Merge branch 'jk/tighten-alloc' into maint
[git] / diff.c
1 /*
2  * Copyright (C) 2005 Junio C Hamano
3  */
4 #include "cache.h"
5 #include "tempfile.h"
6 #include "quote.h"
7 #include "diff.h"
8 #include "diffcore.h"
9 #include "delta.h"
10 #include "xdiff-interface.h"
11 #include "color.h"
12 #include "attr.h"
13 #include "run-command.h"
14 #include "utf8.h"
15 #include "userdiff.h"
16 #include "submodule-config.h"
17 #include "submodule.h"
18 #include "ll-merge.h"
19 #include "string-list.h"
20 #include "argv-array.h"
21
22 #ifdef NO_FAST_WORKING_DIRECTORY
23 #define FAST_WORKING_DIRECTORY 0
24 #else
25 #define FAST_WORKING_DIRECTORY 1
26 #endif
27
28 static int diff_detect_rename_default;
29 static int diff_compaction_heuristic; /* experimental */
30 static int diff_rename_limit_default = 400;
31 static int diff_suppress_blank_empty;
32 static int diff_use_color_default = -1;
33 static int diff_context_default = 3;
34 static const char *diff_word_regex_cfg;
35 static const char *external_diff_cmd_cfg;
36 static const char *diff_order_file_cfg;
37 int diff_auto_refresh_index = 1;
38 static int diff_mnemonic_prefix;
39 static int diff_no_prefix;
40 static int diff_stat_graph_width;
41 static int diff_dirstat_permille_default = 30;
42 static struct diff_options default_diff_options;
43 static long diff_algorithm;
44
45 static char diff_colors[][COLOR_MAXLEN] = {
46         GIT_COLOR_RESET,
47         GIT_COLOR_NORMAL,       /* CONTEXT */
48         GIT_COLOR_BOLD,         /* METAINFO */
49         GIT_COLOR_CYAN,         /* FRAGINFO */
50         GIT_COLOR_RED,          /* OLD */
51         GIT_COLOR_GREEN,        /* NEW */
52         GIT_COLOR_YELLOW,       /* COMMIT */
53         GIT_COLOR_BG_RED,       /* WHITESPACE */
54         GIT_COLOR_NORMAL,       /* FUNCINFO */
55 };
56
57 static int parse_diff_color_slot(const char *var)
58 {
59         if (!strcasecmp(var, "context") || !strcasecmp(var, "plain"))
60                 return DIFF_CONTEXT;
61         if (!strcasecmp(var, "meta"))
62                 return DIFF_METAINFO;
63         if (!strcasecmp(var, "frag"))
64                 return DIFF_FRAGINFO;
65         if (!strcasecmp(var, "old"))
66                 return DIFF_FILE_OLD;
67         if (!strcasecmp(var, "new"))
68                 return DIFF_FILE_NEW;
69         if (!strcasecmp(var, "commit"))
70                 return DIFF_COMMIT;
71         if (!strcasecmp(var, "whitespace"))
72                 return DIFF_WHITESPACE;
73         if (!strcasecmp(var, "func"))
74                 return DIFF_FUNCINFO;
75         return -1;
76 }
77
78 static int parse_dirstat_params(struct diff_options *options, const char *params_string,
79                                 struct strbuf *errmsg)
80 {
81         char *params_copy = xstrdup(params_string);
82         struct string_list params = STRING_LIST_INIT_NODUP;
83         int ret = 0;
84         int i;
85
86         if (*params_copy)
87                 string_list_split_in_place(&params, params_copy, ',', -1);
88         for (i = 0; i < params.nr; i++) {
89                 const char *p = params.items[i].string;
90                 if (!strcmp(p, "changes")) {
91                         DIFF_OPT_CLR(options, DIRSTAT_BY_LINE);
92                         DIFF_OPT_CLR(options, DIRSTAT_BY_FILE);
93                 } else if (!strcmp(p, "lines")) {
94                         DIFF_OPT_SET(options, DIRSTAT_BY_LINE);
95                         DIFF_OPT_CLR(options, DIRSTAT_BY_FILE);
96                 } else if (!strcmp(p, "files")) {
97                         DIFF_OPT_CLR(options, DIRSTAT_BY_LINE);
98                         DIFF_OPT_SET(options, DIRSTAT_BY_FILE);
99                 } else if (!strcmp(p, "noncumulative")) {
100                         DIFF_OPT_CLR(options, DIRSTAT_CUMULATIVE);
101                 } else if (!strcmp(p, "cumulative")) {
102                         DIFF_OPT_SET(options, DIRSTAT_CUMULATIVE);
103                 } else if (isdigit(*p)) {
104                         char *end;
105                         int permille = strtoul(p, &end, 10) * 10;
106                         if (*end == '.' && isdigit(*++end)) {
107                                 /* only use first digit */
108                                 permille += *end - '0';
109                                 /* .. and ignore any further digits */
110                                 while (isdigit(*++end))
111                                         ; /* nothing */
112                         }
113                         if (!*end)
114                                 options->dirstat_permille = permille;
115                         else {
116                                 strbuf_addf(errmsg, _("  Failed to parse dirstat cut-off percentage '%s'\n"),
117                                             p);
118                                 ret++;
119                         }
120                 } else {
121                         strbuf_addf(errmsg, _("  Unknown dirstat parameter '%s'\n"), p);
122                         ret++;
123                 }
124
125         }
126         string_list_clear(&params, 0);
127         free(params_copy);
128         return ret;
129 }
130
131 static int parse_submodule_params(struct diff_options *options, const char *value)
132 {
133         if (!strcmp(value, "log"))
134                 DIFF_OPT_SET(options, SUBMODULE_LOG);
135         else if (!strcmp(value, "short"))
136                 DIFF_OPT_CLR(options, SUBMODULE_LOG);
137         else
138                 return -1;
139         return 0;
140 }
141
142 static int git_config_rename(const char *var, const char *value)
143 {
144         if (!value)
145                 return DIFF_DETECT_RENAME;
146         if (!strcasecmp(value, "copies") || !strcasecmp(value, "copy"))
147                 return  DIFF_DETECT_COPY;
148         return git_config_bool(var,value) ? DIFF_DETECT_RENAME : 0;
149 }
150
151 long parse_algorithm_value(const char *value)
152 {
153         if (!value)
154                 return -1;
155         else if (!strcasecmp(value, "myers") || !strcasecmp(value, "default"))
156                 return 0;
157         else if (!strcasecmp(value, "minimal"))
158                 return XDF_NEED_MINIMAL;
159         else if (!strcasecmp(value, "patience"))
160                 return XDF_PATIENCE_DIFF;
161         else if (!strcasecmp(value, "histogram"))
162                 return XDF_HISTOGRAM_DIFF;
163         return -1;
164 }
165
166 /*
167  * These are to give UI layer defaults.
168  * The core-level commands such as git-diff-files should
169  * never be affected by the setting of diff.renames
170  * the user happens to have in the configuration file.
171  */
172 void init_diff_ui_defaults(void)
173 {
174         diff_detect_rename_default = 1;
175 }
176
177 int git_diff_ui_config(const char *var, const char *value, void *cb)
178 {
179         if (!strcmp(var, "diff.color") || !strcmp(var, "color.diff")) {
180                 diff_use_color_default = git_config_colorbool(var, value);
181                 return 0;
182         }
183         if (!strcmp(var, "diff.context")) {
184                 diff_context_default = git_config_int(var, value);
185                 if (diff_context_default < 0)
186                         return -1;
187                 return 0;
188         }
189         if (!strcmp(var, "diff.renames")) {
190                 diff_detect_rename_default = git_config_rename(var, value);
191                 return 0;
192         }
193         if (!strcmp(var, "diff.compactionheuristic")) {
194                 diff_compaction_heuristic = git_config_bool(var, value);
195                 return 0;
196         }
197         if (!strcmp(var, "diff.autorefreshindex")) {
198                 diff_auto_refresh_index = git_config_bool(var, value);
199                 return 0;
200         }
201         if (!strcmp(var, "diff.mnemonicprefix")) {
202                 diff_mnemonic_prefix = git_config_bool(var, value);
203                 return 0;
204         }
205         if (!strcmp(var, "diff.noprefix")) {
206                 diff_no_prefix = git_config_bool(var, value);
207                 return 0;
208         }
209         if (!strcmp(var, "diff.statgraphwidth")) {
210                 diff_stat_graph_width = git_config_int(var, value);
211                 return 0;
212         }
213         if (!strcmp(var, "diff.external"))
214                 return git_config_string(&external_diff_cmd_cfg, var, value);
215         if (!strcmp(var, "diff.wordregex"))
216                 return git_config_string(&diff_word_regex_cfg, var, value);
217         if (!strcmp(var, "diff.orderfile"))
218                 return git_config_pathname(&diff_order_file_cfg, var, value);
219
220         if (!strcmp(var, "diff.ignoresubmodules"))
221                 handle_ignore_submodules_arg(&default_diff_options, value);
222
223         if (!strcmp(var, "diff.submodule")) {
224                 if (parse_submodule_params(&default_diff_options, value))
225                         warning(_("Unknown value for 'diff.submodule' config variable: '%s'"),
226                                 value);
227                 return 0;
228         }
229
230         if (!strcmp(var, "diff.algorithm")) {
231                 diff_algorithm = parse_algorithm_value(value);
232                 if (diff_algorithm < 0)
233                         return -1;
234                 return 0;
235         }
236
237         if (git_color_config(var, value, cb) < 0)
238                 return -1;
239
240         return git_diff_basic_config(var, value, cb);
241 }
242
243 int git_diff_basic_config(const char *var, const char *value, void *cb)
244 {
245         const char *name;
246
247         if (!strcmp(var, "diff.renamelimit")) {
248                 diff_rename_limit_default = git_config_int(var, value);
249                 return 0;
250         }
251
252         if (userdiff_config(var, value) < 0)
253                 return -1;
254
255         if (skip_prefix(var, "diff.color.", &name) ||
256             skip_prefix(var, "color.diff.", &name)) {
257                 int slot = parse_diff_color_slot(name);
258                 if (slot < 0)
259                         return 0;
260                 if (!value)
261                         return config_error_nonbool(var);
262                 return color_parse(value, diff_colors[slot]);
263         }
264
265         /* like GNU diff's --suppress-blank-empty option  */
266         if (!strcmp(var, "diff.suppressblankempty") ||
267                         /* for backwards compatibility */
268                         !strcmp(var, "diff.suppress-blank-empty")) {
269                 diff_suppress_blank_empty = git_config_bool(var, value);
270                 return 0;
271         }
272
273         if (!strcmp(var, "diff.dirstat")) {
274                 struct strbuf errmsg = STRBUF_INIT;
275                 default_diff_options.dirstat_permille = diff_dirstat_permille_default;
276                 if (parse_dirstat_params(&default_diff_options, value, &errmsg))
277                         warning(_("Found errors in 'diff.dirstat' config variable:\n%s"),
278                                 errmsg.buf);
279                 strbuf_release(&errmsg);
280                 diff_dirstat_permille_default = default_diff_options.dirstat_permille;
281                 return 0;
282         }
283
284         if (starts_with(var, "submodule."))
285                 return parse_submodule_config_option(var, value);
286
287         return git_default_config(var, value, cb);
288 }
289
290 static char *quote_two(const char *one, const char *two)
291 {
292         int need_one = quote_c_style(one, NULL, NULL, 1);
293         int need_two = quote_c_style(two, NULL, NULL, 1);
294         struct strbuf res = STRBUF_INIT;
295
296         if (need_one + need_two) {
297                 strbuf_addch(&res, '"');
298                 quote_c_style(one, &res, NULL, 1);
299                 quote_c_style(two, &res, NULL, 1);
300                 strbuf_addch(&res, '"');
301         } else {
302                 strbuf_addstr(&res, one);
303                 strbuf_addstr(&res, two);
304         }
305         return strbuf_detach(&res, NULL);
306 }
307
308 static const char *external_diff(void)
309 {
310         static const char *external_diff_cmd = NULL;
311         static int done_preparing = 0;
312
313         if (done_preparing)
314                 return external_diff_cmd;
315         external_diff_cmd = getenv("GIT_EXTERNAL_DIFF");
316         if (!external_diff_cmd)
317                 external_diff_cmd = external_diff_cmd_cfg;
318         done_preparing = 1;
319         return external_diff_cmd;
320 }
321
322 /*
323  * Keep track of files used for diffing. Sometimes such an entry
324  * refers to a temporary file, sometimes to an existing file, and
325  * sometimes to "/dev/null".
326  */
327 static struct diff_tempfile {
328         /*
329          * filename external diff should read from, or NULL if this
330          * entry is currently not in use:
331          */
332         const char *name;
333
334         char hex[GIT_SHA1_HEXSZ + 1];
335         char mode[10];
336
337         /*
338          * If this diff_tempfile instance refers to a temporary file,
339          * this tempfile object is used to manage its lifetime.
340          */
341         struct tempfile tempfile;
342 } diff_temp[2];
343
344 typedef unsigned long (*sane_truncate_fn)(char *line, unsigned long len);
345
346 struct emit_callback {
347         int color_diff;
348         unsigned ws_rule;
349         int blank_at_eof_in_preimage;
350         int blank_at_eof_in_postimage;
351         int lno_in_preimage;
352         int lno_in_postimage;
353         sane_truncate_fn truncate;
354         const char **label_path;
355         struct diff_words_data *diff_words;
356         struct diff_options *opt;
357         struct strbuf *header;
358 };
359
360 static int count_lines(const char *data, int size)
361 {
362         int count, ch, completely_empty = 1, nl_just_seen = 0;
363         count = 0;
364         while (0 < size--) {
365                 ch = *data++;
366                 if (ch == '\n') {
367                         count++;
368                         nl_just_seen = 1;
369                         completely_empty = 0;
370                 }
371                 else {
372                         nl_just_seen = 0;
373                         completely_empty = 0;
374                 }
375         }
376         if (completely_empty)
377                 return 0;
378         if (!nl_just_seen)
379                 count++; /* no trailing newline */
380         return count;
381 }
382
383 static int fill_mmfile(mmfile_t *mf, struct diff_filespec *one)
384 {
385         if (!DIFF_FILE_VALID(one)) {
386                 mf->ptr = (char *)""; /* does not matter */
387                 mf->size = 0;
388                 return 0;
389         }
390         else if (diff_populate_filespec(one, 0))
391                 return -1;
392
393         mf->ptr = one->data;
394         mf->size = one->size;
395         return 0;
396 }
397
398 /* like fill_mmfile, but only for size, so we can avoid retrieving blob */
399 static unsigned long diff_filespec_size(struct diff_filespec *one)
400 {
401         if (!DIFF_FILE_VALID(one))
402                 return 0;
403         diff_populate_filespec(one, CHECK_SIZE_ONLY);
404         return one->size;
405 }
406
407 static int count_trailing_blank(mmfile_t *mf, unsigned ws_rule)
408 {
409         char *ptr = mf->ptr;
410         long size = mf->size;
411         int cnt = 0;
412
413         if (!size)
414                 return cnt;
415         ptr += size - 1; /* pointing at the very end */
416         if (*ptr != '\n')
417                 ; /* incomplete line */
418         else
419                 ptr--; /* skip the last LF */
420         while (mf->ptr < ptr) {
421                 char *prev_eol;
422                 for (prev_eol = ptr; mf->ptr <= prev_eol; prev_eol--)
423                         if (*prev_eol == '\n')
424                                 break;
425                 if (!ws_blank_line(prev_eol + 1, ptr - prev_eol, ws_rule))
426                         break;
427                 cnt++;
428                 ptr = prev_eol - 1;
429         }
430         return cnt;
431 }
432
433 static void check_blank_at_eof(mmfile_t *mf1, mmfile_t *mf2,
434                                struct emit_callback *ecbdata)
435 {
436         int l1, l2, at;
437         unsigned ws_rule = ecbdata->ws_rule;
438         l1 = count_trailing_blank(mf1, ws_rule);
439         l2 = count_trailing_blank(mf2, ws_rule);
440         if (l2 <= l1) {
441                 ecbdata->blank_at_eof_in_preimage = 0;
442                 ecbdata->blank_at_eof_in_postimage = 0;
443                 return;
444         }
445         at = count_lines(mf1->ptr, mf1->size);
446         ecbdata->blank_at_eof_in_preimage = (at - l1) + 1;
447
448         at = count_lines(mf2->ptr, mf2->size);
449         ecbdata->blank_at_eof_in_postimage = (at - l2) + 1;
450 }
451
452 static void emit_line_0(struct diff_options *o, const char *set, const char *reset,
453                         int first, const char *line, int len)
454 {
455         int has_trailing_newline, has_trailing_carriage_return;
456         int nofirst;
457         FILE *file = o->file;
458
459         fputs(diff_line_prefix(o), file);
460
461         if (len == 0) {
462                 has_trailing_newline = (first == '\n');
463                 has_trailing_carriage_return = (!has_trailing_newline &&
464                                                 (first == '\r'));
465                 nofirst = has_trailing_newline || has_trailing_carriage_return;
466         } else {
467                 has_trailing_newline = (len > 0 && line[len-1] == '\n');
468                 if (has_trailing_newline)
469                         len--;
470                 has_trailing_carriage_return = (len > 0 && line[len-1] == '\r');
471                 if (has_trailing_carriage_return)
472                         len--;
473                 nofirst = 0;
474         }
475
476         if (len || !nofirst) {
477                 fputs(set, file);
478                 if (!nofirst)
479                         fputc(first, file);
480                 fwrite(line, len, 1, file);
481                 fputs(reset, file);
482         }
483         if (has_trailing_carriage_return)
484                 fputc('\r', file);
485         if (has_trailing_newline)
486                 fputc('\n', file);
487 }
488
489 static void emit_line(struct diff_options *o, const char *set, const char *reset,
490                       const char *line, int len)
491 {
492         emit_line_0(o, set, reset, line[0], line+1, len-1);
493 }
494
495 static int new_blank_line_at_eof(struct emit_callback *ecbdata, const char *line, int len)
496 {
497         if (!((ecbdata->ws_rule & WS_BLANK_AT_EOF) &&
498               ecbdata->blank_at_eof_in_preimage &&
499               ecbdata->blank_at_eof_in_postimage &&
500               ecbdata->blank_at_eof_in_preimage <= ecbdata->lno_in_preimage &&
501               ecbdata->blank_at_eof_in_postimage <= ecbdata->lno_in_postimage))
502                 return 0;
503         return ws_blank_line(line, len, ecbdata->ws_rule);
504 }
505
506 static void emit_line_checked(const char *reset,
507                               struct emit_callback *ecbdata,
508                               const char *line, int len,
509                               enum color_diff color,
510                               unsigned ws_error_highlight,
511                               char sign)
512 {
513         const char *set = diff_get_color(ecbdata->color_diff, color);
514         const char *ws = NULL;
515
516         if (ecbdata->opt->ws_error_highlight & ws_error_highlight) {
517                 ws = diff_get_color(ecbdata->color_diff, DIFF_WHITESPACE);
518                 if (!*ws)
519                         ws = NULL;
520         }
521
522         if (!ws)
523                 emit_line_0(ecbdata->opt, set, reset, sign, line, len);
524         else if (sign == '+' && new_blank_line_at_eof(ecbdata, line, len))
525                 /* Blank line at EOF - paint '+' as well */
526                 emit_line_0(ecbdata->opt, ws, reset, sign, line, len);
527         else {
528                 /* Emit just the prefix, then the rest. */
529                 emit_line_0(ecbdata->opt, set, reset, sign, "", 0);
530                 ws_check_emit(line, len, ecbdata->ws_rule,
531                               ecbdata->opt->file, set, reset, ws);
532         }
533 }
534
535 static void emit_add_line(const char *reset,
536                           struct emit_callback *ecbdata,
537                           const char *line, int len)
538 {
539         emit_line_checked(reset, ecbdata, line, len,
540                           DIFF_FILE_NEW, WSEH_NEW, '+');
541 }
542
543 static void emit_del_line(const char *reset,
544                           struct emit_callback *ecbdata,
545                           const char *line, int len)
546 {
547         emit_line_checked(reset, ecbdata, line, len,
548                           DIFF_FILE_OLD, WSEH_OLD, '-');
549 }
550
551 static void emit_context_line(const char *reset,
552                               struct emit_callback *ecbdata,
553                               const char *line, int len)
554 {
555         emit_line_checked(reset, ecbdata, line, len,
556                           DIFF_CONTEXT, WSEH_CONTEXT, ' ');
557 }
558
559 static void emit_hunk_header(struct emit_callback *ecbdata,
560                              const char *line, int len)
561 {
562         const char *context = diff_get_color(ecbdata->color_diff, DIFF_CONTEXT);
563         const char *frag = diff_get_color(ecbdata->color_diff, DIFF_FRAGINFO);
564         const char *func = diff_get_color(ecbdata->color_diff, DIFF_FUNCINFO);
565         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
566         static const char atat[2] = { '@', '@' };
567         const char *cp, *ep;
568         struct strbuf msgbuf = STRBUF_INIT;
569         int org_len = len;
570         int i = 1;
571
572         /*
573          * As a hunk header must begin with "@@ -<old>, +<new> @@",
574          * it always is at least 10 bytes long.
