archive: complain about path specs that don't match anything
[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 #include "submodule.h"
17
18 #ifdef NO_FAST_WORKING_DIRECTORY
19 #define FAST_WORKING_DIRECTORY 0
20 #else
21 #define FAST_WORKING_DIRECTORY 1
22 #endif
23
24 static int diff_detect_rename_default;
25 static int diff_rename_limit_default = 200;
26 static int diff_suppress_blank_empty;
27 int diff_use_color_default = -1;
28 static const char *diff_word_regex_cfg;
29 static const char *external_diff_cmd_cfg;
30 int diff_auto_refresh_index = 1;
31 static int diff_mnemonic_prefix;
32
33 static char diff_colors[][COLOR_MAXLEN] = {
34         GIT_COLOR_RESET,
35         GIT_COLOR_NORMAL,       /* PLAIN */
36         GIT_COLOR_BOLD,         /* METAINFO */
37         GIT_COLOR_CYAN,         /* FRAGINFO */
38         GIT_COLOR_RED,          /* OLD */
39         GIT_COLOR_GREEN,        /* NEW */
40         GIT_COLOR_YELLOW,       /* COMMIT */
41         GIT_COLOR_BG_RED,       /* WHITESPACE */
42         GIT_COLOR_NORMAL,       /* FUNCINFO */
43 };
44
45 static void diff_filespec_load_driver(struct diff_filespec *one);
46 static char *run_textconv(const char *, struct diff_filespec *, size_t *);
47
48 static int parse_diff_color_slot(const char *var, int ofs)
49 {
50         if (!strcasecmp(var+ofs, "plain"))
51                 return DIFF_PLAIN;
52         if (!strcasecmp(var+ofs, "meta"))
53                 return DIFF_METAINFO;
54         if (!strcasecmp(var+ofs, "frag"))
55                 return DIFF_FRAGINFO;
56         if (!strcasecmp(var+ofs, "old"))
57                 return DIFF_FILE_OLD;
58         if (!strcasecmp(var+ofs, "new"))
59                 return DIFF_FILE_NEW;
60         if (!strcasecmp(var+ofs, "commit"))
61                 return DIFF_COMMIT;
62         if (!strcasecmp(var+ofs, "whitespace"))
63                 return DIFF_WHITESPACE;
64         if (!strcasecmp(var+ofs, "func"))
65                 return DIFF_FUNCINFO;
66         return -1;
67 }
68
69 static int git_config_rename(const char *var, const char *value)
70 {
71         if (!value)
72                 return DIFF_DETECT_RENAME;
73         if (!strcasecmp(value, "copies") || !strcasecmp(value, "copy"))
74                 return  DIFF_DETECT_COPY;
75         return git_config_bool(var,value) ? DIFF_DETECT_RENAME : 0;
76 }
77
78 /*
79  * These are to give UI layer defaults.
80  * The core-level commands such as git-diff-files should
81  * never be affected by the setting of diff.renames
82  * the user happens to have in the configuration file.
83  */
84 int git_diff_ui_config(const char *var, const char *value, void *cb)
85 {
86         if (!strcmp(var, "diff.color") || !strcmp(var, "color.diff")) {
87                 diff_use_color_default = git_config_colorbool(var, value, -1);
88                 return 0;
89         }
90         if (!strcmp(var, "diff.renames")) {
91                 diff_detect_rename_default = git_config_rename(var, value);
92                 return 0;
93         }
94         if (!strcmp(var, "diff.autorefreshindex")) {
95                 diff_auto_refresh_index = git_config_bool(var, value);
96                 return 0;
97         }
98         if (!strcmp(var, "diff.mnemonicprefix")) {
99                 diff_mnemonic_prefix = git_config_bool(var, value);
100                 return 0;
101         }
102         if (!strcmp(var, "diff.external"))
103                 return git_config_string(&external_diff_cmd_cfg, var, value);
104         if (!strcmp(var, "diff.wordregex"))
105                 return git_config_string(&diff_word_regex_cfg, var, value);
106
107         return git_diff_basic_config(var, value, cb);
108 }
109
110 int git_diff_basic_config(const char *var, const char *value, void *cb)
111 {
112         if (!strcmp(var, "diff.renamelimit")) {
113                 diff_rename_limit_default = git_config_int(var, value);
114                 return 0;
115         }
116
117         switch (userdiff_config(var, value)) {
118                 case 0: break;
119                 case -1: return -1;
120                 default: return 0;
121         }
122
123         if (!prefixcmp(var, "diff.color.") || !prefixcmp(var, "color.diff.")) {
124                 int slot = parse_diff_color_slot(var, 11);
125                 if (slot < 0)
126                         return 0;
127                 if (!value)
128                         return config_error_nonbool(var);
129                 color_parse(value, var, diff_colors[slot]);
130                 return 0;
131         }
132
133         /* like GNU diff's --suppress-blank-empty option  */
134         if (!strcmp(var, "diff.suppressblankempty") ||
135                         /* for backwards compatibility */
136                         !strcmp(var, "diff.suppress-blank-empty")) {
137                 diff_suppress_blank_empty = git_config_bool(var, value);
138                 return 0;
139         }
140
141         return git_color_default_config(var, value, cb);
142 }
143
144 static char *quote_two(const char *one, const char *two)
145 {
146         int need_one = quote_c_style(one, NULL, NULL, 1);
147         int need_two = quote_c_style(two, NULL, NULL, 1);
148         struct strbuf res = STRBUF_INIT;
149
150         if (need_one + need_two) {
151                 strbuf_addch(&res, '"');
152                 quote_c_style(one, &res, NULL, 1);
153                 quote_c_style(two, &res, NULL, 1);
154                 strbuf_addch(&res, '"');
155         } else {
156                 strbuf_addstr(&res, one);
157                 strbuf_addstr(&res, two);
158         }
159         return strbuf_detach(&res, NULL);
160 }
161
162 static const char *external_diff(void)
163 {
164         static const char *external_diff_cmd = NULL;
165         static int done_preparing = 0;
166
167         if (done_preparing)
168                 return external_diff_cmd;
169         external_diff_cmd = getenv("GIT_EXTERNAL_DIFF");
170         if (!external_diff_cmd)
171                 external_diff_cmd = external_diff_cmd_cfg;
172         done_preparing = 1;
173         return external_diff_cmd;
174 }
175
176 static struct diff_tempfile {
177         const char *name; /* filename external diff should read from */
178         char hex[41];
179         char mode[10];
180         char tmp_path[PATH_MAX];
181 } diff_temp[2];
182
183 typedef unsigned long (*sane_truncate_fn)(char *line, unsigned long len);
184
185 struct emit_callback {
186         int color_diff;
187         unsigned ws_rule;
188         int blank_at_eof_in_preimage;
189         int blank_at_eof_in_postimage;
190         int lno_in_preimage;
191         int lno_in_postimage;
192         sane_truncate_fn truncate;
193         const char **label_path;
194         struct diff_words_data *diff_words;
195         int *found_changesp;
196         FILE *file;
197 };
198
199 static int count_lines(const char *data, int size)
200 {
201         int count, ch, completely_empty = 1, nl_just_seen = 0;
202         count = 0;
203         while (0 < size--) {
204                 ch = *data++;
205                 if (ch == '\n') {
206                         count++;
207                         nl_just_seen = 1;
208                         completely_empty = 0;
209                 }
210                 else {
211                         nl_just_seen = 0;
212                         completely_empty = 0;
213                 }
214         }
215         if (completely_empty)
216                 return 0;
217         if (!nl_just_seen)
218                 count++; /* no trailing newline */
219         return count;
220 }
221
222 static int fill_mmfile(mmfile_t *mf, struct diff_filespec *one)
223 {
224         if (!DIFF_FILE_VALID(one)) {
225                 mf->ptr = (char *)""; /* does not matter */
226                 mf->size = 0;
227                 return 0;
228         }
229         else if (diff_populate_filespec(one, 0))
230                 return -1;
231
232         mf->ptr = one->data;
233         mf->size = one->size;
234         return 0;
235 }
236
237 static int count_trailing_blank(mmfile_t *mf, unsigned ws_rule)
238 {
239         char *ptr = mf->ptr;
240         long size = mf->size;
241         int cnt = 0;
242
243         if (!size)
244                 return cnt;
245         ptr += size - 1; /* pointing at the very end */
246         if (*ptr != '\n')
247                 ; /* incomplete line */
248         else
249                 ptr--; /* skip the last LF */
250         while (mf->ptr < ptr) {
251                 char *prev_eol;
252                 for (prev_eol = ptr; mf->ptr <= prev_eol; prev_eol--)
253                         if (*prev_eol == '\n')
254                                 break;
255                 if (!ws_blank_line(prev_eol + 1, ptr - prev_eol, ws_rule))
256                         break;
257                 cnt++;
258                 ptr = prev_eol - 1;
259         }
260         return cnt;
261 }
262
263 static void check_blank_at_eof(mmfile_t *mf1, mmfile_t *mf2,
264                                struct emit_callback *ecbdata)
265 {
266         int l1, l2, at;
267         unsigned ws_rule = ecbdata->ws_rule;
268         l1 = count_trailing_blank(mf1, ws_rule);
269         l2 = count_trailing_blank(mf2, ws_rule);
270         if (l2 <= l1) {
271                 ecbdata->blank_at_eof_in_preimage = 0;
272                 ecbdata->blank_at_eof_in_postimage = 0;
273                 return;
274         }
275         at = count_lines(mf1->ptr, mf1->size);
276         ecbdata->blank_at_eof_in_preimage = (at - l1) + 1;
277
278         at = count_lines(mf2->ptr, mf2->size);
279         ecbdata->blank_at_eof_in_postimage = (at - l2) + 1;
280 }
281
282 static void emit_line_0(FILE *file, const char *set, const char *reset,
283                         int first, const char *line, int len)
284 {
285         int has_trailing_newline, has_trailing_carriage_return;
286         int nofirst;
287
288         if (len == 0) {
289                 has_trailing_newline = (first == '\n');
290                 has_trailing_carriage_return = (!has_trailing_newline &&
291                                                 (first == '\r'));
292                 nofirst = has_trailing_newline || has_trailing_carriage_return;
293         } else {
294                 has_trailing_newline = (len > 0 && line[len-1] == '\n');
295                 if (has_trailing_newline)
296                         len--;
297                 has_trailing_carriage_return = (len > 0 && line[len-1] == '\r');
298                 if (has_trailing_carriage_return)
299                         len--;
300                 nofirst = 0;
301         }
302
303         if (len || !nofirst) {
304                 fputs(set, file);
305                 if (!nofirst)
306                         fputc(first, file);
307                 fwrite(line, len, 1, file);
308                 fputs(reset, file);
309         }
310         if (has_trailing_carriage_return)
311                 fputc('\r', file);
312         if (has_trailing_newline)
313                 fputc('\n', file);
314 }
315
316 static void emit_line(FILE *file, const char *set, const char *reset,
317                       const char *line, int len)
318 {
319         emit_line_0(file, set, reset, line[0], line+1, len-1);
320 }
321
322 static int new_blank_line_at_eof(struct emit_callback *ecbdata, const char *line, int len)
323 {
324         if (!((ecbdata->ws_rule & WS_BLANK_AT_EOF) &&
325               ecbdata->blank_at_eof_in_preimage &&
326               ecbdata->blank_at_eof_in_postimage &&
327               ecbdata->blank_at_eof_in_preimage <= ecbdata->lno_in_preimage &&
328               ecbdata->blank_at_eof_in_postimage <= ecbdata->lno_in_postimage))
329                 return 0;
330         return ws_blank_line(line, len, ecbdata->ws_rule);
331 }
332
333 static void emit_add_line(const char *reset,
334                           struct emit_callback *ecbdata,
335                           const char *line, int len)
336 {
337         const char *ws = diff_get_color(ecbdata->color_diff, DIFF_WHITESPACE);
338         const char *set = diff_get_color(ecbdata->color_diff, DIFF_FILE_NEW);
339
340         if (!*ws)
341                 emit_line_0(ecbdata->file, set, reset, '+', line, len);
342         else if (new_blank_line_at_eof(ecbdata, line, len))
343                 /* Blank line at EOF - paint '+' as well */
344                 emit_line_0(ecbdata->file, ws, reset, '+', line, len);
345         else {
346                 /* Emit just the prefix, then the rest. */
347                 emit_line_0(ecbdata->file, set, reset, '+', "", 0);
348                 ws_check_emit(line, len, ecbdata->ws_rule,
349                               ecbdata->file, set, reset, ws);
350         }
351 }
352
353 static void emit_hunk_header(struct emit_callback *ecbdata,
354                              const char *line, int len)
355 {
356         const char *plain = diff_get_color(ecbdata->color_diff, DIFF_PLAIN);
357         const char *frag = diff_get_color(ecbdata->color_diff, DIFF_FRAGINFO);
358         const char *func = diff_get_color(ecbdata->color_diff, DIFF_FUNCINFO);
359         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
360         static const char atat[2] = { '@', '@' };
361         const char *cp, *ep;
362
363         /*
364          * As a hunk header must begin with "@@ -<old>, +<new> @@",
365          * it always is at least 10 bytes long.
