Commit | Line | Data |
---|---|---|
b9c5136f | 1 | /* Caching facts about regions of the buffer, for optimization. |
429ab54e | 2 | Copyright (C) 1985, 1986, 1987, 1988, 1989, 1993, 1995, 2001, 2002, 2003, |
114f9c96 | 3 | 2004, 2005, 2006, 2007, 2008, 2009, 2010 Free Software Foundation, Inc. |
b9c5136f KH |
4 | |
5 | This file is part of GNU Emacs. | |
6 | ||
9ec0b715 | 7 | GNU Emacs is free software: you can redistribute it and/or modify |
b9c5136f | 8 | it under the terms of the GNU General Public License as published by |
9ec0b715 GM |
9 | the Free Software Foundation, either version 3 of the License, or |
10 | (at your option) any later version. | |
b9c5136f KH |
11 | |
12 | GNU Emacs is distributed in the hope that it will be useful, | |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | GNU General Public License for more details. | |
16 | ||
17 | You should have received a copy of the GNU General Public License | |
9ec0b715 | 18 | along with GNU Emacs. If not, see <http://www.gnu.org/licenses/>. */ |
b9c5136f KH |
19 | |
20 | ||
21 | #include <config.h> | |
5890e9f7 | 22 | #include <stdio.h> |
d7306fe6 | 23 | #include <setjmp.h> |
5890e9f7 | 24 | |
b9c5136f KH |
25 | #include "lisp.h" |
26 | #include "buffer.h" | |
27 | #include "region-cache.h" | |
28 | ||
b9c5136f KH |
29 | \f |
30 | /* Data structures. */ | |
31 | ||
32 | /* The region cache. | |
33 | ||
34 | We want something that maps character positions in a buffer onto | |
35 | values. The representation should deal well with long runs of | |
36 | characters with the same value. | |
37 | ||
38 | The tricky part: the representation should be very cheap to | |
39 | maintain in the presence of many insertions and deletions. If the | |
40 | overhead of maintaining the cache is too high, the speedups it | |
41 | offers will be worthless. | |
42 | ||
43 | ||
44 | We represent the region cache as a sorted array of struct | |
45 | boundary's, each of which contains a buffer position and a value; | |
46 | the value applies to all the characters after the buffer position, | |
47 | until the position of the next boundary, or the end of the buffer. | |
48 | ||
49 | The cache always has a boundary whose position is BUF_BEG, so | |
50 | there's always a value associated with every character in the | |
51 | buffer. Since the cache is sorted, this is always the first | |
52 | element of the cache. | |
53 | ||
54 | To facilitate the insertion and deletion of boundaries in the | |
55 | cache, the cache has a gap, just like Emacs's text buffers do. | |
56 | ||
57 | To help boundary positions float along with insertions and | |
58 | deletions, all boundary positions before the cache gap are stored | |
59 | relative to BUF_BEG (buf) (thus they're >= 0), and all boundary | |
60 | positions after the gap are stored relative to BUF_Z (buf) (thus | |
61 | they're <= 0). Look at BOUNDARY_POS to see this in action. See | |
62 | revalidate_region_cache to see how this helps. */ | |
63 | ||
64 | struct boundary { | |
c098fdb8 | 65 | EMACS_INT pos; |
b9c5136f KH |
66 | int value; |
67 | }; | |
68 | ||
69 | struct region_cache { | |
70 | /* A sorted array of locations where the known-ness of the buffer | |
71 | changes. */ | |
72 | struct boundary *boundaries; | |
73 | ||
74 | /* boundaries[gap_start ... gap_start + gap_len - 1] is the gap. */ | |
c098fdb8 | 75 | EMACS_INT gap_start, gap_len; |
b9c5136f KH |
76 | |
77 | /* The number of elements allocated to boundaries, not including the | |
78 | gap. */ | |
f5c7fc27 | 79 | EMACS_INT cache_len; |
b9c5136f KH |
80 | |
81 | /* The areas that haven't changed since the last time we cleaned out | |
82 | invalid entries from the cache. These overlap when the buffer is | |
83 | entirely unchanged. */ | |
c098fdb8 | 84 | EMACS_INT beg_unchanged, end_unchanged; |
b9c5136f KH |
85 | |
86 | /* The first and last positions in the buffer. Because boundaries | |
87 | store their positions relative to the start (BEG) and end (Z) of | |
88 | the buffer, knowing these positions allows us to accurately | |
