rev-list: make "estimate_bisect_steps" non static
[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 #include "userdiff.h"
15 #include "sigchain.h"
16
17 #ifdef NO_FAST_WORKING_DIRECTORY
18 #define FAST_WORKING_DIRECTORY 0
19 #else
20 #define FAST_WORKING_DIRECTORY 1
21 #endif
22
23 static int diff_detect_rename_default;
24 static int diff_rename_limit_default = 200;
25 static int diff_suppress_blank_empty;
26 int diff_use_color_default = -1;
27 static const char *diff_word_regex_cfg;
28 static const char *external_diff_cmd_cfg;
29 int diff_auto_refresh_index = 1;
30 static int diff_mnemonic_prefix;
31
32 static char diff_colors[][COLOR_MAXLEN] = {
33         GIT_COLOR_RESET,
34         GIT_COLOR_NORMAL,       /* PLAIN */
35         GIT_COLOR_BOLD,         /* METAINFO */
36         GIT_COLOR_CYAN,         /* FRAGINFO */
37         GIT_COLOR_RED,          /* OLD */
38         GIT_COLOR_GREEN,        /* NEW */
39         GIT_COLOR_YELLOW,       /* COMMIT */
40         GIT_COLOR_BG_RED,       /* WHITESPACE */
41 };
42
43 static void diff_filespec_load_driver(struct diff_filespec *one);
44 static char *run_textconv(const char *, struct diff_filespec *, size_t *);
45
46 static int parse_diff_color_slot(const char *var, int ofs)
47 {
48         if (!strcasecmp(var+ofs, "plain"))
49                 return DIFF_PLAIN;
50         if (!strcasecmp(var+ofs, "meta"))
51                 return DIFF_METAINFO;
52         if (!strcasecmp(var+ofs, "frag"))
53                 return DIFF_FRAGINFO;
54         if (!strcasecmp(var+ofs, "old"))
55                 return DIFF_FILE_OLD;
56         if (!strcasecmp(var+ofs, "new"))
57                 return DIFF_FILE_NEW;
58         if (!strcasecmp(var+ofs, "commit"))
59                 return DIFF_COMMIT;
60         if (!strcasecmp(var+ofs, "whitespace"))
61                 return DIFF_WHITESPACE;
62         die("bad config variable '%s'", var);
63 }
64
65 static int git_config_rename(const char *var, const char *value)
66 {
67         if (!value)
68                 return DIFF_DETECT_RENAME;
69         if (!strcasecmp(value, "copies") || !strcasecmp(value, "copy"))
70                 return  DIFF_DETECT_COPY;
71         return git_config_bool(var,value) ? DIFF_DETECT_RENAME : 0;
72 }
73
74 /*
75  * These are to give UI layer defaults.
76  * The core-level commands such as git-diff-files should
77  * never be affected by the setting of diff.renames
78  * the user happens to have in the configuration file.
79  */
80 int git_diff_ui_config(const char *var, const char *value, void *cb)
81 {
82         if (!strcmp(var, "diff.color") || !strcmp(var, "color.diff")) {
83                 diff_use_color_default = git_config_colorbool(var, value, -1);
84                 return 0;
85         }
86         if (!strcmp(var, "diff.renames")) {
87                 diff_detect_rename_default = git_config_rename(var, value);
88                 return 0;
89         }
90         if (!strcmp(var, "diff.autorefreshindex")) {
91                 diff_auto_refresh_index = git_config_bool(var, value);
92                 return 0;
93         }
94         if (!strcmp(var, "diff.mnemonicprefix")) {
95                 diff_mnemonic_prefix = git_config_bool(var, value);
96                 return 0;
97         }
98         if (!strcmp(var, "diff.external"))
99                 return git_config_string(&external_diff_cmd_cfg, var, value);
100         if (!strcmp(var, "diff.wordregex"))
101                 return git_config_string(&diff_word_regex_cfg, var, value);
102
103         return git_diff_basic_config(var, value, cb);
104 }
105
106 int git_diff_basic_config(const char *var, const char *value, void *cb)
107 {
108         if (!strcmp(var, "diff.renamelimit")) {
109                 diff_rename_limit_default = git_config_int(var, value);
110                 return 0;
111         }
112
113         switch (userdiff_config(var, value)) {
114                 case 0: break;
115                 case -1: return -1;
116                 default: return 0;
117         }
118
119         if (!prefixcmp(var, "diff.color.") || !prefixcmp(var, "color.diff.")) {
120                 int slot = parse_diff_color_slot(var, 11);
121                 if (!value)
122                         return config_error_nonbool(var);
123                 color_parse(value, var, diff_colors[slot]);
124                 return 0;
125         }
126
127         /* like GNU diff's --suppress-blank-empty option  */
128         if (!strcmp(var, "diff.suppressblankempty") ||
129                         /* for backwards compatibility */
130                         !strcmp(var, "diff.suppress-blank-empty")) {
131                 diff_suppress_blank_empty = git_config_bool(var, value);
132                 return 0;
133         }
134
135         return git_color_default_config(var, value, cb);
136 }
137
138 static char *quote_two(const char *one, const char *two)
139 {
140         int need_one = quote_c_style(one, NULL, NULL, 1);
141         int need_two = quote_c_style(two, NULL, NULL, 1);
142         struct strbuf res = STRBUF_INIT;
143
144         if (need_one + need_two) {
145                 strbuf_addch(&res, '"');
146                 quote_c_style(one, &res, NULL, 1);
147                 quote_c_style(two, &res, NULL, 1);
148                 strbuf_addch(&res, '"');
149         } else {
150                 strbuf_addstr(&res, one);
151                 strbuf_addstr(&res, two);
152         }
153         return strbuf_detach(&res, NULL);
154 }
155
156 static const char *external_diff(void)
157 {
158         static const char *external_diff_cmd = NULL;
159         static int done_preparing = 0;
160
161         if (done_preparing)
162                 return external_diff_cmd;
163         external_diff_cmd = getenv("GIT_EXTERNAL_DIFF");
164         if (!external_diff_cmd)
165                 external_diff_cmd = external_diff_cmd_cfg;
166         done_preparing = 1;
167         return external_diff_cmd;
168 }
169
170 static struct diff_tempfile {
171         const char *name; /* filename external diff should read from */
172         char hex[41];
173         char mode[10];
174         char tmp_path[PATH_MAX];
175 } diff_temp[2];
176
177 static struct diff_tempfile *claim_diff_tempfile(void) {
178         int i;
179         for (i = 0; i < ARRAY_SIZE(diff_temp); i++)
180                 if (!diff_temp[i].name)
181                         return diff_temp + i;
182         die("BUG: diff is failing to clean up its tempfiles");
183 }
184
185 static int remove_tempfile_installed;
186
187 static void remove_tempfile(void)
188 {
189         int i;
190         for (i = 0; i < ARRAY_SIZE(diff_temp); i++) {
191                 if (diff_temp[i].name == diff_temp[i].tmp_path)
192                         unlink(diff_temp[i].name);
193                 diff_temp[i].name = NULL;
194         }
195 }
196
197 static void remove_tempfile_on_signal(int signo)
198 {
199         remove_tempfile();
200         sigchain_pop(signo);
201         raise(signo);
202 }
203
204 static int count_lines(const char *data, int size)
205 {
206         int count, ch, completely_empty = 1, nl_just_seen = 0;
207         count = 0;
208         while (0 < size--) {
209                 ch = *data++;
210                 if (ch == '\n') {
211                         count++;
212                         nl_just_seen = 1;
213                         completely_empty = 0;
214                 }
215                 else {
216                         nl_just_seen = 0;
217                         completely_empty = 0;
218                 }
219         }
220         if (completely_empty)
221                 return 0;
222         if (!nl_just_seen)
223                 count++; /* no trailing newline */
224         return count;
225 }
226
227 static void print_line_count(FILE *file, int count)
228 {
229         switch (count) {
230         case 0:
231                 fprintf(file, "0,0");
232                 break;
233         case 1:
234                 fprintf(file, "1");
235                 break;
236         default:
237                 fprintf(file, "1,%d", count);
238                 break;
239         }
240 }
241
242 static void copy_file_with_prefix(FILE *file,
243                                   int prefix, const char *data, int size,
244                                   const char *set, const char *reset)
245 {
246         int ch, nl_just_seen = 1;
247         while (0 < size--) {
248                 ch = *data++;
249                 if (nl_just_seen) {
250                         fputs(set, file);
251                         putc(prefix, file);
252                 }
253                 if (ch == '\n') {
254                         nl_just_seen = 1;
255                         fputs(reset, file);
256                 } else
257                         nl_just_seen = 0;
258                 putc(ch, file);
259         }
260         if (!nl_just_seen)
261                 fprintf(file, "%s\n\\ No newline at end of file\n", reset);
262 }
263
264 static void emit_rewrite_diff(const char *name_a,
265                               const char *name_b,
266                               struct diff_filespec *one,
267                               struct diff_filespec *two,
268                               const char *textconv_one,
269                               const char *textconv_two,
270                               struct diff_options *o)
271 {
272         int lc_a, lc_b;
273         int color_diff = DIFF_OPT_TST(o, COLOR_DIFF);
274         const char *name_a_tab, *name_b_tab;
275         const char *metainfo = diff_get_color(color_diff, DIFF_METAINFO);
276         const char *fraginfo = diff_get_color(color_diff, DIFF_FRAGINFO);
277         const char *old = diff_get_color(color_diff, DIFF_FILE_OLD);
278         const char *new = diff_get_color(color_diff, DIFF_FILE_NEW);
279         const char *reset = diff_get_color(color_diff, DIFF_RESET);
280         static struct strbuf a_name = STRBUF_INIT, b_name = STRBUF_INIT;
281         const char *a_prefix, *b_prefix;
282         const char *data_one, *data_two;
283         size_t size_one, size_two;
284
285         if (diff_mnemonic_prefix && DIFF_OPT_TST(o, REVERSE_DIFF)) {
286                 a_prefix = o->b_prefix;
287                 b_prefix = o->a_prefix;
288         } else {
289                 a_prefix = o->a_prefix;
290                 b_prefix = o->b_prefix;
291         }
292
293         name_a += (*name_a == '/');
294         name_b += (*name_b == '/');
295         name_a_tab = strchr(name_a, ' ') ? "\t" : "";
296         name_b_tab = strchr(name_b, ' ') ? "\t" : "";
297
298         strbuf_reset(&a_name);
299         strbuf_reset(&b_name);
300         quote_two_c_style(&a_name, a_prefix, name_a, 0);
301         quote_two_c_style(&b_name, b_prefix, name_b, 0);
302
303         diff_populate_filespec(one, 0);
304         diff_populate_filespec(two, 0);
305         if (textconv_one) {
306                 data_one = run_textconv(textconv_one, one, &size_one);
307                 if (!data_one)
308                         die("unable to read files to diff");
309         }
310         else {
311                 data_one = one->data;
312                 size_one = one->size;
313         }
314         if (textconv_two) {
315                 data_two = run_textconv(textconv_two, two, &size_two);
316                 if (!data_two)
317                         die("unable to read files to diff");
318         }
319         else {
320                 data_two = two->data;
321                 size_two = two->size;
322         }
323
324         lc_a = count_lines(data_one, size_one);
325         lc_b = count_lines(data_two, size_two);
326         fprintf(o->file,
327                 "%s--- %s%s%s\n%s+++ %s%s%s\n%s@@ -",
328                 metainfo, a_name.buf, name_a_tab, reset,
329                 metainfo, b_name.buf, name_b_tab, reset, fraginfo);
330         print_line_count(o->file, lc_a);
331         fprintf(o->file, " +");
332         print_line_count(o->file, lc_b);
333         fprintf(o->file, " @@%s\n", reset);
334         if (lc_a)
335                 copy_file_with_prefix(o->file, '-', data_one, size_one, old, reset);
336         if (lc_b)
337                 copy_file_with_prefix(o->file, '+', data_two, size_two, new, reset);
338 }
339
340 static int fill_mmfile(mmfile_t *mf, struct diff_filespec *one)
341 {
342         if (!DIFF_FILE_VALID(one)) {
343                 mf->ptr = (char *)""; /* does not matter */
344                 mf->size = 0;
345                 return 0;
346         }
347         else if (diff_populate_filespec(one, 0))
348                 return -1;
349
350         mf->ptr = one->data;
351         mf->size = one->size;
352         return 0;
353 }
354
355 struct diff_words_buffer {
356         mmfile_t text;
357         long alloc;
358         struct diff_words_orig {
359                 const char *begin, *end;
360         } *orig;
361         int orig_nr, orig_alloc;
362 };
363
364 static void diff_words_append(char *line, unsigned long len,
365                 struct diff_words_buffer *buffer)
366 {
367         ALLOC_GROW(buffer->text.ptr, buffer->text.size + len, buffer->alloc);
368         line++;
369         len--;
370         memcpy(buffer->text.ptr + buffer->text.size, line, len);
371         buffer->text.size += len;
372         buffer->text.ptr[buffer->text.size] = '\0';
373 }
374
375 struct diff_words_data {
376         struct diff_words_buffer minus, plus;
377         const char *current_plus;
378         FILE *file;
379         regex_t *word_regex;
380 };
381
382 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
383 {
384         struct diff_words_data *diff_words = priv;
385         int minus_first, minus_len, plus_first, plus_len;
386         const char *minus_begin, *minus_end, *plus_begin, *plus_end;
387
388         if (line[0] != '@' || parse_hunk_header(line, len,
389                         &minus_first, &minus_len, &plus_first, &plus_len))
390                 return;
391
392         /* POSIX requires that first be decremented by one if len == 0... */
393         if (minus_len) {
394                 minus_begin = diff_words->minus.orig[minus_first].begin;
395                 minus_end =
396                         diff_words->minus.orig[minus_first + minus_len - 1].end;
397         } else
398                 minus_begin = minus_end =
399                         diff_words->minus.orig[minus_first].end;
400
401         if (plus_len) {
402                 plus_begin = diff_words->plus.orig[plus_first].begin;
403                 plus_end = diff_words->plus.orig[plus_first + plus_len - 1].end;
404         } else
405                 plus_begin = plus_end = diff_words->plus.orig[plus_first].end;
406
407         if (diff_words->current_plus != plus_begin)
408                 fwrite(diff_words->current_plus,
409                                 plus_begin - diff_words->current_plus, 1,
410                                 diff_words->file);
411         if (minus_begin != minus_end)
412                 color_fwrite_lines(diff_words->file,
413                                 diff_get_color(1, DIFF_FILE_OLD),
414                                 minus_end - minus_begin, minus_begin);
415         if (plus_begin != plus_end)
416                 color_fwrite_lines(diff_words->file,
417                                 diff_get_color(1, DIFF_FILE_NEW),
418                                 plus_end - plus_begin, plus_begin);
419
420         diff_words->current_plus = plus_end;
421 }
422
423 /* This function starts looking at *begin, and returns 0 iff a word was found. */
424 static int find_word_boundaries(mmfile_t *buffer, regex_t *word_regex,
425                 int *begin, int *end)
426 {
427         if (word_regex && *begin < buffer->size) {
428                 regmatch_t match[1];
429                 if (!regexec(word_regex, buffer->ptr + *begin, 1, match, 0)) {
430                         char *p = memchr(buffer->ptr + *begin + match[0].rm_so,
431                                         '\n', match[0].rm_eo - match[0].rm_so);
432                         *end = p ? p - buffer->ptr : match[0].rm_eo + *begin;
433                         *begin += match[0].rm_so;
434                         return *begin >= *end;
435                 }
436                 return -1;
437         }
438
439         /* find the next word */
440         while (*begin < buffer->size && isspace(buffer->ptr[*begin]))
441                 (*begin)++;
442         if (*begin >= buffer->size)
443                 return -1;
444
445         /* find the end of the word */
446         *end = *begin + 1;
447         while (*end < buffer->size && !isspace(buffer->ptr[*end]))
448                 (*end)++;
449
450         return 0;
451 }
452
453 /*
454  * This function splits the words in buffer->text, stores the list with
455  * newline separator into out, and saves the offsets of the original words
456  * in buffer->orig.
