1 /* Caching facts about regions of the buffer, for optimization.
3 Copyright (C) 1985-1989, 1993, 1995, 2001-2012
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/>. */
27 #include "character.h"
29 #include "region-cache.h"
32 /* Data structures. */
36 We want something that maps character positions in a buffer onto
37 values. The representation should deal well with long runs of
38 characters with the same value.
40 The tricky part: the representation should be very cheap to
41 maintain in the presence of many insertions and deletions. If the
42 overhead of maintaining the cache is too high, the speedups it
43 offers will be worthless.
46 We represent the region cache as a sorted array of struct
47 boundary's, each of which contains a buffer position and a value;
48 the value applies to all the characters after the buffer position,
49 until the position of the next boundary, or the end of the buffer.
51 The cache always has a boundary whose position is BUF_BEG, so
52 there's always a value associated with every character in the
53 buffer. Since the cache is sorted, this is always the first
56 To facilitate the insertion and deletion of boundaries in the
57 cache, the cache has a gap, just like Emacs's text buffers do.
59 To help boundary positions float along with insertions and
60 deletions, all boundary positions before the cache gap are stored
61 relative to BUF_BEG (buf) (thus they're >= 0), and all boundary
62 positions after the gap are stored relative to BUF_Z (buf) (thus
63 they're <= 0). Look at BOUNDARY_POS to see this in action. See
64 revalidate_region_cache to see how this helps. */
72 /* A sorted array of locations where the known-ness of the buffer
74 struct boundary
*boundaries
;
76 /* boundaries[gap_start ... gap_start + gap_len - 1] is the gap. */
77 ptrdiff_t gap_start
, gap_len
;
79 /* The number of elements allocated to boundaries, not including the
83 /* The areas that haven't changed since the last time we cleaned out
84 invalid entries from the cache. These overlap when the buffer is
85 entirely unchanged. */
86 ptrdiff_t beg_unchanged
, end_unchanged
;
88 /* The first and last positions in the buffer. Because boundaries
89 store their positions relative to the start (BEG) and end (Z) of
90 the buffer, knowing these positions allows us to accurately
91 interpret positions without having to pass the buffer structure
92 or its endpoints around all the time.
94 Yes, buffer_beg is always 1. It's there for symmetry with
95 buffer_end and the BEG and BUF_BEG macros. */
96 ptrdiff_t buffer_beg
, buffer_end
;
99 /* Return the position of boundary i in cache c. */
100 #define BOUNDARY_POS(c, i) \
101 ((i) < (c)->gap_start \
102 ? (c)->buffer_beg + (c)->boundaries[(i)].pos \
103 : (c)->buffer_end + (c)->boundaries[(c)->gap_len + (i)].pos)
105 /* Return the value for text after boundary i in cache c. */
106 #define BOUNDARY_VALUE(c, i) \
107 ((i) < (c)->gap_start \
108 ? (c)->boundaries[(i)].value \
109 : (c)->boundaries[(c)->gap_len + (i)].value)
111 /* Set the value for text after boundary i in cache c to v. */
112 #define SET_BOUNDARY_VALUE(c, i, v) \
113 ((i) < (c)->gap_start \
114 ? ((c)->boundaries[(i)].value = (v))\
115 : ((c)->boundaries[(c)->gap_len + (i)].value = (v)))
118 /* How many elements to add to the gap when we resize the buffer. */
119 #define NEW_CACHE_GAP (40)
121 /* See invalidate_region_cache; if an invalidation would throw away
122 information about this many characters, call
123 revalidate_region_cache before doing the new invalidation, to
124 preserve that information, instead of throwing it away. */
125 #define PRESERVE_THRESHOLD (500)
127 static void revalidate_region_cache (struct buffer
*buf
, struct region_cache
*c
);
130 /* Interface: Allocating, initializing, and disposing of region caches. */
132 struct region_cache
*
133 new_region_cache (void)
135 struct region_cache
*c
= xmalloc (sizeof *c
);
138 c
->gap_len
= NEW_CACHE_GAP
;
140 c
->boundaries
= xmalloc ((c
->gap_len
+ c
->cache_len
)
141 * sizeof (*c
->boundaries
));
143 c
->beg_unchanged
= 0;
144 c
->end_unchanged
= 0;
148 /* Insert the boundary for the buffer start. */
152 c
->boundaries
[0].pos
= 0; /* from buffer_beg */
153 c
->boundaries
[0].value
= 0;
159 free_region_cache (struct region_cache
*c
)
161 xfree (c
->boundaries
);
166 /* Finding positions in the cache. */
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).
