Merge branch 'jc/merge-x-theirs-docfix'
[git] / diff.c
1 /*
2  * Copyright (C) 2005 Junio C Hamano
3  */
4 #include "cache.h"
5 #include "config.h"
6 #include "tempfile.h"
7 #include "quote.h"
8 #include "diff.h"
9 #include "diffcore.h"
10 #include "delta.h"
11 #include "xdiff-interface.h"
12 #include "color.h"
13 #include "attr.h"
14 #include "run-command.h"
15 #include "utf8.h"
16 #include "userdiff.h"
17 #include "submodule-config.h"
18 #include "submodule.h"
19 #include "hashmap.h"
20 #include "ll-merge.h"
21 #include "string-list.h"
22 #include "argv-array.h"
23 #include "graph.h"
24 #include "packfile.h"
25
26 #ifdef NO_FAST_WORKING_DIRECTORY
27 #define FAST_WORKING_DIRECTORY 0
28 #else
29 #define FAST_WORKING_DIRECTORY 1
30 #endif
31
32 static int diff_detect_rename_default;
33 static int diff_indent_heuristic = 1;
34 static int diff_rename_limit_default = 400;
35 static int diff_suppress_blank_empty;
36 static int diff_use_color_default = -1;
37 static int diff_color_moved_default;
38 static int diff_context_default = 3;
39 static int diff_interhunk_context_default;
40 static const char *diff_word_regex_cfg;
41 static const char *external_diff_cmd_cfg;
42 static const char *diff_order_file_cfg;
43 int diff_auto_refresh_index = 1;
44 static int diff_mnemonic_prefix;
45 static int diff_no_prefix;
46 static int diff_stat_graph_width;
47 static int diff_dirstat_permille_default = 30;
48 static struct diff_options default_diff_options;
49 static long diff_algorithm;
50 static unsigned ws_error_highlight_default = WSEH_NEW;
51
52 static char diff_colors[][COLOR_MAXLEN] = {
53         GIT_COLOR_RESET,
54         GIT_COLOR_NORMAL,       /* CONTEXT */
55         GIT_COLOR_BOLD,         /* METAINFO */
56         GIT_COLOR_CYAN,         /* FRAGINFO */
57         GIT_COLOR_RED,          /* OLD */
58         GIT_COLOR_GREEN,        /* NEW */
59         GIT_COLOR_YELLOW,       /* COMMIT */
60         GIT_COLOR_BG_RED,       /* WHITESPACE */
61         GIT_COLOR_NORMAL,       /* FUNCINFO */
62         GIT_COLOR_BOLD_MAGENTA, /* OLD_MOVED */
63         GIT_COLOR_BOLD_BLUE,    /* OLD_MOVED ALTERNATIVE */
64         GIT_COLOR_FAINT,        /* OLD_MOVED_DIM */
65         GIT_COLOR_FAINT_ITALIC, /* OLD_MOVED_ALTERNATIVE_DIM */
66         GIT_COLOR_BOLD_CYAN,    /* NEW_MOVED */
67         GIT_COLOR_BOLD_YELLOW,  /* NEW_MOVED ALTERNATIVE */
68         GIT_COLOR_FAINT,        /* NEW_MOVED_DIM */
69         GIT_COLOR_FAINT_ITALIC, /* NEW_MOVED_ALTERNATIVE_DIM */
70 };
71
72 static NORETURN void die_want_option(const char *option_name)
73 {
74         die(_("option '%s' requires a value"), option_name);
75 }
76
77 static int parse_diff_color_slot(const char *var)
78 {
79         if (!strcasecmp(var, "context") || !strcasecmp(var, "plain"))
80                 return DIFF_CONTEXT;
81         if (!strcasecmp(var, "meta"))
82                 return DIFF_METAINFO;
83         if (!strcasecmp(var, "frag"))
84                 return DIFF_FRAGINFO;
85         if (!strcasecmp(var, "old"))
86                 return DIFF_FILE_OLD;
87         if (!strcasecmp(var, "new"))
88                 return DIFF_FILE_NEW;
89         if (!strcasecmp(var, "commit"))
90                 return DIFF_COMMIT;
91         if (!strcasecmp(var, "whitespace"))
92                 return DIFF_WHITESPACE;
93         if (!strcasecmp(var, "func"))
94                 return DIFF_FUNCINFO;
95         if (!strcasecmp(var, "oldmoved"))
96                 return DIFF_FILE_OLD_MOVED;
97         if (!strcasecmp(var, "oldmovedalternative"))
98                 return DIFF_FILE_OLD_MOVED_ALT;
99         if (!strcasecmp(var, "oldmoveddimmed"))
100                 return DIFF_FILE_OLD_MOVED_DIM;
101         if (!strcasecmp(var, "oldmovedalternativedimmed"))
102                 return DIFF_FILE_OLD_MOVED_ALT_DIM;
103         if (!strcasecmp(var, "newmoved"))
104                 return DIFF_FILE_NEW_MOVED;
105         if (!strcasecmp(var, "newmovedalternative"))
106                 return DIFF_FILE_NEW_MOVED_ALT;
107         if (!strcasecmp(var, "newmoveddimmed"))
108                 return DIFF_FILE_NEW_MOVED_DIM;
109         if (!strcasecmp(var, "newmovedalternativedimmed"))
110                 return DIFF_FILE_NEW_MOVED_ALT_DIM;
111         return -1;
112 }
113
114 static int parse_dirstat_params(struct diff_options *options, const char *params_string,
115                                 struct strbuf *errmsg)
116 {
117         char *params_copy = xstrdup(params_string);
118         struct string_list params = STRING_LIST_INIT_NODUP;
119         int ret = 0;
120         int i;
121
122         if (*params_copy)
123                 string_list_split_in_place(&params, params_copy, ',', -1);
124         for (i = 0; i < params.nr; i++) {
125                 const char *p = params.items[i].string;
126                 if (!strcmp(p, "changes")) {
127                         DIFF_OPT_CLR(options, DIRSTAT_BY_LINE);
128                         DIFF_OPT_CLR(options, DIRSTAT_BY_FILE);
129                 } else if (!strcmp(p, "lines")) {
130                         DIFF_OPT_SET(options, DIRSTAT_BY_LINE);
131                         DIFF_OPT_CLR(options, DIRSTAT_BY_FILE);
132                 } else if (!strcmp(p, "files")) {
133                         DIFF_OPT_CLR(options, DIRSTAT_BY_LINE);
134                         DIFF_OPT_SET(options, DIRSTAT_BY_FILE);
135                 } else if (!strcmp(p, "noncumulative")) {
136                         DIFF_OPT_CLR(options, DIRSTAT_CUMULATIVE);
137                 } else if (!strcmp(p, "cumulative")) {
138                         DIFF_OPT_SET(options, DIRSTAT_CUMULATIVE);
139                 } else if (isdigit(*p)) {
140                         char *end;
141                         int permille = strtoul(p, &end, 10) * 10;
142                         if (*end == '.' && isdigit(*++end)) {
143                                 /* only use first digit */
144                                 permille += *end - '0';
145                                 /* .. and ignore any further digits */
146                                 while (isdigit(*++end))
147                                         ; /* nothing */
148                         }
149                         if (!*end)
150                                 options->dirstat_permille = permille;
151                         else {
152                                 strbuf_addf(errmsg, _("  Failed to parse dirstat cut-off percentage '%s'\n"),
153                                             p);
154                                 ret++;
155                         }
156                 } else {
157                         strbuf_addf(errmsg, _("  Unknown dirstat parameter '%s'\n"), p);
158                         ret++;
159                 }
160
161         }
162         string_list_clear(&params, 0);
163         free(params_copy);
164         return ret;
165 }
166
167 static int parse_submodule_params(struct diff_options *options, const char *value)
168 {
169         if (!strcmp(value, "log"))
170                 options->submodule_format = DIFF_SUBMODULE_LOG;
171         else if (!strcmp(value, "short"))
172                 options->submodule_format = DIFF_SUBMODULE_SHORT;
173         else if (!strcmp(value, "diff"))
174                 options->submodule_format = DIFF_SUBMODULE_INLINE_DIFF;
175         else
176                 return -1;
177         return 0;
178 }
179
180 static int git_config_rename(const char *var, const char *value)
181 {
182         if (!value)
183                 return DIFF_DETECT_RENAME;
184         if (!strcasecmp(value, "copies") || !strcasecmp(value, "copy"))
185                 return  DIFF_DETECT_COPY;
186         return git_config_bool(var,value) ? DIFF_DETECT_RENAME : 0;
187 }
188
189 long parse_algorithm_value(const char *value)
190 {
191         if (!value)
192                 return -1;
193         else if (!strcasecmp(value, "myers") || !strcasecmp(value, "default"))
194                 return 0;
195         else if (!strcasecmp(value, "minimal"))
196                 return XDF_NEED_MINIMAL;
197         else if (!strcasecmp(value, "patience"))
198                 return XDF_PATIENCE_DIFF;
199         else if (!strcasecmp(value, "histogram"))
200                 return XDF_HISTOGRAM_DIFF;
201         return -1;
202 }
203
204 static int parse_one_token(const char **arg, const char *token)
205 {
206         const char *rest;
207         if (skip_prefix(*arg, token, &rest) && (!*rest || *rest == ',')) {
208                 *arg = rest;
209                 return 1;
210         }
211         return 0;
212 }
213
214 static int parse_ws_error_highlight(const char *arg)
215 {
216         const char *orig_arg = arg;
217         unsigned val = 0;
218
219         while (*arg) {
220                 if (parse_one_token(&arg, "none"))
221                         val = 0;
222                 else if (parse_one_token(&arg, "default"))
223                         val = WSEH_NEW;
224                 else if (parse_one_token(&arg, "all"))
225                         val = WSEH_NEW | WSEH_OLD | WSEH_CONTEXT;
226                 else if (parse_one_token(&arg, "new"))
227                         val |= WSEH_NEW;
228                 else if (parse_one_token(&arg, "old"))
229                         val |= WSEH_OLD;
230                 else if (parse_one_token(&arg, "context"))
231                         val |= WSEH_CONTEXT;
232                 else {
233                         return -1 - (int)(arg - orig_arg);
234                 }
235                 if (*arg)
236                         arg++;
237         }
238         return val;
239 }
240
241 /*
242  * These are to give UI layer defaults.
243  * The core-level commands such as git-diff-files should
244  * never be affected by the setting of diff.renames
245  * the user happens to have in the configuration file.
246  */
247 void init_diff_ui_defaults(void)
248 {
249         diff_detect_rename_default = 1;
250 }
251
252 int git_diff_heuristic_config(const char *var, const char *value, void *cb)
253 {
254         if (!strcmp(var, "diff.indentheuristic"))
255                 diff_indent_heuristic = git_config_bool(var, value);
256         return 0;
257 }
258
259 static int parse_color_moved(const char *arg)
260 {
261         switch (git_parse_maybe_bool(arg)) {
262         case 0:
263                 return COLOR_MOVED_NO;
264         case 1:
265                 return COLOR_MOVED_DEFAULT;
266         default:
267                 break;
268         }
269
270         if (!strcmp(arg, "no"))
271                 return COLOR_MOVED_NO;
272         else if (!strcmp(arg, "plain"))
273                 return COLOR_MOVED_PLAIN;
274         else if (!strcmp(arg, "zebra"))
275                 return COLOR_MOVED_ZEBRA;
276         else if (!strcmp(arg, "default"))
277                 return COLOR_MOVED_DEFAULT;
278         else if (!strcmp(arg, "dimmed_zebra"))
279                 return COLOR_MOVED_ZEBRA_DIM;
280         else
281                 return error(_("color moved setting must be one of 'no', 'default', 'zebra', 'dimmed_zebra', 'plain'"));
282 }
283
284 int git_diff_ui_config(const char *var, const char *value, void *cb)
285 {
286         if (!strcmp(var, "diff.color") || !strcmp(var, "color.diff")) {
287                 diff_use_color_default = git_config_colorbool(var, value);
288                 return 0;
289         }
290         if (!strcmp(var, "diff.colormoved")) {
291                 int cm = parse_color_moved(value);
292                 if (cm < 0)
293                         return -1;
294                 diff_color_moved_default = cm;
295                 return 0;
296         }
297         if (!strcmp(var, "diff.context")) {
298                 diff_context_default = git_config_int(var, value);
299                 if (diff_context_default < 0)
300                         return -1;
301                 return 0;
302         }
303         if (!strcmp(var, "diff.interhunkcontext")) {
304                 diff_interhunk_context_default = git_config_int(var, value);
305                 if (diff_interhunk_context_default < 0)
306                         return -1;
307                 return 0;
308         }
309         if (!strcmp(var, "diff.renames")) {
310                 diff_detect_rename_default = git_config_rename(var, value);
311                 return 0;
312         }
313         if (!strcmp(var, "diff.autorefreshindex")) {
314                 diff_auto_refresh_index = git_config_bool(var, value);
315                 return 0;
316         }
317         if (!strcmp(var, "diff.mnemonicprefix")) {
318                 diff_mnemonic_prefix = git_config_bool(var, value);
319                 return 0;
320         }
321         if (!strcmp(var, "diff.noprefix")) {
322                 diff_no_prefix = git_config_bool(var, value);
323                 return 0;
324         }
325         if (!strcmp(var, "diff.statgraphwidth")) {
326                 diff_stat_graph_width = git_config_int(var, value);
327                 return 0;
328         }
329         if (!strcmp(var, "diff.external"))
330                 return git_config_string(&external_diff_cmd_cfg, var, value);
331         if (!strcmp(var, "diff.wordregex"))
332                 return git_config_string(&diff_word_regex_cfg, var, value);
333         if (!strcmp(var, "diff.orderfile"))
334                 return git_config_pathname(&diff_order_file_cfg, var, value);
335
336         if (!strcmp(var, "diff.ignoresubmodules"))
337                 handle_ignore_submodules_arg(&default_diff_options, value);
338
339         if (!strcmp(var, "diff.submodule")) {
340                 if (parse_submodule_params(&default_diff_options, value))
341                         warning(_("Unknown value for 'diff.submodule' config variable: '%s'"),
342                                 value);
343                 return 0;
344         }
345
346         if (!strcmp(var, "diff.algorithm")) {
347                 diff_algorithm = parse_algorithm_value(value);
348                 if (diff_algorithm < 0)
349                         return -1;
350                 return 0;
351         }
352
353         if (!strcmp(var, "diff.wserrorhighlight")) {
354                 int val = parse_ws_error_highlight(value);
355                 if (val < 0)
356                         return -1;
357                 ws_error_highlight_default = val;
358                 return 0;
359         }
360
361         return git_diff_basic_config(var, value, cb);
362 }
363
364 int git_diff_basic_config(const char *var, const char *value, void *cb)
365 {
366         const char *name;
367
368         if (!strcmp(var, "diff.renamelimit")) {
369                 diff_rename_limit_default = git_config_int(var, value);
370                 return 0;
371         }
372
373         if (userdiff_config(var, value) < 0)
374                 return -1;
375
376         if (skip_prefix(var, "diff.color.", &name) ||
377             skip_prefix(var, "color.diff.", &name)) {
378                 int slot = parse_diff_color_slot(name);
379                 if (slot < 0)
380                         return 0;
381                 if (!value)
382                         return config_error_nonbool(var);
383                 return color_parse(value, diff_colors[slot]);
384         }
385
386         /* like GNU diff's --suppress-blank-empty option  */
387         if (!strcmp(var, "diff.suppressblankempty") ||
388                         /* for backwards compatibility */
389                         !strcmp(var, "diff.suppress-blank-empty")) {
390                 diff_suppress_blank_empty = git_config_bool(var, value);
391                 return 0;
392         }
393
394         if (!strcmp(var, "diff.dirstat")) {
395                 struct strbuf errmsg = STRBUF_INIT;
396                 default_diff_options.dirstat_permille = diff_dirstat_permille_default;
397                 if (parse_dirstat_params(&default_diff_options, value, &errmsg))
398                         warning(_("Found errors in 'diff.dirstat' config variable:\n%s"),
399                                 errmsg.buf);
400                 strbuf_release(&errmsg);
401                 diff_dirstat_permille_default = default_diff_options.dirstat_permille;
402                 return 0;
403         }
404
405         if (git_diff_heuristic_config(var, value, cb) < 0)
406                 return -1;
407
408         return git_default_config(var, value, cb);
409 }
410
411 static char *quote_two(const char *one, const char *two)
412 {
413         int need_one = quote_c_style(one, NULL, NULL, 1);
414         int need_two = quote_c_style(two, NULL, NULL, 1);
415         struct strbuf res = STRBUF_INIT;
416
417         if (need_one + need_two) {
418                 strbuf_addch(&res, '"');
419                 quote_c_style(one, &res, NULL, 1);
420                 quote_c_style(two, &res, NULL, 1);
421                 strbuf_addch(&res, '"');
422         } else {
423                 strbuf_addstr(&res, one);
424                 strbuf_addstr(&res, two);
425         }
426         return strbuf_detach(&res, NULL);
427 }
428
429 static const char *external_diff(void)
430 {
431         static const char *external_diff_cmd = NULL;
432         static int done_preparing = 0;
433
434         if (done_preparing)
435                 return external_diff_cmd;
436         external_diff_cmd = getenv("GIT_EXTERNAL_DIFF");
437         if (!external_diff_cmd)
438                 external_diff_cmd = external_diff_cmd_cfg;
439         done_preparing = 1;
440         return external_diff_cmd;
441 }
442
443 /*
444  * Keep track of files used for diffing. Sometimes such an entry
445  * refers to a temporary file, sometimes to an existing file, and
446  * sometimes to "/dev/null".
447  */
448 static struct diff_tempfile {
449         /*
450          * filename external diff should read from, or NULL if this
451          * entry is currently not in use:
452          */
453         const char *name;
454
455         char hex[GIT_MAX_HEXSZ + 1];
456         char mode[10];
457
458         /*
459          * If this diff_tempfile instance refers to a temporary file,
460          * this tempfile object is used to manage its lifetime.
461          */
462         struct tempfile *tempfile;
463 } diff_temp[2];
464
465 struct emit_callback {
466         int color_diff;
467         unsigned ws_rule;
468         int blank_at_eof_in_preimage;
469         int blank_at_eof_in_postimage;
470         int lno_in_preimage;
471         int lno_in_postimage;
472         const char **label_path;
473         struct diff_words_data *diff_words;
474         struct diff_options *opt;
475         struct strbuf *header;
476 };
477
478 static int count_lines(const char *data, int size)
479 {
480         int count, ch, completely_empty = 1, nl_just_seen = 0;
481         count = 0;
482         while (0 < size--) {
483                 ch = *data++;
484                 if (ch == '\n') {
485                         count++;
486                         nl_just_seen = 1;
487                         completely_empty = 0;
488                 }
489                 else {
490                         nl_just_seen = 0;
491                         completely_empty = 0;
492                 }
493         }
494         if (completely_empty)
495                 return 0;
496         if (!nl_just_seen)
497                 count++; /* no trailing newline */
498         return count;
499 }
500
501 static int fill_mmfile(mmfile_t *mf, struct diff_filespec *one)
502 {
503         if (!DIFF_FILE_VALID(one)) {
504                 mf->ptr = (char *)""; /* does not matter */
505                 mf->size = 0;
506                 return 0;
507         }
508         else if (diff_populate_filespec(one, 0))
509                 return -1;
510
511         mf->ptr = one->data;
512         mf->size = one->size;
513         return 0;
514 }
515
516 /* like fill_mmfile, but only for size, so we can avoid retrieving blob */
517 static unsigned long diff_filespec_size(struct diff_filespec *one)
518 {
519         if (!DIFF_FILE_VALID(one))
520                 return 0;
521         diff_populate_filespec(one, CHECK_SIZE_ONLY);
522         return one->size;
523 }
524
525 static int count_trailing_blank(mmfile_t *mf, unsigned ws_rule)
526 {
527         char *ptr = mf->ptr;
528         long size = mf->size;
529         int cnt = 0;
530
531         if (!size)
532                 return cnt;
533         ptr += size - 1; /* pointing at the very end */
534         if (*ptr != '\n')
535                 ; /* incomplete line */
536         else
537                 ptr--; /* skip the last LF */
538         while (mf->ptr < ptr) {
539                 char *prev_eol;
540                 for (prev_eol = ptr; mf->ptr <= prev_eol; prev_eol--)
541                         if (*prev_eol == '\n')
542                                 break;
543                 if (!ws_blank_line(prev_eol + 1, ptr - prev_eol, ws_rule))
544                         break;
545                 cnt++;
546                 ptr = prev_eol - 1;
547         }
548         return cnt;
549 }
550
551 static void check_blank_at_eof(mmfile_t *mf1, mmfile_t *mf2,
552                                struct emit_callback *ecbdata)
553 {
554         int l1, l2, at;
555         unsigned ws_rule = ecbdata->ws_rule;
556         l1 = count_trailing_blank(mf1, ws_rule);
557         l2 = count_trailing_blank(mf2, ws_rule);
558         if (l2 <= l1) {
559                 ecbdata->blank_at_eof_in_preimage = 0;
560                 ecbdata->blank_at_eof_in_postimage = 0;
561                 return;
562         }
563         at = count_lines(mf1->ptr, mf1->size);
564         ecbdata->blank_at_eof_in_preimage = (at - l1) + 1;
565
566         at = count_lines(mf2->ptr, mf2->size);
567         ecbdata->blank_at_eof_in_postimage = (at - l2) + 1;
568 }
569
570 static void emit_line_0(struct diff_options *o, const char *set, const char *reset,
571                         int first, const char *line, int len)
572 {
573         int has_trailing_newline, has_trailing_carriage_return;
574         int nofirst;
575         FILE *file = o->file;
576
577         fputs(diff_line_prefix(o), file);
578
579         if (len == 0) {
580                 has_trailing_newline = (first == '\n');
581                 has_trailing_carriage_return = (!has_trailing_newline &&
582                                                 (first == '\r'));
583                 nofirst = has_trailing_newline || has_trailing_carriage_return;
584         } else {
585                 has_trailing_newline = (len > 0 && line[len-1] == '\n');
586                 if (has_trailing_newline)
587                         len--;
588                 has_trailing_carriage_return = (len > 0 && line[len-1] == '\r');
589                 if (has_trailing_carriage_return)
590                         len--;
591                 nofirst = 0;
592         }
593
594         if (len || !nofirst) {
595                 fputs(set, file);
596                 if (!nofirst)
597                         fputc(first, file);
598                 fwrite(line, len, 1, file);
599                 fputs(reset, file);
600         }
601         if (has_trailing_carriage_return)
602                 fputc('\r', file);
603         if (has_trailing_newline)
604                 fputc('\n', file);
605 }
606
607 static void emit_line(struct diff_options *o, const char *set, const char *reset,
608                       const char *line, int len)
609 {
610         emit_line_0(o, set, reset, line[0], line+1, len-1);
611 }
612
613 enum diff_symbol {
614         DIFF_SYMBOL_BINARY_DIFF_HEADER,
615         DIFF_SYMBOL_BINARY_DIFF_HEADER_DELTA,
616         DIFF_SYMBOL_BINARY_DIFF_HEADER_LITERAL,
617         DIFF_SYMBOL_BINARY_DIFF_BODY,
618         DIFF_SYMBOL_BINARY_DIFF_FOOTER,
619         DIFF_SYMBOL_STATS_SUMMARY_NO_FILES,
620         DIFF_SYMBOL_STATS_SUMMARY_ABBREV,
621         DIFF_SYMBOL_STATS_SUMMARY_INSERTS_DELETES,
622         DIFF_SYMBOL_STATS_LINE,
623         DIFF_SYMBOL_WORD_DIFF,
624         DIFF_SYMBOL_STAT_SEP,
625         DIFF_SYMBOL_SUMMARY,
626         DIFF_SYMBOL_SUBMODULE_ADD,
627         DIFF_SYMBOL_SUBMODULE_DEL,
628         DIFF_SYMBOL_SUBMODULE_UNTRACKED,
629         DIFF_SYMBOL_SUBMODULE_MODIFIED,
630         DIFF_SYMBOL_SUBMODULE_HEADER,
631         DIFF_SYMBOL_SUBMODULE_ERROR,
632         DIFF_SYMBOL_SUBMODULE_PIPETHROUGH,
633         DIFF_SYMBOL_REWRITE_DIFF,
634         DIFF_SYMBOL_BINARY_FILES,
635         DIFF_SYMBOL_HEADER,
636         DIFF_SYMBOL_FILEPAIR_PLUS,
637         DIFF_SYMBOL_FILEPAIR_MINUS,
638         DIFF_SYMBOL_WORDS_PORCELAIN,
639         DIFF_SYMBOL_WORDS,
640         DIFF_SYMBOL_CONTEXT,
641         DIFF_SYMBOL_CONTEXT_INCOMPLETE,
642         DIFF_SYMBOL_PLUS,
643         DIFF_SYMBOL_MINUS,
644         DIFF_SYMBOL_NO_LF_EOF,
645         DIFF_SYMBOL_CONTEXT_FRAGINFO,
646         DIFF_SYMBOL_CONTEXT_MARKER,
647         DIFF_SYMBOL_SEPARATOR
648 };
649 /*
650  * Flags for content lines:
651  * 0..12 are whitespace rules
652  * 13-15 are WSEH_NEW | WSEH_OLD | WSEH_CONTEXT
653  * 16 is marking if the line is blank at EOF
654  */
655 #define DIFF_SYMBOL_CONTENT_BLANK_LINE_EOF      (1<<16)
656 #define DIFF_SYMBOL_MOVED_LINE                  (1<<17)
657 #define DIFF_SYMBOL_MOVED_LINE_ALT              (1<<18)
658 #define DIFF_SYMBOL_MOVED_LINE_UNINTERESTING    (1<<19)
659 #define DIFF_SYMBOL_CONTENT_WS_MASK (WSEH_NEW | WSEH_OLD | WSEH_CONTEXT | WS_RULE_MASK)
660
661 /*
662  * This struct is used when we need to buffer the output of the diff output.
663  *
664  * NEEDSWORK: Instead of storing a copy of the line, add an offset pointer
665  * into the pre/post image file. This pointer could be a union with the
666  * line pointer. By storing an offset into the file instead of the literal line,
667  * we can decrease the memory footprint for the buffered output. At first we
668  * may want to only have indirection for the content lines, but we could also
669  * enhance the state for emitting prefabricated lines, e.g. the similarity
670  * score line or hunk/file headers would only need to store a number or path
671  * and then the output can be constructed later on depending on state.
672  */
673 struct emitted_diff_symbol {
674         const char *line;
675         int len;
676         int flags;
677         enum diff_symbol s;
678 };
679 #define EMITTED_DIFF_SYMBOL_INIT {NULL}
680
681 struct emitted_diff_symbols {
682         struct emitted_diff_symbol *buf;
683         int nr, alloc;
684 };
685 #define EMITTED_DIFF_SYMBOLS_INIT {NULL, 0, 0}
686
687 static void append_emitted_diff_symbol(struct diff_options *o,
688                                        struct emitted_diff_symbol *e)
689 {
690         struct emitted_diff_symbol *f;
691
692         ALLOC_GROW(o->emitted_symbols->buf,
693                    o->emitted_symbols->nr + 1,
694                    o->emitted_symbols->alloc);
695         f = &o->emitted_symbols->buf[o->emitted_symbols->nr++];
696
697         memcpy(f, e, sizeof(struct emitted_diff_symbol));
698         f->line = e->line ? xmemdupz(e->line, e->len) : NULL;
699 }
700
701 struct moved_entry {
702         struct hashmap_entry ent;
703         const struct emitted_diff_symbol *es;
704         struct moved_entry *next_line;
705 };
706
707 static int next_byte(const char **cp, const char **endp,
708                      const struct diff_options *diffopt)
709 {
710         int retval;
711
712         if (*cp > *endp)
713                 return -1;
714
715         if (DIFF_XDL_TST(diffopt, IGNORE_WHITESPACE_CHANGE)) {
716                 while (*cp < *endp && isspace(**cp))
717                         (*cp)++;
718                 /*
719                  * After skipping a couple of whitespaces, we still have to
720                  * account for one space.
721                  */
722                 return (int)' ';
723         }
724
725         if (DIFF_XDL_TST(diffopt, IGNORE_WHITESPACE)) {
726                 while (*cp < *endp && isspace(**cp))
727                         (*cp)++;
728                 /* return the first non-ws character via the usual below */
729         }
730
731         retval = (unsigned char)(**cp);
732         (*cp)++;
733         return retval;
734 }
735
736 static int moved_entry_cmp(const struct diff_options *diffopt,
737                            const struct moved_entry *a,
738                            const struct moved_entry *b,
739                            const void *keydata)
740 {
741         const char *ap = a->es->line, *ae = a->es->line + a->es->len;
742         const char *bp = b->es->line, *be = b->es->line + b->es->len;
743
744         if (!(diffopt->xdl_opts & XDF_WHITESPACE_FLAGS))
745                 return a->es->len != b->es->len  || memcmp(ap, bp, a->es->len);
746
747         if (DIFF_XDL_TST(diffopt, IGNORE_WHITESPACE_AT_EOL)) {
748                 while (ae > ap && isspace(*ae))
749                         ae--;
750                 while (be > bp && isspace(*be))
751                         be--;
752         }
753
754         while (1) {
755                 int ca, cb;
756                 ca = next_byte(&ap, &ae, diffopt);
757                 cb = next_byte(&bp, &be, diffopt);
758                 if (ca != cb)
759                         return 1;
760                 if (ca < 0)
761                         return 0;
762         }
763 }
764
765 static unsigned get_string_hash(struct emitted_diff_symbol *es, struct diff_options *o)
766 {
767         if (o->xdl_opts & XDF_WHITESPACE_FLAGS) {
768                 static struct strbuf sb = STRBUF_INIT;
769                 const char *ap = es->line, *ae = es->line + es->len;
770                 int c;
771
772                 strbuf_reset(&sb);
773                 while (ae > ap && isspace(*ae))
774                         ae--;
775                 while ((c = next_byte(&ap, &ae, o)) > 0)
776                         strbuf_addch(&sb, c);
777
778                 return memhash(sb.buf, sb.len);
779         } else {
780                 return memhash(es->line, es->len);
781         }
782 }
783
784 static struct moved_entry *prepare_entry(struct diff_options *o,
785                                          int line_no)
786 {
787         struct moved_entry *ret = xmalloc(sizeof(*ret));
788         struct emitted_diff_symbol *l = &o->emitted_symbols->buf[line_no];
789
790         ret->ent.hash = get_string_hash(l, o);
791         ret->es = l;
792         ret->next_line = NULL;
793
794         return ret;
795 }
796
797 static void add_lines_to_move_detection(struct diff_options *o,
798                                         struct hashmap *add_lines,
799                                         struct hashmap *del_lines)
800 {
801         struct moved_entry *prev_line = NULL;
802
803         int n;
804         for (n = 0; n < o->emitted_symbols->nr; n++) {
805                 struct hashmap *hm;
806                 struct moved_entry *key;
807
808                 switch (o->emitted_symbols->buf[n].s) {
809                 case DIFF_SYMBOL_PLUS:
810                         hm = add_lines;
811                         break;
812                 case DIFF_SYMBOL_MINUS:
813                         hm = del_lines;
814                         break;
815                 default:
816                         prev_line = NULL;
817                         continue;
818                 }
819
820                 key = prepare_entry(o, n);
821                 if (prev_line && prev_line->es->s == o->emitted_symbols->buf[n].s)
822                         prev_line->next_line = key;
823
824                 hashmap_add(hm, key);
825                 prev_line = key;
826         }
827 }
828
829 static int shrink_potential_moved_blocks(struct moved_entry **pmb,
830                                          int pmb_nr)
831 {
832         int lp, rp;
833
834         /* Shrink the set of potential block to the remaining running */
835         for (lp = 0, rp = pmb_nr - 1; lp <= rp;) {
836                 while (lp < pmb_nr && pmb[lp])
837                         lp++;
838                 /* lp points at the first NULL now */
839
840                 while (rp > -1 && !pmb[rp])
841                         rp--;
842                 /* rp points at the last non-NULL */
843
844                 if (lp < pmb_nr && rp > -1 && lp < rp) {
845                         pmb[lp] = pmb[rp];
846                         pmb[rp] = NULL;
847                         rp--;
848                         lp++;
849                 }
850         }
851
852         /* Remember the number of running sets */
853         return rp + 1;
854 }
855
856 /*
857  * If o->color_moved is COLOR_MOVED_PLAIN, this function does nothing.
