Commit 5e5848eb15cc0dd8476d1c6882a9f770e6556586

Russell Belfer 2013-02-14T17:25:10

Change similarity metric to sampled hashes This moves the similarity metric code out of buf_text and into a new file. Also, this implements a different approach to similarity measurement based on a Rabin-Karp rolling hash where we only keep the top 100 and bottom 100 hashes. In theory, that should be sufficient samples to given a fairly accurate measurement while limiting the amount of data we keep for file signatures no matter how large the file is.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
diff --git a/src/buf_text.c b/src/buf_text.c
index 49ec16a..3a8f442 100644
--- a/src/buf_text.c
+++ b/src/buf_text.c
@@ -5,7 +5,6 @@
  * a Linking Exception. For full terms see the included COPYING file.
  */
 #include "buf_text.h"
-#include "fileops.h"
 
 int git_buf_text_puts_escaped(
 	git_buf *buf,
@@ -213,362 +212,3 @@ bool git_buf_text_gather_stats(
 	return (stats->nul > 0 ||
 		((stats->printable >> 7) < stats->nonprintable));
 }
-
-#define SIMILARITY_MAXRUN 256
-#define SIMILARITY_HASH_START  5381
-#define SIMILARITY_HASH_UPDATE(S,N) (((S) << 5) + (S) + (uint32_t)(N))
-
-enum {
-	SIMILARITY_FORMAT_UNKNOWN = 0,
-	SIMILARITY_FORMAT_TEXT = 1,
-	SIMILARITY_FORMAT_BINARY = 2
-};
-
-struct git_buf_text_hashsig {
-	uint32_t *hashes;
-	size_t size;
-	size_t asize;
-	unsigned int format : 2;
-	unsigned int pairs : 1;
-};
-
-static int similarity_record_hash(git_buf_text_hashsig *sig, uint32_t hash)
-{
-	if (sig->size >= sig->asize) {
-		size_t new_asize = sig->asize + 512;
-		uint32_t *new_hashes =
-			git__realloc(sig->hashes, new_asize * sizeof(uint32_t));
-		GITERR_CHECK_ALLOC(new_hashes);
-
-		sig->hashes = new_hashes;
-		sig->asize  = new_asize;
-	}
-
-	sig->hashes[sig->size++] = hash;
-	return 0;
-}
-
-static int similarity_add_hashes_text(
-	git_buf_text_hashsig *sig,
-	uint32_t *hash_start,
-	size_t *hashlen_start,
-	const char *ptr,
-	size_t len)
-{
-	int error;
-	const char *scan = ptr, *scan_end = ptr + len;
-	uint32_t hash = *hash_start;
-	size_t hashlen = *hashlen_start;
-
-	while (scan < scan_end) {
-		char ch = *scan++;
-
-		if (ch == '\r' || ch == '\n' || hashlen >= SIMILARITY_MAXRUN) {
-			if ((error = similarity_record_hash(sig, hash)) < 0)
-				break;
-
-			hash = SIMILARITY_HASH_START;
-			hashlen = 0;
-
-			/* skip all whitespace immediately after line ending */
-			while (scan < scan_end && git__isspace(*scan))
-				scan++;
-		} else {
-			hash = SIMILARITY_HASH_UPDATE(hash, ch);
-			hashlen++;
-		}
-	}
-
-	*hash_start = hash;
-	*hashlen_start = hashlen;
-
-	return error;
-}
-
-static int similarity_add_hashes_binary(
-	git_buf_text_hashsig *sig,
-	uint32_t *hash_start,
-	size_t *hashlen_start,
-	const char *ptr,
-	size_t len)
-{
-	int error;
-	const char *scan = ptr, *scan_end = ptr + len;
-	uint32_t hash = *hash_start;
-	size_t hashlen = *hashlen_start;
-
-	while (scan < scan_end) {
-		char ch = *scan++;
-
-		if (!ch || hashlen >= SIMILARITY_MAXRUN) {
-			if ((error = similarity_record_hash(sig, hash)) < 0)
-				break;
-
-			hash = SIMILARITY_HASH_START;
-			hashlen = 0;
-
-			/* skip run of terminators */
-			while (scan < scan_end && !*scan)
-				scan++;
-		} else {
-			hash = SIMILARITY_HASH_UPDATE(hash, ch);
-			hashlen++;
-		}
-	}
-
-	*hash_start = hash;
-	*hashlen_start = hashlen;
-
-	return error;
-}
-
-static int similarity_add_hashes(
-	git_buf_text_hashsig *sig,
-	uint32_t *hash_start,
-	size_t *hashlen_start,
-	const char *ptr,
-	size_t len)
-{
-	int error = 0;
-	uint32_t hash = hash_start ? *hash_start : SIMILARITY_HASH_START;
-	size_t hashlen = hashlen_start ? *hashlen_start : 0;
-
-	if (sig->format == SIMILARITY_FORMAT_TEXT)
-		error = similarity_add_hashes_text(sig, &hash, &hashlen, ptr, len);
-	else
-		error = similarity_add_hashes_binary(sig, &hash, &hashlen, ptr, len);
-
-	if (hash_start)
-		*hash_start = hash;
-	if (hashlen_start)
-		*hashlen_start = hashlen;
-
-	/* if we're not saving intermediate state, add final hash as needed */
-	if (!error && !hash_start && hashlen > 0)
-		error = similarity_record_hash(sig, hash);
-
-	return error;
-}
-
-/*
- * Decide if \0 or \n terminated runs are a better choice for hashes
- */
-static void similarity_guess_format(
-	git_buf_text_hashsig *sig, const char *ptr, size_t len)
-{
-	size_t lines = 0, line_length = 0, max_line_length = 0;
-	size_t runs = 0, run_length = 0, max_run_length = 0;
-
-	/* don't process more than 4k of data for this */
-	if (len > 4096)
-		len = 4096;
-
-	/* gather some stats */
-	while (len--) {
-		char ch = *ptr++;
-
-		if (ch == '\0') {
-			runs++;
