userdiff.c: remove implicit dependency on the_index
[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 "object-store.h"
17 #include "userdiff.h"
18 #include "submodule-config.h"
19 #include "submodule.h"
20 #include "hashmap.h"
21 #include "ll-merge.h"
22 #include "string-list.h"
23 #include "argv-array.h"
24 #include "graph.h"
25 #include "packfile.h"
26 #include "help.h"
27
28 #ifdef NO_FAST_WORKING_DIRECTORY
29 #define FAST_WORKING_DIRECTORY 0
30 #else
31 #define FAST_WORKING_DIRECTORY 1
32 #endif
33
34 static int diff_detect_rename_default;
35 static int diff_indent_heuristic = 1;
36 static int diff_rename_limit_default = 400;
37 static int diff_suppress_blank_empty;
38 static int diff_use_color_default = -1;
39 static int diff_color_moved_default;
40 static int diff_color_moved_ws_default;
41 static int diff_context_default = 3;
42 static int diff_interhunk_context_default;
43 static const char *diff_word_regex_cfg;
44 static const char *external_diff_cmd_cfg;
45 static const char *diff_order_file_cfg;
46 int diff_auto_refresh_index = 1;
47 static int diff_mnemonic_prefix;
48 static int diff_no_prefix;
49 static int diff_stat_graph_width;
50 static int diff_dirstat_permille_default = 30;
51 static struct diff_options default_diff_options;
52 static long diff_algorithm;
53 static unsigned ws_error_highlight_default = WSEH_NEW;
54
55 static char diff_colors[][COLOR_MAXLEN] = {
56         GIT_COLOR_RESET,
57         GIT_COLOR_NORMAL,       /* CONTEXT */
58         GIT_COLOR_BOLD,         /* METAINFO */
59         GIT_COLOR_CYAN,         /* FRAGINFO */
60         GIT_COLOR_RED,          /* OLD */
61         GIT_COLOR_GREEN,        /* NEW */
62         GIT_COLOR_YELLOW,       /* COMMIT */
63         GIT_COLOR_BG_RED,       /* WHITESPACE */
64         GIT_COLOR_NORMAL,       /* FUNCINFO */
65         GIT_COLOR_BOLD_MAGENTA, /* OLD_MOVED */
66         GIT_COLOR_BOLD_BLUE,    /* OLD_MOVED ALTERNATIVE */
67         GIT_COLOR_FAINT,        /* OLD_MOVED_DIM */
68         GIT_COLOR_FAINT_ITALIC, /* OLD_MOVED_ALTERNATIVE_DIM */
69         GIT_COLOR_BOLD_CYAN,    /* NEW_MOVED */
70         GIT_COLOR_BOLD_YELLOW,  /* NEW_MOVED ALTERNATIVE */
71         GIT_COLOR_FAINT,        /* NEW_MOVED_DIM */
72         GIT_COLOR_FAINT_ITALIC, /* NEW_MOVED_ALTERNATIVE_DIM */
73         GIT_COLOR_FAINT,        /* CONTEXT_DIM */
74         GIT_COLOR_FAINT_RED,    /* OLD_DIM */
75         GIT_COLOR_FAINT_GREEN,  /* NEW_DIM */
76         GIT_COLOR_BOLD,         /* CONTEXT_BOLD */
77         GIT_COLOR_BOLD_RED,     /* OLD_BOLD */
78         GIT_COLOR_BOLD_GREEN,   /* NEW_BOLD */
79 };
80
81 static const char *color_diff_slots[] = {
82         [DIFF_CONTEXT]                = "context",
83         [DIFF_METAINFO]               = "meta",
84         [DIFF_FRAGINFO]               = "frag",
85         [DIFF_FILE_OLD]               = "old",
86         [DIFF_FILE_NEW]               = "new",
87         [DIFF_COMMIT]                 = "commit",
88         [DIFF_WHITESPACE]             = "whitespace",
89         [DIFF_FUNCINFO]               = "func",
90         [DIFF_FILE_OLD_MOVED]         = "oldMoved",
91         [DIFF_FILE_OLD_MOVED_ALT]     = "oldMovedAlternative",
92         [DIFF_FILE_OLD_MOVED_DIM]     = "oldMovedDimmed",
93         [DIFF_FILE_OLD_MOVED_ALT_DIM] = "oldMovedAlternativeDimmed",
94         [DIFF_FILE_NEW_MOVED]         = "newMoved",
95         [DIFF_FILE_NEW_MOVED_ALT]     = "newMovedAlternative",
96         [DIFF_FILE_NEW_MOVED_DIM]     = "newMovedDimmed",
97         [DIFF_FILE_NEW_MOVED_ALT_DIM] = "newMovedAlternativeDimmed",
98         [DIFF_CONTEXT_DIM]            = "contextDimmed",
99         [DIFF_FILE_OLD_DIM]           = "oldDimmed",
100         [DIFF_FILE_NEW_DIM]           = "newDimmed",
101         [DIFF_CONTEXT_BOLD]           = "contextBold",
102         [DIFF_FILE_OLD_BOLD]          = "oldBold",
103         [DIFF_FILE_NEW_BOLD]          = "newBold",
104 };
105
106 static NORETURN void die_want_option(const char *option_name)
107 {
108         die(_("option '%s' requires a value"), option_name);
109 }
110
111 define_list_config_array_extra(color_diff_slots, {"plain"});
112
113 static int parse_diff_color_slot(const char *var)
114 {
115         if (!strcasecmp(var, "plain"))
116                 return DIFF_CONTEXT;
117         return LOOKUP_CONFIG(color_diff_slots, var);
118 }
119
120 static int parse_dirstat_params(struct diff_options *options, const char *params_string,
121                                 struct strbuf *errmsg)
122 {
123         char *params_copy = xstrdup(params_string);
124         struct string_list params = STRING_LIST_INIT_NODUP;
125         int ret = 0;
126         int i;
127
128         if (*params_copy)
129                 string_list_split_in_place(&params, params_copy, ',', -1);
130         for (i = 0; i < params.nr; i++) {
131                 const char *p = params.items[i].string;
132                 if (!strcmp(p, "changes")) {
133                         options->flags.dirstat_by_line = 0;
134                         options->flags.dirstat_by_file = 0;
135                 } else if (!strcmp(p, "lines")) {
136                         options->flags.dirstat_by_line = 1;
137                         options->flags.dirstat_by_file = 0;
138                 } else if (!strcmp(p, "files")) {
139                         options->flags.dirstat_by_line = 0;
140                         options->flags.dirstat_by_file = 1;
141                 } else if (!strcmp(p, "noncumulative")) {
142                         options->flags.dirstat_cumulative = 0;
143                 } else if (!strcmp(p, "cumulative")) {
144                         options->flags.dirstat_cumulative = 1;
145                 } else if (isdigit(*p)) {
146                         char *end;
147                         int permille = strtoul(p, &end, 10) * 10;
148                         if (*end == '.' && isdigit(*++end)) {
149                                 /* only use first digit */
150                                 permille += *end - '0';
151                                 /* .. and ignore any further digits */
152                                 while (isdigit(*++end))
153                                         ; /* nothing */
154                         }
155                         if (!*end)
156                                 options->dirstat_permille = permille;
157                         else {
158                                 strbuf_addf(errmsg, _("  Failed to parse dirstat cut-off percentage '%s'\n"),
159                                             p);
160                                 ret++;
161                         }
162                 } else {
163                         strbuf_addf(errmsg, _("  Unknown dirstat parameter '%s'\n"), p);
164                         ret++;
165                 }
166
167         }
168         string_list_clear(&params, 0);
169         free(params_copy);
170         return ret;
171 }
172
173 static int parse_submodule_params(struct diff_options *options, const char *value)
174 {
175         if (!strcmp(value, "log"))
176                 options->submodule_format = DIFF_SUBMODULE_LOG;
177         else if (!strcmp(value, "short"))
178                 options->submodule_format = DIFF_SUBMODULE_SHORT;
179         else if (!strcmp(value, "diff"))
180                 options->submodule_format = DIFF_SUBMODULE_INLINE_DIFF;
181         else
182                 return -1;
183         return 0;
184 }
185
186 int git_config_rename(const char *var, const char *value)
187 {
188         if (!value)
189                 return DIFF_DETECT_RENAME;
190         if (!strcasecmp(value, "copies") || !strcasecmp(value, "copy"))
191                 return  DIFF_DETECT_COPY;
192         return git_config_bool(var,value) ? DIFF_DETECT_RENAME : 0;
193 }
194
195 long parse_algorithm_value(const char *value)
196 {
197         if (!value)
198                 return -1;
199         else if (!strcasecmp(value, "myers") || !strcasecmp(value, "default"))
200                 return 0;
201         else if (!strcasecmp(value, "minimal"))
202                 return XDF_NEED_MINIMAL;
203         else if (!strcasecmp(value, "patience"))
204                 return XDF_PATIENCE_DIFF;
205         else if (!strcasecmp(value, "histogram"))
206                 return XDF_HISTOGRAM_DIFF;
207         return -1;
208 }
209
210 static int parse_one_token(const char **arg, const char *token)
211 {
212         const char *rest;
213         if (skip_prefix(*arg, token, &rest) && (!*rest || *rest == ',')) {
214                 *arg = rest;
215                 return 1;
216         }
217         return 0;
218 }
219
220 static int parse_ws_error_highlight(const char *arg)
221 {
222         const char *orig_arg = arg;
223         unsigned val = 0;
224
225         while (*arg) {
226                 if (parse_one_token(&arg, "none"))
227                         val = 0;
228                 else if (parse_one_token(&arg, "default"))
229                         val = WSEH_NEW;
230                 else if (parse_one_token(&arg, "all"))
231                         val = WSEH_NEW | WSEH_OLD | WSEH_CONTEXT;
232                 else if (parse_one_token(&arg, "new"))
233                         val |= WSEH_NEW;
234                 else if (parse_one_token(&arg, "old"))
235                         val |= WSEH_OLD;
236                 else if (parse_one_token(&arg, "context"))
237                         val |= WSEH_CONTEXT;
238                 else {
239                         return -1 - (int)(arg - orig_arg);
240                 }
241                 if (*arg)
242                         arg++;
243         }
244         return val;
245 }
246
247 /*
248  * These are to give UI layer defaults.
249  * The core-level commands such as git-diff-files should
250  * never be affected by the setting of diff.renames
251  * the user happens to have in the configuration file.
252  */
253 void init_diff_ui_defaults(void)
254 {
255         diff_detect_rename_default = DIFF_DETECT_RENAME;
256 }
257
258 int git_diff_heuristic_config(const char *var, const char *value, void *cb)
259 {
260         if (!strcmp(var, "diff.indentheuristic"))
261                 diff_indent_heuristic = git_config_bool(var, value);
262         return 0;
263 }
264
265 static int parse_color_moved(const char *arg)
266 {
267         switch (git_parse_maybe_bool(arg)) {
268         case 0:
269                 return COLOR_MOVED_NO;
270         case 1:
271                 return COLOR_MOVED_DEFAULT;
272         default:
273                 break;
274         }
275
276         if (!strcmp(arg, "no"))
277                 return COLOR_MOVED_NO;
278         else if (!strcmp(arg, "plain"))
279                 return COLOR_MOVED_PLAIN;
280         else if (!strcmp(arg, "blocks"))
281                 return COLOR_MOVED_BLOCKS;
282         else if (!strcmp(arg, "zebra"))
283                 return COLOR_MOVED_ZEBRA;
284         else if (!strcmp(arg, "default"))
285                 return COLOR_MOVED_DEFAULT;
286         else if (!strcmp(arg, "dimmed-zebra"))
287                 return COLOR_MOVED_ZEBRA_DIM;
288         else if (!strcmp(arg, "dimmed_zebra"))
289                 return COLOR_MOVED_ZEBRA_DIM;
290         else
291                 return error(_("color moved setting must be one of 'no', 'default', 'blocks', 'zebra', 'dimmed-zebra', 'plain'"));
292 }
293
294 static int parse_color_moved_ws(const char *arg)
295 {
296         int ret = 0;
297         struct string_list l = STRING_LIST_INIT_DUP;
298         struct string_list_item *i;
299
300         string_list_split(&l, arg, ',', -1);
301
302         for_each_string_list_item(i, &l) {
303                 struct strbuf sb = STRBUF_INIT;
304                 strbuf_addstr(&sb, i->string);
305                 strbuf_trim(&sb);
306
307                 if (!strcmp(sb.buf, "ignore-space-change"))
308                         ret |= XDF_IGNORE_WHITESPACE_CHANGE;
309                 else if (!strcmp(sb.buf, "ignore-space-at-eol"))
310                         ret |= XDF_IGNORE_WHITESPACE_AT_EOL;
311                 else if (!strcmp(sb.buf, "ignore-all-space"))
312                         ret |= XDF_IGNORE_WHITESPACE;
313                 else if (!strcmp(sb.buf, "allow-indentation-change"))
314                         ret |= COLOR_MOVED_WS_ALLOW_INDENTATION_CHANGE;
315                 else
316                         error(_("ignoring unknown color-moved-ws mode '%s'"), sb.buf);
317
318                 strbuf_release(&sb);
319         }
320
321         if ((ret & COLOR_MOVED_WS_ALLOW_INDENTATION_CHANGE) &&
322             (ret & XDF_WHITESPACE_FLAGS))
323                 die(_("color-moved-ws: allow-indentation-change cannot be combined with other white space modes"));
324
325         string_list_clear(&l, 0);
326
327         return ret;
328 }
329
330 int git_diff_ui_config(const char *var, const char *value, void *cb)
331 {
332         if (!strcmp(var, "diff.color") || !strcmp(var, "color.diff")) {
333                 diff_use_color_default = git_config_colorbool(var, value);
334                 return 0;
335         }
336         if (!strcmp(var, "diff.colormoved")) {
337                 int cm = parse_color_moved(value);
338                 if (cm < 0)
339                         return -1;
340                 diff_color_moved_default = cm;
341                 return 0;
342         }
343         if (!strcmp(var, "diff.colormovedws")) {
344                 int cm = parse_color_moved_ws(value);
345                 if (cm < 0)
346                         return -1;
347                 diff_color_moved_ws_default = cm;
348                 return 0;
349         }
350         if (!strcmp(var, "diff.context")) {
351                 diff_context_default = git_config_int(var, value);
352                 if (diff_context_default < 0)
353                         return -1;
354                 return 0;
355         }
356         if (!strcmp(var, "diff.interhunkcontext")) {
357                 diff_interhunk_context_default = git_config_int(var, value);
358                 if (diff_interhunk_context_default < 0)
359                         return -1;
360                 return 0;
361         }
362         if (!strcmp(var, "diff.renames")) {
363                 diff_detect_rename_default = git_config_rename(var, value);
364                 return 0;
365         }
366         if (!strcmp(var, "diff.autorefreshindex")) {
367                 diff_auto_refresh_index = git_config_bool(var, value);
368                 return 0;
369         }
370         if (!strcmp(var, "diff.mnemonicprefix")) {
371                 diff_mnemonic_prefix = git_config_bool(var, value);
372                 return 0;
373         }
374         if (!strcmp(var, "diff.noprefix")) {
375                 diff_no_prefix = git_config_bool(var, value);
376                 return 0;
377         }
378         if (!strcmp(var, "diff.statgraphwidth")) {
379                 diff_stat_graph_width = git_config_int(var, value);
380                 return 0;
381         }
382         if (!strcmp(var, "diff.external"))
383                 return git_config_string(&external_diff_cmd_cfg, var, value);
384         if (!strcmp(var, "diff.wordregex"))
385                 return git_config_string(&diff_word_regex_cfg, var, value);
386         if (!strcmp(var, "diff.orderfile"))
387                 return git_config_pathname(&diff_order_file_cfg, var, value);
388
389         if (!strcmp(var, "diff.ignoresubmodules"))
390                 handle_ignore_submodules_arg(&default_diff_options, value);
391
392         if (!strcmp(var, "diff.submodule")) {
393                 if (parse_submodule_params(&default_diff_options, value))
394                         warning(_("Unknown value for 'diff.submodule' config variable: '%s'"),
395                                 value);
396                 return 0;
397         }
398
399         if (!strcmp(var, "diff.algorithm")) {
400                 diff_algorithm = parse_algorithm_value(value);
401                 if (diff_algorithm < 0)
402                         return -1;
403                 return 0;
404         }
405
406         if (!strcmp(var, "diff.wserrorhighlight")) {
407                 int val = parse_ws_error_highlight(value);
408                 if (val < 0)
409                         return -1;
410                 ws_error_highlight_default = val;
411                 return 0;
412         }
413
414         if (git_color_config(var, value, cb) < 0)
415                 return -1;
416
417         return git_diff_basic_config(var, value, cb);
418 }
419
420 int git_diff_basic_config(const char *var, const char *value, void *cb)
421 {
422         const char *name;
423
424         if (!strcmp(var, "diff.renamelimit")) {
425                 diff_rename_limit_default = git_config_int(var, value);
426                 return 0;
427         }
428
429         if (userdiff_config(var, value) < 0)
430                 return -1;
431
432         if (skip_prefix(var, "diff.color.", &name) ||
433             skip_prefix(var, "color.diff.", &name)) {
434                 int slot = parse_diff_color_slot(name);
435                 if (slot < 0)
436                         return 0;
437                 if (!value)
438                         return config_error_nonbool(var);
439                 return color_parse(value, diff_colors[slot]);
440         }
441
442         /* like GNU diff's --suppress-blank-empty option  */
443         if (!strcmp(var, "diff.suppressblankempty") ||
444                         /* for backwards compatibility */
445                         !strcmp(var, "diff.suppress-blank-empty")) {
446                 diff_suppress_blank_empty = git_config_bool(var, value);
447                 return 0;
448         }
449
450         if (!strcmp(var, "diff.dirstat")) {
451                 struct strbuf errmsg = STRBUF_INIT;
452                 default_diff_options.dirstat_permille = diff_dirstat_permille_default;
453                 if (parse_dirstat_params(&default_diff_options, value, &errmsg))
454                         warning(_("Found errors in 'diff.dirstat' config variable:\n%s"),
455                                 errmsg.buf);
456                 strbuf_release(&errmsg);
457                 diff_dirstat_permille_default = default_diff_options.dirstat_permille;
458                 return 0;
459         }
460
461         if (git_diff_heuristic_config(var, value, cb) < 0)
462                 return -1;
463
464         return git_default_config(var, value, cb);
465 }
466
467 static char *quote_two(const char *one, const char *two)
468 {
469         int need_one = quote_c_style(one, NULL, NULL, 1);
470         int need_two = quote_c_style(two, NULL, NULL, 1);
471         struct strbuf res = STRBUF_INIT;
472
473         if (need_one + need_two) {
474                 strbuf_addch(&res, '"');
475                 quote_c_style(one, &res, NULL, 1);
476                 quote_c_style(two, &res, NULL, 1);
477                 strbuf_addch(&res, '"');
478         } else {
479                 strbuf_addstr(&res, one);
480                 strbuf_addstr(&res, two);
481         }
482         return strbuf_detach(&res, NULL);
483 }
484
485 static const char *external_diff(void)
486 {
487         static const char *external_diff_cmd = NULL;
488         static int done_preparing = 0;
489
490         if (done_preparing)
491                 return external_diff_cmd;
492         external_diff_cmd = getenv("GIT_EXTERNAL_DIFF");
493         if (!external_diff_cmd)
494                 external_diff_cmd = external_diff_cmd_cfg;
495         done_preparing = 1;
496         return external_diff_cmd;
497 }
498
499 /*
500  * Keep track of files used for diffing. Sometimes such an entry
501  * refers to a temporary file, sometimes to an existing file, and
502  * sometimes to "/dev/null".
503  */
504 static struct diff_tempfile {
505         /*
506          * filename external diff should read from, or NULL if this
507          * entry is currently not in use:
508          */
509         const char *name;
510
511         char hex[GIT_MAX_HEXSZ + 1];
512         char mode[10];
513
514         /*
515          * If this diff_tempfile instance refers to a temporary file,
516          * this tempfile object is used to manage its lifetime.
