1 /* Caching facts about regions of the buffer, for optimization.
2 
3 Copyright (C) 1985-1989, 1993, 1995, 2001-2021 Free Software Foundation,
4 Inc.
5 
6 This file is part of GNU Emacs.
7 
8 GNU Emacs is free software: you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation, either version 3 of the License, or (at
11 your option) any later version.
12 
13 GNU Emacs is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16 GNU General Public License for more details.
17 
18 You should have received a copy of the GNU General Public License
19 along with GNU Emacs.  If not, see <https://www.gnu.org/licenses/>.  */
20 
21 
22 #include <config.h>
23 #include <stdio.h>
24 
25 #include "lisp.h"
26 #include "buffer.h"
27 #include "region-cache.h"
28 
29 
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 {
65   ptrdiff_t pos;
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.  */
75   ptrdiff_t gap_start, gap_len;
76 
77   /* The number of elements allocated to boundaries, not including the
78      gap.  */
79   ptrdiff_t cache_len;
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.  */
84   ptrdiff_t beg_unchanged, end_unchanged;
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.  */
94   ptrdiff_t buffer_beg, buffer_end;
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 
125 static void revalidate_region_cache (struct buffer *buf, struct region_cache *c);
126 
127 
128 /* Interface: Allocating, initializing, and disposing of region caches.  */
129 
130 struct region_cache *
new_region_cache(void)131 new_region_cache (void)
132 {
133   struct region_cache *c = xmalloc (sizeof *c);
134 
135   c->gap_start = 0;
136   c->gap_len = NEW_CACHE_GAP;
137   c->cache_len = 0;
138   c->boundaries = xmalloc ((c->gap_len + c->cache_len)
139 			   * sizeof (*c->boundaries));
140 
141   c->beg_unchanged = 0;
142   c->end_unchanged = 0;
143   c->buffer_beg = BEG;
144   c->buffer_end = BEG;
145 
146   /* Insert the boundary for the buffer start.  */
147   c->cache_len++;
148   c->gap_len--;
149   c->gap_start++;
150   c->boundaries[0].pos   = 0;  /* from buffer_beg */
151   c->boundaries[0].value = 0;
152 
153   return c;
154 }
155 
156 void
free_region_cache(struct region_cache * c)157 free_region_cache (struct region_cache *c)
158 {
159   xfree (c->boundaries);
160   xfree (c);
161 }
162 
163 
164 /* Finding positions in the cache.  */
165 
166 /* Return the index of the last boundary in cache C at or before POS.
167    In other words, return the boundary that specifies the value for
168    the region POS..(POS + 1).
169 
170    This operation should be logarithmic in the number of cache
171    entries.  It would be nice if it took advantage of locality of
172    reference, too, by searching entries near the last entry found.  */
173 static ptrdiff_t
find_cache_boundary(struct region_cache * c,ptrdiff_t pos)174 find_cache_boundary (struct region_cache *c, ptrdiff_t pos)
175 {
176   ptrdiff_t low = 0, high = c->cache_len;
177 
178   while (low + 1 < high)
179     {
180       /* mid is always a valid index, because low < high and ">> 1"
181          rounds down.  */
182       ptrdiff_t mid = (low >> 1) + (high >> 1) + (low & high & 1);
183       ptrdiff_t boundary = BOUNDARY_POS (c, mid);
184 
185       if (pos < boundary)
186         high = mid;
187       else
188         low = mid;
189     }
190 
191   /* Some testing.  */
192   eassert (!(BOUNDARY_POS (c, low) > pos
193 	     || (low + 1 < c->cache_len
194 		 && BOUNDARY_POS (c, low + 1) <= pos)));
195 
196   return low;
197 }
198 
199 
200 
201 /* Moving the cache gap around, inserting, and deleting.  */
202 
203 
204 /* Move the gap of cache C to index POS, and make sure it has space
205    for at least MIN_SIZE boundaries.  */
206 static void
move_cache_gap(struct region_cache * c,ptrdiff_t pos,ptrdiff_t min_size)207 move_cache_gap (struct region_cache *c, ptrdiff_t pos, ptrdiff_t min_size)
208 {
209   /* Copy these out of the cache and into registers.  */
210   ptrdiff_t gap_start = c->gap_start;
211   ptrdiff_t gap_len = c->gap_len;
212   ptrdiff_t buffer_beg = c->buffer_beg;
213   ptrdiff_t buffer_end = c->buffer_end;
214 
215   /* We mustn't ever try to put the gap before the dummy start
216      boundary.  That must always be start-relative.  */
217   eassert (0 < pos && pos <= c->cache_len);
218 
219   /* Need we move the gap right?  */