-			if (max_run_length < run_length)
-				max_run_length = run_length;
-			run_length = 0;
-		} else if (ch == '\n') {
-			lines++;
-			if (max_line_length < line_length)
-				max_line_length = line_length;
-			line_length = 0;
-		} else {
-			run_length++;
-			line_length++;
-		}
-	}
-
-	/* the following heuristic could probably be improved */
-	if (lines > runs)
-		sig->format = SIMILARITY_FORMAT_TEXT;
-	else if (runs > 0)
-		sig->format = SIMILARITY_FORMAT_BINARY;
-	else
-		sig->format = SIMILARITY_FORMAT_UNKNOWN;
-}
-
-static int similarity_compare_score(const void *a, const void *b)
-{
-	uint32_t av = *(uint32_t *)a, bv = *(uint32_t *)b;
-	return (av < bv) ? -1 : (av > bv) ? 1 : 0;
-}
-
-static int similarity_finalize_hashes(
-	git_buf_text_hashsig *sig, bool generate_pairs)
-{
-	if (!sig->size)
-		return 0;
-
-	/* create pairwise hashes if requested */
-
-	if (generate_pairs) {
-		size_t i, needed_size = sig->size * 2 - 1;
-
-		if (needed_size > sig->asize) {
-			uint32_t *new_hashes =
-				git__realloc(sig->hashes, needed_size * sizeof(uint32_t));
-			GITERR_CHECK_ALLOC(new_hashes);
-
-			sig->hashes = new_hashes;
-			sig->asize  = needed_size;
-		}
-
-		for (i = 1; i < sig->size; ++i)
-			sig->hashes[sig->size + i - 1] =
-				SIMILARITY_HASH_UPDATE(sig->hashes[i - 1], sig->hashes[i]);
-
-		sig->pairs = 1;
-	}
-
-	/* sort all hashes */
-
-	qsort(sig->hashes, sig->size, sizeof(uint32_t), similarity_compare_score);
-
-	if (generate_pairs)
-		qsort(&sig->hashes[sig->size], sig->size - 1, sizeof(uint32_t),
-			similarity_compare_score);
-
-	return 0;
-}
-
-int git_buf_text_hashsig_create(
-	git_buf_text_hashsig **out,
-	const git_buf *buf,
-	bool generate_pairs)
-{
-	int error;
-	git_buf_text_hashsig *sig = git__calloc(1, sizeof(git_buf_text_hashsig));
-	GITERR_CHECK_ALLOC(sig);
-
-	similarity_guess_format(sig, buf->ptr, buf->size);
-
-	error = similarity_add_hashes(sig, NULL, NULL, buf->ptr, buf->size);
-
-	if (!error)
-		error = similarity_finalize_hashes(sig, generate_pairs);
-
-	if (!error)
-		*out = sig;
-	else
-		git_buf_text_hashsig_free(sig);
-
-	return error;
-}
-
-int git_buf_text_hashsig_create_fromfile(
-	git_buf_text_hashsig **out,
-	const char *path,
-	bool generate_pairs)
-{
-	char buf[4096];
-	ssize_t buflen = 0;
-	uint32_t hash = SIMILARITY_HASH_START;
-	size_t hashlen = 0;
-	int error = 0, fd;
-	git_buf_text_hashsig *sig = git__calloc(1, sizeof(git_buf_text_hashsig));
-	GITERR_CHECK_ALLOC(sig);
-
-	if ((fd = git_futils_open_ro(path)) < 0) {
-		git__free(sig);
-		return fd;
-	}
-
-	while (!error && (buflen = p_read(fd, buf, sizeof(buf))) > 0) {
-		if (sig->format == SIMILARITY_FORMAT_UNKNOWN)
-			similarity_guess_format(sig, buf, buflen);
-
-		error = similarity_add_hashes(sig, &hash, &hashlen, buf, buflen);
-	}
-
-	if (buflen < 0) {
-		giterr_set(GITERR_OS,
-			"Read error on '%s' while calculating similarity hashes", path);
-		error = (int)buflen;
-	}
-
-	p_close(fd);
-
-	if (!error && hashlen > 0)
-		error = similarity_record_hash(sig, hash);
-
-	if (!error)
-		error = similarity_finalize_hashes(sig, generate_pairs);
-
-	if (!error)
-		*out = sig;
-	else
-		git_buf_text_hashsig_free(sig);
-
-	return error;
-}
-
-void git_buf_text_hashsig_free(git_buf_text_hashsig *sig)
-{
-	if (!sig)
-		return;
-
-	if (sig->hashes) {
-		git__free(sig->hashes);
-		sig->hashes = NULL;
-	}
-
-	git__free(sig);
-}
-
-int git_buf_text_hashsig_compare(
-	const git_buf_text_hashsig *a,
-	const git_buf_text_hashsig *b,
-	int scale)
-{
-	size_t matches = 0, pairs = 0, total = 0, i, j;
-
-	if (a->format != b->format || !a->size || !b->size)
-		return 0;
-
-	if (scale <= 0)
-		scale = 100;
-
-	/* hash lists are sorted - just look for overlap vs total */
-
-	for (i = 0, j = 0; i < a->size && j < b->size; ) {
-		uint32_t av = a->hashes[i];
-		uint32_t bv = b->hashes[j];
-
-		if (av < bv)
-			++i;
-		else if (av > bv)
-			++j;
-		else {
-			++i; ++j;
-			++matches;
-		}
-	}
-
-	total = (a->size + b->size);
-
-	if (a->pairs && b->pairs) {
-		for (i = 0, j = 0; i < a->size - 1 && j < b->size - 1; ) {
-			uint32_t av = a->hashes[i + a->size];
-			uint32_t bv = b->hashes[j + b->size];
-
-			if (av < bv)
-				++i;
-			else if (av > bv)
-				++j;
-			else {
-				++i; ++j;
-				++pairs;
-			}
-		}
-
-		total += (a->size + b->size - 2);
-	}
-
-	return (int)(scale * 2 * (matches + pairs) / total);
-}
-
diff --git a/src/buf_text.h b/src/buf_text.h
index c48c010..458ee33 100644
--- a/src/buf_text.h
+++ b/src/buf_text.h
@@ -105,52 +105,4 @@ extern int git_buf_text_detect_bom(
 extern bool git_buf_text_gather_stats(
 	git_buf_text_stats *stats, const git_buf *buf, bool skip_bom);
 