366          */
367         if (len < 10 ||
368             memcmp(line, atat, 2) ||
369             !(ep = memmem(line + 2, len - 2, atat, 2))) {
370                 emit_line(ecbdata->file, plain, reset, line, len);
371                 return;
372         }
373         ep += 2; /* skip over @@ */
374
375         /* The hunk header in fraginfo color */
376         emit_line(ecbdata->file, frag, reset, line, ep - line);
377
378         /* blank before the func header */
379         for (cp = ep; ep - line < len; ep++)
380                 if (*ep != ' ' && *ep != '\t')
381                         break;
382         if (ep != cp)
383                 emit_line(ecbdata->file, plain, reset, cp, ep - cp);
384
385         if (ep < line + len)
386                 emit_line(ecbdata->file, func, reset, ep, line + len - ep);
387 }
388
389 static struct diff_tempfile *claim_diff_tempfile(void) {
390         int i;
391         for (i = 0; i < ARRAY_SIZE(diff_temp); i++)
392                 if (!diff_temp[i].name)
393                         return diff_temp + i;
394         die("BUG: diff is failing to clean up its tempfiles");
395 }
396
397 static int remove_tempfile_installed;
398
399 static void remove_tempfile(void)
400 {
401         int i;
402         for (i = 0; i < ARRAY_SIZE(diff_temp); i++) {
403                 if (diff_temp[i].name == diff_temp[i].tmp_path)
404                         unlink_or_warn(diff_temp[i].name);
405                 diff_temp[i].name = NULL;
406         }
407 }
408
409 static void remove_tempfile_on_signal(int signo)
410 {
411         remove_tempfile();
412         sigchain_pop(signo);
413         raise(signo);
414 }
415
416 static void print_line_count(FILE *file, int count)
417 {
418         switch (count) {
419         case 0:
420                 fprintf(file, "0,0");
421                 break;
422         case 1:
423                 fprintf(file, "1");
424                 break;
425         default:
426                 fprintf(file, "1,%d", count);
427                 break;
428         }
429 }
430
431 static void emit_rewrite_lines(struct emit_callback *ecb,
432                                int prefix, const char *data, int size)
433 {
434         const char *endp = NULL;
435         static const char *nneof = " No newline at end of file\n";
436         const char *old = diff_get_color(ecb->color_diff, DIFF_FILE_OLD);
437         const char *reset = diff_get_color(ecb->color_diff, DIFF_RESET);
438
439         while (0 < size) {
440                 int len;
441
442                 endp = memchr(data, '\n', size);
443                 len = endp ? (endp - data + 1) : size;
444                 if (prefix != '+') {
445                         ecb->lno_in_preimage++;
446                         emit_line_0(ecb->file, old, reset, '-',
447                                     data, len);
448                 } else {
449                         ecb->lno_in_postimage++;
450                         emit_add_line(reset, ecb, data, len);
451                 }
452                 size -= len;
453                 data += len;
454         }
455         if (!endp) {
456                 const char *plain = diff_get_color(ecb->color_diff,
457                                                    DIFF_PLAIN);
458                 emit_line_0(ecb->file, plain, reset, '\\',
459                             nneof, strlen(nneof));
460         }
461 }
462
463 static void emit_rewrite_diff(const char *name_a,
464                               const char *name_b,
465                               struct diff_filespec *one,
466                               struct diff_filespec *two,
467                               const char *textconv_one,
468                               const char *textconv_two,
469                               struct diff_options *o)
470 {
471         int lc_a, lc_b;
472         int color_diff = DIFF_OPT_TST(o, COLOR_DIFF);
473         const char *name_a_tab, *name_b_tab;
474         const char *metainfo = diff_get_color(color_diff, DIFF_METAINFO);
475         const char *fraginfo = diff_get_color(color_diff, DIFF_FRAGINFO);
476         const char *reset = diff_get_color(color_diff, DIFF_RESET);
477         static struct strbuf a_name = STRBUF_INIT, b_name = STRBUF_INIT;
478         const char *a_prefix, *b_prefix;
479         const char *data_one, *data_two;
480         size_t size_one, size_two;
481         struct emit_callback ecbdata;
482
483         if (diff_mnemonic_prefix && DIFF_OPT_TST(o, REVERSE_DIFF)) {
484                 a_prefix = o->b_prefix;
485                 b_prefix = o->a_prefix;
486         } else {
487                 a_prefix = o->a_prefix;
488                 b_prefix = o->b_prefix;
489         }
490
491         name_a += (*name_a == '/');
492         name_b += (*name_b == '/');
493         name_a_tab = strchr(name_a, ' ') ? "\t" : "";
494         name_b_tab = strchr(name_b, ' ') ? "\t" : "";
495
496         strbuf_reset(&a_name);
497         strbuf_reset(&b_name);
498         quote_two_c_style(&a_name, a_prefix, name_a, 0);
499         quote_two_c_style(&b_name, b_prefix, name_b, 0);
500
501         diff_populate_filespec(one, 0);
502         diff_populate_filespec(two, 0);
503         if (textconv_one) {
504                 data_one = run_textconv(textconv_one, one, &size_one);
505                 if (!data_one)
506                         die("unable to read files to diff");
507         }
508         else {
509                 data_one = one->data;
510                 size_one = one->size;
511         }
512         if (textconv_two) {
513                 data_two = run_textconv(textconv_two, two, &size_two);
514                 if (!data_two)
515                         die("unable to read files to diff");
516         }
517         else {
518                 data_two = two->data;
519                 size_two = two->size;
520         }
521
522         memset(&ecbdata, 0, sizeof(ecbdata));
523         ecbdata.color_diff = color_diff;
524         ecbdata.found_changesp = &o->found_changes;
525         ecbdata.ws_rule = whitespace_rule(name_b ? name_b : name_a);
526         ecbdata.file = o->file;
527         if (ecbdata.ws_rule & WS_BLANK_AT_EOF) {
528                 mmfile_t mf1, mf2;
529                 mf1.ptr = (char *)data_one;
530                 mf2.ptr = (char *)data_two;
531                 mf1.size = size_one;
532                 mf2.size = size_two;
533                 check_blank_at_eof(&mf1, &mf2, &ecbdata);
534         }
535         ecbdata.lno_in_preimage = 1;
536         ecbdata.lno_in_postimage = 1;
537
538         lc_a = count_lines(data_one, size_one);
539         lc_b = count_lines(data_two, size_two);
540         fprintf(o->file,
541                 "%s--- %s%s%s\n%s+++ %s%s%s\n%s@@ -",
542                 metainfo, a_name.buf, name_a_tab, reset,
543                 metainfo, b_name.buf, name_b_tab, reset, fraginfo);
544         print_line_count(o->file, lc_a);
545         fprintf(o->file, " +");
546         print_line_count(o->file, lc_b);
547         fprintf(o->file, " @@%s\n", reset);
548         if (lc_a)
549                 emit_rewrite_lines(&ecbdata, '-', data_one, size_one);
550         if (lc_b)
551                 emit_rewrite_lines(&ecbdata, '+', data_two, size_two);
552 }
553
554 struct diff_words_buffer {
555         mmfile_t text;
556         long alloc;
557         struct diff_words_orig {
558                 const char *begin, *end;
559         } *orig;
560         int orig_nr, orig_alloc;
561 };
562
563 static void diff_words_append(char *line, unsigned long len,
564                 struct diff_words_buffer *buffer)
565 {
566         ALLOC_GROW(buffer->text.ptr, buffer->text.size + len, buffer->alloc);
567         line++;
568         len--;
569         memcpy(buffer->text.ptr + buffer->text.size, line, len);
570         buffer->text.size += len;
571         buffer->text.ptr[buffer->text.size] = '\0';
572 }
573
574 struct diff_words_data {
575         struct diff_words_buffer minus, plus;
576         const char *current_plus;
577         FILE *file;
578         regex_t *word_regex;
579 };
580
581 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
582 {
583         struct diff_words_data *diff_words = priv;
584         int minus_first, minus_len, plus_first, plus_len;
585         const char *minus_begin, *minus_end, *plus_begin, *plus_end;
586
587         if (line[0] != '@' || parse_hunk_header(line, len,
588                         &minus_first, &minus_len, &plus_first, &plus_len))
589                 return;
590
591         /* POSIX requires that first be decremented by one if len == 0... */
592         if (minus_len) {
593                 minus_begin = diff_words->minus.orig[minus_first].begin;
594                 minus_end =
595                         diff_words->minus.orig[minus_first + minus_len - 1].end;
596         } else
597                 minus_begin = minus_end =
598                         diff_words->minus.orig[minus_first].end;
599
600         if (plus_len) {
601                 plus_begin = diff_words->plus.orig[plus_first].begin;
602                 plus_end = diff_words->plus.orig[plus_first + plus_len - 1].end;
603         } else
604                 plus_begin = plus_end = diff_words->plus.orig[plus_first].end;
605
606         if (diff_words->current_plus != plus_begin)
607                 fwrite(diff_words->current_plus,
608                                 plus_begin - diff_words->current_plus, 1,
609                                 diff_words->file);
610         if (minus_begin != minus_end)
611                 color_fwrite_lines(diff_words->file,
612                                 diff_get_color(1, DIFF_FILE_OLD),
613                                 minus_end - minus_begin, minus_begin);
614         if (plus_begin != plus_end)
615                 color_fwrite_lines(diff_words->file,
616                                 diff_get_color(1, DIFF_FILE_NEW),
617                                 plus_end - plus_begin, plus_begin);
618
619         diff_words->current_plus = plus_end;
620 }
621
622 /* This function starts looking at *begin, and returns 0 iff a word was found. */
623 static int find_word_boundaries(mmfile_t *buffer, regex_t *word_regex,
624                 int *begin, int *end)
625 {
626         if (word_regex && *begin < buffer->size) {
627                 regmatch_t match[1];
628                 if (!regexec(word_regex, buffer->ptr + *begin, 1, match, 0)) {
629                         char *p = memchr(buffer->ptr + *begin + match[0].rm_so,
630                                         '\n', match[0].rm_eo - match[0].rm_so);
631                         *end = p ? p - buffer->ptr : match[0].rm_eo + *begin;
632                         *begin += match[0].rm_so;
633                         return *begin >= *end;
634                 }
635                 return -1;
636         }
637
638         /* find the next word */
639         while (*begin < buffer->size && isspace(buffer->ptr[*begin]))
640                 (*begin)++;
641         if (*begin >= buffer->size)
642                 return -1;
643
644         /* find the end of the word */
645         *end = *begin + 1;
646         while (*end < buffer->size && !isspace(buffer->ptr[*end]))
647                 (*end)++;
648
649         return 0;
650 }
651
652 /*
653  * This function splits the words in buffer->text, stores the list with
654  * newline separator into out, and saves the offsets of the original words
655  * in buffer->orig.
656  */
657 static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out,
658                 regex_t *word_regex)
659 {
660         int i, j;
661         long alloc = 0;
662
663         out->size = 0;
664         out->ptr = NULL;
665
666         /* fake an empty "0th" word */
667         ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
668         buffer->orig[0].begin = buffer->orig[0].end = buffer->text.ptr;
669         buffer->orig_nr = 1;
670
671         for (i = 0; i < buffer->text.size; i++) {
672                 if (find_word_boundaries(&buffer->text, word_regex, &i, &j))
673                         return;
674
675                 /* store original boundaries */
676                 ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
677                                 buffer->orig_alloc);
678                 buffer->orig[buffer->orig_nr].begin = buffer->text.ptr + i;
679                 buffer->orig[buffer->orig_nr].end = buffer->text.ptr + j;
680                 buffer->orig_nr++;
681
682                 /* store one word */
683                 ALLOC_GROW(out->ptr, out->size + j - i + 1, alloc);
684                 memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
685                 out->ptr[out->size + j - i] = '\n';
686                 out->size += j - i + 1;
687
688                 i = j - 1;
689         }
690 }
691
692 /* this executes the word diff on the accumulated buffers */
693 static void diff_words_show(struct diff_words_data *diff_words)
694 {
695         xpparam_t xpp;
696         xdemitconf_t xecfg;
697         xdemitcb_t ecb;
698         mmfile_t minus, plus;
699
700         /* special case: only removal */
701         if (!diff_words->plus.text.size) {
702                 color_fwrite_lines(diff_words->file,
703                         diff_get_color(1, DIFF_FILE_OLD),
704                         diff_words->minus.text.size, diff_words->minus.text.ptr);
705                 diff_words->minus.text.size = 0;
706                 return;
707         }
708
709         diff_words->current_plus = diff_words->plus.text.ptr;
710
711         memset(&xpp, 0, sizeof(xpp));
712         memset(&xecfg, 0, sizeof(xecfg));
713         diff_words_fill(&diff_words->minus, &minus, diff_words->word_regex);
714         diff_words_fill(&diff_words->plus, &plus, diff_words->word_regex);
715         xpp.flags = XDF_NEED_MINIMAL;
716         /* as only the hunk header will be parsed, we need a 0-context */
717         xecfg.ctxlen = 0;
718         xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
719                       &xpp, &xecfg, &ecb);
720         free(minus.ptr);
721         free(plus.ptr);
722         if (diff_words->current_plus != diff_words->plus.text.ptr +
723                         diff_words->plus.text.size)
724                 fwrite(diff_words->current_plus,
725                         diff_words->plus.text.ptr + diff_words->plus.text.size
726                         - diff_words->current_plus, 1,
727                         diff_words->file);
728         diff_words->minus.text.size = diff_words->plus.text.size = 0;
729 }
730
731 /* In "color-words" mode, show word-diff of words accumulated in the buffer */
732 static void diff_words_flush(struct emit_callback *ecbdata)
733 {
734         if (ecbdata->diff_words->minus.text.size ||
735             ecbdata->diff_words->plus.text.size)
736                 diff_words_show(ecbdata->diff_words);
737 }
738
739 static void free_diff_words_data(struct emit_callback *ecbdata)
740 {
741         if (ecbdata->diff_words) {
742                 diff_words_flush(ecbdata);
743                 free (ecbdata->diff_words->minus.text.ptr);
744                 free (ecbdata->diff_words->minus.orig);
745                 free (ecbdata->diff_words->plus.text.ptr);
746                 free (ecbdata->diff_words->plus.orig);
747                 free(ecbdata->diff_words->word_regex);
748                 free(ecbdata->diff_words);
749                 ecbdata->diff_words = NULL;
750         }
751 }
752
753 const char *diff_get_color(int diff_use_color, enum color_diff ix)
754 {
755         if (diff_use_color)
756                 return diff_colors[ix];
757         return "";
758 }
759
760 static unsigned long sane_truncate_line(struct emit_callback *ecb, char *line, unsigned long len)
761 {
762         const char *cp;
763         unsigned long allot;
764         size_t l = len;
765
766         if (ecb->truncate)
767                 return ecb->truncate(line, len);
768         cp = line;
769         allot = l;
770         while (0 < l) {
771                 (void) utf8_width(&cp, &l);
772                 if (!cp)
773                         break; /* truncated in the middle? */
774         }
775         return allot - l;
776 }
777
778 static void find_lno(const char *line, struct emit_callback *ecbdata)
779 {
780         const char *p;
781         ecbdata->lno_in_preimage = 0;
782         ecbdata->lno_in_postimage = 0;
783         p = strchr(line, '-');
784         if (!p)
785                 return; /* cannot happen */
786         ecbdata->lno_in_preimage = strtol(p + 1, NULL, 10);
787         p = strchr(p, '+');
788         if (!p)
789                 return; /* cannot happen */
790         ecbdata->lno_in_postimage = strtol(p + 1, NULL, 10);
791 }
792
793 static void fn_out_consume(void *priv, char *line, unsigned long len)
794 {
795         struct emit_callback *ecbdata = priv;
796         const char *meta = diff_get_color(ecbdata->color_diff, DIFF_METAINFO);
797         const char *plain = diff_get_color(ecbdata->color_diff, DIFF_PLAIN);
798         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
799
800         *(ecbdata->found_changesp) = 1;
801
802         if (ecbdata->label_path[0]) {
803                 const char *name_a_tab, *name_b_tab;
804
805                 name_a_tab = strchr(ecbdata->label_path[0], ' ') ? "\t" : "";
806                 name_b_tab = strchr(ecbdata->label_path[1], ' ') ? "\t" : "";
807
808                 fprintf(ecbdata->file, "%s--- %s%s%s\n",
809                         meta, ecbdata->label_path[0], reset, name_a_tab);
810                 fprintf(ecbdata->file, "%s+++ %s%s%s\n",
811                         meta, ecbdata->label_path[1], reset, name_b_tab);
812                 ecbdata->label_path[0] = ecbdata->label_path[1] = NULL;
813         }
814
815         if (diff_suppress_blank_empty
816             && len == 2 && line[0] == ' ' && line[1] == '\n') {
817                 line[0] = '\n';
818                 len = 1;
819         }
820
821         if (line[0] == '@') {
822                 if (ecbdata->diff_words)
823                         diff_words_flush(ecbdata);
824                 len = sane_truncate_line(ecbdata, line, len);
825                 find_lno(line, ecbdata);
826                 emit_hunk_header(ecbdata, line, len);
827                 if (line[len-1] != '\n')
828                         putc('\n', ecbdata->file);
829                 return;
830         }
831
832         if (len < 1) {
833                 emit_line(ecbdata->file, reset, reset, line, len);
834                 return;
835         }
836
837         if (ecbdata->diff_words) {
838                 if (line[0] == '-') {
839                         diff_words_append(line, len,
840                                           &ecbdata->diff_words->minus);
841                         return;
842                 } else if (line[0] == '+') {
843                         diff_words_append(line, len,
844                                           &ecbdata->diff_words->plus);
845                         return;
846                 }
847                 diff_words_flush(ecbdata);
848                 line++;
849                 len--;
850                 emit_line(ecbdata->file, plain, reset, line, len);
851                 return;
852         }
853
854         if (line[0] != '+') {
855                 const char *color =
856                         diff_get_color(ecbdata->color_diff,
857                                        line[0] == '-' ? DIFF_FILE_OLD : DIFF_PLAIN);
858                 ecbdata->lno_in_preimage++;
859                 if (line[0] == ' ')
860                         ecbdata->lno_in_postimage++;
861                 emit_line(ecbdata->file, color, reset, line, len);
862         } else {
863                 ecbdata->lno_in_postimage++;
864                 emit_add_line(reset, ecbdata, line + 1, len - 1);
865         }
866 }
867
868 static char *pprint_rename(const char *a, const char *b)
869 {
870         const char *old = a;
871         const char *new = b;
872         struct strbuf name = STRBUF_INIT;
873         int pfx_length, sfx_length;
874         int len_a = strlen(a);
875         int len_b = strlen(b);
876         int a_midlen, b_midlen;
877         int qlen_a = quote_c_style(a, NULL, NULL, 0);
878         int qlen_b = quote_c_style(b, NULL, NULL, 0);
879
880         if (qlen_a || qlen_b) {
881                 quote_c_style(a, &name, NULL, 0);
882                 strbuf_addstr(&name, " => ");
883                 quote_c_style(b, &name, NULL, 0);
884                 return strbuf_detach(&name, NULL);
885         }
886
887         /* Find common prefix */
888         pfx_length = 0;
889         while (*old && *new && *old == *new) {
890                 if (*old == '/')
891                         pfx_length = old - a + 1;
892                 old++;
893                 new++;
894         }
895
896         /* Find common suffix */
897         old = a + len_a;
898         new = b + len_b;
899         sfx_length = 0;
900         while (a <= old && b <= new && *old == *new) {
901                 if (*old == '/')
902                         sfx_length = len_a - (old - a);
903                 old--;
904                 new--;
905         }
906
907         /*
908          * pfx{mid-a => mid-b}sfx
909          * {pfx-a => pfx-b}sfx
910          * pfx{sfx-a => sfx-b}
911          * name-a => name-b
912          */
913         a_midlen = len_a - pfx_length - sfx_length;
914         b_midlen = len_b - pfx_length - sfx_length;
915         if (a_midlen < 0)
916                 a_midlen = 0;
917         if (b_midlen < 0)
918                 b_midlen = 0;
919
920         strbuf_grow(&name, pfx_length + a_midlen + b_midlen + sfx_length + 7);
921         if (pfx_length + sfx_length) {
922                 strbuf_add(&name, a, pfx_length);
923                 strbuf_addch(&name, '{');
924         }
925         strbuf_add(&name, a + pfx_length, a_midlen);
926         strbuf_addstr(&name, " => ");
927         strbuf_add(&name, b + pfx_length, b_midlen);
928         if (pfx_length + sfx_length) {
929                 strbuf_addch(&name, '}');
930                 strbuf_add(&name, a + len_a - sfx_length, sfx_length);
931         }
932         return strbuf_detach(&name, NULL);
933 }
934
935 struct diffstat_t {
936         int nr;
937         int alloc;
938         struct diffstat_file {
939                 char *from_name;
940                 char *name;
941                 char *print_name;
942                 unsigned is_unmerged:1;
943                 unsigned is_binary:1;
944                 unsigned is_renamed:1;
945                 unsigned int added, deleted;
946         } **files;
947 };
948
949 static struct diffstat_file *diffstat_add(struct diffstat_t *diffstat,
950                                           const char *name_a,
951                                           const char *name_b)
952 {
953         struct diffstat_file *x;
954         x = xcalloc(sizeof (*x), 1);
955         if (diffstat->nr == diffstat->alloc) {
956                 diffstat->alloc = alloc_nr(diffstat->alloc);
957                 diffstat->files = xrealloc(diffstat->files,
958                                 diffstat->alloc * sizeof(x));
959         }
960         diffstat->files[diffstat->nr++] = x;
961         if (name_b) {
962                 x->from_name = xstrdup(name_a);
963                 x->name = xstrdup(name_b);
964                 x->is_renamed = 1;
965         }
966         else {
967                 x->from_name = NULL;
968                 x->name = xstrdup(name_a);
969         }
970         return x;
971 }
972
973 static void diffstat_consume(void *priv, char *line, unsigned long len)
974 {
975         struct diffstat_t *diffstat = priv;
976         struct diffstat_file *x = diffstat->files[diffstat->nr - 1];
977
978         if (line[0] == '+')
979                 x->added++;
980         else if (line[0] == '-')
981                 x->deleted++;
982 }
983
984 const char mime_boundary_leader[] = "------------";
985
986 static int scale_linear(int it, int width, int max_change)
987 {
988         /*
989          * make sure that at least one '-' is printed if there were deletions,
990          * and likewise for '+'.