220   while (gap_start < pos)
221     {
222       /* Copy one boundary from after to before the gap, and
223          convert its position to start-relative.  */
224       c->boundaries[gap_start].pos
225         = (buffer_end
226            + c->boundaries[gap_start + gap_len].pos
227            - buffer_beg);
228       c->boundaries[gap_start].value
229         = c->boundaries[gap_start + gap_len].value;
230       gap_start++;
231     }
232 
233   /* To enlarge the gap, we need to re-allocate the boundary array, and
234      then shift the area after the gap to the new end.  Since the cost
235      is proportional to the amount of stuff after the gap, we do the
236      enlargement here, after a right shift but before a left shift,
237      when the portion after the gap is smallest.  */
238   if (gap_len < min_size)
239     {
240       ptrdiff_t i, nboundaries = c->cache_len;
241 
242       c->boundaries =
243 	xpalloc (c->boundaries, &nboundaries, min_size - gap_len, -1,
244 		 sizeof *c->boundaries);
245 
246       /* Some systems don't provide a version of the copy routine that
247          can be trusted to shift memory upward into an overlapping
248          region.  memmove isn't widely available.  */
249       min_size = nboundaries - c->cache_len - gap_len;
250       for (i = c->cache_len - 1; i >= gap_start; i--)
251         {
252           c->boundaries[i + min_size].pos   = c->boundaries[i + gap_len].pos;
253           c->boundaries[i + min_size].value = c->boundaries[i + gap_len].value;
254         }
255 
256       gap_len = min_size;
257     }
258 
259   /* Need we move the gap left?  */
260   while (pos < gap_start)
261     {
262       gap_start--;
263 
264       /* Copy one region from before to after the gap, and
265          convert its position to end-relative.  */
266       c->boundaries[gap_start + gap_len].pos
267         = c->boundaries[gap_start].pos + buffer_beg - buffer_end;
268       c->boundaries[gap_start + gap_len].value
269         = c->boundaries[gap_start].value;
270     }
271 
272   /* Assign these back into the cache.  */
273   c->gap_start = gap_start;
274   c->gap_len  = gap_len;
275 }
276 
277 
278 /* Insert a new boundary in cache C; it will have cache index I,
279    and have the specified POS and VALUE.  */
280 static void
insert_cache_boundary(struct region_cache * c,ptrdiff_t i,ptrdiff_t pos,int value)281 insert_cache_boundary (struct region_cache *c, ptrdiff_t i, ptrdiff_t pos,
282 		       int value)
283 {
284   /* I must be a valid cache index, and we must never want
285      to insert something before the dummy first boundary.  */
286   eassert (0 < i && i <= c->cache_len);
287 
288   /* We must only be inserting things in order.  */
289   eassert ((BOUNDARY_POS (c, i - 1) < pos
290 	    && (i == c->cache_len
291 		|| pos < BOUNDARY_POS (c, i))));
292 
293   /* The value must be different from the ones around it.  However, we
294      temporarily create boundaries that establish the same value as
295      the subsequent boundary, so we're not going to flag that case.  */
296   eassert (BOUNDARY_VALUE (c, i - 1) != value);
297 
298   move_cache_gap (c, i, 1);
299 
300   c->boundaries[i].pos = pos - c->buffer_beg;
301   c->boundaries[i].value = value;
302   c->gap_start++;
303   c->gap_len--;
304   c->cache_len++;
305 }
306 
307 
308 /* Delete the i'th entry from cache C if START <= i < END.  */
309 
310 static void
delete_cache_boundaries(struct region_cache * c,ptrdiff_t start,ptrdiff_t end)311 delete_cache_boundaries (struct region_cache *c,
312 			 ptrdiff_t start, ptrdiff_t end)
313 {
314   ptrdiff_t len = end - start;
315 
316   /* Gotta be in range.  */
317   eassert (0 <= start && end <= c->cache_len);
318 
319   /* Gotta be in order.  */
320   eassert (start <= end);
321 
322   /* Can't delete the dummy entry.  */
323   eassert (!(start == 0 && end >= 1));
324 
325   /* Minimize gap motion.  If we're deleting nothing, do nothing.  */
326   if (len == 0)
327     ;
328   /* If the gap is before the region to delete, delete from the start
329      forward.  */
330   else if (c->gap_start <= start)
331     {
332       move_cache_gap (c, start, 0);
333       c->gap_len += len;
334     }
335   /* If the gap is after the region to delete, delete from the end
336      backward.  */
337   else if (end <= c->gap_start)
338     {
339       move_cache_gap (c, end, 0);