457  */
458 static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out,
459                 regex_t *word_regex)
460 {
461         int i, j;
462         long alloc = 0;
463
464         out->size = 0;
465         out->ptr = NULL;
466
467         /* fake an empty "0th" word */
468         ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
469         buffer->orig[0].begin = buffer->orig[0].end = buffer->text.ptr;
470         buffer->orig_nr = 1;
471
472         for (i = 0; i < buffer->text.size; i++) {
473                 if (find_word_boundaries(&buffer->text, word_regex, &i, &j))
474                         return;
475
476                 /* store original boundaries */
477                 ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
478                                 buffer->orig_alloc);
479                 buffer->orig[buffer->orig_nr].begin = buffer->text.ptr + i;
480                 buffer->orig[buffer->orig_nr].end = buffer->text.ptr + j;
481                 buffer->orig_nr++;
482
483                 /* store one word */
484                 ALLOC_GROW(out->ptr, out->size + j - i + 1, alloc);
485                 memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
486                 out->ptr[out->size + j - i] = '\n';
487                 out->size += j - i + 1;
488
489                 i = j - 1;
490         }
491 }
492
493 /* this executes the word diff on the accumulated buffers */
494 static void diff_words_show(struct diff_words_data *diff_words)
495 {
496         xpparam_t xpp;
497         xdemitconf_t xecfg;
498         xdemitcb_t ecb;
499         mmfile_t minus, plus;
500
501         /* special case: only removal */
502         if (!diff_words->plus.text.size) {
503                 color_fwrite_lines(diff_words->file,
504                         diff_get_color(1, DIFF_FILE_OLD),
505                         diff_words->minus.text.size, diff_words->minus.text.ptr);
506                 diff_words->minus.text.size = 0;
507                 return;
508         }
509
510         diff_words->current_plus = diff_words->plus.text.ptr;
511
512         memset(&xpp, 0, sizeof(xpp));
513         memset(&xecfg, 0, sizeof(xecfg));
514         diff_words_fill(&diff_words->minus, &minus, diff_words->word_regex);
515         diff_words_fill(&diff_words->plus, &plus, diff_words->word_regex);
516         xpp.flags = XDF_NEED_MINIMAL;
517         /* as only the hunk header will be parsed, we need a 0-context */
518         xecfg.ctxlen = 0;
519         xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
520                       &xpp, &xecfg, &ecb);
521         free(minus.ptr);
522         free(plus.ptr);
523         if (diff_words->current_plus != diff_words->plus.text.ptr +
524                         diff_words->plus.text.size)
525                 fwrite(diff_words->current_plus,
526                         diff_words->plus.text.ptr + diff_words->plus.text.size
527                         - diff_words->current_plus, 1,
528                         diff_words->file);
529         diff_words->minus.text.size = diff_words->plus.text.size = 0;
530 }
531
532 typedef unsigned long (*sane_truncate_fn)(char *line, unsigned long len);
533
534 struct emit_callback {
535         int nparents, color_diff;
536         unsigned ws_rule;
537         sane_truncate_fn truncate;
538         const char **label_path;
539         struct diff_words_data *diff_words;
540         int *found_changesp;
541         FILE *file;
542 };
543
544 static void free_diff_words_data(struct emit_callback *ecbdata)
545 {
546         if (ecbdata->diff_words) {
547                 /* flush buffers */
548                 if (ecbdata->diff_words->minus.text.size ||
549                                 ecbdata->diff_words->plus.text.size)
550                         diff_words_show(ecbdata->diff_words);
551
552                 free (ecbdata->diff_words->minus.text.ptr);
553                 free (ecbdata->diff_words->minus.orig);
554                 free (ecbdata->diff_words->plus.text.ptr);
555                 free (ecbdata->diff_words->plus.orig);
556                 free(ecbdata->diff_words->word_regex);
557                 free(ecbdata->diff_words);
558                 ecbdata->diff_words = NULL;
559         }
560 }
561
562 const char *diff_get_color(int diff_use_color, enum color_diff ix)
563 {
564         if (diff_use_color)
565                 return diff_colors[ix];
566         return "";
567 }
568
569 static void emit_line(FILE *file, const char *set, const char *reset, const char *line, int len)
570 {
571         int has_trailing_newline, has_trailing_carriage_return;
572
573         has_trailing_newline = (len > 0 && line[len-1] == '\n');
574         if (has_trailing_newline)
575                 len--;
576         has_trailing_carriage_return = (len > 0 && line[len-1] == '\r');
577         if (has_trailing_carriage_return)
578                 len--;
579
580         fputs(set, file);
581         fwrite(line, len, 1, file);
582         fputs(reset, file);
583         if (has_trailing_carriage_return)
584                 fputc('\r', file);
585         if (has_trailing_newline)
586                 fputc('\n', file);
587 }
588
589 static void emit_add_line(const char *reset, struct emit_callback *ecbdata, const char *line, int len)
590 {
591         const char *ws = diff_get_color(ecbdata->color_diff, DIFF_WHITESPACE);
592         const char *set = diff_get_color(ecbdata->color_diff, DIFF_FILE_NEW);
593
594         if (!*ws)
595                 emit_line(ecbdata->file, set, reset, line, len);
596         else {
597                 /* Emit just the prefix, then the rest. */
598                 emit_line(ecbdata->file, set, reset, line, ecbdata->nparents);
599                 ws_check_emit(line + ecbdata->nparents,
600                               len - ecbdata->nparents, ecbdata->ws_rule,
601                               ecbdata->file, set, reset, ws);
602         }
603 }
604
605 static unsigned long sane_truncate_line(struct emit_callback *ecb, char *line, unsigned long len)
606 {
607         const char *cp;
608         unsigned long allot;
609         size_t l = len;
610
611         if (ecb->truncate)
612                 return ecb->truncate(line, len);
613         cp = line;
614         allot = l;
615         while (0 < l) {
616                 (void) utf8_width(&cp, &l);
617                 if (!cp)
618                         break; /* truncated in the middle? */
619         }
620         return allot - l;
621 }
622
623 static void fn_out_consume(void *priv, char *line, unsigned long len)
624 {
625         int i;
626         int color;
627         struct emit_callback *ecbdata = priv;
628         const char *meta = diff_get_color(ecbdata->color_diff, DIFF_METAINFO);
629         const char *plain = diff_get_color(ecbdata->color_diff, DIFF_PLAIN);
630         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
631
632         *(ecbdata->found_changesp) = 1;
633
634         if (ecbdata->label_path[0]) {
635                 const char *name_a_tab, *name_b_tab;
636
637                 name_a_tab = strchr(ecbdata->label_path[0], ' ') ? "\t" : "";
638                 name_b_tab = strchr(ecbdata->label_path[1], ' ') ? "\t" : "";
639
640                 fprintf(ecbdata->file, "%s--- %s%s%s\n",
641                         meta, ecbdata->label_path[0], reset, name_a_tab);
642                 fprintf(ecbdata->file, "%s+++ %s%s%s\n",
643                         meta, ecbdata->label_path[1], reset, name_b_tab);
644                 ecbdata->label_path[0] = ecbdata->label_path[1] = NULL;
645         }
646
647         if (diff_suppress_blank_empty
648             && len == 2 && line[0] == ' ' && line[1] == '\n') {
649                 line[0] = '\n';
650                 len = 1;
651         }
652
653         /* This is not really necessary for now because
654          * this codepath only deals with two-way diffs.
655          */
656         for (i = 0; i < len && line[i] == '@'; i++)
657                 ;
658         if (2 <= i && i < len && line[i] == ' ') {
659                 ecbdata->nparents = i - 1;
660                 len = sane_truncate_line(ecbdata, line, len);
661                 emit_line(ecbdata->file,
662                           diff_get_color(ecbdata->color_diff, DIFF_FRAGINFO),
663                           reset, line, len);
664                 if (line[len-1] != '\n')
665                         putc('\n', ecbdata->file);
666                 return;
667         }
668
669         if (len < ecbdata->nparents) {
670                 emit_line(ecbdata->file, reset, reset, line, len);
671                 return;
672         }
673
674         color = DIFF_PLAIN;
675         if (ecbdata->diff_words && ecbdata->nparents != 1)
676                 /* fall back to normal diff */
677                 free_diff_words_data(ecbdata);
678         if (ecbdata->diff_words) {
679                 if (line[0] == '-') {
680                         diff_words_append(line, len,
681                                           &ecbdata->diff_words->minus);
682                         return;
683                 } else if (line[0] == '+') {
684                         diff_words_append(line, len,
685                                           &ecbdata->diff_words->plus);
686                         return;
687                 }
688                 if (ecbdata->diff_words->minus.text.size ||
689                     ecbdata->diff_words->plus.text.size)
690                         diff_words_show(ecbdata->diff_words);
691                 line++;
692                 len--;
693                 emit_line(ecbdata->file, plain, reset, line, len);
694                 return;
695         }
696         for (i = 0; i < ecbdata->nparents && len; i++) {
697                 if (line[i] == '-')
698                         color = DIFF_FILE_OLD;
699                 else if (line[i] == '+')
700                         color = DIFF_FILE_NEW;
701         }
702
703         if (color != DIFF_FILE_NEW) {
704                 emit_line(ecbdata->file,
705                           diff_get_color(ecbdata->color_diff, color),
706                           reset, line, len);
707                 return;
708         }
709         emit_add_line(reset, ecbdata, line, len);
710 }
711
712 static char *pprint_rename(const char *a, const char *b)
713 {
714         const char *old = a;
715         const char *new = b;
716         struct strbuf name = STRBUF_INIT;
717         int pfx_length, sfx_length;
718         int len_a = strlen(a);
719         int len_b = strlen(b);
720         int a_midlen, b_midlen;
721         int qlen_a = quote_c_style(a, NULL, NULL, 0);
722         int qlen_b = quote_c_style(b, NULL, NULL, 0);
723
724         if (qlen_a || qlen_b) {
725                 quote_c_style(a, &name, NULL, 0);
726                 strbuf_addstr(&name, " => ");
727                 quote_c_style(b, &name, NULL, 0);
728                 return strbuf_detach(&name, NULL);
729         }
730
731         /* Find common prefix */
732         pfx_length = 0;
733         while (*old && *new && *old == *new) {
734                 if (*old == '/')
735                         pfx_length = old - a + 1;
736                 old++;
737                 new++;
738         }
739
740         /* Find common suffix */
741         old = a + len_a;
742         new = b + len_b;
743         sfx_length = 0;
744         while (a <= old && b <= new && *old == *new) {
745                 if (*old == '/')
746                         sfx_length = len_a - (old - a);
747                 old--;
748                 new--;
749         }
750
751         /*
752          * pfx{mid-a => mid-b}sfx
753          * {pfx-a => pfx-b}sfx
754          * pfx{sfx-a => sfx-b}
755          * name-a => name-b
756          */
757         a_midlen = len_a - pfx_length - sfx_length;
758         b_midlen = len_b - pfx_length - sfx_length;
759         if (a_midlen < 0)
760                 a_midlen = 0;
761         if (b_midlen < 0)
762                 b_midlen = 0;
763
764         strbuf_grow(&name, pfx_length + a_midlen + b_midlen + sfx_length + 7);
765         if (pfx_length + sfx_length) {
766                 strbuf_add(&name, a, pfx_length);
767                 strbuf_addch(&name, '{');
768         }
769         strbuf_add(&name, a + pfx_length, a_midlen);
770         strbuf_addstr(&name, " => ");
771         strbuf_add(&name, b + pfx_length, b_midlen);
772         if (pfx_length + sfx_length) {
773                 strbuf_addch(&name, '}');
774                 strbuf_add(&name, a + len_a - sfx_length, sfx_length);
775         }
776         return strbuf_detach(&name, NULL);
777 }
778
779 struct diffstat_t {
780         int nr;
781         int alloc;
782         struct diffstat_file {
783                 char *from_name;
784                 char *name;
785                 char *print_name;
786                 unsigned is_unmerged:1;
787                 unsigned is_binary:1;
788                 unsigned is_renamed:1;
789                 unsigned int added, deleted;
790         } **files;
791 };
792
793 static struct diffstat_file *diffstat_add(struct diffstat_t *diffstat,
794                                           const char *name_a,
795                                           const char *name_b)
796 {
797         struct diffstat_file *x;
798         x = xcalloc(sizeof (*x), 1);
799         if (diffstat->nr == diffstat->alloc) {
800                 diffstat->alloc = alloc_nr(diffstat->alloc);
801                 diffstat->files = xrealloc(diffstat->files,
802                                 diffstat->alloc * sizeof(x));
803         }
804         diffstat->files[diffstat->nr++] = x;
805         if (name_b) {
806                 x->from_name = xstrdup(name_a);
807                 x->name = xstrdup(name_b);
808                 x->is_renamed = 1;
809         }
810         else {
811                 x->from_name = NULL;
812                 x->name = xstrdup(name_a);
813         }
814         return x;
815 }
816
817 static void diffstat_consume(void *priv, char *line, unsigned long len)
818 {
819         struct diffstat_t *diffstat = priv;
820         struct diffstat_file *x = diffstat->files[diffstat->nr - 1];
821
822         if (line[0] == '+')
823                 x->added++;
824         else if (line[0] == '-')
825                 x->deleted++;
826 }
827
828 const char mime_boundary_leader[] = "------------";
829
830 static int scale_linear(int it, int width, int max_change)
831 {
832         /*
833          * make sure that at least one '-' is printed if there were deletions,
834          * and likewise for '+'.
835          */
836         if (max_change < 2)
837                 return it;
838         return ((it - 1) * (width - 1) + max_change - 1) / (max_change - 1);
839 }
840
841 static void show_name(FILE *file,
842                       const char *prefix, const char *name, int len,
843                       const char *reset, const char *set)
844 {
845         fprintf(file, " %s%s%-*s%s |", set, prefix, len, name, reset);
846 }
847
848 static void show_graph(FILE *file, char ch, int cnt, const char *set, const char *reset)
849 {
850         if (cnt <= 0)
851                 return;
852         fprintf(file, "%s", set);
853         while (cnt--)
854                 putc(ch, file);
855         fprintf(file, "%s", reset);
856 }
857
858 static void fill_print_name(struct diffstat_file *file)
859 {
860         char *pname;
861
862         if (file->print_name)
863                 return;
864
865         if (!file->is_renamed) {
866                 struct strbuf buf = STRBUF_INIT;
867                 if (quote_c_style(file->name, &buf, NULL, 0)) {
868                         pname = strbuf_detach(&buf, NULL);
869                 } else {
870                         pname = file->name;
871                         strbuf_release(&buf);
872                 }
873         } else {
874                 pname = pprint_rename(file->from_name, file->name);
875         }
876         file->print_name = pname;
877 }
878
879 static void show_stats(struct diffstat_t *data, struct diff_options *options)
880 {
881         int i, len, add, del, adds = 0, dels = 0;
882         int max_change = 0, max_len = 0;
883         int total_files = data->nr;
884         int width, name_width;
885         const char *reset, *set, *add_c, *del_c;
886
887         if (data->nr == 0)
888                 return;
889
890         width = options->stat_width ? options->stat_width : 80;
891         name_width = options->stat_name_width ? options->stat_name_width : 50;
892
893         /* Sanity: give at least 5 columns to the graph,
894          * but leave at least 10 columns for the name.
895          */
896         if (width < 25)
897                 width = 25;
898         if (name_width < 10)
899                 name_width = 10;
900         else if (width < name_width + 15)
901                 name_width = width - 15;
902
903         /* Find the longest filename and max number of changes */
904         reset = diff_get_color_opt(options, DIFF_RESET);
905         set   = diff_get_color_opt(options, DIFF_PLAIN);
906         add_c = diff_get_color_opt(options, DIFF_FILE_NEW);
907         del_c = diff_get_color_opt(options, DIFF_FILE_OLD);
908
909         for (i = 0; i < data->nr; i++) {
910                 struct diffstat_file *file = data->files[i];
911                 int change = file->added + file->deleted;
912                 fill_print_name(file);
913                 len = strlen(file->print_name);
914                 if (max_len < len)
915                         max_len = len;
916
917                 if (file->is_binary || file->is_unmerged)
918                         continue;
919                 if (max_change < change)
920                         max_change = change;
921         }
922
923         /* Compute the width of the graph part;
924          * 10 is for one blank at the beginning of the line plus
925          * " | count " between the name and the graph.
926          *
927          * From here on, name_width is the width of the name area,
928          * and width is the width of the graph area.
