Commit | Line | Data |
---|---|---|
b9c5136f | 1 | /* Caching facts about regions of the buffer, for optimization. |
f8c25f1b | 2 | Copyright (C) 1985, 1986, 1987, 1988, 1989, 1993, 1995 |
b9c5136f KH |
3 | Free Software Foundation, Inc. |
4 | ||
5 | This file is part of GNU Emacs. | |
6 | ||
7 | GNU Emacs is free software; you can redistribute it and/or modify | |
8 | it under the terms of the GNU General Public License as published by | |
9 | the Free Software Foundation; either version 2, or (at your option) | |
10 | any later version. | |
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 | |
18 | along with GNU Emacs; see the file COPYING. If not, write to | |
3b7ad313 EN |
19 | the Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
20 | Boston, MA 02111-1307, USA. */ | |
b9c5136f KH |
21 | |
22 | ||
23 | #include <config.h> | |
24 | #include "lisp.h" | |
25 | #include "buffer.h" | |
26 | #include "region-cache.h" | |
27 | ||
28 | #include <stdio.h> | |
29 | ||
30 | \f | |
31 | /* Data structures. */ | |
32 | ||
33 | /* The region cache. | |
34 | ||
35 | We want something that maps character positions in a buffer onto | |
36 | values. The representation should deal well with long runs of | |
37 | characters with the same value. | |
38 | ||
39 | The tricky part: the representation should be very cheap to | |
40 | maintain in the presence of many insertions and deletions. If the | |
41 | overhead of maintaining the cache is too high, the speedups it | |
42 | offers will be worthless. | |
43 | ||
44 | ||
45 | We represent the region cache as a sorted array of struct | |
46 | boundary's, each of which contains a buffer position and a value; | |
47 | the value applies to all the characters after the buffer position, | |
48 | until the position of the next boundary, or the end of the buffer. | |
49 | ||
50 | The cache always has a boundary whose position is BUF_BEG, so | |
51 | there's always a value associated with every character in the | |
52 | buffer. Since the cache is sorted, this is always the first | |
53 | element of the cache. | |
54 | ||
55 | To facilitate the insertion and deletion of boundaries in the | |
56 | cache, the cache has a gap, just like Emacs's text buffers do. | |
57 | ||
58 | To help boundary positions float along with insertions and | |
59 | deletions, all boundary positions before the cache gap are stored | |
60 | relative to BUF_BEG (buf) (thus they're >= 0), and all boundary | |
61 | positions after the gap are stored relative to BUF_Z (buf) (thus | |
62 | they're <= 0). Look at BOUNDARY_POS to see this in action. See | |
63 | revalidate_region_cache to see how this helps. */ | |
64 | ||
65 | struct boundary { | |
66 | int pos; | |
67 | int value; | |
68 | }; | |
69 | ||
70 | struct region_cache { | |
71 | /* A sorted array of locations where the known-ness of the buffer | |
72 | changes. */ | |
73 | struct boundary *boundaries; | |
74 | ||
75 | /* boundaries[gap_start ... gap_start + gap_len - 1] is the gap. */ | |
76 | int gap_start, gap_len; | |
77 | ||
78 | /* The number of elements allocated to boundaries, not including the | |
79 | gap. */ | |
80 | int cache_len; | |
81 | ||
82 | /* The areas that haven't changed since the last time we cleaned out | |
83 | invalid entries from the cache. These overlap when the buffer is | |
84 | entirely unchanged. */ | |
85 | int beg_unchanged, end_unchanged; | |
86 | ||
87 | /* The first and last positions in the buffer. Because boundaries | |
88 | store their positions relative to the start (BEG) and end (Z) of | |
89 | the buffer, knowing these positions allows us to accurately | |
90 | interpret positions without having to pass the buffer structure | |
91 | or its endpoints around all the time. | |
92 | ||
93 | Yes, buffer_beg is always 1. It's there for symmetry with | |
94 | buffer_end and the BEG and BUF_BEG macros. */ | |
95 | int buffer_beg, buffer_end; | |