340       c->gap_start -= len;
341       c->gap_len   += len;
342     }
343   /* If the gap is in the region to delete, just expand it.  */
344   else
345     {
346       c->gap_start = start;
347       c->gap_len   += len;
348     }
349 
350   c->cache_len -= len;
351 }
352 
353 
354 
355 /* Set the value for a region.  */
356 
357 /* Set the value in cache C for the region START..END to VALUE.  */
358 static void
set_cache_region(struct region_cache * c,ptrdiff_t start,ptrdiff_t end,int value)359 set_cache_region (struct region_cache *c,
360 		  ptrdiff_t start, ptrdiff_t end, int value)
361 {
362   eassert (start <= end);
363   eassert (c->buffer_beg <= start && end <= c->buffer_end);
364 
365   /* Eliminate this case; then we can assume that start and end-1 are
366      both the locations of real characters in the buffer.  */
367   if (start == end)
368     return;
369 
370   {
371     /* We need to make sure that there are no boundaries in the area
372        between start to end; the whole area will have the same value,
373        so those boundaries will not be necessary.
374 
375        Let start_ix be the cache index of the boundary governing the
376        first character of start..end, and let end_ix be the cache
377        index of the earliest boundary after the last character in
378        start..end.  (This tortured terminology is intended to answer
379        all the "< or <=?" sort of questions.)  */
380     ptrdiff_t start_ix = find_cache_boundary (c, start);
381     ptrdiff_t end_ix   = find_cache_boundary (c, end - 1) + 1;
382 
383     /* We must remember the value established by the last boundary
384        before end; if that boundary's domain stretches beyond end,
385        we'll need to create a new boundary at end, and that boundary
386        must have that remembered value.  */
387     int value_at_end = BOUNDARY_VALUE (c, end_ix - 1);
388 
389     /* Delete all boundaries strictly within start..end; this means
390        those whose indices are between start_ix (exclusive) and end_ix
391        (exclusive).  */
392     delete_cache_boundaries (c, start_ix + 1, end_ix);
393 
394     /* Make sure we have the right value established going in to
395        start..end from the left, and no unnecessary boundaries.  */
396     if (BOUNDARY_POS (c, start_ix) == start)
397       {
398         /* Is this boundary necessary?  If no, remove it; if yes, set
399            its value.  */
400         if (start_ix > 0
401             && BOUNDARY_VALUE (c, start_ix - 1) == value)
402           {
403             delete_cache_boundaries (c, start_ix, start_ix + 1);
404             start_ix--;
405           }
406         else
407           SET_BOUNDARY_VALUE (c, start_ix, value);
408       }
409     else
410       {
411         /* Do we need to add a new boundary here?  */
412         if (BOUNDARY_VALUE (c, start_ix) != value)
413           {
414             insert_cache_boundary (c, start_ix + 1, start, value);
415             start_ix++;
416           }
417       }
418 
419     /* This is equivalent to letting end_ix float (like a buffer
420        marker does) with the insertions and deletions we may have
421        done.  */
422     end_ix = start_ix + 1;
423 
424     /* Make sure we have the correct value established as we leave
425        start..end to the right.  */
426     if (end == c->buffer_end)
427       /* There is no text after start..end; nothing to do.  */
428       ;
429     else if (end_ix >= c->cache_len
430              || end < BOUNDARY_POS (c, end_ix))
431       {
432         /* There is no boundary at end, but we may need one.  */
433         if (value_at_end != value)
434           insert_cache_boundary (c, end_ix, end, value_at_end);
435       }
436     else
437       {
438         /* There is a boundary at end; should it be there?  */
439         if (value == BOUNDARY_VALUE (c, end_ix))
440           delete_cache_boundaries (c, end_ix, end_ix + 1);
441       }
442   }
443 }
444 
445 
446 
447 /* Interface: Invalidating the cache.  Private: Re-validating the cache.  */
448 
449 /* Indicate that a section of BUF has changed, to invalidate CACHE.
450    HEAD is the number of chars unchanged at the beginning of the buffer.
451    TAIL is the number of chars unchanged at the end of the buffer.
452       NOTE: this is *not* the same as the ending position of modified
453       region.