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. */
176 find_cache_boundary (struct region_cache
*c
, ptrdiff_t pos
)
178 ptrdiff_t low
= 0, high
= c
->cache_len
;
180 while (low
+ 1 < high
)
182 /* mid is always a valid index, because low < high and ">> 1"
184 ptrdiff_t mid
= (low
>> 1) + (high
>> 1) + (low
& high
& 1);
185 ptrdiff_t boundary
= BOUNDARY_POS (c
, mid
);
194 if (BOUNDARY_POS (c
, low
) > pos
195 || (low
+ 1 < c
->cache_len
196 && BOUNDARY_POS (c
, low
+ 1) <= pos
))
204 /* Moving the cache gap around, inserting, and deleting. */
207 /* Move the gap of cache C to index POS, and make sure it has space
208 for at least MIN_SIZE boundaries. */
210 move_cache_gap (struct region_cache
*c
, ptrdiff_t pos
, ptrdiff_t min_size
)
212 /* Copy these out of the cache and into registers. */
213 ptrdiff_t gap_start
= c
->gap_start
;
214 ptrdiff_t gap_len
= c
->gap_len
;
215 ptrdiff_t buffer_beg
= c
->buffer_beg
;
216 ptrdiff_t buffer_end
= c
->buffer_end
;
219 || pos
> c
->cache_len
)
222 /* We mustn't ever try to put the gap before the dummy start
223 boundary. That must always be start-relative. */
227 /* Need we move the gap right? */
228 while (gap_start
< pos
)
230 /* Copy one boundary from after to before the gap, and
231 convert its position to start-relative. */
232 c
->boundaries
[gap_start
].pos
234 + c
->boundaries
[gap_start
+ gap_len
].pos
236 c
->boundaries
[gap_start
].value
237 = c
->boundaries
[gap_start
+ gap_len
].value
;
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
)
251 xpalloc (c
->boundaries
, &c
->cache_len
, min_size
, -1,
252 sizeof *c
->boundaries
);
254 /* Some systems don't provide a version of the copy routine that
255 can be trusted to shift memory upward into an overlapping
256 region. memmove isn't widely available. */
258 for (i
= c
->cache_len
- 1; i
>= gap_start
; i
--)
260 c
->boundaries
[i
+ min_size
].pos
= c
->boundaries
[i
+ gap_len
].pos
;
261 c
->boundaries
[i
+ min_size
].value
= c
->boundaries
[i
+ gap_len
].value
;
267 /* Need we move the gap left? */
268 while (pos
< gap_start
)
272 /* Copy one region from before to after the gap, and
273 convert its position to end-relative. */
274 c
->boundaries
[gap_start
+ gap_len
].pos
275 = c
->boundaries
[gap_start
].pos
+ buffer_beg
- buffer_end
;
276 c
->boundaries
[gap_start
+ gap_len
].value
277 = c
->boundaries
[gap_start
].value
;
280 /* Assign these back into the cache. */
281 c
->gap_start
= gap_start
;
282 c
->gap_len
= gap_len
;
286 /* Insert a new boundary in cache C; it will have cache index I,
287 and have the specified POS and VALUE. */
289 insert_cache_boundary (struct region_cache
*c
, ptrdiff_t i
, ptrdiff_t pos
,
292 /* i must be a valid cache index. */
293 if (i
< 0 || i
> c
->cache_len
)
296 /* We must never want to insert something before the dummy first
301 /* We must only be inserting things in order. */
302 if (! (BOUNDARY_POS (c
, i
- 1) < pos
303 && (i
== c
->cache_len
304 || pos
< BOUNDARY_POS (c
, i
))))
307 /* The value must be different from the ones around it. However, we
308 temporarily create boundaries that establish the same value as
309 the subsequent boundary, so we're not going to flag that case. */
310 if (BOUNDARY_VALUE (c
, i
- 1) == value
)
313 move_cache_gap (c
, i
, 1);
315 c
->boundaries
[i
].pos
= pos
- c
->buffer_beg
;
316 c
->boundaries
[i
].value
= value
;
323 /* Delete the i'th entry from cache C if START <= i < END. */
326 delete_cache_boundaries (struct region_cache
*c
,
327 ptrdiff_t start
, ptrdiff_t end
)
329 ptrdiff_t len
