4
4
* This code was greatly inspired by parts of LibXDiff from Davide Libenzi
5
5
* http://www.xmailserver.org/xdiff-lib.html
7
* Rewritten for GIT by Nicolas Pitre <nico@cam.org>, (C) 2005-2007
7
* Rewritten for GIT by Nicolas Pitre <nico@fluxnic.net>, (C) 2005-2007
8
8
* Adapted for Bazaar by John Arbash Meinel <john@arbash-meinel.com> (C) 2009
10
10
* This program is free software; you can redistribute it and/or modify
280
280
if (fit_in_old) {
281
281
// fprintf(stderr, "Fit all %d entries into old index\n",
282
282
// copied_count);
283
/* No need to allocate a new buffer */
284
* No need to allocate a new buffer, but return old_index ptr so
285
* callers can distinguish this from an OOM failure.
286
289
// fprintf(stderr, "Fit only %d entries into old index,"
287
290
// " reallocating\n", copied_count);
374
377
create_delta_index(const struct source_info *src,
375
struct delta_index *old)
378
struct delta_index *old,
379
struct delta_index **fresh,
380
int max_bytes_to_index)
377
382
unsigned int i, hsize, hmask, num_entries, prev_val, *hash_count;
378
unsigned int total_num_entries;
383
unsigned int total_num_entries, stride, max_entries;
379
384
const unsigned char *data, *buffer;
380
385
struct delta_index *index;
381
386
struct unpacked_index_entry *entry, **hash;
383
388
unsigned long memsize;
385
390
if (!src->buf || !src->size)
391
return DELTA_SOURCE_EMPTY;
387
392
buffer = src->buf;
389
394
/* Determine index hash size. Note that indexing skips the
390
first byte to allow for optimizing the Rabin's polynomial
391
initialization in create_delta(). */
395
first byte so we subtract 1 to get the edge cases right.
397
stride = RABIN_WINDOW;
392
398
num_entries = (src->size - 1) / RABIN_WINDOW;
399
if (max_bytes_to_index > 0) {
400
max_entries = (unsigned int) (max_bytes_to_index / RABIN_WINDOW);
401
if (num_entries > max_entries) {
402
/* Limit the max number of matching entries. This reduces the 'best'
403
* possible match, but means we don't consume all of ram.
405
num_entries = max_entries;
406
stride = (src->size - 1) / num_entries;
394
410
total_num_entries = num_entries + old->num_entries;
419
435
hash_count = calloc(hsize, sizeof(*hash_count));
420
436
if (!hash_count) {
438
return DELTA_OUT_OF_MEMORY;
425
441
/* then populate the index for the new data */
427
for (data = buffer + num_entries * RABIN_WINDOW - RABIN_WINDOW;
443
for (data = buffer + num_entries * stride - RABIN_WINDOW;
429
data -= RABIN_WINDOW) {
430
446
unsigned int val = 0;
431
447
for (i = 1; i <= RABIN_WINDOW; i++)
432
448
val = ((val << 8) | data[i]) ^ T[val >> RABIN_SHIFT];
450
466
total_num_entries = limit_hash_buckets(hash, hash_count, hsize,
451
467
total_num_entries);
452
468
free(hash_count);
456
469
index = pack_delta_index(hash, hsize, total_num_entries, old);
471
/* pack_delta_index only returns NULL on malloc failure */
473
return DELTA_OUT_OF_MEMORY;
461
475
index->last_src = src;
465
480
/* Take some entries, and put them into a custom hash.
473
488
unsigned int hsize)
475
490
unsigned int hash_offset, hmask, memsize;
476
struct index_entry *entry, *last_entry;
491
struct index_entry *entry;
477
492
struct index_entry_linked_list *out_entry, **hash;
493
508
/* We know that entries are in the order we want in the output, but they
494
509
* aren't "grouped" by hash bucket yet.