96 | }; | |
97 | ||
98 | /* Return the position of boundary i in cache c. */ | |
99 | #define BOUNDARY_POS(c, i) \ | |
100 | ((i) < (c)->gap_start \ | |
101 | ? (c)->buffer_beg + (c)->boundaries[(i)].pos \ | |
102 | : (c)->buffer_end + (c)->boundaries[(c)->gap_len + (i)].pos) | |
103 | ||
104 | /* Return the value for text after boundary i in cache c. */ | |
105 | #define BOUNDARY_VALUE(c, i) \ | |
106 | ((i) < (c)->gap_start \ | |
107 | ? (c)->boundaries[(i)].value \ | |
108 | : (c)->boundaries[(c)->gap_len + (i)].value) | |
109 | ||
110 | /* Set the value for text after boundary i in cache c to v. */ | |
111 | #define SET_BOUNDARY_VALUE(c, i, v) \ | |
112 | ((i) < (c)->gap_start \ | |
113 | ? ((c)->boundaries[(i)].value = (v))\ | |
114 | : ((c)->boundaries[(c)->gap_len + (i)].value = (v))) | |
115 | ||
116 | ||
117 | /* How many elements to add to the gap when we resize the buffer. */ | |
118 | #define NEW_CACHE_GAP (40) | |
119 | ||
120 | /* See invalidate_region_cache; if an invalidation would throw away | |
121 | information about this many characters, call | |
122 | revalidate_region_cache before doing the new invalidation, to | |
123 | preserve that information, instead of throwing it away. */ | |
124 | #define PRESERVE_THRESHOLD (500) | |
125 | ||
126 | static void revalidate_region_cache (); | |
127 | ||
128 | \f | |
129 | /* Interface: Allocating, initializing, and disposing of region caches. */ | |
130 | ||
131 | struct region_cache * | |
132 | new_region_cache () | |
133 | { | |
134 | struct region_cache *c | |
135 | = (struct region_cache *) xmalloc (sizeof (struct region_cache)); | |
136 | ||
137 | c->gap_start = 0; | |
138 | c->gap_len = NEW_CACHE_GAP; | |
139 | c->cache_len = 0; | |
140 | c->boundaries = | |
141 | (struct boundary *) xmalloc ((c->gap_len + c->cache_len) | |
142 | * sizeof (*c->boundaries)); | |
143 | ||
144 | c->beg_unchanged = 0; | |
145 | c->end_unchanged = 0; | |
146 | c->buffer_beg = 1; | |
147 | c->buffer_end = 1; | |
148 | ||
149 | /* Insert the boundary for the buffer start. */ | |
150 | c->cache_len++; | |
151 | c->gap_len--; | |
152 | c->gap_start++; | |
153 | c->boundaries[0].pos = 0; /* from buffer_beg */ | |
154 | c->boundaries[0].value = 0; | |
155 | ||
156 | return c; | |
157 | } | |
158 | ||
159 | void | |
160 | free_region_cache (c) | |
161 | struct region_cache *c; | |
162 | { | |
163 | xfree (c->boundaries); | |
164 | xfree (c); | |
165 | } | |
166 | ||
167 | \f | |
168 | /* Finding positions in the cache. */ | |
169 | ||
170 | /* Return the index of the last boundary in cache C at or before POS. | |
171 | In other words, return the boundary that specifies the value for | |
172 | the region POS..(POS + 1). | |
173 | ||
174 | This operation should be logarithmic in the number of cache | |
175 | entries. It would be nice if it took advantage of locality of | |
176 | reference, too, by searching entries near the last entry found. */ | |
177 | static int | |
178 | find_cache_boundary (c, pos) | |
179 | struct region_cache *c; | |
180 | int pos; | |
181 | { | |
182 | int low = 0, high = c->cache_len; | |
183 | ||
184 | while (low + 1 < high) | |
185 | { | |
186 | /* mid is always a valid index, because low < high and ">> 1" | |
187 | rounds down. */ | |
188 | int mid = (low + high) >> 1; | |
189 | int boundary = BOUNDARY_POS (c, mid); | |
190 | ||
191 | if (pos < boundary) | |
192 | high = mid; | |
193 | else | |
194 | low = mid; | |
195 | } | |
196 | ||
197 | /* Some testing. */ | |
198 | if (BOUNDARY_POS (c, low) > pos | |
199 | || (low + 1 < c->cache_len | |
200 | && BOUNDARY_POS (c, low + 1) <= pos)) | |
201 | abort (); | |
202 | ||
203 | return low; | |
204 | } | |
205 | ||
206 | ||
207 | \f | |
208 | /* Moving the cache gap around, inserting, and deleting. */ | |
209 | ||
210 | ||
211 | /* Move the gap of cache C to index POS, and make sure it has space | |