991          */
992         if (max_change < 2)
993                 return it;
994         return ((it - 1) * (width - 1) + max_change - 1) / (max_change - 1);
995 }
996
997 static void show_name(FILE *file,
998                       const char *prefix, const char *name, int len)
999 {
1000         fprintf(file, " %s%-*s |", prefix, len, name);
1001 }
1002
1003 static void show_graph(FILE *file, char ch, int cnt, const char *set, const char *reset)
1004 {
1005         if (cnt <= 0)
1006                 return;
1007         fprintf(file, "%s", set);
1008         while (cnt--)
1009                 putc(ch, file);
1010         fprintf(file, "%s", reset);
1011 }
1012
1013 static void fill_print_name(struct diffstat_file *file)
1014 {
1015         char *pname;
1016
1017         if (file->print_name)
1018                 return;
1019
1020         if (!file->is_renamed) {
1021                 struct strbuf buf = STRBUF_INIT;
1022                 if (quote_c_style(file->name, &buf, NULL, 0)) {
1023                         pname = strbuf_detach(&buf, NULL);
1024                 } else {
1025                         pname = file->name;
1026                         strbuf_release(&buf);
1027                 }
1028         } else {
1029                 pname = pprint_rename(file->from_name, file->name);
1030         }
1031         file->print_name = pname;
1032 }
1033
1034 static void show_stats(struct diffstat_t *data, struct diff_options *options)
1035 {
1036         int i, len, add, del, adds = 0, dels = 0;
1037         int max_change = 0, max_len = 0;
1038         int total_files = data->nr;
1039         int width, name_width;
1040         const char *reset, *set, *add_c, *del_c;
1041
1042         if (data->nr == 0)
1043                 return;
1044
1045         width = options->stat_width ? options->stat_width : 80;
1046         name_width = options->stat_name_width ? options->stat_name_width : 50;
1047
1048         /* Sanity: give at least 5 columns to the graph,
1049          * but leave at least 10 columns for the name.
1050          */
1051         if (width < 25)
1052                 width = 25;
1053         if (name_width < 10)
1054                 name_width = 10;
1055         else if (width < name_width + 15)
1056                 name_width = width - 15;
1057
1058         /* Find the longest filename and max number of changes */
1059         reset = diff_get_color_opt(options, DIFF_RESET);
1060         set   = diff_get_color_opt(options, DIFF_PLAIN);
1061         add_c = diff_get_color_opt(options, DIFF_FILE_NEW);
1062         del_c = diff_get_color_opt(options, DIFF_FILE_OLD);
1063
1064         for (i = 0; i < data->nr; i++) {
1065                 struct diffstat_file *file = data->files[i];
1066                 int change = file->added + file->deleted;
1067                 fill_print_name(file);
1068                 len = strlen(file->print_name);
1069                 if (max_len < len)
1070                         max_len = len;
1071
1072                 if (file->is_binary || file->is_unmerged)
1073                         continue;
1074                 if (max_change < change)
1075                         max_change = change;
1076         }
1077
1078         /* Compute the width of the graph part;
1079          * 10 is for one blank at the beginning of the line plus
1080          * " | count " between the name and the graph.
1081          *
1082          * From here on, name_width is the width of the name area,
1083          * and width is the width of the graph area.
1084          */
1085         name_width = (name_width < max_len) ? name_width : max_len;
1086         if (width < (name_width + 10) + max_change)
1087                 width = width - (name_width + 10);
1088         else
1089                 width = max_change;
1090
1091         for (i = 0; i < data->nr; i++) {
1092                 const char *prefix = "";
1093                 char *name = data->files[i]->print_name;
1094                 int added = data->files[i]->added;
1095                 int deleted = data->files[i]->deleted;
1096                 int name_len;
1097
1098                 /*
1099                  * "scale" the filename
1100                  */
1101                 len = name_width;
1102                 name_len = strlen(name);
1103                 if (name_width < name_len) {
1104                         char *slash;
1105                         prefix = "...";
1106                         len -= 3;
1107                         name += name_len - len;
1108                         slash = strchr(name, '/');
1109                         if (slash)
1110                                 name = slash;
1111                 }
1112
1113                 if (data->files[i]->is_binary) {
1114                         show_name(options->file, prefix, name, len);
1115                         fprintf(options->file, "  Bin ");
1116                         fprintf(options->file, "%s%d%s", del_c, deleted, reset);
1117                         fprintf(options->file, " -> ");
1118                         fprintf(options->file, "%s%d%s", add_c, added, reset);
1119                         fprintf(options->file, " bytes");
1120                         fprintf(options->file, "\n");
1121                         continue;
1122                 }
1123                 else if (data->files[i]->is_unmerged) {
1124                         show_name(options->file, prefix, name, len);
1125                         fprintf(options->file, "  Unmerged\n");
1126                         continue;
1127                 }
1128                 else if (!data->files[i]->is_renamed &&
1129                          (added + deleted == 0)) {
1130                         total_files--;
1131                         continue;
1132                 }
1133
1134                 /*
1135                  * scale the add/delete
1136                  */
1137                 add = added;
1138                 del = deleted;
1139                 adds += add;
1140                 dels += del;
1141
1142                 if (width <= max_change) {
1143                         add = scale_linear(add, width, max_change);
1144                         del = scale_linear(del, width, max_change);
1145                 }
1146                 show_name(options->file, prefix, name, len);
1147                 fprintf(options->file, "%5d%s", added + deleted,
1148                                 added + deleted ? " " : "");
1149                 show_graph(options->file, '+', add, add_c, reset);
1150                 show_graph(options->file, '-', del, del_c, reset);
1151                 fprintf(options->file, "\n");
1152         }
1153         fprintf(options->file,
1154                " %d files changed, %d insertions(+), %d deletions(-)\n",
1155                total_files, adds, dels);
1156 }
1157
1158 static void show_shortstats(struct diffstat_t *data, struct diff_options *options)
1159 {
1160         int i, adds = 0, dels = 0, total_files = data->nr;
1161
1162         if (data->nr == 0)
1163                 return;
1164
1165         for (i = 0; i < data->nr; i++) {
1166                 if (!data->files[i]->is_binary &&
1167                     !data->files[i]->is_unmerged) {
1168                         int added = data->files[i]->added;
1169                         int deleted= data->files[i]->deleted;
1170                         if (!data->files[i]->is_renamed &&
1171                             (added + deleted == 0)) {
1172                                 total_files--;
1173                         } else {
1174                                 adds += added;
1175                                 dels += deleted;
1176                         }
1177                 }
1178         }
1179         fprintf(options->file, " %d files changed, %d insertions(+), %d deletions(-)\n",
1180                total_files, adds, dels);
1181 }
1182
1183 static void show_numstat(struct diffstat_t *data, struct diff_options *options)
1184 {
1185         int i;
1186
1187         if (data->nr == 0)
1188                 return;
1189
1190         for (i = 0; i < data->nr; i++) {
1191                 struct diffstat_file *file = data->files[i];
1192
1193                 if (file->is_binary)
1194                         fprintf(options->file, "-\t-\t");
1195                 else
1196                         fprintf(options->file,
1197                                 "%d\t%d\t", file->added, file->deleted);
1198                 if (options->line_termination) {
1199                         fill_print_name(file);
1200                         if (!file->is_renamed)
1201                                 write_name_quoted(file->name, options->file,
1202                                                   options->line_termination);
1203                         else {
1204                                 fputs(file->print_name, options->file);
1205                                 putc(options->line_termination, options->file);
1206                         }
1207                 } else {
1208                         if (file->is_renamed) {
1209                                 putc('\0', options->file);
1210                                 write_name_quoted(file->from_name, options->file, '\0');
1211                         }
1212                         write_name_quoted(file->name, options->file, '\0');
1213                 }
1214         }
1215 }
1216
1217 struct dirstat_file {
1218         const char *name;
1219         unsigned long changed;
1220 };
1221
1222 struct dirstat_dir {
1223         struct dirstat_file *files;
1224         int alloc, nr, percent, cumulative;
1225 };
1226
1227 static long gather_dirstat(FILE *file, struct dirstat_dir *dir, unsigned long changed, const char *base, int baselen)
1228 {
1229         unsigned long this_dir = 0;
1230         unsigned int sources = 0;
1231
1232         while (dir->nr) {
1233                 struct dirstat_file *f = dir->files;
1234                 int namelen = strlen(f->name);
1235                 unsigned long this;
1236                 char *slash;
1237
1238                 if (namelen < baselen)
1239                         break;
1240                 if (memcmp(f->name, base, baselen))
1241                         break;
1242                 slash = strchr(f->name + baselen, '/');
1243                 if (slash) {
1244                         int newbaselen = slash + 1 - f->name;
1245                         this = gather_dirstat(file, dir, changed, f->name, newbaselen);
1246                         sources++;
1247                 } else {
1248                         this = f->changed;
1249                         dir->files++;
1250                         dir->nr--;
1251                         sources += 2;
1252                 }
1253                 this_dir += this;
1254         }
1255
1256         /*
1257          * We don't report dirstat's for
1258          *  - the top level
1259          *  - or cases where everything came from a single directory
1260          *    under this directory (sources == 1).
1261          */
1262         if (baselen && sources != 1) {
1263                 int permille = this_dir * 1000 / changed;
1264                 if (permille) {
1265                         int percent = permille / 10;
1266                         if (percent >= dir->percent) {
1267                                 fprintf(file, "%4d.%01d%% %.*s\n", percent, permille % 10, baselen, base);
1268                                 if (!dir->cumulative)
1269                                         return 0;
1270                         }
1271                 }
1272         }
1273         return this_dir;
1274 }
1275
1276 static int dirstat_compare(const void *_a, const void *_b)
1277 {
1278         const struct dirstat_file *a = _a;
1279         const struct dirstat_file *b = _b;
1280         return strcmp(a->name, b->name);
1281 }
1282
1283 static void show_dirstat(struct diff_options *options)
1284 {
1285         int i;
1286         unsigned long changed;
1287         struct dirstat_dir dir;
1288         struct diff_queue_struct *q = &diff_queued_diff;
1289
1290         dir.files = NULL;
1291         dir.alloc = 0;
1292         dir.nr = 0;
1293         dir.percent = options->dirstat_percent;
1294         dir.cumulative = DIFF_OPT_TST(options, DIRSTAT_CUMULATIVE);
1295
1296         changed = 0;
1297         for (i = 0; i < q->nr; i++) {
1298                 struct diff_filepair *p = q->queue[i];
1299                 const char *name;
1300                 unsigned long copied, added, damage;
1301
1302                 name = p->one->path ? p->one->path : p->two->path;
1303
1304                 if (DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
1305                         diff_populate_filespec(p->one, 0);
1306                         diff_populate_filespec(p->two, 0);
1307                         diffcore_count_changes(p->one, p->two, NULL, NULL, 0,
1308                                                &copied, &added);
1309                         diff_free_filespec_data(p->one);
1310                         diff_free_filespec_data(p->two);
1311                 } else if (DIFF_FILE_VALID(p->one)) {
1312                         diff_populate_filespec(p->one, 1);
1313                         copied = added = 0;
1314                         diff_free_filespec_data(p->one);
1315                 } else if (DIFF_FILE_VALID(p->two)) {
1316                         diff_populate_filespec(p->two, 1);
1317                         copied = 0;
1318                         added = p->two->size;
1319                         diff_free_filespec_data(p->two);
1320                 } else
1321                         continue;
1322
1323                 /*
1324                  * Original minus copied is the removed material,
1325                  * added is the new material.  They are both damages
1326                  * made to the preimage. In --dirstat-by-file mode, count
1327                  * damaged files, not damaged lines. This is done by
1328                  * counting only a single damaged line per file.
1329                  */
1330                 damage = (p->one->size - copied) + added;
1331                 if (DIFF_OPT_TST(options, DIRSTAT_BY_FILE) && damage > 0)
1332                         damage = 1;
1333
1334                 ALLOC_GROW(dir.files, dir.nr + 1, dir.alloc);
1335                 dir.files[dir.nr].name = name;
1336                 dir.files[dir.nr].changed = damage;
1337                 changed += damage;
1338                 dir.nr++;
1339         }
1340
1341         /* This can happen even with many files, if everything was renames */
1342         if (!changed)
1343                 return;
1344
1345         /* Show all directories with more than x% of the changes */
1346         qsort(dir.files, dir.nr, sizeof(dir.files[0]), dirstat_compare);
1347         gather_dirstat(options->file, &dir, changed, "", 0);
1348 }
1349
1350 static void free_diffstat_info(struct diffstat_t *diffstat)
1351 {
1352         int i;
1353         for (i = 0; i < diffstat->nr; i++) {
1354                 struct diffstat_file *f = diffstat->files[i];
1355                 if (f->name != f->print_name)
1356                         free(f->print_name);
1357                 free(f->name);
1358                 free(f->from_name);
1359                 free(f);
1360         }
1361         free(diffstat->files);
1362 }
1363
1364 struct checkdiff_t {
1365         const char *filename;
1366         int lineno;
1367         struct diff_options *o;
1368         unsigned ws_rule;
1369         unsigned status;
1370 };
1371
1372 static int is_conflict_marker(const char *line, unsigned long len)
1373 {
1374         char firstchar;
1375         int cnt;
1376
1377         if (len < 8)
1378                 return 0;
1379         firstchar = line[0];
1380         switch (firstchar) {
1381         case '=': case '>': case '<':
1382                 break;
1383         default:
1384                 return 0;
1385         }
1386         for (cnt = 1; cnt < 7; cnt++)
1387                 if (line[cnt] != firstchar)
1388                         return 0;
1389         /* line[0] thru line[6] are same as firstchar */
1390         if (firstchar == '=') {
1391                 /* divider between ours and theirs? */
1392                 if (len != 8 || line[7] != '\n')
1393                         return 0;
1394         } else if (len < 8 || !isspace(line[7])) {
1395                 /* not divider before ours nor after theirs */
1396                 return 0;
1397         }
1398         return 1;
1399 }
1400
1401 static void checkdiff_consume(void *priv, char *line, unsigned long len)
1402 {
1403         struct checkdiff_t *data = priv;
1404         int color_diff = DIFF_OPT_TST(data->o, COLOR_DIFF);
1405         const char *ws = diff_get_color(color_diff, DIFF_WHITESPACE);
1406         const char *reset = diff_get_color(color_diff, DIFF_RESET);
1407         const char *set = diff_get_color(color_diff, DIFF_FILE_NEW);
1408         char *err;
1409
1410         if (line[0] == '+') {
1411                 unsigned bad;
1412                 data->lineno++;
1413                 if (is_conflict_marker(line + 1, len - 1)) {
1414                         data->status |= 1;
1415                         fprintf(data->o->file,
1416                                 "%s:%d: leftover conflict marker\n",
1417                                 data->filename, data->lineno);
1418                 }
1419                 bad = ws_check(line + 1, len - 1, data->ws_rule);
1420                 if (!bad)
1421                         return;
1422                 data->status |= bad;
1423                 err = whitespace_error_string(bad);
1424                 fprintf(data->o->file, "%s:%d: %s.\n",
1425                         data->filename, data->lineno, err);
1426                 free(err);
1427                 emit_line(data->o->file, set, reset, line, 1);
1428                 ws_check_emit(line + 1, len - 1, data->ws_rule,
1429                               data->o->file, set, reset, ws);
1430         } else if (line[0] == ' ') {
1431                 data->lineno++;
1432         } else if (line[0] == '@') {
1433                 char *plus = strchr(line, '+');
1434                 if (plus)
1435                         data->lineno = strtol(plus, NULL, 10) - 1;
1436                 else
1437                         die("invalid diff");
1438         }
1439 }
1440
1441 static unsigned char *deflate_it(char *data,
1442                                  unsigned long size,
1443                                  unsigned long *result_size)
1444 {
1445         int bound;
1446         unsigned char *deflated;
1447         z_stream stream;
1448
1449         memset(&stream, 0, sizeof(stream));
1450         deflateInit(&stream, zlib_compression_level);
1451         bound = deflateBound(&stream, size);
1452         deflated = xmalloc(bound);
1453         stream.next_out = deflated;
1454         stream.avail_out = bound;
1455
1456         stream.next_in = (unsigned char *)data;
1457         stream.avail_in = size;
1458         while (deflate(&stream, Z_FINISH) == Z_OK)
1459                 ; /* nothing */
1460         deflateEnd(&stream);
1461         *result_size = stream.total_out;
1462         return deflated;
1463 }
1464
1465 static void emit_binary_diff_body(FILE *file, mmfile_t *one, mmfile_t *two)
1466 {
1467         void *cp;
1468         void *delta;
1469         void *deflated;
1470         void *data;
1471         unsigned long orig_size;
1472         unsigned long delta_size;
1473         unsigned long deflate_size;
1474         unsigned long data_size;
1475
1476         /* We could do deflated delta, or we could do just deflated two,
1477          * whichever is smaller.
