Merge branch 'dt/mailinfo'
[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 #include "graph.h"
22
23 #ifdef NO_FAST_WORKING_DIRECTORY
24 #define FAST_WORKING_DIRECTORY 0
25 #else
26 #define FAST_WORKING_DIRECTORY 1
27 #endif
28
29 static int diff_detect_rename_default;
30 static int diff_indent_heuristic; /* experimental */
31 static int diff_compaction_heuristic; /* experimental */
32 static int diff_rename_limit_default = 400;
33 static int diff_suppress_blank_empty;
34 static int diff_use_color_default = -1;
35 static int diff_context_default = 3;
36 static const char *diff_word_regex_cfg;
37 static const char *external_diff_cmd_cfg;
38 static const char *diff_order_file_cfg;
39 int diff_auto_refresh_index = 1;
40 static int diff_mnemonic_prefix;
41 static int diff_no_prefix;
42 static int diff_stat_graph_width;
43 static int diff_dirstat_permille_default = 30;
44 static struct diff_options default_diff_options;
45 static long diff_algorithm;
46
47 static char diff_colors[][COLOR_MAXLEN] = {
48         GIT_COLOR_RESET,
49         GIT_COLOR_NORMAL,       /* CONTEXT */
50         GIT_COLOR_BOLD,         /* METAINFO */
51         GIT_COLOR_CYAN,         /* FRAGINFO */
52         GIT_COLOR_RED,          /* OLD */
53         GIT_COLOR_GREEN,        /* NEW */
54         GIT_COLOR_YELLOW,       /* COMMIT */
55         GIT_COLOR_BG_RED,       /* WHITESPACE */
56         GIT_COLOR_NORMAL,       /* FUNCINFO */
57 };
58
59 static NORETURN void die_want_option(const char *option_name)
60 {
61         die(_("option '%s' requires a value"), option_name);
62 }
63
64 static int parse_diff_color_slot(const char *var)
65 {
66         if (!strcasecmp(var, "context") || !strcasecmp(var, "plain"))
67                 return DIFF_CONTEXT;
68         if (!strcasecmp(var, "meta"))
69                 return DIFF_METAINFO;
70         if (!strcasecmp(var, "frag"))
71                 return DIFF_FRAGINFO;
72         if (!strcasecmp(var, "old"))
73                 return DIFF_FILE_OLD;
74         if (!strcasecmp(var, "new"))
75                 return DIFF_FILE_NEW;
76         if (!strcasecmp(var, "commit"))
77                 return DIFF_COMMIT;
78         if (!strcasecmp(var, "whitespace"))
79                 return DIFF_WHITESPACE;
80         if (!strcasecmp(var, "func"))
81                 return DIFF_FUNCINFO;
82         return -1;
83 }
84
85 static int parse_dirstat_params(struct diff_options *options, const char *params_string,
86                                 struct strbuf *errmsg)
87 {
88         char *params_copy = xstrdup(params_string);
89         struct string_list params = STRING_LIST_INIT_NODUP;
90         int ret = 0;
91         int i;
92
93         if (*params_copy)
94                 string_list_split_in_place(&params, params_copy, ',', -1);
95         for (i = 0; i < params.nr; i++) {
96                 const char *p = params.items[i].string;
97                 if (!strcmp(p, "changes")) {
98                         DIFF_OPT_CLR(options, DIRSTAT_BY_LINE);
99                         DIFF_OPT_CLR(options, DIRSTAT_BY_FILE);
100                 } else if (!strcmp(p, "lines")) {
101                         DIFF_OPT_SET(options, DIRSTAT_BY_LINE);
102                         DIFF_OPT_CLR(options, DIRSTAT_BY_FILE);
103                 } else if (!strcmp(p, "files")) {
104                         DIFF_OPT_CLR(options, DIRSTAT_BY_LINE);
105                         DIFF_OPT_SET(options, DIRSTAT_BY_FILE);
106                 } else if (!strcmp(p, "noncumulative")) {
107                         DIFF_OPT_CLR(options, DIRSTAT_CUMULATIVE);
108                 } else if (!strcmp(p, "cumulative")) {
109                         DIFF_OPT_SET(options, DIRSTAT_CUMULATIVE);
110                 } else if (isdigit(*p)) {
111                         char *end;
112                         int permille = strtoul(p, &end, 10) * 10;
113                         if (*end == '.' && isdigit(*++end)) {
114                                 /* only use first digit */
115                                 permille += *end - '0';
116                                 /* .. and ignore any further digits */
117                                 while (isdigit(*++end))
118                                         ; /* nothing */
119                         }
120                         if (!*end)
121                                 options->dirstat_permille = permille;
122                         else {
123                                 strbuf_addf(errmsg, _("  Failed to parse dirstat cut-off percentage '%s'\n"),
124                                             p);
125                                 ret++;
126                         }
127                 } else {
128                         strbuf_addf(errmsg, _("  Unknown dirstat parameter '%s'\n"), p);
129                         ret++;
130                 }
131
132         }
133         string_list_clear(&params, 0);
134         free(params_copy);
135         return ret;
136 }
137
138 static int parse_submodule_params(struct diff_options *options, const char *value)
139 {
140         if (!strcmp(value, "log"))
141                 options->submodule_format = DIFF_SUBMODULE_LOG;
142         else if (!strcmp(value, "short"))
143                 options->submodule_format = DIFF_SUBMODULE_SHORT;
144         else if (!strcmp(value, "diff"))
145                 options->submodule_format = DIFF_SUBMODULE_INLINE_DIFF;
146         else
147                 return -1;
148         return 0;
149 }
150
151 static int git_config_rename(const char *var, const char *value)
152 {
153         if (!value)
154                 return DIFF_DETECT_RENAME;
155         if (!strcasecmp(value, "copies") || !strcasecmp(value, "copy"))
156                 return  DIFF_DETECT_COPY;
157         return git_config_bool(var,value) ? DIFF_DETECT_RENAME : 0;
158 }
159
160 long parse_algorithm_value(const char *value)
161 {
162         if (!value)
163                 return -1;
164         else if (!strcasecmp(value, "myers") || !strcasecmp(value, "default"))
165                 return 0;
166         else if (!strcasecmp(value, "minimal"))
167                 return XDF_NEED_MINIMAL;
168         else if (!strcasecmp(value, "patience"))
169                 return XDF_PATIENCE_DIFF;
170         else if (!strcasecmp(value, "histogram"))
171                 return XDF_HISTOGRAM_DIFF;
172         return -1;
173 }
174
175 /*
176  * These are to give UI layer defaults.
177  * The core-level commands such as git-diff-files should
178  * never be affected by the setting of diff.renames
179  * the user happens to have in the configuration file.
180  */
181 void init_diff_ui_defaults(void)
182 {
183         diff_detect_rename_default = 1;
184 }
185
186 int git_diff_heuristic_config(const char *var, const char *value, void *cb)
187 {
188         if (!strcmp(var, "diff.indentheuristic")) {
189                 diff_indent_heuristic = git_config_bool(var, value);
190                 if (diff_indent_heuristic)
191                         diff_compaction_heuristic = 0;
192         }
193         if (!strcmp(var, "diff.compactionheuristic")) {
194                 diff_compaction_heuristic = git_config_bool(var, value);
195                 if (diff_compaction_heuristic)
196                         diff_indent_heuristic = 0;
197         }
198         return 0;
199 }
200
201 int git_diff_ui_config(const char *var, const char *value, void *cb)
202 {
203         if (!strcmp(var, "diff.color") || !strcmp(var, "color.diff")) {
204                 diff_use_color_default = git_config_colorbool(var, value);
205                 return 0;
206         }
207         if (!strcmp(var, "diff.context")) {
208                 diff_context_default = git_config_int(var, value);
209                 if (diff_context_default < 0)
210                         return -1;
211                 return 0;
212         }
213         if (!strcmp(var, "diff.renames")) {
214                 diff_detect_rename_default = git_config_rename(var, value);
215                 return 0;
216         }
217         if (!strcmp(var, "diff.autorefreshindex")) {
218                 diff_auto_refresh_index = git_config_bool(var, value);
219                 return 0;
220         }
221         if (!strcmp(var, "diff.mnemonicprefix")) {
222                 diff_mnemonic_prefix = git_config_bool(var, value);
223                 return 0;
224         }
225         if (!strcmp(var, "diff.noprefix")) {
226                 diff_no_prefix = git_config_bool(var, value);
227                 return 0;
228         }
229         if (!strcmp(var, "diff.statgraphwidth")) {
230                 diff_stat_graph_width = git_config_int(var, value);
231                 return 0;
232         }
233         if (!strcmp(var, "diff.external"))
234                 return git_config_string(&external_diff_cmd_cfg, var, value);
235         if (!strcmp(var, "diff.wordregex"))
236                 return git_config_string(&diff_word_regex_cfg, var, value);
237         if (!strcmp(var, "diff.orderfile"))
238                 return git_config_pathname(&diff_order_file_cfg, var, value);
239
240         if (!strcmp(var, "diff.ignoresubmodules"))
241                 handle_ignore_submodules_arg(&default_diff_options, value);
242
243         if (!strcmp(var, "diff.submodule")) {
244                 if (parse_submodule_params(&default_diff_options, value))
245                         warning(_("Unknown value for 'diff.submodule' config variable: '%s'"),
246                                 value);
247                 return 0;
248         }
249
250         if (!strcmp(var, "diff.algorithm")) {
251                 diff_algorithm = parse_algorithm_value(value);
252                 if (diff_algorithm < 0)
253                         return -1;
254                 return 0;
255         }
256
257         if (git_diff_heuristic_config(var, value, cb) < 0)
258                 return -1;
259         if (git_color_config(var, value, cb) < 0)
260                 return -1;
261
262         return git_diff_basic_config(var, value, cb);
263 }
264
265 int git_diff_basic_config(const char *var, const char *value, void *cb)
266 {
267         const char *name;
268
269         if (!strcmp(var, "diff.renamelimit")) {
270                 diff_rename_limit_default = git_config_int(var, value);
271                 return 0;
272         }
273
274         if (userdiff_config(var, value) < 0)
275                 return -1;
276
277         if (skip_prefix(var, "diff.color.", &name) ||
278             skip_prefix(var, "color.diff.", &name)) {
279                 int slot = parse_diff_color_slot(name);
280                 if (slot < 0)
281                         return 0;
282                 if (!value)
283                         return config_error_nonbool(var);
284                 return color_parse(value, diff_colors[slot]);
285         }
286
287         /* like GNU diff's --suppress-blank-empty option  */
288         if (!strcmp(var, "diff.suppressblankempty") ||
289                         /* for backwards compatibility */
290                         !strcmp(var, "diff.suppress-blank-empty")) {
291                 diff_suppress_blank_empty = git_config_bool(var, value);
292                 return 0;
293         }
294
295         if (!strcmp(var, "diff.dirstat")) {
296                 struct strbuf errmsg = STRBUF_INIT;
297                 default_diff_options.dirstat_permille = diff_dirstat_permille_default;
298                 if (parse_dirstat_params(&default_diff_options, value, &errmsg))
299                         warning(_("Found errors in 'diff.dirstat' config variable:\n%s"),
300                                 errmsg.buf);
301                 strbuf_release(&errmsg);
302                 diff_dirstat_permille_default = default_diff_options.dirstat_permille;
303                 return 0;
304         }
305
306         if (starts_with(var, "submodule."))
307                 return parse_submodule_config_option(var, value);
308
309         return git_default_config(var, value, cb);
310 }
311
312 static char *quote_two(const char *one, const char *two)
313 {
314         int need_one = quote_c_style(one, NULL, NULL, 1);
315         int need_two = quote_c_style(two, NULL, NULL, 1);
316         struct strbuf res = STRBUF_INIT;
317
318         if (need_one + need_two) {
319                 strbuf_addch(&res, '"');
320                 quote_c_style(one, &res, NULL, 1);
321                 quote_c_style(two, &res, NULL, 1);
322                 strbuf_addch(&res, '"');
323         } else {
324                 strbuf_addstr(&res, one);
325                 strbuf_addstr(&res, two);
326         }
327         return strbuf_detach(&res, NULL);
328 }
329
330 static const char *external_diff(void)
331 {
332         static const char *external_diff_cmd = NULL;
333         static int done_preparing = 0;
334
335         if (done_preparing)
336                 return external_diff_cmd;
337         external_diff_cmd = getenv("GIT_EXTERNAL_DIFF");
338         if (!external_diff_cmd)
339                 external_diff_cmd = external_diff_cmd_cfg;
340         done_preparing = 1;
341         return external_diff_cmd;
342 }
343
344 /*
345  * Keep track of files used for diffing. Sometimes such an entry
346  * refers to a temporary file, sometimes to an existing file, and
347  * sometimes to "/dev/null".
348  */
349 static struct diff_tempfile {
350         /*
351          * filename external diff should read from, or NULL if this
352          * entry is currently not in use:
353          */
354         const char *name;
355
356         char hex[GIT_SHA1_HEXSZ + 1];
357         char mode[10];
358
359         /*
360          * If this diff_tempfile instance refers to a temporary file,
361          * this tempfile object is used to manage its lifetime.
362          */
363         struct tempfile tempfile;
364 } diff_temp[2];
365
366 typedef unsigned long (*sane_truncate_fn)(char *line, unsigned long len);
367
368 struct emit_callback {
369         int color_diff;
370         unsigned ws_rule;
371         int blank_at_eof_in_preimage;
372         int blank_at_eof_in_postimage;
373         int lno_in_preimage;
374         int lno_in_postimage;
375         sane_truncate_fn truncate;
376         const char **label_path;
377         struct diff_words_data *diff_words;
378         struct diff_options *opt;
379         struct strbuf *header;
380 };
381
382 static int count_lines(const char *data, int size)
383 {
384         int count, ch, completely_empty = 1, nl_just_seen = 0;
385         count = 0;
386         while (0 < size--) {
387                 ch = *data++;
388                 if (ch == '\n') {
389                         count++;
390                         nl_just_seen = 1;
391                         completely_empty = 0;
392                 }
393                 else {
394                         nl_just_seen = 0;
395                         completely_empty = 0;
396                 }
397         }
398         if (completely_empty)
399                 return 0;
400         if (!nl_just_seen)
401                 count++; /* no trailing newline */
402         return count;
403 }
404
405 static int fill_mmfile(mmfile_t *mf, struct diff_filespec *one)
406 {
407         if (!DIFF_FILE_VALID(one)) {
408                 mf->ptr = (char *)""; /* does not matter */
409                 mf->size = 0;
410                 return 0;
411         }
412         else if (diff_populate_filespec(one, 0))
413                 return -1;
414
415         mf->ptr = one->data;
416         mf->size = one->size;
417         return 0;
418 }
419
420 /* like fill_mmfile, but only for size, so we can avoid retrieving blob */
421 static unsigned long diff_filespec_size(struct diff_filespec *one)
422 {
423         if (!DIFF_FILE_VALID(one))
424                 return 0;
425         diff_populate_filespec(one, CHECK_SIZE_ONLY);
426         return one->size;
427 }
428
429 static int count_trailing_blank(mmfile_t *mf, unsigned ws_rule)
430 {
431         char *ptr = mf->ptr;
432         long size = mf->size;
433         int cnt = 0;
434
435         if (!size)
436                 return cnt;
437         ptr += size - 1; /* pointing at the very end */
438         if (*ptr != '\n')
439                 ; /* incomplete line */
440         else
441                 ptr--; /* skip the last LF */
442         while (mf->ptr < ptr) {
443                 char *prev_eol;
444                 for (prev_eol = ptr; mf->ptr <= prev_eol; prev_eol--)
445                         if (*prev_eol == '\n')
446                                 break;
447                 if (!ws_blank_line(prev_eol + 1, ptr - prev_eol, ws_rule))
448                         break;
449                 cnt++;
450                 ptr = prev_eol - 1;
451         }
452         return cnt;
453 }
454
455 static void check_blank_at_eof(mmfile_t *mf1, mmfile_t *mf2,
456                                struct emit_callback *ecbdata)
457 {
458         int l1, l2, at;
459         unsigned ws_rule = ecbdata->ws_rule;
460         l1 = count_trailing_blank(mf1, ws_rule);
461         l2 = count_trailing_blank(mf2, ws_rule);
462         if (l2 <= l1) {
463                 ecbdata->blank_at_eof_in_preimage = 0;
464                 ecbdata->blank_at_eof_in_postimage = 0;
465                 return;
466         }
467         at = count_lines(mf1->ptr, mf1->size);
468         ecbdata->blank_at_eof_in_preimage = (at - l1) + 1;
469
470         at = count_lines(mf2->ptr, mf2->size);
471         ecbdata->blank_at_eof_in_postimage = (at - l2) + 1;
472 }
473
474 static void emit_line_0(struct diff_options *o, const char *set, const char *reset,
475                         int first, const char *line, int len)
476 {
477         int has_trailing_newline, has_trailing_carriage_return;
478         int nofirst;
479         FILE *file = o->file;
480
481         fputs(diff_line_prefix(o), file);
482
483         if (len == 0) {
484                 has_trailing_newline = (first == '\n');
485                 has_trailing_carriage_return = (!has_trailing_newline &&
486                                                 (first == '\r'));
487                 nofirst = has_trailing_newline || has_trailing_carriage_return;
488         } else {
489                 has_trailing_newline = (len > 0 && line[len-1] == '\n');
490                 if (has_trailing_newline)
491                         len--;
492                 has_trailing_carriage_return = (len > 0 && line[len-1] == '\r');
493                 if (has_trailing_carriage_return)
494                         len--;
495                 nofirst = 0;
496         }
497
498         if (len || !nofirst) {
499                 fputs(set, file);
500                 if (!nofirst)
501                         fputc(first, file);
502                 fwrite(line, len, 1, file);
503                 fputs(reset, file);
504         }
505         if (has_trailing_carriage_return)
506                 fputc('\r', file);
507         if (has_trailing_newline)
508                 fputc('\n', file);
509 }
510
511 static void emit_line(struct diff_options *o, const char *set, const char *reset,
512                       const char *line, int len)
513 {
514         emit_line_0(o, set, reset, line[0], line+1, len-1);
515 }
516
517 static int new_blank_line_at_eof(struct emit_callback *ecbdata, const char *line, int len)
518 {
519         if (!((ecbdata->ws_rule & WS_BLANK_AT_EOF) &&
520               ecbdata->blank_at_eof_in_preimage &&
521               ecbdata->blank_at_eof_in_postimage &&
522               ecbdata->blank_at_eof_in_preimage <= ecbdata->lno_in_preimage &&
523               ecbdata->blank_at_eof_in_postimage <= ecbdata->lno_in_postimage))
524                 return 0;
525         return ws_blank_line(line, len, ecbdata->ws_rule);
526 }
527
528 static void emit_line_checked(const char *reset,
529                               struct emit_callback *ecbdata,
530                               const char *line, int len,
531                               enum color_diff color,
532                               unsigned ws_error_highlight,
533                               char sign)
534 {
535         const char *set = diff_get_color(ecbdata->color_diff, color);
536         const char *ws = NULL;
537
538         if (ecbdata->opt->ws_error_highlight & ws_error_highlight) {
539                 ws = diff_get_color(ecbdata->color_diff, DIFF_WHITESPACE);
540                 if (!*ws)
541                         ws = NULL;
542         }
543
544         if (!ws)
545                 emit_line_0(ecbdata->opt, set, reset, sign, line, len);
546         else if (sign == '+' && new_blank_line_at_eof(ecbdata, line, len))
547                 /* Blank line at EOF - paint '+' as well */
548                 emit_line_0(ecbdata->opt, ws, reset, sign, line, len);
549         else {
550                 /* Emit just the prefix, then the rest. */
551                 emit_line_0(ecbdata->opt, set, reset, sign, "", 0);
552                 ws_check_emit(line, len, ecbdata->ws_rule,
553                               ecbdata->opt->file, set, reset, ws);
554         }
555 }
556
557 static void emit_add_line(const char *reset,
558                           struct emit_callback *ecbdata,
559                           const char *line, int len)
560 {
561         emit_line_checked(reset, ecbdata, line, len,
562                           DIFF_FILE_NEW, WSEH_NEW, '+');
563 }
564
565 static void emit_del_line(const char *reset,
566                           struct emit_callback *ecbdata,
567                           const char *line, int len)
568 {
569         emit_line_checked(reset, ecbdata, line, len,
570                           DIFF_FILE_OLD, WSEH_OLD, '-');
571 }
572
573 static void emit_context_line(const char *reset,
574                               struct emit_callback *ecbdata,
575                               const char *line, int len)
576 {
577         emit_line_checked(reset, ecbdata, line, len,
578                           DIFF_CONTEXT, WSEH_CONTEXT, ' ');
579 }
580
581 static void emit_hunk_header(struct emit_callback *ecbdata,
582                              const char *line, int len)
583 {
584         const char *context = diff_get_color(ecbdata->color_diff, DIFF_CONTEXT);
585         const char *frag = diff_get_color(ecbdata->color_diff, DIFF_FRAGINFO);
586         const char *func = diff_get_color(ecbdata->color_diff, DIFF_FUNCINFO);
587         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
588         static const char atat[2] = { '@', '@' };
589         const char *cp, *ep;
590         struct strbuf msgbuf = STRBUF_INIT;
591         int org_len = len;
592         int i = 1;
593
594         /*
595          * As a hunk header must begin with "@@ -<old>, +<new> @@",
596          * it always is at least 10 bytes long.