212 | for at least MIN_SIZE boundaries. */ | |
213 | static void | |
214 | move_cache_gap (c, pos, min_size) | |
215 | struct region_cache *c; | |
216 | int pos; | |
217 | int min_size; | |
218 | { | |
219 | /* Copy these out of the cache and into registers. */ | |
220 | int gap_start = c->gap_start; | |
221 | int gap_len = c->gap_len; | |
222 | int buffer_beg = c->buffer_beg; | |
223 | int buffer_end = c->buffer_end; | |
224 | ||
225 | if (pos < 0 | |
226 | || pos > c->cache_len) | |
227 | abort (); | |
228 | ||
229 | /* We mustn't ever try to put the gap before the dummy start | |
230 | boundary. That must always be start-relative. */ | |
231 | if (pos == 0) | |
232 | abort (); | |
233 | ||
234 | /* Need we move the gap right? */ | |
235 | while (gap_start < pos) | |
236 | { | |
237 | /* Copy one boundary from after to before the gap, and | |
238 | convert its position to start-relative. */ | |
239 | c->boundaries[gap_start].pos | |
240 | = (buffer_end | |
241 | + c->boundaries[gap_start + gap_len].pos | |
242 | - buffer_beg); | |
243 | c->boundaries[gap_start].value | |
244 | = c->boundaries[gap_start + gap_len].value; | |
245 | gap_start++; | |
246 | } | |
247 | ||
248 | /* To enlarge the gap, we need to re-allocate the boundary array, and | |
249 | then shift the area after the gap to the new end. Since the cost | |
250 | is proportional to the amount of stuff after the gap, we do the | |
251 | enlargement here, after a right shift but before a left shift, | |
252 | when the portion after the gap is smallest. */ | |
253 | if (gap_len < min_size) | |
254 | { | |
255 | int i; | |
256 | ||
257 | /* Always make at least NEW_CACHE_GAP elements, as long as we're | |
258 | expanding anyway. */ | |
259 | if (min_size < NEW_CACHE_GAP) | |
260 | min_size = NEW_CACHE_GAP; | |
261 | ||
262 | c->boundaries = | |
263 | (struct boundary *) xrealloc (c->boundaries, | |
264 | ((min_size + c->cache_len) | |
265 | * sizeof (*c->boundaries))); | |
266 | ||
267 | /* Some systems don't provide a version of the copy routine that | |
268 | can be trusted to shift memory upward into an overlapping | |
269 | region. memmove isn't widely available. */ | |
270 | min_size -= gap_len; | |
271 | for (i = c->cache_len - 1; i >= gap_start; i--) | |
272 | { | |
273 | c->boundaries[i + min_size].pos = c->boundaries[i + gap_len].pos; | |
274 | c->boundaries[i + min_size].value = c->boundaries[i + gap_len].value; | |
275 | } | |
276 | ||
277 | gap_len = min_size; | |
278 | } | |
279 | ||
280 | /* Need we move the gap left? */ | |
281 | while (pos < gap_start) | |
282 | { | |
283 | gap_start--; | |
284 | ||
285 | /* Copy one region from before to after the gap, and | |
286 | convert its position to end-relative. */ | |
287 | c->boundaries[gap_start + gap_len].pos | |
288 | = c->boundaries[gap_start].pos + buffer_beg - buffer_end; | |
289 | c->boundaries[gap_start + gap_len].value | |
290 | = c->boundaries[gap_start].value; | |
291 | } | |
292 | ||
293 | /* Assign these back into the cache. */ | |
294 | c->gap_start = gap_start; | |
295 | c->gap_len = gap_len; | |
296 | } | |
297 | ||
298 | ||
299 | /* Insert a new boundary in cache C; it will have cache index INDEX, | |
300 | and have the specified POS and VALUE. */ | |
301 | static void | |
302 | insert_cache_boundary (c, index, pos, value) | |
303 | struct region_cache *c; | |
304 | int index; | |
305 | int pos, value; | |
306 | { | |
307 | /* index must be a valid cache index. */ | |
308 | if (index < 0 || index > c->cache_len) | |
309 | abort (); | |
310 | ||
311 | /* We must never want to insert something before the dummy first | |
312 | boundary. */ | |
313 | if (index == 0) | |
314 | abort (); | |
315 | ||
316 | /* We must only be inserting things in order. */ | |
317 | if (! (BOUNDARY_POS (c, index-1) < pos | |
318 | && (index == c->cache_len | |
319 | || pos < BOUNDARY_POS (c, index)))) | |
320 | abort (); | |
321 | ||