517          */
518         struct tempfile *tempfile;
519 } diff_temp[2];
520
521 struct emit_callback {
522         int color_diff;
523         unsigned ws_rule;
524         int blank_at_eof_in_preimage;
525         int blank_at_eof_in_postimage;
526         int lno_in_preimage;
527         int lno_in_postimage;
528         const char **label_path;
529         struct diff_words_data *diff_words;
530         struct diff_options *opt;
531         struct strbuf *header;
532 };
533
534 static int count_lines(const char *data, int size)
535 {
536         int count, ch, completely_empty = 1, nl_just_seen = 0;
537         count = 0;
538         while (0 < size--) {
539                 ch = *data++;
540                 if (ch == '\n') {
541                         count++;
542                         nl_just_seen = 1;
543                         completely_empty = 0;
544                 }
545                 else {
546                         nl_just_seen = 0;
547                         completely_empty = 0;
548                 }
549         }
550         if (completely_empty)
551                 return 0;
552         if (!nl_just_seen)
553                 count++; /* no trailing newline */
554         return count;
555 }
556
557 static int fill_mmfile(struct repository *r, mmfile_t *mf,
558                        struct diff_filespec *one)
559 {
560         if (!DIFF_FILE_VALID(one)) {
561                 mf->ptr = (char *)""; /* does not matter */
562                 mf->size = 0;
563                 return 0;
564         }
565         else if (diff_populate_filespec(r, one, 0))
566                 return -1;
567
568         mf->ptr = one->data;
569         mf->size = one->size;
570         return 0;
571 }
572
573 /* like fill_mmfile, but only for size, so we can avoid retrieving blob */
574 static unsigned long diff_filespec_size(struct repository *r,
575                                         struct diff_filespec *one)
576 {
577         if (!DIFF_FILE_VALID(one))
578                 return 0;
579         diff_populate_filespec(r, one, CHECK_SIZE_ONLY);
580         return one->size;
581 }
582
583 static int count_trailing_blank(mmfile_t *mf, unsigned ws_rule)
584 {
585         char *ptr = mf->ptr;
586         long size = mf->size;
587         int cnt = 0;
588
589         if (!size)
590                 return cnt;
591         ptr += size - 1; /* pointing at the very end */
592         if (*ptr != '\n')
593                 ; /* incomplete line */
594         else
595                 ptr--; /* skip the last LF */
596         while (mf->ptr < ptr) {
597                 char *prev_eol;
598                 for (prev_eol = ptr; mf->ptr <= prev_eol; prev_eol--)
599                         if (*prev_eol == '\n')
600                                 break;
601                 if (!ws_blank_line(prev_eol + 1, ptr - prev_eol, ws_rule))
602                         break;
603                 cnt++;
604                 ptr = prev_eol - 1;
605         }
606         return cnt;
607 }
608
609 static void check_blank_at_eof(mmfile_t *mf1, mmfile_t *mf2,
610                                struct emit_callback *ecbdata)
611 {
612         int l1, l2, at;
613         unsigned ws_rule = ecbdata->ws_rule;
614         l1 = count_trailing_blank(mf1, ws_rule);
615         l2 = count_trailing_blank(mf2, ws_rule);
616         if (l2 <= l1) {
617                 ecbdata->blank_at_eof_in_preimage = 0;
618                 ecbdata->blank_at_eof_in_postimage = 0;
619                 return;
620         }
621         at = count_lines(mf1->ptr, mf1->size);
622         ecbdata->blank_at_eof_in_preimage = (at - l1) + 1;
623
624         at = count_lines(mf2->ptr, mf2->size);
625         ecbdata->blank_at_eof_in_postimage = (at - l2) + 1;
626 }
627
628 static void emit_line_0(struct diff_options *o,
629                         const char *set, unsigned reverse, const char *reset,
630                         int first, const char *line, int len)
631 {
632         int has_trailing_newline, has_trailing_carriage_return;
633         int nofirst;
634         FILE *file = o->file;
635
636         if (first)
637                 fputs(diff_line_prefix(o), file);
638         else if (!len)
639                 return;
640
641         if (len == 0) {
642                 has_trailing_newline = (first == '\n');
643                 has_trailing_carriage_return = (!has_trailing_newline &&
644                                                 (first == '\r'));
645                 nofirst = has_trailing_newline || has_trailing_carriage_return;
646         } else {
647                 has_trailing_newline = (len > 0 && line[len-1] == '\n');
648                 if (has_trailing_newline)
649                         len--;
650                 has_trailing_carriage_return = (len > 0 && line[len-1] == '\r');
651                 if (has_trailing_carriage_return)
652                         len--;
653                 nofirst = 0;
654         }
655
656         if (len || !nofirst) {
657                 if (reverse && want_color(o->use_color))
658                         fputs(GIT_COLOR_REVERSE, file);
659                 fputs(set, file);
660                 if (first && !nofirst)
661                         fputc(first, file);
662                 fwrite(line, len, 1, file);
663                 fputs(reset, file);
664         }
665         if (has_trailing_carriage_return)
666                 fputc('\r', file);
667         if (has_trailing_newline)
668                 fputc('\n', file);
669 }
670
671 static void emit_line(struct diff_options *o, const char *set, const char *reset,
672                       const char *line, int len)
673 {
674         emit_line_0(o, set, 0, reset, line[0], line+1, len-1);
675 }
676
677 enum diff_symbol {
678         DIFF_SYMBOL_BINARY_DIFF_HEADER,
679         DIFF_SYMBOL_BINARY_DIFF_HEADER_DELTA,
680         DIFF_SYMBOL_BINARY_DIFF_HEADER_LITERAL,
681         DIFF_SYMBOL_BINARY_DIFF_BODY,
682         DIFF_SYMBOL_BINARY_DIFF_FOOTER,
683         DIFF_SYMBOL_STATS_SUMMARY_NO_FILES,
684         DIFF_SYMBOL_STATS_SUMMARY_ABBREV,
685         DIFF_SYMBOL_STATS_SUMMARY_INSERTS_DELETES,
686         DIFF_SYMBOL_STATS_LINE,
687         DIFF_SYMBOL_WORD_DIFF,
688         DIFF_SYMBOL_STAT_SEP,
689         DIFF_SYMBOL_SUMMARY,
690         DIFF_SYMBOL_SUBMODULE_ADD,
691         DIFF_SYMBOL_SUBMODULE_DEL,
692         DIFF_SYMBOL_SUBMODULE_UNTRACKED,
693         DIFF_SYMBOL_SUBMODULE_MODIFIED,
694         DIFF_SYMBOL_SUBMODULE_HEADER,
695         DIFF_SYMBOL_SUBMODULE_ERROR,
696         DIFF_SYMBOL_SUBMODULE_PIPETHROUGH,
697         DIFF_SYMBOL_REWRITE_DIFF,
698         DIFF_SYMBOL_BINARY_FILES,
699         DIFF_SYMBOL_HEADER,
700         DIFF_SYMBOL_FILEPAIR_PLUS,
701         DIFF_SYMBOL_FILEPAIR_MINUS,
702         DIFF_SYMBOL_WORDS_PORCELAIN,
703         DIFF_SYMBOL_WORDS,
704         DIFF_SYMBOL_CONTEXT,
705         DIFF_SYMBOL_CONTEXT_INCOMPLETE,
706         DIFF_SYMBOL_PLUS,
707         DIFF_SYMBOL_MINUS,
708         DIFF_SYMBOL_NO_LF_EOF,
709         DIFF_SYMBOL_CONTEXT_FRAGINFO,
710         DIFF_SYMBOL_CONTEXT_MARKER,
711         DIFF_SYMBOL_SEPARATOR
712 };
713 /*
714  * Flags for content lines:
715  * 0..12 are whitespace rules
716  * 13-15 are WSEH_NEW | WSEH_OLD | WSEH_CONTEXT
717  * 16 is marking if the line is blank at EOF
718  */
719 #define DIFF_SYMBOL_CONTENT_BLANK_LINE_EOF      (1<<16)
720 #define DIFF_SYMBOL_MOVED_LINE                  (1<<17)
721 #define DIFF_SYMBOL_MOVED_LINE_ALT              (1<<18)
722 #define DIFF_SYMBOL_MOVED_LINE_UNINTERESTING    (1<<19)
723 #define DIFF_SYMBOL_CONTENT_WS_MASK (WSEH_NEW | WSEH_OLD | WSEH_CONTEXT | WS_RULE_MASK)
724
725 /*
726  * This struct is used when we need to buffer the output of the diff output.
727  *
728  * NEEDSWORK: Instead of storing a copy of the line, add an offset pointer
729  * into the pre/post image file. This pointer could be a union with the
730  * line pointer. By storing an offset into the file instead of the literal line,
731  * we can decrease the memory footprint for the buffered output. At first we
732  * may want to only have indirection for the content lines, but we could also
733  * enhance the state for emitting prefabricated lines, e.g. the similarity
734  * score line or hunk/file headers would only need to store a number or path
735  * and then the output can be constructed later on depending on state.
736  */
737 struct emitted_diff_symbol {
738         const char *line;
739         int len;
740         int flags;
741         enum diff_symbol s;
742 };
743 #define EMITTED_DIFF_SYMBOL_INIT {NULL}
744
745 struct emitted_diff_symbols {
746         struct emitted_diff_symbol *buf;
747         int nr, alloc;
748 };
749 #define EMITTED_DIFF_SYMBOLS_INIT {NULL, 0, 0}
750
751 static void append_emitted_diff_symbol(struct diff_options *o,
752                                        struct emitted_diff_symbol *e)
753 {
754         struct emitted_diff_symbol *f;
755
756         ALLOC_GROW(o->emitted_symbols->buf,
757                    o->emitted_symbols->nr + 1,
758                    o->emitted_symbols->alloc);
759         f = &o->emitted_symbols->buf[o->emitted_symbols->nr++];
760
761         memcpy(f, e, sizeof(struct emitted_diff_symbol));
762         f->line = e->line ? xmemdupz(e->line, e->len) : NULL;
763 }
764
765 struct moved_entry {
766         struct hashmap_entry ent;
767         const struct emitted_diff_symbol *es;
768         struct moved_entry *next_line;
769         struct ws_delta *wsd;
770 };
771
772 /**
773  * The struct ws_delta holds white space differences between moved lines, i.e.
774  * between '+' and '-' lines that have been detected to be a move.
775  * The string contains the difference in leading white spaces, before the
776  * rest of the line is compared using the white space config for move
777  * coloring. The current_longer indicates if the first string in the
778  * comparision is longer than the second.
779  */
780 struct ws_delta {
781         char *string;
782         unsigned int current_longer : 1;
783 };
784 #define WS_DELTA_INIT { NULL, 0 }
785
786 static int compute_ws_delta(const struct emitted_diff_symbol *a,
787                              const struct emitted_diff_symbol *b,
788                              struct ws_delta *out)
789 {
790         const struct emitted_diff_symbol *longer =  a->len > b->len ? a : b;
791         const struct emitted_diff_symbol *shorter = a->len > b->len ? b : a;
792         int d = longer->len - shorter->len;
793
794         out->string = xmemdupz(longer->line, d);
795         out->current_longer = (a == longer);
796
797         return !strncmp(longer->line + d, shorter->line, shorter->len);
798 }
799
800 static int cmp_in_block_with_wsd(const struct diff_options *o,
801                                  const struct moved_entry *cur,
802                                  const struct moved_entry *match,
803                                  struct moved_entry *pmb,
804                                  int n)
805 {
806         struct emitted_diff_symbol *l = &o->emitted_symbols->buf[n];
807         int al = cur->es->len, cl = l->len;
808         const char *a = cur->es->line,
809                    *b = match->es->line,
810                    *c = l->line;
811
812         int wslen;
813
814         /*
815          * We need to check if 'cur' is equal to 'match'.
816          * As those are from the same (+/-) side, we do not need to adjust for
817          * indent changes. However these were found using fuzzy matching
818          * so we do have to check if they are equal.
819          */
820         if (strcmp(a, b))
821                 return 1;
822
823         if (!pmb->wsd)
824                 /*
825                  * No white space delta was carried forward? This can happen
826                  * when we exit early in this function and do not carry
827                  * forward ws.
828                  */
829                 return 1;
830
831         /*
832          * The indent changes of the block are known and carried forward in
833          * pmb->wsd; however we need to check if the indent changes of the
834          * current line are still the same as before.
835          *
836          * To do so we need to compare 'l' to 'cur', adjusting the
837          * one of them for the white spaces, depending which was longer.
838          */
839
840         wslen = strlen(pmb->wsd->string);
841         if (pmb->wsd->current_longer) {
842                 c += wslen;
843                 cl -= wslen;
844         } else {
845                 a += wslen;
846                 al -= wslen;
847         }
848
849         if (strcmp(a, c))
850                 return 1;
851
852         return 0;
853 }
854
855 static int moved_entry_cmp(const void *hashmap_cmp_fn_data,
856                            const void *entry,
857                            const void *entry_or_key,
858                            const void *keydata)
859 {
860         const struct diff_options *diffopt = hashmap_cmp_fn_data;
861         const struct moved_entry *a = entry;
862         const struct moved_entry *b = entry_or_key;
863         unsigned flags = diffopt->color_moved_ws_handling
864                          & XDF_WHITESPACE_FLAGS;
865
866         if (diffopt->color_moved_ws_handling &
867             COLOR_MOVED_WS_ALLOW_INDENTATION_CHANGE)
868                 /*
869                  * As there is not specific white space config given,
870                  * we'd need to check for a new block, so ignore all
871                  * white space. The setup of the white space
872                  * configuration for the next block is done else where
873                  */
874                 flags |= XDF_IGNORE_WHITESPACE;
875
876         return !xdiff_compare_lines(a->es->line, a->es->len,
877                                     b->es->line, b->es->len,
878                                     flags);
879 }
880
881 static struct moved_entry *prepare_entry(struct diff_options *o,
882                                          int line_no)
883 {
884         struct moved_entry *ret = xmalloc(sizeof(*ret));
885         struct emitted_diff_symbol *l = &o->emitted_symbols->buf[line_no];
886         unsigned flags = o->color_moved_ws_handling & XDF_WHITESPACE_FLAGS;
887
888         ret->ent.hash = xdiff_hash_string(l->line, l->len, flags);
889         ret->es = l;
890         ret->next_line = NULL;
891         ret->wsd = NULL;
892
893         return ret;
894 }
895
896 static void add_lines_to_move_detection(struct diff_options *o,
897                                         struct hashmap *add_lines,
898                                         struct hashmap *del_lines)
899 {
900         struct moved_entry *prev_line = NULL;
901
902         int n;
903         for (n = 0; n < o->emitted_symbols->nr; n++) {
904                 struct hashmap *hm;
905                 struct moved_entry *key;
906
907                 switch (o->emitted_symbols->buf[n].s) {
908                 case DIFF_SYMBOL_PLUS:
909                         hm = add_lines;
910                         break;
911                 case DIFF_SYMBOL_MINUS:
912                         hm = del_lines;
913                         break;
914                 default:
915                         prev_line = NULL;
916                         continue;
917                 }
918
919                 key = prepare_entry(o, n);
920                 if (prev_line && prev_line->es->s == o->emitted_symbols->buf[n].s)
921                         prev_line->next_line = key;
922
923                 hashmap_add(hm, key);
924                 prev_line = key;
925         }
926 }
927
928 static void pmb_advance_or_null(struct diff_options *o,
929                                 struct moved_entry *match,
930                                 struct hashmap *hm,
931                                 struct moved_entry **pmb,
932                                 int pmb_nr)
933 {
934         int i;
935         for (i = 0; i < pmb_nr; i++) {
936                 struct moved_entry *prev = pmb[i];
937                 struct moved_entry *cur = (prev && prev->next_line) ?
938                                 prev->next_line : NULL;
939                 if (cur && !hm->cmpfn(o, cur, match, NULL)) {
940                         pmb[i] = cur;
941                 } else {
942                         pmb[i] = NULL;
943                 }
944         }
945 }
946
947 static void pmb_advance_or_null_multi_match(struct diff_options *o,
948                                             struct moved_entry *match,
949                                             struct hashmap *hm,
950                                             struct moved_entry **pmb,
951                                             int pmb_nr, int n)
952 {
953         int i;
954         char *got_match = xcalloc(1, pmb_nr);
955
956         for (; match; match = hashmap_get_next(hm, match)) {
957                 for (i = 0; i < pmb_nr; i++) {
958                         struct moved_entry *prev = pmb[i];
959                         struct moved_entry *cur = (prev && prev->next_line) ?
960                                         prev->next_line : NULL;
961                         if (!cur)
962                                 continue;
963                         if (!cmp_in_block_with_wsd(o, cur, match, pmb[i], n))
964                                 got_match[i] |= 1;
965                 }
966         }
967
968         for (i = 0; i < pmb_nr; i++) {
969                 if (got_match[i]) {
970                         /* Carry the white space delta forward */
971                         pmb[i]->next_line->wsd = pmb[i]->wsd;
972                         pmb[i] = pmb[i]->next_line;
973                 } else
974                         pmb[i] = NULL;
975         }
976 }
977
978 static int shrink_potential_moved_blocks(struct moved_entry **pmb,
979                                          int pmb_nr)
980 {
981         int lp, rp;
982
983         /* Shrink the set of potential block to the remaining running */
984         for (lp = 0, rp = pmb_nr - 1; lp <= rp;) {
985                 while (lp < pmb_nr && pmb[lp])
986                         lp++;
987                 /* lp points at the first NULL now */
988
989                 while (rp > -1 && !pmb[rp])
990                         rp--;
991                 /* rp points at the last non-NULL */
992
993                 if (lp < pmb_nr && rp > -1 && lp < rp) {
994                         pmb[lp] = pmb[rp];
995                         if (pmb[rp]->wsd) {
996                                 free(pmb[rp]->wsd->string);
997                                 FREE_AND_NULL(pmb[rp]->wsd);
998                         }
999                         pmb[rp] = NULL;
1000                         rp--;
1001                         lp++;
1002                 }
1003         }
1004
1005         /* Remember the number of running sets */
1006         return rp + 1;
1007 }
1008
1009 /*
1010  * If o->color_moved is COLOR_MOVED_PLAIN, this function does nothing.
1011  *
1012  * Otherwise, if the last block has fewer alphanumeric characters than
1013  * COLOR_MOVED_MIN_ALNUM_COUNT, unset DIFF_SYMBOL_MOVED_LINE on all lines in
1014  * that block.
1015  *
1016  * The last block consists of the (n - block_length)'th line up to but not
1017  * including the nth line.
1018  *
1019  * NEEDSWORK: This uses the same heuristic as blame_entry_score() in blame.c.
1020  * Think of a way to unify them.
1021  */
1022 static void adjust_last_block(struct diff_options *o, int n, int block_length)
1023 {
1024         int i, alnum_count = 0;
1025         if (o->color_moved == COLOR_MOVED_PLAIN)
1026                 return;
1027         for (i = 1; i < block_length + 1; i++) {
1028                 const char *c = o->emitted_symbols->buf[n - i].line;
1029                 for (; *c; c++) {
1030                         if (!isalnum(*c))
1031                                 continue;
1032                         alnum_count++;
1033                         if (alnum_count >= COLOR_MOVED_MIN_ALNUM_COUNT)
1034                                 return;
1035                 }
1036         }
1037         for (i = 1; i < block_length + 1; i++)
1038                 o->emitted_symbols->buf[n - i].flags &= ~DIFF_SYMBOL_MOVED_LINE;
1039 }
1040
1041 /* Find blocks of moved code, delegate actual coloring decision to helper */
1042 static void mark_color_as_moved(struct diff_options *o,
1043                                 struct hashmap *add_lines,
1044                                 struct hashmap *del_lines)
1045 {
1046         struct moved_entry **pmb = NULL; /* potentially moved blocks */
1047         int pmb_nr = 0, pmb_alloc = 0;
1048         int n, flipped_block = 1, block_length = 0;
1049
1050
1051         for (n = 0; n < o->emitted_symbols->nr; n++) {
1052                 struct hashmap *hm = NULL;
1053                 struct moved_entry *key;
1054                 struct moved_entry *match = NULL;
1055                 struct emitted_diff_symbol *l = &o->emitted_symbols->buf[n];
1056
1057                 switch (l->s) {
1058                 case DIFF_SYMBOL_PLUS:
1059                         hm = del_lines;
1060                         key = prepare_entry(o, n);
1061                         match = hashmap_get(hm, key, NULL);
1062                         free(key);
1063                         break;
1064                 case DIFF_SYMBOL_MINUS:
1065                         hm = add_lines;
1066                         key = prepare_entry(o, n);
1067                         match = hashmap_get(hm, key, NULL);
1068                         free(key);
1069                         break;
1070                 default:
1071                         flipped_block = 1;
1072                 }
1073
1074                 if (!match) {
1075                         adjust_last_block(o, n, block_length);
1076                         pmb_nr = 0;
1077                         block_length = 0;
1078                         continue;
1079                 }
1080
1081                 l->flags |= DIFF_SYMBOL_MOVED_LINE;
1082
1083                 if (o->color_moved == COLOR_MOVED_PLAIN)
1084                         continue;
1085
1086                 if (o->color_moved_ws_handling &
1087                     COLOR_MOVED_WS_ALLOW_INDENTATION_CHANGE)
1088                         pmb_advance_or_null_multi_match(o, match, hm, pmb, pmb_nr, n);
1089                 else
1090                         pmb_advance_or_null(o, match, hm, pmb, pmb_nr);
1091
1092                 pmb_nr = shrink_potential_moved_blocks(pmb, pmb_nr);
1093
1094                 if (pmb_nr == 0) {
1095                         /*
1096                          * The current line is the start of a new block.
1097                          * Setup the set of potential blocks.