-/**
- * Similarity signature of line hashes for a buffer
- */
-typedef struct git_buf_text_hashsig git_buf_text_hashsig;
-
-/**
- * Build a similarity signature for a buffer
- *
- * This can either generate a simple array of hashed lines/runs in the
- * file, or it can also keep hashes of pairs of runs in sequence.  Adding
- * the pairwise runs means the final score will be sensitive to line
- * ordering changes as well as individual line contents.
- *
- * @param out The array of hashed runs representing the file content
- * @param buf The contents of the file to hash
- * @param generate_pairwise_hashes Should pairwise runs be hashed
- */
-extern int git_buf_text_hashsig_create(
-	git_buf_text_hashsig **out,
-	const git_buf *buf,
-	bool generate_pairwise_hashes);
-
-/**
- * Build a similarity signature from a file
- *
- * This walks through the file, only loading a maximum of 4K of file data at
- * a time.  Otherwise, it acts just like `git_buf_text_hashsig_create`.
- */
-extern int git_buf_text_hashsig_create_fromfile(
-	git_buf_text_hashsig **out,
-	const char *path,
-	bool generate_pairwise_hashes);
-
-/**
- * Release memory for a content similarity signature
- */
-extern void git_buf_text_hashsig_free(git_buf_text_hashsig *sig);
-
-/**
- * Measure similarity between two files
- *
- * @return <0 for error, [0 to scale] as similarity score
- */
-extern int git_buf_text_hashsig_compare(
-	const git_buf_text_hashsig *a,
-	const git_buf_text_hashsig *b,
-	int scale);
-
 #endif
diff --git a/src/diff_tform.c b/src/diff_tform.c
index dbcdc27..3939230 100644
--- a/src/diff_tform.c
+++ b/src/diff_tform.c
@@ -7,7 +7,7 @@
 #include "common.h"
 #include "diff.h"
 #include "git2/config.h"
-#include "buf_text.h"
+#include "hashsig.h"
 
 static git_diff_delta *diff_delta__dup(
 	const git_diff_delta *d, git_pool *pool)
@@ -300,7 +300,7 @@ on_error:
 
 typedef struct {
 	/* array of delta index * 2 + (old_file/new_file) -> file hashes */
-	git_buf_text_hashsig *sigs;
+	git_hashsig *sigs;
 } diff_similarity_cache;
 
 static unsigned int calc_similarity(
@@ -308,6 +308,8 @@ static unsigned int calc_similarity(
 {
 	diff_similarity_cache *cache = ref;
 