89 | interpret positions without having to pass the buffer structure | |
90 | or its endpoints around all the time. | |
91 | ||
92 | Yes, buffer_beg is always 1. It's there for symmetry with | |
93 | buffer_end and the BEG and BUF_BEG macros. */ | |
c098fdb8 | 94 | EMACS_INT buffer_beg, buffer_end; |
b9c5136f KH |
95 | }; |
96 | ||
97 | /* Return the position of boundary i in cache c. */ | |
98 | #define BOUNDARY_POS(c, i) \ | |
99 | ((i) < (c)->gap_start \ | |
100 | ? (c)->buffer_beg + (c)->boundaries[(i)].pos \ | |
101 | : (c)->buffer_end + (c)->boundaries[(c)->gap_len + (i)].pos) | |
102 | ||
103 | /* Return the value for text after boundary i in cache c. */ | |
104 | #define BOUNDARY_VALUE(c, i) \ | |
105 | ((i) < (c)->gap_start \ | |
106 | ? (c)->boundaries[(i)].value \ | |
107 | : (c)->boundaries[(c)->gap_len + (i)].value) | |
108 | ||
109 | /* Set the value for text after boundary i in cache c to v. */ | |
110 | #define SET_BOUNDARY_VALUE(c, i, v) \ | |
111 | ((i) < (c)->gap_start \ | |
112 | ? ((c)->boundaries[(i)].value = (v))\ | |
113 | : ((c)->boundaries[(c)->gap_len + (i)].value = (v))) | |
114 | ||
115 | ||
116 | /* How many elements to add to the gap when we resize the buffer. */ | |
117 | #define NEW_CACHE_GAP (40) | |
118 | ||
119 | /* See invalidate_region_cache; if an invalidation would throw away | |
120 | information about this many characters, call | |
121 | revalidate_region_cache before doing the new invalidation, to | |
122 | preserve that information, instead of throwing it away. */ | |
123 | #define PRESERVE_THRESHOLD (500) | |
124 | ||
971de7fb | 125 | static void revalidate_region_cache (struct buffer *buf, struct region_cache *c); |
b9c5136f KH |
126 | |
127 | \f | |
128 | /* Interface: Allocating, initializing, and disposing of region caches. */ | |
129 | ||
130 | struct region_cache * | |
971de7fb | 131 | new_region_cache (void) |
b9c5136f | 132 | { |
177c0ea7 | 133 | struct region_cache *c |
b9c5136f KH |
134 | = (struct region_cache *) xmalloc (sizeof (struct region_cache)); |
135 | ||
136 | c->gap_start = 0; | |
137 | c->gap_len = NEW_CACHE_GAP; | |
138 | c->cache_len = 0; | |
139 | c->boundaries = | |
140 | (struct boundary *) xmalloc ((c->gap_len + c->cache_len) | |
141 | * sizeof (*c->boundaries)); | |
142 | ||
143 | c->beg_unchanged = 0; | |
144 | c->end_unchanged = 0; | |
b0ead4a8 SM |
145 | c->buffer_beg = BEG; |
146 | c->buffer_end = BEG; | |
b9c5136f KH |
147 | |
148 | /* Insert the boundary for the buffer start. */ | |
149 | c->cache_len++; | |
150 | c->gap_len--; | |
151 | c->gap_start++; | |
152 | c->boundaries[0].pos = 0; /* from buffer_beg */ | |
153 | c->boundaries[0].value = 0; | |
154 | ||
155 | return c; | |
156 | } | |
157 | ||
158 | void | |
971de7fb | 159 | free_region_cache (struct region_cache *c) |
b9c5136f KH |
160 | { |
161 | xfree (c->boundaries); | |
162 | xfree (c); | |
163 | } | |
164 | ||
165 | \f | |
166 | /* Finding positions in the cache. */ | |
167 | ||
168 | /* Return the index of the last boundary in cache C at or before POS. | |
169 | In other words, return the boundary that specifies the value for | |
170 | the region POS..(POS + 1). | |
171 | ||
172 | This operation should be logarithmic in the number of cache | |
173 | entries. It would be nice if it took advantage of locality of | |
174 | reference, too, by searching entries near the last entry found. */ | |
f5c7fc27 | 175 | static EMACS_INT |
c098fdb8 | 176 | find_cache_boundary (struct region_cache *c, EMACS_INT pos) |
b9c5136f | 177 | { |
f5c7fc27 | 178 | EMACS_INT low = 0, high = c->cache_len; |
b9c5136f KH |
179 | |
180 | while (low + 1 < high) | |
181 | { | |
182 | /* mid is always a valid index, because low < high and ">> 1" | |
183 | rounds down. */ | |
f5c7fc27 | 184 | EMACS_INT mid = (low + high) >> 1; |
c098fdb8 | 185 | EMACS_INT boundary = BOUNDARY_POS (c, mid); |
b9c5136f KH |
186 | |
187 | if (pos < boundary) | |
188 | high = mid; | |
189 | else | |
190 | low = mid; | |
191 | } | |
192 | ||
193 | /* Some testing. */ | |
194 | if (BOUNDARY_POS (c, low) > pos | |
195 | || (low + 1 < c->cache_len | |
