token | |
3523 | lookbehind-submatch) | |
3524 | "Like `re-search-forward', but only report matches that are found | |
3525 | in syntactically significant text. I.e. matches in comments, macros | |
3526 | or string literals are ignored. The start point is assumed to be | |
3527 | outside any comment, macro or string literal, or else the content of | |
3528 | that region is taken as syntactically significant text. | |
3529 | ||
3530 | If PAREN-LEVEL is non-nil, an additional restriction is added to | |
2a15eb73 MS |
3531 | ignore matches in nested paren sexps. The search will also not go |
3532 | outside the current list sexp, which has the effect that if the point | |
3533 | should be moved to BOUND when no match is found \(i.e. NOERROR is | |
3534 | neither nil nor t), then it will be at the closing paren if the end of | |
3535 | the current list sexp is encountered first. | |
d9e94c22 MS |
3536 | |
3537 | If NOT-INSIDE-TOKEN is non-nil, matches in the middle of tokens are | |
3538 | ignored. Things like multicharacter operators and special symbols | |
3539 | \(e.g. \"`()\" in Pike) are handled but currently not floating point | |
3540 | constants. | |
3541 | ||
3542 | If LOOKBEHIND-SUBMATCH is non-nil, it's taken as a number of a | |
3543 | subexpression in REGEXP. The end of that submatch is used as the | |
3544 | position to check for syntactic significance. If LOOKBEHIND-SUBMATCH | |
3545 | isn't used or if that subexpression didn't match then the start | |
3546 | position of the whole match is used instead. The \"look behind\" | |
3547 | subexpression is never tested before the starting position, so it | |
3548 | might be a good idea to include \\=\\= as a match alternative in it. | |
3549 | ||
3550 | Optimization note: Matches might be missed if the \"look behind\" | |
2a15eb73 | 3551 | subexpression can match the end of nonwhite syntactic whitespace, |
d9e94c22 | 3552 | i.e. the end of comments or cpp directives. This since the function |
2a15eb73 MS |
3553 | skips over such things before resuming the search. It's on the other |
3554 | hand not safe to assume that the \"look behind\" subexpression never | |
3555 | matches syntactic whitespace. | |
3556 | ||
3557 | Bug: Unbalanced parens inside cpp directives are currently not handled | |
3558 | correctly \(i.e. they don't get ignored as they should) when | |
0386b551 AM |
3559 | PAREN-LEVEL is set. |
3560 | ||
3561 | Note that this function might do hidden buffer changes. See the | |
3562 | comment at the start of cc-engine.el for more info." | |
d9e94c22 MS |
3563 | |
3564 | (or bound (setq bound (point-max))) | |
3565 | (if paren-level (setq paren-level -1)) | |
3566 | ||
3567 | ;;(message "c-syntactic-re-search-forward %s %s %S" (point) bound regexp) | |
3568 | ||
3569 | (let ((start (point)) | |
2a15eb73 MS |
3570 | tmp |
3571 | ;; Start position for the last search. | |
3572 | search-pos | |
3573 | ;; The `parse-partial-sexp' state between the start position | |
3574 | ;; and the point. | |
3575 | state | |
3576 | ;; The current position after the last state update. The next | |
3577 | ;; `parse-partial-sexp' continues from here. | |
3578 | (state-pos (point)) | |
3579 | ;; The position at which to check the state and the state | |
3580 | ;; there. This is separate from `state-pos' since we might | |
3581 | ;; need to back up before doing the next search round. | |
3582 | check-pos check-state | |
3583 | ;; Last position known to end a token. | |
d9e94c22 | 3584 | (last-token-end-pos (point-min)) |
2a15eb73 MS |
3585 | ;; Set when a valid match is found. |
3586 | found) | |
d9e94c22 MS |
3587 | |
3588 | (condition-case err | |
3589 | (while | |
3590 | (and | |
2a15eb73 MS |
3591 | (progn |
3592 | (setq search-pos (point)) | |
3593 | (re-search-forward regexp bound noerror)) | |
d9e94c22 MS |
3594 | |
3595 | (progn | |
2a15eb73 MS |
3596 | (setq state (parse-partial-sexp |
3597 | state-pos (match-beginning 0) paren-level nil state) | |
3598 | state-pos (point)) | |
d9e94c22 | 3599 | (if (setq check-pos (and lookbehind-submatch |
2a15eb73 MS |
3600 | (or (not paren-level) |
3601 | (>= (car state) 0)) | |
d9e94c22 MS |
3602 | (match-end lookbehind-submatch))) |
3603 | (setq check-state (parse-partial-sexp | |
2a15eb73 MS |
3604 | state-pos check-pos paren-level nil state)) |
3605 | (setq check-pos state-pos | |
d9e94c22 MS |
3606 | check-state state)) |
3607 | ||
2a15eb73 MS |
3608 | ;; NOTE: If we got a look behind subexpression and get |
3609 | ;; an insignificant match in something that isn't | |
d9e94c22 MS |
3610 | ;; syntactic whitespace (i.e. strings or in nested |
3611 | ;; parentheses), then we can never skip more than a | |
2a15eb73 MS |
3612 | ;; single character from the match start position |
3613 | ;; (i.e. `state-pos' here) before continuing the | |
3614 | ;; search. That since the look behind subexpression | |
3615 | ;; might match the end of the insignificant region in | |
3616 | ;; the next search. | |
d9e94c22 MS |
3617 | |
3618 | (cond | |
d9e94c22 MS |
3619 | ((elt check-state 7) |
3620 | ;; Match inside a line comment. Skip to eol. Use | |
3621 | ;; `re-search-forward' instead of `skip-chars-forward' to get | |
3622 | ;; the right bound behavior. | |
3623 | (re-search-forward "[\n\r]" bound noerror)) | |
3624 | ||
3625 | ((elt check-state 4) | |
3626 | ;; Match inside a block comment. Skip to the '*/'. | |
3627 | (search-forward "*/" bound noerror)) | |
3628 | ||
3629 | ((and (not (elt check-state 5)) | |
3630 | (eq (char-before check-pos) ?/) | |
2a15eb73 | 3631 | (not (c-get-char-property (1- check-pos) 'syntax-table)) |
d9e94c22 MS |
3632 | (memq (char-after check-pos) '(?/ ?*))) |
3633 | ;; Match in the middle of the opener of a block or line | |
3634 | ;; comment. | |
3635 | (if (= (char-after check-pos) ?/) | |
3636 | (re-search-forward "[\n\r]" bound noerror) | |
3637 | (search-forward "*/" bound noerror))) | |
3638 | ||
2a15eb73 MS |
3639 | ;; The last `parse-partial-sexp' above might have |
3640 | ;; stopped short of the real check position if the end | |
3641 | ;; of the current sexp was encountered in paren-level | |
3642 | ;; mode. The checks above are always false in that | |
3643 | ;; case, and since they can do better skipping in | |
3644 | ;; lookbehind-submatch mode, we do them before | |
3645 | ;; checking the paren level. | |
3646 | ||
3647 | ((and paren-level | |
3648 | (/= (setq tmp (car check-state)) 0)) | |
3649 | ;; Check the paren level first since we're short of the | |
3650 | ;; syntactic checking position if the end of the | |
3651 | ;; current sexp was encountered by `parse-partial-sexp'. | |
3652 | (if (> tmp 0) | |
3653 | ||
3654 | ;; Inside a nested paren sexp. | |
3655 | (if lookbehind-submatch | |
3656 | ;; See the NOTE above. | |
3657 | (progn (goto-char state-pos) t) | |
3658 | ;; Skip out of the paren quickly. | |
3659 | (setq state (parse-partial-sexp state-pos bound 0 nil state) | |
3660 | state-pos (point))) | |
3661 | ||
3662 | ;; Have exited the current paren sexp. | |
3663 | (if noerror | |
3664 | (progn | |
3665 | ;; The last `parse-partial-sexp' call above | |
3666 | ;; has left us just after the closing paren | |
3667 | ;; in this case, so we can modify the bound | |
3668 | ;; to leave the point at the right position | |
3669 | ;; upon return. | |
3670 | (setq bound (1- (point))) | |
3671 | nil) | |
3672 | (signal 'search-failed (list regexp))))) | |
3673 | ||
3674 | ((setq tmp (elt check-state 3)) | |
3675 | ;; Match inside a string. | |
3676 | (if (or lookbehind-submatch | |
3677 | (not (integerp tmp))) | |
3678 | ;; See the NOTE above. | |
3679 | (progn (goto-char state-pos) t) | |
3680 | ;; Skip to the end of the string before continuing. | |
3681 | (let ((ender (make-string 1 tmp)) (continue t)) | |
3682 | (while (if (search-forward ender bound noerror) | |
3683 | (progn | |
3684 | (setq state (parse-partial-sexp | |
3685 | state-pos (point) nil nil state) | |
3686 | state-pos (point)) | |
3687 | (elt state 3)) | |
3688 | (setq continue nil))) | |
3689 | continue))) | |
d9e94c22 MS |
3690 | |
3691 | ((save-excursion | |
3692 | (save-match-data | |
3693 | (c-beginning-of-macro start))) | |
3694 | ;; Match inside a macro. Skip to the end of it. | |
3695 | (c-end-of-macro) | |
3696 | (cond ((<= (point) bound) t) | |
3697 | (noerror nil) | |
2a15eb73 | 3698 | (t (signal 'search-failed (list regexp))))) |
d9e94c22 | 3699 | |
2a15eb73 MS |
3700 | ((and not-inside-token |
3701 | (or (< check-pos last-token-end-pos) | |
3702 | (< check-pos | |
3703 | (save-excursion | |
3704 | (goto-char check-pos) | |
3705 | (save-match-data | |
3706 | (c-end-of-current-token last-token-end-pos)) | |
3707 | (setq last-token-end-pos (point)))))) | |
3708 | ;; Inside a token. | |
3709 | (if lookbehind-submatch | |
3710 | ;; See the NOTE above. | |
3711 | (goto-char state-pos) | |
3712 | (goto-char (min last-token-end-pos bound)))) | |
d9e94c22 MS |
3713 | |
3714 | (t | |
3715 | ;; A real match. | |
3716 | (setq found t) | |
2a15eb73 MS |
3717 | nil))) |
3718 | ||
3719 | ;; Should loop to search again, but take care to avoid | |
3720 | ;; looping on the same spot. | |
3721 | (or (/= search-pos (point)) | |
3722 | (if (= (point) bound) | |
3723 | (if noerror | |
3724 | nil | |
3725 | (signal 'search-failed (list regexp))) | |
3726 | (forward-char) | |
3727 | t)))) | |
d9e94c22 MS |
3728 | |
3729 | (error | |
3730 | (goto-char start) | |
3731 | (signal (car err) (cdr err)))) | |
3732 | ||
2a15eb73 | 3733 | ;;(message "c-syntactic-re-search-forward done %s" (or (match-end 0) (point))) |
d9e94c22 MS |
3734 | |
3735 | (if found | |
3736 | (progn | |
2a15eb73 MS |
3737 | (goto-char (match-end 0)) |
3738 | (match-end 0)) | |
d9e94c22 MS |
3739 | |
3740 | ;; Search failed. Set point as appropriate. | |
2a15eb73 MS |
3741 | (if (eq noerror t) |
3742 | (goto-char start) | |
3743 | (goto-char bound)) | |
d9e94c22 MS |
3744 | nil))) |
3745 | ||
47641aac GM |
3746 | (defvar safe-pos-list) ; bound in c-syntactic-skip-backward |
3747 | ||
d0fcee66 AM |
3748 | (defsubst c-ssb-lit-begin () |
3749 | ;; Return the start of the literal point is in, or nil. | |
3750 | ;; We read and write the variables `safe-pos', `safe-pos-list', `state' | |
3751 | ;; bound in the caller. | |
3752 | ||
3753 | ;; Use `parse-partial-sexp' from a safe position down to the point to check | |
3754 | ;; if it's outside comments and strings. | |
3755 | (save-excursion | |
3756 | (let ((pos (point)) safe-pos state pps-end-pos) | |
3757 | ;; Pick a safe position as close to the point as possible. | |
3758 | ;; | |
3759 | ;; FIXME: Consult `syntax-ppss' here if our cache doesn't give a good | |
3760 | ;; position. | |
47641aac | 3761 | |
d0fcee66 AM |
3762 | (while (and safe-pos-list |
3763 | (> (car safe-pos-list) (point))) | |
3764 | (setq safe-pos-list (cdr safe-pos-list))) | |
3765 | (unless (setq safe-pos (car-safe safe-pos-list)) | |
3766 | (setq safe-pos (max (or (c-safe-position | |
3767 | (point) (or c-state-cache | |
3768 | (c-parse-state))) | |
3769 | 0) | |
3770 | (point-min)) | |
3771 | safe-pos-list (list safe-pos))) | |
3772 | ||
3773 | ;; Cache positions along the way to use if we have to back up more. We | |
3774 | ;; cache every closing paren on the same level. If the paren cache is | |
3775 | ;; relevant in this region then we're typically already on the same | |
3776 | ;; level as the target position. Note that we might cache positions | |
3777 | ;; after opening parens in case safe-pos is in a nested list. That's | |
3778 | ;; both uncommon and harmless. | |
3779 | (while (progn | |
3780 | (setq state (parse-partial-sexp | |
3781 | safe-pos pos 0)) | |
3782 | (< (point) pos)) | |
3783 | (setq safe-pos (point) | |
3784 | safe-pos-list (cons safe-pos safe-pos-list))) | |
3785 | ||
3786 | ;; If the state contains the start of the containing sexp we cache that | |
3787 | ;; position too, so that parse-partial-sexp in the next run has a bigger | |
3788 | ;; chance of starting at the same level as the target position and thus | |
3789 | ;; will get more good safe positions into the list. | |
3790 | (if (elt state 1) | |
3791 | (setq safe-pos (1+ (elt state 1)) | |
3792 | safe-pos-list (cons safe-pos safe-pos-list))) | |
3793 | ||
3794 | (if (or (elt state 3) (elt state 4)) | |
3795 | ;; Inside string or comment. Continue search at the | |
3796 | ;; beginning of it. | |
3797 | (elt state 8))))) | |
3798 | ||
0386b551 | 3799 | (defun c-syntactic-skip-backward (skip-chars &optional limit paren-level) |
d9e94c22 MS |
3800 | "Like `skip-chars-backward' but only look at syntactically relevant chars, |
3801 | i.e. don't stop at positions inside syntactic whitespace or string | |
3802 | literals. Preprocessor directives are also ignored, with the exception | |
3803 | of the one that the point starts within, if any. If LIMIT is given, | |
0386b551 AM |
3804 | it's assumed to be at a syntactically relevant position. |
3805 | ||
3806 | If PAREN-LEVEL is non-nil, the function won't stop in nested paren | |
3807 | sexps, and the search will also not go outside the current paren sexp. | |
3808 | However, if LIMIT or the buffer limit is reached inside a nested paren | |
3809 | then the point will be left at the limit. | |
3810 | ||
3811 | Non-nil is returned if the point moved, nil otherwise. | |
3812 | ||
3813 | Note that this function might do hidden buffer changes. See the | |
3814 | comment at the start of cc-engine.el for more info." | |
d9e94c22 MS |
3815 | |
3816 | (let ((start (point)) | |
d0fcee66 | 3817 | state-2 |
d9e94c22 MS |
3818 | ;; A list of syntactically relevant positions in descending |
3819 | ;; order. It's used to avoid scanning repeatedly over | |
3820 | ;; potentially large regions with `parse-partial-sexp' to verify | |
d0fcee66 | 3821 | ;; each position. Used in `c-ssb-lit-begin' |
d9e94c22 MS |
3822 | safe-pos-list |
3823 | ;; The result from `c-beginning-of-macro' at the start position or the | |
3824 | ;; start position itself if it isn't within a macro. Evaluated on | |
3825 | ;; demand. | |
0386b551 AM |
3826 | start-macro-beg |
3827 | ;; The earliest position after the current one with the same paren | |
3828 | ;; level. Used only when `paren-level' is set. | |
d0fcee66 | 3829 | lit-beg |
0386b551 | 3830 | (paren-level-pos (point))) |
d9e94c22 | 3831 | |
d0fcee66 AM |
3832 | (while |
3833 | (progn | |
3834 | ;; The next loop "tries" to find the end point each time round, | |
3835 | ;; loops when it hasn't succeeded. | |
3836 | (while | |
3837 | (and | |
3838 | (< (skip-chars-backward skip-chars limit) 0) | |
3839 | ||
3840 | (let ((pos (point)) state-2 pps-end-pos) | |
3841 | ||
3842 | (cond | |
3843 | ;; Don't stop inside a literal | |
3844 | ((setq lit-beg (c-ssb-lit-begin)) | |
3845 | (goto-char lit-beg) | |
3846 | t) | |
3847 | ||
3848 | ((and paren-level | |
3849 | (save-excursion | |
3850 | (setq state-2 (parse-partial-sexp | |
3851 | pos paren-level-pos -1) | |
3852 | pps-end-pos (point)) | |
3853 | (/= (car state-2) 0))) | |
3854 | ;; Not at the right level. | |
3855 | ||
3856 | (if (and (< (car state-2) 0) | |
3857 | ;; We stop above if we go out of a paren. | |
3858 | ;; Now check whether it precedes or is | |
3859 | ;; nested in the starting sexp. | |
3860 | (save-excursion | |
3861 | (setq state-2 | |
3862 | (parse-partial-sexp | |
3863 | pps-end-pos paren-level-pos | |
3864 | nil nil state-2)) | |
3865 | (< (car state-2) 0))) | |
3866 | ||
3867 | ;; We've stopped short of the starting position | |
3868 | ;; so the hit was inside a nested list. Go up | |
3869 | ;; until we are at the right level. | |
3870 | (condition-case nil | |
3871 | (progn | |
3872 | (goto-char (scan-lists pos -1 | |
3873 | (- (car state-2)))) | |
3874 | (setq paren-level-pos (point)) | |
3875 | (if (and limit (>= limit paren-level-pos)) | |
0386b551 | 3876 | (progn |
d0fcee66 AM |
3877 | (goto-char limit) |
3878 | nil) | |
3879 | t)) | |
3880 | (error | |
3881 | (goto-char (or limit (point-min))) | |
3882 | nil)) | |
3883 | ||
3884 | ;; The hit was outside the list at the start | |
3885 | ;; position. Go to the start of the list and exit. | |
3886 | (goto-char (1+ (elt state-2 1))) | |
3887 | nil)) | |
3888 | ||
3889 | ((c-beginning-of-macro limit) | |
3890 | ;; Inside a macro. | |
3891 | (if (< (point) | |
3892 | (or start-macro-beg | |
3893 | (setq start-macro-beg | |
3894 | (save-excursion | |
3895 | (goto-char start) | |
3896 | (c-beginning-of-macro limit) | |
3897 | (point))))) | |
3898 | t | |
3899 | ||
3900 | ;; It's inside the same macro we started in so it's | |
3901 | ;; a relevant match. | |
3902 | (goto-char pos) | |
3903 | nil)))))) | |
3904 | ||
3905 | (> (point) | |
3906 | (progn | |
3907 | ;; Skip syntactic ws afterwards so that we don't stop at the | |
3908 | ;; end of a comment if `skip-chars' is something like "^/". | |
3909 | (c-backward-syntactic-ws) | |
3910 | (point))))) | |
d9e94c22 | 3911 | |
0386b551 AM |
3912 | ;; We might want to extend this with more useful return values in |
3913 | ;; the future. | |
3914 | (/= (point) start))) | |
3915 | ||
3916 | ;; The following is an alternative implementation of | |
3917 | ;; `c-syntactic-skip-backward' that uses backward movement to keep | |
3918 | ;; track of the syntactic context. It turned out to be generally | |
3919 | ;; slower than the one above which uses forward checks from earlier | |
3920 | ;; safe positions. | |
3921 | ;; | |
3922 | ;;(defconst c-ssb-stop-re | |
3923 | ;; ;; The regexp matching chars `c-syntactic-skip-backward' needs to | |
3924 | ;; ;; stop at to avoid going into comments and literals. | |
3925 | ;; (concat | |
3926 | ;; ;; Match comment end syntax and string literal syntax. Also match | |
3927 | ;; ;; '/' for block comment endings (not covered by comment end | |
3928 | ;; ;; syntax). | |
3929 | ;; "\\s>\\|/\\|\\s\"" | |
3930 | ;; (if (memq 'gen-string-delim c-emacs-features) | |
3931 | ;; "\\|\\s|" | |
3932 | ;; "") | |
3933 | ;; (if (memq 'gen-comment-delim c-emacs-features) | |
3934 | ;; "\\|\\s!" | |
3935 | ;; ""))) | |
3936 | ;; | |
3937 | ;;(defconst c-ssb-stop-paren-re | |
3938 | ;; ;; Like `c-ssb-stop-re' but also stops at paren chars. | |
3939 | ;; (concat c-ssb-stop-re "\\|\\s(\\|\\s)")) | |
3940 | ;; | |
3941 | ;;(defconst c-ssb-sexp-end-re | |
3942 | ;; ;; Regexp matching the ending syntax of a complex sexp. | |
3943 | ;; (concat c-string-limit-regexp "\\|\\s)")) | |
3944 | ;; | |
3945 | ;;(defun c-syntactic-skip-backward (skip-chars &optional limit paren-level) | |
3946 | ;; "Like `skip-chars-backward' but only look at syntactically relevant chars, | |
3947 | ;;i.e. don't stop at positions inside syntactic whitespace or string | |
3948 | ;;literals. Preprocessor directives are also ignored. However, if the | |
3949 | ;;point is within a comment, string literal or preprocessor directory to | |
3950 | ;;begin with, its contents is treated as syntactically relevant chars. | |
3951 | ;;If LIMIT is given, it limits the backward search and the point will be | |
3952 | ;;left there if no earlier position is found. | |
3953 | ;; | |
3954 | ;;If PAREN-LEVEL is non-nil, the function won't stop in nested paren | |
3955 | ;;sexps, and the search will also not go outside the current paren sexp. | |
3956 | ;;However, if LIMIT or the buffer limit is reached inside a nested paren | |
3957 | ;;then the point will be left at the limit. | |
3958 | ;; | |
3959 | ;;Non-nil is returned if the point moved, nil otherwise. | |
3960 | ;; | |
3961 | ;;Note that this function might do hidden buffer changes. See the | |
3962 | ;;comment at the start of cc-engine.el for more info." | |
3963 | ;; | |
3964 | ;; (save-restriction | |
3965 | ;; (when limit | |
3966 | ;; (narrow-to-region limit (point-max))) | |
3967 | ;; | |
3968 | ;; (let ((start (point))) | |
3969 | ;; (catch 'done | |
3970 | ;; (while (let ((last-pos (point)) | |
3971 | ;; (stop-pos (progn | |
3972 | ;; (skip-chars-backward skip-chars) | |
3973 | ;; (point)))) | |
3974 | ;; | |
3975 | ;; ;; Skip back over the same region as | |
3976 | ;; ;; `skip-chars-backward' above, but keep to | |
3977 | ;; ;; syntactically relevant positions. | |
3978 | ;; (goto-char last-pos) | |
3979 | ;; (while (and | |
3980 | ;; ;; `re-search-backward' with a single char regexp | |
3981 | ;; ;; should be fast. | |
3982 | ;; (re-search-backward | |
3983 | ;; (if paren-level c-ssb-stop-paren-re c-ssb-stop-re) | |
3984 | ;; stop-pos 'move) | |
3985 | ;; | |
3986 | ;; (progn | |
3987 | ;; (cond | |
3988 | ;; ((looking-at "\\s(") | |
3989 | ;; ;; `paren-level' is set and we've found the | |
3990 | ;; ;; start of the containing paren. | |
3991 | ;; (forward-char) | |
3992 | ;; (throw 'done t)) | |
3993 | ;; | |
3994 | ;; ((looking-at c-ssb-sexp-end-re) | |
3995 | ;; ;; We're at the end of a string literal or paren | |
3996 | ;; ;; sexp (if `paren-level' is set). | |
3997 | ;; (forward-char) | |
3998 | ;; (condition-case nil | |
3999 | ;; (c-backward-sexp) | |
4000 | ;; (error | |
4001 | ;; (goto-char limit) | |
4002 | ;; (throw 'done t)))) | |
4003 | ;; | |
4004 | ;; (t | |
4005 | ;; (forward-char) | |
4006 | ;; ;; At the end of some syntactic ws or possibly | |
4007 | ;; ;; after a plain '/' operator. | |
4008 | ;; (let ((pos (point))) | |
4009 | ;; (c-backward-syntactic-ws) | |
4010 | ;; (if (= pos (point)) | |
4011 | ;; ;; Was a plain '/' operator. Go past it. | |
4012 | ;; (backward-char))))) | |
4013 | ;; | |
4014 | ;; (> (point) stop-pos)))) | |
4015 | ;; | |
4016 | ;; ;; Now the point is either at `stop-pos' or at some | |
4017 | ;; ;; position further back if `stop-pos' was at a | |
4018 | ;; ;; syntactically irrelevant place. | |
4019 | ;; | |
4020 | ;; ;; Skip additional syntactic ws so that we don't stop | |
4021 | ;; ;; at the end of a comment if `skip-chars' is | |
4022 | ;; ;; something like "^/". | |
4023 | ;; (c-backward-syntactic-ws) | |
4024 | ;; | |
4025 | ;; (< (point) stop-pos)))) | |
4026 | ;; | |
4027 | ;; ;; We might want to extend this with more useful return values | |
4028 | ;; ;; in the future. | |
4029 | ;; (/= (point) start)))) | |
d9e94c22 MS |
4030 | |
4031 | \f | |
4032 | ;; Tools for handling comments and string literals. | |
4033 | ||
4034 | (defun c-slow-in-literal (&optional lim detect-cpp) | |
4035 | "Return the type of literal point is in, if any. | |
4036 | The return value is `c' if in a C-style comment, `c++' if in a C++ | |
4037 | style comment, `string' if in a string literal, `pound' if DETECT-CPP | |
4038 | is non-nil and in a preprocessor line, or nil if somewhere else. | |
4039 | Optional LIM is used as the backward limit of the search. If omitted, | |
4040 | or nil, `c-beginning-of-defun' is used. | |
4041 | ||
4042 | The last point calculated is cached if the cache is enabled, i.e. if | |
4043 | `c-in-literal-cache' is bound to a two element vector. | |
4044 | ||
0386b551 AM |
4045 | Note that this function might do hidden buffer changes. See the |
4046 | comment at the start of cc-engine.el for more info." | |
4047 | ||
d9e94c22 MS |
4048 | (if (and (vectorp c-in-literal-cache) |
4049 | (= (point) (aref c-in-literal-cache 0))) | |
4050 | (aref c-in-literal-cache 1) | |
4051 | (let ((rtn (save-excursion | |
4052 | (let* ((pos (point)) | |
4053 | (lim (or lim (progn | |
4054 | (c-beginning-of-syntax) | |
4055 | (point)))) | |
4056 | (state (parse-partial-sexp lim pos))) | |
4057 | (cond | |
4058 | ((elt state 3) 'string) | |
4059 | ((elt state 4) (if (elt state 7) 'c++ 'c)) | |
4060 | ((and detect-cpp (c-beginning-of-macro lim)) 'pound) | |
4061 | (t nil)))))) | |
4062 | ;; cache this result if the cache is enabled | |
4063 | (if (not c-in-literal-cache) | |
4064 | (setq c-in-literal-cache (vector (point) rtn))) | |
4065 | rtn))) | |
4066 | ||
4067 | ;; XEmacs has a built-in function that should make this much quicker. | |
4068 | ;; I don't think we even need the cache, which makes our lives more | |
4069 | ;; complicated anyway. In this case, lim is only used to detect | |
4070 | ;; cpp directives. | |
4071 | ;; | |
4072 | ;; Note that there is a bug in Xemacs's buffer-syntactic-context when used in | |
4073 | ;; conjunction with syntax-table-properties. The bug is present in, e.g., | |
4074 | ;; Xemacs 21.4.4. It manifested itself thus: | |
4075 | ;; | |
4076 | ;; Starting with an empty AWK Mode buffer, type | |
4077 | ;; /regexp/ {<C-j> | |