+	GIT_UNUSED(cache);
+
 	if (git_oid_cmp(&old_file->oid, &new_file->oid) == 0)
 		return 100;
 
diff --git a/src/hashsig.c b/src/hashsig.c
new file mode 100644
index 0000000..a9cd8fa
--- /dev/null
+++ b/src/hashsig.c
@@ -0,0 +1,364 @@
+/*
+ * Copyright (C) the libgit2 contributors. All rights reserved.
+ *
+ * This file is part of libgit2, distributed under the GNU GPL v2 with
+ * a Linking Exception. For full terms see the included COPYING file.
+ */
+#include "hashsig.h"
+#include "fileops.h"
+
+typedef uint32_t hashsig_t;
+typedef uint64_t hashsig_state;
+
+#define HASHSIG_SCALE 100
+
+#define HASHSIG_HASH_WINDOW 32
+#define HASHSIG_HASH_START	0
+#define HASHSIG_HASH_SHIFT  5
+#define HASHSIG_HASH_MASK   0x7FFFFFFF
+
+#define HASHSIG_HEAP_SIZE ((1 << 7) - 1)
+
+typedef int (*hashsig_cmp)(const void *a, const void *b);
+
+typedef struct {
+	int size, asize;
+	hashsig_cmp cmp;
+	hashsig_t values[HASHSIG_HEAP_SIZE];
+} hashsig_heap;
+
+typedef struct {
+	hashsig_state state, shift_n;
+	char window[HASHSIG_HASH_WINDOW];
+	int win_len, win_pos, saw_lf;
+} hashsig_in_progress;
+
+#define HASHSIG_IN_PROGRESS_INIT { HASHSIG_HASH_START, 1, {0}, 0, 0, 1 }
+
+struct git_hashsig {
+	hashsig_heap mins;
+	hashsig_heap maxs;
+	git_hashsig_option_t opt;
+	int considered;
+};
+
+#define HEAP_LCHILD_OF(I) (((I)*2)+1)
+#define HEAP_RCHILD_OF(I) (((I)*2)+2)
+#define HEAP_PARENT_OF(I) (((I)-1)>>1)
+
+static void hashsig_heap_init(hashsig_heap *h, hashsig_cmp cmp)
+{
+	h->size  = 0;
+	h->asize = HASHSIG_HEAP_SIZE;
+	h->cmp   = cmp;
+}
+
+static int hashsig_cmp_max(const void *a, const void *b)
+{
+	hashsig_t av = *(const hashsig_t *)a, bv = *(const hashsig_t *)b;
+	return (av < bv) ? -1 : (av > bv) ? 1 : 0;
+}
+
+static int hashsig_cmp_min(const void *a, const void *b)
+{
+	hashsig_t av = *(const hashsig_t *)a, bv = *(const hashsig_t *)b;
+	return (av > bv) ? -1 : (av < bv) ? 1 : 0;
+}
+
+static void hashsig_heap_up(hashsig_heap *h, int el)
+{
+	int parent_el = HEAP_PARENT_OF(el);
+
+	while (el > 0 && h->cmp(&h->values[parent_el], &h->values[el]) > 0) {
+		hashsig_t t = h->values[el];
+		h->values[el] = h->values[parent_el];
+		h->values[parent_el] = t;
+
+		el = parent_el;
+		parent_el = HEAP_PARENT_OF(el);
+	}
+}
+
+static void hashsig_heap_down(hashsig_heap *h, int el)
+{
+	hashsig_t v, lv, rv;
+
+	/* 'el < h->size / 2' tests if el is bottom row of heap */
+
+	while (el < h->size / 2) {
+		int lel = HEAP_LCHILD_OF(el), rel = HEAP_RCHILD_OF(el), swapel;
+
+		v  = h->values[el];
+		lv = h->values[lel];
+		rv = h->values[rel];
+
+		if (h->cmp(&v, &lv) < 0 && h->cmp(&v, &rv) < 0)
+			break;
+
+		swapel = (h->cmp(&lv, &rv) < 0) ? lel : rel;
+
+		h->values[el] = h->values[swapel];
+		h->values[swapel] = v;
+
+		el = swapel;
+	}
+}
+
+static void hashsig_heap_sort(hashsig_heap *h)
+{
+	/* only need to do this at the end for signature comparison */
+	qsort(h->values, h->size, sizeof(hashsig_t), h->cmp);
+}
+
+static void hashsig_heap_insert(hashsig_heap *h, hashsig_t val)
+{
+	/* if heap is full, pop top if new element should replace it */
+	if (h->size == h->asize && h->cmp(&val, &h->values[0]) > 0) {
+		h->size--;
+		h->values[0] = h->values[h->size];
+		hashsig_heap_down(h, 0);
+	}
+
+	/* if heap is not full, insert new element */
+	if (h->size < h->asize) {
+		h->values[h->size++] = val;
+		hashsig_heap_up(h, h->size - 1);
+	}
+}
+
+GIT_INLINE(bool) hashsig_include_char(
+	char ch, git_hashsig_option_t opt, int *saw_lf)
+{
+	if ((opt & GIT_HASHSIG_IGNORE_WHITESPACE) && git__isspace(ch))
+		return false;
+
+	if (opt & GIT_HASHSIG_SMART_WHITESPACE) {
+		if (ch == '\r' || (*saw_lf && git__isspace(ch)))