196 | && BOUNDARY_POS (c, low + 1) <= pos)) | |
197 | abort (); | |
198 | ||
199 | return low; | |
200 | } | |
201 | ||
202 | ||
203 | \f | |
204 | /* Moving the cache gap around, inserting, and deleting. */ | |
205 | ||
206 | ||
207 | /* Move the gap of cache C to index POS, and make sure it has space | |
208 | for at least MIN_SIZE boundaries. */ | |
209 | static void | |
f5c7fc27 | 210 | move_cache_gap (struct region_cache *c, EMACS_INT pos, EMACS_INT min_size) |
b9c5136f KH |
211 | { |
212 | /* Copy these out of the cache and into registers. */ | |
c098fdb8 EZ |
213 | EMACS_INT gap_start = c->gap_start; |
214 | EMACS_INT gap_len = c->gap_len; | |
215 | EMACS_INT buffer_beg = c->buffer_beg; | |
216 | EMACS_INT buffer_end = c->buffer_end; | |
b9c5136f KH |
217 | |
218 | if (pos < 0 | |
219 | || pos > c->cache_len) | |
220 | abort (); | |
221 | ||
222 | /* We mustn't ever try to put the gap before the dummy start | |
223 | boundary. That must always be start-relative. */ | |
224 | if (pos == 0) | |
225 | abort (); | |
226 | ||
227 | /* Need we move the gap right? */ | |
228 | while (gap_start < pos) | |
229 | { | |
230 | /* Copy one boundary from after to before the gap, and | |
231 | convert its position to start-relative. */ | |
232 | c->boundaries[gap_start].pos | |
233 | = (buffer_end | |
234 | + c->boundaries[gap_start + gap_len].pos | |
235 | - buffer_beg); | |
236 | c->boundaries[gap_start].value | |
237 | = c->boundaries[gap_start + gap_len].value; | |
238 | gap_start++; | |
239 | } | |
240 | ||
241 | /* To enlarge the gap, we need to re-allocate the boundary array, and | |
242 | then shift the area after the gap to the new end. Since the cost | |
243 | is proportional to the amount of stuff after the gap, we do the | |
244 | enlargement here, after a right shift but before a left shift, | |
245 | when the portion after the gap is smallest. */ | |
246 | if (gap_len < min_size) | |
247 | { | |
c098fdb8 | 248 | EMACS_INT i; |
b9c5136f KH |
249 | |
250 | /* Always make at least NEW_CACHE_GAP elements, as long as we're | |
251 | expanding anyway. */ | |
252 | if (min_size < NEW_CACHE_GAP) | |
253 | min_size = NEW_CACHE_GAP; | |
254 | ||
255 | c->boundaries = | |
256 | (struct boundary *) xrealloc (c->boundaries, | |
257 | ((min_size + c->cache_len) | |
258 | * sizeof (*c->boundaries))); | |
259 | ||
260 | /* Some systems don't provide a version of the copy routine that | |
261 | can be trusted to shift memory upward into an overlapping | |
262 | region. memmove isn't widely available. */ | |
263 | min_size -= gap_len; | |
264 | for (i = c->cache_len - 1; i >= gap_start; i--) | |
265 | { | |
266 | c->boundaries[i + min_size].pos = c->boundaries[i + gap_len].pos; | |
267 | c->boundaries[i + min_size].value = c->boundaries[i + gap_len].value; | |
268 | } | |
269 | ||
270 | gap_len = min_size; | |
271 | } | |
272 | ||
273 | /* Need we move the gap left? */ | |
274 | while (pos < gap_start) | |
275 | { | |
276 | gap_start--; | |
277 | ||
278 | /* Copy one region from before to after the gap, and | |
279 | convert its position to end-relative. */ | |
280 | c->boundaries[gap_start + gap_len].pos | |
281 | = c->boundaries[gap_start].pos + buffer_beg - buffer_end; | |
282 | c->boundaries[gap_start + gap_len].value | |
283 | = c->boundaries[gap_start].value; | |
284 | } | |
285 | ||
286 | /* Assign these back into the cache. */ | |
287 | c->gap_start = gap_start; | |
288 | c->gap_len = gap_len; | |
289 | } | |
290 | ||
291 | ||
292 | /* Insert a new boundary in cache C; it will have cache index INDEX, | |
293 | and have the specified POS and VALUE. */ | |
294 | static void | |
f5c7fc27 | 295 | insert_cache_boundary (struct region_cache *c, EMACS_INT index, EMACS_INT pos, |
c098fdb8 | 296 | int value) |
b9c5136f KH |
297 | { |
298 | /* index must be a valid cache index. */ | |
299 | if (index < 0 || index > c->cache_len) | |
300 | abort (); | |
301 | ||
302 | /* We must never want to insert something before the dummy first | |
303 | boundary. */ | |
304 | if (index == 0) | |
305 | abort (); | |