4078 | ;; Point gets wrongly left at column 0, rather than being indented to tab-width. | |
4079 | ;; | |
4080 | ;; AWK Mode is designed such that when the first / is typed, it gets the | |
4081 | ;; syntax-table property "string fence". When the second / is typed, BOTH /s | |
4082 | ;; are given the s-t property "string". However, buffer-syntactic-context | |
4083 | ;; fails to take account of the change of the s-t property on the opening / to | |
4084 | ;; "string", and reports that the { is within a string started by the second /. | |
4085 | ;; | |
4086 | ;; The workaround for this is for the AWK Mode initialisation to switch the | |
4087 | ;; defalias for c-in-literal to c-slow-in-literal. This will slow down other | |
4088 | ;; cc-modes in Xemacs whenever an awk-buffer has been initialised. | |
b414f371 | 4089 | ;; |
d9e94c22 MS |
4090 | ;; (Alan Mackenzie, 2003/4/30). |
4091 | ||
4092 | (defun c-fast-in-literal (&optional lim detect-cpp) | |
0386b551 | 4093 | ;; This function might do hidden buffer changes. |
d9e94c22 MS |
4094 | (let ((context (buffer-syntactic-context))) |
4095 | (cond | |
4096 | ((eq context 'string) 'string) | |
4097 | ((eq context 'comment) 'c++) | |
4098 | ((eq context 'block-comment) 'c) | |
4099 | ((and detect-cpp (save-excursion (c-beginning-of-macro lim))) 'pound)))) | |
4100 | ||
4101 | (defalias 'c-in-literal | |
4102 | (if (fboundp 'buffer-syntactic-context) | |
7bfc3fdb | 4103 | 'c-fast-in-literal ; XEmacs |
d9e94c22 MS |
4104 | 'c-slow-in-literal)) ; GNU Emacs |
4105 | ||
4106 | ;; The defalias above isn't enough to shut up the byte compiler. | |
4107 | (cc-bytecomp-defun c-in-literal) | |
4108 | ||
4109 | (defun c-literal-limits (&optional lim near not-in-delimiter) | |
4110 | "Return a cons of the beginning and end positions of the comment or | |
4111 | string surrounding point (including both delimiters), or nil if point | |
4112 | isn't in one. If LIM is non-nil, it's used as the \"safe\" position | |
4113 | to start parsing from. If NEAR is non-nil, then the limits of any | |
4114 | literal next to point is returned. \"Next to\" means there's only | |
4115 | spaces and tabs between point and the literal. The search for such a | |
4116 | literal is done first in forward direction. If NOT-IN-DELIMITER is | |
4117 | non-nil, the case when point is inside a starting delimiter won't be | |
a85fd6da | 4118 | recognized. This only has effect for comments which have starting |
d9e94c22 MS |
4119 | delimiters with more than one character. |
4120 | ||
0386b551 AM |
4121 | Note that this function might do hidden buffer changes. See the |
4122 | comment at the start of cc-engine.el for more info." | |
d9e94c22 MS |
4123 | |
4124 | (save-excursion | |
4125 | (let* ((pos (point)) | |
4126 | (lim (or lim (progn | |
4127 | (c-beginning-of-syntax) | |
4128 | (point)))) | |
4129 | (state (parse-partial-sexp lim pos))) | |
4130 | ||
0386b551 AM |
4131 | (cond ((elt state 3) ; String. |
4132 | (goto-char (elt state 8)) | |
d9e94c22 MS |
4133 | (cons (point) (or (c-safe (c-forward-sexp 1) (point)) |
4134 | (point-max)))) | |
4135 | ||
0386b551 AM |
4136 | ((elt state 4) ; Comment. |
4137 | (goto-char (elt state 8)) | |
d9e94c22 MS |
4138 | (cons (point) (progn (c-forward-single-comment) (point)))) |
4139 | ||
4140 | ((and (not not-in-delimiter) | |
4141 | (not (elt state 5)) | |
4142 | (eq (char-before) ?/) | |
4143 | (looking-at "[/*]")) | |
4144 | ;; We're standing in a comment starter. | |
4145 | (backward-char 1) | |
4146 | (cons (point) (progn (c-forward-single-comment) (point)))) | |
4147 | ||
4148 | (near | |
4149 | (goto-char pos) | |
4150 | ||
4151 | ;; Search forward for a literal. | |
4152 | (skip-chars-forward " \t") | |
4153 | ||
4154 | (cond | |
4155 | ((looking-at c-string-limit-regexp) ; String. | |
4156 | (cons (point) (or (c-safe (c-forward-sexp 1) (point)) | |
4157 | (point-max)))) | |
4158 | ||
4159 | ((looking-at c-comment-start-regexp) ; Line or block comment. | |
4160 | (cons (point) (progn (c-forward-single-comment) (point)))) | |
4161 | ||
4162 | (t | |
4163 | ;; Search backward. | |
4164 | (skip-chars-backward " \t") | |
4165 | ||
4166 | (let ((end (point)) beg) | |
4167 | (cond | |
4168 | ((save-excursion | |
4169 | (< (skip-syntax-backward c-string-syntax) 0)) ; String. | |
4170 | (setq beg (c-safe (c-backward-sexp 1) (point)))) | |
4171 | ||
4172 | ((and (c-safe (forward-char -2) t) | |
4173 | (looking-at "*/")) | |
4174 | ;; Block comment. Due to the nature of line | |
4175 | ;; comments, they will always be covered by the | |
4176 | ;; normal case above. | |
4177 | (goto-char end) | |
4178 | (c-backward-single-comment) | |
4179 | ;; If LIM is bogus, beg will be bogus. | |
4180 | (setq beg (point)))) | |
4181 | ||
4182 | (if beg (cons beg end)))))) | |
4183 | )))) | |
4184 | ||
0386b551 AM |
4185 | ;; In case external callers use this; it did have a docstring. |
4186 | (defalias 'c-literal-limits-fast 'c-literal-limits) | |
d9e94c22 | 4187 | |
0386b551 AM |
4188 | (defun c-collect-line-comments (range) |
4189 | "If the argument is a cons of two buffer positions (such as returned by | |
4190 | `c-literal-limits'), and that range contains a C++ style line comment, | |
4191 | then an extended range is returned that contains all adjacent line | |
4192 | comments (i.e. all comments that starts in the same column with no | |
4193 | empty lines or non-whitespace characters between them). Otherwise the | |
4194 | argument is returned. | |
d9e94c22 | 4195 | |
0386b551 AM |
4196 | Note that this function might do hidden buffer changes. See the |
4197 | comment at the start of cc-engine.el for more info." | |
d9e94c22 MS |
4198 | |
4199 | (save-excursion | |
0386b551 AM |
4200 | (condition-case nil |
4201 | (if (and (consp range) (progn | |
4202 | (goto-char (car range)) | |
4203 | (looking-at c-line-comment-starter))) | |
b414f371 | 4204 | (let ((col (current-column)) |
0386b551 AM |
4205 | (beg (point)) |
4206 | (bopl (c-point 'bopl)) | |
4207 | (end (cdr range))) | |
4208 | ;; Got to take care in the backward direction to handle | |
4209 | ;; comments which are preceded by code. | |
4210 | (while (and (c-backward-single-comment) | |
4211 | (>= (point) bopl) | |
4212 | (looking-at c-line-comment-starter) | |
4213 | (= col (current-column))) | |
4214 | (setq beg (point) | |
4215 | bopl (c-point 'bopl))) | |
4216 | (goto-char end) | |
4217 | (while (and (progn (skip-chars-forward " \t") | |
4218 | (looking-at c-line-comment-starter)) | |
4219 | (= col (current-column)) | |
4220 | (prog1 (zerop (forward-line 1)) | |
4221 | (setq end (point))))) | |
4222 | (cons beg end)) | |
4223 | range) | |
4224 | (error range)))) | |
d9e94c22 MS |
4225 | |
4226 | (defun c-literal-type (range) | |
4227 | "Convenience function that given the result of `c-literal-limits', | |
a85fd6da AM |
4228 | returns nil or the type of literal that the range surrounds, one |
4229 | of the symbols 'c, 'c++ or 'string. It's much faster than using | |
4230 | `c-in-literal' and is intended to be used when you need both the | |
4231 | type of a literal and its limits. | |
d9e94c22 | 4232 | |
0386b551 AM |
4233 | Note that this function might do hidden buffer changes. See the |
4234 | comment at the start of cc-engine.el for more info." | |
4235 | ||
d9e94c22 MS |
4236 | (if (consp range) |
4237 | (save-excursion | |
4238 | (goto-char (car range)) | |
4239 | (cond ((looking-at c-string-limit-regexp) 'string) | |
4240 | ((or (looking-at "//") ; c++ line comment | |
4241 | (and (looking-at "\\s<") ; comment starter | |
4242 | (looking-at "#"))) ; awk comment. | |
4243 | 'c++) | |
4244 | (t 'c))) ; Assuming the range is valid. | |
4245 | range)) | |
4246 | ||
4247 | \f | |
4248 | ;; `c-find-decl-spots' and accompanying stuff. | |
4249 | ||
4250 | ;; Variables used in `c-find-decl-spots' to cache the search done for | |
4251 | ;; the first declaration in the last call. When that function starts, | |
4252 | ;; it needs to back up over syntactic whitespace to look at the last | |
4253 | ;; token before the region being searched. That can sometimes cause | |
4254 | ;; moves back and forth over a quite large region of comments and | |
4255 | ;; macros, which would be repeated for each changed character when | |
4256 | ;; we're called during fontification, since font-lock refontifies the | |
4257 | ;; current line for each change. Thus it's worthwhile to cache the | |
4258 | ;; first match. | |
4259 | ;; | |
4260 | ;; `c-find-decl-syntactic-pos' is a syntactically relevant position in | |
4261 | ;; the syntactic whitespace less or equal to some start position. | |
4262 | ;; There's no cached value if it's nil. | |
4263 | ;; | |
4264 | ;; `c-find-decl-match-pos' is the match position if | |
4265 | ;; `c-find-decl-prefix-search' matched before the syntactic whitespace | |
4266 | ;; at `c-find-decl-syntactic-pos', or nil if there's no such match. | |
4267 | (defvar c-find-decl-syntactic-pos nil) | |
4268 | (make-variable-buffer-local 'c-find-decl-syntactic-pos) | |
4269 | (defvar c-find-decl-match-pos nil) | |
4270 | (make-variable-buffer-local 'c-find-decl-match-pos) | |
4271 | ||
4272 | (defsubst c-invalidate-find-decl-cache (change-min-pos) | |
4273 | (and c-find-decl-syntactic-pos | |
4274 | (< change-min-pos c-find-decl-syntactic-pos) | |
4275 | (setq c-find-decl-syntactic-pos nil))) | |
4276 | ||
4277 | ; (defface c-debug-decl-spot-face | |
4278 | ; '((t (:background "Turquoise"))) | |
4279 | ; "Debug face to mark the spots where `c-find-decl-spots' stopped.") | |
4280 | ; (defface c-debug-decl-sws-face | |
4281 | ; '((t (:background "Khaki"))) | |
4282 | ; "Debug face to mark the syntactic whitespace between the declaration | |
4283 | ; spots and the preceding token end.") | |
4284 | ||
4285 | (defmacro c-debug-put-decl-spot-faces (match-pos decl-pos) | |
4286 | (when (facep 'c-debug-decl-spot-face) | |
0386b551 | 4287 | `(c-save-buffer-state ((match-pos ,match-pos) (decl-pos ,decl-pos)) |
d9e94c22 MS |
4288 | (c-debug-add-face (max match-pos (point-min)) decl-pos |
4289 | 'c-debug-decl-sws-face) | |
4290 | (c-debug-add-face decl-pos (min (1+ decl-pos) (point-max)) | |
4291 | 'c-debug-decl-spot-face)))) | |
4292 | (defmacro c-debug-remove-decl-spot-faces (beg end) | |
4293 | (when (facep 'c-debug-decl-spot-face) | |
0386b551 | 4294 | `(c-save-buffer-state () |
d9e94c22 MS |
4295 | (c-debug-remove-face ,beg ,end 'c-debug-decl-spot-face) |
4296 | (c-debug-remove-face ,beg ,end 'c-debug-decl-sws-face)))) | |
4297 | ||
4298 | (defmacro c-find-decl-prefix-search () | |
4299 | ;; Macro used inside `c-find-decl-spots'. It ought to be a defun, | |
4300 | ;; but it contains lots of free variables that refer to things | |
4301 | ;; inside `c-find-decl-spots'. The point is left at `cfd-match-pos' | |
4302 | ;; if there is a match, otherwise at `cfd-limit'. | |
0386b551 AM |
4303 | ;; |
4304 | ;; This macro might do hidden buffer changes. | |
d9e94c22 MS |
4305 | |
4306 | '(progn | |
4307 | ;; Find the next property match position if we haven't got one already. | |
4308 | (unless cfd-prop-match | |
4309 | (save-excursion | |
4310 | (while (progn | |
4311 | (goto-char (next-single-property-change | |
4312 | (point) 'c-type nil cfd-limit)) | |
4313 | (and (< (point) cfd-limit) | |
4314 | (not (eq (c-get-char-property (1- (point)) 'c-type) | |
4315 | 'c-decl-end))))) | |
4316 | (setq cfd-prop-match (point)))) | |
4317 | ||
0386b551 AM |
4318 | ;; Find the next `c-decl-prefix-or-start-re' match if we haven't |
4319 | ;; got one already. | |
d9e94c22 | 4320 | (unless cfd-re-match |
0386b551 AM |
4321 | |
4322 | (if (> cfd-re-match-end (point)) | |
4323 | (goto-char cfd-re-match-end)) | |
4324 | ||
4325 | (while (if (setq cfd-re-match-end | |
4326 | (re-search-forward c-decl-prefix-or-start-re | |
4327 | cfd-limit 'move)) | |
4328 | ||
4329 | ;; Match. Check if it's inside a comment or string literal. | |
4330 | (c-got-face-at | |
4331 | (if (setq cfd-re-match (match-end 1)) | |
4332 | ;; Matched the end of a token preceding a decl spot. | |
4333 | (progn | |
4334 | (goto-char cfd-re-match) | |
4335 | (1- cfd-re-match)) | |
4336 | ;; Matched a token that start a decl spot. | |
4337 | (goto-char (match-beginning 0)) | |
4338 | (point)) | |
4339 | c-literal-faces) | |
4340 | ||
4341 | ;; No match. Finish up and exit the loop. | |
4342 | (setq cfd-re-match cfd-limit) | |
4343 | nil) | |
4344 | ||
4345 | ;; Skip out of comments and string literals. | |
d9e94c22 MS |
4346 | (while (progn |
4347 | (goto-char (next-single-property-change | |
0386b551 | 4348 | (point) 'face nil cfd-limit)) |
d9e94c22 | 4349 | (and (< (point) cfd-limit) |
0386b551 AM |
4350 | (c-got-face-at (point) c-literal-faces))))) |
4351 | ||
4352 | ;; If we matched at the decl start, we have to back up over the | |
4353 | ;; preceding syntactic ws to set `cfd-match-pos' and to catch | |
4354 | ;; any decl spots in the syntactic ws. | |
d9e94c22 | 4355 | (unless cfd-re-match |
0386b551 AM |
4356 | (c-backward-syntactic-ws) |
4357 | (setq cfd-re-match (point)))) | |
d9e94c22 MS |
4358 | |
4359 | ;; Choose whichever match is closer to the start. | |
4360 | (if (< cfd-re-match cfd-prop-match) | |
4361 | (setq cfd-match-pos cfd-re-match | |
4362 | cfd-re-match nil) | |
4363 | (setq cfd-match-pos cfd-prop-match | |
4364 | cfd-prop-match nil)) | |
4365 | ||
4366 | (goto-char cfd-match-pos) | |
4367 | ||
4368 | (when (< cfd-match-pos cfd-limit) | |
4369 | ;; Skip forward past comments only so we don't skip macros. | |
4370 | (c-forward-comments) | |
4371 | ;; Set the position to continue at. We can avoid going over | |
4372 | ;; the comments skipped above a second time, but it's possible | |
4373 | ;; that the comment skipping has taken us past `cfd-prop-match' | |
4374 | ;; since the property might be used inside comments. | |
4375 | (setq cfd-continue-pos (if cfd-prop-match | |
4376 | (min cfd-prop-match (point)) | |
4377 | (point)))))) | |
4378 | ||
4379 | (defun c-find-decl-spots (cfd-limit cfd-decl-re cfd-face-checklist cfd-fun) | |
0386b551 | 4380 | ;; Call CFD-FUN for each possible spot for a declaration, cast or |
a85fd6da AM |
4381 | ;; label from the point to CFD-LIMIT. |
4382 | ;; | |
4383 | ;; CFD-FUN is called with point at the start of the spot. It's | |
4384 | ;; passed two arguments: The first is the end position of the token | |
4385 | ;; preceding the spot, or 0 for the implicit match at bob. The | |
4386 | ;; second is a flag that is t when the match is inside a macro. If | |
4387 | ;; CFD-FUN adds `c-decl-end' properties somewhere below the current | |
4388 | ;; spot, it should return non-nil to ensure that the next search | |
4389 | ;; will find them. | |
0386b551 | 4390 | ;; |
a85fd6da | 4391 | ;; Such a spot is: |
0386b551 AM |
4392 | ;; o The first token after bob. |
4393 | ;; o The first token after the end of submatch 1 in | |
4394 | ;; `c-decl-prefix-or-start-re' when that submatch matches. | |
4395 | ;; o The start of each `c-decl-prefix-or-start-re' match when | |
4396 | ;; submatch 1 doesn't match. | |
4397 | ;; o The first token after the end of each occurence of the | |
4398 | ;; `c-type' text property with the value `c-decl-end', provided | |
4399 | ;; `c-type-decl-end-used' is set. | |
4400 | ;; | |
4401 | ;; Only a spot that match CFD-DECL-RE and whose face is in the | |
4402 | ;; CFD-FACE-CHECKLIST list causes CFD-FUN to be called. The face | |
4403 | ;; check is disabled if CFD-FACE-CHECKLIST is nil. | |
d9e94c22 MS |
4404 | ;; |
4405 | ;; If the match is inside a macro then the buffer is narrowed to the | |
4406 | ;; end of it, so that CFD-FUN can investigate the following tokens | |
4407 | ;; without matching something that begins inside a macro and ends | |
4408 | ;; outside it. It's to avoid this work that the CFD-DECL-RE and | |
4409 | ;; CFD-FACE-CHECKLIST checks exist. | |
4410 | ;; | |
0386b551 AM |
4411 | ;; The spots are visited approximately in order from top to bottom. |
4412 | ;; It's however the positions where `c-decl-prefix-or-start-re' | |
4413 | ;; matches and where `c-decl-end' properties are found that are in | |
4414 | ;; order. Since the spots often are at the following token, they | |
4415 | ;; might be visited out of order insofar as more spots are reported | |
4416 | ;; later on within the syntactic whitespace between the match | |
4417 | ;; positions and their spots. | |
4418 | ;; | |
4419 | ;; It's assumed that comments and strings are fontified in the | |
d9e94c22 MS |
4420 | ;; searched range. |
4421 | ;; | |
4422 | ;; This is mainly used in fontification, and so has an elaborate | |
4423 | ;; cache to handle repeated calls from the same start position; see | |
4424 | ;; the variables above. | |
4425 | ;; | |
4426 | ;; All variables in this function begin with `cfd-' to avoid name | |
4427 | ;; collision with the (dynamically bound) variables used in CFD-FUN. | |
0386b551 AM |
4428 | ;; |
4429 | ;; This function might do hidden buffer changes. | |
d9e94c22 | 4430 | |
0386b551 AM |
4431 | (let ((cfd-start-pos (point)) |
4432 | (cfd-buffer-end (point-max)) | |
4433 | ;; The end of the token preceding the decl spot last found | |
4434 | ;; with `c-decl-prefix-or-start-re'. `cfd-limit' if there's | |
4435 | ;; no match. | |
d9e94c22 | 4436 | cfd-re-match |
0386b551 AM |
4437 | ;; The end position of the last `c-decl-prefix-or-start-re' |
4438 | ;; match. If this is greater than `cfd-continue-pos', the | |
4439 | ;; next regexp search is started here instead. | |
4440 | (cfd-re-match-end (point-min)) | |
4441 | ;; The end of the last `c-decl-end' found by | |
4442 | ;; `c-find-decl-prefix-search'. `cfd-limit' if there's no | |
4443 | ;; match. If searching for the property isn't needed then we | |
4444 | ;; disable it by setting it to `cfd-limit' directly. | |
d9e94c22 | 4445 | (cfd-prop-match (unless c-type-decl-end-used cfd-limit)) |
0386b551 AM |
4446 | ;; The end of the token preceding the decl spot last found by |
4447 | ;; `c-find-decl-prefix-search'. 0 for the implicit match at | |
4448 | ;; bob. `cfd-limit' if there's no match. In other words, | |
4449 | ;; this is the minimum of `cfd-re-match' and `cfd-prop-match'. | |
d9e94c22 MS |
4450 | (cfd-match-pos cfd-limit) |
4451 | ;; The position to continue searching at. | |
4452 | cfd-continue-pos | |
4453 | ;; The position of the last "real" token we've stopped at. | |
4454 | ;; This can be greater than `cfd-continue-pos' when we get | |
4455 | ;; hits inside macros or at `c-decl-end' positions inside | |
4456 | ;; comments. | |
4457 | (cfd-token-pos 0) | |
4458 | ;; The end position of the last entered macro. | |
4459 | (cfd-macro-end 0)) | |
4460 | ||
4461 | ;; Initialize by finding a syntactically relevant start position | |
0386b551 AM |
4462 | ;; before the point, and do the first `c-decl-prefix-or-start-re' |
4463 | ;; search unless we're at bob. | |
d9e94c22 | 4464 | |
0386b551 | 4465 | (let (start-in-literal start-in-macro syntactic-pos) |
d9e94c22 MS |
4466 | ;; Must back up a bit since we look for the end of the previous |
4467 | ;; statement or declaration, which is earlier than the first | |
4468 | ;; returned match. | |
4469 | ||
0386b551 AM |
4470 | (cond |
4471 | ;; First we need to move to a syntactically relevant position. | |
4472 | ;; Begin by backing out of comment or string literals. | |
4473 | ((and | |
4474 | (when (c-got-face-at (point) c-literal-faces) | |
4475 | ;; Try to use the faces to back up to the start of the | |
4476 | ;; literal. FIXME: What if the point is on a declaration | |
4477 | ;; inside a comment? | |
4478 | (while (and (not (bobp)) | |
4479 | (c-got-face-at (1- (point)) c-literal-faces)) | |
4480 | (goto-char (previous-single-property-change | |
4481 | (point) 'face nil (point-min)))) | |
4482 | ||
4483 | ;; XEmacs doesn't fontify the quotes surrounding string | |
4484 | ;; literals. | |
4485 | (and (featurep 'xemacs) | |
4486 | (eq (get-text-property (point) 'face) | |
4487 | 'font-lock-string-face) | |
4488 | (not (bobp)) | |
4489 | (progn (backward-char) | |
4490 | (not (looking-at c-string-limit-regexp))) | |
4491 | (forward-char)) | |
4492 | ||
4493 | ;; Don't trust the literal to contain only literal faces | |
4494 | ;; (the font lock package might not have fontified the | |
4495 | ;; start of it at all, for instance) so check that we have | |
4496 | ;; arrived at something that looks like a start or else | |
4497 | ;; resort to `c-literal-limits'. | |
4498 | (unless (looking-at c-literal-start-regexp) | |
4499 | (let ((range (c-literal-limits))) | |
4500 | (if range (goto-char (car range))))) | |
4501 | ||
4502 | (setq start-in-literal (point))) | |
4503 | ||
4504 | ;; The start is in a literal. If the limit is in the same | |
4505 | ;; one we don't have to find a syntactic position etc. We | |
4506 | ;; only check that if the limit is at or before bonl to save | |
4507 | ;; time; it covers the by far most common case when font-lock | |
4508 | ;; refontifies the current line only. | |
4509 | (<= cfd-limit (c-point 'bonl cfd-start-pos)) | |
4510 | (save-excursion | |
4511 | (goto-char cfd-start-pos) | |
4512 | (while (progn | |
4513 | (goto-char (next-single-property-change | |
4514 | (point) 'face nil cfd-limit)) | |
4515 | (and (< (point) cfd-limit) | |
4516 | (c-got-face-at (point) c-literal-faces)))) | |
4517 | (= (point) cfd-limit))) | |
4518 | ||
4519 | ;; Completely inside a literal. Set up variables to trig the | |
4520 | ;; (< cfd-continue-pos cfd-start-pos) case below and it'll | |
4521 | ;; find a suitable start position. | |
4522 | (setq cfd-continue-pos start-in-literal)) | |
4523 | ||
4524 | ;; Check if the region might be completely inside a macro, to | |
4525 | ;; optimize that like the completely-inside-literal above. | |
4526 | ((save-excursion | |
4527 | (and (= (forward-line 1) 0) | |
4528 | (bolp) ; forward-line has funny behavior at eob. | |
4529 | (>= (point) cfd-limit) | |
4530 | (progn (backward-char) | |
4531 | (eq (char-before) ?\\)))) | |
4532 | ;; (Maybe) completely inside a macro. Only need to trig the | |
4533 | ;; (< cfd-continue-pos cfd-start-pos) case below to make it | |
4534 | ;; set things up. | |
4535 | (setq cfd-continue-pos (1- cfd-start-pos) | |
4536 | start-in-macro t)) | |
d9e94c22 | 4537 | |
0386b551 AM |
4538 | (t |
4539 | ;; Back out of any macro so we don't miss any declaration | |
4540 | ;; that could follow after it. | |
4541 | (when (c-beginning-of-macro) | |
4542 | (setq start-in-macro t)) | |
4543 | ||
4544 | ;; Now we're at a proper syntactically relevant position so we | |
4545 | ;; can use the cache. But first clear it if it applied | |
4546 | ;; further down. | |
4547 | (c-invalidate-find-decl-cache cfd-start-pos) | |
4548 | ||
4549 | (setq syntactic-pos (point)) | |
4550 | (unless (eq syntactic-pos c-find-decl-syntactic-pos) | |
4551 | ;; Don't have to do this if the cache is relevant here, | |
4552 | ;; typically if the same line is refontified again. If | |
4553 | ;; we're just some syntactic whitespace further down we can | |
4554 | ;; still use the cache to limit the skipping. | |
4555 | (c-backward-syntactic-ws c-find-decl-syntactic-pos)) | |
4556 | ||
4557 | ;; If we hit `c-find-decl-syntactic-pos' and | |
4558 | ;; `c-find-decl-match-pos' is set then we install the cached | |
4559 | ;; values. If we hit `c-find-decl-syntactic-pos' and | |
4560 | ;; `c-find-decl-match-pos' is nil then we know there's no decl | |
4561 | ;; prefix in the whitespace before `c-find-decl-syntactic-pos' | |
4562 | ;; and so we can continue the search from this point. If we | |
4563 | ;; didn't hit `c-find-decl-syntactic-pos' then we're now in | |
4564 | ;; the right spot to begin searching anyway. | |
4565 | (if (and (eq (point) c-find-decl-syntactic-pos) | |
4566 | c-find-decl-match-pos) | |
d9e94c22 MS |
4567 | (setq cfd-match-pos c-find-decl-match-pos |
4568 | cfd-continue-pos syntactic-pos) | |
0386b551 AM |
4569 | |
4570 | (setq c-find-decl-syntactic-pos syntactic-pos) | |
4571 | ||
4572 | (when (if (bobp) | |
4573 | ;; Always consider bob a match to get the first | |
4574 | ;; declaration in the file. Do this separately instead of | |
4575 | ;; letting `c-decl-prefix-or-start-re' match bob, so that | |
4576 | ;; regexp always can consume at least one character to | |
4577 | ;; ensure that we won't get stuck in an infinite loop. | |
4578 | (setq cfd-re-match 0) | |