929          */
930         name_width = (name_width < max_len) ? name_width : max_len;
931         if (width < (name_width + 10) + max_change)
932                 width = width - (name_width + 10);
933         else
934                 width = max_change;
935
936         for (i = 0; i < data->nr; i++) {
937                 const char *prefix = "";
938                 char *name = data->files[i]->print_name;
939                 int added = data->files[i]->added;
940                 int deleted = data->files[i]->deleted;
941                 int name_len;
942
943                 /*
944                  * "scale" the filename
945                  */
946                 len = name_width;
947                 name_len = strlen(name);
948                 if (name_width < name_len) {
949                         char *slash;
950                         prefix = "...";
951                         len -= 3;
952                         name += name_len - len;
953                         slash = strchr(name, '/');
954                         if (slash)
955                                 name = slash;
956                 }
957
958                 if (data->files[i]->is_binary) {
959                         show_name(options->file, prefix, name, len, reset, set);
960                         fprintf(options->file, "  Bin ");
961                         fprintf(options->file, "%s%d%s", del_c, deleted, reset);
962                         fprintf(options->file, " -> ");
963                         fprintf(options->file, "%s%d%s", add_c, added, reset);
964                         fprintf(options->file, " bytes");
965                         fprintf(options->file, "\n");
966                         continue;
967                 }
968                 else if (data->files[i]->is_unmerged) {
969                         show_name(options->file, prefix, name, len, reset, set);
970                         fprintf(options->file, "  Unmerged\n");
971                         continue;
972                 }
973                 else if (!data->files[i]->is_renamed &&
974                          (added + deleted == 0)) {
975                         total_files--;
976                         continue;
977                 }
978
979                 /*
980                  * scale the add/delete
981                  */
982                 add = added;
983                 del = deleted;
984                 adds += add;
985                 dels += del;
986
987                 if (width <= max_change) {
988                         add = scale_linear(add, width, max_change);
989                         del = scale_linear(del, width, max_change);
990                 }
991                 show_name(options->file, prefix, name, len, reset, set);
992                 fprintf(options->file, "%5d%s", added + deleted,
993                                 added + deleted ? " " : "");
994                 show_graph(options->file, '+', add, add_c, reset);
995                 show_graph(options->file, '-', del, del_c, reset);
996                 fprintf(options->file, "\n");
997         }
998         fprintf(options->file,
999                "%s %d files changed, %d insertions(+), %d deletions(-)%s\n",
1000                set, total_files, adds, dels, reset);
1001 }
1002
1003 static void show_shortstats(struct diffstat_t* data, struct diff_options *options)
1004 {
1005         int i, adds = 0, dels = 0, total_files = data->nr;
1006
1007         if (data->nr == 0)
1008                 return;
1009
1010         for (i = 0; i < data->nr; i++) {
1011                 if (!data->files[i]->is_binary &&
1012                     !data->files[i]->is_unmerged) {
1013                         int added = data->files[i]->added;
1014                         int deleted= data->files[i]->deleted;
1015                         if (!data->files[i]->is_renamed &&
1016                             (added + deleted == 0)) {
1017                                 total_files--;
1018                         } else {
1019                                 adds += added;
1020                                 dels += deleted;
1021                         }
1022                 }
1023         }
1024         fprintf(options->file, " %d files changed, %d insertions(+), %d deletions(-)\n",
1025                total_files, adds, dels);
1026 }
1027
1028 static void show_numstat(struct diffstat_t *data, struct diff_options *options)
1029 {
1030         int i;
1031
1032         if (data->nr == 0)
1033                 return;
1034
1035         for (i = 0; i < data->nr; i++) {
1036                 struct diffstat_file *file = data->files[i];
1037
1038                 if (file->is_binary)
1039                         fprintf(options->file, "-\t-\t");
1040                 else
1041                         fprintf(options->file,
1042                                 "%d\t%d\t", file->added, file->deleted);
1043                 if (options->line_termination) {
1044                         fill_print_name(file);
1045                         if (!file->is_renamed)
1046                                 write_name_quoted(file->name, options->file,
1047                                                   options->line_termination);
1048                         else {
1049                                 fputs(file->print_name, options->file);
1050                                 putc(options->line_termination, options->file);
1051                         }
1052                 } else {
1053                         if (file->is_renamed) {
1054                                 putc('\0', options->file);
1055                                 write_name_quoted(file->from_name, options->file, '\0');
1056                         }
1057                         write_name_quoted(file->name, options->file, '\0');
1058                 }
1059         }
1060 }
1061
1062 struct dirstat_file {
1063         const char *name;
1064         unsigned long changed;
1065 };
1066
1067 struct dirstat_dir {
1068         struct dirstat_file *files;
1069         int alloc, nr, percent, cumulative;
1070 };
1071
1072 static long gather_dirstat(FILE *file, struct dirstat_dir *dir, unsigned long changed, const char *base, int baselen)
1073 {
1074         unsigned long this_dir = 0;
1075         unsigned int sources = 0;
1076
1077         while (dir->nr) {
1078                 struct dirstat_file *f = dir->files;
1079                 int namelen = strlen(f->name);
1080                 unsigned long this;
1081                 char *slash;
1082
1083                 if (namelen < baselen)
1084                         break;
1085                 if (memcmp(f->name, base, baselen))
1086                         break;
1087                 slash = strchr(f->name + baselen, '/');
1088                 if (slash) {
1089                         int newbaselen = slash + 1 - f->name;
1090                         this = gather_dirstat(file, dir, changed, f->name, newbaselen);
1091                         sources++;
1092                 } else {
1093                         this = f->changed;
1094                         dir->files++;
1095                         dir->nr--;
1096                         sources += 2;
1097                 }
1098                 this_dir += this;
1099         }
1100
1101         /*
1102          * We don't report dirstat's for
1103          *  - the top level
1104          *  - or cases where everything came from a single directory
1105          *    under this directory (sources == 1).
1106          */
1107         if (baselen && sources != 1) {
1108                 int permille = this_dir * 1000 / changed;
1109                 if (permille) {
1110                         int percent = permille / 10;
1111                         if (percent >= dir->percent) {
1112                                 fprintf(file, "%4d.%01d%% %.*s\n", percent, permille % 10, baselen, base);
1113                                 if (!dir->cumulative)
1114                                         return 0;
1115                         }
1116                 }
1117         }
1118         return this_dir;
1119 }
1120
1121 static int dirstat_compare(const void *_a, const void *_b)
1122 {
1123         const struct dirstat_file *a = _a;
1124         const struct dirstat_file *b = _b;
1125         return strcmp(a->name, b->name);
1126 }
1127
1128 static void show_dirstat(struct diff_options *options)
1129 {
1130         int i;
1131         unsigned long changed;
1132         struct dirstat_dir dir;
1133         struct diff_queue_struct *q = &diff_queued_diff;
1134
1135         dir.files = NULL;
1136         dir.alloc = 0;
1137         dir.nr = 0;
1138         dir.percent = options->dirstat_percent;
1139         dir.cumulative = DIFF_OPT_TST(options, DIRSTAT_CUMULATIVE);
1140
1141         changed = 0;
1142         for (i = 0; i < q->nr; i++) {
1143                 struct diff_filepair *p = q->queue[i];
1144                 const char *name;
1145                 unsigned long copied, added, damage;
1146
1147                 name = p->one->path ? p->one->path : p->two->path;
1148
1149                 if (DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
1150                         diff_populate_filespec(p->one, 0);
1151                         diff_populate_filespec(p->two, 0);
1152                         diffcore_count_changes(p->one, p->two, NULL, NULL, 0,
1153                                                &copied, &added);
1154                         diff_free_filespec_data(p->one);
1155                         diff_free_filespec_data(p->two);
1156                 } else if (DIFF_FILE_VALID(p->one)) {
1157                         diff_populate_filespec(p->one, 1);
1158                         copied = added = 0;
1159                         diff_free_filespec_data(p->one);
1160                 } else if (DIFF_FILE_VALID(p->two)) {
1161                         diff_populate_filespec(p->two, 1);
1162                         copied = 0;
1163                         added = p->two->size;
1164                         diff_free_filespec_data(p->two);
1165                 } else
1166                         continue;
1167
1168                 /*
1169                  * Original minus copied is the removed material,
1170                  * added is the new material.  They are both damages
1171                  * made to the preimage. In --dirstat-by-file mode, count
1172                  * damaged files, not damaged lines. This is done by
1173                  * counting only a single damaged line per file.
1174                  */
1175                 damage = (p->one->size - copied) + added;
1176                 if (DIFF_OPT_TST(options, DIRSTAT_BY_FILE) && damage > 0)
1177                         damage = 1;
1178
1179                 ALLOC_GROW(dir.files, dir.nr + 1, dir.alloc);
1180                 dir.files[dir.nr].name = name;
1181                 dir.files[dir.nr].changed = damage;
1182                 changed += damage;
1183                 dir.nr++;
1184         }
1185
1186         /* This can happen even with many files, if everything was renames */
1187         if (!changed)
1188                 return;
1189
1190         /* Show all directories with more than x% of the changes */
1191         qsort(dir.files, dir.nr, sizeof(dir.files[0]), dirstat_compare);
1192         gather_dirstat(options->file, &dir, changed, "", 0);
1193 }
1194
1195 static void free_diffstat_info(struct diffstat_t *diffstat)
1196 {
1197         int i;
1198         for (i = 0; i < diffstat->nr; i++) {
1199                 struct diffstat_file *f = diffstat->files[i];
1200                 if (f->name != f->print_name)
1201                         free(f->print_name);
1202                 free(f->name);
1203                 free(f->from_name);
1204                 free(f);
1205         }
1206         free(diffstat->files);
1207 }
1208
1209 struct checkdiff_t {
1210         const char *filename;
1211         int lineno;
1212         struct diff_options *o;
1213         unsigned ws_rule;
1214         unsigned status;
1215         int trailing_blanks_start;
1216 };
1217
1218 static int is_conflict_marker(const char *line, unsigned long len)
1219 {
1220         char firstchar;
1221         int cnt;
1222
1223         if (len < 8)
1224                 return 0;
1225         firstchar = line[0];
1226         switch (firstchar) {
1227         case '=': case '>': case '<':
1228                 break;
1229         default:
1230                 return 0;
1231         }
1232         for (cnt = 1; cnt < 7; cnt++)
1233                 if (line[cnt] != firstchar)
1234                         return 0;
1235         /* line[0] thru line[6] are same as firstchar */
1236         if (firstchar == '=') {
1237                 /* divider between ours and theirs? */
1238                 if (len != 8 || line[7] != '\n')
1239                         return 0;
1240         } else if (len < 8 || !isspace(line[7])) {
1241                 /* not divider before ours nor after theirs */
1242                 return 0;
1243         }
1244         return 1;
1245 }
1246
1247 static void checkdiff_consume(void *priv, char *line, unsigned long len)
1248 {
1249         struct checkdiff_t *data = priv;
1250         int color_diff = DIFF_OPT_TST(data->o, COLOR_DIFF);
1251         const char *ws = diff_get_color(color_diff, DIFF_WHITESPACE);
1252         const char *reset = diff_get_color(color_diff, DIFF_RESET);
1253         const char *set = diff_get_color(color_diff, DIFF_FILE_NEW);
1254         char *err;
1255
1256         if (line[0] == '+') {
1257                 unsigned bad;
1258                 data->lineno++;
1259                 if (!ws_blank_line(line + 1, len - 1, data->ws_rule))
1260                         data->trailing_blanks_start = 0;
1261                 else if (!data->trailing_blanks_start)
1262                         data->trailing_blanks_start = data->lineno;
1263                 if (is_conflict_marker(line + 1, len - 1)) {
1264                         data->status |= 1;
1265                         fprintf(data->o->file,
1266                                 "%s:%d: leftover conflict marker\n",
1267                                 data->filename, data->lineno);
1268                 }
1269                 bad = ws_check(line + 1, len - 1, data->ws_rule);
1270                 if (!bad)
1271                         return;
1272                 data->status |= bad;
1273                 err = whitespace_error_string(bad);
1274                 fprintf(data->o->file, "%s:%d: %s.\n",
1275                         data->filename, data->lineno, err);
1276                 free(err);
1277                 emit_line(data->o->file, set, reset, line, 1);
1278                 ws_check_emit(line + 1, len - 1, data->ws_rule,
1279                               data->o->file, set, reset, ws);
1280         } else if (line[0] == ' ') {
1281                 data->lineno++;
1282                 data->trailing_blanks_start = 0;
1283         } else if (line[0] == '@') {
1284                 char *plus = strchr(line, '+');
1285                 if (plus)
1286                         data->lineno = strtol(plus, NULL, 10) - 1;
1287                 else
1288                         die("invalid diff");
1289                 data->trailing_blanks_start = 0;
1290         }
1291 }
1292
1293 static unsigned char *deflate_it(char *data,
1294                                  unsigned long size,
1295                                  unsigned long *result_size)
1296 {
1297         int bound;
1298         unsigned char *deflated;
1299         z_stream stream;
1300
1301         memset(&stream, 0, sizeof(stream));
1302         deflateInit(&stream, zlib_compression_level);
1303         bound = deflateBound(&stream, size);
1304         deflated = xmalloc(bound);
1305         stream.next_out = deflated;
1306         stream.avail_out = bound;
1307
1308         stream.next_in = (unsigned char *)data;
1309         stream.avail_in = size;
1310         while (deflate(&stream, Z_FINISH) == Z_OK)
1311                 ; /* nothing */
1312         deflateEnd(&stream);
1313         *result_size = stream.total_out;
1314         return deflated;
1315 }
1316
1317 static void emit_binary_diff_body(FILE *file, mmfile_t *one, mmfile_t *two)
1318 {
1319         void *cp;
1320         void *delta;
1321         void *deflated;
1322         void *data;
1323         unsigned long orig_size;
1324         unsigned long delta_size;
1325         unsigned long deflate_size;
1326         unsigned long data_size;
1327
1328         /* We could do deflated delta, or we could do just deflated two,
1329          * whichever is smaller.
1330          */
1331         delta = NULL;
1332         deflated = deflate_it(two->ptr, two->size, &deflate_size);
1333         if (one->size && two->size) {
1334                 delta = diff_delta(one->ptr, one->size,
1335                                    two->ptr, two->size,
1336                                    &delta_size, deflate_size);
1337                 if (delta) {
1338                         void *to_free = delta;
1339                         orig_size = delta_size;
1340                         delta = deflate_it(delta, delta_size, &delta_size);
1341                         free(to_free);
1342                 }
1343         }
1344
1345         if (delta && delta_size < deflate_size) {
1346                 fprintf(file, "delta %lu\n", orig_size);
1347                 free(deflated);
1348                 data = delta;
1349                 data_size = delta_size;
1350         }
1351         else {
1352                 fprintf(file, "literal %lu\n", two->size);
1353                 free(delta);
1354                 data = deflated;
1355                 data_size = deflate_size;
1356         }
1357
1358         /* emit data encoded in base85 */
1359         cp = data;
1360         while (data_size) {
1361                 int bytes = (52 < data_size) ? 52 : data_size;
1362                 char line[70];
1363                 data_size -= bytes;
1364                 if (bytes <= 26)
1365                         line[0] = bytes + 'A' - 1;
1366                 else
1367                         line[0] = bytes - 26 + 'a' - 1;
1368                 encode_85(line + 1, cp, bytes);
1369                 cp = (char *) cp + bytes;
1370                 fputs(line, file);
1371                 fputc('\n', file);
1372         }
1373         fprintf(file, "\n");
1374         free(data);
1375 }
1376
1377 static void emit_binary_diff(FILE *file, mmfile_t *one, mmfile_t *two)
1378 {
1379         fprintf(file, "GIT binary patch\n");
1380         emit_binary_diff_body(file, one, two);
1381         emit_binary_diff_body(file, two, one);
1382 }
1383
1384 static void diff_filespec_load_driver(struct diff_filespec *one)
1385 {
1386         if (!one->driver)
1387                 one->driver = userdiff_find_by_path(one->path);
1388         if (!one->driver)
1389                 one->driver = userdiff_find_by_name("default");
1390 }
1391
1392 int diff_filespec_is_binary(struct diff_filespec *one)
1393 {
1394         if (one->is_binary == -1) {
1395                 diff_filespec_load_driver(one);
1396                 if (one->driver->binary != -1)
1397                         one->is_binary = one->driver->binary;
1398                 else {
1399                         if (!one->data && DIFF_FILE_VALID(one))
1400                                 diff_populate_filespec(one, 0);
1401                         if (one->data)
1402                                 one->is_binary = buffer_is_binary(one->data,
1403                                                 one->size);
1404                         if (one->is_binary == -1)
1405                                 one->is_binary = 0;
1406                 }
1407         }
1408         return one->is_binary;
1409 }
1410
1411 static const struct userdiff_funcname *diff_funcname_pattern(struct diff_filespec *one)
1412 {
1413         diff_filespec_load_driver(one);
1414         return one->driver->funcname.pattern ? &one->driver->funcname : NULL;
1415 }
1416
1417 static const char *userdiff_word_regex(struct diff_filespec *one)
1418 {
1419         diff_filespec_load_driver(one);
1420         return one->driver->word_regex;
1421 }
1422
1423 void diff_set_mnemonic_prefix(struct diff_options *options, const char *a, const char *b)
1424 {
1425         if (!options->a_prefix)
1426                 options->a_prefix = a;
1427         if (!options->b_prefix)
1428                 options->b_prefix = b;
1429 }
1430
1431 static const char *get_textconv(struct diff_filespec *one)
1432 {
1433         if (!DIFF_FILE_VALID(one))
1434                 return NULL;
1435         if (!S_ISREG(one->mode))
1436                 return NULL;
1437         diff_filespec_load_driver(one);
1438         return one->driver->textconv;
1439 }
1440
1441 static void builtin_diff(const char *name_a,
1442                          const char *name_b,
1443                          struct diff_filespec *one,
1444                          struct diff_filespec *two,
1445                          const char *xfrm_msg,
1446                          struct diff_options *o,
1447                          int complete_rewrite)
1448 {
1449         mmfile_t mf1, mf2;
1450         const char *lbl[2];
1451         char *a_one, *b_two;
1452         const char *set = diff_get_color_opt(o, DIFF_METAINFO);
1453         const char *reset = diff_get_color_opt(o, DIFF_RESET);
1454         const char *a_prefix, *b_prefix;
1455         const char *textconv_one = NULL, *textconv_two = NULL;
1456
1457         if (DIFF_OPT_TST(o, ALLOW_TEXTCONV)) {
1458                 textconv_one = get_textconv(one);
1459                 textconv_two = get_textconv(two);
1460         }
1461
1462         diff_set_mnemonic_prefix(o, "a/", "b/");
1463         if (DIFF_OPT_TST(o, REVERSE_DIFF)) {
1464                 a_prefix = o->b_prefix;
1465                 b_prefix = o->a_prefix;
1466         } else {
1467                 a_prefix = o->a_prefix;
1468                 b_prefix = o->b_prefix;
1469         }
1470
1471         /* Never use a non-valid filename anywhere if at all possible */
1472         name_a = DIFF_FILE_VALID(one) ? name_a : name_b;
1473         name_b = DIFF_FILE_VALID(two) ? name_b : name_a;
1474
1475         a_one = quote_two(a_prefix, name_a + (*name_a == '/'));
1476         b_two = quote_two(b_prefix, name_b + (*name_b == '/'));
1477         lbl[0] = DIFF_FILE_VALID(one) ? a_one : "/dev/null";
1478         lbl[1] = DIFF_FILE_VALID(two) ? b_two : "/dev/null";
1479         fprintf(o->file, "%sdiff --git %s %s%s\n", set, a_one, b_two, reset);
1480         if (lbl[0][0] == '/') {
1481                 /* /dev/null */
1482                 fprintf(o->file, "%snew file mode %06o%s\n", set, two->mode, reset);
1483                 if (xfrm_msg && xfrm_msg[0])
1484                         fprintf(o->file, "%s%s%s\n", set, xfrm_msg, reset);
1485         }
1486         else if (lbl[1][0] == '/') {
1487                 fprintf(o->file, "%sdeleted file mode %06o%s\n", set, one->mode, reset);
1488                 if (xfrm_msg && xfrm_msg[0])
1489                         fprintf(o->file, "%s%s%s\n", set, xfrm_msg, reset);
1490         }
1491         else {
1492                 if (one->mode != two->mode) {
1493                         fprintf(o->file, "%sold mode %06o%s\n", set, one->mode, reset);
1494                         fprintf(o->file, "%snew mode %06o%s\n", set, two->mode, reset);
1495                 }
1496                 if (xfrm_msg && xfrm_msg[0])
1497                         fprintf(o->file, "%s%s%s\n", set, xfrm_msg, reset);
1498                 /*
1499                  * we do not run diff between different kind
1500                  * of objects.