= end
- start
;
331 /* Gotta be in range. */
333 || end
> c
->cache_len
)
336 /* Gotta be in order. */
340 /* Can't delete the dummy entry. */
345 /* Minimize gap motion. If we're deleting nothing, do nothing. */
348 /* If the gap is before the region to delete, delete from the start
350 else if (c
->gap_start
<= start
)
352 move_cache_gap (c
, start
, 0);
355 /* If the gap is after the region to delete, delete from the end
357 else if (end
<= c
->gap_start
)
359 move_cache_gap (c
, end
, 0);
363 /* If the gap is in the region to delete, just expand it. */
366 c
->gap_start
= start
;
375 /* Set the value for a region. */
377 /* Set the value in cache C for the region START..END to VALUE. */
379 set_cache_region (struct region_cache
*c
,
380 ptrdiff_t start
, ptrdiff_t end
, int value
)
384 if (start
< c
->buffer_beg
385 || end
> c
->buffer_end
)
388 /* Eliminate this case; then we can assume that start and end-1 are
389 both the locations of real characters in the buffer. */
394 /* We need to make sure that there are no boundaries in the area
395 between start to end; the whole area will have the same value,
396 so those boundaries will not be necessary.
398 Let start_ix be the cache index of the boundary governing the
399 first character of start..end, and let end_ix be the cache
400 index of the earliest boundary after the last character in
401 start..end. (This tortured terminology is intended to answer
402 all the "< or <=?" sort of questions.) */
403 ptrdiff_t start_ix
= find_cache_boundary (c
, start
);
404 ptrdiff_t end_ix
= find_cache_boundary (c
, end
- 1) + 1;
406 /* We must remember the value established by the last boundary
407 before end; if that boundary's domain stretches beyond end,
408 we'll need to create a new boundary at end, and that boundary
409 must have that remembered value. */
410 int value_at_end
= BOUNDARY_VALUE (c
, end_ix
- 1);
412 /* Delete all boundaries strictly within start..end; this means
413 those whose indices are between start_ix (exclusive) and end_ix
415 delete_cache_boundaries (c
, start_ix
+ 1, end_ix
);
417 /* Make sure we have the right value established going in to
418 start..end from the left, and no unnecessary boundaries. */
419 if (BOUNDARY_POS (c
, start_ix
) == start
)
421 /* Is this boundary necessary? If no, remove it; if yes, set
424 && BOUNDARY_VALUE (c
, start_ix
- 1) == value
)
426 delete_cache_boundaries (c
, start_ix
, start_ix
+ 1);
430 SET_BOUNDARY_VALUE (c
, start_ix
, value
);
434 /* Do we need to add a new boundary here? */
435 if (BOUNDARY_VALUE (c
, start_ix
) != value
)
437 insert_cache_boundary (c
, start_ix
+ 1, start
, value
);
442 /* This is equivalent to letting end_ix float (like a buffer
443 marker does) with the insertions and deletions we may have
445 end_ix
= start_ix
+ 1;
447 /* Make sure we have the correct value established as we leave
448 start..end to the right. */
449 if (end
== c
->buffer_end
)
450 /* There is no text after start..end; nothing to do. */
452 else if (end_ix
>= c
->cache_len
453 || end
< BOUNDARY_POS (c
, end_ix
))
455 /* There is no boundary at end, but we may need one. */
456 if (value_at_end
!= value
)
457 insert_cache_boundary (c
, end_ix
, end
, value_at_end
);
461 /* There is a boundary at end; should it be there? */
462 if (value
== BOUNDARY_VALUE (c
, end_ix
))
463 delete_cache_boundaries (c
, end_ix
, end_ix
+ 1);
470 /* Interface: Invalidating the cache. Private: Re-validating the cache. */
472 /* Indicate that a section of BUF has changed, to invalidate CACHE.