306 | ||
307 | /* We must only be inserting things in order. */ | |
308 | if (! (BOUNDARY_POS (c, index-1) < pos | |
309 | && (index == c->cache_len | |
310 | || pos < BOUNDARY_POS (c, index)))) | |
311 | abort (); | |
312 | ||
313 | /* The value must be different from the ones around it. However, we | |
314 | temporarily create boundaries that establish the same value as | |
315 | the subsequent boundary, so we're not going to flag that case. */ | |
316 | if (BOUNDARY_VALUE (c, index-1) == value) | |
317 | abort (); | |
318 | ||
319 | move_cache_gap (c, index, 1); | |
320 | ||
321 | c->boundaries[index].pos = pos - c->buffer_beg; | |
322 | c->boundaries[index].value = value; | |
323 | c->gap_start++; | |
324 | c->gap_len--; | |
325 | c->cache_len++; | |
326 | } | |
327 | ||
328 | ||
329 | /* Delete the i'th entry from cache C if START <= i < END. */ | |
330 | ||
331 | static void | |
c098fdb8 EZ |
332 | delete_cache_boundaries (struct region_cache *c, |
333 | EMACS_INT start, EMACS_INT end) | |
b9c5136f | 334 | { |
c098fdb8 | 335 | EMACS_INT len = end - start; |
b9c5136f KH |
336 | |
337 | /* Gotta be in range. */ | |
338 | if (start < 0 | |
339 | || end > c->cache_len) | |
340 | abort (); | |
341 | ||
342 | /* Gotta be in order. */ | |
343 | if (start > end) | |
344 | abort (); | |
345 | ||
346 | /* Can't delete the dummy entry. */ | |
347 | if (start == 0 | |
348 | && end >= 1) | |
349 | abort (); | |
350 | ||
351 | /* Minimize gap motion. If we're deleting nothing, do nothing. */ | |
352 | if (len == 0) | |
353 | ; | |
354 | /* If the gap is before the region to delete, delete from the start | |
355 | forward. */ | |
356 | else if (c->gap_start <= start) | |
357 | { | |
358 | move_cache_gap (c, start, 0); | |
359 | c->gap_len += len; | |
360 | } | |
361 | /* If the gap is after the region to delete, delete from the end | |
362 | backward. */ | |
363 | else if (end <= c->gap_start) | |
364 | { | |
365 | move_cache_gap (c, end, 0); | |
366 | c->gap_start -= len; | |
367 | c->gap_len += len; | |
368 | } | |
369 | /* If the gap is in the region to delete, just expand it. */ | |
370 | else | |
371 | { | |
372 | c->gap_start = start; | |
373 | c->gap_len += len; | |
374 | } | |
375 | ||
376 | c->cache_len -= len; | |
377 | } | |
177c0ea7 | 378 | |
b9c5136f KH |
379 | |
380 | \f | |
381 | /* Set the value for a region. */ | |
382 | ||
383 | /* Set the value in cache C for the region START..END to VALUE. */ | |
384 | static void | |
c098fdb8 EZ |
385 | set_cache_region (struct region_cache *c, |
386 | EMACS_INT start, EMACS_INT end, int value) | |
b9c5136f KH |
387 | { |
388 | if (start > end) | |
389 | abort (); | |
390 | if (start < c->buffer_beg | |
391 | || end > c->buffer_end) | |
392 | abort (); | |
393 | ||
394 | /* Eliminate this case; then we can assume that start and end-1 are | |
395 | both the locations of real characters in the buffer. */ | |
396 | if (start == end) | |
397 | return; | |
177c0ea7 | 398 | |
b9c5136f KH |
399 | { |
400 | /* We need to make sure that there are no boundaries in the area | |
401 | between start to end; the whole area will have the same value, | |
402 | so those boundaries will not be necessary. | |
177c0ea7 | 403 | |
b9c5136f KH |
404 | Let start_ix be the cache index of the boundary governing the |
405 | first character of start..end, and let end_ix be the cache | |
406 | index of the earliest boundary after the last character in | |
407 | start..end. (This tortured terminology is intended to answer | |
408 | all the "< or <=?" sort of questions.) */ | |
f5c7fc27 LMI |
409 | EMACS_INT start_ix = find_cache_boundary (c, start); |
410 | EMACS_INT end_ix = find_cache_boundary (c, end - 1) + 1; | |
b9c5136f KH |
411 | |
412 | /* We must remember the value established by the last boundary | |
413 | before end; if that boundary's domain stretches beyond end, | |
414 | we'll need to create a new boundary at end, and that boundary | |
415 | must have that remembered value. */ | |
416 | int value_at_end = BOUNDARY_VALUE (c, end_ix - 1); | |
417 | ||