597          */
598         if (len < 10 ||
599             memcmp(line, atat, 2) ||
600             !(ep = memmem(line + 2, len - 2, atat, 2))) {
601                 emit_line(ecbdata->opt, context, reset, line, len);
602                 return;
603         }
604         ep += 2; /* skip over @@ */
605
606         /* The hunk header in fraginfo color */
607         strbuf_addstr(&msgbuf, frag);
608         strbuf_add(&msgbuf, line, ep - line);
609         strbuf_addstr(&msgbuf, reset);
610
611         /*
612          * trailing "\r\n"
613          */
614         for ( ; i < 3; i++)
615                 if (line[len - i] == '\r' || line[len - i] == '\n')
616                         len--;
617
618         /* blank before the func header */
619         for (cp = ep; ep - line < len; ep++)
620                 if (*ep != ' ' && *ep != '\t')
621                         break;
622         if (ep != cp) {
623                 strbuf_addstr(&msgbuf, context);
624                 strbuf_add(&msgbuf, cp, ep - cp);
625                 strbuf_addstr(&msgbuf, reset);
626         }
627
628         if (ep < line + len) {
629                 strbuf_addstr(&msgbuf, func);
630                 strbuf_add(&msgbuf, ep, line + len - ep);
631                 strbuf_addstr(&msgbuf, reset);
632         }
633
634         strbuf_add(&msgbuf, line + len, org_len - len);
635         emit_line(ecbdata->opt, "", "", msgbuf.buf, msgbuf.len);
636         strbuf_release(&msgbuf);
637 }
638
639 static struct diff_tempfile *claim_diff_tempfile(void) {
640         int i;
641         for (i = 0; i < ARRAY_SIZE(diff_temp); i++)
642                 if (!diff_temp[i].name)
643                         return diff_temp + i;
644         die("BUG: diff is failing to clean up its tempfiles");
645 }
646
647 static void remove_tempfile(void)
648 {
649         int i;
650         for (i = 0; i < ARRAY_SIZE(diff_temp); i++) {
651                 if (is_tempfile_active(&diff_temp[i].tempfile))
652                         delete_tempfile(&diff_temp[i].tempfile);
653                 diff_temp[i].name = NULL;
654         }
655 }
656
657 static void print_line_count(FILE *file, int count)
658 {
659         switch (count) {
660         case 0:
661                 fprintf(file, "0,0");
662                 break;
663         case 1:
664                 fprintf(file, "1");
665                 break;
666         default:
667                 fprintf(file, "1,%d", count);
668                 break;
669         }
670 }
671
672 static void emit_rewrite_lines(struct emit_callback *ecb,
673                                int prefix, const char *data, int size)
674 {
675         const char *endp = NULL;
676         static const char *nneof = " No newline at end of file\n";
677         const char *reset = diff_get_color(ecb->color_diff, DIFF_RESET);
678
679         while (0 < size) {
680                 int len;
681
682                 endp = memchr(data, '\n', size);
683                 len = endp ? (endp - data + 1) : size;
684                 if (prefix != '+') {
685                         ecb->lno_in_preimage++;
686                         emit_del_line(reset, ecb, data, len);
687                 } else {
688                         ecb->lno_in_postimage++;
689                         emit_add_line(reset, ecb, data, len);
690                 }
691                 size -= len;
692                 data += len;
693         }
694         if (!endp) {
695                 const char *context = diff_get_color(ecb->color_diff,
696                                                      DIFF_CONTEXT);
697                 putc('\n', ecb->opt->file);
698                 emit_line_0(ecb->opt, context, reset, '\\',
699                             nneof, strlen(nneof));
700         }
701 }
702
703 static void emit_rewrite_diff(const char *name_a,
704                               const char *name_b,
705                               struct diff_filespec *one,
706                               struct diff_filespec *two,
707                               struct userdiff_driver *textconv_one,
708                               struct userdiff_driver *textconv_two,
709                               struct diff_options *o)
710 {
711         int lc_a, lc_b;
712         const char *name_a_tab, *name_b_tab;
713         const char *metainfo = diff_get_color(o->use_color, DIFF_METAINFO);
714         const char *fraginfo = diff_get_color(o->use_color, DIFF_FRAGINFO);
715         const char *reset = diff_get_color(o->use_color, DIFF_RESET);
716         static struct strbuf a_name = STRBUF_INIT, b_name = STRBUF_INIT;
717         const char *a_prefix, *b_prefix;
718         char *data_one, *data_two;
719         size_t size_one, size_two;
720         struct emit_callback ecbdata;
721         const char *line_prefix = diff_line_prefix(o);
722
723         if (diff_mnemonic_prefix && DIFF_OPT_TST(o, REVERSE_DIFF)) {
724                 a_prefix = o->b_prefix;
725                 b_prefix = o->a_prefix;
726         } else {
727                 a_prefix = o->a_prefix;
728                 b_prefix = o->b_prefix;
729         }
730
731         name_a += (*name_a == '/');
732         name_b += (*name_b == '/');
733         name_a_tab = strchr(name_a, ' ') ? "\t" : "";
734         name_b_tab = strchr(name_b, ' ') ? "\t" : "";
735
736         strbuf_reset(&a_name);
737         strbuf_reset(&b_name);
738         quote_two_c_style(&a_name, a_prefix, name_a, 0);
739         quote_two_c_style(&b_name, b_prefix, name_b, 0);
740
741         size_one = fill_textconv(textconv_one, one, &data_one);
742         size_two = fill_textconv(textconv_two, two, &data_two);
743
744         memset(&ecbdata, 0, sizeof(ecbdata));
745         ecbdata.color_diff = want_color(o->use_color);
746         ecbdata.ws_rule = whitespace_rule(name_b);
747         ecbdata.opt = o;
748         if (ecbdata.ws_rule & WS_BLANK_AT_EOF) {
749                 mmfile_t mf1, mf2;
750                 mf1.ptr = (char *)data_one;
751                 mf2.ptr = (char *)data_two;
752                 mf1.size = size_one;
753                 mf2.size = size_two;
754                 check_blank_at_eof(&mf1, &mf2, &ecbdata);
755         }
756         ecbdata.lno_in_preimage = 1;
757         ecbdata.lno_in_postimage = 1;
758
759         lc_a = count_lines(data_one, size_one);
760         lc_b = count_lines(data_two, size_two);
761         fprintf(o->file,
762                 "%s%s--- %s%s%s\n%s%s+++ %s%s%s\n%s%s@@ -",
763                 line_prefix, metainfo, a_name.buf, name_a_tab, reset,
764                 line_prefix, metainfo, b_name.buf, name_b_tab, reset,
765                 line_prefix, fraginfo);
766         if (!o->irreversible_delete)
767                 print_line_count(o->file, lc_a);
768         else
769                 fprintf(o->file, "?,?");
770         fprintf(o->file, " +");
771         print_line_count(o->file, lc_b);
772         fprintf(o->file, " @@%s\n", reset);
773         if (lc_a && !o->irreversible_delete)
774                 emit_rewrite_lines(&ecbdata, '-', data_one, size_one);
775         if (lc_b)
776                 emit_rewrite_lines(&ecbdata, '+', data_two, size_two);
777         if (textconv_one)
778                 free((char *)data_one);
779         if (textconv_two)
780                 free((char *)data_two);
781 }
782
783 struct diff_words_buffer {
784         mmfile_t text;
785         long alloc;
786         struct diff_words_orig {
787                 const char *begin, *end;
788         } *orig;
789         int orig_nr, orig_alloc;
790 };
791
792 static void diff_words_append(char *line, unsigned long len,
793                 struct diff_words_buffer *buffer)
794 {
795         ALLOC_GROW(buffer->text.ptr, buffer->text.size + len, buffer->alloc);
796         line++;
797         len--;
798         memcpy(buffer->text.ptr + buffer->text.size, line, len);
799         buffer->text.size += len;
800         buffer->text.ptr[buffer->text.size] = '\0';
801 }
802
803 struct diff_words_style_elem {
804         const char *prefix;
805         const char *suffix;
806         const char *color; /* NULL; filled in by the setup code if
807                             * color is enabled */
808 };
809
810 struct diff_words_style {
811         enum diff_words_type type;
812         struct diff_words_style_elem new, old, ctx;
813         const char *newline;
814 };
815
816 static struct diff_words_style diff_words_styles[] = {
817         { DIFF_WORDS_PORCELAIN, {"+", "\n"}, {"-", "\n"}, {" ", "\n"}, "~\n" },
818         { DIFF_WORDS_PLAIN, {"{+", "+}"}, {"[-", "-]"}, {"", ""}, "\n" },
819         { DIFF_WORDS_COLOR, {"", ""}, {"", ""}, {"", ""}, "\n" }
820 };
821
822 struct diff_words_data {
823         struct diff_words_buffer minus, plus;
824         const char *current_plus;
825         int last_minus;
826         struct diff_options *opt;
827         regex_t *word_regex;
828         enum diff_words_type type;
829         struct diff_words_style *style;
830 };
831
832 static int fn_out_diff_words_write_helper(FILE *fp,
833                                           struct diff_words_style_elem *st_el,
834                                           const char *newline,
835                                           size_t count, const char *buf,
836                                           const char *line_prefix)
837 {
838         int print = 0;
839
840         while (count) {
841                 char *p = memchr(buf, '\n', count);
842                 if (print)
843                         fputs(line_prefix, fp);
844                 if (p != buf) {
845                         if (st_el->color && fputs(st_el->color, fp) < 0)
846                                 return -1;
847                         if (fputs(st_el->prefix, fp) < 0 ||
848                             fwrite(buf, p ? p - buf : count, 1, fp) != 1 ||
849                             fputs(st_el->suffix, fp) < 0)
850                                 return -1;
851                         if (st_el->color && *st_el->color
852                             && fputs(GIT_COLOR_RESET, fp) < 0)
853                                 return -1;
854                 }
855                 if (!p)
856                         return 0;
857                 if (fputs(newline, fp) < 0)
858                         return -1;
859                 count -= p + 1 - buf;
860                 buf = p + 1;
861                 print = 1;
862         }
863         return 0;
864 }
865
866 /*
867  * '--color-words' algorithm can be described as:
868  *
869  *   1. collect a the minus/plus lines of a diff hunk, divided into
870  *      minus-lines and plus-lines;
871  *
872  *   2. break both minus-lines and plus-lines into words and
873  *      place them into two mmfile_t with one word for each line;
874  *
875  *   3. use xdiff to run diff on the two mmfile_t to get the words level diff;
876  *
877  * And for the common parts of the both file, we output the plus side text.
878  * diff_words->current_plus is used to trace the current position of the plus file
879  * which printed. diff_words->last_minus is used to trace the last minus word
880  * printed.
881  *
882  * For '--graph' to work with '--color-words', we need to output the graph prefix
883  * on each line of color words output. Generally, there are two conditions on
884  * which we should output the prefix.
885  *
886  *   1. diff_words->last_minus == 0 &&
887  *      diff_words->current_plus == diff_words->plus.text.ptr
888  *
889  *      that is: the plus text must start as a new line, and if there is no minus
890  *      word printed, a graph prefix must be printed.
891  *
892  *   2. diff_words->current_plus > diff_words->plus.text.ptr &&
893  *      *(diff_words->current_plus - 1) == '\n'
894  *
895  *      that is: a graph prefix must be printed following a '\n'
896  */
897 static int color_words_output_graph_prefix(struct diff_words_data *diff_words)
898 {
899         if ((diff_words->last_minus == 0 &&
900                 diff_words->current_plus == diff_words->plus.text.ptr) ||
901                 (diff_words->current_plus > diff_words->plus.text.ptr &&
902                 *(diff_words->current_plus - 1) == '\n')) {
903                 return 1;
904         } else {
905                 return 0;
906         }
907 }
908
909 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
910 {
911         struct diff_words_data *diff_words = priv;
912         struct diff_words_style *style = diff_words->style;
913         int minus_first, minus_len, plus_first, plus_len;
914         const char *minus_begin, *minus_end, *plus_begin, *plus_end;
915         struct diff_options *opt = diff_words->opt;
916         const char *line_prefix;
917
918         if (line[0] != '@' || parse_hunk_header(line, len,
919                         &minus_first, &minus_len, &plus_first, &plus_len))
920                 return;
921
922         assert(opt);
923         line_prefix = diff_line_prefix(opt);
924
925         /* POSIX requires that first be decremented by one if len == 0... */
926         if (minus_len) {
927                 minus_begin = diff_words->minus.orig[minus_first].begin;
928                 minus_end =
929                         diff_words->minus.orig[minus_first + minus_len - 1].end;
930         } else
931                 minus_begin = minus_end =
932                         diff_words->minus.orig[minus_first].end;
933
934         if (plus_len) {
935                 plus_begin = diff_words->plus.orig[plus_first].begin;
936                 plus_end = diff_words->plus.orig[plus_first + plus_len - 1].end;
937         } else
938                 plus_begin = plus_end = diff_words->plus.orig[plus_first].end;
939
940         if (color_words_output_graph_prefix(diff_words)) {
941                 fputs(line_prefix, diff_words->opt->file);
942         }
943         if (diff_words->current_plus != plus_begin) {
944                 fn_out_diff_words_write_helper(diff_words->opt->file,
945                                 &style->ctx, style->newline,
946                                 plus_begin - diff_words->current_plus,
947                                 diff_words->current_plus, line_prefix);
948                 if (*(plus_begin - 1) == '\n')
949                         fputs(line_prefix, diff_words->opt->file);
950         }
951         if (minus_begin != minus_end) {
952                 fn_out_diff_words_write_helper(diff_words->opt->file,
953                                 &style->old, style->newline,
954                                 minus_end - minus_begin, minus_begin,
955                                 line_prefix);
956         }
957         if (plus_begin != plus_end) {
958                 fn_out_diff_words_write_helper(diff_words->opt->file,
959                                 &style->new, style->newline,
960                                 plus_end - plus_begin, plus_begin,
961                                 line_prefix);
962         }
963
964         diff_words->current_plus = plus_end;
965         diff_words->last_minus = minus_first;
966 }
967
968 /* This function starts looking at *begin, and returns 0 iff a word was found. */
969 static int find_word_boundaries(mmfile_t *buffer, regex_t *word_regex,
970                 int *begin, int *end)
971 {
972         if (word_regex && *begin < buffer->size) {
973                 regmatch_t match[1];
974                 if (!regexec_buf(word_regex, buffer->ptr + *begin,
975                                  buffer->size - *begin, 1, match, 0)) {
976                         char *p = memchr(buffer->ptr + *begin + match[0].rm_so,
977                                         '\n', match[0].rm_eo - match[0].rm_so);
978                         *end = p ? p - buffer->ptr : match[0].rm_eo + *begin;
979                         *begin += match[0].rm_so;
980                         return *begin >= *end;
981                 }
982                 return -1;
983         }
984
985         /* find the next word */
986         while (*begin < buffer->size && isspace(buffer->ptr[*begin]))
987                 (*begin)++;
988         if (*begin >= buffer->size)
989                 return -1;
990
991         /* find the end of the word */
992         *end = *begin + 1;
993         while (*end < buffer->size && !isspace(buffer->ptr[*end]))
994                 (*end)++;
995
996         return 0;
997 }
998
999 /*
1000  * This function splits the words in buffer->text, stores the list with
1001  * newline separator into out, and saves the offsets of the original words
1002  * in buffer->orig.
1003  */
1004 static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out,
1005                 regex_t *word_regex)
1006 {
1007         int i, j;
1008         long alloc = 0;
1009
1010         out->size = 0;
1011         out->ptr = NULL;
1012
1013         /* fake an empty "0th" word */
1014         ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
1015         buffer->orig[0].begin = buffer->orig[0].end = buffer->text.ptr;
1016         buffer->orig_nr = 1;
1017
1018         for (i = 0; i < buffer->text.size; i++) {
1019                 if (find_word_boundaries(&buffer->text, word_regex, &i, &j))
1020                         return;
1021
1022                 /* store original boundaries */
1023                 ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
1024                                 buffer->orig_alloc);
1025                 buffer->orig[buffer->orig_nr].begin = buffer->text.ptr + i;
1026                 buffer->orig[buffer->orig_nr].end = buffer->text.ptr + j;
1027                 buffer->orig_nr++;
1028
1029                 /* store one word */
1030                 ALLOC_GROW(out->ptr, out->size + j - i + 1, alloc);
1031                 memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
1032                 out->ptr[out->size + j - i] = '\n';
1033                 out->size += j - i + 1;
1034
1035                 i = j - 1;
1036         }
1037 }
1038
1039 /* this executes the word diff on the accumulated buffers */
1040 static void diff_words_show(struct diff_words_data *diff_words)
1041 {
1042         xpparam_t xpp;
1043         xdemitconf_t xecfg;
1044         mmfile_t minus, plus;
1045         struct diff_words_style *style = diff_words->style;
1046
1047         struct diff_options *opt = diff_words->opt;
1048         const char *line_prefix;
1049
1050         assert(opt);
1051         line_prefix = diff_line_prefix(opt);
1052
1053         /* special case: only removal */
1054         if (!diff_words->plus.text.size) {
1055                 fputs(line_prefix, diff_words->opt->file);
1056                 fn_out_diff_words_write_helper(diff_words->opt->file,
1057                         &style->old, style->newline,
1058                         diff_words->minus.text.size,
1059                         diff_words->minus.text.ptr, line_prefix);
1060                 diff_words->minus.text.size = 0;
1061                 return;
1062         }
1063
1064         diff_words->current_plus = diff_words->plus.text.ptr;
1065         diff_words->last_minus = 0;
1066
1067         memset(&xpp, 0, sizeof(xpp));
1068         memset(&xecfg, 0, sizeof(xecfg));
1069         diff_words_fill(&diff_words->minus, &minus, diff_words->word_regex);
1070         diff_words_fill(&diff_words->plus, &plus, diff_words->word_regex);
1071         xpp.flags = 0;
1072         /* as only the hunk header will be parsed, we need a 0-context */
1073         xecfg.ctxlen = 0;
1074         if (xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
1075                           &xpp, &xecfg))
1076                 die("unable to generate word diff");
1077         free(minus.ptr);
1078         free(plus.ptr);
1079         if (diff_words->current_plus != diff_words->plus.text.ptr +
1080                         diff_words->plus.text.size) {
1081                 if (color_words_output_graph_prefix(diff_words))
1082                         fputs(line_prefix, diff_words->opt->file);
1083                 fn_out_diff_words_write_helper(diff_words->opt->file,
1084                         &style->ctx, style->newline,
1085                         diff_words->plus.text.ptr + diff_words->plus.text.size
1086                         - diff_words->current_plus, diff_words->current_plus,
1087                         line_prefix);
1088         }
1089         diff_words->minus.text.size = diff_words->plus.text.size = 0;
1090 }
1091
1092 /* In "color-words" mode, show word-diff of words accumulated in the buffer */
1093 static void diff_words_flush(struct emit_callback *ecbdata)
1094 {
1095         if (ecbdata->diff_words->minus.text.size ||
1096             ecbdata->diff_words->plus.text.size)
1097                 diff_words_show(ecbdata->diff_words);
1098 }
1099
1100 static void diff_filespec_load_driver(struct diff_filespec *one)
1101 {
1102         /* Use already-loaded driver */
1103         if (one->driver)
1104                 return;
1105
1106         if (S_ISREG(one->mode))
1107                 one->driver = userdiff_find_by_path(one->path);
1108
1109         /* Fallback to default settings */
1110         if (!one->driver)
1111                 one->driver = userdiff_find_by_name("default");
1112 }
1113
1114 static const char *userdiff_word_regex(struct diff_filespec *one)
1115 {
1116         diff_filespec_load_driver(one);
1117         return one->driver->word_regex;
1118 }
1119
1120 static void init_diff_words_data(struct emit_callback *ecbdata,
1121                                  struct diff_options *orig_opts,
1122                                  struct diff_filespec *one,
1123                                  struct diff_filespec *two)
1124 {
1125         int i;
1126         struct diff_options *o = xmalloc(sizeof(struct diff_options));
1127         memcpy(o, orig_opts, sizeof(struct diff_options));
1128
1129         ecbdata->diff_words =
1130                 xcalloc(1, sizeof(struct diff_words_data));
1131         ecbdata->diff_words->type = o->word_diff;
1132         ecbdata->diff_words->opt = o;
1133         if (!o->word_regex)
1134                 o->word_regex = userdiff_word_regex(one);
1135         if (!o->word_regex)
1136                 o->word_regex = userdiff_word_regex(two);
1137         if (!o->word_regex)
1138                 o->word_regex = diff_word_regex_cfg;
1139         if (o->word_regex) {
1140                 ecbdata->diff_words->word_regex = (regex_t *)
1141                         xmalloc(sizeof(regex_t));
1142                 if (regcomp(ecbdata->diff_words->word_regex,
1143                             o->word_regex,
1144                             REG_EXTENDED | REG_NEWLINE))
1145                         die ("Invalid regular expression: %s",
1146                              o->word_regex);
1147         }
1148         for (i = 0; i < ARRAY_SIZE(diff_words_styles); i++) {
1149                 if (o->word_diff == diff_words_styles[i].type) {
1150                         ecbdata->diff_words->style =
1151                                 &diff_words_styles[i];
1152                         break;
1153                 }
1154         }
1155         if (want_color(o->use_color)) {
1156                 struct diff_words_style *st = ecbdata->diff_words->style;
1157                 st->old.color = diff_get_color_opt(o, DIFF_FILE_OLD);
1158                 st->new.color = diff_get_color_opt(o, DIFF_FILE_NEW);
1159                 st->ctx.color = diff_get_color_opt(o, DIFF_CONTEXT);
1160         }
1161 }
1162
1163 static void free_diff_words_data(struct emit_callback *ecbdata)
1164 {
1165         if (ecbdata->diff_words) {
1166                 diff_words_flush(ecbdata);
1167                 free (ecbdata->diff_words->opt);
1168                 free (ecbdata->diff_words->minus.text.ptr);
1169                 free (ecbdata->diff_words->minus.orig);
1170                 free (ecbdata->diff_words->plus.text.ptr);
1171                 free (ecbdata->diff_words->plus.orig);
1172                 if (ecbdata->diff_words->word_regex) {
1173                         regfree(ecbdata->diff_words->word_regex);
1174                         free(ecbdata->diff_words->word_regex);
1175                 }
1176                 free(ecbdata->diff_words);
1177                 ecbdata->diff_words = NULL;
1178         }
1179 }
1180
1181 const char *diff_get_color(int diff_use_color, enum color_diff ix)
1182 {
1183         if (want_color(diff_use_color))
1184                 return diff_colors[ix];
1185         return "";
1186 }
1187
1188 const char *diff_line_prefix(struct diff_options *opt)
1189 {
1190         struct strbuf *msgbuf;
1191         if (!opt->output_prefix)
1192                 return "";
1193
1194         msgbuf = opt->output_prefix(opt, opt->output_prefix_data);
1195         return msgbuf->buf;
1196 }
1197
1198 static unsigned long sane_truncate_line(struct emit_callback *ecb, char *line, unsigned long len)
1199 {
1200         const char *cp;
1201         unsigned long allot;
1202         size_t l = len;
1203
1204         if (ecb->truncate)
1205                 return ecb->truncate(line, len);
1206         cp = line;
1207         allot = l;
1208         while (0 < l) {
1209                 (void) utf8_width(&cp, &l);
1210                 if (!cp)
1211                         break; /* truncated in the middle? */
1212         }
1213         return allot - l;
1214 }
1215
1216 static void find_lno(const char *line, struct emit_callback *ecbdata)
1217 {
1218         const char *p;
1219         ecbdata->lno_in_preimage = 0;
1220         ecbdata->lno_in_postimage = 0;
1221         p = strchr(line, '-');
1222         if (!p)
1223                 return; /* cannot happen */
1224         ecbdata->lno_in_preimage = strtol(p + 1, NULL, 10);
1225         p = strchr(p, '+');
1226         if (!p)
1227                 return; /* cannot happen */
1228         ecbdata->lno_in_postimage = strtol(p + 1, NULL, 10);
1229 }
1230
1231 static void fn_out_consume(void *priv, char *line, unsigned long len)
1232 {
1233         struct emit_callback *ecbdata = priv;
1234         const char *meta = diff_get_color(ecbdata->color_diff, DIFF_METAINFO);
1235         const char *context = diff_get_color(ecbdata->color_diff, DIFF_CONTEXT);
1236         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
1237         struct diff_options *o = ecbdata->opt;
1238         const char *line_prefix = diff_line_prefix(o);
1239
1240         o->found_changes = 1;
1241
1242         if (ecbdata->header) {
1243                 fprintf(o->file, "%s", ecbdata->header->buf);
1244                 strbuf_reset(ecbdata->header);
1245                 ecbdata->header = NULL;
1246         }
1247
1248         if (ecbdata->label_path[0]) {
1249                 const char *name_a_tab, *name_b_tab;
1250
1251                 name_a_tab = strchr(ecbdata->label_path[0], ' ') ? "\t" : "";
1252                 name_b_tab = strchr(ecbdata->label_path[1], ' ') ? "\t" : "";
1253
1254                 fprintf(o->file, "%s%s--- %s%s%s\n",
1255                         line_prefix, meta, ecbdata->label_path[0], reset, name_a_tab);
1256                 fprintf(o->file, "%s%s+++ %s%s%s\n",
1257                         line_prefix, meta, ecbdata->label_path[1], reset, name_b_tab);
1258                 ecbdata->label_path[0] = ecbdata->label_path[1] = NULL;
1259         }
1260
1261         if (diff_suppress_blank_empty
1262             && len == 2 && line[0] == ' ' && line[1] == '\n') {
1263                 line[0] = '\n';
1264                 len = 1;
1265         }
1266
1267         if (line[0] == '@') {
1268                 if (ecbdata->diff_words)
1269                         diff_words_flush(ecbdata);
1270                 len = sane_truncate_line(ecbdata, line, len);
1271                 find_lno(line, ecbdata);
1272                 emit_hunk_header(ecbdata, line, len);
1273                 if (line[len-1] != '\n')
1274                         putc('\n', o->file);
1275                 return;
1276         }
1277
1278         if (ecbdata->diff_words) {
1279                 if (line[0] == '-') {
1280                         diff_words_append(line, len,
1281                                           &ecbdata->diff_words->minus);
1282                         return;
1283                 } else if (line[0] == '+') {
1284                         diff_words_append(line, len,
1285                                           &ecbdata->diff_words->plus);
1286                         return;
1287                 } else if (starts_with(line, "\\ ")) {
1288                         /*
1289                          * Eat the "no newline at eof" marker as if we
1290                          * saw a "+" or "-" line with nothing on it,
1291                          * and return without diff_words_flush() to
1292                          * defer processing. If this is the end of
1293                          * preimage, more "+" lines may come after it.
1294                          */
1295                         return;
1296                 }
1297                 diff_words_flush(ecbdata);
1298                 if (ecbdata->diff_words->type == DIFF_WORDS_PORCELAIN) {
1299                         emit_line(o, context, reset, line, len);
1300                         fputs("~\n", o->file);
1301                 } else {
1302                         /*
1303                          * Skip the prefix character, if any.  With
1304                          * diff_suppress_blank_empty, there may be
1305                          * none.
1306                          */
1307                         if (line[0] != '\n') {
1308                               line++;
1309                               len--;
1310                         }
1311                         emit_line(o, context, reset, line, len);
1312                 }
1313                 return;
1314         }
1315
1316         switch (line[0]) {
1317         case '+':
1318                 ecbdata->lno_in_postimage++;
1319                 emit_add_line(reset, ecbdata, line + 1, len - 1);
1320                 break;
1321         case '-':
1322                 ecbdata->lno_in_preimage++;
1323                 emit_del_line(reset, ecbdata, line + 1, len - 1);
1324                 break;
1325         case ' ':
1326                 ecbdata->lno_in_postimage++;
1327                 ecbdata->lno_in_preimage++;
1328                 emit_context_line(reset, ecbdata, line + 1, len - 1);
1329                 break;
1330         default:
1331                 /* incomplete line at the end */
1332                 ecbdata->lno_in_preimage++;
1333                 emit_line(o, diff_get_color(ecbdata->color_diff, DIFF_CONTEXT),
1334                           reset, line, len);
1335                 break;
1336         }
1337 }
1338
1339 static char *pprint_rename(const char *a, const char *b)
1340 {
1341         const char *old = a;
1342         const char *new = b;
1343         struct strbuf name = STRBUF_INIT;
1344         int pfx_length, sfx_length;
1345         int pfx_adjust_for_slash;
1346         int len_a = strlen(a);
1347         int len_b = strlen(b);
1348         int a_midlen, b_midlen;
1349         int qlen_a = quote_c_style(a, NULL, NULL, 0);
1350         int qlen_b = quote_c_style(b, NULL, NULL, 0);
1351
1352         if (qlen_a || qlen_b) {
1353                 quote_c_style(a, &name, NULL, 0);
1354                 strbuf_addstr(&name, " => ");
1355                 quote_c_style(b, &name, NULL, 0);
1356                 return strbuf_detach(&name, NULL);
1357         }
1358
1359         /* Find common prefix */
1360         pfx_length = 0;
1361         while (*old && *new && *old == *new) {
1362                 if (*old == '/')
1363                         pfx_length = old - a + 1;
1364                 old++;
1365                 new++;
1366         }
1367
1368         /* Find common suffix */
1369         old = a + len_a;
1370         new = b + len_b;
1371         sfx_length = 0;
1372         /*
1373          * If there is a common prefix, it must end in a slash.  In
1374          * that case we let this loop run 1 into the prefix to see the
1375          * same slash.