1098                          */
1099                         for (; match; match = hashmap_get_next(hm, match)) {
1100                                 ALLOC_GROW(pmb, pmb_nr + 1, pmb_alloc);
1101                                 if (o->color_moved_ws_handling &
1102                                     COLOR_MOVED_WS_ALLOW_INDENTATION_CHANGE) {
1103                                         struct ws_delta *wsd = xmalloc(sizeof(*match->wsd));
1104                                         if (compute_ws_delta(l, match->es, wsd)) {
1105                                                 match->wsd = wsd;
1106                                                 pmb[pmb_nr++] = match;
1107                                         } else
1108                                                 free(wsd);
1109                                 } else {
1110                                         pmb[pmb_nr++] = match;
1111                                 }
1112                         }
1113
1114                         flipped_block = (flipped_block + 1) % 2;
1115
1116                         adjust_last_block(o, n, block_length);
1117                         block_length = 0;
1118                 }
1119
1120                 block_length++;
1121
1122                 if (flipped_block && o->color_moved != COLOR_MOVED_BLOCKS)
1123                         l->flags |= DIFF_SYMBOL_MOVED_LINE_ALT;
1124         }
1125         adjust_last_block(o, n, block_length);
1126
1127         free(pmb);
1128 }
1129
1130 #define DIFF_SYMBOL_MOVED_LINE_ZEBRA_MASK \
1131   (DIFF_SYMBOL_MOVED_LINE | DIFF_SYMBOL_MOVED_LINE_ALT)
1132 static void dim_moved_lines(struct diff_options *o)
1133 {
1134         int n;
1135         for (n = 0; n < o->emitted_symbols->nr; n++) {
1136                 struct emitted_diff_symbol *prev = (n != 0) ?
1137                                 &o->emitted_symbols->buf[n - 1] : NULL;
1138                 struct emitted_diff_symbol *l = &o->emitted_symbols->buf[n];
1139                 struct emitted_diff_symbol *next =
1140                                 (n < o->emitted_symbols->nr - 1) ?
1141                                 &o->emitted_symbols->buf[n + 1] : NULL;
1142
1143                 /* Not a plus or minus line? */
1144                 if (l->s != DIFF_SYMBOL_PLUS && l->s != DIFF_SYMBOL_MINUS)
1145                         continue;
1146
1147                 /* Not a moved line? */
1148                 if (!(l->flags & DIFF_SYMBOL_MOVED_LINE))
1149                         continue;
1150
1151                 /*
1152                  * If prev or next are not a plus or minus line,
1153                  * pretend they don't exist
1154                  */
1155                 if (prev && prev->s != DIFF_SYMBOL_PLUS &&
1156                             prev->s != DIFF_SYMBOL_MINUS)
1157                         prev = NULL;
1158                 if (next && next->s != DIFF_SYMBOL_PLUS &&
1159                             next->s != DIFF_SYMBOL_MINUS)
1160                         next = NULL;
1161
1162                 /* Inside a block? */
1163                 if ((prev &&
1164                     (prev->flags & DIFF_SYMBOL_MOVED_LINE_ZEBRA_MASK) ==
1165                     (l->flags & DIFF_SYMBOL_MOVED_LINE_ZEBRA_MASK)) &&
1166                     (next &&
1167                     (next->flags & DIFF_SYMBOL_MOVED_LINE_ZEBRA_MASK) ==
1168                     (l->flags & DIFF_SYMBOL_MOVED_LINE_ZEBRA_MASK))) {
1169                         l->flags |= DIFF_SYMBOL_MOVED_LINE_UNINTERESTING;
1170                         continue;
1171                 }
1172
1173                 /* Check if we are at an interesting bound: */
1174                 if (prev && (prev->flags & DIFF_SYMBOL_MOVED_LINE) &&
1175                     (prev->flags & DIFF_SYMBOL_MOVED_LINE_ALT) !=
1176                        (l->flags & DIFF_SYMBOL_MOVED_LINE_ALT))
1177                         continue;
1178                 if (next && (next->flags & DIFF_SYMBOL_MOVED_LINE) &&
1179                     (next->flags & DIFF_SYMBOL_MOVED_LINE_ALT) !=
1180                        (l->flags & DIFF_SYMBOL_MOVED_LINE_ALT))
1181                         continue;
1182
1183                 /*
1184                  * The boundary to prev and next are not interesting,
1185                  * so this line is not interesting as a whole
1186                  */
1187                 l->flags |= DIFF_SYMBOL_MOVED_LINE_UNINTERESTING;
1188         }
1189 }
1190
1191 static void emit_line_ws_markup(struct diff_options *o,
1192                                 const char *set, const char *reset,
1193                                 const char *line, int len,
1194                                 const char *set_sign, char sign,
1195                                 unsigned ws_rule, int blank_at_eof)
1196 {
1197         const char *ws = NULL;
1198
1199         if (o->ws_error_highlight & ws_rule) {
1200                 ws = diff_get_color_opt(o, DIFF_WHITESPACE);
1201                 if (!*ws)
1202                         ws = NULL;
1203         }
1204
1205         if (!ws && !set_sign)
1206                 emit_line_0(o, set, 0, reset, sign, line, len);
1207         else if (!ws) {
1208                 /* Emit just the prefix, then the rest. */
1209                 emit_line_0(o, set_sign ? set_sign : set, !!set_sign, reset,
1210                             sign, "", 0);
1211                 emit_line_0(o, set, 0, reset, 0, line, len);
1212         } else if (blank_at_eof)
1213                 /* Blank line at EOF - paint '+' as well */
1214                 emit_line_0(o, ws, 0, reset, sign, line, len);
1215         else {
1216                 /* Emit just the prefix, then the rest. */
1217                 emit_line_0(o, set_sign ? set_sign : set, !!set_sign, reset,
1218                             sign, "", 0);
1219                 ws_check_emit(line, len, ws_rule,
1220                               o->file, set, reset, ws);
1221         }
1222 }
1223
1224 static void emit_diff_symbol_from_struct(struct diff_options *o,
1225                                          struct emitted_diff_symbol *eds)
1226 {
1227         static const char *nneof = " No newline at end of file\n";
1228         const char *context, *reset, *set, *set_sign, *meta, *fraginfo;
1229         struct strbuf sb = STRBUF_INIT;
1230
1231         enum diff_symbol s = eds->s;
1232         const char *line = eds->line;
1233         int len = eds->len;
1234         unsigned flags = eds->flags;
1235
1236         switch (s) {
1237         case DIFF_SYMBOL_NO_LF_EOF:
1238                 context = diff_get_color_opt(o, DIFF_CONTEXT);
1239                 reset = diff_get_color_opt(o, DIFF_RESET);
1240                 putc('\n', o->file);
1241                 emit_line_0(o, context, 0, reset, '\\',
1242                             nneof, strlen(nneof));
1243                 break;
1244         case DIFF_SYMBOL_SUBMODULE_HEADER:
1245         case DIFF_SYMBOL_SUBMODULE_ERROR:
1246         case DIFF_SYMBOL_SUBMODULE_PIPETHROUGH:
1247         case DIFF_SYMBOL_STATS_SUMMARY_INSERTS_DELETES:
1248         case DIFF_SYMBOL_SUMMARY:
1249         case DIFF_SYMBOL_STATS_LINE:
1250         case DIFF_SYMBOL_BINARY_DIFF_BODY:
1251         case DIFF_SYMBOL_CONTEXT_FRAGINFO:
1252                 emit_line(o, "", "", line, len);
1253                 break;
1254         case DIFF_SYMBOL_CONTEXT_INCOMPLETE:
1255         case DIFF_SYMBOL_CONTEXT_MARKER:
1256                 context = diff_get_color_opt(o, DIFF_CONTEXT);
1257                 reset = diff_get_color_opt(o, DIFF_RESET);
1258                 emit_line(o, context, reset, line, len);
1259                 break;
1260         case DIFF_SYMBOL_SEPARATOR:
1261                 fprintf(o->file, "%s%c",
1262                         diff_line_prefix(o),
1263                         o->line_termination);
1264                 break;
1265         case DIFF_SYMBOL_CONTEXT:
1266                 set = diff_get_color_opt(o, DIFF_CONTEXT);
1267                 reset = diff_get_color_opt(o, DIFF_RESET);
1268                 set_sign = NULL;
1269                 if (o->flags.dual_color_diffed_diffs) {
1270                         char c = !len ? 0 : line[0];
1271
1272                         if (c == '+')
1273                                 set = diff_get_color_opt(o, DIFF_FILE_NEW);
1274                         else if (c == '@')
1275                                 set = diff_get_color_opt(o, DIFF_FRAGINFO);
1276                         else if (c == '-')
1277                                 set = diff_get_color_opt(o, DIFF_FILE_OLD);
1278                 }
1279                 emit_line_ws_markup(o, set, reset, line, len, set_sign, ' ',
1280                                     flags & (DIFF_SYMBOL_CONTENT_WS_MASK), 0);
1281                 break;
1282         case DIFF_SYMBOL_PLUS:
1283                 switch (flags & (DIFF_SYMBOL_MOVED_LINE |
1284                                  DIFF_SYMBOL_MOVED_LINE_ALT |
1285                                  DIFF_SYMBOL_MOVED_LINE_UNINTERESTING)) {
1286                 case DIFF_SYMBOL_MOVED_LINE |
1287                      DIFF_SYMBOL_MOVED_LINE_ALT |
1288                      DIFF_SYMBOL_MOVED_LINE_UNINTERESTING:
1289                         set = diff_get_color_opt(o, DIFF_FILE_NEW_MOVED_ALT_DIM);
1290                         break;
1291                 case DIFF_SYMBOL_MOVED_LINE |
1292                      DIFF_SYMBOL_MOVED_LINE_ALT:
1293                         set = diff_get_color_opt(o, DIFF_FILE_NEW_MOVED_ALT);
1294                         break;
1295                 case DIFF_SYMBOL_MOVED_LINE |
1296                      DIFF_SYMBOL_MOVED_LINE_UNINTERESTING:
1297                         set = diff_get_color_opt(o, DIFF_FILE_NEW_MOVED_DIM);
1298                         break;
1299                 case DIFF_SYMBOL_MOVED_LINE:
1300                         set = diff_get_color_opt(o, DIFF_FILE_NEW_MOVED);
1301                         break;
1302                 default:
1303                         set = diff_get_color_opt(o, DIFF_FILE_NEW);
1304                 }
1305                 reset = diff_get_color_opt(o, DIFF_RESET);
1306                 if (!o->flags.dual_color_diffed_diffs)
1307                         set_sign = NULL;
1308                 else {
1309                         char c = !len ? 0 : line[0];
1310
1311                         set_sign = set;
1312                         if (c == '-')
1313                                 set = diff_get_color_opt(o, DIFF_FILE_OLD_BOLD);
1314                         else if (c == '@')
1315                                 set = diff_get_color_opt(o, DIFF_FRAGINFO);
1316                         else if (c == '+')
1317                                 set = diff_get_color_opt(o, DIFF_FILE_NEW_BOLD);
1318                         else
1319                                 set = diff_get_color_opt(o, DIFF_CONTEXT_BOLD);
1320                         flags &= ~DIFF_SYMBOL_CONTENT_WS_MASK;
1321                 }
1322                 emit_line_ws_markup(o, set, reset, line, len, set_sign, '+',
1323                                     flags & DIFF_SYMBOL_CONTENT_WS_MASK,
1324                                     flags & DIFF_SYMBOL_CONTENT_BLANK_LINE_EOF);
1325                 break;
1326         case DIFF_SYMBOL_MINUS:
1327                 switch (flags & (DIFF_SYMBOL_MOVED_LINE |
1328                                  DIFF_SYMBOL_MOVED_LINE_ALT |
1329                                  DIFF_SYMBOL_MOVED_LINE_UNINTERESTING)) {
1330                 case DIFF_SYMBOL_MOVED_LINE |
1331                      DIFF_SYMBOL_MOVED_LINE_ALT |
1332                      DIFF_SYMBOL_MOVED_LINE_UNINTERESTING:
1333                         set = diff_get_color_opt(o, DIFF_FILE_OLD_MOVED_ALT_DIM);
1334                         break;
1335                 case DIFF_SYMBOL_MOVED_LINE |
1336                      DIFF_SYMBOL_MOVED_LINE_ALT:
1337                         set = diff_get_color_opt(o, DIFF_FILE_OLD_MOVED_ALT);
1338                         break;
1339                 case DIFF_SYMBOL_MOVED_LINE |
1340                      DIFF_SYMBOL_MOVED_LINE_UNINTERESTING:
1341                         set = diff_get_color_opt(o, DIFF_FILE_OLD_MOVED_DIM);
1342                         break;
1343                 case DIFF_SYMBOL_MOVED_LINE:
1344                         set = diff_get_color_opt(o, DIFF_FILE_OLD_MOVED);
1345                         break;
1346                 default:
1347                         set = diff_get_color_opt(o, DIFF_FILE_OLD);
1348                 }
1349                 reset = diff_get_color_opt(o, DIFF_RESET);
1350                 if (!o->flags.dual_color_diffed_diffs)
1351                         set_sign = NULL;
1352                 else {
1353                         char c = !len ? 0 : line[0];
1354
1355                         set_sign = set;
1356                         if (c == '+')
1357                                 set = diff_get_color_opt(o, DIFF_FILE_NEW_DIM);
1358                         else if (c == '@')
1359                                 set = diff_get_color_opt(o, DIFF_FRAGINFO);
1360                         else if (c == '-')
1361                                 set = diff_get_color_opt(o, DIFF_FILE_OLD_DIM);
1362                         else
1363                                 set = diff_get_color_opt(o, DIFF_CONTEXT_DIM);
1364                 }
1365                 emit_line_ws_markup(o, set, reset, line, len, set_sign, '-',
1366                                     flags & DIFF_SYMBOL_CONTENT_WS_MASK, 0);
1367                 break;
1368         case DIFF_SYMBOL_WORDS_PORCELAIN:
1369                 context = diff_get_color_opt(o, DIFF_CONTEXT);
1370                 reset = diff_get_color_opt(o, DIFF_RESET);
1371                 emit_line(o, context, reset, line, len);
1372                 fputs("~\n", o->file);
1373                 break;
1374         case DIFF_SYMBOL_WORDS:
1375                 context = diff_get_color_opt(o, DIFF_CONTEXT);
1376                 reset = diff_get_color_opt(o, DIFF_RESET);
1377                 /*
1378                  * Skip the prefix character, if any.  With
1379                  * diff_suppress_blank_empty, there may be
1380                  * none.
1381                  */
1382                 if (line[0] != '\n') {
1383                         line++;
1384                         len--;
1385                 }
1386                 emit_line(o, context, reset, line, len);
1387                 break;
1388         case DIFF_SYMBOL_FILEPAIR_PLUS:
1389                 meta = diff_get_color_opt(o, DIFF_METAINFO);
1390                 reset = diff_get_color_opt(o, DIFF_RESET);
1391                 fprintf(o->file, "%s%s+++ %s%s%s\n", diff_line_prefix(o), meta,
1392                         line, reset,
1393                         strchr(line, ' ') ? "\t" : "");
1394                 break;
1395         case DIFF_SYMBOL_FILEPAIR_MINUS:
1396                 meta = diff_get_color_opt(o, DIFF_METAINFO);
1397                 reset = diff_get_color_opt(o, DIFF_RESET);
1398                 fprintf(o->file, "%s%s--- %s%s%s\n", diff_line_prefix(o), meta,
1399                         line, reset,
1400                         strchr(line, ' ') ? "\t" : "");
1401                 break;
1402         case DIFF_SYMBOL_BINARY_FILES:
1403         case DIFF_SYMBOL_HEADER:
1404                 fprintf(o->file, "%s", line);
1405                 break;
1406         case DIFF_SYMBOL_BINARY_DIFF_HEADER:
1407                 fprintf(o->file, "%sGIT binary patch\n", diff_line_prefix(o));
1408                 break;
1409         case DIFF_SYMBOL_BINARY_DIFF_HEADER_DELTA:
1410                 fprintf(o->file, "%sdelta %s\n", diff_line_prefix(o), line);
1411                 break;
1412         case DIFF_SYMBOL_BINARY_DIFF_HEADER_LITERAL:
1413                 fprintf(o->file, "%sliteral %s\n", diff_line_prefix(o), line);
1414                 break;
1415         case DIFF_SYMBOL_BINARY_DIFF_FOOTER:
1416                 fputs(diff_line_prefix(o), o->file);
1417                 fputc('\n', o->file);
1418                 break;
1419         case DIFF_SYMBOL_REWRITE_DIFF:
1420                 fraginfo = diff_get_color(o->use_color, DIFF_FRAGINFO);
1421                 reset = diff_get_color_opt(o, DIFF_RESET);
1422                 emit_line(o, fraginfo, reset, line, len);
1423                 break;
1424         case DIFF_SYMBOL_SUBMODULE_ADD:
1425                 set = diff_get_color_opt(o, DIFF_FILE_NEW);
1426                 reset = diff_get_color_opt(o, DIFF_RESET);
1427                 emit_line(o, set, reset, line, len);
1428                 break;
1429         case DIFF_SYMBOL_SUBMODULE_DEL:
1430                 set = diff_get_color_opt(o, DIFF_FILE_OLD);
1431                 reset = diff_get_color_opt(o, DIFF_RESET);
1432                 emit_line(o, set, reset, line, len);
1433                 break;
1434         case DIFF_SYMBOL_SUBMODULE_UNTRACKED:
1435                 fprintf(o->file, "%sSubmodule %s contains untracked content\n",
1436                         diff_line_prefix(o), line);
1437                 break;
1438         case DIFF_SYMBOL_SUBMODULE_MODIFIED:
1439                 fprintf(o->file, "%sSubmodule %s contains modified content\n",
1440                         diff_line_prefix(o), line);
1441                 break;
1442         case DIFF_SYMBOL_STATS_SUMMARY_NO_FILES:
1443                 emit_line(o, "", "", " 0 files changed\n",
1444                           strlen(" 0 files changed\n"));
1445                 break;
1446         case DIFF_SYMBOL_STATS_SUMMARY_ABBREV:
1447                 emit_line(o, "", "", " ...\n", strlen(" ...\n"));
1448                 break;
1449         case DIFF_SYMBOL_WORD_DIFF:
1450                 fprintf(o->file, "%.*s", len, line);
1451                 break;
1452         case DIFF_SYMBOL_STAT_SEP:
1453                 fputs(o->stat_sep, o->file);
1454                 break;
1455         default:
1456                 BUG("unknown diff symbol");
1457         }
1458         strbuf_release(&sb);
1459 }
1460
1461 static void emit_diff_symbol(struct diff_options *o, enum diff_symbol s,
1462                              const char *line, int len, unsigned flags)
1463 {
1464         struct emitted_diff_symbol e = {line, len, flags, s};
1465
1466         if (o->emitted_symbols)
1467                 append_emitted_diff_symbol(o, &e);
1468         else
1469                 emit_diff_symbol_from_struct(o, &e);
1470 }
1471
1472 void diff_emit_submodule_del(struct diff_options *o, const char *line)
1473 {
1474         emit_diff_symbol(o, DIFF_SYMBOL_SUBMODULE_DEL, line, strlen(line), 0);
1475 }
1476
1477 void diff_emit_submodule_add(struct diff_options *o, const char *line)
1478 {
1479         emit_diff_symbol(o, DIFF_SYMBOL_SUBMODULE_ADD, line, strlen(line), 0);
1480 }
1481
1482 void diff_emit_submodule_untracked(struct diff_options *o, const char *path)
1483 {
1484         emit_diff_symbol(o, DIFF_SYMBOL_SUBMODULE_UNTRACKED,
1485                          path, strlen(path), 0);
1486 }
1487
1488 void diff_emit_submodule_modified(struct diff_options *o, const char *path)
1489 {
1490         emit_diff_symbol(o, DIFF_SYMBOL_SUBMODULE_MODIFIED,
1491                          path, strlen(path), 0);
1492 }
1493
1494 void diff_emit_submodule_header(struct diff_options *o, const char *header)
1495 {
1496         emit_diff_symbol(o, DIFF_SYMBOL_SUBMODULE_HEADER,
1497                          header, strlen(header), 0);
1498 }
1499
1500 void diff_emit_submodule_error(struct diff_options *o, const char *err)
1501 {
1502         emit_diff_symbol(o, DIFF_SYMBOL_SUBMODULE_ERROR, err, strlen(err), 0);
1503 }
1504
1505 void diff_emit_submodule_pipethrough(struct diff_options *o,
1506                                      const char *line, int len)
1507 {
1508         emit_diff_symbol(o, DIFF_SYMBOL_SUBMODULE_PIPETHROUGH, line, len, 0);
1509 }
1510
1511 static int new_blank_line_at_eof(struct emit_callback *ecbdata, const char *line, int len)
1512 {
1513         if (!((ecbdata->ws_rule & WS_BLANK_AT_EOF) &&
1514               ecbdata->blank_at_eof_in_preimage &&
1515               ecbdata->blank_at_eof_in_postimage &&
1516               ecbdata->blank_at_eof_in_preimage <= ecbdata->lno_in_preimage &&
1517               ecbdata->blank_at_eof_in_postimage <= ecbdata->lno_in_postimage))
1518                 return 0;
1519         return ws_blank_line(line, len, ecbdata->ws_rule);
1520 }
1521
1522 static void emit_add_line(const char *reset,
1523                           struct emit_callback *ecbdata,
1524                           const char *line, int len)
1525 {
1526         unsigned flags = WSEH_NEW | ecbdata->ws_rule;
1527         if (new_blank_line_at_eof(ecbdata, line, len))
1528                 flags |= DIFF_SYMBOL_CONTENT_BLANK_LINE_EOF;
1529
1530         emit_diff_symbol(ecbdata->opt, DIFF_SYMBOL_PLUS, line, len, flags);
1531 }
1532
1533 static void emit_del_line(const char *reset,
1534                           struct emit_callback *ecbdata,
1535                           const char *line, int len)
1536 {
1537         unsigned flags = WSEH_OLD | ecbdata->ws_rule;
1538         emit_diff_symbol(ecbdata->opt, DIFF_SYMBOL_MINUS, line, len, flags);
1539 }
1540
1541 static void emit_context_line(const char *reset,
1542                               struct emit_callback *ecbdata,
1543                               const char *line, int len)
1544 {
1545         unsigned flags = WSEH_CONTEXT | ecbdata->ws_rule;
1546         emit_diff_symbol(ecbdata->opt, DIFF_SYMBOL_CONTEXT, line, len, flags);
1547 }
1548
1549 static void emit_hunk_header(struct emit_callback *ecbdata,
1550                              const char *line, int len)
1551 {
1552         const char *context = diff_get_color(ecbdata->color_diff, DIFF_CONTEXT);
1553         const char *frag = diff_get_color(ecbdata->color_diff, DIFF_FRAGINFO);
1554         const char *func = diff_get_color(ecbdata->color_diff, DIFF_FUNCINFO);
1555         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
1556         const char *reverse = ecbdata->color_diff ? GIT_COLOR_REVERSE : "";
1557         static const char atat[2] = { '@', '@' };
1558         const char *cp, *ep;
1559         struct strbuf msgbuf = STRBUF_INIT;
1560         int org_len = len;
1561         int i = 1;
1562
1563         /*
1564          * As a hunk header must begin with "@@ -<old>, +<new> @@",
1565          * it always is at least 10 bytes long.