1478          */
1479         delta = NULL;
1480         deflated = deflate_it(two->ptr, two->size, &deflate_size);
1481         if (one->size && two->size) {
1482                 delta = diff_delta(one->ptr, one->size,
1483                                    two->ptr, two->size,
1484                                    &delta_size, deflate_size);
1485                 if (delta) {
1486                         void *to_free = delta;
1487                         orig_size = delta_size;
1488                         delta = deflate_it(delta, delta_size, &delta_size);
1489                         free(to_free);
1490                 }
1491         }
1492
1493         if (delta && delta_size < deflate_size) {
1494                 fprintf(file, "delta %lu\n", orig_size);
1495                 free(deflated);
1496                 data = delta;
1497                 data_size = delta_size;
1498         }
1499         else {
1500                 fprintf(file, "literal %lu\n", two->size);
1501                 free(delta);
1502                 data = deflated;
1503                 data_size = deflate_size;
1504         }
1505
1506         /* emit data encoded in base85 */
1507         cp = data;
1508         while (data_size) {
1509                 int bytes = (52 < data_size) ? 52 : data_size;
1510                 char line[70];
1511                 data_size -= bytes;
1512                 if (bytes <= 26)
1513                         line[0] = bytes + 'A' - 1;
1514                 else
1515                         line[0] = bytes - 26 + 'a' - 1;
1516                 encode_85(line + 1, cp, bytes);
1517                 cp = (char *) cp + bytes;
1518                 fputs(line, file);
1519                 fputc('\n', file);
1520         }
1521         fprintf(file, "\n");
1522         free(data);
1523 }
1524
1525 static void emit_binary_diff(FILE *file, mmfile_t *one, mmfile_t *two)
1526 {
1527         fprintf(file, "GIT binary patch\n");
1528         emit_binary_diff_body(file, one, two);
1529         emit_binary_diff_body(file, two, one);
1530 }
1531
1532 static void diff_filespec_load_driver(struct diff_filespec *one)
1533 {
1534         if (!one->driver)
1535                 one->driver = userdiff_find_by_path(one->path);
1536         if (!one->driver)
1537                 one->driver = userdiff_find_by_name("default");
1538 }
1539
1540 int diff_filespec_is_binary(struct diff_filespec *one)
1541 {
1542         if (one->is_binary == -1) {
1543                 diff_filespec_load_driver(one);
1544                 if (one->driver->binary != -1)
1545                         one->is_binary = one->driver->binary;
1546                 else {
1547                         if (!one->data && DIFF_FILE_VALID(one))
1548                                 diff_populate_filespec(one, 0);
1549                         if (one->data)
1550                                 one->is_binary = buffer_is_binary(one->data,
1551                                                 one->size);
1552                         if (one->is_binary == -1)
1553                                 one->is_binary = 0;
1554                 }
1555         }
1556         return one->is_binary;
1557 }
1558
1559 static const struct userdiff_funcname *diff_funcname_pattern(struct diff_filespec *one)
1560 {
1561         diff_filespec_load_driver(one);
1562         return one->driver->funcname.pattern ? &one->driver->funcname : NULL;
1563 }
1564
1565 static const char *userdiff_word_regex(struct diff_filespec *one)
1566 {
1567         diff_filespec_load_driver(one);
1568         return one->driver->word_regex;
1569 }
1570
1571 void diff_set_mnemonic_prefix(struct diff_options *options, const char *a, const char *b)
1572 {
1573         if (!options->a_prefix)
1574                 options->a_prefix = a;
1575         if (!options->b_prefix)
1576                 options->b_prefix = b;
1577 }
1578
1579 static const char *get_textconv(struct diff_filespec *one)
1580 {
1581         if (!DIFF_FILE_VALID(one))
1582                 return NULL;
1583         if (!S_ISREG(one->mode))
1584                 return NULL;
1585         diff_filespec_load_driver(one);
1586         return one->driver->textconv;
1587 }
1588
1589 static void builtin_diff(const char *name_a,
1590                          const char *name_b,
1591                          struct diff_filespec *one,
1592                          struct diff_filespec *two,
1593                          const char *xfrm_msg,
1594                          struct diff_options *o,
1595                          int complete_rewrite)
1596 {
1597         mmfile_t mf1, mf2;
1598         const char *lbl[2];
1599         char *a_one, *b_two;
1600         const char *set = diff_get_color_opt(o, DIFF_METAINFO);
1601         const char *reset = diff_get_color_opt(o, DIFF_RESET);
1602         const char *a_prefix, *b_prefix;
1603         const char *textconv_one = NULL, *textconv_two = NULL;
1604
1605         if (DIFF_OPT_TST(o, SUBMODULE_LOG) &&
1606                         (!one->mode || S_ISGITLINK(one->mode)) &&
1607                         (!two->mode || S_ISGITLINK(two->mode))) {
1608                 const char *del = diff_get_color_opt(o, DIFF_FILE_OLD);
1609                 const char *add = diff_get_color_opt(o, DIFF_FILE_NEW);
1610                 show_submodule_summary(o->file, one ? one->path : two->path,
1611                                 one->sha1, two->sha1,
1612                                 del, add, reset);
1613                 return;
1614         }
1615
1616         if (DIFF_OPT_TST(o, ALLOW_TEXTCONV)) {
1617                 textconv_one = get_textconv(one);
1618                 textconv_two = get_textconv(two);
1619         }
1620
1621         diff_set_mnemonic_prefix(o, "a/", "b/");
1622         if (DIFF_OPT_TST(o, REVERSE_DIFF)) {
1623                 a_prefix = o->b_prefix;
1624                 b_prefix = o->a_prefix;
1625         } else {
1626                 a_prefix = o->a_prefix;
1627                 b_prefix = o->b_prefix;
1628         }
1629
1630         /* Never use a non-valid filename anywhere if at all possible */
1631         name_a = DIFF_FILE_VALID(one) ? name_a : name_b;
1632         name_b = DIFF_FILE_VALID(two) ? name_b : name_a;
1633
1634         a_one = quote_two(a_prefix, name_a + (*name_a == '/'));
1635         b_two = quote_two(b_prefix, name_b + (*name_b == '/'));
1636         lbl[0] = DIFF_FILE_VALID(one) ? a_one : "/dev/null";
1637         lbl[1] = DIFF_FILE_VALID(two) ? b_two : "/dev/null";
1638         fprintf(o->file, "%sdiff --git %s %s%s\n", set, a_one, b_two, reset);
1639         if (lbl[0][0] == '/') {
1640                 /* /dev/null */
1641                 fprintf(o->file, "%snew file mode %06o%s\n", set, two->mode, reset);
1642                 if (xfrm_msg && xfrm_msg[0])
1643                         fprintf(o->file, "%s%s%s\n", set, xfrm_msg, reset);
1644         }
1645         else if (lbl[1][0] == '/') {
1646                 fprintf(o->file, "%sdeleted file mode %06o%s\n", set, one->mode, reset);
1647                 if (xfrm_msg && xfrm_msg[0])
1648                         fprintf(o->file, "%s%s%s\n", set, xfrm_msg, reset);
1649         }
1650         else {
1651                 if (one->mode != two->mode) {
1652                         fprintf(o->file, "%sold mode %06o%s\n", set, one->mode, reset);
1653                         fprintf(o->file, "%snew mode %06o%s\n", set, two->mode, reset);
1654                 }
1655                 if (xfrm_msg && xfrm_msg[0])
1656                         fprintf(o->file, "%s%s%s\n", set, xfrm_msg, reset);
1657                 /*
1658                  * we do not run diff between different kind
1659                  * of objects.
1660                  */
1661                 if ((one->mode ^ two->mode) & S_IFMT)
1662                         goto free_ab_and_return;
1663                 if (complete_rewrite &&
1664                     (textconv_one || !diff_filespec_is_binary(one)) &&
1665                     (textconv_two || !diff_filespec_is_binary(two))) {
1666                         emit_rewrite_diff(name_a, name_b, one, two,
1667                                                 textconv_one, textconv_two, o);
1668                         o->found_changes = 1;
1669                         goto free_ab_and_return;
1670                 }
1671         }
1672
1673         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
1674                 die("unable to read files to diff");
1675
1676         if (!DIFF_OPT_TST(o, TEXT) &&
1677             ( (diff_filespec_is_binary(one) && !textconv_one) ||
1678               (diff_filespec_is_binary(two) && !textconv_two) )) {
1679                 /* Quite common confusing case */
1680                 if (mf1.size == mf2.size &&
1681                     !memcmp(mf1.ptr, mf2.ptr, mf1.size))
1682                         goto free_ab_and_return;
1683                 if (DIFF_OPT_TST(o, BINARY))
1684                         emit_binary_diff(o->file, &mf1, &mf2);
1685                 else
1686                         fprintf(o->file, "Binary files %s and %s differ\n",
1687                                 lbl[0], lbl[1]);
1688                 o->found_changes = 1;
1689         }
1690         else {
1691                 /* Crazy xdl interfaces.. */
1692                 const char *diffopts = getenv("GIT_DIFF_OPTS");
1693                 xpparam_t xpp;
1694                 xdemitconf_t xecfg;
1695                 xdemitcb_t ecb;
1696                 struct emit_callback ecbdata;
1697                 const struct userdiff_funcname *pe;
1698
1699                 if (textconv_one) {
1700                         size_t size;
1701                         mf1.ptr = run_textconv(textconv_one, one, &size);
1702                         if (!mf1.ptr)
1703                                 die("unable to read files to diff");
1704                         mf1.size = size;
1705                 }
1706                 if (textconv_two) {
1707                         size_t size;
1708                         mf2.ptr = run_textconv(textconv_two, two, &size);
1709                         if (!mf2.ptr)
1710                                 die("unable to read files to diff");
1711                         mf2.size = size;
1712                 }
1713
1714                 pe = diff_funcname_pattern(one);
1715                 if (!pe)
1716                         pe = diff_funcname_pattern(two);
1717
1718                 memset(&xpp, 0, sizeof(xpp));
1719                 memset(&xecfg, 0, sizeof(xecfg));
1720                 memset(&ecbdata, 0, sizeof(ecbdata));
1721                 ecbdata.label_path = lbl;
1722                 ecbdata.color_diff = DIFF_OPT_TST(o, COLOR_DIFF);
1723                 ecbdata.found_changesp = &o->found_changes;
1724                 ecbdata.ws_rule = whitespace_rule(name_b ? name_b : name_a);
1725                 if (ecbdata.ws_rule & WS_BLANK_AT_EOF)
1726                         check_blank_at_eof(&mf1, &mf2, &ecbdata);
1727                 ecbdata.file = o->file;
1728                 xpp.flags = XDF_NEED_MINIMAL | o->xdl_opts;
1729                 xecfg.ctxlen = o->context;
1730                 xecfg.interhunkctxlen = o->interhunkcontext;
1731                 xecfg.flags = XDL_EMIT_FUNCNAMES;
1732                 if (pe)
1733                         xdiff_set_find_func(&xecfg, pe->pattern, pe->cflags);
1734                 if (!diffopts)
1735                         ;
1736                 else if (!prefixcmp(diffopts, "--unified="))
1737                         xecfg.ctxlen = strtoul(diffopts + 10, NULL, 10);
1738                 else if (!prefixcmp(diffopts, "-u"))
1739                         xecfg.ctxlen = strtoul(diffopts + 2, NULL, 10);
1740                 if (DIFF_OPT_TST(o, COLOR_DIFF_WORDS)) {
1741                         ecbdata.diff_words =
1742                                 xcalloc(1, sizeof(struct diff_words_data));
1743                         ecbdata.diff_words->file = o->file;
1744                         if (!o->word_regex)
1745                                 o->word_regex = userdiff_word_regex(one);
1746                         if (!o->word_regex)
1747                                 o->word_regex = userdiff_word_regex(two);
1748                         if (!o->word_regex)
1749                                 o->word_regex = diff_word_regex_cfg;
1750                         if (o->word_regex) {
1751                                 ecbdata.diff_words->word_regex = (regex_t *)
1752                                         xmalloc(sizeof(regex_t));
1753                                 if (regcomp(ecbdata.diff_words->word_regex,
1754                                                 o->word_regex,
1755                                                 REG_EXTENDED | REG_NEWLINE))
1756                                         die ("Invalid regular expression: %s",
1757                                                         o->word_regex);
1758                         }
1759                 }
1760                 xdi_diff_outf(&mf1, &mf2, fn_out_consume, &ecbdata,
1761                               &xpp, &xecfg, &ecb);
1762                 if (DIFF_OPT_TST(o, COLOR_DIFF_WORDS))
1763                         free_diff_words_data(&ecbdata);
1764                 if (textconv_one)
1765                         free(mf1.ptr);
1766                 if (textconv_two)
1767                         free(mf2.ptr);
1768                 xdiff_clear_find_func(&xecfg);
1769         }
1770
1771  free_ab_and_return:
1772         diff_free_filespec_data(one);
1773         diff_free_filespec_data(two);
1774         free(a_one);
1775         free(b_two);
1776         return;
1777 }
1778
1779 static void builtin_diffstat(const char *name_a, const char *name_b,
1780                              struct diff_filespec *one,
1781                              struct diff_filespec *two,
1782                              struct diffstat_t *diffstat,
1783                              struct diff_options *o,
1784                              int complete_rewrite)
1785 {
1786         mmfile_t mf1, mf2;
1787         struct diffstat_file *data;
1788
1789         data = diffstat_add(diffstat, name_a, name_b);
1790
1791         if (!one || !two) {
1792                 data->is_unmerged = 1;
1793                 return;
1794         }
1795         if (complete_rewrite) {
1796                 diff_populate_filespec(one, 0);
1797                 diff_populate_filespec(two, 0);
1798                 data->deleted = count_lines(one->data, one->size);
1799                 data->added = count_lines(two->data, two->size);
1800                 goto free_and_return;
1801         }
1802         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
1803                 die("unable to read files to diff");
1804
1805         if (diff_filespec_is_binary(one) || diff_filespec_is_binary(two)) {
1806                 data->is_binary = 1;
1807                 data->added = mf2.size;
1808                 data->deleted = mf1.size;
1809         } else {
1810                 /* Crazy xdl interfaces.. */
1811                 xpparam_t xpp;
1812                 xdemitconf_t xecfg;
1813                 xdemitcb_t ecb;
1814
1815                 memset(&xpp, 0, sizeof(xpp));
1816                 memset(&xecfg, 0, sizeof(xecfg));
1817                 xpp.flags = XDF_NEED_MINIMAL | o->xdl_opts;
1818                 xdi_diff_outf(&mf1, &mf2, diffstat_consume, diffstat,
1819                               &xpp, &xecfg, &ecb);
1820         }
1821
1822  free_and_return:
1823         diff_free_filespec_data(one);
1824         diff_free_filespec_data(two);
1825 }
1826
1827 static void builtin_checkdiff(const char *name_a, const char *name_b,
1828                               const char *attr_path,
1829                               struct diff_filespec *one,
1830                               struct diff_filespec *two,
1831                               struct diff_options *o)
1832 {
1833         mmfile_t mf1, mf2;
1834         struct checkdiff_t data;
1835
1836         if (!two)
1837                 return;
1838
1839         memset(&data, 0, sizeof(data));
1840         data.filename = name_b ? name_b : name_a;
1841         data.lineno = 0;
1842         data.o = o;
1843         data.ws_rule = whitespace_rule(attr_path);
1844
1845         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
1846                 die("unable to read files to diff");
1847
1848         /*
1849          * All the other codepaths check both sides, but not checking
1850          * the "old" side here is deliberate.  We are checking the newly
1851          * introduced changes, and as long as the "new" side is text, we
1852          * can and should check what it introduces.
1853          */
1854         if (diff_filespec_is_binary(two))
1855                 goto free_and_return;
1856         else {
1857                 /* Crazy xdl interfaces.. */
1858                 xpparam_t xpp;
1859                 xdemitconf_t xecfg;
1860                 xdemitcb_t ecb;
1861
1862                 memset(&xpp, 0, sizeof(xpp));
1863                 memset(&xecfg, 0, sizeof(xecfg));
1864                 xecfg.ctxlen = 1; /* at least one context line */
1865                 xpp.flags = XDF_NEED_MINIMAL;
1866                 xdi_diff_outf(&mf1, &mf2, checkdiff_consume, &data,
1867                               &xpp, &xecfg, &ecb);
1868
1869                 if (data.ws_rule & WS_BLANK_AT_EOF) {
1870                         struct emit_callback ecbdata;
1871                         int blank_at_eof;
1872
1873                         ecbdata.ws_rule = data.ws_rule;
1874                         check_blank_at_eof(&mf1, &mf2, &ecbdata);
1875                         blank_at_eof = ecbdata.blank_at_eof_in_preimage;
1876
1877                         if (blank_at_eof) {
1878                                 static char *err;
1879                                 if (!err)
1880                                         err = whitespace_error_string(WS_BLANK_AT_EOF);
1881                                 fprintf(o->file, "%s:%d: %s.\n",
1882                                         data.filename, blank_at_eof, err);
1883                                 data.status = 1; /* report errors */
1884                         }
1885                 }
1886         }
1887  free_and_return:
1888         diff_free_filespec_data(one);
1889         diff_free_filespec_data(two);
1890         if (data.status)
1891                 DIFF_OPT_SET(o, CHECK_FAILED);
1892 }
1893
1894 struct diff_filespec *alloc_filespec(const char *path)
1895 {
1896         int namelen = strlen(path);
1897         struct diff_filespec *spec = xmalloc(sizeof(*spec) + namelen + 1);
1898
1899         memset(spec, 0, sizeof(*spec));
1900         spec->path = (char *)(spec + 1);
1901         memcpy(spec->path, path, namelen+1);
1902         spec->count = 1;
1903         spec->is_binary = -1;
1904         return spec;
1905 }
1906
1907 void free_filespec(struct diff_filespec *spec)
1908 {
1909         if (!--spec->count) {
1910                 diff_free_filespec_data(spec);
1911                 free(spec);
1912         }
1913 }
1914
1915 void fill_filespec(struct diff_filespec *spec, const unsigned char *sha1,
1916                    unsigned short mode)
1917 {
1918         if (mode) {
1919                 spec->mode = canon_mode(mode);
1920                 hashcpy(spec->sha1, sha1);
1921                 spec->sha1_valid = !is_null_sha1(sha1);
1922         }
1923 }
1924
1925 /*
1926  * Given a name and sha1 pair, if the index tells us the file in
1927  * the work tree has that object contents, return true, so that
1928  * prepare_temp_file() does not have to inflate and extract.
1929  */
1930 static int reuse_worktree_file(const char *name, const unsigned char *sha1, int want_file)
1931 {
1932         struct cache_entry *ce;
1933         struct stat st;
1934         int pos, len;
1935
1936         /*
1937          * We do not read the cache ourselves here, because the
1938          * benchmark with my previous version that always reads cache
1939          * shows that it makes things worse for diff-tree comparing
1940          * two linux-2.6 kernel trees in an already checked out work
1941          * tree.  This is because most diff-tree comparisons deal with
1942          * only a small number of files, while reading the cache is
1943          * expensive for a large project, and its cost outweighs the
1944          * savings we get by not inflating the object to a temporary
1945          * file.  Practically, this code only helps when we are used
1946          * by diff-cache --cached, which does read the cache before
1947          * calling us.