1376          *
1377          * If there is no common prefix, we cannot do this as it would
1378          * underrun the input strings.
1379          */
1380         pfx_adjust_for_slash = (pfx_length ? 1 : 0);
1381         while (a + pfx_length - pfx_adjust_for_slash <= old &&
1382                b + pfx_length - pfx_adjust_for_slash <= new &&
1383                *old == *new) {
1384                 if (*old == '/')
1385                         sfx_length = len_a - (old - a);
1386                 old--;
1387                 new--;
1388         }
1389
1390         /*
1391          * pfx{mid-a => mid-b}sfx
1392          * {pfx-a => pfx-b}sfx
1393          * pfx{sfx-a => sfx-b}
1394          * name-a => name-b
1395          */
1396         a_midlen = len_a - pfx_length - sfx_length;
1397         b_midlen = len_b - pfx_length - sfx_length;
1398         if (a_midlen < 0)
1399                 a_midlen = 0;
1400         if (b_midlen < 0)
1401                 b_midlen = 0;
1402
1403         strbuf_grow(&name, pfx_length + a_midlen + b_midlen + sfx_length + 7);
1404         if (pfx_length + sfx_length) {
1405                 strbuf_add(&name, a, pfx_length);
1406                 strbuf_addch(&name, '{');
1407         }
1408         strbuf_add(&name, a + pfx_length, a_midlen);
1409         strbuf_addstr(&name, " => ");
1410         strbuf_add(&name, b + pfx_length, b_midlen);
1411         if (pfx_length + sfx_length) {
1412                 strbuf_addch(&name, '}');
1413                 strbuf_add(&name, a + len_a - sfx_length, sfx_length);
1414         }
1415         return strbuf_detach(&name, NULL);
1416 }
1417
1418 struct diffstat_t {
1419         int nr;
1420         int alloc;
1421         struct diffstat_file {
1422                 char *from_name;
1423                 char *name;
1424                 char *print_name;
1425                 unsigned is_unmerged:1;
1426                 unsigned is_binary:1;
1427                 unsigned is_renamed:1;
1428                 unsigned is_interesting:1;
1429                 uintmax_t added, deleted;
1430         } **files;
1431 };
1432
1433 static struct diffstat_file *diffstat_add(struct diffstat_t *diffstat,
1434                                           const char *name_a,
1435                                           const char *name_b)
1436 {
1437         struct diffstat_file *x;
1438         x = xcalloc(1, sizeof(*x));
1439         ALLOC_GROW(diffstat->files, diffstat->nr + 1, diffstat->alloc);
1440         diffstat->files[diffstat->nr++] = x;
1441         if (name_b) {
1442                 x->from_name = xstrdup(name_a);
1443                 x->name = xstrdup(name_b);
1444                 x->is_renamed = 1;
1445         }
1446         else {
1447                 x->from_name = NULL;
1448                 x->name = xstrdup(name_a);
1449         }
1450         return x;
1451 }
1452
1453 static void diffstat_consume(void *priv, char *line, unsigned long len)
1454 {
1455         struct diffstat_t *diffstat = priv;
1456         struct diffstat_file *x = diffstat->files[diffstat->nr - 1];
1457
1458         if (line[0] == '+')
1459                 x->added++;
1460         else if (line[0] == '-')
1461                 x->deleted++;
1462 }
1463
1464 const char mime_boundary_leader[] = "------------";
1465
1466 static int scale_linear(int it, int width, int max_change)
1467 {
1468         if (!it)
1469                 return 0;
1470         /*
1471          * make sure that at least one '-' or '+' is printed if
1472          * there is any change to this path. The easiest way is to
1473          * scale linearly as if the alloted width is one column shorter
1474          * than it is, and then add 1 to the result.
1475          */
1476         return 1 + (it * (width - 1) / max_change);
1477 }
1478
1479 static void show_name(FILE *file,
1480                       const char *prefix, const char *name, int len)
1481 {
1482         fprintf(file, " %s%-*s |", prefix, len, name);
1483 }
1484
1485 static void show_graph(FILE *file, char ch, int cnt, const char *set, const char *reset)
1486 {
1487         if (cnt <= 0)
1488                 return;
1489         fprintf(file, "%s", set);
1490         while (cnt--)
1491                 putc(ch, file);
1492         fprintf(file, "%s", reset);
1493 }
1494
1495 static void fill_print_name(struct diffstat_file *file)
1496 {
1497         char *pname;
1498
1499         if (file->print_name)
1500                 return;
1501
1502         if (!file->is_renamed) {
1503                 struct strbuf buf = STRBUF_INIT;
1504                 if (quote_c_style(file->name, &buf, NULL, 0)) {
1505                         pname = strbuf_detach(&buf, NULL);
1506                 } else {
1507                         pname = file->name;
1508                         strbuf_release(&buf);
1509                 }
1510         } else {
1511                 pname = pprint_rename(file->from_name, file->name);
1512         }
1513         file->print_name = pname;
1514 }
1515
1516 int print_stat_summary(FILE *fp, int files, int insertions, int deletions)
1517 {
1518         struct strbuf sb = STRBUF_INIT;
1519         int ret;
1520
1521         if (!files) {
1522                 assert(insertions == 0 && deletions == 0);
1523                 return fprintf(fp, "%s\n", " 0 files changed");
1524         }
1525
1526         strbuf_addf(&sb,
1527                     (files == 1) ? " %d file changed" : " %d files changed",
1528                     files);
1529
1530         /*
1531          * For binary diff, the caller may want to print "x files
1532          * changed" with insertions == 0 && deletions == 0.
1533          *
1534          * Not omitting "0 insertions(+), 0 deletions(-)" in this case
1535          * is probably less confusing (i.e skip over "2 files changed
1536          * but nothing about added/removed lines? Is this a bug in Git?").
1537          */
1538         if (insertions || deletions == 0) {
1539                 strbuf_addf(&sb,
1540                             (insertions == 1) ? ", %d insertion(+)" : ", %d insertions(+)",
1541                             insertions);
1542         }
1543
1544         if (deletions || insertions == 0) {
1545                 strbuf_addf(&sb,
1546                             (deletions == 1) ? ", %d deletion(-)" : ", %d deletions(-)",
1547                             deletions);
1548         }
1549         strbuf_addch(&sb, '\n');
1550         ret = fputs(sb.buf, fp);
1551         strbuf_release(&sb);
1552         return ret;
1553 }
1554
1555 static void show_stats(struct diffstat_t *data, struct diff_options *options)
1556 {
1557         int i, len, add, del, adds = 0, dels = 0;
1558         uintmax_t max_change = 0, max_len = 0;
1559         int total_files = data->nr, count;
1560         int width, name_width, graph_width, number_width = 0, bin_width = 0;
1561         const char *reset, *add_c, *del_c;
1562         const char *line_prefix = "";
1563         int extra_shown = 0;
1564
1565         if (data->nr == 0)
1566                 return;
1567
1568         line_prefix = diff_line_prefix(options);
1569         count = options->stat_count ? options->stat_count : data->nr;
1570
1571         reset = diff_get_color_opt(options, DIFF_RESET);
1572         add_c = diff_get_color_opt(options, DIFF_FILE_NEW);
1573         del_c = diff_get_color_opt(options, DIFF_FILE_OLD);
1574
1575         /*
1576          * Find the longest filename and max number of changes
1577          */
1578         for (i = 0; (i < count) && (i < data->nr); i++) {
1579                 struct diffstat_file *file = data->files[i];
1580                 uintmax_t change = file->added + file->deleted;
1581
1582                 if (!file->is_interesting && (change == 0)) {
1583                         count++; /* not shown == room for one more */
1584                         continue;
1585                 }
1586                 fill_print_name(file);
1587                 len = strlen(file->print_name);
1588                 if (max_len < len)
1589                         max_len = len;
1590
1591                 if (file->is_unmerged) {
1592                         /* "Unmerged" is 8 characters */
1593                         bin_width = bin_width < 8 ? 8 : bin_width;
1594                         continue;
1595                 }
1596                 if (file->is_binary) {
1597                         /* "Bin XXX -> YYY bytes" */
1598                         int w = 14 + decimal_width(file->added)
1599                                 + decimal_width(file->deleted);
1600                         bin_width = bin_width < w ? w : bin_width;
1601                         /* Display change counts aligned with "Bin" */
1602                         number_width = 3;
1603                         continue;
1604                 }
1605
1606                 if (max_change < change)
1607                         max_change = change;
1608         }
1609         count = i; /* where we can stop scanning in data->files[] */
1610
1611         /*
1612          * We have width = stat_width or term_columns() columns total.
1613          * We want a maximum of min(max_len, stat_name_width) for the name part.
1614          * We want a maximum of min(max_change, stat_graph_width) for the +- part.
1615          * We also need 1 for " " and 4 + decimal_width(max_change)
1616          * for " | NNNN " and one the empty column at the end, altogether
1617          * 6 + decimal_width(max_change).
1618          *
1619          * If there's not enough space, we will use the smaller of
1620          * stat_name_width (if set) and 5/8*width for the filename,
1621          * and the rest for constant elements + graph part, but no more
1622          * than stat_graph_width for the graph part.
1623          * (5/8 gives 50 for filename and 30 for the constant parts + graph
1624          * for the standard terminal size).
1625          *
1626          * In other words: stat_width limits the maximum width, and
1627          * stat_name_width fixes the maximum width of the filename,
1628          * and is also used to divide available columns if there
1629          * aren't enough.
1630          *
1631          * Binary files are displayed with "Bin XXX -> YYY bytes"
1632          * instead of the change count and graph. This part is treated
1633          * similarly to the graph part, except that it is not
1634          * "scaled". If total width is too small to accommodate the
1635          * guaranteed minimum width of the filename part and the
1636          * separators and this message, this message will "overflow"
1637          * making the line longer than the maximum width.
1638          */
1639
1640         if (options->stat_width == -1)
1641                 width = term_columns() - strlen(line_prefix);
1642         else
1643                 width = options->stat_width ? options->stat_width : 80;
1644         number_width = decimal_width(max_change) > number_width ?
1645                 decimal_width(max_change) : number_width;
1646
1647         if (options->stat_graph_width == -1)
1648                 options->stat_graph_width = diff_stat_graph_width;
1649
1650         /*
1651          * Guarantee 3/8*16==6 for the graph part
1652          * and 5/8*16==10 for the filename part
1653          */
1654         if (width < 16 + 6 + number_width)
1655                 width = 16 + 6 + number_width;
1656
1657         /*
1658          * First assign sizes that are wanted, ignoring available width.
1659          * strlen("Bin XXX -> YYY bytes") == bin_width, and the part
1660          * starting from "XXX" should fit in graph_width.
1661          */
1662         graph_width = max_change + 4 > bin_width ? max_change : bin_width - 4;
1663         if (options->stat_graph_width &&
1664             options->stat_graph_width < graph_width)
1665                 graph_width = options->stat_graph_width;
1666
1667         name_width = (options->stat_name_width > 0 &&
1668                       options->stat_name_width < max_len) ?
1669                 options->stat_name_width : max_len;
1670
1671         /*
1672          * Adjust adjustable widths not to exceed maximum width
1673          */
1674         if (name_width + number_width + 6 + graph_width > width) {
1675                 if (graph_width > width * 3/8 - number_width - 6) {
1676                         graph_width = width * 3/8 - number_width - 6;
1677                         if (graph_width < 6)
1678                                 graph_width = 6;
1679                 }
1680
1681                 if (options->stat_graph_width &&
1682                     graph_width > options->stat_graph_width)
1683                         graph_width = options->stat_graph_width;
1684                 if (name_width > width - number_width - 6 - graph_width)
1685                         name_width = width - number_width - 6 - graph_width;
1686                 else
1687                         graph_width = width - number_width - 6 - name_width;
1688         }
1689
1690         /*
1691          * From here name_width is the width of the name area,
1692          * and graph_width is the width of the graph area.
1693          * max_change is used to scale graph properly.
1694          */
1695         for (i = 0; i < count; i++) {
1696                 const char *prefix = "";
1697                 struct diffstat_file *file = data->files[i];
1698                 char *name = file->print_name;
1699                 uintmax_t added = file->added;
1700                 uintmax_t deleted = file->deleted;
1701                 int name_len;
1702
1703                 if (!file->is_interesting && (added + deleted == 0))
1704                         continue;
1705
1706                 /*
1707                  * "scale" the filename
1708                  */
1709                 len = name_width;
1710                 name_len = strlen(name);
1711                 if (name_width < name_len) {
1712                         char *slash;
1713                         prefix = "...";
1714                         len -= 3;
1715                         name += name_len - len;
1716                         slash = strchr(name, '/');
1717                         if (slash)
1718                                 name = slash;
1719                 }
1720
1721                 if (file->is_binary) {
1722                         fprintf(options->file, "%s", line_prefix);
1723                         show_name(options->file, prefix, name, len);
1724                         fprintf(options->file, " %*s", number_width, "Bin");
1725                         if (!added && !deleted) {
1726                                 putc('\n', options->file);
1727                                 continue;
1728                         }
1729                         fprintf(options->file, " %s%"PRIuMAX"%s",
1730                                 del_c, deleted, reset);
1731                         fprintf(options->file, " -> ");
1732                         fprintf(options->file, "%s%"PRIuMAX"%s",
1733                                 add_c, added, reset);
1734                         fprintf(options->file, " bytes");
1735                         fprintf(options->file, "\n");
1736                         continue;
1737                 }
1738                 else if (file->is_unmerged) {
1739                         fprintf(options->file, "%s", line_prefix);
1740                         show_name(options->file, prefix, name, len);
1741                         fprintf(options->file, " Unmerged\n");
1742                         continue;
1743                 }
1744
1745                 /*
1746                  * scale the add/delete
1747                  */
1748                 add = added;
1749                 del = deleted;
1750
1751                 if (graph_width <= max_change) {
1752                         int total = scale_linear(add + del, graph_width, max_change);
1753                         if (total < 2 && add && del)
1754                                 /* width >= 2 due to the sanity check */
1755                                 total = 2;
1756                         if (add < del) {
1757                                 add = scale_linear(add, graph_width, max_change);
1758                                 del = total - add;
1759                         } else {
1760                                 del = scale_linear(del, graph_width, max_change);
1761                                 add = total - del;
1762                         }
1763                 }
1764                 fprintf(options->file, "%s", line_prefix);
1765                 show_name(options->file, prefix, name, len);
1766                 fprintf(options->file, " %*"PRIuMAX"%s",
1767                         number_width, added + deleted,
1768                         added + deleted ? " " : "");
1769                 show_graph(options->file, '+', add, add_c, reset);
1770                 show_graph(options->file, '-', del, del_c, reset);
1771                 fprintf(options->file, "\n");
1772         }
1773
1774         for (i = 0; i < data->nr; i++) {
1775                 struct diffstat_file *file = data->files[i];
1776                 uintmax_t added = file->added;
1777                 uintmax_t deleted = file->deleted;
1778
1779                 if (file->is_unmerged ||
1780                     (!file->is_interesting && (added + deleted == 0))) {
1781                         total_files--;
1782                         continue;
1783                 }
1784
1785                 if (!file->is_binary) {
1786                         adds += added;
1787                         dels += deleted;
1788                 }
1789                 if (i < count)
1790                         continue;
1791                 if (!extra_shown)
1792                         fprintf(options->file, "%s ...\n", line_prefix);
1793                 extra_shown = 1;
1794         }
1795         fprintf(options->file, "%s", line_prefix);
1796         print_stat_summary(options->file, total_files, adds, dels);
1797 }
1798
1799 static void show_shortstats(struct diffstat_t *data, struct diff_options *options)
1800 {
1801         int i, adds = 0, dels = 0, total_files = data->nr;
1802
1803         if (data->nr == 0)
1804                 return;
1805
1806         for (i = 0; i < data->nr; i++) {
1807                 int added = data->files[i]->added;
1808                 int deleted= data->files[i]->deleted;
1809
1810                 if (data->files[i]->is_unmerged ||
1811                     (!data->files[i]->is_interesting && (added + deleted == 0))) {
1812                         total_files--;
1813                 } else if (!data->files[i]->is_binary) { /* don't count bytes */
1814                         adds += added;
1815                         dels += deleted;
1816                 }
1817         }
1818         fprintf(options->file, "%s", diff_line_prefix(options));
1819         print_stat_summary(options->file, total_files, adds, dels);
1820 }
1821
1822 static void show_numstat(struct diffstat_t *data, struct diff_options *options)
1823 {
1824         int i;
1825
1826         if (data->nr == 0)
1827                 return;
1828
1829         for (i = 0; i < data->nr; i++) {
1830                 struct diffstat_file *file = data->files[i];
1831
1832                 fprintf(options->file, "%s", diff_line_prefix(options));
1833
1834                 if (file->is_binary)
1835                         fprintf(options->file, "-\t-\t");
1836                 else
1837                         fprintf(options->file,
1838                                 "%"PRIuMAX"\t%"PRIuMAX"\t",
1839                                 file->added, file->deleted);
1840                 if (options->line_termination) {
1841                         fill_print_name(file);
1842                         if (!file->is_renamed)
1843                                 write_name_quoted(file->name, options->file,
1844                                                   options->line_termination);
1845                         else {
1846                                 fputs(file->print_name, options->file);
1847                                 putc(options->line_termination, options->file);
1848                         }
1849                 } else {
1850                         if (file->is_renamed) {
1851                                 putc('\0', options->file);
1852                                 write_name_quoted(file->from_name, options->file, '\0');
1853                         }
1854                         write_name_quoted(file->name, options->file, '\0');
1855                 }
1856         }
1857 }
1858
1859 struct dirstat_file {
1860         const char *name;
1861         unsigned long changed;
1862 };
1863
1864 struct dirstat_dir {
1865         struct dirstat_file *files;
1866         int alloc, nr, permille, cumulative;
1867 };
1868
1869 static long gather_dirstat(struct diff_options *opt, struct dirstat_dir *dir,
1870                 unsigned long changed, const char *base, int baselen)
1871 {
1872         unsigned long this_dir = 0;
1873         unsigned int sources = 0;
1874         const char *line_prefix = diff_line_prefix(opt);
1875
1876         while (dir->nr) {
1877                 struct dirstat_file *f = dir->files;
1878                 int namelen = strlen(f->name);
1879                 unsigned long this;
1880                 char *slash;
1881
1882                 if (namelen < baselen)
1883                         break;
1884                 if (memcmp(f->name, base, baselen))
1885                         break;
1886                 slash = strchr(f->name + baselen, '/');
1887                 if (slash) {
1888                         int newbaselen = slash + 1 - f->name;
1889                         this = gather_dirstat(opt, dir, changed, f->name, newbaselen);
1890                         sources++;
1891                 } else {
1892                         this = f->changed;
1893                         dir->files++;
1894                         dir->nr--;
1895                         sources += 2;
1896                 }
1897                 this_dir += this;
1898         }
1899
1900         /*
1901          * We don't report dirstat's for
1902          *  - the top level
1903          *  - or cases where everything came from a single directory
1904          *    under this directory (sources == 1).
1905          */
1906         if (baselen && sources != 1) {
1907                 if (this_dir) {
1908                         int permille = this_dir * 1000 / changed;
1909                         if (permille >= dir->permille) {
1910                                 fprintf(opt->file, "%s%4d.%01d%% %.*s\n", line_prefix,
1911                                         permille / 10, permille % 10, baselen, base);
1912                                 if (!dir->cumulative)
1913                                         return 0;
1914                         }
1915                 }
1916         }
1917         return this_dir;
1918 }
1919
1920 static int dirstat_compare(const void *_a, const void *_b)
1921 {
1922         const struct dirstat_file *a = _a;
1923         const struct dirstat_file *b = _b;
1924         return strcmp(a->name, b->name);
1925 }
1926
1927 static void show_dirstat(struct diff_options *options)
1928 {
1929         int i;
1930         unsigned long changed;
1931         struct dirstat_dir dir;
1932         struct diff_queue_struct *q = &diff_queued_diff;
1933
1934         dir.files = NULL;
1935         dir.alloc = 0;
1936         dir.nr = 0;
1937         dir.permille = options->dirstat_permille;
1938         dir.cumulative = DIFF_OPT_TST(options, DIRSTAT_CUMULATIVE);
1939
1940         changed = 0;
1941         for (i = 0; i < q->nr; i++) {
1942                 struct diff_filepair *p = q->queue[i];
1943                 const char *name;
1944                 unsigned long copied, added, damage;
1945                 int content_changed;
1946
1947                 name = p->two->path ? p->two->path : p->one->path;
1948
1949                 if (p->one->oid_valid && p->two->oid_valid)
1950                         content_changed = oidcmp(&p->one->oid, &p->two->oid);
1951                 else
1952                         content_changed = 1;
1953
1954                 if (!content_changed) {
1955                         /*
1956                          * The SHA1 has not changed, so pre-/post-content is
1957                          * identical. We can therefore skip looking at the
1958                          * file contents altogether.
1959                          */
1960                         damage = 0;
1961                         goto found_damage;
1962                 }
1963
1964                 if (DIFF_OPT_TST(options, DIRSTAT_BY_FILE)) {
1965                         /*
1966                          * In --dirstat-by-file mode, we don't really need to
1967                          * look at the actual file contents at all.
1968                          * The fact that the SHA1 changed is enough for us to
1969                          * add this file to the list of results
1970                          * (with each file contributing equal damage).
1971                          */
1972                         damage = 1;
1973                         goto found_damage;
1974                 }
1975
1976                 if (DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
1977                         diff_populate_filespec(p->one, 0);
1978                         diff_populate_filespec(p->two, 0);
1979                         diffcore_count_changes(p->one, p->two, NULL, NULL, 0,
1980                                                &copied, &added);
1981                         diff_free_filespec_data(p->one);
1982                         diff_free_filespec_data(p->two);
1983                 } else if (DIFF_FILE_VALID(p->one)) {
1984                         diff_populate_filespec(p->one, CHECK_SIZE_ONLY);
1985                         copied = added = 0;
1986                         diff_free_filespec_data(p->one);
1987                 } else if (DIFF_FILE_VALID(p->two)) {
1988                         diff_populate_filespec(p->two, CHECK_SIZE_ONLY);
1989                         copied = 0;
1990                         added = p->two->size;
1991                         diff_free_filespec_data(p->two);
1992                 } else
1993                         continue;
1994
1995                 /*
1996                  * Original minus copied is the removed material,
1997                  * added is the new material.  They are both damages
1998                  * made to the preimage.
1999                  * If the resulting damage is zero, we know that
2000                  * diffcore_count_changes() considers the two entries to
2001                  * be identical, but since content_changed is true, we
2002                  * know that there must have been _some_ kind of change,
2003                  * so we force all entries to have damage > 0.
2004                  */
2005                 damage = (p->one->size - copied) + added;
2006                 if (!damage)
2007                         damage = 1;
2008
2009 found_damage:
2010                 ALLOC_GROW(dir.files, dir.nr + 1, dir.alloc);
2011                 dir.files[dir.nr].name = name;
2012                 dir.files[dir.nr].changed = damage;
2013                 changed += damage;
2014                 dir.nr++;
2015         }
2016
2017         /* This can happen even with many files, if everything was renames */
2018         if (!changed)
2019                 return;
2020
2021         /* Show all directories with more than x% of the changes */
2022         qsort(dir.files, dir.nr, sizeof(dir.files[0]), dirstat_compare);
2023         gather_dirstat(options, &dir, changed, "", 0);
2024 }
2025
2026 static void show_dirstat_by_line(struct diffstat_t *data, struct diff_options *options)
2027 {
2028         int i;
2029         unsigned long changed;
2030         struct dirstat_dir dir;
2031
2032         if (data->nr == 0)
2033                 return;
2034
2035         dir.files = NULL;
2036         dir.alloc = 0;
2037         dir.nr = 0;
2038         dir.permille = options->dirstat_permille;
2039         dir.cumulative = DIFF_OPT_TST(options, DIRSTAT_CUMULATIVE);
2040
2041         changed = 0;
2042         for (i = 0; i < data->nr; i++) {
2043                 struct diffstat_file *file = data->files[i];
2044                 unsigned long damage = file->added + file->deleted;
2045                 if (file->is_binary)
2046                         /*
2047                          * binary files counts bytes, not lines. Must find some
2048                          * way to normalize binary bytes vs. textual lines.
2049                          * The following heuristic assumes that there are 64
2050                          * bytes per "line".
