Add "--dirstat" for some directory statistics
[git] / diff.c
1 /*
2  * Copyright (C) 2005 Junio C Hamano
3  */
4 #include "cache.h"
5 #include "quote.h"
6 #include "diff.h"
7 #include "diffcore.h"
8 #include "delta.h"
9 #include "xdiff-interface.h"
10 #include "color.h"
11 #include "attr.h"
12 #include "run-command.h"
13 #include "utf8.h"
14
15 #ifdef NO_FAST_WORKING_DIRECTORY
16 #define FAST_WORKING_DIRECTORY 0
17 #else
18 #define FAST_WORKING_DIRECTORY 1
19 #endif
20
21 static int diff_detect_rename_default;
22 static int diff_rename_limit_default = 100;
23 static int diff_use_color_default;
24 static const char *external_diff_cmd_cfg;
25 int diff_auto_refresh_index = 1;
26
27 static char diff_colors[][COLOR_MAXLEN] = {
28         "\033[m",       /* reset */
29         "",             /* PLAIN (normal) */
30         "\033[1m",      /* METAINFO (bold) */
31         "\033[36m",     /* FRAGINFO (cyan) */
32         "\033[31m",     /* OLD (red) */
33         "\033[32m",     /* NEW (green) */
34         "\033[33m",     /* COMMIT (yellow) */
35         "\033[41m",     /* WHITESPACE (red background) */
36 };
37
38 static int parse_diff_color_slot(const char *var, int ofs)
39 {
40         if (!strcasecmp(var+ofs, "plain"))
41                 return DIFF_PLAIN;
42         if (!strcasecmp(var+ofs, "meta"))
43                 return DIFF_METAINFO;
44         if (!strcasecmp(var+ofs, "frag"))
45                 return DIFF_FRAGINFO;
46         if (!strcasecmp(var+ofs, "old"))
47                 return DIFF_FILE_OLD;
48         if (!strcasecmp(var+ofs, "new"))
49                 return DIFF_FILE_NEW;
50         if (!strcasecmp(var+ofs, "commit"))
51                 return DIFF_COMMIT;
52         if (!strcasecmp(var+ofs, "whitespace"))
53                 return DIFF_WHITESPACE;
54         die("bad config variable '%s'", var);
55 }
56
57 static struct ll_diff_driver {
58         const char *name;
59         struct ll_diff_driver *next;
60         char *cmd;
61 } *user_diff, **user_diff_tail;
62
63 /*
64  * Currently there is only "diff.<drivername>.command" variable;
65  * because there are "diff.color.<slot>" variables, we are parsing
66  * this in a bit convoluted way to allow low level diff driver
67  * called "color".
68  */
69 static int parse_lldiff_command(const char *var, const char *ep, const char *value)
70 {
71         const char *name;
72         int namelen;
73         struct ll_diff_driver *drv;
74
75         name = var + 5;
76         namelen = ep - name;
77         for (drv = user_diff; drv; drv = drv->next)
78                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
79                         break;
80         if (!drv) {
81                 drv = xcalloc(1, sizeof(struct ll_diff_driver));
82                 drv->name = xmemdupz(name, namelen);
83                 if (!user_diff_tail)
84                         user_diff_tail = &user_diff;
85                 *user_diff_tail = drv;
86                 user_diff_tail = &(drv->next);
87         }
88
89         if (!value)
90                 return error("%s: lacks value", var);
91         drv->cmd = strdup(value);
92         return 0;
93 }
94
95 /*
96  * 'diff.<what>.funcname' attribute can be specified in the configuration
97  * to define a customized regexp to find the beginning of a function to
98  * be used for hunk header lines of "diff -p" style output.
99  */
100 static struct funcname_pattern {
101         char *name;
102         char *pattern;
103         struct funcname_pattern *next;
104 } *funcname_pattern_list;
105
106 static int parse_funcname_pattern(const char *var, const char *ep, const char *value)
107 {
108         const char *name;
109         int namelen;
110         struct funcname_pattern *pp;
111
112         name = var + 5; /* "diff." */
113         namelen = ep - name;
114
115         for (pp = funcname_pattern_list; pp; pp = pp->next)
116                 if (!strncmp(pp->name, name, namelen) && !pp->name[namelen])
117                         break;
118         if (!pp) {
119                 pp = xcalloc(1, sizeof(*pp));
120                 pp->name = xmemdupz(name, namelen);
121                 pp->next = funcname_pattern_list;
122                 funcname_pattern_list = pp;
123         }
124         if (pp->pattern)
125                 free(pp->pattern);
126         pp->pattern = xstrdup(value);
127         return 0;
128 }
129
130 /*
131  * These are to give UI layer defaults.
132  * The core-level commands such as git-diff-files should
133  * never be affected by the setting of diff.renames
134  * the user happens to have in the configuration file.
135  */
136 int git_diff_ui_config(const char *var, const char *value)
137 {
138         if (!strcmp(var, "diff.renamelimit")) {
139                 diff_rename_limit_default = git_config_int(var, value);
140                 return 0;
141         }
142         if (!strcmp(var, "diff.color") || !strcmp(var, "color.diff")) {
143                 diff_use_color_default = git_config_colorbool(var, value, -1);
144                 return 0;
145         }
146         if (!strcmp(var, "diff.renames")) {
147                 if (!value)
148                         diff_detect_rename_default = DIFF_DETECT_RENAME;
149                 else if (!strcasecmp(value, "copies") ||
150                          !strcasecmp(value, "copy"))
151                         diff_detect_rename_default = DIFF_DETECT_COPY;
152                 else if (git_config_bool(var,value))
153                         diff_detect_rename_default = DIFF_DETECT_RENAME;
154                 return 0;
155         }
156         if (!strcmp(var, "diff.autorefreshindex")) {
157                 diff_auto_refresh_index = git_config_bool(var, value);
158                 return 0;
159         }
160         if (!strcmp(var, "diff.external")) {
161                 if (!value)
162                         return config_error_nonbool(var);
163                 external_diff_cmd_cfg = xstrdup(value);
164                 return 0;
165         }
166         if (!prefixcmp(var, "diff.")) {
167                 const char *ep = strrchr(var, '.');
168
169                 if (ep != var + 4) {
170                         if (!strcmp(ep, ".command")) {
171                                 if (!value)
172                                         return config_error_nonbool(var);
173                                 return parse_lldiff_command(var, ep, value);
174                         }
175                 }
176         }
177
178         return git_diff_basic_config(var, value);
179 }
180
181 int git_diff_basic_config(const char *var, const char *value)
182 {
183         if (!prefixcmp(var, "diff.color.") || !prefixcmp(var, "color.diff.")) {
184                 int slot = parse_diff_color_slot(var, 11);
185                 if (!value)
186                         return config_error_nonbool(var);
187                 color_parse(value, var, diff_colors[slot]);
188                 return 0;
189         }
190
191         if (!prefixcmp(var, "diff.")) {
192                 const char *ep = strrchr(var, '.');
193                 if (ep != var + 4) {
194                         if (!strcmp(ep, ".funcname")) {
195                                 if (!value)
196                                         return config_error_nonbool(var);
197                                 return parse_funcname_pattern(var, ep, value);
198                         }
199                 }
200         }
201
202         return git_default_config(var, value);
203 }
204
205 static char *quote_two(const char *one, const char *two)
206 {
207         int need_one = quote_c_style(one, NULL, NULL, 1);
208         int need_two = quote_c_style(two, NULL, NULL, 1);
209         struct strbuf res;
210
211         strbuf_init(&res, 0);
212         if (need_one + need_two) {
213                 strbuf_addch(&res, '"');
214                 quote_c_style(one, &res, NULL, 1);
215                 quote_c_style(two, &res, NULL, 1);
216                 strbuf_addch(&res, '"');
217         } else {
218                 strbuf_addstr(&res, one);
219                 strbuf_addstr(&res, two);
220         }
221         return strbuf_detach(&res, NULL);
222 }
223
224 static const char *external_diff(void)
225 {
226         static const char *external_diff_cmd = NULL;
227         static int done_preparing = 0;
228
229         if (done_preparing)
230                 return external_diff_cmd;
231         external_diff_cmd = getenv("GIT_EXTERNAL_DIFF");
232         if (!external_diff_cmd)
233                 external_diff_cmd = external_diff_cmd_cfg;
234         done_preparing = 1;
235         return external_diff_cmd;
236 }
237
238 static struct diff_tempfile {
239         const char *name; /* filename external diff should read from */
240         char hex[41];
241         char mode[10];
242         char tmp_path[PATH_MAX];
243 } diff_temp[2];
244
245 static int count_lines(const char *data, int size)
246 {
247         int count, ch, completely_empty = 1, nl_just_seen = 0;
248         count = 0;
249         while (0 < size--) {
250                 ch = *data++;
251                 if (ch == '\n') {
252                         count++;
253                         nl_just_seen = 1;
254                         completely_empty = 0;
255                 }
256                 else {
257                         nl_just_seen = 0;
258                         completely_empty = 0;
259                 }
260         }
261         if (completely_empty)
262                 return 0;
263         if (!nl_just_seen)
264                 count++; /* no trailing newline */
265         return count;
266 }
267
268 static void print_line_count(int count)
269 {
270         switch (count) {
271         case 0:
272                 printf("0,0");
273                 break;
274         case 1:
275                 printf("1");
276                 break;
277         default:
278                 printf("1,%d", count);
279                 break;
280         }
281 }
282
283 static void copy_file(int prefix, const char *data, int size,
284                 const char *set, const char *reset)
285 {
286         int ch, nl_just_seen = 1;
287         while (0 < size--) {
288                 ch = *data++;
289                 if (nl_just_seen) {
290                         fputs(set, stdout);
291                         putchar(prefix);
292                 }
293                 if (ch == '\n') {
294                         nl_just_seen = 1;
295                         fputs(reset, stdout);
296                 } else
297                         nl_just_seen = 0;
298                 putchar(ch);
299         }
300         if (!nl_just_seen)
301                 printf("%s\n\\ No newline at end of file\n", reset);
302 }
303
304 static void emit_rewrite_diff(const char *name_a,
305                               const char *name_b,
306                               struct diff_filespec *one,
307                               struct diff_filespec *two,
308                               struct diff_options *o)
309 {
310         int lc_a, lc_b;
311         int color_diff = DIFF_OPT_TST(o, COLOR_DIFF);
312         const char *name_a_tab, *name_b_tab;
313         const char *metainfo = diff_get_color(color_diff, DIFF_METAINFO);
314         const char *fraginfo = diff_get_color(color_diff, DIFF_FRAGINFO);
315         const char *old = diff_get_color(color_diff, DIFF_FILE_OLD);
316         const char *new = diff_get_color(color_diff, DIFF_FILE_NEW);
317         const char *reset = diff_get_color(color_diff, DIFF_RESET);
318         static struct strbuf a_name = STRBUF_INIT, b_name = STRBUF_INIT;
319
320         name_a += (*name_a == '/');
321         name_b += (*name_b == '/');
322         name_a_tab = strchr(name_a, ' ') ? "\t" : "";
323         name_b_tab = strchr(name_b, ' ') ? "\t" : "";
324
325         strbuf_reset(&a_name);
326         strbuf_reset(&b_name);
327         quote_two_c_style(&a_name, o->a_prefix, name_a, 0);
328         quote_two_c_style(&b_name, o->b_prefix, name_b, 0);
329
330         diff_populate_filespec(one, 0);
331         diff_populate_filespec(two, 0);
332         lc_a = count_lines(one->data, one->size);
333         lc_b = count_lines(two->data, two->size);
334         printf("%s--- %s%s%s\n%s+++ %s%s%s\n%s@@ -",
335                metainfo, a_name.buf, name_a_tab, reset,
336                metainfo, b_name.buf, name_b_tab, reset, fraginfo);
337         print_line_count(lc_a);
338         printf(" +");
339         print_line_count(lc_b);
340         printf(" @@%s\n", reset);
341         if (lc_a)
342                 copy_file('-', one->data, one->size, old, reset);
343         if (lc_b)
344                 copy_file('+', two->data, two->size, new, reset);
345 }
346
347 static int fill_mmfile(mmfile_t *mf, struct diff_filespec *one)
348 {
349         if (!DIFF_FILE_VALID(one)) {
350                 mf->ptr = (char *)""; /* does not matter */
351                 mf->size = 0;
352                 return 0;
353         }
354         else if (diff_populate_filespec(one, 0))
355                 return -1;
356         mf->ptr = one->data;
357         mf->size = one->size;
358         return 0;
359 }
360
361 struct diff_words_buffer {
362         mmfile_t text;
363         long alloc;
364         long current; /* output pointer */
365         int suppressed_newline;
366 };
367
368 static void diff_words_append(char *line, unsigned long len,
369                 struct diff_words_buffer *buffer)
370 {
371         if (buffer->text.size + len > buffer->alloc) {
372                 buffer->alloc = (buffer->text.size + len) * 3 / 2;
373                 buffer->text.ptr = xrealloc(buffer->text.ptr, buffer->alloc);
374         }
375         line++;
376         len--;
377         memcpy(buffer->text.ptr + buffer->text.size, line, len);
378         buffer->text.size += len;
379 }
380
381 struct diff_words_data {
382         struct xdiff_emit_state xm;
383         struct diff_words_buffer minus, plus;
384 };
385
386 static void print_word(struct diff_words_buffer *buffer, int len, int color,
387                 int suppress_newline)
388 {
389         const char *ptr;
390         int eol = 0;
391
392         if (len == 0)
393                 return;
394
395         ptr  = buffer->text.ptr + buffer->current;
396         buffer->current += len;
397
398         if (ptr[len - 1] == '\n') {
399                 eol = 1;
400                 len--;
401         }
402
403         fputs(diff_get_color(1, color), stdout);
404         fwrite(ptr, len, 1, stdout);
405         fputs(diff_get_color(1, DIFF_RESET), stdout);
406
407         if (eol) {
408                 if (suppress_newline)
409                         buffer->suppressed_newline = 1;
410                 else
411                         putchar('\n');
412         }
413 }
414
415 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
416 {
417         struct diff_words_data *diff_words = priv;
418
419         if (diff_words->minus.suppressed_newline) {
420                 if (line[0] != '+')
421                         putchar('\n');
422                 diff_words->minus.suppressed_newline = 0;
423         }
424
425         len--;
426         switch (line[0]) {
427                 case '-':
428                         print_word(&diff_words->minus, len, DIFF_FILE_OLD, 1);
429                         break;
430                 case '+':
431                         print_word(&diff_words->plus, len, DIFF_FILE_NEW, 0);
432                         break;
433                 case ' ':
434                         print_word(&diff_words->plus, len, DIFF_PLAIN, 0);
435                         diff_words->minus.current += len;
436                         break;
437         }
438 }
439