418 | /* Delete all boundaries strictly within start..end; this means | |
419 | those whose indices are between start_ix (exclusive) and end_ix | |
420 | (exclusive). */ | |
421 | delete_cache_boundaries (c, start_ix + 1, end_ix); | |
422 | ||
423 | /* Make sure we have the right value established going in to | |
424 | start..end from the left, and no unnecessary boundaries. */ | |
425 | if (BOUNDARY_POS (c, start_ix) == start) | |
426 | { | |
427 | /* Is this boundary necessary? If no, remove it; if yes, set | |
428 | its value. */ | |
429 | if (start_ix > 0 | |
430 | && BOUNDARY_VALUE (c, start_ix - 1) == value) | |
431 | { | |
432 | delete_cache_boundaries (c, start_ix, start_ix + 1); | |
433 | start_ix--; | |
434 | } | |
435 | else | |
436 | SET_BOUNDARY_VALUE (c, start_ix, value); | |
437 | } | |
438 | else | |
439 | { | |
440 | /* Do we need to add a new boundary here? */ | |
441 | if (BOUNDARY_VALUE (c, start_ix) != value) | |
442 | { | |
443 | insert_cache_boundary (c, start_ix + 1, start, value); | |
444 | start_ix++; | |
445 | } | |
446 | } | |
177c0ea7 | 447 | |
b9c5136f KH |
448 | /* This is equivalent to letting end_ix float (like a buffer |
449 | marker does) with the insertions and deletions we may have | |
450 | done. */ | |
451 | end_ix = start_ix + 1; | |
452 | ||
453 | /* Make sure we have the correct value established as we leave | |
454 | start..end to the right. */ | |
455 | if (end == c->buffer_end) | |
456 | /* There is no text after start..end; nothing to do. */ | |
457 | ; | |
458 | else if (end_ix >= c->cache_len | |
459 | || end < BOUNDARY_POS (c, end_ix)) | |
460 | { | |
461 | /* There is no boundary at end, but we may need one. */ | |
462 | if (value_at_end != value) | |
463 | insert_cache_boundary (c, end_ix, end, value_at_end); | |
464 | } | |
465 | else | |
466 | { | |
467 | /* There is a boundary at end; should it be there? */ | |
468 | if (value == BOUNDARY_VALUE (c, end_ix)) | |
469 | delete_cache_boundaries (c, end_ix, end_ix + 1); | |
470 | } | |
471 | } | |
472 | } | |
473 | ||
474 | ||
475 | \f | |
476 | /* Interface: Invalidating the cache. Private: Re-validating the cache. */ | |
477 | ||
478 | /* Indicate that a section of BUF has changed, to invalidate CACHE. | |
479 | HEAD is the number of chars unchanged at the beginning of the buffer. | |
480 | TAIL is the number of chars unchanged at the end of the buffer. | |
481 | NOTE: this is *not* the same as the ending position of modified | |
482 | region. | |
483 | (This way of specifying regions makes more sense than absolute | |
484 | buffer positions in the presence of insertions and deletions; the | |
485 | args to pass are the same before and after such an operation.) */ | |
486 | void | |
c098fdb8 EZ |
487 | invalidate_region_cache (struct buffer *buf, struct region_cache *c, |
488 | EMACS_INT head, EMACS_INT tail) | |
b9c5136f KH |
489 | { |
490 | /* Let chead = c->beg_unchanged, and | |
491 | ctail = c->end_unchanged. | |
492 | If z-tail < beg+chead by a large amount, or | |
493 | z-ctail < beg+head by a large amount, | |
494 | ||
495 | then cutting back chead and ctail to head and tail would lose a | |
496 | lot of information that we could preserve by revalidating the | |
497 | cache before processing this invalidation. Losing that | |
498 | information may be more costly than revalidating the cache now. | |
499 | So go ahead and call revalidate_region_cache if it seems that it | |
500 | might be worthwhile. */ | |
501 | if (((BUF_BEG (buf) + c->beg_unchanged) - (BUF_Z (buf) - tail) | |
502 | > PRESERVE_THRESHOLD) | |
503 | || ((BUF_BEG (buf) + head) - (BUF_Z (buf) - c->end_unchanged) | |
504 | > PRESERVE_THRESHOLD)) | |
505 | revalidate_region_cache (buf, c); | |
506 | ||
507 | ||
508 | if (head < c->beg_unchanged) | |
509 | c->beg_unchanged = head; | |
510 | if (tail < c->end_unchanged) | |
511 | c->end_unchanged = tail; | |
512 | ||
513 | /* We now know nothing about the region between the unchanged head | |
514 | and the unchanged tail (call it the "modified region"), not even | |
515 | its length. | |
516 | ||