858  *
859  * Otherwise, if the last block has fewer alphanumeric characters than
860  * COLOR_MOVED_MIN_ALNUM_COUNT, unset DIFF_SYMBOL_MOVED_LINE on all lines in
861  * that block.
862  *
863  * The last block consists of the (n - block_length)'th line up to but not
864  * including the nth line.
865  *
866  * NEEDSWORK: This uses the same heuristic as blame_entry_score() in blame.c.
867  * Think of a way to unify them.
868  */
869 static void adjust_last_block(struct diff_options *o, int n, int block_length)
870 {
871         int i, alnum_count = 0;
872         if (o->color_moved == COLOR_MOVED_PLAIN)
873                 return;
874         for (i = 1; i < block_length + 1; i++) {
875                 const char *c = o->emitted_symbols->buf[n - i].line;
876                 for (; *c; c++) {
877                         if (!isalnum(*c))
878                                 continue;
879                         alnum_count++;
880                         if (alnum_count >= COLOR_MOVED_MIN_ALNUM_COUNT)
881                                 return;
882                 }
883         }
884         for (i = 1; i < block_length + 1; i++)
885                 o->emitted_symbols->buf[n - i].flags &= ~DIFF_SYMBOL_MOVED_LINE;
886 }
887
888 /* Find blocks of moved code, delegate actual coloring decision to helper */
889 static void mark_color_as_moved(struct diff_options *o,
890                                 struct hashmap *add_lines,
891                                 struct hashmap *del_lines)
892 {
893         struct moved_entry **pmb = NULL; /* potentially moved blocks */
894         int pmb_nr = 0, pmb_alloc = 0;
895         int n, flipped_block = 1, block_length = 0;
896
897
898         for (n = 0; n < o->emitted_symbols->nr; n++) {
899                 struct hashmap *hm = NULL;
900                 struct moved_entry *key;
901                 struct moved_entry *match = NULL;
902                 struct emitted_diff_symbol *l = &o->emitted_symbols->buf[n];
903                 int i;
904
905                 switch (l->s) {
906                 case DIFF_SYMBOL_PLUS:
907                         hm = del_lines;
908                         key = prepare_entry(o, n);
909                         match = hashmap_get(hm, key, o);
910                         free(key);
911                         break;
912                 case DIFF_SYMBOL_MINUS:
913                         hm = add_lines;
914                         key = prepare_entry(o, n);
915                         match = hashmap_get(hm, key, o);
916                         free(key);
917                         break;
918                 default:
919                         flipped_block = 1;
920                 }
921
922                 if (!match) {
923                         adjust_last_block(o, n, block_length);
924                         pmb_nr = 0;
925                         block_length = 0;
926                         continue;
927                 }
928
929                 l->flags |= DIFF_SYMBOL_MOVED_LINE;
930
931                 if (o->color_moved == COLOR_MOVED_PLAIN)
932                         continue;
933
934                 /* Check any potential block runs, advance each or nullify */
935                 for (i = 0; i < pmb_nr; i++) {
936                         struct moved_entry *p = pmb[i];
937                         struct moved_entry *pnext = (p && p->next_line) ?
938                                         p->next_line : NULL;
939                         if (pnext && !hm->cmpfn(o, pnext, match, NULL)) {
940                                 pmb[i] = p->next_line;
941                         } else {
942                                 pmb[i] = NULL;
943                         }
944                 }
945
946                 pmb_nr = shrink_potential_moved_blocks(pmb, pmb_nr);
947
948                 if (pmb_nr == 0) {
949                         /*
950                          * The current line is the start of a new block.
951                          * Setup the set of potential blocks.
952                          */
953                         for (; match; match = hashmap_get_next(hm, match)) {
954                                 ALLOC_GROW(pmb, pmb_nr + 1, pmb_alloc);
955                                 pmb[pmb_nr++] = match;
956                         }
957
958                         flipped_block = (flipped_block + 1) % 2;
959
960                         adjust_last_block(o, n, block_length);
961                         block_length = 0;
962                 }
963
964                 block_length++;
965
966                 if (flipped_block)
967                         l->flags |= DIFF_SYMBOL_MOVED_LINE_ALT;
968         }
969         adjust_last_block(o, n, block_length);
970
971         free(pmb);
972 }
973
974 #define DIFF_SYMBOL_MOVED_LINE_ZEBRA_MASK \
975   (DIFF_SYMBOL_MOVED_LINE | DIFF_SYMBOL_MOVED_LINE_ALT)
976 static void dim_moved_lines(struct diff_options *o)
977 {
978         int n;
979         for (n = 0; n < o->emitted_symbols->nr; n++) {
980                 struct emitted_diff_symbol *prev = (n != 0) ?
981                                 &o->emitted_symbols->buf[n - 1] : NULL;
982                 struct emitted_diff_symbol *l = &o->emitted_symbols->buf[n];
983                 struct emitted_diff_symbol *next =
984                                 (n < o->emitted_symbols->nr - 1) ?
985                                 &o->emitted_symbols->buf[n + 1] : NULL;
986
987                 /* Not a plus or minus line? */
988                 if (l->s != DIFF_SYMBOL_PLUS && l->s != DIFF_SYMBOL_MINUS)
989                         continue;
990
991                 /* Not a moved line? */
992                 if (!(l->flags & DIFF_SYMBOL_MOVED_LINE))
993                         continue;
994
995                 /*
996                  * If prev or next are not a plus or minus line,
997                  * pretend they don't exist
998                  */
999                 if (prev && prev->s != DIFF_SYMBOL_PLUS &&
1000                             prev->s != DIFF_SYMBOL_MINUS)
1001                         prev = NULL;
1002                 if (next && next->s != DIFF_SYMBOL_PLUS &&
1003                             next->s != DIFF_SYMBOL_MINUS)
1004                         next = NULL;
1005
1006                 /* Inside a block? */
1007                 if ((prev &&
1008                     (prev->flags & DIFF_SYMBOL_MOVED_LINE_ZEBRA_MASK) ==
1009                     (l->flags & DIFF_SYMBOL_MOVED_LINE_ZEBRA_MASK)) &&
1010                     (next &&
1011                     (next->flags & DIFF_SYMBOL_MOVED_LINE_ZEBRA_MASK) ==
1012                     (l->flags & DIFF_SYMBOL_MOVED_LINE_ZEBRA_MASK))) {
1013                         l->flags |= DIFF_SYMBOL_MOVED_LINE_UNINTERESTING;
1014                         continue;
1015                 }
1016
1017                 /* Check if we are at an interesting bound: */
1018                 if (prev && (prev->flags & DIFF_SYMBOL_MOVED_LINE) &&
1019                     (prev->flags & DIFF_SYMBOL_MOVED_LINE_ALT) !=
1020                        (l->flags & DIFF_SYMBOL_MOVED_LINE_ALT))
1021                         continue;
1022                 if (next && (next->flags & DIFF_SYMBOL_MOVED_LINE) &&
1023                     (next->flags & DIFF_SYMBOL_MOVED_LINE_ALT) !=
1024                        (l->flags & DIFF_SYMBOL_MOVED_LINE_ALT))
1025                         continue;
1026
1027                 /*
1028                  * The boundary to prev and next are not interesting,
1029                  * so this line is not interesting as a whole
1030                  */
1031                 l->flags |= DIFF_SYMBOL_MOVED_LINE_UNINTERESTING;
1032         }
1033 }
1034
1035 static void emit_line_ws_markup(struct diff_options *o,
1036                                 const char *set, const char *reset,
1037                                 const char *line, int len, char sign,
1038                                 unsigned ws_rule, int blank_at_eof)
1039 {
1040         const char *ws = NULL;
1041
1042         if (o->ws_error_highlight & ws_rule) {
1043                 ws = diff_get_color_opt(o, DIFF_WHITESPACE);
1044                 if (!*ws)
1045                         ws = NULL;
1046         }
1047
1048         if (!ws)
1049                 emit_line_0(o, set, reset, sign, line, len);
1050         else if (blank_at_eof)
1051                 /* Blank line at EOF - paint '+' as well */
1052                 emit_line_0(o, ws, reset, sign, line, len);
1053         else {
1054                 /* Emit just the prefix, then the rest. */
1055                 emit_line_0(o, set, reset, sign, "", 0);
1056                 ws_check_emit(line, len, ws_rule,
1057                               o->file, set, reset, ws);
1058         }
1059 }
1060
1061 static void emit_diff_symbol_from_struct(struct diff_options *o,
1062                                          struct emitted_diff_symbol *eds)
1063 {
1064         static const char *nneof = " No newline at end of file\n";
1065         const char *context, *reset, *set, *meta, *fraginfo;
1066         struct strbuf sb = STRBUF_INIT;
1067
1068         enum diff_symbol s = eds->s;
1069         const char *line = eds->line;
1070         int len = eds->len;
1071         unsigned flags = eds->flags;
1072
1073         switch (s) {
1074         case DIFF_SYMBOL_NO_LF_EOF:
1075                 context = diff_get_color_opt(o, DIFF_CONTEXT);
1076                 reset = diff_get_color_opt(o, DIFF_RESET);
1077                 putc('\n', o->file);
1078                 emit_line_0(o, context, reset, '\\',
1079                             nneof, strlen(nneof));
1080                 break;
1081         case DIFF_SYMBOL_SUBMODULE_HEADER:
1082         case DIFF_SYMBOL_SUBMODULE_ERROR:
1083         case DIFF_SYMBOL_SUBMODULE_PIPETHROUGH:
1084         case DIFF_SYMBOL_STATS_SUMMARY_INSERTS_DELETES:
1085         case DIFF_SYMBOL_SUMMARY:
1086         case DIFF_SYMBOL_STATS_LINE:
1087         case DIFF_SYMBOL_BINARY_DIFF_BODY:
1088         case DIFF_SYMBOL_CONTEXT_FRAGINFO:
1089                 emit_line(o, "", "", line, len);
1090                 break;
1091         case DIFF_SYMBOL_CONTEXT_INCOMPLETE:
1092         case DIFF_SYMBOL_CONTEXT_MARKER:
1093                 context = diff_get_color_opt(o, DIFF_CONTEXT);
1094                 reset = diff_get_color_opt(o, DIFF_RESET);
1095                 emit_line(o, context, reset, line, len);
1096                 break;
1097         case DIFF_SYMBOL_SEPARATOR:
1098                 fprintf(o->file, "%s%c",
1099                         diff_line_prefix(o),
1100                         o->line_termination);
1101                 break;
1102         case DIFF_SYMBOL_CONTEXT:
1103                 set = diff_get_color_opt(o, DIFF_CONTEXT);
1104                 reset = diff_get_color_opt(o, DIFF_RESET);
1105                 emit_line_ws_markup(o, set, reset, line, len, ' ',
1106                                     flags & (DIFF_SYMBOL_CONTENT_WS_MASK), 0);
1107                 break;
1108         case DIFF_SYMBOL_PLUS:
1109                 switch (flags & (DIFF_SYMBOL_MOVED_LINE |
1110                                  DIFF_SYMBOL_MOVED_LINE_ALT |
1111                                  DIFF_SYMBOL_MOVED_LINE_UNINTERESTING)) {
1112                 case DIFF_SYMBOL_MOVED_LINE |
1113                      DIFF_SYMBOL_MOVED_LINE_ALT |
1114                      DIFF_SYMBOL_MOVED_LINE_UNINTERESTING:
1115                         set = diff_get_color_opt(o, DIFF_FILE_NEW_MOVED_ALT_DIM);
1116                         break;
1117                 case DIFF_SYMBOL_MOVED_LINE |
1118                      DIFF_SYMBOL_MOVED_LINE_ALT:
1119                         set = diff_get_color_opt(o, DIFF_FILE_NEW_MOVED_ALT);
1120                         break;
1121                 case DIFF_SYMBOL_MOVED_LINE |
1122                      DIFF_SYMBOL_MOVED_LINE_UNINTERESTING:
1123                         set = diff_get_color_opt(o, DIFF_FILE_NEW_MOVED_DIM);
1124                         break;
1125                 case DIFF_SYMBOL_MOVED_LINE:
1126                         set = diff_get_color_opt(o, DIFF_FILE_NEW_MOVED);
1127                         break;
1128                 default:
1129                         set = diff_get_color_opt(o, DIFF_FILE_NEW);
1130                 }
1131                 reset = diff_get_color_opt(o, DIFF_RESET);
1132                 emit_line_ws_markup(o, set, reset, line, len, '+',
1133                                     flags & DIFF_SYMBOL_CONTENT_WS_MASK,
1134                                     flags & DIFF_SYMBOL_CONTENT_BLANK_LINE_EOF);
1135                 break;
1136         case DIFF_SYMBOL_MINUS:
1137                 switch (flags & (DIFF_SYMBOL_MOVED_LINE |
1138                                  DIFF_SYMBOL_MOVED_LINE_ALT |
1139                                  DIFF_SYMBOL_MOVED_LINE_UNINTERESTING)) {
1140                 case DIFF_SYMBOL_MOVED_LINE |
1141                      DIFF_SYMBOL_MOVED_LINE_ALT |
1142                      DIFF_SYMBOL_MOVED_LINE_UNINTERESTING:
1143                         set = diff_get_color_opt(o, DIFF_FILE_OLD_MOVED_ALT_DIM);
1144                         break;
1145                 case DIFF_SYMBOL_MOVED_LINE |
1146                      DIFF_SYMBOL_MOVED_LINE_ALT:
1147                         set = diff_get_color_opt(o, DIFF_FILE_OLD_MOVED_ALT);
1148                         break;
1149                 case DIFF_SYMBOL_MOVED_LINE |
1150                      DIFF_SYMBOL_MOVED_LINE_UNINTERESTING:
1151                         set = diff_get_color_opt(o, DIFF_FILE_OLD_MOVED_DIM);
1152                         break;
1153                 case DIFF_SYMBOL_MOVED_LINE:
1154                         set = diff_get_color_opt(o, DIFF_FILE_OLD_MOVED);
1155                         break;
1156                 default:
1157                         set = diff_get_color_opt(o, DIFF_FILE_OLD);
1158                 }
1159                 reset = diff_get_color_opt(o, DIFF_RESET);
1160                 emit_line_ws_markup(o, set, reset, line, len, '-',
1161                                     flags & DIFF_SYMBOL_CONTENT_WS_MASK, 0);
1162                 break;
1163         case DIFF_SYMBOL_WORDS_PORCELAIN:
1164                 context = diff_get_color_opt(o, DIFF_CONTEXT);
1165                 reset = diff_get_color_opt(o, DIFF_RESET);
1166                 emit_line(o, context, reset, line, len);
1167                 fputs("~\n", o->file);
1168                 break;
1169         case DIFF_SYMBOL_WORDS:
1170                 context = diff_get_color_opt(o, DIFF_CONTEXT);
1171                 reset = diff_get_color_opt(o, DIFF_RESET);
1172                 /*
1173                  * Skip the prefix character, if any.  With
1174                  * diff_suppress_blank_empty, there may be
1175                  * none.
1176                  */
1177                 if (line[0] != '\n') {
1178                         line++;
1179                         len--;
1180                 }
1181                 emit_line(o, context, reset, line, len);
1182                 break;
1183         case DIFF_SYMBOL_FILEPAIR_PLUS:
1184                 meta = diff_get_color_opt(o, DIFF_METAINFO);
1185                 reset = diff_get_color_opt(o, DIFF_RESET);
1186                 fprintf(o->file, "%s%s+++ %s%s%s\n", diff_line_prefix(o), meta,
1187                         line, reset,
1188                         strchr(line, ' ') ? "\t" : "");
1189                 break;
1190         case DIFF_SYMBOL_FILEPAIR_MINUS:
1191                 meta = diff_get_color_opt(o, DIFF_METAINFO);
1192                 reset = diff_get_color_opt(o, DIFF_RESET);
1193                 fprintf(o->file, "%s%s--- %s%s%s\n", diff_line_prefix(o), meta,
1194                         line, reset,
1195                         strchr(line, ' ') ? "\t" : "");
1196                 break;
1197         case DIFF_SYMBOL_BINARY_FILES:
1198         case DIFF_SYMBOL_HEADER:
1199                 fprintf(o->file, "%s", line);
1200                 break;
1201         case DIFF_SYMBOL_BINARY_DIFF_HEADER:
1202                 fprintf(o->file, "%sGIT binary patch\n", diff_line_prefix(o));
1203                 break;
1204         case DIFF_SYMBOL_BINARY_DIFF_HEADER_DELTA:
1205                 fprintf(o->file, "%sdelta %s\n", diff_line_prefix(o), line);
1206                 break;
1207         case DIFF_SYMBOL_BINARY_DIFF_HEADER_LITERAL:
1208                 fprintf(o->file, "%sliteral %s\n", diff_line_prefix(o), line);
1209                 break;
1210         case DIFF_SYMBOL_BINARY_DIFF_FOOTER:
1211                 fputs(diff_line_prefix(o), o->file);
1212                 fputc('\n', o->file);
1213                 break;
1214         case DIFF_SYMBOL_REWRITE_DIFF:
1215                 fraginfo = diff_get_color(o->use_color, DIFF_FRAGINFO);
1216                 reset = diff_get_color_opt(o, DIFF_RESET);
1217                 emit_line(o, fraginfo, reset, line, len);
1218                 break;
1219         case DIFF_SYMBOL_SUBMODULE_ADD:
1220                 set = diff_get_color_opt(o, DIFF_FILE_NEW);
1221                 reset = diff_get_color_opt(o, DIFF_RESET);
1222                 emit_line(o, set, reset, line, len);
1223                 break;
1224         case DIFF_SYMBOL_SUBMODULE_DEL:
1225                 set = diff_get_color_opt(o, DIFF_FILE_OLD);
1226                 reset = diff_get_color_opt(o, DIFF_RESET);
1227                 emit_line(o, set, reset, line, len);
1228                 break;
1229         case DIFF_SYMBOL_SUBMODULE_UNTRACKED:
1230                 fprintf(o->file, "%sSubmodule %s contains untracked content\n",
1231                         diff_line_prefix(o), line);
1232                 break;
1233         case DIFF_SYMBOL_SUBMODULE_MODIFIED:
1234                 fprintf(o->file, "%sSubmodule %s contains modified content\n",
1235                         diff_line_prefix(o), line);
1236                 break;
1237         case DIFF_SYMBOL_STATS_SUMMARY_NO_FILES:
1238                 emit_line(o, "", "", " 0 files changed\n",
1239                           strlen(" 0 files changed\n"));
1240                 break;
1241         case DIFF_SYMBOL_STATS_SUMMARY_ABBREV:
1242                 emit_line(o, "", "", " ...\n", strlen(" ...\n"));
1243                 break;
1244         case DIFF_SYMBOL_WORD_DIFF:
1245                 fprintf(o->file, "%.*s", len, line);
1246                 break;
1247         case DIFF_SYMBOL_STAT_SEP:
1248                 fputs(o->stat_sep, o->file);
1249                 break;
1250         default:
1251                 die("BUG: unknown diff symbol");
1252         }
1253         strbuf_release(&sb);
1254 }
1255
1256 static void emit_diff_symbol(struct diff_options *o, enum diff_symbol s,
1257                              const char *line, int len, unsigned flags)
1258 {
1259         struct emitted_diff_symbol e = {line, len, flags, s};
1260
1261         if (o->emitted_symbols)
1262                 append_emitted_diff_symbol(o, &e);
1263         else
1264                 emit_diff_symbol_from_struct(o, &e);
1265 }
1266
1267 void diff_emit_submodule_del(struct diff_options *o, const char *line)
1268 {
1269         emit_diff_symbol(o, DIFF_SYMBOL_SUBMODULE_DEL, line, strlen(line), 0);
1270 }
1271
1272 void diff_emit_submodule_add(struct diff_options *o, const char *line)
1273 {
1274         emit_diff_symbol(o, DIFF_SYMBOL_SUBMODULE_ADD, line, strlen(line), 0);
1275 }
1276
1277 void diff_emit_submodule_untracked(struct diff_options *o, const char *path)
1278 {
1279         emit_diff_symbol(o, DIFF_SYMBOL_SUBMODULE_UNTRACKED,
1280                          path, strlen(path), 0);
1281 }
1282
1283 void diff_emit_submodule_modified(struct diff_options *o, const char *path)
1284 {
1285         emit_diff_symbol(o, DIFF_SYMBOL_SUBMODULE_MODIFIED,
1286                          path, strlen(path), 0);
1287 }
1288
1289 void diff_emit_submodule_header(struct diff_options *o, const char *header)
1290 {
1291         emit_diff_symbol(o, DIFF_SYMBOL_SUBMODULE_HEADER,
1292                          header, strlen(header), 0);
1293 }
1294
1295 void diff_emit_submodule_error(struct diff_options *o, const char *err)
1296 {
1297         emit_diff_symbol(o, DIFF_SYMBOL_SUBMODULE_ERROR, err, strlen(err), 0);
1298 }
1299
1300 void diff_emit_submodule_pipethrough(struct diff_options *o,
1301                                      const char *line, int len)
1302 {
1303         emit_diff_symbol(o, DIFF_SYMBOL_SUBMODULE_PIPETHROUGH, line, len, 0);
1304 }
1305
1306 static int new_blank_line_at_eof(struct emit_callback *ecbdata, const char *line, int len)
1307 {
1308         if (!((ecbdata->ws_rule & WS_BLANK_AT_EOF) &&
1309               ecbdata->blank_at_eof_in_preimage &&
1310               ecbdata->blank_at_eof_in_postimage &&
1311               ecbdata->blank_at_eof_in_preimage <= ecbdata->lno_in_preimage &&
1312               ecbdata->blank_at_eof_in_postimage <= ecbdata->lno_in_postimage))
1313                 return 0;
1314         return ws_blank_line(line, len, ecbdata->ws_rule);
1315 }
1316
1317 static void emit_add_line(const char *reset,
1318                           struct emit_callback *ecbdata,
1319                           const char *line, int len)
1320 {
1321         unsigned flags = WSEH_NEW | ecbdata->ws_rule;
1322         if (new_blank_line_at_eof(ecbdata, line, len))
1323                 flags |= DIFF_SYMBOL_CONTENT_BLANK_LINE_EOF;
1324
1325         emit_diff_symbol(ecbdata->opt, DIFF_SYMBOL_PLUS, line, len, flags);
1326 }
1327
1328 static void emit_del_line(const char *reset,
1329                           struct emit_callback *ecbdata,
1330                           const char *line, int len)
1331 {
1332         unsigned flags = WSEH_OLD | ecbdata->ws_rule;
1333         emit_diff_symbol(ecbdata->opt, DIFF_SYMBOL_MINUS, line, len, flags);
1334 }
1335
1336 static void emit_context_line(const char *reset,
1337                               struct emit_callback *ecbdata,
1338                               const char *line, int len)
1339 {
1340         unsigned flags = WSEH_CONTEXT | ecbdata->ws_rule;
1341         emit_diff_symbol(ecbdata->opt, DIFF_SYMBOL_CONTEXT, line, len, flags);
1342 }
1343
1344 static void emit_hunk_header(struct emit_callback *ecbdata,
1345                              const char *line, int len)
1346 {
1347         const char *context = diff_get_color(ecbdata->color_diff, DIFF_CONTEXT);
1348         const char *frag = diff_get_color(ecbdata->color_diff, DIFF_FRAGINFO);
1349         const char *func = diff_get_color(ecbdata->color_diff, DIFF_FUNCINFO);
1350         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
1351         static const char atat[2] = { '@', '@' };
1352         const char *cp, *ep;
1353         struct strbuf msgbuf = STRBUF_INIT;
1354         int org_len = len;
1355         int i = 1;
1356
1357         /*
1358          * As a hunk header must begin with "@@ -<old>, +<new> @@",
1359          * it always is at least 10 bytes long.