+			return false;
+
+		*saw_lf = (ch == '\n');
+	}
+
+	return true;
+}
+
+static void hashsig_initial_window(
+	git_hashsig *sig,
+	const char **data,
+	size_t size,
+	hashsig_in_progress *prog)
+{
+	hashsig_state state, shift_n;
+	int win_len;
+	const char *scan, *end;
+
+	/* init until we have processed at least HASHSIG_HASH_WINDOW data */
+
+	if (prog->win_len >= HASHSIG_HASH_WINDOW)
+		return;
+
+	state   = prog->state;
+	win_len = prog->win_len;
+	shift_n = prog->shift_n;
+
+	scan = *data;
+	end  = scan + size;
+
+	while (scan < end && win_len < HASHSIG_HASH_WINDOW) {
+		char ch = *scan++;
+
+		if (!hashsig_include_char(ch, sig->opt, &prog->saw_lf))
+			continue;
+
+		state = (state * HASHSIG_HASH_SHIFT + ch) & HASHSIG_HASH_MASK;
+
+		if (!win_len)
+			shift_n = 1;
+		else
+			shift_n = (shift_n * HASHSIG_HASH_SHIFT) & HASHSIG_HASH_MASK;
+
+		prog->window[win_len++] = ch;
+	}
+
+	/* insert initial hash if we just finished */
+
+	if (win_len == HASHSIG_HASH_WINDOW) {
+		hashsig_heap_insert(&sig->mins, state);
+		hashsig_heap_insert(&sig->maxs, state);
+		sig->considered = 1;
+	}
+
+	prog->state   = state;
+	prog->win_len = win_len;
+	prog->shift_n = shift_n;
+
+	*data = scan;
+}
+
+static int hashsig_add_hashes(
+	git_hashsig *sig,
+	const char *data,
+	size_t size,
+	hashsig_in_progress *prog)
+{
+	const char *scan = data, *end = data + size;
+	hashsig_state state, shift_n, rmv;
+
+	if (prog->win_len < HASHSIG_HASH_WINDOW)
+		hashsig_initial_window(sig, &scan, size, prog);
+
+	state   = prog->state;
+	shift_n = prog->shift_n;
+
+	/* advance window, adding new chars and removing old */
+
+	for (; scan < end; ++scan) {
+		char ch = *scan;
+
+		if (!hashsig_include_char(ch, sig->opt, &prog->saw_lf))
+			continue;
+
+		rmv = shift_n * prog->window[prog->win_pos];
+
+		state = (state - rmv) & HASHSIG_HASH_MASK;
+		state = (state * HASHSIG_HASH_SHIFT) & HASHSIG_HASH_MASK;
+		state = (state + ch) & HASHSIG_HASH_MASK;
+
+		hashsig_heap_insert(&sig->mins, state);
+		hashsig_heap_insert(&sig->maxs, state);
+		sig->considered++;
+
+		prog->window[prog->win_pos] = ch;
+		prog->win_pos = (prog->win_pos + 1) % HASHSIG_HASH_WINDOW;
+	}
+
+	prog->state = state;
+
+	return 0;
+}
+
+static int hashsig_finalize_hashes(git_hashsig *sig)
+{
+	if (sig->mins.size < HASHSIG_HEAP_SIZE) {
+		giterr_set(GITERR_INVALID,
+			"File too small for similarity signature calculation");
+		return GIT_EBUFS;
+	}
+
+	hashsig_heap_sort(&sig->mins);
+	hashsig_heap_sort(&sig->maxs);
+
+	return 0;
+}
+
+static git_hashsig *hashsig_alloc(git_hashsig_option_t opts)
+{
+	git_hashsig *sig = git__calloc(1, sizeof(git_hashsig));
+	if (!sig)
+		return NULL;
+
+	hashsig_heap_init(&sig->mins, hashsig_cmp_min);
+	hashsig_heap_init(&sig->maxs, hashsig_cmp_max);
+	sig->opt = opts;
+
+	return sig;
+}
+
+int git_hashsig_create(
+	git_hashsig **out,
+	const git_buf *buf,
+	git_hashsig_option_t opts)
+{
+	int error;
+	hashsig_in_progress prog = HASHSIG_IN_PROGRESS_INIT;
+	git_hashsig *sig = hashsig_alloc(opts);
+	GITERR_CHECK_ALLOC(sig);
+
+	error = hashsig_add_hashes(sig, buf->ptr, buf->size, &prog);
+
+	if (!error)
+		error = hashsig_finalize_hashes(sig);
+
+	if (!error)
+		*out = sig;
+	else
+		git_hashsig_free(sig);
+
+	return error;
+}
+
+int git_hashsig_create_fromfile(
+	git_hashsig **out,
+	const char *path,
+	git_hashsig_option_t opts)
+{
+	char buf[4096];
+	ssize_t buflen = 0;
+	int error = 0, fd;
+	hashsig_in_progress prog = HASHSIG_IN_PROGRESS_INIT;
+	git_hashsig *sig = hashsig_alloc(opts);
+	GITERR_CHECK_ALLOC(sig);
+
+	if ((fd = git_futils_open_ro(path)) < 0) {
+		git__free(sig);
+		return fd;