322 | /* The value must be different from the ones around it. However, we | |
323 | temporarily create boundaries that establish the same value as | |
324 | the subsequent boundary, so we're not going to flag that case. */ | |
325 | if (BOUNDARY_VALUE (c, index-1) == value) | |
326 | abort (); | |
327 | ||
328 | move_cache_gap (c, index, 1); | |
329 | ||
330 | c->boundaries[index].pos = pos - c->buffer_beg; | |
331 | c->boundaries[index].value = value; | |
332 | c->gap_start++; | |
333 | c->gap_len--; | |
334 | c->cache_len++; | |
335 | } | |
336 | ||
337 | ||
338 | /* Delete the i'th entry from cache C if START <= i < END. */ | |
339 | ||
340 | static void | |
341 | delete_cache_boundaries (c, start, end) | |
342 | struct region_cache *c; | |
343 | int start, end; | |
344 | { | |
345 | int len = end - start; | |
346 | ||
347 | /* Gotta be in range. */ | |
348 | if (start < 0 | |
349 | || end > c->cache_len) | |
350 | abort (); | |
351 | ||
352 | /* Gotta be in order. */ | |
353 | if (start > end) | |
354 | abort (); | |
355 | ||
356 | /* Can't delete the dummy entry. */ | |
357 | if (start == 0 | |
358 | && end >= 1) | |
359 | abort (); | |
360 | ||
361 | /* Minimize gap motion. If we're deleting nothing, do nothing. */ | |
362 | if (len == 0) | |
363 | ; | |
364 | /* If the gap is before the region to delete, delete from the start | |
365 | forward. */ | |
366 | else if (c->gap_start <= start) | |
367 | { | |
368 | move_cache_gap (c, start, 0); | |
369 | c->gap_len += len; | |
370 | } | |
371 | /* If the gap is after the region to delete, delete from the end | |
372 | backward. */ | |
373 | else if (end <= c->gap_start) | |
374 | { | |
375 | move_cache_gap (c, end, 0); | |
376 | c->gap_start -= len; | |
377 | c->gap_len += len; | |
378 | } | |
379 | /* If the gap is in the region to delete, just expand it. */ | |
380 | else | |
381 | { | |
382 | c->gap_start = start; | |
383 | c->gap_len += len; | |
384 | } | |
385 | ||
386 | c->cache_len -= len; | |
387 | } | |
388 | ||
389 | ||
390 | \f | |
391 | /* Set the value for a region. */ | |
392 | ||
393 | /* Set the value in cache C for the region START..END to VALUE. */ | |
394 | static void | |
395 | set_cache_region (c, start, end, value) | |
396 | struct region_cache *c; | |
397 | int start, end; | |
398 | int value; | |
399 | { | |
400 | if (start > end) | |
401 | abort (); | |
402 | if (start < c->buffer_beg | |
403 | || end > c->buffer_end) | |
404 | abort (); | |
405 | ||
406 | /* Eliminate this case; then we can assume that start and end-1 are | |
407 | both the locations of real characters in the buffer. */ | |
408 | if (start == end) | |
409 | return; | |
410 | ||
411 | { | |
412 | /* We need to make sure that there are no boundaries in the area | |
413 | between start to end; the whole area will have the same value, | |
414 | so those boundaries will not be necessary. | |
415 | ||
416 | Let start_ix be the cache index of the boundary governing the | |
417 | first character of start..end, and let end_ix be the cache | |
418 | index of the earliest boundary after the last character in | |
419 | start..end. (This tortured terminology is intended to answer | |
420 | all the "< or <=?" sort of questions.) */ | |
421 | int start_ix = find_cache_boundary (c, start); | |
422 | int end_ix = find_cache_boundary (c, end - 1) + 1; | |
423 | ||
424 | /* We must remember the value established by the last boundary | |
425 | before end; if that boundary's domain stretches beyond end, | |
426 | we'll need to create a new boundary at end, and that boundary | |
427 | must have that remembered value. */ | |
428 | int value_at_end = BOUNDARY_VALUE (c, end_ix - 1); | |
429 | ||
430 | /* Delete all boundaries strictly within start..end; this means | |
431 | those whose indices are between start_ix (exclusive) and end_ix | |
432 | (exclusive). */ | |
433 | delete_cache_boundaries (c, start_ix + 1, end_ix); | |