575          */
576         if (len < 10 ||
577             memcmp(line, atat, 2) ||
578             !(ep = memmem(line + 2, len - 2, atat, 2))) {
579                 emit_line(ecbdata->opt, context, reset, line, len);
580                 return;
581         }
582         ep += 2; /* skip over @@ */
583
584         /* The hunk header in fraginfo color */
585         strbuf_addstr(&msgbuf, frag);
586         strbuf_add(&msgbuf, line, ep - line);
587         strbuf_addstr(&msgbuf, reset);
588
589         /*
590          * trailing "\r\n"
591          */
592         for ( ; i < 3; i++)
593                 if (line[len - i] == '\r' || line[len - i] == '\n')
594                         len--;
595
596         /* blank before the func header */
597         for (cp = ep; ep - line < len; ep++)
598                 if (*ep != ' ' && *ep != '\t')
599                         break;
600         if (ep != cp) {
601                 strbuf_addstr(&msgbuf, context);
602                 strbuf_add(&msgbuf, cp, ep - cp);
603                 strbuf_addstr(&msgbuf, reset);
604         }
605
606         if (ep < line + len) {
607                 strbuf_addstr(&msgbuf, func);
608                 strbuf_add(&msgbuf, ep, line + len - ep);
609                 strbuf_addstr(&msgbuf, reset);
610         }
611
612         strbuf_add(&msgbuf, line + len, org_len - len);
613         emit_line(ecbdata->opt, "", "", msgbuf.buf, msgbuf.len);
614         strbuf_release(&msgbuf);
615 }
616
617 static struct diff_tempfile *claim_diff_tempfile(void) {
618         int i;
619         for (i = 0; i < ARRAY_SIZE(diff_temp); i++)
620                 if (!diff_temp[i].name)
621                         return diff_temp + i;
622         die("BUG: diff is failing to clean up its tempfiles");
623 }
624
625 static void remove_tempfile(void)
626 {
627         int i;
628         for (i = 0; i < ARRAY_SIZE(diff_temp); i++) {
629                 if (is_tempfile_active(&diff_temp[i].tempfile))
630                         delete_tempfile(&diff_temp[i].tempfile);
631                 diff_temp[i].name = NULL;
632         }
633 }
634
635 static void print_line_count(FILE *file, int count)
636 {
637         switch (count) {
638         case 0:
639                 fprintf(file, "0,0");
640                 break;
641         case 1:
642                 fprintf(file, "1");
643                 break;
644         default:
645                 fprintf(file, "1,%d", count);
646                 break;
647         }
648 }
649
650 static void emit_rewrite_lines(struct emit_callback *ecb,
651                                int prefix, const char *data, int size)
652 {
653         const char *endp = NULL;
654         static const char *nneof = " No newline at end of file\n";
655         const char *reset = diff_get_color(ecb->color_diff, DIFF_RESET);
656
657         while (0 < size) {
658                 int len;
659
660                 endp = memchr(data, '\n', size);
661                 len = endp ? (endp - data + 1) : size;
662                 if (prefix != '+') {
663                         ecb->lno_in_preimage++;
664                         emit_del_line(reset, ecb, data, len);
665                 } else {
666                         ecb->lno_in_postimage++;
667                         emit_add_line(reset, ecb, data, len);
668                 }
669                 size -= len;
670                 data += len;
671         }
672         if (!endp) {
673                 const char *context = diff_get_color(ecb->color_diff,
674                                                      DIFF_CONTEXT);
675                 putc('\n', ecb->opt->file);
676                 emit_line_0(ecb->opt, context, reset, '\\',
677                             nneof, strlen(nneof));
678         }
679 }
680
681 static void emit_rewrite_diff(const char *name_a,
682                               const char *name_b,
683                               struct diff_filespec *one,
684                               struct diff_filespec *two,
685                               struct userdiff_driver *textconv_one,
686                               struct userdiff_driver *textconv_two,
687                               struct diff_options *o)
688 {
689         int lc_a, lc_b;
690         const char *name_a_tab, *name_b_tab;
691         const char *metainfo = diff_get_color(o->use_color, DIFF_METAINFO);
692         const char *fraginfo = diff_get_color(o->use_color, DIFF_FRAGINFO);
693         const char *reset = diff_get_color(o->use_color, DIFF_RESET);
694         static struct strbuf a_name = STRBUF_INIT, b_name = STRBUF_INIT;
695         const char *a_prefix, *b_prefix;
696         char *data_one, *data_two;
697         size_t size_one, size_two;
698         struct emit_callback ecbdata;
699         const char *line_prefix = diff_line_prefix(o);
700
701         if (diff_mnemonic_prefix && DIFF_OPT_TST(o, REVERSE_DIFF)) {
702                 a_prefix = o->b_prefix;
703                 b_prefix = o->a_prefix;
704         } else {
705                 a_prefix = o->a_prefix;
706                 b_prefix = o->b_prefix;
707         }
708
709         name_a += (*name_a == '/');
710         name_b += (*name_b == '/');
711         name_a_tab = strchr(name_a, ' ') ? "\t" : "";
712         name_b_tab = strchr(name_b, ' ') ? "\t" : "";
713
714         strbuf_reset(&a_name);
715         strbuf_reset(&b_name);
716         quote_two_c_style(&a_name, a_prefix, name_a, 0);
717         quote_two_c_style(&b_name, b_prefix, name_b, 0);
718
719         size_one = fill_textconv(textconv_one, one, &data_one);
720         size_two = fill_textconv(textconv_two, two, &data_two);
721
722         memset(&ecbdata, 0, sizeof(ecbdata));
723         ecbdata.color_diff = want_color(o->use_color);
724         ecbdata.ws_rule = whitespace_rule(name_b);
725         ecbdata.opt = o;
726         if (ecbdata.ws_rule & WS_BLANK_AT_EOF) {
727                 mmfile_t mf1, mf2;
728                 mf1.ptr = (char *)data_one;
729                 mf2.ptr = (char *)data_two;
730                 mf1.size = size_one;
731                 mf2.size = size_two;
732                 check_blank_at_eof(&mf1, &mf2, &ecbdata);
733         }
734         ecbdata.lno_in_preimage = 1;
735         ecbdata.lno_in_postimage = 1;
736
737         lc_a = count_lines(data_one, size_one);
738         lc_b = count_lines(data_two, size_two);
739         fprintf(o->file,
740                 "%s%s--- %s%s%s\n%s%s+++ %s%s%s\n%s%s@@ -",
741                 line_prefix, metainfo, a_name.buf, name_a_tab, reset,
742                 line_prefix, metainfo, b_name.buf, name_b_tab, reset,
743                 line_prefix, fraginfo);
744         if (!o->irreversible_delete)
745                 print_line_count(o->file, lc_a);
746         else
747                 fprintf(o->file, "?,?");
748         fprintf(o->file, " +");
749         print_line_count(o->file, lc_b);
750         fprintf(o->file, " @@%s\n", reset);
751         if (lc_a && !o->irreversible_delete)
752                 emit_rewrite_lines(&ecbdata, '-', data_one, size_one);
753         if (lc_b)
754                 emit_rewrite_lines(&ecbdata, '+', data_two, size_two);
755         if (textconv_one)
756                 free((char *)data_one);
757         if (textconv_two)
758                 free((char *)data_two);
759 }
760
761 struct diff_words_buffer {
762         mmfile_t text;
763         long alloc;
764         struct diff_words_orig {
765                 const char *begin, *end;
766         } *orig;
767         int orig_nr, orig_alloc;
768 };
769
770 static void diff_words_append(char *line, unsigned long len,
771                 struct diff_words_buffer *buffer)
772 {
773         ALLOC_GROW(buffer->text.ptr, buffer->text.size + len, buffer->alloc);
774         line++;
775         len--;
776         memcpy(buffer->text.ptr + buffer->text.size, line, len);
777         buffer->text.size += len;
778         buffer->text.ptr[buffer->text.size] = '\0';
779 }
780
781 struct diff_words_style_elem {
782         const char *prefix;
783         const char *suffix;
784         const char *color; /* NULL; filled in by the setup code if
785                             * color is enabled */
786 };
787
788 struct diff_words_style {
789         enum diff_words_type type;
790         struct diff_words_style_elem new, old, ctx;
791         const char *newline;
792 };
793
794 static struct diff_words_style diff_words_styles[] = {
795         { DIFF_WORDS_PORCELAIN, {"+", "\n"}, {"-", "\n"}, {" ", "\n"}, "~\n" },
796         { DIFF_WORDS_PLAIN, {"{+", "+}"}, {"[-", "-]"}, {"", ""}, "\n" },
797         { DIFF_WORDS_COLOR, {"", ""}, {"", ""}, {"", ""}, "\n" }
798 };
799
800 struct diff_words_data {
801         struct diff_words_buffer minus, plus;
802         const char *current_plus;
803         int last_minus;
804         struct diff_options *opt;
805         regex_t *word_regex;
806         enum diff_words_type type;
807         struct diff_words_style *style;
808 };
809
810 static int fn_out_diff_words_write_helper(FILE *fp,
811                                           struct diff_words_style_elem *st_el,
812                                           const char *newline,
813                                           size_t count, const char *buf,
814                                           const char *line_prefix)
815 {
816         int print = 0;
817
818         while (count) {
819                 char *p = memchr(buf, '\n', count);
820                 if (print)
821                         fputs(line_prefix, fp);
822                 if (p != buf) {
823                         if (st_el->color && fputs(st_el->color, fp) < 0)
824                                 return -1;
825                         if (fputs(st_el->prefix, fp) < 0 ||
826                             fwrite(buf, p ? p - buf : count, 1, fp) != 1 ||
827                             fputs(st_el->suffix, fp) < 0)
828                                 return -1;
829                         if (st_el->color && *st_el->color
830                             && fputs(GIT_COLOR_RESET, fp) < 0)
831                                 return -1;
832                 }
833                 if (!p)
834                         return 0;
835                 if (fputs(newline, fp) < 0)
836                         return -1;
837                 count -= p + 1 - buf;
838                 buf = p + 1;
839                 print = 1;
840         }
841         return 0;
842 }
843
844 /*
845  * '--color-words' algorithm can be described as:
846  *
847  *   1. collect a the minus/plus lines of a diff hunk, divided into
848  *      minus-lines and plus-lines;
849  *
850  *   2. break both minus-lines and plus-lines into words and
851  *      place them into two mmfile_t with one word for each line;
852  *
853  *   3. use xdiff to run diff on the two mmfile_t to get the words level diff;
854  *
855  * And for the common parts of the both file, we output the plus side text.
856  * diff_words->current_plus is used to trace the current position of the plus file
857  * which printed. diff_words->last_minus is used to trace the last minus word
858  * printed.
859  *
860  * For '--graph' to work with '--color-words', we need to output the graph prefix
861  * on each line of color words output. Generally, there are two conditions on
862  * which we should output the prefix.
863  *
864  *   1. diff_words->last_minus == 0 &&
865  *      diff_words->current_plus == diff_words->plus.text.ptr
866  *
867  *      that is: the plus text must start as a new line, and if there is no minus
868  *      word printed, a graph prefix must be printed.
869  *
870  *   2. diff_words->current_plus > diff_words->plus.text.ptr &&
871  *      *(diff_words->current_plus - 1) == '\n'
872  *
873  *      that is: a graph prefix must be printed following a '\n'
874  */
875 static int color_words_output_graph_prefix(struct diff_words_data *diff_words)
876 {
877         if ((diff_words->last_minus == 0 &&
878                 diff_words->current_plus == diff_words->plus.text.ptr) ||
879                 (diff_words->current_plus > diff_words->plus.text.ptr &&
880                 *(diff_words->current_plus - 1) == '\n')) {
881                 return 1;
882         } else {
883                 return 0;
884         }
885 }
886
887 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
888 {
889         struct diff_words_data *diff_words = priv;
890         struct diff_words_style *style = diff_words->style;
891         int minus_first, minus_len, plus_first, plus_len;
892         const char *minus_begin, *minus_end, *plus_begin, *plus_end;
893         struct diff_options *opt = diff_words->opt;
894         const char *line_prefix;
895
896         if (line[0] != '@' || parse_hunk_header(line, len,
897                         &minus_first, &minus_len, &plus_first, &plus_len))
898                 return;
899
900         assert(opt);
901         line_prefix = diff_line_prefix(opt);
902
903         /* POSIX requires that first be decremented by one if len == 0... */
904         if (minus_len) {
905                 minus_begin = diff_words->minus.orig[minus_first].begin;
906                 minus_end =
907                         diff_words->minus.orig[minus_first + minus_len - 1].end;
908         } else
909                 minus_begin = minus_end =
910                         diff_words->minus.orig[minus_first].end;
911
912         if (plus_len) {
913                 plus_begin = diff_words->plus.orig[plus_first].begin;
914                 plus_end = diff_words->plus.orig[plus_first + plus_len - 1].end;
915         } else
916                 plus_begin = plus_end = diff_words->plus.orig[plus_first].end;
917
918         if (color_words_output_graph_prefix(diff_words)) {
919                 fputs(line_prefix, diff_words->opt->file);
920         }
921         if (diff_words->current_plus != plus_begin) {
922                 fn_out_diff_words_write_helper(diff_words->opt->file,
923                                 &style->ctx, style->newline,
924                                 plus_begin - diff_words->current_plus,
925                                 diff_words->current_plus, line_prefix);
926                 if (*(plus_begin - 1) == '\n')
927                         fputs(line_prefix, diff_words->opt->file);
928         }
929         if (minus_begin != minus_end) {
930                 fn_out_diff_words_write_helper(diff_words->opt->file,
931                                 &style->old, style->newline,
932                                 minus_end - minus_begin, minus_begin,
933                                 line_prefix);
934         }
935         if (plus_begin != plus_end) {
936                 fn_out_diff_words_write_helper(diff_words->opt->file,
937                                 &style->new, style->newline,
938                                 plus_end - plus_begin, plus_begin,
939                                 line_prefix);
940         }
941
942         diff_words->current_plus = plus_end;
943         diff_words->last_minus = minus_first;
944 }
945
946 /* This function starts looking at *begin, and returns 0 iff a word was found. */
947 static int find_word_boundaries(mmfile_t *buffer, regex_t *word_regex,
948                 int *begin, int *end)
949 {
950         if (word_regex && *begin < buffer->size) {
951                 regmatch_t match[1];
952                 if (!regexec_buf(word_regex, buffer->ptr + *begin,
953                                  buffer->size - *begin, 1, match, 0)) {
954                         char *p = memchr(buffer->ptr + *begin + match[0].rm_so,
955                                         '\n', match[0].rm_eo - match[0].rm_so);
956                         *end = p ? p - buffer->ptr : match[0].rm_eo + *begin;
957                         *begin += match[0].rm_so;
958                         return *begin >= *end;
959                 }
960                 return -1;
961         }
962
963         /* find the next word */
964         while (*begin < buffer->size && isspace(buffer->ptr[*begin]))
965                 (*begin)++;
966         if (*begin >= buffer->size)
967                 return -1;
968
969         /* find the end of the word */
970         *end = *begin + 1;
971         while (*end < buffer->size && !isspace(buffer->ptr[*end]))
972                 (*end)++;
973
974         return 0;
975 }
976
977 /*
978  * This function splits the words in buffer->text, stores the list with
979  * newline separator into out, and saves the offsets of the original words
980  * in buffer->orig.
981  */
982 static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out,
983                 regex_t *word_regex)
984 {
985         int i, j;
986         long alloc = 0;
987
988         out->size = 0;
989         out->ptr = NULL;
990
991         /* fake an empty "0th" word */
992         ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
993         buffer->orig[0].begin = buffer->orig[0].end = buffer->text.ptr;
994         buffer->orig_nr = 1;
995
996         for (i = 0; i < buffer->text.size; i++) {
997                 if (find_word_boundaries(&buffer->text, word_regex, &i, &j))
998                         return;
999
1000                 /* store original boundaries */
1001                 ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
1002                                 buffer->orig_alloc);
1003                 buffer->orig[buffer->orig_nr].begin = buffer->text.ptr + i;
1004                 buffer->orig[buffer->orig_nr].end = buffer->text.ptr + j;
1005                 buffer->orig_nr++;
1006
1007                 /* store one word */
1008                 ALLOC_GROW(out->ptr, out->size + j - i + 1, alloc);
1009                 memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
1010                 out->ptr[out->size + j - i] = '\n';
1011                 out->size += j - i + 1;
1012
1013                 i = j - 1;
1014         }
1015 }
1016
1017 /* this executes the word diff on the accumulated buffers */
1018 static void diff_words_show(struct diff_words_data *diff_words)
1019 {
1020         xpparam_t xpp;
1021         xdemitconf_t xecfg;
1022         mmfile_t minus, plus;
1023         struct diff_words_style *style = diff_words->style;
1024
1025         struct diff_options *opt = diff_words->opt;
1026         const char *line_prefix;
1027
1028         assert(opt);
1029         line_prefix = diff_line_prefix(opt);
1030
1031         /* special case: only removal */
1032         if (!diff_words->plus.text.size) {
1033                 fputs(line_prefix, diff_words->opt->file);
1034                 fn_out_diff_words_write_helper(diff_words->opt->file,
1035                         &style->old, style->newline,
1036                         diff_words->minus.text.size,
1037                         diff_words->minus.text.ptr, line_prefix);
1038                 diff_words->minus.text.size = 0;
1039                 return;
1040         }
1041
1042         diff_words->current_plus = diff_words->plus.text.ptr;
1043         diff_words->last_minus = 0;
1044
1045         memset(&xpp, 0, sizeof(xpp));
1046         memset(&xecfg, 0, sizeof(xecfg));
1047         diff_words_fill(&diff_words->minus, &minus, diff_words->word_regex);
1048         diff_words_fill(&diff_words->plus, &plus, diff_words->word_regex);
1049         xpp.flags = 0;
1050         /* as only the hunk header will be parsed, we need a 0-context */
1051         xecfg.ctxlen = 0;
1052         if (xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
1053                           &xpp, &xecfg))
1054                 die("unable to generate word diff");
1055         free(minus.ptr);
1056         free(plus.ptr);
1057         if (diff_words->current_plus != diff_words->plus.text.ptr +
1058                         diff_words->plus.text.size) {
1059                 if (color_words_output_graph_prefix(diff_words))
1060                         fputs(line_prefix, diff_words->opt->file);
1061                 fn_out_diff_words_write_helper(diff_words->opt->file,
1062                         &style->ctx, style->newline,
1063                         diff_words->plus.text.ptr + diff_words->plus.text.size
1064                         - diff_words->current_plus, diff_words->current_plus,
1065                         line_prefix);
1066         }
1067         diff_words->minus.text.size = diff_words->plus.text.size = 0;
1068 }
1069
1070 /* In "color-words" mode, show word-diff of words accumulated in the buffer */
1071 static void diff_words_flush(struct emit_callback *ecbdata)
1072 {
1073         if (ecbdata->diff_words->minus.text.size ||
1074             ecbdata->diff_words->plus.text.size)
1075                 diff_words_show(ecbdata->diff_words);
1076 }
1077
1078 static void diff_filespec_load_driver(struct diff_filespec *one)
1079 {
1080         /* Use already-loaded driver */
1081         if (one->driver)
1082                 return;
1083
1084         if (S_ISREG(one->mode))
1085                 one->driver = userdiff_find_by_path(one->path);
1086
1087         /* Fallback to default settings */
1088         if (!one->driver)
1089                 one->driver = userdiff_find_by_name("default");
1090 }
1091
1092 static const char *userdiff_word_regex(struct diff_filespec *one)
1093 {
1094         diff_filespec_load_driver(one);
1095         return one->driver->word_regex;
1096 }
1097
1098 static void init_diff_words_data(struct emit_callback *ecbdata,
1099                                  struct diff_options *orig_opts,
1100                                  struct diff_filespec *one,
1101                                  struct diff_filespec *two)
1102 {
1103         int i;
1104         struct diff_options *o = xmalloc(sizeof(struct diff_options));
1105         memcpy(o, orig_opts, sizeof(struct diff_options));
1106
1107         ecbdata->diff_words =
1108                 xcalloc(1, sizeof(struct diff_words_data));
1109         ecbdata->diff_words->type = o->word_diff;
1110         ecbdata->diff_words->opt = o;
1111         if (!o->word_regex)
1112                 o->word_regex = userdiff_word_regex(one);
1113         if (!o->word_regex)
1114                 o->word_regex = userdiff_word_regex(two);
1115         if (!o->word_regex)
1116                 o->word_regex = diff_word_regex_cfg;
1117         if (o->word_regex) {
1118                 ecbdata->diff_words->word_regex = (regex_t *)
1119                         xmalloc(sizeof(regex_t));
1120                 if (regcomp(ecbdata->diff_words->word_regex,
1121                             o->word_regex,
1122                             REG_EXTENDED | REG_NEWLINE))
1123                         die ("Invalid regular expression: %s",
1124                              o->word_regex);
1125         }
1126         for (i = 0; i < ARRAY_SIZE(diff_words_styles); i++) {
1127                 if (o->word_diff == diff_words_styles[i].type) {
1128                         ecbdata->diff_words->style =
1129                                 &diff_words_styles[i];
1130                         break;
1131                 }
1132         }
1133         if (want_color(o->use_color)) {
1134                 struct diff_words_style *st = ecbdata->diff_words->style;
1135                 st->old.color = diff_get_color_opt(o, DIFF_FILE_OLD);
1136                 st->new.color = diff_get_color_opt(o, DIFF_FILE_NEW);
1137                 st->ctx.color = diff_get_color_opt(o, DIFF_CONTEXT);
1138         }
1139 }
1140
1141 static void free_diff_words_data(struct emit_callback *ecbdata)
1142 {
1143         if (ecbdata->diff_words) {
1144                 diff_words_flush(ecbdata);
1145                 free (ecbdata->diff_words->opt);
1146                 free (ecbdata->diff_words->minus.text.ptr);
1147                 free (ecbdata->diff_words->minus.orig);
1148                 free (ecbdata->diff_words->plus.text.ptr);
1149                 free (ecbdata->diff_words->plus.orig);
1150                 if (ecbdata->diff_words->word_regex) {
1151                         regfree(ecbdata->diff_words->word_regex);
1152                         free(ecbdata->diff_words->word_regex);
1153                 }
1154                 free(ecbdata->diff_words);
1155                 ecbdata->diff_words = NULL;
1156         }
1157 }
1158
1159 const char *diff_get_color(int diff_use_color, enum color_diff ix)
1160 {
1161         if (want_color(diff_use_color))
1162                 return diff_colors[ix];
1163         return "";
1164 }
1165
1166 const char *diff_line_prefix(struct diff_options *opt)
1167 {
1168         struct strbuf *msgbuf;
1169         if (!opt->output_prefix)
1170                 return "";
1171
1172         msgbuf = opt->output_prefix(opt, opt->output_prefix_data);
1173         return msgbuf->buf;
1174 }
1175
1176 static unsigned long sane_truncate_line(struct emit_callback *ecb, char *line, unsigned long len)
1177 {
1178         const char *cp;
1179         unsigned long allot;
1180         size_t l = len;
1181
1182         if (ecb->truncate)
1183                 return ecb->truncate(line, len);
1184         cp = line;
1185         allot = l;
1186         while (0 < l) {
1187                 (void) utf8_width(&cp, &l);
1188                 if (!cp)
1189                         break; /* truncated in the middle? */
1190         }
1191         return allot - l;
1192 }
1193
1194 static void find_lno(const char *line, struct emit_callback *ecbdata)
1195 {
1196         const char *p;
1197         ecbdata->lno_in_preimage = 0;
1198         ecbdata->lno_in_postimage = 0;
1199         p = strchr(line, '-');
1200         if (!p)
1201                 return; /* cannot happen */
1202         ecbdata->lno_in_preimage = strtol(p + 1, NULL, 10);
1203         p = strchr(p, '+');
1204         if (!p)
1205                 return; /* cannot happen */
1206         ecbdata->lno_in_postimage = strtol(p + 1, NULL, 10);
1207 }
1208
1209 static void fn_out_consume(void *priv, char *line, unsigned long len)
1210 {
1211         struct emit_callback *ecbdata = priv;
1212         const char *meta = diff_get_color(ecbdata->color_diff, DIFF_METAINFO);
1213         const char *context = diff_get_color(ecbdata->color_diff, DIFF_CONTEXT);
1214         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
1215         struct diff_options *o = ecbdata->opt;
1216         const char *line_prefix = diff_line_prefix(o);
1217
1218         o->found_changes = 1;
1219
1220         if (ecbdata->header) {
1221                 fprintf(o->file, "%s", ecbdata->header->buf);
1222                 strbuf_reset(ecbdata->header);
1223                 ecbdata->header = NULL;
1224         }
1225
1226         if (ecbdata->label_path[0]) {
1227                 const char *name_a_tab, *name_b_tab;
1228
1229                 name_a_tab = strchr(ecbdata->label_path[0], ' ') ? "\t" : "";
1230                 name_b_tab = strchr(ecbdata->label_path[1], ' ') ? "\t" : "";
1231
1232                 fprintf(o->file, "%s%s--- %s%s%s\n",
1233                         line_prefix, meta, ecbdata->label_path[0], reset, name_a_tab);
1234                 fprintf(o->file, "%s%s+++ %s%s%s\n",
1235                         line_prefix, meta, ecbdata->label_path[1], reset, name_b_tab);
1236                 ecbdata->label_path[0] = ecbdata->label_path[1] = NULL;
1237         }
1238
1239         if (diff_suppress_blank_empty
1240             && len == 2 && line[0] == ' ' && line[1] == '\n') {
1241                 line[0] = '\n';
1242                 len = 1;
1243         }
1244
1245         if (line[0] == '@') {
1246                 if (ecbdata->diff_words)
1247                         diff_words_flush(ecbdata);
1248                 len = sane_truncate_line(ecbdata, line, len);
1249                 find_lno(line, ecbdata);
1250                 emit_hunk_header(ecbdata, line, len);
1251                 if (line[len-1] != '\n')
1252                         putc('\n', o->file);
1253                 return;
1254         }
1255
1256         if (ecbdata->diff_words) {
1257                 if (line[0] == '-') {
1258                         diff_words_append(line, len,
1259                                           &ecbdata->diff_words->minus);
1260                         return;
1261                 } else if (line[0] == '+') {
1262                         diff_words_append(line, len,
1263                                           &ecbdata->diff_words->plus);
1264                         return;
1265                 } else if (starts_with(line, "\\ ")) {
1266                         /*
1267                          * Eat the "no newline at eof" marker as if we
1268                          * saw a "+" or "-" line with nothing on it,
1269                          * and return without diff_words_flush() to
1270                          * defer processing. If this is the end of
1271                          * preimage, more "+" lines may come after it.
1272                          */
1273                         return;
1274                 }
1275                 diff_words_flush(ecbdata);
1276                 if (ecbdata->diff_words->type == DIFF_WORDS_PORCELAIN) {
1277                         emit_line(o, context, reset, line, len);
1278                         fputs("~\n", o->file);
1279                 } else {
1280                         /*
1281                          * Skip the prefix character, if any.  With
1282                          * diff_suppress_blank_empty, there may be
1283                          * none.
1284                          */
1285                         if (line[0] != '\n') {
1286                               line++;
1287                               len--;
1288                         }
1289                         emit_line(o, context, reset, line, len);
1290                 }
1291                 return;
1292         }
1293
1294         switch (line[0]) {
1295         case '+':
1296                 ecbdata->lno_in_postimage++;
1297                 emit_add_line(reset, ecbdata, line + 1, len - 1);
1298                 break;
1299         case '-':
1300                 ecbdata->lno_in_preimage++;
1301                 emit_del_line(reset, ecbdata, line + 1, len - 1);
1302                 break;
1303         case ' ':
1304                 ecbdata->lno_in_postimage++;
1305                 ecbdata->lno_in_preimage++;
1306                 emit_context_line(reset, ecbdata, line + 1, len - 1);
1307                 break;
1308         default:
1309                 /* incomplete line at the end */
1310                 ecbdata->lno_in_preimage++;
1311                 emit_line(o, diff_get_color(ecbdata->color_diff, DIFF_CONTEXT),
1312                           reset, line, len);
1313                 break;
1314         }
1315 }
1316
1317 static char *pprint_rename(const char *a, const char *b)
1318 {
1319         const char *old = a;
1320         const char *new = b;
1321         struct strbuf name = STRBUF_INIT;
1322         int pfx_length, sfx_length;
1323         int pfx_adjust_for_slash;
1324         int len_a = strlen(a);
1325         int len_b = strlen(b);
1326         int a_midlen, b_midlen;
1327         int qlen_a = quote_c_style(a, NULL, NULL, 0);
1328         int qlen_b = quote_c_style(b, NULL, NULL, 0);
1329
1330         if (qlen_a || qlen_b) {
1331                 quote_c_style(a, &name, NULL, 0);
1332                 strbuf_addstr(&name, " => ");
1333                 quote_c_style(b, &name, NULL, 0);
1334                 return strbuf_detach(&name, NULL);
1335         }
1336
1337         /* Find common prefix */
1338         pfx_length = 0;
1339         while (*old && *new && *old == *new) {
1340                 if (*old == '/')
1341                         pfx_length = old - a + 1;
1342                 old++;
1343                 new++;
1344         }
1345
1346         /* Find common suffix */
1347         old = a + len_a;
1348         new = b + len_b;
1349         sfx_length = 0;
1350         /*
1351          * If there is a common prefix, it must end in a slash.  In
1352          * that case we let this loop run 1 into the prefix to see the
1353          * same slash.