496
last_entry = entries + num_entries;
497
511
for (entry = entries + num_entries - 1; entry >= entries; --entry) {
498
512
hash_offset = entry->val & hmask;
499
513
out_entry->p_entry = entry;
518
532
unsigned int i, j, hsize, hmask, total_num_entries;
519
533
struct delta_index *index;
520
534
struct index_entry *entry, *packed_entry, **packed_hash;
521
struct index_entry *last_entry, null_entry = {0};
535
struct index_entry null_entry = {0};
523
537
unsigned long memsize;
524
538
struct index_entry_linked_list *unpacked_entry, **mini_hash;
683
696
create_delta_index_from_delta(const struct source_info *src,
684
struct delta_index *old_index)
697
struct delta_index *old_index,
698
struct delta_index **fresh)
686
700
unsigned int i, num_entries, max_num_entries, prev_val, num_inserted;
687
701
unsigned int hash_offset;
690
704
struct delta_index *new_index;
691
705
struct index_entry *entry, *entries;
708
return DELTA_INDEX_NEEDED;
693
709
if (!src->buf || !src->size)
710
return DELTA_SOURCE_EMPTY;
695
711
buffer = src->buf;
696
712
top = buffer + src->size;
705
721
max_num_entries = (src->size - 1) / RABIN_WINDOW;
723
if (!max_num_entries) {
707
728
/* allocate an array to hold whatever entries we find */
708
729
entries = malloc(sizeof(*entry) * max_num_entries);
709
730
if (!entries) /* malloc failure */
731
return DELTA_OUT_OF_MEMORY;
712
733
/* then populate the index for the new data */
776
797
if (data != top) {
777
/* Something was wrong with this delta */
798
/* The source_info data passed was corrupted or otherwise invalid */
800
return DELTA_SOURCE_BAD;
781
802
if (num_entries == 0) {
782
803
/** Nothing to index **/
786
assert(old_index != NULL);
787
808
old_index->last_src = src;
788
809
/* See if we can fill in these values into the holes in the array */
841
862
new_index = create_index_from_old_and_new_entries(old_index,
842
863
entry, num_entries);
865
new_index = old_index;
845
866
// fprintf(stderr, "inserted %d without resizing\n", num_inserted);
869
/* create_index_from_old_and_new_entries returns NULL on malloc failure */
871
return DELTA_OUT_OF_MEMORY;
851
876
void free_delta_index(struct delta_index *index)
869
894
#define MAX_OP_SIZE (5 + 5 + 1 + RABIN_WINDOW + 7)
872
897
create_delta(const struct delta_index *index,
873
898
const void *trg_buf, unsigned long trg_size,
874
unsigned long *delta_size, unsigned long max_size)
899
unsigned long *delta_size, unsigned long max_size,
876
902
unsigned int i, outpos, outsize, moff, val;
1082
1108
if (max_size && outpos > max_size) {
1110
return DELTA_SIZE_TOO_BIG;
1087
1113
*delta_size = outpos;
1120
get_entry_summary(const struct delta_index *index, int pos,
1121
unsigned int *text_offset, unsigned int *hash_val)
1124
const struct index_entry *entry;
1125
const struct index_entry *start_of_entries;
1126
unsigned int offset;
1127
if (pos < 0 || text_offset == NULL || hash_val == NULL
1132
hsize = index->hash_mask + 1;
1133
start_of_entries = (struct index_entry *)(((struct index_entry **)index->hash) + (hsize + 1));
1134
entry = start_of_entries + pos;
1135
if (entry > index->last_entry) {
1138
if (entry->ptr == NULL) {
1142
offset = entry->src->agg_offset;
1143
offset += (entry->ptr - ((unsigned char *)entry->src->buf));
1144
*text_offset = offset;
1145
*hash_val = entry->val;
1152
get_hash_offset(const struct delta_index *index, int pos,
1153
unsigned int *entry_offset)
1156
const struct index_entry *entry;
1157
const struct index_entry *start_of_entries;
1158
if (pos < 0 || index == NULL || entry_offset == NULL)
1162
hsize = index->hash_mask + 1;
1163
start_of_entries = (struct index_entry *)(((struct index_entry **)index->hash) + (hsize + 1));
1167
entry = index->hash[pos];
1168
if (entry == NULL) {
1171
*entry_offset = (entry - start_of_entries);
1178
rabin_hash(const unsigned char *data)
1181
unsigned int val = 0;
1182
for (i = 0; i < RABIN_WINDOW; i++)
1183
val = ((val << 8) | data[i]) ^ T[val >> RABIN_SHIFT];
1091
1187
/* vim: et ts=4 sw=4 sts=4