2051                          * This is stupid and ugly, but very cheap...
2052                          */
2053                         damage = (damage + 63) / 64;
2054                 ALLOC_GROW(dir.files, dir.nr + 1, dir.alloc);
2055                 dir.files[dir.nr].name = file->name;
2056                 dir.files[dir.nr].changed = damage;
2057                 changed += damage;
2058                 dir.nr++;
2059         }
2060
2061         /* This can happen even with many files, if everything was renames */
2062         if (!changed)
2063                 return;
2064
2065         /* Show all directories with more than x% of the changes */
2066         qsort(dir.files, dir.nr, sizeof(dir.files[0]), dirstat_compare);
2067         gather_dirstat(options, &dir, changed, "", 0);
2068 }
2069
2070 static void free_diffstat_info(struct diffstat_t *diffstat)
2071 {
2072         int i;
2073         for (i = 0; i < diffstat->nr; i++) {
2074                 struct diffstat_file *f = diffstat->files[i];
2075                 if (f->name != f->print_name)
2076                         free(f->print_name);
2077                 free(f->name);
2078                 free(f->from_name);
2079                 free(f);
2080         }
2081         free(diffstat->files);
2082 }
2083
2084 struct checkdiff_t {
2085         const char *filename;
2086         int lineno;
2087         int conflict_marker_size;
2088         struct diff_options *o;
2089         unsigned ws_rule;
2090         unsigned status;
2091 };
2092
2093 static int is_conflict_marker(const char *line, int marker_size, unsigned long len)
2094 {
2095         char firstchar;
2096         int cnt;
2097
2098         if (len < marker_size + 1)
2099                 return 0;
2100         firstchar = line[0];
2101         switch (firstchar) {
2102         case '=': case '>': case '<': case '|':
2103                 break;
2104         default:
2105                 return 0;
2106         }
2107         for (cnt = 1; cnt < marker_size; cnt++)
2108                 if (line[cnt] != firstchar)
2109                         return 0;
2110         /* line[1] thru line[marker_size-1] are same as firstchar */
2111         if (len < marker_size + 1 || !isspace(line[marker_size]))
2112                 return 0;
2113         return 1;
2114 }
2115
2116 static void checkdiff_consume(void *priv, char *line, unsigned long len)
2117 {
2118         struct checkdiff_t *data = priv;
2119         int marker_size = data->conflict_marker_size;
2120         const char *ws = diff_get_color(data->o->use_color, DIFF_WHITESPACE);
2121         const char *reset = diff_get_color(data->o->use_color, DIFF_RESET);
2122         const char *set = diff_get_color(data->o->use_color, DIFF_FILE_NEW);
2123         char *err;
2124         const char *line_prefix;
2125
2126         assert(data->o);
2127         line_prefix = diff_line_prefix(data->o);
2128
2129         if (line[0] == '+') {
2130                 unsigned bad;
2131                 data->lineno++;
2132                 if (is_conflict_marker(line + 1, marker_size, len - 1)) {
2133                         data->status |= 1;
2134                         fprintf(data->o->file,
2135                                 "%s%s:%d: leftover conflict marker\n",
2136                                 line_prefix, data->filename, data->lineno);
2137                 }
2138                 bad = ws_check(line + 1, len - 1, data->ws_rule);
2139                 if (!bad)
2140                         return;
2141                 data->status |= bad;
2142                 err = whitespace_error_string(bad);
2143                 fprintf(data->o->file, "%s%s:%d: %s.\n",
2144                         line_prefix, data->filename, data->lineno, err);
2145                 free(err);
2146                 emit_line(data->o, set, reset, line, 1);
2147                 ws_check_emit(line + 1, len - 1, data->ws_rule,
2148                               data->o->file, set, reset, ws);
2149         } else if (line[0] == ' ') {
2150                 data->lineno++;
2151         } else if (line[0] == '@') {
2152                 char *plus = strchr(line, '+');
2153                 if (plus)
2154                         data->lineno = strtol(plus, NULL, 10) - 1;
2155                 else
2156                         die("invalid diff");
2157         }
2158 }
2159
2160 static unsigned char *deflate_it(char *data,
2161                                  unsigned long size,
2162                                  unsigned long *result_size)
2163 {
2164         int bound;
2165         unsigned char *deflated;
2166         git_zstream stream;
2167
2168         git_deflate_init(&stream, zlib_compression_level);
2169         bound = git_deflate_bound(&stream, size);
2170         deflated = xmalloc(bound);
2171         stream.next_out = deflated;
2172         stream.avail_out = bound;
2173
2174         stream.next_in = (unsigned char *)data;
2175         stream.avail_in = size;
2176         while (git_deflate(&stream, Z_FINISH) == Z_OK)
2177                 ; /* nothing */
2178         git_deflate_end(&stream);
2179         *result_size = stream.total_out;
2180         return deflated;
2181 }
2182
2183 static void emit_binary_diff_body(FILE *file, mmfile_t *one, mmfile_t *two,
2184                                   const char *prefix)
2185 {
2186         void *cp;
2187         void *delta;
2188         void *deflated;
2189         void *data;
2190         unsigned long orig_size;
2191         unsigned long delta_size;
2192         unsigned long deflate_size;
2193         unsigned long data_size;
2194
2195         /* We could do deflated delta, or we could do just deflated two,
2196          * whichever is smaller.
2197          */
2198         delta = NULL;
2199         deflated = deflate_it(two->ptr, two->size, &deflate_size);
2200         if (one->size && two->size) {
2201                 delta = diff_delta(one->ptr, one->size,
2202                                    two->ptr, two->size,
2203                                    &delta_size, deflate_size);
2204                 if (delta) {
2205                         void *to_free = delta;
2206                         orig_size = delta_size;
2207                         delta = deflate_it(delta, delta_size, &delta_size);
2208                         free(to_free);
2209                 }
2210         }
2211
2212         if (delta && delta_size < deflate_size) {
2213                 fprintf(file, "%sdelta %lu\n", prefix, orig_size);
2214                 free(deflated);
2215                 data = delta;
2216                 data_size = delta_size;
2217         }
2218         else {
2219                 fprintf(file, "%sliteral %lu\n", prefix, two->size);
2220                 free(delta);
2221                 data = deflated;
2222                 data_size = deflate_size;
2223         }
2224
2225         /* emit data encoded in base85 */
2226         cp = data;
2227         while (data_size) {
2228                 int bytes = (52 < data_size) ? 52 : data_size;
2229                 char line[70];
2230                 data_size -= bytes;
2231                 if (bytes <= 26)
2232                         line[0] = bytes + 'A' - 1;
2233                 else
2234                         line[0] = bytes - 26 + 'a' - 1;
2235                 encode_85(line + 1, cp, bytes);
2236                 cp = (char *) cp + bytes;
2237                 fprintf(file, "%s", prefix);
2238                 fputs(line, file);
2239                 fputc('\n', file);
2240         }
2241         fprintf(file, "%s\n", prefix);
2242         free(data);
2243 }
2244
2245 static void emit_binary_diff(FILE *file, mmfile_t *one, mmfile_t *two,
2246                              const char *prefix)
2247 {
2248         fprintf(file, "%sGIT binary patch\n", prefix);
2249         emit_binary_diff_body(file, one, two, prefix);
2250         emit_binary_diff_body(file, two, one, prefix);
2251 }
2252
2253 int diff_filespec_is_binary(struct diff_filespec *one)
2254 {
2255         if (one->is_binary == -1) {
2256                 diff_filespec_load_driver(one);
2257                 if (one->driver->binary != -1)
2258                         one->is_binary = one->driver->binary;
2259                 else {
2260                         if (!one->data && DIFF_FILE_VALID(one))
2261                                 diff_populate_filespec(one, CHECK_BINARY);
2262                         if (one->is_binary == -1 && one->data)
2263                                 one->is_binary = buffer_is_binary(one->data,
2264                                                 one->size);
2265                         if (one->is_binary == -1)
2266                                 one->is_binary = 0;
2267                 }
2268         }
2269         return one->is_binary;
2270 }
2271
2272 static const struct userdiff_funcname *diff_funcname_pattern(struct diff_filespec *one)
2273 {
2274         diff_filespec_load_driver(one);
2275         return one->driver->funcname.pattern ? &one->driver->funcname : NULL;
2276 }
2277
2278 void diff_set_mnemonic_prefix(struct diff_options *options, const char *a, const char *b)
2279 {
2280         if (!options->a_prefix)
2281                 options->a_prefix = a;
2282         if (!options->b_prefix)
2283                 options->b_prefix = b;
2284 }
2285
2286 struct userdiff_driver *get_textconv(struct diff_filespec *one)
2287 {
2288         if (!DIFF_FILE_VALID(one))
2289                 return NULL;
2290
2291         diff_filespec_load_driver(one);
2292         return userdiff_get_textconv(one->driver);
2293 }
2294
2295 static void builtin_diff(const char *name_a,
2296                          const char *name_b,
2297                          struct diff_filespec *one,
2298                          struct diff_filespec *two,
2299                          const char *xfrm_msg,
2300                          int must_show_header,
2301                          struct diff_options *o,
2302                          int complete_rewrite)
2303 {
2304         mmfile_t mf1, mf2;
2305         const char *lbl[2];
2306         char *a_one, *b_two;
2307         const char *meta = diff_get_color_opt(o, DIFF_METAINFO);
2308         const char *reset = diff_get_color_opt(o, DIFF_RESET);
2309         const char *a_prefix, *b_prefix;
2310         struct userdiff_driver *textconv_one = NULL;
2311         struct userdiff_driver *textconv_two = NULL;
2312         struct strbuf header = STRBUF_INIT;
2313         const char *line_prefix = diff_line_prefix(o);
2314
2315         diff_set_mnemonic_prefix(o, "a/", "b/");
2316         if (DIFF_OPT_TST(o, REVERSE_DIFF)) {
2317                 a_prefix = o->b_prefix;
2318                 b_prefix = o->a_prefix;
2319         } else {
2320                 a_prefix = o->a_prefix;
2321                 b_prefix = o->b_prefix;
2322         }
2323
2324         if (o->submodule_format == DIFF_SUBMODULE_LOG &&
2325             (!one->mode || S_ISGITLINK(one->mode)) &&
2326             (!two->mode || S_ISGITLINK(two->mode))) {
2327                 const char *del = diff_get_color_opt(o, DIFF_FILE_OLD);
2328                 const char *add = diff_get_color_opt(o, DIFF_FILE_NEW);
2329                 show_submodule_summary(o->file, one->path ? one->path : two->path,
2330                                 line_prefix,
2331                                 &one->oid, &two->oid,
2332                                 two->dirty_submodule,
2333                                 meta, del, add, reset);
2334                 return;
2335         } else if (o->submodule_format == DIFF_SUBMODULE_INLINE_DIFF &&
2336                    (!one->mode || S_ISGITLINK(one->mode)) &&
2337                    (!two->mode || S_ISGITLINK(two->mode))) {
2338                 const char *del = diff_get_color_opt(o, DIFF_FILE_OLD);
2339                 const char *add = diff_get_color_opt(o, DIFF_FILE_NEW);
2340                 show_submodule_inline_diff(o->file, one->path ? one->path : two->path,
2341                                 line_prefix,
2342                                 &one->oid, &two->oid,
2343                                 two->dirty_submodule,
2344                                 meta, del, add, reset, o);
2345                 return;
2346         }
2347
2348         if (DIFF_OPT_TST(o, ALLOW_TEXTCONV)) {
2349                 textconv_one = get_textconv(one);
2350                 textconv_two = get_textconv(two);
2351         }
2352
2353         /* Never use a non-valid filename anywhere if at all possible */
2354         name_a = DIFF_FILE_VALID(one) ? name_a : name_b;
2355         name_b = DIFF_FILE_VALID(two) ? name_b : name_a;
2356
2357         a_one = quote_two(a_prefix, name_a + (*name_a == '/'));
2358         b_two = quote_two(b_prefix, name_b + (*name_b == '/'));
2359         lbl[0] = DIFF_FILE_VALID(one) ? a_one : "/dev/null";
2360         lbl[1] = DIFF_FILE_VALID(two) ? b_two : "/dev/null";
2361         strbuf_addf(&header, "%s%sdiff --git %s %s%s\n", line_prefix, meta, a_one, b_two, reset);
2362         if (lbl[0][0] == '/') {
2363                 /* /dev/null */
2364                 strbuf_addf(&header, "%s%snew file mode %06o%s\n", line_prefix, meta, two->mode, reset);
2365                 if (xfrm_msg)
2366                         strbuf_addstr(&header, xfrm_msg);
2367                 must_show_header = 1;
2368         }
2369         else if (lbl[1][0] == '/') {
2370                 strbuf_addf(&header, "%s%sdeleted file mode %06o%s\n", line_prefix, meta, one->mode, reset);
2371                 if (xfrm_msg)
2372                         strbuf_addstr(&header, xfrm_msg);
2373                 must_show_header = 1;
2374         }
2375         else {
2376                 if (one->mode != two->mode) {
2377                         strbuf_addf(&header, "%s%sold mode %06o%s\n", line_prefix, meta, one->mode, reset);
2378                         strbuf_addf(&header, "%s%snew mode %06o%s\n", line_prefix, meta, two->mode, reset);
2379                         must_show_header = 1;
2380                 }
2381                 if (xfrm_msg)
2382                         strbuf_addstr(&header, xfrm_msg);
2383
2384                 /*
2385                  * we do not run diff between different kind
2386                  * of objects.
2387                  */
2388                 if ((one->mode ^ two->mode) & S_IFMT)
2389                         goto free_ab_and_return;
2390                 if (complete_rewrite &&
2391                     (textconv_one || !diff_filespec_is_binary(one)) &&
2392                     (textconv_two || !diff_filespec_is_binary(two))) {
2393                         fprintf(o->file, "%s", header.buf);
2394                         strbuf_reset(&header);
2395                         emit_rewrite_diff(name_a, name_b, one, two,
2396                                                 textconv_one, textconv_two, o);
2397                         o->found_changes = 1;
2398                         goto free_ab_and_return;
2399                 }
2400         }
2401
2402         if (o->irreversible_delete && lbl[1][0] == '/') {
2403                 fprintf(o->file, "%s", header.buf);
2404                 strbuf_reset(&header);
2405                 goto free_ab_and_return;
2406         } else if (!DIFF_OPT_TST(o, TEXT) &&
2407             ( (!textconv_one && diff_filespec_is_binary(one)) ||
2408               (!textconv_two && diff_filespec_is_binary(two)) )) {
2409                 if (!one->data && !two->data &&
2410                     S_ISREG(one->mode) && S_ISREG(two->mode) &&
2411                     !DIFF_OPT_TST(o, BINARY)) {
2412                         if (!oidcmp(&one->oid, &two->oid)) {
2413                                 if (must_show_header)
2414                                         fprintf(o->file, "%s", header.buf);
2415                                 goto free_ab_and_return;
2416                         }
2417                         fprintf(o->file, "%s", header.buf);
2418                         fprintf(o->file, "%sBinary files %s and %s differ\n",
2419                                 line_prefix, lbl[0], lbl[1]);
2420                         goto free_ab_and_return;
2421                 }
2422                 if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
2423                         die("unable to read files to diff");
2424                 /* Quite common confusing case */
2425                 if (mf1.size == mf2.size &&
2426                     !memcmp(mf1.ptr, mf2.ptr, mf1.size)) {
2427                         if (must_show_header)
2428                                 fprintf(o->file, "%s", header.buf);
2429                         goto free_ab_and_return;
2430                 }
2431                 fprintf(o->file, "%s", header.buf);
2432                 strbuf_reset(&header);
2433                 if (DIFF_OPT_TST(o, BINARY))
2434                         emit_binary_diff(o->file, &mf1, &mf2, line_prefix);
2435                 else
2436                         fprintf(o->file, "%sBinary files %s and %s differ\n",
2437                                 line_prefix, lbl[0], lbl[1]);
2438                 o->found_changes = 1;
2439         } else {
2440                 /* Crazy xdl interfaces.. */
2441                 const char *diffopts = getenv("GIT_DIFF_OPTS");
2442                 const char *v;
2443                 xpparam_t xpp;
2444                 xdemitconf_t xecfg;
2445                 struct emit_callback ecbdata;
2446                 const struct userdiff_funcname *pe;
2447
2448                 if (must_show_header) {
2449                         fprintf(o->file, "%s", header.buf);
2450                         strbuf_reset(&header);
2451                 }
2452
2453                 mf1.size = fill_textconv(textconv_one, one, &mf1.ptr);
2454                 mf2.size = fill_textconv(textconv_two, two, &mf2.ptr);
2455
2456                 pe = diff_funcname_pattern(one);
2457                 if (!pe)
2458                         pe = diff_funcname_pattern(two);
2459
2460                 memset(&xpp, 0, sizeof(xpp));
2461                 memset(&xecfg, 0, sizeof(xecfg));
2462                 memset(&ecbdata, 0, sizeof(ecbdata));
2463                 ecbdata.label_path = lbl;
2464                 ecbdata.color_diff = want_color(o->use_color);
2465                 ecbdata.ws_rule = whitespace_rule(name_b);
2466                 if (ecbdata.ws_rule & WS_BLANK_AT_EOF)
2467                         check_blank_at_eof(&mf1, &mf2, &ecbdata);
2468                 ecbdata.opt = o;
2469                 ecbdata.header = header.len ? &header : NULL;
2470                 xpp.flags = o->xdl_opts;
2471                 xecfg.ctxlen = o->context;
2472                 xecfg.interhunkctxlen = o->interhunkcontext;
2473                 xecfg.flags = XDL_EMIT_FUNCNAMES;
2474                 if (DIFF_OPT_TST(o, FUNCCONTEXT))
2475                         xecfg.flags |= XDL_EMIT_FUNCCONTEXT;
2476                 if (pe)
2477                         xdiff_set_find_func(&xecfg, pe->pattern, pe->cflags);
2478                 if (!diffopts)
2479                         ;
2480                 else if (skip_prefix(diffopts, "--unified=", &v))
2481                         xecfg.ctxlen = strtoul(v, NULL, 10);
2482                 else if (skip_prefix(diffopts, "-u", &v))
2483                         xecfg.ctxlen = strtoul(v, NULL, 10);
2484                 if (o->word_diff)
2485                         init_diff_words_data(&ecbdata, o, one, two);
2486                 if (xdi_diff_outf(&mf1, &mf2, fn_out_consume, &ecbdata,
2487                                   &xpp, &xecfg))
2488                         die("unable to generate diff for %s", one->path);
2489                 if (o->word_diff)
2490                         free_diff_words_data(&ecbdata);
2491                 if (textconv_one)
2492                         free(mf1.ptr);
2493                 if (textconv_two)
2494                         free(mf2.ptr);
2495                 xdiff_clear_find_func(&xecfg);
2496         }
2497
2498  free_ab_and_return:
2499         strbuf_release(&header);
2500         diff_free_filespec_data(one);
2501         diff_free_filespec_data(two);
2502         free(a_one);
2503         free(b_two);
2504         return;
2505 }
2506
2507 static void builtin_diffstat(const char *name_a, const char *name_b,
2508                              struct diff_filespec *one,
2509                              struct diff_filespec *two,
2510                              struct diffstat_t *diffstat,
2511                              struct diff_options *o,
2512                              struct diff_filepair *p)
2513 {
2514         mmfile_t mf1, mf2;
2515         struct diffstat_file *data;
2516         int same_contents;
2517         int complete_rewrite = 0;
2518
2519         if (!DIFF_PAIR_UNMERGED(p)) {
2520                 if (p->status == DIFF_STATUS_MODIFIED && p->score)
2521                         complete_rewrite = 1;
2522         }
2523
2524         data = diffstat_add(diffstat, name_a, name_b);
2525         data->is_interesting = p->status != DIFF_STATUS_UNKNOWN;
2526
2527         if (!one || !two) {
2528                 data->is_unmerged = 1;
2529                 return;
2530         }
2531
2532         same_contents = !oidcmp(&one->oid, &two->oid);
2533
2534         if (diff_filespec_is_binary(one) || diff_filespec_is_binary(two)) {
2535                 data->is_binary = 1;
2536                 if (same_contents) {
2537                         data->added = 0;
2538                         data->deleted = 0;
2539                 } else {
2540                         data->added = diff_filespec_size(two);
2541                         data->deleted = diff_filespec_size(one);
2542                 }
2543         }
2544
2545         else if (complete_rewrite) {
2546                 diff_populate_filespec(one, 0);
2547                 diff_populate_filespec(two, 0);
2548                 data->deleted = count_lines(one->data, one->size);
2549                 data->added = count_lines(two->data, two->size);
2550         }
2551
2552         else if (!same_contents) {
2553                 /* Crazy xdl interfaces.. */
2554                 xpparam_t xpp;
2555                 xdemitconf_t xecfg;
2556
2557                 if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
2558                         die("unable to read files to diff");
2559
2560                 memset(&xpp, 0, sizeof(xpp));
2561                 memset(&xecfg, 0, sizeof(xecfg));
2562                 xpp.flags = o->xdl_opts;
2563                 xecfg.ctxlen = o->context;
2564                 xecfg.interhunkctxlen = o->interhunkcontext;
2565                 if (xdi_diff_outf(&mf1, &mf2, diffstat_consume, diffstat,
2566                                   &xpp, &xecfg))
2567                         die("unable to generate diffstat for %s", one->path);
2568         }
2569
2570         diff_free_filespec_data(one);
2571         diff_free_filespec_data(two);
2572 }
2573
2574 static void builtin_checkdiff(const char *name_a, const char *name_b,
2575                               const char *attr_path,
2576                               struct diff_filespec *one,
2577                               struct diff_filespec *two,
2578                               struct diff_options *o)
2579 {
2580         mmfile_t mf1, mf2;
2581         struct checkdiff_t data;
2582
2583         if (!two)
2584                 return;
2585
2586         memset(&data, 0, sizeof(data));
2587         data.filename = name_b ? name_b : name_a;
2588         data.lineno = 0;
2589         data.o = o;
2590         data.ws_rule = whitespace_rule(attr_path);
2591         data.conflict_marker_size = ll_merge_marker_size(attr_path);
2592
2593         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
2594                 die("unable to read files to diff");
2595
2596         /*
2597          * All the other codepaths check both sides, but not checking
2598          * the "old" side here is deliberate.  We are checking the newly
2599          * introduced changes, and as long as the "new" side is text, we
2600          * can and should check what it introduces.
2601          */
2602         if (diff_filespec_is_binary(two))
2603                 goto free_and_return;
2604         else {
2605                 /* Crazy xdl interfaces.. */
2606                 xpparam_t xpp;
2607                 xdemitconf_t xecfg;
2608
2609                 memset(&xpp, 0, sizeof(xpp));
2610                 memset(&xecfg, 0, sizeof(xecfg));
2611                 xecfg.ctxlen = 1; /* at least one context line */
2612                 xpp.flags = 0;
2613                 if (xdi_diff_outf(&mf1, &mf2, checkdiff_consume, &data,
2614                                   &xpp, &xecfg))
2615                         die("unable to generate checkdiff for %s", one->path);
2616
2617                 if (data.ws_rule & WS_BLANK_AT_EOF) {
2618                         struct emit_callback ecbdata;
2619                         int blank_at_eof;
2620
2621                         ecbdata.ws_rule = data.ws_rule;
2622                         check_blank_at_eof(&mf1, &mf2, &ecbdata);
2623                         blank_at_eof = ecbdata.blank_at_eof_in_postimage;
2624
2625                         if (blank_at_eof) {
2626                                 static char *err;
2627                                 if (!err)
2628                                         err = whitespace_error_string(WS_BLANK_AT_EOF);
2629                                 fprintf(o->file, "%s:%d: %s.\n",
2630                                         data.filename, blank_at_eof, err);
2631                                 data.status = 1; /* report errors */
2632                         }
2633                 }
2634         }
2635  free_and_return:
2636         diff_free_filespec_data(one);
2637         diff_free_filespec_data(two);
2638         if (data.status)
2639                 DIFF_OPT_SET(o, CHECK_FAILED);
2640 }
2641
2642 struct diff_filespec *alloc_filespec(const char *path)
2643 {
2644         struct diff_filespec *spec;
2645
2646         FLEXPTR_ALLOC_STR(spec, path, path);
2647         spec->count = 1;
2648         spec->is_binary = -1;
2649         return spec;
2650 }
2651
2652 void free_filespec(struct diff_filespec *spec)
2653 {
2654         if (!--spec->count) {
2655                 diff_free_filespec_data(spec);
2656                 free(spec);
2657         }
2658 }
2659
2660 void fill_filespec(struct diff_filespec *spec, const unsigned char *sha1,
2661                    int sha1_valid, unsigned short mode)
2662 {
2663         if (mode) {
2664                 spec->mode = canon_mode(mode);
2665                 hashcpy(spec->oid.hash, sha1);
2666                 spec->oid_valid = sha1_valid;
2667         }
2668 }
2669
2670 /*
2671  * Given a name and sha1 pair, if the index tells us the file in
2672  * the work tree has that object contents, return true, so that
2673  * prepare_temp_file() does not have to inflate and extract.