1354          *
1355          * If there is no common prefix, we cannot do this as it would
1356          * underrun the input strings.
1357          */
1358         pfx_adjust_for_slash = (pfx_length ? 1 : 0);
1359         while (a + pfx_length - pfx_adjust_for_slash <= old &&
1360                b + pfx_length - pfx_adjust_for_slash <= new &&
1361                *old == *new) {
1362                 if (*old == '/')
1363                         sfx_length = len_a - (old - a);
1364                 old--;
1365                 new--;
1366         }
1367
1368         /*
1369          * pfx{mid-a => mid-b}sfx
1370          * {pfx-a => pfx-b}sfx
1371          * pfx{sfx-a => sfx-b}
1372          * name-a => name-b
1373          */
1374         a_midlen = len_a - pfx_length - sfx_length;
1375         b_midlen = len_b - pfx_length - sfx_length;
1376         if (a_midlen < 0)
1377                 a_midlen = 0;
1378         if (b_midlen < 0)
1379                 b_midlen = 0;
1380
1381         strbuf_grow(&name, pfx_length + a_midlen + b_midlen + sfx_length + 7);
1382         if (pfx_length + sfx_length) {
1383                 strbuf_add(&name, a, pfx_length);
1384                 strbuf_addch(&name, '{');
1385         }
1386         strbuf_add(&name, a + pfx_length, a_midlen);
1387         strbuf_addstr(&name, " => ");
1388         strbuf_add(&name, b + pfx_length, b_midlen);
1389         if (pfx_length + sfx_length) {
1390                 strbuf_addch(&name, '}');
1391                 strbuf_add(&name, a + len_a - sfx_length, sfx_length);
1392         }
1393         return strbuf_detach(&name, NULL);
1394 }
1395
1396 struct diffstat_t {
1397         int nr;
1398         int alloc;
1399         struct diffstat_file {
1400                 char *from_name;
1401                 char *name;
1402                 char *print_name;
1403                 unsigned is_unmerged:1;
1404                 unsigned is_binary:1;
1405                 unsigned is_renamed:1;
1406                 unsigned is_interesting:1;
1407                 uintmax_t added, deleted;
1408         } **files;
1409 };
1410
1411 static struct diffstat_file *diffstat_add(struct diffstat_t *diffstat,
1412                                           const char *name_a,
1413                                           const char *name_b)
1414 {
1415         struct diffstat_file *x;
1416         x = xcalloc(1, sizeof(*x));
1417         ALLOC_GROW(diffstat->files, diffstat->nr + 1, diffstat->alloc);
1418         diffstat->files[diffstat->nr++] = x;
1419         if (name_b) {
1420                 x->from_name = xstrdup(name_a);
1421                 x->name = xstrdup(name_b);
1422                 x->is_renamed = 1;
1423         }
1424         else {
1425                 x->from_name = NULL;
1426                 x->name = xstrdup(name_a);
1427         }
1428         return x;
1429 }
1430
1431 static void diffstat_consume(void *priv, char *line, unsigned long len)
1432 {
1433         struct diffstat_t *diffstat = priv;
1434         struct diffstat_file *x = diffstat->files[diffstat->nr - 1];
1435
1436         if (line[0] == '+')
1437                 x->added++;
1438         else if (line[0] == '-')
1439                 x->deleted++;
1440 }
1441
1442 const char mime_boundary_leader[] = "------------";
1443
1444 static int scale_linear(int it, int width, int max_change)
1445 {
1446         if (!it)
1447                 return 0;
1448         /*
1449          * make sure that at least one '-' or '+' is printed if
1450          * there is any change to this path. The easiest way is to
1451          * scale linearly as if the alloted width is one column shorter
1452          * than it is, and then add 1 to the result.
1453          */
1454         return 1 + (it * (width - 1) / max_change);
1455 }
1456
1457 static void show_name(FILE *file,
1458                       const char *prefix, const char *name, int len)
1459 {
1460         fprintf(file, " %s%-*s |", prefix, len, name);
1461 }
1462
1463 static void show_graph(FILE *file, char ch, int cnt, const char *set, const char *reset)
1464 {
1465         if (cnt <= 0)
1466                 return;
1467         fprintf(file, "%s", set);
1468         while (cnt--)
1469                 putc(ch, file);
1470         fprintf(file, "%s", reset);
1471 }
1472
1473 static void fill_print_name(struct diffstat_file *file)
1474 {
1475         char *pname;
1476
1477         if (file->print_name)
1478                 return;
1479
1480         if (!file->is_renamed) {
1481                 struct strbuf buf = STRBUF_INIT;
1482                 if (quote_c_style(file->name, &buf, NULL, 0)) {
1483                         pname = strbuf_detach(&buf, NULL);
1484                 } else {
1485                         pname = file->name;
1486                         strbuf_release(&buf);
1487                 }
1488         } else {
1489                 pname = pprint_rename(file->from_name, file->name);
1490         }
1491         file->print_name = pname;
1492 }
1493
1494 int print_stat_summary(FILE *fp, int files, int insertions, int deletions)
1495 {
1496         struct strbuf sb = STRBUF_INIT;
1497         int ret;
1498
1499         if (!files) {
1500                 assert(insertions == 0 && deletions == 0);
1501                 return fprintf(fp, "%s\n", " 0 files changed");
1502         }
1503
1504         strbuf_addf(&sb,
1505                     (files == 1) ? " %d file changed" : " %d files changed",
1506                     files);
1507
1508         /*
1509          * For binary diff, the caller may want to print "x files
1510          * changed" with insertions == 0 && deletions == 0.
1511          *
1512          * Not omitting "0 insertions(+), 0 deletions(-)" in this case
1513          * is probably less confusing (i.e skip over "2 files changed
1514          * but nothing about added/removed lines? Is this a bug in Git?").
1515          */
1516         if (insertions || deletions == 0) {
1517                 strbuf_addf(&sb,
1518                             (insertions == 1) ? ", %d insertion(+)" : ", %d insertions(+)",
1519                             insertions);
1520         }
1521
1522         if (deletions || insertions == 0) {
1523                 strbuf_addf(&sb,
1524                             (deletions == 1) ? ", %d deletion(-)" : ", %d deletions(-)",
1525                             deletions);
1526         }
1527         strbuf_addch(&sb, '\n');
1528         ret = fputs(sb.buf, fp);
1529         strbuf_release(&sb);
1530         return ret;
1531 }
1532
1533 static void show_stats(struct diffstat_t *data, struct diff_options *options)
1534 {
1535         int i, len, add, del, adds = 0, dels = 0;
1536         uintmax_t max_change = 0, max_len = 0;
1537         int total_files = data->nr, count;
1538         int width, name_width, graph_width, number_width = 0, bin_width = 0;
1539         const char *reset, *add_c, *del_c;
1540         const char *line_prefix = "";
1541         int extra_shown = 0;
1542
1543         if (data->nr == 0)
1544                 return;
1545
1546         line_prefix = diff_line_prefix(options);
1547         count = options->stat_count ? options->stat_count : data->nr;
1548
1549         reset = diff_get_color_opt(options, DIFF_RESET);
1550         add_c = diff_get_color_opt(options, DIFF_FILE_NEW);
1551         del_c = diff_get_color_opt(options, DIFF_FILE_OLD);
1552
1553         /*
1554          * Find the longest filename and max number of changes
1555          */
1556         for (i = 0; (i < count) && (i < data->nr); i++) {
1557                 struct diffstat_file *file = data->files[i];
1558                 uintmax_t change = file->added + file->deleted;
1559
1560                 if (!file->is_interesting && (change == 0)) {
1561                         count++; /* not shown == room for one more */
1562                         continue;
1563                 }
1564                 fill_print_name(file);
1565                 len = strlen(file->print_name);
1566                 if (max_len < len)
1567                         max_len = len;
1568
1569                 if (file->is_unmerged) {
1570                         /* "Unmerged" is 8 characters */
1571                         bin_width = bin_width < 8 ? 8 : bin_width;
1572                         continue;
1573                 }
1574                 if (file->is_binary) {
1575                         /* "Bin XXX -> YYY bytes" */
1576                         int w = 14 + decimal_width(file->added)
1577                                 + decimal_width(file->deleted);
1578                         bin_width = bin_width < w ? w : bin_width;
1579                         /* Display change counts aligned with "Bin" */
1580                         number_width = 3;
1581                         continue;
1582                 }
1583
1584                 if (max_change < change)
1585                         max_change = change;
1586         }
1587         count = i; /* where we can stop scanning in data->files[] */
1588
1589         /*
1590          * We have width = stat_width or term_columns() columns total.
1591          * We want a maximum of min(max_len, stat_name_width) for the name part.
1592          * We want a maximum of min(max_change, stat_graph_width) for the +- part.
1593          * We also need 1 for " " and 4 + decimal_width(max_change)
1594          * for " | NNNN " and one the empty column at the end, altogether
1595          * 6 + decimal_width(max_change).
1596          *
1597          * If there's not enough space, we will use the smaller of
1598          * stat_name_width (if set) and 5/8*width for the filename,
1599          * and the rest for constant elements + graph part, but no more
1600          * than stat_graph_width for the graph part.
1601          * (5/8 gives 50 for filename and 30 for the constant parts + graph
1602          * for the standard terminal size).
1603          *
1604          * In other words: stat_width limits the maximum width, and
1605          * stat_name_width fixes the maximum width of the filename,
1606          * and is also used to divide available columns if there
1607          * aren't enough.
1608          *
1609          * Binary files are displayed with "Bin XXX -> YYY bytes"
1610          * instead of the change count and graph. This part is treated
1611          * similarly to the graph part, except that it is not
1612          * "scaled". If total width is too small to accommodate the
1613          * guaranteed minimum width of the filename part and the
1614          * separators and this message, this message will "overflow"
1615          * making the line longer than the maximum width.
1616          */
1617
1618         if (options->stat_width == -1)
1619                 width = term_columns() - options->output_prefix_length;
1620         else
1621                 width = options->stat_width ? options->stat_width : 80;
1622         number_width = decimal_width(max_change) > number_width ?
1623                 decimal_width(max_change) : number_width;
1624
1625         if (options->stat_graph_width == -1)
1626                 options->stat_graph_width = diff_stat_graph_width;
1627
1628         /*
1629          * Guarantee 3/8*16==6 for the graph part
1630          * and 5/8*16==10 for the filename part
1631          */
1632         if (width < 16 + 6 + number_width)
1633                 width = 16 + 6 + number_width;
1634
1635         /*
1636          * First assign sizes that are wanted, ignoring available width.
1637          * strlen("Bin XXX -> YYY bytes") == bin_width, and the part
1638          * starting from "XXX" should fit in graph_width.
1639          */
1640         graph_width = max_change + 4 > bin_width ? max_change : bin_width - 4;
1641         if (options->stat_graph_width &&
1642             options->stat_graph_width < graph_width)
1643                 graph_width = options->stat_graph_width;
1644
1645         name_width = (options->stat_name_width > 0 &&
1646                       options->stat_name_width < max_len) ?
1647                 options->stat_name_width : max_len;
1648
1649         /*
1650          * Adjust adjustable widths not to exceed maximum width
1651          */
1652         if (name_width + number_width + 6 + graph_width > width) {
1653                 if (graph_width > width * 3/8 - number_width - 6) {
1654                         graph_width = width * 3/8 - number_width - 6;
1655                         if (graph_width < 6)
1656                                 graph_width = 6;
1657                 }
1658
1659                 if (options->stat_graph_width &&
1660                     graph_width > options->stat_graph_width)
1661                         graph_width = options->stat_graph_width;
1662                 if (name_width > width - number_width - 6 - graph_width)
1663                         name_width = width - number_width - 6 - graph_width;
1664                 else
1665                         graph_width = width - number_width - 6 - name_width;
1666         }
1667
1668         /*
1669          * From here name_width is the width of the name area,
1670          * and graph_width is the width of the graph area.
1671          * max_change is used to scale graph properly.
1672          */
1673         for (i = 0; i < count; i++) {
1674                 const char *prefix = "";
1675                 struct diffstat_file *file = data->files[i];
1676                 char *name = file->print_name;
1677                 uintmax_t added = file->added;
1678                 uintmax_t deleted = file->deleted;
1679                 int name_len;
1680
1681                 if (!file->is_interesting && (added + deleted == 0))
1682                         continue;
1683
1684                 /*
1685                  * "scale" the filename
1686                  */
1687                 len = name_width;
1688                 name_len = strlen(name);
1689                 if (name_width < name_len) {
1690                         char *slash;
1691                         prefix = "...";
1692                         len -= 3;
1693                         name += name_len - len;
1694                         slash = strchr(name, '/');
1695                         if (slash)
1696                                 name = slash;
1697                 }
1698
1699                 if (file->is_binary) {
1700                         fprintf(options->file, "%s", line_prefix);
1701                         show_name(options->file, prefix, name, len);
1702                         fprintf(options->file, " %*s", number_width, "Bin");
1703                         if (!added && !deleted) {
1704                                 putc('\n', options->file);
1705                                 continue;
1706                         }
1707                         fprintf(options->file, " %s%"PRIuMAX"%s",
1708                                 del_c, deleted, reset);
1709                         fprintf(options->file, " -> ");
1710                         fprintf(options->file, "%s%"PRIuMAX"%s",
1711                                 add_c, added, reset);
1712                         fprintf(options->file, " bytes");
1713                         fprintf(options->file, "\n");
1714                         continue;
1715                 }
1716                 else if (file->is_unmerged) {
1717                         fprintf(options->file, "%s", line_prefix);
1718                         show_name(options->file, prefix, name, len);
1719                         fprintf(options->file, " Unmerged\n");
1720                         continue;
1721                 }
1722
1723                 /*
1724                  * scale the add/delete
1725                  */
1726                 add = added;
1727                 del = deleted;
1728
1729                 if (graph_width <= max_change) {
1730                         int total = scale_linear(add + del, graph_width, max_change);
1731                         if (total < 2 && add && del)
1732                                 /* width >= 2 due to the sanity check */
1733                                 total = 2;
1734                         if (add < del) {
1735                                 add = scale_linear(add, graph_width, max_change);
1736                                 del = total - add;
1737                         } else {
1738                                 del = scale_linear(del, graph_width, max_change);
1739                                 add = total - del;
1740                         }
1741                 }
1742                 fprintf(options->file, "%s", line_prefix);
1743                 show_name(options->file, prefix, name, len);
1744                 fprintf(options->file, " %*"PRIuMAX"%s",
1745                         number_width, added + deleted,
1746                         added + deleted ? " " : "");
1747                 show_graph(options->file, '+', add, add_c, reset);
1748                 show_graph(options->file, '-', del, del_c, reset);
1749                 fprintf(options->file, "\n");
1750         }
1751
1752         for (i = 0; i < data->nr; i++) {
1753                 struct diffstat_file *file = data->files[i];
1754                 uintmax_t added = file->added;
1755                 uintmax_t deleted = file->deleted;
1756
1757                 if (file->is_unmerged ||
1758                     (!file->is_interesting && (added + deleted == 0))) {
1759                         total_files--;
1760                         continue;
1761                 }
1762
1763                 if (!file->is_binary) {
1764                         adds += added;
1765                         dels += deleted;
1766                 }
1767                 if (i < count)
1768                         continue;
1769                 if (!extra_shown)
1770                         fprintf(options->file, "%s ...\n", line_prefix);
1771                 extra_shown = 1;
1772         }
1773         fprintf(options->file, "%s", line_prefix);
1774         print_stat_summary(options->file, total_files, adds, dels);
1775 }
1776
1777 static void show_shortstats(struct diffstat_t *data, struct diff_options *options)
1778 {
1779         int i, adds = 0, dels = 0, total_files = data->nr;
1780
1781         if (data->nr == 0)
1782                 return;
1783
1784         for (i = 0; i < data->nr; i++) {
1785                 int added = data->files[i]->added;
1786                 int deleted= data->files[i]->deleted;
1787
1788                 if (data->files[i]->is_unmerged ||
1789                     (!data->files[i]->is_interesting && (added + deleted == 0))) {
1790                         total_files--;
1791                 } else if (!data->files[i]->is_binary) { /* don't count bytes */
1792                         adds += added;
1793                         dels += deleted;
1794                 }
1795         }
1796         fprintf(options->file, "%s", diff_line_prefix(options));
1797         print_stat_summary(options->file, total_files, adds, dels);
1798 }
1799
1800 static void show_numstat(struct diffstat_t *data, struct diff_options *options)
1801 {
1802         int i;
1803
1804         if (data->nr == 0)
1805                 return;
1806
1807         for (i = 0; i < data->nr; i++) {
1808                 struct diffstat_file *file = data->files[i];
1809
1810                 fprintf(options->file, "%s", diff_line_prefix(options));
1811
1812                 if (file->is_binary)
1813                         fprintf(options->file, "-\t-\t");
1814                 else
1815                         fprintf(options->file,
1816                                 "%"PRIuMAX"\t%"PRIuMAX"\t",
1817                                 file->added, file->deleted);
1818                 if (options->line_termination) {
1819                         fill_print_name(file);
1820                         if (!file->is_renamed)
1821                                 write_name_quoted(file->name, options->file,
1822                                                   options->line_termination);
1823                         else {
1824                                 fputs(file->print_name, options->file);
1825                                 putc(options->line_termination, options->file);
1826                         }
1827                 } else {
1828                         if (file->is_renamed) {
1829                                 putc('\0', options->file);
1830                                 write_name_quoted(file->from_name, options->file, '\0');
1831                         }
1832                         write_name_quoted(file->name, options->file, '\0');
1833                 }
1834         }
1835 }
1836
1837 struct dirstat_file {
1838         const char *name;
1839         unsigned long changed;
1840 };
1841
1842 struct dirstat_dir {
1843         struct dirstat_file *files;
1844         int alloc, nr, permille, cumulative;
1845 };
1846
1847 static long gather_dirstat(struct diff_options *opt, struct dirstat_dir *dir,
1848                 unsigned long changed, const char *base, int baselen)
1849 {
1850         unsigned long this_dir = 0;
1851         unsigned int sources = 0;
1852         const char *line_prefix = diff_line_prefix(opt);
1853
1854         while (dir->nr) {
1855                 struct dirstat_file *f = dir->files;
1856                 int namelen = strlen(f->name);
1857                 unsigned long this;
1858                 char *slash;
1859
1860                 if (namelen < baselen)
1861                         break;
1862                 if (memcmp(f->name, base, baselen))
1863                         break;
1864                 slash = strchr(f->name + baselen, '/');
1865                 if (slash) {
1866                         int newbaselen = slash + 1 - f->name;
1867                         this = gather_dirstat(opt, dir, changed, f->name, newbaselen);
1868                         sources++;
1869                 } else {
1870                         this = f->changed;
1871                         dir->files++;
1872                         dir->nr--;
1873                         sources += 2;
1874                 }
1875                 this_dir += this;
1876         }
1877
1878         /*
1879          * We don't report dirstat's for
1880          *  - the top level
1881          *  - or cases where everything came from a single directory
1882          *    under this directory (sources == 1).
1883          */
1884         if (baselen && sources != 1) {
1885                 if (this_dir) {
1886                         int permille = this_dir * 1000 / changed;
1887                         if (permille >= dir->permille) {
1888                                 fprintf(opt->file, "%s%4d.%01d%% %.*s\n", line_prefix,
1889                                         permille / 10, permille % 10, baselen, base);
1890                                 if (!dir->cumulative)
1891                                         return 0;
1892                         }
1893                 }
1894         }
1895         return this_dir;
1896 }
1897
1898 static int dirstat_compare(const void *_a, const void *_b)
1899 {
1900         const struct dirstat_file *a = _a;
1901         const struct dirstat_file *b = _b;
1902         return strcmp(a->name, b->name);
1903 }
1904
1905 static void show_dirstat(struct diff_options *options)
1906 {
1907         int i;
1908         unsigned long changed;
1909         struct dirstat_dir dir;
1910         struct diff_queue_struct *q = &diff_queued_diff;
1911
1912         dir.files = NULL;
1913         dir.alloc = 0;
1914         dir.nr = 0;
1915         dir.permille = options->dirstat_permille;
1916         dir.cumulative = DIFF_OPT_TST(options, DIRSTAT_CUMULATIVE);
1917
1918         changed = 0;
1919         for (i = 0; i < q->nr; i++) {
1920                 struct diff_filepair *p = q->queue[i];
1921                 const char *name;
1922                 unsigned long copied, added, damage;
1923                 int content_changed;
1924
1925                 name = p->two->path ? p->two->path : p->one->path;
1926
1927                 if (p->one->oid_valid && p->two->oid_valid)
1928                         content_changed = oidcmp(&p->one->oid, &p->two->oid);
1929                 else
1930                         content_changed = 1;
1931
1932                 if (!content_changed) {
1933                         /*
1934                          * The SHA1 has not changed, so pre-/post-content is
1935                          * identical. We can therefore skip looking at the
1936                          * file contents altogether.
1937                          */
1938                         damage = 0;
1939                         goto found_damage;
1940                 }
1941
1942                 if (DIFF_OPT_TST(options, DIRSTAT_BY_FILE)) {
1943                         /*
1944                          * In --dirstat-by-file mode, we don't really need to
1945                          * look at the actual file contents at all.
1946                          * The fact that the SHA1 changed is enough for us to
1947                          * add this file to the list of results
1948                          * (with each file contributing equal damage).
1949                          */
1950                         damage = 1;
1951                         goto found_damage;
1952                 }
1953
1954                 if (DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
1955                         diff_populate_filespec(p->one, 0);
1956                         diff_populate_filespec(p->two, 0);
1957                         diffcore_count_changes(p->one, p->two, NULL, NULL, 0,
1958                                                &copied, &added);
1959                         diff_free_filespec_data(p->one);
1960                         diff_free_filespec_data(p->two);
1961                 } else if (DIFF_FILE_VALID(p->one)) {
1962                         diff_populate_filespec(p->one, CHECK_SIZE_ONLY);
1963                         copied = added = 0;
1964                         diff_free_filespec_data(p->one);
1965                 } else if (DIFF_FILE_VALID(p->two)) {
1966                         diff_populate_filespec(p->two, CHECK_SIZE_ONLY);
1967                         copied = 0;
1968                         added = p->two->size;
1969                         diff_free_filespec_data(p->two);
1970                 } else
1971                         continue;
1972
1973                 /*
1974                  * Original minus copied is the removed material,
1975                  * added is the new material.  They are both damages
1976                  * made to the preimage.
1977                  * If the resulting damage is zero, we know that
1978                  * diffcore_count_changes() considers the two entries to
1979                  * be identical, but since content_changed is true, we
1980                  * know that there must have been _some_ kind of change,
1981                  * so we force all entries to have damage > 0.
1982                  */
1983                 damage = (p->one->size - copied) + added;
1984                 if (!damage)
1985                         damage = 1;
1986
1987 found_damage:
1988                 ALLOC_GROW(dir.files, dir.nr + 1, dir.alloc);
1989                 dir.files[dir.nr].name = name;
1990                 dir.files[dir.nr].changed = damage;
1991                 changed += damage;
1992                 dir.nr++;
1993         }
1994
1995         /* This can happen even with many files, if everything was renames */
1996         if (!changed)
1997                 return;
1998
1999         /* Show all directories with more than x% of the changes */
2000         qsort(dir.files, dir.nr, sizeof(dir.files[0]), dirstat_compare);
2001         gather_dirstat(options, &dir, changed, "", 0);
2002 }
2003
2004 static void show_dirstat_by_line(struct diffstat_t *data, struct diff_options *options)
2005 {
2006         int i;
2007         unsigned long changed;
2008         struct dirstat_dir dir;
2009
2010         if (data->nr == 0)
2011                 return;
2012
2013         dir.files = NULL;
2014         dir.alloc = 0;
2015         dir.nr = 0;
2016         dir.permille = options->dirstat_permille;
2017         dir.cumulative = DIFF_OPT_TST(options, DIRSTAT_CUMULATIVE);
2018
2019         changed = 0;
2020         for (i = 0; i < data->nr; i++) {
2021                 struct diffstat_file *file = data->files[i];
2022                 unsigned long damage = file->added + file->deleted;
2023                 if (file->is_binary)
2024                         /*
2025                          * binary files counts bytes, not lines. Must find some
2026                          * way to normalize binary bytes vs. textual lines.