440 /* this executes the word diff on the accumulated buffers */
441 static void diff_words_show(struct diff_words_data *diff_words)
442 {
443         xpparam_t xpp;
444         xdemitconf_t xecfg;
445         xdemitcb_t ecb;
446         mmfile_t minus, plus;
447         int i;
448
449         memset(&xecfg, 0, sizeof(xecfg));
450         minus.size = diff_words->minus.text.size;
451         minus.ptr = xmalloc(minus.size);
452         memcpy(minus.ptr, diff_words->minus.text.ptr, minus.size);
453         for (i = 0; i < minus.size; i++)
454                 if (isspace(minus.ptr[i]))
455                         minus.ptr[i] = '\n';
456         diff_words->minus.current = 0;
457
458         plus.size = diff_words->plus.text.size;
459         plus.ptr = xmalloc(plus.size);
460         memcpy(plus.ptr, diff_words->plus.text.ptr, plus.size);
461         for (i = 0; i < plus.size; i++)
462                 if (isspace(plus.ptr[i]))
463                         plus.ptr[i] = '\n';
464         diff_words->plus.current = 0;
465
466         xpp.flags = XDF_NEED_MINIMAL;
467         xecfg.ctxlen = diff_words->minus.alloc + diff_words->plus.alloc;
468         ecb.outf = xdiff_outf;
469         ecb.priv = diff_words;
470         diff_words->xm.consume = fn_out_diff_words_aux;
471         xdi_diff(&minus, &plus, &xpp, &xecfg, &ecb);
472
473         free(minus.ptr);
474         free(plus.ptr);
475         diff_words->minus.text.size = diff_words->plus.text.size = 0;
476
477         if (diff_words->minus.suppressed_newline) {
478                 putchar('\n');
479                 diff_words->minus.suppressed_newline = 0;
480         }
481 }
482
483 typedef unsigned long (*sane_truncate_fn)(char *line, unsigned long len);
484
485 struct emit_callback {
486         struct xdiff_emit_state xm;
487         int nparents, color_diff;
488         unsigned ws_rule;
489         sane_truncate_fn truncate;
490         const char **label_path;
491         struct diff_words_data *diff_words;
492         int *found_changesp;
493 };
494
495 static void free_diff_words_data(struct emit_callback *ecbdata)
496 {
497         if (ecbdata->diff_words) {
498                 /* flush buffers */
499                 if (ecbdata->diff_words->minus.text.size ||
500                                 ecbdata->diff_words->plus.text.size)
501                         diff_words_show(ecbdata->diff_words);
502
503                 if (ecbdata->diff_words->minus.text.ptr)
504                         free (ecbdata->diff_words->minus.text.ptr);
505                 if (ecbdata->diff_words->plus.text.ptr)
506                         free (ecbdata->diff_words->plus.text.ptr);
507                 free(ecbdata->diff_words);
508                 ecbdata->diff_words = NULL;
509         }
510 }
511
512 const char *diff_get_color(int diff_use_color, enum color_diff ix)
513 {
514         if (diff_use_color)
515                 return diff_colors[ix];
516         return "";
517 }
518
519 static void emit_line(const char *set, const char *reset, const char *line, int len)
520 {
521         fputs(set, stdout);
522         fwrite(line, len, 1, stdout);
523         fputs(reset, stdout);
524 }
525
526 static void emit_add_line(const char *reset, struct emit_callback *ecbdata, const char *line, int len)
527 {
528         const char *ws = diff_get_color(ecbdata->color_diff, DIFF_WHITESPACE);
529         const char *set = diff_get_color(ecbdata->color_diff, DIFF_FILE_NEW);
530
531         if (!*ws)
532                 emit_line(set, reset, line, len);
533         else {
534                 /* Emit just the prefix, then the rest. */
535                 emit_line(set, reset, line, ecbdata->nparents);
536                 (void)check_and_emit_line(line + ecbdata->nparents,
537                     len - ecbdata->nparents, ecbdata->ws_rule,
538                     stdout, set, reset, ws);
539         }
540 }
541
542 static unsigned long sane_truncate_line(struct emit_callback *ecb, char *line, unsigned long len)
543 {
544         const char *cp;
545         unsigned long allot;
546         size_t l = len;
547
548         if (ecb->truncate)
549                 return ecb->truncate(line, len);
550         cp = line;
551         allot = l;
552         while (0 < l) {
553                 (void) utf8_width(&cp, &l);
554                 if (!cp)
555                         break; /* truncated in the middle? */
556         }
557         return allot - l;
558 }
559
560 static void fn_out_consume(void *priv, char *line, unsigned long len)
561 {
562         int i;
563         int color;
564         struct emit_callback *ecbdata = priv;
565         const char *meta = diff_get_color(ecbdata->color_diff, DIFF_METAINFO);
566         const char *plain = diff_get_color(ecbdata->color_diff, DIFF_PLAIN);
567         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
568
569         *(ecbdata->found_changesp) = 1;
570
571         if (ecbdata->label_path[0]) {
572                 const char *name_a_tab, *name_b_tab;
573
574                 name_a_tab = strchr(ecbdata->label_path[0], ' ') ? "\t" : "";
575                 name_b_tab = strchr(ecbdata->label_path[1], ' ') ? "\t" : "";
576
577                 printf("%s--- %s%s%s\n",
578                        meta, ecbdata->label_path[0], reset, name_a_tab);
579                 printf("%s+++ %s%s%s\n",
580                        meta, ecbdata->label_path[1], reset, name_b_tab);
581                 ecbdata->label_path[0] = ecbdata->label_path[1] = NULL;
582         }
583
584         /* This is not really necessary for now because
585          * this codepath only deals with two-way diffs.
586          */
587         for (i = 0; i < len && line[i] == '@'; i++)
588                 ;
589         if (2 <= i && i < len && line[i] == ' ') {
590                 ecbdata->nparents = i - 1;
591                 len = sane_truncate_line(ecbdata, line, len);
592                 emit_line(diff_get_color(ecbdata->color_diff, DIFF_FRAGINFO),
593                           reset, line, len);
594                 if (line[len-1] != '\n')
595                         putchar('\n');
596                 return;
597         }
598
599         if (len < ecbdata->nparents) {
600                 emit_line(reset, reset, line, len);
601                 return;
602         }
603
604         color = DIFF_PLAIN;
605         if (ecbdata->diff_words && ecbdata->nparents != 1)
606                 /* fall back to normal diff */
607                 free_diff_words_data(ecbdata);
608         if (ecbdata->diff_words) {
609                 if (line[0] == '-') {
610                         diff_words_append(line, len,
611                                           &ecbdata->diff_words->minus);
612                         return;
613                 } else if (line[0] == '+') {
614                         diff_words_append(line, len,
615                                           &ecbdata->diff_words->plus);
616                         return;
617                 }
618                 if (ecbdata->diff_words->minus.text.size ||
619                     ecbdata->diff_words->plus.text.size)
620                         diff_words_show(ecbdata->diff_words);
621                 line++;
622                 len--;
623                 emit_line(plain, reset, line, len);
624                 return;
625         }
626         for (i = 0; i < ecbdata->nparents && len; i++) {
627                 if (line[i] == '-')
628                         color = DIFF_FILE_OLD;
629                 else if (line[i] == '+')
630                         color = DIFF_FILE_NEW;
631         }
632
633         if (color != DIFF_FILE_NEW) {
634                 emit_line(diff_get_color(ecbdata->color_diff, color),
635                           reset, line, len);
636                 return;
637         }
638         emit_add_line(reset, ecbdata, line, len);
639 }
640
641 static char *pprint_rename(const char *a, const char *b)
642 {
643         const char *old = a;
644         const char *new = b;
645         struct strbuf name;
646         int pfx_length, sfx_length;
647         int len_a = strlen(a);
648         int len_b = strlen(b);
649         int a_midlen, b_midlen;
650         int qlen_a = quote_c_style(a, NULL, NULL, 0);
651         int qlen_b = quote_c_style(b, NULL, NULL, 0);
652
653         strbuf_init(&name, 0);
654         if (qlen_a || qlen_b) {
655                 quote_c_style(a, &name, NULL, 0);
656                 strbuf_addstr(&name, " => ");
657                 quote_c_style(b, &name, NULL, 0);
658                 return strbuf_detach(&name, NULL);
659         }
660
661         /* Find common prefix */
662         pfx_length = 0;
663         while (*old && *new && *old == *new) {
664                 if (*old == '/')
665                         pfx_length = old - a + 1;
666                 old++;
667                 new++;
668         }
669
670         /* Find common suffix */
671         old = a + len_a;
672         new = b + len_b;
673         sfx_length = 0;
674         while (a <= old && b <= new && *old == *new) {
675                 if (*old == '/')
676                         sfx_length = len_a - (old - a);
677                 old--;
678                 new--;
679         }
680
681         /*
682          * pfx{mid-a => mid-b}sfx
683          * {pfx-a => pfx-b}sfx
684          * pfx{sfx-a => sfx-b}
685          * name-a => name-b
686          */
687         a_midlen = len_a - pfx_length - sfx_length;
688         b_midlen = len_b - pfx_length - sfx_length;
689         if (a_midlen < 0)
690                 a_midlen = 0;
691         if (b_midlen < 0)
692                 b_midlen = 0;
693
694         strbuf_grow(&name, pfx_length + a_midlen + b_midlen + sfx_length + 7);
695         if (pfx_length + sfx_length) {
696                 strbuf_add(&name, a, pfx_length);
697                 strbuf_addch(&name, '{');
698         }
699         strbuf_add(&name, a + pfx_length, a_midlen);
700         strbuf_addstr(&name, " => ");
701         strbuf_add(&name, b + pfx_length, b_midlen);
702         if (pfx_length + sfx_length) {
703                 strbuf_addch(&name, '}');
704                 strbuf_add(&name, a + len_a - sfx_length, sfx_length);
705         }
706         return strbuf_detach(&name, NULL);
707 }
708
709 struct diffstat_t {
710         struct xdiff_emit_state xm;
711
712         int nr;
713         int alloc;
714         struct diffstat_file {
715                 char *from_name;
716                 char *name;
717                 char *print_name;
718                 unsigned is_unmerged:1;
719                 unsigned is_binary:1;
720                 unsigned is_renamed:1;
721                 unsigned int added, deleted;
722         } **files;
723 };
724
725 static struct diffstat_file *diffstat_add(struct diffstat_t *diffstat,
726                                           const char *name_a,
727                                           const char *name_b)
728 {
729         struct diffstat_file *x;
730         x = xcalloc(sizeof (*x), 1);
731         if (diffstat->nr == diffstat->alloc) {
732                 diffstat->alloc = alloc_nr(diffstat->alloc);
733                 diffstat->files = xrealloc(diffstat->files,
734                                 diffstat->alloc * sizeof(x));
735         }
736         diffstat->files[diffstat->nr++] = x;
737         if (name_b) {
738                 x->from_name = xstrdup(name_a);
739                 x->name = xstrdup(name_b);
740                 x->is_renamed = 1;
741         }
742         else {
743                 x->from_name = NULL;
744                 x->name = xstrdup(name_a);
745         }
746         return x;
747 }
748
749 static void diffstat_consume(void *priv, char *line, unsigned long len)
750 {
751         struct diffstat_t *diffstat = priv;
752         struct diffstat_file *x = diffstat->files[diffstat->nr - 1];
753
754         if (line[0] == '+')
755                 x->added++;
756         else if (line[0] == '-')
757                 x->deleted++;
758 }
759
760 const char mime_boundary_leader[] = "------------";
761
762 static int scale_linear(int it, int width, int max_change)
763 {
764         /*
765          * make sure that at least one '-' is printed if there were deletions,
766          * and likewise for '+'.
767          */
768         if (max_change < 2)
769                 return it;
770         return ((it - 1) * (width - 1) + max_change - 1) / (max_change - 1);
771 }
772
773 static void show_name(const char *prefix, const char *name, int len,
774                       const char *reset, const char *set)
775 {
776         printf(" %s%s%-*s%s |", set, prefix, len, name, reset);
777 }
778
779 static void show_graph(char ch, int cnt, const char *set, const char *reset)
780 {
781         if (cnt <= 0)
782                 return;
783         printf("%s", set);
784         while (cnt--)
785                 putchar(ch);
786         printf("%s", reset);
787 }
788
789 static void fill_print_name(struct diffstat_file *file)
790 {
791         char *pname;
792
793         if (file->print_name)
794                 return;
795
796         if (!file->is_renamed) {
797                 struct strbuf buf;
798                 strbuf_init(&buf, 0);
799                 if (quote_c_style(file->name, &buf, NULL, 0)) {
800                         pname = strbuf_detach(&buf, NULL);
801                 } else {
802                         pname = file->name;
803                         strbuf_release(&buf);
804                 }
805         } else {
806                 pname = pprint_rename(file->from_name, file->name);
807         }
808         file->print_name = pname;
809 }
810
811 static void show_stats(struct diffstat_t* data, struct diff_options *options)
812 {
813         int i, len, add, del, total, adds = 0, dels = 0;
814         int max_change = 0, max_len = 0;
815         int total_files = data->nr;
816         int width, name_width;
817         const char *reset, *set, *add_c, *del_c;
818
819         if (data->nr == 0)
820                 return;
821
822         width = options->stat_width ? options->stat_width : 80;
823         name_width = options->stat_name_width ? options->stat_name_width : 50;
824
825         /* Sanity: give at least 5 columns to the graph,
826          * but leave at least 10 columns for the name.
827          */
828         if (width < name_width + 15) {
829                 if (name_width <= 25)
830                         width = name_width + 15;
831                 else
832                         name_width = width - 15;
833         }
834
835         /* Find the longest filename and max number of changes */
836         reset = diff_get_color_opt(options, DIFF_RESET);
837         set   = diff_get_color_opt(options, DIFF_PLAIN);
838         add_c = diff_get_color_opt(options, DIFF_FILE_NEW);
839         del_c = diff_get_color_opt(options, DIFF_FILE_OLD);
840
841         for (i = 0; i < data->nr; i++) {
842                 struct diffstat_file *file = data->files[i];
843                 int change = file->added + file->deleted;
844                 fill_print_name(file);
845                 len = strlen(file->print_name);
846                 if (max_len < len)
847                         max_len = len;
848
849                 if (file->is_binary || file->is_unmerged)
850                         continue;
851                 if (max_change < change)
852                         max_change = change;
853         }
854
855         /* Compute the width of the graph part;
856          * 10 is for one blank at the beginning of the line plus
857          * " | count " between the name and the graph.
858          *
859          * From here on, name_width is the width of the name area,
860          * and width is the width of the graph area.