2674  */
2675 static int reuse_worktree_file(const char *name, const unsigned char *sha1, int want_file)
2676 {
2677         const struct cache_entry *ce;
2678         struct stat st;
2679         int pos, len;
2680
2681         /*
2682          * We do not read the cache ourselves here, because the
2683          * benchmark with my previous version that always reads cache
2684          * shows that it makes things worse for diff-tree comparing
2685          * two linux-2.6 kernel trees in an already checked out work
2686          * tree.  This is because most diff-tree comparisons deal with
2687          * only a small number of files, while reading the cache is
2688          * expensive for a large project, and its cost outweighs the
2689          * savings we get by not inflating the object to a temporary
2690          * file.  Practically, this code only helps when we are used
2691          * by diff-cache --cached, which does read the cache before
2692          * calling us.
2693          */
2694         if (!active_cache)
2695                 return 0;
2696
2697         /* We want to avoid the working directory if our caller
2698          * doesn't need the data in a normal file, this system
2699          * is rather slow with its stat/open/mmap/close syscalls,
2700          * and the object is contained in a pack file.  The pack
2701          * is probably already open and will be faster to obtain
2702          * the data through than the working directory.  Loose
2703          * objects however would tend to be slower as they need
2704          * to be individually opened and inflated.
2705          */
2706         if (!FAST_WORKING_DIRECTORY && !want_file && has_sha1_pack(sha1))
2707                 return 0;
2708
2709         /*
2710          * Similarly, if we'd have to convert the file contents anyway, that
2711          * makes the optimization not worthwhile.
2712          */
2713         if (!want_file && would_convert_to_git(name))
2714                 return 0;
2715
2716         len = strlen(name);
2717         pos = cache_name_pos(name, len);
2718         if (pos < 0)
2719                 return 0;
2720         ce = active_cache[pos];
2721
2722         /*
2723          * This is not the sha1 we are looking for, or
2724          * unreusable because it is not a regular file.
2725          */
2726         if (hashcmp(sha1, ce->oid.hash) || !S_ISREG(ce->ce_mode))
2727                 return 0;
2728
2729         /*
2730          * If ce is marked as "assume unchanged", there is no
2731          * guarantee that work tree matches what we are looking for.
2732          */
2733         if ((ce->ce_flags & CE_VALID) || ce_skip_worktree(ce))
2734                 return 0;
2735
2736         /*
2737          * If ce matches the file in the work tree, we can reuse it.
2738          */
2739         if (ce_uptodate(ce) ||
2740             (!lstat(name, &st) && !ce_match_stat(ce, &st, 0)))
2741                 return 1;
2742
2743         return 0;
2744 }
2745
2746 static int diff_populate_gitlink(struct diff_filespec *s, int size_only)
2747 {
2748         struct strbuf buf = STRBUF_INIT;
2749         char *dirty = "";
2750
2751         /* Are we looking at the work tree? */
2752         if (s->dirty_submodule)
2753                 dirty = "-dirty";
2754
2755         strbuf_addf(&buf, "Subproject commit %s%s\n",
2756                     oid_to_hex(&s->oid), dirty);
2757         s->size = buf.len;
2758         if (size_only) {
2759                 s->data = NULL;
2760                 strbuf_release(&buf);
2761         } else {
2762                 s->data = strbuf_detach(&buf, NULL);
2763                 s->should_free = 1;
2764         }
2765         return 0;
2766 }
2767
2768 /*
2769  * While doing rename detection and pickaxe operation, we may need to
2770  * grab the data for the blob (or file) for our own in-core comparison.
2771  * diff_filespec has data and size fields for this purpose.
2772  */
2773 int diff_populate_filespec(struct diff_filespec *s, unsigned int flags)
2774 {
2775         int size_only = flags & CHECK_SIZE_ONLY;
2776         int err = 0;
2777         /*
2778          * demote FAIL to WARN to allow inspecting the situation
2779          * instead of refusing.
2780          */
2781         enum safe_crlf crlf_warn = (safe_crlf == SAFE_CRLF_FAIL
2782                                     ? SAFE_CRLF_WARN
2783                                     : safe_crlf);
2784
2785         if (!DIFF_FILE_VALID(s))
2786                 die("internal error: asking to populate invalid file.");
2787         if (S_ISDIR(s->mode))
2788                 return -1;
2789
2790         if (s->data)
2791                 return 0;
2792
2793         if (size_only && 0 < s->size)
2794                 return 0;
2795
2796         if (S_ISGITLINK(s->mode))
2797                 return diff_populate_gitlink(s, size_only);
2798
2799         if (!s->oid_valid ||
2800             reuse_worktree_file(s->path, s->oid.hash, 0)) {
2801                 struct strbuf buf = STRBUF_INIT;
2802                 struct stat st;
2803                 int fd;
2804
2805                 if (lstat(s->path, &st) < 0) {
2806                         if (errno == ENOENT) {
2807                         err_empty:
2808                                 err = -1;
2809                         empty:
2810                                 s->data = (char *)"";
2811                                 s->size = 0;
2812                                 return err;
2813                         }
2814                 }
2815                 s->size = xsize_t(st.st_size);
2816                 if (!s->size)
2817                         goto empty;
2818                 if (S_ISLNK(st.st_mode)) {
2819                         struct strbuf sb = STRBUF_INIT;
2820
2821                         if (strbuf_readlink(&sb, s->path, s->size))
2822                                 goto err_empty;
2823                         s->size = sb.len;
2824                         s->data = strbuf_detach(&sb, NULL);
2825                         s->should_free = 1;
2826                         return 0;
2827                 }
2828                 if (size_only)
2829                         return 0;
2830                 if ((flags & CHECK_BINARY) &&
2831                     s->size > big_file_threshold && s->is_binary == -1) {
2832                         s->is_binary = 1;
2833                         return 0;
2834                 }
2835                 fd = open(s->path, O_RDONLY);
2836                 if (fd < 0)
2837                         goto err_empty;
2838                 s->data = xmmap(NULL, s->size, PROT_READ, MAP_PRIVATE, fd, 0);
2839                 close(fd);
2840                 s->should_munmap = 1;
2841
2842                 /*
2843                  * Convert from working tree format to canonical git format
2844                  */
2845                 if (convert_to_git(s->path, s->data, s->size, &buf, crlf_warn)) {
2846                         size_t size = 0;
2847                         munmap(s->data, s->size);
2848                         s->should_munmap = 0;
2849                         s->data = strbuf_detach(&buf, &size);
2850                         s->size = size;
2851                         s->should_free = 1;
2852                 }
2853         }
2854         else {
2855                 enum object_type type;
2856                 if (size_only || (flags & CHECK_BINARY)) {
2857                         type = sha1_object_info(s->oid.hash, &s->size);
2858                         if (type < 0)
2859                                 die("unable to read %s",
2860                                     oid_to_hex(&s->oid));
2861                         if (size_only)
2862                                 return 0;
2863                         if (s->size > big_file_threshold && s->is_binary == -1) {
2864                                 s->is_binary = 1;
2865                                 return 0;
2866                         }
2867                 }
2868                 s->data = read_sha1_file(s->oid.hash, &type, &s->size);
2869                 if (!s->data)
2870                         die("unable to read %s", oid_to_hex(&s->oid));
2871                 s->should_free = 1;
2872         }
2873         return 0;
2874 }
2875
2876 void diff_free_filespec_blob(struct diff_filespec *s)
2877 {
2878         if (s->should_free)
2879                 free(s->data);
2880         else if (s->should_munmap)
2881                 munmap(s->data, s->size);
2882
2883         if (s->should_free || s->should_munmap) {
2884                 s->should_free = s->should_munmap = 0;
2885                 s->data = NULL;
2886         }
2887 }
2888
2889 void diff_free_filespec_data(struct diff_filespec *s)
2890 {
2891         diff_free_filespec_blob(s);
2892         free(s->cnt_data);
2893         s->cnt_data = NULL;
2894 }
2895
2896 static void prep_temp_blob(const char *path, struct diff_tempfile *temp,
2897                            void *blob,
2898                            unsigned long size,
2899                            const struct object_id *oid,
2900                            int mode)
2901 {
2902         int fd;
2903         struct strbuf buf = STRBUF_INIT;
2904         struct strbuf template = STRBUF_INIT;
2905         char *path_dup = xstrdup(path);
2906         const char *base = basename(path_dup);
2907
2908         /* Generate "XXXXXX_basename.ext" */
2909         strbuf_addstr(&template, "XXXXXX_");
2910         strbuf_addstr(&template, base);
2911
2912         fd = mks_tempfile_ts(&temp->tempfile, template.buf, strlen(base) + 1);
2913         if (fd < 0)
2914                 die_errno("unable to create temp-file");
2915         if (convert_to_working_tree(path,
2916                         (const char *)blob, (size_t)size, &buf)) {
2917                 blob = buf.buf;
2918                 size = buf.len;
2919         }
2920         if (write_in_full(fd, blob, size) != size)
2921                 die_errno("unable to write temp-file");
2922         close_tempfile(&temp->tempfile);
2923         temp->name = get_tempfile_path(&temp->tempfile);
2924         oid_to_hex_r(temp->hex, oid);
2925         xsnprintf(temp->mode, sizeof(temp->mode), "%06o", mode);
2926         strbuf_release(&buf);
2927         strbuf_release(&template);
2928         free(path_dup);
2929 }
2930
2931 static struct diff_tempfile *prepare_temp_file(const char *name,
2932                 struct diff_filespec *one)
2933 {
2934         struct diff_tempfile *temp = claim_diff_tempfile();
2935
2936         if (!DIFF_FILE_VALID(one)) {
2937         not_a_valid_file:
2938                 /* A '-' entry produces this for file-2, and
2939                  * a '+' entry produces this for file-1.
2940                  */
2941                 temp->name = "/dev/null";
2942                 xsnprintf(temp->hex, sizeof(temp->hex), ".");
2943                 xsnprintf(temp->mode, sizeof(temp->mode), ".");
2944                 return temp;
2945         }
2946
2947         if (!S_ISGITLINK(one->mode) &&
2948             (!one->oid_valid ||
2949              reuse_worktree_file(name, one->oid.hash, 1))) {
2950                 struct stat st;
2951                 if (lstat(name, &st) < 0) {
2952                         if (errno == ENOENT)
2953                                 goto not_a_valid_file;
2954                         die_errno("stat(%s)", name);
2955                 }
2956                 if (S_ISLNK(st.st_mode)) {
2957                         struct strbuf sb = STRBUF_INIT;
2958                         if (strbuf_readlink(&sb, name, st.st_size) < 0)
2959                                 die_errno("readlink(%s)", name);
2960                         prep_temp_blob(name, temp, sb.buf, sb.len,
2961                                        (one->oid_valid ?
2962                                         &one->oid : &null_oid),
2963                                        (one->oid_valid ?
2964                                         one->mode : S_IFLNK));
2965                         strbuf_release(&sb);
2966                 }
2967                 else {
2968                         /* we can borrow from the file in the work tree */
2969                         temp->name = name;
2970                         if (!one->oid_valid)
2971                                 sha1_to_hex_r(temp->hex, null_sha1);
2972                         else
2973                                 sha1_to_hex_r(temp->hex, one->oid.hash);
2974                         /* Even though we may sometimes borrow the
2975                          * contents from the work tree, we always want
2976                          * one->mode.  mode is trustworthy even when
2977                          * !(one->sha1_valid), as long as
2978                          * DIFF_FILE_VALID(one).
2979                          */
2980                         xsnprintf(temp->mode, sizeof(temp->mode), "%06o", one->mode);
2981                 }
2982                 return temp;
2983         }
2984         else {
2985                 if (diff_populate_filespec(one, 0))
2986                         die("cannot read data blob for %s", one->path);
2987                 prep_temp_blob(name, temp, one->data, one->size,
2988                                &one->oid, one->mode);
2989         }
2990         return temp;
2991 }
2992
2993 static void add_external_diff_name(struct argv_array *argv,
2994                                    const char *name,
2995                                    struct diff_filespec *df)
2996 {
2997         struct diff_tempfile *temp = prepare_temp_file(name, df);
2998         argv_array_push(argv, temp->name);
2999         argv_array_push(argv, temp->hex);
3000         argv_array_push(argv, temp->mode);
3001 }
3002
3003 /* An external diff command takes:
3004  *
3005  * diff-cmd name infile1 infile1-sha1 infile1-mode \
3006  *               infile2 infile2-sha1 infile2-mode [ rename-to ]
3007  *
3008  */
3009 static void run_external_diff(const char *pgm,
3010                               const char *name,
3011                               const char *other,
3012                               struct diff_filespec *one,
3013                               struct diff_filespec *two,
3014                               const char *xfrm_msg,
3015                               int complete_rewrite,
3016                               struct diff_options *o)
3017 {
3018         struct argv_array argv = ARGV_ARRAY_INIT;
3019         struct argv_array env = ARGV_ARRAY_INIT;
3020         struct diff_queue_struct *q = &diff_queued_diff;
3021
3022         argv_array_push(&argv, pgm);
3023         argv_array_push(&argv, name);
3024
3025         if (one && two) {
3026                 add_external_diff_name(&argv, name, one);
3027                 if (!other)
3028                         add_external_diff_name(&argv, name, two);
3029                 else {
3030                         add_external_diff_name(&argv, other, two);
3031                         argv_array_push(&argv, other);
3032                         argv_array_push(&argv, xfrm_msg);
3033                 }
3034         }
3035
3036         argv_array_pushf(&env, "GIT_DIFF_PATH_COUNTER=%d", ++o->diff_path_counter);
3037         argv_array_pushf(&env, "GIT_DIFF_PATH_TOTAL=%d", q->nr);
3038
3039         if (run_command_v_opt_cd_env(argv.argv, RUN_USING_SHELL, NULL, env.argv))
3040                 die(_("external diff died, stopping at %s"), name);
3041
3042         remove_tempfile();
3043         argv_array_clear(&argv);
3044         argv_array_clear(&env);
3045 }
3046
3047 static int similarity_index(struct diff_filepair *p)
3048 {
3049         return p->score * 100 / MAX_SCORE;
3050 }
3051
3052 static void fill_metainfo(struct strbuf *msg,
3053                           const char *name,
3054                           const char *other,
3055                           struct diff_filespec *one,
3056                           struct diff_filespec *two,
3057                           struct diff_options *o,
3058                           struct diff_filepair *p,
3059                           int *must_show_header,
3060                           int use_color)
3061 {
3062         const char *set = diff_get_color(use_color, DIFF_METAINFO);
3063         const char *reset = diff_get_color(use_color, DIFF_RESET);
3064         const char *line_prefix = diff_line_prefix(o);
3065
3066         *must_show_header = 1;
3067         strbuf_init(msg, PATH_MAX * 2 + 300);
3068         switch (p->status) {
3069         case DIFF_STATUS_COPIED:
3070                 strbuf_addf(msg, "%s%ssimilarity index %d%%",
3071                             line_prefix, set, similarity_index(p));
3072                 strbuf_addf(msg, "%s\n%s%scopy from ",
3073                             reset,  line_prefix, set);
3074                 quote_c_style(name, msg, NULL, 0);
3075                 strbuf_addf(msg, "%s\n%s%scopy to ", reset, line_prefix, set);
3076                 quote_c_style(other, msg, NULL, 0);
3077                 strbuf_addf(msg, "%s\n", reset);
3078                 break;
3079         case DIFF_STATUS_RENAMED:
3080                 strbuf_addf(msg, "%s%ssimilarity index %d%%",
3081                             line_prefix, set, similarity_index(p));
3082                 strbuf_addf(msg, "%s\n%s%srename from ",
3083                             reset, line_prefix, set);
3084                 quote_c_style(name, msg, NULL, 0);
3085                 strbuf_addf(msg, "%s\n%s%srename to ",
3086                             reset, line_prefix, set);
3087                 quote_c_style(other, msg, NULL, 0);
3088                 strbuf_addf(msg, "%s\n", reset);
3089                 break;
3090         case DIFF_STATUS_MODIFIED:
3091                 if (p->score) {
3092                         strbuf_addf(msg, "%s%sdissimilarity index %d%%%s\n",
3093                                     line_prefix,
3094                                     set, similarity_index(p), reset);
3095                         break;
3096                 }
3097                 /* fallthru */
3098         default:
3099                 *must_show_header = 0;
3100         }
3101         if (one && two && oidcmp(&one->oid, &two->oid)) {
3102                 int abbrev = DIFF_OPT_TST(o, FULL_INDEX) ? 40 : DEFAULT_ABBREV;
3103
3104                 if (DIFF_OPT_TST(o, BINARY)) {
3105                         mmfile_t mf;
3106                         if ((!fill_mmfile(&mf, one) && diff_filespec_is_binary(one)) ||
3107                             (!fill_mmfile(&mf, two) && diff_filespec_is_binary(two)))
3108                                 abbrev = 40;
3109                 }
3110                 strbuf_addf(msg, "%s%sindex %s..", line_prefix, set,
3111                             find_unique_abbrev(one->oid.hash, abbrev));
3112                 strbuf_addstr(msg, find_unique_abbrev(two->oid.hash, abbrev));
3113                 if (one->mode == two->mode)
3114                         strbuf_addf(msg, " %06o", one->mode);
3115                 strbuf_addf(msg, "%s\n", reset);
3116         }
3117 }
3118
3119 static void run_diff_cmd(const char *pgm,
3120                          const char *name,
3121                          const char *other,
3122                          const char *attr_path,
3123                          struct diff_filespec *one,
3124                          struct diff_filespec *two,
3125                          struct strbuf *msg,
3126                          struct diff_options *o,
3127                          struct diff_filepair *p)
3128 {
3129         const char *xfrm_msg = NULL;
3130         int complete_rewrite = (p->status == DIFF_STATUS_MODIFIED) && p->score;
3131         int must_show_header = 0;
3132
3133
3134         if (DIFF_OPT_TST(o, ALLOW_EXTERNAL)) {
3135                 struct userdiff_driver *drv = userdiff_find_by_path(attr_path);
3136                 if (drv && drv->external)
3137                         pgm = drv->external;
3138         }
3139
3140         if (msg) {
3141                 /*
3142                  * don't use colors when the header is intended for an
3143                  * external diff driver
3144                  */
3145                 fill_metainfo(msg, name, other, one, two, o, p,
3146                               &must_show_header,
3147                               want_color(o->use_color) && !pgm);
3148                 xfrm_msg = msg->len ? msg->buf : NULL;
3149         }
3150
3151         if (pgm) {
3152                 run_external_diff(pgm, name, other, one, two, xfrm_msg,
3153                                   complete_rewrite, o);
3154                 return;
3155         }
3156         if (one && two)
3157                 builtin_diff(name, other ? other : name,
3158                              one, two, xfrm_msg, must_show_header,
3159                              o, complete_rewrite);
3160         else
3161                 fprintf(o->file, "* Unmerged path %s\n", name);
3162 }
3163
3164 static void diff_fill_sha1_info(struct diff_filespec *one)
3165 {
3166         if (DIFF_FILE_VALID(one)) {
3167                 if (!one->oid_valid) {
3168                         struct stat st;
3169                         if (one->is_stdin) {
3170                                 oidclr(&one->oid);
3171                                 return;
3172                         }
3173                         if (lstat(one->path, &st) < 0)
3174                                 die_errno("stat '%s'", one->path);
3175                         if (index_path(one->oid.hash, one->path, &st, 0))
3176                                 die("cannot hash %s", one->path);
3177                 }
3178         }
3179         else
3180                 oidclr(&one->oid);
3181 }
3182
3183 static void strip_prefix(int prefix_length, const char **namep, const char **otherp)
3184 {
3185         /* Strip the prefix but do not molest /dev/null and absolute paths */
3186         if (*namep && **namep != '/') {
3187                 *namep += prefix_length;
3188                 if (**namep == '/')
3189                         ++*namep;
3190         }
3191         if (*otherp && **otherp != '/') {
3192                 *otherp += prefix_length;
3193                 if (**otherp == '/')
3194                         ++*otherp;
3195         }
3196 }
3197
3198 static void run_diff(struct diff_filepair *p, struct diff_options *o)
3199 {
3200         const char *pgm = external_diff();
3201         struct strbuf msg;
3202         struct diff_filespec *one = p->one;
3203         struct diff_filespec *two = p->two;
3204         const char *name;
3205         const char *other;
3206         const char *attr_path;
3207
3208         name  = p->one->path;
3209         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
3210         attr_path = name;
3211         if (o->prefix_length)
3212                 strip_prefix(o->prefix_length, &name, &other);
3213
3214         if (!DIFF_OPT_TST(o, ALLOW_EXTERNAL))
3215                 pgm = NULL;
3216
3217         if (DIFF_PAIR_UNMERGED(p)) {
3218                 run_diff_cmd(pgm, name, NULL, attr_path,
3219                              NULL, NULL, NULL, o, p);
3220                 return;
3221         }
3222
3223         diff_fill_sha1_info(one);
3224         diff_fill_sha1_info(two);
3225
3226         if (!pgm &&
3227             DIFF_FILE_VALID(one) && DIFF_FILE_VALID(two) &&
3228             (S_IFMT & one->mode) != (S_IFMT & two->mode)) {
3229                 /*
3230                  * a filepair that changes between file and symlink
3231                  * needs to be split into deletion and creation.
3232                  */
3233                 struct diff_filespec *null = alloc_filespec(two->path);
3234                 run_diff_cmd(NULL, name, other, attr_path,
3235                              one, null, &msg, o, p);
3236                 free(null);
3237                 strbuf_release(&msg);
3238
3239                 null = alloc_filespec(one->path);
3240                 run_diff_cmd(NULL, name, other, attr_path,
3241                              null, two, &msg, o, p);
3242                 free(null);
3243         }
3244         else
3245                 run_diff_cmd(pgm, name, other, attr_path,
3246                              one, two, &msg, o, p);
3247
3248         strbuf_release(&msg);
3249 }
3250
3251 static void run_diffstat(struct diff_filepair *p, struct diff_options *o,
3252                          struct diffstat_t *diffstat)
3253 {
3254         const char *name;
3255         const char *other;
3256
3257         if (DIFF_PAIR_UNMERGED(p)) {
3258                 /* unmerged */
3259                 builtin_diffstat(p->one->path, NULL, NULL, NULL, diffstat, o, p);
3260                 return;
3261         }
3262
3263         name = p->one->path;
3264         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
3265
3266         if (o->prefix_length)
3267                 strip_prefix(o->prefix_length, &name, &other);
3268
3269         diff_fill_sha1_info(p->one);
3270         diff_fill_sha1_info(p->two);
3271
3272         builtin_diffstat(name, other, p->one, p->two, diffstat, o, p);
3273 }
3274
3275 static void run_checkdiff(struct diff_filepair *p, struct diff_options *o)
3276 {
3277         const char *name;
3278         const char *other;
3279         const char *attr_path;
3280
3281         if (DIFF_PAIR_UNMERGED(p)) {
3282                 /* unmerged */
3283                 return;
3284         }
3285
3286         name = p->one->path;
3287         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
3288         attr_path = other ? other : name;
3289
3290         if (o->prefix_length)
3291                 strip_prefix(o->prefix_length, &name, &other);
3292
3293         diff_fill_sha1_info(p->one);
3294         diff_fill_sha1_info(p->two);
3295
3296         builtin_checkdiff(name, other, attr_path, p->one, p->two, o);
3297 }
3298
3299 void diff_setup(struct diff_options *options)
3300 {
3301         memcpy(options, &default_diff_options, sizeof(*options));
3302
3303         options->file = stdout;
3304
3305         options->line_termination = '\n';
3306         options->break_opt = -1;
3307         options->rename_limit = -1;
3308         options->dirstat_permille = diff_dirstat_permille_default;
3309         options->context = diff_context_default;
3310         options->ws_error_highlight = WSEH_NEW;
3311         DIFF_OPT_SET(options, RENAME_EMPTY);
3312
3313         /* pathchange left =NULL by default */
3314         options->change = diff_change;
3315         options->add_remove = diff_addremove;
3316         options->use_color = diff_use_color_default;
3317         options->detect_rename = diff_detect_rename_default;
3318         options->xdl_opts |= diff_algorithm;
3319         if (diff_indent_heuristic)
3320                 DIFF_XDL_SET(options, INDENT_HEURISTIC);
3321         else if (diff_compaction_heuristic)
3322                 DIFF_XDL_SET(options, COMPACTION_HEURISTIC);
3323
3324         options->orderfile = diff_order_file_cfg;
3325
3326         if (diff_no_prefix) {
3327                 options->a_prefix = options->b_prefix = "";
3328         } else if (!diff_mnemonic_prefix) {
3329                 options->a_prefix = "a/";
3330                 options->b_prefix = "b/";
3331         }
3332 }
3333
3334 void diff_setup_done(struct diff_options *options)
3335 {
3336         int count = 0;
3337
3338         if (options->set_default)
3339                 options->set_default(options);
3340
3341         if (options->output_format & DIFF_FORMAT_NAME)
3342                 count++;
3343         if (options->output_format & DIFF_FORMAT_NAME_STATUS)
3344                 count++;
3345         if (options->output_format & DIFF_FORMAT_CHECKDIFF)
3346                 count++;
3347         if (options->output_format & DIFF_FORMAT_NO_OUTPUT)
3348                 count++;
3349         if (count > 1)
3350                 die(_("--name-only, --name-status, --check and -s are mutually exclusive"));
3351
3352         /*
3353          * Most of the time we can say "there are changes"
3354          * only by checking if there are changed paths, but
3355          * --ignore-whitespace* options force us to look
3356          * inside contents.