+	}
+
+	while (!error) {
+		if ((buflen = p_read(fd, buf, sizeof(buf))) <= 0) {
+			if ((error = buflen) < 0)
+				giterr_set(GITERR_OS,
+					"Read error on '%s' calculating similarity hashes", path);
+			break;
+		}
+
+		error = hashsig_add_hashes(sig, buf, buflen, &prog);
+	}
+
+	p_close(fd);
+
+	if (!error)
+		error = hashsig_finalize_hashes(sig);
+
+	if (!error)
+		*out = sig;
+	else
+		git_hashsig_free(sig);
+
+	return error;
+}
+
+void git_hashsig_free(git_hashsig *sig)
+{
+	git__free(sig);
+}
+
+static int hashsig_heap_compare(const hashsig_heap *a, const hashsig_heap *b)
+{
+	int matches = 0, i, j, cmp;
+
+	assert(a->cmp == b->cmp);
+
+	/* hash heaps are sorted - just look for overlap vs total */
+
+	for (i = 0, j = 0; i < a->size && j < b->size; ) {
+		cmp = a->cmp(&a->values[i], &b->values[j]);
+
+		if (cmp < 0)
+			++i;
+		else if (cmp > 0)
+			++j;
+		else {
+			++i; ++j; ++matches;
+		}
+	}
+
+	return HASHSIG_SCALE * (matches * 2) / (a->size + b->size);
+}
+
+int git_hashsig_compare(const git_hashsig *a, const git_hashsig *b)
+{
+	return (hashsig_heap_compare(&a->mins, &b->mins) +
+			hashsig_heap_compare(&a->maxs, &b->maxs)) / 2;
+}
+
diff --git a/src/hashsig.h b/src/hashsig.h
new file mode 100644
index 0000000..70b47f5
--- /dev/null
+++ b/src/hashsig.h
@@ -0,0 +1,70 @@
+/*
+ * Copyright (C) the libgit2 contributors. All rights reserved.
+ *
+ * This file is part of libgit2, distributed under the GNU GPL v2 with
+ * a Linking Exception. For full terms see the included COPYING file.
+ */
+#ifndef INCLUDE_hashsig_h__
+#define INCLUDE_hashsig_h__
+
+#include "buffer.h"
+
+/**
+ * Similarity signature of line hashes for a buffer
+ */
+typedef struct git_hashsig git_hashsig;
+
+typedef enum {
+	GIT_HASHSIG_NORMAL = 0, /* use all data */
+	GIT_HASHSIG_IGNORE_WHITESPACE = 1, /* ignore whitespace */
+	GIT_HASHSIG_SMART_WHITESPACE = 2, /* ignore \r and all space after \n */
+} git_hashsig_option_t;
+
+/**
+ * Build a similarity signature for a buffer
+ *
+ * If you have passed a whitespace-ignoring buffer, then the whitespace
+ * will be removed from the buffer while it is being processed, modifying
+ * the buffer in place.  Sorry about that!
+ *
+ * This will return an error if the buffer doesn't contain enough data to
+ * compute a valid signature.
+ *
+ * @param out The array of hashed runs representing the file content
+ * @param buf The contents of the file to hash
+ * @param generate_pairwise_hashes Should pairwise runs be hashed
+ */
+extern int git_hashsig_create(
+	git_hashsig **out,
+	const git_buf *buf,
+	git_hashsig_option_t opts);
+
+/**
+ * Build a similarity signature from a file
+ *
+ * This walks through the file, only loading a maximum of 4K of file data at
+ * a time.  Otherwise, it acts just like `git_hashsig_create`.
+ *
+ * This will return an error if the file doesn't contain enough data to
+ * compute a valid signature.
+ */
+extern int git_hashsig_create_fromfile(
+	git_hashsig **out,
+	const char *path,
+	git_hashsig_option_t opts);
+
+/**
+ * Release memory for a content similarity signature
+ */
+extern void git_hashsig_free(git_hashsig *sig);
+
+/**
+ * Measure similarity between two files
+ *
+ * @return <0 for error, [0 to 100] as similarity score
+ */
+extern int git_hashsig_compare(
+	const git_hashsig *a,
+	const git_hashsig *b);
+
+#endif
diff --git a/tests-clar/core/buffer.c b/tests-clar/core/buffer.c
index 63753bb..6cad05c 100644
--- a/tests-clar/core/buffer.c
+++ b/tests-clar/core/buffer.c
@@ -1,6 +1,7 @@
 #include "clar_libgit2.h"
 #include "buffer.h"
 #include "buf_text.h"
+#include "hashsig.h"
 #include "fileops.h"
 