1360          */
1361         if (len < 10 ||
1362             memcmp(line, atat, 2) ||
1363             !(ep = memmem(line + 2, len - 2, atat, 2))) {
1364                 emit_diff_symbol(ecbdata->opt,
1365                                  DIFF_SYMBOL_CONTEXT_MARKER, line, len, 0);
1366                 return;
1367         }
1368         ep += 2; /* skip over @@ */
1369
1370         /* The hunk header in fraginfo color */
1371         strbuf_addstr(&msgbuf, frag);
1372         strbuf_add(&msgbuf, line, ep - line);
1373         strbuf_addstr(&msgbuf, reset);
1374
1375         /*
1376          * trailing "\r\n"
1377          */
1378         for ( ; i < 3; i++)
1379                 if (line[len - i] == '\r' || line[len - i] == '\n')
1380                         len--;
1381
1382         /* blank before the func header */
1383         for (cp = ep; ep - line < len; ep++)
1384                 if (*ep != ' ' && *ep != '\t')
1385                         break;
1386         if (ep != cp) {
1387                 strbuf_addstr(&msgbuf, context);
1388                 strbuf_add(&msgbuf, cp, ep - cp);
1389                 strbuf_addstr(&msgbuf, reset);
1390         }
1391
1392         if (ep < line + len) {
1393                 strbuf_addstr(&msgbuf, func);
1394                 strbuf_add(&msgbuf, ep, line + len - ep);
1395                 strbuf_addstr(&msgbuf, reset);
1396         }
1397
1398         strbuf_add(&msgbuf, line + len, org_len - len);
1399         strbuf_complete_line(&msgbuf);
1400         emit_diff_symbol(ecbdata->opt,
1401                          DIFF_SYMBOL_CONTEXT_FRAGINFO, msgbuf.buf, msgbuf.len, 0);
1402         strbuf_release(&msgbuf);
1403 }
1404
1405 static struct diff_tempfile *claim_diff_tempfile(void) {
1406         int i;
1407         for (i = 0; i < ARRAY_SIZE(diff_temp); i++)
1408                 if (!diff_temp[i].name)
1409                         return diff_temp + i;
1410         die("BUG: diff is failing to clean up its tempfiles");
1411 }
1412
1413 static void remove_tempfile(void)
1414 {
1415         int i;
1416         for (i = 0; i < ARRAY_SIZE(diff_temp); i++) {
1417                 if (is_tempfile_active(diff_temp[i].tempfile))
1418                         delete_tempfile(&diff_temp[i].tempfile);
1419                 diff_temp[i].name = NULL;
1420         }
1421 }
1422
1423 static void add_line_count(struct strbuf *out, int count)
1424 {
1425         switch (count) {
1426         case 0:
1427                 strbuf_addstr(out, "0,0");
1428                 break;
1429         case 1:
1430                 strbuf_addstr(out, "1");
1431                 break;
1432         default:
1433                 strbuf_addf(out, "1,%d", count);
1434                 break;
1435         }
1436 }
1437
1438 static void emit_rewrite_lines(struct emit_callback *ecb,
1439                                int prefix, const char *data, int size)
1440 {
1441         const char *endp = NULL;
1442         const char *reset = diff_get_color(ecb->color_diff, DIFF_RESET);
1443
1444         while (0 < size) {
1445                 int len;
1446
1447                 endp = memchr(data, '\n', size);
1448                 len = endp ? (endp - data + 1) : size;
1449                 if (prefix != '+') {
1450                         ecb->lno_in_preimage++;
1451                         emit_del_line(reset, ecb, data, len);
1452                 } else {
1453                         ecb->lno_in_postimage++;
1454                         emit_add_line(reset, ecb, data, len);
1455                 }
1456                 size -= len;
1457                 data += len;
1458         }
1459         if (!endp)
1460                 emit_diff_symbol(ecb->opt, DIFF_SYMBOL_NO_LF_EOF, NULL, 0, 0);
1461 }
1462
1463 static void emit_rewrite_diff(const char *name_a,
1464                               const char *name_b,
1465                               struct diff_filespec *one,
1466                               struct diff_filespec *two,
1467                               struct userdiff_driver *textconv_one,
1468                               struct userdiff_driver *textconv_two,
1469                               struct diff_options *o)
1470 {
1471         int lc_a, lc_b;
1472         static struct strbuf a_name = STRBUF_INIT, b_name = STRBUF_INIT;
1473         const char *a_prefix, *b_prefix;
1474         char *data_one, *data_two;
1475         size_t size_one, size_two;
1476         struct emit_callback ecbdata;
1477         struct strbuf out = STRBUF_INIT;
1478
1479         if (diff_mnemonic_prefix && DIFF_OPT_TST(o, REVERSE_DIFF)) {
1480                 a_prefix = o->b_prefix;
1481                 b_prefix = o->a_prefix;
1482         } else {
1483                 a_prefix = o->a_prefix;
1484                 b_prefix = o->b_prefix;
1485         }
1486
1487         name_a += (*name_a == '/');
1488         name_b += (*name_b == '/');
1489
1490         strbuf_reset(&a_name);
1491         strbuf_reset(&b_name);
1492         quote_two_c_style(&a_name, a_prefix, name_a, 0);
1493         quote_two_c_style(&b_name, b_prefix, name_b, 0);
1494
1495         size_one = fill_textconv(textconv_one, one, &data_one);
1496         size_two = fill_textconv(textconv_two, two, &data_two);
1497
1498         memset(&ecbdata, 0, sizeof(ecbdata));
1499         ecbdata.color_diff = want_color(o->use_color);
1500         ecbdata.ws_rule = whitespace_rule(name_b);
1501         ecbdata.opt = o;
1502         if (ecbdata.ws_rule & WS_BLANK_AT_EOF) {
1503                 mmfile_t mf1, mf2;
1504                 mf1.ptr = (char *)data_one;
1505                 mf2.ptr = (char *)data_two;
1506                 mf1.size = size_one;
1507                 mf2.size = size_two;
1508                 check_blank_at_eof(&mf1, &mf2, &ecbdata);
1509         }
1510         ecbdata.lno_in_preimage = 1;
1511         ecbdata.lno_in_postimage = 1;
1512
1513         lc_a = count_lines(data_one, size_one);
1514         lc_b = count_lines(data_two, size_two);
1515
1516         emit_diff_symbol(o, DIFF_SYMBOL_FILEPAIR_MINUS,
1517                          a_name.buf, a_name.len, 0);
1518         emit_diff_symbol(o, DIFF_SYMBOL_FILEPAIR_PLUS,
1519                          b_name.buf, b_name.len, 0);
1520
1521         strbuf_addstr(&out, "@@ -");
1522         if (!o->irreversible_delete)
1523                 add_line_count(&out, lc_a);
1524         else
1525                 strbuf_addstr(&out, "?,?");
1526         strbuf_addstr(&out, " +");
1527         add_line_count(&out, lc_b);
1528         strbuf_addstr(&out, " @@\n");
1529         emit_diff_symbol(o, DIFF_SYMBOL_REWRITE_DIFF, out.buf, out.len, 0);
1530         strbuf_release(&out);
1531
1532         if (lc_a && !o->irreversible_delete)
1533                 emit_rewrite_lines(&ecbdata, '-', data_one, size_one);
1534         if (lc_b)
1535                 emit_rewrite_lines(&ecbdata, '+', data_two, size_two);
1536         if (textconv_one)
1537                 free((char *)data_one);
1538         if (textconv_two)
1539                 free((char *)data_two);
1540 }
1541
1542 struct diff_words_buffer {
1543         mmfile_t text;
1544         long alloc;
1545         struct diff_words_orig {
1546                 const char *begin, *end;
1547         } *orig;
1548         int orig_nr, orig_alloc;
1549 };
1550
1551 static void diff_words_append(char *line, unsigned long len,
1552                 struct diff_words_buffer *buffer)
1553 {
1554         ALLOC_GROW(buffer->text.ptr, buffer->text.size + len, buffer->alloc);
1555         line++;
1556         len--;
1557         memcpy(buffer->text.ptr + buffer->text.size, line, len);
1558         buffer->text.size += len;
1559         buffer->text.ptr[buffer->text.size] = '\0';
1560 }
1561
1562 struct diff_words_style_elem {
1563         const char *prefix;
1564         const char *suffix;
1565         const char *color; /* NULL; filled in by the setup code if
1566                             * color is enabled */
1567 };
1568
1569 struct diff_words_style {
1570         enum diff_words_type type;
1571         struct diff_words_style_elem new, old, ctx;
1572         const char *newline;
1573 };
1574
1575 static struct diff_words_style diff_words_styles[] = {
1576         { DIFF_WORDS_PORCELAIN, {"+", "\n"}, {"-", "\n"}, {" ", "\n"}, "~\n" },
1577         { DIFF_WORDS_PLAIN, {"{+", "+}"}, {"[-", "-]"}, {"", ""}, "\n" },
1578         { DIFF_WORDS_COLOR, {"", ""}, {"", ""}, {"", ""}, "\n" }
1579 };
1580
1581 struct diff_words_data {
1582         struct diff_words_buffer minus, plus;
1583         const char *current_plus;
1584         int last_minus;
1585         struct diff_options *opt;
1586         regex_t *word_regex;
1587         enum diff_words_type type;
1588         struct diff_words_style *style;
1589 };
1590
1591 static int fn_out_diff_words_write_helper(struct diff_options *o,
1592                                           struct diff_words_style_elem *st_el,
1593                                           const char *newline,
1594                                           size_t count, const char *buf)
1595 {
1596         int print = 0;
1597         struct strbuf sb = STRBUF_INIT;
1598
1599         while (count) {
1600                 char *p = memchr(buf, '\n', count);
1601                 if (print)
1602                         strbuf_addstr(&sb, diff_line_prefix(o));
1603
1604                 if (p != buf) {
1605                         const char *reset = st_el->color && *st_el->color ?
1606                                             GIT_COLOR_RESET : NULL;
1607                         if (st_el->color && *st_el->color)
1608                                 strbuf_addstr(&sb, st_el->color);
1609                         strbuf_addstr(&sb, st_el->prefix);
1610                         strbuf_add(&sb, buf, p ? p - buf : count);
1611                         strbuf_addstr(&sb, st_el->suffix);
1612                         if (reset)
1613                                 strbuf_addstr(&sb, reset);
1614                 }
1615                 if (!p)
1616                         goto out;
1617
1618                 strbuf_addstr(&sb, newline);
1619                 count -= p + 1 - buf;
1620                 buf = p + 1;
1621                 print = 1;
1622                 if (count) {
1623                         emit_diff_symbol(o, DIFF_SYMBOL_WORD_DIFF,
1624                                          sb.buf, sb.len, 0);
1625                         strbuf_reset(&sb);
1626                 }
1627         }
1628
1629 out:
1630         if (sb.len)
1631                 emit_diff_symbol(o, DIFF_SYMBOL_WORD_DIFF,
1632                                  sb.buf, sb.len, 0);
1633         strbuf_release(&sb);
1634         return 0;
1635 }
1636
1637 /*
1638  * '--color-words' algorithm can be described as:
1639  *
1640  *   1. collect the minus/plus lines of a diff hunk, divided into
1641  *      minus-lines and plus-lines;
1642  *
1643  *   2. break both minus-lines and plus-lines into words and
1644  *      place them into two mmfile_t with one word for each line;
1645  *
1646  *   3. use xdiff to run diff on the two mmfile_t to get the words level diff;
1647  *
1648  * And for the common parts of the both file, we output the plus side text.
1649  * diff_words->current_plus is used to trace the current position of the plus file
1650  * which printed. diff_words->last_minus is used to trace the last minus word
1651  * printed.
1652  *
1653  * For '--graph' to work with '--color-words', we need to output the graph prefix
1654  * on each line of color words output. Generally, there are two conditions on
1655  * which we should output the prefix.
1656  *
1657  *   1. diff_words->last_minus == 0 &&
1658  *      diff_words->current_plus == diff_words->plus.text.ptr
1659  *
1660  *      that is: the plus text must start as a new line, and if there is no minus
1661  *      word printed, a graph prefix must be printed.
1662  *
1663  *   2. diff_words->current_plus > diff_words->plus.text.ptr &&
1664  *      *(diff_words->current_plus - 1) == '\n'
1665  *
1666  *      that is: a graph prefix must be printed following a '\n'
1667  */
1668 static int color_words_output_graph_prefix(struct diff_words_data *diff_words)
1669 {
1670         if ((diff_words->last_minus == 0 &&
1671                 diff_words->current_plus == diff_words->plus.text.ptr) ||
1672                 (diff_words->current_plus > diff_words->plus.text.ptr &&
1673                 *(diff_words->current_plus - 1) == '\n')) {
1674                 return 1;
1675         } else {
1676                 return 0;
1677         }
1678 }
1679
1680 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
1681 {
1682         struct diff_words_data *diff_words = priv;
1683         struct diff_words_style *style = diff_words->style;
1684         int minus_first, minus_len, plus_first, plus_len;
1685         const char *minus_begin, *minus_end, *plus_begin, *plus_end;
1686         struct diff_options *opt = diff_words->opt;
1687         const char *line_prefix;
1688
1689         if (line[0] != '@' || parse_hunk_header(line, len,
1690                         &minus_first, &minus_len, &plus_first, &plus_len))
1691                 return;
1692
1693         assert(opt);
1694         line_prefix = diff_line_prefix(opt);
1695
1696         /* POSIX requires that first be decremented by one if len == 0... */
1697         if (minus_len) {
1698                 minus_begin = diff_words->minus.orig[minus_first].begin;
1699                 minus_end =
1700                         diff_words->minus.orig[minus_first + minus_len - 1].end;
1701         } else
1702                 minus_begin = minus_end =
1703                         diff_words->minus.orig[minus_first].end;
1704
1705         if (plus_len) {
1706                 plus_begin = diff_words->plus.orig[plus_first].begin;
1707                 plus_end = diff_words->plus.orig[plus_first + plus_len - 1].end;
1708         } else
1709                 plus_begin = plus_end = diff_words->plus.orig[plus_first].end;
1710
1711         if (color_words_output_graph_prefix(diff_words)) {
1712                 fputs(line_prefix, diff_words->opt->file);
1713         }
1714         if (diff_words->current_plus != plus_begin) {
1715                 fn_out_diff_words_write_helper(diff_words->opt,
1716                                 &style->ctx, style->newline,
1717                                 plus_begin - diff_words->current_plus,
1718                                 diff_words->current_plus);
1719         }
1720         if (minus_begin != minus_end) {
1721                 fn_out_diff_words_write_helper(diff_words->opt,
1722                                 &style->old, style->newline,
1723                                 minus_end - minus_begin, minus_begin);
1724         }
1725         if (plus_begin != plus_end) {
1726                 fn_out_diff_words_write_helper(diff_words->opt,
1727                                 &style->new, style->newline,
1728                                 plus_end - plus_begin, plus_begin);
1729         }
1730
1731         diff_words->current_plus = plus_end;
1732         diff_words->last_minus = minus_first;
1733 }
1734
1735 /* This function starts looking at *begin, and returns 0 iff a word was found. */
1736 static int find_word_boundaries(mmfile_t *buffer, regex_t *word_regex,
1737                 int *begin, int *end)
1738 {
1739         if (word_regex && *begin < buffer->size) {
1740                 regmatch_t match[1];
1741                 if (!regexec_buf(word_regex, buffer->ptr + *begin,
1742                                  buffer->size - *begin, 1, match, 0)) {
1743                         char *p = memchr(buffer->ptr + *begin + match[0].rm_so,
1744                                         '\n', match[0].rm_eo - match[0].rm_so);
1745                         *end = p ? p - buffer->ptr : match[0].rm_eo + *begin;
1746                         *begin += match[0].rm_so;
1747                         return *begin >= *end;
1748                 }
1749                 return -1;
1750         }
1751
1752         /* find the next word */
1753         while (*begin < buffer->size && isspace(buffer->ptr[*begin]))
1754                 (*begin)++;
1755         if (*begin >= buffer->size)
1756                 return -1;
1757
1758         /* find the end of the word */
1759         *end = *begin + 1;
1760         while (*end < buffer->size && !isspace(buffer->ptr[*end]))
1761                 (*end)++;
1762
1763         return 0;
1764 }
1765
1766 /*
1767  * This function splits the words in buffer->text, stores the list with
1768  * newline separator into out, and saves the offsets of the original words
1769  * in buffer->orig.
1770  */
1771 static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out,
1772                 regex_t *word_regex)
1773 {
1774         int i, j;
1775         long alloc = 0;
1776
1777         out->size = 0;
1778         out->ptr = NULL;
1779
1780         /* fake an empty "0th" word */
1781         ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
1782         buffer->orig[0].begin = buffer->orig[0].end = buffer->text.ptr;
1783         buffer->orig_nr = 1;
1784
1785         for (i = 0; i < buffer->text.size; i++) {
1786                 if (find_word_boundaries(&buffer->text, word_regex, &i, &j))
1787                         return;
1788
1789                 /* store original boundaries */
1790                 ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
1791                                 buffer->orig_alloc);
1792                 buffer->orig[buffer->orig_nr].begin = buffer->text.ptr + i;
1793                 buffer->orig[buffer->orig_nr].end = buffer->text.ptr + j;
1794                 buffer->orig_nr++;
1795
1796                 /* store one word */
1797                 ALLOC_GROW(out->ptr, out->size + j - i + 1, alloc);
1798                 memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
1799                 out->ptr[out->size + j - i] = '\n';
1800                 out->size += j - i + 1;
1801
1802                 i = j - 1;
1803         }
1804 }
1805
1806 /* this executes the word diff on the accumulated buffers */
1807 static void diff_words_show(struct diff_words_data *diff_words)
1808 {
1809         xpparam_t xpp;
1810         xdemitconf_t xecfg;
1811         mmfile_t minus, plus;
1812         struct diff_words_style *style = diff_words->style;
1813
1814         struct diff_options *opt = diff_words->opt;
1815         const char *line_prefix;
1816
1817         assert(opt);
1818         line_prefix = diff_line_prefix(opt);
1819
1820         /* special case: only removal */
1821         if (!diff_words->plus.text.size) {
1822                 emit_diff_symbol(diff_words->opt, DIFF_SYMBOL_WORD_DIFF,
1823                                  line_prefix, strlen(line_prefix), 0);
1824                 fn_out_diff_words_write_helper(diff_words->opt,
1825                         &style->old, style->newline,
1826                         diff_words->minus.text.size,
1827                         diff_words->minus.text.ptr);
1828                 diff_words->minus.text.size = 0;
1829                 return;
1830         }
1831
1832         diff_words->current_plus = diff_words->plus.text.ptr;
1833         diff_words->last_minus = 0;
1834
1835         memset(&xpp, 0, sizeof(xpp));
1836         memset(&xecfg, 0, sizeof(xecfg));
1837         diff_words_fill(&diff_words->minus, &minus, diff_words->word_regex);
1838         diff_words_fill(&diff_words->plus, &plus, diff_words->word_regex);
1839         xpp.flags = 0;
1840         /* as only the hunk header will be parsed, we need a 0-context */
1841         xecfg.ctxlen = 0;
1842         if (xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
1843                           &xpp, &xecfg))
1844                 die("unable to generate word diff");
1845         free(minus.ptr);
1846         free(plus.ptr);
1847         if (diff_words->current_plus != diff_words->plus.text.ptr +
1848                         diff_words->plus.text.size) {
1849                 if (color_words_output_graph_prefix(diff_words))
1850                         emit_diff_symbol(diff_words->opt, DIFF_SYMBOL_WORD_DIFF,
1851                                          line_prefix, strlen(line_prefix), 0);
1852                 fn_out_diff_words_write_helper(diff_words->opt,
1853                         &style->ctx, style->newline,
1854                         diff_words->plus.text.ptr + diff_words->plus.text.size
1855                         - diff_words->current_plus, diff_words->current_plus);
1856         }
1857         diff_words->minus.text.size = diff_words->plus.text.size = 0;
1858 }
1859
1860 /* In "color-words" mode, show word-diff of words accumulated in the buffer */
1861 static void diff_words_flush(struct emit_callback *ecbdata)
1862 {
1863         struct diff_options *wo = ecbdata->diff_words->opt;
1864
1865         if (ecbdata->diff_words->minus.text.size ||
1866             ecbdata->diff_words->plus.text.size)
1867                 diff_words_show(ecbdata->diff_words);
1868
1869         if (wo->emitted_symbols) {
1870                 struct diff_options *o = ecbdata->opt;
1871                 struct emitted_diff_symbols *wol = wo->emitted_symbols;
1872                 int i;
1873
1874                 /*
1875                  * NEEDSWORK:
1876                  * Instead of appending each, concat all words to a line?
1877                  */
1878                 for (i = 0; i < wol->nr; i++)
1879                         append_emitted_diff_symbol(o, &wol->buf[i]);
1880
1881                 for (i = 0; i < wol->nr; i++)
1882                         free((void *)wol->buf[i].line);
1883
1884                 wol->nr = 0;
1885         }
1886 }
1887
1888 static void diff_filespec_load_driver(struct diff_filespec *one)
1889 {
1890         /* Use already-loaded driver */
1891         if (one->driver)
1892                 return;
1893
1894         if (S_ISREG(one->mode))
1895                 one->driver = userdiff_find_by_path(one->path);
1896
1897         /* Fallback to default settings */
1898         if (!one->driver)
1899                 one->driver = userdiff_find_by_name("default");
1900 }
1901
1902 static const char *userdiff_word_regex(struct diff_filespec *one)
1903 {
1904         diff_filespec_load_driver(one);
1905         return one->driver->word_regex;
1906 }
1907
1908 static void init_diff_words_data(struct emit_callback *ecbdata,
1909                                  struct diff_options *orig_opts,
1910                                  struct diff_filespec *one,
1911                                  struct diff_filespec *two)
1912 {
1913         int i;
1914         struct diff_options *o = xmalloc(sizeof(struct diff_options));
1915         memcpy(o, orig_opts, sizeof(struct diff_options));
1916
1917         ecbdata->diff_words =
1918                 xcalloc(1, sizeof(struct diff_words_data));
1919         ecbdata->diff_words->type = o->word_diff;
1920         ecbdata->diff_words->opt = o;
1921
1922         if (orig_opts->emitted_symbols)
1923                 o->emitted_symbols =
1924                         xcalloc(1, sizeof(struct emitted_diff_symbols));
1925
1926         if (!o->word_regex)
1927                 o->word_regex = userdiff_word_regex(one);
1928         if (!o->word_regex)
1929                 o->word_regex = userdiff_word_regex(two);
1930         if (!o->word_regex)
1931                 o->word_regex = diff_word_regex_cfg;
1932         if (o->word_regex) {
1933                 ecbdata->diff_words->word_regex = (regex_t *)
1934                         xmalloc(sizeof(regex_t));
1935                 if (regcomp(ecbdata->diff_words->word_regex,
1936                             o->word_regex,
1937                             REG_EXTENDED | REG_NEWLINE))
1938                         die ("Invalid regular expression: %s",
1939                              o->word_regex);
1940         }
1941         for (i = 0; i < ARRAY_SIZE(diff_words_styles); i++) {
1942                 if (o->word_diff == diff_words_styles[i].type) {
1943                         ecbdata->diff_words->style =
1944                                 &diff_words_styles[i];
1945                         break;
1946                 }
1947         }
1948         if (want_color(o->use_color)) {
1949                 struct diff_words_style *st = ecbdata->diff_words->style;
1950                 st->old.color = diff_get_color_opt(o, DIFF_FILE_OLD);
1951                 st->new.color = diff_get_color_opt(o, DIFF_FILE_NEW);
1952                 st->ctx.color = diff_get_color_opt(o, DIFF_CONTEXT);
1953         }
1954 }
1955
1956 static void free_diff_words_data(struct emit_callback *ecbdata)
1957 {
1958         if (ecbdata->diff_words) {
1959                 diff_words_flush(ecbdata);
1960                 free (ecbdata->diff_words->opt->emitted_symbols);
1961                 free (ecbdata->diff_words->opt);
1962                 free (ecbdata->diff_words->minus.text.ptr);
1963                 free (ecbdata->diff_words->minus.orig);
1964                 free (ecbdata->diff_words->plus.text.ptr);
1965                 free (ecbdata->diff_words->plus.orig);
1966                 if (ecbdata->diff_words->word_regex) {
1967                         regfree(ecbdata->diff_words->word_regex);
1968                         free(ecbdata->diff_words->word_regex);
1969                 }
1970                 FREE_AND_NULL(ecbdata->diff_words);
1971         }
1972 }
1973
1974 const char *diff_get_color(int diff_use_color, enum color_diff ix)
1975 {
1976         if (want_color(diff_use_color))
1977                 return diff_colors[ix];
1978         return "";
1979 }
1980
1981 const char *diff_line_prefix(struct diff_options *opt)
1982 {
1983         struct strbuf *msgbuf;
1984         if (!opt->output_prefix)
1985                 return "";
1986
1987         msgbuf = opt->output_prefix(opt, opt->output_prefix_data);
1988         return msgbuf->buf;
1989 }
1990
1991 static unsigned long sane_truncate_line(struct emit_callback *ecb, char *line, unsigned long len)
1992 {
1993         const char *cp;
1994         unsigned long allot;
1995         size_t l = len;
1996
1997         cp = line;
1998         allot = l;
1999         while (0 < l) {
2000                 (void) utf8_width(&cp, &l);
2001                 if (!cp)
2002                         break; /* truncated in the middle? */
2003         }
2004         return allot - l;
2005 }
2006
2007 static void find_lno(const char *line, struct emit_callback *ecbdata)
2008 {
2009         const char *p;
2010         ecbdata->lno_in_preimage = 0;
2011         ecbdata->lno_in_postimage = 0;
2012         p = strchr(line, '-');
2013         if (!p)
2014                 return; /* cannot happen */
2015         ecbdata->lno_in_preimage = strtol(p + 1, NULL, 10);
2016         p = strchr(p, '+');
2017         if (!p)
2018                 return; /* cannot happen */
2019         ecbdata->lno_in_postimage = strtol(p + 1, NULL, 10);
2020 }
2021
2022 static void fn_out_consume(void *priv, char *line, unsigned long len)
2023 {
2024         struct emit_callback *ecbdata = priv;
2025         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
2026         struct diff_options *o = ecbdata->opt;
2027
2028         o->found_changes = 1;
2029
2030         if (ecbdata->header) {
2031                 emit_diff_symbol(o, DIFF_SYMBOL_HEADER,
2032                                  ecbdata->header->buf, ecbdata->header->len, 0);
2033                 strbuf_reset(ecbdata->header);
2034                 ecbdata->header = NULL;
2035         }
2036
2037         if (ecbdata->label_path[0]) {
2038                 emit_diff_symbol(o, DIFF_SYMBOL_FILEPAIR_MINUS,
2039                                  ecbdata->label_path[0],
2040                                  strlen(ecbdata->label_path[0]), 0);
2041                 emit_diff_symbol(o, DIFF_SYMBOL_FILEPAIR_PLUS,
2042                                  ecbdata->label_path[1],
2043                                  strlen(ecbdata->label_path[1]), 0);
2044                 ecbdata->label_path[0] = ecbdata->label_path[1] = NULL;
2045         }
2046
2047         if (diff_suppress_blank_empty
2048             && len == 2 && line[0] == ' ' && line[1] == '\n') {
2049                 line[0] = '\n';
2050                 len = 1;
2051         }
2052
2053         if (line[0] == '@') {
2054                 if (ecbdata->diff_words)
2055                         diff_words_flush(ecbdata);
2056                 len = sane_truncate_line(ecbdata, line, len);
2057                 find_lno(line, ecbdata);
2058                 emit_hunk_header(ecbdata, line, len);
2059                 return;
2060         }
2061
2062         if (ecbdata->diff_words) {
2063                 enum diff_symbol s =
2064                         ecbdata->diff_words->type == DIFF_WORDS_PORCELAIN ?
2065                         DIFF_SYMBOL_WORDS_PORCELAIN : DIFF_SYMBOL_WORDS;
2066                 if (line[0] == '-') {
2067                         diff_words_append(line, len,
2068                                           &ecbdata->diff_words->minus);
2069                         return;
2070                 } else if (line[0] == '+') {
2071                         diff_words_append(line, len,
2072                                           &ecbdata->diff_words->plus);
2073                         return;
2074                 } else if (starts_with(line, "\\ ")) {
2075                         /*
2076                          * Eat the "no newline at eof" marker as if we
2077                          * saw a "+" or "-" line with nothing on it,
2078                          * and return without diff_words_flush() to
2079                          * defer processing. If this is the end of
2080                          * preimage, more "+" lines may come after it.
2081                          */
2082                         return;
2083                 }
2084                 diff_words_flush(ecbdata);
2085                 emit_diff_symbol(o, s, line, len, 0);
2086                 return;
2087         }
2088
2089         switch (line[0]) {
2090         case '+':
2091                 ecbdata->lno_in_postimage++;
2092                 emit_add_line(reset, ecbdata, line + 1, len - 1);
2093                 break;
2094         case '-':
2095                 ecbdata->lno_in_preimage++;
2096                 emit_del_line(reset, ecbdata, line + 1, len - 1);
2097                 break;
2098         case ' ':
2099                 ecbdata->lno_in_postimage++;
2100                 ecbdata->lno_in_preimage++;
2101                 emit_context_line(reset, ecbdata, line + 1, len - 1);
2102                 break;
2103         default:
2104                 /* incomplete line at the end */
2105                 ecbdata->lno_in_preimage++;
2106                 emit_diff_symbol(o, DIFF_SYMBOL_CONTEXT_INCOMPLETE,
2107                                  line, len, 0);
2108                 break;
2109         }
2110 }
2111
2112 static char *pprint_rename(const char *a, const char *b)
2113 {
2114         const char *old = a;
2115         const char *new = b;
2116         struct strbuf name = STRBUF_INIT;
2117         int pfx_length, sfx_length;
2118         int pfx_adjust_for_slash;
2119         int len_a = strlen(a);
2120         int len_b = strlen(b);
2121         int a_midlen, b_midlen;
2122         int qlen_a = quote_c_style(a, NULL, NULL, 0);
2123         int qlen_b = quote_c_style(b, NULL, NULL, 0);
2124
2125         if (qlen_a || qlen_b) {
2126                 quote_c_style(a, &name, NULL, 0);
2127                 strbuf_addstr(&name, " => ");
2128                 quote_c_style(b, &name, NULL, 0);
2129                 return strbuf_detach(&name, NULL);
2130         }
2131
2132         /* Find common prefix */
2133         pfx_length = 0;
2134         while (*old && *new && *old == *new) {
2135                 if (*old == '/')
2136                         pfx_length = old - a + 1;
2137                 old++;
2138                 new++;
2139         }
2140
2141         /* Find common suffix */
2142         old = a + len_a;
2143         new = b + len_b;
2144         sfx_length = 0;
2145         /*
2146          * If there is a common prefix, it must end in a slash.  In
2147          * that case we let this loop run 1 into the prefix to see the
2148          * same slash.
2149          *
2150          * If there is no common prefix, we cannot do this as it would
2151          * underrun the input strings.
2152          */
2153         pfx_adjust_for_slash = (pfx_length ? 1 : 0);
2154         while (a + pfx_length - pfx_adjust_for_slash <= old &&
2155                b + pfx_length - pfx_adjust_for_slash <= new &&
2156                *old == *new) {
2157                 if (*old == '/')
2158                         sfx_length = len_a - (old - a);
2159                 old--;
2160                 new--;
2161         }
2162
2163         /*
2164          * pfx{mid-a => mid-b}sfx
2165          * {pfx-a => pfx-b}sfx
2166          * pfx{sfx-a => sfx-b}
2167          * name-a => name-b
2168          */
2169         a_midlen = len_a - pfx_length - sfx_length;
2170         b_midlen = len_b - pfx_length - sfx_length;
2171         if (a_midlen < 0)
2172                 a_midlen = 0;
2173         if (b_midlen < 0)
2174                 b_midlen = 0;
2175
2176         strbuf_grow(&name, pfx_length + a_midlen + b_midlen + sfx_length + 7);
2177         if (pfx_length + sfx_length) {
2178                 strbuf_add(&name, a, pfx_length);
2179                 strbuf_addch(&name, '{');
2180         }
2181         strbuf_add(&name, a + pfx_length, a_midlen);
2182         strbuf_addstr(&name, " => ");
2183         strbuf_add(&name, b + pfx_length, b_midlen);
2184         if (pfx_length + sfx_length) {
2185                 strbuf_addch(&name, '}');
2186                 strbuf_add(&name, a + len_a - sfx_length, sfx_length);
2187         }
2188         return strbuf_detach(&name, NULL);
2189 }
2190
2191 struct diffstat_t {
2192         int nr;
2193         int alloc;
2194         struct diffstat_file {
2195                 char *from_name;
2196                 char *name;
2197                 char *print_name;
2198                 unsigned is_unmerged:1;
2199                 unsigned is_binary:1;
2200                 unsigned is_renamed:1;
2201                 unsigned is_interesting:1;
2202                 uintmax_t added, deleted;
2203         } **files;
2204 };
2205
2206 static struct diffstat_file *diffstat_add(struct diffstat_t *diffstat,
2207                                           const char *name_a,
2208                                           const char *name_b)
2209 {
2210         struct diffstat_file *x;
2211         x = xcalloc(1, sizeof(*x));
2212         ALLOC_GROW(diffstat->files, diffstat->nr + 1, diffstat->alloc);
2213         diffstat->files[diffstat->nr++] = x;
2214         if (name_b) {
2215                 x->from_name = xstrdup(name_a);
2216                 x->name = xstrdup(name_b);
2217                 x->is_renamed = 1;
2218         }
2219         else {
2220                 x->from_name = NULL;
2221                 x->name = xstrdup(name_a);
2222         }
2223         return x;
2224 }
2225
2226 static void diffstat_consume(void *priv, char *line, unsigned long len)
2227 {
2228         struct diffstat_t *diffstat = priv;
2229         struct diffstat_file *x = diffstat->files[diffstat->nr - 1];
2230
2231         if (line[0] == '+')
2232                 x->added++;
2233         else if (line[0] == '-')
2234                 x->deleted++;
2235 }
2236
2237 const char mime_boundary_leader[] = "------------";
2238
2239 static int scale_linear(int it, int width, int max_change)
2240 {
2241         if (!it)
2242                 return 0;
2243         /*
2244          * make sure that at least one '-' or '+' is printed if
2245          * there is any change to this path. The easiest way is to
2246          * scale linearly as if the alloted width is one column shorter
2247          * than it is, and then add 1 to the result.