454    (This way of specifying regions makes more sense than absolute
455    buffer positions in the presence of insertions and deletions; the
456    args to pass are the same before and after such an operation.)  */
457 void
invalidate_region_cache(struct buffer * buf,struct region_cache * c,ptrdiff_t head,ptrdiff_t tail)458 invalidate_region_cache (struct buffer *buf, struct region_cache *c,
459 			 ptrdiff_t head, ptrdiff_t tail)
460 {
461   /* Let chead = c->beg_unchanged, and
462          ctail = c->end_unchanged.
463      If z-tail < beg+chead by a large amount, or
464         z-ctail < beg+head by a large amount,
465 
466      then cutting back chead and ctail to head and tail would lose a
467      lot of information that we could preserve by revalidating the
468      cache before processing this invalidation.  Losing that
469      information may be more costly than revalidating the cache now.
470      So go ahead and call revalidate_region_cache if it seems that it
471      might be worthwhile.  */
472   if (((BUF_BEG (buf) + c->beg_unchanged) - (BUF_Z (buf) - tail)
473        > PRESERVE_THRESHOLD)
474       || ((BUF_BEG (buf) + head) - (BUF_Z (buf) - c->end_unchanged)
475           > PRESERVE_THRESHOLD))
476     revalidate_region_cache (buf, c);
477 
478 
479   if (head < c->beg_unchanged)
480     c->beg_unchanged = head;
481   if (tail < c->end_unchanged)
482     c->end_unchanged = tail;
483 
484   /* We now know nothing about the region between the unchanged head
485      and the unchanged tail (call it the "modified region"), not even
486      its length.
487 
488      If the modified region has shrunk in size (deletions do this),
489      then the cache may now contain boundaries originally located in
490      text that doesn't exist any more.
491 
492      If the modified region has increased in size (insertions do
493      this), then there may now be boundaries in the modified region
494      whose positions are wrong.
495 
496      Even calling BOUNDARY_POS on boundaries still in the unchanged
497      head or tail may well give incorrect answers now, since
498      c->buffer_beg and c->buffer_end may well be wrong now.  (Well,
499      okay, c->buffer_beg never changes, so boundaries in the unchanged
500      head will still be okay.  But it's the principle of the thing.)
501 
502      So things are generally a mess.
503 
504      But we don't clean up this mess here; that would be expensive,
505      and this function gets called every time any buffer modification
506      occurs.  Rather, we can clean up everything in one swell foop,
507      accounting for all the modifications at once, by calling
508      revalidate_region_cache before we try to consult the cache the
509      next time.  */
510 }
511 
512 
513 /* Clean out any cache entries applying to the modified region, and
514    make the positions of the remaining entries accurate again.
515 
516    After calling this function, the mess described in the comment in
517    invalidate_region_cache is cleaned up.
518 
519    This function operates by simply throwing away everything it knows
520    about the modified region.  It doesn't care exactly which
521    insertions and deletions took place; it just tosses it all.
522 
523    For example, if you insert a single character at the beginning of
524    the buffer, and a single character at the end of the buffer (for
525    example), without calling this function in between the two
526    insertions, then the entire cache will be freed of useful
527    information.  On the other hand, if you do manage to call this
528    function in between the two insertions, then the modified regions
529    will be small in both cases, no information will be tossed, and the
530    cache will know that it doesn't have knowledge of the first and
531    last characters any more.
532 
533    Calling this function may be expensive; it does binary searches in
534    the cache, and causes cache gap motion.  */
535 
536 static void
revalidate_region_cache(struct buffer * buf,struct region_cache * c)537 revalidate_region_cache (struct buffer *buf, struct region_cache *c)
538 {
539   /* The boundaries now in the cache are expressed relative to the
540      buffer_beg and buffer_end values stored in the cache.  Now,
541      buffer_beg and buffer_end may not be the same as BUF_BEG (buf)
542      and BUF_Z (buf), so we have two different "bases" to deal with
543      --- the cache's, and the buffer's.  */
544 
545   /* If the entire buffer is still valid, don't waste time.  Yes, this
546      should be a >, not a >=; think about what beg_unchanged and
547      end_unchanged get set to when the only change has been an
548      insertion.  */
549   if (c->buffer_beg + c->beg_unchanged
550       > c->buffer_end - c->end_unchanged)
551     return;
552 
553   /* If all the text we knew about as of the last cache revalidation
554      is still there, then all of the information in the cache is still
555      valid.  Because c->buffer_beg and c->buffer_end are out-of-date,
556      the modified region appears from the cache's point of view to be
557      a null region located someplace in the buffer.