861          */
862         name_width = (name_width < max_len) ? name_width : max_len;
863         if (width < (name_width + 10) + max_change)
864                 width = width - (name_width + 10);
865         else
866                 width = max_change;
867
868         for (i = 0; i < data->nr; i++) {
869                 const char *prefix = "";
870                 char *name = data->files[i]->print_name;
871                 int added = data->files[i]->added;
872                 int deleted = data->files[i]->deleted;
873                 int name_len;
874
875                 /*
876                  * "scale" the filename
877                  */
878                 len = name_width;
879                 name_len = strlen(name);
880                 if (name_width < name_len) {
881                         char *slash;
882                         prefix = "...";
883                         len -= 3;
884                         name += name_len - len;
885                         slash = strchr(name, '/');
886                         if (slash)
887                                 name = slash;
888                 }
889
890                 if (data->files[i]->is_binary) {
891                         show_name(prefix, name, len, reset, set);
892                         printf("  Bin ");
893                         printf("%s%d%s", del_c, deleted, reset);
894                         printf(" -> ");
895                         printf("%s%d%s", add_c, added, reset);
896                         printf(" bytes");
897                         printf("\n");
898                         continue;
899                 }
900                 else if (data->files[i]->is_unmerged) {
901                         show_name(prefix, name, len, reset, set);
902                         printf("  Unmerged\n");
903                         continue;
904                 }
905                 else if (!data->files[i]->is_renamed &&
906                          (added + deleted == 0)) {
907                         total_files--;
908                         continue;
909                 }
910
911                 /*
912                  * scale the add/delete
913                  */
914                 add = added;
915                 del = deleted;
916                 total = add + del;
917                 adds += add;
918                 dels += del;
919
920                 if (width <= max_change) {
921                         add = scale_linear(add, width, max_change);
922                         del = scale_linear(del, width, max_change);
923                         total = add + del;
924                 }
925                 show_name(prefix, name, len, reset, set);
926                 printf("%5d ", added + deleted);
927                 show_graph('+', add, add_c, reset);
928                 show_graph('-', del, del_c, reset);
929                 putchar('\n');
930         }
931         printf("%s %d files changed, %d insertions(+), %d deletions(-)%s\n",
932                set, total_files, adds, dels, reset);
933 }
934
935 static void show_shortstats(struct diffstat_t* data)
936 {
937         int i, adds = 0, dels = 0, total_files = data->nr;
938
939         if (data->nr == 0)
940                 return;
941
942         for (i = 0; i < data->nr; i++) {
943                 if (!data->files[i]->is_binary &&
944                     !data->files[i]->is_unmerged) {
945                         int added = data->files[i]->added;
946                         int deleted= data->files[i]->deleted;
947                         if (!data->files[i]->is_renamed &&
948                             (added + deleted == 0)) {
949                                 total_files--;
950                         } else {
951                                 adds += added;
952                                 dels += deleted;
953                         }
954                 }
955         }
956         printf(" %d files changed, %d insertions(+), %d deletions(-)\n",
957                total_files, adds, dels);
958 }
959
960 static void show_numstat(struct diffstat_t* data, struct diff_options *options)
961 {
962         int i;
963
964         if (data->nr == 0)
965                 return;
966
967         for (i = 0; i < data->nr; i++) {
968                 struct diffstat_file *file = data->files[i];
969
970                 if (file->is_binary)
971                         printf("-\t-\t");
972                 else
973                         printf("%d\t%d\t", file->added, file->deleted);
974                 if (options->line_termination) {
975                         fill_print_name(file);
976                         if (!file->is_renamed)
977                                 write_name_quoted(file->name, stdout,
978                                                   options->line_termination);
979                         else {
980                                 fputs(file->print_name, stdout);
981                                 putchar(options->line_termination);
982                         }
983                 } else {
984                         if (file->is_renamed) {
985                                 putchar('\0');
986                                 write_name_quoted(file->from_name, stdout, '\0');
987                         }
988                         write_name_quoted(file->name, stdout, '\0');
989                 }
990         }
991 }
992
993 struct diffstat_dir {
994         struct diffstat_file **files;
995         int nr, percent, cumulative;
996 };
997
998 static long gather_dirstat(struct diffstat_dir *dir, unsigned long changed, const char *base, int baselen)
999 {
1000         unsigned long this_dir = 0;
1001         unsigned int sources = 0;
1002
1003         while (dir->nr) {
1004                 struct diffstat_file *f = *dir->files;
1005                 int namelen = strlen(f->name);
1006                 unsigned long this;
1007                 char *slash;
1008
1009                 if (namelen < baselen)
1010                         break;
1011                 if (memcmp(f->name, base, baselen))
1012                         break;
1013                 slash = strchr(f->name + baselen, '/');
1014                 if (slash) {
1015                         int newbaselen = slash + 1 - f->name;
1016                         this = gather_dirstat(dir, changed, f->name, newbaselen);
1017                         sources++;
1018                 } else {
1019                         this = f->added + f->deleted;
1020                         dir->files++;
1021                         dir->nr--;
1022                         sources += 2;
1023                 }
1024                 this_dir += this;
1025         }
1026
1027         /*
1028          * We don't report dirstat's for
1029          *  - the top level
1030          *  - or cases where everything came from a single directory
1031          *    under this directory (sources == 1).
1032          */
1033         if (baselen && sources != 1) {
1034                 int permille = this_dir * 1000 / changed;
1035                 if (permille) {
1036                         int percent = permille / 10;
1037                         if (percent >= dir->percent) {
1038                                 printf("%4d.%01d%% %.*s\n", percent, permille % 10, baselen, base);
1039                                 if (!dir->cumulative)
1040                                         return 0;
1041                         }
1042                 }
1043         }
1044         return this_dir;
1045 }
1046
1047 static void show_dirstat(struct diffstat_t *data, struct diff_options *options)
1048 {
1049         int i;
1050         unsigned long changed;
1051         struct diffstat_dir dir;
1052
1053         /* Calculate total changes */
1054         changed = 0;
1055         for (i = 0; i < data->nr; i++) {
1056                 changed += data->files[i]->added;
1057                 changed += data->files[i]->deleted;
1058         }
1059
1060         /* This can happen even with many files, if everything was renames */
1061         if (!changed)
1062                 return;
1063
1064         /* Show all directories with more than x% of the changes */
1065         dir.files = data->files;
1066         dir.nr = data->nr;
1067         dir.percent = options->dirstat_percent;
1068         dir.cumulative = options->output_format & DIFF_FORMAT_CUMULATIVE;
1069         gather_dirstat(&dir, changed, "", 0);
1070 }
1071
1072 static void free_diffstat_info(struct diffstat_t *diffstat)
1073 {
1074         int i;
1075         for (i = 0; i < diffstat->nr; i++) {
1076                 struct diffstat_file *f = diffstat->files[i];
1077                 if (f->name != f->print_name)
1078                         free(f->print_name);
1079                 free(f->name);
1080                 free(f->from_name);
1081                 free(f);
1082         }
1083         free(diffstat->files);
1084 }
1085
1086 struct checkdiff_t {
1087         struct xdiff_emit_state xm;
1088         const char *filename;
1089         int lineno, color_diff;
1090         unsigned ws_rule;
1091         unsigned status;
1092 };
1093
1094 static void checkdiff_consume(void *priv, char *line, unsigned long len)
1095 {
1096         struct checkdiff_t *data = priv;
1097         const char *ws = diff_get_color(data->color_diff, DIFF_WHITESPACE);
1098         const char *reset = diff_get_color(data->color_diff, DIFF_RESET);
1099         const char *set = diff_get_color(data->color_diff, DIFF_FILE_NEW);
1100         char *err;
1101
1102         if (line[0] == '+') {
1103                 data->status = check_and_emit_line(line + 1, len - 1,
1104                     data->ws_rule, NULL, NULL, NULL, NULL);
1105                 if (!data->status)
1106                         return;
1107                 err = whitespace_error_string(data->status);
1108                 printf("%s:%d: %s.\n", data->filename, data->lineno, err);
1109                 free(err);
1110                 emit_line(set, reset, line, 1);
1111                 (void)check_and_emit_line(line + 1, len - 1, data->ws_rule,
1112                     stdout, set, reset, ws);
1113                 data->lineno++;
1114         } else if (line[0] == ' ')
1115                 data->lineno++;
1116         else if (line[0] == '@') {
1117                 char *plus = strchr(line, '+');
1118                 if (plus)
1119                         data->lineno = strtol(plus, NULL, 10);
1120                 else
1121                         die("invalid diff");
1122         }
1123 }
1124
1125 static unsigned char *deflate_it(char *data,
1126                                  unsigned long size,
1127                                  unsigned long *result_size)
1128 {
1129         int bound;
1130         unsigned char *deflated;
1131         z_stream stream;
1132
1133         memset(&stream, 0, sizeof(stream));
1134         deflateInit(&stream, zlib_compression_level);
1135         bound = deflateBound(&stream, size);
1136         deflated = xmalloc(bound);
1137         stream.next_out = deflated;
1138         stream.avail_out = bound;
1139
1140         stream.next_in = (unsigned char *)data;
1141         stream.avail_in = size;
1142         while (deflate(&stream, Z_FINISH) == Z_OK)
1143                 ; /* nothing */
1144         deflateEnd(&stream);
1145         *result_size = stream.total_out;
1146         return deflated;
1147 }
1148
1149 static void emit_binary_diff_body(mmfile_t *one, mmfile_t *two)
1150 {
1151         void *cp;
1152         void *delta;
1153         void *deflated;
1154         void *data;
1155         unsigned long orig_size;
1156         unsigned long delta_size;
1157         unsigned long deflate_size;
1158         unsigned long data_size;
1159
1160         /* We could do deflated delta, or we could do just deflated two,
1161          * whichever is smaller.
1162          */
1163         delta = NULL;
1164         deflated = deflate_it(two->ptr, two->size, &deflate_size);
1165         if (one->size && two->size) {
1166                 delta = diff_delta(one->ptr, one->size,
1167                                    two->ptr, two->size,
1168                                    &delta_size, deflate_size);
1169                 if (delta) {
1170                         void *to_free = delta;
1171                         orig_size = delta_size;
1172                         delta = deflate_it(delta, delta_size, &delta_size);
1173                         free(to_free);
1174                 }
1175         }
1176
1177         if (delta && delta_size < deflate_size) {
1178                 printf("delta %lu\n", orig_size);
1179                 free(deflated);
1180                 data = delta;
1181                 data_size = delta_size;
1182         }
1183         else {
1184                 printf("literal %lu\n", two->size);
1185                 free(delta);
1186                 data = deflated;
1187                 data_size = deflate_size;
1188         }
1189
1190         /* emit data encoded in base85 */
1191         cp = data;
1192         while (data_size) {
1193                 int bytes = (52 < data_size) ? 52 : data_size;
1194                 char line[70];
1195                 data_size -= bytes;
1196                 if (bytes <= 26)
1197                         line[0] = bytes + 'A' - 1;
1198                 else
1199                         line[0] = bytes - 26 + 'a' - 1;
1200                 encode_85(line + 1, cp, bytes);
1201                 cp = (char *) cp + bytes;
1202                 puts(line);
1203         }
1204         printf("\n");
1205         free(data);
1206 }
1207
1208 static void emit_binary_diff(mmfile_t *one, mmfile_t *two)
1209 {
1210         printf("GIT binary patch\n");
1211         emit_binary_diff_body(one, two);
1212         emit_binary_diff_body(two, one);
1213 }
1214
1215 static void setup_diff_attr_check(struct git_attr_check *check)
1216 {
1217         static struct git_attr *attr_diff;
1218
1219         if (!attr_diff) {
1220                 attr_diff = git_attr("diff", 4);
1221         }
1222         check[0].attr = attr_diff;
1223 }
1224
1225 static void diff_filespec_check_attr(struct diff_filespec *one)
1226 {
1227         struct git_attr_check attr_diff_check;
1228         int check_from_data = 0;
1229
1230         if (one->checked_attr)
1231                 return;
1232
1233         setup_diff_attr_check(&attr_diff_check);
1234         one->is_binary = 0;
1235         one->funcname_pattern_ident = NULL;
1236
1237         if (!git_checkattr(one->path, 1, &attr_diff_check)) {
1238                 const char *value;
1239
1240                 /* binaryness */
1241                 value = attr_diff_check.value;
1242                 if (ATTR_TRUE(value))
1243                         ;
1244                 else if (ATTR_FALSE(value))
1245                         one->is_binary = 1;
1246                 else
1247                         check_from_data = 1;
1248
1249                 /* funcname pattern ident */
1250                 if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
1251                         ;
1252                 else
1253                         one->funcname_pattern_ident = value;
1254         }
1255
1256         if (check_from_data) {
1257                 if (!one->data && DIFF_FILE_VALID(one))
1258                         diff_populate_filespec(one, 0);
1259
1260                 if (one->data)
1261                         one->is_binary = buffer_is_binary(one->data, one->size);
1262         }
1263 }
1264
1265 int diff_filespec_is_binary(struct diff_filespec *one)
1266 {
1267         diff_filespec_check_attr(one);
1268         return one->is_binary;
1269 }
1270
1271 static const char *funcname_pattern(const char *ident)
1272 {
1273         struct funcname_pattern *pp;
1274
1275         for (pp = funcname_pattern_list; pp; pp = pp->next)
1276                 if (!strcmp(ident, pp->name))
1277                         return pp->pattern;
1278         return NULL;
1279 }
1280
1281 static struct builtin_funcname_pattern {
1282         const char *name;
1283         const char *pattern;
1284 } builtin_funcname_pattern[] = {
1285         { "java", "!^[  ]*\\(catch\\|do\\|for\\|if\\|instanceof\\|"
1286                         "new\\|return\\|switch\\|throw\\|while\\)\n"
1287                         "^[     ]*\\(\\([       ]*"
1288                         "[A-Za-z_][A-Za-z_0-9]*\\)\\{2,\\}"
1289                         "[      ]*([^;]*$\\)" },
1290         { "tex", "^\\(\\\\\\(sub\\)*section{.*\\)$" },
1291 };
1292
1293 static const char *diff_funcname_pattern(struct diff_filespec *one)
1294 {
1295         const char *ident, *pattern;
1296         int i;
1297
1298         diff_filespec_check_attr(one);
1299         ident = one->funcname_pattern_ident;
1300
1301         if (!ident)
1302                 /*
1303                  * If the config file has "funcname.default" defined, that
1304                  * regexp is used; otherwise NULL is returned and xemit uses
1305                  * the built-in default.
1306                  */
1307                 return funcname_pattern("default");
1308
1309         /* Look up custom "funcname.$ident" regexp from config. */
1310         pattern = funcname_pattern(ident);
1311         if (pattern)
1312                 return pattern;
1313
1314         /*
1315          * And define built-in fallback patterns here.  Note that
1316          * these can be overridden by the user's config settings.
1317          */
1318         for (i = 0; i < ARRAY_SIZE(builtin_funcname_pattern); i++)
1319                 if (!strcmp(ident, builtin_funcname_pattern[i].name))
1320                         return builtin_funcname_pattern[i].pattern;
1321
1322         return NULL;
1323 }
1324
1325 static void builtin_diff(const char *name_a,
1326                          const char *name_b,
1327                          struct diff_filespec *one,
1328                          struct diff_filespec *two,
1329                          const char *xfrm_msg,
1330                          struct diff_options *o,
1331                          int complete_rewrite)
1332 {
1333         mmfile_t mf1, mf2;
1334         const char *lbl[2];
1335         char *a_one, *b_two;
1336         const char *set = diff_get_color_opt(o, DIFF_METAINFO);
1337         const char *reset = diff_get_color_opt(o, DIFF_RESET);
1338
1339         a_one = quote_two(o->a_prefix, name_a + (*name_a == '/'));
1340         b_two = quote_two(o->b_prefix, name_b + (*name_b == '/'));
1341         lbl[0] = DIFF_FILE_VALID(one) ? a_one : "/dev/null";
1342         lbl[1] = DIFF_FILE_VALID(two) ? b_two : "/dev/null";
1343         printf("%sdiff --git %s %s%s\n", set, a_one, b_two, reset);
1344         if (lbl[0][0] == '/') {
1345                 /* /dev/null */
1346                 printf("%snew file mode %06o%s\n", set, two->mode, reset);
1347                 if (xfrm_msg && xfrm_msg[0])
1348                         printf("%s%s%s\n", set, xfrm_msg, reset);
1349         }
1350         else if (lbl[1][0] == '/') {
1351                 printf("%sdeleted file mode %06o%s\n", set, one->mode, reset);
1352                 if (xfrm_msg && xfrm_msg[0])
1353                         printf("%s%s%s\n", set, xfrm_msg, reset);
1354         }
1355         else {
1356                 if (one->mode != two->mode) {
1357                         printf("%sold mode %06o%s\n", set, one->mode, reset);
1358                         printf("%snew mode %06o%s\n", set, two->mode, reset);
1359                 }
1360                 if (xfrm_msg && xfrm_msg[0])
1361                         printf("%s%s%s\n", set, xfrm_msg, reset);
1362                 /*
1363                  * we do not run diff between different kind
1364                  * of objects.