2248          */
2249         return 1 + (it * (width - 1) / max_change);
2250 }
2251
2252 static void show_graph(struct strbuf *out, char ch, int cnt,
2253                        const char *set, const char *reset)
2254 {
2255         if (cnt <= 0)
2256                 return;
2257         strbuf_addstr(out, set);
2258         strbuf_addchars(out, ch, cnt);
2259         strbuf_addstr(out, reset);
2260 }
2261
2262 static void fill_print_name(struct diffstat_file *file)
2263 {
2264         char *pname;
2265
2266         if (file->print_name)
2267                 return;
2268
2269         if (!file->is_renamed) {
2270                 struct strbuf buf = STRBUF_INIT;
2271                 if (quote_c_style(file->name, &buf, NULL, 0)) {
2272                         pname = strbuf_detach(&buf, NULL);
2273                 } else {
2274                         pname = file->name;
2275                         strbuf_release(&buf);
2276                 }
2277         } else {
2278                 pname = pprint_rename(file->from_name, file->name);
2279         }
2280         file->print_name = pname;
2281 }
2282
2283 static void print_stat_summary_inserts_deletes(struct diff_options *options,
2284                 int files, int insertions, int deletions)
2285 {
2286         struct strbuf sb = STRBUF_INIT;
2287
2288         if (!files) {
2289                 assert(insertions == 0 && deletions == 0);
2290                 emit_diff_symbol(options, DIFF_SYMBOL_STATS_SUMMARY_NO_FILES,
2291                                  NULL, 0, 0);
2292                 return;
2293         }
2294
2295         strbuf_addf(&sb,
2296                     (files == 1) ? " %d file changed" : " %d files changed",
2297                     files);
2298
2299         /*
2300          * For binary diff, the caller may want to print "x files
2301          * changed" with insertions == 0 && deletions == 0.
2302          *
2303          * Not omitting "0 insertions(+), 0 deletions(-)" in this case
2304          * is probably less confusing (i.e skip over "2 files changed
2305          * but nothing about added/removed lines? Is this a bug in Git?").
2306          */
2307         if (insertions || deletions == 0) {
2308                 strbuf_addf(&sb,
2309                             (insertions == 1) ? ", %d insertion(+)" : ", %d insertions(+)",
2310                             insertions);
2311         }
2312
2313         if (deletions || insertions == 0) {
2314                 strbuf_addf(&sb,
2315                             (deletions == 1) ? ", %d deletion(-)" : ", %d deletions(-)",
2316                             deletions);
2317         }
2318         strbuf_addch(&sb, '\n');
2319         emit_diff_symbol(options, DIFF_SYMBOL_STATS_SUMMARY_INSERTS_DELETES,
2320                          sb.buf, sb.len, 0);
2321         strbuf_release(&sb);
2322 }
2323
2324 void print_stat_summary(FILE *fp, int files,
2325                         int insertions, int deletions)
2326 {
2327         struct diff_options o;
2328         memset(&o, 0, sizeof(o));
2329         o.file = fp;
2330
2331         print_stat_summary_inserts_deletes(&o, files, insertions, deletions);
2332 }
2333
2334 static void show_stats(struct diffstat_t *data, struct diff_options *options)
2335 {
2336         int i, len, add, del, adds = 0, dels = 0;
2337         uintmax_t max_change = 0, max_len = 0;
2338         int total_files = data->nr, count;
2339         int width, name_width, graph_width, number_width = 0, bin_width = 0;
2340         const char *reset, *add_c, *del_c;
2341         int extra_shown = 0;
2342         const char *line_prefix = diff_line_prefix(options);
2343         struct strbuf out = STRBUF_INIT;
2344
2345         if (data->nr == 0)
2346                 return;
2347
2348         count = options->stat_count ? options->stat_count : data->nr;
2349
2350         reset = diff_get_color_opt(options, DIFF_RESET);
2351         add_c = diff_get_color_opt(options, DIFF_FILE_NEW);
2352         del_c = diff_get_color_opt(options, DIFF_FILE_OLD);
2353
2354         /*
2355          * Find the longest filename and max number of changes
2356          */
2357         for (i = 0; (i < count) && (i < data->nr); i++) {
2358                 struct diffstat_file *file = data->files[i];
2359                 uintmax_t change = file->added + file->deleted;
2360
2361                 if (!file->is_interesting && (change == 0)) {
2362                         count++; /* not shown == room for one more */
2363                         continue;
2364                 }
2365                 fill_print_name(file);
2366                 len = strlen(file->print_name);
2367                 if (max_len < len)
2368                         max_len = len;
2369
2370                 if (file->is_unmerged) {
2371                         /* "Unmerged" is 8 characters */
2372                         bin_width = bin_width < 8 ? 8 : bin_width;
2373                         continue;
2374                 }
2375                 if (file->is_binary) {
2376                         /* "Bin XXX -> YYY bytes" */
2377                         int w = 14 + decimal_width(file->added)
2378                                 + decimal_width(file->deleted);
2379                         bin_width = bin_width < w ? w : bin_width;
2380                         /* Display change counts aligned with "Bin" */
2381                         number_width = 3;
2382                         continue;
2383                 }
2384
2385                 if (max_change < change)
2386                         max_change = change;
2387         }
2388         count = i; /* where we can stop scanning in data->files[] */
2389
2390         /*
2391          * We have width = stat_width or term_columns() columns total.
2392          * We want a maximum of min(max_len, stat_name_width) for the name part.
2393          * We want a maximum of min(max_change, stat_graph_width) for the +- part.
2394          * We also need 1 for " " and 4 + decimal_width(max_change)
2395          * for " | NNNN " and one the empty column at the end, altogether
2396          * 6 + decimal_width(max_change).
2397          *
2398          * If there's not enough space, we will use the smaller of
2399          * stat_name_width (if set) and 5/8*width for the filename,
2400          * and the rest for constant elements + graph part, but no more
2401          * than stat_graph_width for the graph part.
2402          * (5/8 gives 50 for filename and 30 for the constant parts + graph
2403          * for the standard terminal size).
2404          *
2405          * In other words: stat_width limits the maximum width, and
2406          * stat_name_width fixes the maximum width of the filename,
2407          * and is also used to divide available columns if there
2408          * aren't enough.
2409          *
2410          * Binary files are displayed with "Bin XXX -> YYY bytes"
2411          * instead of the change count and graph. This part is treated
2412          * similarly to the graph part, except that it is not
2413          * "scaled". If total width is too small to accommodate the
2414          * guaranteed minimum width of the filename part and the
2415          * separators and this message, this message will "overflow"
2416          * making the line longer than the maximum width.
2417          */
2418
2419         if (options->stat_width == -1)
2420                 width = term_columns() - strlen(line_prefix);
2421         else
2422                 width = options->stat_width ? options->stat_width : 80;
2423         number_width = decimal_width(max_change) > number_width ?
2424                 decimal_width(max_change) : number_width;
2425
2426         if (options->stat_graph_width == -1)
2427                 options->stat_graph_width = diff_stat_graph_width;
2428
2429         /*
2430          * Guarantee 3/8*16==6 for the graph part
2431          * and 5/8*16==10 for the filename part
2432          */
2433         if (width < 16 + 6 + number_width)
2434                 width = 16 + 6 + number_width;
2435
2436         /*
2437          * First assign sizes that are wanted, ignoring available width.
2438          * strlen("Bin XXX -> YYY bytes") == bin_width, and the part
2439          * starting from "XXX" should fit in graph_width.
2440          */
2441         graph_width = max_change + 4 > bin_width ? max_change : bin_width - 4;
2442         if (options->stat_graph_width &&
2443             options->stat_graph_width < graph_width)
2444                 graph_width = options->stat_graph_width;
2445
2446         name_width = (options->stat_name_width > 0 &&
2447                       options->stat_name_width < max_len) ?
2448                 options->stat_name_width : max_len;
2449
2450         /*
2451          * Adjust adjustable widths not to exceed maximum width
2452          */
2453         if (name_width + number_width + 6 + graph_width > width) {
2454                 if (graph_width > width * 3/8 - number_width - 6) {
2455                         graph_width = width * 3/8 - number_width - 6;
2456                         if (graph_width < 6)
2457                                 graph_width = 6;
2458                 }
2459
2460                 if (options->stat_graph_width &&
2461                     graph_width > options->stat_graph_width)
2462                         graph_width = options->stat_graph_width;
2463                 if (name_width > width - number_width - 6 - graph_width)
2464                         name_width = width - number_width - 6 - graph_width;
2465                 else
2466                         graph_width = width - number_width - 6 - name_width;
2467         }
2468
2469         /*
2470          * From here name_width is the width of the name area,
2471          * and graph_width is the width of the graph area.
2472          * max_change is used to scale graph properly.
2473          */
2474         for (i = 0; i < count; i++) {
2475                 const char *prefix = "";
2476                 struct diffstat_file *file = data->files[i];
2477                 char *name = file->print_name;
2478                 uintmax_t added = file->added;
2479                 uintmax_t deleted = file->deleted;
2480                 int name_len;
2481
2482                 if (!file->is_interesting && (added + deleted == 0))
2483                         continue;
2484
2485                 /*
2486                  * "scale" the filename
2487                  */
2488                 len = name_width;
2489                 name_len = strlen(name);
2490                 if (name_width < name_len) {
2491                         char *slash;
2492                         prefix = "...";
2493                         len -= 3;
2494                         name += name_len - len;
2495                         slash = strchr(name, '/');
2496                         if (slash)
2497                                 name = slash;
2498                 }
2499
2500                 if (file->is_binary) {
2501                         strbuf_addf(&out, " %s%-*s |", prefix, len, name);
2502                         strbuf_addf(&out, " %*s", number_width, "Bin");
2503                         if (!added && !deleted) {
2504                                 strbuf_addch(&out, '\n');
2505                                 emit_diff_symbol(options, DIFF_SYMBOL_STATS_LINE,
2506                                                  out.buf, out.len, 0);
2507                                 strbuf_reset(&out);
2508                                 continue;
2509                         }
2510                         strbuf_addf(&out, " %s%"PRIuMAX"%s",
2511                                 del_c, deleted, reset);
2512                         strbuf_addstr(&out, " -> ");
2513                         strbuf_addf(&out, "%s%"PRIuMAX"%s",
2514                                 add_c, added, reset);
2515                         strbuf_addstr(&out, " bytes\n");
2516                         emit_diff_symbol(options, DIFF_SYMBOL_STATS_LINE,
2517                                          out.buf, out.len, 0);
2518                         strbuf_reset(&out);
2519                         continue;
2520                 }
2521                 else if (file->is_unmerged) {
2522                         strbuf_addf(&out, " %s%-*s |", prefix, len, name);
2523                         strbuf_addstr(&out, " Unmerged\n");
2524                         emit_diff_symbol(options, DIFF_SYMBOL_STATS_LINE,
2525                                          out.buf, out.len, 0);
2526                         strbuf_reset(&out);
2527                         continue;
2528                 }
2529
2530                 /*
2531                  * scale the add/delete
2532                  */
2533                 add = added;
2534                 del = deleted;
2535
2536                 if (graph_width <= max_change) {
2537                         int total = scale_linear(add + del, graph_width, max_change);
2538                         if (total < 2 && add && del)
2539                                 /* width >= 2 due to the sanity check */
2540                                 total = 2;
2541                         if (add < del) {
2542                                 add = scale_linear(add, graph_width, max_change);
2543                                 del = total - add;
2544                         } else {
2545                                 del = scale_linear(del, graph_width, max_change);
2546                                 add = total - del;
2547                         }
2548                 }
2549                 strbuf_addf(&out, " %s%-*s |", prefix, len, name);
2550                 strbuf_addf(&out, " %*"PRIuMAX"%s",
2551                         number_width, added + deleted,
2552                         added + deleted ? " " : "");
2553                 show_graph(&out, '+', add, add_c, reset);
2554                 show_graph(&out, '-', del, del_c, reset);
2555                 strbuf_addch(&out, '\n');
2556                 emit_diff_symbol(options, DIFF_SYMBOL_STATS_LINE,
2557                                  out.buf, out.len, 0);
2558                 strbuf_reset(&out);
2559         }
2560
2561         for (i = 0; i < data->nr; i++) {
2562                 struct diffstat_file *file = data->files[i];
2563                 uintmax_t added = file->added;
2564                 uintmax_t deleted = file->deleted;
2565
2566                 if (file->is_unmerged ||
2567                     (!file->is_interesting && (added + deleted == 0))) {
2568                         total_files--;
2569                         continue;
2570                 }
2571
2572                 if (!file->is_binary) {
2573                         adds += added;
2574                         dels += deleted;
2575                 }
2576                 if (i < count)
2577                         continue;
2578                 if (!extra_shown)
2579                         emit_diff_symbol(options,
2580                                          DIFF_SYMBOL_STATS_SUMMARY_ABBREV,
2581                                          NULL, 0, 0);
2582                 extra_shown = 1;
2583         }
2584
2585         print_stat_summary_inserts_deletes(options, total_files, adds, dels);
2586         strbuf_release(&out);
2587 }
2588
2589 static void show_shortstats(struct diffstat_t *data, struct diff_options *options)
2590 {
2591         int i, adds = 0, dels = 0, total_files = data->nr;
2592
2593         if (data->nr == 0)
2594                 return;
2595
2596         for (i = 0; i < data->nr; i++) {
2597                 int added = data->files[i]->added;
2598                 int deleted = data->files[i]->deleted;
2599
2600                 if (data->files[i]->is_unmerged ||
2601                     (!data->files[i]->is_interesting && (added + deleted == 0))) {
2602                         total_files--;
2603                 } else if (!data->files[i]->is_binary) { /* don't count bytes */
2604                         adds += added;
2605                         dels += deleted;
2606                 }
2607         }
2608         print_stat_summary_inserts_deletes(options, total_files, adds, dels);
2609 }
2610
2611 static void show_numstat(struct diffstat_t *data, struct diff_options *options)
2612 {
2613         int i;
2614
2615         if (data->nr == 0)
2616                 return;
2617
2618         for (i = 0; i < data->nr; i++) {
2619                 struct diffstat_file *file = data->files[i];
2620
2621                 fprintf(options->file, "%s", diff_line_prefix(options));
2622
2623                 if (file->is_binary)
2624                         fprintf(options->file, "-\t-\t");
2625                 else
2626                         fprintf(options->file,
2627                                 "%"PRIuMAX"\t%"PRIuMAX"\t",
2628                                 file->added, file->deleted);
2629                 if (options->line_termination) {
2630                         fill_print_name(file);
2631                         if (!file->is_renamed)
2632                                 write_name_quoted(file->name, options->file,
2633                                                   options->line_termination);
2634                         else {
2635                                 fputs(file->print_name, options->file);
2636                                 putc(options->line_termination, options->file);
2637                         }
2638                 } else {
2639                         if (file->is_renamed) {
2640                                 putc('\0', options->file);
2641                                 write_name_quoted(file->from_name, options->file, '\0');
2642                         }
2643                         write_name_quoted(file->name, options->file, '\0');
2644                 }
2645         }
2646 }
2647
2648 struct dirstat_file {
2649         const char *name;
2650         unsigned long changed;
2651 };
2652
2653 struct dirstat_dir {
2654         struct dirstat_file *files;
2655         int alloc, nr, permille, cumulative;
2656 };
2657
2658 static long gather_dirstat(struct diff_options *opt, struct dirstat_dir *dir,
2659                 unsigned long changed, const char *base, int baselen)
2660 {
2661         unsigned long this_dir = 0;
2662         unsigned int sources = 0;
2663         const char *line_prefix = diff_line_prefix(opt);
2664
2665         while (dir->nr) {
2666                 struct dirstat_file *f = dir->files;
2667                 int namelen = strlen(f->name);
2668                 unsigned long this;
2669                 char *slash;
2670
2671                 if (namelen < baselen)
2672                         break;
2673                 if (memcmp(f->name, base, baselen))
2674                         break;
2675                 slash = strchr(f->name + baselen, '/');
2676                 if (slash) {
2677                         int newbaselen = slash + 1 - f->name;
2678                         this = gather_dirstat(opt, dir, changed, f->name, newbaselen);
2679                         sources++;
2680                 } else {
2681                         this = f->changed;
2682                         dir->files++;
2683                         dir->nr--;
2684                         sources += 2;
2685                 }
2686                 this_dir += this;
2687         }
2688
2689         /*
2690          * We don't report dirstat's for
2691          *  - the top level
2692          *  - or cases where everything came from a single directory
2693          *    under this directory (sources == 1).
2694          */
2695         if (baselen && sources != 1) {
2696                 if (this_dir) {
2697                         int permille = this_dir * 1000 / changed;
2698                         if (permille >= dir->permille) {
2699                                 fprintf(opt->file, "%s%4d.%01d%% %.*s\n", line_prefix,
2700                                         permille / 10, permille % 10, baselen, base);
2701                                 if (!dir->cumulative)
2702                                         return 0;
2703                         }
2704                 }
2705         }
2706         return this_dir;
2707 }
2708
2709 static int dirstat_compare(const void *_a, const void *_b)
2710 {
2711         const struct dirstat_file *a = _a;
2712         const struct dirstat_file *b = _b;
2713         return strcmp(a->name, b->name);
2714 }
2715
2716 static void show_dirstat(struct diff_options *options)
2717 {
2718         int i;
2719         unsigned long changed;
2720         struct dirstat_dir dir;
2721         struct diff_queue_struct *q = &diff_queued_diff;
2722
2723         dir.files = NULL;
2724         dir.alloc = 0;
2725         dir.nr = 0;
2726         dir.permille = options->dirstat_permille;
2727         dir.cumulative = DIFF_OPT_TST(options, DIRSTAT_CUMULATIVE);
2728
2729         changed = 0;
2730         for (i = 0; i < q->nr; i++) {
2731                 struct diff_filepair *p = q->queue[i];
2732                 const char *name;
2733                 unsigned long copied, added, damage;
2734                 int content_changed;
2735
2736                 name = p->two->path ? p->two->path : p->one->path;
2737
2738                 if (p->one->oid_valid && p->two->oid_valid)
2739                         content_changed = oidcmp(&p->one->oid, &p->two->oid);
2740                 else
2741                         content_changed = 1;
2742
2743                 if (!content_changed) {
2744                         /*
2745                          * The SHA1 has not changed, so pre-/post-content is
2746                          * identical. We can therefore skip looking at the
2747                          * file contents altogether.
2748                          */
2749                         damage = 0;
2750                         goto found_damage;
2751                 }
2752
2753                 if (DIFF_OPT_TST(options, DIRSTAT_BY_FILE)) {
2754                         /*
2755                          * In --dirstat-by-file mode, we don't really need to
2756                          * look at the actual file contents at all.
2757                          * The fact that the SHA1 changed is enough for us to
2758                          * add this file to the list of results
2759                          * (with each file contributing equal damage).
2760                          */
2761                         damage = 1;
2762                         goto found_damage;
2763                 }
2764
2765                 if (DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
2766                         diff_populate_filespec(p->one, 0);
2767                         diff_populate_filespec(p->two, 0);
2768                         diffcore_count_changes(p->one, p->two, NULL, NULL,
2769                                                &copied, &added);
2770                         diff_free_filespec_data(p->one);
2771                         diff_free_filespec_data(p->two);
2772                 } else if (DIFF_FILE_VALID(p->one)) {
2773                         diff_populate_filespec(p->one, CHECK_SIZE_ONLY);
2774                         copied = added = 0;
2775                         diff_free_filespec_data(p->one);
2776                 } else if (DIFF_FILE_VALID(p->two)) {
2777                         diff_populate_filespec(p->two, CHECK_SIZE_ONLY);
2778                         copied = 0;
2779                         added = p->two->size;
2780                         diff_free_filespec_data(p->two);
2781                 } else
2782                         continue;
2783
2784                 /*
2785                  * Original minus copied is the removed material,
2786                  * added is the new material.  They are both damages
2787                  * made to the preimage.
2788                  * If the resulting damage is zero, we know that
2789                  * diffcore_count_changes() considers the two entries to
2790                  * be identical, but since content_changed is true, we
2791                  * know that there must have been _some_ kind of change,
2792                  * so we force all entries to have damage > 0.
2793                  */
2794                 damage = (p->one->size - copied) + added;
2795                 if (!damage)
2796                         damage = 1;
2797
2798 found_damage:
2799                 ALLOC_GROW(dir.files, dir.nr + 1, dir.alloc);
2800                 dir.files[dir.nr].name = name;
2801                 dir.files[dir.nr].changed = damage;
2802                 changed += damage;
2803                 dir.nr++;
2804         }
2805
2806         /* This can happen even with many files, if everything was renames */
2807         if (!changed)
2808                 return;
2809
2810         /* Show all directories with more than x% of the changes */
2811         QSORT(dir.files, dir.nr, dirstat_compare);
2812         gather_dirstat(options, &dir, changed, "", 0);
2813 }
2814
2815 static void show_dirstat_by_line(struct diffstat_t *data, struct diff_options *options)
2816 {
2817         int i;
2818         unsigned long changed;
2819         struct dirstat_dir dir;
2820
2821         if (data->nr == 0)
2822                 return;
2823
2824         dir.files = NULL;
2825         dir.alloc = 0;
2826         dir.nr = 0;
2827         dir.permille = options->dirstat_permille;
2828         dir.cumulative = DIFF_OPT_TST(options, DIRSTAT_CUMULATIVE);
2829
2830         changed = 0;
2831         for (i = 0; i < data->nr; i++) {
2832                 struct diffstat_file *file = data->files[i];
2833                 unsigned long damage = file->added + file->deleted;
2834                 if (file->is_binary)
2835                         /*
2836                          * binary files counts bytes, not lines. Must find some
2837                          * way to normalize binary bytes vs. textual lines.
2838                          * The following heuristic assumes that there are 64
2839                          * bytes per "line".
2840                          * This is stupid and ugly, but very cheap...
2841                          */
2842                         damage = DIV_ROUND_UP(damage, 64);
2843                 ALLOC_GROW(dir.files, dir.nr + 1, dir.alloc);
2844                 dir.files[dir.nr].name = file->name;
2845                 dir.files[dir.nr].changed = damage;
2846                 changed += damage;
2847                 dir.nr++;
2848         }
2849
2850         /* This can happen even with many files, if everything was renames */
2851         if (!changed)
2852                 return;
2853
2854         /* Show all directories with more than x% of the changes */
2855         QSORT(dir.files, dir.nr, dirstat_compare);
2856         gather_dirstat(options, &dir, changed, "", 0);
2857 }
2858
2859 static void free_diffstat_info(struct diffstat_t *diffstat)
2860 {
2861         int i;
2862         for (i = 0; i < diffstat->nr; i++) {
2863                 struct diffstat_file *f = diffstat->files[i];
2864                 if (f->name != f->print_name)
2865                         free(f->print_name);
2866                 free(f->name);
2867                 free(f->from_name);
2868                 free(f);
2869         }
2870         free(diffstat->files);
2871 }
2872
2873 struct checkdiff_t {
2874         const char *filename;
2875         int lineno;
2876         int conflict_marker_size;
2877         struct diff_options *o;
2878         unsigned ws_rule;
2879         unsigned status;
2880 };
2881
2882 static int is_conflict_marker(const char *line, int marker_size, unsigned long len)
2883 {
2884         char firstchar;
2885         int cnt;
2886
2887         if (len < marker_size + 1)
2888                 return 0;
2889         firstchar = line[0];
2890         switch (firstchar) {
2891         case '=': case '>': case '<': case '|':
2892                 break;
2893         default:
2894                 return 0;
2895         }
2896         for (cnt = 1; cnt < marker_size; cnt++)
2897                 if (line[cnt] != firstchar)
2898                         return 0;
2899         /* line[1] thru line[marker_size-1] are same as firstchar */
2900         if (len < marker_size + 1 || !isspace(line[marker_size]))
2901                 return 0;
2902         return 1;
2903 }
2904
2905 static void checkdiff_consume(void *priv, char *line, unsigned long len)
2906 {
2907         struct checkdiff_t *data = priv;
2908         int marker_size = data->conflict_marker_size;
2909         const char *ws = diff_get_color(data->o->use_color, DIFF_WHITESPACE);
2910         const char *reset = diff_get_color(data->o->use_color, DIFF_RESET);
2911         const char *set = diff_get_color(data->o->use_color, DIFF_FILE_NEW);
2912         char *err;
2913         const char *line_prefix;
2914
2915         assert(data->o);
2916         line_prefix = diff_line_prefix(data->o);
2917
2918         if (line[0] == '+') {
2919                 unsigned bad;
2920                 data->lineno++;
2921                 if (is_conflict_marker(line + 1, marker_size, len - 1)) {
2922                         data->status |= 1;
2923                         fprintf(data->o->file,
2924                                 "%s%s:%d: leftover conflict marker\n",
2925                                 line_prefix, data->filename, data->lineno);
2926                 }
2927                 bad = ws_check(line + 1, len - 1, data->ws_rule);
2928                 if (!bad)
2929                         return;
2930                 data->status |= bad;
2931                 err = whitespace_error_string(bad);
2932                 fprintf(data->o->file, "%s%s:%d: %s.\n",
2933                         line_prefix, data->filename, data->lineno, err);
2934                 free(err);
2935                 emit_line(data->o, set, reset, line, 1);
2936                 ws_check_emit(line + 1, len - 1, data->ws_rule,
2937                               data->o->file, set, reset, ws);
2938         } else if (line[0] == ' ') {
2939                 data->lineno++;
2940         } else if (line[0] == '@') {
2941                 char *plus = strchr(line, '+');
2942                 if (plus)
2943                         data->lineno = strtol(plus, NULL, 10) - 1;
2944                 else
2945                         die("invalid diff");
2946         }
2947 }
2948
2949 static unsigned char *deflate_it(char *data,
2950                                  unsigned long size,
2951                                  unsigned long *result_size)
2952 {
2953         int bound;
2954         unsigned char *deflated;
2955         git_zstream stream;
2956
2957         git_deflate_init(&stream, zlib_compression_level);
2958         bound = git_deflate_bound(&stream, size);
2959         deflated = xmalloc(bound);
2960         stream.next_out = deflated;
2961         stream.avail_out = bound;
2962
2963         stream.next_in = (unsigned char *)data;
2964         stream.avail_in = size;
2965         while (git_deflate(&stream, Z_FINISH) == Z_OK)
2966                 ; /* nothing */
2967         git_deflate_end(&stream);
2968         *result_size = stream.total_out;
2969         return deflated;
2970 }
2971
2972 static void emit_binary_diff_body(struct diff_options *o,
2973                                   mmfile_t *one, mmfile_t *two)
2974 {
2975         void *cp;
2976         void *delta;
2977         void *deflated;
2978         void *data;
2979         unsigned long orig_size;
2980         unsigned long delta_size;
2981         unsigned long deflate_size;
2982         unsigned long data_size;
2983
2984         /* We could do deflated delta, or we could do just deflated two,
2985          * whichever is smaller.
2986          */
2987         delta = NULL;
2988         deflated = deflate_it(two->ptr, two->size, &deflate_size);
2989         if (one->size && two->size) {
2990                 delta = diff_delta(one->ptr, one->size,
2991                                    two->ptr, two->size,
2992                                    &delta_size, deflate_size);
2993                 if (delta) {
2994                         void *to_free = delta;
2995                         orig_size = delta_size;
2996                         delta = deflate_it(delta, delta_size, &delta_size);
2997                         free(to_free);
2998                 }
2999         }
3000
3001         if (delta && delta_size < deflate_size) {
3002                 char *s = xstrfmt("%lu", orig_size);
3003                 emit_diff_symbol(o, DIFF_SYMBOL_BINARY_DIFF_HEADER_DELTA,
3004                                  s, strlen(s), 0);
3005                 free(s);
3006                 free(deflated);
3007                 data = delta;
3008                 data_size = delta_size;
3009         } else {
3010                 char *s = xstrfmt("%lu", two->size);
3011                 emit_diff_symbol(o, DIFF_SYMBOL_BINARY_DIFF_HEADER_LITERAL,
3012                                  s, strlen(s), 0);
3013                 free(s);
3014                 free(delta);
3015                 data = deflated;
3016                 data_size = deflate_size;
3017         }
3018
3019         /* emit data encoded in base85 */
3020         cp = data;
3021         while (data_size) {
3022                 int len;
3023                 int bytes = (52 < data_size) ? 52 : data_size;
3024                 char line[71];
3025                 data_size -= bytes;
3026                 if (bytes <= 26)
3027                         line[0] = bytes + 'A' - 1;
3028                 else
3029                         line[0] = bytes - 26 + 'a' - 1;
3030                 encode_85(line + 1, cp, bytes);
3031                 cp = (char *) cp + bytes;
3032
3033                 len = strlen(line);
3034                 line[len++] = '\n';
3035                 line[len] = '\0';
3036
3037                 emit_diff_symbol(o, DIFF_SYMBOL_BINARY_DIFF_BODY,
3038                                  line, len, 0);
3039         }
3040         emit_diff_symbol(o, DIFF_SYMBOL_BINARY_DIFF_FOOTER, NULL, 0, 0);
3041         free(data);
3042 }
3043
3044 static void emit_binary_diff(struct diff_options *o,
3045                              mmfile_t *one, mmfile_t *two)
3046 {
3047         emit_diff_symbol(o, DIFF_SYMBOL_BINARY_DIFF_HEADER, NULL, 0, 0);
3048         emit_binary_diff_body(o, one, two);
3049         emit_binary_diff_body(o, two, one);
3050 }
3051
3052 int diff_filespec_is_binary(struct diff_filespec *one)
3053 {
3054         if (one->is_binary == -1) {
3055                 diff_filespec_load_driver(one);
3056                 if (one->driver->binary != -1)
3057                         one->is_binary = one->driver->binary;
3058                 else {
3059                         if (!one->data && DIFF_FILE_VALID(one))
3060                                 diff_populate_filespec(one, CHECK_BINARY);
3061                         if (one->is_binary == -1 && one->data)
3062                                 one->is_binary = buffer_is_binary(one->data,
3063                                                 one->size);
3064                         if (one->is_binary == -1)
3065                                 one->is_binary = 0;
3066                 }
3067         }
3068         return one->is_binary;
3069 }
3070
3071 static const struct userdiff_funcname *diff_funcname_pattern(struct diff_filespec *one)
3072 {
3073         diff_filespec_load_driver(one);
3074         return one->driver->funcname.pattern ? &one->driver->funcname : NULL;
3075 }
3076
3077 void diff_set_mnemonic_prefix(struct diff_options *options, const char *a, const char *b)
3078 {
3079         if (!options->a_prefix)
3080                 options->a_prefix = a;
3081         if (!options->b_prefix)
3082                 options->b_prefix = b;
3083 }
3084
3085 struct userdiff_driver *get_textconv(struct diff_filespec *one)
3086 {
3087         if (!DIFF_FILE_VALID(one))
3088                 return NULL;
3089
3090         diff_filespec_load_driver(one);
3091         return userdiff_get_textconv(one->driver);
3092 }
3093
3094 static void builtin_diff(const char *name_a,
3095                          const char *name_b,
3096                          struct diff_filespec *one,
3097                          struct diff_filespec *two,
3098                          const char *xfrm_msg,
3099                          int must_show_header,
3100                          struct diff_options *o,
3101                          int complete_rewrite)
3102 {
3103         mmfile_t mf1, mf2;
3104         const char *lbl[2];
3105         char *a_one, *b_two;
3106         const char *meta = diff_get_color_opt(o, DIFF_METAINFO);
3107         const char *reset = diff_get_color_opt(o, DIFF_RESET);
3108         const char *a_prefix, *b_prefix;
3109         struct userdiff_driver *textconv_one = NULL;
3110         struct userdiff_driver *textconv_two = NULL;
3111         struct strbuf header = STRBUF_INIT;
3112         const char *line_prefix = diff_line_prefix(o);
3113
3114         diff_set_mnemonic_prefix(o, "a/", "b/");
3115         if (DIFF_OPT_TST(o, REVERSE_DIFF)) {
3116                 a_prefix = o->b_prefix;
3117                 b_prefix = o->a_prefix;
3118         } else {
3119                 a_prefix = o->a_prefix;
3120                 b_prefix = o->b_prefix;
3121         }
3122
3123         if (o->submodule_format == DIFF_SUBMODULE_LOG &&
3124             (!one->mode || S_ISGITLINK(one->mode)) &&
3125             (!two->mode || S_ISGITLINK(two->mode))) {
3126                 show_submodule_summary(o, one->path ? one->path : two->path,
3127                                 &one->oid, &two->oid,
3128                                 two->dirty_submodule);
3129                 return;
3130         } else if (o->submodule_format == DIFF_SUBMODULE_INLINE_DIFF &&
3131                    (!one->mode || S_ISGITLINK(one->mode)) &&
3132                    (!two->mode || S_ISGITLINK(two->mode))) {
3133                 show_submodule_inline_diff(o, one->path ? one->path : two->path,
3134                                 &one->oid, &two->oid,
3135                                 two->dirty_submodule);
3136                 return;
3137         }
3138
3139         if (DIFF_OPT_TST(o, ALLOW_TEXTCONV)) {
3140                 textconv_one = get_textconv(one);
3141                 textconv_two = get_textconv(two);
3142         }
3143
3144         /* Never use a non-valid filename anywhere if at all possible */
3145         name_a = DIFF_FILE_VALID(one) ? name_a : name_b;
3146         name_b = DIFF_FILE_VALID(two) ? name_b : name_a;
3147
3148         a_one = quote_two(a_prefix, name_a + (*name_a == '/'));
3149         b_two = quote_two(b_prefix, name_b + (*name_b == '/'));
3150         lbl[0] = DIFF_FILE_VALID(one) ? a_one : "/dev/null";
3151         lbl[1] = DIFF_FILE_VALID(two) ? b_two : "/dev/null";
3152         strbuf_addf(&header, "%s%sdiff --git %s %s%s\n", line_prefix, meta, a_one, b_two, reset);
3153         if (lbl[0][0] == '/') {
3154                 /* /dev/null */
3155                 strbuf_addf(&header, "%s%snew file mode %06o%s\n", line_prefix, meta, two->mode, reset);
3156                 if (xfrm_msg)
3157                         strbuf_addstr(&header, xfrm_msg);
3158                 must_show_header = 1;
3159         }
3160         else if (lbl[1][0] == '/') {
3161                 strbuf_addf(&header, "%s%sdeleted file mode %06o%s\n", line_prefix, meta, one->mode, reset);
3162                 if (xfrm_msg)
3163                         strbuf_addstr(&header, xfrm_msg);
3164                 must_show_header = 1;
3165         }
3166         else {
3167                 if (one->mode != two->mode) {
3168                         strbuf_addf(&header, "%s%sold mode %06o%s\n", line_prefix, meta, one->mode, reset);
3169                         strbuf_addf(&header, "%s%snew mode %06o%s\n", line_prefix, meta, two->mode, reset);
3170                         must_show_header = 1;
3171                 }
3172                 if (xfrm_msg)
3173                         strbuf_addstr(&header, xfrm_msg);
3174
3175                 /*
3176                  * we do not run diff between different kind
3177                  * of objects.
