1 /* Caching facts about regions of the buffer, for optimization.
3 Copyright (C) 1985-1989, 1993, 1995, 2001-2011
4 Free Software Foundation, Inc.
6 This file is part of GNU Emacs.
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
11 (at your option) any later version.
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.
18 You should have received a copy of the GNU General Public License
19 along with GNU Emacs. If not, see <http://www.gnu.org/licenses/>. */
28 #include "region-cache.h"
31 /* Data structures. */
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.
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.
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.
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
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.
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. */
71 /* A sorted array of locations where the known-ness of the buffer
73 struct boundary
*boundaries
;
75 /* boundaries[gap_start ... gap_start + gap_len - 1] is the gap. */
76 ptrdiff_t gap_start
, gap_len
;
78 /* The number of elements allocated to boundaries, not including the
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 ptrdiff_t beg_unchanged
, end_unchanged
;
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.
93 Yes, buffer_beg is always 1. It's there for symmetry with
94 buffer_end and the BEG and BUF_BEG macros. */
95 ptrdiff_t buffer_beg
, buffer_end
;
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)
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)
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)))
117 /* How many elements to add to the gap when we resize the buffer. */
118 #define NEW_CACHE_GAP (40)
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)
126 static void revalidate_region_cache (struct buffer
*buf
, struct region_cache
*c
);
129 /* Interface: Allocating, initializing, and disposing of region caches. */
131 struct region_cache
*
132 new_region_cache (void)
134 struct region_cache
*c
135 = (struct region_cache
*) xmalloc (sizeof (struct region_cache
));
138 c
->gap_len
= NEW_CACHE_GAP
;
141 (struct boundary
*) xmalloc ((c
->gap_len
+ c
->cache_len
)
142 * sizeof (*c
->boundaries
));
144 c
->beg_unchanged
= 0;
145 c
->end_unchanged
= 0;
149 /* Insert the boundary for the buffer start. */
153 c
->boundaries
[0].pos
= 0; /* from buffer_beg */
154 c
->boundaries
[0].value
= 0;
160 free_region_cache (struct region_cache
*c
)
162 xfree (c
->boundaries
);
167 /* Finding positions in the cache. */
169 /* Return the index of the last boundary in cache C at or before POS.
170 In other words, return the boundary that specifies the value for
171 the region POS..(POS + 1).
173 This operation should be logarithmic in the number of cache
174 entries. It would be nice if it took advantage of locality of
175 reference, too, by searching entries near the last entry found. */
177 find_cache_boundary (struct region_cache
*c
, ptrdiff_t pos
)
179 ptrdiff_t low
= 0, high
= c
->cache_len
;
181 while (low
+ 1 < high
)
183 /* mid is always a valid index, because low < high and ">> 1"
185 ptrdiff_t mid
= (low
>> 1) + (high
>> 1) + (low
& high
& 1);
186 ptrdiff_t boundary
= BOUNDARY_POS (c
, mid
);
195 if (BOUNDARY_POS (c
, low
) > pos
196 || (low
+ 1 < c
->cache_len
197 && BOUNDARY_POS (c
, low
+ 1) <= pos
))
205 /* Moving the cache gap around, inserting, and deleting. */
208 /* Move the gap of cache C to index POS, and make sure it has space
209 for at least MIN_SIZE boundaries. */
211 move_cache_gap (struct region_cache
*c
, ptrdiff_t pos
, ptrdiff_t min_size
)
213 /* Copy these out of the cache and into registers. */
214 ptrdiff_t gap_start
= c
->gap_start
;
215 ptrdiff_t gap_len
= c
->gap_len
;
216 ptrdiff_t buffer_beg
= c
->buffer_beg
;
217 ptrdiff_t buffer_end
= c
->buffer_end
;
220 || pos
> c
->cache_len
)
223 /* We mustn't ever try to put the gap before the dummy start