1365                  */
1366                 if ((one->mode ^ two->mode) & S_IFMT)
1367                         goto free_ab_and_return;
1368                 if (complete_rewrite) {
1369                         emit_rewrite_diff(name_a, name_b, one, two, o);
1370                         o->found_changes = 1;
1371                         goto free_ab_and_return;
1372                 }
1373         }
1374
1375         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
1376                 die("unable to read files to diff");
1377
1378         if (!DIFF_OPT_TST(o, TEXT) &&
1379             (diff_filespec_is_binary(one) || diff_filespec_is_binary(two))) {
1380                 /* Quite common confusing case */
1381                 if (mf1.size == mf2.size &&
1382                     !memcmp(mf1.ptr, mf2.ptr, mf1.size))
1383                         goto free_ab_and_return;
1384                 if (DIFF_OPT_TST(o, BINARY))
1385                         emit_binary_diff(&mf1, &mf2);
1386                 else
1387                         printf("Binary files %s and %s differ\n",
1388                                lbl[0], lbl[1]);
1389                 o->found_changes = 1;
1390         }
1391         else {
1392                 /* Crazy xdl interfaces.. */
1393                 const char *diffopts = getenv("GIT_DIFF_OPTS");
1394                 xpparam_t xpp;
1395                 xdemitconf_t xecfg;
1396                 xdemitcb_t ecb;
1397                 struct emit_callback ecbdata;
1398                 const char *funcname_pattern;
1399
1400                 funcname_pattern = diff_funcname_pattern(one);
1401                 if (!funcname_pattern)
1402                         funcname_pattern = diff_funcname_pattern(two);
1403
1404                 memset(&xecfg, 0, sizeof(xecfg));
1405                 memset(&ecbdata, 0, sizeof(ecbdata));
1406                 ecbdata.label_path = lbl;
1407                 ecbdata.color_diff = DIFF_OPT_TST(o, COLOR_DIFF);
1408                 ecbdata.found_changesp = &o->found_changes;
1409                 ecbdata.ws_rule = whitespace_rule(name_b ? name_b : name_a);
1410                 xpp.flags = XDF_NEED_MINIMAL | o->xdl_opts;
1411                 xecfg.ctxlen = o->context;
1412                 xecfg.flags = XDL_EMIT_FUNCNAMES;
1413                 if (funcname_pattern)
1414                         xdiff_set_find_func(&xecfg, funcname_pattern);
1415                 if (!diffopts)
1416                         ;
1417                 else if (!prefixcmp(diffopts, "--unified="))
1418                         xecfg.ctxlen = strtoul(diffopts + 10, NULL, 10);
1419                 else if (!prefixcmp(diffopts, "-u"))
1420                         xecfg.ctxlen = strtoul(diffopts + 2, NULL, 10);
1421                 ecb.outf = xdiff_outf;
1422                 ecb.priv = &ecbdata;
1423                 ecbdata.xm.consume = fn_out_consume;
1424                 if (DIFF_OPT_TST(o, COLOR_DIFF_WORDS))
1425                         ecbdata.diff_words =
1426                                 xcalloc(1, sizeof(struct diff_words_data));
1427                 xdi_diff(&mf1, &mf2, &xpp, &xecfg, &ecb);
1428                 if (DIFF_OPT_TST(o, COLOR_DIFF_WORDS))
1429                         free_diff_words_data(&ecbdata);
1430         }
1431
1432  free_ab_and_return:
1433         diff_free_filespec_data(one);
1434         diff_free_filespec_data(two);
1435         free(a_one);
1436         free(b_two);
1437         return;
1438 }
1439
1440 static void builtin_diffstat(const char *name_a, const char *name_b,
1441                              struct diff_filespec *one,
1442                              struct diff_filespec *two,
1443                              struct diffstat_t *diffstat,
1444                              struct diff_options *o,
1445                              int complete_rewrite)
1446 {
1447         mmfile_t mf1, mf2;
1448         struct diffstat_file *data;
1449
1450         data = diffstat_add(diffstat, name_a, name_b);
1451
1452         if (!one || !two) {
1453                 data->is_unmerged = 1;
1454                 return;
1455         }
1456         if (complete_rewrite) {
1457                 diff_populate_filespec(one, 0);
1458                 diff_populate_filespec(two, 0);
1459                 data->deleted = count_lines(one->data, one->size);
1460                 data->added = count_lines(two->data, two->size);
1461                 goto free_and_return;
1462         }
1463         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
1464                 die("unable to read files to diff");
1465
1466         if (diff_filespec_is_binary(one) || diff_filespec_is_binary(two)) {
1467                 data->is_binary = 1;
1468                 data->added = mf2.size;
1469                 data->deleted = mf1.size;
1470         } else {
1471                 /* Crazy xdl interfaces.. */
1472                 xpparam_t xpp;
1473                 xdemitconf_t xecfg;
1474                 xdemitcb_t ecb;
1475
1476                 memset(&xecfg, 0, sizeof(xecfg));
1477                 xpp.flags = XDF_NEED_MINIMAL | o->xdl_opts;
1478                 ecb.outf = xdiff_outf;
1479                 ecb.priv = diffstat;
1480                 xdi_diff(&mf1, &mf2, &xpp, &xecfg, &ecb);
1481         }
1482
1483  free_and_return:
1484         diff_free_filespec_data(one);
1485         diff_free_filespec_data(two);
1486 }
1487
1488 static void builtin_checkdiff(const char *name_a, const char *name_b,
1489                              struct diff_filespec *one,
1490                              struct diff_filespec *two, struct diff_options *o)
1491 {
1492         mmfile_t mf1, mf2;
1493         struct checkdiff_t data;
1494
1495         if (!two)
1496                 return;
1497
1498         memset(&data, 0, sizeof(data));
1499         data.xm.consume = checkdiff_consume;
1500         data.filename = name_b ? name_b : name_a;
1501         data.lineno = 0;
1502         data.color_diff = DIFF_OPT_TST(o, COLOR_DIFF);
1503         data.ws_rule = whitespace_rule(data.filename);
1504
1505         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
1506                 die("unable to read files to diff");
1507
1508         if (diff_filespec_is_binary(two))
1509                 goto free_and_return;
1510         else {
1511                 /* Crazy xdl interfaces.. */
1512                 xpparam_t xpp;
1513                 xdemitconf_t xecfg;
1514                 xdemitcb_t ecb;
1515
1516                 memset(&xecfg, 0, sizeof(xecfg));
1517                 xpp.flags = XDF_NEED_MINIMAL;
1518                 ecb.outf = xdiff_outf;
1519                 ecb.priv = &data;
1520                 xdi_diff(&mf1, &mf2, &xpp, &xecfg, &ecb);
1521         }
1522  free_and_return:
1523         diff_free_filespec_data(one);
1524         diff_free_filespec_data(two);
1525         if (data.status)
1526                 DIFF_OPT_SET(o, CHECK_FAILED);
1527 }
1528
1529 struct diff_filespec *alloc_filespec(const char *path)
1530 {
1531         int namelen = strlen(path);
1532         struct diff_filespec *spec = xmalloc(sizeof(*spec) + namelen + 1);
1533
1534         memset(spec, 0, sizeof(*spec));
1535         spec->path = (char *)(spec + 1);
1536         memcpy(spec->path, path, namelen+1);
1537         spec->count = 1;
1538         return spec;
1539 }
1540
1541 void free_filespec(struct diff_filespec *spec)
1542 {
1543         if (!--spec->count) {
1544                 diff_free_filespec_data(spec);
1545                 free(spec);
1546         }
1547 }
1548
1549 void fill_filespec(struct diff_filespec *spec, const unsigned char *sha1,
1550                    unsigned short mode)
1551 {
1552         if (mode) {
1553                 spec->mode = canon_mode(mode);
1554                 hashcpy(spec->sha1, sha1);
1555                 spec->sha1_valid = !is_null_sha1(sha1);
1556         }
1557 }
1558
1559 /*
1560  * Given a name and sha1 pair, if the index tells us the file in
1561  * the work tree has that object contents, return true, so that
1562  * prepare_temp_file() does not have to inflate and extract.
1563  */
1564 static int reuse_worktree_file(const char *name, const unsigned char *sha1, int want_file)
1565 {
1566         struct cache_entry *ce;
1567         struct stat st;
1568         int pos, len;
1569
1570         /* We do not read the cache ourselves here, because the
1571          * benchmark with my previous version that always reads cache
1572          * shows that it makes things worse for diff-tree comparing
1573          * two linux-2.6 kernel trees in an already checked out work
1574          * tree.  This is because most diff-tree comparisons deal with
1575          * only a small number of files, while reading the cache is
1576          * expensive for a large project, and its cost outweighs the
1577          * savings we get by not inflating the object to a temporary
1578          * file.  Practically, this code only helps when we are used
1579          * by diff-cache --cached, which does read the cache before
1580          * calling us.
1581          */
1582         if (!active_cache)
1583                 return 0;
1584
1585         /* We want to avoid the working directory if our caller
1586          * doesn't need the data in a normal file, this system
1587          * is rather slow with its stat/open/mmap/close syscalls,
1588          * and the object is contained in a pack file.  The pack
1589          * is probably already open and will be faster to obtain
1590          * the data through than the working directory.  Loose
1591          * objects however would tend to be slower as they need
1592          * to be individually opened and inflated.
1593          */
1594         if (!FAST_WORKING_DIRECTORY && !want_file && has_sha1_pack(sha1, NULL))
1595                 return 0;
1596
1597         len = strlen(name);
1598         pos = cache_name_pos(name, len);
1599         if (pos < 0)
1600                 return 0;
1601         ce = active_cache[pos];
1602
1603         /*
1604          * This is not the sha1 we are looking for, or
1605          * unreusable because it is not a regular file.
1606          */
1607         if (hashcmp(sha1, ce->sha1) || !S_ISREG(ce->ce_mode))
1608                 return 0;
1609
1610         /*
1611          * If ce matches the file in the work tree, we can reuse it.
1612          */
1613         if (ce_uptodate(ce) ||
1614             (!lstat(name, &st) && !ce_match_stat(ce, &st, 0)))
1615                 return 1;
1616
1617         return 0;
1618 }
1619
1620 static int populate_from_stdin(struct diff_filespec *s)
1621 {
1622         struct strbuf buf;
1623         size_t size = 0;
1624
1625         strbuf_init(&buf, 0);
1626         if (strbuf_read(&buf, 0, 0) < 0)
1627                 return error("error while reading from stdin %s",
1628                                      strerror(errno));
1629
1630         s->should_munmap = 0;
1631         s->data = strbuf_detach(&buf, &size);
1632         s->size = size;
1633         s->should_free = 1;
1634         return 0;
1635 }
1636
1637 static int diff_populate_gitlink(struct diff_filespec *s, int size_only)
1638 {
1639         int len;
1640         char *data = xmalloc(100);
1641         len = snprintf(data, 100,
1642                 "Subproject commit %s\n", sha1_to_hex(s->sha1));
1643         s->data = data;
1644         s->size = len;
1645         s->should_free = 1;
1646         if (size_only) {
1647                 s->data = NULL;
1648                 free(data);
1649         }
1650         return 0;
1651 }
1652
1653 /*
1654  * While doing rename detection and pickaxe operation, we may need to
1655  * grab the data for the blob (or file) for our own in-core comparison.
1656  * diff_filespec has data and size fields for this purpose.
1657  */
1658 int diff_populate_filespec(struct diff_filespec *s, int size_only)
1659 {
1660         int err = 0;
1661         if (!DIFF_FILE_VALID(s))
1662                 die("internal error: asking to populate invalid file.");
1663         if (S_ISDIR(s->mode))
1664                 return -1;
1665
1666         if (s->data)
1667                 return 0;
1668
1669         if (size_only && 0 < s->size)
1670                 return 0;
1671
1672         if (S_ISGITLINK(s->mode))
1673                 return diff_populate_gitlink(s, size_only);
1674
1675         if (!s->sha1_valid ||
1676             reuse_worktree_file(s->path, s->sha1, 0)) {
1677                 struct strbuf buf;
1678                 struct stat st;
1679                 int fd;
1680
1681                 if (!strcmp(s->path, "-"))
1682                         return populate_from_stdin(s);
1683
1684                 if (lstat(s->path, &st) < 0) {
1685                         if (errno == ENOENT) {
1686                         err_empty:
1687                                 err = -1;
1688                         empty:
1689                                 s->data = (char *)"";
1690                                 s->size = 0;
1691                                 return err;
1692                         }
1693                 }
1694                 s->size = xsize_t(st.st_size);
1695                 if (!s->size)
1696                         goto empty;
1697                 if (size_only)
1698                         return 0;
1699                 if (S_ISLNK(st.st_mode)) {
1700                         int ret;
1701                         s->data = xmalloc(s->size);
1702                         s->should_free = 1;
1703                         ret = readlink(s->path, s->data, s->size);
1704                         if (ret < 0) {
1705                                 free(s->data);
1706                                 goto err_empty;
1707                         }
1708                         return 0;
1709                 }
1710                 fd = open(s->path, O_RDONLY);
1711                 if (fd < 0)
1712                         goto err_empty;
1713                 s->data = xmmap(NULL, s->size, PROT_READ, MAP_PRIVATE, fd, 0);
1714                 close(fd);
1715                 s->should_munmap = 1;
1716
1717                 /*
1718                  * Convert from working tree format to canonical git format
1719                  */
1720                 strbuf_init(&buf, 0);
1721                 if (convert_to_git(s->path, s->data, s->size, &buf)) {
1722                         size_t size = 0;
1723                         munmap(s->data, s->size);
1724                         s->should_munmap = 0;
1725                         s->data = strbuf_detach(&buf, &size);
1726                         s->size = size;
1727                         s->should_free = 1;
1728                 }
1729         }
1730         else {
1731                 enum object_type type;
1732                 if (size_only)
1733                         type = sha1_object_info(s->sha1, &s->size);
1734                 else {
1735                         s->data = read_sha1_file(s->sha1, &type, &s->size);
1736                         s->should_free = 1;
1737                 }
1738         }
1739         return 0;
1740 }
1741
1742 void diff_free_filespec_blob(struct diff_filespec *s)
1743 {
1744         if (s->should_free)
1745                 free(s->data);
1746         else if (s->should_munmap)
1747                 munmap(s->data, s->size);
1748
1749         if (s->should_free || s->should_munmap) {
1750                 s->should_free = s->should_munmap = 0;
1751                 s->data = NULL;
1752         }
1753 }
1754
1755 void diff_free_filespec_data(struct diff_filespec *s)
1756 {
1757         diff_free_filespec_blob(s);
1758         free(s->cnt_data);
1759         s->cnt_data = NULL;
1760 }
1761
1762 static void prep_temp_blob(struct diff_tempfile *temp,
1763                            void *blob,
1764                            unsigned long size,
1765                            const unsigned char *sha1,
1766                            int mode)
1767 {
1768         int fd;
1769
1770         fd = git_mkstemp(temp->tmp_path, PATH_MAX, ".diff_XXXXXX");
1771         if (fd < 0)
1772                 die("unable to create temp-file: %s", strerror(errno));
1773         if (write_in_full(fd, blob, size) != size)
1774                 die("unable to write temp-file");
1775         close(fd);
1776         temp->name = temp->tmp_path;
1777         strcpy(temp->hex, sha1_to_hex(sha1));
1778         temp->hex[40] = 0;
1779         sprintf(temp->mode, "%06o", mode);
1780 }
1781
1782 static void prepare_temp_file(const char *name,
1783                               struct diff_tempfile *temp,
1784                               struct diff_filespec *one)
1785 {
1786         if (!DIFF_FILE_VALID(one)) {
1787         not_a_valid_file:
1788                 /* A '-' entry produces this for file-2, and
1789                  * a '+' entry produces this for file-1.