 #define TESTSTR "Have you seen that? Have you seeeen that??"
@@ -732,89 +733,94 @@ void test_core_buffer__classify_with_utf8(void)
 	cl_assert(git_buf_text_contains_nul(&b));
 }
 
+#define SIMILARITY_TEST_DATA_1 \
+	"test data\nright here\ninline\ntada\nneeds more data\nlots of data\n" \
+	"is this enough?\nthere has to be enough data to fill the hash array!\n" \
+	"Apparently 191 bytes is the minimum amount of data needed.\nHere goes!\n" \
+	"Let's make sure we've got plenty to go with here.\n   smile   \n"
+
 void test_core_buffer__similarity_metric(void)
 {
-	git_buf_text_hashsig *a, *b;
+	git_hashsig *a, *b;
 	git_buf buf = GIT_BUF_INIT;
 	int sim;
 
 	/* in the first case, we compare data to itself and expect 100% match */
 
-	cl_git_pass(git_buf_sets(&buf, "test data\nright here\ninline\ntada"));
-	cl_git_pass(git_buf_text_hashsig_create(&a, &buf, true));
-	cl_git_pass(git_buf_text_hashsig_create(&b, &buf, true));
+	cl_git_pass(git_buf_sets(&buf, SIMILARITY_TEST_DATA_1));
+	cl_git_pass(git_hashsig_create(&a, &buf, GIT_HASHSIG_NORMAL));
+	cl_git_pass(git_hashsig_create(&b, &buf, GIT_HASHSIG_NORMAL));
 
-	cl_assert_equal_i(100, git_buf_text_hashsig_compare(a, b, 100));
+	cl_assert_equal_i(100, git_hashsig_compare(a, b));
 
-	git_buf_text_hashsig_free(a);
-	git_buf_text_hashsig_free(b);
+	git_hashsig_free(a);
+	git_hashsig_free(b);
 
-	/* in the second case, half of a is matched and all of b is matched, so
-	 * we'll expect a score of around 66% to be the similarity score
-	 */
+	/* if we change just a single byte, how much does that change magnify? */
 
-	cl_git_pass(
-		git_buf_sets(&buf, "a\nb\nc\nd\ne\nf\ng\nh\ni\nj\nk\nl\nm\nn\no\np\n"));
-	cl_git_pass(git_buf_text_hashsig_create(&a, &buf, true));
+	cl_git_pass(git_buf_sets(&buf, SIMILARITY_TEST_DATA_1));
+	cl_git_pass(git_hashsig_create(&a, &buf, GIT_HASHSIG_NORMAL));
+	cl_git_pass(git_buf_sets(&buf,
+		"Test data\nright here\ninline\ntada\nneeds more data\nlots of data\n"
+		"is this enough?\nthere has to be enough data to fill the hash array!\n"
+		"Apparently 191 bytes is the minimum amount of data needed.\nHere goes!\n"
+		 "Let's make sure we've got plenty to go with here.\n   smile   \n"));
+	cl_git_pass(git_hashsig_create(&b, &buf, GIT_HASHSIG_NORMAL));
 
-	cl_git_pass(git_buf_sets(&buf, "a\nb\nc\nd\ne\nf\ng\nh"));
-	cl_git_pass(git_buf_text_hashsig_create(&b, &buf, true));
+	sim = git_hashsig_compare(a, b);
 
-	sim = git_buf_text_hashsig_compare(a, b, 100);
-	cl_assert(sim > 60 && sim < 70);
+	cl_assert(95 < sim && sim < 100); /* expect >95% similarity */
 
-	git_buf_text_hashsig_free(a);
-	git_buf_text_hashsig_free(b);
+	git_hashsig_free(a);
+	git_hashsig_free(b);
 
-	/* in the reversed case, 100% of line hashes match, but no pairwise hashes
-	 * match, so we'll expect about a 50% match for a reversed file
-	 */
+	/* let's try comparing data to a superset of itself */
 