1948          */
1949         if (!active_cache)
1950                 return 0;
1951
1952         /* We want to avoid the working directory if our caller
1953          * doesn't need the data in a normal file, this system
1954          * is rather slow with its stat/open/mmap/close syscalls,
1955          * and the object is contained in a pack file.  The pack
1956          * is probably already open and will be faster to obtain
1957          * the data through than the working directory.  Loose
1958          * objects however would tend to be slower as they need
1959          * to be individually opened and inflated.
1960          */
1961         if (!FAST_WORKING_DIRECTORY && !want_file && has_sha1_pack(sha1))
1962                 return 0;
1963
1964         len = strlen(name);
1965         pos = cache_name_pos(name, len);
1966         if (pos < 0)
1967                 return 0;
1968         ce = active_cache[pos];
1969
1970         /*
1971          * This is not the sha1 we are looking for, or
1972          * unreusable because it is not a regular file.
1973          */
1974         if (hashcmp(sha1, ce->sha1) || !S_ISREG(ce->ce_mode))
1975                 return 0;
1976
1977         /*
1978          * If ce is marked as "assume unchanged", there is no
1979          * guarantee that work tree matches what we are looking for.
1980          */
1981         if (ce->ce_flags & CE_VALID)
1982                 return 0;
1983
1984         /*
1985          * If ce matches the file in the work tree, we can reuse it.
1986          */
1987         if (ce_uptodate(ce) ||
1988             (!lstat(name, &st) && !ce_match_stat(ce, &st, 0)))
1989                 return 1;
1990
1991         return 0;
1992 }
1993
1994 static int populate_from_stdin(struct diff_filespec *s)
1995 {
1996         struct strbuf buf = STRBUF_INIT;
1997         size_t size = 0;
1998
1999         if (strbuf_read(&buf, 0, 0) < 0)
2000                 return error("error while reading from stdin %s",
2001                                      strerror(errno));
2002
2003         s->should_munmap = 0;
2004         s->data = strbuf_detach(&buf, &size);
2005         s->size = size;
2006         s->should_free = 1;
2007         return 0;
2008 }
2009
2010 static int diff_populate_gitlink(struct diff_filespec *s, int size_only)
2011 {
2012         int len;
2013         char *data = xmalloc(100);
2014         len = snprintf(data, 100,
2015                 "Subproject commit %s\n", sha1_to_hex(s->sha1));
2016         s->data = data;
2017         s->size = len;
2018         s->should_free = 1;
2019         if (size_only) {
2020                 s->data = NULL;
2021                 free(data);
2022         }
2023         return 0;
2024 }
2025
2026 /*
2027  * While doing rename detection and pickaxe operation, we may need to
2028  * grab the data for the blob (or file) for our own in-core comparison.
2029  * diff_filespec has data and size fields for this purpose.
2030  */
2031 int diff_populate_filespec(struct diff_filespec *s, int size_only)
2032 {
2033         int err = 0;
2034         if (!DIFF_FILE_VALID(s))
2035                 die("internal error: asking to populate invalid file.");
2036         if (S_ISDIR(s->mode))
2037                 return -1;
2038
2039         if (s->data)
2040                 return 0;
2041
2042         if (size_only && 0 < s->size)
2043                 return 0;
2044
2045         if (S_ISGITLINK(s->mode))
2046                 return diff_populate_gitlink(s, size_only);
2047
2048         if (!s->sha1_valid ||
2049             reuse_worktree_file(s->path, s->sha1, 0)) {
2050                 struct strbuf buf = STRBUF_INIT;
2051                 struct stat st;
2052                 int fd;
2053
2054                 if (!strcmp(s->path, "-"))
2055                         return populate_from_stdin(s);
2056
2057                 if (lstat(s->path, &st) < 0) {
2058                         if (errno == ENOENT) {
2059                         err_empty:
2060                                 err = -1;
2061                         empty:
2062                                 s->data = (char *)"";
2063                                 s->size = 0;
2064                                 return err;
2065                         }
2066                 }
2067                 s->size = xsize_t(st.st_size);
2068                 if (!s->size)
2069                         goto empty;
2070                 if (S_ISLNK(st.st_mode)) {
2071                         struct strbuf sb = STRBUF_INIT;
2072
2073                         if (strbuf_readlink(&sb, s->path, s->size))
2074                                 goto err_empty;
2075                         s->size = sb.len;
2076                         s->data = strbuf_detach(&sb, NULL);
2077                         s->should_free = 1;
2078                         return 0;
2079                 }
2080                 if (size_only)
2081                         return 0;
2082                 fd = open(s->path, O_RDONLY);
2083                 if (fd < 0)
2084                         goto err_empty;
2085                 s->data = xmmap(NULL, s->size, PROT_READ, MAP_PRIVATE, fd, 0);
2086                 close(fd);
2087                 s->should_munmap = 1;
2088
2089                 /*
2090                  * Convert from working tree format to canonical git format
2091                  */
2092                 if (convert_to_git(s->path, s->data, s->size, &buf, safe_crlf)) {
2093                         size_t size = 0;
2094                         munmap(s->data, s->size);
2095                         s->should_munmap = 0;
2096                         s->data = strbuf_detach(&buf, &size);
2097                         s->size = size;
2098                         s->should_free = 1;
2099                 }
2100         }
2101         else {
2102                 enum object_type type;
2103                 if (size_only)
2104                         type = sha1_object_info(s->sha1, &s->size);
2105                 else {
2106                         s->data = read_sha1_file(s->sha1, &type, &s->size);
2107                         s->should_free = 1;
2108                 }
2109         }
2110         return 0;
2111 }
2112
2113 void diff_free_filespec_blob(struct diff_filespec *s)
2114 {
2115         if (s->should_free)
2116                 free(s->data);
2117         else if (s->should_munmap)
2118                 munmap(s->data, s->size);
2119
2120         if (s->should_free || s->should_munmap) {
2121                 s->should_free = s->should_munmap = 0;
2122                 s->data = NULL;
2123         }
2124 }
2125
2126 void diff_free_filespec_data(struct diff_filespec *s)
2127 {
2128         diff_free_filespec_blob(s);
2129         free(s->cnt_data);
2130         s->cnt_data = NULL;
2131 }
2132
2133 static void prep_temp_blob(const char *path, struct diff_tempfile *temp,
2134                            void *blob,
2135                            unsigned long size,
2136                            const unsigned char *sha1,
2137                            int mode)
2138 {
2139         int fd;
2140         struct strbuf buf = STRBUF_INIT;
2141         struct strbuf template = STRBUF_INIT;
2142         char *path_dup = xstrdup(path);
2143         const char *base = basename(path_dup);
2144
2145         /* Generate "XXXXXX_basename.ext" */
2146         strbuf_addstr(&template, "XXXXXX_");
2147         strbuf_addstr(&template, base);
2148
2149         fd = git_mkstemps(temp->tmp_path, PATH_MAX, template.buf,
2150                         strlen(base) + 1);
2151         if (fd < 0)
2152                 die_errno("unable to create temp-file");
2153         if (convert_to_working_tree(path,
2154                         (const char *)blob, (size_t)size, &buf)) {
2155                 blob = buf.buf;
2156                 size = buf.len;
2157         }
2158         if (write_in_full(fd, blob, size) != size)
2159                 die_errno("unable to write temp-file");
2160         close(fd);
2161         temp->name = temp->tmp_path;
2162         strcpy(temp->hex, sha1_to_hex(sha1));
2163         temp->hex[40] = 0;
2164         sprintf(temp->mode, "%06o", mode);
2165         strbuf_release(&buf);
2166         strbuf_release(&template);
2167         free(path_dup);
2168 }
2169
2170 static struct diff_tempfile *prepare_temp_file(const char *name,
2171                 struct diff_filespec *one)
2172 {
2173         struct diff_tempfile *temp = claim_diff_tempfile();
2174
2175         if (!DIFF_FILE_VALID(one)) {
2176         not_a_valid_file:
2177                 /* A '-' entry produces this for file-2, and
2178                  * a '+' entry produces this for file-1.
2179                  */
2180                 temp->name = "/dev/null";
2181                 strcpy(temp->hex, ".");
2182                 strcpy(temp->mode, ".");
2183                 return temp;
2184         }
2185
2186         if (!remove_tempfile_installed) {
2187                 atexit(remove_tempfile);
2188                 sigchain_push_common(remove_tempfile_on_signal);
2189                 remove_tempfile_installed = 1;
2190         }
2191
2192         if (!one->sha1_valid ||
2193             reuse_worktree_file(name, one->sha1, 1)) {
2194                 struct stat st;
2195                 if (lstat(name, &st) < 0) {
2196                         if (errno == ENOENT)
2197                                 goto not_a_valid_file;
2198                         die_errno("stat(%s)", name);
2199                 }
2200                 if (S_ISLNK(st.st_mode)) {
2201                         struct strbuf sb = STRBUF_INIT;
2202                         if (strbuf_readlink(&sb, name, st.st_size) < 0)
2203                                 die_errno("readlink(%s)", name);
2204                         prep_temp_blob(name, temp, sb.buf, sb.len,
2205                                        (one->sha1_valid ?
2206                                         one->sha1 : null_sha1),
2207                                        (one->sha1_valid ?
2208                                         one->mode : S_IFLNK));
2209                         strbuf_release(&sb);
2210                 }
2211                 else {
2212                         /* we can borrow from the file in the work tree */
2213                         temp->name = name;
2214                         if (!one->sha1_valid)
2215                                 strcpy(temp->hex, sha1_to_hex(null_sha1));
2216                         else
2217                                 strcpy(temp->hex, sha1_to_hex(one->sha1));
2218                         /* Even though we may sometimes borrow the
2219                          * contents from the work tree, we always want
2220                          * one->mode.  mode is trustworthy even when
2221                          * !(one->sha1_valid), as long as
2222                          * DIFF_FILE_VALID(one).
2223                          */
2224                         sprintf(temp->mode, "%06o", one->mode);
2225                 }
2226                 return temp;
2227         }
2228         else {
2229                 if (diff_populate_filespec(one, 0))
2230                         die("cannot read data blob for %s", one->path);
2231                 prep_temp_blob(name, temp, one->data, one->size,
2232                                one->sha1, one->mode);
2233         }
2234         return temp;
2235 }
2236
2237 /* An external diff command takes:
2238  *
2239  * diff-cmd name infile1 infile1-sha1 infile1-mode \
2240  *               infile2 infile2-sha1 infile2-mode [ rename-to ]
2241  *
2242  */
2243 static void run_external_diff(const char *pgm,
2244                               const char *name,
2245                               const char *other,
2246                               struct diff_filespec *one,
2247                               struct diff_filespec *two,
2248                               const char *xfrm_msg,
2249                               int complete_rewrite)
2250 {
2251         const char *spawn_arg[10];
2252         int retval;
2253         const char **arg = &spawn_arg[0];
2254
2255         if (one && two) {
2256                 struct diff_tempfile *temp_one, *temp_two;
2257                 const char *othername = (other ? other : name);
2258                 temp_one = prepare_temp_file(name, one);
2259                 temp_two = prepare_temp_file(othername, two);
2260                 *arg++ = pgm;
2261                 *arg++ = name;
2262                 *arg++ = temp_one->name;
2263                 *arg++ = temp_one->hex;
2264                 *arg++ = temp_one->mode;
2265                 *arg++ = temp_two->name;
2266                 *arg++ = temp_two->hex;
2267                 *arg++ = temp_two->mode;
2268                 if (other) {
2269                         *arg++ = other;
2270                         *arg++ = xfrm_msg;
2271                 }
2272         } else {
2273                 *arg++ = pgm;
2274                 *arg++ = name;
2275         }
2276         *arg = NULL;
2277         fflush(NULL);
2278         retval = run_command_v_opt(spawn_arg, 0);
2279         remove_tempfile();
2280         if (retval) {
2281                 fprintf(stderr, "external diff died, stopping at %s.\n", name);
2282                 exit(1);
2283         }
2284 }
2285
2286 static int similarity_index(struct diff_filepair *p)
2287 {
2288         return p->score * 100 / MAX_SCORE;
2289 }
2290
2291 static void fill_metainfo(struct strbuf *msg,
2292                           const char *name,
2293                           const char *other,
2294                           struct diff_filespec *one,
2295                           struct diff_filespec *two,
2296                           struct diff_options *o,
2297                           struct diff_filepair *p)
2298 {
2299         strbuf_init(msg, PATH_MAX * 2 + 300);
2300         switch (p->status) {
2301         case DIFF_STATUS_COPIED:
2302                 strbuf_addf(msg, "similarity index %d%%", similarity_index(p));
2303                 strbuf_addstr(msg, "\ncopy from ");
2304                 quote_c_style(name, msg, NULL, 0);
2305                 strbuf_addstr(msg, "\ncopy to ");
2306                 quote_c_style(other, msg, NULL, 0);
2307                 strbuf_addch(msg, '\n');
2308                 break;
2309         case DIFF_STATUS_RENAMED:
2310                 strbuf_addf(msg, "similarity index %d%%", similarity_index(p));
2311                 strbuf_addstr(msg, "\nrename from ");
2312                 quote_c_style(name, msg, NULL, 0);
2313                 strbuf_addstr(msg, "\nrename to ");
2314                 quote_c_style(other, msg, NULL, 0);
2315                 strbuf_addch(msg, '\n');
2316                 break;
2317         case DIFF_STATUS_MODIFIED:
2318                 if (p->score) {
2319                         strbuf_addf(msg, "dissimilarity index %d%%\n",
2320                                     similarity_index(p));
2321                         break;
2322                 }
2323                 /* fallthru */
2324         default:
2325                 /* nothing */
2326                 ;
2327         }
2328         if (one && two && hashcmp(one->sha1, two->sha1)) {
2329                 int abbrev = DIFF_OPT_TST(o, FULL_INDEX) ? 40 : DEFAULT_ABBREV;
2330
2331                 if (DIFF_OPT_TST(o, BINARY)) {
2332                         mmfile_t mf;
2333                         if ((!fill_mmfile(&mf, one) && diff_filespec_is_binary(one)) ||
2334                             (!fill_mmfile(&mf, two) && diff_filespec_is_binary(two)))
2335                                 abbrev = 40;
2336                 }
2337                 strbuf_addf(msg, "index %.*s..%.*s",
2338                             abbrev, sha1_to_hex(one->sha1),
2339                             abbrev, sha1_to_hex(two->sha1));
2340                 if (one->mode == two->mode)
2341                         strbuf_addf(msg, " %06o", one->mode);
2342                 strbuf_addch(msg, '\n');
2343         }
2344         if (msg->len)
2345                 strbuf_setlen(msg, msg->len - 1);
2346 }
2347
2348 static void run_diff_cmd(const char *pgm,
2349                          const char *name,
2350                          const char *other,
2351                          const char *attr_path,
2352                          struct diff_filespec *one,
2353                          struct diff_filespec *two,
2354                          struct strbuf *msg,
2355                          struct diff_options *o,
2356                          struct diff_filepair *p)
2357 {
2358         const char *xfrm_msg = NULL;
2359         int complete_rewrite = (p->status == DIFF_STATUS_MODIFIED) && p->score;
2360
2361         if (msg) {
2362                 fill_metainfo(msg, name, other, one, two, o, p);
2363                 xfrm_msg = msg->len ? msg->buf : NULL;
2364         }
2365
2366         if (!DIFF_OPT_TST(o, ALLOW_EXTERNAL))
2367                 pgm = NULL;
2368         else {
2369                 struct userdiff_driver *drv = userdiff_find_by_path(attr_path);
2370                 if (drv && drv->external)
2371                         pgm = drv->external;
2372         }
2373
2374         if (pgm) {
2375                 run_external_diff(pgm, name, other, one, two, xfrm_msg,
2376                                   complete_rewrite);
2377                 return;
2378         }
2379         if (one && two)
2380                 builtin_diff(name, other ? other : name,
2381                              one, two, xfrm_msg, o, complete_rewrite);
2382         else
2383                 fprintf(o->file, "* Unmerged path %s\n", name);
2384 }
2385
2386 static void diff_fill_sha1_info(struct diff_filespec *one)
2387 {
2388         if (DIFF_FILE_VALID(one)) {
2389                 if (!one->sha1_valid) {
2390                         struct stat st;
2391                         if (!strcmp(one->path, "-")) {
2392                                 hashcpy(one->sha1, null_sha1);
2393                                 return;
2394                         }
2395                         if (lstat(one->path, &st) < 0)
2396                                 die_errno("stat '%s'", one->path);
2397                         if (index_path(one->sha1, one->path, &st, 0))
2398                                 die("cannot hash %s", one->path);
2399                 }
2400         }
2401         else
2402                 hashclr(one->sha1);
2403 }
2404
2405 static void strip_prefix(int prefix_length, const char **namep, const char **otherp)
2406 {
2407         /* Strip the prefix but do not molest /dev/null and absolute paths */
2408         if (*namep && **namep != '/')
2409                 *namep += prefix_length;
2410         if (*otherp && **otherp != '/')
2411                 *otherp += prefix_length;
2412 }
2413
2414 static void run_diff(struct diff_filepair *p, struct diff_options *o)
2415 {
2416         const char *pgm = external_diff();
2417         struct strbuf msg;
2418         struct diff_filespec *one = p->one;
2419         struct diff_filespec *two = p->two;
2420         const char *name;
2421         const char *other;
2422         const char *attr_path;
2423
2424         name  = p->one->path;
2425         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
2426         attr_path = name;
2427         if (o->prefix_length)
2428                 strip_prefix(o->prefix_length, &name, &other);
2429
2430         if (DIFF_PAIR_UNMERGED(p)) {
2431                 run_diff_cmd(pgm, name, NULL, attr_path,
2432                              NULL, NULL, NULL, o, p);
2433                 return;
2434         }
2435
2436         diff_fill_sha1_info(one);
2437         diff_fill_sha1_info(two);
2438
2439         if (!pgm &&
2440             DIFF_FILE_VALID(one) && DIFF_FILE_VALID(two) &&
2441             (S_IFMT & one->mode) != (S_IFMT & two->mode)) {
2442                 /*
2443                  * a filepair that changes between file and symlink
2444                  * needs to be split into deletion and creation.