473 HEAD is the number of chars unchanged at the beginning of the buffer.
474 TAIL is the number of chars unchanged at the end of the buffer.
475 NOTE: this is *not* the same as the ending position of modified
477 (This way of specifying regions makes more sense than absolute
478 buffer positions in the presence of insertions and deletions; the
479 args to pass are the same before and after such an operation.) */
481 invalidate_region_cache (struct buffer
*buf
, struct region_cache
*c
,
482 ptrdiff_t head
, ptrdiff_t tail
)
484 /* Let chead = c->beg_unchanged, and
485 ctail = c->end_unchanged.
486 If z-tail < beg+chead by a large amount, or
487 z-ctail < beg+head by a large amount,
489 then cutting back chead and ctail to head and tail would lose a
490 lot of information that we could preserve by revalidating the
491 cache before processing this invalidation. Losing that
492 information may be more costly than revalidating the cache now.
493 So go ahead and call revalidate_region_cache if it seems that it
494 might be worthwhile. */
495 if (((BUF_BEG (buf
) + c
->beg_unchanged
) - (BUF_Z (buf
) - tail
)
496 > PRESERVE_THRESHOLD
)
497 || ((BUF_BEG (buf
) + head
) - (BUF_Z (buf
) - c
->end_unchanged
)
498 > PRESERVE_THRESHOLD
))
499 revalidate_region_cache (buf
, c
);
502 if (head
< c
->beg_unchanged
)
503 c
->beg_unchanged
= head
;
504 if (tail
< c
->end_unchanged
)
505 c
->end_unchanged
= tail
;
507 /* We now know nothing about the region between the unchanged head
508 and the unchanged tail (call it the "modified region"), not even
511 If the modified region has shrunk in size (deletions do this),
512 then the cache may now contain boundaries originally located in
513 text that doesn't exist any more.
515 If the modified region has increased in size (insertions do
516 this), then there may now be boundaries in the modified region
517 whose positions are wrong.
519 Even calling BOUNDARY_POS on boundaries still in the unchanged
520 head or tail may well give incorrect answers now, since
521 c->buffer_beg and c->buffer_end may well be wrong now. (Well,
522 okay, c->buffer_beg never changes, so boundaries in the unchanged
523 head will still be okay. But it's the principle of the thing.)
525 So things are generally a mess.
527 But we don't clean up this mess here; that would be expensive,
528 and this function gets called every time any buffer modification
529 occurs. Rather, we can clean up everything in one swell foop,
530 accounting for all the modifications at once, by calling
531 revalidate_region_cache before we try to consult the cache the
536 /* Clean out any cache entries applying to the modified region, and
537 make the positions of the remaining entries accurate again.
539 After calling this function, the mess described in the comment in
540 invalidate_region_cache is cleaned up.
542 This function operates by simply throwing away everything it knows
543 about the modified region. It doesn't care exactly which
544 insertions and deletions took place; it just tosses it all.
546 For example, if you insert a single character at the beginning of
547 the buffer, and a single character at the end of the buffer (for
548 example), without calling this function in between the two
549 insertions, then the entire cache will be freed of useful
550 information. On the other hand, if you do manage to call this
551 function in between the two insertions, then the modified regions
552 will be small in both cases, no information will be tossed, and the
553 cache will know that it doesn't have knowledge of the first and
554 last characters any more.