558 
559      Now, invalidating that empty string will have no actual affect on
560      the cache; instead, we need to update the cache's basis first
561      (which will give the modified region the same size in the cache
562      as it has in the buffer), and then invalidate the modified
563      region. */
564   if (c->buffer_beg + c->beg_unchanged
565       == c->buffer_end - c->end_unchanged)
566     {
567       /* Move the gap so that all the boundaries in the unchanged head
568          are expressed beg-relative, and all the boundaries in the
569          unchanged tail are expressed end-relative.  That done, we can
570          plug in the new buffer beg and end, and all the positions
571          will be accurate.
572 
573          The boundary which has jurisdiction over the modified region
574          should be left before the gap.  */
575       move_cache_gap (c,
576                       (find_cache_boundary (c, (c->buffer_beg
577                                                 + c->beg_unchanged))
578                        + 1),
579                       0);
580 
581       c->buffer_beg = BUF_BEG (buf);
582       c->buffer_end = BUF_Z   (buf);
583 
584       /* Now that the cache's basis has been changed, the modified
585          region actually takes up some space in the cache, so we can
586          invalidate it.  */
587       set_cache_region (c,
588                         c->buffer_beg + c->beg_unchanged,
589                         c->buffer_end - c->end_unchanged,
590                         0);
591     }
592 
593   /* Otherwise, there is a non-empty region in the cache which
594      corresponds to the modified region of the buffer.  */
595   else
596     {
597       ptrdiff_t modified_ix;
598 
599       /* These positions are correct, relative to both the cache basis
600          and the buffer basis.  */
601       set_cache_region (c,
602                         c->buffer_beg + c->beg_unchanged,
603                         c->buffer_end - c->end_unchanged,
604                         0);
605 
606       /* Now the cache contains only boundaries that are in the
607          unchanged head and tail; we've disposed of any boundaries
608          whose positions we can't be sure of given the information
609          we've saved.
610 
611          If we put the cache gap between the unchanged head and the
612          unchanged tail, we can adjust all the boundary positions at
613          once, simply by setting buffer_beg and buffer_end.
614 
615          The boundary which has jurisdiction over the modified region
616          should be left before the gap.  */
617       modified_ix =
618         find_cache_boundary (c, (c->buffer_beg + c->beg_unchanged)) + 1;
619       move_cache_gap (c, modified_ix, 0);
620 
621       c->buffer_beg = BUF_BEG (buf);
622       c->buffer_end = BUF_Z   (buf);
623 
624       /* Now, we may have shrunk the buffer when we changed the basis,
625          and brought the boundaries we created for the start and end
626          of the modified region together, giving them the same
627          position.  If that's the case, we should collapse them into
628          one boundary.  Or we may even delete them both, if the values
629          before and after them are the same.  */
630       if (modified_ix < c->cache_len
631           && (BOUNDARY_POS (c, modified_ix - 1)
632               == BOUNDARY_POS (c, modified_ix)))
633         {
634           int value_after = BOUNDARY_VALUE (c, modified_ix);
635 
636           /* Should we remove both of the boundaries?  Yes, if the
637              latter boundary is now establishing the same value that
638              the former boundary's predecessor does.  */
639           if (modified_ix - 1 > 0
640               && value_after == BOUNDARY_VALUE (c, modified_ix - 2))
641             delete_cache_boundaries (c, modified_ix - 1, modified_ix + 1);
642           else
643             {
644               /* We do need a boundary here; collapse the two
645                  boundaries into one.  */
646               SET_BOUNDARY_VALUE (c, modified_ix - 1, value_after);
647               delete_cache_boundaries (c, modified_ix, modified_ix + 1);
648             }
649         }
650     }
651 
652   /* Now the entire cache is valid.  */
653   c->beg_unchanged
654     = c->end_unchanged
655       = c->buffer_end - c->buffer_beg;
656 }
657 
658 
659 /* Interface: Adding information to the cache.  */
660 
661 /* Assert that the region of BUF between START and END (absolute
662    buffer positions) is "known," for the purposes of CACHE (e.g. "has
663    no newlines", in the case of the line cache).  */
664 void
know_region_cache(struct buffer * buf,struct region_cache * c,ptrdiff_t start,ptrdiff_t end)665 know_region_cache (struct buffer *buf, struct region_cache *c,
666 		   ptrdiff_t start, ptrdiff_t end)
667 {
668   revalidate_region_cache (buf, c);
669 
670   set_cache_region (c, start, end, 1);
671 }
672 
673 
674 /* Interface: using the cache.  */
675 
676 /* Return the value for the text immediately after POS in BUF if the value
677    is known, for the purposes of CACHE, and return zero otherwise.