1501                  */
1502                 if ((one->mode ^ two->mode) & S_IFMT)
1503                         goto free_ab_and_return;
1504                 if (complete_rewrite &&
1505                     (textconv_one || !diff_filespec_is_binary(one)) &&
1506                     (textconv_two || !diff_filespec_is_binary(two))) {
1507                         emit_rewrite_diff(name_a, name_b, one, two,
1508                                                 textconv_one, textconv_two, o);
1509                         o->found_changes = 1;
1510                         goto free_ab_and_return;
1511                 }
1512         }
1513
1514         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
1515                 die("unable to read files to diff");
1516
1517         if (!DIFF_OPT_TST(o, TEXT) &&
1518             ( (diff_filespec_is_binary(one) && !textconv_one) ||
1519               (diff_filespec_is_binary(two) && !textconv_two) )) {
1520                 /* Quite common confusing case */
1521                 if (mf1.size == mf2.size &&
1522                     !memcmp(mf1.ptr, mf2.ptr, mf1.size))
1523                         goto free_ab_and_return;
1524                 if (DIFF_OPT_TST(o, BINARY))
1525                         emit_binary_diff(o->file, &mf1, &mf2);
1526                 else
1527                         fprintf(o->file, "Binary files %s and %s differ\n",
1528                                 lbl[0], lbl[1]);
1529                 o->found_changes = 1;
1530         }
1531         else {
1532                 /* Crazy xdl interfaces.. */
1533                 const char *diffopts = getenv("GIT_DIFF_OPTS");
1534                 xpparam_t xpp;
1535                 xdemitconf_t xecfg;
1536                 xdemitcb_t ecb;
1537                 struct emit_callback ecbdata;
1538                 const struct userdiff_funcname *pe;
1539
1540                 if (textconv_one) {
1541                         size_t size;
1542                         mf1.ptr = run_textconv(textconv_one, one, &size);
1543                         if (!mf1.ptr)
1544                                 die("unable to read files to diff");
1545                         mf1.size = size;
1546                 }
1547                 if (textconv_two) {
1548                         size_t size;
1549                         mf2.ptr = run_textconv(textconv_two, two, &size);
1550                         if (!mf2.ptr)
1551                                 die("unable to read files to diff");
1552                         mf2.size = size;
1553                 }
1554
1555                 pe = diff_funcname_pattern(one);
1556                 if (!pe)
1557                         pe = diff_funcname_pattern(two);
1558
1559                 memset(&xpp, 0, sizeof(xpp));
1560                 memset(&xecfg, 0, sizeof(xecfg));
1561                 memset(&ecbdata, 0, sizeof(ecbdata));
1562                 ecbdata.label_path = lbl;
1563                 ecbdata.color_diff = DIFF_OPT_TST(o, COLOR_DIFF);
1564                 ecbdata.found_changesp = &o->found_changes;
1565                 ecbdata.ws_rule = whitespace_rule(name_b ? name_b : name_a);
1566                 ecbdata.file = o->file;
1567                 xpp.flags = XDF_NEED_MINIMAL | o->xdl_opts;
1568                 xecfg.ctxlen = o->context;
1569                 xecfg.interhunkctxlen = o->interhunkcontext;
1570                 xecfg.flags = XDL_EMIT_FUNCNAMES;
1571                 if (pe)
1572                         xdiff_set_find_func(&xecfg, pe->pattern, pe->cflags);
1573                 if (!diffopts)
1574                         ;
1575                 else if (!prefixcmp(diffopts, "--unified="))
1576                         xecfg.ctxlen = strtoul(diffopts + 10, NULL, 10);
1577                 else if (!prefixcmp(diffopts, "-u"))
1578                         xecfg.ctxlen = strtoul(diffopts + 2, NULL, 10);
1579                 if (DIFF_OPT_TST(o, COLOR_DIFF_WORDS)) {
1580                         ecbdata.diff_words =
1581                                 xcalloc(1, sizeof(struct diff_words_data));
1582                         ecbdata.diff_words->file = o->file;
1583                         if (!o->word_regex)
1584                                 o->word_regex = userdiff_word_regex(one);
1585                         if (!o->word_regex)
1586                                 o->word_regex = userdiff_word_regex(two);
1587                         if (!o->word_regex)
1588                                 o->word_regex = diff_word_regex_cfg;
1589                         if (o->word_regex) {
1590                                 ecbdata.diff_words->word_regex = (regex_t *)
1591                                         xmalloc(sizeof(regex_t));
1592                                 if (regcomp(ecbdata.diff_words->word_regex,
1593                                                 o->word_regex,
1594                                                 REG_EXTENDED | REG_NEWLINE))
1595                                         die ("Invalid regular expression: %s",
1596                                                         o->word_regex);
1597                         }
1598                 }
1599                 xdi_diff_outf(&mf1, &mf2, fn_out_consume, &ecbdata,
1600                               &xpp, &xecfg, &ecb);
1601                 if (DIFF_OPT_TST(o, COLOR_DIFF_WORDS))
1602                         free_diff_words_data(&ecbdata);
1603                 if (textconv_one)
1604                         free(mf1.ptr);
1605                 if (textconv_two)
1606                         free(mf2.ptr);
1607         }
1608
1609  free_ab_and_return:
1610         diff_free_filespec_data(one);
1611         diff_free_filespec_data(two);
1612         free(a_one);
1613         free(b_two);
1614         return;
1615 }
1616
1617 static void builtin_diffstat(const char *name_a, const char *name_b,
1618                              struct diff_filespec *one,
1619                              struct diff_filespec *two,
1620                              struct diffstat_t *diffstat,
1621                              struct diff_options *o,
1622                              int complete_rewrite)
1623 {
1624         mmfile_t mf1, mf2;
1625         struct diffstat_file *data;
1626
1627         data = diffstat_add(diffstat, name_a, name_b);
1628
1629         if (!one || !two) {
1630                 data->is_unmerged = 1;
1631                 return;
1632         }
1633         if (complete_rewrite) {
1634                 diff_populate_filespec(one, 0);
1635                 diff_populate_filespec(two, 0);
1636                 data->deleted = count_lines(one->data, one->size);
1637                 data->added = count_lines(two->data, two->size);
1638                 goto free_and_return;
1639         }
1640         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
1641                 die("unable to read files to diff");
1642
1643         if (diff_filespec_is_binary(one) || diff_filespec_is_binary(two)) {
1644                 data->is_binary = 1;
1645                 data->added = mf2.size;
1646                 data->deleted = mf1.size;
1647         } else {
1648                 /* Crazy xdl interfaces.. */
1649                 xpparam_t xpp;
1650                 xdemitconf_t xecfg;
1651                 xdemitcb_t ecb;
1652
1653                 memset(&xpp, 0, sizeof(xpp));
1654                 memset(&xecfg, 0, sizeof(xecfg));
1655                 xpp.flags = XDF_NEED_MINIMAL | o->xdl_opts;
1656                 xdi_diff_outf(&mf1, &mf2, diffstat_consume, diffstat,
1657                               &xpp, &xecfg, &ecb);
1658         }
1659
1660  free_and_return:
1661         diff_free_filespec_data(one);
1662         diff_free_filespec_data(two);
1663 }
1664
1665 static void builtin_checkdiff(const char *name_a, const char *name_b,
1666                               const char *attr_path,
1667                               struct diff_filespec *one,
1668                               struct diff_filespec *two,
1669                               struct diff_options *o)
1670 {
1671         mmfile_t mf1, mf2;
1672         struct checkdiff_t data;
1673
1674         if (!two)
1675                 return;
1676
1677         memset(&data, 0, sizeof(data));
1678         data.filename = name_b ? name_b : name_a;
1679         data.lineno = 0;
1680         data.o = o;
1681         data.ws_rule = whitespace_rule(attr_path);
1682
1683         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
1684                 die("unable to read files to diff");
1685
1686         /*
1687          * All the other codepaths check both sides, but not checking
1688          * the "old" side here is deliberate.  We are checking the newly
1689          * introduced changes, and as long as the "new" side is text, we
1690          * can and should check what it introduces.
1691          */
1692         if (diff_filespec_is_binary(two))
1693                 goto free_and_return;
1694         else {
1695                 /* Crazy xdl interfaces.. */
1696                 xpparam_t xpp;
1697                 xdemitconf_t xecfg;
1698                 xdemitcb_t ecb;
1699
1700                 memset(&xpp, 0, sizeof(xpp));
1701                 memset(&xecfg, 0, sizeof(xecfg));
1702                 xecfg.ctxlen = 1; /* at least one context line */
1703                 xpp.flags = XDF_NEED_MINIMAL;
1704                 xdi_diff_outf(&mf1, &mf2, checkdiff_consume, &data,
1705                               &xpp, &xecfg, &ecb);
1706
1707                 if ((data.ws_rule & WS_TRAILING_SPACE) &&
1708                     data.trailing_blanks_start) {
1709                         fprintf(o->file, "%s:%d: ends with blank lines.\n",
1710                                 data.filename, data.trailing_blanks_start);
1711                         data.status = 1; /* report errors */
1712                 }
1713         }
1714  free_and_return:
1715         diff_free_filespec_data(one);
1716         diff_free_filespec_data(two);
1717         if (data.status)
1718                 DIFF_OPT_SET(o, CHECK_FAILED);
1719 }
1720
1721 struct diff_filespec *alloc_filespec(const char *path)
1722 {
1723         int namelen = strlen(path);
1724         struct diff_filespec *spec = xmalloc(sizeof(*spec) + namelen + 1);
1725
1726         memset(spec, 0, sizeof(*spec));
1727         spec->path = (char *)(spec + 1);
1728         memcpy(spec->path, path, namelen+1);
1729         spec->count = 1;
1730         spec->is_binary = -1;
1731         return spec;
1732 }
1733
1734 void free_filespec(struct diff_filespec *spec)
1735 {
1736         if (!--spec->count) {
1737                 diff_free_filespec_data(spec);
1738                 free(spec);
1739         }
1740 }
1741
1742 void fill_filespec(struct diff_filespec *spec, const unsigned char *sha1,
1743                    unsigned short mode)
1744 {
1745         if (mode) {
1746                 spec->mode = canon_mode(mode);
1747                 hashcpy(spec->sha1, sha1);
1748                 spec->sha1_valid = !is_null_sha1(sha1);
1749         }
1750 }
1751
1752 /*
1753  * Given a name and sha1 pair, if the index tells us the file in
1754  * the work tree has that object contents, return true, so that
1755  * prepare_temp_file() does not have to inflate and extract.
1756  */
1757 static int reuse_worktree_file(const char *name, const unsigned char *sha1, int want_file)
1758 {
1759         struct cache_entry *ce;
1760         struct stat st;
1761         int pos, len;
1762
1763         /*
1764          * We do not read the cache ourselves here, because the
1765          * benchmark with my previous version that always reads cache
1766          * shows that it makes things worse for diff-tree comparing
1767          * two linux-2.6 kernel trees in an already checked out work
1768          * tree.  This is because most diff-tree comparisons deal with
1769          * only a small number of files, while reading the cache is
1770          * expensive for a large project, and its cost outweighs the
1771          * savings we get by not inflating the object to a temporary
1772          * file.  Practically, this code only helps when we are used
1773          * by diff-cache --cached, which does read the cache before
1774          * calling us.
1775          */
1776         if (!active_cache)
1777                 return 0;
1778
1779         /* We want to avoid the working directory if our caller
1780          * doesn't need the data in a normal file, this system
1781          * is rather slow with its stat/open/mmap/close syscalls,
1782          * and the object is contained in a pack file.  The pack
1783          * is probably already open and will be faster to obtain
1784          * the data through than the working directory.  Loose
1785          * objects however would tend to be slower as they need
1786          * to be individually opened and inflated.
1787          */
1788         if (!FAST_WORKING_DIRECTORY && !want_file && has_sha1_pack(sha1))
1789                 return 0;
1790
1791         len = strlen(name);
1792         pos = cache_name_pos(name, len);
1793         if (pos < 0)
1794                 return 0;
1795         ce = active_cache[pos];
1796
1797         /*
1798          * This is not the sha1 we are looking for, or
1799          * unreusable because it is not a regular file.
1800          */
1801         if (hashcmp(sha1, ce->sha1) || !S_ISREG(ce->ce_mode))
1802                 return 0;
1803
1804         /*
1805          * If ce is marked as "assume unchanged", there is no
1806          * guarantee that work tree matches what we are looking for.
1807          */
1808         if (ce->ce_flags & CE_VALID)
1809                 return 0;
1810
1811         /*
1812          * If ce matches the file in the work tree, we can reuse it.
1813          */
1814         if (ce_uptodate(ce) ||
1815             (!lstat(name, &st) && !ce_match_stat(ce, &st, 0)))
1816                 return 1;
1817
1818         return 0;
1819 }
1820
1821 static int populate_from_stdin(struct diff_filespec *s)
1822 {
1823         struct strbuf buf = STRBUF_INIT;
1824         size_t size = 0;
1825
1826         if (strbuf_read(&buf, 0, 0) < 0)
1827                 return error("error while reading from stdin %s",
1828                                      strerror(errno));
1829
1830         s->should_munmap = 0;
1831         s->data = strbuf_detach(&buf, &size);
1832         s->size = size;
1833         s->should_free = 1;
1834         return 0;
1835 }
1836
1837 static int diff_populate_gitlink(struct diff_filespec *s, int size_only)
1838 {
1839         int len;
1840         char *data = xmalloc(100);
1841         len = snprintf(data, 100,
1842                 "Subproject commit %s\n", sha1_to_hex(s->sha1));
1843         s->data = data;
1844         s->size = len;
1845         s->should_free = 1;
1846         if (size_only) {
1847                 s->data = NULL;
1848                 free(data);
1849         }
1850         return 0;
1851 }
1852
1853 /*
1854  * While doing rename detection and pickaxe operation, we may need to
1855  * grab the data for the blob (or file) for our own in-core comparison.
1856  * diff_filespec has data and size fields for this purpose.