3357          */
3358
3359         if (DIFF_XDL_TST(options, IGNORE_WHITESPACE) ||
3360             DIFF_XDL_TST(options, IGNORE_WHITESPACE_CHANGE) ||
3361             DIFF_XDL_TST(options, IGNORE_WHITESPACE_AT_EOL))
3362                 DIFF_OPT_SET(options, DIFF_FROM_CONTENTS);
3363         else
3364                 DIFF_OPT_CLR(options, DIFF_FROM_CONTENTS);
3365
3366         if (DIFF_OPT_TST(options, FIND_COPIES_HARDER))
3367                 options->detect_rename = DIFF_DETECT_COPY;
3368
3369         if (!DIFF_OPT_TST(options, RELATIVE_NAME))
3370                 options->prefix = NULL;
3371         if (options->prefix)
3372                 options->prefix_length = strlen(options->prefix);
3373         else
3374                 options->prefix_length = 0;
3375
3376         if (options->output_format & (DIFF_FORMAT_NAME |
3377                                       DIFF_FORMAT_NAME_STATUS |
3378                                       DIFF_FORMAT_CHECKDIFF |
3379                                       DIFF_FORMAT_NO_OUTPUT))
3380                 options->output_format &= ~(DIFF_FORMAT_RAW |
3381                                             DIFF_FORMAT_NUMSTAT |
3382                                             DIFF_FORMAT_DIFFSTAT |
3383                                             DIFF_FORMAT_SHORTSTAT |
3384                                             DIFF_FORMAT_DIRSTAT |
3385                                             DIFF_FORMAT_SUMMARY |
3386                                             DIFF_FORMAT_PATCH);
3387
3388         /*
3389          * These cases always need recursive; we do not drop caller-supplied
3390          * recursive bits for other formats here.
3391          */
3392         if (options->output_format & (DIFF_FORMAT_PATCH |
3393                                       DIFF_FORMAT_NUMSTAT |
3394                                       DIFF_FORMAT_DIFFSTAT |
3395                                       DIFF_FORMAT_SHORTSTAT |
3396                                       DIFF_FORMAT_DIRSTAT |
3397                                       DIFF_FORMAT_SUMMARY |
3398                                       DIFF_FORMAT_CHECKDIFF))
3399                 DIFF_OPT_SET(options, RECURSIVE);
3400         /*
3401          * Also pickaxe would not work very well if you do not say recursive
3402          */
3403         if (options->pickaxe)
3404                 DIFF_OPT_SET(options, RECURSIVE);
3405         /*
3406          * When patches are generated, submodules diffed against the work tree
3407          * must be checked for dirtiness too so it can be shown in the output
3408          */
3409         if (options->output_format & DIFF_FORMAT_PATCH)
3410                 DIFF_OPT_SET(options, DIRTY_SUBMODULES);
3411
3412         if (options->detect_rename && options->rename_limit < 0)
3413                 options->rename_limit = diff_rename_limit_default;
3414         if (options->setup & DIFF_SETUP_USE_CACHE) {
3415                 if (!active_cache)
3416                         /* read-cache does not die even when it fails
3417                          * so it is safe for us to do this here.  Also
3418                          * it does not smudge active_cache or active_nr
3419                          * when it fails, so we do not have to worry about
3420                          * cleaning it up ourselves either.
3421                          */
3422                         read_cache();
3423         }
3424         if (options->abbrev <= 0 || 40 < options->abbrev)
3425                 options->abbrev = 40; /* full */
3426
3427         /*
3428          * It does not make sense to show the first hit we happened
3429          * to have found.  It does not make sense not to return with
3430          * exit code in such a case either.
3431          */
3432         if (DIFF_OPT_TST(options, QUICK)) {
3433                 options->output_format = DIFF_FORMAT_NO_OUTPUT;
3434                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
3435         }
3436
3437         options->diff_path_counter = 0;
3438
3439         if (DIFF_OPT_TST(options, FOLLOW_RENAMES) && options->pathspec.nr != 1)
3440                 die(_("--follow requires exactly one pathspec"));
3441 }
3442
3443 static int opt_arg(const char *arg, int arg_short, const char *arg_long, int *val)
3444 {
3445         char c, *eq;
3446         int len;
3447
3448         if (*arg != '-')
3449                 return 0;
3450         c = *++arg;
3451         if (!c)
3452                 return 0;
3453         if (c == arg_short) {
3454                 c = *++arg;
3455                 if (!c)
3456                         return 1;
3457                 if (val && isdigit(c)) {
3458                         char *end;
3459                         int n = strtoul(arg, &end, 10);
3460                         if (*end)
3461                                 return 0;
3462                         *val = n;
3463                         return 1;
3464                 }
3465                 return 0;
3466         }
3467         if (c != '-')
3468                 return 0;
3469         arg++;
3470         eq = strchrnul(arg, '=');
3471         len = eq - arg;
3472         if (!len || strncmp(arg, arg_long, len))
3473                 return 0;
3474         if (*eq) {
3475                 int n;
3476                 char *end;
3477                 if (!isdigit(*++eq))
3478                         return 0;
3479                 n = strtoul(eq, &end, 10);
3480                 if (*end)
3481                         return 0;
3482                 *val = n;
3483         }
3484         return 1;
3485 }
3486
3487 static int diff_scoreopt_parse(const char *opt);
3488
3489 static inline int short_opt(char opt, const char **argv,
3490                             const char **optarg)
3491 {
3492         const char *arg = argv[0];
3493         if (arg[0] != '-' || arg[1] != opt)
3494                 return 0;
3495         if (arg[2] != '\0') {
3496                 *optarg = arg + 2;
3497                 return 1;
3498         }
3499         if (!argv[1])
3500                 die("Option '%c' requires a value", opt);
3501         *optarg = argv[1];
3502         return 2;
3503 }
3504
3505 int parse_long_opt(const char *opt, const char **argv,
3506                    const char **optarg)
3507 {
3508         const char *arg = argv[0];
3509         if (!skip_prefix(arg, "--", &arg))
3510                 return 0;
3511         if (!skip_prefix(arg, opt, &arg))
3512                 return 0;
3513         if (*arg == '=') { /* stuck form: --option=value */
3514                 *optarg = arg + 1;
3515                 return 1;
3516         }
3517         if (*arg != '\0')
3518                 return 0;
3519         /* separate form: --option value */
3520         if (!argv[1])
3521                 die("Option '--%s' requires a value", opt);
3522         *optarg = argv[1];
3523         return 2;
3524 }
3525
3526 static int stat_opt(struct diff_options *options, const char **av)
3527 {
3528         const char *arg = av[0];
3529         char *end;
3530         int width = options->stat_width;
3531         int name_width = options->stat_name_width;
3532         int graph_width = options->stat_graph_width;
3533         int count = options->stat_count;
3534         int argcount = 1;
3535
3536         if (!skip_prefix(arg, "--stat", &arg))
3537                 die("BUG: stat option does not begin with --stat: %s", arg);
3538         end = (char *)arg;
3539
3540         switch (*arg) {
3541         case '-':
3542                 if (skip_prefix(arg, "-width", &arg)) {
3543                         if (*arg == '=')
3544                                 width = strtoul(arg + 1, &end, 10);
3545                         else if (!*arg && !av[1])
3546                                 die_want_option("--stat-width");
3547                         else if (!*arg) {
3548                                 width = strtoul(av[1], &end, 10);
3549                                 argcount = 2;
3550                         }
3551                 } else if (skip_prefix(arg, "-name-width", &arg)) {
3552                         if (*arg == '=')
3553                                 name_width = strtoul(arg + 1, &end, 10);
3554                         else if (!*arg && !av[1])
3555                                 die_want_option("--stat-name-width");
3556                         else if (!*arg) {
3557                                 name_width = strtoul(av[1], &end, 10);
3558                                 argcount = 2;
3559                         }
3560                 } else if (skip_prefix(arg, "-graph-width", &arg)) {
3561                         if (*arg == '=')
3562                                 graph_width = strtoul(arg + 1, &end, 10);
3563                         else if (!*arg && !av[1])
3564                                 die_want_option("--stat-graph-width");
3565                         else if (!*arg) {
3566                                 graph_width = strtoul(av[1], &end, 10);
3567                                 argcount = 2;
3568                         }
3569                 } else if (skip_prefix(arg, "-count", &arg)) {
3570                         if (*arg == '=')
3571                                 count = strtoul(arg + 1, &end, 10);
3572                         else if (!*arg && !av[1])
3573                                 die_want_option("--stat-count");
3574                         else if (!*arg) {
3575                                 count = strtoul(av[1], &end, 10);
3576                                 argcount = 2;
3577                         }
3578                 }
3579                 break;
3580         case '=':
3581                 width = strtoul(arg+1, &end, 10);
3582                 if (*end == ',')
3583                         name_width = strtoul(end+1, &end, 10);
3584                 if (*end == ',')
3585                         count = strtoul(end+1, &end, 10);
3586         }
3587
3588         /* Important! This checks all the error cases! */
3589         if (*end)
3590                 return 0;
3591         options->output_format |= DIFF_FORMAT_DIFFSTAT;
3592         options->stat_name_width = name_width;
3593         options->stat_graph_width = graph_width;
3594         options->stat_width = width;
3595         options->stat_count = count;
3596         return argcount;
3597 }
3598
3599 static int parse_dirstat_opt(struct diff_options *options, const char *params)
3600 {
3601         struct strbuf errmsg = STRBUF_INIT;
3602         if (parse_dirstat_params(options, params, &errmsg))
3603                 die(_("Failed to parse --dirstat/-X option parameter:\n%s"),
3604                     errmsg.buf);
3605         strbuf_release(&errmsg);
3606         /*
3607          * The caller knows a dirstat-related option is given from the command
3608          * line; allow it to say "return this_function();"
3609          */
3610         options->output_format |= DIFF_FORMAT_DIRSTAT;
3611         return 1;
3612 }
3613
3614 static int parse_submodule_opt(struct diff_options *options, const char *value)
3615 {
3616         if (parse_submodule_params(options, value))
3617                 die(_("Failed to parse --submodule option parameter: '%s'"),
3618                         value);
3619         return 1;
3620 }
3621
3622 static const char diff_status_letters[] = {
3623         DIFF_STATUS_ADDED,
3624         DIFF_STATUS_COPIED,
3625         DIFF_STATUS_DELETED,
3626         DIFF_STATUS_MODIFIED,
3627         DIFF_STATUS_RENAMED,
3628         DIFF_STATUS_TYPE_CHANGED,
3629         DIFF_STATUS_UNKNOWN,
3630         DIFF_STATUS_UNMERGED,
3631         DIFF_STATUS_FILTER_AON,
3632         DIFF_STATUS_FILTER_BROKEN,
3633         '\0',
3634 };
3635
3636 static unsigned int filter_bit['Z' + 1];
3637
3638 static void prepare_filter_bits(void)
3639 {
3640         int i;
3641
3642         if (!filter_bit[DIFF_STATUS_ADDED]) {
3643                 for (i = 0; diff_status_letters[i]; i++)
3644                         filter_bit[(int) diff_status_letters[i]] = (1 << i);
3645         }
3646 }
3647
3648 static unsigned filter_bit_tst(char status, const struct diff_options *opt)
3649 {
3650         return opt->filter & filter_bit[(int) status];
3651 }
3652
3653 static int parse_diff_filter_opt(const char *optarg, struct diff_options *opt)
3654 {
3655         int i, optch;
3656
3657         prepare_filter_bits();
3658
3659         /*
3660          * If there is a negation e.g. 'd' in the input, and we haven't
3661          * initialized the filter field with another --diff-filter, start
3662          * from full set of bits, except for AON.
3663          */
3664         if (!opt->filter) {
3665                 for (i = 0; (optch = optarg[i]) != '\0'; i++) {
3666                         if (optch < 'a' || 'z' < optch)
3667                                 continue;
3668                         opt->filter = (1 << (ARRAY_SIZE(diff_status_letters) - 1)) - 1;
3669                         opt->filter &= ~filter_bit[DIFF_STATUS_FILTER_AON];
3670                         break;
3671                 }
3672         }
3673
3674         for (i = 0; (optch = optarg[i]) != '\0'; i++) {
3675                 unsigned int bit;
3676                 int negate;
3677
3678                 if ('a' <= optch && optch <= 'z') {
3679                         negate = 1;
3680                         optch = toupper(optch);
3681                 } else {
3682                         negate = 0;
3683                 }
3684
3685                 bit = (0 <= optch && optch <= 'Z') ? filter_bit[optch] : 0;
3686                 if (!bit)
3687                         return optarg[i];
3688                 if (negate)
3689                         opt->filter &= ~bit;
3690                 else
3691                         opt->filter |= bit;
3692         }
3693         return 0;
3694 }
3695
3696 static void enable_patch_output(int *fmt) {
3697         *fmt &= ~DIFF_FORMAT_NO_OUTPUT;
3698         *fmt |= DIFF_FORMAT_PATCH;
3699 }
3700
3701 static int parse_one_token(const char **arg, const char *token)
3702 {
3703         const char *rest;
3704         if (skip_prefix(*arg, token, &rest) && (!*rest || *rest == ',')) {
3705                 *arg = rest;
3706                 return 1;
3707         }
3708         return 0;
3709 }
3710
3711 static int parse_ws_error_highlight(struct diff_options *opt, const char *arg)
3712 {
3713         const char *orig_arg = arg;
3714         unsigned val = 0;
3715         while (*arg) {
3716                 if (parse_one_token(&arg, "none"))
3717                         val = 0;
3718                 else if (parse_one_token(&arg, "default"))
3719                         val = WSEH_NEW;
3720                 else if (parse_one_token(&arg, "all"))
3721                         val = WSEH_NEW | WSEH_OLD | WSEH_CONTEXT;
3722                 else if (parse_one_token(&arg, "new"))
3723                         val |= WSEH_NEW;
3724                 else if (parse_one_token(&arg, "old"))
3725                         val |= WSEH_OLD;
3726                 else if (parse_one_token(&arg, "context"))
3727                         val |= WSEH_CONTEXT;
3728                 else {
3729                         error("unknown value after ws-error-highlight=%.*s",
3730                               (int)(arg - orig_arg), orig_arg);
3731                         return 0;
3732                 }
3733                 if (*arg)
3734                         arg++;
3735         }
3736         opt->ws_error_highlight = val;
3737         return 1;
3738 }
3739
3740 int diff_opt_parse(struct diff_options *options,
3741                    const char **av, int ac, const char *prefix)
3742 {
3743         const char *arg = av[0];
3744         const char *optarg;
3745         int argcount;
3746
3747         if (!prefix)
3748                 prefix = "";
3749
3750         /* Output format options */
3751         if (!strcmp(arg, "-p") || !strcmp(arg, "-u") || !strcmp(arg, "--patch")
3752             || opt_arg(arg, 'U', "unified", &options->context))
3753                 enable_patch_output(&options->output_format);
3754         else if (!strcmp(arg, "--raw"))
3755                 options->output_format |= DIFF_FORMAT_RAW;
3756         else if (!strcmp(arg, "--patch-with-raw")) {
3757                 enable_patch_output(&options->output_format);
3758                 options->output_format |= DIFF_FORMAT_RAW;
3759         } else if (!strcmp(arg, "--numstat"))
3760                 options->output_format |= DIFF_FORMAT_NUMSTAT;
3761         else if (!strcmp(arg, "--shortstat"))
3762                 options->output_format |= DIFF_FORMAT_SHORTSTAT;
3763         else if (!strcmp(arg, "-X") || !strcmp(arg, "--dirstat"))
3764                 return parse_dirstat_opt(options, "");
3765         else if (skip_prefix(arg, "-X", &arg))
3766                 return parse_dirstat_opt(options, arg);
3767         else if (skip_prefix(arg, "--dirstat=", &arg))
3768                 return parse_dirstat_opt(options, arg);
3769         else if (!strcmp(arg, "--cumulative"))
3770                 return parse_dirstat_opt(options, "cumulative");
3771         else if (!strcmp(arg, "--dirstat-by-file"))
3772                 return parse_dirstat_opt(options, "files");
3773         else if (skip_prefix(arg, "--dirstat-by-file=", &arg)) {
3774                 parse_dirstat_opt(options, "files");
3775                 return parse_dirstat_opt(options, arg);
3776         }
3777         else if (!strcmp(arg, "--check"))
3778                 options->output_format |= DIFF_FORMAT_CHECKDIFF;
3779         else if (!strcmp(arg, "--summary"))
3780                 options->output_format |= DIFF_FORMAT_SUMMARY;
3781         else if (!strcmp(arg, "--patch-with-stat")) {
3782                 enable_patch_output(&options->output_format);
3783                 options->output_format |= DIFF_FORMAT_DIFFSTAT;
3784         } else if (!strcmp(arg, "--name-only"))
3785                 options->output_format |= DIFF_FORMAT_NAME;
3786         else if (!strcmp(arg, "--name-status"))
3787                 options->output_format |= DIFF_FORMAT_NAME_STATUS;
3788         else if (!strcmp(arg, "-s") || !strcmp(arg, "--no-patch"))
3789                 options->output_format |= DIFF_FORMAT_NO_OUTPUT;
3790         else if (starts_with(arg, "--stat"))
3791                 /* --stat, --stat-width, --stat-name-width, or --stat-count */
3792                 return stat_opt(options, av);
3793
3794         /* renames options */
3795         else if (starts_with(arg, "-B") || starts_with(arg, "--break-rewrites=") ||
3796                  !strcmp(arg, "--break-rewrites")) {
3797                 if ((options->break_opt = diff_scoreopt_parse(arg)) == -1)
3798                         return error("invalid argument to -B: %s", arg+2);
3799         }
3800         else if (starts_with(arg, "-M") || starts_with(arg, "--find-renames=") ||
3801                  !strcmp(arg, "--find-renames")) {
3802                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
3803                         return error("invalid argument to -M: %s", arg+2);
3804                 options->detect_rename = DIFF_DETECT_RENAME;
3805         }
3806         else if (!strcmp(arg, "-D") || !strcmp(arg, "--irreversible-delete")) {
3807                 options->irreversible_delete = 1;
3808         }
3809         else if (starts_with(arg, "-C") || starts_with(arg, "--find-copies=") ||
3810                  !strcmp(arg, "--find-copies")) {
3811                 if (options->detect_rename == DIFF_DETECT_COPY)
3812                         DIFF_OPT_SET(options, FIND_COPIES_HARDER);
3813                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
3814                         return error("invalid argument to -C: %s", arg+2);
3815                 options->detect_rename = DIFF_DETECT_COPY;
3816         }
3817         else if (!strcmp(arg, "--no-renames"))
3818                 options->detect_rename = 0;
3819         else if (!strcmp(arg, "--rename-empty"))
3820                 DIFF_OPT_SET(options, RENAME_EMPTY);
3821         else if (!strcmp(arg, "--no-rename-empty"))
3822                 DIFF_OPT_CLR(options, RENAME_EMPTY);
3823         else if (!strcmp(arg, "--relative"))
3824                 DIFF_OPT_SET(options, RELATIVE_NAME);
3825         else if (skip_prefix(arg, "--relative=", &arg)) {
3826                 DIFF_OPT_SET(options, RELATIVE_NAME);
3827                 options->prefix = arg;
3828         }
3829
3830         /* xdiff options */
3831         else if (!strcmp(arg, "--minimal"))
3832                 DIFF_XDL_SET(options, NEED_MINIMAL);
3833         else if (!strcmp(arg, "--no-minimal"))
3834                 DIFF_XDL_CLR(options, NEED_MINIMAL);
3835         else if (!strcmp(arg, "-w") || !strcmp(arg, "--ignore-all-space"))
3836                 DIFF_XDL_SET(options, IGNORE_WHITESPACE);
3837         else if (!strcmp(arg, "-b") || !strcmp(arg, "--ignore-space-change"))
3838                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_CHANGE);
3839         else if (!strcmp(arg, "--ignore-space-at-eol"))
3840                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_AT_EOL);
3841         else if (!strcmp(arg, "--ignore-blank-lines"))
3842                 DIFF_XDL_SET(options, IGNORE_BLANK_LINES);
3843         else if (!strcmp(arg, "--indent-heuristic")) {
3844                 DIFF_XDL_SET(options, INDENT_HEURISTIC);
3845                 DIFF_XDL_CLR(options, COMPACTION_HEURISTIC);
3846         } else if (!strcmp(arg, "--no-indent-heuristic"))
3847                 DIFF_XDL_CLR(options, INDENT_HEURISTIC);
3848         else if (!strcmp(arg, "--compaction-heuristic")) {
3849                 DIFF_XDL_SET(options, COMPACTION_HEURISTIC);
3850                 DIFF_XDL_CLR(options, INDENT_HEURISTIC);
3851         } else if (!strcmp(arg, "--no-compaction-heuristic"))
3852                 DIFF_XDL_CLR(options, COMPACTION_HEURISTIC);
3853         else if (!strcmp(arg, "--patience"))
3854                 options->xdl_opts = DIFF_WITH_ALG(options, PATIENCE_DIFF);
3855         else if (!strcmp(arg, "--histogram"))
3856                 options->xdl_opts = DIFF_WITH_ALG(options, HISTOGRAM_DIFF);
3857         else if ((argcount = parse_long_opt("diff-algorithm", av, &optarg))) {
3858                 long value = parse_algorithm_value(optarg);
3859                 if (value < 0)
3860                         return error("option diff-algorithm accepts \"myers\", "
3861                                      "\"minimal\", \"patience\" and \"histogram\"");
3862                 /* clear out previous settings */
3863                 DIFF_XDL_CLR(options, NEED_MINIMAL);
3864                 options->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3865                 options->xdl_opts |= value;
3866                 return argcount;
3867         }
3868
3869         /* flags options */
3870         else if (!strcmp(arg, "--binary")) {
3871                 enable_patch_output(&options->output_format);
3872                 DIFF_OPT_SET(options, BINARY);
3873         }
3874         else if (!strcmp(arg, "--full-index"))
3875                 DIFF_OPT_SET(options, FULL_INDEX);
3876         else if (!strcmp(arg, "-a") || !strcmp(arg, "--text"))
3877                 DIFF_OPT_SET(options, TEXT);
3878         else if (!strcmp(arg, "-R"))
3879                 DIFF_OPT_SET(options, REVERSE_DIFF);
3880         else if (!strcmp(arg, "--find-copies-harder"))
3881                 DIFF_OPT_SET(options, FIND_COPIES_HARDER);
3882         else if (!strcmp(arg, "--follow"))
3883                 DIFF_OPT_SET(options, FOLLOW_RENAMES);
3884         else if (!strcmp(arg, "--no-follow")) {
3885                 DIFF_OPT_CLR(options, FOLLOW_RENAMES);
3886                 DIFF_OPT_CLR(options, DEFAULT_FOLLOW_RENAMES);
3887         } else if (!strcmp(arg, "--color"))
3888                 options->use_color = 1;
3889         else if (skip_prefix(arg, "--color=", &arg)) {
3890                 int value = git_config_colorbool(NULL, arg);
3891                 if (value < 0)
3892                         return error("option `color' expects \"always\", \"auto\", or \"never\"");
3893                 options->use_color = value;
3894         }
3895         else if (!strcmp(arg, "--no-color"))
3896                 options->use_color = 0;
3897         else if (!strcmp(arg, "--color-words")) {
3898                 options->use_color = 1;
3899                 options->word_diff = DIFF_WORDS_COLOR;
3900         }
3901         else if (skip_prefix(arg, "--color-words=", &arg)) {
3902                 options->use_color = 1;
3903                 options->word_diff = DIFF_WORDS_COLOR;
3904                 options->word_regex = arg;
3905         }
3906         else if (!strcmp(arg, "--word-diff")) {
3907                 if (options->word_diff == DIFF_WORDS_NONE)
3908                         options->word_diff = DIFF_WORDS_PLAIN;
3909         }
3910         else if (skip_prefix(arg, "--word-diff=", &arg)) {
3911                 if (!strcmp(arg, "plain"))
3912                         options->word_diff = DIFF_WORDS_PLAIN;
3913                 else if (!strcmp(arg, "color")) {
3914                         options->use_color = 1;
3915                         options->word_diff = DIFF_WORDS_COLOR;
3916                 }
3917                 else if (!strcmp(arg, "porcelain"))
3918                         options->word_diff = DIFF_WORDS_PORCELAIN;
3919                 else if (!strcmp(arg, "none"))
3920                         options->word_diff = DIFF_WORDS_NONE;
3921                 else
3922                         die("bad --word-diff argument: %s", arg);
3923         }
3924         else if ((argcount = parse_long_opt("word-diff-regex", av, &optarg))) {
3925                 if (options->word_diff == DIFF_WORDS_NONE)
3926                         options->word_diff = DIFF_WORDS_PLAIN;
3927                 options->word_regex = optarg;
3928                 return argcount;
3929         }
3930         else if (!strcmp(arg, "--exit-code"))
3931                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
3932         else if (!strcmp(arg, "--quiet"))
3933                 DIFF_OPT_SET(options, QUICK);
3934         else if (!strcmp(arg, "--ext-diff"))
3935                 DIFF_OPT_SET(options, ALLOW_EXTERNAL);
3936         else if (!strcmp(arg, "--no-ext-diff"))
3937                 DIFF_OPT_CLR(options, ALLOW_EXTERNAL);
3938         else if (!strcmp(arg, "--textconv"))
3939                 DIFF_OPT_SET(options, ALLOW_TEXTCONV);
3940         else if (!strcmp(arg, "--no-textconv"))
3941                 DIFF_OPT_CLR(options, ALLOW_TEXTCONV);
3942         else if (!strcmp(arg, "--ignore-submodules")) {
3943                 DIFF_OPT_SET(options, OVERRIDE_SUBMODULE_CONFIG);
3944                 handle_ignore_submodules_arg(options, "all");
3945         } else if (skip_prefix(arg, "--ignore-submodules=", &arg)) {
3946                 DIFF_OPT_SET(options, OVERRIDE_SUBMODULE_CONFIG);
3947                 handle_ignore_submodules_arg(options, arg);
3948         } else if (!strcmp(arg, "--submodule"))
3949                 options->submodule_format = DIFF_SUBMODULE_LOG;
3950         else if (skip_prefix(arg, "--submodule=", &arg))
3951                 return parse_submodule_opt(options, arg);
3952         else if (skip_prefix(arg, "--ws-error-highlight=", &arg))
3953                 return parse_ws_error_highlight(options, arg);
3954
3955         /* misc options */
3956         else if (!strcmp(arg, "-z"))
3957                 options->line_termination = 0;
3958         else if ((argcount = short_opt('l', av, &optarg))) {
3959                 options->rename_limit = strtoul(optarg, NULL, 10);
3960                 return argcount;
3961         }
3962         else if ((argcount = short_opt('S', av, &optarg))) {
3963                 options->pickaxe = optarg;
3964                 options->pickaxe_opts |= DIFF_PICKAXE_KIND_S;
3965                 return argcount;
3966         } else if ((argcount = short_opt('G', av, &optarg))) {
3967                 options->pickaxe = optarg;
3968                 options->pickaxe_opts |= DIFF_PICKAXE_KIND_G;
3969                 return argcount;
3970         }
3971         else if (!strcmp(arg, "--pickaxe-all"))
3972                 options->pickaxe_opts |= DIFF_PICKAXE_ALL;
3973         else if (!strcmp(arg, "--pickaxe-regex"))
3974                 options->pickaxe_opts |= DIFF_PICKAXE_REGEX;
3975         else if ((argcount = short_opt('O', av, &optarg))) {
3976                 const char *path = prefix_filename(prefix, strlen(prefix), optarg);
3977                 options->orderfile = xstrdup(path);
3978                 return argcount;
3979         }
3980         else if ((argcount = parse_long_opt("diff-filter", av, &optarg))) {
3981                 int offending = parse_diff_filter_opt(optarg, options);
3982                 if (offending)
3983                         die("unknown change class '%c' in --diff-filter=%s",
3984                             offending, optarg);
3985                 return argcount;
3986         }
3987         else if (!strcmp(arg, "--abbrev"))
3988                 options->abbrev = DEFAULT_ABBREV;
3989         else if (skip_prefix(arg, "--abbrev=", &arg)) {
3990                 options->abbrev = strtoul(arg, NULL, 10);
3991                 if (options->abbrev < MINIMUM_ABBREV)
3992                         options->abbrev = MINIMUM_ABBREV;
3993                 else if (40 < options->abbrev)
3994                         options->abbrev = 40;
3995         }
3996         else if ((argcount = parse_long_opt("src-prefix", av, &optarg))) {
3997                 options->a_prefix = optarg;
3998                 return argcount;
3999         }
4000         else if ((argcount = parse_long_opt("line-prefix", av, &optarg))) {
4001                 options->line_prefix = optarg;
4002                 options->line_prefix_length = strlen(options->line_prefix);
4003                 graph_setup_line_prefix(options);
4004                 return argcount;
4005         }
4006         else if ((argcount = parse_long_opt("dst-prefix", av, &optarg))) {
4007                 options->b_prefix = optarg;
4008                 return argcount;
4009         }
4010         else if (!strcmp(arg, "--no-prefix"))
4011                 options->a_prefix = options->b_prefix = "";
4012         else if (opt_arg(arg, '\0', "inter-hunk-context",
4013                          &options->interhunkcontext))
4014                 ;
4015         else if (!strcmp(arg, "-W"))
4016                 DIFF_OPT_SET(options, FUNCCONTEXT);
4017         else if (!strcmp(arg, "--function-context"))
4018                 DIFF_OPT_SET(options, FUNCCONTEXT);
4019         else if (!strcmp(arg, "--no-function-context"))
4020                 DIFF_OPT_CLR(options, FUNCCONTEXT);
4021         else if ((argcount = parse_long_opt("output", av, &optarg))) {
4022                 const char *path = prefix_filename(prefix, strlen(prefix), optarg);
4023                 options->file = fopen(path, "w");
4024                 if (!options->file)
4025                         die_errno("Could not open '%s'", path);
4026                 options->close_file = 1;
4027                 if (options->use_color != GIT_COLOR_ALWAYS)
4028                         options->use_color = GIT_COLOR_NEVER;
4029                 return argcount;
4030         } else
4031                 return 0;
4032         return 1;
4033 }
4034
4035 int parse_rename_score(const char **cp_p)
4036 {
4037         unsigned long num, scale;
4038         int ch, dot;
4039         const char *cp = *cp_p;
4040
4041         num = 0;
4042         scale = 1;
4043         dot = 0;
4044         for (;;) {
4045                 ch = *cp;
4046                 if ( !dot && ch == '.' ) {
4047                         scale = 1;
4048                         dot = 1;
4049                 } else if ( ch == '%' ) {
4050                         scale = dot ? scale*100 : 100;
4051                         cp++;   /* % is always at the end */
4052                         break;
4053                 } else if ( ch >= '0' && ch <= '9' ) {
4054                         if ( scale < 100000 ) {
4055                                 scale *= 10;
4056                                 num = (num*10) + (ch-'0');
4057                         }
4058                 } else {
4059                         break;
4060                 }
4061                 cp++;
4062         }
4063         *cp_p = cp;
4064
4065         /* user says num divided by scale and we say internally that
4066          * is MAX_SCORE * num / scale.