2445                  */
2446                 struct diff_filespec *null = alloc_filespec(two->path);
2447                 run_diff_cmd(NULL, name, other, attr_path,
2448                              one, null, &msg, o, p);
2449                 free(null);
2450                 strbuf_release(&msg);
2451
2452                 null = alloc_filespec(one->path);
2453                 run_diff_cmd(NULL, name, other, attr_path,
2454                              null, two, &msg, o, p);
2455                 free(null);
2456         }
2457         else
2458                 run_diff_cmd(pgm, name, other, attr_path,
2459                              one, two, &msg, o, p);
2460
2461         strbuf_release(&msg);
2462 }
2463
2464 static void run_diffstat(struct diff_filepair *p, struct diff_options *o,
2465                          struct diffstat_t *diffstat)
2466 {
2467         const char *name;
2468         const char *other;
2469         int complete_rewrite = 0;
2470
2471         if (DIFF_PAIR_UNMERGED(p)) {
2472                 /* unmerged */
2473                 builtin_diffstat(p->one->path, NULL, NULL, NULL, diffstat, o, 0);
2474                 return;
2475         }
2476
2477         name = p->one->path;
2478         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
2479
2480         if (o->prefix_length)
2481                 strip_prefix(o->prefix_length, &name, &other);
2482
2483         diff_fill_sha1_info(p->one);
2484         diff_fill_sha1_info(p->two);
2485
2486         if (p->status == DIFF_STATUS_MODIFIED && p->score)
2487                 complete_rewrite = 1;
2488         builtin_diffstat(name, other, p->one, p->two, diffstat, o, complete_rewrite);
2489 }
2490
2491 static void run_checkdiff(struct diff_filepair *p, struct diff_options *o)
2492 {
2493         const char *name;
2494         const char *other;
2495         const char *attr_path;
2496
2497         if (DIFF_PAIR_UNMERGED(p)) {
2498                 /* unmerged */
2499                 return;
2500         }
2501
2502         name = p->one->path;
2503         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
2504         attr_path = other ? other : name;
2505
2506         if (o->prefix_length)
2507                 strip_prefix(o->prefix_length, &name, &other);
2508
2509         diff_fill_sha1_info(p->one);
2510         diff_fill_sha1_info(p->two);
2511
2512         builtin_checkdiff(name, other, attr_path, p->one, p->two, o);
2513 }
2514
2515 void diff_setup(struct diff_options *options)
2516 {
2517         memset(options, 0, sizeof(*options));
2518
2519         options->file = stdout;
2520
2521         options->line_termination = '\n';
2522         options->break_opt = -1;
2523         options->rename_limit = -1;
2524         options->dirstat_percent = 3;
2525         options->context = 3;
2526
2527         options->change = diff_change;
2528         options->add_remove = diff_addremove;
2529         if (diff_use_color_default > 0)
2530                 DIFF_OPT_SET(options, COLOR_DIFF);
2531         options->detect_rename = diff_detect_rename_default;
2532
2533         if (!diff_mnemonic_prefix) {
2534                 options->a_prefix = "a/";
2535                 options->b_prefix = "b/";
2536         }
2537 }
2538
2539 int diff_setup_done(struct diff_options *options)
2540 {
2541         int count = 0;
2542
2543         if (options->output_format & DIFF_FORMAT_NAME)
2544                 count++;
2545         if (options->output_format & DIFF_FORMAT_NAME_STATUS)
2546                 count++;
2547         if (options->output_format & DIFF_FORMAT_CHECKDIFF)
2548                 count++;
2549         if (options->output_format & DIFF_FORMAT_NO_OUTPUT)
2550                 count++;
2551         if (count > 1)
2552                 die("--name-only, --name-status, --check and -s are mutually exclusive");
2553
2554         if (DIFF_OPT_TST(options, FIND_COPIES_HARDER))
2555                 options->detect_rename = DIFF_DETECT_COPY;
2556
2557         if (!DIFF_OPT_TST(options, RELATIVE_NAME))
2558                 options->prefix = NULL;
2559         if (options->prefix)
2560                 options->prefix_length = strlen(options->prefix);
2561         else
2562                 options->prefix_length = 0;
2563
2564         if (options->output_format & (DIFF_FORMAT_NAME |
2565                                       DIFF_FORMAT_NAME_STATUS |
2566                                       DIFF_FORMAT_CHECKDIFF |
2567                                       DIFF_FORMAT_NO_OUTPUT))
2568                 options->output_format &= ~(DIFF_FORMAT_RAW |
2569                                             DIFF_FORMAT_NUMSTAT |
2570                                             DIFF_FORMAT_DIFFSTAT |
2571                                             DIFF_FORMAT_SHORTSTAT |
2572                                             DIFF_FORMAT_DIRSTAT |
2573                                             DIFF_FORMAT_SUMMARY |
2574                                             DIFF_FORMAT_PATCH);
2575
2576         /*
2577          * These cases always need recursive; we do not drop caller-supplied
2578          * recursive bits for other formats here.
2579          */
2580         if (options->output_format & (DIFF_FORMAT_PATCH |
2581                                       DIFF_FORMAT_NUMSTAT |
2582                                       DIFF_FORMAT_DIFFSTAT |
2583                                       DIFF_FORMAT_SHORTSTAT |
2584                                       DIFF_FORMAT_DIRSTAT |
2585                                       DIFF_FORMAT_SUMMARY |
2586                                       DIFF_FORMAT_CHECKDIFF))
2587                 DIFF_OPT_SET(options, RECURSIVE);
2588         /*
2589          * Also pickaxe would not work very well if you do not say recursive
2590          */
2591         if (options->pickaxe)
2592                 DIFF_OPT_SET(options, RECURSIVE);
2593
2594         if (options->detect_rename && options->rename_limit < 0)
2595                 options->rename_limit = diff_rename_limit_default;
2596         if (options->setup & DIFF_SETUP_USE_CACHE) {
2597                 if (!active_cache)
2598                         /* read-cache does not die even when it fails
2599                          * so it is safe for us to do this here.  Also
2600                          * it does not smudge active_cache or active_nr
2601                          * when it fails, so we do not have to worry about
2602                          * cleaning it up ourselves either.
2603                          */
2604                         read_cache();
2605         }
2606         if (options->abbrev <= 0 || 40 < options->abbrev)
2607                 options->abbrev = 40; /* full */
2608
2609         /*
2610          * It does not make sense to show the first hit we happened
2611          * to have found.  It does not make sense not to return with
2612          * exit code in such a case either.
2613          */
2614         if (DIFF_OPT_TST(options, QUIET)) {
2615                 options->output_format = DIFF_FORMAT_NO_OUTPUT;
2616                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
2617         }
2618
2619         return 0;
2620 }
2621
2622 static int opt_arg(const char *arg, int arg_short, const char *arg_long, int *val)
2623 {
2624         char c, *eq;
2625         int len;
2626
2627         if (*arg != '-')
2628                 return 0;
2629         c = *++arg;
2630         if (!c)
2631                 return 0;
2632         if (c == arg_short) {
2633                 c = *++arg;
2634                 if (!c)
2635                         return 1;
2636                 if (val && isdigit(c)) {
2637                         char *end;
2638                         int n = strtoul(arg, &end, 10);
2639                         if (*end)
2640                                 return 0;
2641                         *val = n;
2642                         return 1;
2643                 }
2644                 return 0;
2645         }
2646         if (c != '-')
2647                 return 0;
2648         arg++;
2649         eq = strchr(arg, '=');
2650         if (eq)
2651                 len = eq - arg;
2652         else
2653                 len = strlen(arg);
2654         if (!len || strncmp(arg, arg_long, len))
2655                 return 0;
2656         if (eq) {
2657                 int n;
2658                 char *end;
2659                 if (!isdigit(*++eq))
2660                         return 0;
2661                 n = strtoul(eq, &end, 10);
2662                 if (*end)
2663                         return 0;
2664                 *val = n;
2665         }
2666         return 1;
2667 }
2668
2669 static int diff_scoreopt_parse(const char *opt);
2670
2671 int diff_opt_parse(struct diff_options *options, const char **av, int ac)
2672 {
2673         const char *arg = av[0];
2674
2675         /* Output format options */
2676         if (!strcmp(arg, "-p") || !strcmp(arg, "-u"))
2677                 options->output_format |= DIFF_FORMAT_PATCH;
2678         else if (opt_arg(arg, 'U', "unified", &options->context))
2679                 options->output_format |= DIFF_FORMAT_PATCH;
2680         else if (!strcmp(arg, "--raw"))
2681                 options->output_format |= DIFF_FORMAT_RAW;
2682         else if (!strcmp(arg, "--patch-with-raw"))
2683                 options->output_format |= DIFF_FORMAT_PATCH | DIFF_FORMAT_RAW;
2684         else if (!strcmp(arg, "--numstat"))
2685                 options->output_format |= DIFF_FORMAT_NUMSTAT;
2686         else if (!strcmp(arg, "--shortstat"))
2687                 options->output_format |= DIFF_FORMAT_SHORTSTAT;
2688         else if (opt_arg(arg, 'X', "dirstat", &options->dirstat_percent))
2689                 options->output_format |= DIFF_FORMAT_DIRSTAT;
2690         else if (!strcmp(arg, "--cumulative")) {
2691                 options->output_format |= DIFF_FORMAT_DIRSTAT;
2692                 DIFF_OPT_SET(options, DIRSTAT_CUMULATIVE);
2693         } else if (opt_arg(arg, 0, "dirstat-by-file",
2694                            &options->dirstat_percent)) {
2695                 options->output_format |= DIFF_FORMAT_DIRSTAT;
2696                 DIFF_OPT_SET(options, DIRSTAT_BY_FILE);
2697         }
2698         else if (!strcmp(arg, "--check"))
2699                 options->output_format |= DIFF_FORMAT_CHECKDIFF;
2700         else if (!strcmp(arg, "--summary"))
2701                 options->output_format |= DIFF_FORMAT_SUMMARY;
2702         else if (!strcmp(arg, "--patch-with-stat"))
2703                 options->output_format |= DIFF_FORMAT_PATCH | DIFF_FORMAT_DIFFSTAT;
2704         else if (!strcmp(arg, "--name-only"))
2705                 options->output_format |= DIFF_FORMAT_NAME;
2706         else if (!strcmp(arg, "--name-status"))
2707                 options->output_format |= DIFF_FORMAT_NAME_STATUS;
2708         else if (!strcmp(arg, "-s"))
2709                 options->output_format |= DIFF_FORMAT_NO_OUTPUT;
2710         else if (!prefixcmp(arg, "--stat")) {
2711                 char *end;
2712                 int width = options->stat_width;
2713                 int name_width = options->stat_name_width;
2714                 arg += 6;
2715                 end = (char *)arg;
2716
2717                 switch (*arg) {
2718                 case '-':
2719                         if (!prefixcmp(arg, "-width="))
2720                                 width = strtoul(arg + 7, &end, 10);
2721                         else if (!prefixcmp(arg, "-name-width="))
2722                                 name_width = strtoul(arg + 12, &end, 10);
2723                         break;
2724                 case '=':
2725                         width = strtoul(arg+1, &end, 10);
2726                         if (*end == ',')
2727                                 name_width = strtoul(end+1, &end, 10);
2728                 }
2729
2730                 /* Important! This checks all the error cases! */
2731                 if (*end)
2732                         return 0;
2733                 options->output_format |= DIFF_FORMAT_DIFFSTAT;
2734                 options->stat_name_width = name_width;
2735                 options->stat_width = width;
2736         }
2737
2738         /* renames options */
2739         else if (!prefixcmp(arg, "-B")) {
2740                 if ((options->break_opt = diff_scoreopt_parse(arg)) == -1)
2741                         return -1;
2742         }
2743         else if (!prefixcmp(arg, "-M")) {
2744                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
2745                         return -1;
2746                 options->detect_rename = DIFF_DETECT_RENAME;
2747         }
2748         else if (!prefixcmp(arg, "-C")) {
2749                 if (options->detect_rename == DIFF_DETECT_COPY)
2750                         DIFF_OPT_SET(options, FIND_COPIES_HARDER);
2751                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
2752                         return -1;
2753                 options->detect_rename = DIFF_DETECT_COPY;
2754         }
2755         else if (!strcmp(arg, "--no-renames"))
2756                 options->detect_rename = 0;
2757         else if (!strcmp(arg, "--relative"))
2758                 DIFF_OPT_SET(options, RELATIVE_NAME);
2759         else if (!prefixcmp(arg, "--relative=")) {
2760                 DIFF_OPT_SET(options, RELATIVE_NAME);
2761                 options->prefix = arg + 11;
2762         }
2763
2764         /* xdiff options */
2765         else if (!strcmp(arg, "-w") || !strcmp(arg, "--ignore-all-space"))
2766                 DIFF_XDL_SET(options, IGNORE_WHITESPACE);
2767         else if (!strcmp(arg, "-b") || !strcmp(arg, "--ignore-space-change"))
2768                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_CHANGE);
2769         else if (!strcmp(arg, "--ignore-space-at-eol"))
2770                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_AT_EOL);
2771         else if (!strcmp(arg, "--patience"))
2772                 DIFF_XDL_SET(options, PATIENCE_DIFF);
2773
2774         /* flags options */
2775         else if (!strcmp(arg, "--binary")) {
2776                 options->output_format |= DIFF_FORMAT_PATCH;
2777                 DIFF_OPT_SET(options, BINARY);
2778         }
2779         else if (!strcmp(arg, "--full-index"))
2780                 DIFF_OPT_SET(options, FULL_INDEX);
2781         else if (!strcmp(arg, "-a") || !strcmp(arg, "--text"))
2782                 DIFF_OPT_SET(options, TEXT);
2783         else if (!strcmp(arg, "-R"))
2784                 DIFF_OPT_SET(options, REVERSE_DIFF);
2785         else if (!strcmp(arg, "--find-copies-harder"))
2786                 DIFF_OPT_SET(options, FIND_COPIES_HARDER);
2787         else if (!strcmp(arg, "--follow"))
2788                 DIFF_OPT_SET(options, FOLLOW_RENAMES);
2789         else if (!strcmp(arg, "--color"))
2790                 DIFF_OPT_SET(options, COLOR_DIFF);
2791         else if (!strcmp(arg, "--no-color"))
2792                 DIFF_OPT_CLR(options, COLOR_DIFF);
2793         else if (!strcmp(arg, "--color-words")) {
2794                 DIFF_OPT_SET(options, COLOR_DIFF);
2795                 DIFF_OPT_SET(options, COLOR_DIFF_WORDS);
2796         }
2797         else if (!prefixcmp(arg, "--color-words=")) {
2798                 DIFF_OPT_SET(options, COLOR_DIFF);
2799                 DIFF_OPT_SET(options, COLOR_DIFF_WORDS);
2800                 options->word_regex = arg + 14;
2801         }
2802         else if (!strcmp(arg, "--exit-code"))
2803                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
2804         else if (!strcmp(arg, "--quiet"))
2805                 DIFF_OPT_SET(options, QUIET);
2806         else if (!strcmp(arg, "--ext-diff"))
2807                 DIFF_OPT_SET(options, ALLOW_EXTERNAL);
2808         else if (!strcmp(arg, "--no-ext-diff"))
2809                 DIFF_OPT_CLR(options, ALLOW_EXTERNAL);
2810         else if (!strcmp(arg, "--textconv"))
2811                 DIFF_OPT_SET(options, ALLOW_TEXTCONV);
2812         else if (!strcmp(arg, "--no-textconv"))
2813                 DIFF_OPT_CLR(options, ALLOW_TEXTCONV);
2814         else if (!strcmp(arg, "--ignore-submodules"))
2815                 DIFF_OPT_SET(options, IGNORE_SUBMODULES);
2816         else if (!strcmp(arg, "--submodule"))
2817                 DIFF_OPT_SET(options, SUBMODULE_LOG);
2818         else if (!prefixcmp(arg, "--submodule=")) {
2819                 if (!strcmp(arg + 12, "log"))
2820                         DIFF_OPT_SET(options, SUBMODULE_LOG);
2821         }
2822
2823         /* misc options */
2824         else if (!strcmp(arg, "-z"))
2825                 options->line_termination = 0;
2826         else if (!prefixcmp(arg, "-l"))
2827                 options->rename_limit = strtoul(arg+2, NULL, 10);
2828         else if (!prefixcmp(arg, "-S"))
2829                 options->pickaxe = arg + 2;
2830         else if (!strcmp(arg, "--pickaxe-all"))
2831                 options->pickaxe_opts = DIFF_PICKAXE_ALL;
2832         else if (!strcmp(arg, "--pickaxe-regex"))
2833                 options->pickaxe_opts = DIFF_PICKAXE_REGEX;
2834         else if (!prefixcmp(arg, "-O"))
2835                 options->orderfile = arg + 2;
2836         else if (!prefixcmp(arg, "--diff-filter="))
2837                 options->filter = arg + 14;
2838         else if (!strcmp(arg, "--abbrev"))
2839                 options->abbrev = DEFAULT_ABBREV;
2840         else if (!prefixcmp(arg, "--abbrev=")) {
2841                 options->abbrev = strtoul(arg + 9, NULL, 10);
2842                 if (options->abbrev < MINIMUM_ABBREV)
2843                         options->abbrev = MINIMUM_ABBREV;
2844                 else if (40 < options->abbrev)
2845                         options->abbrev = 40;
2846         }
2847         else if (!prefixcmp(arg, "--src-prefix="))
2848                 options->a_prefix = arg + 13;
2849         else if (!prefixcmp(arg, "--dst-prefix="))
2850                 options->b_prefix = arg + 13;
2851         else if (!strcmp(arg, "--no-prefix"))
2852                 options->a_prefix = options->b_prefix = "";
2853         else if (opt_arg(arg, '\0', "inter-hunk-context",
2854                          &options->interhunkcontext))
2855                 ;
2856         else if (!prefixcmp(arg, "--output=")) {
2857                 options->file = fopen(arg + strlen("--output="), "w");
2858                 options->close_file = 1;
2859         } else
2860                 return 0;
2861         return 1;
2862 }
2863
2864 static int parse_num(const char **cp_p)
2865 {
2866         unsigned long num, scale;
2867         int ch, dot;
2868         const char *cp = *cp_p;
2869
2870         num = 0;
2871         scale = 1;
2872         dot = 0;
2873         for (;;) {
2874                 ch = *cp;
2875                 if ( !dot && ch == '.' ) {
2876                         scale = 1;
2877                         dot = 1;
2878                 } else if ( ch == '%' ) {
2879                         scale = dot ? scale*100 : 100;
2880                         cp++;   /* % is always at the end */
2881                         break;
2882                 } else if ( ch >= '0' && ch <= '9' ) {
2883                         if ( scale < 100000 ) {
2884                                 scale *= 10;
2885                                 num = (num*10) + (ch-'0');
2886                         }
2887                 } else {
2888                         break;
2889                 }
2890                 cp++;
2891         }
2892         *cp_p = cp;
2893
2894         /* user says num divided by scale and we say internally that
2895          * is MAX_SCORE * num / scale.