3178                  */
3179                 if ((one->mode ^ two->mode) & S_IFMT)
3180                         goto free_ab_and_return;
3181                 if (complete_rewrite &&
3182                     (textconv_one || !diff_filespec_is_binary(one)) &&
3183                     (textconv_two || !diff_filespec_is_binary(two))) {
3184                         emit_diff_symbol(o, DIFF_SYMBOL_HEADER,
3185                                          header.buf, header.len, 0);
3186                         strbuf_reset(&header);
3187                         emit_rewrite_diff(name_a, name_b, one, two,
3188                                                 textconv_one, textconv_two, o);
3189                         o->found_changes = 1;
3190                         goto free_ab_and_return;
3191                 }
3192         }
3193
3194         if (o->irreversible_delete && lbl[1][0] == '/') {
3195                 emit_diff_symbol(o, DIFF_SYMBOL_HEADER, header.buf,
3196                                  header.len, 0);
3197                 strbuf_reset(&header);
3198                 goto free_ab_and_return;
3199         } else if (!DIFF_OPT_TST(o, TEXT) &&
3200             ( (!textconv_one && diff_filespec_is_binary(one)) ||
3201               (!textconv_two && diff_filespec_is_binary(two)) )) {
3202                 struct strbuf sb = STRBUF_INIT;
3203                 if (!one->data && !two->data &&
3204                     S_ISREG(one->mode) && S_ISREG(two->mode) &&
3205                     !DIFF_OPT_TST(o, BINARY)) {
3206                         if (!oidcmp(&one->oid, &two->oid)) {
3207                                 if (must_show_header)
3208                                         emit_diff_symbol(o, DIFF_SYMBOL_HEADER,
3209                                                          header.buf, header.len,
3210                                                          0);
3211                                 goto free_ab_and_return;
3212                         }
3213                         emit_diff_symbol(o, DIFF_SYMBOL_HEADER,
3214                                          header.buf, header.len, 0);
3215                         strbuf_addf(&sb, "%sBinary files %s and %s differ\n",
3216                                     diff_line_prefix(o), lbl[0], lbl[1]);
3217                         emit_diff_symbol(o, DIFF_SYMBOL_BINARY_FILES,
3218                                          sb.buf, sb.len, 0);
3219                         strbuf_release(&sb);
3220                         goto free_ab_and_return;
3221                 }
3222                 if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
3223                         die("unable to read files to diff");
3224                 /* Quite common confusing case */
3225                 if (mf1.size == mf2.size &&
3226                     !memcmp(mf1.ptr, mf2.ptr, mf1.size)) {
3227                         if (must_show_header)
3228                                 emit_diff_symbol(o, DIFF_SYMBOL_HEADER,
3229                                                  header.buf, header.len, 0);
3230                         goto free_ab_and_return;
3231                 }
3232                 emit_diff_symbol(o, DIFF_SYMBOL_HEADER, header.buf, header.len, 0);
3233                 strbuf_reset(&header);
3234                 if (DIFF_OPT_TST(o, BINARY))
3235                         emit_binary_diff(o, &mf1, &mf2);
3236                 else {
3237                         strbuf_addf(&sb, "%sBinary files %s and %s differ\n",
3238                                     diff_line_prefix(o), lbl[0], lbl[1]);
3239                         emit_diff_symbol(o, DIFF_SYMBOL_BINARY_FILES,
3240                                          sb.buf, sb.len, 0);
3241                         strbuf_release(&sb);
3242                 }
3243                 o->found_changes = 1;
3244         } else {
3245                 /* Crazy xdl interfaces.. */
3246                 const char *diffopts = getenv("GIT_DIFF_OPTS");
3247                 const char *v;
3248                 xpparam_t xpp;
3249                 xdemitconf_t xecfg;
3250                 struct emit_callback ecbdata;
3251                 const struct userdiff_funcname *pe;
3252
3253                 if (must_show_header) {
3254                         emit_diff_symbol(o, DIFF_SYMBOL_HEADER,
3255                                          header.buf, header.len, 0);
3256                         strbuf_reset(&header);
3257                 }
3258
3259                 mf1.size = fill_textconv(textconv_one, one, &mf1.ptr);
3260                 mf2.size = fill_textconv(textconv_two, two, &mf2.ptr);
3261
3262                 pe = diff_funcname_pattern(one);
3263                 if (!pe)
3264                         pe = diff_funcname_pattern(two);
3265
3266                 memset(&xpp, 0, sizeof(xpp));
3267                 memset(&xecfg, 0, sizeof(xecfg));
3268                 memset(&ecbdata, 0, sizeof(ecbdata));
3269                 ecbdata.label_path = lbl;
3270                 ecbdata.color_diff = want_color(o->use_color);
3271                 ecbdata.ws_rule = whitespace_rule(name_b);
3272                 if (ecbdata.ws_rule & WS_BLANK_AT_EOF)
3273                         check_blank_at_eof(&mf1, &mf2, &ecbdata);
3274                 ecbdata.opt = o;
3275                 ecbdata.header = header.len ? &header : NULL;
3276                 xpp.flags = o->xdl_opts;
3277                 xecfg.ctxlen = o->context;
3278                 xecfg.interhunkctxlen = o->interhunkcontext;
3279                 xecfg.flags = XDL_EMIT_FUNCNAMES;
3280                 if (DIFF_OPT_TST(o, FUNCCONTEXT))
3281                         xecfg.flags |= XDL_EMIT_FUNCCONTEXT;
3282                 if (pe)
3283                         xdiff_set_find_func(&xecfg, pe->pattern, pe->cflags);
3284                 if (!diffopts)
3285                         ;
3286                 else if (skip_prefix(diffopts, "--unified=", &v))
3287                         xecfg.ctxlen = strtoul(v, NULL, 10);
3288                 else if (skip_prefix(diffopts, "-u", &v))
3289                         xecfg.ctxlen = strtoul(v, NULL, 10);
3290                 if (o->word_diff)
3291                         init_diff_words_data(&ecbdata, o, one, two);
3292                 if (xdi_diff_outf(&mf1, &mf2, fn_out_consume, &ecbdata,
3293                                   &xpp, &xecfg))
3294                         die("unable to generate diff for %s", one->path);
3295                 if (o->word_diff)
3296                         free_diff_words_data(&ecbdata);
3297                 if (textconv_one)
3298                         free(mf1.ptr);
3299                 if (textconv_two)
3300                         free(mf2.ptr);
3301                 xdiff_clear_find_func(&xecfg);
3302         }
3303
3304  free_ab_and_return:
3305         strbuf_release(&header);
3306         diff_free_filespec_data(one);
3307         diff_free_filespec_data(two);
3308         free(a_one);
3309         free(b_two);
3310         return;
3311 }
3312
3313 static void builtin_diffstat(const char *name_a, const char *name_b,
3314                              struct diff_filespec *one,
3315                              struct diff_filespec *two,
3316                              struct diffstat_t *diffstat,
3317                              struct diff_options *o,
3318                              struct diff_filepair *p)
3319 {
3320         mmfile_t mf1, mf2;
3321         struct diffstat_file *data;
3322         int same_contents;
3323         int complete_rewrite = 0;
3324
3325         if (!DIFF_PAIR_UNMERGED(p)) {
3326                 if (p->status == DIFF_STATUS_MODIFIED && p->score)
3327                         complete_rewrite = 1;
3328         }
3329
3330         data = diffstat_add(diffstat, name_a, name_b);
3331         data->is_interesting = p->status != DIFF_STATUS_UNKNOWN;
3332
3333         if (!one || !two) {
3334                 data->is_unmerged = 1;
3335                 return;
3336         }
3337
3338         same_contents = !oidcmp(&one->oid, &two->oid);
3339
3340         if (diff_filespec_is_binary(one) || diff_filespec_is_binary(two)) {
3341                 data->is_binary = 1;
3342                 if (same_contents) {
3343                         data->added = 0;
3344                         data->deleted = 0;
3345                 } else {
3346                         data->added = diff_filespec_size(two);
3347                         data->deleted = diff_filespec_size(one);
3348                 }
3349         }
3350
3351         else if (complete_rewrite) {
3352                 diff_populate_filespec(one, 0);
3353                 diff_populate_filespec(two, 0);
3354                 data->deleted = count_lines(one->data, one->size);
3355                 data->added = count_lines(two->data, two->size);
3356         }
3357
3358         else if (!same_contents) {
3359                 /* Crazy xdl interfaces.. */
3360                 xpparam_t xpp;
3361                 xdemitconf_t xecfg;
3362
3363                 if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
3364                         die("unable to read files to diff");
3365
3366                 memset(&xpp, 0, sizeof(xpp));
3367                 memset(&xecfg, 0, sizeof(xecfg));
3368                 xpp.flags = o->xdl_opts;
3369                 xecfg.ctxlen = o->context;
3370                 xecfg.interhunkctxlen = o->interhunkcontext;
3371                 if (xdi_diff_outf(&mf1, &mf2, diffstat_consume, diffstat,
3372                                   &xpp, &xecfg))
3373                         die("unable to generate diffstat for %s", one->path);
3374         }
3375
3376         diff_free_filespec_data(one);
3377         diff_free_filespec_data(two);
3378 }
3379
3380 static void builtin_checkdiff(const char *name_a, const char *name_b,
3381                               const char *attr_path,
3382                               struct diff_filespec *one,
3383                               struct diff_filespec *two,
3384                               struct diff_options *o)
3385 {
3386         mmfile_t mf1, mf2;
3387         struct checkdiff_t data;
3388
3389         if (!two)
3390                 return;
3391
3392         memset(&data, 0, sizeof(data));
3393         data.filename = name_b ? name_b : name_a;
3394         data.lineno = 0;
3395         data.o = o;
3396         data.ws_rule = whitespace_rule(attr_path);
3397         data.conflict_marker_size = ll_merge_marker_size(attr_path);
3398
3399         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
3400                 die("unable to read files to diff");
3401
3402         /*
3403          * All the other codepaths check both sides, but not checking
3404          * the "old" side here is deliberate.  We are checking the newly
3405          * introduced changes, and as long as the "new" side is text, we
3406          * can and should check what it introduces.
3407          */
3408         if (diff_filespec_is_binary(two))
3409                 goto free_and_return;
3410         else {
3411                 /* Crazy xdl interfaces.. */
3412                 xpparam_t xpp;
3413                 xdemitconf_t xecfg;
3414
3415                 memset(&xpp, 0, sizeof(xpp));
3416                 memset(&xecfg, 0, sizeof(xecfg));
3417                 xecfg.ctxlen = 1; /* at least one context line */
3418                 xpp.flags = 0;
3419                 if (xdi_diff_outf(&mf1, &mf2, checkdiff_consume, &data,
3420                                   &xpp, &xecfg))
3421                         die("unable to generate checkdiff for %s", one->path);
3422
3423                 if (data.ws_rule & WS_BLANK_AT_EOF) {
3424                         struct emit_callback ecbdata;
3425                         int blank_at_eof;
3426
3427                         ecbdata.ws_rule = data.ws_rule;
3428                         check_blank_at_eof(&mf1, &mf2, &ecbdata);
3429                         blank_at_eof = ecbdata.blank_at_eof_in_postimage;
3430
3431                         if (blank_at_eof) {
3432                                 static char *err;
3433                                 if (!err)
3434                                         err = whitespace_error_string(WS_BLANK_AT_EOF);
3435                                 fprintf(o->file, "%s:%d: %s.\n",
3436                                         data.filename, blank_at_eof, err);
3437                                 data.status = 1; /* report errors */
3438                         }
3439                 }
3440         }
3441  free_and_return:
3442         diff_free_filespec_data(one);
3443         diff_free_filespec_data(two);
3444         if (data.status)
3445                 DIFF_OPT_SET(o, CHECK_FAILED);
3446 }
3447
3448 struct diff_filespec *alloc_filespec(const char *path)
3449 {
3450         struct diff_filespec *spec;
3451
3452         FLEXPTR_ALLOC_STR(spec, path, path);
3453         spec->count = 1;
3454         spec->is_binary = -1;
3455         return spec;
3456 }
3457
3458 void free_filespec(struct diff_filespec *spec)
3459 {
3460         if (!--spec->count) {
3461                 diff_free_filespec_data(spec);
3462                 free(spec);
3463         }
3464 }
3465
3466 void fill_filespec(struct diff_filespec *spec, const struct object_id *oid,
3467                    int oid_valid, unsigned short mode)
3468 {
3469         if (mode) {
3470                 spec->mode = canon_mode(mode);
3471                 oidcpy(&spec->oid, oid);
3472                 spec->oid_valid = oid_valid;
3473         }
3474 }
3475
3476 /*
3477  * Given a name and sha1 pair, if the index tells us the file in
3478  * the work tree has that object contents, return true, so that
3479  * prepare_temp_file() does not have to inflate and extract.
3480  */
3481 static int reuse_worktree_file(const char *name, const struct object_id *oid, int want_file)
3482 {
3483         const struct cache_entry *ce;
3484         struct stat st;
3485         int pos, len;
3486
3487         /*
3488          * We do not read the cache ourselves here, because the
3489          * benchmark with my previous version that always reads cache
3490          * shows that it makes things worse for diff-tree comparing
3491          * two linux-2.6 kernel trees in an already checked out work
3492          * tree.  This is because most diff-tree comparisons deal with
3493          * only a small number of files, while reading the cache is
3494          * expensive for a large project, and its cost outweighs the
3495          * savings we get by not inflating the object to a temporary
3496          * file.  Practically, this code only helps when we are used
3497          * by diff-cache --cached, which does read the cache before
3498          * calling us.
3499          */
3500         if (!active_cache)
3501                 return 0;
3502
3503         /* We want to avoid the working directory if our caller
3504          * doesn't need the data in a normal file, this system
3505          * is rather slow with its stat/open/mmap/close syscalls,
3506          * and the object is contained in a pack file.  The pack
3507          * is probably already open and will be faster to obtain
3508          * the data through than the working directory.  Loose
3509          * objects however would tend to be slower as they need
3510          * to be individually opened and inflated.
3511          */
3512         if (!FAST_WORKING_DIRECTORY && !want_file && has_sha1_pack(oid->hash))
3513                 return 0;
3514
3515         /*
3516          * Similarly, if we'd have to convert the file contents anyway, that
3517          * makes the optimization not worthwhile.
3518          */
3519         if (!want_file && would_convert_to_git(&the_index, name))
3520                 return 0;
3521
3522         len = strlen(name);
3523         pos = cache_name_pos(name, len);
3524         if (pos < 0)
3525                 return 0;
3526         ce = active_cache[pos];
3527
3528         /*
3529          * This is not the sha1 we are looking for, or
3530          * unreusable because it is not a regular file.
3531          */
3532         if (oidcmp(oid, &ce->oid) || !S_ISREG(ce->ce_mode))
3533                 return 0;
3534
3535         /*
3536          * If ce is marked as "assume unchanged", there is no
3537          * guarantee that work tree matches what we are looking for.
3538          */
3539         if ((ce->ce_flags & CE_VALID) || ce_skip_worktree(ce))
3540                 return 0;
3541
3542         /*
3543          * If ce matches the file in the work tree, we can reuse it.
3544          */
3545         if (ce_uptodate(ce) ||
3546             (!lstat(name, &st) && !ce_match_stat(ce, &st, 0)))
3547                 return 1;
3548
3549         return 0;
3550 }
3551
3552 static int diff_populate_gitlink(struct diff_filespec *s, int size_only)
3553 {
3554         struct strbuf buf = STRBUF_INIT;
3555         char *dirty = "";
3556
3557         /* Are we looking at the work tree? */
3558         if (s->dirty_submodule)
3559                 dirty = "-dirty";
3560
3561         strbuf_addf(&buf, "Subproject commit %s%s\n",
3562                     oid_to_hex(&s->oid), dirty);
3563         s->size = buf.len;
3564         if (size_only) {
3565                 s->data = NULL;
3566                 strbuf_release(&buf);
3567         } else {
3568                 s->data = strbuf_detach(&buf, NULL);
3569                 s->should_free = 1;
3570         }
3571         return 0;
3572 }
3573
3574 /*
3575  * While doing rename detection and pickaxe operation, we may need to
3576  * grab the data for the blob (or file) for our own in-core comparison.
3577  * diff_filespec has data and size fields for this purpose.
3578  */
3579 int diff_populate_filespec(struct diff_filespec *s, unsigned int flags)
3580 {
3581         int size_only = flags & CHECK_SIZE_ONLY;
3582         int err = 0;
3583         /*
3584          * demote FAIL to WARN to allow inspecting the situation
3585          * instead of refusing.
3586          */
3587         enum safe_crlf crlf_warn = (safe_crlf == SAFE_CRLF_FAIL
3588                                     ? SAFE_CRLF_WARN
3589                                     : safe_crlf);
3590
3591         if (!DIFF_FILE_VALID(s))
3592                 die("internal error: asking to populate invalid file.");
3593         if (S_ISDIR(s->mode))
3594                 return -1;
3595
3596         if (s->data)
3597                 return 0;
3598
3599         if (size_only && 0 < s->size)
3600                 return 0;
3601
3602         if (S_ISGITLINK(s->mode))
3603                 return diff_populate_gitlink(s, size_only);
3604
3605         if (!s->oid_valid ||
3606             reuse_worktree_file(s->path, &s->oid, 0)) {
3607                 struct strbuf buf = STRBUF_INIT;
3608                 struct stat st;
3609                 int fd;
3610
3611                 if (lstat(s->path, &st) < 0) {
3612                         if (errno == ENOENT) {
3613                         err_empty:
3614                                 err = -1;
3615                         empty:
3616                                 s->data = (char *)"";
3617                                 s->size = 0;
3618                                 return err;
3619                         }
3620                 }
3621                 s->size = xsize_t(st.st_size);
3622                 if (!s->size)
3623                         goto empty;
3624                 if (S_ISLNK(st.st_mode)) {
3625                         struct strbuf sb = STRBUF_INIT;
3626
3627                         if (strbuf_readlink(&sb, s->path, s->size))
3628                                 goto err_empty;
3629                         s->size = sb.len;
3630                         s->data = strbuf_detach(&sb, NULL);
3631                         s->should_free = 1;
3632                         return 0;
3633                 }
3634
3635                 /*
3636                  * Even if the caller would be happy with getting
3637                  * only the size, we cannot return early at this
3638                  * point if the path requires us to run the content
3639                  * conversion.
3640                  */
3641                 if (size_only && !would_convert_to_git(&the_index, s->path))
3642                         return 0;
3643
3644                 /*
3645                  * Note: this check uses xsize_t(st.st_size) that may
3646                  * not be the true size of the blob after it goes
3647                  * through convert_to_git().  This may not strictly be
3648                  * correct, but the whole point of big_file_threshold
3649                  * and is_binary check being that we want to avoid
3650                  * opening the file and inspecting the contents, this
3651                  * is probably fine.
3652                  */
3653                 if ((flags & CHECK_BINARY) &&
3654                     s->size > big_file_threshold && s->is_binary == -1) {
3655                         s->is_binary = 1;
3656                         return 0;
3657                 }
3658                 fd = open(s->path, O_RDONLY);
3659                 if (fd < 0)
3660                         goto err_empty;
3661                 s->data = xmmap(NULL, s->size, PROT_READ, MAP_PRIVATE, fd, 0);
3662                 close(fd);
3663                 s->should_munmap = 1;
3664
3665                 /*
3666                  * Convert from working tree format to canonical git format
3667                  */
3668                 if (convert_to_git(&the_index, s->path, s->data, s->size, &buf, crlf_warn)) {
3669                         size_t size = 0;
3670                         munmap(s->data, s->size);
3671                         s->should_munmap = 0;
3672                         s->data = strbuf_detach(&buf, &size);
3673                         s->size = size;
3674                         s->should_free = 1;
3675                 }
3676         }
3677         else {
3678                 enum object_type type;
3679                 if (size_only || (flags & CHECK_BINARY)) {
3680                         type = sha1_object_info(s->oid.hash, &s->size);
3681                         if (type < 0)
3682                                 die("unable to read %s",
3683                                     oid_to_hex(&s->oid));
3684                         if (size_only)
3685                                 return 0;
3686                         if (s->size > big_file_threshold && s->is_binary == -1) {
3687                                 s->is_binary = 1;
3688                                 return 0;
3689                         }
3690                 }
3691                 s->data = read_sha1_file(s->oid.hash, &type, &s->size);
3692                 if (!s->data)
3693                         die("unable to read %s", oid_to_hex(&s->oid));
3694                 s->should_free = 1;
3695         }
3696         return 0;
3697 }
3698
3699 void diff_free_filespec_blob(struct diff_filespec *s)
3700 {
3701         if (s->should_free)
3702                 free(s->data);
3703         else if (s->should_munmap)
3704                 munmap(s->data, s->size);
3705
3706         if (s->should_free || s->should_munmap) {
3707                 s->should_free = s->should_munmap = 0;
3708                 s->data = NULL;
3709         }
3710 }
3711
3712 void diff_free_filespec_data(struct diff_filespec *s)
3713 {
3714         diff_free_filespec_blob(s);
3715         FREE_AND_NULL(s->cnt_data);
3716 }
3717
3718 static void prep_temp_blob(const char *path, struct diff_tempfile *temp,
3719                            void *blob,
3720                            unsigned long size,
3721                            const struct object_id *oid,
3722                            int mode)
3723 {
3724         struct strbuf buf = STRBUF_INIT;
3725         struct strbuf template = STRBUF_INIT;
3726         char *path_dup = xstrdup(path);
3727         const char *base = basename(path_dup);
3728
3729         /* Generate "XXXXXX_basename.ext" */
3730         strbuf_addstr(&template, "XXXXXX_");
3731         strbuf_addstr(&template, base);
3732
3733         temp->tempfile = mks_tempfile_ts(template.buf, strlen(base) + 1);
3734         if (!temp->tempfile)
3735                 die_errno("unable to create temp-file");
3736         if (convert_to_working_tree(path,
3737                         (const char *)blob, (size_t)size, &buf)) {
3738                 blob = buf.buf;
3739                 size = buf.len;
3740         }
3741         if (write_in_full(temp->tempfile->fd, blob, size) < 0 ||
3742             close_tempfile_gently(temp->tempfile))
3743                 die_errno("unable to write temp-file");
3744         temp->name = get_tempfile_path(temp->tempfile);
3745         oid_to_hex_r(temp->hex, oid);
3746         xsnprintf(temp->mode, sizeof(temp->mode), "%06o", mode);
3747         strbuf_release(&buf);
3748         strbuf_release(&template);
3749         free(path_dup);
3750 }
3751
3752 static struct diff_tempfile *prepare_temp_file(const char *name,
3753                 struct diff_filespec *one)
3754 {
3755         struct diff_tempfile *temp = claim_diff_tempfile();
3756
3757         if (!DIFF_FILE_VALID(one)) {
3758         not_a_valid_file:
3759                 /* A '-' entry produces this for file-2, and
3760                  * a '+' entry produces this for file-1.
3761                  */
3762                 temp->name = "/dev/null";
3763                 xsnprintf(temp->hex, sizeof(temp->hex), ".");
3764                 xsnprintf(temp->mode, sizeof(temp->mode), ".");
3765                 return temp;
3766         }
3767
3768         if (!S_ISGITLINK(one->mode) &&
3769             (!one->oid_valid ||
3770              reuse_worktree_file(name, &one->oid, 1))) {
3771                 struct stat st;
3772                 if (lstat(name, &st) < 0) {
3773                         if (errno == ENOENT)
3774                                 goto not_a_valid_file;
3775                         die_errno("stat(%s)", name);
3776                 }
3777                 if (S_ISLNK(st.st_mode)) {
3778                         struct strbuf sb = STRBUF_INIT;
3779                         if (strbuf_readlink(&sb, name, st.st_size) < 0)
3780                                 die_errno("readlink(%s)", name);
3781                         prep_temp_blob(name, temp, sb.buf, sb.len,
3782                                        (one->oid_valid ?
3783                                         &one->oid : &null_oid),
3784                                        (one->oid_valid ?
3785                                         one->mode : S_IFLNK));
3786                         strbuf_release(&sb);
3787                 }
3788                 else {
3789                         /* we can borrow from the file in the work tree */
3790                         temp->name = name;
3791                         if (!one->oid_valid)
3792                                 oid_to_hex_r(temp->hex, &null_oid);
3793                         else
3794                                 oid_to_hex_r(temp->hex, &one->oid);
3795                         /* Even though we may sometimes borrow the
3796                          * contents from the work tree, we always want
3797                          * one->mode.  mode is trustworthy even when
3798                          * !(one->oid_valid), as long as
3799                          * DIFF_FILE_VALID(one).