1857  */
1858 int diff_populate_filespec(struct diff_filespec *s, int size_only)
1859 {
1860         int err = 0;
1861         if (!DIFF_FILE_VALID(s))
1862                 die("internal error: asking to populate invalid file.");
1863         if (S_ISDIR(s->mode))
1864                 return -1;
1865
1866         if (s->data)
1867                 return 0;
1868
1869         if (size_only && 0 < s->size)
1870                 return 0;
1871
1872         if (S_ISGITLINK(s->mode))
1873                 return diff_populate_gitlink(s, size_only);
1874
1875         if (!s->sha1_valid ||
1876             reuse_worktree_file(s->path, s->sha1, 0)) {
1877                 struct strbuf buf = STRBUF_INIT;
1878                 struct stat st;
1879                 int fd;
1880
1881                 if (!strcmp(s->path, "-"))
1882                         return populate_from_stdin(s);
1883
1884                 if (lstat(s->path, &st) < 0) {
1885                         if (errno == ENOENT) {
1886                         err_empty:
1887                                 err = -1;
1888                         empty:
1889                                 s->data = (char *)"";
1890                                 s->size = 0;
1891                                 return err;
1892                         }
1893                 }
1894                 s->size = xsize_t(st.st_size);
1895                 if (!s->size)
1896                         goto empty;
1897                 if (S_ISLNK(st.st_mode)) {
1898                         struct strbuf sb = STRBUF_INIT;
1899
1900                         if (strbuf_readlink(&sb, s->path, s->size))
1901                                 goto err_empty;
1902                         s->size = sb.len;
1903                         s->data = strbuf_detach(&sb, NULL);
1904                         s->should_free = 1;
1905                         return 0;
1906                 }
1907                 if (size_only)
1908                         return 0;
1909                 fd = open(s->path, O_RDONLY);
1910                 if (fd < 0)
1911                         goto err_empty;
1912                 s->data = xmmap(NULL, s->size, PROT_READ, MAP_PRIVATE, fd, 0);
1913                 close(fd);
1914                 s->should_munmap = 1;
1915
1916                 /*
1917                  * Convert from working tree format to canonical git format
1918                  */
1919                 if (convert_to_git(s->path, s->data, s->size, &buf, safe_crlf)) {
1920                         size_t size = 0;
1921                         munmap(s->data, s->size);
1922                         s->should_munmap = 0;
1923                         s->data = strbuf_detach(&buf, &size);
1924                         s->size = size;
1925                         s->should_free = 1;
1926                 }
1927         }
1928         else {
1929                 enum object_type type;
1930                 if (size_only)
1931                         type = sha1_object_info(s->sha1, &s->size);
1932                 else {
1933                         s->data = read_sha1_file(s->sha1, &type, &s->size);
1934                         s->should_free = 1;
1935                 }
1936         }
1937         return 0;
1938 }
1939
1940 void diff_free_filespec_blob(struct diff_filespec *s)
1941 {
1942         if (s->should_free)
1943                 free(s->data);
1944         else if (s->should_munmap)
1945                 munmap(s->data, s->size);
1946
1947         if (s->should_free || s->should_munmap) {
1948                 s->should_free = s->should_munmap = 0;
1949                 s->data = NULL;
1950         }
1951 }
1952
1953 void diff_free_filespec_data(struct diff_filespec *s)
1954 {
1955         diff_free_filespec_blob(s);
1956         free(s->cnt_data);
1957         s->cnt_data = NULL;
1958 }
1959
1960 static void prep_temp_blob(const char *path, struct diff_tempfile *temp,
1961                            void *blob,
1962                            unsigned long size,
1963                            const unsigned char *sha1,
1964                            int mode)
1965 {
1966         int fd;
1967         struct strbuf buf = STRBUF_INIT;
1968
1969         fd = git_mkstemp(temp->tmp_path, PATH_MAX, ".diff_XXXXXX");
1970         if (fd < 0)
1971                 die("unable to create temp-file: %s", strerror(errno));
1972         if (convert_to_working_tree(path,
1973                         (const char *)blob, (size_t)size, &buf)) {
1974                 blob = buf.buf;
1975                 size = buf.len;
1976         }
1977         if (write_in_full(fd, blob, size) != size)
1978                 die("unable to write temp-file");
1979         close(fd);
1980         temp->name = temp->tmp_path;
1981         strcpy(temp->hex, sha1_to_hex(sha1));
1982         temp->hex[40] = 0;
1983         sprintf(temp->mode, "%06o", mode);
1984         strbuf_release(&buf);
1985 }
1986
1987 static struct diff_tempfile *prepare_temp_file(const char *name,
1988                 struct diff_filespec *one)
1989 {
1990         struct diff_tempfile *temp = claim_diff_tempfile();
1991
1992         if (!DIFF_FILE_VALID(one)) {
1993         not_a_valid_file:
1994                 /* A '-' entry produces this for file-2, and
1995                  * a '+' entry produces this for file-1.
1996                  */
1997                 temp->name = "/dev/null";
1998                 strcpy(temp->hex, ".");
1999                 strcpy(temp->mode, ".");
2000                 return temp;
2001         }
2002
2003         if (!remove_tempfile_installed) {
2004                 atexit(remove_tempfile);
2005                 sigchain_push_common(remove_tempfile_on_signal);
2006                 remove_tempfile_installed = 1;
2007         }
2008
2009         if (!one->sha1_valid ||
2010             reuse_worktree_file(name, one->sha1, 1)) {
2011                 struct stat st;
2012                 if (lstat(name, &st) < 0) {
2013                         if (errno == ENOENT)
2014                                 goto not_a_valid_file;
2015                         die("stat(%s): %s", name, strerror(errno));
2016                 }
2017                 if (S_ISLNK(st.st_mode)) {
2018                         int ret;
2019                         char buf[PATH_MAX + 1]; /* ought to be SYMLINK_MAX */
2020                         ret = readlink(name, buf, sizeof(buf));
2021                         if (ret < 0)
2022                                 die("readlink(%s)", name);
2023                         if (ret == sizeof(buf))
2024                                 die("symlink too long: %s", name);
2025                         prep_temp_blob(name, temp, buf, ret,
2026                                        (one->sha1_valid ?
2027                                         one->sha1 : null_sha1),
2028                                        (one->sha1_valid ?
2029                                         one->mode : S_IFLNK));
2030                 }
2031                 else {
2032                         /* we can borrow from the file in the work tree */
2033                         temp->name = name;
2034                         if (!one->sha1_valid)
2035                                 strcpy(temp->hex, sha1_to_hex(null_sha1));
2036                         else
2037                                 strcpy(temp->hex, sha1_to_hex(one->sha1));
2038                         /* Even though we may sometimes borrow the
2039                          * contents from the work tree, we always want
2040                          * one->mode.  mode is trustworthy even when
2041                          * !(one->sha1_valid), as long as
2042                          * DIFF_FILE_VALID(one).
2043                          */
2044                         sprintf(temp->mode, "%06o", one->mode);
2045                 }
2046                 return temp;
2047         }
2048         else {
2049                 if (diff_populate_filespec(one, 0))
2050                         die("cannot read data blob for %s", one->path);
2051                 prep_temp_blob(name, temp, one->data, one->size,
2052                                one->sha1, one->mode);
2053         }
2054         return temp;
2055 }
2056
2057 /* An external diff command takes:
2058  *
2059  * diff-cmd name infile1 infile1-sha1 infile1-mode \
2060  *               infile2 infile2-sha1 infile2-mode [ rename-to ]
2061  *
2062  */
2063 static void run_external_diff(const char *pgm,
2064                               const char *name,
2065                               const char *other,
2066                               struct diff_filespec *one,
2067                               struct diff_filespec *two,
2068                               const char *xfrm_msg,
2069                               int complete_rewrite)
2070 {
2071         const char *spawn_arg[10];
2072         int retval;
2073         const char **arg = &spawn_arg[0];
2074
2075         if (one && two) {
2076                 struct diff_tempfile *temp_one, *temp_two;
2077                 const char *othername = (other ? other : name);
2078                 temp_one = prepare_temp_file(name, one);
2079                 temp_two = prepare_temp_file(othername, two);
2080                 *arg++ = pgm;
2081                 *arg++ = name;
2082                 *arg++ = temp_one->name;
2083                 *arg++ = temp_one->hex;
2084                 *arg++ = temp_one->mode;
2085                 *arg++ = temp_two->name;
2086                 *arg++ = temp_two->hex;
2087                 *arg++ = temp_two->mode;
2088                 if (other) {
2089                         *arg++ = other;
2090                         *arg++ = xfrm_msg;
2091                 }
2092         } else {
2093                 *arg++ = pgm;
2094                 *arg++ = name;
2095         }
2096         *arg = NULL;
2097         fflush(NULL);
2098         retval = run_command_v_opt(spawn_arg, 0);
2099         remove_tempfile();
2100         if (retval) {
2101                 fprintf(stderr, "external diff died, stopping at %s.\n", name);
2102                 exit(1);
2103         }
2104 }
2105
2106 static int similarity_index(struct diff_filepair *p)
2107 {
2108         return p->score * 100 / MAX_SCORE;
2109 }
2110
2111 static void fill_metainfo(struct strbuf *msg,
2112                           const char *name,
2113                           const char *other,
2114                           struct diff_filespec *one,
2115                           struct diff_filespec *two,
2116                           struct diff_options *o,
2117                           struct diff_filepair *p)
2118 {
2119         strbuf_init(msg, PATH_MAX * 2 + 300);
2120         switch (p->status) {
2121         case DIFF_STATUS_COPIED:
2122                 strbuf_addf(msg, "similarity index %d%%", similarity_index(p));
2123                 strbuf_addstr(msg, "\ncopy from ");
2124                 quote_c_style(name, msg, NULL, 0);
2125                 strbuf_addstr(msg, "\ncopy to ");
2126                 quote_c_style(other, msg, NULL, 0);
2127                 strbuf_addch(msg, '\n');
2128                 break;
2129         case DIFF_STATUS_RENAMED:
2130                 strbuf_addf(msg, "similarity index %d%%", similarity_index(p));
2131                 strbuf_addstr(msg, "\nrename from ");
2132                 quote_c_style(name, msg, NULL, 0);
2133                 strbuf_addstr(msg, "\nrename to ");
2134                 quote_c_style(other, msg, NULL, 0);
2135                 strbuf_addch(msg, '\n');
2136                 break;
2137         case DIFF_STATUS_MODIFIED:
2138                 if (p->score) {
2139                         strbuf_addf(msg, "dissimilarity index %d%%\n",
2140                                     similarity_index(p));
2141                         break;
2142                 }
2143                 /* fallthru */
2144         default:
2145                 /* nothing */
2146                 ;
2147         }
2148         if (one && two && hashcmp(one->sha1, two->sha1)) {
2149                 int abbrev = DIFF_OPT_TST(o, FULL_INDEX) ? 40 : DEFAULT_ABBREV;
2150
2151                 if (DIFF_OPT_TST(o, BINARY)) {
2152                         mmfile_t mf;
2153                         if ((!fill_mmfile(&mf, one) && diff_filespec_is_binary(one)) ||
2154                             (!fill_mmfile(&mf, two) && diff_filespec_is_binary(two)))
2155                                 abbrev = 40;
2156                 }
2157                 strbuf_addf(msg, "index %.*s..%.*s",
2158                             abbrev, sha1_to_hex(one->sha1),
2159                             abbrev, sha1_to_hex(two->sha1));
2160                 if (one->mode == two->mode)
2161                         strbuf_addf(msg, " %06o", one->mode);
2162                 strbuf_addch(msg, '\n');
2163         }
2164         if (msg->len)
2165                 strbuf_setlen(msg, msg->len - 1);
2166 }
2167
2168 static void run_diff_cmd(const char *pgm,
2169                          const char *name,
2170                          const char *other,
2171                          const char *attr_path,
2172                          struct diff_filespec *one,
2173                          struct diff_filespec *two,
2174                          struct strbuf *msg,
2175                          struct diff_options *o,
2176                          struct diff_filepair *p)
2177 {
2178         const char *xfrm_msg = NULL;
2179         int complete_rewrite = (p->status == DIFF_STATUS_MODIFIED) && p->score;
2180
2181         if (msg) {
2182                 fill_metainfo(msg, name, other, one, two, o, p);
2183                 xfrm_msg = msg->len ? msg->buf : NULL;
2184         }
2185
2186         if (!DIFF_OPT_TST(o, ALLOW_EXTERNAL))
2187                 pgm = NULL;
2188         else {
2189                 struct userdiff_driver *drv = userdiff_find_by_path(attr_path);
2190                 if (drv && drv->external)
2191                         pgm = drv->external;
2192         }
2193
2194         if (pgm) {
2195                 run_external_diff(pgm, name, other, one, two, xfrm_msg,
2196                                   complete_rewrite);
2197                 return;
2198         }
2199         if (one && two)
2200                 builtin_diff(name, other ? other : name,
2201                              one, two, xfrm_msg, o, complete_rewrite);
2202         else
2203                 fprintf(o->file, "* Unmerged path %s\n", name);
2204 }
2205
2206 static void diff_fill_sha1_info(struct diff_filespec *one)
2207 {
2208         if (DIFF_FILE_VALID(one)) {
2209                 if (!one->sha1_valid) {
2210                         struct stat st;
2211                         if (!strcmp(one->path, "-")) {
2212                                 hashcpy(one->sha1, null_sha1);
2213                                 return;
2214                         }
2215                         if (lstat(one->path, &st) < 0)
2216                                 die("stat %s", one->path);
2217                         if (index_path(one->sha1, one->path, &st, 0))
2218                                 die("cannot hash %s", one->path);
2219                 }
2220         }
2221         else
2222                 hashclr(one->sha1);
2223 }
2224
2225 static void strip_prefix(int prefix_length, const char **namep, const char **otherp)
2226 {
2227         /* Strip the prefix but do not molest /dev/null and absolute paths */
2228         if (*namep && **namep != '/')
2229                 *namep += prefix_length;
2230         if (*otherp && **otherp != '/')
2231                 *otherp += prefix_length;
2232 }
2233
2234 static void run_diff(struct diff_filepair *p, struct diff_options *o)
2235 {
2236         const char *pgm = external_diff();
2237         struct strbuf msg;
2238         struct diff_filespec *one = p->one;
2239         struct diff_filespec *two = p->two;
2240         const char *name;
2241         const char *other;
2242         const char *attr_path;
2243
2244         name  = p->one->path;
2245         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
2246         attr_path = name;
2247         if (o->prefix_length)
2248                 strip_prefix(o->prefix_length, &name, &other);
2249
2250         if (DIFF_PAIR_UNMERGED(p)) {
2251                 run_diff_cmd(pgm, name, NULL, attr_path,
2252                              NULL, NULL, NULL, o, p);
2253                 return;
2254         }
2255
2256         diff_fill_sha1_info(one);
2257         diff_fill_sha1_info(two);
2258
2259         if (!pgm &&
2260             DIFF_FILE_VALID(one) && DIFF_FILE_VALID(two) &&
2261             (S_IFMT & one->mode) != (S_IFMT & two->mode)) {
2262                 /*
2263                  * a filepair that changes between file and symlink
2264                  * needs to be split into deletion and creation.