2896          */
2897         return (int)((num >= scale) ? MAX_SCORE : (MAX_SCORE * num / scale));
2898 }
2899
2900 static int diff_scoreopt_parse(const char *opt)
2901 {
2902         int opt1, opt2, cmd;
2903
2904         if (*opt++ != '-')
2905                 return -1;
2906         cmd = *opt++;
2907         if (cmd != 'M' && cmd != 'C' && cmd != 'B')
2908                 return -1; /* that is not a -M, -C nor -B option */
2909
2910         opt1 = parse_num(&opt);
2911         if (cmd != 'B')
2912                 opt2 = 0;
2913         else {
2914                 if (*opt == 0)
2915                         opt2 = 0;
2916                 else if (*opt != '/')
2917                         return -1; /* we expect -B80/99 or -B80 */
2918                 else {
2919                         opt++;
2920                         opt2 = parse_num(&opt);
2921                 }
2922         }
2923         if (*opt != 0)
2924                 return -1;
2925         return opt1 | (opt2 << 16);
2926 }
2927
2928 struct diff_queue_struct diff_queued_diff;
2929
2930 void diff_q(struct diff_queue_struct *queue, struct diff_filepair *dp)
2931 {
2932         if (queue->alloc <= queue->nr) {
2933                 queue->alloc = alloc_nr(queue->alloc);
2934                 queue->queue = xrealloc(queue->queue,
2935                                         sizeof(dp) * queue->alloc);
2936         }
2937         queue->queue[queue->nr++] = dp;
2938 }
2939
2940 struct diff_filepair *diff_queue(struct diff_queue_struct *queue,
2941                                  struct diff_filespec *one,
2942                                  struct diff_filespec *two)
2943 {
2944         struct diff_filepair *dp = xcalloc(1, sizeof(*dp));
2945         dp->one = one;
2946         dp->two = two;
2947         if (queue)
2948                 diff_q(queue, dp);
2949         return dp;
2950 }
2951
2952 void diff_free_filepair(struct diff_filepair *p)
2953 {
2954         free_filespec(p->one);
2955         free_filespec(p->two);
2956         free(p);
2957 }
2958
2959 /* This is different from find_unique_abbrev() in that
2960  * it stuffs the result with dots for alignment.
2961  */
2962 const char *diff_unique_abbrev(const unsigned char *sha1, int len)
2963 {
2964         int abblen;
2965         const char *abbrev;
2966         if (len == 40)
2967                 return sha1_to_hex(sha1);
2968
2969         abbrev = find_unique_abbrev(sha1, len);
2970         abblen = strlen(abbrev);
2971         if (abblen < 37) {
2972                 static char hex[41];
2973                 if (len < abblen && abblen <= len + 2)
2974                         sprintf(hex, "%s%.*s", abbrev, len+3-abblen, "..");
2975                 else
2976                         sprintf(hex, "%s...", abbrev);
2977                 return hex;
2978         }
2979         return sha1_to_hex(sha1);
2980 }
2981
2982 static void diff_flush_raw(struct diff_filepair *p, struct diff_options *opt)
2983 {
2984         int line_termination = opt->line_termination;
2985         int inter_name_termination = line_termination ? '\t' : '\0';
2986
2987         if (!(opt->output_format & DIFF_FORMAT_NAME_STATUS)) {
2988                 fprintf(opt->file, ":%06o %06o %s ", p->one->mode, p->two->mode,
2989                         diff_unique_abbrev(p->one->sha1, opt->abbrev));
2990                 fprintf(opt->file, "%s ", diff_unique_abbrev(p->two->sha1, opt->abbrev));
2991         }
2992         if (p->score) {
2993                 fprintf(opt->file, "%c%03d%c", p->status, similarity_index(p),
2994                         inter_name_termination);
2995         } else {
2996                 fprintf(opt->file, "%c%c", p->status, inter_name_termination);
2997         }
2998
2999         if (p->status == DIFF_STATUS_COPIED ||
3000             p->status == DIFF_STATUS_RENAMED) {
3001                 const char *name_a, *name_b;
3002                 name_a = p->one->path;
3003                 name_b = p->two->path;
3004                 strip_prefix(opt->prefix_length, &name_a, &name_b);
3005                 write_name_quoted(name_a, opt->file, inter_name_termination);
3006                 write_name_quoted(name_b, opt->file, line_termination);
3007         } else {
3008                 const char *name_a, *name_b;
3009                 name_a = p->one->mode ? p->one->path : p->two->path;
3010                 name_b = NULL;
3011                 strip_prefix(opt->prefix_length, &name_a, &name_b);
3012                 write_name_quoted(name_a, opt->file, line_termination);
3013         }
3014 }
3015
3016 int diff_unmodified_pair(struct diff_filepair *p)
3017 {
3018         /* This function is written stricter than necessary to support
3019          * the currently implemented transformers, but the idea is to
3020          * let transformers to produce diff_filepairs any way they want,
3021          * and filter and clean them up here before producing the output.
3022          */
3023         struct diff_filespec *one = p->one, *two = p->two;
3024
3025         if (DIFF_PAIR_UNMERGED(p))
3026                 return 0; /* unmerged is interesting */
3027
3028         /* deletion, addition, mode or type change
3029          * and rename are all interesting.
3030          */
3031         if (DIFF_FILE_VALID(one) != DIFF_FILE_VALID(two) ||
3032             DIFF_PAIR_MODE_CHANGED(p) ||
3033             strcmp(one->path, two->path))
3034                 return 0;
3035
3036         /* both are valid and point at the same path.  that is, we are
3037          * dealing with a change.
3038          */
3039         if (one->sha1_valid && two->sha1_valid &&
3040             !hashcmp(one->sha1, two->sha1))
3041                 return 1; /* no change */
3042         if (!one->sha1_valid && !two->sha1_valid)
3043                 return 1; /* both look at the same file on the filesystem. */
3044         return 0;
3045 }
3046
3047 static void diff_flush_patch(struct diff_filepair *p, struct diff_options *o)
3048 {
3049         if (diff_unmodified_pair(p))
3050                 return;
3051
3052         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
3053             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
3054                 return; /* no tree diffs in patch format */
3055
3056         run_diff(p, o);
3057 }
3058
3059 static void diff_flush_stat(struct diff_filepair *p, struct diff_options *o,
3060                             struct diffstat_t *diffstat)
3061 {
3062         if (diff_unmodified_pair(p))
3063                 return;
3064
3065         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
3066             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
3067                 return; /* no tree diffs in patch format */
3068
3069         run_diffstat(p, o, diffstat);
3070 }
3071
3072 static void diff_flush_checkdiff(struct diff_filepair *p,
3073                 struct diff_options *o)
3074 {
3075         if (diff_unmodified_pair(p))
3076                 return;
3077
3078         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
3079             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
3080                 return; /* no tree diffs in patch format */
3081
3082         run_checkdiff(p, o);
3083 }
3084
3085 int diff_queue_is_empty(void)
3086 {
3087         struct diff_queue_struct *q = &diff_queued_diff;
3088         int i;
3089         for (i = 0; i < q->nr; i++)
3090                 if (!diff_unmodified_pair(q->queue[i]))
3091                         return 0;
3092         return 1;
3093 }
3094
3095 #if DIFF_DEBUG
3096 void diff_debug_filespec(struct diff_filespec *s, int x, const char *one)
3097 {
3098         fprintf(stderr, "queue[%d] %s (%s) %s %06o %s\n",
3099                 x, one ? one : "",
3100                 s->path,
3101                 DIFF_FILE_VALID(s) ? "valid" : "invalid",
3102                 s->mode,
3103                 s->sha1_valid ? sha1_to_hex(s->sha1) : "");
3104         fprintf(stderr, "queue[%d] %s size %lu flags %d\n",
3105                 x, one ? one : "",
3106                 s->size, s->xfrm_flags);
3107 }
3108
3109 void diff_debug_filepair(const struct diff_filepair *p, int i)
3110 {
3111         diff_debug_filespec(p->one, i, "one");
3112         diff_debug_filespec(p->two, i, "two");
3113         fprintf(stderr, "score %d, status %c rename_used %d broken %d\n",
3114                 p->score, p->status ? p->status : '?',
3115                 p->one->rename_used, p->broken_pair);
3116 }
3117
3118 void diff_debug_queue(const char *msg, struct diff_queue_struct *q)
3119 {
3120         int i;
3121         if (msg)
3122                 fprintf(stderr, "%s\n", msg);
3123         fprintf(stderr, "q->nr = %d\n", q->nr);
3124         for (i = 0; i < q->nr; i++) {
3125                 struct diff_filepair *p = q->queue[i];
3126                 diff_debug_filepair(p, i);
3127         }
3128 }
3129 #endif
3130
3131 static void diff_resolve_rename_copy(void)
3132 {
3133         int i;
3134         struct diff_filepair *p;
3135         struct diff_queue_struct *q = &diff_queued_diff;
3136
3137         diff_debug_queue("resolve-rename-copy", q);
3138
3139         for (i = 0; i < q->nr; i++) {
3140                 p = q->queue[i];
3141                 p->status = 0; /* undecided */
3142                 if (DIFF_PAIR_UNMERGED(p))
3143                         p->status = DIFF_STATUS_UNMERGED;
3144                 else if (!DIFF_FILE_VALID(p->one))
3145                         p->status = DIFF_STATUS_ADDED;
3146                 else if (!DIFF_FILE_VALID(p->two))
3147                         p->status = DIFF_STATUS_DELETED;
3148                 else if (DIFF_PAIR_TYPE_CHANGED(p))
3149                         p->status = DIFF_STATUS_TYPE_CHANGED;
3150
3151                 /* from this point on, we are dealing with a pair
3152                  * whose both sides are valid and of the same type, i.e.
3153                  * either in-place edit or rename/copy edit.
3154                  */
3155                 else if (DIFF_PAIR_RENAME(p)) {
3156                         /*
3157                          * A rename might have re-connected a broken
3158                          * pair up, causing the pathnames to be the
3159                          * same again. If so, that's not a rename at
3160                          * all, just a modification..
3161                          *
3162                          * Otherwise, see if this source was used for
3163                          * multiple renames, in which case we decrement
3164                          * the count, and call it a copy.
3165                          */
3166                         if (!strcmp(p->one->path, p->two->path))
3167                                 p->status = DIFF_STATUS_MODIFIED;
3168                         else if (--p->one->rename_used > 0)
3169                                 p->status = DIFF_STATUS_COPIED;
3170                         else
3171                                 p->status = DIFF_STATUS_RENAMED;
3172                 }
3173                 else if (hashcmp(p->one->sha1, p->two->sha1) ||
3174                          p->one->mode != p->two->mode ||
3175                          is_null_sha1(p->one->sha1))
3176                         p->status = DIFF_STATUS_MODIFIED;
3177                 else {
3178                         /* This is a "no-change" entry and should not
3179                          * happen anymore, but prepare for broken callers.