3800                          */
3801                         xsnprintf(temp->mode, sizeof(temp->mode), "%06o", one->mode);
3802                 }
3803                 return temp;
3804         }
3805         else {
3806                 if (diff_populate_filespec(one, 0))
3807                         die("cannot read data blob for %s", one->path);
3808                 prep_temp_blob(name, temp, one->data, one->size,
3809                                &one->oid, one->mode);
3810         }
3811         return temp;
3812 }
3813
3814 static void add_external_diff_name(struct argv_array *argv,
3815                                    const char *name,
3816                                    struct diff_filespec *df)
3817 {
3818         struct diff_tempfile *temp = prepare_temp_file(name, df);
3819         argv_array_push(argv, temp->name);
3820         argv_array_push(argv, temp->hex);
3821         argv_array_push(argv, temp->mode);
3822 }
3823
3824 /* An external diff command takes:
3825  *
3826  * diff-cmd name infile1 infile1-sha1 infile1-mode \
3827  *               infile2 infile2-sha1 infile2-mode [ rename-to ]
3828  *
3829  */
3830 static void run_external_diff(const char *pgm,
3831                               const char *name,
3832                               const char *other,
3833                               struct diff_filespec *one,
3834                               struct diff_filespec *two,
3835                               const char *xfrm_msg,
3836                               int complete_rewrite,
3837                               struct diff_options *o)
3838 {
3839         struct argv_array argv = ARGV_ARRAY_INIT;
3840         struct argv_array env = ARGV_ARRAY_INIT;
3841         struct diff_queue_struct *q = &diff_queued_diff;
3842
3843         argv_array_push(&argv, pgm);
3844         argv_array_push(&argv, name);
3845
3846         if (one && two) {
3847                 add_external_diff_name(&argv, name, one);
3848                 if (!other)
3849                         add_external_diff_name(&argv, name, two);
3850                 else {
3851                         add_external_diff_name(&argv, other, two);
3852                         argv_array_push(&argv, other);
3853                         argv_array_push(&argv, xfrm_msg);
3854                 }
3855         }
3856
3857         argv_array_pushf(&env, "GIT_DIFF_PATH_COUNTER=%d", ++o->diff_path_counter);
3858         argv_array_pushf(&env, "GIT_DIFF_PATH_TOTAL=%d", q->nr);
3859
3860         if (run_command_v_opt_cd_env(argv.argv, RUN_USING_SHELL, NULL, env.argv))
3861                 die(_("external diff died, stopping at %s"), name);
3862
3863         remove_tempfile();
3864         argv_array_clear(&argv);
3865         argv_array_clear(&env);
3866 }
3867
3868 static int similarity_index(struct diff_filepair *p)
3869 {
3870         return p->score * 100 / MAX_SCORE;
3871 }
3872
3873 static const char *diff_abbrev_oid(const struct object_id *oid, int abbrev)
3874 {
3875         if (startup_info->have_repository)
3876                 return find_unique_abbrev(oid->hash, abbrev);
3877         else {
3878                 char *hex = oid_to_hex(oid);
3879                 if (abbrev < 0)
3880                         abbrev = FALLBACK_DEFAULT_ABBREV;
3881                 if (abbrev > GIT_SHA1_HEXSZ)
3882                         die("BUG: oid abbreviation out of range: %d", abbrev);
3883                 if (abbrev)
3884                         hex[abbrev] = '\0';
3885                 return hex;
3886         }
3887 }
3888
3889 static void fill_metainfo(struct strbuf *msg,
3890                           const char *name,
3891                           const char *other,
3892                           struct diff_filespec *one,
3893                           struct diff_filespec *two,
3894                           struct diff_options *o,
3895                           struct diff_filepair *p,
3896                           int *must_show_header,
3897                           int use_color)
3898 {
3899         const char *set = diff_get_color(use_color, DIFF_METAINFO);
3900         const char *reset = diff_get_color(use_color, DIFF_RESET);
3901         const char *line_prefix = diff_line_prefix(o);
3902
3903         *must_show_header = 1;
3904         strbuf_init(msg, PATH_MAX * 2 + 300);
3905         switch (p->status) {
3906         case DIFF_STATUS_COPIED:
3907                 strbuf_addf(msg, "%s%ssimilarity index %d%%",
3908                             line_prefix, set, similarity_index(p));
3909                 strbuf_addf(msg, "%s\n%s%scopy from ",
3910                             reset,  line_prefix, set);
3911                 quote_c_style(name, msg, NULL, 0);
3912                 strbuf_addf(msg, "%s\n%s%scopy to ", reset, line_prefix, set);
3913                 quote_c_style(other, msg, NULL, 0);
3914                 strbuf_addf(msg, "%s\n", reset);
3915                 break;
3916         case DIFF_STATUS_RENAMED:
3917                 strbuf_addf(msg, "%s%ssimilarity index %d%%",
3918                             line_prefix, set, similarity_index(p));
3919                 strbuf_addf(msg, "%s\n%s%srename from ",
3920                             reset, line_prefix, set);
3921                 quote_c_style(name, msg, NULL, 0);
3922                 strbuf_addf(msg, "%s\n%s%srename to ",
3923                             reset, line_prefix, set);
3924                 quote_c_style(other, msg, NULL, 0);
3925                 strbuf_addf(msg, "%s\n", reset);
3926                 break;
3927         case DIFF_STATUS_MODIFIED:
3928                 if (p->score) {
3929                         strbuf_addf(msg, "%s%sdissimilarity index %d%%%s\n",
3930                                     line_prefix,
3931                                     set, similarity_index(p), reset);
3932                         break;
3933                 }
3934                 /* fallthru */
3935         default:
3936                 *must_show_header = 0;
3937         }
3938         if (one && two && oidcmp(&one->oid, &two->oid)) {
3939                 int abbrev = DIFF_OPT_TST(o, FULL_INDEX) ? 40 : DEFAULT_ABBREV;
3940
3941                 if (DIFF_OPT_TST(o, BINARY)) {
3942                         mmfile_t mf;
3943                         if ((!fill_mmfile(&mf, one) && diff_filespec_is_binary(one)) ||
3944                             (!fill_mmfile(&mf, two) && diff_filespec_is_binary(two)))
3945                                 abbrev = 40;
3946                 }
3947                 strbuf_addf(msg, "%s%sindex %s..%s", line_prefix, set,
3948                             diff_abbrev_oid(&one->oid, abbrev),
3949                             diff_abbrev_oid(&two->oid, abbrev));
3950                 if (one->mode == two->mode)
3951                         strbuf_addf(msg, " %06o", one->mode);
3952                 strbuf_addf(msg, "%s\n", reset);
3953         }
3954 }
3955
3956 static void run_diff_cmd(const char *pgm,
3957                          const char *name,
3958                          const char *other,
3959                          const char *attr_path,
3960                          struct diff_filespec *one,
3961                          struct diff_filespec *two,
3962                          struct strbuf *msg,
3963                          struct diff_options *o,
3964                          struct diff_filepair *p)
3965 {
3966         const char *xfrm_msg = NULL;
3967         int complete_rewrite = (p->status == DIFF_STATUS_MODIFIED) && p->score;
3968         int must_show_header = 0;
3969
3970
3971         if (DIFF_OPT_TST(o, ALLOW_EXTERNAL)) {
3972                 struct userdiff_driver *drv = userdiff_find_by_path(attr_path);
3973                 if (drv && drv->external)
3974                         pgm = drv->external;
3975         }
3976
3977         if (msg) {
3978                 /*
3979                  * don't use colors when the header is intended for an
3980                  * external diff driver
3981                  */
3982                 fill_metainfo(msg, name, other, one, two, o, p,
3983                               &must_show_header,
3984                               want_color(o->use_color) && !pgm);
3985                 xfrm_msg = msg->len ? msg->buf : NULL;
3986         }
3987
3988         if (pgm) {
3989                 run_external_diff(pgm, name, other, one, two, xfrm_msg,
3990                                   complete_rewrite, o);
3991                 return;
3992         }
3993         if (one && two)
3994                 builtin_diff(name, other ? other : name,
3995                              one, two, xfrm_msg, must_show_header,
3996                              o, complete_rewrite);
3997         else
3998                 fprintf(o->file, "* Unmerged path %s\n", name);
3999 }
4000
4001 static void diff_fill_oid_info(struct diff_filespec *one)
4002 {
4003         if (DIFF_FILE_VALID(one)) {
4004                 if (!one->oid_valid) {
4005                         struct stat st;
4006                         if (one->is_stdin) {
4007                                 oidclr(&one->oid);
4008                                 return;
4009                         }
4010                         if (lstat(one->path, &st) < 0)
4011                                 die_errno("stat '%s'", one->path);
4012                         if (index_path(&one->oid, one->path, &st, 0))
4013                                 die("cannot hash %s", one->path);
4014                 }
4015         }
4016         else
4017                 oidclr(&one->oid);
4018 }
4019
4020 static void strip_prefix(int prefix_length, const char **namep, const char **otherp)
4021 {
4022         /* Strip the prefix but do not molest /dev/null and absolute paths */
4023         if (*namep && **namep != '/') {
4024                 *namep += prefix_length;
4025                 if (**namep == '/')
4026                         ++*namep;
4027         }
4028         if (*otherp && **otherp != '/') {
4029                 *otherp += prefix_length;
4030                 if (**otherp == '/')
4031                         ++*otherp;
4032         }
4033 }
4034
4035 static void run_diff(struct diff_filepair *p, struct diff_options *o)
4036 {
4037         const char *pgm = external_diff();
4038         struct strbuf msg;
4039         struct diff_filespec *one = p->one;
4040         struct diff_filespec *two = p->two;
4041         const char *name;
4042         const char *other;
4043         const char *attr_path;
4044
4045         name  = one->path;
4046         other = (strcmp(name, two->path) ? two->path : NULL);
4047         attr_path = name;
4048         if (o->prefix_length)
4049                 strip_prefix(o->prefix_length, &name, &other);
4050
4051         if (!DIFF_OPT_TST(o, ALLOW_EXTERNAL))
4052                 pgm = NULL;
4053
4054         if (DIFF_PAIR_UNMERGED(p)) {
4055                 run_diff_cmd(pgm, name, NULL, attr_path,
4056                              NULL, NULL, NULL, o, p);
4057                 return;
4058         }
4059
4060         diff_fill_oid_info(one);
4061         diff_fill_oid_info(two);
4062
4063         if (!pgm &&
4064             DIFF_FILE_VALID(one) && DIFF_FILE_VALID(two) &&
4065             (S_IFMT & one->mode) != (S_IFMT & two->mode)) {
4066                 /*
4067                  * a filepair that changes between file and symlink
4068                  * needs to be split into deletion and creation.
4069                  */
4070                 struct diff_filespec *null = alloc_filespec(two->path);
4071                 run_diff_cmd(NULL, name, other, attr_path,
4072                              one, null, &msg, o, p);
4073                 free(null);
4074                 strbuf_release(&msg);
4075
4076                 null = alloc_filespec(one->path);
4077                 run_diff_cmd(NULL, name, other, attr_path,
4078                              null, two, &msg, o, p);
4079                 free(null);
4080         }
4081         else
4082                 run_diff_cmd(pgm, name, other, attr_path,
4083                              one, two, &msg, o, p);
4084
4085         strbuf_release(&msg);
4086 }
4087
4088 static void run_diffstat(struct diff_filepair *p, struct diff_options *o,
4089                          struct diffstat_t *diffstat)
4090 {
4091         const char *name;
4092         const char *other;
4093
4094         if (DIFF_PAIR_UNMERGED(p)) {
4095                 /* unmerged */
4096                 builtin_diffstat(p->one->path, NULL, NULL, NULL, diffstat, o, p);
4097                 return;
4098         }
4099
4100         name = p->one->path;
4101         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
4102
4103         if (o->prefix_length)
4104                 strip_prefix(o->prefix_length, &name, &other);
4105
4106         diff_fill_oid_info(p->one);
4107         diff_fill_oid_info(p->two);
4108
4109         builtin_diffstat(name, other, p->one, p->two, diffstat, o, p);
4110 }
4111
4112 static void run_checkdiff(struct diff_filepair *p, struct diff_options *o)
4113 {
4114         const char *name;
4115         const char *other;
4116         const char *attr_path;
4117
4118         if (DIFF_PAIR_UNMERGED(p)) {
4119                 /* unmerged */
4120                 return;
4121         }
4122
4123         name = p->one->path;
4124         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
4125         attr_path = other ? other : name;
4126
4127         if (o->prefix_length)
4128                 strip_prefix(o->prefix_length, &name, &other);
4129
4130         diff_fill_oid_info(p->one);
4131         diff_fill_oid_info(p->two);
4132
4133         builtin_checkdiff(name, other, attr_path, p->one, p->two, o);
4134 }
4135
4136 void diff_setup(struct diff_options *options)
4137 {
4138         memcpy(options, &default_diff_options, sizeof(*options));
4139
4140         options->file = stdout;
4141
4142         options->abbrev = DEFAULT_ABBREV;
4143         options->line_termination = '\n';
4144         options->break_opt = -1;
4145         options->rename_limit = -1;
4146         options->dirstat_permille = diff_dirstat_permille_default;
4147         options->context = diff_context_default;
4148         options->interhunkcontext = diff_interhunk_context_default;
4149         options->ws_error_highlight = ws_error_highlight_default;
4150         DIFF_OPT_SET(options, RENAME_EMPTY);
4151
4152         /* pathchange left =NULL by default */
4153         options->change = diff_change;
4154         options->add_remove = diff_addremove;
4155         options->use_color = diff_use_color_default;
4156         options->detect_rename = diff_detect_rename_default;
4157         options->xdl_opts |= diff_algorithm;
4158         if (diff_indent_heuristic)
4159                 DIFF_XDL_SET(options, INDENT_HEURISTIC);
4160
4161         options->orderfile = diff_order_file_cfg;
4162
4163         if (diff_no_prefix) {
4164                 options->a_prefix = options->b_prefix = "";
4165         } else if (!diff_mnemonic_prefix) {
4166                 options->a_prefix = "a/";
4167                 options->b_prefix = "b/";
4168         }
4169
4170         options->color_moved = diff_color_moved_default;
4171 }
4172
4173 void diff_setup_done(struct diff_options *options)
4174 {
4175         int count = 0;
4176
4177         if (options->set_default)
4178                 options->set_default(options);
4179
4180         if (options->output_format & DIFF_FORMAT_NAME)
4181                 count++;
4182         if (options->output_format & DIFF_FORMAT_NAME_STATUS)
4183                 count++;
4184         if (options->output_format & DIFF_FORMAT_CHECKDIFF)
4185                 count++;
4186         if (options->output_format & DIFF_FORMAT_NO_OUTPUT)
4187                 count++;
4188         if (count > 1)
4189                 die(_("--name-only, --name-status, --check and -s are mutually exclusive"));
4190
4191         /*
4192          * Most of the time we can say "there are changes"
4193          * only by checking if there are changed paths, but
4194          * --ignore-whitespace* options force us to look
4195          * inside contents.
4196          */
4197
4198         if (DIFF_XDL_TST(options, IGNORE_WHITESPACE) ||
4199             DIFF_XDL_TST(options, IGNORE_WHITESPACE_CHANGE) ||
4200             DIFF_XDL_TST(options, IGNORE_WHITESPACE_AT_EOL))
4201                 DIFF_OPT_SET(options, DIFF_FROM_CONTENTS);
4202         else
4203                 DIFF_OPT_CLR(options, DIFF_FROM_CONTENTS);
4204
4205         if (DIFF_OPT_TST(options, FIND_COPIES_HARDER))
4206                 options->detect_rename = DIFF_DETECT_COPY;
4207
4208         if (!DIFF_OPT_TST(options, RELATIVE_NAME))
4209                 options->prefix = NULL;
4210         if (options->prefix)
4211                 options->prefix_length = strlen(options->prefix);
4212         else
4213                 options->prefix_length = 0;
4214
4215         if (options->output_format & (DIFF_FORMAT_NAME |
4216                                       DIFF_FORMAT_NAME_STATUS |
4217                                       DIFF_FORMAT_CHECKDIFF |
4218                                       DIFF_FORMAT_NO_OUTPUT))
4219                 options->output_format &= ~(DIFF_FORMAT_RAW |
4220                                             DIFF_FORMAT_NUMSTAT |
4221                                             DIFF_FORMAT_DIFFSTAT |
4222                                             DIFF_FORMAT_SHORTSTAT |
4223                                             DIFF_FORMAT_DIRSTAT |
4224                                             DIFF_FORMAT_SUMMARY |
4225                                             DIFF_FORMAT_PATCH);
4226
4227         /*
4228          * These cases always need recursive; we do not drop caller-supplied
4229          * recursive bits for other formats here.
4230          */
4231         if (options->output_format & (DIFF_FORMAT_PATCH |
4232                                       DIFF_FORMAT_NUMSTAT |
4233                                       DIFF_FORMAT_DIFFSTAT |
4234                                       DIFF_FORMAT_SHORTSTAT |
4235                                       DIFF_FORMAT_DIRSTAT |
4236                                       DIFF_FORMAT_SUMMARY |
4237                                       DIFF_FORMAT_CHECKDIFF))
4238                 DIFF_OPT_SET(options, RECURSIVE);
4239         /*
4240          * Also pickaxe would not work very well if you do not say recursive
4241          */
4242         if (options->pickaxe)
4243                 DIFF_OPT_SET(options, RECURSIVE);
4244         /*
4245          * When patches are generated, submodules diffed against the work tree
4246          * must be checked for dirtiness too so it can be shown in the output
4247          */
4248         if (options->output_format & DIFF_FORMAT_PATCH)
4249                 DIFF_OPT_SET(options, DIRTY_SUBMODULES);
4250
4251         if (options->detect_rename && options->rename_limit < 0)
4252                 options->rename_limit = diff_rename_limit_default;
4253         if (options->setup & DIFF_SETUP_USE_CACHE) {
4254                 if (!active_cache)
4255                         /* read-cache does not die even when it fails
4256                          * so it is safe for us to do this here.  Also
4257                          * it does not smudge active_cache or active_nr
4258                          * when it fails, so we do not have to worry about
4259                          * cleaning it up ourselves either.
4260                          */
4261                         read_cache();
4262         }
4263         if (40 < options->abbrev)
4264                 options->abbrev = 40; /* full */
4265
4266         /*
4267          * It does not make sense to show the first hit we happened
4268          * to have found.  It does not make sense not to return with
4269          * exit code in such a case either.
4270          */
4271         if (DIFF_OPT_TST(options, QUICK)) {
4272                 options->output_format = DIFF_FORMAT_NO_OUTPUT;
4273                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
4274         }
4275
4276         options->diff_path_counter = 0;
4277
4278         if (DIFF_OPT_TST(options, FOLLOW_RENAMES) && options->pathspec.nr != 1)
4279                 die(_("--follow requires exactly one pathspec"));
4280
4281         if (!options->use_color || external_diff())
4282                 options->color_moved = 0;
4283 }
4284
4285 static int opt_arg(const char *arg, int arg_short, const char *arg_long, int *val)
4286 {
4287         char c, *eq;
4288         int len;
4289
4290         if (*arg != '-')
4291                 return 0;
4292         c = *++arg;
4293         if (!c)
4294                 return 0;
4295         if (c == arg_short) {
4296                 c = *++arg;
4297                 if (!c)
4298                         return 1;
4299                 if (val && isdigit(c)) {
4300                         char *end;
4301                         int n = strtoul(arg, &end, 10);
4302                         if (*end)
4303                                 return 0;
4304                         *val = n;
4305                         return 1;
4306                 }
4307                 return 0;
4308         }
4309         if (c != '-')
4310                 return 0;
4311         arg++;
4312         eq = strchrnul(arg, '=');
4313         len = eq - arg;
4314         if (!len || strncmp(arg, arg_long, len))
4315                 return 0;
4316         if (*eq) {
4317                 int n;
4318                 char *end;
4319                 if (!isdigit(*++eq))
4320                         return 0;
4321                 n = strtoul(eq, &end, 10);
4322                 if (*end)
4323                         return 0;
4324                 *val = n;
4325         }
4326         return 1;
4327 }
4328
4329 static int diff_scoreopt_parse(const char *opt);
4330
4331 static inline int short_opt(char opt, const char **argv,
4332                             const char **optarg)
4333 {
4334         const char *arg = argv[0];
4335         if (arg[0] != '-' || arg[1] != opt)
4336                 return 0;
4337         if (arg[2] != '\0') {
4338                 *optarg = arg + 2;
4339                 return 1;
4340         }
4341         if (!argv[1])
4342                 die("Option '%c' requires a value", opt);
4343         *optarg = argv[1];
4344         return 2;
4345 }
4346
4347 int parse_long_opt(const char *opt, const char **argv,
4348                    const char **optarg)
4349 {
4350         const char *arg = argv[0];
4351         if (!skip_prefix(arg, "--", &arg))
4352                 return 0;
4353         if (!skip_prefix(arg, opt, &arg))
4354                 return 0;
4355         if (*arg == '=') { /* stuck form: --option=value */
4356                 *optarg = arg + 1;
4357                 return 1;
4358         }
4359         if (*arg != '\0')
4360                 return 0;
4361         /* separate form: --option value */
4362         if (!argv[1])
4363                 die("Option '--%s' requires a value", opt);
4364         *optarg = argv[1];
4365         return 2;
4366 }
4367
4368 static int stat_opt(struct diff_options *options, const char **av)
4369 {
4370         const char *arg = av[0];
4371         char *end;
4372         int width = options->stat_width;
4373         int name_width = options->stat_name_width;
4374         int graph_width = options->stat_graph_width;
4375         int count = options->stat_count;
4376         int argcount = 1;
4377
4378         if (!skip_prefix(arg, "--stat", &arg))
4379                 die("BUG: stat option does not begin with --stat: %s", arg);
4380         end = (char *)arg;
4381
4382         switch (*arg) {
4383         case '-':
4384                 if (skip_prefix(arg, "-width", &arg)) {
4385                         if (*arg == '=')
4386                                 width = strtoul(arg + 1, &end, 10);
4387                         else if (!*arg && !av[1])
4388                                 die_want_option("--stat-width");
4389                         else if (!*arg) {
4390                                 width = strtoul(av[1], &end, 10);
4391                                 argcount = 2;
4392                         }
4393                 } else if (skip_prefix(arg, "-name-width", &arg)) {
4394                         if (*arg == '=')
4395                                 name_width = strtoul(arg + 1, &end, 10);
4396                         else if (!*arg && !av[1])
4397                                 die_want_option("--stat-name-width");
4398                         else if (!*arg) {
4399                                 name_width = strtoul(av[1], &end, 10);
4400                                 argcount = 2;
4401                         }
4402                 } else if (skip_prefix(arg, "-graph-width", &arg)) {
4403                         if (*arg == '=')
4404                                 graph_width = strtoul(arg + 1, &end, 10);
4405                         else if (!*arg && !av[1])
4406                                 die_want_option("--stat-graph-width");
4407                         else if (!*arg) {
4408                                 graph_width = strtoul(av[1], &end, 10);
4409                                 argcount = 2;
4410                         }
4411                 } else if (skip_prefix(arg, "-count", &arg)) {
4412                         if (*arg == '=')
4413                                 count = strtoul(arg + 1, &end, 10);
4414                         else if (!*arg && !av[1])
4415                                 die_want_option("--stat-count");
4416                         else if (!*arg) {
4417                                 count = strtoul(av[1], &end, 10);
4418                                 argcount = 2;
4419                         }
4420                 }
4421                 break;
4422         case '=':
4423                 width = strtoul(arg+1, &end, 10);
4424                 if (*end == ',')
4425                         name_width = strtoul(end+1, &end, 10);
4426                 if (*end == ',')
4427                         count = strtoul(end+1, &end, 10);
4428         }
4429
4430         /* Important! This checks all the error cases! */
4431         if (*end)
4432                 return 0;
4433         options->output_format |= DIFF_FORMAT_DIFFSTAT;
4434         options->stat_name_width = name_width;
4435         options->stat_graph_width = graph_width;
4436         options->stat_width = width;
4437         options->stat_count = count;
4438         return argcount;
4439 }
4440
4441 static int parse_dirstat_opt(struct diff_options *options, const char *params)
4442 {
4443         struct strbuf errmsg = STRBUF_INIT;
4444         if (parse_dirstat_params(options, params, &errmsg))
4445                 die(_("Failed to parse --dirstat/-X option parameter:\n%s"),
4446                     errmsg.buf);
4447         strbuf_release(&errmsg);
4448         /*
4449          * The caller knows a dirstat-related option is given from the command
4450          * line; allow it to say "return this_function();"
4451          */
4452         options->output_format |= DIFF_FORMAT_DIRSTAT;
4453         return 1;
4454 }
4455
4456 static int parse_submodule_opt(struct diff_options *options, const char *value)
4457 {
4458         if (parse_submodule_params(options, value))
4459                 die(_("Failed to parse --submodule option parameter: '%s'"),
4460                         value);
4461         return 1;
4462 }
4463
4464 static const char diff_status_letters[] = {
4465         DIFF_STATUS_ADDED,
4466         DIFF_STATUS_COPIED,
4467         DIFF_STATUS_DELETED,
4468         DIFF_STATUS_MODIFIED,
4469         DIFF_STATUS_RENAMED,
4470         DIFF_STATUS_TYPE_CHANGED,
4471         DIFF_STATUS_UNKNOWN,
4472         DIFF_STATUS_UNMERGED,
4473         DIFF_STATUS_FILTER_AON,
4474         DIFF_STATUS_FILTER_BROKEN,
4475         '\0',
4476 };
4477
4478 static unsigned int filter_bit['Z' + 1];
4479
4480 static void prepare_filter_bits(void)
4481 {
4482         int i;
4483
4484         if (!filter_bit[DIFF_STATUS_ADDED]) {
4485                 for (i = 0; diff_status_letters[i]; i++)
4486                         filter_bit[(int) diff_status_letters[i]] = (1 << i);
4487         }
4488 }
4489
4490 static unsigned filter_bit_tst(char status, const struct diff_options *opt)
4491 {
4492         return opt->filter & filter_bit[(int) status];
4493 }
4494
4495 static int parse_diff_filter_opt(const char *optarg, struct diff_options *opt)
4496 {
4497         int i, optch;
4498
4499         prepare_filter_bits();
4500
4501         /*
4502          * If there is a negation e.g. 'd' in the input, and we haven't
4503          * initialized the filter field with another --diff-filter, start
4504          * from full set of bits, except for AON.