4579 | (backward-char) | |
4580 | (c-beginning-of-current-token) | |
4581 | (< (point) cfd-limit)) | |
4582 | ;; Do an initial search now. In the bob case above it's | |
4583 | ;; only done to search for a `c-decl-end' spot. | |
4584 | (c-find-decl-prefix-search)) | |
4585 | ||
4586 | (setq c-find-decl-match-pos (and (< cfd-match-pos cfd-start-pos) | |
4587 | cfd-match-pos))))) | |
4588 | ||
4589 | ;; Advance `cfd-continue-pos' if it's before the start position. | |
4590 | ;; The closest continue position that might have effect at or | |
4591 | ;; after the start depends on what we started in. This also | |
4592 | ;; finds a suitable start position in the special cases when the | |
4593 | ;; region is completely within a literal or macro. | |
4594 | (when (and cfd-continue-pos (< cfd-continue-pos cfd-start-pos)) | |
4595 | ||
4596 | (cond | |
4597 | (start-in-macro | |
4598 | ;; If we're in a macro then it's the closest preceding token | |
4599 | ;; in the macro. Check this before `start-in-literal', | |
4600 | ;; since if we're inside a literal in a macro, the preceding | |
4601 | ;; token is earlier than any `c-decl-end' spot inside the | |
4602 | ;; literal (comment). | |
4603 | (goto-char (or start-in-literal cfd-start-pos)) | |
4604 | ;; The only syntactic ws in macros are comments. | |
d9e94c22 | 4605 | (c-backward-comments) |
0386b551 AM |
4606 | (backward-char) |
4607 | (c-beginning-of-current-token)) | |
4608 | ||
4609 | (start-in-literal | |
4610 | ;; If we're in a comment it can only be the closest | |
4611 | ;; preceding `c-decl-end' position within that comment, if | |
4612 | ;; any. Go back to the beginning of such a property so that | |
4613 | ;; `c-find-decl-prefix-search' will find the end of it. | |
4614 | ;; (Can't stop at the end and install it directly on | |
4615 | ;; `cfd-prop-match' since that variable might be cleared | |
4616 | ;; after `cfd-fun' below.) | |
4617 | ;; | |
4618 | ;; Note that if the literal is a string then the property | |
4619 | ;; search will simply skip to the beginning of it right | |
4620 | ;; away. | |
4621 | (if (not c-type-decl-end-used) | |
4622 | (goto-char start-in-literal) | |
4623 | (goto-char cfd-start-pos) | |
4624 | (while (progn | |
4625 | (goto-char (previous-single-property-change | |
4626 | (point) 'c-type nil start-in-literal)) | |
4627 | (and (> (point) start-in-literal) | |
4628 | (not (eq (c-get-char-property (point) 'c-type) | |
4629 | 'c-decl-end)))))) | |
4630 | ||
4631 | (when (= (point) start-in-literal) | |
4632 | ;; Didn't find any property inside the comment, so we can | |
4633 | ;; skip it entirely. (This won't skip past a string, but | |
4634 | ;; that'll be handled quickly by the next | |
4635 | ;; `c-find-decl-prefix-search' anyway.) | |
4636 | (c-forward-single-comment) | |
4637 | (if (> (point) cfd-limit) | |
4638 | (goto-char cfd-limit)))) | |
d9e94c22 | 4639 | |
0386b551 AM |
4640 | (t |
4641 | ;; If we started in normal code, the only match that might | |
4642 | ;; apply before the start is what we already got in | |
4643 | ;; `cfd-match-pos' so we can continue at the start position. | |
4644 | ;; (Note that we don't get here if the first match is below | |
4645 | ;; it.) | |
4646 | (goto-char cfd-start-pos))) | |
4647 | ||
4648 | ;; Delete found matches if they are before our new continue | |
4649 | ;; position, so that `c-find-decl-prefix-search' won't back up | |
4650 | ;; to them later on. | |
4651 | (setq cfd-continue-pos (point)) | |
4652 | (when (and cfd-re-match (< cfd-re-match cfd-continue-pos)) | |
4653 | (setq cfd-re-match nil)) | |
4654 | (when (and cfd-prop-match (< cfd-prop-match cfd-continue-pos)) | |
4655 | (setq cfd-prop-match nil))) | |
4656 | ||
4657 | (if syntactic-pos | |
4658 | ;; This is the normal case and we got a proper syntactic | |
4659 | ;; position. If there's a match then it's always outside | |
4660 | ;; macros and comments, so advance to the next token and set | |
4661 | ;; `cfd-token-pos'. The loop below will later go back using | |
4662 | ;; `cfd-continue-pos' to fix declarations inside the | |
4663 | ;; syntactic ws. | |
4664 | (when (and cfd-match-pos (< cfd-match-pos syntactic-pos)) | |
4665 | (goto-char syntactic-pos) | |
4666 | (c-forward-syntactic-ws) | |
4667 | (and cfd-continue-pos | |
4668 | (< cfd-continue-pos (point)) | |
4669 | (setq cfd-token-pos (point)))) | |
4670 | ||
4671 | ;; Have one of the special cases when the region is completely | |
4672 | ;; within a literal or macro. `cfd-continue-pos' is set to a | |
4673 | ;; good start position for the search, so do it. | |
4674 | (c-find-decl-prefix-search))) | |
d9e94c22 | 4675 | |
51c9af45 | 4676 | ;; Now loop. Round what? (ACM, 2006/7/5). We already got the first match. |
d9e94c22 MS |
4677 | |
4678 | (while (progn | |
4679 | (while (and | |
4680 | (< cfd-match-pos cfd-limit) | |
4681 | ||
4682 | (or | |
4683 | ;; Kludge to filter out matches on the "<" that | |
4684 | ;; aren't open parens, for the sake of languages | |
4685 | ;; that got `c-recognize-<>-arglists' set. | |
4686 | (and (eq (char-before cfd-match-pos) ?<) | |
4687 | (not (c-get-char-property (1- cfd-match-pos) | |
4688 | 'syntax-table))) | |
4689 | ||
4690 | ;; If `cfd-continue-pos' is less or equal to | |
4691 | ;; `cfd-token-pos', we've got a hit inside a macro | |
4692 | ;; that's in the syntactic whitespace before the last | |
4693 | ;; "real" declaration we've checked. If they're equal | |
4694 | ;; we've arrived at the declaration a second time, so | |
4695 | ;; there's nothing to do. | |
4696 | (= cfd-continue-pos cfd-token-pos) | |
4697 | ||
4698 | (progn | |
4699 | ;; If `cfd-continue-pos' is less than `cfd-token-pos' | |
4700 | ;; we're still searching for declarations embedded in | |
4701 | ;; the syntactic whitespace. In that case we need | |
4702 | ;; only to skip comments and not macros, since they | |
4703 | ;; can't be nested, and that's already been done in | |
4704 | ;; `c-find-decl-prefix-search'. | |
4705 | (when (> cfd-continue-pos cfd-token-pos) | |
4706 | (c-forward-syntactic-ws) | |
4707 | (setq cfd-token-pos (point))) | |
4708 | ||
4709 | ;; Continue if the following token fails the | |
4710 | ;; CFD-DECL-RE and CFD-FACE-CHECKLIST checks. | |
4711 | (when (or (>= (point) cfd-limit) | |
4712 | (not (looking-at cfd-decl-re)) | |
4713 | (and cfd-face-checklist | |
4714 | (not (c-got-face-at | |
4715 | (point) cfd-face-checklist)))) | |
4716 | (goto-char cfd-continue-pos) | |
4717 | t))) | |
4718 | ||
4719 | (< (point) cfd-limit)) | |
4720 | (c-find-decl-prefix-search)) | |
4721 | ||
4722 | (< (point) cfd-limit)) | |
4723 | ||
0386b551 AM |
4724 | (when (and |
4725 | (>= (point) cfd-start-pos) | |
d9e94c22 | 4726 | |
0386b551 AM |
4727 | (progn |
4728 | ;; Narrow to the end of the macro if we got a hit inside | |
4729 | ;; one, to avoid recognizing things that start inside the | |
4730 | ;; macro and end outside it. | |
4731 | (when (> cfd-match-pos cfd-macro-end) | |
4732 | ;; Not in the same macro as in the previous round. | |
4733 | (save-excursion | |
4734 | (goto-char cfd-match-pos) | |
4735 | (setq cfd-macro-end | |
4736 | (if (save-excursion (and (c-beginning-of-macro) | |
4737 | (< (point) cfd-match-pos))) | |
4738 | (progn (c-end-of-macro) | |
4739 | (point)) | |
4740 | 0)))) | |
4741 | ||
4742 | (if (zerop cfd-macro-end) | |
4743 | t | |
4744 | (if (> cfd-macro-end (point)) | |
4745 | (progn (narrow-to-region (point-min) cfd-macro-end) | |
4746 | t) | |
4747 | ;; The matched token was the last thing in the macro, | |
4748 | ;; so the whole match is bogus. | |
4749 | (setq cfd-macro-end 0) | |
4750 | nil)))) | |
d9e94c22 MS |
4751 | |
4752 | (c-debug-put-decl-spot-faces cfd-match-pos (point)) | |
0386b551 AM |
4753 | (if (funcall cfd-fun cfd-match-pos (/= cfd-macro-end 0)) |
4754 | (setq cfd-prop-match nil)) | |
d9e94c22 MS |
4755 | |
4756 | (when (/= cfd-macro-end 0) | |
4757 | ;; Restore limits if we did macro narrowment above. | |
4758 | (narrow-to-region (point-min) cfd-buffer-end))) | |
4759 | ||
4760 | (goto-char cfd-continue-pos) | |
4761 | (if (= cfd-continue-pos cfd-limit) | |
4762 | (setq cfd-match-pos cfd-limit) | |
4763 | (c-find-decl-prefix-search))))) | |
4764 | ||
4765 | \f | |
4766 | ;; A cache for found types. | |
4767 | ||
4768 | ;; Buffer local variable that contains an obarray with the types we've | |
4769 | ;; found. If a declaration is recognized somewhere we record the | |
4770 | ;; fully qualified identifier in it to recognize it as a type | |
4771 | ;; elsewhere in the file too. This is not accurate since we do not | |
4772 | ;; bother with the scoping rules of the languages, but in practice the | |
4773 | ;; same name is seldom used as both a type and something else in a | |
4774 | ;; file, and we only use this as a last resort in ambiguous cases (see | |
0386b551 AM |
4775 | ;; `c-forward-decl-or-cast-1'). |
4776 | ;; | |
580fba94 AM |
4777 | ;; Not every type need be in this cache. However, things which have |
4778 | ;; ceased to be types must be removed from it. | |
4779 | ;; | |
0386b551 AM |
4780 | ;; Template types in C++ are added here too but with the template |
4781 | ;; arglist replaced with "<>" in references or "<" for the one in the | |
4782 | ;; primary type. E.g. the type "Foo<A,B>::Bar<C>" is stored as | |
4783 | ;; "Foo<>::Bar<". This avoids storing very long strings (since C++ | |
4784 | ;; template specs can be fairly sized programs in themselves) and | |
4785 | ;; improves the hit ratio (it's a type regardless of the template | |
4786 | ;; args; it's just not the same type, but we're only interested in | |
4787 | ;; recognizing types, not telling distinct types apart). Note that | |
4788 | ;; template types in references are added here too; from the example | |
4789 | ;; above there will also be an entry "Foo<". | |
d9e94c22 MS |
4790 | (defvar c-found-types nil) |
4791 | (make-variable-buffer-local 'c-found-types) | |
4792 | ||
4793 | (defsubst c-clear-found-types () | |
4794 | ;; Clears `c-found-types'. | |
d9e94c22 MS |
4795 | (setq c-found-types (make-vector 53 0))) |
4796 | ||
4797 | (defun c-add-type (from to) | |
4798 | ;; Add the given region as a type in `c-found-types'. If the region | |
4799 | ;; doesn't match an existing type but there is a type which is equal | |
4800 | ;; to the given one except that the last character is missing, then | |
4801 | ;; the shorter type is removed. That's done to avoid adding all | |
4802 | ;; prefixes of a type as it's being entered and font locked. This | |
4803 | ;; doesn't cover cases like when characters are removed from a type | |
4804 | ;; or added in the middle. We'd need the position of point when the | |
4805 | ;; font locking is invoked to solve this well. | |
0386b551 AM |
4806 | ;; |
4807 | ;; This function might do hidden buffer changes. | |
4808 | (let ((type (c-syntactic-content from to c-recognize-<>-arglists))) | |
4809 | (unless (intern-soft type c-found-types) | |
4810 | (unintern (substring type 0 -1) c-found-types) | |
4811 | (intern type c-found-types)))) | |
d9e94c22 | 4812 | |
580fba94 AM |
4813 | (defun c-unfind-type (name) |
4814 | ;; Remove the "NAME" from c-found-types, if present. | |
4815 | (unintern name c-found-types)) | |
4816 | ||
d9e94c22 MS |
4817 | (defsubst c-check-type (from to) |
4818 | ;; Return non-nil if the given region contains a type in | |
4819 | ;; `c-found-types'. | |
0386b551 AM |
4820 | ;; |
4821 | ;; This function might do hidden buffer changes. | |
4822 | (intern-soft (c-syntactic-content from to c-recognize-<>-arglists) | |
4823 | c-found-types)) | |
d9e94c22 MS |
4824 | |
4825 | (defun c-list-found-types () | |
4826 | ;; Return all the types in `c-found-types' as a sorted list of | |
4827 | ;; strings. | |
4828 | (let (type-list) | |
4829 | (mapatoms (lambda (type) | |
4830 | (setq type-list (cons (symbol-name type) | |
4831 | type-list))) | |
4832 | c-found-types) | |
4833 | (sort type-list 'string-lessp))) | |
a66cd3ee | 4834 | |
2f42c75f DN |
4835 | ;; Shut up the byte compiler. |
4836 | (defvar c-maybe-stale-found-type) | |
4837 | ||
580fba94 AM |
4838 | (defun c-trim-found-types (beg end old-len) |
4839 | ;; An after change function which, in conjunction with the info in | |
4840 | ;; c-maybe-stale-found-type (set in c-before-change), removes a type | |
4841 | ;; from `c-found-types', should this type have become stale. For | |
4842 | ;; example, this happens to "foo" when "foo \n bar();" becomes | |
4843 | ;; "foo(); \n bar();". Such stale types, if not removed, foul up | |
4844 | ;; the fontification. | |
b414f371 | 4845 | ;; |
580fba94 AM |
4846 | ;; Have we, perhaps, added non-ws characters to the front/back of a found |
4847 | ;; type? | |
4848 | (when (> end beg) | |
4849 | (save-excursion | |
4850 | (when (< end (point-max)) | |
4851 | (goto-char end) | |
4852 | (if (and (c-beginning-of-current-token) ; only moves when we started in the middle | |
4853 | (progn (goto-char end) | |
4854 | (c-end-of-current-token))) | |
4855 | (c-unfind-type (buffer-substring-no-properties | |
4856 | end (point))))) | |
4857 | (when (> beg (point-min)) | |
4858 | (goto-char beg) | |
4859 | (if (and (c-end-of-current-token) ; only moves when we started in the middle | |
4860 | (progn (goto-char beg) | |
4861 | (c-beginning-of-current-token))) | |
4862 | (c-unfind-type (buffer-substring-no-properties | |
4863 | (point) beg)))))) | |
b414f371 | 4864 | |
580fba94 AM |
4865 | (if c-maybe-stale-found-type ; e.g. (c-decl-id-start "foo" 97 107 " (* ooka) " "o") |
4866 | (cond | |
4867 | ;; Changing the amount of (already existing) whitespace - don't do anything. | |
4868 | ((and (c-partial-ws-p beg end) | |
4869 | (or (= beg end) ; removal of WS | |
580fba94 AM |
4870 | (string-match "^[ \t\n\r\f\v]*$" (nth 5 c-maybe-stale-found-type))))) |
4871 | ||
4872 | ;; The syntactic relationship which defined a "found type" has been | |
4873 | ;; destroyed. | |
4874 | ((eq (car c-maybe-stale-found-type) 'c-decl-id-start) | |
4875 | (c-unfind-type (cadr c-maybe-stale-found-type))) | |
4876 | ;; ((eq (car c-maybe-stale-found-type) 'c-decl-type-start) FIXME!!! | |
4877 | ))) | |
4878 | ||
d9e94c22 MS |
4879 | \f |
4880 | ;; Handling of small scale constructs like types and names. | |
4881 | ||
0386b551 AM |
4882 | (defun c-after-change-check-<>-operators (beg end) |
4883 | ;; This is called from `after-change-functions' when | |
4884 | ;; c-recognize-<>-arglists' is set. It ensures that no "<" or ">" | |
4885 | ;; chars with paren syntax become part of another operator like "<<" | |
4886 | ;; or ">=". | |
4887 | ;; | |
4888 | ;; This function might do hidden buffer changes. | |
4889 | ||
cb694ab7 AM |
4890 | (save-excursion |
4891 | (goto-char beg) | |
4892 | (when (or (looking-at "[<>]") | |
4893 | (< (skip-chars-backward "<>") 0)) | |
4894 | ||
0386b551 | 4895 | (goto-char beg) |
cb694ab7 AM |
4896 | (c-beginning-of-current-token) |
4897 | (when (and (< (point) beg) | |
4898 | (looking-at c-<>-multichar-token-regexp) | |
4899 | (< beg (setq beg (match-end 0)))) | |
4900 | (while (progn (skip-chars-forward "^<>" beg) | |
4901 | (< (point) beg)) | |
4902 | (c-clear-char-property (point) 'syntax-table) | |
4903 | (forward-char)))) | |
4904 | ||
4905 | (when (< beg end) | |
4906 | (goto-char end) | |
0386b551 AM |
4907 | (when (or (looking-at "[<>]") |
4908 | (< (skip-chars-backward "<>") 0)) | |
4909 | ||
cb694ab7 | 4910 | (goto-char end) |
0386b551 | 4911 | (c-beginning-of-current-token) |
cb694ab7 | 4912 | (when (and (< (point) end) |
0386b551 | 4913 | (looking-at c-<>-multichar-token-regexp) |
cb694ab7 AM |
4914 | (< end (setq end (match-end 0)))) |
4915 | (while (progn (skip-chars-forward "^<>" end) | |
4916 | (< (point) end)) | |
0386b551 | 4917 | (c-clear-char-property (point) 'syntax-table) |
cb694ab7 | 4918 | (forward-char))))))) |
d9e94c22 MS |
4919 | |
4920 | ;; Dynamically bound variable that instructs `c-forward-type' to also | |
4921 | ;; treat possible types (i.e. those that it normally returns 'maybe or | |
4922 | ;; 'found for) as actual types (and always return 'found for them). | |
4923 | ;; This means that it records them in `c-record-type-identifiers' if | |
4924 | ;; that is set, and that it adds them to `c-found-types'. | |
4925 | (defvar c-promote-possible-types nil) | |
4926 | ||
0386b551 AM |
4927 | ;; Dynamically bound variable that instructs `c-forward-<>-arglist' to |
4928 | ;; mark up successfully parsed arglists with paren syntax properties on | |
4929 | ;; the surrounding angle brackets and with `c-<>-arg-sep' in the | |
4930 | ;; `c-type' property of each argument separating comma. | |
4931 | ;; | |
4932 | ;; Setting this variable also makes `c-forward-<>-arglist' recurse into | |
4933 | ;; all arglists for side effects (i.e. recording types), otherwise it | |
4934 | ;; exploits any existing paren syntax properties to quickly jump to the | |
4935 | ;; end of already parsed arglists. | |
4936 | ;; | |
4937 | ;; Marking up the arglists is not the default since doing that correctly | |
4938 | ;; depends on a proper value for `c-restricted-<>-arglists'. | |
4939 | (defvar c-parse-and-markup-<>-arglists nil) | |
4940 | ||
d9e94c22 | 4941 | ;; Dynamically bound variable that instructs `c-forward-<>-arglist' to |
037558bf MS |
4942 | ;; not accept arglists that contain binary operators. |
4943 | ;; | |
4944 | ;; This is primarily used to handle C++ template arglists. C++ | |
4945 | ;; disambiguates them by checking whether the preceding name is a | |
4946 | ;; template or not. We can't do that, so we assume it is a template | |
4947 | ;; if it can be parsed as one. That usually works well since | |
4948 | ;; comparison expressions on the forms "a < b > c" or "a < b, c > d" | |
4949 | ;; in almost all cases would be pointless. | |
4950 | ;; | |
4951 | ;; However, in function arglists, e.g. in "foo (a < b, c > d)", we | |
4952 | ;; should let the comma separate the function arguments instead. And | |
4953 | ;; in a context where the value of the expression is taken, e.g. in | |
4954 | ;; "if (a < b || c > d)", it's probably not a template. | |
4955 | (defvar c-restricted-<>-arglists nil) | |
d9e94c22 | 4956 | |
0386b551 AM |
4957 | ;; Dynamically bound variables that instructs |
4958 | ;; `c-forward-keyword-clause', `c-forward-<>-arglist', | |
4959 | ;; `c-forward-name', `c-forward-type', `c-forward-decl-or-cast-1', and | |
4960 | ;; `c-forward-label' to record the ranges of all the type and | |
4961 | ;; reference identifiers they encounter. They will build lists on | |
4962 | ;; these variables where each element is a cons of the buffer | |
4963 | ;; positions surrounding each identifier. This recording is only | |
4964 | ;; activated when `c-record-type-identifiers' is non-nil. | |
d9e94c22 MS |
4965 | ;; |
4966 | ;; All known types that can't be identifiers are recorded, and also | |
4967 | ;; other possible types if `c-promote-possible-types' is set. | |
4968 | ;; Recording is however disabled inside angle bracket arglists that | |
4969 | ;; are encountered inside names and other angle bracket arglists. | |
0386b551 | 4970 | ;; Such occurrences are taken care of by `c-font-lock-<>-arglists' |
d9e94c22 MS |
4971 | ;; instead. |
4972 | ;; | |
4973 | ;; Only the names in C++ template style references (e.g. "tmpl" in | |
4974 | ;; "tmpl<a,b>::foo") are recorded as references, other references | |
4975 | ;; aren't handled here. | |
0386b551 AM |
4976 | ;; |
4977 | ;; `c-forward-label' records the label identifier(s) on | |
4978 | ;; `c-record-ref-identifiers'. | |
d9e94c22 MS |
4979 | (defvar c-record-type-identifiers nil) |
4980 | (defvar c-record-ref-identifiers nil) | |
4981 | ||
0386b551 AM |
4982 | ;; This variable will receive a cons cell of the range of the last |
4983 | ;; single identifier symbol stepped over by `c-forward-name' if it's | |
4984 | ;; successful. This is the range that should be put on one of the | |
4985 | ;; record lists above by the caller. It's assigned nil if there's no | |
4986 | ;; such symbol in the name. | |
d9e94c22 MS |
4987 | (defvar c-last-identifier-range nil) |
4988 | ||
4989 | (defmacro c-record-type-id (range) | |
4990 | (if (eq (car-safe range) 'cons) | |
4991 | ;; Always true. | |
4992 | `(setq c-record-type-identifiers | |
4993 | (cons ,range c-record-type-identifiers)) | |
4994 | `(let ((range ,range)) | |
4995 | (if range | |
4996 | (setq c-record-type-identifiers | |
4997 | (cons range c-record-type-identifiers)))))) | |
4998 | ||
4999 | (defmacro c-record-ref-id (range) | |
5000 | (if (eq (car-safe range) 'cons) | |
5001 | ;; Always true. | |
5002 | `(setq c-record-ref-identifiers | |
5003 | (cons ,range c-record-ref-identifiers)) | |
5004 | `(let ((range ,range)) | |
5005 | (if range | |
5006 | (setq c-record-ref-identifiers | |
5007 | (cons range c-record-ref-identifiers)))))) | |
5008 | ||
5009 | ;; Dynamically bound variable that instructs `c-forward-type' to | |
5010 | ;; record the ranges of types that only are found. Behaves otherwise | |
5011 | ;; like `c-record-type-identifiers'. | |
5012 | (defvar c-record-found-types nil) | |
5013 | ||
5014 | (defmacro c-forward-keyword-prefixed-id (type) | |
5015 | ;; Used internally in `c-forward-keyword-clause' to move forward | |
5016 | ;; over a type (if TYPE is 'type) or a name (otherwise) which | |
5017 | ;; possibly is prefixed by keywords and their associated clauses. | |
5018 | ;; Try with a type/name first to not trip up on those that begin | |
5019 | ;; with a keyword. Return t if a known or found type is moved | |
5020 | ;; over. The point is clobbered if nil is returned. If range | |
5021 | ;; recording is enabled, the identifier is recorded on as a type | |
5022 | ;; if TYPE is 'type or as a reference if TYPE is 'ref. | |
0386b551 AM |
5023 | ;; |
5024 | ;; This macro might do hidden buffer changes. | |
d9e94c22 MS |
5025 | `(let (res) |
5026 | (while (if (setq res ,(if (eq type 'type) | |
5027 | `(c-forward-type) | |
5028 | `(c-forward-name))) | |
5029 | nil | |
5030 | (and (looking-at c-keywords-regexp) | |
0386b551 | 5031 | (c-forward-keyword-clause 1)))) |
d9e94c22 MS |
5032 | (when (memq res '(t known found prefix)) |
5033 | ,(when (eq type 'ref) | |
5034 | `(when c-record-type-identifiers | |
5035 | (c-record-ref-id c-last-identifier-range))) | |
5036 | t))) | |
5037 | ||
0386b551 | 5038 | (defmacro c-forward-id-comma-list (type update-safe-pos) |
d9e94c22 MS |
5039 | ;; Used internally in `c-forward-keyword-clause' to move forward |
5040 | ;; over a comma separated list of types or names using | |
5041 | ;; `c-forward-keyword-prefixed-id'. | |
0386b551 AM |
5042 | ;; |
5043 | ;; This macro might do hidden buffer changes. | |
d9e94c22 | 5044 | `(while (and (progn |
0386b551 AM |
5045 | ,(when update-safe-pos |
5046 | `(setq safe-pos (point))) | |
d9e94c22 MS |
5047 | (eq (char-after) ?,)) |
5048 | (progn | |
5049 | (forward-char) | |
5050 | (c-forward-syntactic-ws) | |
5051 | (c-forward-keyword-prefixed-id ,type))))) | |
5052 | ||
0386b551 AM |
5053 | (defun c-forward-keyword-clause (match) |
5054 | ;; Submatch MATCH in the current match data is assumed to surround a | |
5055 | ;; token. If it's a keyword, move over it and any immediately | |
5056 | ;; following clauses associated with it, stopping at the start of | |
5057 | ;; the next token. t is returned in that case, otherwise the point | |
d9e94c22 MS |
5058 | ;; stays and nil is returned. The kind of clauses that are |
5059 | ;; recognized are those specified by `c-type-list-kwds', | |
5060 | ;; `c-ref-list-kwds', `c-colon-type-list-kwds', | |
5061 | ;; `c-paren-nontype-kwds', `c-paren-type-kwds', `c-<>-type-kwds', | |
5062 | ;; and `c-<>-arglist-kwds'. | |
0386b551 AM |
5063 | ;; |
5064 | ;; This function records identifier ranges on | |
5065 | ;; `c-record-type-identifiers' and `c-record-ref-identifiers' if | |
5066 | ;; `c-record-type-identifiers' is non-nil. | |
5067 | ;; | |
5068 | ;; Note that for `c-colon-type-list-kwds', which doesn't necessary | |
5069 | ;; apply directly after the keyword, the type list is moved over | |
5070 | ;; only when there is no unaccounted token before it (i.e. a token | |
5071 | ;; that isn't moved over due to some other keyword list). The | |
5072 | ;; identifier ranges in the list are still recorded if that should | |
5073 | ;; be done, though. | |
5074 | ;; | |
5075 | ;; This function might do hidden buffer changes. | |