224 boundary. That must always be start-relative. */
228 /* Need we move the gap right? */
229 while (gap_start
< pos
)
231 /* Copy one boundary from after to before the gap, and
232 convert its position to start-relative. */
233 c
->boundaries
[gap_start
].pos
235 + c
->boundaries
[gap_start
+ gap_len
].pos
237 c
->boundaries
[gap_start
].value
238 = c
->boundaries
[gap_start
+ gap_len
].value
;
242 /* To enlarge the gap, we need to re-allocate the boundary array, and
243 then shift the area after the gap to the new end. Since the cost
244 is proportional to the amount of stuff after the gap, we do the
245 enlargement here, after a right shift but before a left shift,
246 when the portion after the gap is smallest. */
247 if (gap_len
< min_size
)
252 xpalloc (c
->boundaries
, &c
->cache_len
, min_size
, -1,
253 sizeof *c
->boundaries
);
255 /* Some systems don't provide a version of the copy routine that
256 can be trusted to shift memory upward into an overlapping
257 region. memmove isn't widely available. */
259 for (i
= c
->cache_len
- 1; i
>= gap_start
; i
--)
261 c
->boundaries
[i
+ min_size
].pos
= c
->boundaries
[i
+ gap_len
].pos
;
262 c
->boundaries
[i
+ min_size
].value
= c
->boundaries
[i
+ gap_len
].value
;
268 /* Need we move the gap left? */
269 while (pos
< gap_start
)
273 /* Copy one region from before to after the gap, and
274 convert its position to end-relative. */
275 c
->boundaries
[gap_start
+ gap_len
].pos
276 = c
->boundaries
[gap_start
].pos
+ buffer_beg
- buffer_end
;
277 c
->boundaries
[gap_start
+ gap_len
].value
278 = c
->boundaries
[gap_start
].value
;
281 /* Assign these back into the cache. */
282 c
->gap_start
= gap_start
;
283 c
->gap_len
= gap_len
;
287 /* Insert a new boundary in cache C; it will have cache index I,
288 and have the specified POS and VALUE. */
290 insert_cache_boundary (struct region_cache
*c
, ptrdiff_t i
, ptrdiff_t pos
,
293 /* i must be a valid cache index. */
294 if (i
< 0 || i
> c
->cache_len
)
297 /* We must never want to insert something before the dummy first
302 /* We must only be inserting things in order. */
303 if (! (BOUNDARY_POS (c
, i
- 1) < pos
304 && (i
== c
->cache_len
305 || pos
< BOUNDARY_POS (c
, i
))))
308 /* The value must be different from the ones around it. However, we
309 temporarily create boundaries that establish the same value as
310 the subsequent boundary, so we're not going to flag that case. */
311 if (BOUNDARY_VALUE (c
, i
- 1) == value
)
314 move_cache_gap (c
, i
, 1);
316 c
->boundaries
[i
].pos
= pos
- c
->buffer_beg
;
317 c
->boundaries
[i
].value
= value
;
324 /* Delete the i'th entry from cache C if START <= i < END. */
327 delete_cache_boundaries (struct region_cache
*c
,
328 ptrdiff_t start
, ptrdiff_t end
)
330 ptrdiff_t len
= end
- start
;
332 /* Gotta be in range. */
334 || end
> c
->cache_len
)
337 /* Gotta be in order. */
341 /* Can't delete the dummy entry. */
346 /* Minimize gap motion. If we're deleting nothing, do nothing. */
349 /* If the gap is before the region to delete, delete from the start
351 else if (c
->gap_start
<= start
)
353 move_cache_gap (c
, start
, 0);
356 /* If the gap is after the region to delete, delete from the end
358 else if (end
<= c
->gap_start
)
360 move_cache_gap (c
, end
, 0);
364 /* If the gap is in the region to delete, just expand it. */
367 c
->gap_start
= start
;
376 /* Set the value for a region. */
378 /* Set the value in cache C for the region START..END to VALUE. */
380 set_cache_region (struct region_cache
*c
,
381 ptrdiff_t start
, ptrdiff_t end
, int value
)
385 if (start
< c
->buffer_beg
386 || end
> c
->buffer_end
)
389 /* Eliminate this case; then we can assume that start and end-1 are
390 both the locations of real characters in the buffer. */
395 /* We need to make sure that there are no boundaries in the area
396 between start to end; the whole area will have the same value,
397 so those boundaries will not be necessary.