2027                          * The following heuristic assumes that there are 64
2028                          * bytes per "line".
2029                          * This is stupid and ugly, but very cheap...
2030                          */
2031                         damage = (damage + 63) / 64;
2032                 ALLOC_GROW(dir.files, dir.nr + 1, dir.alloc);
2033                 dir.files[dir.nr].name = file->name;
2034                 dir.files[dir.nr].changed = damage;
2035                 changed += damage;
2036                 dir.nr++;
2037         }
2038
2039         /* This can happen even with many files, if everything was renames */
2040         if (!changed)
2041                 return;
2042
2043         /* Show all directories with more than x% of the changes */
2044         qsort(dir.files, dir.nr, sizeof(dir.files[0]), dirstat_compare);
2045         gather_dirstat(options, &dir, changed, "", 0);
2046 }
2047
2048 static void free_diffstat_info(struct diffstat_t *diffstat)
2049 {
2050         int i;
2051         for (i = 0; i < diffstat->nr; i++) {
2052                 struct diffstat_file *f = diffstat->files[i];
2053                 if (f->name != f->print_name)
2054                         free(f->print_name);
2055                 free(f->name);
2056                 free(f->from_name);
2057                 free(f);
2058         }
2059         free(diffstat->files);
2060 }
2061
2062 struct checkdiff_t {
2063         const char *filename;
2064         int lineno;
2065         int conflict_marker_size;
2066         struct diff_options *o;
2067         unsigned ws_rule;
2068         unsigned status;
2069 };
2070
2071 static int is_conflict_marker(const char *line, int marker_size, unsigned long len)
2072 {
2073         char firstchar;
2074         int cnt;
2075
2076         if (len < marker_size + 1)
2077                 return 0;
2078         firstchar = line[0];
2079         switch (firstchar) {
2080         case '=': case '>': case '<': case '|':
2081                 break;
2082         default:
2083                 return 0;
2084         }
2085         for (cnt = 1; cnt < marker_size; cnt++)
2086                 if (line[cnt] != firstchar)
2087                         return 0;
2088         /* line[1] thru line[marker_size-1] are same as firstchar */
2089         if (len < marker_size + 1 || !isspace(line[marker_size]))
2090                 return 0;
2091         return 1;
2092 }
2093
2094 static void checkdiff_consume(void *priv, char *line, unsigned long len)
2095 {
2096         struct checkdiff_t *data = priv;
2097         int marker_size = data->conflict_marker_size;
2098         const char *ws = diff_get_color(data->o->use_color, DIFF_WHITESPACE);
2099         const char *reset = diff_get_color(data->o->use_color, DIFF_RESET);
2100         const char *set = diff_get_color(data->o->use_color, DIFF_FILE_NEW);
2101         char *err;
2102         const char *line_prefix;
2103
2104         assert(data->o);
2105         line_prefix = diff_line_prefix(data->o);
2106
2107         if (line[0] == '+') {
2108                 unsigned bad;
2109                 data->lineno++;
2110                 if (is_conflict_marker(line + 1, marker_size, len - 1)) {
2111                         data->status |= 1;
2112                         fprintf(data->o->file,
2113                                 "%s%s:%d: leftover conflict marker\n",
2114                                 line_prefix, data->filename, data->lineno);
2115                 }
2116                 bad = ws_check(line + 1, len - 1, data->ws_rule);
2117                 if (!bad)
2118                         return;
2119                 data->status |= bad;
2120                 err = whitespace_error_string(bad);
2121                 fprintf(data->o->file, "%s%s:%d: %s.\n",
2122                         line_prefix, data->filename, data->lineno, err);
2123                 free(err);
2124                 emit_line(data->o, set, reset, line, 1);
2125                 ws_check_emit(line + 1, len - 1, data->ws_rule,
2126                               data->o->file, set, reset, ws);
2127         } else if (line[0] == ' ') {
2128                 data->lineno++;
2129         } else if (line[0] == '@') {
2130                 char *plus = strchr(line, '+');
2131                 if (plus)
2132                         data->lineno = strtol(plus, NULL, 10) - 1;
2133                 else
2134                         die("invalid diff");
2135         }
2136 }
2137
2138 static unsigned char *deflate_it(char *data,
2139                                  unsigned long size,
2140                                  unsigned long *result_size)
2141 {
2142         int bound;
2143         unsigned char *deflated;
2144         git_zstream stream;
2145
2146         git_deflate_init(&stream, zlib_compression_level);
2147         bound = git_deflate_bound(&stream, size);
2148         deflated = xmalloc(bound);
2149         stream.next_out = deflated;
2150         stream.avail_out = bound;
2151
2152         stream.next_in = (unsigned char *)data;
2153         stream.avail_in = size;
2154         while (git_deflate(&stream, Z_FINISH) == Z_OK)
2155                 ; /* nothing */
2156         git_deflate_end(&stream);
2157         *result_size = stream.total_out;
2158         return deflated;
2159 }
2160
2161 static void emit_binary_diff_body(FILE *file, mmfile_t *one, mmfile_t *two,
2162                                   const char *prefix)
2163 {
2164         void *cp;
2165         void *delta;
2166         void *deflated;
2167         void *data;
2168         unsigned long orig_size;
2169         unsigned long delta_size;
2170         unsigned long deflate_size;
2171         unsigned long data_size;
2172
2173         /* We could do deflated delta, or we could do just deflated two,
2174          * whichever is smaller.
2175          */
2176         delta = NULL;
2177         deflated = deflate_it(two->ptr, two->size, &deflate_size);
2178         if (one->size && two->size) {
2179                 delta = diff_delta(one->ptr, one->size,
2180                                    two->ptr, two->size,
2181                                    &delta_size, deflate_size);
2182                 if (delta) {
2183                         void *to_free = delta;
2184                         orig_size = delta_size;
2185                         delta = deflate_it(delta, delta_size, &delta_size);
2186                         free(to_free);
2187                 }
2188         }
2189
2190         if (delta && delta_size < deflate_size) {
2191                 fprintf(file, "%sdelta %lu\n", prefix, orig_size);
2192                 free(deflated);
2193                 data = delta;
2194                 data_size = delta_size;
2195         }
2196         else {
2197                 fprintf(file, "%sliteral %lu\n", prefix, two->size);
2198                 free(delta);
2199                 data = deflated;
2200                 data_size = deflate_size;
2201         }
2202
2203         /* emit data encoded in base85 */
2204         cp = data;
2205         while (data_size) {
2206                 int bytes = (52 < data_size) ? 52 : data_size;
2207                 char line[70];
2208                 data_size -= bytes;
2209                 if (bytes <= 26)
2210                         line[0] = bytes + 'A' - 1;
2211                 else
2212                         line[0] = bytes - 26 + 'a' - 1;
2213                 encode_85(line + 1, cp, bytes);
2214                 cp = (char *) cp + bytes;
2215                 fprintf(file, "%s", prefix);
2216                 fputs(line, file);
2217                 fputc('\n', file);
2218         }
2219         fprintf(file, "%s\n", prefix);
2220         free(data);
2221 }
2222
2223 static void emit_binary_diff(FILE *file, mmfile_t *one, mmfile_t *two,
2224                              const char *prefix)
2225 {
2226         fprintf(file, "%sGIT binary patch\n", prefix);
2227         emit_binary_diff_body(file, one, two, prefix);
2228         emit_binary_diff_body(file, two, one, prefix);
2229 }
2230
2231 int diff_filespec_is_binary(struct diff_filespec *one)
2232 {
2233         if (one->is_binary == -1) {
2234                 diff_filespec_load_driver(one);
2235                 if (one->driver->binary != -1)
2236                         one->is_binary = one->driver->binary;
2237                 else {
2238                         if (!one->data && DIFF_FILE_VALID(one))
2239                                 diff_populate_filespec(one, CHECK_BINARY);
2240                         if (one->is_binary == -1 && one->data)
2241                                 one->is_binary = buffer_is_binary(one->data,
2242                                                 one->size);
2243                         if (one->is_binary == -1)
2244                                 one->is_binary = 0;
2245                 }
2246         }
2247         return one->is_binary;
2248 }
2249
2250 static const struct userdiff_funcname *diff_funcname_pattern(struct diff_filespec *one)
2251 {
2252         diff_filespec_load_driver(one);
2253         return one->driver->funcname.pattern ? &one->driver->funcname : NULL;
2254 }
2255
2256 void diff_set_mnemonic_prefix(struct diff_options *options, const char *a, const char *b)
2257 {
2258         if (!options->a_prefix)
2259                 options->a_prefix = a;
2260         if (!options->b_prefix)
2261                 options->b_prefix = b;
2262 }
2263
2264 struct userdiff_driver *get_textconv(struct diff_filespec *one)
2265 {
2266         if (!DIFF_FILE_VALID(one))
2267                 return NULL;
2268
2269         diff_filespec_load_driver(one);
2270         return userdiff_get_textconv(one->driver);
2271 }
2272
2273 static void builtin_diff(const char *name_a,
2274                          const char *name_b,
2275                          struct diff_filespec *one,
2276                          struct diff_filespec *two,
2277                          const char *xfrm_msg,
2278                          int must_show_header,
2279                          struct diff_options *o,
2280                          int complete_rewrite)
2281 {
2282         mmfile_t mf1, mf2;
2283         const char *lbl[2];
2284         char *a_one, *b_two;
2285         const char *meta = diff_get_color_opt(o, DIFF_METAINFO);
2286         const char *reset = diff_get_color_opt(o, DIFF_RESET);
2287         const char *a_prefix, *b_prefix;
2288         struct userdiff_driver *textconv_one = NULL;
2289         struct userdiff_driver *textconv_two = NULL;
2290         struct strbuf header = STRBUF_INIT;
2291         const char *line_prefix = diff_line_prefix(o);
2292
2293         if (DIFF_OPT_TST(o, SUBMODULE_LOG) &&
2294                         (!one->mode || S_ISGITLINK(one->mode)) &&
2295                         (!two->mode || S_ISGITLINK(two->mode))) {
2296                 const char *del = diff_get_color_opt(o, DIFF_FILE_OLD);
2297                 const char *add = diff_get_color_opt(o, DIFF_FILE_NEW);
2298                 show_submodule_summary(o->file, one->path ? one->path : two->path,
2299                                 line_prefix,
2300                                 one->oid.hash, two->oid.hash,
2301                                 two->dirty_submodule,
2302                                 meta, del, add, reset);
2303                 return;
2304         }
2305
2306         if (DIFF_OPT_TST(o, ALLOW_TEXTCONV)) {
2307                 textconv_one = get_textconv(one);
2308                 textconv_two = get_textconv(two);
2309         }
2310
2311         diff_set_mnemonic_prefix(o, "a/", "b/");
2312         if (DIFF_OPT_TST(o, REVERSE_DIFF)) {
2313                 a_prefix = o->b_prefix;
2314                 b_prefix = o->a_prefix;
2315         } else {
2316                 a_prefix = o->a_prefix;
2317                 b_prefix = o->b_prefix;
2318         }
2319
2320         /* Never use a non-valid filename anywhere if at all possible */
2321         name_a = DIFF_FILE_VALID(one) ? name_a : name_b;
2322         name_b = DIFF_FILE_VALID(two) ? name_b : name_a;
2323
2324         a_one = quote_two(a_prefix, name_a + (*name_a == '/'));
2325         b_two = quote_two(b_prefix, name_b + (*name_b == '/'));
2326         lbl[0] = DIFF_FILE_VALID(one) ? a_one : "/dev/null";
2327         lbl[1] = DIFF_FILE_VALID(two) ? b_two : "/dev/null";
2328         strbuf_addf(&header, "%s%sdiff --git %s %s%s\n", line_prefix, meta, a_one, b_two, reset);
2329         if (lbl[0][0] == '/') {
2330                 /* /dev/null */
2331                 strbuf_addf(&header, "%s%snew file mode %06o%s\n", line_prefix, meta, two->mode, reset);
2332                 if (xfrm_msg)
2333                         strbuf_addstr(&header, xfrm_msg);
2334                 must_show_header = 1;
2335         }
2336         else if (lbl[1][0] == '/') {
2337                 strbuf_addf(&header, "%s%sdeleted file mode %06o%s\n", line_prefix, meta, one->mode, reset);
2338                 if (xfrm_msg)
2339                         strbuf_addstr(&header, xfrm_msg);
2340                 must_show_header = 1;
2341         }
2342         else {
2343                 if (one->mode != two->mode) {
2344                         strbuf_addf(&header, "%s%sold mode %06o%s\n", line_prefix, meta, one->mode, reset);
2345                         strbuf_addf(&header, "%s%snew mode %06o%s\n", line_prefix, meta, two->mode, reset);
2346                         must_show_header = 1;
2347                 }
2348                 if (xfrm_msg)
2349                         strbuf_addstr(&header, xfrm_msg);
2350
2351                 /*
2352                  * we do not run diff between different kind
2353                  * of objects.
2354                  */
2355                 if ((one->mode ^ two->mode) & S_IFMT)
2356                         goto free_ab_and_return;
2357                 if (complete_rewrite &&
2358                     (textconv_one || !diff_filespec_is_binary(one)) &&
2359                     (textconv_two || !diff_filespec_is_binary(two))) {
2360                         fprintf(o->file, "%s", header.buf);
2361                         strbuf_reset(&header);
2362                         emit_rewrite_diff(name_a, name_b, one, two,
2363                                                 textconv_one, textconv_two, o);
2364                         o->found_changes = 1;
2365                         goto free_ab_and_return;
2366                 }
2367         }
2368
2369         if (o->irreversible_delete && lbl[1][0] == '/') {
2370                 fprintf(o->file, "%s", header.buf);
2371                 strbuf_reset(&header);
2372                 goto free_ab_and_return;
2373         } else if (!DIFF_OPT_TST(o, TEXT) &&
2374             ( (!textconv_one && diff_filespec_is_binary(one)) ||
2375               (!textconv_two && diff_filespec_is_binary(two)) )) {
2376                 if (!one->data && !two->data &&
2377                     S_ISREG(one->mode) && S_ISREG(two->mode) &&
2378                     !DIFF_OPT_TST(o, BINARY)) {
2379                         if (!oidcmp(&one->oid, &two->oid)) {
2380                                 if (must_show_header)
2381                                         fprintf(o->file, "%s", header.buf);
2382                                 goto free_ab_and_return;
2383                         }
2384                         fprintf(o->file, "%s", header.buf);
2385                         fprintf(o->file, "%sBinary files %s and %s differ\n",
2386                                 line_prefix, lbl[0], lbl[1]);
2387                         goto free_ab_and_return;
2388                 }
2389                 if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
2390                         die("unable to read files to diff");
2391                 /* Quite common confusing case */
2392                 if (mf1.size == mf2.size &&
2393                     !memcmp(mf1.ptr, mf2.ptr, mf1.size)) {
2394                         if (must_show_header)
2395                                 fprintf(o->file, "%s", header.buf);
2396                         goto free_ab_and_return;
2397                 }
2398                 fprintf(o->file, "%s", header.buf);
2399                 strbuf_reset(&header);
2400                 if (DIFF_OPT_TST(o, BINARY))
2401                         emit_binary_diff(o->file, &mf1, &mf2, line_prefix);
2402                 else
2403                         fprintf(o->file, "%sBinary files %s and %s differ\n",
2404                                 line_prefix, lbl[0], lbl[1]);
2405                 o->found_changes = 1;
2406         } else {
2407                 /* Crazy xdl interfaces.. */
2408                 const char *diffopts = getenv("GIT_DIFF_OPTS");
2409                 const char *v;
2410                 xpparam_t xpp;
2411                 xdemitconf_t xecfg;
2412                 struct emit_callback ecbdata;
2413                 const struct userdiff_funcname *pe;
2414
2415                 if (must_show_header) {
2416                         fprintf(o->file, "%s", header.buf);
2417                         strbuf_reset(&header);
2418                 }
2419
2420                 mf1.size = fill_textconv(textconv_one, one, &mf1.ptr);
2421                 mf2.size = fill_textconv(textconv_two, two, &mf2.ptr);
2422
2423                 pe = diff_funcname_pattern(one);
2424                 if (!pe)
2425                         pe = diff_funcname_pattern(two);
2426
2427                 memset(&xpp, 0, sizeof(xpp));
2428                 memset(&xecfg, 0, sizeof(xecfg));
2429                 memset(&ecbdata, 0, sizeof(ecbdata));
2430                 ecbdata.label_path = lbl;
2431                 ecbdata.color_diff = want_color(o->use_color);
2432                 ecbdata.ws_rule = whitespace_rule(name_b);
2433                 if (ecbdata.ws_rule & WS_BLANK_AT_EOF)
2434                         check_blank_at_eof(&mf1, &mf2, &ecbdata);
2435                 ecbdata.opt = o;
2436                 ecbdata.header = header.len ? &header : NULL;
2437                 xpp.flags = o->xdl_opts;
2438                 xecfg.ctxlen = o->context;
2439                 xecfg.interhunkctxlen = o->interhunkcontext;
2440                 xecfg.flags = XDL_EMIT_FUNCNAMES;
2441                 if (DIFF_OPT_TST(o, FUNCCONTEXT))
2442                         xecfg.flags |= XDL_EMIT_FUNCCONTEXT;
2443                 if (pe)
2444                         xdiff_set_find_func(&xecfg, pe->pattern, pe->cflags);
2445                 if (!diffopts)
2446                         ;
2447                 else if (skip_prefix(diffopts, "--unified=", &v))
2448                         xecfg.ctxlen = strtoul(v, NULL, 10);
2449                 else if (skip_prefix(diffopts, "-u", &v))
2450                         xecfg.ctxlen = strtoul(v, NULL, 10);
2451                 if (o->word_diff)
2452                         init_diff_words_data(&ecbdata, o, one, two);
2453                 if (xdi_diff_outf(&mf1, &mf2, fn_out_consume, &ecbdata,
2454                                   &xpp, &xecfg))
2455                         die("unable to generate diff for %s", one->path);
2456                 if (o->word_diff)
2457                         free_diff_words_data(&ecbdata);
2458                 if (textconv_one)
2459                         free(mf1.ptr);
2460                 if (textconv_two)
2461                         free(mf2.ptr);
2462                 xdiff_clear_find_func(&xecfg);
2463         }
2464
2465  free_ab_and_return:
2466         strbuf_release(&header);
2467         diff_free_filespec_data(one);
2468         diff_free_filespec_data(two);
2469         free(a_one);
2470         free(b_two);
2471         return;
2472 }
2473
2474 static void builtin_diffstat(const char *name_a, const char *name_b,
2475                              struct diff_filespec *one,
2476                              struct diff_filespec *two,
2477                              struct diffstat_t *diffstat,
2478                              struct diff_options *o,
2479                              struct diff_filepair *p)
2480 {
2481         mmfile_t mf1, mf2;
2482         struct diffstat_file *data;
2483         int same_contents;
2484         int complete_rewrite = 0;
2485
2486         if (!DIFF_PAIR_UNMERGED(p)) {
2487                 if (p->status == DIFF_STATUS_MODIFIED && p->score)
2488                         complete_rewrite = 1;
2489         }
2490
2491         data = diffstat_add(diffstat, name_a, name_b);
2492         data->is_interesting = p->status != DIFF_STATUS_UNKNOWN;
2493
2494         if (!one || !two) {
2495                 data->is_unmerged = 1;
2496                 return;
2497         }
2498
2499         same_contents = !oidcmp(&one->oid, &two->oid);
2500
2501         if (diff_filespec_is_binary(one) || diff_filespec_is_binary(two)) {
2502                 data->is_binary = 1;
2503                 if (same_contents) {
2504                         data->added = 0;
2505                         data->deleted = 0;
2506                 } else {
2507                         data->added = diff_filespec_size(two);
2508                         data->deleted = diff_filespec_size(one);
2509                 }
2510         }
2511
2512         else if (complete_rewrite) {
2513                 diff_populate_filespec(one, 0);
2514                 diff_populate_filespec(two, 0);
2515                 data->deleted = count_lines(one->data, one->size);
2516                 data->added = count_lines(two->data, two->size);
2517         }
2518
2519         else if (!same_contents) {
2520                 /* Crazy xdl interfaces.. */
2521                 xpparam_t xpp;
2522                 xdemitconf_t xecfg;
2523
2524                 if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
2525                         die("unable to read files to diff");
2526
2527                 memset(&xpp, 0, sizeof(xpp));
2528                 memset(&xecfg, 0, sizeof(xecfg));
2529                 xpp.flags = o->xdl_opts;
2530                 xecfg.ctxlen = o->context;
2531                 xecfg.interhunkctxlen = o->interhunkcontext;
2532                 if (xdi_diff_outf(&mf1, &mf2, diffstat_consume, diffstat,
2533                                   &xpp, &xecfg))
2534                         die("unable to generate diffstat for %s", one->path);
2535         }
2536
2537         diff_free_filespec_data(one);
2538         diff_free_filespec_data(two);
2539 }
2540
2541 static void builtin_checkdiff(const char *name_a, const char *name_b,
2542                               const char *attr_path,
2543                               struct diff_filespec *one,
2544                               struct diff_filespec *two,
2545                               struct diff_options *o)
2546 {
2547         mmfile_t mf1, mf2;
2548         struct checkdiff_t data;
2549
2550         if (!two)
2551                 return;
2552
2553         memset(&data, 0, sizeof(data));
2554         data.filename = name_b ? name_b : name_a;
2555         data.lineno = 0;
2556         data.o = o;
2557         data.ws_rule = whitespace_rule(attr_path);
2558         data.conflict_marker_size = ll_merge_marker_size(attr_path);
2559
2560         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
2561                 die("unable to read files to diff");
2562
2563         /*
2564          * All the other codepaths check both sides, but not checking
2565          * the "old" side here is deliberate.  We are checking the newly
2566          * introduced changes, and as long as the "new" side is text, we
2567          * can and should check what it introduces.
2568          */
2569         if (diff_filespec_is_binary(two))
2570                 goto free_and_return;
2571         else {
2572                 /* Crazy xdl interfaces.. */
2573                 xpparam_t xpp;
2574                 xdemitconf_t xecfg;
2575
2576                 memset(&xpp, 0, sizeof(xpp));
2577                 memset(&xecfg, 0, sizeof(xecfg));
2578                 xecfg.ctxlen = 1; /* at least one context line */
2579                 xpp.flags = 0;
2580                 if (xdi_diff_outf(&mf1, &mf2, checkdiff_consume, &data,
2581                                   &xpp, &xecfg))
2582                         die("unable to generate checkdiff for %s", one->path);
2583
2584                 if (data.ws_rule & WS_BLANK_AT_EOF) {
2585                         struct emit_callback ecbdata;
2586                         int blank_at_eof;
2587
2588                         ecbdata.ws_rule = data.ws_rule;
2589                         check_blank_at_eof(&mf1, &mf2, &ecbdata);
2590                         blank_at_eof = ecbdata.blank_at_eof_in_postimage;
2591
2592                         if (blank_at_eof) {
2593                                 static char *err;
2594                                 if (!err)
2595                                         err = whitespace_error_string(WS_BLANK_AT_EOF);
2596                                 fprintf(o->file, "%s:%d: %s.\n",
2597                                         data.filename, blank_at_eof, err);
2598                                 data.status = 1; /* report errors */
2599                         }
2600                 }
2601         }
2602  free_and_return:
2603         diff_free_filespec_data(one);
2604         diff_free_filespec_data(two);
2605         if (data.status)
2606                 DIFF_OPT_SET(o, CHECK_FAILED);
2607 }
2608
2609 struct diff_filespec *alloc_filespec(const char *path)
2610 {
2611         struct diff_filespec *spec;
2612
2613         FLEXPTR_ALLOC_STR(spec, path, path);
2614         spec->count = 1;
2615         spec->is_binary = -1;
2616         return spec;
2617 }
2618
2619 void free_filespec(struct diff_filespec *spec)
2620 {
2621         if (!--spec->count) {
2622                 diff_free_filespec_data(spec);
2623                 free(spec);
2624         }
2625 }
2626
2627 void fill_filespec(struct diff_filespec *spec, const unsigned char *sha1,
2628                    int sha1_valid, unsigned short mode)
2629 {
2630         if (mode) {
2631                 spec->mode = canon_mode(mode);
2632                 hashcpy(spec->oid.hash, sha1);
2633                 spec->oid_valid = sha1_valid;
2634         }
2635 }
2636
2637 /*
2638  * Given a name and sha1 pair, if the index tells us the file in
2639  * the work tree has that object contents, return true, so that
2640  * prepare_temp_file() does not have to inflate and extract.