4505          */
4506         if (!opt->filter) {
4507                 for (i = 0; (optch = optarg[i]) != '\0'; i++) {
4508                         if (optch < 'a' || 'z' < optch)
4509                                 continue;
4510                         opt->filter = (1 << (ARRAY_SIZE(diff_status_letters) - 1)) - 1;
4511                         opt->filter &= ~filter_bit[DIFF_STATUS_FILTER_AON];
4512                         break;
4513                 }
4514         }
4515
4516         for (i = 0; (optch = optarg[i]) != '\0'; i++) {
4517                 unsigned int bit;
4518                 int negate;
4519
4520                 if ('a' <= optch && optch <= 'z') {
4521                         negate = 1;
4522                         optch = toupper(optch);
4523                 } else {
4524                         negate = 0;
4525                 }
4526
4527                 bit = (0 <= optch && optch <= 'Z') ? filter_bit[optch] : 0;
4528                 if (!bit)
4529                         return optarg[i];
4530                 if (negate)
4531                         opt->filter &= ~bit;
4532                 else
4533                         opt->filter |= bit;
4534         }
4535         return 0;
4536 }
4537
4538 static void enable_patch_output(int *fmt) {
4539         *fmt &= ~DIFF_FORMAT_NO_OUTPUT;
4540         *fmt |= DIFF_FORMAT_PATCH;
4541 }
4542
4543 static int parse_ws_error_highlight_opt(struct diff_options *opt, const char *arg)
4544 {
4545         int val = parse_ws_error_highlight(arg);
4546
4547         if (val < 0) {
4548                 error("unknown value after ws-error-highlight=%.*s",
4549                       -1 - val, arg);
4550                 return 0;
4551         }
4552         opt->ws_error_highlight = val;
4553         return 1;
4554 }
4555
4556 int diff_opt_parse(struct diff_options *options,
4557                    const char **av, int ac, const char *prefix)
4558 {
4559         const char *arg = av[0];
4560         const char *optarg;
4561         int argcount;
4562
4563         if (!prefix)
4564                 prefix = "";
4565
4566         /* Output format options */
4567         if (!strcmp(arg, "-p") || !strcmp(arg, "-u") || !strcmp(arg, "--patch")
4568             || opt_arg(arg, 'U', "unified", &options->context))
4569                 enable_patch_output(&options->output_format);
4570         else if (!strcmp(arg, "--raw"))
4571                 options->output_format |= DIFF_FORMAT_RAW;
4572         else if (!strcmp(arg, "--patch-with-raw")) {
4573                 enable_patch_output(&options->output_format);
4574                 options->output_format |= DIFF_FORMAT_RAW;
4575         } else if (!strcmp(arg, "--numstat"))
4576                 options->output_format |= DIFF_FORMAT_NUMSTAT;
4577         else if (!strcmp(arg, "--shortstat"))
4578                 options->output_format |= DIFF_FORMAT_SHORTSTAT;
4579         else if (!strcmp(arg, "-X") || !strcmp(arg, "--dirstat"))
4580                 return parse_dirstat_opt(options, "");
4581         else if (skip_prefix(arg, "-X", &arg))
4582                 return parse_dirstat_opt(options, arg);
4583         else if (skip_prefix(arg, "--dirstat=", &arg))
4584                 return parse_dirstat_opt(options, arg);
4585         else if (!strcmp(arg, "--cumulative"))
4586                 return parse_dirstat_opt(options, "cumulative");
4587         else if (!strcmp(arg, "--dirstat-by-file"))
4588                 return parse_dirstat_opt(options, "files");
4589         else if (skip_prefix(arg, "--dirstat-by-file=", &arg)) {
4590                 parse_dirstat_opt(options, "files");
4591                 return parse_dirstat_opt(options, arg);
4592         }
4593         else if (!strcmp(arg, "--check"))
4594                 options->output_format |= DIFF_FORMAT_CHECKDIFF;
4595         else if (!strcmp(arg, "--summary"))
4596                 options->output_format |= DIFF_FORMAT_SUMMARY;
4597         else if (!strcmp(arg, "--patch-with-stat")) {
4598                 enable_patch_output(&options->output_format);
4599                 options->output_format |= DIFF_FORMAT_DIFFSTAT;
4600         } else if (!strcmp(arg, "--name-only"))
4601                 options->output_format |= DIFF_FORMAT_NAME;
4602         else if (!strcmp(arg, "--name-status"))
4603                 options->output_format |= DIFF_FORMAT_NAME_STATUS;
4604         else if (!strcmp(arg, "-s") || !strcmp(arg, "--no-patch"))
4605                 options->output_format |= DIFF_FORMAT_NO_OUTPUT;
4606         else if (starts_with(arg, "--stat"))
4607                 /* --stat, --stat-width, --stat-name-width, or --stat-count */
4608                 return stat_opt(options, av);
4609
4610         /* renames options */
4611         else if (starts_with(arg, "-B") || starts_with(arg, "--break-rewrites=") ||
4612                  !strcmp(arg, "--break-rewrites")) {
4613                 if ((options->break_opt = diff_scoreopt_parse(arg)) == -1)
4614                         return error("invalid argument to -B: %s", arg+2);
4615         }
4616         else if (starts_with(arg, "-M") || starts_with(arg, "--find-renames=") ||
4617                  !strcmp(arg, "--find-renames")) {
4618                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
4619                         return error("invalid argument to -M: %s", arg+2);
4620                 options->detect_rename = DIFF_DETECT_RENAME;
4621         }
4622         else if (!strcmp(arg, "-D") || !strcmp(arg, "--irreversible-delete")) {
4623                 options->irreversible_delete = 1;
4624         }
4625         else if (starts_with(arg, "-C") || starts_with(arg, "--find-copies=") ||
4626                  !strcmp(arg, "--find-copies")) {
4627                 if (options->detect_rename == DIFF_DETECT_COPY)
4628                         DIFF_OPT_SET(options, FIND_COPIES_HARDER);
4629                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
4630                         return error("invalid argument to -C: %s", arg+2);
4631                 options->detect_rename = DIFF_DETECT_COPY;
4632         }
4633         else if (!strcmp(arg, "--no-renames"))
4634                 options->detect_rename = 0;
4635         else if (!strcmp(arg, "--rename-empty"))
4636                 DIFF_OPT_SET(options, RENAME_EMPTY);
4637         else if (!strcmp(arg, "--no-rename-empty"))
4638                 DIFF_OPT_CLR(options, RENAME_EMPTY);
4639         else if (!strcmp(arg, "--relative"))
4640                 DIFF_OPT_SET(options, RELATIVE_NAME);
4641         else if (skip_prefix(arg, "--relative=", &arg)) {
4642                 DIFF_OPT_SET(options, RELATIVE_NAME);
4643                 options->prefix = arg;
4644         }
4645
4646         /* xdiff options */
4647         else if (!strcmp(arg, "--minimal"))
4648                 DIFF_XDL_SET(options, NEED_MINIMAL);
4649         else if (!strcmp(arg, "--no-minimal"))
4650                 DIFF_XDL_CLR(options, NEED_MINIMAL);
4651         else if (!strcmp(arg, "-w") || !strcmp(arg, "--ignore-all-space"))
4652                 DIFF_XDL_SET(options, IGNORE_WHITESPACE);
4653         else if (!strcmp(arg, "-b") || !strcmp(arg, "--ignore-space-change"))
4654                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_CHANGE);
4655         else if (!strcmp(arg, "--ignore-space-at-eol"))
4656                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_AT_EOL);
4657         else if (!strcmp(arg, "--ignore-blank-lines"))
4658                 DIFF_XDL_SET(options, IGNORE_BLANK_LINES);
4659         else if (!strcmp(arg, "--indent-heuristic"))
4660                 DIFF_XDL_SET(options, INDENT_HEURISTIC);
4661         else if (!strcmp(arg, "--no-indent-heuristic"))
4662                 DIFF_XDL_CLR(options, INDENT_HEURISTIC);
4663         else if (!strcmp(arg, "--patience"))
4664                 options->xdl_opts = DIFF_WITH_ALG(options, PATIENCE_DIFF);
4665         else if (!strcmp(arg, "--histogram"))
4666                 options->xdl_opts = DIFF_WITH_ALG(options, HISTOGRAM_DIFF);
4667         else if ((argcount = parse_long_opt("diff-algorithm", av, &optarg))) {
4668                 long value = parse_algorithm_value(optarg);
4669                 if (value < 0)
4670                         return error("option diff-algorithm accepts \"myers\", "
4671                                      "\"minimal\", \"patience\" and \"histogram\"");
4672                 /* clear out previous settings */
4673                 DIFF_XDL_CLR(options, NEED_MINIMAL);
4674                 options->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
4675                 options->xdl_opts |= value;
4676                 return argcount;
4677         }
4678
4679         /* flags options */
4680         else if (!strcmp(arg, "--binary")) {
4681                 enable_patch_output(&options->output_format);
4682                 DIFF_OPT_SET(options, BINARY);
4683         }
4684         else if (!strcmp(arg, "--full-index"))
4685                 DIFF_OPT_SET(options, FULL_INDEX);
4686         else if (!strcmp(arg, "-a") || !strcmp(arg, "--text"))
4687                 DIFF_OPT_SET(options, TEXT);
4688         else if (!strcmp(arg, "-R"))
4689                 DIFF_OPT_SET(options, REVERSE_DIFF);
4690         else if (!strcmp(arg, "--find-copies-harder"))
4691                 DIFF_OPT_SET(options, FIND_COPIES_HARDER);
4692         else if (!strcmp(arg, "--follow"))
4693                 DIFF_OPT_SET(options, FOLLOW_RENAMES);
4694         else if (!strcmp(arg, "--no-follow")) {
4695                 DIFF_OPT_CLR(options, FOLLOW_RENAMES);
4696                 DIFF_OPT_CLR(options, DEFAULT_FOLLOW_RENAMES);
4697         } else if (!strcmp(arg, "--color"))
4698                 options->use_color = 1;
4699         else if (skip_prefix(arg, "--color=", &arg)) {
4700                 int value = git_config_colorbool(NULL, arg);
4701                 if (value < 0)
4702                         return error("option `color' expects \"always\", \"auto\", or \"never\"");
4703                 options->use_color = value;
4704         }
4705         else if (!strcmp(arg, "--no-color"))
4706                 options->use_color = 0;
4707         else if (!strcmp(arg, "--color-moved")) {
4708                 if (diff_color_moved_default)
4709                         options->color_moved = diff_color_moved_default;
4710                 if (options->color_moved == COLOR_MOVED_NO)
4711                         options->color_moved = COLOR_MOVED_DEFAULT;
4712         } else if (!strcmp(arg, "--no-color-moved"))
4713                 options->color_moved = COLOR_MOVED_NO;
4714         else if (skip_prefix(arg, "--color-moved=", &arg)) {
4715                 int cm = parse_color_moved(arg);
4716                 if (cm < 0)
4717                         die("bad --color-moved argument: %s", arg);
4718                 options->color_moved = cm;
4719         } else if (!strcmp(arg, "--color-words")) {
4720                 options->use_color = 1;
4721                 options->word_diff = DIFF_WORDS_COLOR;
4722         }
4723         else if (skip_prefix(arg, "--color-words=", &arg)) {
4724                 options->use_color = 1;
4725                 options->word_diff = DIFF_WORDS_COLOR;
4726                 options->word_regex = arg;
4727         }
4728         else if (!strcmp(arg, "--word-diff")) {
4729                 if (options->word_diff == DIFF_WORDS_NONE)
4730                         options->word_diff = DIFF_WORDS_PLAIN;
4731         }
4732         else if (skip_prefix(arg, "--word-diff=", &arg)) {
4733                 if (!strcmp(arg, "plain"))
4734                         options->word_diff = DIFF_WORDS_PLAIN;
4735                 else if (!strcmp(arg, "color")) {
4736                         options->use_color = 1;
4737                         options->word_diff = DIFF_WORDS_COLOR;
4738                 }
4739                 else if (!strcmp(arg, "porcelain"))
4740                         options->word_diff = DIFF_WORDS_PORCELAIN;
4741                 else if (!strcmp(arg, "none"))
4742                         options->word_diff = DIFF_WORDS_NONE;
4743                 else
4744                         die("bad --word-diff argument: %s", arg);
4745         }
4746         else if ((argcount = parse_long_opt("word-diff-regex", av, &optarg))) {
4747                 if (options->word_diff == DIFF_WORDS_NONE)
4748                         options->word_diff = DIFF_WORDS_PLAIN;
4749                 options->word_regex = optarg;
4750                 return argcount;
4751         }
4752         else if (!strcmp(arg, "--exit-code"))
4753                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
4754         else if (!strcmp(arg, "--quiet"))
4755                 DIFF_OPT_SET(options, QUICK);
4756         else if (!strcmp(arg, "--ext-diff"))
4757                 DIFF_OPT_SET(options, ALLOW_EXTERNAL);
4758         else if (!strcmp(arg, "--no-ext-diff"))
4759                 DIFF_OPT_CLR(options, ALLOW_EXTERNAL);
4760         else if (!strcmp(arg, "--textconv"))
4761                 DIFF_OPT_SET(options, ALLOW_TEXTCONV);
4762         else if (!strcmp(arg, "--no-textconv"))
4763                 DIFF_OPT_CLR(options, ALLOW_TEXTCONV);
4764         else if (!strcmp(arg, "--ignore-submodules")) {
4765                 DIFF_OPT_SET(options, OVERRIDE_SUBMODULE_CONFIG);
4766                 handle_ignore_submodules_arg(options, "all");
4767         } else if (skip_prefix(arg, "--ignore-submodules=", &arg)) {
4768                 DIFF_OPT_SET(options, OVERRIDE_SUBMODULE_CONFIG);
4769                 handle_ignore_submodules_arg(options, arg);
4770         } else if (!strcmp(arg, "--submodule"))
4771                 options->submodule_format = DIFF_SUBMODULE_LOG;
4772         else if (skip_prefix(arg, "--submodule=", &arg))
4773                 return parse_submodule_opt(options, arg);
4774         else if (skip_prefix(arg, "--ws-error-highlight=", &arg))
4775                 return parse_ws_error_highlight_opt(options, arg);
4776         else if (!strcmp(arg, "--ita-invisible-in-index"))
4777                 options->ita_invisible_in_index = 1;
4778         else if (!strcmp(arg, "--ita-visible-in-index"))
4779                 options->ita_invisible_in_index = 0;
4780
4781         /* misc options */
4782         else if (!strcmp(arg, "-z"))
4783                 options->line_termination = 0;
4784         else if ((argcount = short_opt('l', av, &optarg))) {
4785                 options->rename_limit = strtoul(optarg, NULL, 10);
4786                 return argcount;
4787         }
4788         else if ((argcount = short_opt('S', av, &optarg))) {
4789                 options->pickaxe = optarg;
4790                 options->pickaxe_opts |= DIFF_PICKAXE_KIND_S;
4791                 return argcount;
4792         } else if ((argcount = short_opt('G', av, &optarg))) {
4793                 options->pickaxe = optarg;
4794                 options->pickaxe_opts |= DIFF_PICKAXE_KIND_G;
4795                 return argcount;
4796         }
4797         else if (!strcmp(arg, "--pickaxe-all"))
4798                 options->pickaxe_opts |= DIFF_PICKAXE_ALL;
4799         else if (!strcmp(arg, "--pickaxe-regex"))
4800                 options->pickaxe_opts |= DIFF_PICKAXE_REGEX;
4801         else if ((argcount = short_opt('O', av, &optarg))) {
4802                 options->orderfile = prefix_filename(prefix, optarg);
4803                 return argcount;
4804         }
4805         else if ((argcount = parse_long_opt("diff-filter", av, &optarg))) {
4806                 int offending = parse_diff_filter_opt(optarg, options);
4807                 if (offending)
4808                         die("unknown change class '%c' in --diff-filter=%s",
4809                             offending, optarg);
4810                 return argcount;
4811         }
4812         else if (!strcmp(arg, "--no-abbrev"))
4813                 options->abbrev = 0;
4814         else if (!strcmp(arg, "--abbrev"))
4815                 options->abbrev = DEFAULT_ABBREV;
4816         else if (skip_prefix(arg, "--abbrev=", &arg)) {
4817                 options->abbrev = strtoul(arg, NULL, 10);
4818                 if (options->abbrev < MINIMUM_ABBREV)
4819                         options->abbrev = MINIMUM_ABBREV;
4820                 else if (40 < options->abbrev)
4821                         options->abbrev = 40;
4822         }
4823         else if ((argcount = parse_long_opt("src-prefix", av, &optarg))) {
4824                 options->a_prefix = optarg;
4825                 return argcount;
4826         }
4827         else if ((argcount = parse_long_opt("line-prefix", av, &optarg))) {
4828                 options->line_prefix = optarg;
4829                 options->line_prefix_length = strlen(options->line_prefix);
4830                 graph_setup_line_prefix(options);
4831                 return argcount;
4832         }
4833         else if ((argcount = parse_long_opt("dst-prefix", av, &optarg))) {
4834                 options->b_prefix = optarg;
4835                 return argcount;
4836         }
4837         else if (!strcmp(arg, "--no-prefix"))
4838                 options->a_prefix = options->b_prefix = "";
4839         else if (opt_arg(arg, '\0', "inter-hunk-context",
4840                          &options->interhunkcontext))
4841                 ;
4842         else if (!strcmp(arg, "-W"))
4843                 DIFF_OPT_SET(options, FUNCCONTEXT);
4844         else if (!strcmp(arg, "--function-context"))
4845                 DIFF_OPT_SET(options, FUNCCONTEXT);
4846         else if (!strcmp(arg, "--no-function-context"))
4847                 DIFF_OPT_CLR(options, FUNCCONTEXT);
4848         else if ((argcount = parse_long_opt("output", av, &optarg))) {
4849                 char *path = prefix_filename(prefix, optarg);
4850                 options->file = xfopen(path, "w");
4851                 options->close_file = 1;
4852                 if (options->use_color != GIT_COLOR_ALWAYS)
4853                         options->use_color = GIT_COLOR_NEVER;
4854                 free(path);
4855                 return argcount;
4856         } else
4857                 return 0;
4858         return 1;
4859 }
4860
4861 int parse_rename_score(const char **cp_p)
4862 {
4863         unsigned long num, scale;
4864         int ch, dot;
4865         const char *cp = *cp_p;
4866
4867         num = 0;
4868         scale = 1;
4869         dot = 0;
4870         for (;;) {
4871                 ch = *cp;
4872                 if ( !dot && ch == '.' ) {
4873                         scale = 1;
4874                         dot = 1;
4875                 } else if ( ch == '%' ) {
4876                         scale = dot ? scale*100 : 100;
4877                         cp++;   /* % is always at the end */
4878                         break;
4879                 } else if ( ch >= '0' && ch <= '9' ) {
4880                         if ( scale < 100000 ) {
4881                                 scale *= 10;
4882                                 num = (num*10) + (ch-'0');
4883                         }
4884                 } else {
4885                         break;
4886                 }
4887                 cp++;
4888         }
4889         *cp_p = cp;
4890
4891         /* user says num divided by scale and we say internally that
4892          * is MAX_SCORE * num / scale.
4893          */
4894         return (int)((num >= scale) ? MAX_SCORE : (MAX_SCORE * num / scale));
4895 }
4896
4897 static int diff_scoreopt_parse(const char *opt)
4898 {
4899         int opt1, opt2, cmd;
4900
4901         if (*opt++ != '-')
4902                 return -1;
4903         cmd = *opt++;
4904         if (cmd == '-') {
4905                 /* convert the long-form arguments into short-form versions */
4906                 if (skip_prefix(opt, "break-rewrites", &opt)) {
4907                         if (*opt == 0 || *opt++ == '=')
4908                                 cmd = 'B';
4909                 } else if (skip_prefix(opt, "find-copies", &opt)) {
4910                         if (*opt == 0 || *opt++ == '=')
4911                                 cmd = 'C';
4912                 } else if (skip_prefix(opt, "find-renames", &opt)) {
4913                         if (*opt == 0 || *opt++ == '=')
4914                                 cmd = 'M';
4915                 }
4916         }
4917         if (cmd != 'M' && cmd != 'C' && cmd != 'B')
4918                 return -1; /* that is not a -M, -C, or -B option */
4919
4920         opt1 = parse_rename_score(&opt);
4921         if (cmd != 'B')
4922                 opt2 = 0;
4923         else {
4924                 if (*opt == 0)
4925                         opt2 = 0;
4926                 else if (*opt != '/')
4927                         return -1; /* we expect -B80/99 or -B80 */
4928                 else {
4929                         opt++;
4930                         opt2 = parse_rename_score(&opt);
4931                 }
4932         }
4933         if (*opt != 0)
4934                 return -1;
4935         return opt1 | (opt2 << 16);
4936 }
4937
4938 struct diff_queue_struct diff_queued_diff;
4939
4940 void diff_q(struct diff_queue_struct *queue, struct diff_filepair *dp)
4941 {
4942         ALLOC_GROW(queue->queue, queue->nr + 1, queue->alloc);
4943         queue->queue[queue->nr++] = dp;
4944 }
4945
4946 struct diff_filepair *diff_queue(struct diff_queue_struct *queue,
4947                                  struct diff_filespec *one,
4948                                  struct diff_filespec *two)
4949 {
4950         struct diff_filepair *dp = xcalloc(1, sizeof(*dp));
4951         dp->one = one;
4952         dp->two = two;
4953         if (queue)
4954                 diff_q(queue, dp);
4955         return dp;
4956 }
4957
4958 void diff_free_filepair(struct diff_filepair *p)
4959 {
4960         free_filespec(p->one);
4961         free_filespec(p->two);
4962         free(p);
4963 }
4964
4965 const char *diff_aligned_abbrev(const struct object_id *oid, int len)
4966 {
4967         int abblen;
4968         const char *abbrev;
4969
4970         if (len == GIT_SHA1_HEXSZ)
4971                 return oid_to_hex(oid);
4972
4973         abbrev = diff_abbrev_oid(oid, len);
4974         abblen = strlen(abbrev);
4975
4976         /*
4977          * In well-behaved cases, where the abbbreviated result is the
4978          * same as the requested length, append three dots after the
4979          * abbreviation (hence the whole logic is limited to the case
4980          * where abblen < 37); when the actual abbreviated result is a
4981          * bit longer than the requested length, we reduce the number
4982          * of dots so that they match the well-behaved ones.  However,
4983          * if the actual abbreviation is longer than the requested
4984          * length by more than three, we give up on aligning, and add
4985          * three dots anyway, to indicate that the output is not the
4986          * full object name.  Yes, this may be suboptimal, but this
4987          * appears only in "diff --raw --abbrev" output and it is not
4988          * worth the effort to change it now.  Note that this would
4989          * likely to work fine when the automatic sizing of default
4990          * abbreviation length is used--we would be fed -1 in "len" in
4991          * that case, and will end up always appending three-dots, but
4992          * the automatic sizing is supposed to give abblen that ensures
4993          * uniqueness across all objects (statistically speaking).
4994          */
4995         if (abblen < GIT_SHA1_HEXSZ - 3) {
4996                 static char hex[GIT_MAX_HEXSZ + 1];
4997                 if (len < abblen && abblen <= len + 2)
4998                         xsnprintf(hex, sizeof(hex), "%s%.*s", abbrev, len+3-abblen, "..");
4999                 else
5000                         xsnprintf(hex, sizeof(hex), "%s...", abbrev);
5001                 return hex;
5002         }
5003
5004         return oid_to_hex(oid);
5005 }
5006
5007 static void diff_flush_raw(struct diff_filepair *p, struct diff_options *opt)
5008 {
5009         int line_termination = opt->line_termination;
5010         int inter_name_termination = line_termination ? '\t' : '\0';
5011
5012         fprintf(opt->file, "%s", diff_line_prefix(opt));
5013         if (!(opt->output_format & DIFF_FORMAT_NAME_STATUS)) {
5014                 fprintf(opt->file, ":%06o %06o %s ", p->one->mode, p->two->mode,
5015                         diff_aligned_abbrev(&p->one->oid, opt->abbrev));
5016                 fprintf(opt->file, "%s ",
5017                         diff_aligned_abbrev(&p->two->oid, opt->abbrev));
5018         }
5019         if (p->score) {
5020                 fprintf(opt->file, "%c%03d%c", p->status, similarity_index(p),
5021                         inter_name_termination);
5022         } else {
5023                 fprintf(opt->file, "%c%c", p->status, inter_name_termination);
5024         }
5025
5026         if (p->status == DIFF_STATUS_COPIED ||
5027             p->status == DIFF_STATUS_RENAMED) {
5028                 const char *name_a, *name_b;
5029                 name_a = p->one->path;
5030                 name_b = p->two->path;
5031                 strip_prefix(opt->prefix_length, &name_a, &name_b);
5032                 write_name_quoted(name_a, opt->file, inter_name_termination);
5033                 write_name_quoted(name_b, opt->file, line_termination);
5034         } else {
5035                 const char *name_a, *name_b;
5036                 name_a = p->one->mode ? p->one->path : p->two->path;
5037                 name_b = NULL;
5038                 strip_prefix(opt->prefix_length, &name_a, &name_b);
5039                 write_name_quoted(name_a, opt->file, line_termination);
5040         }
5041 }
5042
5043 int diff_unmodified_pair(struct diff_filepair *p)
5044 {
5045         /* This function is written stricter than necessary to support
5046          * the currently implemented transformers, but the idea is to
5047          * let transformers to produce diff_filepairs any way they want,
5048          * and filter and clean them up here before producing the output.
5049          */
5050         struct diff_filespec *one = p->one, *two = p->two;
5051
5052         if (DIFF_PAIR_UNMERGED(p))
5053                 return 0; /* unmerged is interesting */
5054
5055         /* deletion, addition, mode or type change
5056          * and rename are all interesting.
5057          */
5058         if (DIFF_FILE_VALID(one) != DIFF_FILE_VALID(two) ||
5059             DIFF_PAIR_MODE_CHANGED(p) ||
5060             strcmp(one->path, two->path))
5061                 return 0;
5062
5063         /* both are valid and point at the same path.  that is, we are
5064          * dealing with a change.
5065          */
5066         if (one->oid_valid && two->oid_valid &&
5067             !oidcmp(&one->oid, &two->oid) &&
5068             !one->dirty_submodule && !two->dirty_submodule)
5069                 return 1; /* no change */
5070         if (!one->oid_valid && !two->oid_valid)
5071                 return 1; /* both look at the same file on the filesystem. */
5072         return 0;
5073 }
5074
5075 static void diff_flush_patch(struct diff_filepair *p, struct diff_options *o)
5076 {
5077         if (diff_unmodified_pair(p))
5078                 return;
5079
5080         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
5081             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
5082                 return; /* no tree diffs in patch format */
5083
5084         run_diff(p, o);
5085 }
5086
5087 static void diff_flush_stat(struct diff_filepair *p, struct diff_options *o,
5088                             struct diffstat_t *diffstat)
5089 {
5090         if (diff_unmodified_pair(p))
5091                 return;
5092
5093         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
5094             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
5095                 return; /* no useful stat for tree diffs */
5096
5097         run_diffstat(p, o, diffstat);
5098 }
5099
5100 static void diff_flush_checkdiff(struct diff_filepair *p,
5101                 struct diff_options *o)
5102 {
5103         if (diff_unmodified_pair(p))
5104                 return;
5105
5106         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
5107             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
5108                 return; /* nothing to check in tree diffs */
5109
5110         run_checkdiff(p, o);
5111 }
5112
5113 int diff_queue_is_empty(void)
5114 {
5115         struct diff_queue_struct *q = &diff_queued_diff;
5116         int i;
5117         for (i = 0; i < q->nr; i++)
5118                 if (!diff_unmodified_pair(q->queue[i]))
5119                         return 0;
5120         return 1;
5121 }
5122
5123 #if DIFF_DEBUG
5124 void diff_debug_filespec(struct diff_filespec *s, int x, const char *one)
5125 {
5126         fprintf(stderr, "queue[%d] %s (%s) %s %06o %s\n",
5127                 x, one ? one : "",
5128                 s->path,
5129                 DIFF_FILE_VALID(s) ? "valid" : "invalid",
5130                 s->mode,
5131                 s->oid_valid ? oid_to_hex(&s->oid) : "");
5132         fprintf(stderr, "queue[%d] %s size %lu\n",
5133                 x, one ? one : "",
5134                 s->size);
5135 }
5136
5137 void diff_debug_filepair(const struct diff_filepair *p, int i)
5138 {
5139         diff_debug_filespec(p->one, i, "one");
5140         diff_debug_filespec(p->two, i, "two");
5141         fprintf(stderr, "score %d, status %c rename_used %d broken %d\n",
5142                 p->score, p->status ? p->status : '?',
5143                 p->one->rename_used, p->broken_pair);
5144 }
5145
5146 void diff_debug_queue(const char *msg, struct diff_queue_struct *q)
5147 {
5148         int i;
5149         if (msg)
5150                 fprintf(stderr, "%s\n", msg);
5151         fprintf(stderr, "q->nr = %d\n", q->nr);
5152         for (i = 0; i < q->nr; i++) {
5153                 struct diff_filepair *p = q->queue[i];
5154                 diff_debug_filepair(p, i);
5155         }
5156 }
5157 #endif
5158
5159 static void diff_resolve_rename_copy(void)
5160 {
5161         int i;
5162         struct diff_filepair *p;
5163         struct diff_queue_struct *q = &diff_queued_diff;
5164
5165         diff_debug_queue("resolve-rename-copy", q);
5166
5167         for (i = 0; i < q->nr; i++) {
5168                 p = q->queue[i];
5169                 p->status = 0; /* undecided */
5170                 if (DIFF_PAIR_UNMERGED(p))
5171                         p->status = DIFF_STATUS_UNMERGED;
5172                 else if (!DIFF_FILE_VALID(p->one))
5173                         p->status = DIFF_STATUS_ADDED;
5174                 else if (!DIFF_FILE_VALID(p->two))
5175                         p->status = DIFF_STATUS_DELETED;
5176                 else if (DIFF_PAIR_TYPE_CHANGED(p))
5177                         p->status = DIFF_STATUS_TYPE_CHANGED;
5178
5179                 /* from this point on, we are dealing with a pair
5180                  * whose both sides are valid and of the same type, i.e.
5181                  * either in-place edit or rename/copy edit.
5182                  */
5183                 else if (DIFF_PAIR_RENAME(p)) {
5184                         /*
5185                          * A rename might have re-connected a broken
5186                          * pair up, causing the pathnames to be the
5187                          * same again. If so, that's not a rename at
5188                          * all, just a modification..
5189                          *
5190                          * Otherwise, see if this source was used for
5191                          * multiple renames, in which case we decrement
5192                          * the count, and call it a copy.
5193                          */
5194                         if (!strcmp(p->one->path, p->two->path))
5195                                 p->status = DIFF_STATUS_MODIFIED;
5196                         else if (--p->one->rename_used > 0)
5197                                 p->status = DIFF_STATUS_COPIED;
5198                         else
5199                                 p->status = DIFF_STATUS_RENAMED;
5200                 }
5201                 else if (oidcmp(&p->one->oid, &p->two->oid) ||
5202                          p->one->mode != p->two->mode ||
5203                          p->one->dirty_submodule ||
5204                          p->two->dirty_submodule ||
5205                          is_null_oid(&p->one->oid))
5206                         p->status = DIFF_STATUS_MODIFIED;
5207                 else {
5208                         /* This is a "no-change" entry and should not
5209                          * happen anymore, but prepare for broken callers.