399 Let start_ix be the cache index of the boundary governing the
400 first character of start..end, and let end_ix be the cache
401 index of the earliest boundary after the last character in
402 start..end. (This tortured terminology is intended to answer
403 all the "< or <=?" sort of questions.) */
404 ptrdiff_t start_ix
= find_cache_boundary (c
, start
);
405 ptrdiff_t end_ix
= find_cache_boundary (c
, end
- 1) + 1;
407 /* We must remember the value established by the last boundary
408 before end; if that boundary's domain stretches beyond end,
409 we'll need to create a new boundary at end, and that boundary
410 must have that remembered value. */
411 int value_at_end
= BOUNDARY_VALUE (c
, end_ix
- 1);
413 /* Delete all boundaries strictly within start..end; this means
414 those whose indices are between start_ix (exclusive) and end_ix
416 delete_cache_boundaries (c
, start_ix
+ 1, end_ix
);
418 /* Make sure we have the right value established going in to
419 start..end from the left, and no unnecessary boundaries. */
420 if (BOUNDARY_POS (c
, start_ix
) == start
)
422 /* Is this boundary necessary? If no, remove it; if yes, set
425 && BOUNDARY_VALUE (c
, start_ix
- 1) == value
)
427 delete_cache_boundaries (c
, start_ix
, start_ix
+ 1);
431 SET_BOUNDARY_VALUE (c
, start_ix
, value
);
435 /* Do we need to add a new boundary here? */
436 if (BOUNDARY_VALUE (c
, start_ix
) != value
)
438 insert_cache_boundary (c
, start_ix
+ 1, start
, value
);
443 /* This is equivalent to letting end_ix float (like a buffer
444 marker does) with the insertions and deletions we may have
446 end_ix
= start_ix
+ 1;
448 /* Make sure we have the correct value established as we leave
449 start..end to the right. */
450 if (end
== c
->buffer_end
)
451 /* There is no text after start..end; nothing to do. */
453 else if (end_ix
>= c
->cache_len
454 || end
< BOUNDARY_POS (c
, end_ix
))
456 /* There is no boundary at end, but we may need one. */
457 if (value_at_end
!= value
)
458 insert_cache_boundary (c
, end_ix
, end
, value_at_end
);
462 /* There is a boundary at end; should it be there? */
463 if (value
== BOUNDARY_VALUE (c
, end_ix
))
464 delete_cache_boundaries (c
, end_ix
, end_ix
+ 1);
471 /* Interface: Invalidating the cache. Private: Re-validating the cache. */
473 /* Indicate that a section of BUF has changed, to invalidate CACHE.
474 HEAD is the number of chars unchanged at the beginning of the buffer.
475 TAIL is the number of chars unchanged at the end of the buffer.
476 NOTE: this is *not* the same as the ending position of modified
478 (This way of specifying regions makes more sense than absolute
479 buffer positions in the presence of insertions and deletions; the
480 args to pass are the same before and after such an operation.) */
482 invalidate_region_cache (struct buffer
*buf
, struct region_cache
*c
,
483 ptrdiff_t head
, ptrdiff_t tail
)
485 /* Let chead = c->beg_unchanged, and
486 ctail = c->end_unchanged.
487 If z-tail < beg+chead by a large amount, or
488 z-ctail < beg+head by a large amount,
490 then cutting back chead and ctail to head and tail would lose a
491 lot of information that we could preserve by revalidating the
492 cache before processing this invalidation. Losing that
493 information may be more costly than revalidating the cache now.
494 So go ahead and call revalidate_region_cache if it seems that it
495 might be worthwhile. */
496 if (((BUF_BEG (buf
) + c
->beg_unchanged
) - (BUF_Z (buf
) - tail
)
497 > PRESERVE_THRESHOLD
)
498 || ((BUF_BEG (buf
) + head
) - (BUF_Z (buf
) - c
->end_unchanged
)
499 > PRESERVE_THRESHOLD
))
500 revalidate_region_cache (buf
, c
);
503 if (head
< c
->beg_unchanged
)
504 c
->beg_unchanged
= head
;
505 if (tail
< c
->end_unchanged
)
506 c
->end_unchanged
= tail
;
508 /* We now know nothing about the region between the unchanged head
509 and the unchanged tail (call it the "modified region"), not even
512 If the modified region has shrunk in size (deletions do this),
513 then the cache may now contain boundaries originally located in
514 text that doesn't exist any more.
516 If the modified region has increased in size (insertions do
517 this), then there may now be boundaries in the modified region
518 whose positions are wrong.
520 Even calling BOUNDARY_POS on boundaries still in the unchanged
521 head or tail may well give incorrect answers now, since
522 c->buffer_beg and c->buffer_end may well be wrong now. (Well,
523 okay, c->buffer_beg never changes, so boundaries in the unchanged
524 head will still be okay. But it's the principle of the thing.)
526 So things are generally a mess.
528 But we don't clean up this mess here; that would be expensive,
529 and this function gets called every time any buffer modification
530 occurs. Rather, we can clean up everything in one swell foop,
531 accounting for all the modifications at once, by calling
532 revalidate_region_cache before we try to consult the cache the
537 /* Clean out any cache entries applying to the modified region, and
538 make the positions of the remaining entries accurate again.