-	cl_git_pass(
-		git_buf_sets(&buf, "a\nb\nc\nd\ne\nf\ng\nh\ni\nj\nk\nl\nm\nn\no\np\n"));
-	cl_git_pass(git_buf_text_hashsig_create(&a, &buf, true));
-	cl_git_pass(
-		git_buf_sets(&buf, "p\no\nn\nm\nl\nk\nj\ni\nh\ng\nf\ne\nd\nc\nb\na\n"));
-	cl_git_pass(git_buf_text_hashsig_create(&b, &buf, true));
+	cl_git_pass(git_buf_sets(&buf, SIMILARITY_TEST_DATA_1));
+	cl_git_pass(git_hashsig_create(&a, &buf, GIT_HASHSIG_NORMAL));
+	cl_git_pass(git_buf_sets(&buf, SIMILARITY_TEST_DATA_1
+		"and if I add some more, it should still be pretty similar, yes?\n"));
+	cl_git_pass(git_hashsig_create(&b, &buf, GIT_HASHSIG_NORMAL));
 
-	sim = git_buf_text_hashsig_compare(a, b, 100);
-	cl_assert(sim > 45 && sim < 55);
+	sim = git_hashsig_compare(a, b);
 
-	git_buf_text_hashsig_free(a);
-	git_buf_text_hashsig_free(b);
+	cl_assert(70 < sim && sim < 80); /* expect in the 70-80% similarity range */
 
-	/* if we don't use pairwise signatures, then a reversed file should
-	 * match 100%
-	 */
+	git_hashsig_free(a);
+	git_hashsig_free(b);
+
+	/* what if we keep about half the original data and add half new */
+
+	cl_git_pass(git_buf_sets(&buf, SIMILARITY_TEST_DATA_1));
+	cl_git_pass(git_hashsig_create(&a, &buf, GIT_HASHSIG_NORMAL));
+	cl_git_pass(git_buf_sets(&buf,
+		"test data\nright here\ninline\ntada\nneeds more data\nlots of data\n"
+		"is this enough?\nthere has to be enough data to fill the hash array!\n"
+		"okay, that's half the original\nwhat else can we add?\nmore data\n"
+		 "one more line will complete this\nshort\nlines\ndon't\nmatter\n"));
+	cl_git_pass(git_hashsig_create(&b, &buf, GIT_HASHSIG_NORMAL));
 
-	cl_git_pass(
-		git_buf_sets(&buf, "a\nb\nc\nd\ne\nf\ng\nh\ni\nj\nk\nl\nm\nn\no\np\n"));
-	cl_git_pass(git_buf_text_hashsig_create(&a, &buf, false));
-	cl_git_pass(
-		git_buf_sets(&buf, "p\no\nn\nm\nl\nk\nj\ni\nh\ng\nf\ne\nd\nc\nb\na\n"));
-	cl_git_pass(git_buf_text_hashsig_create(&b, &buf, false));
+	sim = git_hashsig_compare(a, b);
 
-	sim = git_buf_text_hashsig_compare(a, b, 100);
-	cl_assert_equal_i(100, sim);
+	cl_assert(40 < sim && sim < 60); /* expect in the 40-60% similarity range */
 
-	git_buf_text_hashsig_free(a);
-	git_buf_text_hashsig_free(b);
+	git_hashsig_free(a);
+	git_hashsig_free(b);
 
 	/* lastly, let's check that we can hash file content as well */
 
-	cl_git_pass(
-		git_buf_sets(&buf, "a\nb\nc\nd\ne\nf\ng\nh\ni\nj\nk\nl\nm\nn\no\np\n"));
-	cl_git_pass(git_buf_text_hashsig_create(&a, &buf, true));
+	cl_git_pass(git_buf_sets(&buf, SIMILARITY_TEST_DATA_1));
+	cl_git_pass(git_hashsig_create(&a, &buf, GIT_HASHSIG_NORMAL));
 
 	cl_git_pass(git_futils_mkdir("scratch", NULL, 0755, GIT_MKDIR_PATH));
-	cl_git_mkfile("scratch/testdata",
-		"a\nb\nc\nd\ne\nf\ng\nh\ni\nj\nk\nl\nm\nn\no\np\n");
-	cl_git_pass(git_buf_text_hashsig_create_fromfile(&b, "scratch/testdata", true));
+	cl_git_mkfile("scratch/testdata", SIMILARITY_TEST_DATA_1);
+	cl_git_pass(git_hashsig_create_fromfile(
+		&b, "scratch/testdata", GIT_HASHSIG_NORMAL));
 
-	cl_assert_equal_i(100, git_buf_text_hashsig_compare(a, b, 100));
+	cl_assert_equal_i(100, git_hashsig_compare(a, b));
 
-	git_buf_text_hashsig_free(a);
-	git_buf_text_hashsig_free(b);
+	git_hashsig_free(a);
+	git_hashsig_free(b);
 
 	git_buf_free(&buf);
 	git_futils_rmdir_r("scratch", NULL, GIT_RMDIR_REMOVE_FILES);