1790                  */
1791                 temp->name = "/dev/null";
1792                 strcpy(temp->hex, ".");
1793                 strcpy(temp->mode, ".");
1794                 return;
1795         }
1796
1797         if (!one->sha1_valid ||
1798             reuse_worktree_file(name, one->sha1, 1)) {
1799                 struct stat st;
1800                 if (lstat(name, &st) < 0) {
1801                         if (errno == ENOENT)
1802                                 goto not_a_valid_file;
1803                         die("stat(%s): %s", name, strerror(errno));
1804                 }
1805                 if (S_ISLNK(st.st_mode)) {
1806                         int ret;
1807                         char buf[PATH_MAX + 1]; /* ought to be SYMLINK_MAX */
1808                         size_t sz = xsize_t(st.st_size);
1809                         if (sizeof(buf) <= st.st_size)
1810                                 die("symlink too long: %s", name);
1811                         ret = readlink(name, buf, sz);
1812                         if (ret < 0)
1813                                 die("readlink(%s)", name);
1814                         prep_temp_blob(temp, buf, sz,
1815                                        (one->sha1_valid ?
1816                                         one->sha1 : null_sha1),
1817                                        (one->sha1_valid ?
1818                                         one->mode : S_IFLNK));
1819                 }
1820                 else {
1821                         /* we can borrow from the file in the work tree */
1822                         temp->name = name;
1823                         if (!one->sha1_valid)
1824                                 strcpy(temp->hex, sha1_to_hex(null_sha1));
1825                         else
1826                                 strcpy(temp->hex, sha1_to_hex(one->sha1));
1827                         /* Even though we may sometimes borrow the
1828                          * contents from the work tree, we always want
1829                          * one->mode.  mode is trustworthy even when
1830                          * !(one->sha1_valid), as long as
1831                          * DIFF_FILE_VALID(one).
1832                          */
1833                         sprintf(temp->mode, "%06o", one->mode);
1834                 }
1835                 return;
1836         }
1837         else {
1838                 if (diff_populate_filespec(one, 0))
1839                         die("cannot read data blob for %s", one->path);
1840                 prep_temp_blob(temp, one->data, one->size,
1841                                one->sha1, one->mode);
1842         }
1843 }
1844
1845 static void remove_tempfile(void)
1846 {
1847         int i;
1848
1849         for (i = 0; i < 2; i++)
1850                 if (diff_temp[i].name == diff_temp[i].tmp_path) {
1851                         unlink(diff_temp[i].name);
1852                         diff_temp[i].name = NULL;
1853                 }
1854 }
1855
1856 static void remove_tempfile_on_signal(int signo)
1857 {
1858         remove_tempfile();
1859         signal(SIGINT, SIG_DFL);
1860         raise(signo);
1861 }
1862
1863 /* An external diff command takes:
1864  *
1865  * diff-cmd name infile1 infile1-sha1 infile1-mode \
1866  *               infile2 infile2-sha1 infile2-mode [ rename-to ]
1867  *
1868  */
1869 static void run_external_diff(const char *pgm,
1870                               const char *name,
1871                               const char *other,
1872                               struct diff_filespec *one,
1873                               struct diff_filespec *two,
1874                               const char *xfrm_msg,
1875                               int complete_rewrite)
1876 {
1877         const char *spawn_arg[10];
1878         struct diff_tempfile *temp = diff_temp;
1879         int retval;
1880         static int atexit_asked = 0;
1881         const char *othername;
1882         const char **arg = &spawn_arg[0];
1883
1884         othername = (other? other : name);
1885         if (one && two) {
1886                 prepare_temp_file(name, &temp[0], one);
1887                 prepare_temp_file(othername, &temp[1], two);
1888                 if (! atexit_asked &&
1889                     (temp[0].name == temp[0].tmp_path ||
1890                      temp[1].name == temp[1].tmp_path)) {
1891                         atexit_asked = 1;
1892                         atexit(remove_tempfile);
1893                 }
1894                 signal(SIGINT, remove_tempfile_on_signal);
1895         }
1896
1897         if (one && two) {
1898                 *arg++ = pgm;
1899                 *arg++ = name;
1900                 *arg++ = temp[0].name;
1901                 *arg++ = temp[0].hex;
1902                 *arg++ = temp[0].mode;
1903                 *arg++ = temp[1].name;
1904                 *arg++ = temp[1].hex;
1905                 *arg++ = temp[1].mode;
1906                 if (other) {
1907                         *arg++ = other;
1908                         *arg++ = xfrm_msg;
1909                 }
1910         } else {
1911                 *arg++ = pgm;
1912                 *arg++ = name;
1913         }
1914         *arg = NULL;
1915         fflush(NULL);
1916         retval = run_command_v_opt(spawn_arg, 0);
1917         remove_tempfile();
1918         if (retval) {
1919                 fprintf(stderr, "external diff died, stopping at %s.\n", name);
1920                 exit(1);
1921         }
1922 }
1923
1924 static const char *external_diff_attr(const char *name)
1925 {
1926         struct git_attr_check attr_diff_check;
1927
1928         setup_diff_attr_check(&attr_diff_check);
1929         if (!git_checkattr(name, 1, &attr_diff_check)) {
1930                 const char *value = attr_diff_check.value;
1931                 if (!ATTR_TRUE(value) &&
1932                     !ATTR_FALSE(value) &&
1933                     !ATTR_UNSET(value)) {
1934                         struct ll_diff_driver *drv;
1935
1936                         for (drv = user_diff; drv; drv = drv->next)
1937                                 if (!strcmp(drv->name, value))
1938                                         return drv->cmd;
1939                 }
1940         }
1941         return NULL;
1942 }
1943
1944 static void run_diff_cmd(const char *pgm,
1945                          const char *name,
1946                          const char *other,
1947                          struct diff_filespec *one,
1948                          struct diff_filespec *two,
1949                          const char *xfrm_msg,
1950                          struct diff_options *o,
1951                          int complete_rewrite)
1952 {
1953         if (!DIFF_OPT_TST(o, ALLOW_EXTERNAL))
1954                 pgm = NULL;
1955         else {
1956                 const char *cmd = external_diff_attr(name);
1957                 if (cmd)
1958                         pgm = cmd;
1959         }
1960
1961         if (pgm) {
1962                 run_external_diff(pgm, name, other, one, two, xfrm_msg,
1963                                   complete_rewrite);
1964                 return;
1965         }
1966         if (one && two)
1967                 builtin_diff(name, other ? other : name,
1968                              one, two, xfrm_msg, o, complete_rewrite);
1969         else
1970                 printf("* Unmerged path %s\n", name);
1971 }
1972
1973 static void diff_fill_sha1_info(struct diff_filespec *one)
1974 {
1975         if (DIFF_FILE_VALID(one)) {
1976                 if (!one->sha1_valid) {
1977                         struct stat st;
1978                         if (!strcmp(one->path, "-")) {
1979                                 hashcpy(one->sha1, null_sha1);
1980                                 return;
1981                         }
1982                         if (lstat(one->path, &st) < 0)
1983                                 die("stat %s", one->path);
1984                         if (index_path(one->sha1, one->path, &st, 0))
1985                                 die("cannot hash %s\n", one->path);
1986                 }
1987         }
1988         else
1989                 hashclr(one->sha1);
1990 }
1991
1992 static int similarity_index(struct diff_filepair *p)
1993 {
1994         return p->score * 100 / MAX_SCORE;
1995 }
1996
1997 static void run_diff(struct diff_filepair *p, struct diff_options *o)
1998 {
1999         const char *pgm = external_diff();
2000         struct strbuf msg;
2001         char *xfrm_msg;
2002         struct diff_filespec *one = p->one;
2003         struct diff_filespec *two = p->two;
2004         const char *name;
2005         const char *other;
2006         int complete_rewrite = 0;
2007
2008
2009         if (DIFF_PAIR_UNMERGED(p)) {
2010                 run_diff_cmd(pgm, p->one->path, NULL, NULL, NULL, NULL, o, 0);
2011                 return;
2012         }
2013
2014         name  = p->one->path;
2015         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
2016         diff_fill_sha1_info(one);
2017         diff_fill_sha1_info(two);
2018
2019         strbuf_init(&msg, PATH_MAX * 2 + 300);
2020         switch (p->status) {
2021         case DIFF_STATUS_COPIED:
2022                 strbuf_addf(&msg, "similarity index %d%%", similarity_index(p));
2023                 strbuf_addstr(&msg, "\ncopy from ");
2024                 quote_c_style(name, &msg, NULL, 0);
2025                 strbuf_addstr(&msg, "\ncopy to ");
2026                 quote_c_style(other, &msg, NULL, 0);
2027                 strbuf_addch(&msg, '\n');
2028                 break;
2029         case DIFF_STATUS_RENAMED:
2030                 strbuf_addf(&msg, "similarity index %d%%", similarity_index(p));
2031                 strbuf_addstr(&msg, "\nrename from ");
2032                 quote_c_style(name, &msg, NULL, 0);
2033                 strbuf_addstr(&msg, "\nrename to ");
2034                 quote_c_style(other, &msg, NULL, 0);
2035                 strbuf_addch(&msg, '\n');
2036                 break;
2037         case DIFF_STATUS_MODIFIED:
2038                 if (p->score) {
2039                         strbuf_addf(&msg, "dissimilarity index %d%%\n",
2040                                         similarity_index(p));
2041                         complete_rewrite = 1;
2042                         break;
2043                 }
2044                 /* fallthru */
2045         default:
2046                 /* nothing */
2047                 ;
2048         }
2049
2050         if (hashcmp(one->sha1, two->sha1)) {
2051                 int abbrev = DIFF_OPT_TST(o, FULL_INDEX) ? 40 : DEFAULT_ABBREV;
2052
2053                 if (DIFF_OPT_TST(o, BINARY)) {
2054                         mmfile_t mf;
2055                         if ((!fill_mmfile(&mf, one) && diff_filespec_is_binary(one)) ||
2056                             (!fill_mmfile(&mf, two) && diff_filespec_is_binary(two)))
2057                                 abbrev = 40;
2058                 }
2059                 strbuf_addf(&msg, "index %.*s..%.*s",
2060                                 abbrev, sha1_to_hex(one->sha1),
2061                                 abbrev, sha1_to_hex(two->sha1));
2062                 if (one->mode == two->mode)
2063                         strbuf_addf(&msg, " %06o", one->mode);
2064                 strbuf_addch(&msg, '\n');
2065         }
2066
2067         if (msg.len)
2068                 strbuf_setlen(&msg, msg.len - 1);
2069         xfrm_msg = msg.len ? msg.buf : NULL;
2070
2071         if (!pgm &&
2072             DIFF_FILE_VALID(one) && DIFF_FILE_VALID(two) &&
2073             (S_IFMT & one->mode) != (S_IFMT & two->mode)) {
2074                 /* a filepair that changes between file and symlink
2075                  * needs to be split into deletion and creation.
2076                  */
2077                 struct diff_filespec *null = alloc_filespec(two->path);
2078                 run_diff_cmd(NULL, name, other, one, null, xfrm_msg, o, 0);
2079                 free(null);
2080                 null = alloc_filespec(one->path);
2081                 run_diff_cmd(NULL, name, other, null, two, xfrm_msg, o, 0);
2082                 free(null);
2083         }
2084         else
2085                 run_diff_cmd(pgm, name, other, one, two, xfrm_msg, o,
2086                              complete_rewrite);
2087
2088         strbuf_release(&msg);
2089 }
2090
2091 static void run_diffstat(struct diff_filepair *p, struct diff_options *o,
2092                          struct diffstat_t *diffstat)
2093 {
2094         const char *name;
2095         const char *other;
2096         int complete_rewrite = 0;
2097
2098         if (DIFF_PAIR_UNMERGED(p)) {
2099                 /* unmerged */
2100                 builtin_diffstat(p->one->path, NULL, NULL, NULL, diffstat, o, 0);
2101                 return;
2102         }
2103
2104         name = p->one->path;
2105         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
2106
2107         diff_fill_sha1_info(p->one);
2108         diff_fill_sha1_info(p->two);
2109
2110         if (p->status == DIFF_STATUS_MODIFIED && p->score)
2111                 complete_rewrite = 1;
2112         builtin_diffstat(name, other, p->one, p->two, diffstat, o, complete_rewrite);
2113 }
2114
2115 static void run_checkdiff(struct diff_filepair *p, struct diff_options *o)
2116 {
2117         const char *name;
2118         const char *other;
2119
2120         if (DIFF_PAIR_UNMERGED(p)) {
2121                 /* unmerged */
2122                 return;
2123         }
2124
2125         name = p->one->path;
2126         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
2127
2128         diff_fill_sha1_info(p->one);
2129         diff_fill_sha1_info(p->two);
2130
2131         builtin_checkdiff(name, other, p->one, p->two, o);
2132 }
2133
2134 void diff_setup(struct diff_options *options)
2135 {
2136         memset(options, 0, sizeof(*options));
2137         options->line_termination = '\n';
2138         options->break_opt = -1;
2139         options->rename_limit = -1;
2140         options->dirstat_percent = 3;
2141         options->context = 3;
2142         options->msg_sep = "";
2143
2144         options->change = diff_change;
2145         options->add_remove = diff_addremove;
2146         if (diff_use_color_default)
2147                 DIFF_OPT_SET(options, COLOR_DIFF);
2148         else
2149                 DIFF_OPT_CLR(options, COLOR_DIFF);
2150         options->detect_rename = diff_detect_rename_default;
2151
2152         options->a_prefix = "a/";
2153         options->b_prefix = "b/";
2154 }
2155
2156 int diff_setup_done(struct diff_options *options)
2157 {
2158         int count = 0;
2159
2160         if (options->output_format & DIFF_FORMAT_NAME)
2161                 count++;
2162         if (options->output_format & DIFF_FORMAT_NAME_STATUS)
2163                 count++;
2164         if (options->output_format & DIFF_FORMAT_CHECKDIFF)
2165                 count++;
2166         if (options->output_format & DIFF_FORMAT_NO_OUTPUT)
2167                 count++;
2168         if (count > 1)
2169                 die("--name-only, --name-status, --check and -s are mutually exclusive");
2170
2171         if (DIFF_OPT_TST(options, FIND_COPIES_HARDER))
2172                 options->detect_rename = DIFF_DETECT_COPY;
2173
2174         if (options->output_format & (DIFF_FORMAT_NAME |
2175                                       DIFF_FORMAT_NAME_STATUS |
2176                                       DIFF_FORMAT_CHECKDIFF |
2177                                       DIFF_FORMAT_NO_OUTPUT))
2178                 options->output_format &= ~(DIFF_FORMAT_RAW |
2179                                             DIFF_FORMAT_NUMSTAT |
2180                                             DIFF_FORMAT_DIFFSTAT |
2181                                             DIFF_FORMAT_SHORTSTAT |
2182                                             DIFF_FORMAT_DIRSTAT |
2183                                             DIFF_FORMAT_SUMMARY |
2184                                             DIFF_FORMAT_PATCH);
2185
2186         /*
2187          * These cases always need recursive; we do not drop caller-supplied
2188          * recursive bits for other formats here.