5076 | ||
5077 | (let ((kwd-sym (c-keyword-sym (match-string match))) safe-pos pos | |
5078 | ;; The call to `c-forward-<>-arglist' below is made after | |
5079 | ;; `c-<>-sexp-kwds' keywords, so we're certain they actually | |
5080 | ;; are angle bracket arglists and `c-restricted-<>-arglists' | |
5081 | ;; should therefore be nil. | |
5082 | (c-parse-and-markup-<>-arglists t) | |
5083 | c-restricted-<>-arglists) | |
d9e94c22 | 5084 | |
d9e94c22 | 5085 | (when kwd-sym |
0386b551 | 5086 | (goto-char (match-end match)) |
d9e94c22 MS |
5087 | (c-forward-syntactic-ws) |
5088 | (setq safe-pos (point)) | |
a66cd3ee | 5089 | |
d9e94c22 MS |
5090 | (cond |
5091 | ((and (c-keyword-member kwd-sym 'c-type-list-kwds) | |
5092 | (c-forward-keyword-prefixed-id type)) | |
5093 | ;; There's a type directly after a keyword in `c-type-list-kwds'. | |
0386b551 | 5094 | (c-forward-id-comma-list type t)) |
d9e94c22 MS |
5095 | |
5096 | ((and (c-keyword-member kwd-sym 'c-ref-list-kwds) | |
5097 | (c-forward-keyword-prefixed-id ref)) | |
5098 | ;; There's a name directly after a keyword in `c-ref-list-kwds'. | |
0386b551 | 5099 | (c-forward-id-comma-list ref t)) |
d9e94c22 MS |
5100 | |
5101 | ((and (c-keyword-member kwd-sym 'c-paren-any-kwds) | |
5102 | (eq (char-after) ?\()) | |
5103 | ;; There's an open paren after a keyword in `c-paren-any-kwds'. | |
5104 | ||
5105 | (forward-char) | |
5106 | (when (and (setq pos (c-up-list-forward)) | |
5107 | (eq (char-before pos) ?\))) | |
5108 | (when (and c-record-type-identifiers | |
5109 | (c-keyword-member kwd-sym 'c-paren-type-kwds)) | |
5110 | ;; Use `c-forward-type' on every identifier we can find | |
5111 | ;; inside the paren, to record the types. | |
5112 | (while (c-syntactic-re-search-forward c-symbol-start pos t) | |
5113 | (goto-char (match-beginning 0)) | |
5114 | (unless (c-forward-type) | |
5115 | (looking-at c-symbol-key) ; Always matches. | |
5116 | (goto-char (match-end 0))))) | |
5117 | ||
5118 | (goto-char pos) | |
5119 | (c-forward-syntactic-ws) | |
5120 | (setq safe-pos (point)))) | |
5121 | ||
5122 | ((and (c-keyword-member kwd-sym 'c-<>-sexp-kwds) | |
5123 | (eq (char-after) ?<) | |
0386b551 | 5124 | (c-forward-<>-arglist (c-keyword-member kwd-sym 'c-<>-type-kwds))) |
d9e94c22 MS |
5125 | (c-forward-syntactic-ws) |
5126 | (setq safe-pos (point))) | |
5127 | ||
5128 | ((and (c-keyword-member kwd-sym 'c-nonsymbol-sexp-kwds) | |
449a2b0d MS |
5129 | (not (looking-at c-symbol-start)) |
5130 | (c-safe (c-forward-sexp) t)) | |
d9e94c22 MS |
5131 | (c-forward-syntactic-ws) |
5132 | (setq safe-pos (point)))) | |
5133 | ||
0386b551 AM |
5134 | (when (c-keyword-member kwd-sym 'c-colon-type-list-kwds) |
5135 | (if (eq (char-after) ?:) | |
5136 | ;; If we are at the colon already, we move over the type | |
5137 | ;; list after it. | |
5138 | (progn | |
5139 | (forward-char) | |
5140 | (c-forward-syntactic-ws) | |
5141 | (when (c-forward-keyword-prefixed-id type) | |
5142 | (c-forward-id-comma-list type t))) | |
5143 | ;; Not at the colon, so stop here. But the identifier | |
5144 | ;; ranges in the type list later on should still be | |
5145 | ;; recorded. | |
5146 | (and c-record-type-identifiers | |
5147 | (progn | |
5148 | ;; If a keyword matched both one of the types above and | |
5149 | ;; this one, we match `c-colon-type-list-re' after the | |
5150 | ;; clause matched above. | |
5151 | (goto-char safe-pos) | |
5152 | (looking-at c-colon-type-list-re)) | |
5153 | (progn | |
5154 | (goto-char (match-end 0)) | |
5155 | (c-forward-syntactic-ws) | |
5156 | (c-forward-keyword-prefixed-id type)) | |
5157 | ;; There's a type after the `c-colon-type-list-re' match | |
5158 | ;; after a keyword in `c-colon-type-list-kwds'. | |
5159 | (c-forward-id-comma-list type nil)))) | |
d9e94c22 MS |
5160 | |
5161 | (goto-char safe-pos) | |
5162 | t))) | |
5163 | ||
0386b551 AM |
5164 | (defun c-forward-<>-arglist (all-types) |
5165 | ;; The point is assumed to be at a "<". Try to treat it as the open | |
b4dc7d98 | 5166 | ;; paren of an angle bracket arglist and move forward to the |
0386b551 AM |
5167 | ;; corresponding ">". If successful, the point is left after the |
5168 | ;; ">" and t is returned, otherwise the point isn't moved and nil is | |
d9e94c22 MS |
5169 | ;; returned. If ALL-TYPES is t then all encountered arguments in |
5170 | ;; the arglist that might be types are treated as found types. | |
5171 | ;; | |
0386b551 AM |
5172 | ;; The variable `c-parse-and-markup-<>-arglists' controls how this |
5173 | ;; function handles text properties on the angle brackets and argument | |
5174 | ;; separating commas. | |
d9e94c22 | 5175 | ;; |
0386b551 AM |
5176 | ;; `c-restricted-<>-arglists' controls how lenient the template |
5177 | ;; arglist recognition should be. | |
5178 | ;; | |
5179 | ;; This function records identifier ranges on | |
5180 | ;; `c-record-type-identifiers' and `c-record-ref-identifiers' if | |
5181 | ;; `c-record-type-identifiers' is non-nil. | |
5182 | ;; | |
5183 | ;; This function might do hidden buffer changes. | |
d9e94c22 MS |
5184 | |
5185 | (let ((start (point)) | |
5186 | ;; If `c-record-type-identifiers' is set then activate | |
5187 | ;; recording of any found types that constitute an argument in | |
5188 | ;; the arglist. | |
5189 | (c-record-found-types (if c-record-type-identifiers t))) | |
5190 | (if (catch 'angle-bracket-arglist-escape | |
5191 | (setq c-record-found-types | |
0386b551 | 5192 | (c-forward-<>-arglist-recur all-types))) |
d9e94c22 MS |
5193 | (progn |
5194 | (when (consp c-record-found-types) | |
5195 | (setq c-record-type-identifiers | |
5196 | ;; `nconc' doesn't mind that the tail of | |
5197 | ;; `c-record-found-types' is t. | |
5198 | (nconc c-record-found-types c-record-type-identifiers))) | |
5199 | t) | |
5200 | ||
5201 | (goto-char start) | |
a66cd3ee | 5202 | nil))) |
785eecbb | 5203 | |
0386b551 | 5204 | (defun c-forward-<>-arglist-recur (all-types) |
d9e94c22 | 5205 | ;; Recursive part of `c-forward-<>-arglist'. |
0386b551 AM |
5206 | ;; |
5207 | ;; This function might do hidden buffer changes. | |
d9e94c22 MS |
5208 | |
5209 | (let ((start (point)) res pos tmp | |
5210 | ;; Cover this so that any recorded found type ranges are | |
5211 | ;; automatically lost if it turns out to not be an angle | |
5212 | ;; bracket arglist. It's propagated through the return value | |
5213 | ;; on successful completion. | |
5214 | (c-record-found-types c-record-found-types) | |
5215 | ;; List that collects the positions after the argument | |
5216 | ;; separating ',' in the arglist. | |
5217 | arg-start-pos) | |
5218 | ||
0386b551 AM |
5219 | ;; If the '<' has paren open syntax then we've marked it as an angle |
5220 | ;; bracket arglist before, so skip to the end. | |
5221 | (if (and (not c-parse-and-markup-<>-arglists) | |
5222 | (c-get-char-property (point) 'syntax-table)) | |
5223 | ||
5224 | (progn | |
5225 | (forward-char) | |
5226 | (if (and (c-go-up-list-forward) | |
5227 | (eq (char-before) ?>)) | |
5228 | t | |
5229 | ||
5230 | ;; Got unmatched paren angle brackets. We don't clear the paren | |
5231 | ;; syntax properties and retry, on the basis that it's very | |
5232 | ;; unlikely that paren angle brackets become operators by code | |
5233 | ;; manipulation. It's far more likely that it doesn't match due | |
5234 | ;; to narrowing or some temporary change. | |
5235 | (goto-char start) | |
5236 | nil)) | |
d9e94c22 MS |
5237 | |
5238 | (forward-char) | |
5239 | (unless (looking-at c-<-op-cont-regexp) | |
5240 | (while (and | |
5241 | (progn | |
5242 | ||
5243 | (when c-record-type-identifiers | |
5244 | (if all-types | |
5245 | ||
5246 | ;; All encountered identifiers are types, so set the | |
5247 | ;; promote flag and parse the type. | |
5248 | (progn | |
5249 | (c-forward-syntactic-ws) | |
5250 | (when (looking-at c-identifier-start) | |
5251 | (let ((c-promote-possible-types t)) | |
5252 | (c-forward-type)))) | |
5253 | ||
5254 | ;; Check if this arglist argument is a sole type. If | |
5255 | ;; it's known then it's recorded in | |
5256 | ;; `c-record-type-identifiers'. If it only is found | |
5257 | ;; then it's recorded in `c-record-found-types' which we | |
5258 | ;; might roll back if it turns out that this isn't an | |
5259 | ;; angle bracket arglist afterall. | |
5260 | (when (memq (char-before) '(?, ?<)) | |
5261 | (let ((orig-record-found-types c-record-found-types)) | |
5262 | (c-forward-syntactic-ws) | |
5263 | (and (memq (c-forward-type) '(known found)) | |
5264 | (not (looking-at "[,>]")) | |
5265 | ;; A found type was recorded but it's not the | |
5266 | ;; only thing in the arglist argument, so reset | |
5267 | ;; `c-record-found-types'. | |
5268 | (setq c-record-found-types | |
5269 | orig-record-found-types)))))) | |
5270 | ||
5271 | (setq pos (point)) | |
5272 | (or (when (eq (char-after) ?>) | |
5273 | ;; Must check for '>' at the very start separately, | |
5274 | ;; since the regexp below has to avoid ">>" without | |
5275 | ;; using \\=. | |
5276 | (forward-char) | |
5277 | t) | |
5278 | ||
037558bf MS |
5279 | ;; Note: These regexps exploit the match order in \| so |
5280 | ;; that "<>" is matched by "<" rather than "[^>:-]>". | |
d9e94c22 | 5281 | (c-syntactic-re-search-forward |
037558bf MS |
5282 | (if c-restricted-<>-arglists |
5283 | ;; Stop on ',', '|', '&', '+' and '-' to catch | |
5284 | ;; common binary operators that could be between | |
5285 | ;; two comparison expressions "a<b" and "c>d". | |
5286 | "[<;{},|&+-]\\|\\([^>:-]>\\)" | |
5287 | ;; Otherwise we still stop on ',' to find the | |
5288 | ;; argument start positions. | |
5289 | "[<;{},]\\|\\([^>:-]>\\)") | |
5290 | nil 'move t t 1) | |
d9e94c22 MS |
5291 | |
5292 | ;; If the arglist starter has lost its open paren | |
5293 | ;; syntax but not the closer, we won't find the | |
5294 | ;; closer above since we only search in the | |
5295 | ;; balanced sexp. In that case we stop just short | |
5296 | ;; of it so check if the following char is the closer. | |
5297 | (when (eq (char-after) ?>) | |
d9e94c22 MS |
5298 | (forward-char) |
5299 | t))) | |
5300 | ||
5301 | (cond | |
5302 | ((eq (char-before) ?>) | |
5303 | ;; Either an operator starting with '>' or the end of | |
5304 | ;; the angle bracket arglist. | |
5305 | ||
0386b551 | 5306 | (if (looking-at c->-op-cont-regexp) |
d9e94c22 | 5307 | (progn |
0386b551 AM |
5308 | (goto-char (match-end 0)) |
5309 | t) ; Continue the loop. | |
d9e94c22 | 5310 | |
0386b551 AM |
5311 | ;; The angle bracket arglist is finished. |
5312 | (when c-parse-and-markup-<>-arglists | |
d9e94c22 | 5313 | (while arg-start-pos |
0386b551 AM |
5314 | (c-put-c-type-property (1- (car arg-start-pos)) |
5315 | 'c-<>-arg-sep) | |
d9e94c22 MS |
5316 | (setq arg-start-pos (cdr arg-start-pos))) |
5317 | (c-mark-<-as-paren start) | |
0386b551 AM |
5318 | (c-mark->-as-paren (1- (point)))) |
5319 | (setq res t) | |
5320 | nil)) ; Exit the loop. | |
d9e94c22 MS |
5321 | |
5322 | ((eq (char-before) ?<) | |
5323 | ;; Either an operator starting with '<' or a nested arglist. | |
5324 | ||
5325 | (setq pos (point)) | |
5326 | (let (id-start id-end subres keyword-match) | |
5327 | (if (if (looking-at c-<-op-cont-regexp) | |
5328 | (setq tmp (match-end 0)) | |
5329 | (setq tmp pos) | |
5330 | (backward-char) | |
5331 | (not | |
5332 | (and | |
5333 | ||
5334 | (save-excursion | |
0386b551 | 5335 | ;; There's always an identifier before an angle |
d9e94c22 MS |
5336 | ;; bracket arglist, or a keyword in |
5337 | ;; `c-<>-type-kwds' or `c-<>-arglist-kwds'. | |
5338 | (c-backward-syntactic-ws) | |
5339 | (setq id-end (point)) | |
5340 | (c-simple-skip-symbol-backward) | |
5341 | (when (or (setq keyword-match | |
5342 | (looking-at c-opt-<>-sexp-key)) | |
5343 | (not (looking-at c-keywords-regexp))) | |
5344 | (setq id-start (point)))) | |
5345 | ||
5346 | (setq subres | |
5347 | (let ((c-record-type-identifiers nil) | |
5348 | (c-record-found-types nil)) | |
5349 | (c-forward-<>-arglist-recur | |
5350 | (and keyword-match | |
5351 | (c-keyword-member | |
5352 | (c-keyword-sym (match-string 1)) | |
0386b551 | 5353 | 'c-<>-type-kwds))))) |
d9e94c22 MS |
5354 | ))) |
5355 | ||
5356 | ;; It was not an angle bracket arglist. | |
0386b551 | 5357 | (goto-char tmp) |
d9e94c22 MS |
5358 | |
5359 | ;; It was an angle bracket arglist. | |
5360 | (setq c-record-found-types subres) | |
5361 | ||
5362 | ;; Record the identifier before the template as a type | |
5363 | ;; or reference depending on whether the arglist is last | |
5364 | ;; in a qualified identifier. | |
5365 | (when (and c-record-type-identifiers | |
5366 | (not keyword-match)) | |
5367 | (if (and c-opt-identifier-concat-key | |
5368 | (progn | |
5369 | (c-forward-syntactic-ws) | |
5370 | (looking-at c-opt-identifier-concat-key))) | |
5371 | (c-record-ref-id (cons id-start id-end)) | |
5372 | (c-record-type-id (cons id-start id-end)))))) | |
5373 | t) | |
5374 | ||
5375 | ((and (eq (char-before) ?,) | |
037558bf | 5376 | (not c-restricted-<>-arglists)) |
d9e94c22 MS |
5377 | ;; Just another argument. Record the position. The |
5378 | ;; type check stuff that made us stop at it is at | |
5379 | ;; the top of the loop. | |
5380 | (setq arg-start-pos (cons (point) arg-start-pos))) | |
5381 | ||
5382 | (t | |
5383 | ;; Got a character that can't be in an angle bracket | |
5384 | ;; arglist argument. Abort using `throw', since | |
5385 | ;; it's useless to try to find a surrounding arglist | |
5386 | ;; if we're nested. | |
5387 | (throw 'angle-bracket-arglist-escape nil)))))) | |
5388 | ||
5389 | (if res | |
5390 | (or c-record-found-types t))))) | |
5391 | ||
0386b551 AM |
5392 | (defun c-backward-<>-arglist (all-types &optional limit) |
5393 | ;; The point is assumed to be directly after a ">". Try to treat it | |
5394 | ;; as the close paren of an angle bracket arglist and move back to | |
5395 | ;; the corresponding "<". If successful, the point is left at | |
5396 | ;; the "<" and t is returned, otherwise the point isn't moved and | |
5397 | ;; nil is returned. ALL-TYPES is passed on to | |
5398 | ;; `c-forward-<>-arglist'. | |
5399 | ;; | |
5400 | ;; If the optional LIMIT is given, it bounds the backward search. | |
5401 | ;; It's then assumed to be at a syntactically relevant position. | |
5402 | ;; | |
5403 | ;; This is a wrapper around `c-forward-<>-arglist'. See that | |
5404 | ;; function for more details. | |
5405 | ||
5406 | (let ((start (point))) | |
5407 | (backward-char) | |
5408 | (if (and (not c-parse-and-markup-<>-arglists) | |
5409 | (c-get-char-property (point) 'syntax-table)) | |
5410 | ||
5411 | (if (and (c-go-up-list-backward) | |
5412 | (eq (char-after) ?<)) | |
5413 | t | |
5414 | ;; See corresponding note in `c-forward-<>-arglist'. | |
5415 | (goto-char start) | |
5416 | nil) | |
5417 | ||
51c9af45 | 5418 | (while (progn |
0386b551 AM |
5419 | (c-syntactic-skip-backward "^<;{}" limit t) |
5420 | ||
51c9af45 AM |
5421 | (and |
5422 | (if (eq (char-before) ?<) | |
5423 | t | |
5424 | ;; Stopped at bob or a char that isn't allowed in an | |
5425 | ;; arglist, so we've failed. | |
5426 | (goto-char start) | |
5427 | nil) | |
0386b551 | 5428 | |
51c9af45 AM |
5429 | (if (> (point) |
5430 | (progn (c-beginning-of-current-token) | |
5431 | (point))) | |
5432 | ;; If we moved then the "<" was part of some | |
5433 | ;; multicharacter token. | |
5434 | t | |
0386b551 | 5435 | |
51c9af45 AM |
5436 | (backward-char) |
5437 | (let ((beg-pos (point))) | |
5438 | (if (c-forward-<>-arglist all-types) | |
5439 | (cond ((= (point) start) | |
5440 | ;; Matched the arglist. Break the while. | |
5441 | (goto-char beg-pos) | |
5442 | nil) | |
5443 | ((> (point) start) | |
5444 | ;; We started from a non-paren ">" inside an | |
5445 | ;; arglist. | |
5446 | (goto-char start) | |
5447 | nil) | |
5448 | (t | |
5449 | ;; Matched a shorter arglist. Can be a nested | |
5450 | ;; one so continue looking. | |
5451 | (goto-char beg-pos) | |
5452 | t)) | |
5453 | t)))))) | |
0386b551 AM |
5454 | |
5455 | (/= (point) start)))) | |
5456 | ||
d9e94c22 MS |
5457 | (defun c-forward-name () |
5458 | ;; Move forward over a complete name if at the beginning of one, | |
5459 | ;; stopping at the next following token. If the point is not at | |
5460 | ;; something that are recognized as name then it stays put. A name | |
5461 | ;; could be something as simple as "foo" in C or something as | |
5462 | ;; complex as "X<Y<class A<int>::B, BIT_MAX >> b>, ::operator<> :: | |
5463 | ;; Z<(a>b)> :: operator const X<&foo>::T Q::G<unsigned short | |
5464 | ;; int>::*volatile const" in C++ (this function is actually little | |
5465 | ;; more than a `looking-at' call in all modes except those that, | |
5466 | ;; like C++, have `c-recognize-<>-arglists' set). Return nil if no | |
5467 | ;; name is found, 'template if it's an identifier ending with an | |
5468 | ;; angle bracket arglist, 'operator of it's an operator identifier, | |
5469 | ;; or t if it's some other kind of name. | |
0386b551 AM |
5470 | ;; |
5471 | ;; This function records identifier ranges on | |
5472 | ;; `c-record-type-identifiers' and `c-record-ref-identifiers' if | |
5473 | ;; `c-record-type-identifiers' is non-nil. | |
5474 | ;; | |
5475 | ;; This function might do hidden buffer changes. | |
d9e94c22 | 5476 | |
0386b551 | 5477 | (let ((pos (point)) (start (point)) res id-start id-end |
d9e94c22 MS |
5478 | ;; Turn off `c-promote-possible-types' here since we might |
5479 | ;; call `c-forward-<>-arglist' and we don't want it to promote | |
5480 | ;; every suspect thing in the arglist to a type. We're | |
5481 | ;; typically called from `c-forward-type' in this case, and | |
5482 | ;; the caller only wants the top level type that it finds to | |
5483 | ;; be promoted. | |
5484 | c-promote-possible-types) | |
5485 | (while | |
5486 | (and | |
5487 | (looking-at c-identifier-key) | |
5488 | ||
5489 | (progn | |
5490 | ;; Check for keyword. We go to the last symbol in | |
5491 | ;; `c-identifier-key' first. | |
0386b551 AM |
5492 | (goto-char (setq id-end (match-end 0))) |
5493 | (c-simple-skip-symbol-backward) | |
5494 | (setq id-start (point)) | |
d9e94c22 MS |
5495 | |
5496 | (if (looking-at c-keywords-regexp) | |
5497 | (when (and (c-major-mode-is 'c++-mode) | |
5498 | (looking-at | |
5499 | (cc-eval-when-compile | |
5500 | (concat "\\(operator\\|\\(template\\)\\)" | |
5501 | "\\(" (c-lang-const c-nonsymbol-key c++) | |
5502 | "\\|$\\)"))) | |
5503 | (if (match-beginning 2) | |
5504 | ;; "template" is only valid inside an | |
5505 | ;; identifier if preceded by "::". | |
5506 | (save-excursion | |
5507 | (c-backward-syntactic-ws) | |
5508 | (and (c-safe (backward-char 2) t) | |
5509 | (looking-at "::"))) | |
5510 | t)) | |
5511 | ||
5512 | ;; Handle a C++ operator or template identifier. | |
5513 | (goto-char id-end) | |
5514 | (c-forward-syntactic-ws) | |
5515 | (cond ((eq (char-before id-end) ?e) | |
5516 | ;; Got "... ::template". | |
5517 | (let ((subres (c-forward-name))) | |
5518 | (when subres | |
5519 | (setq pos (point) | |
5520 | res subres)))) | |
5521 | ||
5522 | ((looking-at c-identifier-start) | |
5523 | ;; Got a cast operator. | |
5524 | (when (c-forward-type) | |
5525 | (setq pos (point) | |
5526 | res 'operator) | |
5527 | ;; Now we should match a sequence of either | |
5528 | ;; '*', '&' or a name followed by ":: *", | |
5529 | ;; where each can be followed by a sequence | |
5530 | ;; of `c-opt-type-modifier-key'. | |
5531 | (while (cond ((looking-at "[*&]") | |
5532 | (goto-char (match-end 0)) | |
5533 | t) | |
5534 | ((looking-at c-identifier-start) | |
5535 | (and (c-forward-name) | |
5536 | (looking-at "::") | |
5537 | (progn | |
5538 | (goto-char (match-end 0)) | |
5539 | (c-forward-syntactic-ws) | |
5540 | (eq (char-after) ?*)) | |
5541 | (progn | |
5542 | (forward-char) | |
5543 | t)))) | |
5544 | (while (progn | |
5545 | (c-forward-syntactic-ws) | |
5546 | (setq pos (point)) | |
5547 | (looking-at c-opt-type-modifier-key)) | |
5548 | (goto-char (match-end 1)))))) | |
5549 | ||
5550 | ((looking-at c-overloadable-operators-regexp) | |
5551 | ;; Got some other operator. | |
0386b551 AM |
5552 | (setq c-last-identifier-range |
5553 | (cons (point) (match-end 0))) | |
d9e94c22 MS |
5554 | (goto-char (match-end 0)) |
5555 | (c-forward-syntactic-ws) | |
5556 | (setq pos (point) | |
5557 | res 'operator))) | |
5558 | ||
5559 | nil) | |
5560 | ||
0386b551 AM |
5561 | ;; `id-start' is equal to `id-end' if we've jumped over |
5562 | ;; an identifier that doesn't end with a symbol token. | |
5563 | ;; That can occur e.g. for Java import directives on the | |
5564 | ;; form "foo.bar.*". | |
5565 | (when (and id-start (/= id-start id-end)) | |
d9e94c22 MS |
5566 | (setq c-last-identifier-range |
5567 | (cons id-start id-end))) | |
5568 | (goto-char id-end) | |
5569 | (c-forward-syntactic-ws) | |
5570 | (setq pos (point) | |
5571 | res t))) | |
5572 | ||
5573 | (progn | |
5574 | (goto-char pos) | |
5575 | (when (or c-opt-identifier-concat-key | |
5576 | c-recognize-<>-arglists) | |
5577 | ||
5578 | (cond | |
5579 | ((and c-opt-identifier-concat-key | |
5580 | (looking-at c-opt-identifier-concat-key)) | |
5581 | ;; Got a concatenated identifier. This handles the | |
5582 | ;; cases with tricky syntactic whitespace that aren't | |
5583 | ;; covered in `c-identifier-key'. | |
5584 | (goto-char (match-end 0)) | |
5585 | (c-forward-syntactic-ws) | |
5586 | t) | |
5587 | ||
5588 | ((and c-recognize-<>-arglists | |
5589 | (eq (char-after) ?<)) | |
5590 | ;; Maybe an angle bracket arglist. | |
0386b551 AM |
5591 | |
5592 | (when (let (c-record-type-identifiers | |
5593 | c-record-found-types) | |
5594 | (c-forward-<>-arglist nil)) | |
5595 | ||
5596 | (c-add-type start (1+ pos)) | |
d9e94c22 | 5597 | (c-forward-syntactic-ws) |
0386b551 AM |
5598 | (setq pos (point) |
5599 | c-last-identifier-range nil) | |
5600 | ||
d9e94c22 MS |
5601 | (if (and c-opt-identifier-concat-key |
5602 | (looking-at c-opt-identifier-concat-key)) | |
0386b551 | 5603 | |
d9e94c22 MS |
5604 | ;; Continue if there's an identifier concatenation |
5605 | ;; operator after the template argument. | |
5606 | (progn | |
0386b551 AM |
5607 | (when (and c-record-type-identifiers id-start) |
5608 | (c-record-ref-id (cons id-start id-end))) | |
d9e94c22 MS |
5609 | (forward-char 2) |
5610 | (c-forward-syntactic-ws) | |
5611 | t) | |
0386b551 AM |
5612 | |
5613 | (when (and c-record-type-identifiers id-start) | |
5614 | (c-record-type-id (cons id-start id-end))) | |
d9e94c22 MS |
5615 | (setq res 'template) |
5616 | nil))) | |
5617 | ))))) | |
5618 | ||
5619 | (goto-char pos) | |
5620 | res)) | |
5621 | ||
5622 | (defun c-forward-type () | |
5623 | ;; Move forward over a type spec if at the beginning of one, | |
5624 | ;; stopping at the next following token. Return t if it's a known | |
2a15eb73 MS |
5625 | ;; type that can't be a name or other expression, 'known if it's an |
5626 | ;; otherwise known type (according to `*-font-lock-extra-types'), | |
5627 | ;; 'prefix if it's a known prefix of a type, 'found if it's a type | |
5628 | ;; that matches one in `c-found-types', 'maybe if it's an identfier | |
5629 | ;; that might be a type, or nil if it can't be a type (the point | |
5630 | ;; isn't moved then). The point is assumed to be at the beginning | |
5631 | ;; of a token. | |
d9e94c22 MS |
5632 | ;; |
5633 | ;; Note that this function doesn't skip past the brace definition | |
5634 | ;; that might be considered part of the type, e.g. | |
5635 | ;; "enum {a, b, c} foo". | |
0386b551 AM |
5636 | ;; |
5637 | ;; This function records identifier ranges on | |
5638 | ;; `c-record-type-identifiers' and `c-record-ref-identifiers' if | |
5639 | ;; `c-record-type-identifiers' is non-nil. | |
5640 | ;; | |
5641 | ;; This function might do hidden buffer changes. | |
5642 | ||
5643 | (let ((start (point)) pos res name-res id-start id-end id-range) | |
d9e94c22 MS |
5644 | |
5645 | ;; Skip leading type modifiers. If any are found we know it's a | |
5646 | ;; prefix of a type. | |
5647 | (when c-opt-type-modifier-key | |
5648 | (while (looking-at c-opt-type-modifier-key) | |
5649 | (goto-char (match-end 1)) | |
5650 | (c-forward-syntactic-ws) | |
5651 | (setq res 'prefix))) | |
5652 | ||
5653 | (cond | |
5654 | ((looking-at c-type-prefix-key) | |