2641  */
2642 static int reuse_worktree_file(const char *name, const unsigned char *sha1, int want_file)
2643 {
2644         const struct cache_entry *ce;
2645         struct stat st;
2646         int pos, len;
2647
2648         /*
2649          * We do not read the cache ourselves here, because the
2650          * benchmark with my previous version that always reads cache
2651          * shows that it makes things worse for diff-tree comparing
2652          * two linux-2.6 kernel trees in an already checked out work
2653          * tree.  This is because most diff-tree comparisons deal with
2654          * only a small number of files, while reading the cache is
2655          * expensive for a large project, and its cost outweighs the
2656          * savings we get by not inflating the object to a temporary
2657          * file.  Practically, this code only helps when we are used
2658          * by diff-cache --cached, which does read the cache before
2659          * calling us.
2660          */
2661         if (!active_cache)
2662                 return 0;
2663
2664         /* We want to avoid the working directory if our caller
2665          * doesn't need the data in a normal file, this system
2666          * is rather slow with its stat/open/mmap/close syscalls,
2667          * and the object is contained in a pack file.  The pack
2668          * is probably already open and will be faster to obtain
2669          * the data through than the working directory.  Loose
2670          * objects however would tend to be slower as they need
2671          * to be individually opened and inflated.
2672          */
2673         if (!FAST_WORKING_DIRECTORY && !want_file && has_sha1_pack(sha1))
2674                 return 0;
2675
2676         /*
2677          * Similarly, if we'd have to convert the file contents anyway, that
2678          * makes the optimization not worthwhile.
2679          */
2680         if (!want_file && would_convert_to_git(name))
2681                 return 0;
2682
2683         len = strlen(name);
2684         pos = cache_name_pos(name, len);
2685         if (pos < 0)
2686                 return 0;
2687         ce = active_cache[pos];
2688
2689         /*
2690          * This is not the sha1 we are looking for, or
2691          * unreusable because it is not a regular file.
2692          */
2693         if (hashcmp(sha1, ce->sha1) || !S_ISREG(ce->ce_mode))
2694                 return 0;
2695
2696         /*
2697          * If ce is marked as "assume unchanged", there is no
2698          * guarantee that work tree matches what we are looking for.
2699          */
2700         if ((ce->ce_flags & CE_VALID) || ce_skip_worktree(ce))
2701                 return 0;
2702
2703         /*
2704          * If ce matches the file in the work tree, we can reuse it.
2705          */
2706         if (ce_uptodate(ce) ||
2707             (!lstat(name, &st) && !ce_match_stat(ce, &st, 0)))
2708                 return 1;
2709
2710         return 0;
2711 }
2712
2713 static int diff_populate_gitlink(struct diff_filespec *s, int size_only)
2714 {
2715         struct strbuf buf = STRBUF_INIT;
2716         char *dirty = "";
2717
2718         /* Are we looking at the work tree? */
2719         if (s->dirty_submodule)
2720                 dirty = "-dirty";
2721
2722         strbuf_addf(&buf, "Subproject commit %s%s\n",
2723                     oid_to_hex(&s->oid), dirty);
2724         s->size = buf.len;
2725         if (size_only) {
2726                 s->data = NULL;
2727                 strbuf_release(&buf);
2728         } else {
2729                 s->data = strbuf_detach(&buf, NULL);
2730                 s->should_free = 1;
2731         }
2732         return 0;
2733 }
2734
2735 /*
2736  * While doing rename detection and pickaxe operation, we may need to
2737  * grab the data for the blob (or file) for our own in-core comparison.
2738  * diff_filespec has data and size fields for this purpose.
2739  */
2740 int diff_populate_filespec(struct diff_filespec *s, unsigned int flags)
2741 {
2742         int size_only = flags & CHECK_SIZE_ONLY;
2743         int err = 0;
2744         /*
2745          * demote FAIL to WARN to allow inspecting the situation
2746          * instead of refusing.
2747          */
2748         enum safe_crlf crlf_warn = (safe_crlf == SAFE_CRLF_FAIL
2749                                     ? SAFE_CRLF_WARN
2750                                     : safe_crlf);
2751
2752         if (!DIFF_FILE_VALID(s))
2753                 die("internal error: asking to populate invalid file.");
2754         if (S_ISDIR(s->mode))
2755                 return -1;
2756
2757         if (s->data)
2758                 return 0;
2759
2760         if (size_only && 0 < s->size)
2761                 return 0;
2762
2763         if (S_ISGITLINK(s->mode))
2764                 return diff_populate_gitlink(s, size_only);
2765
2766         if (!s->oid_valid ||
2767             reuse_worktree_file(s->path, s->oid.hash, 0)) {
2768                 struct strbuf buf = STRBUF_INIT;
2769                 struct stat st;
2770                 int fd;
2771
2772                 if (lstat(s->path, &st) < 0) {
2773                         if (errno == ENOENT) {
2774                         err_empty:
2775                                 err = -1;
2776                         empty:
2777                                 s->data = (char *)"";
2778                                 s->size = 0;
2779                                 return err;
2780                         }
2781                 }
2782                 s->size = xsize_t(st.st_size);
2783                 if (!s->size)
2784                         goto empty;
2785                 if (S_ISLNK(st.st_mode)) {
2786                         struct strbuf sb = STRBUF_INIT;
2787
2788                         if (strbuf_readlink(&sb, s->path, s->size))
2789                                 goto err_empty;
2790                         s->size = sb.len;
2791                         s->data = strbuf_detach(&sb, NULL);
2792                         s->should_free = 1;
2793                         return 0;
2794                 }
2795                 if (size_only)
2796                         return 0;
2797                 if ((flags & CHECK_BINARY) &&
2798                     s->size > big_file_threshold && s->is_binary == -1) {
2799                         s->is_binary = 1;
2800                         return 0;
2801                 }
2802                 fd = open(s->path, O_RDONLY);
2803                 if (fd < 0)
2804                         goto err_empty;
2805                 s->data = xmmap(NULL, s->size, PROT_READ, MAP_PRIVATE, fd, 0);
2806                 close(fd);
2807                 s->should_munmap = 1;
2808
2809                 /*
2810                  * Convert from working tree format to canonical git format
2811                  */
2812                 if (convert_to_git(s->path, s->data, s->size, &buf, crlf_warn)) {
2813                         size_t size = 0;
2814                         munmap(s->data, s->size);
2815                         s->should_munmap = 0;
2816                         s->data = strbuf_detach(&buf, &size);
2817                         s->size = size;
2818                         s->should_free = 1;
2819                 }
2820         }
2821         else {
2822                 enum object_type type;
2823                 if (size_only || (flags & CHECK_BINARY)) {
2824                         type = sha1_object_info(s->oid.hash, &s->size);
2825                         if (type < 0)
2826                                 die("unable to read %s",
2827                                     oid_to_hex(&s->oid));
2828                         if (size_only)
2829                                 return 0;
2830                         if (s->size > big_file_threshold && s->is_binary == -1) {
2831                                 s->is_binary = 1;
2832                                 return 0;
2833                         }
2834                 }
2835                 s->data = read_sha1_file(s->oid.hash, &type, &s->size);
2836                 if (!s->data)
2837                         die("unable to read %s", oid_to_hex(&s->oid));
2838                 s->should_free = 1;
2839         }
2840         return 0;
2841 }
2842
2843 void diff_free_filespec_blob(struct diff_filespec *s)
2844 {
2845         if (s->should_free)
2846                 free(s->data);
2847         else if (s->should_munmap)
2848                 munmap(s->data, s->size);
2849
2850         if (s->should_free || s->should_munmap) {
2851                 s->should_free = s->should_munmap = 0;
2852                 s->data = NULL;
2853         }
2854 }
2855
2856 void diff_free_filespec_data(struct diff_filespec *s)
2857 {
2858         diff_free_filespec_blob(s);
2859         free(s->cnt_data);
2860         s->cnt_data = NULL;
2861 }
2862
2863 static void prep_temp_blob(const char *path, struct diff_tempfile *temp,
2864                            void *blob,
2865                            unsigned long size,
2866                            const struct object_id *oid,
2867                            int mode)
2868 {
2869         int fd;
2870         struct strbuf buf = STRBUF_INIT;
2871         struct strbuf template = STRBUF_INIT;
2872         char *path_dup = xstrdup(path);
2873         const char *base = basename(path_dup);
2874
2875         /* Generate "XXXXXX_basename.ext" */
2876         strbuf_addstr(&template, "XXXXXX_");
2877         strbuf_addstr(&template, base);
2878
2879         fd = mks_tempfile_ts(&temp->tempfile, template.buf, strlen(base) + 1);
2880         if (fd < 0)
2881                 die_errno("unable to create temp-file");
2882         if (convert_to_working_tree(path,
2883                         (const char *)blob, (size_t)size, &buf)) {
2884                 blob = buf.buf;
2885                 size = buf.len;
2886         }
2887         if (write_in_full(fd, blob, size) != size)
2888                 die_errno("unable to write temp-file");
2889         close_tempfile(&temp->tempfile);
2890         temp->name = get_tempfile_path(&temp->tempfile);
2891         oid_to_hex_r(temp->hex, oid);
2892         xsnprintf(temp->mode, sizeof(temp->mode), "%06o", mode);
2893         strbuf_release(&buf);
2894         strbuf_release(&template);
2895         free(path_dup);
2896 }
2897
2898 static struct diff_tempfile *prepare_temp_file(const char *name,
2899                 struct diff_filespec *one)
2900 {
2901         struct diff_tempfile *temp = claim_diff_tempfile();
2902
2903         if (!DIFF_FILE_VALID(one)) {
2904         not_a_valid_file:
2905                 /* A '-' entry produces this for file-2, and
2906                  * a '+' entry produces this for file-1.
2907                  */
2908                 temp->name = "/dev/null";
2909                 xsnprintf(temp->hex, sizeof(temp->hex), ".");
2910                 xsnprintf(temp->mode, sizeof(temp->mode), ".");
2911                 return temp;
2912         }
2913
2914         if (!S_ISGITLINK(one->mode) &&
2915             (!one->oid_valid ||
2916              reuse_worktree_file(name, one->oid.hash, 1))) {
2917                 struct stat st;
2918                 if (lstat(name, &st) < 0) {
2919                         if (errno == ENOENT)
2920                                 goto not_a_valid_file;
2921                         die_errno("stat(%s)", name);
2922                 }
2923                 if (S_ISLNK(st.st_mode)) {
2924                         struct strbuf sb = STRBUF_INIT;
2925                         if (strbuf_readlink(&sb, name, st.st_size) < 0)
2926                                 die_errno("readlink(%s)", name);
2927                         prep_temp_blob(name, temp, sb.buf, sb.len,
2928                                        (one->oid_valid ?
2929                                         &one->oid : &null_oid),
2930                                        (one->oid_valid ?
2931                                         one->mode : S_IFLNK));
2932                         strbuf_release(&sb);
2933                 }
2934                 else {
2935                         /* we can borrow from the file in the work tree */
2936                         temp->name = name;
2937                         if (!one->oid_valid)
2938                                 sha1_to_hex_r(temp->hex, null_sha1);
2939                         else
2940                                 sha1_to_hex_r(temp->hex, one->oid.hash);
2941                         /* Even though we may sometimes borrow the
2942                          * contents from the work tree, we always want
2943                          * one->mode.  mode is trustworthy even when
2944                          * !(one->sha1_valid), as long as
2945                          * DIFF_FILE_VALID(one).
2946                          */
2947                         xsnprintf(temp->mode, sizeof(temp->mode), "%06o", one->mode);
2948                 }
2949                 return temp;
2950         }
2951         else {
2952                 if (diff_populate_filespec(one, 0))
2953                         die("cannot read data blob for %s", one->path);
2954                 prep_temp_blob(name, temp, one->data, one->size,
2955                                &one->oid, one->mode);
2956         }
2957         return temp;
2958 }
2959
2960 static void add_external_diff_name(struct argv_array *argv,
2961                                    const char *name,
2962                                    struct diff_filespec *df)
2963 {
2964         struct diff_tempfile *temp = prepare_temp_file(name, df);
2965         argv_array_push(argv, temp->name);
2966         argv_array_push(argv, temp->hex);
2967         argv_array_push(argv, temp->mode);
2968 }
2969
2970 /* An external diff command takes:
2971  *
2972  * diff-cmd name infile1 infile1-sha1 infile1-mode \
2973  *               infile2 infile2-sha1 infile2-mode [ rename-to ]
2974  *
2975  */
2976 static void run_external_diff(const char *pgm,
2977                               const char *name,
2978                               const char *other,
2979                               struct diff_filespec *one,
2980                               struct diff_filespec *two,
2981                               const char *xfrm_msg,
2982                               int complete_rewrite,
2983                               struct diff_options *o)
2984 {
2985         struct argv_array argv = ARGV_ARRAY_INIT;
2986         struct argv_array env = ARGV_ARRAY_INIT;
2987         struct diff_queue_struct *q = &diff_queued_diff;
2988
2989         argv_array_push(&argv, pgm);
2990         argv_array_push(&argv, name);
2991
2992         if (one && two) {
2993                 add_external_diff_name(&argv, name, one);
2994                 if (!other)
2995                         add_external_diff_name(&argv, name, two);
2996                 else {
2997                         add_external_diff_name(&argv, other, two);
2998                         argv_array_push(&argv, other);
2999                         argv_array_push(&argv, xfrm_msg);
3000                 }
3001         }
3002
3003         argv_array_pushf(&env, "GIT_DIFF_PATH_COUNTER=%d", ++o->diff_path_counter);
3004         argv_array_pushf(&env, "GIT_DIFF_PATH_TOTAL=%d", q->nr);
3005
3006         if (run_command_v_opt_cd_env(argv.argv, RUN_USING_SHELL, NULL, env.argv))
3007                 die(_("external diff died, stopping at %s"), name);
3008
3009         remove_tempfile();
3010         argv_array_clear(&argv);
3011         argv_array_clear(&env);
3012 }
3013
3014 static int similarity_index(struct diff_filepair *p)
3015 {
3016         return p->score * 100 / MAX_SCORE;
3017 }
3018
3019 static void fill_metainfo(struct strbuf *msg,
3020                           const char *name,
3021                           const char *other,
3022                           struct diff_filespec *one,
3023                           struct diff_filespec *two,
3024                           struct diff_options *o,
3025                           struct diff_filepair *p,
3026                           int *must_show_header,
3027                           int use_color)
3028 {
3029         const char *set = diff_get_color(use_color, DIFF_METAINFO);
3030         const char *reset = diff_get_color(use_color, DIFF_RESET);
3031         const char *line_prefix = diff_line_prefix(o);
3032
3033         *must_show_header = 1;
3034         strbuf_init(msg, PATH_MAX * 2 + 300);
3035         switch (p->status) {
3036         case DIFF_STATUS_COPIED:
3037                 strbuf_addf(msg, "%s%ssimilarity index %d%%",
3038                             line_prefix, set, similarity_index(p));
3039                 strbuf_addf(msg, "%s\n%s%scopy from ",
3040                             reset,  line_prefix, set);
3041                 quote_c_style(name, msg, NULL, 0);
3042                 strbuf_addf(msg, "%s\n%s%scopy to ", reset, line_prefix, set);
3043                 quote_c_style(other, msg, NULL, 0);
3044                 strbuf_addf(msg, "%s\n", reset);
3045                 break;
3046         case DIFF_STATUS_RENAMED:
3047                 strbuf_addf(msg, "%s%ssimilarity index %d%%",
3048                             line_prefix, set, similarity_index(p));
3049                 strbuf_addf(msg, "%s\n%s%srename from ",
3050                             reset, line_prefix, set);
3051                 quote_c_style(name, msg, NULL, 0);
3052                 strbuf_addf(msg, "%s\n%s%srename to ",
3053                             reset, line_prefix, set);
3054                 quote_c_style(other, msg, NULL, 0);
3055                 strbuf_addf(msg, "%s\n", reset);
3056                 break;
3057         case DIFF_STATUS_MODIFIED:
3058                 if (p->score) {
3059                         strbuf_addf(msg, "%s%sdissimilarity index %d%%%s\n",
3060                                     line_prefix,
3061                                     set, similarity_index(p), reset);
3062                         break;
3063                 }
3064                 /* fallthru */
3065         default:
3066                 *must_show_header = 0;
3067         }
3068         if (one && two && oidcmp(&one->oid, &two->oid)) {
3069                 int abbrev = DIFF_OPT_TST(o, FULL_INDEX) ? 40 : DEFAULT_ABBREV;
3070
3071                 if (DIFF_OPT_TST(o, BINARY)) {
3072                         mmfile_t mf;
3073                         if ((!fill_mmfile(&mf, one) && diff_filespec_is_binary(one)) ||
3074                             (!fill_mmfile(&mf, two) && diff_filespec_is_binary(two)))
3075                                 abbrev = 40;
3076                 }
3077                 strbuf_addf(msg, "%s%sindex %s..", line_prefix, set,
3078                             find_unique_abbrev(one->oid.hash, abbrev));
3079                 strbuf_addstr(msg, find_unique_abbrev(two->oid.hash, abbrev));
3080                 if (one->mode == two->mode)
3081                         strbuf_addf(msg, " %06o", one->mode);
3082                 strbuf_addf(msg, "%s\n", reset);
3083         }
3084 }
3085
3086 static void run_diff_cmd(const char *pgm,
3087                          const char *name,
3088                          const char *other,
3089                          const char *attr_path,
3090                          struct diff_filespec *one,
3091                          struct diff_filespec *two,
3092                          struct strbuf *msg,
3093                          struct diff_options *o,
3094                          struct diff_filepair *p)
3095 {
3096         const char *xfrm_msg = NULL;
3097         int complete_rewrite = (p->status == DIFF_STATUS_MODIFIED) && p->score;
3098         int must_show_header = 0;
3099
3100
3101         if (DIFF_OPT_TST(o, ALLOW_EXTERNAL)) {
3102                 struct userdiff_driver *drv = userdiff_find_by_path(attr_path);
3103                 if (drv && drv->external)
3104                         pgm = drv->external;
3105         }
3106
3107         if (msg) {
3108                 /*
3109                  * don't use colors when the header is intended for an
3110                  * external diff driver
3111                  */
3112                 fill_metainfo(msg, name, other, one, two, o, p,
3113                               &must_show_header,
3114                               want_color(o->use_color) && !pgm);
3115                 xfrm_msg = msg->len ? msg->buf : NULL;
3116         }
3117
3118         if (pgm) {
3119                 run_external_diff(pgm, name, other, one, two, xfrm_msg,
3120                                   complete_rewrite, o);
3121                 return;
3122         }
3123         if (one && two)
3124                 builtin_diff(name, other ? other : name,
3125                              one, two, xfrm_msg, must_show_header,
3126                              o, complete_rewrite);
3127         else
3128                 fprintf(o->file, "* Unmerged path %s\n", name);
3129 }
3130
3131 static void diff_fill_sha1_info(struct diff_filespec *one)
3132 {
3133         if (DIFF_FILE_VALID(one)) {
3134                 if (!one->oid_valid) {
3135                         struct stat st;
3136                         if (one->is_stdin) {
3137                                 oidclr(&one->oid);
3138                                 return;
3139                         }
3140                         if (lstat(one->path, &st) < 0)
3141                                 die_errno("stat '%s'", one->path);
3142                         if (index_path(one->oid.hash, one->path, &st, 0))
3143                                 die("cannot hash %s", one->path);
3144                 }
3145         }
3146         else
3147                 oidclr(&one->oid);
3148 }
3149
3150 static void strip_prefix(int prefix_length, const char **namep, const char **otherp)
3151 {
3152         /* Strip the prefix but do not molest /dev/null and absolute paths */
3153         if (*namep && **namep != '/') {
3154                 *namep += prefix_length;
3155                 if (**namep == '/')
3156                         ++*namep;
3157         }
3158         if (*otherp && **otherp != '/') {
3159                 *otherp += prefix_length;
3160                 if (**otherp == '/')
3161                         ++*otherp;
3162         }
3163 }
3164
3165 static void run_diff(struct diff_filepair *p, struct diff_options *o)
3166 {
3167         const char *pgm = external_diff();
3168         struct strbuf msg;
3169         struct diff_filespec *one = p->one;
3170         struct diff_filespec *two = p->two;
3171         const char *name;
3172         const char *other;
3173         const char *attr_path;
3174
3175         name  = p->one->path;
3176         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
3177         attr_path = name;
3178         if (o->prefix_length)
3179                 strip_prefix(o->prefix_length, &name, &other);
3180
3181         if (!DIFF_OPT_TST(o, ALLOW_EXTERNAL))
3182                 pgm = NULL;
3183
3184         if (DIFF_PAIR_UNMERGED(p)) {
3185                 run_diff_cmd(pgm, name, NULL, attr_path,
3186                              NULL, NULL, NULL, o, p);
3187                 return;
3188         }
3189
3190         diff_fill_sha1_info(one);
3191         diff_fill_sha1_info(two);
3192
3193         if (!pgm &&
3194             DIFF_FILE_VALID(one) && DIFF_FILE_VALID(two) &&
3195             (S_IFMT & one->mode) != (S_IFMT & two->mode)) {
3196                 /*
3197                  * a filepair that changes between file and symlink
3198                  * needs to be split into deletion and creation.
3199                  */
3200                 struct diff_filespec *null = alloc_filespec(two->path);
3201                 run_diff_cmd(NULL, name, other, attr_path,
3202                              one, null, &msg, o, p);
3203                 free(null);
3204                 strbuf_release(&msg);
3205
3206                 null = alloc_filespec(one->path);
3207                 run_diff_cmd(NULL, name, other, attr_path,
3208                              null, two, &msg, o, p);
3209                 free(null);
3210         }
3211         else
3212                 run_diff_cmd(pgm, name, other, attr_path,
3213                              one, two, &msg, o, p);
3214
3215         strbuf_release(&msg);
3216 }
3217
3218 static void run_diffstat(struct diff_filepair *p, struct diff_options *o,
3219                          struct diffstat_t *diffstat)
3220 {
3221         const char *name;
3222         const char *other;
3223
3224         if (DIFF_PAIR_UNMERGED(p)) {
3225                 /* unmerged */
3226                 builtin_diffstat(p->one->path, NULL, NULL, NULL, diffstat, o, p);
3227                 return;
3228         }
3229
3230         name = p->one->path;
3231         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
3232
3233         if (o->prefix_length)
3234                 strip_prefix(o->prefix_length, &name, &other);
3235
3236         diff_fill_sha1_info(p->one);
3237         diff_fill_sha1_info(p->two);
3238
3239         builtin_diffstat(name, other, p->one, p->two, diffstat, o, p);
3240 }
3241
3242 static void run_checkdiff(struct diff_filepair *p, struct diff_options *o)
3243 {
3244         const char *name;
3245         const char *other;
3246         const char *attr_path;
3247
3248         if (DIFF_PAIR_UNMERGED(p)) {
3249                 /* unmerged */
3250                 return;
3251         }
3252
3253         name = p->one->path;
3254         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
3255         attr_path = other ? other : name;
3256
3257         if (o->prefix_length)
3258                 strip_prefix(o->prefix_length, &name, &other);
3259
3260         diff_fill_sha1_info(p->one);
3261         diff_fill_sha1_info(p->two);
3262
3263         builtin_checkdiff(name, other, attr_path, p->one, p->two, o);
3264 }
3265
3266 void diff_setup(struct diff_options *options)
3267 {
3268         memcpy(options, &default_diff_options, sizeof(*options));
3269
3270         options->file = stdout;
3271
3272         options->line_termination = '\n';
3273         options->break_opt = -1;
3274         options->rename_limit = -1;
3275         options->dirstat_permille = diff_dirstat_permille_default;
3276         options->context = diff_context_default;
3277         options->ws_error_highlight = WSEH_NEW;
3278         DIFF_OPT_SET(options, RENAME_EMPTY);
3279
3280         /* pathchange left =NULL by default */
3281         options->change = diff_change;
3282         options->add_remove = diff_addremove;
3283         options->use_color = diff_use_color_default;
3284         options->detect_rename = diff_detect_rename_default;
3285         options->xdl_opts |= diff_algorithm;
3286         if (diff_compaction_heuristic)
3287                 DIFF_XDL_SET(options, COMPACTION_HEURISTIC);
3288
3289         options->orderfile = diff_order_file_cfg;
3290
3291         if (diff_no_prefix) {
3292                 options->a_prefix = options->b_prefix = "";
3293         } else if (!diff_mnemonic_prefix) {
3294                 options->a_prefix = "a/";
3295                 options->b_prefix = "b/";
3296         }
3297 }
3298
3299 void diff_setup_done(struct diff_options *options)
3300 {
3301         int count = 0;
3302
3303         if (options->set_default)
3304                 options->set_default(options);
3305
3306         if (options->output_format & DIFF_FORMAT_NAME)
3307                 count++;
3308         if (options->output_format & DIFF_FORMAT_NAME_STATUS)
3309                 count++;
3310         if (options->output_format & DIFF_FORMAT_CHECKDIFF)
3311                 count++;
3312         if (options->output_format & DIFF_FORMAT_NO_OUTPUT)
3313                 count++;
3314         if (count > 1)
3315                 die("--name-only, --name-status, --check and -s are mutually exclusive");
3316
3317         /*
3318          * Most of the time we can say "there are changes"
3319          * only by checking if there are changed paths, but
3320          * --ignore-whitespace* options force us to look
3321          * inside contents.