2265                  */
2266                 struct diff_filespec *null = alloc_filespec(two->path);
2267                 run_diff_cmd(NULL, name, other, attr_path,
2268                              one, null, &msg, o, p);
2269                 free(null);
2270                 strbuf_release(&msg);
2271
2272                 null = alloc_filespec(one->path);
2273                 run_diff_cmd(NULL, name, other, attr_path,
2274                              null, two, &msg, o, p);
2275                 free(null);
2276         }
2277         else
2278                 run_diff_cmd(pgm, name, other, attr_path,
2279                              one, two, &msg, o, p);
2280
2281         strbuf_release(&msg);
2282 }
2283
2284 static void run_diffstat(struct diff_filepair *p, struct diff_options *o,
2285                          struct diffstat_t *diffstat)
2286 {
2287         const char *name;
2288         const char *other;
2289         int complete_rewrite = 0;
2290
2291         if (DIFF_PAIR_UNMERGED(p)) {
2292                 /* unmerged */
2293                 builtin_diffstat(p->one->path, NULL, NULL, NULL, diffstat, o, 0);
2294                 return;
2295         }
2296
2297         name = p->one->path;
2298         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
2299
2300         if (o->prefix_length)
2301                 strip_prefix(o->prefix_length, &name, &other);
2302
2303         diff_fill_sha1_info(p->one);
2304         diff_fill_sha1_info(p->two);
2305
2306         if (p->status == DIFF_STATUS_MODIFIED && p->score)
2307                 complete_rewrite = 1;
2308         builtin_diffstat(name, other, p->one, p->two, diffstat, o, complete_rewrite);
2309 }
2310
2311 static void run_checkdiff(struct diff_filepair *p, struct diff_options *o)
2312 {
2313         const char *name;
2314         const char *other;
2315         const char *attr_path;
2316
2317         if (DIFF_PAIR_UNMERGED(p)) {
2318                 /* unmerged */
2319                 return;
2320         }
2321
2322         name = p->one->path;
2323         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
2324         attr_path = other ? other : name;
2325
2326         if (o->prefix_length)
2327                 strip_prefix(o->prefix_length, &name, &other);
2328
2329         diff_fill_sha1_info(p->one);
2330         diff_fill_sha1_info(p->two);
2331
2332         builtin_checkdiff(name, other, attr_path, p->one, p->two, o);
2333 }
2334
2335 void diff_setup(struct diff_options *options)
2336 {
2337         memset(options, 0, sizeof(*options));
2338
2339         options->file = stdout;
2340
2341         options->line_termination = '\n';
2342         options->break_opt = -1;
2343         options->rename_limit = -1;
2344         options->dirstat_percent = 3;
2345         options->context = 3;
2346
2347         options->change = diff_change;
2348         options->add_remove = diff_addremove;
2349         if (diff_use_color_default > 0)
2350                 DIFF_OPT_SET(options, COLOR_DIFF);
2351         options->detect_rename = diff_detect_rename_default;
2352
2353         if (!diff_mnemonic_prefix) {
2354                 options->a_prefix = "a/";
2355                 options->b_prefix = "b/";
2356         }
2357 }
2358
2359 int diff_setup_done(struct diff_options *options)
2360 {
2361         int count = 0;
2362
2363         if (options->output_format & DIFF_FORMAT_NAME)
2364                 count++;
2365         if (options->output_format & DIFF_FORMAT_NAME_STATUS)
2366                 count++;
2367         if (options->output_format & DIFF_FORMAT_CHECKDIFF)
2368                 count++;
2369         if (options->output_format & DIFF_FORMAT_NO_OUTPUT)
2370                 count++;
2371         if (count > 1)
2372                 die("--name-only, --name-status, --check and -s are mutually exclusive");
2373
2374         if (DIFF_OPT_TST(options, FIND_COPIES_HARDER))
2375                 options->detect_rename = DIFF_DETECT_COPY;
2376
2377         if (!DIFF_OPT_TST(options, RELATIVE_NAME))
2378                 options->prefix = NULL;
2379         if (options->prefix)
2380                 options->prefix_length = strlen(options->prefix);
2381         else
2382                 options->prefix_length = 0;
2383
2384         if (options->output_format & (DIFF_FORMAT_NAME |
2385                                       DIFF_FORMAT_NAME_STATUS |
2386                                       DIFF_FORMAT_CHECKDIFF |
2387                                       DIFF_FORMAT_NO_OUTPUT))
2388                 options->output_format &= ~(DIFF_FORMAT_RAW |
2389                                             DIFF_FORMAT_NUMSTAT |
2390                                             DIFF_FORMAT_DIFFSTAT |
2391                                             DIFF_FORMAT_SHORTSTAT |
2392                                             DIFF_FORMAT_DIRSTAT |
2393                                             DIFF_FORMAT_SUMMARY |
2394                                             DIFF_FORMAT_PATCH);
2395
2396         /*
2397          * These cases always need recursive; we do not drop caller-supplied
2398          * recursive bits for other formats here.
2399          */
2400         if (options->output_format & (DIFF_FORMAT_PATCH |
2401                                       DIFF_FORMAT_NUMSTAT |
2402                                       DIFF_FORMAT_DIFFSTAT |
2403                                       DIFF_FORMAT_SHORTSTAT |
2404                                       DIFF_FORMAT_DIRSTAT |
2405                                       DIFF_FORMAT_SUMMARY |
2406                                       DIFF_FORMAT_CHECKDIFF))
2407                 DIFF_OPT_SET(options, RECURSIVE);
2408         /*
2409          * Also pickaxe would not work very well if you do not say recursive
2410          */
2411         if (options->pickaxe)
2412                 DIFF_OPT_SET(options, RECURSIVE);
2413
2414         if (options->detect_rename && options->rename_limit < 0)
2415                 options->rename_limit = diff_rename_limit_default;
2416         if (options->setup & DIFF_SETUP_USE_CACHE) {
2417                 if (!active_cache)
2418                         /* read-cache does not die even when it fails
2419                          * so it is safe for us to do this here.  Also
2420                          * it does not smudge active_cache or active_nr
2421                          * when it fails, so we do not have to worry about
2422                          * cleaning it up ourselves either.
2423                          */
2424                         read_cache();
2425         }
2426         if (options->abbrev <= 0 || 40 < options->abbrev)
2427                 options->abbrev = 40; /* full */
2428
2429         /*
2430          * It does not make sense to show the first hit we happened
2431          * to have found.  It does not make sense not to return with
2432          * exit code in such a case either.
2433          */
2434         if (DIFF_OPT_TST(options, QUIET)) {
2435                 options->output_format = DIFF_FORMAT_NO_OUTPUT;
2436                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
2437         }
2438
2439         return 0;
2440 }
2441
2442 static int opt_arg(const char *arg, int arg_short, const char *arg_long, int *val)
2443 {
2444         char c, *eq;
2445         int len;
2446
2447         if (*arg != '-')
2448                 return 0;
2449         c = *++arg;
2450         if (!c)
2451                 return 0;
2452         if (c == arg_short) {
2453                 c = *++arg;
2454                 if (!c)
2455                         return 1;
2456                 if (val && isdigit(c)) {
2457                         char *end;
2458                         int n = strtoul(arg, &end, 10);
2459                         if (*end)
2460                                 return 0;
2461                         *val = n;
2462                         return 1;
2463                 }
2464                 return 0;
2465         }
2466         if (c != '-')
2467                 return 0;
2468         arg++;
2469         eq = strchr(arg, '=');
2470         if (eq)
2471                 len = eq - arg;
2472         else
2473                 len = strlen(arg);
2474         if (!len || strncmp(arg, arg_long, len))
2475                 return 0;
2476         if (eq) {
2477                 int n;
2478                 char *end;
2479                 if (!isdigit(*++eq))
2480                         return 0;
2481                 n = strtoul(eq, &end, 10);
2482                 if (*end)
2483                         return 0;
2484                 *val = n;
2485         }
2486         return 1;
2487 }
2488
2489 static int diff_scoreopt_parse(const char *opt);
2490
2491 int diff_opt_parse(struct diff_options *options, const char **av, int ac)
2492 {
2493         const char *arg = av[0];
2494
2495         /* Output format options */
2496         if (!strcmp(arg, "-p") || !strcmp(arg, "-u"))
2497                 options->output_format |= DIFF_FORMAT_PATCH;
2498         else if (opt_arg(arg, 'U', "unified", &options->context))
2499                 options->output_format |= DIFF_FORMAT_PATCH;
2500         else if (!strcmp(arg, "--raw"))
2501                 options->output_format |= DIFF_FORMAT_RAW;
2502         else if (!strcmp(arg, "--patch-with-raw"))
2503                 options->output_format |= DIFF_FORMAT_PATCH | DIFF_FORMAT_RAW;
2504         else if (!strcmp(arg, "--numstat"))
2505                 options->output_format |= DIFF_FORMAT_NUMSTAT;
2506         else if (!strcmp(arg, "--shortstat"))
2507                 options->output_format |= DIFF_FORMAT_SHORTSTAT;
2508         else if (opt_arg(arg, 'X', "dirstat", &options->dirstat_percent))
2509                 options->output_format |= DIFF_FORMAT_DIRSTAT;
2510         else if (!strcmp(arg, "--cumulative")) {
2511                 options->output_format |= DIFF_FORMAT_DIRSTAT;
2512                 DIFF_OPT_SET(options, DIRSTAT_CUMULATIVE);
2513         } else if (opt_arg(arg, 0, "dirstat-by-file",
2514                            &options->dirstat_percent)) {
2515                 options->output_format |= DIFF_FORMAT_DIRSTAT;
2516                 DIFF_OPT_SET(options, DIRSTAT_BY_FILE);
2517         }
2518         else if (!strcmp(arg, "--check"))
2519                 options->output_format |= DIFF_FORMAT_CHECKDIFF;
2520         else if (!strcmp(arg, "--summary"))
2521                 options->output_format |= DIFF_FORMAT_SUMMARY;
2522         else if (!strcmp(arg, "--patch-with-stat"))
2523                 options->output_format |= DIFF_FORMAT_PATCH | DIFF_FORMAT_DIFFSTAT;
2524         else if (!strcmp(arg, "--name-only"))
2525                 options->output_format |= DIFF_FORMAT_NAME;
2526         else if (!strcmp(arg, "--name-status"))
2527                 options->output_format |= DIFF_FORMAT_NAME_STATUS;
2528         else if (!strcmp(arg, "-s"))
2529                 options->output_format |= DIFF_FORMAT_NO_OUTPUT;
2530         else if (!prefixcmp(arg, "--stat")) {
2531                 char *end;
2532                 int width = options->stat_width;
2533                 int name_width = options->stat_name_width;
2534                 arg += 6;
2535                 end = (char *)arg;
2536
2537                 switch (*arg) {
2538                 case '-':
2539                         if (!prefixcmp(arg, "-width="))
2540                                 width = strtoul(arg + 7, &end, 10);
2541                         else if (!prefixcmp(arg, "-name-width="))
2542                                 name_width = strtoul(arg + 12, &end, 10);
2543                         break;
2544                 case '=':
2545                         width = strtoul(arg+1, &end, 10);
2546                         if (*end == ',')
2547                                 name_width = strtoul(end+1, &end, 10);
2548                 }
2549
2550                 /* Important! This checks all the error cases! */
2551                 if (*end)
2552                         return 0;
2553                 options->output_format |= DIFF_FORMAT_DIFFSTAT;
2554                 options->stat_name_width = name_width;
2555                 options->stat_width = width;
2556         }
2557
2558         /* renames options */
2559         else if (!prefixcmp(arg, "-B")) {
2560                 if ((options->break_opt = diff_scoreopt_parse(arg)) == -1)
2561                         return -1;
2562         }
2563         else if (!prefixcmp(arg, "-M")) {
2564                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
2565                         return -1;
2566                 options->detect_rename = DIFF_DETECT_RENAME;
2567         }
2568         else if (!prefixcmp(arg, "-C")) {
2569                 if (options->detect_rename == DIFF_DETECT_COPY)
2570                         DIFF_OPT_SET(options, FIND_COPIES_HARDER);
2571                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
2572                         return -1;
2573                 options->detect_rename = DIFF_DETECT_COPY;
2574         }
2575         else if (!strcmp(arg, "--no-renames"))
2576                 options->detect_rename = 0;
2577         else if (!strcmp(arg, "--relative"))
2578                 DIFF_OPT_SET(options, RELATIVE_NAME);
2579         else if (!prefixcmp(arg, "--relative=")) {
2580                 DIFF_OPT_SET(options, RELATIVE_NAME);
2581                 options->prefix = arg + 11;
2582         }
2583
2584         /* xdiff options */
2585         else if (!strcmp(arg, "-w") || !strcmp(arg, "--ignore-all-space"))
2586                 DIFF_XDL_SET(options, IGNORE_WHITESPACE);
2587         else if (!strcmp(arg, "-b") || !strcmp(arg, "--ignore-space-change"))
2588                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_CHANGE);
2589         else if (!strcmp(arg, "--ignore-space-at-eol"))
2590                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_AT_EOL);
2591         else if (!strcmp(arg, "--patience"))
2592                 DIFF_XDL_SET(options, PATIENCE_DIFF);
2593
2594         /* flags options */
2595         else if (!strcmp(arg, "--binary")) {
2596                 options->output_format |= DIFF_FORMAT_PATCH;
2597                 DIFF_OPT_SET(options, BINARY);
2598         }
2599         else if (!strcmp(arg, "--full-index"))
2600                 DIFF_OPT_SET(options, FULL_INDEX);
2601         else if (!strcmp(arg, "-a") || !strcmp(arg, "--text"))
2602                 DIFF_OPT_SET(options, TEXT);
2603         else if (!strcmp(arg, "-R"))
2604                 DIFF_OPT_SET(options, REVERSE_DIFF);
2605         else if (!strcmp(arg, "--find-copies-harder"))
2606                 DIFF_OPT_SET(options, FIND_COPIES_HARDER);
2607         else if (!strcmp(arg, "--follow"))
2608                 DIFF_OPT_SET(options, FOLLOW_RENAMES);
2609         else if (!strcmp(arg, "--color"))
2610                 DIFF_OPT_SET(options, COLOR_DIFF);
2611         else if (!strcmp(arg, "--no-color"))
2612                 DIFF_OPT_CLR(options, COLOR_DIFF);
2613         else if (!strcmp(arg, "--color-words")) {
2614                 DIFF_OPT_SET(options, COLOR_DIFF);
2615                 DIFF_OPT_SET(options, COLOR_DIFF_WORDS);
2616         }
2617         else if (!prefixcmp(arg, "--color-words=")) {
2618                 DIFF_OPT_SET(options, COLOR_DIFF);
2619                 DIFF_OPT_SET(options, COLOR_DIFF_WORDS);
2620                 options->word_regex = arg + 14;
2621         }
2622         else if (!strcmp(arg, "--exit-code"))
2623                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
2624         else if (!strcmp(arg, "--quiet"))
2625                 DIFF_OPT_SET(options, QUIET);
2626         else if (!strcmp(arg, "--ext-diff"))
2627                 DIFF_OPT_SET(options, ALLOW_EXTERNAL);
2628         else if (!strcmp(arg, "--no-ext-diff"))
2629                 DIFF_OPT_CLR(options, ALLOW_EXTERNAL);
2630         else if (!strcmp(arg, "--textconv"))
2631                 DIFF_OPT_SET(options, ALLOW_TEXTCONV);
2632         else if (!strcmp(arg, "--no-textconv"))
2633                 DIFF_OPT_CLR(options, ALLOW_TEXTCONV);
2634         else if (!strcmp(arg, "--ignore-submodules"))
2635                 DIFF_OPT_SET(options, IGNORE_SUBMODULES);
2636
2637         /* misc options */
2638         else if (!strcmp(arg, "-z"))
2639                 options->line_termination = 0;
2640         else if (!prefixcmp(arg, "-l"))
2641                 options->rename_limit = strtoul(arg+2, NULL, 10);
2642         else if (!prefixcmp(arg, "-S"))
2643                 options->pickaxe = arg + 2;
2644         else if (!strcmp(arg, "--pickaxe-all"))
2645                 options->pickaxe_opts = DIFF_PICKAXE_ALL;
2646         else if (!strcmp(arg, "--pickaxe-regex"))
2647                 options->pickaxe_opts = DIFF_PICKAXE_REGEX;
2648         else if (!prefixcmp(arg, "-O"))
2649                 options->orderfile = arg + 2;
2650         else if (!prefixcmp(arg, "--diff-filter="))
2651                 options->filter = arg + 14;
2652         else if (!strcmp(arg, "--abbrev"))
2653                 options->abbrev = DEFAULT_ABBREV;
2654         else if (!prefixcmp(arg, "--abbrev=")) {
2655                 options->abbrev = strtoul(arg + 9, NULL, 10);
2656                 if (options->abbrev < MINIMUM_ABBREV)
2657                         options->abbrev = MINIMUM_ABBREV;
2658                 else if (40 < options->abbrev)
2659                         options->abbrev = 40;
2660         }
2661         else if (!prefixcmp(arg, "--src-prefix="))
2662                 options->a_prefix = arg + 13;
2663         else if (!prefixcmp(arg, "--dst-prefix="))
2664                 options->b_prefix = arg + 13;
2665         else if (!strcmp(arg, "--no-prefix"))
2666                 options->a_prefix = options->b_prefix = "";
2667         else if (opt_arg(arg, '\0', "inter-hunk-context",
2668                          &options->interhunkcontext))
2669                 ;
2670         else if (!prefixcmp(arg, "--output=")) {
2671                 options->file = fopen(arg + strlen("--output="), "w");
2672                 options->close_file = 1;
2673         } else
2674                 return 0;
2675         return 1;
2676 }
2677
2678 static int parse_num(const char **cp_p)
2679 {
2680         unsigned long num, scale;
2681         int ch, dot;
2682         const char *cp = *cp_p;
2683
2684         num = 0;
2685         scale = 1;
2686         dot = 0;
2687         for(;;) {
2688                 ch = *cp;
2689                 if ( !dot && ch == '.' ) {
2690                         scale = 1;
2691                         dot = 1;
2692                 } else if ( ch == '%' ) {
2693                         scale = dot ? scale*100 : 100;
2694                         cp++;   /* % is always at the end */
2695                         break;
2696                 } else if ( ch >= '0' && ch <= '9' ) {
2697                         if ( scale < 100000 ) {
2698                                 scale *= 10;
2699                                 num = (num*10) + (ch-'0');
2700                         }
2701                 } else {
2702                         break;
2703                 }
2704                 cp++;
2705         }
2706         *cp_p = cp;
2707
2708         /* user says num divided by scale and we say internally that
2709          * is MAX_SCORE * num / scale.