5655 | ;; Looking at a keyword that prefixes a type identifier, | |
5656 | ;; e.g. "class". | |
5657 | (goto-char (match-end 1)) | |
5658 | (c-forward-syntactic-ws) | |
5659 | (setq pos (point)) | |
0386b551 | 5660 | (if (memq (setq name-res (c-forward-name)) '(t template)) |
d9e94c22 | 5661 | (progn |
0386b551 | 5662 | (when (eq name-res t) |
d9e94c22 MS |
5663 | ;; In many languages the name can be used without the |
5664 | ;; prefix, so we add it to `c-found-types'. | |
5665 | (c-add-type pos (point)) | |
0386b551 AM |
5666 | (when (and c-record-type-identifiers |
5667 | c-last-identifier-range) | |
d9e94c22 MS |
5668 | (c-record-type-id c-last-identifier-range))) |
5669 | (setq res t)) | |
5670 | ;; Invalid syntax. | |
5671 | (goto-char start) | |
5672 | (setq res nil))) | |
5673 | ||
5674 | ((progn | |
5675 | (setq pos nil) | |
5676 | (if (looking-at c-identifier-start) | |
5677 | (save-excursion | |
5678 | (setq id-start (point) | |
0386b551 AM |
5679 | name-res (c-forward-name)) |
5680 | (when name-res | |
d9e94c22 MS |
5681 | (setq id-end (point) |
5682 | id-range c-last-identifier-range)))) | |
5683 | (and (cond ((looking-at c-primitive-type-key) | |
5684 | (setq res t)) | |
5685 | ((c-with-syntax-table c-identifier-syntax-table | |
5686 | (looking-at c-known-type-key)) | |
5687 | (setq res 'known))) | |
5688 | (or (not id-end) | |
5689 | (>= (save-excursion | |
5690 | (save-match-data | |
5691 | (goto-char (match-end 1)) | |
5692 | (c-forward-syntactic-ws) | |
5693 | (setq pos (point)))) | |
5694 | id-end) | |
5695 | (setq res nil)))) | |
5696 | ;; Looking at a primitive or known type identifier. We've | |
5697 | ;; checked for a name first so that we don't go here if the | |
5698 | ;; known type match only is a prefix of another name. | |
5699 | ||
5700 | (setq id-end (match-end 1)) | |
5701 | ||
5702 | (when (and c-record-type-identifiers | |
5703 | (or c-promote-possible-types (eq res t))) | |
5704 | (c-record-type-id (cons (match-beginning 1) (match-end 1)))) | |
5705 | ||
5706 | (if (and c-opt-type-component-key | |
5707 | (save-match-data | |
5708 | (looking-at c-opt-type-component-key))) | |
5709 | ;; There might be more keywords for the type. | |
5710 | (let (safe-pos) | |
0386b551 | 5711 | (c-forward-keyword-clause 1) |
d9e94c22 MS |
5712 | (while (progn |
5713 | (setq safe-pos (point)) | |
5714 | (looking-at c-opt-type-component-key)) | |
5715 | (when (and c-record-type-identifiers | |
5716 | (looking-at c-primitive-type-key)) | |
5717 | (c-record-type-id (cons (match-beginning 1) | |
5718 | (match-end 1)))) | |
0386b551 | 5719 | (c-forward-keyword-clause 1)) |
d9e94c22 MS |
5720 | (if (looking-at c-primitive-type-key) |
5721 | (progn | |
5722 | (when c-record-type-identifiers | |
5723 | (c-record-type-id (cons (match-beginning 1) | |
5724 | (match-end 1)))) | |
0386b551 | 5725 | (c-forward-keyword-clause 1) |
d9e94c22 MS |
5726 | (setq res t)) |
5727 | (goto-char safe-pos) | |
5728 | (setq res 'prefix))) | |
0386b551 | 5729 | (unless (save-match-data (c-forward-keyword-clause 1)) |
d9e94c22 MS |
5730 | (if pos |
5731 | (goto-char pos) | |
5732 | (goto-char (match-end 1)) | |
5733 | (c-forward-syntactic-ws))))) | |
5734 | ||
0386b551 AM |
5735 | (name-res |
5736 | (cond ((eq name-res t) | |
d9e94c22 MS |
5737 | ;; A normal identifier. |
5738 | (goto-char id-end) | |
5739 | (if (or res c-promote-possible-types) | |
5740 | (progn | |
5741 | (c-add-type id-start id-end) | |
0386b551 | 5742 | (when (and c-record-type-identifiers id-range) |
d9e94c22 MS |
5743 | (c-record-type-id id-range)) |
5744 | (unless res | |
5745 | (setq res 'found))) | |
5746 | (setq res (if (c-check-type id-start id-end) | |
5747 | ;; It's an identifier that has been used as | |
5748 | ;; a type somewhere else. | |
5749 | 'found | |
5750 | ;; It's an identifier that might be a type. | |
5751 | 'maybe)))) | |
0386b551 | 5752 | ((eq name-res 'template) |
d9e94c22 MS |
5753 | ;; A template is a type. |
5754 | (goto-char id-end) | |
5755 | (setq res t)) | |
5756 | (t | |
5757 | ;; Otherwise it's an operator identifier, which is not a type. | |
5758 | (goto-char start) | |
5759 | (setq res nil))))) | |
5760 | ||
5761 | (when res | |
5762 | ;; Skip trailing type modifiers. If any are found we know it's | |
5763 | ;; a type. | |
5764 | (when c-opt-type-modifier-key | |
5765 | (while (looking-at c-opt-type-modifier-key) | |
5766 | (goto-char (match-end 1)) | |
5767 | (c-forward-syntactic-ws) | |
5768 | (setq res t))) | |
5769 | ||
5770 | ;; Step over any type suffix operator. Do not let the existence | |
5771 | ;; of these alter the classification of the found type, since | |
5772 | ;; these operators typically are allowed in normal expressions | |
5773 | ;; too. | |
5774 | (when c-opt-type-suffix-key | |
5775 | (while (looking-at c-opt-type-suffix-key) | |
5776 | (goto-char (match-end 1)) | |
5777 | (c-forward-syntactic-ws))) | |
5778 | ||
5779 | (when c-opt-type-concat-key | |
0386b551 AM |
5780 | ;; Look for a trailing operator that concatenates the type |
5781 | ;; with a following one, and if so step past that one through | |
5782 | ;; a recursive call. Note that we don't record concatenated | |
5783 | ;; types in `c-found-types' - it's the component types that | |
5784 | ;; are recorded when appropriate. | |
d9e94c22 MS |
5785 | (setq pos (point)) |
5786 | (let* ((c-promote-possible-types (or (memq res '(t known)) | |
5787 | c-promote-possible-types)) | |
5788 | ;; If we can't promote then set `c-record-found-types' so that | |
5789 | ;; we can merge in the types from the second part afterwards if | |
5790 | ;; it turns out to be a known type there. | |
5791 | (c-record-found-types (and c-record-type-identifiers | |
0386b551 AM |
5792 | (not c-promote-possible-types))) |
5793 | subres) | |
d9e94c22 MS |
5794 | (if (and (looking-at c-opt-type-concat-key) |
5795 | ||
5796 | (progn | |
5797 | (goto-char (match-end 1)) | |
5798 | (c-forward-syntactic-ws) | |
0386b551 | 5799 | (setq subres (c-forward-type)))) |
d9e94c22 MS |
5800 | |
5801 | (progn | |
5802 | ;; If either operand certainly is a type then both are, but we | |
5803 | ;; don't let the existence of the operator itself promote two | |
5804 | ;; uncertain types to a certain one. | |
5805 | (cond ((eq res t)) | |
0386b551 AM |
5806 | ((eq subres t) |
5807 | (unless (eq name-res 'template) | |
5808 | (c-add-type id-start id-end)) | |
5809 | (when (and c-record-type-identifiers id-range) | |
d9e94c22 MS |
5810 | (c-record-type-id id-range)) |
5811 | (setq res t)) | |
2a15eb73 | 5812 | ((eq res 'known)) |
0386b551 | 5813 | ((eq subres 'known) |
2a15eb73 | 5814 | (setq res 'known)) |
d9e94c22 | 5815 | ((eq res 'found)) |
0386b551 | 5816 | ((eq subres 'found) |
d9e94c22 MS |
5817 | (setq res 'found)) |
5818 | (t | |
5819 | (setq res 'maybe))) | |
5820 | ||
5821 | (when (and (eq res t) | |
5822 | (consp c-record-found-types)) | |
5823 | ;; Merge in the ranges of any types found by the second | |
5824 | ;; `c-forward-type'. | |
5825 | (setq c-record-type-identifiers | |
5826 | ;; `nconc' doesn't mind that the tail of | |
5827 | ;; `c-record-found-types' is t. | |
5828 | (nconc c-record-found-types | |
5829 | c-record-type-identifiers)))) | |
5830 | ||
5831 | (goto-char pos)))) | |
5832 | ||
5833 | (when (and c-record-found-types (memq res '(known found)) id-range) | |
5834 | (setq c-record-found-types | |
5835 | (cons id-range c-record-found-types)))) | |
5836 | ||
5837 | ;;(message "c-forward-type %s -> %s: %s" start (point) res) | |
5838 | ||
5839 | res)) | |
5840 | ||
785eecbb | 5841 | \f |
d9e94c22 MS |
5842 | ;; Handling of large scale constructs like statements and declarations. |
5843 | ||
0386b551 AM |
5844 | ;; Macro used inside `c-forward-decl-or-cast-1'. It ought to be a |
5845 | ;; defsubst or perhaps even a defun, but it contains lots of free | |
5846 | ;; variables that refer to things inside `c-forward-decl-or-cast-1'. | |
5847 | (defmacro c-fdoc-shift-type-backward (&optional short) | |
5848 | ;; `c-forward-decl-or-cast-1' can consume an arbitrary length list | |
5849 | ;; of types when parsing a declaration, which means that it | |
5850 | ;; sometimes consumes the identifier in the declaration as a type. | |
5851 | ;; This is used to "backtrack" and make the last type be treated as | |
5852 | ;; an identifier instead. | |
5853 | `(progn | |
5854 | ,(unless short | |
5855 | ;; These identifiers are bound only in the inner let. | |
5856 | '(setq identifier-type at-type | |
5857 | identifier-start type-start | |
5858 | got-parens nil | |
5859 | got-identifier t | |
5860 | got-suffix t | |
5861 | got-suffix-after-parens id-start | |
5862 | paren-depth 0)) | |
5863 | ||
5864 | (if (setq at-type (if (eq backup-at-type 'prefix) | |
5865 | t | |
5866 | backup-at-type)) | |
5867 | (setq type-start backup-type-start | |
5868 | id-start backup-id-start) | |
5869 | (setq type-start start-pos | |
5870 | id-start start-pos)) | |
5871 | ||
5872 | ;; When these flags already are set we've found specifiers that | |
5873 | ;; unconditionally signal these attributes - backtracking doesn't | |
5874 | ;; change that. So keep them set in that case. | |
5875 | (or at-type-decl | |
5876 | (setq at-type-decl backup-at-type-decl)) | |
5877 | (or maybe-typeless | |
5878 | (setq maybe-typeless backup-maybe-typeless)) | |
5879 | ||
5880 | ,(unless short | |
5881 | ;; This identifier is bound only in the inner let. | |
5882 | '(setq start id-start)))) | |
5883 | ||
5884 | (defun c-forward-decl-or-cast-1 (preceding-token-end context last-cast-end) | |
5885 | ;; Move forward over a declaration or a cast if at the start of one. | |
5886 | ;; The point is assumed to be at the start of some token. Nil is | |
5887 | ;; returned if no declaration or cast is recognized, and the point | |
5888 | ;; is clobbered in that case. | |
5889 | ;; | |
5890 | ;; If a declaration is parsed: | |
5891 | ;; | |
5892 | ;; The point is left at the first token after the first complete | |
5893 | ;; declarator, if there is one. The return value is a cons where | |
022d0cf4 AM |
5894 | ;; the car is the position of the first token in the declarator. (See |
5895 | ;; below for the cdr.) | |
0386b551 AM |
5896 | ;; Some examples: |
5897 | ;; | |
5898 | ;; void foo (int a, char *b) stuff ... | |
5899 | ;; car ^ ^ point | |
5900 | ;; float (*a)[], b; | |
5901 | ;; car ^ ^ point | |
5902 | ;; unsigned int a = c_style_initializer, b; | |
5903 | ;; car ^ ^ point | |
5904 | ;; unsigned int a (cplusplus_style_initializer), b; | |
5905 | ;; car ^ ^ point (might change) | |
5906 | ;; class Foo : public Bar {} | |
5907 | ;; car ^ ^ point | |
5908 | ;; class PikeClass (int a, string b) stuff ... | |
5909 | ;; car ^ ^ point | |
5910 | ;; enum bool; | |
5911 | ;; car ^ ^ point | |
5912 | ;; enum bool flag; | |
5913 | ;; car ^ ^ point | |
5914 | ;; void cplusplus_function (int x) throw (Bad); | |
5915 | ;; car ^ ^ point | |
5916 | ;; Foo::Foo (int b) : Base (b) {} | |
5917 | ;; car ^ ^ point | |
5918 | ;; | |
022d0cf4 AM |
5919 | ;; The cdr of the return value is non-nil iff a `c-typedef-decl-kwds' |
5920 | ;; specifier (e.g. class, struct, enum, typedef) is found in the | |
5921 | ;; declaration, i.e. the declared identifier(s) are types. | |
0386b551 AM |
5922 | ;; |
5923 | ;; If a cast is parsed: | |
5924 | ;; | |
5925 | ;; The point is left at the first token after the closing paren of | |
5926 | ;; the cast. The return value is `cast'. Note that the start | |
5927 | ;; position must be at the first token inside the cast parenthesis | |
5928 | ;; to recognize it. | |
5929 | ;; | |
5930 | ;; PRECEDING-TOKEN-END is the first position after the preceding | |
5931 | ;; token, i.e. on the other side of the syntactic ws from the point. | |
5932 | ;; Use a value less than or equal to (point-min) if the point is at | |
5933 | ;; the first token in (the visible part of) the buffer. | |
5934 | ;; | |
5935 | ;; CONTEXT is a symbol that describes the context at the point: | |
022d0cf4 | 5936 | ;; 'decl In a comma-separated declaration context (typically |
0386b551 AM |
5937 | ;; inside a function declaration arglist). |
5938 | ;; '<> In an angle bracket arglist. | |
5939 | ;; 'arglist Some other type of arglist. | |
a85fd6da AM |
5940 | ;; nil Some other context or unknown context. Includes |
5941 | ;; within the parens of an if, for, ... construct. | |
0386b551 AM |
5942 | ;; |
5943 | ;; LAST-CAST-END is the first token after the closing paren of a | |
5944 | ;; preceding cast, or nil if none is known. If | |
5945 | ;; `c-forward-decl-or-cast-1' is used in succession, it should be | |
5946 | ;; the position after the closest preceding call where a cast was | |
5947 | ;; matched. In that case it's used to discover chains of casts like | |
5948 | ;; "(a) (b) c". | |
5949 | ;; | |
5950 | ;; This function records identifier ranges on | |
5951 | ;; `c-record-type-identifiers' and `c-record-ref-identifiers' if | |
5952 | ;; `c-record-type-identifiers' is non-nil. | |
5953 | ;; | |
5954 | ;; This function might do hidden buffer changes. | |
5955 | ||
5956 | (let (;; `start-pos' is used below to point to the start of the | |
5957 | ;; first type, i.e. after any leading specifiers. It might | |
5958 | ;; also point at the beginning of the preceding syntactic | |
5959 | ;; whitespace. | |
5960 | (start-pos (point)) | |
5961 | ;; Set to the result of `c-forward-type'. | |
5962 | at-type | |
5963 | ;; The position of the first token in what we currently | |
5964 | ;; believe is the type in the declaration or cast, after any | |
5965 | ;; specifiers and their associated clauses. | |
5966 | type-start | |
5967 | ;; The position of the first token in what we currently | |
5968 | ;; believe is the declarator for the first identifier. Set | |
5969 | ;; when the type is found, and moved forward over any | |
5970 | ;; `c-decl-hangon-kwds' and their associated clauses that | |
5971 | ;; occurs after the type. | |
5972 | id-start | |
5973 | ;; These store `at-type', `type-start' and `id-start' of the | |
5974 | ;; identifier before the one in those variables. The previous | |
5975 | ;; identifier might turn out to be the real type in a | |
5976 | ;; declaration if the last one has to be the declarator in it. | |
5977 | ;; If `backup-at-type' is nil then the other variables have | |
5978 | ;; undefined values. | |
5979 | backup-at-type backup-type-start backup-id-start | |
5980 | ;; Set if we've found a specifier that makes the defined | |
5981 | ;; identifier(s) types. | |
5982 | at-type-decl | |
5983 | ;; Set if we've found a specifier that can start a declaration | |
5984 | ;; where there's no type. | |
5985 | maybe-typeless | |
5986 | ;; If a specifier is found that also can be a type prefix, | |
5987 | ;; these flags are set instead of those above. If we need to | |
5988 | ;; back up an identifier, they are copied to the real flag | |
5989 | ;; variables. Thus they only take effect if we fail to | |
5990 | ;; interpret it as a type. | |
5991 | backup-at-type-decl backup-maybe-typeless | |
5992 | ;; Whether we've found a declaration or a cast. We might know | |
5993 | ;; this before we've found the type in it. It's 'ids if we've | |
5994 | ;; found two consecutive identifiers (usually a sure sign, but | |
5995 | ;; we should allow that in labels too), and t if we've found a | |
5996 | ;; specifier keyword (a 100% sure sign). | |
5997 | at-decl-or-cast | |
5998 | ;; Set when we need to back up to parse this as a declaration | |
5999 | ;; but not as a cast. | |
6000 | backup-if-not-cast | |
6001 | ;; For casts, the return position. | |
6002 | cast-end | |
6003 | ;; Save `c-record-type-identifiers' and | |
6004 | ;; `c-record-ref-identifiers' since ranges are recorded | |
6005 | ;; speculatively and should be thrown away if it turns out | |
6006 | ;; that it isn't a declaration or cast. | |
6007 | (save-rec-type-ids c-record-type-identifiers) | |
6008 | (save-rec-ref-ids c-record-ref-identifiers)) | |
6009 | ||
6010 | ;; Check for a type. Unknown symbols are treated as possible | |
6011 | ;; types, but they could also be specifiers disguised through | |
6012 | ;; macros like __INLINE__, so we recognize both types and known | |
6013 | ;; specifiers after them too. | |
6014 | (while | |
6015 | (let* ((start (point)) kwd-sym kwd-clause-end found-type) | |
6016 | ||
6017 | ;; Look for a specifier keyword clause. | |
6018 | (when (looking-at c-prefix-spec-kwds-re) | |
6019 | (setq kwd-sym (c-keyword-sym (match-string 1))) | |
6020 | (save-excursion | |
6021 | (c-forward-keyword-clause 1) | |
6022 | (setq kwd-clause-end (point)))) | |
6023 | ||
6024 | (when (setq found-type (c-forward-type)) | |
6025 | ;; Found a known or possible type or a prefix of a known type. | |
6026 | ||
6027 | (when at-type | |
6028 | ;; Got two identifiers with nothing but whitespace | |
6029 | ;; between them. That can only happen in declarations. | |
6030 | (setq at-decl-or-cast 'ids) | |
6031 | ||
6032 | (when (eq at-type 'found) | |
6033 | ;; If the previous identifier is a found type we | |
6034 | ;; record it as a real one; it might be some sort of | |
6035 | ;; alias for a prefix like "unsigned". | |
6036 | (save-excursion | |
6037 | (goto-char type-start) | |
6038 | (let ((c-promote-possible-types t)) | |
6039 | (c-forward-type))))) | |
6040 | ||
6041 | (setq backup-at-type at-type | |
6042 | backup-type-start type-start | |
6043 | backup-id-start id-start | |
6044 | at-type found-type | |
6045 | type-start start | |
6046 | id-start (point) | |
6047 | ;; The previous ambiguous specifier/type turned out | |
6048 | ;; to be a type since we've parsed another one after | |
6049 | ;; it, so clear these backup flags. | |
6050 | backup-at-type-decl nil | |
6051 | backup-maybe-typeless nil)) | |
6052 | ||
6053 | (if kwd-sym | |
6054 | (progn | |
6055 | ;; Handle known specifier keywords and | |
6056 | ;; `c-decl-hangon-kwds' which can occur after known | |
6057 | ;; types. | |
6058 | ||
6059 | (if (c-keyword-member kwd-sym 'c-decl-hangon-kwds) | |
6060 | ;; It's a hang-on keyword that can occur anywhere. | |
6061 | (progn | |
6062 | (setq at-decl-or-cast t) | |
6063 | (if at-type | |
6064 | ;; Move the identifier start position if | |
6065 | ;; we've passed a type. | |
6066 | (setq id-start kwd-clause-end) | |
6067 | ;; Otherwise treat this as a specifier and | |
6068 | ;; move the fallback position. | |
6069 | (setq start-pos kwd-clause-end)) | |
6070 | (goto-char kwd-clause-end)) | |
6071 | ||
6072 | ;; It's an ordinary specifier so we know that | |
6073 | ;; anything before this can't be the type. | |
6074 | (setq backup-at-type nil | |
6075 | start-pos kwd-clause-end) | |
6076 | ||
6077 | (if found-type | |
6078 | ;; It's ambiguous whether this keyword is a | |
6079 | ;; specifier or a type prefix, so set the backup | |
6080 | ;; flags. (It's assumed that `c-forward-type' | |
6081 | ;; moved further than `c-forward-keyword-clause'.) | |
6082 | (progn | |
6083 | (when (c-keyword-member kwd-sym 'c-typedef-decl-kwds) | |
6084 | (setq backup-at-type-decl t)) | |
6085 | (when (c-keyword-member kwd-sym 'c-typeless-decl-kwds) | |
6086 | (setq backup-maybe-typeless t))) | |
6087 | ||
6088 | (when (c-keyword-member kwd-sym 'c-typedef-decl-kwds) | |
6089 | (setq at-type-decl t)) | |
6090 | (when (c-keyword-member kwd-sym 'c-typeless-decl-kwds) | |
6091 | (setq maybe-typeless t)) | |
6092 | ||
6093 | ;; Haven't matched a type so it's an umambiguous | |
6094 | ;; specifier keyword and we know we're in a | |
6095 | ;; declaration. | |
6096 | (setq at-decl-or-cast t) | |
6097 | ||
6098 | (goto-char kwd-clause-end)))) | |
6099 | ||
6100 | ;; If the type isn't known we continue so that we'll jump | |
6101 | ;; over all specifiers and type identifiers. The reason | |
6102 | ;; to do this for a known type prefix is to make things | |
6103 | ;; like "unsigned INT16" work. | |
6104 | (and found-type (not (eq found-type t)))))) | |
6105 | ||
6106 | (cond | |
6107 | ((eq at-type t) | |
6108 | ;; If a known type was found, we still need to skip over any | |
6109 | ;; hangon keyword clauses after it. Otherwise it has already | |
6110 | ;; been done in the loop above. | |
6111 | (while (looking-at c-decl-hangon-key) | |
6112 | (c-forward-keyword-clause 1)) | |
6113 | (setq id-start (point))) | |
6114 | ||
6115 | ((eq at-type 'prefix) | |
6116 | ;; A prefix type is itself a primitive type when it's not | |
6117 | ;; followed by another type. | |
6118 | (setq at-type t)) | |
6119 | ||
6120 | ((not at-type) | |
6121 | ;; Got no type but set things up to continue anyway to handle | |
6122 | ;; the various cases when a declaration doesn't start with a | |
6123 | ;; type. | |
6124 | (setq id-start start-pos)) | |
6125 | ||
6126 | ((and (eq at-type 'maybe) | |
6127 | (c-major-mode-is 'c++-mode)) | |
6128 | ;; If it's C++ then check if the last "type" ends on the form | |
6129 | ;; "foo::foo" or "foo::~foo", i.e. if it's the name of a | |
6130 | ;; (con|de)structor. | |
6131 | (save-excursion | |
6132 | (let (name end-2 end-1) | |
6133 | (goto-char id-start) | |
6134 | (c-backward-syntactic-ws) | |
6135 | (setq end-2 (point)) | |
6136 | (when (and | |
6137 | (c-simple-skip-symbol-backward) | |
6138 | (progn | |
6139 | (setq name | |
6140 | (buffer-substring-no-properties (point) end-2)) | |
6141 | ;; Cheating in the handling of syntactic ws below. | |
6142 | (< (skip-chars-backward ":~ \t\n\r\v\f") 0)) | |
6143 | (progn | |
6144 | (setq end-1 (point)) | |
6145 | (c-simple-skip-symbol-backward)) | |
6146 | (>= (point) type-start) | |
6147 | (equal (buffer-substring-no-properties (point) end-1) | |
6148 | name)) | |
6149 | ;; It is a (con|de)structor name. In that case the | |
6150 | ;; declaration is typeless so zap out any preceding | |
6151 | ;; identifier(s) that we might have taken as types. | |
6152 | (goto-char type-start) | |
6153 | (setq at-type nil | |
6154 | backup-at-type nil | |
6155 | id-start type-start)))))) | |
6156 | ||
6157 | ;; Check for and step over a type decl expression after the thing | |
6158 | ;; that is or might be a type. This can't be skipped since we | |
6159 | ;; need the correct end position of the declarator for | |
6160 | ;; `max-type-decl-end-*'. | |
6161 | (let ((start (point)) (paren-depth 0) pos | |
6162 | ;; True if there's a non-open-paren match of | |
6163 | ;; `c-type-decl-prefix-key'. | |
6164 | got-prefix | |
6165 | ;; True if the declarator is surrounded by a parenthesis pair. | |
6166 | got-parens | |
6167 | ;; True if there is an identifier in the declarator. | |
6168 | got-identifier | |
6169 | ;; True if there's a non-close-paren match of | |
6170 | ;; `c-type-decl-suffix-key'. | |
6171 | got-suffix | |
6172 | ;; True if there's a prefix match outside the outermost | |
6173 | ;; paren pair that surrounds the declarator. | |
6174 | got-prefix-before-parens | |
1379f2c5 | 6175 | ;; True if there's a suffix match outside the outermost |
0386b551 AM |
6176 | ;; paren pair that surrounds the declarator. The value is |
6177 | ;; the position of the first suffix match. | |
6178 | got-suffix-after-parens | |
6179 | ;; True if we've parsed the type decl to a token that is | |
6180 | ;; known to end declarations in this context. | |
6181 | at-decl-end | |
6182 | ;; The earlier values of `at-type' and `type-start' if we've | |
6183 | ;; shifted the type backwards. | |
6184 | identifier-type identifier-start | |
6185 | ;; If `c-parse-and-markup-<>-arglists' is set we need to | |
6186 | ;; turn it off during the name skipping below to avoid | |
6187 | ;; getting `c-type' properties that might be bogus. That | |
6188 | ;; can happen since we don't know if | |
6189 | ;; `c-restricted-<>-arglists' will be correct inside the | |
6190 | ;; arglist paren that gets entered. | |
6191 | c-parse-and-markup-<>-arglists) | |
6192 | ||
6193 | (goto-char id-start) | |
6194 | ||
6195 | ;; Skip over type decl prefix operators. (Note similar code in | |
6196 | ;; `c-font-lock-declarators'.) | |
6197 | (while (and (looking-at c-type-decl-prefix-key) | |
6198 | (if (and (c-major-mode-is 'c++-mode) | |
6199 | (match-beginning 2)) | |
6200 | ;; If the second submatch matches in C++ then | |
6201 | ;; we're looking at an identifier that's a | |
6202 | ;; prefix only if it specifies a member pointer. | |
6203 | (when (setq got-identifier (c-forward-name)) | |
6204 | (if (looking-at "\\(::\\)") | |