540 After calling this function, the mess described in the comment in
541 invalidate_region_cache is cleaned up.
543 This function operates by simply throwing away everything it knows
544 about the modified region. It doesn't care exactly which
545 insertions and deletions took place; it just tosses it all.
547 For example, if you insert a single character at the beginning of
548 the buffer, and a single character at the end of the buffer (for
549 example), without calling this function in between the two
550 insertions, then the entire cache will be freed of useful
551 information. On the other hand, if you do manage to call this
552 function in between the two insertions, then the modified regions
553 will be small in both cases, no information will be tossed, and the
554 cache will know that it doesn't have knowledge of the first and
555 last characters any more.
557 Calling this function may be expensive; it does binary searches in
558 the cache, and causes cache gap motion. */
561 revalidate_region_cache (struct buffer
*buf
, struct region_cache
*c
)
563 /* The boundaries now in the cache are expressed relative to the
564 buffer_beg and buffer_end values stored in the cache. Now,
565 buffer_beg and buffer_end may not be the same as BUF_BEG (buf)
566 and BUF_Z (buf), so we have two different "bases" to deal with
567 --- the cache's, and the buffer's. */
569 /* If the entire buffer is still valid, don't waste time. Yes, this
570 should be a >, not a >=; think about what beg_unchanged and
571 end_unchanged get set to when the only change has been an
573 if (c
->buffer_beg
+ c
->beg_unchanged
574 > c
->buffer_end
- c
->end_unchanged
)
577 /* If all the text we knew about as of the last cache revalidation
578 is still there, then all of the information in the cache is still
579 valid. Because c->buffer_beg and c->buffer_end are out-of-date,
580 the modified region appears from the cache's point of view to be
581 a null region located someplace in the buffer.
583 Now, invalidating that empty string will have no actual affect on
584 the cache; instead, we need to update the cache's basis first
585 (which will give the modified region the same size in the cache
586 as it has in the buffer), and then invalidate the modified
588 if (c
->buffer_beg
+ c
->beg_unchanged
589 == c
->buffer_end
- c
->end_unchanged
)
591 /* Move the gap so that all the boundaries in the unchanged head
592 are expressed beg-relative, and all the boundaries in the
593 unchanged tail are expressed end-relative. That done, we can
594 plug in the new buffer beg and end, and all the positions
597 The boundary which has jurisdiction over the modified region
598 should be left before the gap. */
600 (find_cache_boundary (c
, (c
->buffer_beg
605 c
->buffer_beg
= BUF_BEG (buf
);
606 c
->buffer_end
= BUF_Z (buf
);
608 /* Now that the cache's basis has been changed, the modified
609 region actually takes up some space in the cache, so we can
612 c
->buffer_beg
+ c
->beg_unchanged
,
613 c
->buffer_end
- c
->end_unchanged
,
617 /* Otherwise, there is a non-empty region in the cache which
618 corresponds to the modified region of the buffer. */
621 ptrdiff_t modified_ix
;
623 /* These positions are correct, relative to both the cache basis
624 and the buffer basis. */
626 c
->buffer_beg
+ c
->beg_unchanged
,
627 c
->buffer_end
- c
->end_unchanged
,
630 /* Now the cache contains only boundaries that are in the
631 unchanged head and tail; we've disposed of any boundaries
632 whose positions we can't be sure of given the information
635 If we put the cache gap between the unchanged head and the
636 unchanged tail, we can adjust all the boundary positions at
637 once, simply by setting buffer_beg and buffer_end.