434 | ||
435 | /* Make sure we have the right value established going in to | |
436 | start..end from the left, and no unnecessary boundaries. */ | |
437 | if (BOUNDARY_POS (c, start_ix) == start) | |
438 | { | |
439 | /* Is this boundary necessary? If no, remove it; if yes, set | |
440 | its value. */ | |
441 | if (start_ix > 0 | |
442 | && BOUNDARY_VALUE (c, start_ix - 1) == value) | |
443 | { | |
444 | delete_cache_boundaries (c, start_ix, start_ix + 1); | |
445 | start_ix--; | |
446 | } | |
447 | else | |
448 | SET_BOUNDARY_VALUE (c, start_ix, value); | |
449 | } | |
450 | else | |
451 | { | |
452 | /* Do we need to add a new boundary here? */ | |
453 | if (BOUNDARY_VALUE (c, start_ix) != value) | |
454 | { | |
455 | insert_cache_boundary (c, start_ix + 1, start, value); | |
456 | start_ix++; | |
457 | } | |
458 | } | |
459 | ||
460 | /* This is equivalent to letting end_ix float (like a buffer | |
461 | marker does) with the insertions and deletions we may have | |
462 | done. */ | |
463 | end_ix = start_ix + 1; | |
464 | ||
465 | /* Make sure we have the correct value established as we leave | |
466 | start..end to the right. */ | |
467 | if (end == c->buffer_end) | |
468 | /* There is no text after start..end; nothing to do. */ | |
469 | ; | |
470 | else if (end_ix >= c->cache_len | |
471 | || end < BOUNDARY_POS (c, end_ix)) | |
472 | { | |
473 | /* There is no boundary at end, but we may need one. */ | |
474 | if (value_at_end != value) | |
475 | insert_cache_boundary (c, end_ix, end, value_at_end); | |
476 | } | |
477 | else | |
478 | { | |
479 | /* There is a boundary at end; should it be there? */ | |
480 | if (value == BOUNDARY_VALUE (c, end_ix)) | |
481 | delete_cache_boundaries (c, end_ix, end_ix + 1); | |
482 | } | |
483 | } | |
484 | } | |
485 | ||
486 | ||
487 | \f | |
488 | /* Interface: Invalidating the cache. Private: Re-validating the cache. */ | |
489 | ||
490 | /* Indicate that a section of BUF has changed, to invalidate CACHE. | |
491 | HEAD is the number of chars unchanged at the beginning of the buffer. | |
492 | TAIL is the number of chars unchanged at the end of the buffer. | |
493 | NOTE: this is *not* the same as the ending position of modified | |
494 | region. | |
495 | (This way of specifying regions makes more sense than absolute | |
496 | buffer positions in the presence of insertions and deletions; the | |
497 | args to pass are the same before and after such an operation.) */ | |
498 | void | |
499 | invalidate_region_cache (buf, c, head, tail) | |
500 | struct buffer *buf; | |
501 | struct region_cache *c; | |
502 | int head, tail; | |
503 | { | |
504 | /* Let chead = c->beg_unchanged, and | |
505 | ctail = c->end_unchanged. | |
506 | If z-tail < beg+chead by a large amount, or | |
507 | z-ctail < beg+head by a large amount, | |
508 | ||
509 | then cutting back chead and ctail to head and tail would lose a | |
510 | lot of information that we could preserve by revalidating the | |
511 | cache before processing this invalidation. Losing that | |
512 | information may be more costly than revalidating the cache now. | |
513 | So go ahead and call revalidate_region_cache if it seems that it | |
514 | might be worthwhile. */ | |
515 | if (((BUF_BEG (buf) + c->beg_unchanged) - (BUF_Z (buf) - tail) | |
516 | > PRESERVE_THRESHOLD) | |
517 | || ((BUF_BEG (buf) + head) - (BUF_Z (buf) - c->end_unchanged) | |
518 | > PRESERVE_THRESHOLD)) | |
519 | revalidate_region_cache (buf, c); | |
520 | ||
521 | ||
522 | if (head < c->beg_unchanged) | |
523 | c->beg_unchanged = head; | |
524 | if (tail < c->end_unchanged) | |
525 | c->end_unchanged = tail; | |
526 | ||
527 | /* We now know nothing about the region between the unchanged head | |
528 | and the unchanged tail (call it the "modified region"), not even | |
529 | its length. | |
530 | ||
531 | If the modified region has shrunk in size (deletions do this), | |