6205 | ;; We only check for a trailing "::" and | |
6206 | ;; let the "*" that should follow be | |
6207 | ;; matched in the next round. | |
6208 | (progn (setq got-identifier nil) t) | |
6209 | ;; It turned out to be the real identifier, | |
6210 | ;; so stop. | |
6211 | nil)) | |
6212 | t)) | |
6213 | ||
6214 | (if (eq (char-after) ?\() | |
6215 | (progn | |
6216 | (setq paren-depth (1+ paren-depth)) | |
6217 | (forward-char)) | |
6218 | (unless got-prefix-before-parens | |
6219 | (setq got-prefix-before-parens (= paren-depth 0))) | |
6220 | (setq got-prefix t) | |
6221 | (goto-char (match-end 1))) | |
6222 | (c-forward-syntactic-ws)) | |
6223 | ||
6224 | (setq got-parens (> paren-depth 0)) | |
6225 | ||
6226 | ;; Skip over an identifier. | |
6227 | (or got-identifier | |
6228 | (and (looking-at c-identifier-start) | |
6229 | (setq got-identifier (c-forward-name)))) | |
6230 | ||
6231 | ;; Skip over type decl suffix operators. | |
6232 | (while (if (looking-at c-type-decl-suffix-key) | |
6233 | ||
6234 | (if (eq (char-after) ?\)) | |
6235 | (when (> paren-depth 0) | |
6236 | (setq paren-depth (1- paren-depth)) | |
6237 | (forward-char) | |
6238 | t) | |
6239 | (when (if (save-match-data (looking-at "\\s\(")) | |
6240 | (c-safe (c-forward-sexp 1) t) | |
6241 | (goto-char (match-end 1)) | |
6242 | t) | |
6243 | (when (and (not got-suffix-after-parens) | |
6244 | (= paren-depth 0)) | |
6245 | (setq got-suffix-after-parens (match-beginning 0))) | |
6246 | (setq got-suffix t))) | |
6247 | ||
6248 | ;; No suffix matched. We might have matched the | |
6249 | ;; identifier as a type and the open paren of a | |
6250 | ;; function arglist as a type decl prefix. In that | |
6251 | ;; case we should "backtrack": Reinterpret the last | |
6252 | ;; type as the identifier, move out of the arglist and | |
6253 | ;; continue searching for suffix operators. | |
6254 | ;; | |
6255 | ;; Do this even if there's no preceding type, to cope | |
6256 | ;; with old style function declarations in K&R C, | |
6257 | ;; (con|de)structors in C++ and `c-typeless-decl-kwds' | |
6258 | ;; style declarations. That isn't applicable in an | |
6259 | ;; arglist context, though. | |
6260 | (when (and (= paren-depth 1) | |
6261 | (not got-prefix-before-parens) | |
6262 | (not (eq at-type t)) | |
6263 | (or backup-at-type | |
6264 | maybe-typeless | |
6265 | backup-maybe-typeless | |
6266 | (when c-recognize-typeless-decls | |
6267 | (not context))) | |
6268 | (setq pos (c-up-list-forward (point))) | |
6269 | (eq (char-before pos) ?\))) | |
6270 | (c-fdoc-shift-type-backward) | |
6271 | (goto-char pos) | |
6272 | t)) | |
6273 | ||
6274 | (c-forward-syntactic-ws)) | |
6275 | ||
6276 | (when (and (or maybe-typeless backup-maybe-typeless) | |
6277 | (not got-identifier) | |
6278 | (not got-prefix) | |
6279 | at-type) | |
6280 | ;; Have found no identifier but `c-typeless-decl-kwds' has | |
6281 | ;; matched so we know we're inside a declaration. The | |
6282 | ;; preceding type must be the identifier instead. | |
6283 | (c-fdoc-shift-type-backward)) | |
6284 | ||
6285 | (setq | |
6286 | at-decl-or-cast | |
6287 | (catch 'at-decl-or-cast | |
6288 | ||
a85fd6da | 6289 | ;; CASE 1 |
0386b551 AM |
6290 | (when (> paren-depth 0) |
6291 | ;; Encountered something inside parens that isn't matched by | |
6292 | ;; the `c-type-decl-*' regexps, so it's not a type decl | |
6293 | ;; expression. Try to skip out to the same paren depth to | |
6294 | ;; not confuse the cast check below. | |
6295 | (c-safe (goto-char (scan-lists (point) 1 paren-depth))) | |
6296 | ;; If we've found a specifier keyword then it's a | |
6297 | ;; declaration regardless. | |
6298 | (throw 'at-decl-or-cast (eq at-decl-or-cast t))) | |
6299 | ||
6300 | (setq at-decl-end | |
6301 | (looking-at (cond ((eq context '<>) "[,>]") | |
6302 | (context "[,\)]") | |
6303 | (t "[,;]")))) | |
6304 | ||
6305 | ;; Now we've collected info about various characteristics of | |
6306 | ;; the construct we're looking at. Below follows a decision | |
6307 | ;; tree based on that. It's ordered to check more certain | |
6308 | ;; signs before less certain ones. | |
6309 | ||
6310 | (if got-identifier | |
6311 | (progn | |
6312 | ||
a85fd6da | 6313 | ;; CASE 2 |
0386b551 AM |
6314 | (when (and (or at-type maybe-typeless) |
6315 | (not (or got-prefix got-parens))) | |
6316 | ;; Got another identifier directly after the type, so it's a | |
6317 | ;; declaration. | |
6318 | (throw 'at-decl-or-cast t)) | |
6319 | ||
6320 | (when (and got-parens | |
6321 | (not got-prefix) | |
6322 | (not got-suffix-after-parens) | |
6323 | (or backup-at-type | |
6324 | maybe-typeless | |
6325 | backup-maybe-typeless)) | |
6326 | ;; Got a declaration of the form "foo bar (gnu);" where we've | |
6327 | ;; recognized "bar" as the type and "gnu" as the declarator. | |
6328 | ;; In this case it's however more likely that "bar" is the | |
6329 | ;; declarator and "gnu" a function argument or initializer (if | |
6330 | ;; `c-recognize-paren-inits' is set), since the parens around | |
6331 | ;; "gnu" would be superfluous if it's a declarator. Shift the | |
6332 | ;; type one step backward. | |
6333 | (c-fdoc-shift-type-backward))) | |
6334 | ||
6335 | ;; Found no identifier. | |
6336 | ||
6337 | (if backup-at-type | |
6338 | (progn | |
6339 | ||
a85fd6da | 6340 | ;; CASE 3 |
0386b551 AM |
6341 | (when (= (point) start) |
6342 | ;; Got a plain list of identifiers. If a colon follows it's | |
6343 | ;; a valid label. Otherwise the last one probably is the | |
6344 | ;; declared identifier and we should back up to the previous | |
6345 | ;; type, providing it isn't a cast. | |
6346 | (if (eq (char-after) ?:) | |
6347 | ;; If we've found a specifier keyword then it's a | |
6348 | ;; declaration regardless. | |
6349 | (throw 'at-decl-or-cast (eq at-decl-or-cast t)) | |
6350 | (setq backup-if-not-cast t) | |
6351 | (throw 'at-decl-or-cast t))) | |
6352 | ||
a85fd6da | 6353 | ;; CASE 4 |
0386b551 AM |
6354 | (when (and got-suffix |
6355 | (not got-prefix) | |
6356 | (not got-parens)) | |
6357 | ;; Got a plain list of identifiers followed by some suffix. | |
6358 | ;; If this isn't a cast then the last identifier probably is | |
6359 | ;; the declared one and we should back up to the previous | |
6360 | ;; type. | |
6361 | (setq backup-if-not-cast t) | |
6362 | (throw 'at-decl-or-cast t))) | |
6363 | ||
a85fd6da | 6364 | ;; CASE 5 |
0386b551 AM |
6365 | (when (eq at-type t) |
6366 | ;; If the type is known we know that there can't be any | |
6367 | ;; identifier somewhere else, and it's only in declarations in | |
6368 | ;; e.g. function prototypes and in casts that the identifier may | |
6369 | ;; be left out. | |
6370 | (throw 'at-decl-or-cast t)) | |
6371 | ||
6372 | (when (= (point) start) | |
6373 | ;; Only got a single identifier (parsed as a type so far). | |
a85fd6da | 6374 | ;; CASE 6 |
0386b551 AM |
6375 | (if (and |
6376 | ;; Check that the identifier isn't at the start of an | |
6377 | ;; expression. | |
6378 | at-decl-end | |
6379 | (cond | |
6380 | ((eq context 'decl) | |
6381 | ;; Inside an arglist that contains declarations. If K&R | |
6382 | ;; style declarations and parenthesis style initializers | |
6383 | ;; aren't allowed then the single identifier must be a | |
6384 | ;; type, else we require that it's known or found | |
6385 | ;; (primitive types are handled above). | |
6386 | (or (and (not c-recognize-knr-p) | |
6387 | (not c-recognize-paren-inits)) | |
6388 | (memq at-type '(known found)))) | |
6389 | ((eq context '<>) | |
6390 | ;; Inside a template arglist. Accept known and found | |
6391 | ;; types; other identifiers could just as well be | |
6392 | ;; constants in C++. | |
6393 | (memq at-type '(known found))))) | |
6394 | (throw 'at-decl-or-cast t) | |
a85fd6da | 6395 | ;; CASE 7 |
0386b551 AM |
6396 | ;; Can't be a valid declaration or cast, but if we've found a |
6397 | ;; specifier it can't be anything else either, so treat it as | |
6398 | ;; an invalid/unfinished declaration or cast. | |
6399 | (throw 'at-decl-or-cast at-decl-or-cast)))) | |
6400 | ||
6401 | (if (and got-parens | |
6402 | (not got-prefix) | |
6403 | (not context) | |
6404 | (not (eq at-type t)) | |
6405 | (or backup-at-type | |
6406 | maybe-typeless | |
6407 | backup-maybe-typeless | |
6408 | (when c-recognize-typeless-decls | |
6409 | (or (not got-suffix) | |
6410 | (not (looking-at | |
6411 | c-after-suffixed-type-maybe-decl-key)))))) | |
6412 | ;; Got an empty paren pair and a preceding type that probably | |
6413 | ;; really is the identifier. Shift the type backwards to make | |
6414 | ;; the last one the identifier. This is analogous to the | |
6415 | ;; "backtracking" done inside the `c-type-decl-suffix-key' loop | |
6416 | ;; above. | |
6417 | ;; | |
6418 | ;; Exception: In addition to the conditions in that | |
6419 | ;; "backtracking" code, do not shift backward if we're not | |
6420 | ;; looking at either `c-after-suffixed-type-decl-key' or "[;,]". | |
6421 | ;; Since there's no preceding type, the shift would mean that | |
6422 | ;; the declaration is typeless. But if the regexp doesn't match | |
6423 | ;; then we will simply fall through in the tests below and not | |
6424 | ;; recognize it at all, so it's better to try it as an abstract | |
6425 | ;; declarator instead. | |
6426 | (c-fdoc-shift-type-backward) | |
6427 | ||
6428 | ;; Still no identifier. | |
a85fd6da | 6429 | ;; CASE 8 |
0386b551 AM |
6430 | (when (and got-prefix (or got-parens got-suffix)) |
6431 | ;; Require `got-prefix' together with either `got-parens' or | |
6432 | ;; `got-suffix' to recognize it as an abstract declarator: | |
6433 | ;; `got-parens' only is probably an empty function call. | |
6434 | ;; `got-suffix' only can build an ordinary expression together | |
6435 | ;; with the preceding identifier which we've taken as a type. | |
6436 | ;; We could actually accept on `got-prefix' only, but that can | |
6437 | ;; easily occur temporarily while writing an expression so we | |
6438 | ;; avoid that case anyway. We could do a better job if we knew | |
6439 | ;; the point when the fontification was invoked. | |
6440 | (throw 'at-decl-or-cast t)) | |
6441 | ||
a85fd6da | 6442 | ;; CASE 9 |
0386b551 AM |
6443 | (when (and at-type |
6444 | (not got-prefix) | |
6445 | (not got-parens) | |
6446 | got-suffix-after-parens | |
6447 | (eq (char-after got-suffix-after-parens) ?\()) | |
6448 | ;; Got a type, no declarator but a paren suffix. I.e. it's a | |
6449 | ;; normal function call afterall (or perhaps a C++ style object | |
6450 | ;; instantiation expression). | |
6451 | (throw 'at-decl-or-cast nil)))) | |
6452 | ||
a85fd6da | 6453 | ;; CASE 10 |
0386b551 AM |
6454 | (when at-decl-or-cast |
6455 | ;; By now we've located the type in the declaration that we know | |
6456 | ;; we're in. | |
6457 | (throw 'at-decl-or-cast t)) | |
6458 | ||
a85fd6da | 6459 | ;; CASE 11 |
0386b551 AM |
6460 | (when (and got-identifier |
6461 | (not context) | |
6462 | (looking-at c-after-suffixed-type-decl-key) | |
6463 | (if (and got-parens | |
6464 | (not got-prefix) | |
6465 | (not got-suffix) | |
6466 | (not (eq at-type t))) | |
6467 | ;; Shift the type backward in the case that there's a | |
6468 | ;; single identifier inside parens. That can only | |
6469 | ;; occur in K&R style function declarations so it's | |
6470 | ;; more likely that it really is a function call. | |
6471 | ;; Therefore we only do this after | |
6472 | ;; `c-after-suffixed-type-decl-key' has matched. | |
6473 | (progn (c-fdoc-shift-type-backward) t) | |
6474 | got-suffix-after-parens)) | |
6475 | ;; A declaration according to `c-after-suffixed-type-decl-key'. | |
6476 | (throw 'at-decl-or-cast t)) | |
6477 | ||
a85fd6da | 6478 | ;; CASE 12 |
0386b551 AM |
6479 | (when (and (or got-prefix (not got-parens)) |
6480 | (memq at-type '(t known))) | |
6481 | ;; It's a declaration if a known type precedes it and it can't be a | |
6482 | ;; function call. | |
6483 | (throw 'at-decl-or-cast t)) | |
6484 | ||
6485 | ;; If we get here we can't tell if this is a type decl or a normal | |
6486 | ;; expression by looking at it alone. (That's under the assumption | |
6487 | ;; that normal expressions always can look like type decl expressions, | |
6488 | ;; which isn't really true but the cases where it doesn't hold are so | |
6489 | ;; uncommon (e.g. some placements of "const" in C++) it's not worth | |
6490 | ;; the effort to look for them.) | |
6491 | ||
6492 | (unless (or at-decl-end (looking-at "=[^=]")) | |
6493 | ;; If this is a declaration it should end here or its initializer(*) | |
6494 | ;; should start here, so check for allowed separation tokens. Note | |
6495 | ;; that this rule doesn't work e.g. with a K&R arglist after a | |
6496 | ;; function header. | |
6497 | ;; | |
6498 | ;; *) Don't check for C++ style initializers using parens | |
6499 | ;; since those already have been matched as suffixes. | |
6500 | ;; | |
6501 | ;; If `at-decl-or-cast' is then we've found some other sign that | |
6502 | ;; it's a declaration or cast, so then it's probably an | |
6503 | ;; invalid/unfinished one. | |
6504 | (throw 'at-decl-or-cast at-decl-or-cast)) | |
6505 | ||
6506 | ;; Below are tests that only should be applied when we're certain to | |
6507 | ;; not have parsed halfway through an expression. | |
6508 | ||
a85fd6da | 6509 | ;; CASE 14 |
0386b551 AM |
6510 | (when (memq at-type '(t known)) |
6511 | ;; The expression starts with a known type so treat it as a | |
6512 | ;; declaration. | |
6513 | (throw 'at-decl-or-cast t)) | |
6514 | ||
a85fd6da | 6515 | ;; CASE 15 |
0386b551 AM |
6516 | (when (and (c-major-mode-is 'c++-mode) |
6517 | ;; In C++ we check if the identifier is a known type, since | |
6518 | ;; (con|de)structors use the class name as identifier. | |
6519 | ;; We've always shifted over the identifier as a type and | |
6520 | ;; then backed up again in this case. | |
6521 | identifier-type | |
6522 | (or (memq identifier-type '(found known)) | |
6523 | (and (eq (char-after identifier-start) ?~) | |
6524 | ;; `at-type' probably won't be 'found for | |
6525 | ;; destructors since the "~" is then part of the | |
6526 | ;; type name being checked against the list of | |
6527 | ;; known types, so do a check without that | |
6528 | ;; operator. | |
6529 | (or (save-excursion | |
6530 | (goto-char (1+ identifier-start)) | |
6531 | (c-forward-syntactic-ws) | |
6532 | (c-with-syntax-table | |
6533 | c-identifier-syntax-table | |
6534 | (looking-at c-known-type-key))) | |
6535 | (save-excursion | |
6536 | (goto-char (1+ identifier-start)) | |
6537 | ;; We have already parsed the type earlier, | |
6538 | ;; so it'd be possible to cache the end | |
6539 | ;; position instead of redoing it here, but | |
6540 | ;; then we'd need to keep track of another | |
6541 | ;; position everywhere. | |
6542 | (c-check-type (point) | |
6543 | (progn (c-forward-type) | |
6544 | (point)))))))) | |
6545 | (throw 'at-decl-or-cast t)) | |
6546 | ||
6547 | (if got-identifier | |
6548 | (progn | |
a85fd6da | 6549 | ;; CASE 16 |
0386b551 AM |
6550 | (when (and got-prefix-before-parens |
6551 | at-type | |
6552 | (or at-decl-end (looking-at "=[^=]")) | |
6553 | (not context) | |
6554 | (not got-suffix)) | |
6555 | ;; Got something like "foo * bar;". Since we're not inside an | |
6556 | ;; arglist it would be a meaningless expression because the | |
6557 | ;; result isn't used. We therefore choose to recognize it as | |
6558 | ;; a declaration. Do not allow a suffix since it could then | |
6559 | ;; be a function call. | |
6560 | (throw 'at-decl-or-cast t)) | |
6561 | ||
a85fd6da | 6562 | ;; CASE 17 |
0386b551 AM |
6563 | (when (and (or got-suffix-after-parens |
6564 | (looking-at "=[^=]")) | |
6565 | (eq at-type 'found) | |
6566 | (not (eq context 'arglist))) | |
6567 | ;; Got something like "a (*b) (c);" or "a (b) = c;". It could | |
6568 | ;; be an odd expression or it could be a declaration. Treat | |
6569 | ;; it as a declaration if "a" has been used as a type | |
6570 | ;; somewhere else (if it's a known type we won't get here). | |
6571 | (throw 'at-decl-or-cast t))) | |
6572 | ||
a85fd6da | 6573 | ;; CASE 18 |
0386b551 AM |
6574 | (when (and context |
6575 | (or got-prefix | |
6576 | (and (eq context 'decl) | |
6577 | (not c-recognize-paren-inits) | |
6578 | (or got-parens got-suffix)))) | |
6579 | ;; Got a type followed by an abstract declarator. If `got-prefix' | |
6580 | ;; is set it's something like "a *" without anything after it. If | |
6581 | ;; `got-parens' or `got-suffix' is set it's "a()", "a[]", "a()[]", | |
6582 | ;; or similar, which we accept only if the context rules out | |
6583 | ;; expressions. | |
6584 | (throw 'at-decl-or-cast t))) | |
6585 | ||
6586 | ;; If we had a complete symbol table here (which rules out | |
6587 | ;; `c-found-types') we should return t due to the disambiguation rule | |
6588 | ;; (in at least C++) that anything that can be parsed as a declaration | |
6589 | ;; is a declaration. Now we're being more defensive and prefer to | |
6590 | ;; highlight things like "foo (bar);" as a declaration only if we're | |
6591 | ;; inside an arglist that contains declarations. | |
6592 | (eq context 'decl)))) | |
6593 | ||
6594 | ;; The point is now after the type decl expression. | |
6595 | ||
6596 | (cond | |
6597 | ;; Check for a cast. | |
6598 | ((save-excursion | |
6599 | (and | |
6600 | c-cast-parens | |
6601 | ||
6602 | ;; Should be the first type/identifier in a cast paren. | |
6603 | (> preceding-token-end (point-min)) | |
6604 | (memq (char-before preceding-token-end) c-cast-parens) | |
6605 | ||
6606 | ;; The closing paren should follow. | |
6607 | (progn | |
6608 | (c-forward-syntactic-ws) | |
6609 | (looking-at "\\s\)")) | |
6610 | ||
6611 | ;; There should be a primary expression after it. | |
6612 | (let (pos) | |
6613 | (forward-char) | |
6614 | (c-forward-syntactic-ws) | |
6615 | (setq cast-end (point)) | |
6616 | (and (looking-at c-primary-expr-regexp) | |
6617 | (progn | |
6618 | (setq pos (match-end 0)) | |
6619 | (or | |
6620 | ;; Check if the expression begins with a prefix keyword. | |
6621 | (match-beginning 2) | |
6622 | (if (match-beginning 1) | |
6623 | ;; Expression begins with an ambiguous operator. Treat | |
6624 | ;; it as a cast if it's a type decl or if we've | |
6625 | ;; recognized the type somewhere else. | |
6626 | (or at-decl-or-cast | |
6627 | (memq at-type '(t known found))) | |
6628 | ;; Unless it's a keyword, it's the beginning of a primary | |
6629 | ;; expression. | |
6630 | (not (looking-at c-keywords-regexp))))) | |
6631 | ;; If `c-primary-expr-regexp' matched a nonsymbol token, check | |
6632 | ;; that it matched a whole one so that we don't e.g. confuse | |
6633 | ;; the operator '-' with '->'. It's ok if it matches further, | |
6634 | ;; though, since it e.g. can match the float '.5' while the | |
6635 | ;; operator regexp only matches '.'. | |
6636 | (or (not (looking-at c-nonsymbol-token-regexp)) | |
6637 | (<= (match-end 0) pos)))) | |
6638 | ||
6639 | ;; There should either be a cast before it or something that isn't an | |
6640 | ;; identifier or close paren. | |
6641 | (> preceding-token-end (point-min)) | |
6642 | (progn | |
6643 | (goto-char (1- preceding-token-end)) | |
6644 | (or (eq (point) last-cast-end) | |
6645 | (progn | |
6646 | (c-backward-syntactic-ws) | |
6647 | (if (< (skip-syntax-backward "w_") 0) | |
6648 | ;; It's a symbol. Accept it only if it's one of the | |
6649 | ;; keywords that can precede an expression (without | |
6650 | ;; surrounding parens). | |
6651 | (looking-at c-simple-stmt-key) | |
6652 | (and | |
6653 | ;; Check that it isn't a close paren (block close is ok, | |
6654 | ;; though). | |
6655 | (not (memq (char-before) '(?\) ?\]))) | |
6656 | ;; Check that it isn't a nonsymbol identifier. | |
6657 | (not (c-on-identifier))))))))) | |
6658 | ||
6659 | ;; Handle the cast. | |
6660 | (when (and c-record-type-identifiers at-type (not (eq at-type t))) | |
6661 | (let ((c-promote-possible-types t)) | |
6662 | (goto-char type-start) | |
6663 | (c-forward-type))) | |
6664 | ||
6665 | (goto-char cast-end) | |
6666 | 'cast) | |
6667 | ||
6668 | (at-decl-or-cast | |
6669 | ;; We're at a declaration. Highlight the type and the following | |
6670 | ;; declarators. | |
6671 | ||
6672 | (when backup-if-not-cast | |
6673 | (c-fdoc-shift-type-backward t)) | |
6674 | ||
6675 | (when (and (eq context 'decl) (looking-at ",")) | |
6676 | ;; Make sure to propagate the `c-decl-arg-start' property to | |
6677 | ;; the next argument if it's set in this one, to cope with | |
6678 | ;; interactive refontification. | |
6679 | (c-put-c-type-property (point) 'c-decl-arg-start)) | |
6680 | ||
6681 | (when (and c-record-type-identifiers at-type (not (eq at-type t))) | |
6682 | (let ((c-promote-possible-types t)) | |
6683 | (save-excursion | |
6684 | (goto-char type-start) | |
6685 | (c-forward-type)))) | |
6686 | ||
6687 | (cons id-start at-type-decl)) | |
6688 | ||
6689 | (t | |
6690 | ;; False alarm. Restore the recorded ranges. | |
6691 | (setq c-record-type-identifiers save-rec-type-ids | |
6692 | c-record-ref-identifiers save-rec-ref-ids) | |
6693 | nil)))) | |
6694 | ||
6695 | (defun c-forward-label (&optional assume-markup preceding-token-end limit) | |
51c9af45 | 6696 | ;; Assuming that point is at the beginning of a token, check if it starts a |
1379f2c5 AM |
6697 | ;; label and if so move over it and return non-nil (t in default situations, |
6698 | ;; specific symbols (see below) for interesting situations), otherwise don't | |
6699 | ;; move and return nil. "Label" here means "most things with a colon". | |
51c9af45 AM |
6700 | ;; |
6701 | ;; More precisely, a "label" is regarded as one of: | |
1379f2c5 AM |
6702 | ;; (i) a goto target like "foo:" - returns the symbol `goto-target'; |
6703 | ;; (ii) A case label - either the entire construct "case FOO:", or just the | |
6704 | ;; bare "case", should the colon be missing. We return t; | |
6705 | ;; (iii) a keyword which needs a colon, like "default:" or "private:"; We | |
6706 | ;; return t; | |
51c9af45 | 6707 | ;; (iv) One of QT's "extended" C++ variants of |
1379f2c5 AM |
6708 | ;; "private:"/"protected:"/"public:"/"more:" looking like "public slots:". |
6709 | ;; Returns the symbol `qt-2kwds-colon'. | |
6710 | ;; (v) QT's construct "signals:". Returns the symbol `qt-1kwd-colon'. | |
82ba65cf | 6711 | ;; (vi) One of the keywords matched by `c-opt-extra-label-key' (without any |
51c9af45 | 6712 | ;; colon). Currently (2006-03), this applies only to Objective C's |
1379f2c5 | 6713 | ;; keywords "@private", "@protected", and "@public". Returns t. |
51c9af45 AM |
6714 | ;; |
6715 | ;; One of the things which will NOT be recognised as a label is a bit-field | |
6716 | ;; element of a struct, something like "int foo:5". | |
6717 | ;; | |
6718 | ;; The end of the label is taken to be just after the colon, or the end of | |
6719 | ;; the first submatch in `c-opt-extra-label-key'. The point is directly | |
6720 | ;; after the end on return. The terminating char gets marked with | |
6721 | ;; `c-decl-end' to improve recognition of the following declaration or | |
6722 | ;; statement. | |
0386b551 AM |
6723 | ;; |
6724 | ;; If ASSUME-MARKUP is non-nil, it's assumed that the preceding | |
51c9af45 | 6725 | ;; label, if any, has already been marked up like that. |
0386b551 AM |
6726 | ;; |
6727 | ;; If PRECEDING-TOKEN-END is given, it should be the first position | |
6728 | ;; after the preceding token, i.e. on the other side of the | |
6729 | ;; syntactic ws from the point. Use a value less than or equal to | |
6730 | ;; (point-min) if the point is at the first token in (the visible | |
6731 | ;; part of) the buffer. | |
6732 | ;; | |
6733 | ;; The optional LIMIT limits the forward scan for the colon. | |
6734 | ;; | |
6735 | ;; This function records the ranges of the label symbols on | |
6736 | ;; `c-record-ref-identifiers' if `c-record-type-identifiers' (!) is | |
6737 | ;; non-nil. | |
6738 | ;; | |
6739 | ;; This function might do hidden buffer changes. | |
6740 | ||
51c9af45 | 6741 | (let ((start (point)) |
1379f2c5 | 6742 | label-end |
51c9af45 | 6743 | qt-symbol-idx |
1379f2c5 | 6744 | macro-start ; if we're in one. |
f412a567 AM |
6745 | label-type |
6746 | kwd) | |
0386b551 | 6747 | (cond |