3180                          */
3181                         error("feeding unmodified %s to diffcore",
3182                               p->one->path);
3183                         p->status = DIFF_STATUS_UNKNOWN;
3184                 }
3185         }
3186         diff_debug_queue("resolve-rename-copy done", q);
3187 }
3188
3189 static int check_pair_status(struct diff_filepair *p)
3190 {
3191         switch (p->status) {
3192         case DIFF_STATUS_UNKNOWN:
3193                 return 0;
3194         case 0:
3195                 die("internal error in diff-resolve-rename-copy");
3196         default:
3197                 return 1;
3198         }
3199 }
3200
3201 static void flush_one_pair(struct diff_filepair *p, struct diff_options *opt)
3202 {
3203         int fmt = opt->output_format;
3204
3205         if (fmt & DIFF_FORMAT_CHECKDIFF)
3206                 diff_flush_checkdiff(p, opt);
3207         else if (fmt & (DIFF_FORMAT_RAW | DIFF_FORMAT_NAME_STATUS))
3208                 diff_flush_raw(p, opt);
3209         else if (fmt & DIFF_FORMAT_NAME) {
3210                 const char *name_a, *name_b;
3211                 name_a = p->two->path;
3212                 name_b = NULL;
3213                 strip_prefix(opt->prefix_length, &name_a, &name_b);
3214                 write_name_quoted(name_a, opt->file, opt->line_termination);
3215         }
3216 }
3217
3218 static void show_file_mode_name(FILE *file, const char *newdelete, struct diff_filespec *fs)
3219 {
3220         if (fs->mode)
3221                 fprintf(file, " %s mode %06o ", newdelete, fs->mode);
3222         else
3223                 fprintf(file, " %s ", newdelete);
3224         write_name_quoted(fs->path, file, '\n');
3225 }
3226
3227
3228 static void show_mode_change(FILE *file, struct diff_filepair *p, int show_name)
3229 {
3230         if (p->one->mode && p->two->mode && p->one->mode != p->two->mode) {
3231                 fprintf(file, " mode change %06o => %06o%c", p->one->mode, p->two->mode,
3232                         show_name ? ' ' : '\n');
3233                 if (show_name) {
3234                         write_name_quoted(p->two->path, file, '\n');
3235                 }
3236         }
3237 }
3238
3239 static void show_rename_copy(FILE *file, const char *renamecopy, struct diff_filepair *p)
3240 {
3241         char *names = pprint_rename(p->one->path, p->two->path);
3242
3243         fprintf(file, " %s %s (%d%%)\n", renamecopy, names, similarity_index(p));
3244         free(names);
3245         show_mode_change(file, p, 0);
3246 }
3247
3248 static void diff_summary(FILE *file, struct diff_filepair *p)
3249 {
3250         switch(p->status) {
3251         case DIFF_STATUS_DELETED:
3252                 show_file_mode_name(file, "delete", p->one);
3253                 break;
3254         case DIFF_STATUS_ADDED:
3255                 show_file_mode_name(file, "create", p->two);
3256                 break;
3257         case DIFF_STATUS_COPIED:
3258                 show_rename_copy(file, "copy", p);
3259                 break;
3260         case DIFF_STATUS_RENAMED:
3261                 show_rename_copy(file, "rename", p);
3262                 break;
3263         default:
3264                 if (p->score) {
3265                         fputs(" rewrite ", file);
3266                         write_name_quoted(p->two->path, file, ' ');
3267                         fprintf(file, "(%d%%)\n", similarity_index(p));
3268                 }
3269                 show_mode_change(file, p, !p->score);
3270                 break;
3271         }
3272 }
3273
3274 struct patch_id_t {
3275         git_SHA_CTX *ctx;
3276         int patchlen;
3277 };
3278
3279 static int remove_space(char *line, int len)
3280 {
3281         int i;
3282         char *dst = line;
3283         unsigned char c;
3284
3285         for (i = 0; i < len; i++)
3286                 if (!isspace((c = line[i])))
3287                         *dst++ = c;
3288
3289         return dst - line;
3290 }
3291
3292 static void patch_id_consume(void *priv, char *line, unsigned long len)
3293 {
3294         struct patch_id_t *data = priv;
3295         int new_len;
3296
3297         /* Ignore line numbers when computing the SHA1 of the patch */
3298         if (!prefixcmp(line, "@@ -"))
3299                 return;
3300
3301         new_len = remove_space(line, len);
3302
3303         git_SHA1_Update(data->ctx, line, new_len);
3304         data->patchlen += new_len;
3305 }
3306
3307 /* returns 0 upon success, and writes result into sha1 */
3308 static int diff_get_patch_id(struct diff_options *options, unsigned char *sha1)
3309 {
3310         struct diff_queue_struct *q = &diff_queued_diff;
3311         int i;
3312         git_SHA_CTX ctx;
3313         struct patch_id_t data;
3314         char buffer[PATH_MAX * 4 + 20];
3315
3316         git_SHA1_Init(&ctx);
3317         memset(&data, 0, sizeof(struct patch_id_t));
3318         data.ctx = &ctx;
3319
3320         for (i = 0; i < q->nr; i++) {
3321                 xpparam_t xpp;
3322                 xdemitconf_t xecfg;
3323                 xdemitcb_t ecb;
3324                 mmfile_t mf1, mf2;
3325                 struct diff_filepair *p = q->queue[i];
3326                 int len1, len2;
3327
3328                 memset(&xpp, 0, sizeof(xpp));
3329                 memset(&xecfg, 0, sizeof(xecfg));
3330                 if (p->status == 0)
3331                         return error("internal diff status error");
3332                 if (p->status == DIFF_STATUS_UNKNOWN)
3333                         continue;
3334                 if (diff_unmodified_pair(p))
3335                         continue;
3336                 if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
3337                     (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
3338                         continue;
3339                 if (DIFF_PAIR_UNMERGED(p))
3340                         continue;
3341
3342                 diff_fill_sha1_info(p->one);
3343                 diff_fill_sha1_info(p->two);
3344                 if (fill_mmfile(&mf1, p->one) < 0 ||
3345                                 fill_mmfile(&mf2, p->two) < 0)
3346                         return error("unable to read files to diff");
3347
3348                 len1 = remove_space(p->one->path, strlen(p->one->path));
3349                 len2 = remove_space(p->two->path, strlen(p->two->path));
3350                 if (p->one->mode == 0)
3351                         len1 = snprintf(buffer, sizeof(buffer),
3352                                         "diff--gita/%.*sb/%.*s"
3353                                         "newfilemode%06o"
3354                                         "---/dev/null"
3355                                         "+++b/%.*s",
3356                                         len1, p->one->path,
3357                                         len2, p->two->path,
3358                                         p->two->mode,
3359                                         len2, p->two->path);
3360                 else if (p->two->mode == 0)
3361                         len1 = snprintf(buffer, sizeof(buffer),
3362                                         "diff--gita/%.*sb/%.*s"
3363                                         "deletedfilemode%06o"
3364                                         "---a/%.*s"
3365                                         "+++/dev/null",
3366                                         len1, p->one->path,
3367                                         len2, p->two->path,
3368                                         p->one->mode,
3369                                         len1, p->one->path);
3370                 else
3371                         len1 = snprintf(buffer, sizeof(buffer),
3372                                         "diff--gita/%.*sb/%.*s"
3373                                         "---a/%.*s"
3374                                         "+++b/%.*s",
3375                                         len1, p->one->path,
3376                                         len2, p->two->path,
3377                                         len1, p->one->path,
3378                                         len2, p->two->path);
3379                 git_SHA1_Update(&ctx, buffer, len1);
3380
3381                 xpp.flags = XDF_NEED_MINIMAL;
3382                 xecfg.ctxlen = 3;
3383                 xecfg.flags = XDL_EMIT_FUNCNAMES;
3384                 xdi_diff_outf(&mf1, &mf2, patch_id_consume, &data,
3385                               &xpp, &xecfg, &ecb);
3386         }
3387
3388         git_SHA1_Final(sha1, &ctx);
3389         return 0;
3390 }
3391
3392 int diff_flush_patch_id(struct diff_options *options, unsigned char *sha1)
3393 {
3394         struct diff_queue_struct *q = &diff_queued_diff;
3395         int i;
3396         int result = diff_get_patch_id(options, sha1);
3397
3398         for (i = 0; i < q->nr; i++)
3399                 diff_free_filepair(q->queue[i]);
3400
3401         free(q->queue);
3402         q->queue = NULL;
3403         q->nr = q->alloc = 0;
3404
3405         return result;
3406 }
3407
3408 static int is_summary_empty(const struct diff_queue_struct *q)
3409 {
3410         int i;
3411
3412         for (i = 0; i < q->nr; i++) {
3413                 const struct diff_filepair *p = q->queue[i];
3414
3415                 switch (p->status) {
3416                 case DIFF_STATUS_DELETED:
3417                 case DIFF_STATUS_ADDED:
3418                 case DIFF_STATUS_COPIED:
3419                 case DIFF_STATUS_RENAMED:
3420                         return 0;
3421                 default:
3422                         if (p->score)
3423                                 return 0;
3424                         if (p->one->mode && p->two->mode &&
3425                             p->one->mode != p->two->mode)
3426                                 return 0;
3427                         break;
3428                 }
3429         }
3430         return 1;
3431 }
3432
3433 void diff_flush(struct diff_options *options)
3434 {
3435         struct diff_queue_struct *q = &diff_queued_diff;
3436         int i, output_format = options->output_format;
3437         int separator = 0;
3438
3439         /*
3440          * Order: raw, stat, summary, patch
3441          * or:    name/name-status/checkdiff (other bits clear)
3442          */
3443         if (!q->nr)
3444                 goto free_queue;
3445
3446         if (output_format & (DIFF_FORMAT_RAW |
3447                              DIFF_FORMAT_NAME |
3448                              DIFF_FORMAT_NAME_STATUS |
3449                              DIFF_FORMAT_CHECKDIFF)) {
3450                 for (i = 0; i < q->nr; i++) {
3451                         struct diff_filepair *p = q->queue[i];
3452                         if (check_pair_status(p))
3453                                 flush_one_pair(p, options);
3454                 }
3455                 separator++;
3456         }
3457
3458         if (output_format & (DIFF_FORMAT_DIFFSTAT|DIFF_FORMAT_SHORTSTAT|DIFF_FORMAT_NUMSTAT)) {
3459                 struct diffstat_t diffstat;
3460
3461                 memset(&diffstat, 0, sizeof(struct diffstat_t));
3462                 for (i = 0; i < q->nr; i++) {
3463                         struct diff_filepair *p = q->queue[i];
3464                         if (check_pair_status(p))
3465                                 diff_flush_stat(p, options, &diffstat);
3466                 }
3467                 if (output_format & DIFF_FORMAT_NUMSTAT)
3468                         show_numstat(&diffstat, options);
3469                 if (output_format & DIFF_FORMAT_DIFFSTAT)
3470                         show_stats(&diffstat, options);
3471                 if (output_format & DIFF_FORMAT_SHORTSTAT)
3472                         show_shortstats(&diffstat, options);
3473                 free_diffstat_info(&diffstat);
3474                 separator++;
3475         }
3476         if (output_format & DIFF_FORMAT_DIRSTAT)
3477                 show_dirstat(options);
3478
3479         if (output_format & DIFF_FORMAT_SUMMARY && !is_summary_empty(q)) {
3480                 for (i = 0; i < q->nr; i++)
3481                         diff_summary(options->file, q->queue[i]);
3482                 separator++;
3483         }
3484
3485         if (output_format & DIFF_FORMAT_PATCH) {
3486                 if (separator) {
3487                         putc(options->line_termination, options->file);
3488                         if (options->stat_sep) {
3489                                 /* attach patch instead of inline */
3490                                 fputs(options->stat_sep, options->file);
3491                         }
3492                 }
3493
3494                 for (i = 0; i < q->nr; i++) {
3495                         struct diff_filepair *p = q->queue[i];
3496                         if (check_pair_status(p))
3497                                 diff_flush_patch(p, options);
3498                 }
3499         }
3500
3501         if (output_format & DIFF_FORMAT_CALLBACK)
3502                 options->format_callback(q, options, options->format_callback_data);
3503
3504         for (i = 0; i < q->nr; i++)
3505                 diff_free_filepair(q->queue[i]);
3506 free_queue:
3507         free(q->queue);
3508         q->queue = NULL;
3509         q->nr = q->alloc = 0;
3510         if (options->close_file)
3511                 fclose(options->file);
3512 }
3513
3514 static void diffcore_apply_filter(const char *filter)
3515 {
3516         int i;
3517         struct diff_queue_struct *q = &diff_queued_diff;
3518         struct diff_queue_struct outq;
3519         outq.queue = NULL;
3520         outq.nr = outq.alloc = 0;
3521
3522         if (!filter)
3523                 return;
3524
3525         if (strchr(filter, DIFF_STATUS_FILTER_AON)) {
3526                 int found;
3527                 for (i = found = 0; !found && i < q->nr; i++) {
3528                         struct diff_filepair *p = q->queue[i];
3529                         if (((p->status == DIFF_STATUS_MODIFIED) &&
3530                              ((p->score &&
3531                                strchr(filter, DIFF_STATUS_FILTER_BROKEN)) ||
3532                               (!p->score &&
3533                                strchr(filter, DIFF_STATUS_MODIFIED)))) ||
3534                             ((p->status != DIFF_STATUS_MODIFIED) &&
3535                              strchr(filter, p->status)))
3536                                 found++;
3537                 }
3538                 if (found)
3539                         return;
3540
3541                 /* otherwise we will clear the whole queue
3542                  * by copying the empty outq at the end of this
3543                  * function, but first clear the current entries
3544                  * in the queue.
3545                  */
3546                 for (i = 0; i < q->nr; i++)
3547                         diff_free_filepair(q->queue[i]);
3548         }
3549         else {
3550                 /* Only the matching ones */
3551                 for (i = 0; i < q->nr; i++) {
3552                         struct diff_filepair *p = q->queue[i];
3553
3554                         if (((p->status == DIFF_STATUS_MODIFIED) &&
3555                              ((p->score &&
3556                                strchr(filter, DIFF_STATUS_FILTER_BROKEN)) ||
3557                               (!p->score &&
3558                                strchr(filter, DIFF_STATUS_MODIFIED)))) ||
3559                             ((p->status != DIFF_STATUS_MODIFIED) &&
3560                              strchr(filter, p->status)))
3561                                 diff_q(&outq, p);
3562                         else
3563                                 diff_free_filepair(p);
3564                 }
3565         }
3566         free(q->queue);
3567         *q = outq;
3568 }
3569
3570 /* Check whether two filespecs with the same mode and size are identical */
3571 static int diff_filespec_is_identical(struct diff_filespec *one,
3572                                       struct diff_filespec *two)
3573 {
3574         if (S_ISGITLINK(one->mode))
3575                 return 0;
3576         if (diff_populate_filespec(one, 0))
3577                 return 0;
3578         if (diff_populate_filespec(two, 0))
3579                 return 0;
3580         return !memcmp(one->data, two->data, one->size);
3581 }
3582
3583 static void diffcore_skip_stat_unmatch(struct diff_options *diffopt)
3584 {
3585         int i;
3586         struct diff_queue_struct *q = &diff_queued_diff;
3587         struct diff_queue_struct outq;
3588         outq.queue = NULL;
3589         outq.nr = outq.alloc = 0;
3590
3591         for (i = 0; i < q->nr; i++) {
3592                 struct diff_filepair *p = q->queue[i];
3593
3594                 /*
3595                  * 1. Entries that come from stat info dirtyness
3596                  *    always have both sides (iow, not create/delete),
3597                  *    one side of the object name is unknown, with
3598                  *    the same mode and size.  Keep the ones that
3599                  *    do not match these criteria.  They have real
3600                  *    differences.
3601                  *
3602                  * 2. At this point, the file is known to be modified,
3603                  *    with the same mode and size, and the object
3604                  *    name of one side is unknown.  Need to inspect
3605                  *    the identical contents.
3606                  */
3607                 if (!DIFF_FILE_VALID(p->one) || /* (1) */
3608                     !DIFF_FILE_VALID(p->two) ||
3609                     (p->one->sha1_valid && p->two->sha1_valid) ||
3610                     (p->one->mode != p->two->mode) ||
3611                     diff_populate_filespec(p->one, 1) ||
3612                     diff_populate_filespec(p->two, 1) ||
3613                     (p->one->size != p->two->size) ||
3614                     !diff_filespec_is_identical(p->one, p->two)) /* (2) */
3615                         diff_q(&outq, p);
3616                 else {
3617                         /*
3618                          * The caller can subtract 1 from skip_stat_unmatch
3619                          * to determine how many paths were dirty only
3620                          * due to stat info mismatch.
3621                          */
3622                         if (!DIFF_OPT_TST(diffopt, NO_INDEX))
3623                                 diffopt->skip_stat_unmatch++;
3624                         diff_free_filepair(p);
3625                 }
3626         }
3627         free(q->queue);
3628         *q = outq;
3629 }
3630
3631 void diffcore_std(struct diff_options *options)
3632 {
3633         if (options->skip_stat_unmatch)
3634                 diffcore_skip_stat_unmatch(options);
3635         if (options->break_opt != -1)
3636                 diffcore_break(options->break_opt);
3637         if (options->detect_rename)
3638                 diffcore_rename(options);
3639         if (options->break_opt != -1)
3640                 diffcore_merge_broken();
3641         if (options->pickaxe)
3642                 diffcore_pickaxe(options->pickaxe, options->pickaxe_opts);
3643         if (options->orderfile)
3644                 diffcore_order(options->orderfile);
3645         diff_resolve_rename_copy();
3646         diffcore_apply_filter(options->filter);
3647
3648         if (diff_queued_diff.nr)
3649                 DIFF_OPT_SET(options, HAS_CHANGES);
3650         else
3651                 DIFF_OPT_CLR(options, HAS_CHANGES);
3652 }
3653
3654 int diff_result_code(struct diff_options *opt, int status)
3655 {
3656         int result = 0;
3657         if (!DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
3658             !(opt->output_format & DIFF_FORMAT_CHECKDIFF))
3659                 return status;
3660         if (DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
3661             DIFF_OPT_TST(opt, HAS_CHANGES))
3662                 result |= 01;
3663         if ((opt->output_format & DIFF_FORMAT_CHECKDIFF) &&
3664             DIFF_OPT_TST(opt, CHECK_FAILED))
3665                 result |= 02;
3666         return result;
3667 }
3668
3669 void diff_addremove(struct diff_options *options,
3670                     int addremove, unsigned mode,
3671                     const unsigned char *sha1,
3672                     const char *concatpath)
3673 {
3674         struct diff_filespec *one, *two;
3675
3676         if (DIFF_OPT_TST(options, IGNORE_SUBMODULES) && S_ISGITLINK(mode))
3677                 return;
3678
3679         /* This may look odd, but it is a preparation for
3680          * feeding "there are unchanged files which should
3681          * not produce diffs, but when you are doing copy
3682          * detection you would need them, so here they are"
3683          * entries to the diff-core.  They will be prefixed
3684          * with something like '=' or '*' (I haven't decided
3685          * which but should not make any difference).
3686          * Feeding the same new and old to diff_change()
3687          * also has the same effect.
3688          * Before the final output happens, they are pruned after
3689          * merged into rename/copy pairs as appropriate.
3690          */
3691         if (DIFF_OPT_TST(options, REVERSE_DIFF))
3692                 addremove = (addremove == '+' ? '-' :
3693                              addremove == '-' ? '+' : addremove);
3694
3695         if (options->prefix &&
3696             strncmp(concatpath, options->prefix, options->prefix_length))
3697                 return;
3698
3699         one = alloc_filespec(concatpath);
3700         two = alloc_filespec(concatpath);
3701
3702         if (addremove != '+')
3703                 fill_filespec(one, sha1, mode);
3704         if (addremove != '-')
3705                 fill_filespec(two, sha1, mode);
3706
3707         diff_queue(&diff_queued_diff, one, two);
3708         DIFF_OPT_SET(options, HAS_CHANGES);
3709 }
3710
3711 void diff_change(struct diff_options *options,
3712                  unsigned old_mode, unsigned new_mode,
3713                  const unsigned char *old_sha1,
3714                  const unsigned char *new_sha1,
3715                  const char *concatpath)
3716 {
3717         struct diff_filespec *one, *two;
3718
3719         if (DIFF_OPT_TST(options, IGNORE_SUBMODULES) && S_ISGITLINK(old_mode)
3720                         && S_ISGITLINK(new_mode))
3721                 return;
3722
3723         if (DIFF_OPT_TST(options, REVERSE_DIFF)) {
3724                 unsigned tmp;
3725                 const unsigned char *tmp_c;
3726                 tmp = old_mode; old_mode = new_mode; new_mode = tmp;
3727                 tmp_c = old_sha1; old_sha1 = new_sha1; new_sha1 = tmp_c;
3728         }
3729
3730         if (options->prefix &&
3731             strncmp(concatpath, options->prefix, options->prefix_length))
3732                 return;
3733
3734         one = alloc_filespec(concatpath);
3735         two = alloc_filespec(concatpath);
3736         fill_filespec(one, old_sha1, old_mode);
3737         fill_filespec(two, new_sha1, new_mode);
3738
3739         diff_queue(&diff_queued_diff, one, two);
3740         DIFF_OPT_SET(options, HAS_CHANGES);
3741 }
3742
3743 void diff_unmerge(struct diff_options *options,
3744                   const char *path,
3745                   unsigned mode, const unsigned char *sha1)
3746 {
3747         struct diff_filespec *one, *two;
3748
3749         if (options->prefix &&
3750             strncmp(path, options->prefix, options->prefix_length))
3751                 return;
3752
3753         one = alloc_filespec(path);
3754         two = alloc_filespec(path);
3755         fill_filespec(one, sha1, mode);
3756         diff_queue(&diff_queued_diff, one, two)->is_unmerged = 1;
3757 }
3758
3759 static char *run_textconv(const char *pgm, struct diff_filespec *spec,
3760                 size_t *outsize)
3761 {
3762         struct diff_tempfile *temp;
3763         const char *argv[3];
3764         const char **arg = argv;
3765         struct child_process child;
3766         struct strbuf buf = STRBUF_INIT;
3767
3768         temp = prepare_temp_file(spec->path, spec);
3769         *arg++ = pgm;
3770         *arg++ = temp->name;
3771         *arg = NULL;
3772
3773         memset(&child, 0, sizeof(child));
3774         child.argv = argv;
3775         child.out = -1;
3776         if (start_command(&child) != 0 ||
3777             strbuf_read(&buf, child.out, 0) < 0 ||
3778             finish_command(&child) != 0) {
3779                 strbuf_release(&buf);
3780                 remove_tempfile();
3781                 error("error running textconv command '%s'", pgm);
3782                 return NULL;
3783         }
3784         remove_tempfile();
3785
3786         return strbuf_detach(&buf, outsize);
3787 }