532 | then the cache may now contain boundaries originally located in | |
533 | text that doesn't exist any more. | |
534 | ||
535 | If the modified region has increased in size (insertions do | |
536 | this), then there may now be boundaries in the modified region | |
537 | whose positions are wrong. | |
538 | ||
539 | Even calling BOUNDARY_POS on boundaries still in the unchanged | |
540 | head or tail may well give incorrect answers now, since | |
541 | c->buffer_beg and c->buffer_end may well be wrong now. (Well, | |
542 | okay, c->buffer_beg never changes, so boundaries in the unchanged | |
543 | head will still be okay. But it's the principle of the thing.) | |
544 | ||
545 | So things are generally a mess. | |
546 | ||
547 | But we don't clean up this mess here; that would be expensive, | |
548 | and this function gets called every time any buffer modification | |
549 | occurs. Rather, we can clean up everything in one swell foop, | |
550 | accounting for all the modifications at once, by calling | |
551 | revalidate_region_cache before we try to consult the cache the | |
552 | next time. */ | |
553 | } | |
554 | ||
555 | ||
556 | /* Clean out any cache entries applying to the modified region, and | |
557 | make the positions of the remaining entries accurate again. | |
558 | ||
559 | After calling this function, the mess described in the comment in | |
560 | invalidate_region_cache is cleaned up. | |
561 | ||
562 | This function operates by simply throwing away everything it knows | |
563 | about the modified region. It doesn't care exactly which | |
564 | insertions and deletions took place; it just tosses it all. | |
565 | ||
566 | For example, if you insert a single character at the beginning of | |
567 | the buffer, and a single character at the end of the buffer (for | |
568 | example), without calling this function in between the two | |
569 | insertions, then the entire cache will be freed of useful | |
570 | information. On the other hand, if you do manage to call this | |
571 | function in between the two insertions, then the modified regions | |
572 | will be small in both cases, no information will be tossed, and the | |
573 | cache will know that it doesn't have knowledge of the first and | |
574 | last characters any more. | |
575 | ||
576 | Calling this function may be expensive; it does binary searches in | |
577 | the cache, and causes cache gap motion. */ | |
578 | ||
579 | static void | |
580 | revalidate_region_cache (buf, c) | |
581 | struct buffer *buf; | |
582 | struct region_cache *c; | |
583 | { | |
584 | /* The boundaries now in the cache are expressed relative to the | |
585 | buffer_beg and buffer_end values stored in the cache. Now, | |
586 | buffer_beg and buffer_end may not be the same as BUF_BEG (buf) | |
587 | and BUF_Z (buf), so we have two different "bases" to deal with | |
588 | --- the cache's, and the buffer's. */ | |
589 | ||
590 | /* If the entire buffer is still valid, don't waste time. Yes, this | |
591 | should be a >, not a >=; think about what beg_unchanged and | |
592 | end_unchanged get set to when the only change has been an | |
593 | insertion. */ | |
594 | if (c->buffer_beg + c->beg_unchanged | |
595 | > c->buffer_end - c->end_unchanged) | |
596 | return; | |
597 | ||
598 | /* If all the text we knew about as of the last cache revalidation | |
599 | is still there, then all of the information in the cache is still | |
600 | valid. Because c->buffer_beg and c->buffer_end are out-of-date, | |
601 | the modified region appears from the cache's point of view to be | |
602 | a null region located someplace in the buffer. | |
603 | ||
604 | Now, invalidating that empty string will have no actual affect on | |
605 | the cache; instead, we need to update the cache's basis first | |
606 | (which will give the modified region the same size in the cache | |