b414f371 | 6748 | ;; "case" or "default" (Doesn't apply to AWK). |
0386b551 AM |
6749 | ((looking-at c-label-kwds-regexp) |
6750 | (let ((kwd-end (match-end 1))) | |
6751 | ;; Record only the keyword itself for fontification, since in | |
6752 | ;; case labels the following is a constant expression and not | |
6753 | ;; a label. | |
6754 | (when c-record-type-identifiers | |
6755 | (c-record-ref-id (cons (match-beginning 1) kwd-end))) | |
6756 | ||
6757 | ;; Find the label end. | |
6758 | (goto-char kwd-end) | |
1379f2c5 AM |
6759 | (setq label-type |
6760 | (if (and (c-syntactic-re-search-forward | |
6761 | ;; Stop on chars that aren't allowed in expressions, | |
6762 | ;; and on operator chars that would be meaningless | |
6763 | ;; there. FIXME: This doesn't cope with ?: operators. | |
6764 | "[;{=,@]\\|\\(\\=\\|[^:]\\):\\([^:]\\|\\'\\)" | |
6765 | limit t t nil 1) | |
6766 | (match-beginning 2)) | |
6767 | ||
6768 | (progn ; there's a proper : | |
6769 | (goto-char (match-beginning 2)) ; just after the : | |
6770 | (c-put-c-type-property (1- (point)) 'c-decl-end) | |
6771 | t) | |
6772 | ||
6773 | ;; It's an unfinished label. We consider the keyword enough | |
6774 | ;; to recognize it as a label, so that it gets fontified. | |
6775 | ;; Leave the point at the end of it, but don't put any | |
6776 | ;; `c-decl-end' marker. | |
6777 | (goto-char kwd-end) | |
6778 | t)))) | |
0386b551 | 6779 | |
51c9af45 | 6780 | ;; @private, @protected, @public, in Objective C, or similar. |
0386b551 AM |
6781 | ((and c-opt-extra-label-key |
6782 | (looking-at c-opt-extra-label-key)) | |
6783 | ;; For a `c-opt-extra-label-key' match, we record the whole | |
6784 | ;; thing for fontification. That's to get the leading '@' in | |
6785 | ;; Objective-C protection labels fontified. | |
6786 | (goto-char (match-end 1)) | |
6787 | (when c-record-type-identifiers | |
6788 | (c-record-ref-id (cons (match-beginning 1) (point)))) | |
6789 | (c-put-c-type-property (1- (point)) 'c-decl-end) | |
1379f2c5 | 6790 | (setq label-type t)) |
0386b551 | 6791 | |
51c9af45 AM |
6792 | ;; All other cases of labels. |
6793 | ((and c-recognize-colon-labels ; nil for AWK and IDL, otherwise t. | |
0386b551 AM |
6794 | |
6795 | ;; A colon label must have something before the colon. | |
6796 | (not (eq (char-after) ?:)) | |
6797 | ||
6798 | ;; Check that we're not after a token that can't precede a label. | |
6799 | (or | |
6800 | ;; Trivially succeeds when there's no preceding token. | |
6801 | (if preceding-token-end | |
6802 | (<= preceding-token-end (point-min)) | |
6803 | (save-excursion | |
6804 | (c-backward-syntactic-ws) | |
6805 | (setq preceding-token-end (point)) | |
6806 | (bobp))) | |
6807 | ||
6808 | ;; Check if we're after a label, if we're after a closing | |
6809 | ;; paren that belong to statement, and with | |
6810 | ;; `c-label-prefix-re'. It's done in different order | |
6811 | ;; depending on `assume-markup' since the checks have | |
6812 | ;; different expensiveness. | |
6813 | (if assume-markup | |
6814 | (or | |
6815 | (eq (c-get-char-property (1- preceding-token-end) 'c-type) | |
6816 | 'c-decl-end) | |
6817 | ||
6818 | (save-excursion | |
6819 | (goto-char (1- preceding-token-end)) | |
6820 | (c-beginning-of-current-token) | |
51c9af45 AM |
6821 | (or (looking-at c-label-prefix-re) |
6822 | (looking-at c-block-stmt-1-key))) | |
0386b551 AM |
6823 | |
6824 | (and (eq (char-before preceding-token-end) ?\)) | |
6825 | (c-after-conditional))) | |
6826 | ||
6827 | (or | |
6828 | (save-excursion | |
6829 | (goto-char (1- preceding-token-end)) | |
6830 | (c-beginning-of-current-token) | |
51c9af45 AM |
6831 | (or (looking-at c-label-prefix-re) |
6832 | (looking-at c-block-stmt-1-key))) | |
0386b551 AM |
6833 | |
6834 | (cond | |
6835 | ((eq (char-before preceding-token-end) ?\)) | |
6836 | (c-after-conditional)) | |
6837 | ||
6838 | ((eq (char-before preceding-token-end) ?:) | |
6839 | ;; Might be after another label, so check it recursively. | |
51c9af45 AM |
6840 | (save-restriction |
6841 | (save-excursion | |
6842 | (goto-char (1- preceding-token-end)) | |
6843 | ;; Essentially the same as the | |
6844 | ;; `c-syntactic-re-search-forward' regexp below. | |
6845 | (setq macro-start | |
6846 | (save-excursion (and (c-beginning-of-macro) | |
6847 | (point)))) | |
6848 | (if macro-start (narrow-to-region macro-start (point-max))) | |
6849 | (c-syntactic-skip-backward "^-]:?;}=*/%&|,<>!@+" nil t) | |
6850 | ;; Note: the following should work instead of the | |
6851 | ;; narrow-to-region above. Investigate why not, | |
6852 | ;; sometime. ACM, 2006-03-31. | |
6853 | ;; (c-syntactic-skip-backward "^-]:?;}=*/%&|,<>!@+" | |
6854 | ;; macro-start t) | |
6855 | (let ((pte (point)) | |
6856 | ;; If the caller turned on recording for us, | |
6857 | ;; it shouldn't apply when we check the | |
6858 | ;; preceding label. | |
6859 | c-record-type-identifiers) | |
6860 | ;; A label can't start at a cpp directive. Check for | |
6861 | ;; this, since c-forward-syntactic-ws would foul up on it. | |
6862 | (unless (and c-opt-cpp-prefix (looking-at c-opt-cpp-prefix)) | |
6863 | (c-forward-syntactic-ws) | |
6864 | (c-forward-label nil pte start)))))))))) | |
6865 | ||
1379f2c5 | 6866 | ;; Point is still at the beginning of the possible label construct. |
b414f371 | 6867 | ;; |
51c9af45 AM |
6868 | ;; Check that the next nonsymbol token is ":", or that we're in one |
6869 | ;; of QT's "slots" declarations. Allow '(' for the sake of macro | |
6870 | ;; arguments. FIXME: Should build this regexp from the language | |
6871 | ;; constants. | |
1379f2c5 AM |
6872 | (cond |
6873 | ;; public: protected: private: | |
6874 | ((and | |
6875 | (c-major-mode-is 'c++-mode) | |
6876 | (search-forward-regexp | |
6877 | "\\=p\\(r\\(ivate\\|otected\\)\\|ublic\\)\\>[^_]" nil t) | |
6878 | (progn (backward-char) | |
6879 | (c-forward-syntactic-ws limit) | |
6880 | (looking-at ":\\([^:]\\|\\'\\)"))) ; A single colon. | |
6881 | (forward-char) | |
6882 | (setq label-type t)) | |
6883 | ;; QT double keyword like "protected slots:" or goto target. | |
6884 | ((progn (goto-char start) nil)) | |
6885 | ((when (c-syntactic-re-search-forward | |
6886 | "[ \t\n[:?;{=*/%&|,<>!@+-]" limit t t) ; not at EOB | |
6887 | (backward-char) | |
6888 | (setq label-end (point)) | |
6889 | (setq qt-symbol-idx | |
6890 | (and (c-major-mode-is 'c++-mode) | |
6891 | (string-match | |
6892 | "\\(p\\(r\\(ivate\\|otected\\)\\|ublic\\)\\|more\\)\\>" | |
b414f371 | 6893 | (buffer-substring start (point))))) |
1379f2c5 AM |
6894 | (c-forward-syntactic-ws limit) |
6895 | (cond | |
6896 | ((looking-at ":\\([^:]\\|\\'\\)") ; A single colon. | |
6897 | (forward-char) | |
6898 | (setq label-type | |
f412a567 AM |
6899 | (if (or (string= "signals" ; Special QT macro |
6900 | (setq kwd (buffer-substring-no-properties start label-end))) | |
6901 | (string= "Q_SIGNALS" kwd)) | |
1379f2c5 AM |
6902 | 'qt-1kwd-colon |
6903 | 'goto-target))) | |
6904 | ((and qt-symbol-idx | |
f412a567 | 6905 | (search-forward-regexp "\\=\\(slots\\|Q_SLOTS\\)\\>" limit t) |
1379f2c5 AM |
6906 | (progn (c-forward-syntactic-ws limit) |
6907 | (looking-at ":\\([^:]\\|\\'\\)"))) ; A single colon | |
6908 | (forward-char) | |
6909 | (setq label-type 'qt-2kwds-colon))))))) | |
0386b551 AM |
6910 | |
6911 | (save-restriction | |
6912 | (narrow-to-region start (point)) | |
6913 | ||
6914 | ;; Check that `c-nonlabel-token-key' doesn't match anywhere. | |
6915 | (catch 'check-label | |
6916 | (goto-char start) | |
6917 | (while (progn | |
6918 | (when (looking-at c-nonlabel-token-key) | |
6919 | (goto-char start) | |
1379f2c5 | 6920 | (setq label-type nil) |
0386b551 AM |
6921 | (throw 'check-label nil)) |
6922 | (and (c-safe (c-forward-sexp) | |
6923 | (c-forward-syntactic-ws) | |
6924 | t) | |
6925 | (not (eobp))))) | |
6926 | ||
6927 | ;; Record the identifiers in the label for fontification, unless | |
6928 | ;; it begins with `c-label-kwds' in which case the following | |
6929 | ;; identifiers are part of a (constant) expression that | |
6930 | ;; shouldn't be fontified. | |
6931 | (when (and c-record-type-identifiers | |
6932 | (progn (goto-char start) | |
6933 | (not (looking-at c-label-kwds-regexp)))) | |
6934 | (while (c-syntactic-re-search-forward c-symbol-key nil t) | |
6935 | (c-record-ref-id (cons (match-beginning 0) | |
6936 | (match-end 0))))) | |
6937 | ||
6938 | (c-put-c-type-property (1- (point-max)) 'c-decl-end) | |
82ba65cf | 6939 | (goto-char (point-max))))) |
0386b551 AM |
6940 | |
6941 | (t | |
6942 | ;; Not a label. | |
1379f2c5 AM |
6943 | (goto-char start))) |
6944 | label-type)) | |
0386b551 AM |
6945 | |
6946 | (defun c-forward-objc-directive () | |
6947 | ;; Assuming the point is at the beginning of a token, try to move | |
6948 | ;; forward to the end of the Objective-C directive that starts | |
6949 | ;; there. Return t if a directive was fully recognized, otherwise | |
6950 | ;; the point is moved as far as one could be successfully parsed and | |
6951 | ;; nil is returned. | |
6952 | ;; | |
6953 | ;; This function records identifier ranges on | |
6954 | ;; `c-record-type-identifiers' and `c-record-ref-identifiers' if | |
6955 | ;; `c-record-type-identifiers' is non-nil. | |
6956 | ;; | |
6957 | ;; This function might do hidden buffer changes. | |
6958 | ||
6959 | (let ((start (point)) | |
6960 | start-char | |
6961 | (c-promote-possible-types t) | |
6962 | ;; Turn off recognition of angle bracket arglists while parsing | |
6963 | ;; types here since the protocol reference list might then be | |
6964 | ;; considered part of the preceding name or superclass-name. | |
6965 | c-recognize-<>-arglists) | |
6966 | ||
6967 | (if (or | |
6968 | (when (looking-at | |
6969 | (eval-when-compile | |
6970 | (c-make-keywords-re t | |
6971 | (append (c-lang-const c-protection-kwds objc) | |
6972 | '("@end")) | |
6973 | 'objc-mode))) | |
6974 | (goto-char (match-end 1)) | |
6975 | t) | |
6976 | ||
6977 | (and | |
6978 | (looking-at | |
6979 | (eval-when-compile | |
6980 | (c-make-keywords-re t | |
6981 | '("@interface" "@implementation" "@protocol") | |
6982 | 'objc-mode))) | |
6983 | ||
6984 | ;; Handle the name of the class itself. | |
6985 | (progn | |
cb694ab7 AM |
6986 | ; (c-forward-token-2) ; 2006/1/13 This doesn't move if the token's |
6987 | ; at EOB. | |
6988 | (goto-char (match-end 0)) | |
6989 | (c-skip-ws-forward) | |
0386b551 AM |
6990 | (c-forward-type)) |
6991 | ||
6992 | (catch 'break | |
6993 | ;; Look for ": superclass-name" or "( category-name )". | |
6994 | (when (looking-at "[:\(]") | |
6995 | (setq start-char (char-after)) | |
6996 | (forward-char) | |
6997 | (c-forward-syntactic-ws) | |
6998 | (unless (c-forward-type) (throw 'break nil)) | |
6999 | (when (eq start-char ?\() | |
7000 | (unless (eq (char-after) ?\)) (throw 'break nil)) | |
7001 | (forward-char) | |
7002 | (c-forward-syntactic-ws))) | |
7003 | ||
7004 | ;; Look for a protocol reference list. | |
7005 | (if (eq (char-after) ?<) | |
7006 | (let ((c-recognize-<>-arglists t) | |
7007 | (c-parse-and-markup-<>-arglists t) | |
7008 | c-restricted-<>-arglists) | |
7009 | (c-forward-<>-arglist t)) | |
7010 | t)))) | |
7011 | ||
7012 | (progn | |
7013 | (c-backward-syntactic-ws) | |
7014 | (c-clear-c-type-property start (1- (point)) 'c-decl-end) | |
7015 | (c-put-c-type-property (1- (point)) 'c-decl-end) | |
7016 | t) | |
7017 | ||
7018 | (c-clear-c-type-property start (point) 'c-decl-end) | |
7019 | nil))) | |
7020 | ||
785eecbb RS |
7021 | (defun c-beginning-of-inheritance-list (&optional lim) |
7022 | ;; Go to the first non-whitespace after the colon that starts a | |
7023 | ;; multiple inheritance introduction. Optional LIM is the farthest | |
7024 | ;; back we should search. | |
0386b551 AM |
7025 | ;; |
7026 | ;; This function might do hidden buffer changes. | |
7027 | (c-with-syntax-table c++-template-syntax-table | |
7028 | (c-backward-token-2 0 t lim) | |
7029 | (while (and (or (looking-at c-symbol-start) | |
7030 | (looking-at "[<,]\\|::")) | |
7031 | (zerop (c-backward-token-2 1 t lim)))))) | |
785eecbb | 7032 | |
785eecbb RS |
7033 | (defun c-in-method-def-p () |
7034 | ;; Return nil if we aren't in a method definition, otherwise the | |
7035 | ;; position of the initial [+-]. | |
0386b551 AM |
7036 | ;; |
7037 | ;; This function might do hidden buffer changes. | |
785eecbb RS |
7038 | (save-excursion |
7039 | (beginning-of-line) | |
a66cd3ee MS |
7040 | (and c-opt-method-key |
7041 | (looking-at c-opt-method-key) | |
785eecbb RS |
7042 | (point)) |
7043 | )) | |
7044 | ||
a66cd3ee MS |
7045 | ;; Contributed by Kevin Ryde <user42@zip.com.au>. |
7046 | (defun c-in-gcc-asm-p () | |
7047 | ;; Return non-nil if point is within a gcc \"asm\" block. | |
7048 | ;; | |
7049 | ;; This should be called with point inside an argument list. | |
7050 | ;; | |
7051 | ;; Only one level of enclosing parentheses is considered, so for | |
7052 | ;; instance `nil' is returned when in a function call within an asm | |
7053 | ;; operand. | |
0386b551 AM |
7054 | ;; |
7055 | ;; This function might do hidden buffer changes. | |
a66cd3ee MS |
7056 | |
7057 | (and c-opt-asm-stmt-key | |
7058 | (save-excursion | |
7059 | (beginning-of-line) | |
7060 | (backward-up-list 1) | |
7061 | (c-beginning-of-statement-1 (point-min) nil t) | |
7062 | (looking-at c-opt-asm-stmt-key)))) | |
7063 | ||
abb7e5cf | 7064 | (defun c-at-toplevel-p () |
a85fd6da AM |
7065 | "Return a determination as to whether point is \"at the top level\". |
7066 | Informally, \"at the top level\" is anywhere where you can write | |
7067 | a function. | |
7068 | ||
7069 | More precisely, being at the top-level means that point is either | |
7070 | outside any enclosing block (such as a function definition), or | |
7071 | directly inside a class, namespace or other block that contains | |
7072 | another declaration level. | |
abb7e5cf SM |
7073 | |
7074 | If point is not at the top-level (e.g. it is inside a method | |
7075 | definition), then nil is returned. Otherwise, if point is at a | |
7076 | top-level not enclosed within a class definition, t is returned. | |
7077 | Otherwise, a 2-vector is returned where the zeroth element is the | |
7078 | buffer position of the start of the class declaration, and the first | |
7079 | element is the buffer position of the enclosing class's opening | |
0386b551 AM |
7080 | brace. |
7081 | ||
7082 | Note that this function might do hidden buffer changes. See the | |
7083 | comment at the start of cc-engine.el for more info." | |
a66cd3ee MS |
7084 | (let ((paren-state (c-parse-state))) |
7085 | (or (not (c-most-enclosing-brace paren-state)) | |
7086 | (c-search-uplist-for-classkey paren-state)))) | |
7087 | ||
d9e94c22 | 7088 | (defun c-just-after-func-arglist-p (&optional lim) |
0386b551 AM |
7089 | ;; Return non-nil if the point is in the region after the argument |
7090 | ;; list of a function and its opening brace (or semicolon in case it | |
7091 | ;; got no body). If there are K&R style argument declarations in | |
7092 | ;; that region, the point has to be inside the first one for this | |
7093 | ;; function to recognize it. | |
a66cd3ee | 7094 | ;; |
0386b551 AM |
7095 | ;; If successful, the point is moved to the first token after the |
7096 | ;; function header (see `c-forward-decl-or-cast-1' for details) and | |
7097 | ;; the position of the opening paren of the function arglist is | |
7098 | ;; returned. | |
7099 | ;; | |
7100 | ;; The point is clobbered if not successful. | |
7101 | ;; | |
7102 | ;; LIM is used as bound for backward buffer searches. | |
7103 | ;; | |
7104 | ;; This function might do hidden buffer changes. | |
7105 | ||
7106 | (let ((beg (point)) end id-start) | |
7107 | (and | |
7108 | (eq (c-beginning-of-statement-1 lim) 'same) | |
7109 | ||
7110 | (not (or (c-major-mode-is 'objc-mode) | |
7111 | (c-forward-objc-directive))) | |
7112 | ||
7113 | (setq id-start | |
7114 | (car-safe (c-forward-decl-or-cast-1 (c-point 'bosws) nil nil))) | |
7115 | (< id-start beg) | |
7116 | ||
7117 | ;; There should not be a '=' or ',' between beg and the | |
7118 | ;; start of the declaration since that means we were in the | |
7119 | ;; "expression part" of the declaration. | |
7120 | (or (> (point) beg) | |
7121 | (not (looking-at "[=,]"))) | |
7122 | ||
7123 | (save-excursion | |
7124 | ;; Check that there's an arglist paren in the | |
7125 | ;; declaration. | |
7126 | (goto-char id-start) | |
7127 | (cond ((eq (char-after) ?\() | |
7128 | ;; The declarator is a paren expression, so skip past it | |
7129 | ;; so that we don't get stuck on that instead of the | |
7130 | ;; function arglist. | |
7131 | (c-forward-sexp)) | |
51c9af45 AM |
7132 | ((and c-opt-op-identifier-prefix |
7133 | (looking-at c-opt-op-identifier-prefix)) | |
0386b551 AM |
7134 | ;; Don't trip up on "operator ()". |
7135 | (c-forward-token-2 2 t))) | |
7136 | (and (< (point) beg) | |
7137 | (c-syntactic-re-search-forward "(" beg t t) | |
7138 | (1- (point))))))) | |
785eecbb | 7139 | |
a66cd3ee MS |
7140 | (defun c-in-knr-argdecl (&optional lim) |
7141 | ;; Return the position of the first argument declaration if point is | |
7142 | ;; inside a K&R style argument declaration list, nil otherwise. | |
7143 | ;; `c-recognize-knr-p' is not checked. If LIM is non-nil, it's a | |
7144 | ;; position that bounds the backward search for the argument list. | |
7145 | ;; | |
9cf17ef1 AM |
7146 | ;; Point must be within a possible K&R region, e.g. just before a top-level |
7147 | ;; "{". It must be outside of parens and brackets. The test can return | |
7148 | ;; false positives otherwise. | |
0386b551 AM |
7149 | ;; |
7150 | ;; This function might do hidden buffer changes. | |
d9e94c22 | 7151 | |
a66cd3ee MS |
7152 | (save-excursion |
7153 | (save-restriction | |
9cf17ef1 AM |
7154 | ;; If we're in a macro, our search range is restricted to it. Narrow to |
7155 | ;; the searchable range. | |
7156 | (let* ((macro-start (c-query-macro-start)) | |
7157 | (lim (max (or lim (point-min)) (or macro-start (point-min)))) | |
6b6481ed AM |
7158 | before-lparen after-rparen |
7159 | (pp-count-out 20)) ; Max number of paren/brace constructs before we give up | |
9cf17ef1 AM |
7160 | (narrow-to-region lim (c-point 'eol)) |
7161 | ||
7162 | ;; Search backwards for the defun's argument list. We give up if we | |
7163 | ;; encounter a "}" (end of a previous defun) or BOB. | |
7164 | ;; | |
7165 | ;; The criterion for a paren structure being the arg list is: | |
7166 | ;; o - there is non-WS stuff after it but before any "{"; AND | |
7167 | ;; o - the token after it isn't a ";" AND | |
7168 | ;; o - it is preceded by either an identifier (the function name) or | |
7169 | ;; a macro expansion like "DEFUN (...)"; AND | |
7170 | ;; o - its content is a non-empty comma-separated list of identifiers | |
7171 | ;; (an empty arg list won't have a knr region). | |
7172 | ;; | |
7173 | ;; The following snippet illustrates these rules: | |
7174 | ;; int foo (bar, baz, yuk) | |
7175 | ;; int bar [] ; | |
7176 | ;; int (*baz) (my_type) ; | |
7177 | ;; int (*) (void) (*yuk) (void) ; | |
7178 | ;; { | |
7179 | ||
7180 | (catch 'knr | |
6b6481ed AM |
7181 | (while (> pp-count-out 0) ; go back one paren/bracket pair each time. |
7182 | (setq pp-count-out (1- pp-count-out)) | |
10489fcb | 7183 | (c-syntactic-skip-backward "^)]}") |
9cf17ef1 AM |
7184 | (cond ((eq (char-before) ?\)) |
7185 | (setq after-rparen (point))) | |
10489fcb AM |
7186 | ((eq (char-before) ?\]) |
7187 | (setq after-rparen nil)) | |
7188 | (t ; either } (hit previous defun) or no more parens/brackets | |
7189 | (throw 'knr nil))) | |
9cf17ef1 AM |
7190 | |
7191 | (if after-rparen | |
7192 | ;; We're inside a paren. Could it be our argument list....? | |
7193 | (if | |
7194 | (and | |
7195 | (progn | |
7196 | (goto-char after-rparen) | |
7197 | (unless (c-go-list-backward) (throw 'knr nil)) ; | |
7198 | ;; FIXME!!! What about macros between the parens? 2007/01/20 | |
7199 | (setq before-lparen (point))) | |
d9e94c22 | 7200 | |
9cf17ef1 AM |
7201 | ;; It can't be the arg list if next token is ; or { |
7202 | (progn (goto-char after-rparen) | |
7203 | (c-forward-syntactic-ws) | |
7204 | (not (memq (char-after) '(?\; ?\{)))) | |
d9e94c22 | 7205 | |
9cf17ef1 AM |
7206 | ;; Is the thing preceding the list an identifier (the |
7207 | ;; function name), or a macro expansion? | |
7208 | (progn | |
7209 | (goto-char before-lparen) | |
7210 | (eq (c-backward-token-2) 0) | |
7211 | (or (c-on-identifier) | |
7212 | (and (eq (char-after) ?\)) | |
7213 | (c-go-up-list-backward) | |
7214 | (eq (c-backward-token-2) 0) | |
7215 | (c-on-identifier)))) | |
7216 | ||
7217 | ;; Have we got a non-empty list of comma-separated | |
7218 | ;; identifiers? | |
7219 | (progn | |
7220 | (goto-char before-lparen) | |
7221 | (c-forward-token-2) ; to first token inside parens | |
7222 | (and | |
7223 | (c-on-identifier) | |
7224 | (c-forward-token-2) | |
7225 | (catch 'id-list | |
7226 | (while (eq (char-after) ?\,) | |
7227 | (c-forward-token-2) | |
7228 | (unless (c-on-identifier) (throw 'id-list nil)) | |
7229 | (c-forward-token-2)) | |
7230 | (eq (char-after) ?\)))))) | |
7231 | ||
7232 | ;; ...Yes. We've identified the function's argument list. | |
7233 | (throw 'knr | |
7234 | (progn (goto-char after-rparen) | |
7235 | (c-forward-syntactic-ws) | |
7236 | (point))) | |
7237 | ||
7238 | ;; ...No. The current parens aren't the function's arg list. | |
7239 | (goto-char before-lparen)) | |
7240 | ||
7241 | (or (c-go-list-backward) ; backwards over [ .... ] | |
7242 | (throw 'knr nil))))))))) | |
785eecbb RS |
7243 | |
7244 | (defun c-skip-conditional () | |
7245 | ;; skip forward over conditional at point, including any predicate | |
7246 | ;; statements in parentheses. No error checking is performed. | |
0386b551 AM |
7247 | ;; |
7248 | ;; This function might do hidden buffer changes. | |
0ec8351b BW |
7249 | (c-forward-sexp (cond |
7250 | ;; else if() | |
a66cd3ee MS |
7251 | ((looking-at (concat "\\<else" |
7252 | "\\([ \t\n]\\|\\\\\n\\)+" | |
7253 | "if\\>\\([^_]\\|$\\)")) | |
7254 | 3) | |
0ec8351b | 7255 | ;; do, else, try, finally |
a66cd3ee MS |
7256 | ((looking-at (concat "\\<\\(" |
7257 | "do\\|else\\|try\\|finally" | |
7258 | "\\)\\>\\([^_]\\|$\\)")) | |
130c507e | 7259 | 1) |
ce8c7486 | 7260 | ;; for, if, while, switch, catch, synchronized, foreach |
0ec8351b | 7261 | (t 2)))) |
785eecbb | 7262 | |
a66cd3ee MS |
7263 | (defun c-after-conditional (&optional lim) |
7264 | ;; If looking at the token after a conditional then return the | |
7265 | ;; position of its start, otherwise return nil. | |
0386b551 AM |
7266 | ;; |
7267 | ;; This function might do hidden buffer changes. | |
a66cd3ee | 7268 | (save-excursion |
d9e94c22 | 7269 | (and (zerop (c-backward-token-2 1 t lim)) |
a66cd3ee MS |
7270 | (or (looking-at c-block-stmt-1-key) |
7271 | (and (eq (char-after) ?\() | |
d9e94c22 | 7272 | (zerop (c-backward-token-2 1 t lim)) |
a66cd3ee MS |
7273 | (looking-at c-block-stmt-2-key))) |
7274 | (point)))) | |
7275 | ||
0386b551 AM |
7276 | (defun c-after-special-operator-id (&optional lim) |
7277 | ;; If the point is after an operator identifier that isn't handled | |
7278 | ;; like an ordinary symbol (i.e. like "operator =" in C++) then the | |
7279 | ;; position of the start of that identifier is returned. nil is | |
7280 | ;; returned otherwise. The point may be anywhere in the syntactic | |
7281 | ;; whitespace after the last token of the operator identifier. | |
7282 | ;; | |
7283 | ;; This function might do hidden buffer changes. | |
7284 | (save-excursion | |
7285 | (and c-overloadable-operators-regexp | |
7286 | (zerop (c-backward-token-2 1 nil lim)) | |
7287 | (looking-at c-overloadable-operators-regexp) | |
51c9af45 | 7288 | (or (not c-opt-op-identifier-prefix) |
0386b551 AM |
7289 | (and |
7290 | (zerop (c-backward-token-2 1 nil lim)) | |
51c9af45 | 7291 | (looking-at c-opt-op-identifier-prefix))) |
0386b551 AM |
7292 | (point)))) |
7293 | ||
a66cd3ee MS |
7294 | (defsubst c-backward-to-block-anchor (&optional lim) |
7295 | ;; Assuming point is at a brace that opens a statement block of some | |
7296 | ;; kind, move to the proper anchor point for that block. It might | |
7297 | ;; need to be adjusted further by c-add-stmt-syntax, but the | |