3322          */
3323
3324         if (DIFF_XDL_TST(options, IGNORE_WHITESPACE) ||
3325             DIFF_XDL_TST(options, IGNORE_WHITESPACE_CHANGE) ||
3326             DIFF_XDL_TST(options, IGNORE_WHITESPACE_AT_EOL))
3327                 DIFF_OPT_SET(options, DIFF_FROM_CONTENTS);
3328         else
3329                 DIFF_OPT_CLR(options, DIFF_FROM_CONTENTS);
3330
3331         if (DIFF_OPT_TST(options, FIND_COPIES_HARDER))
3332                 options->detect_rename = DIFF_DETECT_COPY;
3333
3334         if (!DIFF_OPT_TST(options, RELATIVE_NAME))
3335                 options->prefix = NULL;
3336         if (options->prefix)
3337                 options->prefix_length = strlen(options->prefix);
3338         else
3339                 options->prefix_length = 0;
3340
3341         if (options->output_format & (DIFF_FORMAT_NAME |
3342                                       DIFF_FORMAT_NAME_STATUS |
3343                                       DIFF_FORMAT_CHECKDIFF |
3344                                       DIFF_FORMAT_NO_OUTPUT))
3345                 options->output_format &= ~(DIFF_FORMAT_RAW |
3346                                             DIFF_FORMAT_NUMSTAT |
3347                                             DIFF_FORMAT_DIFFSTAT |
3348                                             DIFF_FORMAT_SHORTSTAT |
3349                                             DIFF_FORMAT_DIRSTAT |
3350                                             DIFF_FORMAT_SUMMARY |
3351                                             DIFF_FORMAT_PATCH);
3352
3353         /*
3354          * These cases always need recursive; we do not drop caller-supplied
3355          * recursive bits for other formats here.
3356          */
3357         if (options->output_format & (DIFF_FORMAT_PATCH |
3358                                       DIFF_FORMAT_NUMSTAT |
3359                                       DIFF_FORMAT_DIFFSTAT |
3360                                       DIFF_FORMAT_SHORTSTAT |
3361                                       DIFF_FORMAT_DIRSTAT |
3362                                       DIFF_FORMAT_SUMMARY |
3363                                       DIFF_FORMAT_CHECKDIFF))
3364                 DIFF_OPT_SET(options, RECURSIVE);
3365         /*
3366          * Also pickaxe would not work very well if you do not say recursive
3367          */
3368         if (options->pickaxe)
3369                 DIFF_OPT_SET(options, RECURSIVE);
3370         /*
3371          * When patches are generated, submodules diffed against the work tree
3372          * must be checked for dirtiness too so it can be shown in the output
3373          */
3374         if (options->output_format & DIFF_FORMAT_PATCH)
3375                 DIFF_OPT_SET(options, DIRTY_SUBMODULES);
3376
3377         if (options->detect_rename && options->rename_limit < 0)
3378                 options->rename_limit = diff_rename_limit_default;
3379         if (options->setup & DIFF_SETUP_USE_CACHE) {
3380                 if (!active_cache)
3381                         /* read-cache does not die even when it fails
3382                          * so it is safe for us to do this here.  Also
3383                          * it does not smudge active_cache or active_nr
3384                          * when it fails, so we do not have to worry about
3385                          * cleaning it up ourselves either.
3386                          */
3387                         read_cache();
3388         }
3389         if (options->abbrev <= 0 || 40 < options->abbrev)
3390                 options->abbrev = 40; /* full */
3391
3392         /*
3393          * It does not make sense to show the first hit we happened
3394          * to have found.  It does not make sense not to return with
3395          * exit code in such a case either.
3396          */
3397         if (DIFF_OPT_TST(options, QUICK)) {
3398                 options->output_format = DIFF_FORMAT_NO_OUTPUT;
3399                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
3400         }
3401
3402         options->diff_path_counter = 0;
3403
3404         if (DIFF_OPT_TST(options, FOLLOW_RENAMES) && options->pathspec.nr != 1)
3405                 die(_("--follow requires exactly one pathspec"));
3406 }
3407
3408 static int opt_arg(const char *arg, int arg_short, const char *arg_long, int *val)
3409 {
3410         char c, *eq;
3411         int len;
3412
3413         if (*arg != '-')
3414                 return 0;
3415         c = *++arg;
3416         if (!c)
3417                 return 0;
3418         if (c == arg_short) {
3419                 c = *++arg;
3420                 if (!c)
3421                         return 1;
3422                 if (val && isdigit(c)) {
3423                         char *end;
3424                         int n = strtoul(arg, &end, 10);
3425                         if (*end)
3426                                 return 0;
3427                         *val = n;
3428                         return 1;
3429                 }
3430                 return 0;
3431         }
3432         if (c != '-')
3433                 return 0;
3434         arg++;
3435         eq = strchrnul(arg, '=');
3436         len = eq - arg;
3437         if (!len || strncmp(arg, arg_long, len))
3438                 return 0;
3439         if (*eq) {
3440                 int n;
3441                 char *end;
3442                 if (!isdigit(*++eq))
3443                         return 0;
3444                 n = strtoul(eq, &end, 10);
3445                 if (*end)
3446                         return 0;
3447                 *val = n;
3448         }
3449         return 1;
3450 }
3451
3452 static int diff_scoreopt_parse(const char *opt);
3453
3454 static inline int short_opt(char opt, const char **argv,
3455                             const char **optarg)
3456 {
3457         const char *arg = argv[0];
3458         if (arg[0] != '-' || arg[1] != opt)
3459                 return 0;
3460         if (arg[2] != '\0') {
3461                 *optarg = arg + 2;
3462                 return 1;
3463         }
3464         if (!argv[1])
3465                 die("Option '%c' requires a value", opt);
3466         *optarg = argv[1];
3467         return 2;
3468 }
3469
3470 int parse_long_opt(const char *opt, const char **argv,
3471                    const char **optarg)
3472 {
3473         const char *arg = argv[0];
3474         if (!skip_prefix(arg, "--", &arg))
3475                 return 0;
3476         if (!skip_prefix(arg, opt, &arg))
3477                 return 0;
3478         if (*arg == '=') { /* stuck form: --option=value */
3479                 *optarg = arg + 1;
3480                 return 1;
3481         }
3482         if (*arg != '\0')
3483                 return 0;
3484         /* separate form: --option value */
3485         if (!argv[1])
3486                 die("Option '--%s' requires a value", opt);
3487         *optarg = argv[1];
3488         return 2;
3489 }
3490
3491 static int stat_opt(struct diff_options *options, const char **av)
3492 {
3493         const char *arg = av[0];
3494         char *end;
3495         int width = options->stat_width;
3496         int name_width = options->stat_name_width;
3497         int graph_width = options->stat_graph_width;
3498         int count = options->stat_count;
3499         int argcount = 1;
3500
3501         if (!skip_prefix(arg, "--stat", &arg))
3502                 die("BUG: stat option does not begin with --stat: %s", arg);
3503         end = (char *)arg;
3504
3505         switch (*arg) {
3506         case '-':
3507                 if (skip_prefix(arg, "-width", &arg)) {
3508                         if (*arg == '=')
3509                                 width = strtoul(arg + 1, &end, 10);
3510                         else if (!*arg && !av[1])
3511                                 die("Option '--stat-width' requires a value");
3512                         else if (!*arg) {
3513                                 width = strtoul(av[1], &end, 10);
3514                                 argcount = 2;
3515                         }
3516                 } else if (skip_prefix(arg, "-name-width", &arg)) {
3517                         if (*arg == '=')
3518                                 name_width = strtoul(arg + 1, &end, 10);
3519                         else if (!*arg && !av[1])
3520                                 die("Option '--stat-name-width' requires a value");
3521                         else if (!*arg) {
3522                                 name_width = strtoul(av[1], &end, 10);
3523                                 argcount = 2;
3524                         }
3525                 } else if (skip_prefix(arg, "-graph-width", &arg)) {
3526                         if (*arg == '=')
3527                                 graph_width = strtoul(arg + 1, &end, 10);
3528                         else if (!*arg && !av[1])
3529                                 die("Option '--stat-graph-width' requires a value");
3530                         else if (!*arg) {
3531                                 graph_width = strtoul(av[1], &end, 10);
3532                                 argcount = 2;
3533                         }
3534                 } else if (skip_prefix(arg, "-count", &arg)) {
3535                         if (*arg == '=')
3536                                 count = strtoul(arg + 1, &end, 10);
3537                         else if (!*arg && !av[1])
3538                                 die("Option '--stat-count' requires a value");
3539                         else if (!*arg) {
3540                                 count = strtoul(av[1], &end, 10);
3541                                 argcount = 2;
3542                         }
3543                 }
3544                 break;
3545         case '=':
3546                 width = strtoul(arg+1, &end, 10);
3547                 if (*end == ',')
3548                         name_width = strtoul(end+1, &end, 10);
3549                 if (*end == ',')
3550                         count = strtoul(end+1, &end, 10);
3551         }
3552
3553         /* Important! This checks all the error cases! */
3554         if (*end)
3555                 return 0;
3556         options->output_format |= DIFF_FORMAT_DIFFSTAT;
3557         options->stat_name_width = name_width;
3558         options->stat_graph_width = graph_width;
3559         options->stat_width = width;
3560         options->stat_count = count;
3561         return argcount;
3562 }
3563
3564 static int parse_dirstat_opt(struct diff_options *options, const char *params)
3565 {
3566         struct strbuf errmsg = STRBUF_INIT;
3567         if (parse_dirstat_params(options, params, &errmsg))
3568                 die(_("Failed to parse --dirstat/-X option parameter:\n%s"),
3569                     errmsg.buf);
3570         strbuf_release(&errmsg);
3571         /*
3572          * The caller knows a dirstat-related option is given from the command
3573          * line; allow it to say "return this_function();"
3574          */
3575         options->output_format |= DIFF_FORMAT_DIRSTAT;
3576         return 1;
3577 }
3578
3579 static int parse_submodule_opt(struct diff_options *options, const char *value)
3580 {
3581         if (parse_submodule_params(options, value))
3582                 die(_("Failed to parse --submodule option parameter: '%s'"),
3583                         value);
3584         return 1;
3585 }
3586
3587 static const char diff_status_letters[] = {
3588         DIFF_STATUS_ADDED,
3589         DIFF_STATUS_COPIED,
3590         DIFF_STATUS_DELETED,
3591         DIFF_STATUS_MODIFIED,
3592         DIFF_STATUS_RENAMED,
3593         DIFF_STATUS_TYPE_CHANGED,
3594         DIFF_STATUS_UNKNOWN,
3595         DIFF_STATUS_UNMERGED,
3596         DIFF_STATUS_FILTER_AON,
3597         DIFF_STATUS_FILTER_BROKEN,
3598         '\0',
3599 };
3600
3601 static unsigned int filter_bit['Z' + 1];
3602
3603 static void prepare_filter_bits(void)
3604 {
3605         int i;
3606
3607         if (!filter_bit[DIFF_STATUS_ADDED]) {
3608                 for (i = 0; diff_status_letters[i]; i++)
3609                         filter_bit[(int) diff_status_letters[i]] = (1 << i);
3610         }
3611 }
3612
3613 static unsigned filter_bit_tst(char status, const struct diff_options *opt)
3614 {
3615         return opt->filter & filter_bit[(int) status];
3616 }
3617
3618 static int parse_diff_filter_opt(const char *optarg, struct diff_options *opt)
3619 {
3620         int i, optch;
3621
3622         prepare_filter_bits();
3623
3624         /*
3625          * If there is a negation e.g. 'd' in the input, and we haven't
3626          * initialized the filter field with another --diff-filter, start
3627          * from full set of bits, except for AON.
3628          */
3629         if (!opt->filter) {
3630                 for (i = 0; (optch = optarg[i]) != '\0'; i++) {
3631                         if (optch < 'a' || 'z' < optch)
3632                                 continue;
3633                         opt->filter = (1 << (ARRAY_SIZE(diff_status_letters) - 1)) - 1;
3634                         opt->filter &= ~filter_bit[DIFF_STATUS_FILTER_AON];
3635                         break;
3636                 }
3637         }
3638
3639         for (i = 0; (optch = optarg[i]) != '\0'; i++) {
3640                 unsigned int bit;
3641                 int negate;
3642
3643                 if ('a' <= optch && optch <= 'z') {
3644                         negate = 1;
3645                         optch = toupper(optch);
3646                 } else {
3647                         negate = 0;
3648                 }
3649
3650                 bit = (0 <= optch && optch <= 'Z') ? filter_bit[optch] : 0;
3651                 if (!bit)
3652                         return optarg[i];
3653                 if (negate)
3654                         opt->filter &= ~bit;
3655                 else
3656                         opt->filter |= bit;
3657         }
3658         return 0;
3659 }
3660
3661 static void enable_patch_output(int *fmt) {
3662         *fmt &= ~DIFF_FORMAT_NO_OUTPUT;
3663         *fmt |= DIFF_FORMAT_PATCH;
3664 }
3665
3666 static int parse_one_token(const char **arg, const char *token)
3667 {
3668         const char *rest;
3669         if (skip_prefix(*arg, token, &rest) && (!*rest || *rest == ',')) {
3670                 *arg = rest;
3671                 return 1;
3672         }
3673         return 0;
3674 }
3675
3676 static int parse_ws_error_highlight(struct diff_options *opt, const char *arg)
3677 {
3678         const char *orig_arg = arg;
3679         unsigned val = 0;
3680         while (*arg) {
3681                 if (parse_one_token(&arg, "none"))
3682                         val = 0;
3683                 else if (parse_one_token(&arg, "default"))
3684                         val = WSEH_NEW;
3685                 else if (parse_one_token(&arg, "all"))
3686                         val = WSEH_NEW | WSEH_OLD | WSEH_CONTEXT;
3687                 else if (parse_one_token(&arg, "new"))
3688                         val |= WSEH_NEW;
3689                 else if (parse_one_token(&arg, "old"))
3690                         val |= WSEH_OLD;
3691                 else if (parse_one_token(&arg, "context"))
3692                         val |= WSEH_CONTEXT;
3693                 else {
3694                         error("unknown value after ws-error-highlight=%.*s",
3695                               (int)(arg - orig_arg), orig_arg);
3696                         return 0;
3697                 }
3698                 if (*arg)
3699                         arg++;
3700         }
3701         opt->ws_error_highlight = val;
3702         return 1;
3703 }
3704
3705 int diff_opt_parse(struct diff_options *options,
3706                    const char **av, int ac, const char *prefix)
3707 {
3708         const char *arg = av[0];
3709         const char *optarg;
3710         int argcount;
3711
3712         if (!prefix)
3713                 prefix = "";
3714
3715         /* Output format options */
3716         if (!strcmp(arg, "-p") || !strcmp(arg, "-u") || !strcmp(arg, "--patch")
3717             || opt_arg(arg, 'U', "unified", &options->context))
3718                 enable_patch_output(&options->output_format);
3719         else if (!strcmp(arg, "--raw"))
3720                 options->output_format |= DIFF_FORMAT_RAW;
3721         else if (!strcmp(arg, "--patch-with-raw")) {
3722                 enable_patch_output(&options->output_format);
3723                 options->output_format |= DIFF_FORMAT_RAW;
3724         } else if (!strcmp(arg, "--numstat"))
3725                 options->output_format |= DIFF_FORMAT_NUMSTAT;
3726         else if (!strcmp(arg, "--shortstat"))
3727                 options->output_format |= DIFF_FORMAT_SHORTSTAT;
3728         else if (!strcmp(arg, "-X") || !strcmp(arg, "--dirstat"))
3729                 return parse_dirstat_opt(options, "");
3730         else if (skip_prefix(arg, "-X", &arg))
3731                 return parse_dirstat_opt(options, arg);
3732         else if (skip_prefix(arg, "--dirstat=", &arg))
3733                 return parse_dirstat_opt(options, arg);
3734         else if (!strcmp(arg, "--cumulative"))
3735                 return parse_dirstat_opt(options, "cumulative");
3736         else if (!strcmp(arg, "--dirstat-by-file"))
3737                 return parse_dirstat_opt(options, "files");
3738         else if (skip_prefix(arg, "--dirstat-by-file=", &arg)) {
3739                 parse_dirstat_opt(options, "files");
3740                 return parse_dirstat_opt(options, arg);
3741         }
3742         else if (!strcmp(arg, "--check"))
3743                 options->output_format |= DIFF_FORMAT_CHECKDIFF;
3744         else if (!strcmp(arg, "--summary"))
3745                 options->output_format |= DIFF_FORMAT_SUMMARY;
3746         else if (!strcmp(arg, "--patch-with-stat")) {
3747                 enable_patch_output(&options->output_format);
3748                 options->output_format |= DIFF_FORMAT_DIFFSTAT;
3749         } else if (!strcmp(arg, "--name-only"))
3750                 options->output_format |= DIFF_FORMAT_NAME;
3751         else if (!strcmp(arg, "--name-status"))
3752                 options->output_format |= DIFF_FORMAT_NAME_STATUS;
3753         else if (!strcmp(arg, "-s") || !strcmp(arg, "--no-patch"))
3754                 options->output_format |= DIFF_FORMAT_NO_OUTPUT;
3755         else if (starts_with(arg, "--stat"))
3756                 /* --stat, --stat-width, --stat-name-width, or --stat-count */
3757                 return stat_opt(options, av);
3758
3759         /* renames options */
3760         else if (starts_with(arg, "-B") || starts_with(arg, "--break-rewrites=") ||
3761                  !strcmp(arg, "--break-rewrites")) {
3762                 if ((options->break_opt = diff_scoreopt_parse(arg)) == -1)
3763                         return error("invalid argument to -B: %s", arg+2);
3764         }
3765         else if (starts_with(arg, "-M") || starts_with(arg, "--find-renames=") ||
3766                  !strcmp(arg, "--find-renames")) {
3767                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
3768                         return error("invalid argument to -M: %s", arg+2);
3769                 options->detect_rename = DIFF_DETECT_RENAME;
3770         }
3771         else if (!strcmp(arg, "-D") || !strcmp(arg, "--irreversible-delete")) {
3772                 options->irreversible_delete = 1;
3773         }
3774         else if (starts_with(arg, "-C") || starts_with(arg, "--find-copies=") ||
3775                  !strcmp(arg, "--find-copies")) {
3776                 if (options->detect_rename == DIFF_DETECT_COPY)
3777                         DIFF_OPT_SET(options, FIND_COPIES_HARDER);
3778                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
3779                         return error("invalid argument to -C: %s", arg+2);
3780                 options->detect_rename = DIFF_DETECT_COPY;
3781         }
3782         else if (!strcmp(arg, "--no-renames"))
3783                 options->detect_rename = 0;
3784         else if (!strcmp(arg, "--rename-empty"))
3785                 DIFF_OPT_SET(options, RENAME_EMPTY);
3786         else if (!strcmp(arg, "--no-rename-empty"))
3787                 DIFF_OPT_CLR(options, RENAME_EMPTY);
3788         else if (!strcmp(arg, "--relative"))
3789                 DIFF_OPT_SET(options, RELATIVE_NAME);
3790         else if (skip_prefix(arg, "--relative=", &arg)) {
3791                 DIFF_OPT_SET(options, RELATIVE_NAME);
3792                 options->prefix = arg;
3793         }
3794
3795         /* xdiff options */
3796         else if (!strcmp(arg, "--minimal"))
3797                 DIFF_XDL_SET(options, NEED_MINIMAL);
3798         else if (!strcmp(arg, "--no-minimal"))
3799                 DIFF_XDL_CLR(options, NEED_MINIMAL);
3800         else if (!strcmp(arg, "-w") || !strcmp(arg, "--ignore-all-space"))
3801                 DIFF_XDL_SET(options, IGNORE_WHITESPACE);
3802         else if (!strcmp(arg, "-b") || !strcmp(arg, "--ignore-space-change"))
3803                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_CHANGE);
3804         else if (!strcmp(arg, "--ignore-space-at-eol"))
3805                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_AT_EOL);
3806         else if (!strcmp(arg, "--ignore-blank-lines"))
3807                 DIFF_XDL_SET(options, IGNORE_BLANK_LINES);
3808         else if (!strcmp(arg, "--compaction-heuristic"))
3809                 DIFF_XDL_SET(options, COMPACTION_HEURISTIC);
3810         else if (!strcmp(arg, "--no-compaction-heuristic"))
3811                 DIFF_XDL_CLR(options, COMPACTION_HEURISTIC);
3812         else if (!strcmp(arg, "--patience"))
3813                 options->xdl_opts = DIFF_WITH_ALG(options, PATIENCE_DIFF);
3814         else if (!strcmp(arg, "--histogram"))
3815                 options->xdl_opts = DIFF_WITH_ALG(options, HISTOGRAM_DIFF);
3816         else if ((argcount = parse_long_opt("diff-algorithm", av, &optarg))) {
3817                 long value = parse_algorithm_value(optarg);
3818                 if (value < 0)
3819                         return error("option diff-algorithm accepts \"myers\", "
3820                                      "\"minimal\", \"patience\" and \"histogram\"");
3821                 /* clear out previous settings */
3822                 DIFF_XDL_CLR(options, NEED_MINIMAL);
3823                 options->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3824                 options->xdl_opts |= value;
3825                 return argcount;
3826         }
3827
3828         /* flags options */
3829         else if (!strcmp(arg, "--binary")) {
3830                 enable_patch_output(&options->output_format);
3831                 DIFF_OPT_SET(options, BINARY);
3832         }
3833         else if (!strcmp(arg, "--full-index"))
3834                 DIFF_OPT_SET(options, FULL_INDEX);
3835         else if (!strcmp(arg, "-a") || !strcmp(arg, "--text"))
3836                 DIFF_OPT_SET(options, TEXT);
3837         else if (!strcmp(arg, "-R"))
3838                 DIFF_OPT_SET(options, REVERSE_DIFF);
3839         else if (!strcmp(arg, "--find-copies-harder"))
3840                 DIFF_OPT_SET(options, FIND_COPIES_HARDER);
3841         else if (!strcmp(arg, "--follow"))
3842                 DIFF_OPT_SET(options, FOLLOW_RENAMES);
3843         else if (!strcmp(arg, "--no-follow")) {
3844                 DIFF_OPT_CLR(options, FOLLOW_RENAMES);
3845                 DIFF_OPT_CLR(options, DEFAULT_FOLLOW_RENAMES);
3846         } else if (!strcmp(arg, "--color"))
3847                 options->use_color = 1;
3848         else if (skip_prefix(arg, "--color=", &arg)) {
3849                 int value = git_config_colorbool(NULL, arg);
3850                 if (value < 0)
3851                         return error("option `color' expects \"always\", \"auto\", or \"never\"");
3852                 options->use_color = value;
3853         }
3854         else if (!strcmp(arg, "--no-color"))
3855                 options->use_color = 0;
3856         else if (!strcmp(arg, "--color-words")) {
3857                 options->use_color = 1;
3858                 options->word_diff = DIFF_WORDS_COLOR;
3859         }
3860         else if (skip_prefix(arg, "--color-words=", &arg)) {
3861                 options->use_color = 1;
3862                 options->word_diff = DIFF_WORDS_COLOR;
3863                 options->word_regex = arg;
3864         }
3865         else if (!strcmp(arg, "--word-diff")) {
3866                 if (options->word_diff == DIFF_WORDS_NONE)
3867                         options->word_diff = DIFF_WORDS_PLAIN;
3868         }
3869         else if (skip_prefix(arg, "--word-diff=", &arg)) {
3870                 if (!strcmp(arg, "plain"))
3871                         options->word_diff = DIFF_WORDS_PLAIN;
3872                 else if (!strcmp(arg, "color")) {
3873                         options->use_color = 1;
3874                         options->word_diff = DIFF_WORDS_COLOR;
3875                 }
3876                 else if (!strcmp(arg, "porcelain"))
3877                         options->word_diff = DIFF_WORDS_PORCELAIN;
3878                 else if (!strcmp(arg, "none"))
3879                         options->word_diff = DIFF_WORDS_NONE;
3880                 else
3881                         die("bad --word-diff argument: %s", arg);
3882         }
3883         else if ((argcount = parse_long_opt("word-diff-regex", av, &optarg))) {
3884                 if (options->word_diff == DIFF_WORDS_NONE)
3885                         options->word_diff = DIFF_WORDS_PLAIN;
3886                 options->word_regex = optarg;
3887                 return argcount;
3888         }
3889         else if (!strcmp(arg, "--exit-code"))
3890                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
3891         else if (!strcmp(arg, "--quiet"))
3892                 DIFF_OPT_SET(options, QUICK);
3893         else if (!strcmp(arg, "--ext-diff"))
3894                 DIFF_OPT_SET(options, ALLOW_EXTERNAL);
3895         else if (!strcmp(arg, "--no-ext-diff"))
3896                 DIFF_OPT_CLR(options, ALLOW_EXTERNAL);
3897         else if (!strcmp(arg, "--textconv"))
3898                 DIFF_OPT_SET(options, ALLOW_TEXTCONV);
3899         else if (!strcmp(arg, "--no-textconv"))
3900                 DIFF_OPT_CLR(options, ALLOW_TEXTCONV);
3901         else if (!strcmp(arg, "--ignore-submodules")) {
3902                 DIFF_OPT_SET(options, OVERRIDE_SUBMODULE_CONFIG);
3903                 handle_ignore_submodules_arg(options, "all");
3904         } else if (skip_prefix(arg, "--ignore-submodules=", &arg)) {
3905                 DIFF_OPT_SET(options, OVERRIDE_SUBMODULE_CONFIG);
3906                 handle_ignore_submodules_arg(options, arg);
3907         } else if (!strcmp(arg, "--submodule"))
3908                 DIFF_OPT_SET(options, SUBMODULE_LOG);
3909         else if (skip_prefix(arg, "--submodule=", &arg))
3910                 return parse_submodule_opt(options, arg);
3911         else if (skip_prefix(arg, "--ws-error-highlight=", &arg))
3912                 return parse_ws_error_highlight(options, arg);
3913
3914         /* misc options */
3915         else if (!strcmp(arg, "-z"))
3916                 options->line_termination = 0;
3917         else if ((argcount = short_opt('l', av, &optarg))) {
3918                 options->rename_limit = strtoul(optarg, NULL, 10);
3919                 return argcount;
3920         }
3921         else if ((argcount = short_opt('S', av, &optarg))) {
3922                 options->pickaxe = optarg;
3923                 options->pickaxe_opts |= DIFF_PICKAXE_KIND_S;
3924                 return argcount;
3925         } else if ((argcount = short_opt('G', av, &optarg))) {
3926                 options->pickaxe = optarg;
3927                 options->pickaxe_opts |= DIFF_PICKAXE_KIND_G;
3928                 return argcount;
3929         }
3930         else if (!strcmp(arg, "--pickaxe-all"))
3931                 options->pickaxe_opts |= DIFF_PICKAXE_ALL;
3932         else if (!strcmp(arg, "--pickaxe-regex"))
3933                 options->pickaxe_opts |= DIFF_PICKAXE_REGEX;
3934         else if ((argcount = short_opt('O', av, &optarg))) {
3935                 const char *path = prefix_filename(prefix, strlen(prefix), optarg);
3936                 options->orderfile = xstrdup(path);
3937                 return argcount;
3938         }
3939         else if ((argcount = parse_long_opt("diff-filter", av, &optarg))) {
3940                 int offending = parse_diff_filter_opt(optarg, options);
3941                 if (offending)
3942                         die("unknown change class '%c' in --diff-filter=%s",
3943                             offending, optarg);
3944                 return argcount;
3945         }
3946         else if (!strcmp(arg, "--abbrev"))
3947                 options->abbrev = DEFAULT_ABBREV;
3948         else if (skip_prefix(arg, "--abbrev=", &arg)) {
3949                 options->abbrev = strtoul(arg, NULL, 10);
3950                 if (options->abbrev < MINIMUM_ABBREV)
3951                         options->abbrev = MINIMUM_ABBREV;
3952                 else if (40 < options->abbrev)
3953                         options->abbrev = 40;
3954         }
3955         else if ((argcount = parse_long_opt("src-prefix", av, &optarg))) {
3956                 options->a_prefix = optarg;
3957                 return argcount;
3958         }
3959         else if ((argcount = parse_long_opt("dst-prefix", av, &optarg))) {
3960                 options->b_prefix = optarg;
3961                 return argcount;
3962         }
3963         else if (!strcmp(arg, "--no-prefix"))
3964                 options->a_prefix = options->b_prefix = "";
3965         else if (opt_arg(arg, '\0', "inter-hunk-context",
3966                          &options->interhunkcontext))
3967                 ;
3968         else if (!strcmp(arg, "-W"))
3969                 DIFF_OPT_SET(options, FUNCCONTEXT);
3970         else if (!strcmp(arg, "--function-context"))
3971                 DIFF_OPT_SET(options, FUNCCONTEXT);
3972         else if (!strcmp(arg, "--no-function-context"))
3973                 DIFF_OPT_CLR(options, FUNCCONTEXT);
3974         else if ((argcount = parse_long_opt("output", av, &optarg))) {
3975                 const char *path = prefix_filename(prefix, strlen(prefix), optarg);
3976                 options->file = fopen(path, "w");
3977                 if (!options->file)
3978                         die_errno("Could not open '%s'", path);
3979                 options->close_file = 1;
3980                 if (options->use_color != GIT_COLOR_ALWAYS)
3981                         options->use_color = GIT_COLOR_NEVER;
3982                 return argcount;
3983         } else
3984                 return 0;
3985         return 1;
3986 }
3987
3988 int parse_rename_score(const char **cp_p)
3989 {
3990         unsigned long num, scale;
3991         int ch, dot;
3992         const char *cp = *cp_p;
3993
3994         num = 0;
3995         scale = 1;
3996         dot = 0;
3997         for (;;) {
3998                 ch = *cp;
3999                 if ( !dot && ch == '.' ) {
4000                         scale = 1;
4001                         dot = 1;
4002                 } else if ( ch == '%' ) {
4003                         scale = dot ? scale*100 : 100;
4004                         cp++;   /* % is always at the end */
4005                         break;
4006                 } else if ( ch >= '0' && ch <= '9' ) {
4007                         if ( scale < 100000 ) {
4008                                 scale *= 10;
4009                                 num = (num*10) + (ch-'0');
4010                         }
4011                 } else {
4012                         break;
4013                 }
4014                 cp++;
4015         }
4016         *cp_p = cp;
4017
4018         /* user says num divided by scale and we say internally that
4019          * is MAX_SCORE * num / scale.