517 | If the modified region has shrunk in size (deletions do this), | |
518 | then the cache may now contain boundaries originally located in | |
519 | text that doesn't exist any more. | |
520 | ||
521 | If the modified region has increased in size (insertions do | |
522 | this), then there may now be boundaries in the modified region | |
523 | whose positions are wrong. | |
524 | ||
525 | Even calling BOUNDARY_POS on boundaries still in the unchanged | |
526 | head or tail may well give incorrect answers now, since | |
527 | c->buffer_beg and c->buffer_end may well be wrong now. (Well, | |
528 | okay, c->buffer_beg never changes, so boundaries in the unchanged | |
529 | head will still be okay. But it's the principle of the thing.) | |
530 | ||
531 | So things are generally a mess. | |
532 | ||
533 | But we don't clean up this mess here; that would be expensive, | |
534 | and this function gets called every time any buffer modification | |
535 | occurs. Rather, we can clean up everything in one swell foop, | |
536 | accounting for all the modifications at once, by calling | |
537 | revalidate_region_cache before we try to consult the cache the | |
538 | next time. */ | |
539 | } | |
540 | ||
541 | ||
177c0ea7 | 542 | /* Clean out any cache entries applying to the modified region, and |
b9c5136f KH |
543 | make the positions of the remaining entries accurate again. |
544 | ||
545 | After calling this function, the mess described in the comment in | |
546 | invalidate_region_cache is cleaned up. | |
547 | ||
548 | This function operates by simply throwing away everything it knows | |
549 | about the modified region. It doesn't care exactly which | |
550 | insertions and deletions took place; it just tosses it all. | |
551 | ||
552 | For example, if you insert a single character at the beginning of | |
553 | the buffer, and a single character at the end of the buffer (for | |
554 | example), without calling this function in between the two | |
555 | insertions, then the entire cache will be freed of useful | |
556 | information. On the other hand, if you do manage to call this | |
557 | function in between the two insertions, then the modified regions | |
558 | will be small in both cases, no information will be tossed, and the | |
559 | cache will know that it doesn't have knowledge of the first and | |
560 | last characters any more. | |
561 | ||
562 | Calling this function may be expensive; it does binary searches in | |
563 | the cache, and causes cache gap motion. */ | |
564 | ||
565 | static void | |
971de7fb | 566 | revalidate_region_cache (struct buffer *buf, struct region_cache *c) |
b9c5136f KH |
567 | { |
568 | /* The boundaries now in the cache are expressed relative to the | |
569 | buffer_beg and buffer_end values stored in the cache. Now, | |
570 | buffer_beg and buffer_end may not be the same as BUF_BEG (buf) | |
571 | and BUF_Z (buf), so we have two different "bases" to deal with | |
572 | --- the cache's, and the buffer's. */ | |
573 | ||
574 | /* If the entire buffer is still valid, don't waste time. Yes, this | |
575 | should be a >, not a >=; think about what beg_unchanged and | |
576 | end_unchanged get set to when the only change has been an | |
577 | insertion. */ | |
578 | if (c->buffer_beg + c->beg_unchanged | |
579 | > c->buffer_end - c->end_unchanged) | |
580 | return; | |
581 | ||
582 | /* If all the text we knew about as of the last cache revalidation | |
583 | is still there, then all of the information in the cache is still | |
584 | valid. Because c->buffer_beg and c->buffer_end are out-of-date, | |
585 | the modified region appears from the cache's point of view to be | |
586 | a null region located someplace in the buffer. | |
587 | ||
588 | Now, invalidating that empty string will have no actual affect on | |
589 | the cache; instead, we need to update the cache's basis first | |
590 | (which will give the modified region the same size in the cache | |
591 | as it has in the buffer), and then invalidate the modified | |
592 | region. */ | |