2710          */
2711         return (int)((num >= scale) ? MAX_SCORE : (MAX_SCORE * num / scale));
2712 }
2713
2714 static int diff_scoreopt_parse(const char *opt)
2715 {
2716         int opt1, opt2, cmd;
2717
2718         if (*opt++ != '-')
2719                 return -1;
2720         cmd = *opt++;
2721         if (cmd != 'M' && cmd != 'C' && cmd != 'B')
2722                 return -1; /* that is not a -M, -C nor -B option */
2723
2724         opt1 = parse_num(&opt);
2725         if (cmd != 'B')
2726                 opt2 = 0;
2727         else {
2728                 if (*opt == 0)
2729                         opt2 = 0;
2730                 else if (*opt != '/')
2731                         return -1; /* we expect -B80/99 or -B80 */
2732                 else {
2733                         opt++;
2734                         opt2 = parse_num(&opt);
2735                 }
2736         }
2737         if (*opt != 0)
2738                 return -1;
2739         return opt1 | (opt2 << 16);
2740 }
2741
2742 struct diff_queue_struct diff_queued_diff;
2743
2744 void diff_q(struct diff_queue_struct *queue, struct diff_filepair *dp)
2745 {
2746         if (queue->alloc <= queue->nr) {
2747                 queue->alloc = alloc_nr(queue->alloc);
2748                 queue->queue = xrealloc(queue->queue,
2749                                         sizeof(dp) * queue->alloc);
2750         }
2751         queue->queue[queue->nr++] = dp;
2752 }
2753
2754 struct diff_filepair *diff_queue(struct diff_queue_struct *queue,
2755                                  struct diff_filespec *one,
2756                                  struct diff_filespec *two)
2757 {
2758         struct diff_filepair *dp = xcalloc(1, sizeof(*dp));
2759         dp->one = one;
2760         dp->two = two;
2761         if (queue)
2762                 diff_q(queue, dp);
2763         return dp;
2764 }
2765
2766 void diff_free_filepair(struct diff_filepair *p)
2767 {
2768         free_filespec(p->one);
2769         free_filespec(p->two);
2770         free(p);
2771 }
2772
2773 /* This is different from find_unique_abbrev() in that
2774  * it stuffs the result with dots for alignment.
2775  */
2776 const char *diff_unique_abbrev(const unsigned char *sha1, int len)
2777 {
2778         int abblen;
2779         const char *abbrev;
2780         if (len == 40)
2781                 return sha1_to_hex(sha1);
2782
2783         abbrev = find_unique_abbrev(sha1, len);
2784         abblen = strlen(abbrev);
2785         if (abblen < 37) {
2786                 static char hex[41];
2787                 if (len < abblen && abblen <= len + 2)
2788                         sprintf(hex, "%s%.*s", abbrev, len+3-abblen, "..");
2789                 else
2790                         sprintf(hex, "%s...", abbrev);
2791                 return hex;
2792         }
2793         return sha1_to_hex(sha1);
2794 }
2795
2796 static void diff_flush_raw(struct diff_filepair *p, struct diff_options *opt)
2797 {
2798         int line_termination = opt->line_termination;
2799         int inter_name_termination = line_termination ? '\t' : '\0';
2800
2801         if (!(opt->output_format & DIFF_FORMAT_NAME_STATUS)) {
2802                 fprintf(opt->file, ":%06o %06o %s ", p->one->mode, p->two->mode,
2803                         diff_unique_abbrev(p->one->sha1, opt->abbrev));
2804                 fprintf(opt->file, "%s ", diff_unique_abbrev(p->two->sha1, opt->abbrev));
2805         }
2806         if (p->score) {
2807                 fprintf(opt->file, "%c%03d%c", p->status, similarity_index(p),
2808                         inter_name_termination);
2809         } else {
2810                 fprintf(opt->file, "%c%c", p->status, inter_name_termination);
2811         }
2812
2813         if (p->status == DIFF_STATUS_COPIED ||
2814             p->status == DIFF_STATUS_RENAMED) {
2815                 const char *name_a, *name_b;
2816                 name_a = p->one->path;
2817                 name_b = p->two->path;
2818                 strip_prefix(opt->prefix_length, &name_a, &name_b);
2819                 write_name_quoted(name_a, opt->file, inter_name_termination);
2820                 write_name_quoted(name_b, opt->file, line_termination);
2821         } else {
2822                 const char *name_a, *name_b;
2823                 name_a = p->one->mode ? p->one->path : p->two->path;
2824                 name_b = NULL;
2825                 strip_prefix(opt->prefix_length, &name_a, &name_b);
2826                 write_name_quoted(name_a, opt->file, line_termination);
2827         }
2828 }
2829
2830 int diff_unmodified_pair(struct diff_filepair *p)
2831 {
2832         /* This function is written stricter than necessary to support
2833          * the currently implemented transformers, but the idea is to
2834          * let transformers to produce diff_filepairs any way they want,
2835          * and filter and clean them up here before producing the output.
2836          */
2837         struct diff_filespec *one = p->one, *two = p->two;
2838
2839         if (DIFF_PAIR_UNMERGED(p))
2840                 return 0; /* unmerged is interesting */
2841
2842         /* deletion, addition, mode or type change
2843          * and rename are all interesting.
2844          */
2845         if (DIFF_FILE_VALID(one) != DIFF_FILE_VALID(two) ||
2846             DIFF_PAIR_MODE_CHANGED(p) ||
2847             strcmp(one->path, two->path))
2848                 return 0;
2849
2850         /* both are valid and point at the same path.  that is, we are
2851          * dealing with a change.
2852          */
2853         if (one->sha1_valid && two->sha1_valid &&
2854             !hashcmp(one->sha1, two->sha1))
2855                 return 1; /* no change */
2856         if (!one->sha1_valid && !two->sha1_valid)
2857                 return 1; /* both look at the same file on the filesystem. */
2858         return 0;
2859 }
2860
2861 static void diff_flush_patch(struct diff_filepair *p, struct diff_options *o)
2862 {
2863         if (diff_unmodified_pair(p))
2864                 return;
2865
2866         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
2867             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
2868                 return; /* no tree diffs in patch format */
2869
2870         run_diff(p, o);
2871 }
2872
2873 static void diff_flush_stat(struct diff_filepair *p, struct diff_options *o,
2874                             struct diffstat_t *diffstat)
2875 {
2876         if (diff_unmodified_pair(p))
2877                 return;
2878
2879         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
2880             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
2881                 return; /* no tree diffs in patch format */
2882
2883         run_diffstat(p, o, diffstat);
2884 }
2885
2886 static void diff_flush_checkdiff(struct diff_filepair *p,
2887                 struct diff_options *o)
2888 {
2889         if (diff_unmodified_pair(p))
2890                 return;
2891
2892         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
2893             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
2894                 return; /* no tree diffs in patch format */
2895
2896         run_checkdiff(p, o);
2897 }
2898
2899 int diff_queue_is_empty(void)
2900 {
2901         struct diff_queue_struct *q = &diff_queued_diff;
2902         int i;
2903         for (i = 0; i < q->nr; i++)
2904                 if (!diff_unmodified_pair(q->queue[i]))
2905                         return 0;
2906         return 1;
2907 }
2908
2909 #if DIFF_DEBUG
2910 void diff_debug_filespec(struct diff_filespec *s, int x, const char *one)
2911 {
2912         fprintf(stderr, "queue[%d] %s (%s) %s %06o %s\n",
2913                 x, one ? one : "",
2914                 s->path,
2915                 DIFF_FILE_VALID(s) ? "valid" : "invalid",
2916                 s->mode,
2917                 s->sha1_valid ? sha1_to_hex(s->sha1) : "");
2918         fprintf(stderr, "queue[%d] %s size %lu flags %d\n",
2919                 x, one ? one : "",
2920                 s->size, s->xfrm_flags);
2921 }
2922
2923 void diff_debug_filepair(const struct diff_filepair *p, int i)
2924 {
2925         diff_debug_filespec(p->one, i, "one");
2926         diff_debug_filespec(p->two, i, "two");
2927         fprintf(stderr, "score %d, status %c rename_used %d broken %d\n",
2928                 p->score, p->status ? p->status : '?',
2929                 p->one->rename_used, p->broken_pair);
2930 }
2931
2932 void diff_debug_queue(const char *msg, struct diff_queue_struct *q)
2933 {
2934         int i;
2935         if (msg)
2936                 fprintf(stderr, "%s\n", msg);
2937         fprintf(stderr, "q->nr = %d\n", q->nr);
2938         for (i = 0; i < q->nr; i++) {
2939                 struct diff_filepair *p = q->queue[i];
2940                 diff_debug_filepair(p, i);
2941         }
2942 }
2943 #endif
2944
2945 static void diff_resolve_rename_copy(void)
2946 {
2947         int i;
2948         struct diff_filepair *p;
2949         struct diff_queue_struct *q = &diff_queued_diff;
2950
2951         diff_debug_queue("resolve-rename-copy", q);
2952
2953         for (i = 0; i < q->nr; i++) {
2954                 p = q->queue[i];
2955                 p->status = 0; /* undecided */
2956                 if (DIFF_PAIR_UNMERGED(p))
2957                         p->status = DIFF_STATUS_UNMERGED;
2958                 else if (!DIFF_FILE_VALID(p->one))
2959                         p->status = DIFF_STATUS_ADDED;
2960                 else if (!DIFF_FILE_VALID(p->two))
2961                         p->status = DIFF_STATUS_DELETED;
2962                 else if (DIFF_PAIR_TYPE_CHANGED(p))
2963                         p->status = DIFF_STATUS_TYPE_CHANGED;
2964
2965                 /* from this point on, we are dealing with a pair
2966                  * whose both sides are valid and of the same type, i.e.
2967                  * either in-place edit or rename/copy edit.
2968                  */
2969                 else if (DIFF_PAIR_RENAME(p)) {
2970                         /*
2971                          * A rename might have re-connected a broken
2972                          * pair up, causing the pathnames to be the
2973                          * same again. If so, that's not a rename at
2974                          * all, just a modification..
2975                          *
2976                          * Otherwise, see if this source was used for
2977                          * multiple renames, in which case we decrement
2978                          * the count, and call it a copy.
2979                          */
2980                         if (!strcmp(p->one->path, p->two->path))
2981                                 p->status = DIFF_STATUS_MODIFIED;
2982                         else if (--p->one->rename_used > 0)
2983                                 p->status = DIFF_STATUS_COPIED;
2984                         else
2985                                 p->status = DIFF_STATUS_RENAMED;
2986                 }
2987                 else if (hashcmp(p->one->sha1, p->two->sha1) ||
2988                          p->one->mode != p->two->mode ||
2989                          is_null_sha1(p->one->sha1))
2990                         p->status = DIFF_STATUS_MODIFIED;
2991                 else {
2992                         /* This is a "no-change" entry and should not
2993                          * happen anymore, but prepare for broken callers.
2994                          */
2995                         error("feeding unmodified %s to diffcore",
2996                               p->one->path);
2997                         p->status = DIFF_STATUS_UNKNOWN;
2998                 }
2999         }
3000         diff_debug_queue("resolve-rename-copy done", q);
3001 }
3002
3003 static int check_pair_status(struct diff_filepair *p)
3004 {
3005         switch (p->status) {
3006         case DIFF_STATUS_UNKNOWN:
3007                 return 0;
3008         case 0:
3009                 die("internal error in diff-resolve-rename-copy");
3010         default:
3011                 return 1;
3012         }
3013 }
3014
3015 static void flush_one_pair(struct diff_filepair *p, struct diff_options *opt)
3016 {
3017         int fmt = opt->output_format;
3018
3019         if (fmt & DIFF_FORMAT_CHECKDIFF)
3020                 diff_flush_checkdiff(p, opt);
3021         else if (fmt & (DIFF_FORMAT_RAW | DIFF_FORMAT_NAME_STATUS))
3022                 diff_flush_raw(p, opt);
3023         else if (fmt & DIFF_FORMAT_NAME) {
3024                 const char *name_a, *name_b;
3025                 name_a = p->two->path;
3026                 name_b = NULL;
3027                 strip_prefix(opt->prefix_length, &name_a, &name_b);
3028                 write_name_quoted(name_a, opt->file, opt->line_termination);
3029         }
3030 }
3031
3032 static void show_file_mode_name(FILE *file, const char *newdelete, struct diff_filespec *fs)
3033 {
3034         if (fs->mode)
3035                 fprintf(file, " %s mode %06o ", newdelete, fs->mode);
3036         else
3037                 fprintf(file, " %s ", newdelete);
3038         write_name_quoted(fs->path, file, '\n');
3039 }
3040
3041
3042 static void show_mode_change(FILE *file, struct diff_filepair *p, int show_name)
3043 {
3044         if (p->one->mode && p->two->mode && p->one->mode != p->two->mode) {
3045                 fprintf(file, " mode change %06o => %06o%c", p->one->mode, p->two->mode,
3046                         show_name ? ' ' : '\n');
3047                 if (show_name) {
3048                         write_name_quoted(p->two->path, file, '\n');
3049                 }
3050         }
3051 }
3052
3053 static void show_rename_copy(FILE *file, const char *renamecopy, struct diff_filepair *p)
3054 {
3055         char *names = pprint_rename(p->one->path, p->two->path);
3056
3057         fprintf(file, " %s %s (%d%%)\n", renamecopy, names, similarity_index(p));
3058         free(names);
3059         show_mode_change(file, p, 0);
3060 }
3061
3062 static void diff_summary(FILE *file, struct diff_filepair *p)
3063 {
3064         switch(p->status) {
3065         case DIFF_STATUS_DELETED:
3066                 show_file_mode_name(file, "delete", p->one);
3067                 break;
3068         case DIFF_STATUS_ADDED:
3069                 show_file_mode_name(file, "create", p->two);
3070                 break;
3071         case DIFF_STATUS_COPIED:
3072                 show_rename_copy(file, "copy", p);
3073                 break;
3074         case DIFF_STATUS_RENAMED:
3075                 show_rename_copy(file, "rename", p);
3076                 break;
3077         default:
3078                 if (p->score) {
3079                         fputs(" rewrite ", file);
3080                         write_name_quoted(p->two->path, file, ' ');
3081                         fprintf(file, "(%d%%)\n", similarity_index(p));
3082                 }
3083                 show_mode_change(file, p, !p->score);
3084                 break;
3085         }
3086 }
3087
3088 struct patch_id_t {
3089         git_SHA_CTX *ctx;
3090         int patchlen;
3091 };
3092
3093 static int remove_space(char *line, int len)
3094 {
3095         int i;
3096         char *dst = line;
3097         unsigned char c;
3098
3099         for (i = 0; i < len; i++)
3100                 if (!isspace((c = line[i])))
3101                         *dst++ = c;
3102
3103         return dst - line;
3104 }
3105
3106 static void patch_id_consume(void *priv, char *line, unsigned long len)
3107 {
3108         struct patch_id_t *data = priv;
3109         int new_len;
3110
3111         /* Ignore line numbers when computing the SHA1 of the patch */
3112         if (!prefixcmp(line, "@@ -"))
3113                 return;
3114
3115         new_len = remove_space(line, len);
3116
3117         git_SHA1_Update(data->ctx, line, new_len);
3118         data->patchlen += new_len;
3119 }
3120
3121 /* returns 0 upon success, and writes result into sha1 */
3122 static int diff_get_patch_id(struct diff_options *options, unsigned char *sha1)
3123 {
3124         struct diff_queue_struct *q = &diff_queued_diff;
3125         int i;
3126         git_SHA_CTX ctx;
3127         struct patch_id_t data;
3128         char buffer[PATH_MAX * 4 + 20];
3129
3130         git_SHA1_Init(&ctx);
3131         memset(&data, 0, sizeof(struct patch_id_t));
3132         data.ctx = &ctx;
3133
3134         for (i = 0; i < q->nr; i++) {
3135                 xpparam_t xpp;
3136                 xdemitconf_t xecfg;
3137                 xdemitcb_t ecb;
3138                 mmfile_t mf1, mf2;
3139                 struct diff_filepair *p = q->queue[i];
3140                 int len1, len2;
3141
3142                 memset(&xpp, 0, sizeof(xpp));
3143                 memset(&xecfg, 0, sizeof(xecfg));
3144                 if (p->status == 0)
3145                         return error("internal diff status error");
3146                 if (p->status == DIFF_STATUS_UNKNOWN)
3147                         continue;
3148                 if (diff_unmodified_pair(p))
3149                         continue;
3150                 if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
3151                     (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
3152                         continue;
3153                 if (DIFF_PAIR_UNMERGED(p))
3154                         continue;
3155
3156                 diff_fill_sha1_info(p->one);
3157                 diff_fill_sha1_info(p->two);
3158                 if (fill_mmfile(&mf1, p->one) < 0 ||
3159                                 fill_mmfile(&mf2, p->two) < 0)
3160                         return error("unable to read files to diff");
3161
3162                 len1 = remove_space(p->one->path, strlen(p->one->path));
3163                 len2 = remove_space(p->two->path, strlen(p->two->path));
3164                 if (p->one->mode == 0)
3165                         len1 = snprintf(buffer, sizeof(buffer),
3166                                         "diff--gita/%.*sb/%.*s"
3167                                         "newfilemode%06o"
3168                                         "---/dev/null"