4020          */
4021         return (int)((num >= scale) ? MAX_SCORE : (MAX_SCORE * num / scale));
4022 }
4023
4024 static int diff_scoreopt_parse(const char *opt)
4025 {
4026         int opt1, opt2, cmd;
4027
4028         if (*opt++ != '-')
4029                 return -1;
4030         cmd = *opt++;
4031         if (cmd == '-') {
4032                 /* convert the long-form arguments into short-form versions */
4033                 if (skip_prefix(opt, "break-rewrites", &opt)) {
4034                         if (*opt == 0 || *opt++ == '=')
4035                                 cmd = 'B';
4036                 } else if (skip_prefix(opt, "find-copies", &opt)) {
4037                         if (*opt == 0 || *opt++ == '=')
4038                                 cmd = 'C';
4039                 } else if (skip_prefix(opt, "find-renames", &opt)) {
4040                         if (*opt == 0 || *opt++ == '=')
4041                                 cmd = 'M';
4042                 }
4043         }
4044         if (cmd != 'M' && cmd != 'C' && cmd != 'B')
4045                 return -1; /* that is not a -M, -C, or -B option */
4046
4047         opt1 = parse_rename_score(&opt);
4048         if (cmd != 'B')
4049                 opt2 = 0;
4050         else {
4051                 if (*opt == 0)
4052                         opt2 = 0;
4053                 else if (*opt != '/')
4054                         return -1; /* we expect -B80/99 or -B80 */
4055                 else {
4056                         opt++;
4057                         opt2 = parse_rename_score(&opt);
4058                 }
4059         }
4060         if (*opt != 0)
4061                 return -1;
4062         return opt1 | (opt2 << 16);
4063 }
4064
4065 struct diff_queue_struct diff_queued_diff;
4066
4067 void diff_q(struct diff_queue_struct *queue, struct diff_filepair *dp)
4068 {
4069         ALLOC_GROW(queue->queue, queue->nr + 1, queue->alloc);
4070         queue->queue[queue->nr++] = dp;
4071 }
4072
4073 struct diff_filepair *diff_queue(struct diff_queue_struct *queue,
4074                                  struct diff_filespec *one,
4075                                  struct diff_filespec *two)
4076 {
4077         struct diff_filepair *dp = xcalloc(1, sizeof(*dp));
4078         dp->one = one;
4079         dp->two = two;
4080         if (queue)
4081                 diff_q(queue, dp);
4082         return dp;
4083 }
4084
4085 void diff_free_filepair(struct diff_filepair *p)
4086 {
4087         free_filespec(p->one);
4088         free_filespec(p->two);
4089         free(p);
4090 }
4091
4092 /* This is different from find_unique_abbrev() in that
4093  * it stuffs the result with dots for alignment.
4094  */
4095 const char *diff_unique_abbrev(const unsigned char *sha1, int len)
4096 {
4097         int abblen;
4098         const char *abbrev;
4099         if (len == 40)
4100                 return sha1_to_hex(sha1);
4101
4102         abbrev = find_unique_abbrev(sha1, len);
4103         abblen = strlen(abbrev);
4104         if (abblen < 37) {
4105                 static char hex[41];
4106                 if (len < abblen && abblen <= len + 2)
4107                         xsnprintf(hex, sizeof(hex), "%s%.*s", abbrev, len+3-abblen, "..");
4108                 else
4109                         xsnprintf(hex, sizeof(hex), "%s...", abbrev);
4110                 return hex;
4111         }
4112         return sha1_to_hex(sha1);
4113 }
4114
4115 static void diff_flush_raw(struct diff_filepair *p, struct diff_options *opt)
4116 {
4117         int line_termination = opt->line_termination;
4118         int inter_name_termination = line_termination ? '\t' : '\0';
4119
4120         fprintf(opt->file, "%s", diff_line_prefix(opt));
4121         if (!(opt->output_format & DIFF_FORMAT_NAME_STATUS)) {
4122                 fprintf(opt->file, ":%06o %06o %s ", p->one->mode, p->two->mode,
4123                         diff_unique_abbrev(p->one->oid.hash, opt->abbrev));
4124                 fprintf(opt->file, "%s ",
4125                         diff_unique_abbrev(p->two->oid.hash, opt->abbrev));
4126         }
4127         if (p->score) {
4128                 fprintf(opt->file, "%c%03d%c", p->status, similarity_index(p),
4129                         inter_name_termination);
4130         } else {
4131                 fprintf(opt->file, "%c%c", p->status, inter_name_termination);
4132         }
4133
4134         if (p->status == DIFF_STATUS_COPIED ||
4135             p->status == DIFF_STATUS_RENAMED) {
4136                 const char *name_a, *name_b;
4137                 name_a = p->one->path;
4138                 name_b = p->two->path;
4139                 strip_prefix(opt->prefix_length, &name_a, &name_b);
4140                 write_name_quoted(name_a, opt->file, inter_name_termination);
4141                 write_name_quoted(name_b, opt->file, line_termination);
4142         } else {
4143                 const char *name_a, *name_b;
4144                 name_a = p->one->mode ? p->one->path : p->two->path;
4145                 name_b = NULL;
4146                 strip_prefix(opt->prefix_length, &name_a, &name_b);
4147                 write_name_quoted(name_a, opt->file, line_termination);
4148         }
4149 }
4150
4151 int diff_unmodified_pair(struct diff_filepair *p)
4152 {
4153         /* This function is written stricter than necessary to support
4154          * the currently implemented transformers, but the idea is to
4155          * let transformers to produce diff_filepairs any way they want,
4156          * and filter and clean them up here before producing the output.
4157          */
4158         struct diff_filespec *one = p->one, *two = p->two;
4159
4160         if (DIFF_PAIR_UNMERGED(p))
4161                 return 0; /* unmerged is interesting */
4162
4163         /* deletion, addition, mode or type change
4164          * and rename are all interesting.
4165          */
4166         if (DIFF_FILE_VALID(one) != DIFF_FILE_VALID(two) ||
4167             DIFF_PAIR_MODE_CHANGED(p) ||
4168             strcmp(one->path, two->path))
4169                 return 0;
4170
4171         /* both are valid and point at the same path.  that is, we are
4172          * dealing with a change.
4173          */
4174         if (one->oid_valid && two->oid_valid &&
4175             !oidcmp(&one->oid, &two->oid) &&
4176             !one->dirty_submodule && !two->dirty_submodule)
4177                 return 1; /* no change */
4178         if (!one->oid_valid && !two->oid_valid)
4179                 return 1; /* both look at the same file on the filesystem. */
4180         return 0;
4181 }
4182
4183 static void diff_flush_patch(struct diff_filepair *p, struct diff_options *o)
4184 {
4185         if (diff_unmodified_pair(p))
4186                 return;
4187
4188         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
4189             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
4190                 return; /* no tree diffs in patch format */
4191
4192         run_diff(p, o);
4193 }
4194
4195 static void diff_flush_stat(struct diff_filepair *p, struct diff_options *o,
4196                             struct diffstat_t *diffstat)
4197 {
4198         if (diff_unmodified_pair(p))
4199                 return;
4200
4201         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
4202             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
4203                 return; /* no useful stat for tree diffs */
4204
4205         run_diffstat(p, o, diffstat);
4206 }
4207
4208 static void diff_flush_checkdiff(struct diff_filepair *p,
4209                 struct diff_options *o)
4210 {
4211         if (diff_unmodified_pair(p))
4212                 return;
4213
4214         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
4215             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
4216                 return; /* nothing to check in tree diffs */
4217
4218         run_checkdiff(p, o);
4219 }
4220
4221 int diff_queue_is_empty(void)
4222 {
4223         struct diff_queue_struct *q = &diff_queued_diff;
4224         int i;
4225         for (i = 0; i < q->nr; i++)
4226                 if (!diff_unmodified_pair(q->queue[i]))
4227                         return 0;
4228         return 1;
4229 }
4230
4231 #if DIFF_DEBUG
4232 void diff_debug_filespec(struct diff_filespec *s, int x, const char *one)
4233 {
4234         fprintf(stderr, "queue[%d] %s (%s) %s %06o %s\n",
4235                 x, one ? one : "",
4236                 s->path,
4237                 DIFF_FILE_VALID(s) ? "valid" : "invalid",
4238                 s->mode,
4239                 s->oid_valid ? oid_to_hex(&s->oid) : "");
4240         fprintf(stderr, "queue[%d] %s size %lu\n",
4241                 x, one ? one : "",
4242                 s->size);
4243 }
4244
4245 void diff_debug_filepair(const struct diff_filepair *p, int i)
4246 {
4247         diff_debug_filespec(p->one, i, "one");
4248         diff_debug_filespec(p->two, i, "two");
4249         fprintf(stderr, "score %d, status %c rename_used %d broken %d\n",
4250                 p->score, p->status ? p->status : '?',
4251                 p->one->rename_used, p->broken_pair);
4252 }
4253
4254 void diff_debug_queue(const char *msg, struct diff_queue_struct *q)
4255 {
4256         int i;
4257         if (msg)
4258                 fprintf(stderr, "%s\n", msg);
4259         fprintf(stderr, "q->nr = %d\n", q->nr);
4260         for (i = 0; i < q->nr; i++) {
4261                 struct diff_filepair *p = q->queue[i];
4262                 diff_debug_filepair(p, i);
4263         }
4264 }
4265 #endif
4266
4267 static void diff_resolve_rename_copy(void)
4268 {
4269         int i;
4270         struct diff_filepair *p;
4271         struct diff_queue_struct *q = &diff_queued_diff;
4272
4273         diff_debug_queue("resolve-rename-copy", q);
4274
4275         for (i = 0; i < q->nr; i++) {
4276                 p = q->queue[i];
4277                 p->status = 0; /* undecided */
4278                 if (DIFF_PAIR_UNMERGED(p))
4279                         p->status = DIFF_STATUS_UNMERGED;
4280                 else if (!DIFF_FILE_VALID(p->one))
4281                         p->status = DIFF_STATUS_ADDED;
4282                 else if (!DIFF_FILE_VALID(p->two))
4283                         p->status = DIFF_STATUS_DELETED;
4284                 else if (DIFF_PAIR_TYPE_CHANGED(p))
4285                         p->status = DIFF_STATUS_TYPE_CHANGED;
4286
4287                 /* from this point on, we are dealing with a pair
4288                  * whose both sides are valid and of the same type, i.e.
4289                  * either in-place edit or rename/copy edit.
4290                  */
4291                 else if (DIFF_PAIR_RENAME(p)) {
4292                         /*
4293                          * A rename might have re-connected a broken
4294                          * pair up, causing the pathnames to be the
4295                          * same again. If so, that's not a rename at
4296                          * all, just a modification..
4297                          *
4298                          * Otherwise, see if this source was used for
4299                          * multiple renames, in which case we decrement
4300                          * the count, and call it a copy.
4301                          */
4302                         if (!strcmp(p->one->path, p->two->path))
4303                                 p->status = DIFF_STATUS_MODIFIED;
4304                         else if (--p->one->rename_used > 0)
4305                                 p->status = DIFF_STATUS_COPIED;
4306                         else
4307                                 p->status = DIFF_STATUS_RENAMED;
4308                 }
4309                 else if (oidcmp(&p->one->oid, &p->two->oid) ||
4310                          p->one->mode != p->two->mode ||
4311                          p->one->dirty_submodule ||
4312                          p->two->dirty_submodule ||
4313                          is_null_oid(&p->one->oid))
4314                         p->status = DIFF_STATUS_MODIFIED;
4315                 else {
4316                         /* This is a "no-change" entry and should not
4317                          * happen anymore, but prepare for broken callers.
4318                          */
4319                         error("feeding unmodified %s to diffcore",
4320                               p->one->path);
4321                         p->status = DIFF_STATUS_UNKNOWN;
4322                 }
4323         }
4324         diff_debug_queue("resolve-rename-copy done", q);
4325 }
4326
4327 static int check_pair_status(struct diff_filepair *p)
4328 {
4329         switch (p->status) {
4330         case DIFF_STATUS_UNKNOWN:
4331                 return 0;
4332         case 0:
4333                 die("internal error in diff-resolve-rename-copy");
4334         default:
4335                 return 1;
4336         }
4337 }
4338
4339 static void flush_one_pair(struct diff_filepair *p, struct diff_options *opt)
4340 {
4341         int fmt = opt->output_format;
4342
4343         if (fmt & DIFF_FORMAT_CHECKDIFF)
4344                 diff_flush_checkdiff(p, opt);
4345         else if (fmt & (DIFF_FORMAT_RAW | DIFF_FORMAT_NAME_STATUS))
4346                 diff_flush_raw(p, opt);
4347         else if (fmt & DIFF_FORMAT_NAME) {
4348                 const char *name_a, *name_b;
4349                 name_a = p->two->path;
4350                 name_b = NULL;
4351                 strip_prefix(opt->prefix_length, &name_a, &name_b);
4352                 write_name_quoted(name_a, opt->file, opt->line_termination);
4353         }
4354 }
4355
4356 static void show_file_mode_name(FILE *file, const char *newdelete, struct diff_filespec *fs)
4357 {
4358         if (fs->mode)
4359                 fprintf(file, " %s mode %06o ", newdelete, fs->mode);
4360         else
4361                 fprintf(file, " %s ", newdelete);
4362         write_name_quoted(fs->path, file, '\n');
4363 }
4364
4365
4366 static void show_mode_change(FILE *file, struct diff_filepair *p, int show_name,
4367                 const char *line_prefix)
4368 {
4369         if (p->one->mode && p->two->mode && p->one->mode != p->two->mode) {
4370                 fprintf(file, "%s mode change %06o => %06o%c", line_prefix, p->one->mode,
4371                         p->two->mode, show_name ? ' ' : '\n');
4372                 if (show_name) {
4373                         write_name_quoted(p->two->path, file, '\n');
4374                 }
4375         }
4376 }
4377
4378 static void show_rename_copy(FILE *file, const char *renamecopy, struct diff_filepair *p,
4379                         const char *line_prefix)
4380 {
4381         char *names = pprint_rename(p->one->path, p->two->path);
4382
4383         fprintf(file, " %s %s (%d%%)\n", renamecopy, names, similarity_index(p));
4384         free(names);
4385         show_mode_change(file, p, 0, line_prefix);
4386 }
4387
4388 static void diff_summary(struct diff_options *opt, struct diff_filepair *p)
4389 {
4390         FILE *file = opt->file;
4391         const char *line_prefix = diff_line_prefix(opt);
4392
4393         switch(p->status) {
4394         case DIFF_STATUS_DELETED:
4395                 fputs(line_prefix, file);
4396                 show_file_mode_name(file, "delete", p->one);
4397                 break;
4398         case DIFF_STATUS_ADDED:
4399                 fputs(line_prefix, file);
4400                 show_file_mode_name(file, "create", p->two);
4401                 break;
4402         case DIFF_STATUS_COPIED:
4403                 fputs(line_prefix, file);
4404                 show_rename_copy(file, "copy", p, line_prefix);
4405                 break;
4406         case DIFF_STATUS_RENAMED:
4407                 fputs(line_prefix, file);
4408                 show_rename_copy(file, "rename", p, line_prefix);
4409                 break;
4410         default:
4411                 if (p->score) {
4412                         fprintf(file, "%s rewrite ", line_prefix);
4413                         write_name_quoted(p->two->path, file, ' ');
4414                         fprintf(file, "(%d%%)\n", similarity_index(p));
4415                 }
4416                 show_mode_change(file, p, !p->score, line_prefix);
4417                 break;
4418         }
4419 }
4420
4421 struct patch_id_t {
4422         git_SHA_CTX *ctx;
4423         int patchlen;
4424 };
4425
4426 static int remove_space(char *line, int len)
4427 {
4428         int i;
4429         char *dst = line;
4430         unsigned char c;
4431
4432         for (i = 0; i < len; i++)
4433                 if (!isspace((c = line[i])))
4434                         *dst++ = c;
4435
4436         return dst - line;
4437 }
4438
4439 static void patch_id_consume(void *priv, char *line, unsigned long len)
4440 {
4441         struct patch_id_t *data = priv;
4442         int new_len;
4443
4444         /* Ignore line numbers when computing the SHA1 of the patch */
4445         if (starts_with(line, "@@ -"))
4446                 return;
4447
4448         new_len = remove_space(line, len);
4449
4450         git_SHA1_Update(data->ctx, line, new_len);
4451         data->patchlen += new_len;
4452 }
4453
4454 /* returns 0 upon success, and writes result into sha1 */
4455 static int diff_get_patch_id(struct diff_options *options, unsigned char *sha1, int diff_header_only)
4456 {
4457         struct diff_queue_struct *q = &diff_queued_diff;
4458         int i;
4459         git_SHA_CTX ctx;
4460         struct patch_id_t data;
4461         char buffer[PATH_MAX * 4 + 20];
4462
4463         git_SHA1_Init(&ctx);
4464         memset(&data, 0, sizeof(struct patch_id_t));
4465         data.ctx = &ctx;
4466
4467         for (i = 0; i < q->nr; i++) {
4468                 xpparam_t xpp;
4469                 xdemitconf_t xecfg;
4470                 mmfile_t mf1, mf2;
4471                 struct diff_filepair *p = q->queue[i];
4472                 int len1, len2;
4473
4474                 memset(&xpp, 0, sizeof(xpp));
4475                 memset(&xecfg, 0, sizeof(xecfg));
4476                 if (p->status == 0)
4477                         return error("internal diff status error");
4478                 if (p->status == DIFF_STATUS_UNKNOWN)
4479                         continue;
4480                 if (diff_unmodified_pair(p))
4481                         continue;
4482                 if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
4483                     (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
4484                         continue;
4485                 if (DIFF_PAIR_UNMERGED(p))
4486                         continue;
4487
4488                 diff_fill_sha1_info(p->one);
4489                 diff_fill_sha1_info(p->two);
4490
4491                 len1 = remove_space(p->one->path, strlen(p->one->path));
4492                 len2 = remove_space(p->two->path, strlen(p->two->path));
4493                 if (p->one->mode == 0)
4494                         len1 = snprintf(buffer, sizeof(buffer),
4495                                         "diff--gita/%.*sb/%.*s"
4496                                         "newfilemode%06o"
4497                                         "---/dev/null"
4498                                         "+++b/%.*s",
4499                                         len1, p->one->path,
4500                                         len2, p->two->path,
4501                                         p->two->mode,
4502                                         len2, p->two->path);
4503                 else if (p->two->mode == 0)
4504                         len1 = snprintf(buffer, sizeof(buffer),
4505                                         "diff--gita/%.*sb/%.*s"
4506                                         "deletedfilemode%06o"
4507                                         "---a/%.*s"
4508                                         "+++/dev/null",
4509                                         len1, p->one->path,
4510                                         len2, p->two->path,
4511                                         p->one->mode,
4512                                         len1, p->one->path);
4513                 else
4514                         len1 = snprintf(buffer, sizeof(buffer),
4515                                         "diff--gita/%.*sb/%.*s"
4516                                         "---a/%.*s"
4517                                         "+++b/%.*s",
4518                                         len1, p->one->path,
4519                                         len2, p->two->path,
4520                                         len1, p->one->path,
4521                                         len2, p->two->path);
4522                 git_SHA1_Update(&ctx, buffer, len1);
4523
4524                 if (diff_header_only)
4525                         continue;
4526
4527                 if (fill_mmfile(&mf1, p->one) < 0 ||
4528                     fill_mmfile(&mf2, p->two) < 0)
4529                         return error("unable to read files to diff");
4530
4531                 if (diff_filespec_is_binary(p->one) ||
4532                     diff_filespec_is_binary(p->two)) {
4533                         git_SHA1_Update(&ctx, oid_to_hex(&p->one->oid),
4534                                         40);
4535                         git_SHA1_Update(&ctx, oid_to_hex(&p->two->oid),
4536                                         40);
4537                         continue;
4538                 }
4539
4540                 xpp.flags = 0;
4541                 xecfg.ctxlen = 3;
4542                 xecfg.flags = 0;
4543                 if (xdi_diff_outf(&mf1, &mf2, patch_id_consume, &data,
4544                                   &xpp, &xecfg))
4545                         return error("unable to generate patch-id diff for %s",
4546                                      p->one->path);
4547         }
4548
4549         git_SHA1_Final(sha1, &ctx);
4550         return 0;
4551 }
4552
4553 int diff_flush_patch_id(struct diff_options *options, unsigned char *sha1, int diff_header_only)
4554 {
4555         struct diff_queue_struct *q = &diff_queued_diff;
4556         int i;
4557         int result = diff_get_patch_id(options, sha1, diff_header_only);
4558
4559         for (i = 0; i < q->nr; i++)
4560                 diff_free_filepair(q->queue[i]);
4561
4562         free(q->queue);
4563         DIFF_QUEUE_CLEAR(q);
4564
4565         return result;
4566 }
4567
4568 static int is_summary_empty(const struct diff_queue_struct *q)
4569 {
4570         int i;
4571
4572         for (i = 0; i < q->nr; i++) {
4573                 const struct diff_filepair *p = q->queue[i];
4574
4575                 switch (p->status) {
4576                 case DIFF_STATUS_DELETED:
4577                 case DIFF_STATUS_ADDED:
4578                 case DIFF_STATUS_COPIED:
4579                 case DIFF_STATUS_RENAMED:
4580                         return 0;
4581                 default:
4582                         if (p->score)
4583                                 return 0;
4584                         if (p->one->mode && p->two->mode &&
4585                             p->one->mode != p->two->mode)
4586                                 return 0;
4587                         break;
4588                 }
4589         }
4590         return 1;
4591 }
4592
4593 static const char rename_limit_warning[] =
4594 "inexact rename detection was skipped due to too many files.";
4595
4596 static const char degrade_cc_to_c_warning[] =
4597 "only found copies from modified paths due to too many files.";
4598
4599 static const char rename_limit_advice[] =
4600 "you may want to set your %s variable to at least "
4601 "%d and retry the command.";
4602
4603 void diff_warn_rename_limit(const char *varname, int needed, int degraded_cc)
4604 {
4605         if (degraded_cc)
4606                 warning(degrade_cc_to_c_warning);
4607         else if (needed)
4608                 warning(rename_limit_warning);
4609         else
4610                 return;
4611         if (0 < needed && needed < 32767)
4612                 warning(rename_limit_advice, varname, needed);
4613 }
4614
4615 void diff_flush(struct diff_options *options)
4616 {
4617         struct diff_queue_struct *q = &diff_queued_diff;
4618         int i, output_format = options->output_format;
4619         int separator = 0;
4620         int dirstat_by_line = 0;
4621
4622         /*
4623          * Order: raw, stat, summary, patch
4624          * or:    name/name-status/checkdiff (other bits clear)
4625          */
4626         if (!q->nr)
4627                 goto free_queue;
4628
4629         if (output_format & (DIFF_FORMAT_RAW |
4630                              DIFF_FORMAT_NAME |
4631                              DIFF_FORMAT_NAME_STATUS |
4632                              DIFF_FORMAT_CHECKDIFF)) {
4633                 for (i = 0; i < q->nr; i++) {
4634                         struct diff_filepair *p = q->queue[i];
4635                         if (check_pair_status(p))
4636                                 flush_one_pair(p, options);
4637                 }
4638                 separator++;
4639         }
4640
4641         if (output_format & DIFF_FORMAT_DIRSTAT && DIFF_OPT_TST(options, DIRSTAT_BY_LINE))
4642                 dirstat_by_line = 1;
4643
4644         if (output_format & (DIFF_FORMAT_DIFFSTAT|DIFF_FORMAT_SHORTSTAT|DIFF_FORMAT_NUMSTAT) ||
4645             dirstat_by_line) {
4646                 struct diffstat_t diffstat;
4647
4648                 memset(&diffstat, 0, sizeof(struct diffstat_t));
4649                 for (i = 0; i < q->nr; i++) {
4650                         struct diff_filepair *p = q->queue[i];
4651                         if (check_pair_status(p))
4652                                 diff_flush_stat(p, options, &diffstat);
4653                 }
4654                 if (output_format & DIFF_FORMAT_NUMSTAT)
4655                         show_numstat(&diffstat, options);
4656                 if (output_format & DIFF_FORMAT_DIFFSTAT)
4657                         show_stats(&diffstat, options);
4658                 if (output_format & DIFF_FORMAT_SHORTSTAT)
4659                         show_shortstats(&diffstat, options);
4660                 if (output_format & DIFF_FORMAT_DIRSTAT && dirstat_by_line)
4661                         show_dirstat_by_line(&diffstat, options);
4662                 free_diffstat_info(&diffstat);
4663                 separator++;
4664         }
4665         if ((output_format & DIFF_FORMAT_DIRSTAT) && !dirstat_by_line)
4666                 show_dirstat(options);
4667
4668         if (output_format & DIFF_FORMAT_SUMMARY && !is_summary_empty(q)) {
4669                 for (i = 0; i < q->nr; i++) {
4670                         diff_summary(options, q->queue[i]);
4671                 }
4672                 separator++;
4673         }
4674
4675         if (output_format & DIFF_FORMAT_NO_OUTPUT &&
4676             DIFF_OPT_TST(options, EXIT_WITH_STATUS) &&
4677             DIFF_OPT_TST(options, DIFF_FROM_CONTENTS)) {
4678                 /*
4679                  * run diff_flush_patch for the exit status. setting
4680                  * options->file to /dev/null should be safe, because we
4681                  * aren't supposed to produce any output anyway.