1566          */
1567         if (len < 10 ||
1568             memcmp(line, atat, 2) ||
1569             !(ep = memmem(line + 2, len - 2, atat, 2))) {
1570                 emit_diff_symbol(ecbdata->opt,
1571                                  DIFF_SYMBOL_CONTEXT_MARKER, line, len, 0);
1572                 return;
1573         }
1574         ep += 2; /* skip over @@ */
1575
1576         /* The hunk header in fraginfo color */
1577         if (ecbdata->opt->flags.dual_color_diffed_diffs)
1578                 strbuf_addstr(&msgbuf, reverse);
1579         strbuf_addstr(&msgbuf, frag);
1580         strbuf_add(&msgbuf, line, ep - line);
1581         strbuf_addstr(&msgbuf, reset);
1582
1583         /*
1584          * trailing "\r\n"
1585          */
1586         for ( ; i < 3; i++)
1587                 if (line[len - i] == '\r' || line[len - i] == '\n')
1588                         len--;
1589
1590         /* blank before the func header */
1591         for (cp = ep; ep - line < len; ep++)
1592                 if (*ep != ' ' && *ep != '\t')
1593                         break;
1594         if (ep != cp) {
1595                 strbuf_addstr(&msgbuf, context);
1596                 strbuf_add(&msgbuf, cp, ep - cp);
1597                 strbuf_addstr(&msgbuf, reset);
1598         }
1599
1600         if (ep < line + len) {
1601                 strbuf_addstr(&msgbuf, func);
1602                 strbuf_add(&msgbuf, ep, line + len - ep);
1603                 strbuf_addstr(&msgbuf, reset);
1604         }
1605
1606         strbuf_add(&msgbuf, line + len, org_len - len);
1607         strbuf_complete_line(&msgbuf);
1608         emit_diff_symbol(ecbdata->opt,
1609                          DIFF_SYMBOL_CONTEXT_FRAGINFO, msgbuf.buf, msgbuf.len, 0);
1610         strbuf_release(&msgbuf);
1611 }
1612
1613 static struct diff_tempfile *claim_diff_tempfile(void) {
1614         int i;
1615         for (i = 0; i < ARRAY_SIZE(diff_temp); i++)
1616                 if (!diff_temp[i].name)
1617                         return diff_temp + i;
1618         BUG("diff is failing to clean up its tempfiles");
1619 }
1620
1621 static void remove_tempfile(void)
1622 {
1623         int i;
1624         for (i = 0; i < ARRAY_SIZE(diff_temp); i++) {
1625                 if (is_tempfile_active(diff_temp[i].tempfile))
1626                         delete_tempfile(&diff_temp[i].tempfile);
1627                 diff_temp[i].name = NULL;
1628         }
1629 }
1630
1631 static void add_line_count(struct strbuf *out, int count)
1632 {
1633         switch (count) {
1634         case 0:
1635                 strbuf_addstr(out, "0,0");
1636                 break;
1637         case 1:
1638                 strbuf_addstr(out, "1");
1639                 break;
1640         default:
1641                 strbuf_addf(out, "1,%d", count);
1642                 break;
1643         }
1644 }
1645
1646 static void emit_rewrite_lines(struct emit_callback *ecb,
1647                                int prefix, const char *data, int size)
1648 {
1649         const char *endp = NULL;
1650         const char *reset = diff_get_color(ecb->color_diff, DIFF_RESET);
1651
1652         while (0 < size) {
1653                 int len;
1654
1655                 endp = memchr(data, '\n', size);
1656                 len = endp ? (endp - data + 1) : size;
1657                 if (prefix != '+') {
1658                         ecb->lno_in_preimage++;
1659                         emit_del_line(reset, ecb, data, len);
1660                 } else {
1661                         ecb->lno_in_postimage++;
1662                         emit_add_line(reset, ecb, data, len);
1663                 }
1664                 size -= len;
1665                 data += len;
1666         }
1667         if (!endp)
1668                 emit_diff_symbol(ecb->opt, DIFF_SYMBOL_NO_LF_EOF, NULL, 0, 0);
1669 }
1670
1671 static void emit_rewrite_diff(const char *name_a,
1672                               const char *name_b,
1673                               struct diff_filespec *one,
1674                               struct diff_filespec *two,
1675                               struct userdiff_driver *textconv_one,
1676                               struct userdiff_driver *textconv_two,
1677                               struct diff_options *o)
1678 {
1679         int lc_a, lc_b;
1680         static struct strbuf a_name = STRBUF_INIT, b_name = STRBUF_INIT;
1681         const char *a_prefix, *b_prefix;
1682         char *data_one, *data_two;
1683         size_t size_one, size_two;
1684         struct emit_callback ecbdata;
1685         struct strbuf out = STRBUF_INIT;
1686
1687         if (diff_mnemonic_prefix && o->flags.reverse_diff) {
1688                 a_prefix = o->b_prefix;
1689                 b_prefix = o->a_prefix;
1690         } else {
1691                 a_prefix = o->a_prefix;
1692                 b_prefix = o->b_prefix;
1693         }
1694
1695         name_a += (*name_a == '/');
1696         name_b += (*name_b == '/');
1697
1698         strbuf_reset(&a_name);
1699         strbuf_reset(&b_name);
1700         quote_two_c_style(&a_name, a_prefix, name_a, 0);
1701         quote_two_c_style(&b_name, b_prefix, name_b, 0);
1702
1703         size_one = fill_textconv(o->repo, textconv_one, one, &data_one);
1704         size_two = fill_textconv(o->repo, textconv_two, two, &data_two);
1705
1706         memset(&ecbdata, 0, sizeof(ecbdata));
1707         ecbdata.color_diff = want_color(o->use_color);
1708         ecbdata.ws_rule = whitespace_rule(name_b);
1709         ecbdata.opt = o;
1710         if (ecbdata.ws_rule & WS_BLANK_AT_EOF) {
1711                 mmfile_t mf1, mf2;
1712                 mf1.ptr = (char *)data_one;
1713                 mf2.ptr = (char *)data_two;
1714                 mf1.size = size_one;
1715                 mf2.size = size_two;
1716                 check_blank_at_eof(&mf1, &mf2, &ecbdata);
1717         }
1718         ecbdata.lno_in_preimage = 1;
1719         ecbdata.lno_in_postimage = 1;
1720
1721         lc_a = count_lines(data_one, size_one);
1722         lc_b = count_lines(data_two, size_two);
1723
1724         emit_diff_symbol(o, DIFF_SYMBOL_FILEPAIR_MINUS,
1725                          a_name.buf, a_name.len, 0);
1726         emit_diff_symbol(o, DIFF_SYMBOL_FILEPAIR_PLUS,
1727                          b_name.buf, b_name.len, 0);
1728
1729         strbuf_addstr(&out, "@@ -");
1730         if (!o->irreversible_delete)
1731                 add_line_count(&out, lc_a);
1732         else
1733                 strbuf_addstr(&out, "?,?");
1734         strbuf_addstr(&out, " +");
1735         add_line_count(&out, lc_b);
1736         strbuf_addstr(&out, " @@\n");
1737         emit_diff_symbol(o, DIFF_SYMBOL_REWRITE_DIFF, out.buf, out.len, 0);
1738         strbuf_release(&out);
1739
1740         if (lc_a && !o->irreversible_delete)
1741                 emit_rewrite_lines(&ecbdata, '-', data_one, size_one);
1742         if (lc_b)
1743                 emit_rewrite_lines(&ecbdata, '+', data_two, size_two);
1744         if (textconv_one)
1745                 free((char *)data_one);
1746         if (textconv_two)
1747                 free((char *)data_two);
1748 }
1749
1750 struct diff_words_buffer {
1751         mmfile_t text;
1752         unsigned long alloc;
1753         struct diff_words_orig {
1754                 const char *begin, *end;
1755         } *orig;
1756         int orig_nr, orig_alloc;
1757 };
1758
1759 static void diff_words_append(char *line, unsigned long len,
1760                 struct diff_words_buffer *buffer)
1761 {
1762         ALLOC_GROW(buffer->text.ptr, buffer->text.size + len, buffer->alloc);
1763         line++;
1764         len--;
1765         memcpy(buffer->text.ptr + buffer->text.size, line, len);
1766         buffer->text.size += len;
1767         buffer->text.ptr[buffer->text.size] = '\0';
1768 }
1769
1770 struct diff_words_style_elem {
1771         const char *prefix;
1772         const char *suffix;
1773         const char *color; /* NULL; filled in by the setup code if
1774                             * color is enabled */
1775 };
1776
1777 struct diff_words_style {
1778         enum diff_words_type type;
1779         struct diff_words_style_elem new_word, old_word, ctx;
1780         const char *newline;
1781 };
1782
1783 static struct diff_words_style diff_words_styles[] = {
1784         { DIFF_WORDS_PORCELAIN, {"+", "\n"}, {"-", "\n"}, {" ", "\n"}, "~\n" },
1785         { DIFF_WORDS_PLAIN, {"{+", "+}"}, {"[-", "-]"}, {"", ""}, "\n" },
1786         { DIFF_WORDS_COLOR, {"", ""}, {"", ""}, {"", ""}, "\n" }
1787 };
1788
1789 struct diff_words_data {
1790         struct diff_words_buffer minus, plus;
1791         const char *current_plus;
1792         int last_minus;
1793         struct diff_options *opt;
1794         regex_t *word_regex;
1795         enum diff_words_type type;
1796         struct diff_words_style *style;
1797 };
1798
1799 static int fn_out_diff_words_write_helper(struct diff_options *o,
1800                                           struct diff_words_style_elem *st_el,
1801                                           const char *newline,
1802                                           size_t count, const char *buf)
1803 {
1804         int print = 0;
1805         struct strbuf sb = STRBUF_INIT;
1806
1807         while (count) {
1808                 char *p = memchr(buf, '\n', count);
1809                 if (print)
1810                         strbuf_addstr(&sb, diff_line_prefix(o));
1811
1812                 if (p != buf) {
1813                         const char *reset = st_el->color && *st_el->color ?
1814                                             GIT_COLOR_RESET : NULL;
1815                         if (st_el->color && *st_el->color)
1816                                 strbuf_addstr(&sb, st_el->color);
1817                         strbuf_addstr(&sb, st_el->prefix);
1818                         strbuf_add(&sb, buf, p ? p - buf : count);
1819                         strbuf_addstr(&sb, st_el->suffix);
1820                         if (reset)
1821                                 strbuf_addstr(&sb, reset);
1822                 }
1823                 if (!p)
1824                         goto out;
1825
1826                 strbuf_addstr(&sb, newline);
1827                 count -= p + 1 - buf;
1828                 buf = p + 1;
1829                 print = 1;
1830                 if (count) {
1831                         emit_diff_symbol(o, DIFF_SYMBOL_WORD_DIFF,
1832                                          sb.buf, sb.len, 0);
1833                         strbuf_reset(&sb);
1834                 }
1835         }
1836
1837 out:
1838         if (sb.len)
1839                 emit_diff_symbol(o, DIFF_SYMBOL_WORD_DIFF,
1840                                  sb.buf, sb.len, 0);
1841         strbuf_release(&sb);
1842         return 0;
1843 }
1844
1845 /*
1846  * '--color-words' algorithm can be described as:
1847  *
1848  *   1. collect the minus/plus lines of a diff hunk, divided into
1849  *      minus-lines and plus-lines;
1850  *
1851  *   2. break both minus-lines and plus-lines into words and
1852  *      place them into two mmfile_t with one word for each line;
1853  *
1854  *   3. use xdiff to run diff on the two mmfile_t to get the words level diff;
1855  *
1856  * And for the common parts of the both file, we output the plus side text.
1857  * diff_words->current_plus is used to trace the current position of the plus file
1858  * which printed. diff_words->last_minus is used to trace the last minus word
1859  * printed.
1860  *
1861  * For '--graph' to work with '--color-words', we need to output the graph prefix
1862  * on each line of color words output. Generally, there are two conditions on
1863  * which we should output the prefix.
1864  *
1865  *   1. diff_words->last_minus == 0 &&
1866  *      diff_words->current_plus == diff_words->plus.text.ptr
1867  *
1868  *      that is: the plus text must start as a new line, and if there is no minus
1869  *      word printed, a graph prefix must be printed.
1870  *
1871  *   2. diff_words->current_plus > diff_words->plus.text.ptr &&
1872  *      *(diff_words->current_plus - 1) == '\n'
1873  *
1874  *      that is: a graph prefix must be printed following a '\n'
1875  */
1876 static int color_words_output_graph_prefix(struct diff_words_data *diff_words)
1877 {
1878         if ((diff_words->last_minus == 0 &&
1879                 diff_words->current_plus == diff_words->plus.text.ptr) ||
1880                 (diff_words->current_plus > diff_words->plus.text.ptr &&
1881                 *(diff_words->current_plus - 1) == '\n')) {
1882                 return 1;
1883         } else {
1884                 return 0;
1885         }
1886 }
1887
1888 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
1889 {
1890         struct diff_words_data *diff_words = priv;
1891         struct diff_words_style *style = diff_words->style;
1892         int minus_first, minus_len, plus_first, plus_len;
1893         const char *minus_begin, *minus_end, *plus_begin, *plus_end;
1894         struct diff_options *opt = diff_words->opt;
1895         const char *line_prefix;
1896
1897         if (line[0] != '@' || parse_hunk_header(line, len,
1898                         &minus_first, &minus_len, &plus_first, &plus_len))
1899                 return;
1900
1901         assert(opt);
1902         line_prefix = diff_line_prefix(opt);
1903
1904         /* POSIX requires that first be decremented by one if len == 0... */
1905         if (minus_len) {
1906                 minus_begin = diff_words->minus.orig[minus_first].begin;
1907                 minus_end =
1908                         diff_words->minus.orig[minus_first + minus_len - 1].end;
1909         } else
1910                 minus_begin = minus_end =
1911                         diff_words->minus.orig[minus_first].end;
1912
1913         if (plus_len) {
1914                 plus_begin = diff_words->plus.orig[plus_first].begin;
1915                 plus_end = diff_words->plus.orig[plus_first + plus_len - 1].end;
1916         } else
1917                 plus_begin = plus_end = diff_words->plus.orig[plus_first].end;
1918
1919         if (color_words_output_graph_prefix(diff_words)) {
1920                 fputs(line_prefix, diff_words->opt->file);
1921         }
1922         if (diff_words->current_plus != plus_begin) {
1923                 fn_out_diff_words_write_helper(diff_words->opt,
1924                                 &style->ctx, style->newline,
1925                                 plus_begin - diff_words->current_plus,
1926                                 diff_words->current_plus);
1927         }
1928         if (minus_begin != minus_end) {
1929                 fn_out_diff_words_write_helper(diff_words->opt,
1930                                 &style->old_word, style->newline,
1931                                 minus_end - minus_begin, minus_begin);
1932         }
1933         if (plus_begin != plus_end) {
1934                 fn_out_diff_words_write_helper(diff_words->opt,
1935                                 &style->new_word, style->newline,
1936                                 plus_end - plus_begin, plus_begin);
1937         }
1938
1939         diff_words->current_plus = plus_end;
1940         diff_words->last_minus = minus_first;
1941 }
1942
1943 /* This function starts looking at *begin, and returns 0 iff a word was found. */
1944 static int find_word_boundaries(mmfile_t *buffer, regex_t *word_regex,
1945                 int *begin, int *end)
1946 {
1947         if (word_regex && *begin < buffer->size) {
1948                 regmatch_t match[1];
1949                 if (!regexec_buf(word_regex, buffer->ptr + *begin,
1950                                  buffer->size - *begin, 1, match, 0)) {
1951                         char *p = memchr(buffer->ptr + *begin + match[0].rm_so,
1952                                         '\n', match[0].rm_eo - match[0].rm_so);
1953                         *end = p ? p - buffer->ptr : match[0].rm_eo + *begin;
1954                         *begin += match[0].rm_so;
1955                         return *begin >= *end;
1956                 }
1957                 return -1;
1958         }
1959
1960         /* find the next word */
1961         while (*begin < buffer->size && isspace(buffer->ptr[*begin]))
1962                 (*begin)++;
1963         if (*begin >= buffer->size)
1964                 return -1;
1965
1966         /* find the end of the word */
1967         *end = *begin + 1;
1968         while (*end < buffer->size && !isspace(buffer->ptr[*end]))
1969                 (*end)++;
1970
1971         return 0;
1972 }
1973
1974 /*
1975  * This function splits the words in buffer->text, stores the list with
1976  * newline separator into out, and saves the offsets of the original words
1977  * in buffer->orig.
1978  */
1979 static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out,
1980                 regex_t *word_regex)
1981 {
1982         int i, j;
1983         long alloc = 0;
1984
1985         out->size = 0;
1986         out->ptr = NULL;
1987
1988         /* fake an empty "0th" word */
1989         ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
1990         buffer->orig[0].begin = buffer->orig[0].end = buffer->text.ptr;
1991         buffer->orig_nr = 1;
1992
1993         for (i = 0; i < buffer->text.size; i++) {
1994                 if (find_word_boundaries(&buffer->text, word_regex, &i, &j))
1995                         return;
1996
1997                 /* store original boundaries */
1998                 ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
1999                                 buffer->orig_alloc);
2000                 buffer->orig[buffer->orig_nr].begin = buffer->text.ptr + i;
2001                 buffer->orig[buffer->orig_nr].end = buffer->text.ptr + j;
2002                 buffer->orig_nr++;
2003
2004                 /* store one word */
2005                 ALLOC_GROW(out->ptr, out->size + j - i + 1, alloc);
2006                 memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
2007                 out->ptr[out->size + j - i] = '\n';
2008                 out->size += j - i + 1;
2009
2010                 i = j - 1;
2011         }
2012 }
2013
2014 /* this executes the word diff on the accumulated buffers */
2015 static void diff_words_show(struct diff_words_data *diff_words)
2016 {
2017         xpparam_t xpp;
2018         xdemitconf_t xecfg;
2019         mmfile_t minus, plus;
2020         struct diff_words_style *style = diff_words->style;
2021
2022         struct diff_options *opt = diff_words->opt;
2023         const char *line_prefix;
2024
2025         assert(opt);
2026         line_prefix = diff_line_prefix(opt);
2027
2028         /* special case: only removal */
2029         if (!diff_words->plus.text.size) {
2030                 emit_diff_symbol(diff_words->opt, DIFF_SYMBOL_WORD_DIFF,
2031                                  line_prefix, strlen(line_prefix), 0);
2032                 fn_out_diff_words_write_helper(diff_words->opt,
2033                         &style->old_word, style->newline,
2034                         diff_words->minus.text.size,
2035                         diff_words->minus.text.ptr);
2036                 diff_words->minus.text.size = 0;
2037                 return;
2038         }
2039
2040         diff_words->current_plus = diff_words->plus.text.ptr;
2041         diff_words->last_minus = 0;
2042
2043         memset(&xpp, 0, sizeof(xpp));
2044         memset(&xecfg, 0, sizeof(xecfg));
2045         diff_words_fill(&diff_words->minus, &minus, diff_words->word_regex);
2046         diff_words_fill(&diff_words->plus, &plus, diff_words->word_regex);
2047         xpp.flags = 0;
2048         /* as only the hunk header will be parsed, we need a 0-context */
2049         xecfg.ctxlen = 0;
2050         if (xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
2051                           &xpp, &xecfg))
2052                 die("unable to generate word diff");
2053         free(minus.ptr);
2054         free(plus.ptr);
2055         if (diff_words->current_plus != diff_words->plus.text.ptr +
2056                         diff_words->plus.text.size) {
2057                 if (color_words_output_graph_prefix(diff_words))
2058                         emit_diff_symbol(diff_words->opt, DIFF_SYMBOL_WORD_DIFF,
2059                                          line_prefix, strlen(line_prefix), 0);
2060                 fn_out_diff_words_write_helper(diff_words->opt,
2061                         &style->ctx, style->newline,
2062                         diff_words->plus.text.ptr + diff_words->plus.text.size
2063                         - diff_words->current_plus, diff_words->current_plus);
2064         }
2065         diff_words->minus.text.size = diff_words->plus.text.size = 0;
2066 }
2067
2068 /* In "color-words" mode, show word-diff of words accumulated in the buffer */
2069 static void diff_words_flush(struct emit_callback *ecbdata)
2070 {
2071         struct diff_options *wo = ecbdata->diff_words->opt;
2072
2073         if (ecbdata->diff_words->minus.text.size ||
2074             ecbdata->diff_words->plus.text.size)
2075                 diff_words_show(ecbdata->diff_words);
2076
2077         if (wo->emitted_symbols) {
2078                 struct diff_options *o = ecbdata->opt;
2079                 struct emitted_diff_symbols *wol = wo->emitted_symbols;
2080                 int i;
2081
2082                 /*
2083                  * NEEDSWORK:
2084                  * Instead of appending each, concat all words to a line?