4067          */
4068         return (int)((num >= scale) ? MAX_SCORE : (MAX_SCORE * num / scale));
4069 }
4070
4071 static int diff_scoreopt_parse(const char *opt)
4072 {
4073         int opt1, opt2, cmd;
4074
4075         if (*opt++ != '-')
4076                 return -1;
4077         cmd = *opt++;
4078         if (cmd == '-') {
4079                 /* convert the long-form arguments into short-form versions */
4080                 if (skip_prefix(opt, "break-rewrites", &opt)) {
4081                         if (*opt == 0 || *opt++ == '=')
4082                                 cmd = 'B';
4083                 } else if (skip_prefix(opt, "find-copies", &opt)) {
4084                         if (*opt == 0 || *opt++ == '=')
4085                                 cmd = 'C';
4086                 } else if (skip_prefix(opt, "find-renames", &opt)) {
4087                         if (*opt == 0 || *opt++ == '=')
4088                                 cmd = 'M';
4089                 }
4090         }
4091         if (cmd != 'M' && cmd != 'C' && cmd != 'B')
4092                 return -1; /* that is not a -M, -C, or -B option */
4093
4094         opt1 = parse_rename_score(&opt);
4095         if (cmd != 'B')
4096                 opt2 = 0;
4097         else {
4098                 if (*opt == 0)
4099                         opt2 = 0;
4100                 else if (*opt != '/')
4101                         return -1; /* we expect -B80/99 or -B80 */
4102                 else {
4103                         opt++;
4104                         opt2 = parse_rename_score(&opt);
4105                 }
4106         }
4107         if (*opt != 0)
4108                 return -1;
4109         return opt1 | (opt2 << 16);
4110 }
4111
4112 struct diff_queue_struct diff_queued_diff;
4113
4114 void diff_q(struct diff_queue_struct *queue, struct diff_filepair *dp)
4115 {
4116         ALLOC_GROW(queue->queue, queue->nr + 1, queue->alloc);
4117         queue->queue[queue->nr++] = dp;
4118 }
4119
4120 struct diff_filepair *diff_queue(struct diff_queue_struct *queue,
4121                                  struct diff_filespec *one,
4122                                  struct diff_filespec *two)
4123 {
4124         struct diff_filepair *dp = xcalloc(1, sizeof(*dp));
4125         dp->one = one;
4126         dp->two = two;
4127         if (queue)
4128                 diff_q(queue, dp);
4129         return dp;
4130 }
4131
4132 void diff_free_filepair(struct diff_filepair *p)
4133 {
4134         free_filespec(p->one);
4135         free_filespec(p->two);
4136         free(p);
4137 }
4138
4139 /* This is different from find_unique_abbrev() in that
4140  * it stuffs the result with dots for alignment.
4141  */
4142 const char *diff_unique_abbrev(const unsigned char *sha1, int len)
4143 {
4144         int abblen;
4145         const char *abbrev;
4146         if (len == 40)
4147                 return sha1_to_hex(sha1);
4148
4149         abbrev = find_unique_abbrev(sha1, len);
4150         abblen = strlen(abbrev);
4151         if (abblen < 37) {
4152                 static char hex[41];
4153                 if (len < abblen && abblen <= len + 2)
4154                         xsnprintf(hex, sizeof(hex), "%s%.*s", abbrev, len+3-abblen, "..");
4155                 else
4156                         xsnprintf(hex, sizeof(hex), "%s...", abbrev);
4157                 return hex;
4158         }
4159         return sha1_to_hex(sha1);
4160 }
4161
4162 static void diff_flush_raw(struct diff_filepair *p, struct diff_options *opt)
4163 {
4164         int line_termination = opt->line_termination;
4165         int inter_name_termination = line_termination ? '\t' : '\0';
4166
4167         fprintf(opt->file, "%s", diff_line_prefix(opt));
4168         if (!(opt->output_format & DIFF_FORMAT_NAME_STATUS)) {
4169                 fprintf(opt->file, ":%06o %06o %s ", p->one->mode, p->two->mode,
4170                         diff_unique_abbrev(p->one->oid.hash, opt->abbrev));
4171                 fprintf(opt->file, "%s ",
4172                         diff_unique_abbrev(p->two->oid.hash, opt->abbrev));
4173         }
4174         if (p->score) {
4175                 fprintf(opt->file, "%c%03d%c", p->status, similarity_index(p),
4176                         inter_name_termination);
4177         } else {
4178                 fprintf(opt->file, "%c%c", p->status, inter_name_termination);
4179         }
4180
4181         if (p->status == DIFF_STATUS_COPIED ||
4182             p->status == DIFF_STATUS_RENAMED) {
4183                 const char *name_a, *name_b;
4184                 name_a = p->one->path;
4185                 name_b = p->two->path;
4186                 strip_prefix(opt->prefix_length, &name_a, &name_b);
4187                 write_name_quoted(name_a, opt->file, inter_name_termination);
4188                 write_name_quoted(name_b, opt->file, line_termination);
4189         } else {
4190                 const char *name_a, *name_b;
4191                 name_a = p->one->mode ? p->one->path : p->two->path;
4192                 name_b = NULL;
4193                 strip_prefix(opt->prefix_length, &name_a, &name_b);
4194                 write_name_quoted(name_a, opt->file, line_termination);
4195         }
4196 }
4197
4198 int diff_unmodified_pair(struct diff_filepair *p)
4199 {
4200         /* This function is written stricter than necessary to support
4201          * the currently implemented transformers, but the idea is to
4202          * let transformers to produce diff_filepairs any way they want,
4203          * and filter and clean them up here before producing the output.
4204          */
4205         struct diff_filespec *one = p->one, *two = p->two;
4206
4207         if (DIFF_PAIR_UNMERGED(p))
4208                 return 0; /* unmerged is interesting */
4209
4210         /* deletion, addition, mode or type change
4211          * and rename are all interesting.
4212          */
4213         if (DIFF_FILE_VALID(one) != DIFF_FILE_VALID(two) ||
4214             DIFF_PAIR_MODE_CHANGED(p) ||
4215             strcmp(one->path, two->path))
4216                 return 0;
4217
4218         /* both are valid and point at the same path.  that is, we are
4219          * dealing with a change.
4220          */
4221         if (one->oid_valid && two->oid_valid &&
4222             !oidcmp(&one->oid, &two->oid) &&
4223             !one->dirty_submodule && !two->dirty_submodule)
4224                 return 1; /* no change */
4225         if (!one->oid_valid && !two->oid_valid)
4226                 return 1; /* both look at the same file on the filesystem. */
4227         return 0;
4228 }
4229
4230 static void diff_flush_patch(struct diff_filepair *p, struct diff_options *o)
4231 {
4232         if (diff_unmodified_pair(p))
4233                 return;
4234
4235         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
4236             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
4237                 return; /* no tree diffs in patch format */
4238
4239         run_diff(p, o);
4240 }
4241
4242 static void diff_flush_stat(struct diff_filepair *p, struct diff_options *o,
4243                             struct diffstat_t *diffstat)
4244 {
4245         if (diff_unmodified_pair(p))
4246                 return;
4247
4248         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
4249             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
4250                 return; /* no useful stat for tree diffs */
4251
4252         run_diffstat(p, o, diffstat);
4253 }
4254
4255 static void diff_flush_checkdiff(struct diff_filepair *p,
4256                 struct diff_options *o)
4257 {
4258         if (diff_unmodified_pair(p))
4259                 return;
4260
4261         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
4262             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
4263                 return; /* nothing to check in tree diffs */
4264
4265         run_checkdiff(p, o);
4266 }
4267
4268 int diff_queue_is_empty(void)
4269 {
4270         struct diff_queue_struct *q = &diff_queued_diff;
4271         int i;
4272         for (i = 0; i < q->nr; i++)
4273                 if (!diff_unmodified_pair(q->queue[i]))
4274                         return 0;
4275         return 1;
4276 }
4277
4278 #if DIFF_DEBUG
4279 void diff_debug_filespec(struct diff_filespec *s, int x, const char *one)
4280 {
4281         fprintf(stderr, "queue[%d] %s (%s) %s %06o %s\n",
4282                 x, one ? one : "",
4283                 s->path,
4284                 DIFF_FILE_VALID(s) ? "valid" : "invalid",
4285                 s->mode,
4286                 s->oid_valid ? oid_to_hex(&s->oid) : "");
4287         fprintf(stderr, "queue[%d] %s size %lu\n",
4288                 x, one ? one : "",
4289                 s->size);
4290 }
4291
4292 void diff_debug_filepair(const struct diff_filepair *p, int i)
4293 {
4294         diff_debug_filespec(p->one, i, "one");
4295         diff_debug_filespec(p->two, i, "two");
4296         fprintf(stderr, "score %d, status %c rename_used %d broken %d\n",
4297                 p->score, p->status ? p->status : '?',
4298                 p->one->rename_used, p->broken_pair);
4299 }
4300
4301 void diff_debug_queue(const char *msg, struct diff_queue_struct *q)
4302 {
4303         int i;
4304         if (msg)
4305                 fprintf(stderr, "%s\n", msg);
4306         fprintf(stderr, "q->nr = %d\n", q->nr);
4307         for (i = 0; i < q->nr; i++) {
4308                 struct diff_filepair *p = q->queue[i];
4309                 diff_debug_filepair(p, i);
4310         }
4311 }
4312 #endif
4313
4314 static void diff_resolve_rename_copy(void)
4315 {
4316         int i;
4317         struct diff_filepair *p;
4318         struct diff_queue_struct *q = &diff_queued_diff;
4319
4320         diff_debug_queue("resolve-rename-copy", q);
4321
4322         for (i = 0; i < q->nr; i++) {
4323                 p = q->queue[i];
4324                 p->status = 0; /* undecided */
4325                 if (DIFF_PAIR_UNMERGED(p))
4326                         p->status = DIFF_STATUS_UNMERGED;
4327                 else if (!DIFF_FILE_VALID(p->one))
4328                         p->status = DIFF_STATUS_ADDED;
4329                 else if (!DIFF_FILE_VALID(p->two))
4330                         p->status = DIFF_STATUS_DELETED;
4331                 else if (DIFF_PAIR_TYPE_CHANGED(p))
4332                         p->status = DIFF_STATUS_TYPE_CHANGED;
4333
4334                 /* from this point on, we are dealing with a pair
4335                  * whose both sides are valid and of the same type, i.e.
4336                  * either in-place edit or rename/copy edit.
4337                  */
4338                 else if (DIFF_PAIR_RENAME(p)) {
4339                         /*
4340                          * A rename might have re-connected a broken
4341                          * pair up, causing the pathnames to be the
4342                          * same again. If so, that's not a rename at
4343                          * all, just a modification..
4344                          *
4345                          * Otherwise, see if this source was used for
4346                          * multiple renames, in which case we decrement
4347                          * the count, and call it a copy.
4348                          */
4349                         if (!strcmp(p->one->path, p->two->path))
4350                                 p->status = DIFF_STATUS_MODIFIED;
4351                         else if (--p->one->rename_used > 0)
4352                                 p->status = DIFF_STATUS_COPIED;
4353                         else
4354                                 p->status = DIFF_STATUS_RENAMED;
4355                 }
4356                 else if (oidcmp(&p->one->oid, &p->two->oid) ||
4357                          p->one->mode != p->two->mode ||
4358                          p->one->dirty_submodule ||
4359                          p->two->dirty_submodule ||
4360                          is_null_oid(&p->one->oid))
4361                         p->status = DIFF_STATUS_MODIFIED;
4362                 else {
4363                         /* This is a "no-change" entry and should not
4364                          * happen anymore, but prepare for broken callers.
4365                          */
4366                         error("feeding unmodified %s to diffcore",
4367                               p->one->path);
4368                         p->status = DIFF_STATUS_UNKNOWN;
4369                 }
4370         }
4371         diff_debug_queue("resolve-rename-copy done", q);
4372 }
4373
4374 static int check_pair_status(struct diff_filepair *p)
4375 {
4376         switch (p->status) {
4377         case DIFF_STATUS_UNKNOWN:
4378                 return 0;
4379         case 0:
4380                 die("internal error in diff-resolve-rename-copy");
4381         default:
4382                 return 1;
4383         }
4384 }
4385
4386 static void flush_one_pair(struct diff_filepair *p, struct diff_options *opt)
4387 {
4388         int fmt = opt->output_format;
4389
4390         if (fmt & DIFF_FORMAT_CHECKDIFF)
4391                 diff_flush_checkdiff(p, opt);
4392         else if (fmt & (DIFF_FORMAT_RAW | DIFF_FORMAT_NAME_STATUS))
4393                 diff_flush_raw(p, opt);
4394         else if (fmt & DIFF_FORMAT_NAME) {
4395                 const char *name_a, *name_b;
4396                 name_a = p->two->path;
4397                 name_b = NULL;
4398                 strip_prefix(opt->prefix_length, &name_a, &name_b);
4399                 write_name_quoted(name_a, opt->file, opt->line_termination);
4400         }
4401 }
4402
4403 static void show_file_mode_name(FILE *file, const char *newdelete, struct diff_filespec *fs)
4404 {
4405         if (fs->mode)
4406                 fprintf(file, " %s mode %06o ", newdelete, fs->mode);
4407         else
4408                 fprintf(file, " %s ", newdelete);
4409         write_name_quoted(fs->path, file, '\n');
4410 }
4411
4412
4413 static void show_mode_change(FILE *file, struct diff_filepair *p, int show_name,
4414                 const char *line_prefix)
4415 {
4416         if (p->one->mode && p->two->mode && p->one->mode != p->two->mode) {
4417                 fprintf(file, "%s mode change %06o => %06o%c", line_prefix, p->one->mode,
4418                         p->two->mode, show_name ? ' ' : '\n');
4419                 if (show_name) {
4420                         write_name_quoted(p->two->path, file, '\n');
4421                 }
4422         }
4423 }
4424
4425 static void show_rename_copy(FILE *file, const char *renamecopy, struct diff_filepair *p,
4426                         const char *line_prefix)
4427 {
4428         char *names = pprint_rename(p->one->path, p->two->path);
4429
4430         fprintf(file, " %s %s (%d%%)\n", renamecopy, names, similarity_index(p));
4431         free(names);
4432         show_mode_change(file, p, 0, line_prefix);
4433 }
4434
4435 static void diff_summary(struct diff_options *opt, struct diff_filepair *p)
4436 {
4437         FILE *file = opt->file;
4438         const char *line_prefix = diff_line_prefix(opt);
4439
4440         switch(p->status) {
4441         case DIFF_STATUS_DELETED:
4442                 fputs(line_prefix, file);
4443                 show_file_mode_name(file, "delete", p->one);
4444                 break;
4445         case DIFF_STATUS_ADDED:
4446                 fputs(line_prefix, file);
4447                 show_file_mode_name(file, "create", p->two);
4448                 break;
4449         case DIFF_STATUS_COPIED:
4450                 fputs(line_prefix, file);
4451                 show_rename_copy(file, "copy", p, line_prefix);
4452                 break;
4453         case DIFF_STATUS_RENAMED:
4454                 fputs(line_prefix, file);
4455                 show_rename_copy(file, "rename", p, line_prefix);
4456                 break;
4457         default:
4458                 if (p->score) {
4459                         fprintf(file, "%s rewrite ", line_prefix);
4460                         write_name_quoted(p->two->path, file, ' ');
4461                         fprintf(file, "(%d%%)\n", similarity_index(p));
4462                 }
4463                 show_mode_change(file, p, !p->score, line_prefix);
4464                 break;
4465         }
4466 }
4467
4468 struct patch_id_t {
4469         git_SHA_CTX *ctx;
4470         int patchlen;
4471 };
4472
4473 static int remove_space(char *line, int len)
4474 {
4475         int i;
4476         char *dst = line;
4477         unsigned char c;
4478
4479         for (i = 0; i < len; i++)
4480                 if (!isspace((c = line[i])))
4481                         *dst++ = c;
4482
4483         return dst - line;
4484 }
4485
4486 static void patch_id_consume(void *priv, char *line, unsigned long len)
4487 {
4488         struct patch_id_t *data = priv;
4489         int new_len;
4490
4491         /* Ignore line numbers when computing the SHA1 of the patch */
4492         if (starts_with(line, "@@ -"))
4493                 return;
4494
4495         new_len = remove_space(line, len);
4496
4497         git_SHA1_Update(data->ctx, line, new_len);
4498         data->patchlen += new_len;
4499 }
4500
4501 /* returns 0 upon success, and writes result into sha1 */
4502 static int diff_get_patch_id(struct diff_options *options, unsigned char *sha1, int diff_header_only)
4503 {
4504         struct diff_queue_struct *q = &diff_queued_diff;
4505         int i;
4506         git_SHA_CTX ctx;
4507         struct patch_id_t data;
4508         char buffer[PATH_MAX * 4 + 20];
4509
4510         git_SHA1_Init(&ctx);
4511         memset(&data, 0, sizeof(struct patch_id_t));
4512         data.ctx = &ctx;
4513
4514         for (i = 0; i < q->nr; i++) {
4515                 xpparam_t xpp;
4516                 xdemitconf_t xecfg;
4517                 mmfile_t mf1, mf2;
4518                 struct diff_filepair *p = q->queue[i];
4519                 int len1, len2;
4520
4521                 memset(&xpp, 0, sizeof(xpp));
4522                 memset(&xecfg, 0, sizeof(xecfg));
4523                 if (p->status == 0)
4524                         return error("internal diff status error");
4525                 if (p->status == DIFF_STATUS_UNKNOWN)
4526                         continue;
4527                 if (diff_unmodified_pair(p))
4528                         continue;
4529                 if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
4530                     (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
4531                         continue;
4532                 if (DIFF_PAIR_UNMERGED(p))
4533                         continue;
4534
4535                 diff_fill_sha1_info(p->one);
4536                 diff_fill_sha1_info(p->two);
4537
4538                 len1 = remove_space(p->one->path, strlen(p->one->path));
4539                 len2 = remove_space(p->two->path, strlen(p->two->path));
4540                 if (p->one->mode == 0)
4541                         len1 = snprintf(buffer, sizeof(buffer),
4542                                         "diff--gita/%.*sb/%.*s"
4543                                         "newfilemode%06o"
4544                                         "---/dev/null"
4545                                         "+++b/%.*s",
4546                                         len1, p->one->path,
4547                                         len2, p->two->path,
4548                                         p->two->mode,
4549                                         len2, p->two->path);
4550                 else if (p->two->mode == 0)
4551                         len1 = snprintf(buffer, sizeof(buffer),
4552                                         "diff--gita/%.*sb/%.*s"
4553                                         "deletedfilemode%06o"
4554                                         "---a/%.*s"
4555                                         "+++/dev/null",
4556                                         len1, p->one->path,
4557                                         len2, p->two->path,
4558                                         p->one->mode,
4559                                         len1, p->one->path);
4560                 else
4561                         len1 = snprintf(buffer, sizeof(buffer),
4562                                         "diff--gita/%.*sb/%.*s"
4563                                         "---a/%.*s"
4564                                         "+++b/%.*s",
4565                                         len1, p->one->path,
4566                                         len2, p->two->path,
4567                                         len1, p->one->path,
4568                                         len2, p->two->path);
4569                 git_SHA1_Update(&ctx, buffer, len1);
4570
4571                 if (diff_header_only)
4572                         continue;
4573
4574                 if (fill_mmfile(&mf1, p->one) < 0 ||
4575                     fill_mmfile(&mf2, p->two) < 0)
4576                         return error("unable to read files to diff");
4577
4578                 if (diff_filespec_is_binary(p->one) ||
4579                     diff_filespec_is_binary(p->two)) {
4580                         git_SHA1_Update(&ctx, oid_to_hex(&p->one->oid),
4581                                         40);
4582                         git_SHA1_Update(&ctx, oid_to_hex(&p->two->oid),
4583                                         40);
4584                         continue;
4585                 }
4586
4587                 xpp.flags = 0;
4588                 xecfg.ctxlen = 3;
4589                 xecfg.flags = 0;
4590                 if (xdi_diff_outf(&mf1, &mf2, patch_id_consume, &data,
4591                                   &xpp, &xecfg))
4592                         return error("unable to generate patch-id diff for %s",
4593                                      p->one->path);
4594         }
4595
4596         git_SHA1_Final(sha1, &ctx);
4597         return 0;
4598 }
4599
4600 int diff_flush_patch_id(struct diff_options *options, unsigned char *sha1, int diff_header_only)
4601 {
4602         struct diff_queue_struct *q = &diff_queued_diff;
4603         int i;
4604         int result = diff_get_patch_id(options, sha1, diff_header_only);
4605
4606         for (i = 0; i < q->nr; i++)
4607                 diff_free_filepair(q->queue[i]);
4608
4609         free(q->queue);
4610         DIFF_QUEUE_CLEAR(q);
4611
4612         return result;
4613 }
4614
4615 static int is_summary_empty(const struct diff_queue_struct *q)
4616 {
4617         int i;
4618
4619         for (i = 0; i < q->nr; i++) {
4620                 const struct diff_filepair *p = q->queue[i];
4621
4622                 switch (p->status) {
4623                 case DIFF_STATUS_DELETED:
4624                 case DIFF_STATUS_ADDED:
4625                 case DIFF_STATUS_COPIED:
4626                 case DIFF_STATUS_RENAMED:
4627                         return 0;
4628                 default:
4629                         if (p->score)
4630                                 return 0;
4631                         if (p->one->mode && p->two->mode &&
4632                             p->one->mode != p->two->mode)
4633                                 return 0;
4634                         break;
4635                 }
4636         }
4637         return 1;
4638 }
4639
4640 static const char rename_limit_warning[] =
4641 "inexact rename detection was skipped due to too many files.";
4642
4643 static const char degrade_cc_to_c_warning[] =
4644 "only found copies from modified paths due to too many files.";
4645
4646 static const char rename_limit_advice[] =
4647 "you may want to set your %s variable to at least "
4648 "%d and retry the command.";
4649
4650 void diff_warn_rename_limit(const char *varname, int needed, int degraded_cc)
4651 {
4652         if (degraded_cc)
4653                 warning(degrade_cc_to_c_warning);
4654         else if (needed)
4655                 warning(rename_limit_warning);
4656         else
4657                 return;
4658         if (0 < needed && needed < 32767)
4659                 warning(rename_limit_advice, varname, needed);
4660 }
4661
4662 void diff_flush(struct diff_options *options)
4663 {
4664         struct diff_queue_struct *q = &diff_queued_diff;
4665         int i, output_format = options->output_format;
4666         int separator = 0;
4667         int dirstat_by_line = 0;
4668
4669         /*
4670          * Order: raw, stat, summary, patch
4671          * or:    name/name-status/checkdiff (other bits clear)
4672          */
4673         if (!q->nr)
4674                 goto free_queue;
4675
4676         if (output_format & (DIFF_FORMAT_RAW |
4677                              DIFF_FORMAT_NAME |
4678                              DIFF_FORMAT_NAME_STATUS |
4679                              DIFF_FORMAT_CHECKDIFF)) {
4680                 for (i = 0; i < q->nr; i++) {
4681                         struct diff_filepair *p = q->queue[i];
4682                         if (check_pair_status(p))
4683                                 flush_one_pair(p, options);
4684                 }
4685                 separator++;
4686         }
4687
4688         if (output_format & DIFF_FORMAT_DIRSTAT && DIFF_OPT_TST(options, DIRSTAT_BY_LINE))
4689                 dirstat_by_line = 1;
4690
4691         if (output_format & (DIFF_FORMAT_DIFFSTAT|DIFF_FORMAT_SHORTSTAT|DIFF_FORMAT_NUMSTAT) ||
4692             dirstat_by_line) {
4693                 struct diffstat_t diffstat;
4694
4695                 memset(&diffstat, 0, sizeof(struct diffstat_t));
4696                 for (i = 0; i < q->nr; i++) {
4697                         struct diff_filepair *p = q->queue[i];
4698                         if (check_pair_status(p))
4699                                 diff_flush_stat(p, options, &diffstat);
4700                 }
4701                 if (output_format & DIFF_FORMAT_NUMSTAT)
4702                         show_numstat(&diffstat, options);
4703                 if (output_format & DIFF_FORMAT_DIFFSTAT)
4704                         show_stats(&diffstat, options);
4705                 if (output_format & DIFF_FORMAT_SHORTSTAT)
4706                         show_shortstats(&diffstat, options);
4707                 if (output_format & DIFF_FORMAT_DIRSTAT && dirstat_by_line)
4708                         show_dirstat_by_line(&diffstat, options);
4709                 free_diffstat_info(&diffstat);
4710                 separator++;
4711         }
4712         if ((output_format & DIFF_FORMAT_DIRSTAT) && !dirstat_by_line)
4713                 show_dirstat(options);
4714
4715         if (output_format & DIFF_FORMAT_SUMMARY && !is_summary_empty(q)) {
4716                 for (i = 0; i < q->nr; i++) {
4717                         diff_summary(options, q->queue[i]);
4718                 }
4719                 separator++;
4720         }
4721
4722         if (output_format & DIFF_FORMAT_NO_OUTPUT &&
4723             DIFF_OPT_TST(options, EXIT_WITH_STATUS) &&
4724             DIFF_OPT_TST(options, DIFF_FROM_CONTENTS)) {
4725                 /*
4726                  * run diff_flush_patch for the exit status. setting
4727                  * options->file to /dev/null should be safe, because we
4728                  * aren't supposed to produce any output anyway.