607 | as it has in the buffer), and then invalidate the modified | |
608 | region. */ | |
609 | if (c->buffer_beg + c->beg_unchanged | |
610 | == c->buffer_end - c->end_unchanged) | |
611 | { | |
612 | /* Move the gap so that all the boundaries in the unchanged head | |
613 | are expressed beg-relative, and all the boundaries in the | |
614 | unchanged tail are expressed end-relative. That done, we can | |
615 | plug in the new buffer beg and end, and all the positions | |
616 | will be accurate. | |
617 | ||
618 | The boundary which has jurisdiction over the modified region | |
619 | should be left before the gap. */ | |
620 | move_cache_gap (c, | |
621 | (find_cache_boundary (c, (c->buffer_beg | |
622 | + c->beg_unchanged)) | |
623 | + 1), | |
624 | 0); | |
625 | ||
626 | c->buffer_beg = BUF_BEG (buf); | |
627 | c->buffer_end = BUF_Z (buf); | |
628 | ||
629 | /* Now that the cache's basis has been changed, the modified | |
630 | region actually takes up some space in the cache, so we can | |
631 | invalidate it. */ | |
632 | set_cache_region (c, | |
633 | c->buffer_beg + c->beg_unchanged, | |
634 | c->buffer_end - c->end_unchanged, | |
635 | 0); | |
636 | } | |
637 | ||
638 | /* Otherwise, there is a non-empty region in the cache which | |
639 | corresponds to the modified region of the buffer. */ | |
640 | else | |
641 | { | |
642 | int modified_ix; | |
643 | ||
644 | /* These positions are correct, relative to both the cache basis | |
645 | and the buffer basis. */ | |
646 | set_cache_region (c, | |
647 | c->buffer_beg + c->beg_unchanged, | |
648 | c->buffer_end - c->end_unchanged, | |
649 | 0); | |
650 | ||
651 | /* Now the cache contains only boundaries that are in the | |
652 | unchanged head and tail; we've disposed of any boundaries | |
653 | whose positions we can't be sure of given the information | |
654 | we've saved. | |
655 | ||
656 | If we put the cache gap between the unchanged head and the | |
657 | unchanged tail, we can adjust all the boundary positions at | |
658 | once, simply by setting buffer_beg and buffer_end. | |
659 | ||
660 | The boundary which has jurisdiction over the modified region | |
661 | should be left before the gap. */ | |
662 | modified_ix = | |
663 | find_cache_boundary (c, (c->buffer_beg + c->beg_unchanged)) + 1; | |
664 | move_cache_gap (c, modified_ix, 0); | |
665 | ||
666 | c->buffer_beg = BUF_BEG (buf); | |
667 | c->buffer_end = BUF_Z (buf); | |
668 | ||
669 | /* Now, we may have shrunk the buffer when we changed the basis, | |
670 | and brought the boundaries we created for the start and end | |
671 | of the modified region together, giving them the same | |
672 | position. If that's the case, we should collapse them into | |
673 | one boundary. Or we may even delete them both, if the values | |
674 | before and after them are the same. */ | |
675 | if (modified_ix < c->cache_len | |
676 | && (BOUNDARY_POS (c, modified_ix - 1) | |
677 | == BOUNDARY_POS (c, modified_ix))) | |
678 | { | |
679 | int value_after = BOUNDARY_VALUE (c, modified_ix); | |
680 | ||
681 | /* Should we remove both of the boundaries? Yes, if the | |
682 | latter boundary is now establishing the same value that | |
683 | the former boundary's predecessor does. */ | |
684 | if (modified_ix - 1 > 0 | |
685 | && value_after == BOUNDARY_VALUE (c, modified_ix - 2)) | |
686 | delete_cache_boundaries (c, modified_ix - 1, modified_ix + 1); | |
687 | else | |
688 | { | |
689 | /* We do need a boundary here; collapse the two | |
690 | boundaries into one. */ | |
691 | SET_BOUNDARY_VALUE (c, modified_ix - 1, value_after); | |
692 | delete_cache_boundaries (c, modified_ix, modified_ix + 1); | |
693 | } | |
694 | } | |
695 | } | |
696 | ||
697 | /* Now the entire cache is valid. */ | |
698 | c->beg_unchanged | |
699 | = c->end_unchanged | |
700 | = c->buffer_end - c->buffer_beg; | |
701 | } | |
702 | ||
703 | \f | |
704 | /* Interface: Adding information to the cache. */ | |