7298 | ;; position at return is suitable as start position for that | |
7299 | ;; function. | |
0386b551 AM |
7300 | ;; |
7301 | ;; This function might do hidden buffer changes. | |
a66cd3ee MS |
7302 | (unless (= (point) (c-point 'boi)) |
7303 | (let ((start (c-after-conditional lim))) | |
7304 | (if start | |
7305 | (goto-char start))))) | |
7306 | ||
037558bf | 7307 | (defsubst c-backward-to-decl-anchor (&optional lim) |
a66cd3ee MS |
7308 | ;; Assuming point is at a brace that opens the block of a top level |
7309 | ;; declaration of some kind, move to the proper anchor point for | |
7310 | ;; that block. | |
0386b551 AM |
7311 | ;; |
7312 | ;; This function might do hidden buffer changes. | |
a66cd3ee | 7313 | (unless (= (point) (c-point 'boi)) |
037558bf | 7314 | (c-beginning-of-statement-1 lim))) |
a66cd3ee | 7315 | |
ff959bab | 7316 | (defun c-search-decl-header-end () |
a66cd3ee MS |
7317 | ;; Search forward for the end of the "header" of the current |
7318 | ;; declaration. That's the position where the definition body | |
7319 | ;; starts, or the first variable initializer, or the ending | |
7320 | ;; semicolon. I.e. search forward for the closest following | |
7321 | ;; (syntactically relevant) '{', '=' or ';' token. Point is left | |
7322 | ;; _after_ the first found token, or at point-max if none is found. | |
0386b551 AM |
7323 | ;; |
7324 | ;; This function might do hidden buffer changes. | |
ff959bab MS |
7325 | |
7326 | (let ((base (point))) | |
7327 | (if (c-major-mode-is 'c++-mode) | |
7328 | ||
7329 | ;; In C++ we need to take special care to handle operator | |
7330 | ;; tokens and those pesky template brackets. | |
7331 | (while (and | |
7332 | (c-syntactic-re-search-forward "[;{<=]" nil 'move t t) | |
7333 | (or | |
7334 | (c-end-of-current-token base) | |
7335 | ;; Handle operator identifiers, i.e. ignore any | |
7336 | ;; operator token preceded by "operator". | |
7337 | (save-excursion | |
7338 | (and (c-safe (c-backward-sexp) t) | |
51c9af45 | 7339 | (looking-at c-opt-op-identifier-prefix))) |
ff959bab MS |
7340 | (and (eq (char-before) ?<) |
7341 | (c-with-syntax-table c++-template-syntax-table | |
7342 | (if (c-safe (goto-char (c-up-list-forward (point)))) | |
7343 | t | |
7344 | (goto-char (point-max)) | |
7345 | nil))))) | |
7346 | (setq base (point))) | |
7347 | ||
7348 | (while (and | |
7349 | (c-syntactic-re-search-forward "[;{=]" nil 'move t t) | |
7350 | (c-end-of-current-token base)) | |
7351 | (setq base (point)))))) | |
a66cd3ee MS |
7352 | |
7353 | (defun c-beginning-of-decl-1 (&optional lim) | |
7354 | ;; Go to the beginning of the current declaration, or the beginning | |
7355 | ;; of the previous one if already at the start of it. Point won't | |
0386b551 | 7356 | ;; be moved out of any surrounding paren. Return a cons cell of the |
a66cd3ee MS |
7357 | ;; form (MOVE . KNR-POS). MOVE is like the return value from |
7358 | ;; `c-beginning-of-statement-1'. If point skipped over some K&R | |
7359 | ;; style argument declarations (and they are to be recognized) then | |
7360 | ;; KNR-POS is set to the start of the first such argument | |
7361 | ;; declaration, otherwise KNR-POS is nil. If LIM is non-nil, it's a | |
7362 | ;; position that bounds the backward search. | |
7363 | ;; | |
7364 | ;; NB: Cases where the declaration continues after the block, as in | |
7365 | ;; "struct foo { ... } bar;", are currently recognized as two | |
7366 | ;; declarations, e.g. "struct foo { ... }" and "bar;" in this case. | |
0386b551 AM |
7367 | ;; |
7368 | ;; This function might do hidden buffer changes. | |
a66cd3ee MS |
7369 | (catch 'return |
7370 | (let* ((start (point)) | |
d9e94c22 | 7371 | (last-stmt-start (point)) |
0386b551 | 7372 | (move (c-beginning-of-statement-1 lim nil t))) |
a66cd3ee | 7373 | |
a66cd3ee MS |
7374 | ;; `c-beginning-of-statement-1' stops at a block start, but we |
7375 | ;; want to continue if the block doesn't begin a top level | |
2a15eb73 MS |
7376 | ;; construct, i.e. if it isn't preceded by ';', '}', ':', bob, |
7377 | ;; or an open paren. | |
d9e94c22 | 7378 | (let ((beg (point)) tentative-move) |
51c9af45 AM |
7379 | ;; Go back one "statement" each time round the loop until we're just |
7380 | ;; after a ;, }, or :, or at BOB or the start of a macro or start of | |
7381 | ;; an ObjC method. This will move over a multiple declaration whose | |
7382 | ;; components are comma separated. | |
d9e94c22 MS |
7383 | (while (and |
7384 | ;; Must check with c-opt-method-key in ObjC mode. | |
7385 | (not (and c-opt-method-key | |
7386 | (looking-at c-opt-method-key))) | |
7387 | (/= last-stmt-start (point)) | |
7388 | (progn | |
7389 | (c-backward-syntactic-ws lim) | |
7390 | (not (memq (char-before) '(?\; ?} ?: nil)))) | |
2a15eb73 MS |
7391 | (save-excursion |
7392 | (backward-char) | |
7393 | (not (looking-at "\\s("))) | |
d9e94c22 MS |
7394 | ;; Check that we don't move from the first thing in a |
7395 | ;; macro to its header. | |
7396 | (not (eq (setq tentative-move | |
0386b551 | 7397 | (c-beginning-of-statement-1 lim nil t)) |
d9e94c22 MS |
7398 | 'macro))) |
7399 | (setq last-stmt-start beg | |
7400 | beg (point) | |
7401 | move tentative-move)) | |
7402 | (goto-char beg)) | |
7403 | ||
7404 | (when c-recognize-knr-p | |
7405 | (let ((fallback-pos (point)) knr-argdecl-start) | |
7406 | ;; Handle K&R argdecls. Back up after the "statement" jumped | |
7407 | ;; over by `c-beginning-of-statement-1', unless it was the | |
7408 | ;; function body, in which case we're sitting on the opening | |
7409 | ;; brace now. Then test if we're in a K&R argdecl region and | |
7410 | ;; that we started at the other side of the first argdecl in | |
7411 | ;; it. | |
7412 | (unless (eq (char-after) ?{) | |
7413 | (goto-char last-stmt-start)) | |
7414 | (if (and (setq knr-argdecl-start (c-in-knr-argdecl lim)) | |
7415 | (< knr-argdecl-start start) | |
7416 | (progn | |
7417 | (goto-char knr-argdecl-start) | |
0386b551 | 7418 | (not (eq (c-beginning-of-statement-1 lim nil t) 'macro)))) |
d9e94c22 MS |
7419 | (throw 'return |
7420 | (cons (if (eq (char-after fallback-pos) ?{) | |
7421 | 'previous | |
7422 | 'same) | |
7423 | knr-argdecl-start)) | |
7424 | (goto-char fallback-pos)))) | |
7425 | ||
51c9af45 AM |
7426 | ;; `c-beginning-of-statement-1' counts each brace block as a separate |
7427 | ;; statement, so the result will be 'previous if we've moved over any. | |
7428 | ;; So change our result back to 'same if necessary. | |
7429 | ;; | |
7430 | ;; If they were brace list initializers we might not have moved over a | |
7431 | ;; declaration boundary though, so change it to 'same if we've moved | |
7432 | ;; past a '=' before '{', but not ';'. (This ought to be integrated | |
7433 | ;; into `c-beginning-of-statement-1', so we avoid this extra pass which | |
7434 | ;; potentially can search over a large amount of text.). Take special | |
7435 | ;; pains not to get mislead by C++'s "operator=", and the like. | |
d9e94c22 MS |
7436 | (if (and (eq move 'previous) |
7437 | (c-with-syntax-table (if (c-major-mode-is 'c++-mode) | |
7438 | c++-template-syntax-table | |
7439 | (syntax-table)) | |
7440 | (save-excursion | |
51c9af45 AM |
7441 | (and |
7442 | (progn | |
7443 | (while ; keep going back to "[;={"s until we either find | |
7444 | ; no more, or get to one which isn't an "operator =" | |
7445 | (and (c-syntactic-re-search-forward "[;={]" start t t t) | |
7446 | (eq (char-before) ?=) | |
7447 | c-overloadable-operators-regexp | |
7448 | c-opt-op-identifier-prefix | |
7449 | (save-excursion | |
7450 | (eq (c-backward-token-2) 0) | |
7451 | (looking-at c-overloadable-operators-regexp) | |
7452 | (eq (c-backward-token-2) 0) | |
7453 | (looking-at c-opt-op-identifier-prefix)))) | |
7454 | (eq (char-before) ?=)) | |
7455 | (c-syntactic-re-search-forward "[;{]" start t t) | |
7456 | (eq (char-before) ?{) | |
7457 | (c-safe (goto-char (c-up-list-forward (point))) t) | |
7458 | (not (c-syntactic-re-search-forward ";" start t t)))))) | |
d9e94c22 MS |
7459 | (cons 'same nil) |
7460 | (cons move nil))))) | |
a66cd3ee MS |
7461 | |
7462 | (defun c-end-of-decl-1 () | |
7463 | ;; Assuming point is at the start of a declaration (as detected by | |
7464 | ;; e.g. `c-beginning-of-decl-1'), go to the end of it. Unlike | |
7465 | ;; `c-beginning-of-decl-1', this function handles the case when a | |
7466 | ;; block is followed by identifiers in e.g. struct declarations in C | |
7467 | ;; or C++. If a proper end was found then t is returned, otherwise | |
7468 | ;; point is moved as far as possible within the current sexp and nil | |
7469 | ;; is returned. This function doesn't handle macros; use | |
7470 | ;; `c-end-of-macro' instead in those cases. | |
0386b551 AM |
7471 | ;; |
7472 | ;; This function might do hidden buffer changes. | |
ce8c7486 | 7473 | (let ((start (point)) |
a66cd3ee MS |
7474 | (decl-syntax-table (if (c-major-mode-is 'c++-mode) |
7475 | c++-template-syntax-table | |
7476 | (syntax-table)))) | |
7477 | (catch 'return | |
7478 | (c-search-decl-header-end) | |
7479 | ||
7480 | (when (and c-recognize-knr-p | |
7481 | (eq (char-before) ?\;) | |
7482 | (c-in-knr-argdecl start)) | |
7483 | ;; Stopped at the ';' in a K&R argdecl section which is | |
7484 | ;; detected using the same criteria as in | |
7485 | ;; `c-beginning-of-decl-1'. Move to the following block | |
7486 | ;; start. | |
d9e94c22 | 7487 | (c-syntactic-re-search-forward "{" nil 'move t)) |
a66cd3ee MS |
7488 | |
7489 | (when (eq (char-before) ?{) | |
7490 | ;; Encountered a block in the declaration. Jump over it. | |
7491 | (condition-case nil | |
7492 | (goto-char (c-up-list-forward (point))) | |
d9e94c22 MS |
7493 | (error (goto-char (point-max)) |
7494 | (throw 'return nil))) | |
a66cd3ee MS |
7495 | (if (or (not c-opt-block-decls-with-vars-key) |
7496 | (save-excursion | |
7497 | (c-with-syntax-table decl-syntax-table | |
7498 | (let ((lim (point))) | |
7499 | (goto-char start) | |
b3cf7e18 MS |
7500 | (not (and |
7501 | ;; Check for `c-opt-block-decls-with-vars-key' | |
7502 | ;; before the first paren. | |
7503 | (c-syntactic-re-search-forward | |
d9e94c22 | 7504 | (concat "[;=\(\[{]\\|\\(" |
b3cf7e18 MS |
7505 | c-opt-block-decls-with-vars-key |
7506 | "\\)") | |
d9e94c22 | 7507 | lim t t t) |
b3cf7e18 MS |
7508 | (match-beginning 1) |
7509 | (not (eq (char-before) ?_)) | |
d9e94c22 MS |
7510 | ;; Check that the first following paren is |
7511 | ;; the block. | |
7512 | (c-syntactic-re-search-forward "[;=\(\[{]" | |
7513 | lim t t t) | |
b3cf7e18 | 7514 | (eq (char-before) ?{))))))) |
a66cd3ee MS |
7515 | ;; The declaration doesn't have any of the |
7516 | ;; `c-opt-block-decls-with-vars' keywords in the | |
7517 | ;; beginning, so it ends here at the end of the block. | |
7518 | (throw 'return t))) | |
7519 | ||
7520 | (c-with-syntax-table decl-syntax-table | |
7521 | (while (progn | |
7522 | (if (eq (char-before) ?\;) | |
7523 | (throw 'return t)) | |
d9e94c22 | 7524 | (c-syntactic-re-search-forward ";" nil 'move t)))) |
a66cd3ee | 7525 | nil))) |
ce8c7486 | 7526 | |
0386b551 AM |
7527 | (defun c-looking-at-decl-block (containing-sexp goto-start &optional limit) |
7528 | ;; Assuming the point is at an open brace, check if it starts a | |
7529 | ;; block that contains another declaration level, i.e. that isn't a | |
7530 | ;; statement block or a brace list, and if so return non-nil. | |
7531 | ;; | |
7532 | ;; If the check is successful, the return value is the start of the | |
7533 | ;; keyword that tells what kind of construct it is, i.e. typically | |
7534 | ;; what `c-decl-block-key' matched. Also, if GOTO-START is set then | |
7535 | ;; the point will be at the start of the construct, before any | |
7536 | ;; leading specifiers, otherwise it's at the returned position. | |
7537 | ;; | |
7538 | ;; The point is clobbered if the check is unsuccessful. | |
7539 | ;; | |
7540 | ;; CONTAINING-SEXP is the position of the open of the surrounding | |
7541 | ;; paren, or nil if none. | |
7542 | ;; | |
7543 | ;; The optional LIMIT limits the backward search for the start of | |
7544 | ;; the construct. It's assumed to be at a syntactically relevant | |
7545 | ;; position. | |
7546 | ;; | |
7547 | ;; If any template arglists are found in the searched region before | |
7548 | ;; the open brace, they get marked with paren syntax. | |
7549 | ;; | |
7550 | ;; This function might do hidden buffer changes. | |
7551 | ||
7552 | (let ((open-brace (point)) kwd-start first-specifier-pos) | |
7553 | (c-syntactic-skip-backward c-block-prefix-charset limit t) | |
7554 | ||
7555 | (when (and c-recognize-<>-arglists | |
7556 | (eq (char-before) ?>)) | |
7557 | ;; Could be at the end of a template arglist. | |
7558 | (let ((c-parse-and-markup-<>-arglists t) | |
7559 | (c-disallow-comma-in-<>-arglists | |
7560 | (and containing-sexp | |
7561 | (not (eq (char-after containing-sexp) ?{))))) | |
7562 | (while (and | |
7563 | (c-backward-<>-arglist nil limit) | |
7564 | (progn | |
7565 | (c-syntactic-skip-backward c-block-prefix-charset limit t) | |
7566 | (eq (char-before) ?>)))))) | |
7567 | ||
7568 | ;; Note: Can't get bogus hits inside template arglists below since they | |
7569 | ;; have gotten paren syntax above. | |
7570 | (when (and | |
7571 | ;; If `goto-start' is set we begin by searching for the | |
7572 | ;; first possible position of a leading specifier list. | |
7573 | ;; The `c-decl-block-key' search continues from there since | |
7574 | ;; we know it can't match earlier. | |
7575 | (if goto-start | |
7576 | (when (c-syntactic-re-search-forward c-symbol-start | |
7577 | open-brace t t) | |
7578 | (goto-char (setq first-specifier-pos (match-beginning 0))) | |
7579 | t) | |
7580 | t) | |
7581 | ||
7582 | (cond | |
7583 | ((c-syntactic-re-search-forward c-decl-block-key open-brace t t t) | |
7584 | (goto-char (setq kwd-start (match-beginning 0))) | |
7585 | (or | |
7586 | ||
7587 | ;; Found a keyword that can't be a type? | |
7588 | (match-beginning 1) | |
7589 | ||
7590 | ;; Can be a type too, in which case it's the return type of a | |
7591 | ;; function (under the assumption that no declaration level | |
7592 | ;; block construct starts with a type). | |
7593 | (not (c-forward-type)) | |
7594 | ||
7595 | ;; Jumped over a type, but it could be a declaration keyword | |
7596 | ;; followed by the declared identifier that we've jumped over | |
7597 | ;; instead (e.g. in "class Foo {"). If it indeed is a type | |
7598 | ;; then we should be at the declarator now, so check for a | |
7599 | ;; valid declarator start. | |
7600 | ;; | |
7601 | ;; Note: This doesn't cope with the case when a declared | |
7602 | ;; identifier is followed by e.g. '(' in a language where '(' | |
7603 | ;; also might be part of a declarator expression. Currently | |
7604 | ;; there's no such language. | |
7605 | (not (or (looking-at c-symbol-start) | |
7606 | (looking-at c-type-decl-prefix-key))))) | |
7607 | ||
7608 | ;; In Pike a list of modifiers may be followed by a brace | |
7609 | ;; to make them apply to many identifiers. Note that the | |
7610 | ;; match data will be empty on return in this case. | |
7611 | ((and (c-major-mode-is 'pike-mode) | |
7612 | (progn | |
7613 | (goto-char open-brace) | |
7614 | (= (c-backward-token-2) 0)) | |
7615 | (looking-at c-specifier-key) | |
7616 | ;; Use this variant to avoid yet another special regexp. | |
7617 | (c-keyword-member (c-keyword-sym (match-string 1)) | |
7618 | 'c-modifier-kwds)) | |
7619 | (setq kwd-start (point)) | |
7620 | t))) | |
7621 | ||
7622 | ;; Got a match. | |
7623 | ||
7624 | (if goto-start | |
7625 | ;; Back up over any preceding specifiers and their clauses | |
7626 | ;; by going forward from `first-specifier-pos', which is the | |
7627 | ;; earliest possible position where the specifier list can | |
7628 | ;; start. | |
7629 | (progn | |
7630 | (goto-char first-specifier-pos) | |
7631 | ||
7632 | (while (< (point) kwd-start) | |
7633 | (if (looking-at c-symbol-key) | |
7634 | ;; Accept any plain symbol token on the ground that | |
7635 | ;; it's a specifier masked through a macro (just | |
7636 | ;; like `c-forward-decl-or-cast-1' skip forward over | |
7637 | ;; such tokens). | |
7638 | ;; | |
7639 | ;; Could be more restrictive wrt invalid keywords, | |
7640 | ;; but that'd only occur in invalid code so there's | |
7641 | ;; no use spending effort on it. | |
7642 | (let ((end (match-end 0))) | |
7643 | (unless (c-forward-keyword-clause 0) | |
7644 | (goto-char end) | |
7645 | (c-forward-syntactic-ws))) | |
7646 | ||
7647 | ;; Can't parse a declaration preamble and is still | |
7648 | ;; before `kwd-start'. That means `first-specifier-pos' | |
7649 | ;; was in some earlier construct. Search again. | |
7650 | (if (c-syntactic-re-search-forward c-symbol-start | |
7651 | kwd-start 'move t) | |
7652 | (goto-char (setq first-specifier-pos (match-beginning 0))) | |
7653 | ;; Got no preamble before the block declaration keyword. | |
7654 | (setq first-specifier-pos kwd-start)))) | |
7655 | ||
7656 | (goto-char first-specifier-pos)) | |
7657 | (goto-char kwd-start)) | |
7658 | ||
7659 | kwd-start))) | |
ce8c7486 | 7660 | |
a66cd3ee | 7661 | (defun c-search-uplist-for-classkey (paren-state) |
0386b551 AM |
7662 | ;; Check if the closest containing paren sexp is a declaration |
7663 | ;; block, returning a 2 element vector in that case. Aref 0 | |
7664 | ;; contains the bufpos at boi of the class key line, and aref 1 | |
7665 | ;; contains the bufpos of the open brace. This function is an | |
7666 | ;; obsolete wrapper for `c-looking-at-decl-block'. | |
7667 | ;; | |
7668 | ;; This function might do hidden buffer changes. | |
7669 | (let ((open-paren-pos (c-most-enclosing-brace paren-state))) | |
7670 | (when open-paren-pos | |
7671 | (save-excursion | |
7672 | (goto-char open-paren-pos) | |
7673 | (when (and (eq (char-after) ?{) | |
7674 | (c-looking-at-decl-block | |
7675 | (c-safe-position open-paren-pos paren-state) | |
7676 | nil)) | |
7677 | (back-to-indentation) | |
7678 | (vector (point) open-paren-pos)))))) | |
785eecbb | 7679 | |
a66cd3ee | 7680 | (defun c-inside-bracelist-p (containing-sexp paren-state) |
785eecbb RS |
7681 | ;; return the buffer position of the beginning of the brace list |
7682 | ;; statement if we're inside a brace list, otherwise return nil. | |
7683 | ;; CONTAINING-SEXP is the buffer pos of the innermost containing | |
0386b551 | 7684 | ;; paren. PAREN-STATE is the remainder of the state of enclosing |
130c507e | 7685 | ;; braces |
785eecbb RS |
7686 | ;; |
7687 | ;; N.B.: This algorithm can potentially get confused by cpp macros | |
0386b551 | 7688 | ;; placed in inconvenient locations. It's a trade-off we make for |
785eecbb | 7689 | ;; speed. |
0386b551 AM |
7690 | ;; |
7691 | ;; This function might do hidden buffer changes. | |
785eecbb | 7692 | (or |
d9e94c22 | 7693 | ;; This will pick up brace list declarations. |
b2acd789 RS |
7694 | (c-safe |
7695 | (save-excursion | |
7696 | (goto-char containing-sexp) | |
0ec8351b | 7697 | (c-forward-sexp -1) |
b2acd789 | 7698 | (let (bracepos) |
d9e94c22 | 7699 | (if (and (or (looking-at c-brace-list-key) |
0ec8351b | 7700 | (progn (c-forward-sexp -1) |
d9e94c22 | 7701 | (looking-at c-brace-list-key))) |
a66cd3ee | 7702 | (setq bracepos (c-down-list-forward (point))) |
b2acd789 RS |
7703 | (not (c-crosses-statement-barrier-p (point) |
7704 | (- bracepos 2)))) | |
7705 | (point))))) | |
785eecbb RS |
7706 | ;; this will pick up array/aggregate init lists, even if they are nested. |
7707 | (save-excursion | |
0ec8351b BW |
7708 | (let ((class-key |
7709 | ;; Pike can have class definitions anywhere, so we must | |
7710 | ;; check for the class key here. | |
7711 | (and (c-major-mode-is 'pike-mode) | |
a66cd3ee MS |
7712 | c-decl-block-key)) |
7713 | bufpos braceassignp lim next-containing) | |
785eecbb RS |
7714 | (while (and (not bufpos) |
7715 | containing-sexp) | |
a66cd3ee MS |
7716 | (when paren-state |
7717 | (if (consp (car paren-state)) | |
7718 | (setq lim (cdr (car paren-state)) | |
7719 | paren-state (cdr paren-state)) | |
7720 | (setq lim (car paren-state))) | |
7721 | (when paren-state | |
7722 | (setq next-containing (car paren-state) | |
7723 | paren-state (cdr paren-state)))) | |
785eecbb | 7724 | (goto-char containing-sexp) |
a66cd3ee MS |
7725 | (if (c-looking-at-inexpr-block next-containing next-containing) |
7726 | ;; We're in an in-expression block of some kind. Do not | |
7727 | ;; check nesting. We deliberately set the limit to the | |
7728 | ;; containing sexp, so that c-looking-at-inexpr-block | |
7729 | ;; doesn't check for an identifier before it. | |
0ec8351b BW |
7730 | (setq containing-sexp nil) |
7731 | ;; see if the open brace is preceded by = or [...] in | |
7732 | ;; this statement, but watch out for operator= | |
a66cd3ee | 7733 | (setq braceassignp 'dontknow) |
d9e94c22 | 7734 | (c-backward-token-2 1 t lim) |
6393fef2 | 7735 | ;; Checks to do only on the first sexp before the brace. |
d9e94c22 | 7736 | (when (and c-opt-inexpr-brace-list-key |
6393fef2 RS |
7737 | (eq (char-after) ?\[)) |
7738 | ;; In Java, an initialization brace list may follow | |
7739 | ;; directly after "new Foo[]", so check for a "new" | |
7740 | ;; earlier. | |
7741 | (while (eq braceassignp 'dontknow) | |
7742 | (setq braceassignp | |
d9e94c22 MS |
7743 | (cond ((/= (c-backward-token-2 1 t lim) 0) nil) |
7744 | ((looking-at c-opt-inexpr-brace-list-key) t) | |
6393fef2 RS |
7745 | ((looking-at "\\sw\\|\\s_\\|[.[]") |
7746 | ;; Carry on looking if this is an | |
7747 | ;; identifier (may contain "." in Java) | |
7748 | ;; or another "[]" sexp. | |
7749 | 'dontknow) | |
7750 | (t nil))))) | |
7751 | ;; Checks to do on all sexps before the brace, up to the | |
7752 | ;; beginning of the statement. | |
7753 | (while (eq braceassignp 'dontknow) | |
0ec8351b BW |
7754 | (cond ((eq (char-after) ?\;) |
7755 | (setq braceassignp nil)) | |
7756 | ((and class-key | |
7757 | (looking-at class-key)) | |
7758 | (setq braceassignp nil)) | |
7759 | ((eq (char-after) ?=) | |
7760 | ;; We've seen a =, but must check earlier tokens so | |
7761 | ;; that it isn't something that should be ignored. | |
7762 | (setq braceassignp 'maybe) | |
7763 | (while (and (eq braceassignp 'maybe) | |
d9e94c22 | 7764 | (zerop (c-backward-token-2 1 t lim))) |
0ec8351b BW |
7765 | (setq braceassignp |
7766 | (cond | |
7767 | ;; Check for operator = | |
51c9af45 AM |
7768 | ((and c-opt-op-identifier-prefix |
7769 | (looking-at c-opt-op-identifier-prefix)) | |
0386b551 | 7770 | nil) |
130c507e GM |
7771 | ;; Check for `<opchar>= in Pike. |
7772 | ((and (c-major-mode-is 'pike-mode) | |
7773 | (or (eq (char-after) ?`) | |
7774 | ;; Special case for Pikes | |
7775 | ;; `[]=, since '[' is not in | |
7776 | ;; the punctuation class. | |
7777 | (and (eq (char-after) ?\[) | |
7778 | (eq (char-before) ?`)))) | |
7779 | nil) | |
0ec8351b BW |
7780 | ((looking-at "\\s.") 'maybe) |
7781 | ;; make sure we're not in a C++ template | |
7782 | ;; argument assignment | |
a66cd3ee MS |
7783 | ((and |
7784 | (c-major-mode-is 'c++-mode) | |
7785 | (save-excursion | |
7786 | (let ((here (point)) | |
7787 | (pos< (progn | |
7788 | (skip-chars-backward "^<>") | |
7789 | (point)))) | |
7790 | (and (eq (char-before) ?<) | |
7791 | (not (c-crosses-statement-barrier-p | |
7792 | pos< here)) | |
7793 | (not (c-in-literal)) | |
7794 | )))) | |
0ec8351b | 7795 | nil) |
6393fef2 RS |
7796 | (t t)))))) |
7797 | (if (and (eq braceassignp 'dontknow) | |
d9e94c22 | 7798 | (/= (c-backward-token-2 1 t lim) 0)) |
6393fef2 RS |
7799 | (setq braceassignp nil))) |
7800 | (if (not braceassignp) | |
0ec8351b BW |
7801 | (if (eq (char-after) ?\;) |
7802 | ;; Brace lists can't contain a semicolon, so we're done. | |
7803 | (setq containing-sexp nil) | |
a66cd3ee MS |
7804 | ;; Go up one level. |
7805 | (setq containing-sexp next-containing | |
7806 | lim nil | |
7807 | next-containing nil)) | |
0ec8351b BW |
7808 | ;; we've hit the beginning of the aggregate list |
7809 | (c-beginning-of-statement-1 | |
a66cd3ee | 7810 | (c-most-enclosing-brace paren-state)) |
0ec8351b | 7811 | (setq bufpos (point)))) |
a66cd3ee | 7812 | ) |
785eecbb RS |
7813 | bufpos)) |
7814 | )) | |
7815 | ||
0ec8351b BW |
7816 | (defun c-looking-at-special-brace-list (&optional lim) |
7817 | Content-type: text/html