5210                          */
5211                         error("feeding unmodified %s to diffcore",
5212                               p->one->path);
5213                         p->status = DIFF_STATUS_UNKNOWN;
5214                 }
5215         }
5216         diff_debug_queue("resolve-rename-copy done", q);
5217 }
5218
5219 static int check_pair_status(struct diff_filepair *p)
5220 {
5221         switch (p->status) {
5222         case DIFF_STATUS_UNKNOWN:
5223                 return 0;
5224         case 0:
5225                 die("internal error in diff-resolve-rename-copy");
5226         default:
5227                 return 1;
5228         }
5229 }
5230
5231 static void flush_one_pair(struct diff_filepair *p, struct diff_options *opt)
5232 {
5233         int fmt = opt->output_format;
5234
5235         if (fmt & DIFF_FORMAT_CHECKDIFF)
5236                 diff_flush_checkdiff(p, opt);
5237         else if (fmt & (DIFF_FORMAT_RAW | DIFF_FORMAT_NAME_STATUS))
5238                 diff_flush_raw(p, opt);
5239         else if (fmt & DIFF_FORMAT_NAME) {
5240                 const char *name_a, *name_b;
5241                 name_a = p->two->path;
5242                 name_b = NULL;
5243                 strip_prefix(opt->prefix_length, &name_a, &name_b);
5244                 fprintf(opt->file, "%s", diff_line_prefix(opt));
5245                 write_name_quoted(name_a, opt->file, opt->line_termination);
5246         }
5247 }
5248
5249 static void show_file_mode_name(struct diff_options *opt, const char *newdelete, struct diff_filespec *fs)
5250 {
5251         struct strbuf sb = STRBUF_INIT;
5252         if (fs->mode)
5253                 strbuf_addf(&sb, " %s mode %06o ", newdelete, fs->mode);
5254         else
5255                 strbuf_addf(&sb, " %s ", newdelete);
5256
5257         quote_c_style(fs->path, &sb, NULL, 0);
5258         strbuf_addch(&sb, '\n');
5259         emit_diff_symbol(opt, DIFF_SYMBOL_SUMMARY,
5260                          sb.buf, sb.len, 0);
5261         strbuf_release(&sb);
5262 }
5263
5264 static void show_mode_change(struct diff_options *opt, struct diff_filepair *p,
5265                 int show_name)
5266 {
5267         if (p->one->mode && p->two->mode && p->one->mode != p->two->mode) {
5268                 struct strbuf sb = STRBUF_INIT;
5269                 strbuf_addf(&sb, " mode change %06o => %06o",
5270                             p->one->mode, p->two->mode);
5271                 if (show_name) {
5272                         strbuf_addch(&sb, ' ');
5273                         quote_c_style(p->two->path, &sb, NULL, 0);
5274                 }
5275                 emit_diff_symbol(opt, DIFF_SYMBOL_SUMMARY,
5276                                  sb.buf, sb.len, 0);
5277                 strbuf_release(&sb);
5278         }
5279 }
5280
5281 static void show_rename_copy(struct diff_options *opt, const char *renamecopy,
5282                 struct diff_filepair *p)
5283 {
5284         struct strbuf sb = STRBUF_INIT;
5285         char *names = pprint_rename(p->one->path, p->two->path);
5286         strbuf_addf(&sb, " %s %s (%d%%)\n",
5287                         renamecopy, names, similarity_index(p));
5288         free(names);
5289         emit_diff_symbol(opt, DIFF_SYMBOL_SUMMARY,
5290                                  sb.buf, sb.len, 0);
5291         show_mode_change(opt, p, 0);
5292         strbuf_release(&sb);
5293 }
5294
5295 static void diff_summary(struct diff_options *opt, struct diff_filepair *p)
5296 {
5297         switch(p->status) {
5298         case DIFF_STATUS_DELETED:
5299                 show_file_mode_name(opt, "delete", p->one);
5300                 break;
5301         case DIFF_STATUS_ADDED:
5302                 show_file_mode_name(opt, "create", p->two);
5303                 break;
5304         case DIFF_STATUS_COPIED:
5305                 show_rename_copy(opt, "copy", p);
5306                 break;
5307         case DIFF_STATUS_RENAMED:
5308                 show_rename_copy(opt, "rename", p);
5309                 break;
5310         default:
5311                 if (p->score) {
5312                         struct strbuf sb = STRBUF_INIT;
5313                         strbuf_addstr(&sb, " rewrite ");
5314                         quote_c_style(p->two->path, &sb, NULL, 0);
5315                         strbuf_addf(&sb, " (%d%%)\n", similarity_index(p));
5316                         emit_diff_symbol(opt, DIFF_SYMBOL_SUMMARY,
5317                                          sb.buf, sb.len, 0);
5318                         strbuf_release(&sb);
5319                 }
5320                 show_mode_change(opt, p, !p->score);
5321                 break;
5322         }
5323 }
5324
5325 struct patch_id_t {
5326         git_SHA_CTX *ctx;
5327         int patchlen;
5328 };
5329
5330 static int remove_space(char *line, int len)
5331 {
5332         int i;
5333         char *dst = line;
5334         unsigned char c;
5335
5336         for (i = 0; i < len; i++)
5337                 if (!isspace((c = line[i])))
5338                         *dst++ = c;
5339
5340         return dst - line;
5341 }
5342
5343 static void patch_id_consume(void *priv, char *line, unsigned long len)
5344 {
5345         struct patch_id_t *data = priv;
5346         int new_len;
5347
5348         /* Ignore line numbers when computing the SHA1 of the patch */
5349         if (starts_with(line, "@@ -"))
5350                 return;
5351
5352         new_len = remove_space(line, len);
5353
5354         git_SHA1_Update(data->ctx, line, new_len);
5355         data->patchlen += new_len;
5356 }
5357
5358 static void patch_id_add_string(git_SHA_CTX *ctx, const char *str)
5359 {
5360         git_SHA1_Update(ctx, str, strlen(str));
5361 }
5362
5363 static void patch_id_add_mode(git_SHA_CTX *ctx, unsigned mode)
5364 {
5365         /* large enough for 2^32 in octal */
5366         char buf[12];
5367         int len = xsnprintf(buf, sizeof(buf), "%06o", mode);
5368         git_SHA1_Update(ctx, buf, len);
5369 }
5370
5371 /* returns 0 upon success, and writes result into sha1 */
5372 static int diff_get_patch_id(struct diff_options *options, struct object_id *oid, int diff_header_only)
5373 {
5374         struct diff_queue_struct *q = &diff_queued_diff;
5375         int i;
5376         git_SHA_CTX ctx;
5377         struct patch_id_t data;
5378
5379         git_SHA1_Init(&ctx);
5380         memset(&data, 0, sizeof(struct patch_id_t));
5381         data.ctx = &ctx;
5382
5383         for (i = 0; i < q->nr; i++) {
5384                 xpparam_t xpp;
5385                 xdemitconf_t xecfg;
5386                 mmfile_t mf1, mf2;
5387                 struct diff_filepair *p = q->queue[i];
5388                 int len1, len2;
5389
5390                 memset(&xpp, 0, sizeof(xpp));
5391                 memset(&xecfg, 0, sizeof(xecfg));
5392                 if (p->status == 0)
5393                         return error("internal diff status error");
5394                 if (p->status == DIFF_STATUS_UNKNOWN)
5395                         continue;
5396                 if (diff_unmodified_pair(p))
5397                         continue;
5398                 if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
5399                     (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
5400                         continue;
5401                 if (DIFF_PAIR_UNMERGED(p))
5402                         continue;
5403
5404                 diff_fill_oid_info(p->one);
5405                 diff_fill_oid_info(p->two);
5406
5407                 len1 = remove_space(p->one->path, strlen(p->one->path));
5408                 len2 = remove_space(p->two->path, strlen(p->two->path));
5409                 patch_id_add_string(&ctx, "diff--git");
5410                 patch_id_add_string(&ctx, "a/");
5411                 git_SHA1_Update(&ctx, p->one->path, len1);
5412                 patch_id_add_string(&ctx, "b/");
5413                 git_SHA1_Update(&ctx, p->two->path, len2);
5414
5415                 if (p->one->mode == 0) {
5416                         patch_id_add_string(&ctx, "newfilemode");
5417                         patch_id_add_mode(&ctx, p->two->mode);
5418                         patch_id_add_string(&ctx, "---/dev/null");
5419                         patch_id_add_string(&ctx, "+++b/");
5420                         git_SHA1_Update(&ctx, p->two->path, len2);
5421                 } else if (p->two->mode == 0) {
5422                         patch_id_add_string(&ctx, "deletedfilemode");
5423                         patch_id_add_mode(&ctx, p->one->mode);
5424                         patch_id_add_string(&ctx, "---a/");
5425                         git_SHA1_Update(&ctx, p->one->path, len1);
5426                         patch_id_add_string(&ctx, "+++/dev/null");
5427                 } else {
5428                         patch_id_add_string(&ctx, "---a/");
5429                         git_SHA1_Update(&ctx, p->one->path, len1);
5430                         patch_id_add_string(&ctx, "+++b/");
5431                         git_SHA1_Update(&ctx, p->two->path, len2);
5432                 }
5433
5434                 if (diff_header_only)
5435                         continue;
5436
5437                 if (fill_mmfile(&mf1, p->one) < 0 ||
5438                     fill_mmfile(&mf2, p->two) < 0)
5439                         return error("unable to read files to diff");
5440
5441                 if (diff_filespec_is_binary(p->one) ||
5442                     diff_filespec_is_binary(p->two)) {
5443                         git_SHA1_Update(&ctx, oid_to_hex(&p->one->oid),
5444                                         GIT_SHA1_HEXSZ);
5445                         git_SHA1_Update(&ctx, oid_to_hex(&p->two->oid),
5446                                         GIT_SHA1_HEXSZ);
5447                         continue;
5448                 }
5449
5450                 xpp.flags = 0;
5451                 xecfg.ctxlen = 3;
5452                 xecfg.flags = 0;
5453                 if (xdi_diff_outf(&mf1, &mf2, patch_id_consume, &data,
5454                                   &xpp, &xecfg))
5455                         return error("unable to generate patch-id diff for %s",
5456                                      p->one->path);
5457         }
5458
5459         git_SHA1_Final(oid->hash, &ctx);
5460         return 0;
5461 }
5462
5463 int diff_flush_patch_id(struct diff_options *options, struct object_id *oid, int diff_header_only)
5464 {
5465         struct diff_queue_struct *q = &diff_queued_diff;
5466         int i;
5467         int result = diff_get_patch_id(options, oid, diff_header_only);
5468
5469         for (i = 0; i < q->nr; i++)
5470                 diff_free_filepair(q->queue[i]);
5471
5472         free(q->queue);
5473         DIFF_QUEUE_CLEAR(q);
5474
5475         return result;
5476 }
5477
5478 static int is_summary_empty(const struct diff_queue_struct *q)
5479 {
5480         int i;
5481
5482         for (i = 0; i < q->nr; i++) {
5483                 const struct diff_filepair *p = q->queue[i];
5484
5485                 switch (p->status) {
5486                 case DIFF_STATUS_DELETED:
5487                 case DIFF_STATUS_ADDED:
5488                 case DIFF_STATUS_COPIED:
5489                 case DIFF_STATUS_RENAMED:
5490                         return 0;
5491                 default:
5492                         if (p->score)
5493                                 return 0;
5494                         if (p->one->mode && p->two->mode &&
5495                             p->one->mode != p->two->mode)
5496                                 return 0;
5497                         break;
5498                 }
5499         }
5500         return 1;
5501 }
5502
5503 static const char rename_limit_warning[] =
5504 N_("inexact rename detection was skipped due to too many files.");
5505
5506 static const char degrade_cc_to_c_warning[] =
5507 N_("only found copies from modified paths due to too many files.");
5508
5509 static const char rename_limit_advice[] =
5510 N_("you may want to set your %s variable to at least "
5511    "%d and retry the command.");
5512
5513 void diff_warn_rename_limit(const char *varname, int needed, int degraded_cc)
5514 {
5515         if (degraded_cc)
5516                 warning(_(degrade_cc_to_c_warning));
5517         else if (needed)
5518                 warning(_(rename_limit_warning));
5519         else
5520                 return;
5521         if (0 < needed && needed < 32767)
5522                 warning(_(rename_limit_advice), varname, needed);
5523 }
5524
5525 static void diff_flush_patch_all_file_pairs(struct diff_options *o)
5526 {
5527         int i;
5528         static struct emitted_diff_symbols esm = EMITTED_DIFF_SYMBOLS_INIT;
5529         struct diff_queue_struct *q = &diff_queued_diff;
5530
5531         if (WSEH_NEW & WS_RULE_MASK)
5532                 die("BUG: WS rules bit mask overlaps with diff symbol flags");
5533
5534         if (o->color_moved)
5535                 o->emitted_symbols = &esm;
5536
5537         for (i = 0; i < q->nr; i++) {
5538                 struct diff_filepair *p = q->queue[i];
5539                 if (check_pair_status(p))
5540                         diff_flush_patch(p, o);
5541         }
5542
5543         if (o->emitted_symbols) {
5544                 if (o->color_moved) {
5545                         struct hashmap add_lines, del_lines;
5546
5547                         hashmap_init(&del_lines,
5548                                      (hashmap_cmp_fn)moved_entry_cmp, o, 0);
5549                         hashmap_init(&add_lines,
5550                                      (hashmap_cmp_fn)moved_entry_cmp, o, 0);
5551
5552                         add_lines_to_move_detection(o, &add_lines, &del_lines);
5553                         mark_color_as_moved(o, &add_lines, &del_lines);
5554                         if (o->color_moved == COLOR_MOVED_ZEBRA_DIM)
5555                                 dim_moved_lines(o);
5556
5557                         hashmap_free(&add_lines, 0);
5558                         hashmap_free(&del_lines, 0);
5559                 }
5560
5561                 for (i = 0; i < esm.nr; i++)
5562                         emit_diff_symbol_from_struct(o, &esm.buf[i]);
5563
5564                 for (i = 0; i < esm.nr; i++)
5565                         free((void *)esm.buf[i].line);
5566         }
5567         esm.nr = 0;
5568 }
5569
5570 void diff_flush(struct diff_options *options)
5571 {
5572         struct diff_queue_struct *q = &diff_queued_diff;
5573         int i, output_format = options->output_format;
5574         int separator = 0;
5575         int dirstat_by_line = 0;
5576
5577         /*
5578          * Order: raw, stat, summary, patch
5579          * or:    name/name-status/checkdiff (other bits clear)
5580          */
5581         if (!q->nr)
5582                 goto free_queue;
5583
5584         if (output_format & (DIFF_FORMAT_RAW |
5585                              DIFF_FORMAT_NAME |
5586                              DIFF_FORMAT_NAME_STATUS |
5587                              DIFF_FORMAT_CHECKDIFF)) {
5588                 for (i = 0; i < q->nr; i++) {
5589                         struct diff_filepair *p = q->queue[i];
5590                         if (check_pair_status(p))
5591                                 flush_one_pair(p, options);
5592                 }
5593                 separator++;
5594         }
5595
5596         if (output_format & DIFF_FORMAT_DIRSTAT && DIFF_OPT_TST(options, DIRSTAT_BY_LINE))
5597                 dirstat_by_line = 1;
5598
5599         if (output_format & (DIFF_FORMAT_DIFFSTAT|DIFF_FORMAT_SHORTSTAT|DIFF_FORMAT_NUMSTAT) ||
5600             dirstat_by_line) {
5601                 struct diffstat_t diffstat;
5602
5603                 memset(&diffstat, 0, sizeof(struct diffstat_t));
5604                 for (i = 0; i < q->nr; i++) {
5605                         struct diff_filepair *p = q->queue[i];
5606                         if (check_pair_status(p))
5607                                 diff_flush_stat(p, options, &diffstat);
5608                 }
5609                 if (output_format & DIFF_FORMAT_NUMSTAT)
5610                         show_numstat(&diffstat, options);
5611                 if (output_format & DIFF_FORMAT_DIFFSTAT)
5612                         show_stats(&diffstat, options);
5613                 if (output_format & DIFF_FORMAT_SHORTSTAT)
5614                         show_shortstats(&diffstat, options);
5615                 if (output_format & DIFF_FORMAT_DIRSTAT && dirstat_by_line)
5616                         show_dirstat_by_line(&diffstat, options);
5617                 free_diffstat_info(&diffstat);
5618                 separator++;
5619         }
5620         if ((output_format & DIFF_FORMAT_DIRSTAT) && !dirstat_by_line)
5621                 show_dirstat(options);
5622
5623         if (output_format & DIFF_FORMAT_SUMMARY && !is_summary_empty(q)) {
5624                 for (i = 0; i < q->nr; i++) {
5625                         diff_summary(options, q->queue[i]);
5626                 }
5627                 separator++;
5628         }
5629
5630         if (output_format & DIFF_FORMAT_NO_OUTPUT &&
5631             DIFF_OPT_TST(options, EXIT_WITH_STATUS) &&
5632             DIFF_OPT_TST(options, DIFF_FROM_CONTENTS)) {
5633                 /*
5634                  * run diff_flush_patch for the exit status. setting
5635                  * options->file to /dev/null should be safe, because we
5636                  * aren't supposed to produce any output anyway.
5637                  */
5638                 if (options->close_file)
5639                         fclose(options->file);
5640                 options->file = xfopen("/dev/null", "w");
5641                 options->close_file = 1;
5642                 options->color_moved = 0;
5643                 for (i = 0; i < q->nr; i++) {
5644                         struct diff_filepair *p = q->queue[i];
5645                         if (check_pair_status(p))
5646                                 diff_flush_patch(p, options);
5647                         if (options->found_changes)
5648                                 break;
5649                 }
5650         }
5651
5652         if (output_format & DIFF_FORMAT_PATCH) {
5653                 if (separator) {
5654                         emit_diff_symbol(options, DIFF_SYMBOL_SEPARATOR, NULL, 0, 0);
5655                         if (options->stat_sep)
5656                                 /* attach patch instead of inline */
5657                                 emit_diff_symbol(options, DIFF_SYMBOL_STAT_SEP,
5658                                                  NULL, 0, 0);
5659                 }
5660
5661                 diff_flush_patch_all_file_pairs(options);
5662         }
5663
5664         if (output_format & DIFF_FORMAT_CALLBACK)
5665                 options->format_callback(q, options, options->format_callback_data);
5666
5667         for (i = 0; i < q->nr; i++)
5668                 diff_free_filepair(q->queue[i]);
5669 free_queue:
5670         free(q->queue);
5671         DIFF_QUEUE_CLEAR(q);
5672         if (options->close_file)
5673                 fclose(options->file);
5674
5675         /*
5676          * Report the content-level differences with HAS_CHANGES;
5677          * diff_addremove/diff_change does not set the bit when
5678          * DIFF_FROM_CONTENTS is in effect (e.g. with -w).
5679          */
5680         if (DIFF_OPT_TST(options, DIFF_FROM_CONTENTS)) {
5681                 if (options->found_changes)
5682                         DIFF_OPT_SET(options, HAS_CHANGES);
5683                 else
5684                         DIFF_OPT_CLR(options, HAS_CHANGES);
5685         }
5686 }
5687
5688 static int match_filter(const struct diff_options *options, const struct diff_filepair *p)
5689 {
5690         return (((p->status == DIFF_STATUS_MODIFIED) &&
5691                  ((p->score &&
5692                    filter_bit_tst(DIFF_STATUS_FILTER_BROKEN, options)) ||
5693                   (!p->score &&
5694                    filter_bit_tst(DIFF_STATUS_MODIFIED, options)))) ||
5695                 ((p->status != DIFF_STATUS_MODIFIED) &&
5696                  filter_bit_tst(p->status, options)));
5697 }
5698
5699 static void diffcore_apply_filter(struct diff_options *options)
5700 {
5701         int i;
5702         struct diff_queue_struct *q = &diff_queued_diff;
5703         struct diff_queue_struct outq;
5704
5705         DIFF_QUEUE_CLEAR(&outq);
5706
5707         if (!options->filter)
5708                 return;
5709
5710         if (filter_bit_tst(DIFF_STATUS_FILTER_AON, options)) {
5711                 int found;
5712                 for (i = found = 0; !found && i < q->nr; i++) {
5713                         if (match_filter(options, q->queue[i]))
5714                                 found++;
5715                 }
5716                 if (found)
5717                         return;
5718
5719                 /* otherwise we will clear the whole queue
5720                  * by copying the empty outq at the end of this
5721                  * function, but first clear the current entries
5722                  * in the queue.
5723                  */
5724                 for (i = 0; i < q->nr; i++)
5725                         diff_free_filepair(q->queue[i]);
5726         }
5727         else {
5728                 /* Only the matching ones */
5729                 for (i = 0; i < q->nr; i++) {
5730                         struct diff_filepair *p = q->queue[i];
5731                         if (match_filter(options, p))
5732                                 diff_q(&outq, p);
5733                         else
5734                                 diff_free_filepair(p);
5735                 }
5736         }
5737         free(q->queue);
5738         *q = outq;
5739 }
5740
5741 /* Check whether two filespecs with the same mode and size are identical */
5742 static int diff_filespec_is_identical(struct diff_filespec *one,
5743                                       struct diff_filespec *two)
5744 {
5745         if (S_ISGITLINK(one->mode))
5746                 return 0;
5747         if (diff_populate_filespec(one, 0))
5748                 return 0;
5749         if (diff_populate_filespec(two, 0))
5750                 return 0;
5751         return !memcmp(one->data, two->data, one->size);
5752 }
5753
5754 static int diff_filespec_check_stat_unmatch(struct diff_filepair *p)
5755 {
5756         if (p->done_skip_stat_unmatch)
5757                 return p->skip_stat_unmatch_result;
5758
5759         p->done_skip_stat_unmatch = 1;
5760         p->skip_stat_unmatch_result = 0;
5761         /*
5762          * 1. Entries that come from stat info dirtiness
5763          *    always have both sides (iow, not create/delete),
5764          *    one side of the object name is unknown, with
5765          *    the same mode and size.  Keep the ones that
5766          *    do not match these criteria.  They have real
5767          *    differences.
5768          *
5769          * 2. At this point, the file is known to be modified,
5770          *    with the same mode and size, and the object
5771          *    name of one side is unknown.  Need to inspect
5772          *    the identical contents.
5773          */
5774         if (!DIFF_FILE_VALID(p->one) || /* (1) */
5775             !DIFF_FILE_VALID(p->two) ||
5776             (p->one->oid_valid && p->two->oid_valid) ||
5777             (p->one->mode != p->two->mode) ||
5778             diff_populate_filespec(p->one, CHECK_SIZE_ONLY) ||
5779             diff_populate_filespec(p->two, CHECK_SIZE_ONLY) ||
5780             (p->one->size != p->two->size) ||
5781             !diff_filespec_is_identical(p->one, p->two)) /* (2) */
5782                 p->skip_stat_unmatch_result = 1;
5783         return p->skip_stat_unmatch_result;
5784 }
5785
5786 static void diffcore_skip_stat_unmatch(struct diff_options *diffopt)
5787 {
5788         int i;
5789         struct diff_queue_struct *q = &diff_queued_diff;
5790         struct diff_queue_struct outq;
5791         DIFF_QUEUE_CLEAR(&outq);
5792
5793         for (i = 0; i < q->nr; i++) {
5794                 struct diff_filepair *p = q->queue[i];
5795
5796                 if (diff_filespec_check_stat_unmatch(p))
5797                         diff_q(&outq, p);
5798                 else {
5799                         /*
5800                          * The caller can subtract 1 from skip_stat_unmatch
5801                          * to determine how many paths were dirty only
5802                          * due to stat info mismatch.
5803                          */
5804                         if (!DIFF_OPT_TST(diffopt, NO_INDEX))
5805                                 diffopt->skip_stat_unmatch++;
5806                         diff_free_filepair(p);
5807                 }
5808         }
5809         free(q->queue);
5810         *q = outq;
5811 }
5812
5813 static int diffnamecmp(const void *a_, const void *b_)
5814 {
5815         const struct diff_filepair *a = *((const struct diff_filepair **)a_);
5816         const struct diff_filepair *b = *((const struct diff_filepair **)b_);
5817         const char *name_a, *name_b;
5818
5819         name_a = a->one ? a->one->path : a->two->path;
5820         name_b = b->one ? b->one->path : b->two->path;
5821         return strcmp(name_a, name_b);
5822 }
5823
5824 void diffcore_fix_diff_index(struct diff_options *options)
5825 {
5826         struct diff_queue_struct *q = &diff_queued_diff;
5827         QSORT(q->queue, q->nr, diffnamecmp);
5828 }
5829
5830 void diffcore_std(struct diff_options *options)
5831 {
5832         /* NOTE please keep the following in sync with diff_tree_combined() */
5833         if (options->skip_stat_unmatch)
5834                 diffcore_skip_stat_unmatch(options);
5835         if (!options->found_follow) {
5836                 /* See try_to_follow_renames() in tree-diff.c */
5837                 if (options->break_opt != -1)
5838                         diffcore_break(options->break_opt);
5839                 if (options->detect_rename)
5840                         diffcore_rename(options);
5841                 if (options->break_opt != -1)
5842                         diffcore_merge_broken();
5843         }
5844         if (options->pickaxe)
5845                 diffcore_pickaxe(options);
5846         if (options->orderfile)
5847                 diffcore_order(options->orderfile);
5848         if (!options->found_follow)
5849                 /* See try_to_follow_renames() in tree-diff.c */
5850                 diff_resolve_rename_copy();
5851         diffcore_apply_filter(options);
5852
5853         if (diff_queued_diff.nr && !DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
5854                 DIFF_OPT_SET(options, HAS_CHANGES);
5855         else
5856                 DIFF_OPT_CLR(options, HAS_CHANGES);
5857
5858         options->found_follow = 0;
5859 }
5860
5861 int diff_result_code(struct diff_options *opt, int status)
5862 {
5863         int result = 0;
5864
5865         diff_warn_rename_limit("diff.renameLimit",
5866                                opt->needed_rename_limit,
5867                                opt->degraded_cc_to_c);
5868         if (!DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
5869             !(opt->output_format & DIFF_FORMAT_CHECKDIFF))
5870                 return status;
5871         if (DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
5872             DIFF_OPT_TST(opt, HAS_CHANGES))
5873                 result |= 01;
5874         if ((opt->output_format & DIFF_FORMAT_CHECKDIFF) &&
5875             DIFF_OPT_TST(opt, CHECK_FAILED))
5876                 result |= 02;
5877         return result;
5878 }
5879
5880 int diff_can_quit_early(struct diff_options *opt)
5881 {
5882         return (DIFF_OPT_TST(opt, QUICK) &&
5883                 !opt->filter &&
5884                 DIFF_OPT_TST(opt, HAS_CHANGES));
5885 }
5886
5887 /*
5888  * Shall changes to this submodule be ignored?
5889  *
5890  * Submodule changes can be configured to be ignored separately for each path,
5891  * but that configuration can be overridden from the command line.
5892  */
5893 static int is_submodule_ignored(const char *path, struct diff_options *options)
5894 {
5895         int ignored = 0;
5896         unsigned orig_flags = options->flags;
5897         if (!DIFF_OPT_TST(options, OVERRIDE_SUBMODULE_CONFIG))
5898                 set_diffopt_flags_from_submodule_config(options, path);
5899         if (DIFF_OPT_TST(options, IGNORE_SUBMODULES))
5900                 ignored = 1;
5901         options->flags = orig_flags;
5902         return ignored;
5903 }
5904
5905 void diff_addremove(struct diff_options *options,
5906                     int addremove, unsigned mode,
5907                     const struct object_id *oid,
5908                     int oid_valid,
5909                     const char *concatpath, unsigned dirty_submodule)
5910 {
5911         struct diff_filespec *one, *two;
5912
5913         if (S_ISGITLINK(mode) && is_submodule_ignored(concatpath, options))
5914                 return;
5915
5916         /* This may look odd, but it is a preparation for
5917          * feeding "there are unchanged files which should
5918          * not produce diffs, but when you are doing copy
5919          * detection you would need them, so here they are"
5920          * entries to the diff-core.  They will be prefixed
5921          * with something like '=' or '*' (I haven't decided
5922          * which but should not make any difference).
5923          * Feeding the same new and old to diff_change()
5924          * also has the same effect.
5925          * Before the final output happens, they are pruned after
5926          * merged into rename/copy pairs as appropriate.
5927          */
5928         if (DIFF_OPT_TST(options, REVERSE_DIFF))
5929                 addremove = (addremove == '+' ? '-' :
5930                              addremove == '-' ? '+' : addremove);
5931
5932         if (options->prefix &&
5933             strncmp(concatpath, options->prefix, options->prefix_length))
5934                 return;
5935
5936         one = alloc_filespec(concatpath);
5937         two = alloc_filespec(concatpath);
5938
5939         if (addremove != '+')
5940                 fill_filespec(one, oid, oid_valid, mode);
5941         if (addremove != '-') {
5942                 fill_filespec(two, oid, oid_valid, mode);
5943                 two->dirty_submodule = dirty_submodule;
5944         }
5945
5946         diff_queue(&diff_queued_diff, one, two);
5947         if (!DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
5948                 DIFF_OPT_SET(options, HAS_CHANGES);
5949 }
5950
5951 void diff_change(struct diff_options *options,
5952                  unsigned old_mode, unsigned new_mode,
5953                  const struct object_id *old_oid,
5954                  const struct object_id *new_oid,
5955                  int old_oid_valid, int new_oid_valid,
5956                  const char *concatpath,
5957                  unsigned old_dirty_submodule, unsigned new_dirty_submodule)
5958 {
5959         struct diff_filespec *one, *two;
5960         struct diff_filepair *p;
5961
5962         if (S_ISGITLINK(old_mode) && S_ISGITLINK(new_mode) &&
5963             is_submodule_ignored(concatpath, options))
5964                 return;
5965
5966         if (DIFF_OPT_TST(options, REVERSE_DIFF)) {
5967                 SWAP(old_mode, new_mode);
5968                 SWAP(old_oid, new_oid);
5969                 SWAP(old_oid_valid, new_oid_valid);
5970                 SWAP(old_dirty_submodule, new_dirty_submodule);
5971         }
5972
5973         if (options->prefix &&
5974             strncmp(concatpath, options->prefix, options->prefix_length))
5975                 return;
5976
5977         one = alloc_filespec(concatpath);
5978         two = alloc_filespec(concatpath);
5979         fill_filespec(one, old_oid, old_oid_valid, old_mode);
5980         fill_filespec(two, new_oid, new_oid_valid, new_mode);
5981         one->dirty_submodule = old_dirty_submodule;
5982         two->dirty_submodule = new_dirty_submodule;
5983         p = diff_queue(&diff_queued_diff, one, two);
5984
5985         if (DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
5986                 return;
5987
5988         if (DIFF_OPT_TST(options, QUICK) && options->skip_stat_unmatch &&
5989             !diff_filespec_check_stat_unmatch(p))
5990                 return;
5991
5992         DIFF_OPT_SET(options, HAS_CHANGES);
5993 }
5994
5995 struct diff_filepair *diff_unmerge(struct diff_options *options, const char *path)
5996 {
5997         struct diff_filepair *pair;
5998         struct diff_filespec *one, *two;
5999
6000         if (options->prefix &&
6001             strncmp(path, options->prefix, options->prefix_length))
6002                 return NULL;
6003
6004         one = alloc_filespec(path);
6005         two = alloc_filespec(path);
6006         pair = diff_queue(&diff_queued_diff, one, two);
6007         pair->is_unmerged = 1;
6008         return pair;
6009 }
6010
6011 static char *run_textconv(const char *pgm, struct diff_filespec *spec,
6012                 size_t *outsize)
6013 {
6014         struct diff_tempfile *temp;
6015         const char *argv[3];
6016         const char **arg = argv;
6017         struct child_process child = CHILD_PROCESS_INIT;
6018         struct strbuf buf = STRBUF_INIT;
6019         int err = 0;
6020
6021         temp = prepare_temp_file(spec->path, spec);
6022         *arg++ = pgm;
6023         *arg++ = temp->name;
6024         *arg = NULL;
6025
6026         child.use_shell = 1;
6027         child.argv = argv;
6028         child.out = -1;
6029         if (start_command(&child)) {
6030                 remove_tempfile();
6031                 return NULL;
6032         }
6033
6034         if (strbuf_read(&buf, child.out, 0) < 0)
6035                 err = error("error reading from textconv command '%s'", pgm);
6036         close(child.out);
6037
6038         if (finish_command(&child) || err) {
6039                 strbuf_release(&buf);
6040                 remove_tempfile();
6041                 return NULL;
6042         }
6043         remove_tempfile();
6044
6045         return strbuf_detach(&buf, outsize);
6046 }
6047
6048 size_t fill_textconv(struct userdiff_driver *driver,
6049                      struct diff_filespec *df,
6050                      char **outbuf)
6051 {
6052         size_t size;
6053
6054         if (!driver) {
6055                 if (!DIFF_FILE_VALID(df)) {
6056                         *outbuf = "";
6057                         return 0;
6058                 }
6059                 if (diff_populate_filespec(df, 0))
6060                         die("unable to read files to diff");
6061                 *outbuf = df->data;
6062                 return df->size;
6063         }
6064
6065         if (!driver->textconv)
6066                 die("BUG: fill_textconv called with non-textconv driver");
6067
6068         if (driver->textconv_cache && df->oid_valid) {
6069                 *outbuf = notes_cache_get(driver->textconv_cache,
6070                                           &df->oid,
6071                                           &size);
6072                 if (*outbuf)
6073                         return size;
6074         }
6075
6076         *outbuf = run_textconv(driver->textconv, df, &size);
6077         if (!*outbuf)
6078                 die("unable to read files to diff");
6079
6080         if (driver->textconv_cache && df->oid_valid) {
6081                 /* ignore errors, as we might be in a readonly repository */
6082                 notes_cache_put(driver->textconv_cache, &df->oid, *outbuf,
6083                                 size);
6084                 /*
6085                  * we could save up changes and flush them all at the end,
6086                  * but we would need an extra call after all diffing is done.
6087                  * Since generating a cache entry is the slow path anyway,
6088                  * this extra overhead probably isn't a big deal.
6089                  */
6090                 notes_cache_write(driver->textconv_cache);
6091         }
6092
6093         return size;
6094 }
6095
6096 int textconv_object(const char *path,
6097                     unsigned mode,
6098                     const struct object_id *oid,
6099                     int oid_valid,
6100                     char **buf,
6101                     unsigned long *buf_size)
6102 {
6103         struct diff_filespec *df;
6104         struct userdiff_driver *textconv;
6105
6106         df = alloc_filespec(path);
6107         fill_filespec(df, oid, oid_valid, mode);
6108         textconv = get_textconv(df);
6109         if (!textconv) {
6110                 free_filespec(df);
6111                 return 0;
6112         }
6113
6114         *buf_size = fill_textconv(textconv, df, buf);
6115         free_filespec(df);
6116         return 1;
6117 }
6118
6119 void setup_diff_pager(struct diff_options *opt)
6120 {
6121         /*
6122          * If the user asked for our exit code, then either they want --quiet
6123          * or --exit-code. We should definitely not bother with a pager in the
6124          * former case, as we will generate no output. Since we still properly
6125          * report our exit code even when a pager is run, we _could_ run a
6126          * pager with --exit-code. But since we have not done so historically,
6127          * and because it is easy to find people oneline advising "git diff
6128          * --exit-code" in hooks and other scripts, we do not do so.
6129          */
6130         if (!DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
6131             check_pager_config("diff") != 0)
6132                 setup_pager();
6133 }