2085                  */
2086                 for (i = 0; i < wol->nr; i++)
2087                         append_emitted_diff_symbol(o, &wol->buf[i]);
2088
2089                 for (i = 0; i < wol->nr; i++)
2090                         free((void *)wol->buf[i].line);
2091
2092                 wol->nr = 0;
2093         }
2094 }
2095
2096 static void diff_filespec_load_driver(struct diff_filespec *one,
2097                                       struct index_state *istate)
2098 {
2099         /* Use already-loaded driver */
2100         if (one->driver)
2101                 return;
2102
2103         if (S_ISREG(one->mode))
2104                 one->driver = userdiff_find_by_path(istate, one->path);
2105
2106         /* Fallback to default settings */
2107         if (!one->driver)
2108                 one->driver = userdiff_find_by_name("default");
2109 }
2110
2111 static const char *userdiff_word_regex(struct diff_filespec *one,
2112                                        struct index_state *istate)
2113 {
2114         diff_filespec_load_driver(one, istate);
2115         return one->driver->word_regex;
2116 }
2117
2118 static void init_diff_words_data(struct emit_callback *ecbdata,
2119                                  struct diff_options *orig_opts,
2120                                  struct diff_filespec *one,
2121                                  struct diff_filespec *two)
2122 {
2123         int i;
2124         struct diff_options *o = xmalloc(sizeof(struct diff_options));
2125         memcpy(o, orig_opts, sizeof(struct diff_options));
2126
2127         ecbdata->diff_words =
2128                 xcalloc(1, sizeof(struct diff_words_data));
2129         ecbdata->diff_words->type = o->word_diff;
2130         ecbdata->diff_words->opt = o;
2131
2132         if (orig_opts->emitted_symbols)
2133                 o->emitted_symbols =
2134                         xcalloc(1, sizeof(struct emitted_diff_symbols));
2135
2136         if (!o->word_regex)
2137                 o->word_regex = userdiff_word_regex(one, o->repo->index);
2138         if (!o->word_regex)
2139                 o->word_regex = userdiff_word_regex(two, o->repo->index);
2140         if (!o->word_regex)
2141                 o->word_regex = diff_word_regex_cfg;
2142         if (o->word_regex) {
2143                 ecbdata->diff_words->word_regex = (regex_t *)
2144                         xmalloc(sizeof(regex_t));
2145                 if (regcomp(ecbdata->diff_words->word_regex,
2146                             o->word_regex,
2147                             REG_EXTENDED | REG_NEWLINE))
2148                         die("invalid regular expression: %s",
2149                             o->word_regex);
2150         }
2151         for (i = 0; i < ARRAY_SIZE(diff_words_styles); i++) {
2152                 if (o->word_diff == diff_words_styles[i].type) {
2153                         ecbdata->diff_words->style =
2154                                 &diff_words_styles[i];
2155                         break;
2156                 }
2157         }
2158         if (want_color(o->use_color)) {
2159                 struct diff_words_style *st = ecbdata->diff_words->style;
2160                 st->old_word.color = diff_get_color_opt(o, DIFF_FILE_OLD);
2161                 st->new_word.color = diff_get_color_opt(o, DIFF_FILE_NEW);
2162                 st->ctx.color = diff_get_color_opt(o, DIFF_CONTEXT);
2163         }
2164 }
2165
2166 static void free_diff_words_data(struct emit_callback *ecbdata)
2167 {
2168         if (ecbdata->diff_words) {
2169                 diff_words_flush(ecbdata);
2170                 free (ecbdata->diff_words->opt->emitted_symbols);
2171                 free (ecbdata->diff_words->opt);
2172                 free (ecbdata->diff_words->minus.text.ptr);
2173                 free (ecbdata->diff_words->minus.orig);
2174                 free (ecbdata->diff_words->plus.text.ptr);
2175                 free (ecbdata->diff_words->plus.orig);
2176                 if (ecbdata->diff_words->word_regex) {
2177                         regfree(ecbdata->diff_words->word_regex);
2178                         free(ecbdata->diff_words->word_regex);
2179                 }
2180                 FREE_AND_NULL(ecbdata->diff_words);
2181         }
2182 }
2183
2184 const char *diff_get_color(int diff_use_color, enum color_diff ix)
2185 {
2186         if (want_color(diff_use_color))
2187                 return diff_colors[ix];
2188         return "";
2189 }
2190
2191 const char *diff_line_prefix(struct diff_options *opt)
2192 {
2193         struct strbuf *msgbuf;
2194         if (!opt->output_prefix)
2195                 return "";
2196
2197         msgbuf = opt->output_prefix(opt, opt->output_prefix_data);
2198         return msgbuf->buf;
2199 }
2200
2201 static unsigned long sane_truncate_line(struct emit_callback *ecb, char *line, unsigned long len)
2202 {
2203         const char *cp;
2204         unsigned long allot;
2205         size_t l = len;
2206
2207         cp = line;
2208         allot = l;
2209         while (0 < l) {
2210                 (void) utf8_width(&cp, &l);
2211                 if (!cp)
2212                         break; /* truncated in the middle? */
2213         }
2214         return allot - l;
2215 }
2216
2217 static void find_lno(const char *line, struct emit_callback *ecbdata)
2218 {
2219         const char *p;
2220         ecbdata->lno_in_preimage = 0;
2221         ecbdata->lno_in_postimage = 0;
2222         p = strchr(line, '-');
2223         if (!p)
2224                 return; /* cannot happen */
2225         ecbdata->lno_in_preimage = strtol(p + 1, NULL, 10);
2226         p = strchr(p, '+');
2227         if (!p)
2228                 return; /* cannot happen */
2229         ecbdata->lno_in_postimage = strtol(p + 1, NULL, 10);
2230 }
2231
2232 static void fn_out_consume(void *priv, char *line, unsigned long len)
2233 {
2234         struct emit_callback *ecbdata = priv;
2235         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
2236         struct diff_options *o = ecbdata->opt;
2237
2238         o->found_changes = 1;
2239
2240         if (ecbdata->header) {
2241                 emit_diff_symbol(o, DIFF_SYMBOL_HEADER,
2242                                  ecbdata->header->buf, ecbdata->header->len, 0);
2243                 strbuf_reset(ecbdata->header);
2244                 ecbdata->header = NULL;
2245         }
2246
2247         if (ecbdata->label_path[0]) {
2248                 emit_diff_symbol(o, DIFF_SYMBOL_FILEPAIR_MINUS,
2249                                  ecbdata->label_path[0],
2250                                  strlen(ecbdata->label_path[0]), 0);
2251                 emit_diff_symbol(o, DIFF_SYMBOL_FILEPAIR_PLUS,
2252                                  ecbdata->label_path[1],
2253                                  strlen(ecbdata->label_path[1]), 0);
2254                 ecbdata->label_path[0] = ecbdata->label_path[1] = NULL;
2255         }
2256
2257         if (diff_suppress_blank_empty
2258             && len == 2 && line[0] == ' ' && line[1] == '\n') {
2259                 line[0] = '\n';
2260                 len = 1;
2261         }
2262
2263         if (line[0] == '@') {
2264                 if (ecbdata->diff_words)
2265                         diff_words_flush(ecbdata);
2266                 len = sane_truncate_line(ecbdata, line, len);
2267                 find_lno(line, ecbdata);
2268                 emit_hunk_header(ecbdata, line, len);
2269                 return;
2270         }
2271
2272         if (ecbdata->diff_words) {
2273                 enum diff_symbol s =
2274                         ecbdata->diff_words->type == DIFF_WORDS_PORCELAIN ?
2275                         DIFF_SYMBOL_WORDS_PORCELAIN : DIFF_SYMBOL_WORDS;
2276                 if (line[0] == '-') {
2277                         diff_words_append(line, len,
2278                                           &ecbdata->diff_words->minus);
2279                         return;
2280                 } else if (line[0] == '+') {
2281                         diff_words_append(line, len,
2282                                           &ecbdata->diff_words->plus);
2283                         return;
2284                 } else if (starts_with(line, "\\ ")) {
2285                         /*
2286                          * Eat the "no newline at eof" marker as if we
2287                          * saw a "+" or "-" line with nothing on it,
2288                          * and return without diff_words_flush() to
2289                          * defer processing. If this is the end of
2290                          * preimage, more "+" lines may come after it.
2291                          */
2292                         return;
2293                 }
2294                 diff_words_flush(ecbdata);
2295                 emit_diff_symbol(o, s, line, len, 0);
2296                 return;
2297         }
2298
2299         switch (line[0]) {
2300         case '+':
2301                 ecbdata->lno_in_postimage++;
2302                 emit_add_line(reset, ecbdata, line + 1, len - 1);
2303                 break;
2304         case '-':
2305                 ecbdata->lno_in_preimage++;
2306                 emit_del_line(reset, ecbdata, line + 1, len - 1);
2307                 break;
2308         case ' ':
2309                 ecbdata->lno_in_postimage++;
2310                 ecbdata->lno_in_preimage++;
2311                 emit_context_line(reset, ecbdata, line + 1, len - 1);
2312                 break;
2313         default:
2314                 /* incomplete line at the end */
2315                 ecbdata->lno_in_preimage++;
2316                 emit_diff_symbol(o, DIFF_SYMBOL_CONTEXT_INCOMPLETE,
2317                                  line, len, 0);
2318                 break;
2319         }
2320 }
2321
2322 static void pprint_rename(struct strbuf *name, const char *a, const char *b)
2323 {
2324         const char *old_name = a;
2325         const char *new_name = b;
2326         int pfx_length, sfx_length;
2327         int pfx_adjust_for_slash;
2328         int len_a = strlen(a);
2329         int len_b = strlen(b);
2330         int a_midlen, b_midlen;
2331         int qlen_a = quote_c_style(a, NULL, NULL, 0);
2332         int qlen_b = quote_c_style(b, NULL, NULL, 0);
2333
2334         if (qlen_a || qlen_b) {
2335                 quote_c_style(a, name, NULL, 0);
2336                 strbuf_addstr(name, " => ");
2337                 quote_c_style(b, name, NULL, 0);
2338                 return;
2339         }
2340
2341         /* Find common prefix */
2342         pfx_length = 0;
2343         while (*old_name && *new_name && *old_name == *new_name) {
2344                 if (*old_name == '/')
2345                         pfx_length = old_name - a + 1;
2346                 old_name++;
2347                 new_name++;
2348         }
2349
2350         /* Find common suffix */
2351         old_name = a + len_a;
2352         new_name = b + len_b;
2353         sfx_length = 0;
2354         /*
2355          * If there is a common prefix, it must end in a slash.  In
2356          * that case we let this loop run 1 into the prefix to see the
2357          * same slash.
2358          *
2359          * If there is no common prefix, we cannot do this as it would
2360          * underrun the input strings.
2361          */
2362         pfx_adjust_for_slash = (pfx_length ? 1 : 0);
2363         while (a + pfx_length - pfx_adjust_for_slash <= old_name &&
2364                b + pfx_length - pfx_adjust_for_slash <= new_name &&
2365                *old_name == *new_name) {
2366                 if (*old_name == '/')
2367                         sfx_length = len_a - (old_name - a);
2368                 old_name--;
2369                 new_name--;
2370         }
2371
2372         /*
2373          * pfx{mid-a => mid-b}sfx
2374          * {pfx-a => pfx-b}sfx
2375          * pfx{sfx-a => sfx-b}
2376          * name-a => name-b
2377          */
2378         a_midlen = len_a - pfx_length - sfx_length;
2379         b_midlen = len_b - pfx_length - sfx_length;
2380         if (a_midlen < 0)
2381                 a_midlen = 0;
2382         if (b_midlen < 0)
2383                 b_midlen = 0;
2384
2385         strbuf_grow(name, pfx_length + a_midlen + b_midlen + sfx_length + 7);
2386         if (pfx_length + sfx_length) {
2387                 strbuf_add(name, a, pfx_length);
2388                 strbuf_addch(name, '{');
2389         }
2390         strbuf_add(name, a + pfx_length, a_midlen);
2391         strbuf_addstr(name, " => ");
2392         strbuf_add(name, b + pfx_length, b_midlen);
2393         if (pfx_length + sfx_length) {
2394                 strbuf_addch(name, '}');
2395                 strbuf_add(name, a + len_a - sfx_length, sfx_length);
2396         }
2397 }
2398
2399 struct diffstat_t {
2400         int nr;
2401         int alloc;
2402         struct diffstat_file {
2403                 char *from_name;
2404                 char *name;
2405                 char *print_name;
2406                 const char *comments;
2407                 unsigned is_unmerged:1;
2408                 unsigned is_binary:1;
2409                 unsigned is_renamed:1;
2410                 unsigned is_interesting:1;
2411                 uintmax_t added, deleted;
2412         } **files;
2413 };
2414
2415 static struct diffstat_file *diffstat_add(struct diffstat_t *diffstat,
2416                                           const char *name_a,
2417                                           const char *name_b)
2418 {
2419         struct diffstat_file *x;
2420         x = xcalloc(1, sizeof(*x));
2421         ALLOC_GROW(diffstat->files, diffstat->nr + 1, diffstat->alloc);
2422         diffstat->files[diffstat->nr++] = x;
2423         if (name_b) {
2424                 x->from_name = xstrdup(name_a);
2425                 x->name = xstrdup(name_b);
2426                 x->is_renamed = 1;
2427         }
2428         else {
2429                 x->from_name = NULL;
2430                 x->name = xstrdup(name_a);
2431         }
2432         return x;
2433 }
2434
2435 static void diffstat_consume(void *priv, char *line, unsigned long len)
2436 {
2437         struct diffstat_t *diffstat = priv;
2438         struct diffstat_file *x = diffstat->files[diffstat->nr - 1];
2439
2440         if (line[0] == '+')
2441                 x->added++;
2442         else if (line[0] == '-')
2443                 x->deleted++;
2444 }
2445
2446 const char mime_boundary_leader[] = "------------";
2447
2448 static int scale_linear(int it, int width, int max_change)
2449 {
2450         if (!it)
2451                 return 0;
2452         /*
2453          * make sure that at least one '-' or '+' is printed if
2454          * there is any change to this path. The easiest way is to
2455          * scale linearly as if the alloted width is one column shorter
2456          * than it is, and then add 1 to the result.
2457          */
2458         return 1 + (it * (width - 1) / max_change);
2459 }
2460
2461 static void show_graph(struct strbuf *out, char ch, int cnt,
2462                        const char *set, const char *reset)
2463 {
2464         if (cnt <= 0)
2465                 return;
2466         strbuf_addstr(out, set);
2467         strbuf_addchars(out, ch, cnt);
2468         strbuf_addstr(out, reset);
2469 }
2470
2471 static void fill_print_name(struct diffstat_file *file)
2472 {
2473         struct strbuf pname = STRBUF_INIT;
2474
2475         if (file->print_name)
2476                 return;
2477
2478         if (file->is_renamed)
2479                 pprint_rename(&pname, file->from_name, file->name);
2480         else
2481                 quote_c_style(file->name, &pname, NULL, 0);
2482
2483         if (file->comments)
2484                 strbuf_addf(&pname, " (%s)", file->comments);
2485
2486         file->print_name = strbuf_detach(&pname, NULL);
2487 }
2488
2489 static void print_stat_summary_inserts_deletes(struct diff_options *options,
2490                 int files, int insertions, int deletions)
2491 {
2492         struct strbuf sb = STRBUF_INIT;
2493
2494         if (!files) {
2495                 assert(insertions == 0 && deletions == 0);
2496                 emit_diff_symbol(options, DIFF_SYMBOL_STATS_SUMMARY_NO_FILES,
2497                                  NULL, 0, 0);
2498                 return;
2499         }
2500
2501         strbuf_addf(&sb,
2502                     (files == 1) ? " %d file changed" : " %d files changed",
2503                     files);
2504
2505         /*
2506          * For binary diff, the caller may want to print "x files
2507          * changed" with insertions == 0 && deletions == 0.
2508          *
2509          * Not omitting "0 insertions(+), 0 deletions(-)" in this case
2510          * is probably less confusing (i.e skip over "2 files changed
2511          * but nothing about added/removed lines? Is this a bug in Git?").
2512          */
2513         if (insertions || deletions == 0) {
2514                 strbuf_addf(&sb,
2515                             (insertions == 1) ? ", %d insertion(+)" : ", %d insertions(+)",
2516                             insertions);
2517         }
2518
2519         if (deletions || insertions == 0) {
2520                 strbuf_addf(&sb,
2521                             (deletions == 1) ? ", %d deletion(-)" : ", %d deletions(-)",
2522                             deletions);
2523         }
2524         strbuf_addch(&sb, '\n');
2525         emit_diff_symbol(options, DIFF_SYMBOL_STATS_SUMMARY_INSERTS_DELETES,
2526                          sb.buf, sb.len, 0);
2527         strbuf_release(&sb);
2528 }
2529
2530 void print_stat_summary(FILE *fp, int files,
2531                         int insertions, int deletions)
2532 {
2533         struct diff_options o;
2534         memset(&o, 0, sizeof(o));
2535         o.file = fp;
2536
2537         print_stat_summary_inserts_deletes(&o, files, insertions, deletions);
2538 }
2539
2540 static void show_stats(struct diffstat_t *data, struct diff_options *options)
2541 {
2542         int i, len, add, del, adds = 0, dels = 0;
2543         uintmax_t max_change = 0, max_len = 0;
2544         int total_files = data->nr, count;
2545         int width, name_width, graph_width, number_width = 0, bin_width = 0;
2546         const char *reset, *add_c, *del_c;
2547         int extra_shown = 0;
2548         const char *line_prefix = diff_line_prefix(options);
2549         struct strbuf out = STRBUF_INIT;
2550
2551         if (data->nr == 0)
2552                 return;
2553
2554         count = options->stat_count ? options->stat_count : data->nr;
2555
2556         reset = diff_get_color_opt(options, DIFF_RESET);
2557         add_c = diff_get_color_opt(options, DIFF_FILE_NEW);
2558         del_c = diff_get_color_opt(options, DIFF_FILE_OLD);
2559
2560         /*
2561          * Find the longest filename and max number of changes
2562          */
2563         for (i = 0; (i < count) && (i < data->nr); i++) {
2564                 struct diffstat_file *file = data->files[i];
2565                 uintmax_t change = file->added + file->deleted;
2566
2567                 if (!file->is_interesting && (change == 0)) {
2568                         count++; /* not shown == room for one more */
2569                         continue;
2570                 }
2571                 fill_print_name(file);
2572                 len = strlen(file->print_name);
2573                 if (max_len < len)
2574                         max_len = len;
2575
2576                 if (file->is_unmerged) {
2577                         /* "Unmerged" is 8 characters */
2578                         bin_width = bin_width < 8 ? 8 : bin_width;
2579                         continue;
2580                 }
2581                 if (file->is_binary) {
2582                         /* "Bin XXX -> YYY bytes" */
2583                         int w = 14 + decimal_width(file->added)
2584                                 + decimal_width(file->deleted);
2585                         bin_width = bin_width < w ? w : bin_width;
2586                         /* Display change counts aligned with "Bin" */
2587                         number_width = 3;
2588                         continue;
2589                 }
2590
2591                 if (max_change < change)
2592                         max_change = change;
2593         }
2594         count = i; /* where we can stop scanning in data->files[] */
2595
2596         /*
2597          * We have width = stat_width or term_columns() columns total.
2598          * We want a maximum of min(max_len, stat_name_width) for the name part.
2599          * We want a maximum of min(max_change, stat_graph_width) for the +- part.
2600          * We also need 1 for " " and 4 + decimal_width(max_change)
2601          * for " | NNNN " and one the empty column at the end, altogether
2602          * 6 + decimal_width(max_change).
2603          *
2604          * If there's not enough space, we will use the smaller of
2605          * stat_name_width (if set) and 5/8*width for the filename,
2606          * and the rest for constant elements + graph part, but no more
2607          * than stat_graph_width for the graph part.
2608          * (5/8 gives 50 for filename and 30 for the constant parts + graph
2609          * for the standard terminal size).