177c0ea7 | 593 | if (c->buffer_beg + c->beg_unchanged |
b9c5136f KH |
594 | == c->buffer_end - c->end_unchanged) |
595 | { | |
596 | /* Move the gap so that all the boundaries in the unchanged head | |
597 | are expressed beg-relative, and all the boundaries in the | |
598 | unchanged tail are expressed end-relative. That done, we can | |
599 | plug in the new buffer beg and end, and all the positions | |
600 | will be accurate. | |
601 | ||
602 | The boundary which has jurisdiction over the modified region | |
603 | should be left before the gap. */ | |
604 | move_cache_gap (c, | |
605 | (find_cache_boundary (c, (c->buffer_beg | |
606 | + c->beg_unchanged)) | |
607 | + 1), | |
608 | 0); | |
609 | ||
610 | c->buffer_beg = BUF_BEG (buf); | |
611 | c->buffer_end = BUF_Z (buf); | |
612 | ||
613 | /* Now that the cache's basis has been changed, the modified | |
614 | region actually takes up some space in the cache, so we can | |
615 | invalidate it. */ | |
616 | set_cache_region (c, | |
617 | c->buffer_beg + c->beg_unchanged, | |
618 | c->buffer_end - c->end_unchanged, | |
619 | 0); | |
620 | } | |
621 | ||
622 | /* Otherwise, there is a non-empty region in the cache which | |
623 | corresponds to the modified region of the buffer. */ | |
624 | else | |
625 | { | |
f5c7fc27 | 626 | EMACS_INT modified_ix; |
b9c5136f KH |
627 | |
628 | /* These positions are correct, relative to both the cache basis | |
629 | and the buffer basis. */ | |
630 | set_cache_region (c, | |
631 | c->buffer_beg + c->beg_unchanged, | |
632 | c->buffer_end - c->end_unchanged, | |
633 | 0); | |
634 | ||
635 | /* Now the cache contains only boundaries that are in the | |
636 | unchanged head and tail; we've disposed of any boundaries | |
637 | whose positions we can't be sure of given the information | |
638 | we've saved. | |
639 | ||
640 | If we put the cache gap between the unchanged head and the | |
641 | unchanged tail, we can adjust all the boundary positions at | |
642 | once, simply by setting buffer_beg and buffer_end. | |
643 | ||
644 | The boundary which has jurisdiction over the modified region | |
645 | should be left before the gap. */ | |
646 | modified_ix = | |
647 | find_cache_boundary (c, (c->buffer_beg + c->beg_unchanged)) + 1; | |
648 | move_cache_gap (c, modified_ix, 0); | |
649 | ||
650 | c->buffer_beg = BUF_BEG (buf); | |
651 | c->buffer_end = BUF_Z (buf); | |
652 | ||
653 | /* Now, we may have shrunk the buffer when we changed the basis, | |
654 | and brought the boundaries we created for the start and end | |
655 | of the modified region together, giving them the same | |
656 | position. If that's the case, we should collapse them into | |
657 | one boundary. Or we may even delete them both, if the values | |
658 | before and after them are the same. */ | |
659 | if (modified_ix < c->cache_len | |
660 | && (BOUNDARY_POS (c, modified_ix - 1) | |
661 | == BOUNDARY_POS (c, modified_ix))) | |
662 | { | |
663 | int value_after = BOUNDARY_VALUE (c, modified_ix); | |
664 | ||
665 | /* Should we remove both of the boundaries? Yes, if the | |
666 | latter boundary is now establishing the same value that | |
667 | the former boundary's predecessor does. */ | |
668 | if (modified_ix - 1 > 0 | |
669 | && value_after == BOUNDARY_VALUE (c, modified_ix - 2)) | |
670 | delete_cache_boundaries (c, modified_ix - 1, modified_ix + 1); | |
671 | else | |
672 | { | |
673 | /* We do need a boundary here; collapse the two | |
674 | boundaries into one. */ | |
675 | SET_BOUNDARY_VALUE (c, modified_ix - 1, value_after); | |
676 | delete_cache_boundaries (c, modified_ix, modified_ix + 1); | |
677 | } | |
678 | } | |
679 | } | |
680 | ||
681 | /* Now the entire cache is valid. */ | |
682 | c->beg_unchanged | |
683 | = c->end_unchanged | |
684 | = c->buffer_end - c->buffer_beg; | |
685 | } | |
686 | ||
687 | \f | |
688 | /* Interface: Adding information to the cache. */ | |
689 | ||
690 | /* Assert that the region of BUF between START and END (absolute | |
691 | buffer positions) is "known," for the purposes of CACHE (e.g. "has | |