639 The boundary which has jurisdiction over the modified region
640 should be left before the gap. */
642 find_cache_boundary (c
, (c
->buffer_beg
+ c
->beg_unchanged
)) + 1;
643 move_cache_gap (c
, modified_ix
, 0);
645 c
->buffer_beg
= BUF_BEG (buf
);
646 c
->buffer_end
= BUF_Z (buf
);
648 /* Now, we may have shrunk the buffer when we changed the basis,
649 and brought the boundaries we created for the start and end
650 of the modified region together, giving them the same
651 position. If that's the case, we should collapse them into
652 one boundary. Or we may even delete them both, if the values
653 before and after them are the same. */
654 if (modified_ix
< c
->cache_len
655 && (BOUNDARY_POS (c
, modified_ix
- 1)
656 == BOUNDARY_POS (c
, modified_ix
)))
658 int value_after
= BOUNDARY_VALUE (c
, modified_ix
);
660 /* Should we remove both of the boundaries? Yes, if the
661 latter boundary is now establishing the same value that
662 the former boundary's predecessor does. */
663 if (modified_ix
- 1 > 0
664 && value_after
== BOUNDARY_VALUE (c
, modified_ix
- 2))
665 delete_cache_boundaries (c
, modified_ix
- 1, modified_ix
+ 1);
668 /* We do need a boundary here; collapse the two
669 boundaries into one. */
670 SET_BOUNDARY_VALUE (c
, modified_ix
- 1, value_after
);
671 delete_cache_boundaries (c
, modified_ix
, modified_ix
+ 1);
676 /* Now the entire cache is valid. */
679 = c
->buffer_end
- c
->buffer_beg
;
683 /* Interface: Adding information to the cache. */
685 /* Assert that the region of BUF between START and END (absolute
686 buffer positions) is "known," for the purposes of CACHE (e.g. "has
687 no newlines", in the case of the line cache). */
689 know_region_cache (struct buffer
*buf
, struct region_cache
*c
,
690 ptrdiff_t start
, ptrdiff_t end
)
692 revalidate_region_cache (buf
, c
);
694 set_cache_region (c
, start
, end
, 1);
698 /* Interface: using the cache. */
700 /* Return true if the text immediately after POS in BUF is known, for
701 the purposes of CACHE. If NEXT is non-zero, set *NEXT to the nearest
702 position after POS where the knownness changes. */
704 region_cache_forward (struct buffer
*buf
, struct region_cache
*c
,
705 ptrdiff_t pos
, ptrdiff_t *next
)
707 revalidate_region_cache (buf
, c
);
710 ptrdiff_t i
= find_cache_boundary (c
, pos
);
711 int i_value
= BOUNDARY_VALUE (c
, i
);
714 /* Beyond the end of the buffer is unknown, by definition. */
715 if (pos
>= BUF_Z (buf
))
717 if (next
) *next
= BUF_Z (buf
);
722 /* Scan forward from i to find the next differing position. */
723 for (j
= i
+ 1; j
< c
->cache_len
; j
++)
724 if (BOUNDARY_VALUE (c
, j
) != i_value
)
727 if (j
< c
->cache_len
)
728 *next
= BOUNDARY_POS (c
, j
);
737 /* Return true if the text immediately before POS in BUF is known, for
738 the purposes of CACHE. If NEXT is non-zero, set *NEXT to the nearest
739 position before POS where the knownness changes. */
740 int region_cache_backward (struct buffer
*buf
, struct region_cache
*c
,
741 ptrdiff_t pos
, ptrdiff_t *next
)
743 revalidate_region_cache (buf
, c
);
745 /* Before the beginning of the buffer is unknown, by
747 if (pos
<= BUF_BEG (buf
))
749 if (next
) *next
= BUF_BEG (buf
);
754 ptrdiff_t i
= find_cache_boundary (c
, pos
- 1);
755 int i_value
= BOUNDARY_VALUE (c
, i
);
760 /* Scan backward from i to find the next differing position. */
761 for (j
= i
- 1; j
>= 0; j
--)
762 if (BOUNDARY_VALUE (c
, j
) != i_value
)
766 *next
= BOUNDARY_POS (c
, j
+ 1);
768 *next
= BUF_BEG (buf
);
776 /* Debugging: pretty-print a cache to the standard error output. */
778 void pp_cache (struct region_cache
*) EXTERNALLY_VISIBLE
;
780 pp_cache (struct region_cache
*c
)
783 ptrdiff_t beg_u
= c
->buffer_beg
+ c
->beg_unchanged
;
784 ptrdiff_t end_u
= c
->buffer_end
- c
->end_unchanged
;
787 "basis: %"pD
"d..%"pD
"d modified: %"pD
"d..%"pD
"d\n",
788 c
->buffer_beg
, c
->buffer_end
,
791 for (i
= 0; i
< c
->cache_len
; i
++)
793 ptrdiff_t pos
= BOUNDARY_POS (c
, i
);
795 putc (((pos
< beg_u
) ? 'v'
796 : (pos
== beg_u
) ? '-'
799 putc (((pos
> end_u
) ? '^'
800 : (pos
== end_u
) ? '-'
803 fprintf (stderr
, "%"pD
"d : %d\n", pos
, BOUNDARY_VALUE (c
, i
));