2610          *
2611          * In other words: stat_width limits the maximum width, and
2612          * stat_name_width fixes the maximum width of the filename,
2613          * and is also used to divide available columns if there
2614          * aren't enough.
2615          *
2616          * Binary files are displayed with "Bin XXX -> YYY bytes"
2617          * instead of the change count and graph. This part is treated
2618          * similarly to the graph part, except that it is not
2619          * "scaled". If total width is too small to accommodate the
2620          * guaranteed minimum width of the filename part and the
2621          * separators and this message, this message will "overflow"
2622          * making the line longer than the maximum width.
2623          */
2624
2625         if (options->stat_width == -1)
2626                 width = term_columns() - strlen(line_prefix);
2627         else
2628                 width = options->stat_width ? options->stat_width : 80;
2629         number_width = decimal_width(max_change) > number_width ?
2630                 decimal_width(max_change) : number_width;
2631
2632         if (options->stat_graph_width == -1)
2633                 options->stat_graph_width = diff_stat_graph_width;
2634
2635         /*
2636          * Guarantee 3/8*16==6 for the graph part
2637          * and 5/8*16==10 for the filename part
2638          */
2639         if (width < 16 + 6 + number_width)
2640                 width = 16 + 6 + number_width;
2641
2642         /*
2643          * First assign sizes that are wanted, ignoring available width.
2644          * strlen("Bin XXX -> YYY bytes") == bin_width, and the part
2645          * starting from "XXX" should fit in graph_width.
2646          */
2647         graph_width = max_change + 4 > bin_width ? max_change : bin_width - 4;
2648         if (options->stat_graph_width &&
2649             options->stat_graph_width < graph_width)
2650                 graph_width = options->stat_graph_width;
2651
2652         name_width = (options->stat_name_width > 0 &&
2653                       options->stat_name_width < max_len) ?
2654                 options->stat_name_width : max_len;
2655
2656         /*
2657          * Adjust adjustable widths not to exceed maximum width
2658          */
2659         if (name_width + number_width + 6 + graph_width > width) {
2660                 if (graph_width > width * 3/8 - number_width - 6) {
2661                         graph_width = width * 3/8 - number_width - 6;
2662                         if (graph_width < 6)
2663                                 graph_width = 6;
2664                 }
2665
2666                 if (options->stat_graph_width &&
2667                     graph_width > options->stat_graph_width)
2668                         graph_width = options->stat_graph_width;
2669                 if (name_width > width - number_width - 6 - graph_width)
2670                         name_width = width - number_width - 6 - graph_width;
2671                 else
2672                         graph_width = width - number_width - 6 - name_width;
2673         }
2674
2675         /*
2676          * From here name_width is the width of the name area,
2677          * and graph_width is the width of the graph area.
2678          * max_change is used to scale graph properly.
2679          */
2680         for (i = 0; i < count; i++) {
2681                 const char *prefix = "";
2682                 struct diffstat_file *file = data->files[i];
2683                 char *name = file->print_name;
2684                 uintmax_t added = file->added;
2685                 uintmax_t deleted = file->deleted;
2686                 int name_len;
2687
2688                 if (!file->is_interesting && (added + deleted == 0))
2689                         continue;
2690
2691                 /*
2692                  * "scale" the filename
2693                  */
2694                 len = name_width;
2695                 name_len = strlen(name);
2696                 if (name_width < name_len) {
2697                         char *slash;
2698                         prefix = "...";
2699                         len -= 3;
2700                         name += name_len - len;
2701                         slash = strchr(name, '/');
2702                         if (slash)
2703                                 name = slash;
2704                 }
2705
2706                 if (file->is_binary) {
2707                         strbuf_addf(&out, " %s%-*s |", prefix, len, name);
2708                         strbuf_addf(&out, " %*s", number_width, "Bin");
2709                         if (!added && !deleted) {
2710                                 strbuf_addch(&out, '\n');
2711                                 emit_diff_symbol(options, DIFF_SYMBOL_STATS_LINE,
2712                                                  out.buf, out.len, 0);
2713                                 strbuf_reset(&out);
2714                                 continue;
2715                         }
2716                         strbuf_addf(&out, " %s%"PRIuMAX"%s",
2717                                 del_c, deleted, reset);
2718                         strbuf_addstr(&out, " -> ");
2719                         strbuf_addf(&out, "%s%"PRIuMAX"%s",
2720                                 add_c, added, reset);
2721                         strbuf_addstr(&out, " bytes\n");
2722                         emit_diff_symbol(options, DIFF_SYMBOL_STATS_LINE,
2723                                          out.buf, out.len, 0);
2724                         strbuf_reset(&out);
2725                         continue;
2726                 }
2727                 else if (file->is_unmerged) {
2728                         strbuf_addf(&out, " %s%-*s |", prefix, len, name);
2729                         strbuf_addstr(&out, " Unmerged\n");
2730                         emit_diff_symbol(options, DIFF_SYMBOL_STATS_LINE,
2731                                          out.buf, out.len, 0);
2732                         strbuf_reset(&out);
2733                         continue;
2734                 }
2735
2736                 /*
2737                  * scale the add/delete
2738                  */
2739                 add = added;
2740                 del = deleted;
2741
2742                 if (graph_width <= max_change) {
2743                         int total = scale_linear(add + del, graph_width, max_change);
2744                         if (total < 2 && add && del)
2745                                 /* width >= 2 due to the sanity check */
2746                                 total = 2;
2747                         if (add < del) {
2748                                 add = scale_linear(add, graph_width, max_change);
2749                                 del = total - add;
2750                         } else {
2751                                 del = scale_linear(del, graph_width, max_change);
2752                                 add = total - del;
2753                         }
2754                 }
2755                 strbuf_addf(&out, " %s%-*s |", prefix, len, name);
2756                 strbuf_addf(&out, " %*"PRIuMAX"%s",
2757                         number_width, added + deleted,
2758                         added + deleted ? " " : "");
2759                 show_graph(&out, '+', add, add_c, reset);
2760                 show_graph(&out, '-', del, del_c, reset);
2761                 strbuf_addch(&out, '\n');
2762                 emit_diff_symbol(options, DIFF_SYMBOL_STATS_LINE,
2763                                  out.buf, out.len, 0);
2764                 strbuf_reset(&out);
2765         }
2766
2767         for (i = 0; i < data->nr; i++) {
2768                 struct diffstat_file *file = data->files[i];
2769                 uintmax_t added = file->added;
2770                 uintmax_t deleted = file->deleted;
2771
2772                 if (file->is_unmerged ||
2773                     (!file->is_interesting && (added + deleted == 0))) {
2774                         total_files--;
2775                         continue;
2776                 }
2777
2778                 if (!file->is_binary) {
2779                         adds += added;
2780                         dels += deleted;
2781                 }
2782                 if (i < count)
2783                         continue;
2784                 if (!extra_shown)
2785                         emit_diff_symbol(options,
2786                                          DIFF_SYMBOL_STATS_SUMMARY_ABBREV,
2787                                          NULL, 0, 0);
2788                 extra_shown = 1;
2789         }
2790
2791         print_stat_summary_inserts_deletes(options, total_files, adds, dels);
2792         strbuf_release(&out);
2793 }
2794
2795 static void show_shortstats(struct diffstat_t *data, struct diff_options *options)
2796 {
2797         int i, adds = 0, dels = 0, total_files = data->nr;
2798
2799         if (data->nr == 0)
2800                 return;
2801
2802         for (i = 0; i < data->nr; i++) {
2803                 int added = data->files[i]->added;
2804                 int deleted = data->files[i]->deleted;
2805
2806                 if (data->files[i]->is_unmerged ||
2807                     (!data->files[i]->is_interesting && (added + deleted == 0))) {
2808                         total_files--;
2809                 } else if (!data->files[i]->is_binary) { /* don't count bytes */
2810                         adds += added;
2811                         dels += deleted;
2812                 }
2813         }
2814         print_stat_summary_inserts_deletes(options, total_files, adds, dels);
2815 }
2816
2817 static void show_numstat(struct diffstat_t *data, struct diff_options *options)
2818 {
2819         int i;
2820
2821         if (data->nr == 0)
2822                 return;
2823
2824         for (i = 0; i < data->nr; i++) {
2825                 struct diffstat_file *file = data->files[i];
2826
2827                 fprintf(options->file, "%s", diff_line_prefix(options));
2828
2829                 if (file->is_binary)
2830                         fprintf(options->file, "-\t-\t");
2831                 else
2832                         fprintf(options->file,
2833                                 "%"PRIuMAX"\t%"PRIuMAX"\t",
2834                                 file->added, file->deleted);
2835                 if (options->line_termination) {
2836                         fill_print_name(file);
2837                         if (!file->is_renamed)
2838                                 write_name_quoted(file->name, options->file,
2839                                                   options->line_termination);
2840                         else {
2841                                 fputs(file->print_name, options->file);
2842                                 putc(options->line_termination, options->file);
2843                         }
2844                 } else {
2845                         if (file->is_renamed) {
2846                                 putc('\0', options->file);
2847                                 write_name_quoted(file->from_name, options->file, '\0');
2848                         }
2849                         write_name_quoted(file->name, options->file, '\0');
2850                 }
2851         }
2852 }
2853
2854 struct dirstat_file {
2855         const char *name;
2856         unsigned long changed;
2857 };
2858
2859 struct dirstat_dir {
2860         struct dirstat_file *files;
2861         int alloc, nr, permille, cumulative;
2862 };
2863
2864 static long gather_dirstat(struct diff_options *opt, struct dirstat_dir *dir,
2865                 unsigned long changed, const char *base, int baselen)
2866 {
2867         unsigned long sum_changes = 0;
2868         unsigned int sources = 0;
2869         const char *line_prefix = diff_line_prefix(opt);
2870
2871         while (dir->nr) {
2872                 struct dirstat_file *f = dir->files;
2873                 int namelen = strlen(f->name);
2874                 unsigned long changes;
2875                 char *slash;
2876
2877                 if (namelen < baselen)
2878                         break;
2879                 if (memcmp(f->name, base, baselen))
2880                         break;
2881                 slash = strchr(f->name + baselen, '/');
2882                 if (slash) {
2883                         int newbaselen = slash + 1 - f->name;
2884                         changes = gather_dirstat(opt, dir, changed, f->name, newbaselen);
2885                         sources++;
2886                 } else {
2887                         changes = f->changed;
2888                         dir->files++;
2889                         dir->nr--;
2890                         sources += 2;
2891                 }
2892                 sum_changes += changes;
2893         }
2894
2895         /*
2896          * We don't report dirstat's for
2897          *  - the top level
2898          *  - or cases where everything came from a single directory
2899          *    under this directory (sources == 1).
2900          */
2901         if (baselen && sources != 1) {
2902                 if (sum_changes) {
2903                         int permille = sum_changes * 1000 / changed;
2904                         if (permille >= dir->permille) {
2905                                 fprintf(opt->file, "%s%4d.%01d%% %.*s\n", line_prefix,
2906                                         permille / 10, permille % 10, baselen, base);
2907                                 if (!dir->cumulative)
2908                                         return 0;
2909                         }
2910                 }
2911         }
2912         return sum_changes;
2913 }
2914
2915 static int dirstat_compare(const void *_a, const void *_b)
2916 {
2917         const struct dirstat_file *a = _a;
2918         const struct dirstat_file *b = _b;
2919         return strcmp(a->name, b->name);
2920 }
2921
2922 static void show_dirstat(struct diff_options *options)
2923 {
2924         int i;
2925         unsigned long changed;
2926         struct dirstat_dir dir;
2927         struct diff_queue_struct *q = &diff_queued_diff;
2928
2929         dir.files = NULL;
2930         dir.alloc = 0;
2931         dir.nr = 0;
2932         dir.permille = options->dirstat_permille;
2933         dir.cumulative = options->flags.dirstat_cumulative;
2934
2935         changed = 0;
2936         for (i = 0; i < q->nr; i++) {
2937                 struct diff_filepair *p = q->queue[i];
2938                 const char *name;
2939                 unsigned long copied, added, damage;
2940                 int content_changed;
2941
2942                 name = p->two->path ? p->two->path : p->one->path;
2943
2944                 if (p->one->oid_valid && p->two->oid_valid)
2945                         content_changed = oidcmp(&p->one->oid, &p->two->oid);
2946                 else
2947                         content_changed = 1;
2948
2949                 if (!content_changed) {
2950                         /*
2951                          * The SHA1 has not changed, so pre-/post-content is
2952                          * identical. We can therefore skip looking at the
2953                          * file contents altogether.
2954                          */
2955                         damage = 0;
2956                         goto found_damage;
2957                 }
2958
2959                 if (options->flags.dirstat_by_file) {
2960                         /*
2961                          * In --dirstat-by-file mode, we don't really need to
2962                          * look at the actual file contents at all.
2963                          * The fact that the SHA1 changed is enough for us to
2964                          * add this file to the list of results
2965                          * (with each file contributing equal damage).
2966                          */
2967                         damage = 1;
2968                         goto found_damage;
2969                 }
2970
2971                 if (DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
2972                         diff_populate_filespec(options->repo, p->one, 0);
2973                         diff_populate_filespec(options->repo, p->two, 0);
2974                         diffcore_count_changes(options->repo,
2975                                                p->one, p->two, NULL, NULL,
2976                                                &copied, &added);
2977                         diff_free_filespec_data(p->one);
2978                         diff_free_filespec_data(p->two);
2979                 } else if (DIFF_FILE_VALID(p->one)) {
2980                         diff_populate_filespec(options->repo, p->one, CHECK_SIZE_ONLY);
2981                         copied = added = 0;
2982                         diff_free_filespec_data(p->one);
2983                 } else if (DIFF_FILE_VALID(p->two)) {
2984                         diff_populate_filespec(options->repo, p->two, CHECK_SIZE_ONLY);
2985                         copied = 0;
2986                         added = p->two->size;
2987                         diff_free_filespec_data(p->two);
2988                 } else
2989                         continue;
2990
2991                 /*
2992                  * Original minus copied is the removed material,
2993                  * added is the new material.  They are both damages
2994                  * made to the preimage.
2995                  * If the resulting damage is zero, we know that
2996                  * diffcore_count_changes() considers the two entries to
2997                  * be identical, but since content_changed is true, we
2998                  * know that there must have been _some_ kind of change,
2999                  * so we force all entries to have damage > 0.
3000                  */
3001                 damage = (p->one->size - copied) + added;
3002                 if (!damage)
3003                         damage = 1;
3004
3005 found_damage:
3006                 ALLOC_GROW(dir.files, dir.nr + 1, dir.alloc);
3007                 dir.files[dir.nr].name = name;
3008                 dir.files[dir.nr].changed = damage;
3009                 changed += damage;
3010                 dir.nr++;
3011         }
3012
3013         /* This can happen even with many files, if everything was renames */
3014         if (!changed)
3015                 return;
3016
3017         /* Show all directories with more than x% of the changes */
3018         QSORT(dir.files, dir.nr, dirstat_compare);
3019         gather_dirstat(options, &dir, changed, "", 0);
3020 }
3021
3022 static void show_dirstat_by_line(struct diffstat_t *data, struct diff_options *options)
3023 {
3024         int i;
3025         unsigned long changed;
3026         struct dirstat_dir dir;
3027
3028         if (data->nr == 0)
3029                 return;
3030
3031         dir.files = NULL;
3032         dir.alloc = 0;
3033         dir.nr = 0;
3034         dir.permille = options->dirstat_permille;
3035         dir.cumulative = options->flags.dirstat_cumulative;
3036
3037         changed = 0;
3038         for (i = 0; i < data->nr; i++) {
3039                 struct diffstat_file *file = data->files[i];
3040                 unsigned long damage = file->added + file->deleted;
3041                 if (file->is_binary)
3042                         /*
3043                          * binary files counts bytes, not lines. Must find some
3044                          * way to normalize binary bytes vs. textual lines.
3045                          * The following heuristic assumes that there are 64
3046                          * bytes per "line".
3047                          * This is stupid and ugly, but very cheap...
3048                          */
3049                         damage = DIV_ROUND_UP(damage, 64);
3050                 ALLOC_GROW(dir.files, dir.nr + 1, dir.alloc);
3051                 dir.files[dir.nr].name = file->name;
3052                 dir.files[dir.nr].changed = damage;
3053                 changed += damage;
3054                 dir.nr++;
3055         }
3056
3057         /* This can happen even with many files, if everything was renames */
3058         if (!changed)
3059                 return;
3060
3061         /* Show all directories with more than x% of the changes */
3062         QSORT(dir.files, dir.nr, dirstat_compare);
3063         gather_dirstat(options, &dir, changed, "", 0);
3064 }
3065
3066 static void free_diffstat_info(struct diffstat_t *diffstat)
3067 {
3068         int i;
3069         for (i = 0; i < diffstat->nr; i++) {
3070                 struct diffstat_file *f = diffstat->files[i];
3071                 free(f->print_name);
3072                 free(f->name);
3073                 free(f->from_name);
3074                 free(f);
3075         }
3076         free(diffstat->files);
3077 }
3078
3079 struct checkdiff_t {
3080         const char *filename;
3081         int lineno;
3082         int conflict_marker_size;
3083         struct diff_options *o;
3084         unsigned ws_rule;
3085         unsigned status;
3086 };
3087
3088 static int is_conflict_marker(const char *line, int marker_size, unsigned long len)
3089 {
3090         char firstchar;
3091         int cnt;
3092
3093         if (len < marker_size + 1)
3094                 return 0;
3095         firstchar = line[0];
3096         switch (firstchar) {
3097         case '=': case '>': case '<': case '|':
3098                 break;
3099         default:
3100                 return 0;
3101         }
3102         for (cnt = 1; cnt < marker_size; cnt++)
3103                 if (line[cnt] != firstchar)
3104                         return 0;
3105         /* line[1] thru line[marker_size-1] are same as firstchar */
3106         if (len < marker_size + 1 || !isspace(line[marker_size]))
3107                 return 0;
3108         return 1;
3109 }
3110
3111 static void checkdiff_consume(void *priv, char *line, unsigned long len)
3112 {
3113         struct checkdiff_t *data = priv;
3114         int marker_size = data->conflict_marker_size;
3115         const char *ws = diff_get_color(data->o->use_color, DIFF_WHITESPACE);
3116         const char *reset = diff_get_color(data->o->use_color, DIFF_RESET);
3117         const char *set = diff_get_color(data->o->use_color, DIFF_FILE_NEW);
3118         char *err;
3119         const char *line_prefix;
3120
3121         assert(data->o);
3122         line_prefix = diff_line_prefix(data->o);
3123
3124         if (line[0] == '+') {
3125                 unsigned bad;
3126                 data->lineno++;
3127                 if (is_conflict_marker(line + 1, marker_size, len - 1)) {
3128                         data->status |= 1;
3129                         fprintf(data->o->file,
3130                                 "%s%s:%d: leftover conflict marker\n",
3131                                 line_prefix, data->filename, data->lineno);
3132                 }
3133                 bad = ws_check(line + 1, len - 1, data->ws_rule);
3134                 if (!bad)
3135                         return;
3136                 data->status |= bad;
3137                 err = whitespace_error_string(bad);
3138                 fprintf(data->o->file, "%s%s:%d: %s.\n",
3139                         line_prefix, data->filename, data->lineno, err);
3140                 free(err);
3141                 emit_line(data->o, set, reset, line, 1);
3142                 ws_check_emit(line + 1, len - 1, data->ws_rule,
3143                               data->o->file, set, reset, ws);
3144         } else if (line[0] == ' ') {
3145                 data->lineno++;
3146         } else if (line[0] == '@') {
3147                 char *plus = strchr(line, '+');
3148                 if (plus)
3149                         data->lineno = strtol(plus, NULL, 10) - 1;
3150                 else
3151                         die("invalid diff");
3152         }
3153 }
3154
3155 static unsigned char *deflate_it(char *data,
3156                                  unsigned long size,
3157                                  unsigned long *result_size)
3158 {
3159         int bound;
3160         unsigned char *deflated;
3161         git_zstream stream;
3162
3163         git_deflate_init(&stream, zlib_compression_level);
3164         bound = git_deflate_bound(&stream, size);
3165         deflated = xmalloc(bound);
3166         stream.next_out = deflated;
3167         stream.avail_out = bound;
3168
3169         stream.next_in = (unsigned char *)data;
3170         stream.avail_in = size;
3171         while (git_deflate(&stream, Z_FINISH) == Z_OK)
3172                 ; /* nothing */
3173         git_deflate_end(&stream);
3174         *result_size = stream.total_out;
3175         return deflated;
3176 }
3177
3178 static void emit_binary_diff_body(struct diff_options *o,
3179                                   mmfile_t *one, mmfile_t *two)
3180 {
3181         void *cp;
3182         void *delta;
3183         void *deflated;
3184         void *data;
3185         unsigned long orig_size;
3186         unsigned long delta_size;
3187         unsigned long deflate_size;
3188         unsigned long data_size;
3189
3190         /* We could do deflated delta, or we could do just deflated two,
3191          * whichever is smaller.