705 | ||
706 | /* Assert that the region of BUF between START and END (absolute | |
707 | buffer positions) is "known," for the purposes of CACHE (e.g. "has | |
708 | no newlines", in the case of the line cache). */ | |
709 | void | |
710 | know_region_cache (buf, c, start, end) | |
711 | struct buffer *buf; | |
712 | struct region_cache *c; | |
713 | int start, end; | |
714 | { | |
715 | revalidate_region_cache (buf, c); | |
716 | ||
717 | set_cache_region (c, start, end, 1); | |
718 | } | |
719 | ||
720 | \f | |
721 | /* Interface: using the cache. */ | |
722 | ||
723 | /* Return true if the text immediately after POS in BUF is known, for | |
724 | the purposes of CACHE. If NEXT is non-zero, set *NEXT to the nearest | |
725 | position after POS where the knownness changes. */ | |
726 | int | |
727 | region_cache_forward (buf, c, pos, next) | |
728 | struct buffer *buf; | |
729 | struct region_cache *c; | |
730 | int pos; | |
731 | int *next; | |
732 | { | |
733 | revalidate_region_cache (buf, c); | |
734 | ||
735 | { | |
736 | int i = find_cache_boundary (c, pos); | |
737 | int i_value = BOUNDARY_VALUE (c, i); | |
738 | int j; | |
739 | ||
740 | /* Beyond the end of the buffer is unknown, by definition. */ | |
741 | if (pos >= BUF_Z (buf)) | |
742 | { | |
743 | if (next) *next = BUF_Z (buf); | |
744 | i_value = 0; | |
745 | } | |
746 | else if (next) | |
747 | { | |
748 | /* Scan forward from i to find the next differing position. */ | |
749 | for (j = i + 1; j < c->cache_len; j++) | |
750 | if (BOUNDARY_VALUE (c, j) != i_value) | |
751 | break; | |
752 | ||
753 | if (j < c->cache_len) | |
754 | *next = BOUNDARY_POS (c, j); | |
755 | else | |
756 | *next = BUF_Z (buf); | |
757 | } | |
758 | ||
759 | return i_value; | |
760 | } | |
761 | } | |
762 | ||
763 | /* Return true if the text immediately before POS in BUF is known, for | |
764 | the purposes of CACHE. If NEXT is non-zero, set *NEXT to the nearest | |
765 | position before POS where the knownness changes. */ | |
766 | int region_cache_backward (buf, c, pos, next) | |
767 | struct buffer *buf; | |
768 | struct region_cache *c; | |
769 | int pos; | |
770 | int *next; | |
771 | { | |
772 | revalidate_region_cache (buf, c); | |
773 | ||
774 | /* Before the beginning of the buffer is unknown, by | |
775 | definition. */ | |
776 | if (pos <= BUF_BEG (buf)) | |
777 | { | |
778 | if (next) *next = BUF_BEG (buf); | |
779 | return 0; | |
780 | } | |
781 | ||
782 | { | |
783 | int i = find_cache_boundary (c, pos - 1); | |
784 | int i_value = BOUNDARY_VALUE (c, i); | |
785 | int j; | |
786 | ||
787 | if (next) | |
788 | { | |
789 | /* Scan backward from i to find the next differing position. */ | |
790 | for (j = i - 1; j >= 0; j--) | |
791 | if (BOUNDARY_VALUE (c, j) != i_value) | |
792 | break; | |
793 | ||
794 | if (j >= 0) | |
795 | *next = BOUNDARY_POS (c, j + 1); | |
796 | else | |
797 | *next = BUF_BEG (buf); | |
798 | } | |
799 | ||
800 | return i_value; | |
801 | } | |
802 | } | |
803 | ||
804 | \f | |
805 | /* Debugging: pretty-print a cache to the standard error output. */ | |
806 | ||
807 | void | |
808 | pp_cache (c) | |
809 | struct region_cache *c; | |
810 | { | |
811 | int i; | |
812 | int beg_u = c->buffer_beg + c->beg_unchanged; | |
813 | int end_u = c->buffer_end - c->end_unchanged; | |
814 | ||
815 | fprintf (stderr, | |
816 | "basis: %d..%d modified: %d..%d\n", | |
817 | c->buffer_beg, c->buffer_end, | |
818 | beg_u, end_u); | |
819 | ||
820 | for (i = 0; i < c->cache_len; i++) | |
821 | { | |
822 | int pos = BOUNDARY_POS (c, i); | |
823 | ||
824 | putc (((pos < beg_u) ? 'v' | |
825 | : (pos == beg_u) ? '-' | |
826 | : ' '), | |
827 | stderr); | |
828 | putc (((pos > end_u) ? '^' | |
829 | : (pos == end_u) ? '-' | |
830 | : ' '), | |
831 | stderr); | |
832 | fprintf (stderr, "%d : %d\n", pos, BOUNDARY_VALUE (c, i)); | |
833 | } | |
834 | } |