678    If NEXT is non-zero, set *NEXT to the nearest
679    position after POS where the knowledge changes.  */
680 int
region_cache_forward(struct buffer * buf,struct region_cache * c,ptrdiff_t pos,ptrdiff_t * next)681 region_cache_forward (struct buffer *buf, struct region_cache *c,
682 		      ptrdiff_t pos, ptrdiff_t *next)
683 {
684   revalidate_region_cache (buf, c);
685 
686   {
687     ptrdiff_t i = find_cache_boundary (c, pos);
688     int i_value = BOUNDARY_VALUE (c, i);
689     ptrdiff_t j;
690 
691     /* Beyond the end of the buffer is unknown, by definition.  */
692     if (pos >= BUF_Z (buf))
693       {
694         if (next) *next = BUF_Z (buf);
695         i_value = 0;
696       }
697     else if (next)
698       {
699         /* Scan forward from i to find the next differing position.  */
700         for (j = i + 1; j < c->cache_len; j++)
701           if (BOUNDARY_VALUE (c, j) != i_value)
702             break;
703 
704         if (j < c->cache_len)
705           *next = BOUNDARY_POS (c, j);
706         else
707           *next = BUF_Z (buf);
708       }
709 
710     return i_value;
711   }
712 }
713 
714 /* Return the value for the text immediately before POS in BUF if the
715    value is known, for the purposes of CACHE, and return zero
716    otherwise.  If NEXT is non-zero, set *NEXT to the nearest
717    position before POS where the knowledge changes.  */
718 int
region_cache_backward(struct buffer * buf,struct region_cache * c,ptrdiff_t pos,ptrdiff_t * next)719 region_cache_backward (struct buffer *buf, struct region_cache *c,
720 		       ptrdiff_t pos, ptrdiff_t *next)
721 {
722   revalidate_region_cache (buf, c);
723 
724   /* Before the beginning of the buffer is unknown, by
725      definition. */
726   if (pos <= BUF_BEG (buf))
727     {
728       if (next) *next = BUF_BEG (buf);
729       return 0;
730     }
731 
732   {
733     ptrdiff_t i = find_cache_boundary (c, pos - 1);
734     int i_value = BOUNDARY_VALUE (c, i);
735     ptrdiff_t j;
736 
737     if (next)
738       {
739         /* Scan backward from i to find the next differing position.  */
740         for (j = i - 1; j >= 0; j--)
741           if (BOUNDARY_VALUE (c, j) != i_value)
742             break;
743 
744         if (j >= 0)
745           *next = BOUNDARY_POS (c, j + 1);
746         else
747           *next = BUF_BEG (buf);
748       }
749 
750     return i_value;
751   }
752 }
753 
754 #ifdef ENABLE_CHECKING
755 
756 /* Debugging: pretty-print a cache to the standard error output.  */
757 
758 void pp_cache (struct region_cache *) EXTERNALLY_VISIBLE;
759 void
pp_cache(struct region_cache * c)760 pp_cache (struct region_cache *c)
761 {
762   ptrdiff_t beg_u = c->buffer_beg + c->beg_unchanged;
763   ptrdiff_t end_u = c->buffer_end - c->end_unchanged;
764 
765   fprintf (stderr,
766            "basis: %"pD"d..%"pD"d    modified: %"pD"d..%"pD"d\n",
767            c->buffer_beg, c->buffer_end,
768            beg_u, end_u);
769 
770   for (ptrdiff_t i = 0; i < c->cache_len; i++)
771     {
772       ptrdiff_t pos = BOUNDARY_POS (c, i);
773 
774       fprintf (stderr, "%c%c%"pD"d : %d\n",
775 	       pos < beg_u ? 'v' : pos == beg_u ? '-' : ' ',
776 	       pos > end_u ? '^' : pos == end_u ? '-' : ' ',
777 	       pos, BOUNDARY_VALUE (c, i));
778     }
779 }
780 
781 #endif /* ENABLE_CHECKING */
782