3192          */
3193         delta = NULL;
3194         deflated = deflate_it(two->ptr, two->size, &deflate_size);
3195         if (one->size && two->size) {
3196                 delta = diff_delta(one->ptr, one->size,
3197                                    two->ptr, two->size,
3198                                    &delta_size, deflate_size);
3199                 if (delta) {
3200                         void *to_free = delta;
3201                         orig_size = delta_size;
3202                         delta = deflate_it(delta, delta_size, &delta_size);
3203                         free(to_free);
3204                 }
3205         }
3206
3207         if (delta && delta_size < deflate_size) {
3208                 char *s = xstrfmt("%lu", orig_size);
3209                 emit_diff_symbol(o, DIFF_SYMBOL_BINARY_DIFF_HEADER_DELTA,
3210                                  s, strlen(s), 0);
3211                 free(s);
3212                 free(deflated);
3213                 data = delta;
3214                 data_size = delta_size;
3215         } else {
3216                 char *s = xstrfmt("%lu", two->size);
3217                 emit_diff_symbol(o, DIFF_SYMBOL_BINARY_DIFF_HEADER_LITERAL,
3218                                  s, strlen(s), 0);
3219                 free(s);
3220                 free(delta);
3221                 data = deflated;
3222                 data_size = deflate_size;
3223         }
3224
3225         /* emit data encoded in base85 */
3226         cp = data;
3227         while (data_size) {
3228                 int len;
3229                 int bytes = (52 < data_size) ? 52 : data_size;
3230                 char line[71];
3231                 data_size -= bytes;
3232                 if (bytes <= 26)
3233                         line[0] = bytes + 'A' - 1;
3234                 else
3235                         line[0] = bytes - 26 + 'a' - 1;
3236                 encode_85(line + 1, cp, bytes);
3237                 cp = (char *) cp + bytes;
3238
3239                 len = strlen(line);
3240                 line[len++] = '\n';
3241                 line[len] = '\0';
3242
3243                 emit_diff_symbol(o, DIFF_SYMBOL_BINARY_DIFF_BODY,
3244                                  line, len, 0);
3245         }
3246         emit_diff_symbol(o, DIFF_SYMBOL_BINARY_DIFF_FOOTER, NULL, 0, 0);
3247         free(data);
3248 }
3249
3250 static void emit_binary_diff(struct diff_options *o,
3251                              mmfile_t *one, mmfile_t *two)
3252 {
3253         emit_diff_symbol(o, DIFF_SYMBOL_BINARY_DIFF_HEADER, NULL, 0, 0);
3254         emit_binary_diff_body(o, one, two);
3255         emit_binary_diff_body(o, two, one);
3256 }
3257
3258 int diff_filespec_is_binary(struct repository *r,
3259                             struct diff_filespec *one)
3260 {
3261         if (one->is_binary == -1) {
3262                 diff_filespec_load_driver(one, r->index);
3263                 if (one->driver->binary != -1)
3264                         one->is_binary = one->driver->binary;
3265                 else {
3266                         if (!one->data && DIFF_FILE_VALID(one))
3267                                 diff_populate_filespec(r, one, CHECK_BINARY);
3268                         if (one->is_binary == -1 && one->data)
3269                                 one->is_binary = buffer_is_binary(one->data,
3270                                                 one->size);
3271                         if (one->is_binary == -1)
3272                                 one->is_binary = 0;
3273                 }
3274         }
3275         return one->is_binary;
3276 }
3277
3278 static const struct userdiff_funcname *
3279 diff_funcname_pattern(struct diff_options *o, struct diff_filespec *one)
3280 {
3281         diff_filespec_load_driver(one, o->repo->index);
3282         return one->driver->funcname.pattern ? &one->driver->funcname : NULL;
3283 }
3284
3285 void diff_set_mnemonic_prefix(struct diff_options *options, const char *a, const char *b)
3286 {
3287         if (!options->a_prefix)
3288                 options->a_prefix = a;
3289         if (!options->b_prefix)
3290                 options->b_prefix = b;
3291 }
3292
3293 struct userdiff_driver *get_textconv(struct index_state *istate,
3294                                      struct diff_filespec *one)
3295 {
3296         if (!DIFF_FILE_VALID(one))
3297                 return NULL;
3298
3299         diff_filespec_load_driver(one, istate);
3300         return userdiff_get_textconv(one->driver);
3301 }
3302
3303 static void builtin_diff(const char *name_a,
3304                          const char *name_b,
3305                          struct diff_filespec *one,
3306                          struct diff_filespec *two,
3307                          const char *xfrm_msg,
3308                          int must_show_header,
3309                          struct diff_options *o,
3310                          int complete_rewrite)
3311 {
3312         mmfile_t mf1, mf2;
3313         const char *lbl[2];
3314         char *a_one, *b_two;
3315         const char *meta = diff_get_color_opt(o, DIFF_METAINFO);
3316         const char *reset = diff_get_color_opt(o, DIFF_RESET);
3317         const char *a_prefix, *b_prefix;
3318         struct userdiff_driver *textconv_one = NULL;
3319         struct userdiff_driver *textconv_two = NULL;
3320         struct strbuf header = STRBUF_INIT;
3321         const char *line_prefix = diff_line_prefix(o);
3322
3323         diff_set_mnemonic_prefix(o, "a/", "b/");
3324         if (o->flags.reverse_diff) {
3325                 a_prefix = o->b_prefix;
3326                 b_prefix = o->a_prefix;
3327         } else {
3328                 a_prefix = o->a_prefix;
3329                 b_prefix = o->b_prefix;
3330         }
3331
3332         if (o->submodule_format == DIFF_SUBMODULE_LOG &&
3333             (!one->mode || S_ISGITLINK(one->mode)) &&
3334             (!two->mode || S_ISGITLINK(two->mode))) {
3335                 show_submodule_summary(o, one->path ? one->path : two->path,
3336                                 &one->oid, &two->oid,
3337                                 two->dirty_submodule);
3338                 return;
3339         } else if (o->submodule_format == DIFF_SUBMODULE_INLINE_DIFF &&
3340                    (!one->mode || S_ISGITLINK(one->mode)) &&
3341                    (!two->mode || S_ISGITLINK(two->mode))) {
3342                 show_submodule_inline_diff(o, one->path ? one->path : two->path,
3343                                 &one->oid, &two->oid,
3344                                 two->dirty_submodule);
3345                 return;
3346         }
3347
3348         if (o->flags.allow_textconv) {
3349                 textconv_one = get_textconv(o->repo->index, one);
3350                 textconv_two = get_textconv(o->repo->index, two);
3351         }
3352
3353         /* Never use a non-valid filename anywhere if at all possible */
3354         name_a = DIFF_FILE_VALID(one) ? name_a : name_b;
3355         name_b = DIFF_FILE_VALID(two) ? name_b : name_a;
3356
3357         a_one = quote_two(a_prefix, name_a + (*name_a == '/'));
3358         b_two = quote_two(b_prefix, name_b + (*name_b == '/'));
3359         lbl[0] = DIFF_FILE_VALID(one) ? a_one : "/dev/null";
3360         lbl[1] = DIFF_FILE_VALID(two) ? b_two : "/dev/null";
3361         strbuf_addf(&header, "%s%sdiff --git %s %s%s\n", line_prefix, meta, a_one, b_two, reset);
3362         if (lbl[0][0] == '/') {
3363                 /* /dev/null */
3364                 strbuf_addf(&header, "%s%snew file mode %06o%s\n", line_prefix, meta, two->mode, reset);
3365                 if (xfrm_msg)
3366                         strbuf_addstr(&header, xfrm_msg);
3367                 must_show_header = 1;
3368         }
3369         else if (lbl[1][0] == '/') {
3370                 strbuf_addf(&header, "%s%sdeleted file mode %06o%s\n", line_prefix, meta, one->mode, reset);
3371                 if (xfrm_msg)
3372                         strbuf_addstr(&header, xfrm_msg);
3373                 must_show_header = 1;
3374         }
3375         else {
3376                 if (one->mode != two->mode) {
3377                         strbuf_addf(&header, "%s%sold mode %06o%s\n", line_prefix, meta, one->mode, reset);
3378                         strbuf_addf(&header, "%s%snew mode %06o%s\n", line_prefix, meta, two->mode, reset);
3379                         must_show_header = 1;
3380                 }
3381                 if (xfrm_msg)
3382                         strbuf_addstr(&header, xfrm_msg);
3383
3384                 /*
3385                  * we do not run diff between different kind
3386                  * of objects.
3387                  */
3388                 if ((one->mode ^ two->mode) & S_IFMT)
3389                         goto free_ab_and_return;
3390                 if (complete_rewrite &&
3391                     (textconv_one || !diff_filespec_is_binary(o->repo, one)) &&
3392                     (textconv_two || !diff_filespec_is_binary(o->repo, two))) {
3393                         emit_diff_symbol(o, DIFF_SYMBOL_HEADER,
3394                                          header.buf, header.len, 0);
3395                         strbuf_reset(&header);
3396                         emit_rewrite_diff(name_a, name_b, one, two,
3397                                           textconv_one, textconv_two, o);
3398                         o->found_changes = 1;
3399                         goto free_ab_and_return;
3400                 }
3401         }
3402
3403         if (o->irreversible_delete && lbl[1][0] == '/') {
3404                 emit_diff_symbol(o, DIFF_SYMBOL_HEADER, header.buf,
3405                                  header.len, 0);
3406                 strbuf_reset(&header);
3407                 goto free_ab_and_return;
3408         } else if (!o->flags.text &&
3409                    ( (!textconv_one && diff_filespec_is_binary(o->repo, one)) ||
3410                      (!textconv_two && diff_filespec_is_binary(o->repo, two)) )) {
3411                 struct strbuf sb = STRBUF_INIT;
3412                 if (!one->data && !two->data &&
3413                     S_ISREG(one->mode) && S_ISREG(two->mode) &&
3414                     !o->flags.binary) {
3415                         if (!oidcmp(&one->oid, &two->oid)) {
3416                                 if (must_show_header)
3417                                         emit_diff_symbol(o, DIFF_SYMBOL_HEADER,
3418                                                          header.buf, header.len,
3419                                                          0);
3420                                 goto free_ab_and_return;
3421                         }
3422                         emit_diff_symbol(o, DIFF_SYMBOL_HEADER,
3423                                          header.buf, header.len, 0);
3424                         strbuf_addf(&sb, "%sBinary files %s and %s differ\n",
3425                                     diff_line_prefix(o), lbl[0], lbl[1]);
3426                         emit_diff_symbol(o, DIFF_SYMBOL_BINARY_FILES,
3427                                          sb.buf, sb.len, 0);
3428                         strbuf_release(&sb);
3429                         goto free_ab_and_return;
3430                 }
3431                 if (fill_mmfile(o->repo, &mf1, one) < 0 ||
3432                     fill_mmfile(o->repo, &mf2, two) < 0)
3433                         die("unable to read files to diff");
3434                 /* Quite common confusing case */
3435                 if (mf1.size == mf2.size &&
3436                     !memcmp(mf1.ptr, mf2.ptr, mf1.size)) {
3437                         if (must_show_header)
3438                                 emit_diff_symbol(o, DIFF_SYMBOL_HEADER,
3439                                                  header.buf, header.len, 0);
3440                         goto free_ab_and_return;
3441                 }
3442                 emit_diff_symbol(o, DIFF_SYMBOL_HEADER, header.buf, header.len, 0);
3443                 strbuf_reset(&header);
3444                 if (o->flags.binary)
3445                         emit_binary_diff(o, &mf1, &mf2);
3446                 else {
3447                         strbuf_addf(&sb, "%sBinary files %s and %s differ\n",
3448                                     diff_line_prefix(o), lbl[0], lbl[1]);
3449                         emit_diff_symbol(o, DIFF_SYMBOL_BINARY_FILES,
3450                                          sb.buf, sb.len, 0);
3451                         strbuf_release(&sb);
3452                 }
3453                 o->found_changes = 1;
3454         } else {
3455                 /* Crazy xdl interfaces.. */
3456                 const char *diffopts = getenv("GIT_DIFF_OPTS");
3457                 const char *v;
3458                 xpparam_t xpp;
3459                 xdemitconf_t xecfg;
3460                 struct emit_callback ecbdata;
3461                 const struct userdiff_funcname *pe;
3462
3463                 if (must_show_header) {
3464                         emit_diff_symbol(o, DIFF_SYMBOL_HEADER,
3465                                          header.buf, header.len, 0);
3466                         strbuf_reset(&header);
3467                 }
3468
3469                 mf1.size = fill_textconv(o->repo, textconv_one, one, &mf1.ptr);
3470                 mf2.size = fill_textconv(o->repo, textconv_two, two, &mf2.ptr);
3471
3472                 pe = diff_funcname_pattern(o, one);
3473                 if (!pe)
3474                         pe = diff_funcname_pattern(o, two);
3475
3476                 memset(&xpp, 0, sizeof(xpp));
3477                 memset(&xecfg, 0, sizeof(xecfg));
3478                 memset(&ecbdata, 0, sizeof(ecbdata));
3479                 if (o->flags.suppress_diff_headers)
3480                         lbl[0] = NULL;
3481                 ecbdata.label_path = lbl;
3482                 ecbdata.color_diff = want_color(o->use_color);
3483                 ecbdata.ws_rule = whitespace_rule(name_b);
3484                 if (ecbdata.ws_rule & WS_BLANK_AT_EOF)
3485                         check_blank_at_eof(&mf1, &mf2, &ecbdata);
3486                 ecbdata.opt = o;
3487                 if (header.len && !o->flags.suppress_diff_headers)
3488                         ecbdata.header = &header;
3489                 xpp.flags = o->xdl_opts;
3490                 xpp.anchors = o->anchors;
3491                 xpp.anchors_nr = o->anchors_nr;
3492                 xecfg.ctxlen = o->context;
3493                 xecfg.interhunkctxlen = o->interhunkcontext;
3494                 xecfg.flags = XDL_EMIT_FUNCNAMES;
3495                 if (o->flags.funccontext)
3496                         xecfg.flags |= XDL_EMIT_FUNCCONTEXT;
3497                 if (pe)
3498                         xdiff_set_find_func(&xecfg, pe->pattern, pe->cflags);
3499                 if (!diffopts)
3500                         ;
3501                 else if (skip_prefix(diffopts, "--unified=", &v))
3502                         xecfg.ctxlen = strtoul(v, NULL, 10);
3503                 else if (skip_prefix(diffopts, "-u", &v))
3504                         xecfg.ctxlen = strtoul(v, NULL, 10);
3505                 if (o->word_diff)
3506                         init_diff_words_data(&ecbdata, o, one, two);
3507                 if (xdi_diff_outf(&mf1, &mf2, fn_out_consume, &ecbdata,
3508                                   &xpp, &xecfg))
3509                         die("unable to generate diff for %s", one->path);
3510                 if (o->word_diff)
3511                         free_diff_words_data(&ecbdata);
3512                 if (textconv_one)
3513                         free(mf1.ptr);
3514                 if (textconv_two)
3515                         free(mf2.ptr);
3516                 xdiff_clear_find_func(&xecfg);
3517         }
3518
3519  free_ab_and_return:
3520         strbuf_release(&header);
3521         diff_free_filespec_data(one);
3522         diff_free_filespec_data(two);
3523         free(a_one);
3524         free(b_two);
3525         return;
3526 }
3527
3528 static char *get_compact_summary(const struct diff_filepair *p, int is_renamed)
3529 {
3530         if (!is_renamed) {
3531                 if (p->status == DIFF_STATUS_ADDED) {
3532                         if (S_ISLNK(p->two->mode))
3533                                 return "new +l";
3534                         else if ((p->two->mode & 0777) == 0755)
3535                                 return "new +x";
3536                         else
3537                                 return "new";
3538                 } else if (p->status == DIFF_STATUS_DELETED)
3539                         return "gone";
3540         }
3541         if (S_ISLNK(p->one->mode) && !S_ISLNK(p->two->mode))
3542                 return "mode -l";
3543         else if (!S_ISLNK(p->one->mode) && S_ISLNK(p->two->mode))
3544                 return "mode +l";
3545         else if ((p->one->mode & 0777) == 0644 &&
3546                  (p->two->mode & 0777) == 0755)
3547                 return "mode +x";
3548         else if ((p->one->mode & 0777) == 0755 &&
3549                  (p->two->mode & 0777) == 0644)
3550                 return "mode -x";
3551         return NULL;
3552 }
3553
3554 static void builtin_diffstat(const char *name_a, const char *name_b,
3555                              struct diff_filespec *one,
3556                              struct diff_filespec *two,
3557                              struct diffstat_t *diffstat,
3558                              struct diff_options *o,
3559                              struct diff_filepair *p)
3560 {
3561         mmfile_t mf1, mf2;
3562         struct diffstat_file *data;
3563         int same_contents;
3564         int complete_rewrite = 0;
3565
3566         if (!DIFF_PAIR_UNMERGED(p)) {
3567                 if (p->status == DIFF_STATUS_MODIFIED && p->score)
3568                         complete_rewrite = 1;
3569         }
3570
3571         data = diffstat_add(diffstat, name_a, name_b);
3572         data->is_interesting = p->status != DIFF_STATUS_UNKNOWN;
3573         if (o->flags.stat_with_summary)
3574                 data->comments = get_compact_summary(p, data->is_renamed);
3575
3576         if (!one || !two) {
3577                 data->is_unmerged = 1;
3578                 return;
3579         }
3580
3581         same_contents = !oidcmp(&one->oid, &two->oid);
3582
3583         if (diff_filespec_is_binary(o->repo, one) ||
3584             diff_filespec_is_binary(o->repo, two)) {
3585                 data->is_binary = 1;
3586                 if (same_contents) {
3587                         data->added = 0;
3588                         data->deleted = 0;
3589                 } else {
3590                         data->added = diff_filespec_size(o->repo, two);
3591                         data->deleted = diff_filespec_size(o->repo, one);
3592                 }
3593         }
3594
3595         else if (complete_rewrite) {
3596                 diff_populate_filespec(o->repo, one, 0);
3597                 diff_populate_filespec(o->repo, two, 0);
3598                 data->deleted = count_lines(one->data, one->size);
3599                 data->added = count_lines(two->data, two->size);
3600         }
3601
3602         else if (!same_contents) {
3603                 /* Crazy xdl interfaces.. */
3604                 xpparam_t xpp;
3605                 xdemitconf_t xecfg;
3606
3607                 if (fill_mmfile(o->repo, &mf1, one) < 0 ||
3608                     fill_mmfile(o->repo, &mf2, two) < 0)
3609                         die("unable to read files to diff");
3610
3611                 memset(&xpp, 0, sizeof(xpp));
3612                 memset(&xecfg, 0, sizeof(xecfg));
3613                 xpp.flags = o->xdl_opts;
3614                 xpp.anchors = o->anchors;
3615                 xpp.anchors_nr = o->anchors_nr;
3616                 xecfg.ctxlen = o->context;
3617                 xecfg.interhunkctxlen = o->interhunkcontext;
3618                 if (xdi_diff_outf(&mf1, &mf2, diffstat_consume, diffstat,
3619                                   &xpp, &xecfg))
3620                         die("unable to generate diffstat for %s", one->path);
3621         }
3622
3623         diff_free_filespec_data(one);
3624         diff_free_filespec_data(two);
3625 }
3626
3627 static void builtin_checkdiff(const char *name_a, const char *name_b,
3628                               const char *attr_path,
3629                               struct diff_filespec *one,
3630                               struct diff_filespec *two,
3631                               struct diff_options *o)
3632 {
3633         mmfile_t mf1, mf2;
3634         struct checkdiff_t data;
3635
3636         if (!two)
3637                 return;
3638
3639         memset(&data, 0, sizeof(data));
3640         data.filename = name_b ? name_b : name_a;
3641         data.lineno = 0;
3642         data.o = o;
3643         data.ws_rule = whitespace_rule(attr_path);
3644         data.conflict_marker_size = ll_merge_marker_size(o->repo->index, attr_path);
3645
3646         if (fill_mmfile(o->repo, &mf1, one) < 0 ||
3647             fill_mmfile(o->repo, &mf2, two) < 0)
3648                 die("unable to read files to diff");
3649
3650         /*
3651          * All the other codepaths check both sides, but not checking
3652          * the "old" side here is deliberate.  We are checking the newly
3653          * introduced changes, and as long as the "new" side is text, we
3654          * can and should check what it introduces.