556 Calling this function may be expensive; it does binary searches in
557 the cache, and causes cache gap motion. */
560 revalidate_region_cache (struct buffer
*buf
, struct region_cache
*c
)
562 /* The boundaries now in the cache are expressed relative to the
563 buffer_beg and buffer_end values stored in the cache. Now,
564 buffer_beg and buffer_end may not be the same as BUF_BEG (buf)
565 and BUF_Z (buf), so we have two different "bases" to deal with
566 --- the cache's, and the buffer's. */
568 /* If the entire buffer is still valid, don't waste time. Yes, this
569 should be a >, not a >=; think about what beg_unchanged and
570 end_unchanged get set to when the only change has been an
572 if (c
->buffer_beg
+ c
->beg_unchanged
573 > c
->buffer_end
- c
->end_unchanged
)
576 /* If all the text we knew about as of the last cache revalidation
577 is still there, then all of the information in the cache is still
578 valid. Because c->buffer_beg and c->buffer_end are out-of-date,
579 the modified region appears from the cache's point of view to be
580 a null region located someplace in the buffer.
582 Now, invalidating that empty string will have no actual affect on
583 the cache; instead, we need to update the cache's basis first
584 (which will give the modified region the same size in the cache
585 as it has in the buffer), and then invalidate the modified
587 if (c
->buffer_beg
+ c
->beg_unchanged
588 == c
->buffer_end
- c
->end_unchanged
)
590 /* Move the gap so that all the boundaries in the unchanged head
591 are expressed beg-relative, and all the boundaries in the
592 unchanged tail are expressed end-relative. That done, we can
593 plug in the new buffer beg and end, and all the positions
596 The boundary which has jurisdiction over the modified region
597 should be left before the gap. */
599 (find_cache_boundary (c
, (c
->buffer_beg
604 c
->buffer_beg
= BUF_BEG (buf
);
605 c
->buffer_end
= BUF_Z (buf
);
607 /* Now that the cache's basis has been changed, the modified
608 region actually takes up some space in the cache, so we can
611 c
->buffer_beg
+ c
->beg_unchanged
,
612 c
->buffer_end
- c
->end_unchanged
,
616 /* Otherwise, there is a non-empty region in the cache which
617 corresponds to the modified region of the buffer. */
620 ptrdiff_t modified_ix
;
622 /* These positions are correct, relative to both the cache basis
623 and the buffer basis. */
625 c
->buffer_beg
+ c
->beg_unchanged
,
626 c
->buffer_end
- c
->end_unchanged
,
629 /* Now the cache contains only boundaries that are in the
630 unchanged head and tail; we've disposed of any boundaries
631 whose positions we can't be sure of given the information
634 If we put the cache gap between the unchanged head and the
635 unchanged tail, we can adjust all the boundary positions at
636 once, simply by setting buffer_beg and buffer_end.
638 The boundary which has jurisdiction over the modified region
639 should be left before the gap. */
641 find_cache_boundary (c
, (c
->buffer_beg
+ c
->beg_unchanged
)) + 1;
642 move_cache_gap (c
, modified_ix
, 0);
644 c
->buffer_beg
= BUF_BEG (buf
);
645 c
->buffer_end
= BUF_Z (buf
);
647 /* Now, we may have shrunk the buffer when we changed the basis,
648 and brought the boundaries we created for the start and end
649 of the modified region together, giving them the same
650 position. If that's the case, we should collapse them into
651 one boundary. Or we may even delete them both, if the values
652 before and after them are the same. */