2189          */
2190         if (options->output_format & (DIFF_FORMAT_PATCH |
2191                                       DIFF_FORMAT_NUMSTAT |
2192                                       DIFF_FORMAT_DIFFSTAT |
2193                                       DIFF_FORMAT_SHORTSTAT |
2194                                       DIFF_FORMAT_DIRSTAT |
2195                                       DIFF_FORMAT_SUMMARY |
2196                                       DIFF_FORMAT_CHECKDIFF))
2197                 DIFF_OPT_SET(options, RECURSIVE);
2198         /*
2199          * Also pickaxe would not work very well if you do not say recursive
2200          */
2201         if (options->pickaxe)
2202                 DIFF_OPT_SET(options, RECURSIVE);
2203
2204         if (options->detect_rename && options->rename_limit < 0)
2205                 options->rename_limit = diff_rename_limit_default;
2206         if (options->setup & DIFF_SETUP_USE_CACHE) {
2207                 if (!active_cache)
2208                         /* read-cache does not die even when it fails
2209                          * so it is safe for us to do this here.  Also
2210                          * it does not smudge active_cache or active_nr
2211                          * when it fails, so we do not have to worry about
2212                          * cleaning it up ourselves either.
2213                          */
2214                         read_cache();
2215         }
2216         if (options->abbrev <= 0 || 40 < options->abbrev)
2217                 options->abbrev = 40; /* full */
2218
2219         /*
2220          * It does not make sense to show the first hit we happened
2221          * to have found.  It does not make sense not to return with
2222          * exit code in such a case either.
2223          */
2224         if (DIFF_OPT_TST(options, QUIET)) {
2225                 options->output_format = DIFF_FORMAT_NO_OUTPUT;
2226                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
2227         }
2228
2229         /*
2230          * If we postprocess in diffcore, we cannot simply return
2231          * upon the first hit.  We need to run diff as usual.
2232          */
2233         if (options->pickaxe || options->filter)
2234                 DIFF_OPT_CLR(options, QUIET);
2235
2236         return 0;
2237 }
2238
2239 static int opt_arg(const char *arg, int arg_short, const char *arg_long, int *val)
2240 {
2241         char c, *eq;
2242         int len;
2243
2244         if (*arg != '-')
2245                 return 0;
2246         c = *++arg;
2247         if (!c)
2248                 return 0;
2249         if (c == arg_short) {
2250                 c = *++arg;
2251                 if (!c)
2252                         return 1;
2253                 if (val && isdigit(c)) {
2254                         char *end;
2255                         int n = strtoul(arg, &end, 10);
2256                         if (*end)
2257                                 return 0;
2258                         *val = n;
2259                         return 1;
2260                 }
2261                 return 0;
2262         }
2263         if (c != '-')
2264                 return 0;
2265         arg++;
2266         eq = strchr(arg, '=');
2267         if (eq)
2268                 len = eq - arg;
2269         else
2270                 len = strlen(arg);
2271         if (!len || strncmp(arg, arg_long, len))
2272                 return 0;
2273         if (eq) {
2274                 int n;
2275                 char *end;
2276                 if (!isdigit(*++eq))
2277                         return 0;
2278                 n = strtoul(eq, &end, 10);
2279                 if (*end)
2280                         return 0;
2281                 *val = n;
2282         }
2283         return 1;
2284 }
2285
2286 static int diff_scoreopt_parse(const char *opt);
2287
2288 int diff_opt_parse(struct diff_options *options, const char **av, int ac)
2289 {
2290         const char *arg = av[0];
2291
2292         /* Output format options */
2293         if (!strcmp(arg, "-p") || !strcmp(arg, "-u"))
2294                 options->output_format |= DIFF_FORMAT_PATCH;
2295         else if (opt_arg(arg, 'U', "unified", &options->context))
2296                 options->output_format |= DIFF_FORMAT_PATCH;
2297         else if (!strcmp(arg, "--raw"))
2298                 options->output_format |= DIFF_FORMAT_RAW;
2299         else if (!strcmp(arg, "--patch-with-raw"))
2300                 options->output_format |= DIFF_FORMAT_PATCH | DIFF_FORMAT_RAW;
2301         else if (!strcmp(arg, "--numstat"))
2302                 options->output_format |= DIFF_FORMAT_NUMSTAT;
2303         else if (!strcmp(arg, "--shortstat"))
2304                 options->output_format |= DIFF_FORMAT_SHORTSTAT;
2305         else if (opt_arg(arg, 'X', "dirstat", &options->dirstat_percent))
2306                 options->output_format |= DIFF_FORMAT_DIRSTAT;
2307         else if (!strcmp(arg, "--cumulative"))
2308                 options->output_format |= DIFF_FORMAT_CUMULATIVE;
2309         else if (!strcmp(arg, "--check"))
2310                 options->output_format |= DIFF_FORMAT_CHECKDIFF;
2311         else if (!strcmp(arg, "--summary"))
2312                 options->output_format |= DIFF_FORMAT_SUMMARY;
2313         else if (!strcmp(arg, "--patch-with-stat"))
2314                 options->output_format |= DIFF_FORMAT_PATCH | DIFF_FORMAT_DIFFSTAT;
2315         else if (!strcmp(arg, "--name-only"))
2316                 options->output_format |= DIFF_FORMAT_NAME;
2317         else if (!strcmp(arg, "--name-status"))
2318                 options->output_format |= DIFF_FORMAT_NAME_STATUS;
2319         else if (!strcmp(arg, "-s"))
2320                 options->output_format |= DIFF_FORMAT_NO_OUTPUT;
2321         else if (!prefixcmp(arg, "--stat")) {
2322                 char *end;
2323                 int width = options->stat_width;
2324                 int name_width = options->stat_name_width;
2325                 arg += 6;
2326                 end = (char *)arg;
2327
2328                 switch (*arg) {
2329                 case '-':
2330                         if (!prefixcmp(arg, "-width="))
2331                                 width = strtoul(arg + 7, &end, 10);
2332                         else if (!prefixcmp(arg, "-name-width="))
2333                                 name_width = strtoul(arg + 12, &end, 10);
2334                         break;
2335                 case '=':
2336                         width = strtoul(arg+1, &end, 10);
2337                         if (*end == ',')
2338                                 name_width = strtoul(end+1, &end, 10);
2339                 }
2340
2341                 /* Important! This checks all the error cases! */
2342                 if (*end)
2343                         return 0;
2344                 options->output_format |= DIFF_FORMAT_DIFFSTAT;
2345                 options->stat_name_width = name_width;
2346                 options->stat_width = width;
2347         }
2348
2349         /* renames options */
2350         else if (!prefixcmp(arg, "-B")) {
2351                 if ((options->break_opt = diff_scoreopt_parse(arg)) == -1)
2352                         return -1;
2353         }
2354         else if (!prefixcmp(arg, "-M")) {
2355                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
2356                         return -1;
2357                 options->detect_rename = DIFF_DETECT_RENAME;
2358         }
2359         else if (!prefixcmp(arg, "-C")) {
2360                 if (options->detect_rename == DIFF_DETECT_COPY)
2361                         DIFF_OPT_SET(options, FIND_COPIES_HARDER);
2362                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
2363                         return -1;
2364                 options->detect_rename = DIFF_DETECT_COPY;
2365         }
2366         else if (!strcmp(arg, "--no-renames"))
2367                 options->detect_rename = 0;
2368
2369         /* xdiff options */
2370         else if (!strcmp(arg, "-w") || !strcmp(arg, "--ignore-all-space"))
2371                 options->xdl_opts |= XDF_IGNORE_WHITESPACE;
2372         else if (!strcmp(arg, "-b") || !strcmp(arg, "--ignore-space-change"))
2373                 options->xdl_opts |= XDF_IGNORE_WHITESPACE_CHANGE;
2374         else if (!strcmp(arg, "--ignore-space-at-eol"))
2375                 options->xdl_opts |= XDF_IGNORE_WHITESPACE_AT_EOL;
2376
2377         /* flags options */
2378         else if (!strcmp(arg, "--binary")) {
2379                 options->output_format |= DIFF_FORMAT_PATCH;
2380                 DIFF_OPT_SET(options, BINARY);
2381         }
2382         else if (!strcmp(arg, "--full-index"))
2383                 DIFF_OPT_SET(options, FULL_INDEX);
2384         else if (!strcmp(arg, "-a") || !strcmp(arg, "--text"))
2385                 DIFF_OPT_SET(options, TEXT);
2386         else if (!strcmp(arg, "-R"))
2387                 DIFF_OPT_SET(options, REVERSE_DIFF);
2388         else if (!strcmp(arg, "--find-copies-harder"))
2389                 DIFF_OPT_SET(options, FIND_COPIES_HARDER);
2390         else if (!strcmp(arg, "--follow"))
2391                 DIFF_OPT_SET(options, FOLLOW_RENAMES);
2392         else if (!strcmp(arg, "--color"))
2393                 DIFF_OPT_SET(options, COLOR_DIFF);
2394         else if (!strcmp(arg, "--no-color"))
2395                 DIFF_OPT_CLR(options, COLOR_DIFF);
2396         else if (!strcmp(arg, "--color-words"))
2397                 options->flags |= DIFF_OPT_COLOR_DIFF | DIFF_OPT_COLOR_DIFF_WORDS;
2398         else if (!strcmp(arg, "--exit-code"))
2399                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
2400         else if (!strcmp(arg, "--quiet"))
2401                 DIFF_OPT_SET(options, QUIET);
2402         else if (!strcmp(arg, "--ext-diff"))
2403                 DIFF_OPT_SET(options, ALLOW_EXTERNAL);
2404         else if (!strcmp(arg, "--no-ext-diff"))
2405                 DIFF_OPT_CLR(options, ALLOW_EXTERNAL);
2406
2407         /* misc options */
2408         else if (!strcmp(arg, "-z"))
2409                 options->line_termination = 0;
2410         else if (!prefixcmp(arg, "-l"))
2411                 options->rename_limit = strtoul(arg+2, NULL, 10);
2412         else if (!prefixcmp(arg, "-S"))
2413                 options->pickaxe = arg + 2;
2414         else if (!strcmp(arg, "--pickaxe-all"))
2415                 options->pickaxe_opts = DIFF_PICKAXE_ALL;
2416         else if (!strcmp(arg, "--pickaxe-regex"))
2417                 options->pickaxe_opts = DIFF_PICKAXE_REGEX;
2418         else if (!prefixcmp(arg, "-O"))
2419                 options->orderfile = arg + 2;
2420         else if (!prefixcmp(arg, "--diff-filter="))
2421                 options->filter = arg + 14;
2422         else if (!strcmp(arg, "--abbrev"))
2423                 options->abbrev = DEFAULT_ABBREV;
2424         else if (!prefixcmp(arg, "--abbrev=")) {
2425                 options->abbrev = strtoul(arg + 9, NULL, 10);
2426                 if (options->abbrev < MINIMUM_ABBREV)
2427                         options->abbrev = MINIMUM_ABBREV;
2428                 else if (40 < options->abbrev)
2429                         options->abbrev = 40;
2430         }
2431         else if (!prefixcmp(arg, "--src-prefix="))
2432                 options->a_prefix = arg + 13;
2433         else if (!prefixcmp(arg, "--dst-prefix="))
2434                 options->b_prefix = arg + 13;
2435         else if (!strcmp(arg, "--no-prefix"))
2436                 options->a_prefix = options->b_prefix = "";
2437         else
2438                 return 0;
2439         return 1;
2440 }
2441
2442 static int parse_num(const char **cp_p)
2443 {
2444         unsigned long num, scale;
2445         int ch, dot;
2446         const char *cp = *cp_p;
2447
2448         num = 0;
2449         scale = 1;
2450         dot = 0;
2451         for(;;) {
2452                 ch = *cp;
2453                 if ( !dot && ch == '.' ) {
2454                         scale = 1;
2455                         dot = 1;
2456                 } else if ( ch == '%' ) {
2457                         scale = dot ? scale*100 : 100;
2458                         cp++;   /* % is always at the end */
2459                         break;
2460                 } else if ( ch >= '0' && ch <= '9' ) {
2461                         if ( scale < 100000 ) {
2462                                 scale *= 10;
2463                                 num = (num*10) + (ch-'0');
2464                         }
2465                 } else {
2466                         break;
2467                 }
2468                 cp++;
2469         }
2470         *cp_p = cp;
2471
2472         /* user says num divided by scale and we say internally that
2473          * is MAX_SCORE * num / scale.
2474          */
2475         return (int)((num >= scale) ? MAX_SCORE : (MAX_SCORE * num / scale));
2476 }
2477
2478 static int diff_scoreopt_parse(const char *opt)
2479 {
2480         int opt1, opt2, cmd;
2481
2482         if (*opt++ != '-')
2483                 return -1;
2484         cmd = *opt++;
2485         if (cmd != 'M' && cmd != 'C' && cmd != 'B')
2486                 return -1; /* that is not a -M, -C nor -B option */
2487
2488         opt1 = parse_num(&opt);
2489         if (cmd != 'B')
2490                 opt2 = 0;
2491         else {
2492                 if (*opt == 0)
2493                         opt2 = 0;
2494                 else if (*opt != '/')
2495                         return -1; /* we expect -B80/99 or -B80 */
2496                 else {
2497                         opt++;
2498                         opt2 = parse_num(&opt);
2499                 }
2500         }
2501         if (*opt != 0)
2502                 return -1;
2503         return opt1 | (opt2 << 16);
2504 }
2505
2506 struct diff_queue_struct diff_queued_diff;
2507
2508 void diff_q(struct diff_queue_struct *queue, struct diff_filepair *dp)
2509 {
2510         if (queue->alloc <= queue->nr) {
2511                 queue->alloc = alloc_nr(queue->alloc);
2512                 queue->queue = xrealloc(queue->queue,
2513                                         sizeof(dp) * queue->alloc);
2514         }
2515         queue->queue[queue->nr++] = dp;
2516 }
2517
2518 struct diff_filepair *diff_queue(struct diff_queue_struct *queue,
2519                                  struct diff_filespec *one,
2520                                  struct diff_filespec *two)
2521 {
2522         struct diff_filepair *dp = xcalloc(1, sizeof(*dp));
2523         dp->one = one;
2524         dp->two = two;
2525         if (queue)
2526                 diff_q(queue, dp);
2527         return dp;
2528 }
2529
2530 void diff_free_filepair(struct diff_filepair *p)
2531 {
2532         free_filespec(p->one);
2533         free_filespec(p->two);
2534         free(p);
2535 }
2536
2537 /* This is different from find_unique_abbrev() in that
2538  * it stuffs the result with dots for alignment.