3169                                         "+++b/%.*s",
3170                                         len1, p->one->path,
3171                                         len2, p->two->path,
3172                                         p->two->mode,
3173                                         len2, p->two->path);
3174                 else if (p->two->mode == 0)
3175                         len1 = snprintf(buffer, sizeof(buffer),
3176                                         "diff--gita/%.*sb/%.*s"
3177                                         "deletedfilemode%06o"
3178                                         "---a/%.*s"
3179                                         "+++/dev/null",
3180                                         len1, p->one->path,
3181                                         len2, p->two->path,
3182                                         p->one->mode,
3183                                         len1, p->one->path);
3184                 else
3185                         len1 = snprintf(buffer, sizeof(buffer),
3186                                         "diff--gita/%.*sb/%.*s"
3187                                         "---a/%.*s"
3188                                         "+++b/%.*s",
3189                                         len1, p->one->path,
3190                                         len2, p->two->path,
3191                                         len1, p->one->path,
3192                                         len2, p->two->path);
3193                 git_SHA1_Update(&ctx, buffer, len1);
3194
3195                 xpp.flags = XDF_NEED_MINIMAL;
3196                 xecfg.ctxlen = 3;
3197                 xecfg.flags = XDL_EMIT_FUNCNAMES;
3198                 xdi_diff_outf(&mf1, &mf2, patch_id_consume, &data,
3199                               &xpp, &xecfg, &ecb);
3200         }
3201
3202         git_SHA1_Final(sha1, &ctx);
3203         return 0;
3204 }
3205
3206 int diff_flush_patch_id(struct diff_options *options, unsigned char *sha1)
3207 {
3208         struct diff_queue_struct *q = &diff_queued_diff;
3209         int i;
3210         int result = diff_get_patch_id(options, sha1);
3211
3212         for (i = 0; i < q->nr; i++)
3213                 diff_free_filepair(q->queue[i]);
3214
3215         free(q->queue);
3216         q->queue = NULL;
3217         q->nr = q->alloc = 0;
3218
3219         return result;
3220 }
3221
3222 static int is_summary_empty(const struct diff_queue_struct *q)
3223 {
3224         int i;
3225
3226         for (i = 0; i < q->nr; i++) {
3227                 const struct diff_filepair *p = q->queue[i];
3228
3229                 switch (p->status) {
3230                 case DIFF_STATUS_DELETED:
3231                 case DIFF_STATUS_ADDED:
3232                 case DIFF_STATUS_COPIED:
3233                 case DIFF_STATUS_RENAMED:
3234                         return 0;
3235                 default:
3236                         if (p->score)
3237                                 return 0;
3238                         if (p->one->mode && p->two->mode &&
3239                             p->one->mode != p->two->mode)
3240                                 return 0;
3241                         break;
3242                 }
3243         }
3244         return 1;
3245 }
3246
3247 void diff_flush(struct diff_options *options)
3248 {
3249         struct diff_queue_struct *q = &diff_queued_diff;
3250         int i, output_format = options->output_format;
3251         int separator = 0;
3252
3253         /*
3254          * Order: raw, stat, summary, patch
3255          * or:    name/name-status/checkdiff (other bits clear)
3256          */
3257         if (!q->nr)
3258                 goto free_queue;
3259
3260         if (output_format & (DIFF_FORMAT_RAW |
3261                              DIFF_FORMAT_NAME |
3262                              DIFF_FORMAT_NAME_STATUS |
3263                              DIFF_FORMAT_CHECKDIFF)) {
3264                 for (i = 0; i < q->nr; i++) {
3265                         struct diff_filepair *p = q->queue[i];
3266                         if (check_pair_status(p))
3267                                 flush_one_pair(p, options);
3268                 }
3269                 separator++;
3270         }
3271
3272         if (output_format & (DIFF_FORMAT_DIFFSTAT|DIFF_FORMAT_SHORTSTAT|DIFF_FORMAT_NUMSTAT)) {
3273                 struct diffstat_t diffstat;
3274
3275                 memset(&diffstat, 0, sizeof(struct diffstat_t));
3276                 for (i = 0; i < q->nr; i++) {
3277                         struct diff_filepair *p = q->queue[i];
3278                         if (check_pair_status(p))
3279                                 diff_flush_stat(p, options, &diffstat);
3280                 }
3281                 if (output_format & DIFF_FORMAT_NUMSTAT)
3282                         show_numstat(&diffstat, options);
3283                 if (output_format & DIFF_FORMAT_DIFFSTAT)
3284                         show_stats(&diffstat, options);
3285                 if (output_format & DIFF_FORMAT_SHORTSTAT)
3286                         show_shortstats(&diffstat, options);
3287                 free_diffstat_info(&diffstat);
3288                 separator++;
3289         }
3290         if (output_format & DIFF_FORMAT_DIRSTAT)
3291                 show_dirstat(options);
3292
3293         if (output_format & DIFF_FORMAT_SUMMARY && !is_summary_empty(q)) {
3294                 for (i = 0; i < q->nr; i++)
3295                         diff_summary(options->file, q->queue[i]);
3296                 separator++;
3297         }
3298
3299         if (output_format & DIFF_FORMAT_PATCH) {
3300                 if (separator) {
3301                         putc(options->line_termination, options->file);
3302                         if (options->stat_sep) {
3303                                 /* attach patch instead of inline */
3304                                 fputs(options->stat_sep, options->file);
3305                         }
3306                 }
3307
3308                 for (i = 0; i < q->nr; i++) {
3309                         struct diff_filepair *p = q->queue[i];
3310                         if (check_pair_status(p))
3311                                 diff_flush_patch(p, options);
3312                 }
3313         }
3314
3315         if (output_format & DIFF_FORMAT_CALLBACK)
3316                 options->format_callback(q, options, options->format_callback_data);
3317
3318         for (i = 0; i < q->nr; i++)
3319                 diff_free_filepair(q->queue[i]);
3320 free_queue:
3321         free(q->queue);
3322         q->queue = NULL;
3323         q->nr = q->alloc = 0;
3324         if (options->close_file)
3325                 fclose(options->file);
3326 }
3327
3328 static void diffcore_apply_filter(const char *filter)
3329 {
3330         int i;
3331         struct diff_queue_struct *q = &diff_queued_diff;
3332         struct diff_queue_struct outq;
3333         outq.queue = NULL;
3334         outq.nr = outq.alloc = 0;
3335
3336         if (!filter)
3337                 return;
3338
3339         if (strchr(filter, DIFF_STATUS_FILTER_AON)) {
3340                 int found;
3341                 for (i = found = 0; !found && i < q->nr; i++) {
3342                         struct diff_filepair *p = q->queue[i];
3343                         if (((p->status == DIFF_STATUS_MODIFIED) &&
3344                              ((p->score &&
3345                                strchr(filter, DIFF_STATUS_FILTER_BROKEN)) ||
3346                               (!p->score &&
3347                                strchr(filter, DIFF_STATUS_MODIFIED)))) ||
3348                             ((p->status != DIFF_STATUS_MODIFIED) &&
3349                              strchr(filter, p->status)))
3350                                 found++;
3351                 }
3352                 if (found)
3353                         return;
3354
3355                 /* otherwise we will clear the whole queue
3356                  * by copying the empty outq at the end of this
3357                  * function, but first clear the current entries
3358                  * in the queue.
3359                  */
3360                 for (i = 0; i < q->nr; i++)
3361                         diff_free_filepair(q->queue[i]);
3362         }
3363         else {
3364                 /* Only the matching ones */
3365                 for (i = 0; i < q->nr; i++) {
3366                         struct diff_filepair *p = q->queue[i];
3367
3368                         if (((p->status == DIFF_STATUS_MODIFIED) &&
3369                              ((p->score &&
3370                                strchr(filter, DIFF_STATUS_FILTER_BROKEN)) ||
3371                               (!p->score &&
3372                                strchr(filter, DIFF_STATUS_MODIFIED)))) ||
3373                             ((p->status != DIFF_STATUS_MODIFIED) &&
3374                              strchr(filter, p->status)))
3375                                 diff_q(&outq, p);
3376                         else
3377                                 diff_free_filepair(p);
3378                 }
3379         }
3380         free(q->queue);
3381         *q = outq;
3382 }
3383
3384 /* Check whether two filespecs with the same mode and size are identical */
3385 static int diff_filespec_is_identical(struct diff_filespec *one,
3386                                       struct diff_filespec *two)
3387 {
3388         if (S_ISGITLINK(one->mode))
3389                 return 0;
3390         if (diff_populate_filespec(one, 0))
3391                 return 0;
3392         if (diff_populate_filespec(two, 0))
3393                 return 0;
3394         return !memcmp(one->data, two->data, one->size);
3395 }
3396
3397 static void diffcore_skip_stat_unmatch(struct diff_options *diffopt)
3398 {
3399         int i;
3400         struct diff_queue_struct *q = &diff_queued_diff;
3401         struct diff_queue_struct outq;
3402         outq.queue = NULL;
3403         outq.nr = outq.alloc = 0;
3404
3405         for (i = 0; i < q->nr; i++) {
3406                 struct diff_filepair *p = q->queue[i];
3407
3408                 /*
3409                  * 1. Entries that come from stat info dirtyness
3410                  *    always have both sides (iow, not create/delete),
3411                  *    one side of the object name is unknown, with
3412                  *    the same mode and size.  Keep the ones that
3413                  *    do not match these criteria.  They have real
3414                  *    differences.
3415                  *
3416                  * 2. At this point, the file is known to be modified,
3417                  *    with the same mode and size, and the object
3418                  *    name of one side is unknown.  Need to inspect
3419                  *    the identical contents.
3420                  */
3421                 if (!DIFF_FILE_VALID(p->one) || /* (1) */
3422                     !DIFF_FILE_VALID(p->two) ||
3423                     (p->one->sha1_valid && p->two->sha1_valid) ||
3424                     (p->one->mode != p->two->mode) ||
3425                     diff_populate_filespec(p->one, 1) ||
3426                     diff_populate_filespec(p->two, 1) ||
3427                     (p->one->size != p->two->size) ||
3428                     !diff_filespec_is_identical(p->one, p->two)) /* (2) */
3429                         diff_q(&outq, p);
3430                 else {
3431                         /*
3432                          * The caller can subtract 1 from skip_stat_unmatch
3433                          * to determine how many paths were dirty only
3434                          * due to stat info mismatch.
3435                          */
3436                         if (!DIFF_OPT_TST(diffopt, NO_INDEX))
3437                                 diffopt->skip_stat_unmatch++;
3438                         diff_free_filepair(p);
3439                 }
3440         }
3441         free(q->queue);
3442         *q = outq;
3443 }
3444
3445 void diffcore_std(struct diff_options *options)
3446 {
3447         if (options->skip_stat_unmatch)
3448                 diffcore_skip_stat_unmatch(options);
3449         if (options->break_opt != -1)
3450                 diffcore_break(options->break_opt);
3451         if (options->detect_rename)
3452                 diffcore_rename(options);
3453         if (options->break_opt != -1)
3454                 diffcore_merge_broken();
3455         if (options->pickaxe)
3456                 diffcore_pickaxe(options->pickaxe, options->pickaxe_opts);
3457         if (options->orderfile)
3458                 diffcore_order(options->orderfile);
3459         diff_resolve_rename_copy();
3460         diffcore_apply_filter(options->filter);
3461
3462         if (diff_queued_diff.nr)
3463                 DIFF_OPT_SET(options, HAS_CHANGES);
3464         else
3465                 DIFF_OPT_CLR(options, HAS_CHANGES);
3466 }
3467
3468 int diff_result_code(struct diff_options *opt, int status)
3469 {
3470         int result = 0;
3471         if (!DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
3472             !(opt->output_format & DIFF_FORMAT_CHECKDIFF))
3473                 return status;
3474         if (DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
3475             DIFF_OPT_TST(opt, HAS_CHANGES))
3476                 result |= 01;
3477         if ((opt->output_format & DIFF_FORMAT_CHECKDIFF) &&
3478             DIFF_OPT_TST(opt, CHECK_FAILED))
3479                 result |= 02;
3480         return result;
3481 }
3482
3483 void diff_addremove(struct diff_options *options,
3484                     int addremove, unsigned mode,
3485                     const unsigned char *sha1,
3486                     const char *concatpath)
3487 {
3488         struct diff_filespec *one, *two;
3489
3490         if (DIFF_OPT_TST(options, IGNORE_SUBMODULES) && S_ISGITLINK(mode))
3491                 return;
3492
3493         /* This may look odd, but it is a preparation for
3494          * feeding "there are unchanged files which should
3495          * not produce diffs, but when you are doing copy
3496          * detection you would need them, so here they are"
3497          * entries to the diff-core.  They will be prefixed
3498          * with something like '=' or '*' (I haven't decided
3499          * which but should not make any difference).
3500          * Feeding the same new and old to diff_change()
3501          * also has the same effect.
3502          * Before the final output happens, they are pruned after
3503          * merged into rename/copy pairs as appropriate.
3504          */
3505         if (DIFF_OPT_TST(options, REVERSE_DIFF))
3506                 addremove = (addremove == '+' ? '-' :
3507                              addremove == '-' ? '+' : addremove);
3508
3509         if (options->prefix &&
3510             strncmp(concatpath, options->prefix, options->prefix_length))
3511                 return;
3512
3513         one = alloc_filespec(concatpath);
3514         two = alloc_filespec(concatpath);
3515
3516         if (addremove != '+')
3517                 fill_filespec(one, sha1, mode);
3518         if (addremove != '-')
3519                 fill_filespec(two, sha1, mode);
3520
3521         diff_queue(&diff_queued_diff, one, two);
3522         DIFF_OPT_SET(options, HAS_CHANGES);
3523 }
3524
3525 void diff_change(struct diff_options *options,
3526                  unsigned old_mode, unsigned new_mode,
3527                  const unsigned char *old_sha1,
3528                  const unsigned char *new_sha1,
3529                  const char *concatpath)
3530 {
3531         struct diff_filespec *one, *two;
3532
3533         if (DIFF_OPT_TST(options, IGNORE_SUBMODULES) && S_ISGITLINK(old_mode)
3534                         && S_ISGITLINK(new_mode))
3535                 return;
3536
3537         if (DIFF_OPT_TST(options, REVERSE_DIFF)) {
3538                 unsigned tmp;
3539                 const unsigned char *tmp_c;
3540                 tmp = old_mode; old_mode = new_mode; new_mode = tmp;
3541                 tmp_c = old_sha1; old_sha1 = new_sha1; new_sha1 = tmp_c;
3542         }
3543
3544         if (options->prefix &&
3545             strncmp(concatpath, options->prefix, options->prefix_length))
3546                 return;
3547
3548         one = alloc_filespec(concatpath);
3549         two = alloc_filespec(concatpath);
3550         fill_filespec(one, old_sha1, old_mode);
3551         fill_filespec(two, new_sha1, new_mode);
3552
3553         diff_queue(&diff_queued_diff, one, two);
3554         DIFF_OPT_SET(options, HAS_CHANGES);
3555 }
3556
3557 void diff_unmerge(struct diff_options *options,
3558                   const char *path,
3559                   unsigned mode, const unsigned char *sha1)
3560 {
3561         struct diff_filespec *one, *two;
3562
3563         if (options->prefix &&
3564             strncmp(path, options->prefix, options->prefix_length))
3565                 return;
3566
3567         one = alloc_filespec(path);
3568         two = alloc_filespec(path);
3569         fill_filespec(one, sha1, mode);
3570         diff_queue(&diff_queued_diff, one, two)->is_unmerged = 1;
3571 }
3572
3573 static char *run_textconv(const char *pgm, struct diff_filespec *spec,
3574                 size_t *outsize)
3575 {
3576         struct diff_tempfile *temp;
3577         const char *argv[3];
3578         const char **arg = argv;
3579         struct child_process child;
3580         struct strbuf buf = STRBUF_INIT;
3581
3582         temp = prepare_temp_file(spec->path, spec);
3583         *arg++ = pgm;
3584         *arg++ = temp->name;
3585         *arg = NULL;
3586
3587         memset(&child, 0, sizeof(child));
3588         child.argv = argv;
3589         child.out = -1;
3590         if (start_command(&child) != 0 ||
3591             strbuf_read(&buf, child.out, 0) < 0 ||
3592             finish_command(&child) != 0) {
3593                 remove_tempfile();
3594                 error("error running textconv command '%s'", pgm);
3595                 return NULL;
3596         }
3597         remove_tempfile();
3598
3599         return strbuf_detach(&buf, outsize);
3600 }