4682                  */
4683                 if (options->close_file)
4684                         fclose(options->file);
4685                 options->file = fopen("/dev/null", "w");
4686                 if (!options->file)
4687                         die_errno("Could not open /dev/null");
4688                 options->close_file = 1;
4689                 for (i = 0; i < q->nr; i++) {
4690                         struct diff_filepair *p = q->queue[i];
4691                         if (check_pair_status(p))
4692                                 diff_flush_patch(p, options);
4693                         if (options->found_changes)
4694                                 break;
4695                 }
4696         }
4697
4698         if (output_format & DIFF_FORMAT_PATCH) {
4699                 if (separator) {
4700                         fprintf(options->file, "%s%c",
4701                                 diff_line_prefix(options),
4702                                 options->line_termination);
4703                         if (options->stat_sep) {
4704                                 /* attach patch instead of inline */
4705                                 fputs(options->stat_sep, options->file);
4706                         }
4707                 }
4708
4709                 for (i = 0; i < q->nr; i++) {
4710                         struct diff_filepair *p = q->queue[i];
4711                         if (check_pair_status(p))
4712                                 diff_flush_patch(p, options);
4713                 }
4714         }
4715
4716         if (output_format & DIFF_FORMAT_CALLBACK)
4717                 options->format_callback(q, options, options->format_callback_data);
4718
4719         for (i = 0; i < q->nr; i++)
4720                 diff_free_filepair(q->queue[i]);
4721 free_queue:
4722         free(q->queue);
4723         DIFF_QUEUE_CLEAR(q);
4724         if (options->close_file)
4725                 fclose(options->file);
4726
4727         /*
4728          * Report the content-level differences with HAS_CHANGES;
4729          * diff_addremove/diff_change does not set the bit when
4730          * DIFF_FROM_CONTENTS is in effect (e.g. with -w).
4731          */
4732         if (DIFF_OPT_TST(options, DIFF_FROM_CONTENTS)) {
4733                 if (options->found_changes)
4734                         DIFF_OPT_SET(options, HAS_CHANGES);
4735                 else
4736                         DIFF_OPT_CLR(options, HAS_CHANGES);
4737         }
4738 }
4739
4740 static int match_filter(const struct diff_options *options, const struct diff_filepair *p)
4741 {
4742         return (((p->status == DIFF_STATUS_MODIFIED) &&
4743                  ((p->score &&
4744                    filter_bit_tst(DIFF_STATUS_FILTER_BROKEN, options)) ||
4745                   (!p->score &&
4746                    filter_bit_tst(DIFF_STATUS_MODIFIED, options)))) ||
4747                 ((p->status != DIFF_STATUS_MODIFIED) &&
4748                  filter_bit_tst(p->status, options)));
4749 }
4750
4751 static void diffcore_apply_filter(struct diff_options *options)
4752 {
4753         int i;
4754         struct diff_queue_struct *q = &diff_queued_diff;
4755         struct diff_queue_struct outq;
4756
4757         DIFF_QUEUE_CLEAR(&outq);
4758
4759         if (!options->filter)
4760                 return;
4761
4762         if (filter_bit_tst(DIFF_STATUS_FILTER_AON, options)) {
4763                 int found;
4764                 for (i = found = 0; !found && i < q->nr; i++) {
4765                         if (match_filter(options, q->queue[i]))
4766                                 found++;
4767                 }
4768                 if (found)
4769                         return;
4770
4771                 /* otherwise we will clear the whole queue
4772                  * by copying the empty outq at the end of this
4773                  * function, but first clear the current entries
4774                  * in the queue.
4775                  */
4776                 for (i = 0; i < q->nr; i++)
4777                         diff_free_filepair(q->queue[i]);
4778         }
4779         else {
4780                 /* Only the matching ones */
4781                 for (i = 0; i < q->nr; i++) {
4782                         struct diff_filepair *p = q->queue[i];
4783                         if (match_filter(options, p))
4784                                 diff_q(&outq, p);
4785                         else
4786                                 diff_free_filepair(p);
4787                 }
4788         }
4789         free(q->queue);
4790         *q = outq;
4791 }
4792
4793 /* Check whether two filespecs with the same mode and size are identical */
4794 static int diff_filespec_is_identical(struct diff_filespec *one,
4795                                       struct diff_filespec *two)
4796 {
4797         if (S_ISGITLINK(one->mode))
4798                 return 0;
4799         if (diff_populate_filespec(one, 0))
4800                 return 0;
4801         if (diff_populate_filespec(two, 0))
4802                 return 0;
4803         return !memcmp(one->data, two->data, one->size);
4804 }
4805
4806 static int diff_filespec_check_stat_unmatch(struct diff_filepair *p)
4807 {
4808         if (p->done_skip_stat_unmatch)
4809                 return p->skip_stat_unmatch_result;
4810
4811         p->done_skip_stat_unmatch = 1;
4812         p->skip_stat_unmatch_result = 0;
4813         /*
4814          * 1. Entries that come from stat info dirtiness
4815          *    always have both sides (iow, not create/delete),
4816          *    one side of the object name is unknown, with
4817          *    the same mode and size.  Keep the ones that
4818          *    do not match these criteria.  They have real
4819          *    differences.
4820          *
4821          * 2. At this point, the file is known to be modified,
4822          *    with the same mode and size, and the object
4823          *    name of one side is unknown.  Need to inspect
4824          *    the identical contents.
4825          */
4826         if (!DIFF_FILE_VALID(p->one) || /* (1) */
4827             !DIFF_FILE_VALID(p->two) ||
4828             (p->one->oid_valid && p->two->oid_valid) ||
4829             (p->one->mode != p->two->mode) ||
4830             diff_populate_filespec(p->one, CHECK_SIZE_ONLY) ||
4831             diff_populate_filespec(p->two, CHECK_SIZE_ONLY) ||
4832             (p->one->size != p->two->size) ||
4833             !diff_filespec_is_identical(p->one, p->two)) /* (2) */
4834                 p->skip_stat_unmatch_result = 1;
4835         return p->skip_stat_unmatch_result;
4836 }
4837
4838 static void diffcore_skip_stat_unmatch(struct diff_options *diffopt)
4839 {
4840         int i;
4841         struct diff_queue_struct *q = &diff_queued_diff;
4842         struct diff_queue_struct outq;
4843         DIFF_QUEUE_CLEAR(&outq);
4844
4845         for (i = 0; i < q->nr; i++) {
4846                 struct diff_filepair *p = q->queue[i];
4847
4848                 if (diff_filespec_check_stat_unmatch(p))
4849                         diff_q(&outq, p);
4850                 else {
4851                         /*
4852                          * The caller can subtract 1 from skip_stat_unmatch
4853                          * to determine how many paths were dirty only
4854                          * due to stat info mismatch.
4855                          */
4856                         if (!DIFF_OPT_TST(diffopt, NO_INDEX))
4857                                 diffopt->skip_stat_unmatch++;
4858                         diff_free_filepair(p);
4859                 }
4860         }
4861         free(q->queue);
4862         *q = outq;
4863 }
4864
4865 static int diffnamecmp(const void *a_, const void *b_)
4866 {
4867         const struct diff_filepair *a = *((const struct diff_filepair **)a_);
4868         const struct diff_filepair *b = *((const struct diff_filepair **)b_);
4869         const char *name_a, *name_b;
4870
4871         name_a = a->one ? a->one->path : a->two->path;
4872         name_b = b->one ? b->one->path : b->two->path;
4873         return strcmp(name_a, name_b);
4874 }
4875
4876 void diffcore_fix_diff_index(struct diff_options *options)
4877 {
4878         struct diff_queue_struct *q = &diff_queued_diff;
4879         qsort(q->queue, q->nr, sizeof(q->queue[0]), diffnamecmp);
4880 }
4881
4882 void diffcore_std(struct diff_options *options)
4883 {
4884         /* NOTE please keep the following in sync with diff_tree_combined() */
4885         if (options->skip_stat_unmatch)
4886                 diffcore_skip_stat_unmatch(options);
4887         if (!options->found_follow) {
4888                 /* See try_to_follow_renames() in tree-diff.c */
4889                 if (options->break_opt != -1)
4890                         diffcore_break(options->break_opt);
4891                 if (options->detect_rename)
4892                         diffcore_rename(options);
4893                 if (options->break_opt != -1)
4894                         diffcore_merge_broken();
4895         }
4896         if (options->pickaxe)
4897                 diffcore_pickaxe(options);
4898         if (options->orderfile)
4899                 diffcore_order(options->orderfile);
4900         if (!options->found_follow)
4901                 /* See try_to_follow_renames() in tree-diff.c */
4902                 diff_resolve_rename_copy();
4903         diffcore_apply_filter(options);
4904
4905         if (diff_queued_diff.nr && !DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
4906                 DIFF_OPT_SET(options, HAS_CHANGES);
4907         else
4908                 DIFF_OPT_CLR(options, HAS_CHANGES);
4909
4910         options->found_follow = 0;
4911 }
4912
4913 int diff_result_code(struct diff_options *opt, int status)
4914 {
4915         int result = 0;
4916
4917         diff_warn_rename_limit("diff.renameLimit",
4918                                opt->needed_rename_limit,
4919                                opt->degraded_cc_to_c);
4920         if (!DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
4921             !(opt->output_format & DIFF_FORMAT_CHECKDIFF))
4922                 return status;
4923         if (DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
4924             DIFF_OPT_TST(opt, HAS_CHANGES))
4925                 result |= 01;
4926         if ((opt->output_format & DIFF_FORMAT_CHECKDIFF) &&
4927             DIFF_OPT_TST(opt, CHECK_FAILED))
4928                 result |= 02;
4929         return result;
4930 }
4931
4932 int diff_can_quit_early(struct diff_options *opt)
4933 {
4934         return (DIFF_OPT_TST(opt, QUICK) &&
4935                 !opt->filter &&
4936                 DIFF_OPT_TST(opt, HAS_CHANGES));
4937 }
4938
4939 /*
4940  * Shall changes to this submodule be ignored?
4941  *
4942  * Submodule changes can be configured to be ignored separately for each path,
4943  * but that configuration can be overridden from the command line.
4944  */
4945 static int is_submodule_ignored(const char *path, struct diff_options *options)
4946 {
4947         int ignored = 0;
4948         unsigned orig_flags = options->flags;
4949         if (!DIFF_OPT_TST(options, OVERRIDE_SUBMODULE_CONFIG))
4950                 set_diffopt_flags_from_submodule_config(options, path);
4951         if (DIFF_OPT_TST(options, IGNORE_SUBMODULES))
4952                 ignored = 1;
4953         options->flags = orig_flags;
4954         return ignored;
4955 }
4956
4957 void diff_addremove(struct diff_options *options,
4958                     int addremove, unsigned mode,
4959                     const unsigned char *sha1,
4960                     int sha1_valid,
4961                     const char *concatpath, unsigned dirty_submodule)
4962 {
4963         struct diff_filespec *one, *two;
4964
4965         if (S_ISGITLINK(mode) && is_submodule_ignored(concatpath, options))
4966                 return;
4967
4968         /* This may look odd, but it is a preparation for
4969          * feeding "there are unchanged files which should
4970          * not produce diffs, but when you are doing copy
4971          * detection you would need them, so here they are"
4972          * entries to the diff-core.  They will be prefixed
4973          * with something like '=' or '*' (I haven't decided
4974          * which but should not make any difference).
4975          * Feeding the same new and old to diff_change()
4976          * also has the same effect.
4977          * Before the final output happens, they are pruned after
4978          * merged into rename/copy pairs as appropriate.
4979          */
4980         if (DIFF_OPT_TST(options, REVERSE_DIFF))
4981                 addremove = (addremove == '+' ? '-' :
4982                              addremove == '-' ? '+' : addremove);
4983
4984         if (options->prefix &&
4985             strncmp(concatpath, options->prefix, options->prefix_length))
4986                 return;
4987
4988         one = alloc_filespec(concatpath);
4989         two = alloc_filespec(concatpath);
4990
4991         if (addremove != '+')
4992                 fill_filespec(one, sha1, sha1_valid, mode);
4993         if (addremove != '-') {
4994                 fill_filespec(two, sha1, sha1_valid, mode);
4995                 two->dirty_submodule = dirty_submodule;
4996         }
4997
4998         diff_queue(&diff_queued_diff, one, two);
4999         if (!DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
5000                 DIFF_OPT_SET(options, HAS_CHANGES);
5001 }
5002
5003 void diff_change(struct diff_options *options,
5004                  unsigned old_mode, unsigned new_mode,
5005                  const unsigned char *old_sha1,
5006                  const unsigned char *new_sha1,
5007                  int old_sha1_valid, int new_sha1_valid,
5008                  const char *concatpath,
5009                  unsigned old_dirty_submodule, unsigned new_dirty_submodule)
5010 {
5011         struct diff_filespec *one, *two;
5012         struct diff_filepair *p;
5013
5014         if (S_ISGITLINK(old_mode) && S_ISGITLINK(new_mode) &&
5015             is_submodule_ignored(concatpath, options))
5016                 return;
5017
5018         if (DIFF_OPT_TST(options, REVERSE_DIFF)) {
5019                 unsigned tmp;
5020                 const unsigned char *tmp_c;
5021                 tmp = old_mode; old_mode = new_mode; new_mode = tmp;
5022                 tmp_c = old_sha1; old_sha1 = new_sha1; new_sha1 = tmp_c;
5023                 tmp = old_sha1_valid; old_sha1_valid = new_sha1_valid;
5024                         new_sha1_valid = tmp;
5025                 tmp = old_dirty_submodule; old_dirty_submodule = new_dirty_submodule;
5026                         new_dirty_submodule = tmp;
5027         }
5028
5029         if (options->prefix &&
5030             strncmp(concatpath, options->prefix, options->prefix_length))
5031                 return;
5032
5033         one = alloc_filespec(concatpath);
5034         two = alloc_filespec(concatpath);
5035         fill_filespec(one, old_sha1, old_sha1_valid, old_mode);
5036         fill_filespec(two, new_sha1, new_sha1_valid, new_mode);
5037         one->dirty_submodule = old_dirty_submodule;
5038         two->dirty_submodule = new_dirty_submodule;
5039         p = diff_queue(&diff_queued_diff, one, two);
5040
5041         if (DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
5042                 return;
5043
5044         if (DIFF_OPT_TST(options, QUICK) && options->skip_stat_unmatch &&
5045             !diff_filespec_check_stat_unmatch(p))
5046                 return;
5047
5048         DIFF_OPT_SET(options, HAS_CHANGES);
5049 }
5050
5051 struct diff_filepair *diff_unmerge(struct diff_options *options, const char *path)
5052 {
5053         struct diff_filepair *pair;
5054         struct diff_filespec *one, *two;
5055
5056         if (options->prefix &&
5057             strncmp(path, options->prefix, options->prefix_length))
5058                 return NULL;
5059
5060         one = alloc_filespec(path);
5061         two = alloc_filespec(path);
5062         pair = diff_queue(&diff_queued_diff, one, two);
5063         pair->is_unmerged = 1;
5064         return pair;
5065 }
5066
5067 static char *run_textconv(const char *pgm, struct diff_filespec *spec,
5068                 size_t *outsize)
5069 {
5070         struct diff_tempfile *temp;
5071         const char *argv[3];
5072         const char **arg = argv;
5073         struct child_process child = CHILD_PROCESS_INIT;
5074         struct strbuf buf = STRBUF_INIT;
5075         int err = 0;
5076
5077         temp = prepare_temp_file(spec->path, spec);
5078         *arg++ = pgm;
5079         *arg++ = temp->name;
5080         *arg = NULL;
5081
5082         child.use_shell = 1;
5083         child.argv = argv;
5084         child.out = -1;
5085         if (start_command(&child)) {
5086                 remove_tempfile();
5087                 return NULL;
5088         }
5089
5090         if (strbuf_read(&buf, child.out, 0) < 0)
5091                 err = error("error reading from textconv command '%s'", pgm);
5092         close(child.out);
5093
5094         if (finish_command(&child) || err) {
5095                 strbuf_release(&buf);
5096                 remove_tempfile();
5097                 return NULL;
5098         }
5099         remove_tempfile();
5100
5101         return strbuf_detach(&buf, outsize);
5102 }
5103
5104 size_t fill_textconv(struct userdiff_driver *driver,
5105                      struct diff_filespec *df,
5106                      char **outbuf)
5107 {
5108         size_t size;
5109
5110         if (!driver) {
5111                 if (!DIFF_FILE_VALID(df)) {
5112                         *outbuf = "";
5113                         return 0;
5114                 }
5115                 if (diff_populate_filespec(df, 0))
5116                         die("unable to read files to diff");
5117                 *outbuf = df->data;
5118                 return df->size;
5119         }
5120
5121         if (!driver->textconv)
5122                 die("BUG: fill_textconv called with non-textconv driver");
5123
5124         if (driver->textconv_cache && df->oid_valid) {
5125                 *outbuf = notes_cache_get(driver->textconv_cache,
5126                                           df->oid.hash,
5127                                           &size);
5128                 if (*outbuf)
5129                         return size;
5130         }
5131
5132         *outbuf = run_textconv(driver->textconv, df, &size);
5133         if (!*outbuf)
5134                 die("unable to read files to diff");
5135
5136         if (driver->textconv_cache && df->oid_valid) {
5137                 /* ignore errors, as we might be in a readonly repository */
5138                 notes_cache_put(driver->textconv_cache, df->oid.hash, *outbuf,
5139                                 size);
5140                 /*
5141                  * we could save up changes and flush them all at the end,
5142                  * but we would need an extra call after all diffing is done.
5143                  * Since generating a cache entry is the slow path anyway,
5144                  * this extra overhead probably isn't a big deal.
5145                  */
5146                 notes_cache_write(driver->textconv_cache);
5147         }
5148
5149         return size;
5150 }
5151
5152 void setup_diff_pager(struct diff_options *opt)
5153 {
5154         /*
5155          * If the user asked for our exit code, then either they want --quiet
5156          * or --exit-code. We should definitely not bother with a pager in the
5157          * former case, as we will generate no output. Since we still properly
5158          * report our exit code even when a pager is run, we _could_ run a
5159          * pager with --exit-code. But since we have not done so historically,
5160          * and because it is easy to find people oneline advising "git diff
5161          * --exit-code" in hooks and other scripts, we do not do so.
5162          */
5163         if (!DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
5164             check_pager_config("diff") != 0)
5165                 setup_pager();
5166 }