692 | no newlines", in the case of the line cache). */ | |
693 | void | |
c098fdb8 EZ |
694 | know_region_cache (struct buffer *buf, struct region_cache *c, |
695 | EMACS_INT start, EMACS_INT end) | |
b9c5136f KH |
696 | { |
697 | revalidate_region_cache (buf, c); | |
698 | ||
699 | set_cache_region (c, start, end, 1); | |
700 | } | |
701 | ||
702 | \f | |
703 | /* Interface: using the cache. */ | |
704 | ||
705 | /* Return true if the text immediately after POS in BUF is known, for | |
177c0ea7 | 706 | the purposes of CACHE. If NEXT is non-zero, set *NEXT to the nearest |
b9c5136f KH |
707 | position after POS where the knownness changes. */ |
708 | int | |
c098fdb8 EZ |
709 | region_cache_forward (struct buffer *buf, struct region_cache *c, |
710 | EMACS_INT pos, EMACS_INT *next) | |
b9c5136f KH |
711 | { |
712 | revalidate_region_cache (buf, c); | |
713 | ||
714 | { | |
f5c7fc27 | 715 | EMACS_INT i = find_cache_boundary (c, pos); |
b9c5136f | 716 | int i_value = BOUNDARY_VALUE (c, i); |
f5c7fc27 | 717 | EMACS_INT j; |
b9c5136f KH |
718 | |
719 | /* Beyond the end of the buffer is unknown, by definition. */ | |
720 | if (pos >= BUF_Z (buf)) | |
721 | { | |
722 | if (next) *next = BUF_Z (buf); | |
723 | i_value = 0; | |
724 | } | |
725 | else if (next) | |
726 | { | |
727 | /* Scan forward from i to find the next differing position. */ | |
728 | for (j = i + 1; j < c->cache_len; j++) | |
729 | if (BOUNDARY_VALUE (c, j) != i_value) | |
730 | break; | |
731 | ||
732 | if (j < c->cache_len) | |
733 | *next = BOUNDARY_POS (c, j); | |
734 | else | |
735 | *next = BUF_Z (buf); | |
736 | } | |
737 | ||
738 | return i_value; | |
739 | } | |
740 | } | |
741 | ||
742 | /* Return true if the text immediately before POS in BUF is known, for | |
743 | the purposes of CACHE. If NEXT is non-zero, set *NEXT to the nearest | |
744 | position before POS where the knownness changes. */ | |
c098fdb8 EZ |
745 | int region_cache_backward (struct buffer *buf, struct region_cache *c, |
746 | EMACS_INT pos, EMACS_INT *next) | |
b9c5136f KH |
747 | { |
748 | revalidate_region_cache (buf, c); | |
749 | ||
750 | /* Before the beginning of the buffer is unknown, by | |
751 | definition. */ | |
752 | if (pos <= BUF_BEG (buf)) | |
753 | { | |
754 | if (next) *next = BUF_BEG (buf); | |
755 | return 0; | |
756 | } | |
757 | ||
758 | { | |
f5c7fc27 | 759 | EMACS_INT i = find_cache_boundary (c, pos - 1); |
b9c5136f | 760 | int i_value = BOUNDARY_VALUE (c, i); |
f5c7fc27 | 761 | EMACS_INT j; |
b9c5136f KH |
762 | |
763 | if (next) | |
764 | { | |
765 | /* Scan backward from i to find the next differing position. */ | |
766 | for (j = i - 1; j >= 0; j--) | |
767 | if (BOUNDARY_VALUE (c, j) != i_value) | |
768 | break; | |
769 | ||
770 | if (j >= 0) | |
771 | *next = BOUNDARY_POS (c, j + 1); | |
772 | else | |
773 | *next = BUF_BEG (buf); | |
774 | } | |
775 | ||
776 | return i_value; | |
777 | } | |
778 | } | |
779 | ||
780 | \f | |
781 | /* Debugging: pretty-print a cache to the standard error output. */ | |
782 | ||
783 | void | |
971de7fb | 784 | pp_cache (struct region_cache *c) |
b9c5136f KH |
785 | { |
786 | int i; | |
c098fdb8 EZ |
787 | EMACS_INT beg_u = c->buffer_beg + c->beg_unchanged; |
788 | EMACS_INT end_u = c->buffer_end - c->end_unchanged; | |
b9c5136f KH |
789 | |
790 | fprintf (stderr, | |
5816888b EZ |
791 | "basis: %ld..%ld modified: %ld..%ld\n", |
792 | (long)c->buffer_beg, (long)c->buffer_end, | |
793 | (long)beg_u, (long)end_u); | |
b9c5136f KH |
794 | |
795 | for (i = 0; i < c->cache_len; i++) | |
796 | { | |
f5c7fc27 | 797 | EMACS_INT pos = BOUNDARY_POS (c, i); |
b9c5136f KH |
798 | |
799 | putc (((pos < beg_u) ? 'v' | |
800 | : (pos == beg_u) ? '-' | |
801 | : ' '), | |
802 | stderr); | |
803 | putc (((pos > end_u) ? '^' | |
804 | : (pos == end_u) ? '-' | |
805 | : ' '), | |
806 | stderr); | |
5816888b | 807 | fprintf (stderr, "%ld : %d\n", (long)pos, BOUNDARY_VALUE (c, i)); |
b9c5136f KH |
808 | } |
809 | } | |
ab5796a9 MB |
810 | |
811 | /* arch-tag: 98c29f3f-2ca2-4e3a-92f0-f2249200a17d | |
812 | (do not change this comment) */ |