2539  */
2540 const char *diff_unique_abbrev(const unsigned char *sha1, int len)
2541 {
2542         int abblen;
2543         const char *abbrev;
2544         if (len == 40)
2545                 return sha1_to_hex(sha1);
2546
2547         abbrev = find_unique_abbrev(sha1, len);
2548         if (!abbrev)
2549                 return sha1_to_hex(sha1);
2550         abblen = strlen(abbrev);
2551         if (abblen < 37) {
2552                 static char hex[41];
2553                 if (len < abblen && abblen <= len + 2)
2554                         sprintf(hex, "%s%.*s", abbrev, len+3-abblen, "..");
2555                 else
2556                         sprintf(hex, "%s...", abbrev);
2557                 return hex;
2558         }
2559         return sha1_to_hex(sha1);
2560 }
2561
2562 static void diff_flush_raw(struct diff_filepair *p, struct diff_options *opt)
2563 {
2564         int line_termination = opt->line_termination;
2565         int inter_name_termination = line_termination ? '\t' : '\0';
2566
2567         if (!(opt->output_format & DIFF_FORMAT_NAME_STATUS)) {
2568                 printf(":%06o %06o %s ", p->one->mode, p->two->mode,
2569                        diff_unique_abbrev(p->one->sha1, opt->abbrev));
2570                 printf("%s ", diff_unique_abbrev(p->two->sha1, opt->abbrev));
2571         }
2572         if (p->score) {
2573                 printf("%c%03d%c", p->status, similarity_index(p),
2574                            inter_name_termination);
2575         } else {
2576                 printf("%c%c", p->status, inter_name_termination);
2577         }
2578
2579         if (p->status == DIFF_STATUS_COPIED || p->status == DIFF_STATUS_RENAMED) {
2580                 write_name_quoted(p->one->path, stdout, inter_name_termination);
2581                 write_name_quoted(p->two->path, stdout, line_termination);
2582         } else {
2583                 const char *path = p->one->mode ? p->one->path : p->two->path;
2584                 write_name_quoted(path, stdout, line_termination);
2585         }
2586 }
2587
2588 int diff_unmodified_pair(struct diff_filepair *p)
2589 {
2590         /* This function is written stricter than necessary to support
2591          * the currently implemented transformers, but the idea is to
2592          * let transformers to produce diff_filepairs any way they want,
2593          * and filter and clean them up here before producing the output.
2594          */
2595         struct diff_filespec *one = p->one, *two = p->two;
2596
2597         if (DIFF_PAIR_UNMERGED(p))
2598                 return 0; /* unmerged is interesting */
2599
2600         /* deletion, addition, mode or type change
2601          * and rename are all interesting.
2602          */
2603         if (DIFF_FILE_VALID(one) != DIFF_FILE_VALID(two) ||
2604             DIFF_PAIR_MODE_CHANGED(p) ||
2605             strcmp(one->path, two->path))
2606                 return 0;
2607
2608         /* both are valid and point at the same path.  that is, we are
2609          * dealing with a change.
2610          */
2611         if (one->sha1_valid && two->sha1_valid &&
2612             !hashcmp(one->sha1, two->sha1))
2613                 return 1; /* no change */
2614         if (!one->sha1_valid && !two->sha1_valid)
2615                 return 1; /* both look at the same file on the filesystem. */
2616         return 0;
2617 }
2618
2619 static void diff_flush_patch(struct diff_filepair *p, struct diff_options *o)
2620 {
2621         if (diff_unmodified_pair(p))
2622                 return;
2623
2624         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
2625             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
2626                 return; /* no tree diffs in patch format */
2627
2628         run_diff(p, o);
2629 }
2630
2631 static void diff_flush_stat(struct diff_filepair *p, struct diff_options *o,
2632                             struct diffstat_t *diffstat)
2633 {
2634         if (diff_unmodified_pair(p))
2635                 return;
2636
2637         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
2638             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
2639                 return; /* no tree diffs in patch format */
2640
2641         run_diffstat(p, o, diffstat);
2642 }
2643
2644 static void diff_flush_checkdiff(struct diff_filepair *p,
2645                 struct diff_options *o)
2646 {
2647         if (diff_unmodified_pair(p))
2648                 return;
2649
2650         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
2651             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
2652                 return; /* no tree diffs in patch format */
2653
2654         run_checkdiff(p, o);
2655 }
2656
2657 int diff_queue_is_empty(void)
2658 {
2659         struct diff_queue_struct *q = &diff_queued_diff;
2660         int i;
2661         for (i = 0; i < q->nr; i++)
2662                 if (!diff_unmodified_pair(q->queue[i]))
2663                         return 0;
2664         return 1;
2665 }
2666
2667 #if DIFF_DEBUG
2668 void diff_debug_filespec(struct diff_filespec *s, int x, const char *one)
2669 {
2670         fprintf(stderr, "queue[%d] %s (%s) %s %06o %s\n",
2671                 x, one ? one : "",
2672                 s->path,
2673                 DIFF_FILE_VALID(s) ? "valid" : "invalid",
2674                 s->mode,
2675                 s->sha1_valid ? sha1_to_hex(s->sha1) : "");
2676         fprintf(stderr, "queue[%d] %s size %lu flags %d\n",
2677                 x, one ? one : "",
2678                 s->size, s->xfrm_flags);
2679 }
2680
2681 void diff_debug_filepair(const struct diff_filepair *p, int i)
2682 {
2683         diff_debug_filespec(p->one, i, "one");
2684         diff_debug_filespec(p->two, i, "two");
2685         fprintf(stderr, "score %d, status %c rename_used %d broken %d\n",
2686                 p->score, p->status ? p->status : '?',
2687                 p->one->rename_used, p->broken_pair);
2688 }
2689
2690 void diff_debug_queue(const char *msg, struct diff_queue_struct *q)
2691 {
2692         int i;
2693         if (msg)
2694                 fprintf(stderr, "%s\n", msg);
2695         fprintf(stderr, "q->nr = %d\n", q->nr);
2696         for (i = 0; i < q->nr; i++) {
2697                 struct diff_filepair *p = q->queue[i];
2698                 diff_debug_filepair(p, i);
2699         }
2700 }
2701 #endif
2702
2703 static void diff_resolve_rename_copy(void)
2704 {
2705         int i;
2706         struct diff_filepair *p;
2707         struct diff_queue_struct *q = &diff_queued_diff;
2708
2709         diff_debug_queue("resolve-rename-copy", q);
2710
2711         for (i = 0; i < q->nr; i++) {
2712                 p = q->queue[i];
2713                 p->status = 0; /* undecided */
2714                 if (DIFF_PAIR_UNMERGED(p))
2715                         p->status = DIFF_STATUS_UNMERGED;
2716                 else if (!DIFF_FILE_VALID(p->one))
2717                         p->status = DIFF_STATUS_ADDED;
2718                 else if (!DIFF_FILE_VALID(p->two))
2719                         p->status = DIFF_STATUS_DELETED;
2720                 else if (DIFF_PAIR_TYPE_CHANGED(p))
2721                         p->status = DIFF_STATUS_TYPE_CHANGED;
2722
2723                 /* from this point on, we are dealing with a pair
2724                  * whose both sides are valid and of the same type, i.e.
2725                  * either in-place edit or rename/copy edit.
2726                  */
2727                 else if (DIFF_PAIR_RENAME(p)) {
2728                         /*
2729                          * A rename might have re-connected a broken
2730                          * pair up, causing the pathnames to be the
2731                          * same again. If so, that's not a rename at
2732                          * all, just a modification..
2733                          *
2734                          * Otherwise, see if this source was used for
2735                          * multiple renames, in which case we decrement
2736                          * the count, and call it a copy.
2737                          */
2738                         if (!strcmp(p->one->path, p->two->path))
2739                                 p->status = DIFF_STATUS_MODIFIED;
2740                         else if (--p->one->rename_used > 0)
2741                                 p->status = DIFF_STATUS_COPIED;
2742                         else
2743                                 p->status = DIFF_STATUS_RENAMED;
2744                 }
2745                 else if (hashcmp(p->one->sha1, p->two->sha1) ||
2746                          p->one->mode != p->two->mode ||
2747                          is_null_sha1(p->one->sha1))
2748                         p->status = DIFF_STATUS_MODIFIED;
2749                 else {
2750                         /* This is a "no-change" entry and should not
2751                          * happen anymore, but prepare for broken callers.
2752                          */
2753                         error("feeding unmodified %s to diffcore",
2754                               p->one->path);
2755                         p->status = DIFF_STATUS_UNKNOWN;
2756                 }
2757         }
2758         diff_debug_queue("resolve-rename-copy done", q);
2759 }
2760
2761 static int check_pair_status(struct diff_filepair *p)
2762 {
2763         switch (p->status) {
2764         case DIFF_STATUS_UNKNOWN:
2765                 return 0;
2766         case 0:
2767                 die("internal error in diff-resolve-rename-copy");
2768         default:
2769                 return 1;
2770         }
2771 }
2772
2773 static void flush_one_pair(struct diff_filepair *p, struct diff_options *opt)
2774 {
2775         int fmt = opt->output_format;
2776
2777         if (fmt & DIFF_FORMAT_CHECKDIFF)
2778                 diff_flush_checkdiff(p, opt);
2779         else if (fmt & (DIFF_FORMAT_RAW | DIFF_FORMAT_NAME_STATUS))
2780                 diff_flush_raw(p, opt);
2781         else if (fmt & DIFF_FORMAT_NAME)
2782                 write_name_quoted(p->two->path, stdout, opt->line_termination);
2783 }
2784
2785 static void show_file_mode_name(const char *newdelete, struct diff_filespec *fs)
2786 {
2787         if (fs->mode)
2788                 printf(" %s mode %06o ", newdelete, fs->mode);
2789         else
2790                 printf(" %s ", newdelete);
2791         write_name_quoted(fs->path, stdout, '\n');
2792 }
2793
2794
2795 static void show_mode_change(struct diff_filepair *p, int show_name)
2796 {
2797         if (p->one->mode && p->two->mode && p->one->mode != p->two->mode) {
2798                 printf(" mode change %06o => %06o%c", p->one->mode, p->two->mode,
2799                         show_name ? ' ' : '\n');
2800                 if (show_name) {
2801                         write_name_quoted(p->two->path, stdout, '\n');
2802                 }
2803         }
2804 }
2805
2806 static void show_rename_copy(const char *renamecopy, struct diff_filepair *p)
2807 {
2808         char *names = pprint_rename(p->one->path, p->two->path);
2809
2810         printf(" %s %s (%d%%)\n", renamecopy, names, similarity_index(p));
2811         free(names);
2812         show_mode_change(p, 0);
2813 }
2814
2815 static void diff_summary(struct diff_filepair *p)
2816 {
2817         switch(p->status) {
2818         case DIFF_STATUS_DELETED:
2819                 show_file_mode_name("delete", p->one);
2820                 break;
2821         case DIFF_STATUS_ADDED:
2822                 show_file_mode_name("create", p->two);
2823                 break;
2824         case DIFF_STATUS_COPIED:
2825                 show_rename_copy("copy", p);
2826                 break;
2827         case DIFF_STATUS_RENAMED:
2828                 show_rename_copy("rename", p);
2829                 break;
2830         default:
2831                 if (p->score) {
2832                         fputs(" rewrite ", stdout);
2833                         write_name_quoted(p->two->path, stdout, ' ');
2834                         printf("(%d%%)\n", similarity_index(p));
2835                 }
2836                 show_mode_change(p, !p->score);
2837                 break;
2838         }
2839 }
2840
2841 struct patch_id_t {
2842         struct xdiff_emit_state xm;
2843         SHA_CTX *ctx;
2844         int patchlen;
2845 };
2846
2847 static int remove_space(char *line, int len)
2848 {
2849         int i;
2850         char *dst = line;
2851         unsigned char c;
2852
2853         for (i = 0; i < len; i++)
2854                 if (!isspace((c = line[i])))
2855                         *dst++ = c;
2856
2857         return dst - line;
2858 }
2859
2860 static void patch_id_consume(void *priv, char *line, unsigned long len)
2861 {
2862         struct patch_id_t *data = priv;
2863         int new_len;
2864
2865         /* Ignore line numbers when computing the SHA1 of the patch */
2866         if (!prefixcmp(line, "@@ -"))
2867                 return;
2868
2869         new_len = remove_space(line, len);
2870
2871         SHA1_Update(data->ctx, line, new_len);
2872         data->patchlen += new_len;
2873 }
2874
2875 /* returns 0 upon success, and writes result into sha1 */
2876 static int diff_get_patch_id(struct diff_options *options, unsigned char *sha1)
2877 {
2878         struct diff_queue_struct *q = &diff_queued_diff;
2879         int i;
2880         SHA_CTX ctx;
2881         struct patch_id_t data;
2882         char buffer[PATH_MAX * 4 + 20];
2883
2884         SHA1_Init(&ctx);
2885         memset(&data, 0, sizeof(struct patch_id_t));
2886         data.ctx = &ctx;
2887         data.xm.consume = patch_id_consume;
2888
2889         for (i = 0; i < q->nr; i++) {
2890                 xpparam_t xpp;
2891                 xdemitconf_t xecfg;
2892                 xdemitcb_t ecb;
2893                 mmfile_t mf1, mf2;
2894                 struct diff_filepair *p = q->queue[i];
2895                 int len1, len2;
2896
2897                 memset(&xecfg, 0, sizeof(xecfg));
2898                 if (p->status == 0)
2899                         return error("internal diff status error");
2900                 if (p->status == DIFF_STATUS_UNKNOWN)
2901                         continue;
2902                 if (diff_unmodified_pair(p))
2903                         continue;
2904                 if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
2905                     (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
2906                         continue;
2907                 if (DIFF_PAIR_UNMERGED(p))
2908                         continue;
2909
2910                 diff_fill_sha1_info(p->one);
2911                 diff_fill_sha1_info(p->two);
2912                 if (fill_mmfile(&mf1, p->one) < 0 ||
2913                                 fill_mmfile(&mf2, p->two) < 0)
2914                         return error("unable to read files to diff");
2915
2916                 len1 = remove_space(p->one->path, strlen(p->one->path));
2917                 len2 = remove_space(p->two->path, strlen(p->two->path));
2918                 if (p->one->mode == 0)
2919                         len1 = snprintf(buffer, sizeof(buffer),
2920                                         "diff--gita/%.*sb/%.*s"
2921                                         "newfilemode%06o"
2922                                         "---/dev/null"
2923                                         "+++b/%.*s",
2924                                         len1, p->one->path,
2925                                         len2, p->two->path,
2926                                         p->two->mode,
2927                                         len2, p->two->path);
2928                 else if (p->two->mode == 0)
2929                         len1 = snprintf(buffer, sizeof(buffer),
2930                                         "diff--gita/%.*sb/%.*s"
2931                                         "deletedfilemode%06o"
2932                                         "---a/%.*s"
2933                                         "+++/dev/null",
2934                                         len1, p->one->path,
2935                                         len2, p->two->path,
2936                                         p->one->mode,