4729                  */
4730                 if (options->close_file)
4731                         fclose(options->file);
4732                 options->file = fopen("/dev/null", "w");
4733                 if (!options->file)
4734                         die_errno("Could not open /dev/null");
4735                 options->close_file = 1;
4736                 for (i = 0; i < q->nr; i++) {
4737                         struct diff_filepair *p = q->queue[i];
4738                         if (check_pair_status(p))
4739                                 diff_flush_patch(p, options);
4740                         if (options->found_changes)
4741                                 break;
4742                 }
4743         }
4744
4745         if (output_format & DIFF_FORMAT_PATCH) {
4746                 if (separator) {
4747                         fprintf(options->file, "%s%c",
4748                                 diff_line_prefix(options),
4749                                 options->line_termination);
4750                         if (options->stat_sep) {
4751                                 /* attach patch instead of inline */
4752                                 fputs(options->stat_sep, options->file);
4753                         }
4754                 }
4755
4756                 for (i = 0; i < q->nr; i++) {
4757                         struct diff_filepair *p = q->queue[i];
4758                         if (check_pair_status(p))
4759                                 diff_flush_patch(p, options);
4760                 }
4761         }
4762
4763         if (output_format & DIFF_FORMAT_CALLBACK)
4764                 options->format_callback(q, options, options->format_callback_data);
4765
4766         for (i = 0; i < q->nr; i++)
4767                 diff_free_filepair(q->queue[i]);
4768 free_queue:
4769         free(q->queue);
4770         DIFF_QUEUE_CLEAR(q);
4771         if (options->close_file)
4772                 fclose(options->file);
4773
4774         /*
4775          * Report the content-level differences with HAS_CHANGES;
4776          * diff_addremove/diff_change does not set the bit when
4777          * DIFF_FROM_CONTENTS is in effect (e.g. with -w).
4778          */
4779         if (DIFF_OPT_TST(options, DIFF_FROM_CONTENTS)) {
4780                 if (options->found_changes)
4781                         DIFF_OPT_SET(options, HAS_CHANGES);
4782                 else
4783                         DIFF_OPT_CLR(options, HAS_CHANGES);
4784         }
4785 }
4786
4787 static int match_filter(const struct diff_options *options, const struct diff_filepair *p)
4788 {
4789         return (((p->status == DIFF_STATUS_MODIFIED) &&
4790                  ((p->score &&
4791                    filter_bit_tst(DIFF_STATUS_FILTER_BROKEN, options)) ||
4792                   (!p->score &&
4793                    filter_bit_tst(DIFF_STATUS_MODIFIED, options)))) ||
4794                 ((p->status != DIFF_STATUS_MODIFIED) &&
4795                  filter_bit_tst(p->status, options)));
4796 }
4797
4798 static void diffcore_apply_filter(struct diff_options *options)
4799 {
4800         int i;
4801         struct diff_queue_struct *q = &diff_queued_diff;
4802         struct diff_queue_struct outq;
4803
4804         DIFF_QUEUE_CLEAR(&outq);
4805
4806         if (!options->filter)
4807                 return;
4808
4809         if (filter_bit_tst(DIFF_STATUS_FILTER_AON, options)) {
4810                 int found;
4811                 for (i = found = 0; !found && i < q->nr; i++) {
4812                         if (match_filter(options, q->queue[i]))
4813                                 found++;
4814                 }
4815                 if (found)
4816                         return;
4817
4818                 /* otherwise we will clear the whole queue
4819                  * by copying the empty outq at the end of this
4820                  * function, but first clear the current entries
4821                  * in the queue.
4822                  */
4823                 for (i = 0; i < q->nr; i++)
4824                         diff_free_filepair(q->queue[i]);
4825         }
4826         else {
4827                 /* Only the matching ones */
4828                 for (i = 0; i < q->nr; i++) {
4829                         struct diff_filepair *p = q->queue[i];
4830                         if (match_filter(options, p))
4831                                 diff_q(&outq, p);
4832                         else
4833                                 diff_free_filepair(p);
4834                 }
4835         }
4836         free(q->queue);
4837         *q = outq;
4838 }
4839
4840 /* Check whether two filespecs with the same mode and size are identical */
4841 static int diff_filespec_is_identical(struct diff_filespec *one,
4842                                       struct diff_filespec *two)
4843 {
4844         if (S_ISGITLINK(one->mode))
4845                 return 0;
4846         if (diff_populate_filespec(one, 0))
4847                 return 0;
4848         if (diff_populate_filespec(two, 0))
4849                 return 0;
4850         return !memcmp(one->data, two->data, one->size);
4851 }
4852
4853 static int diff_filespec_check_stat_unmatch(struct diff_filepair *p)
4854 {
4855         if (p->done_skip_stat_unmatch)
4856                 return p->skip_stat_unmatch_result;
4857
4858         p->done_skip_stat_unmatch = 1;
4859         p->skip_stat_unmatch_result = 0;
4860         /*
4861          * 1. Entries that come from stat info dirtiness
4862          *    always have both sides (iow, not create/delete),
4863          *    one side of the object name is unknown, with
4864          *    the same mode and size.  Keep the ones that
4865          *    do not match these criteria.  They have real
4866          *    differences.
4867          *
4868          * 2. At this point, the file is known to be modified,
4869          *    with the same mode and size, and the object
4870          *    name of one side is unknown.  Need to inspect
4871          *    the identical contents.
4872          */
4873         if (!DIFF_FILE_VALID(p->one) || /* (1) */
4874             !DIFF_FILE_VALID(p->two) ||
4875             (p->one->oid_valid && p->two->oid_valid) ||
4876             (p->one->mode != p->two->mode) ||
4877             diff_populate_filespec(p->one, CHECK_SIZE_ONLY) ||
4878             diff_populate_filespec(p->two, CHECK_SIZE_ONLY) ||
4879             (p->one->size != p->two->size) ||
4880             !diff_filespec_is_identical(p->one, p->two)) /* (2) */
4881                 p->skip_stat_unmatch_result = 1;
4882         return p->skip_stat_unmatch_result;
4883 }
4884
4885 static void diffcore_skip_stat_unmatch(struct diff_options *diffopt)
4886 {
4887         int i;
4888         struct diff_queue_struct *q = &diff_queued_diff;
4889         struct diff_queue_struct outq;
4890         DIFF_QUEUE_CLEAR(&outq);
4891
4892         for (i = 0; i < q->nr; i++) {
4893                 struct diff_filepair *p = q->queue[i];
4894
4895                 if (diff_filespec_check_stat_unmatch(p))
4896                         diff_q(&outq, p);
4897                 else {
4898                         /*
4899                          * The caller can subtract 1 from skip_stat_unmatch
4900                          * to determine how many paths were dirty only
4901                          * due to stat info mismatch.
4902                          */
4903                         if (!DIFF_OPT_TST(diffopt, NO_INDEX))
4904                                 diffopt->skip_stat_unmatch++;
4905                         diff_free_filepair(p);
4906                 }
4907         }
4908         free(q->queue);
4909         *q = outq;
4910 }
4911
4912 static int diffnamecmp(const void *a_, const void *b_)
4913 {
4914         const struct diff_filepair *a = *((const struct diff_filepair **)a_);
4915         const struct diff_filepair *b = *((const struct diff_filepair **)b_);
4916         const char *name_a, *name_b;
4917
4918         name_a = a->one ? a->one->path : a->two->path;
4919         name_b = b->one ? b->one->path : b->two->path;
4920         return strcmp(name_a, name_b);
4921 }
4922
4923 void diffcore_fix_diff_index(struct diff_options *options)
4924 {
4925         struct diff_queue_struct *q = &diff_queued_diff;
4926         qsort(q->queue, q->nr, sizeof(q->queue[0]), diffnamecmp);
4927 }
4928
4929 void diffcore_std(struct diff_options *options)
4930 {
4931         /* NOTE please keep the following in sync with diff_tree_combined() */
4932         if (options->skip_stat_unmatch)
4933                 diffcore_skip_stat_unmatch(options);
4934         if (!options->found_follow) {
4935                 /* See try_to_follow_renames() in tree-diff.c */
4936                 if (options->break_opt != -1)
4937                         diffcore_break(options->break_opt);
4938                 if (options->detect_rename)
4939                         diffcore_rename(options);
4940                 if (options->break_opt != -1)
4941                         diffcore_merge_broken();
4942         }
4943         if (options->pickaxe)
4944                 diffcore_pickaxe(options);
4945         if (options->orderfile)
4946                 diffcore_order(options->orderfile);
4947         if (!options->found_follow)
4948                 /* See try_to_follow_renames() in tree-diff.c */
4949                 diff_resolve_rename_copy();
4950         diffcore_apply_filter(options);
4951
4952         if (diff_queued_diff.nr && !DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
4953                 DIFF_OPT_SET(options, HAS_CHANGES);
4954         else
4955                 DIFF_OPT_CLR(options, HAS_CHANGES);
4956
4957         options->found_follow = 0;
4958 }
4959
4960 int diff_result_code(struct diff_options *opt, int status)
4961 {
4962         int result = 0;
4963
4964         diff_warn_rename_limit("diff.renameLimit",
4965                                opt->needed_rename_limit,
4966                                opt->degraded_cc_to_c);
4967         if (!DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
4968             !(opt->output_format & DIFF_FORMAT_CHECKDIFF))
4969                 return status;
4970         if (DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
4971             DIFF_OPT_TST(opt, HAS_CHANGES))
4972                 result |= 01;
4973         if ((opt->output_format & DIFF_FORMAT_CHECKDIFF) &&
4974             DIFF_OPT_TST(opt, CHECK_FAILED))
4975                 result |= 02;
4976         return result;
4977 }
4978
4979 int diff_can_quit_early(struct diff_options *opt)
4980 {
4981         return (DIFF_OPT_TST(opt, QUICK) &&
4982                 !opt->filter &&
4983                 DIFF_OPT_TST(opt, HAS_CHANGES));
4984 }
4985
4986 /*
4987  * Shall changes to this submodule be ignored?
4988  *
4989  * Submodule changes can be configured to be ignored separately for each path,
4990  * but that configuration can be overridden from the command line.
4991  */
4992 static int is_submodule_ignored(const char *path, struct diff_options *options)
4993 {
4994         int ignored = 0;
4995         unsigned orig_flags = options->flags;
4996         if (!DIFF_OPT_TST(options, OVERRIDE_SUBMODULE_CONFIG))
4997                 set_diffopt_flags_from_submodule_config(options, path);
4998         if (DIFF_OPT_TST(options, IGNORE_SUBMODULES))
4999                 ignored = 1;
5000         options->flags = orig_flags;
5001         return ignored;
5002 }
5003
5004 void diff_addremove(struct diff_options *options,
5005                     int addremove, unsigned mode,
5006                     const unsigned char *sha1,
5007                     int sha1_valid,
5008                     const char *concatpath, unsigned dirty_submodule)
5009 {
5010         struct diff_filespec *one, *two;
5011
5012         if (S_ISGITLINK(mode) && is_submodule_ignored(concatpath, options))
5013                 return;
5014
5015         /* This may look odd, but it is a preparation for
5016          * feeding "there are unchanged files which should
5017          * not produce diffs, but when you are doing copy
5018          * detection you would need them, so here they are"
5019          * entries to the diff-core.  They will be prefixed
5020          * with something like '=' or '*' (I haven't decided
5021          * which but should not make any difference).
5022          * Feeding the same new and old to diff_change()
5023          * also has the same effect.
5024          * Before the final output happens, they are pruned after
5025          * merged into rename/copy pairs as appropriate.
5026          */
5027         if (DIFF_OPT_TST(options, REVERSE_DIFF))
5028                 addremove = (addremove == '+' ? '-' :
5029                              addremove == '-' ? '+' : addremove);
5030
5031         if (options->prefix &&
5032             strncmp(concatpath, options->prefix, options->prefix_length))
5033                 return;
5034
5035         one = alloc_filespec(concatpath);
5036         two = alloc_filespec(concatpath);
5037
5038         if (addremove != '+')
5039                 fill_filespec(one, sha1, sha1_valid, mode);
5040         if (addremove != '-') {
5041                 fill_filespec(two, sha1, sha1_valid, mode);
5042                 two->dirty_submodule = dirty_submodule;
5043         }
5044
5045         diff_queue(&diff_queued_diff, one, two);
5046         if (!DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
5047                 DIFF_OPT_SET(options, HAS_CHANGES);
5048 }
5049
5050 void diff_change(struct diff_options *options,
5051                  unsigned old_mode, unsigned new_mode,
5052                  const unsigned char *old_sha1,
5053                  const unsigned char *new_sha1,
5054                  int old_sha1_valid, int new_sha1_valid,
5055                  const char *concatpath,
5056                  unsigned old_dirty_submodule, unsigned new_dirty_submodule)
5057 {
5058         struct diff_filespec *one, *two;
5059         struct diff_filepair *p;
5060
5061         if (S_ISGITLINK(old_mode) && S_ISGITLINK(new_mode) &&
5062             is_submodule_ignored(concatpath, options))
5063                 return;
5064
5065         if (DIFF_OPT_TST(options, REVERSE_DIFF)) {
5066                 unsigned tmp;
5067                 const unsigned char *tmp_c;
5068                 tmp = old_mode; old_mode = new_mode; new_mode = tmp;
5069                 tmp_c = old_sha1; old_sha1 = new_sha1; new_sha1 = tmp_c;
5070                 tmp = old_sha1_valid; old_sha1_valid = new_sha1_valid;
5071                         new_sha1_valid = tmp;
5072                 tmp = old_dirty_submodule; old_dirty_submodule = new_dirty_submodule;
5073                         new_dirty_submodule = tmp;
5074         }
5075
5076         if (options->prefix &&
5077             strncmp(concatpath, options->prefix, options->prefix_length))
5078                 return;
5079
5080         one = alloc_filespec(concatpath);
5081         two = alloc_filespec(concatpath);
5082         fill_filespec(one, old_sha1, old_sha1_valid, old_mode);
5083         fill_filespec(two, new_sha1, new_sha1_valid, new_mode);
5084         one->dirty_submodule = old_dirty_submodule;
5085         two->dirty_submodule = new_dirty_submodule;
5086         p = diff_queue(&diff_queued_diff, one, two);
5087
5088         if (DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
5089                 return;
5090
5091         if (DIFF_OPT_TST(options, QUICK) && options->skip_stat_unmatch &&
5092             !diff_filespec_check_stat_unmatch(p))
5093                 return;
5094
5095         DIFF_OPT_SET(options, HAS_CHANGES);
5096 }
5097
5098 struct diff_filepair *diff_unmerge(struct diff_options *options, const char *path)
5099 {
5100         struct diff_filepair *pair;
5101         struct diff_filespec *one, *two;
5102
5103         if (options->prefix &&
5104             strncmp(path, options->prefix, options->prefix_length))
5105                 return NULL;
5106
5107         one = alloc_filespec(path);
5108         two = alloc_filespec(path);
5109         pair = diff_queue(&diff_queued_diff, one, two);
5110         pair->is_unmerged = 1;
5111         return pair;
5112 }
5113
5114 static char *run_textconv(const char *pgm, struct diff_filespec *spec,
5115                 size_t *outsize)
5116 {
5117         struct diff_tempfile *temp;
5118         const char *argv[3];
5119         const char **arg = argv;
5120         struct child_process child = CHILD_PROCESS_INIT;
5121         struct strbuf buf = STRBUF_INIT;
5122         int err = 0;
5123
5124         temp = prepare_temp_file(spec->path, spec);
5125         *arg++ = pgm;
5126         *arg++ = temp->name;
5127         *arg = NULL;
5128
5129         child.use_shell = 1;
5130         child.argv = argv;
5131         child.out = -1;
5132         if (start_command(&child)) {
5133                 remove_tempfile();
5134                 return NULL;
5135         }
5136
5137         if (strbuf_read(&buf, child.out, 0) < 0)
5138                 err = error("error reading from textconv command '%s'", pgm);
5139         close(child.out);
5140
5141         if (finish_command(&child) || err) {
5142                 strbuf_release(&buf);
5143                 remove_tempfile();
5144                 return NULL;
5145         }
5146         remove_tempfile();
5147
5148         return strbuf_detach(&buf, outsize);
5149 }
5150
5151 size_t fill_textconv(struct userdiff_driver *driver,
5152                      struct diff_filespec *df,
5153                      char **outbuf)
5154 {
5155         size_t size;
5156
5157         if (!driver) {
5158                 if (!DIFF_FILE_VALID(df)) {
5159                         *outbuf = "";
5160                         return 0;
5161                 }
5162                 if (diff_populate_filespec(df, 0))
5163                         die("unable to read files to diff");
5164                 *outbuf = df->data;
5165                 return df->size;
5166         }
5167
5168         if (!driver->textconv)
5169                 die("BUG: fill_textconv called with non-textconv driver");
5170
5171         if (driver->textconv_cache && df->oid_valid) {
5172                 *outbuf = notes_cache_get(driver->textconv_cache,
5173                                           df->oid.hash,
5174                                           &size);
5175                 if (*outbuf)
5176                         return size;
5177         }
5178
5179         *outbuf = run_textconv(driver->textconv, df, &size);
5180         if (!*outbuf)
5181                 die("unable to read files to diff");
5182
5183         if (driver->textconv_cache && df->oid_valid) {
5184                 /* ignore errors, as we might be in a readonly repository */
5185                 notes_cache_put(driver->textconv_cache, df->oid.hash, *outbuf,
5186                                 size);
5187                 /*
5188                  * we could save up changes and flush them all at the end,
5189                  * but we would need an extra call after all diffing is done.
5190                  * Since generating a cache entry is the slow path anyway,
5191                  * this extra overhead probably isn't a big deal.
5192                  */
5193                 notes_cache_write(driver->textconv_cache);
5194         }
5195
5196         return size;
5197 }
5198
5199 void setup_diff_pager(struct diff_options *opt)
5200 {
5201         /*
5202          * If the user asked for our exit code, then either they want --quiet
5203          * or --exit-code. We should definitely not bother with a pager in the
5204          * former case, as we will generate no output. Since we still properly
5205          * report our exit code even when a pager is run, we _could_ run a
5206          * pager with --exit-code. But since we have not done so historically,
5207          * and because it is easy to find people oneline advising "git diff
5208          * --exit-code" in hooks and other scripts, we do not do so.
5209          */
5210         if (!DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
5211             check_pager_config("diff") != 0)
5212                 setup_pager();
5213 }