653 if (modified_ix
< c
->cache_len
654 && (BOUNDARY_POS (c
, modified_ix
- 1)
655 == BOUNDARY_POS (c
, modified_ix
)))
657 int value_after
= BOUNDARY_VALUE (c
, modified_ix
);
659 /* Should we remove both of the boundaries? Yes, if the
660 latter boundary is now establishing the same value that
661 the former boundary's predecessor does. */
662 if (modified_ix
- 1 > 0
663 && value_after
== BOUNDARY_VALUE (c
, modified_ix
- 2))
664 delete_cache_boundaries (c
, modified_ix
- 1, modified_ix
+ 1);
667 /* We do need a boundary here; collapse the two
668 boundaries into one. */
669 SET_BOUNDARY_VALUE (c
, modified_ix
- 1, value_after
);
670 delete_cache_boundaries (c
, modified_ix
, modified_ix
+ 1);
675 /* Now the entire cache is valid. */
678 = c
->buffer_end
- c
->buffer_beg
;
682 /* Interface: Adding information to the cache. */
684 /* Assert that the region of BUF between START and END (absolute
685 buffer positions) is "known," for the purposes of CACHE (e.g. "has
686 no newlines", in the case of the line cache). */
688 know_region_cache (struct buffer
*buf
, struct region_cache
*c
,
689 ptrdiff_t start
, ptrdiff_t end
)
691 revalidate_region_cache (buf
, c
);
693 set_cache_region (c
, start
, end
, 1);
697 /* Interface: using the cache. */
699 /* Return true if the text immediately after POS in BUF is known, for
700 the purposes of CACHE. If NEXT is non-zero, set *NEXT to the nearest
701 position after POS where the knowledge changes. */
703 region_cache_forward (struct buffer
*buf
, struct region_cache
*c
,
704 ptrdiff_t pos
, ptrdiff_t *next
)
706 revalidate_region_cache (buf
, c
);
709 ptrdiff_t i
= find_cache_boundary (c
, pos
);
710 int i_value
= BOUNDARY_VALUE (c
, i
);
713 /* Beyond the end of the buffer is unknown, by definition. */
714 if (pos
>= BUF_Z (buf
))
716 if (next
) *next
= BUF_Z (buf
);
721 /* Scan forward from i to find the next differing position. */
722 for (j
= i
+ 1; j
< c
->cache_len
; j
++)
723 if (BOUNDARY_VALUE (c
, j
) != i_value
)
726 if (j
< c
->cache_len
)
727 *next
= BOUNDARY_POS (c
, j
);
736 /* Return true if the text immediately before POS in BUF is known, for
737 the purposes of CACHE. If NEXT is non-zero, set *NEXT to the nearest
738 position before POS where the knowledge changes. */
739 int region_cache_backward (struct buffer
*buf
, struct region_cache
*c
,
740 ptrdiff_t pos
, ptrdiff_t *next
)
742 revalidate_region_cache (buf
, c
);
744 /* Before the beginning of the buffer is unknown, by
746 if (pos
<= BUF_BEG (buf
))
748 if (next
) *next
= BUF_BEG (buf
);
753 ptrdiff_t i
= find_cache_boundary (c
, pos
- 1);
754 int i_value
= BOUNDARY_VALUE (c
, i
);
759 /* Scan backward from i to find the next differing position. */
760 for (j
= i
- 1; j
>= 0; j
--)
761 if (BOUNDARY_VALUE (c
, j
) != i_value
)
765 *next
= BOUNDARY_POS (c
, j
+ 1);
767 *next
= BUF_BEG (buf
);
775 /* Debugging: pretty-print a cache to the standard error output. */
777 void pp_cache (struct region_cache
*) EXTERNALLY_VISIBLE
;
779 pp_cache (struct region_cache
*c
)
782 ptrdiff_t beg_u
= c
->buffer_beg
+ c
->beg_unchanged
;
783 ptrdiff_t end_u
= c
->buffer_end
- c
->end_unchanged
;
786 "basis: %"pD
"d..%"pD
"d modified: %"pD
"d..%"pD
"d\n",
787 c
->buffer_beg
, c
->buffer_end
,
790 for (i
= 0; i
< c
->cache_len
; i
++)
792 ptrdiff_t pos
= BOUNDARY_POS (c
, i
);
794 putc (((pos
< beg_u
) ? 'v'
795 : (pos
== beg_u
) ? '-'
798 putc (((pos
> end_u
) ? '^'
799 : (pos
== end_u
) ? '-'
802 fprintf (stderr
, "%"pD
"d : %d\n", pos
, BOUNDARY_VALUE (c
, i
));