2937                                         len1, p->one->path);
2938                 else
2939                         len1 = snprintf(buffer, sizeof(buffer),
2940                                         "diff--gita/%.*sb/%.*s"
2941                                         "---a/%.*s"
2942                                         "+++b/%.*s",
2943                                         len1, p->one->path,
2944                                         len2, p->two->path,
2945                                         len1, p->one->path,
2946                                         len2, p->two->path);
2947                 SHA1_Update(&ctx, buffer, len1);
2948
2949                 xpp.flags = XDF_NEED_MINIMAL;
2950                 xecfg.ctxlen = 3;
2951                 xecfg.flags = XDL_EMIT_FUNCNAMES;
2952                 ecb.outf = xdiff_outf;
2953                 ecb.priv = &data;
2954                 xdi_diff(&mf1, &mf2, &xpp, &xecfg, &ecb);
2955         }
2956
2957         SHA1_Final(sha1, &ctx);
2958         return 0;
2959 }
2960
2961 int diff_flush_patch_id(struct diff_options *options, unsigned char *sha1)
2962 {
2963         struct diff_queue_struct *q = &diff_queued_diff;
2964         int i;
2965         int result = diff_get_patch_id(options, sha1);
2966
2967         for (i = 0; i < q->nr; i++)
2968                 diff_free_filepair(q->queue[i]);
2969
2970         free(q->queue);
2971         q->queue = NULL;
2972         q->nr = q->alloc = 0;
2973
2974         return result;
2975 }
2976
2977 static int is_summary_empty(const struct diff_queue_struct *q)
2978 {
2979         int i;
2980
2981         for (i = 0; i < q->nr; i++) {
2982                 const struct diff_filepair *p = q->queue[i];
2983
2984                 switch (p->status) {
2985                 case DIFF_STATUS_DELETED:
2986                 case DIFF_STATUS_ADDED:
2987                 case DIFF_STATUS_COPIED:
2988                 case DIFF_STATUS_RENAMED:
2989                         return 0;
2990                 default:
2991                         if (p->score)
2992                                 return 0;
2993                         if (p->one->mode && p->two->mode &&
2994                             p->one->mode != p->two->mode)
2995                                 return 0;
2996                         break;
2997                 }
2998         }
2999         return 1;
3000 }
3001
3002 void diff_flush(struct diff_options *options)
3003 {
3004         struct diff_queue_struct *q = &diff_queued_diff;
3005         int i, output_format = options->output_format;
3006         int separator = 0;
3007
3008         /*
3009          * Order: raw, stat, summary, patch
3010          * or:    name/name-status/checkdiff (other bits clear)
3011          */
3012         if (!q->nr)
3013                 goto free_queue;
3014
3015         if (output_format & (DIFF_FORMAT_RAW |
3016                              DIFF_FORMAT_NAME |
3017                              DIFF_FORMAT_NAME_STATUS |
3018                              DIFF_FORMAT_CHECKDIFF)) {
3019                 for (i = 0; i < q->nr; i++) {
3020                         struct diff_filepair *p = q->queue[i];
3021                         if (check_pair_status(p))
3022                                 flush_one_pair(p, options);
3023                 }
3024                 separator++;
3025         }
3026
3027         if (output_format & (DIFF_FORMAT_DIFFSTAT|DIFF_FORMAT_SHORTSTAT|DIFF_FORMAT_NUMSTAT|DIFF_FORMAT_DIRSTAT)) {
3028                 struct diffstat_t diffstat;
3029
3030                 memset(&diffstat, 0, sizeof(struct diffstat_t));
3031                 diffstat.xm.consume = diffstat_consume;
3032                 for (i = 0; i < q->nr; i++) {
3033                         struct diff_filepair *p = q->queue[i];
3034                         if (check_pair_status(p))
3035                                 diff_flush_stat(p, options, &diffstat);
3036                 }
3037                 if (output_format & DIFF_FORMAT_DIRSTAT)
3038                         show_dirstat(&diffstat, options);
3039                 if (output_format & DIFF_FORMAT_NUMSTAT)
3040                         show_numstat(&diffstat, options);
3041                 if (output_format & DIFF_FORMAT_DIFFSTAT)
3042                         show_stats(&diffstat, options);
3043                 if (output_format & DIFF_FORMAT_SHORTSTAT)
3044                         show_shortstats(&diffstat);
3045                 free_diffstat_info(&diffstat);
3046                 separator++;
3047         }
3048
3049         if (output_format & DIFF_FORMAT_SUMMARY && !is_summary_empty(q)) {
3050                 for (i = 0; i < q->nr; i++)
3051                         diff_summary(q->queue[i]);
3052                 separator++;
3053         }
3054
3055         if (output_format & DIFF_FORMAT_PATCH) {
3056                 if (separator) {
3057                         if (options->stat_sep) {
3058                                 /* attach patch instead of inline */
3059                                 fputs(options->stat_sep, stdout);
3060                         } else {
3061                                 putchar(options->line_termination);
3062                         }
3063                 }
3064
3065                 for (i = 0; i < q->nr; i++) {
3066                         struct diff_filepair *p = q->queue[i];
3067                         if (check_pair_status(p))
3068                                 diff_flush_patch(p, options);
3069                 }
3070         }
3071
3072         if (output_format & DIFF_FORMAT_CALLBACK)
3073                 options->format_callback(q, options, options->format_callback_data);
3074
3075         for (i = 0; i < q->nr; i++)
3076                 diff_free_filepair(q->queue[i]);
3077 free_queue:
3078         free(q->queue);
3079         q->queue = NULL;
3080         q->nr = q->alloc = 0;
3081 }
3082
3083 static void diffcore_apply_filter(const char *filter)
3084 {
3085         int i;
3086         struct diff_queue_struct *q = &diff_queued_diff;
3087         struct diff_queue_struct outq;
3088         outq.queue = NULL;
3089         outq.nr = outq.alloc = 0;
3090
3091         if (!filter)
3092                 return;
3093
3094         if (strchr(filter, DIFF_STATUS_FILTER_AON)) {
3095                 int found;
3096                 for (i = found = 0; !found && i < q->nr; i++) {
3097                         struct diff_filepair *p = q->queue[i];
3098                         if (((p->status == DIFF_STATUS_MODIFIED) &&
3099                              ((p->score &&
3100                                strchr(filter, DIFF_STATUS_FILTER_BROKEN)) ||
3101                               (!p->score &&
3102                                strchr(filter, DIFF_STATUS_MODIFIED)))) ||
3103                             ((p->status != DIFF_STATUS_MODIFIED) &&
3104                              strchr(filter, p->status)))
3105                                 found++;
3106                 }
3107                 if (found)
3108                         return;
3109
3110                 /* otherwise we will clear the whole queue
3111                  * by copying the empty outq at the end of this
3112                  * function, but first clear the current entries
3113                  * in the queue.
3114                  */
3115                 for (i = 0; i < q->nr; i++)
3116                         diff_free_filepair(q->queue[i]);
3117         }
3118         else {
3119                 /* Only the matching ones */
3120                 for (i = 0; i < q->nr; i++) {
3121                         struct diff_filepair *p = q->queue[i];
3122
3123                         if (((p->status == DIFF_STATUS_MODIFIED) &&
3124                              ((p->score &&
3125                                strchr(filter, DIFF_STATUS_FILTER_BROKEN)) ||
3126                               (!p->score &&
3127                                strchr(filter, DIFF_STATUS_MODIFIED)))) ||
3128                             ((p->status != DIFF_STATUS_MODIFIED) &&
3129                              strchr(filter, p->status)))
3130                                 diff_q(&outq, p);
3131                         else
3132                                 diff_free_filepair(p);
3133                 }
3134         }
3135         free(q->queue);
3136         *q = outq;
3137 }
3138
3139 /* Check whether two filespecs with the same mode and size are identical */
3140 static int diff_filespec_is_identical(struct diff_filespec *one,
3141                                       struct diff_filespec *two)
3142 {
3143         if (S_ISGITLINK(one->mode)) {
3144                 diff_fill_sha1_info(one);
3145                 diff_fill_sha1_info(two);
3146                 return !hashcmp(one->sha1, two->sha1);
3147         }
3148         if (diff_populate_filespec(one, 0))
3149                 return 0;
3150         if (diff_populate_filespec(two, 0))
3151                 return 0;
3152         return !memcmp(one->data, two->data, one->size);
3153 }
3154
3155 static void diffcore_skip_stat_unmatch(struct diff_options *diffopt)
3156 {
3157         int i;
3158         struct diff_queue_struct *q = &diff_queued_diff;
3159         struct diff_queue_struct outq;
3160         outq.queue = NULL;
3161         outq.nr = outq.alloc = 0;
3162
3163         for (i = 0; i < q->nr; i++) {
3164                 struct diff_filepair *p = q->queue[i];
3165
3166                 /*
3167                  * 1. Entries that come from stat info dirtyness
3168                  *    always have both sides (iow, not create/delete),
3169                  *    one side of the object name is unknown, with
3170                  *    the same mode and size.  Keep the ones that
3171                  *    do not match these criteria.  They have real
3172                  *    differences.
3173                  *
3174                  * 2. At this point, the file is known to be modified,
3175                  *    with the same mode and size, and the object
3176                  *    name of one side is unknown.  Need to inspect
3177                  *    the identical contents.
3178                  */
3179                 if (!DIFF_FILE_VALID(p->one) || /* (1) */
3180                     !DIFF_FILE_VALID(p->two) ||
3181                     (p->one->sha1_valid && p->two->sha1_valid) ||
3182                     (p->one->mode != p->two->mode) ||
3183                     diff_populate_filespec(p->one, 1) ||
3184                     diff_populate_filespec(p->two, 1) ||
3185                     (p->one->size != p->two->size) ||
3186                     !diff_filespec_is_identical(p->one, p->two)) /* (2) */
3187                         diff_q(&outq, p);
3188                 else {
3189                         /*
3190                          * The caller can subtract 1 from skip_stat_unmatch
3191                          * to determine how many paths were dirty only
3192                          * due to stat info mismatch.
3193                          */
3194                         if (!DIFF_OPT_TST(diffopt, NO_INDEX))
3195                                 diffopt->skip_stat_unmatch++;
3196                         diff_free_filepair(p);
3197                 }
3198         }
3199         free(q->queue);
3200         *q = outq;
3201 }
3202
3203 void diffcore_std(struct diff_options *options)
3204 {
3205         if (DIFF_OPT_TST(options, QUIET))
3206                 return;
3207
3208         if (options->skip_stat_unmatch && !DIFF_OPT_TST(options, FIND_COPIES_HARDER))
3209                 diffcore_skip_stat_unmatch(options);
3210         if (options->break_opt != -1)
3211                 diffcore_break(options->break_opt);
3212         if (options->detect_rename)
3213                 diffcore_rename(options);
3214         if (options->break_opt != -1)
3215                 diffcore_merge_broken();
3216         if (options->pickaxe)
3217                 diffcore_pickaxe(options->pickaxe, options->pickaxe_opts);
3218         if (options->orderfile)
3219                 diffcore_order(options->orderfile);
3220         diff_resolve_rename_copy();
3221         diffcore_apply_filter(options->filter);
3222
3223         if (diff_queued_diff.nr)
3224                 DIFF_OPT_SET(options, HAS_CHANGES);
3225         else
3226                 DIFF_OPT_CLR(options, HAS_CHANGES);
3227 }
3228
3229 int diff_result_code(struct diff_options *opt, int status)
3230 {
3231         int result = 0;
3232         if (!DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
3233             !(opt->output_format & DIFF_FORMAT_CHECKDIFF))
3234                 return status;
3235         if (DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
3236             DIFF_OPT_TST(opt, HAS_CHANGES))
3237                 result |= 01;
3238         if ((opt->output_format & DIFF_FORMAT_CHECKDIFF) &&
3239             DIFF_OPT_TST(opt, CHECK_FAILED))
3240                 result |= 02;
3241         return result;
3242 }
3243
3244 void diff_addremove(struct diff_options *options,
3245                     int addremove, unsigned mode,
3246                     const unsigned char *sha1,
3247                     const char *base, const char *path)
3248 {
3249         char concatpath[PATH_MAX];
3250         struct diff_filespec *one, *two;
3251
3252         /* This may look odd, but it is a preparation for
3253          * feeding "there are unchanged files which should
3254          * not produce diffs, but when you are doing copy
3255          * detection you would need them, so here they are"
3256          * entries to the diff-core.  They will be prefixed
3257          * with something like '=' or '*' (I haven't decided
3258          * which but should not make any difference).
3259          * Feeding the same new and old to diff_change()
3260          * also has the same effect.
3261          * Before the final output happens, they are pruned after
3262          * merged into rename/copy pairs as appropriate.
3263          */
3264         if (DIFF_OPT_TST(options, REVERSE_DIFF))
3265                 addremove = (addremove == '+' ? '-' :
3266                              addremove == '-' ? '+' : addremove);
3267
3268         if (!path) path = "";
3269         sprintf(concatpath, "%s%s", base, path);
3270         one = alloc_filespec(concatpath);
3271         two = alloc_filespec(concatpath);
3272
3273         if (addremove != '+')
3274                 fill_filespec(one, sha1, mode);
3275         if (addremove != '-')
3276                 fill_filespec(two, sha1, mode);
3277
3278         diff_queue(&diff_queued_diff, one, two);
3279         DIFF_OPT_SET(options, HAS_CHANGES);
3280 }
3281
3282 void diff_change(struct diff_options *options,
3283                  unsigned old_mode, unsigned new_mode,
3284                  const unsigned char *old_sha1,
3285                  const unsigned char *new_sha1,
3286                  const char *base, const char *path)
3287 {
3288         char concatpath[PATH_MAX];
3289         struct diff_filespec *one, *two;
3290
3291         if (DIFF_OPT_TST(options, REVERSE_DIFF)) {
3292                 unsigned tmp;
3293                 const unsigned char *tmp_c;
3294                 tmp = old_mode; old_mode = new_mode; new_mode = tmp;
3295                 tmp_c = old_sha1; old_sha1 = new_sha1; new_sha1 = tmp_c;
3296         }
3297         if (!path) path = "";
3298         sprintf(concatpath, "%s%s", base, path);
3299         one = alloc_filespec(concatpath);
3300         two = alloc_filespec(concatpath);
3301         fill_filespec(one, old_sha1, old_mode);
3302         fill_filespec(two, new_sha1, new_mode);
3303
3304         diff_queue(&diff_queued_diff, one, two);
3305         DIFF_OPT_SET(options, HAS_CHANGES);
3306 }
3307
3308 void diff_unmerge(struct diff_options *options,
3309                   const char *path,
3310                   unsigned mode, const unsigned char *sha1)
3311 {
3312         struct diff_filespec *one, *two;
3313         one = alloc_filespec(path);
3314         two = alloc_filespec(path);
3315         fill_filespec(one, sha1, mode);
3316         diff_queue(&diff_queued_diff, one, two)->is_unmerged = 1;
3317 }