3655          */
3656         if (diff_filespec_is_binary(o->repo, two))
3657                 goto free_and_return;
3658         else {
3659                 /* Crazy xdl interfaces.. */
3660                 xpparam_t xpp;
3661                 xdemitconf_t xecfg;
3662
3663                 memset(&xpp, 0, sizeof(xpp));
3664                 memset(&xecfg, 0, sizeof(xecfg));
3665                 xecfg.ctxlen = 1; /* at least one context line */
3666                 xpp.flags = 0;
3667                 if (xdi_diff_outf(&mf1, &mf2, checkdiff_consume, &data,
3668                                   &xpp, &xecfg))
3669                         die("unable to generate checkdiff for %s", one->path);
3670
3671                 if (data.ws_rule & WS_BLANK_AT_EOF) {
3672                         struct emit_callback ecbdata;
3673                         int blank_at_eof;
3674
3675                         ecbdata.ws_rule = data.ws_rule;
3676                         check_blank_at_eof(&mf1, &mf2, &ecbdata);
3677                         blank_at_eof = ecbdata.blank_at_eof_in_postimage;
3678
3679                         if (blank_at_eof) {
3680                                 static char *err;
3681                                 if (!err)
3682                                         err = whitespace_error_string(WS_BLANK_AT_EOF);
3683                                 fprintf(o->file, "%s:%d: %s.\n",
3684                                         data.filename, blank_at_eof, err);
3685                                 data.status = 1; /* report errors */
3686                         }
3687                 }
3688         }
3689  free_and_return:
3690         diff_free_filespec_data(one);
3691         diff_free_filespec_data(two);
3692         if (data.status)
3693                 o->flags.check_failed = 1;
3694 }
3695
3696 struct diff_filespec *alloc_filespec(const char *path)
3697 {
3698         struct diff_filespec *spec;
3699
3700         FLEXPTR_ALLOC_STR(spec, path, path);
3701         spec->count = 1;
3702         spec->is_binary = -1;
3703         return spec;
3704 }
3705
3706 void free_filespec(struct diff_filespec *spec)
3707 {
3708         if (!--spec->count) {
3709                 diff_free_filespec_data(spec);
3710                 free(spec);
3711         }
3712 }
3713
3714 void fill_filespec(struct diff_filespec *spec, const struct object_id *oid,
3715                    int oid_valid, unsigned short mode)
3716 {
3717         if (mode) {
3718                 spec->mode = canon_mode(mode);
3719                 oidcpy(&spec->oid, oid);
3720                 spec->oid_valid = oid_valid;
3721         }
3722 }
3723
3724 /*
3725  * Given a name and sha1 pair, if the index tells us the file in
3726  * the work tree has that object contents, return true, so that
3727  * prepare_temp_file() does not have to inflate and extract.
3728  */
3729 static int reuse_worktree_file(struct index_state *istate,
3730                                const char *name,
3731                                const struct object_id *oid,
3732                                int want_file)
3733 {
3734         const struct cache_entry *ce;
3735         struct stat st;
3736         int pos, len;
3737
3738         /*
3739          * We do not read the cache ourselves here, because the
3740          * benchmark with my previous version that always reads cache
3741          * shows that it makes things worse for diff-tree comparing
3742          * two linux-2.6 kernel trees in an already checked out work
3743          * tree.  This is because most diff-tree comparisons deal with
3744          * only a small number of files, while reading the cache is
3745          * expensive for a large project, and its cost outweighs the
3746          * savings we get by not inflating the object to a temporary
3747          * file.  Practically, this code only helps when we are used
3748          * by diff-cache --cached, which does read the cache before
3749          * calling us.
3750          */
3751         if (!istate->cache)
3752                 return 0;
3753
3754         /* We want to avoid the working directory if our caller
3755          * doesn't need the data in a normal file, this system
3756          * is rather slow with its stat/open/mmap/close syscalls,
3757          * and the object is contained in a pack file.  The pack
3758          * is probably already open and will be faster to obtain
3759          * the data through than the working directory.  Loose
3760          * objects however would tend to be slower as they need
3761          * to be individually opened and inflated.
3762          */
3763         if (!FAST_WORKING_DIRECTORY && !want_file && has_object_pack(oid))
3764                 return 0;
3765
3766         /*
3767          * Similarly, if we'd have to convert the file contents anyway, that
3768          * makes the optimization not worthwhile.
3769          */
3770         if (!want_file && would_convert_to_git(istate, name))
3771                 return 0;
3772
3773         len = strlen(name);
3774         pos = index_name_pos(istate, name, len);
3775         if (pos < 0)
3776                 return 0;
3777         ce = istate->cache[pos];
3778
3779         /*
3780          * This is not the sha1 we are looking for, or
3781          * unreusable because it is not a regular file.
3782          */
3783         if (oidcmp(oid, &ce->oid) || !S_ISREG(ce->ce_mode))
3784                 return 0;
3785
3786         /*
3787          * If ce is marked as "assume unchanged", there is no
3788          * guarantee that work tree matches what we are looking for.
3789          */
3790         if ((ce->ce_flags & CE_VALID) || ce_skip_worktree(ce))
3791                 return 0;
3792
3793         /*
3794          * If ce matches the file in the work tree, we can reuse it.
3795          */
3796         if (ce_uptodate(ce) ||
3797             (!lstat(name, &st) && !ie_match_stat(istate, ce, &st, 0)))
3798                 return 1;
3799
3800         return 0;
3801 }
3802
3803 static int diff_populate_gitlink(struct diff_filespec *s, int size_only)
3804 {
3805         struct strbuf buf = STRBUF_INIT;
3806         char *dirty = "";
3807
3808         /* Are we looking at the work tree? */
3809         if (s->dirty_submodule)
3810                 dirty = "-dirty";
3811
3812         strbuf_addf(&buf, "Subproject commit %s%s\n",
3813                     oid_to_hex(&s->oid), dirty);
3814         s->size = buf.len;
3815         if (size_only) {
3816                 s->data = NULL;
3817                 strbuf_release(&buf);
3818         } else {
3819                 s->data = strbuf_detach(&buf, NULL);
3820                 s->should_free = 1;
3821         }
3822         return 0;
3823 }
3824
3825 /*
3826  * While doing rename detection and pickaxe operation, we may need to
3827  * grab the data for the blob (or file) for our own in-core comparison.
3828  * diff_filespec has data and size fields for this purpose.
3829  */
3830 int diff_populate_filespec(struct repository *r,
3831                            struct diff_filespec *s,
3832                            unsigned int flags)
3833 {
3834         int size_only = flags & CHECK_SIZE_ONLY;
3835         int err = 0;
3836         int conv_flags = global_conv_flags_eol;
3837         /*
3838          * demote FAIL to WARN to allow inspecting the situation
3839          * instead of refusing.
3840          */
3841         if (conv_flags & CONV_EOL_RNDTRP_DIE)
3842                 conv_flags = CONV_EOL_RNDTRP_WARN;
3843
3844         if (!DIFF_FILE_VALID(s))
3845                 die("internal error: asking to populate invalid file.");
3846         if (S_ISDIR(s->mode))
3847                 return -1;
3848
3849         if (s->data)
3850                 return 0;
3851
3852         if (size_only && 0 < s->size)
3853                 return 0;
3854
3855         if (S_ISGITLINK(s->mode))
3856                 return diff_populate_gitlink(s, size_only);
3857
3858         if (!s->oid_valid ||
3859             reuse_worktree_file(r->index, s->path, &s->oid, 0)) {
3860                 struct strbuf buf = STRBUF_INIT;
3861                 struct stat st;
3862                 int fd;
3863
3864                 if (lstat(s->path, &st) < 0) {
3865                 err_empty:
3866                         err = -1;
3867                 empty:
3868                         s->data = (char *)"";
3869                         s->size = 0;
3870                         return err;
3871                 }
3872                 s->size = xsize_t(st.st_size);
3873                 if (!s->size)
3874                         goto empty;
3875                 if (S_ISLNK(st.st_mode)) {
3876                         struct strbuf sb = STRBUF_INIT;
3877
3878                         if (strbuf_readlink(&sb, s->path, s->size))
3879                                 goto err_empty;
3880                         s->size = sb.len;
3881                         s->data = strbuf_detach(&sb, NULL);
3882                         s->should_free = 1;
3883                         return 0;
3884                 }
3885
3886                 /*
3887                  * Even if the caller would be happy with getting
3888                  * only the size, we cannot return early at this
3889                  * point if the path requires us to run the content
3890                  * conversion.
3891                  */
3892                 if (size_only && !would_convert_to_git(r->index, s->path))
3893                         return 0;
3894
3895                 /*
3896                  * Note: this check uses xsize_t(st.st_size) that may
3897                  * not be the true size of the blob after it goes
3898                  * through convert_to_git().  This may not strictly be
3899                  * correct, but the whole point of big_file_threshold
3900                  * and is_binary check being that we want to avoid
3901                  * opening the file and inspecting the contents, this
3902                  * is probably fine.
3903                  */
3904                 if ((flags & CHECK_BINARY) &&
3905                     s->size > big_file_threshold && s->is_binary == -1) {
3906                         s->is_binary = 1;
3907                         return 0;
3908                 }
3909                 fd = open(s->path, O_RDONLY);
3910                 if (fd < 0)
3911                         goto err_empty;
3912                 s->data = xmmap(NULL, s->size, PROT_READ, MAP_PRIVATE, fd, 0);
3913                 close(fd);
3914                 s->should_munmap = 1;
3915
3916                 /*
3917                  * Convert from working tree format to canonical git format
3918                  */
3919                 if (convert_to_git(r->index, s->path, s->data, s->size, &buf, conv_flags)) {
3920                         size_t size = 0;
3921                         munmap(s->data, s->size);
3922                         s->should_munmap = 0;
3923                         s->data = strbuf_detach(&buf, &size);
3924                         s->size = size;
3925                         s->should_free = 1;
3926                 }
3927         }
3928         else {
3929                 enum object_type type;
3930                 if (size_only || (flags & CHECK_BINARY)) {
3931                         type = oid_object_info(r, &s->oid, &s->size);
3932                         if (type < 0)
3933                                 die("unable to read %s",
3934                                     oid_to_hex(&s->oid));
3935                         if (size_only)
3936                                 return 0;
3937                         if (s->size > big_file_threshold && s->is_binary == -1) {
3938                                 s->is_binary = 1;
3939                                 return 0;
3940                         }
3941                 }
3942                 s->data = read_object_file(&s->oid, &type, &s->size);
3943                 if (!s->data)
3944                         die("unable to read %s", oid_to_hex(&s->oid));
3945                 s->should_free = 1;
3946         }
3947         return 0;
3948 }
3949
3950 void diff_free_filespec_blob(struct diff_filespec *s)
3951 {
3952         if (s->should_free)
3953                 free(s->data);
3954         else if (s->should_munmap)
3955                 munmap(s->data, s->size);
3956
3957         if (s->should_free || s->should_munmap) {
3958                 s->should_free = s->should_munmap = 0;
3959                 s->data = NULL;
3960         }
3961 }
3962
3963 void diff_free_filespec_data(struct diff_filespec *s)
3964 {
3965         diff_free_filespec_blob(s);
3966         FREE_AND_NULL(s->cnt_data);
3967 }
3968
3969 static void prep_temp_blob(struct index_state *istate,
3970                            const char *path, struct diff_tempfile *temp,
3971                            void *blob,
3972                            unsigned long size,
3973                            const struct object_id *oid,
3974                            int mode)
3975 {
3976         struct strbuf buf = STRBUF_INIT;
3977         struct strbuf tempfile = STRBUF_INIT;
3978         char *path_dup = xstrdup(path);
3979         const char *base = basename(path_dup);
3980
3981         /* Generate "XXXXXX_basename.ext" */
3982         strbuf_addstr(&tempfile, "XXXXXX_");
3983         strbuf_addstr(&tempfile, base);
3984
3985         temp->tempfile = mks_tempfile_ts(tempfile.buf, strlen(base) + 1);
3986         if (!temp->tempfile)
3987                 die_errno("unable to create temp-file");
3988         if (convert_to_working_tree(istate, path,
3989                         (const char *)blob, (size_t)size, &buf)) {
3990                 blob = buf.buf;
3991                 size = buf.len;
3992         }
3993         if (write_in_full(temp->tempfile->fd, blob, size) < 0 ||
3994             close_tempfile_gently(temp->tempfile))
3995                 die_errno("unable to write temp-file");
3996         temp->name = get_tempfile_path(temp->tempfile);
3997         oid_to_hex_r(temp->hex, oid);
3998         xsnprintf(temp->mode, sizeof(temp->mode), "%06o", mode);
3999         strbuf_release(&buf);
4000         strbuf_release(&tempfile);
4001         free(path_dup);
4002 }
4003
4004 static struct diff_tempfile *prepare_temp_file(struct repository *r,
4005                                                const char *name,
4006                                                struct diff_filespec *one)
4007 {
4008         struct diff_tempfile *temp = claim_diff_tempfile();
4009
4010         if (!DIFF_FILE_VALID(one)) {
4011         not_a_valid_file:
4012                 /* A '-' entry produces this for file-2, and
4013                  * a '+' entry produces this for file-1.
4014                  */
4015                 temp->name = "/dev/null";
4016                 xsnprintf(temp->hex, sizeof(temp->hex), ".");
4017                 xsnprintf(temp->mode, sizeof(temp->mode), ".");
4018                 return temp;
4019         }
4020
4021         if (!S_ISGITLINK(one->mode) &&
4022             (!one->oid_valid ||
4023              reuse_worktree_file(r->index, name, &one->oid, 1))) {
4024                 struct stat st;
4025                 if (lstat(name, &st) < 0) {
4026                         if (errno == ENOENT)
4027                                 goto not_a_valid_file;
4028                         die_errno("stat(%s)", name);
4029                 }
4030                 if (S_ISLNK(st.st_mode)) {
4031                         struct strbuf sb = STRBUF_INIT;
4032                         if (strbuf_readlink(&sb, name, st.st_size) < 0)
4033                                 die_errno("readlink(%s)", name);
4034                         prep_temp_blob(r->index, name, temp, sb.buf, sb.len,
4035                                        (one->oid_valid ?
4036                                         &one->oid : &null_oid),
4037                                        (one->oid_valid ?
4038                                         one->mode : S_IFLNK));
4039                         strbuf_release(&sb);
4040                 }
4041                 else {
4042                         /* we can borrow from the file in the work tree */
4043                         temp->name = name;
4044                         if (!one->oid_valid)
4045                                 oid_to_hex_r(temp->hex, &null_oid);
4046                         else
4047                                 oid_to_hex_r(temp->hex, &one->oid);
4048                         /* Even though we may sometimes borrow the
4049                          * contents from the work tree, we always want
4050                          * one->mode.  mode is trustworthy even when
4051                          * !(one->oid_valid), as long as
4052                          * DIFF_FILE_VALID(one).
4053                          */
4054                         xsnprintf(temp->mode, sizeof(temp->mode), "%06o", one->mode);
4055                 }
4056                 return temp;
4057         }
4058         else {
4059                 if (diff_populate_filespec(r, one, 0))
4060                         die("cannot read data blob for %s", one->path);
4061                 prep_temp_blob(r->index, name, temp,
4062                                one->data, one->size,
4063                                &one->oid, one->mode);
4064         }
4065         return temp;
4066 }
4067
4068 static void add_external_diff_name(struct repository *r,
4069                                    struct argv_array *argv,
4070                                    const char *name,
4071                                    struct diff_filespec *df)
4072 {
4073         struct diff_tempfile *temp = prepare_temp_file(r, name, df);
4074         argv_array_push(argv, temp->name);
4075         argv_array_push(argv, temp->hex);
4076         argv_array_push(argv, temp->mode);
4077 }
4078
4079 /* An external diff command takes:
4080  *
4081  * diff-cmd name infile1 infile1-sha1 infile1-mode \
4082  *               infile2 infile2-sha1 infile2-mode [ rename-to ]
4083  *
4084  */
4085 static void run_external_diff(const char *pgm,
4086                               const char *name,
4087                               const char *other,
4088                               struct diff_filespec *one,
4089                               struct diff_filespec *two,
4090                               const char *xfrm_msg,
4091                               int complete_rewrite,
4092                               struct diff_options *o)
4093 {
4094         struct argv_array argv = ARGV_ARRAY_INIT;
4095         struct argv_array env = ARGV_ARRAY_INIT;
4096         struct diff_queue_struct *q = &diff_queued_diff;
4097
4098         argv_array_push(&argv, pgm);
4099         argv_array_push(&argv, name);
4100
4101         if (one && two) {
4102                 add_external_diff_name(o->repo, &argv, name, one);
4103                 if (!other)
4104                         add_external_diff_name(o->repo, &argv, name, two);
4105                 else {
4106                         add_external_diff_name(o->repo, &argv, other, two);
4107                         argv_array_push(&argv, other);
4108                         argv_array_push(&argv, xfrm_msg);
4109                 }
4110         }
4111
4112         argv_array_pushf(&env, "GIT_DIFF_PATH_COUNTER=%d", ++o->diff_path_counter);
4113         argv_array_pushf(&env, "GIT_DIFF_PATH_TOTAL=%d", q->nr);
4114
4115         if (run_command_v_opt_cd_env(argv.argv, RUN_USING_SHELL, NULL, env.argv))
4116                 die(_("external diff died, stopping at %s"), name);
4117
4118         remove_tempfile();
4119         argv_array_clear(&argv);
4120         argv_array_clear(&env);
4121 }
4122
4123 static int similarity_index(struct diff_filepair *p)
4124 {
4125         return p->score * 100 / MAX_SCORE;
4126 }
4127
4128 static const char *diff_abbrev_oid(const struct object_id *oid, int abbrev)
4129 {
4130         if (startup_info->have_repository)
4131                 return find_unique_abbrev(oid, abbrev);
4132         else {
4133                 char *hex = oid_to_hex(oid);
4134                 if (abbrev < 0)
4135                         abbrev = FALLBACK_DEFAULT_ABBREV;
4136                 if (abbrev > the_hash_algo->hexsz)
4137                         BUG("oid abbreviation out of range: %d", abbrev);
4138                 if (abbrev)
4139                         hex[abbrev] = '\0';
4140                 return hex;
4141         }
4142 }
4143
4144 static void fill_metainfo(struct strbuf *msg,
4145                           const char *name,
4146                           const char *other,
4147                           struct diff_filespec *one,
4148                           struct diff_filespec *two,
4149                           struct diff_options *o,
4150                           struct diff_filepair *p,
4151                           int *must_show_header,
4152                           int use_color)
4153 {
4154         const char *set = diff_get_color(use_color, DIFF_METAINFO);
4155         const char *reset = diff_get_color(use_color, DIFF_RESET);
4156         const char *line_prefix = diff_line_prefix(o);
4157
4158         *must_show_header = 1;
4159         strbuf_init(msg, PATH_MAX * 2 + 300);
4160         switch (p->status) {
4161         case DIFF_STATUS_COPIED:
4162                 strbuf_addf(msg, "%s%ssimilarity index %d%%",
4163                             line_prefix, set, similarity_index(p));
4164                 strbuf_addf(msg, "%s\n%s%scopy from ",
4165                             reset,  line_prefix, set);
4166                 quote_c_style(name, msg, NULL, 0);
4167                 strbuf_addf(msg, "%s\n%s%scopy to ", reset, line_prefix, set);
4168                 quote_c_style(other, msg, NULL, 0);
4169                 strbuf_addf(msg, "%s\n", reset);
4170                 break;
4171         case DIFF_STATUS_RENAMED:
4172                 strbuf_addf(msg, "%s%ssimilarity index %d%%",
4173                             line_prefix, set, similarity_index(p));
4174                 strbuf_addf(msg, "%s\n%s%srename from ",
4175                             reset, line_prefix, set);
4176                 quote_c_style(name, msg, NULL, 0);
4177                 strbuf_addf(msg, "%s\n%s%srename to ",
4178                             reset, line_prefix, set);
4179                 quote_c_style(other, msg, NULL, 0);
4180                 strbuf_addf(msg, "%s\n", reset);
4181                 break;
4182         case DIFF_STATUS_MODIFIED:
4183                 if (p->score) {
4184                         strbuf_addf(msg, "%s%sdissimilarity index %d%%%s\n",
4185                                     line_prefix,
4186                                     set, similarity_index(p), reset);
4187                         break;
4188                 }
4189                 /* fallthru */
4190         default:
4191                 *must_show_header = 0;
4192         }
4193         if (one && two && oidcmp(&one->oid, &two->oid)) {
4194                 const unsigned hexsz = the_hash_algo->hexsz;
4195                 int abbrev = o->flags.full_index ? hexsz : DEFAULT_ABBREV;
4196
4197                 if (o->flags.binary) {
4198                         mmfile_t mf;
4199                         if ((!fill_mmfile(o->repo, &mf, one) &&
4200                              diff_filespec_is_binary(o->repo, one)) ||
4201                             (!fill_mmfile(o->repo, &mf, two) &&
4202                              diff_filespec_is_binary(o->repo, two)))
4203                                 abbrev = hexsz;
4204                 }
4205                 strbuf_addf(msg, "%s%sindex %s..%s", line_prefix, set,
4206                             diff_abbrev_oid(&one->oid, abbrev),
4207                             diff_abbrev_oid(&two->oid, abbrev));
4208                 if (one->mode == two->mode)
4209                         strbuf_addf(msg, " %06o", one->mode);
4210                 strbuf_addf(msg, "%s\n", reset);
4211         }
4212 }
4213
4214 static void run_diff_cmd(const char *pgm,
4215                          const char *name,
4216                          const char *other,
4217                          const char *attr_path,
4218                          struct diff_filespec *one,
4219                          struct diff_filespec *two,
4220                          struct strbuf *msg,
4221                          struct diff_options *o,
4222                          struct diff_filepair *p)
4223 {
4224         const char *xfrm_msg = NULL;
4225         int complete_rewrite = (p->status == DIFF_STATUS_MODIFIED) && p->score;
4226         int must_show_header = 0;
4227
4228
4229         if (o->flags.allow_external) {
4230                 struct userdiff_driver *drv;
4231
4232                 drv = userdiff_find_by_path(o->repo->index, attr_path);
4233                 if (drv && drv->external)
4234                         pgm = drv->external;
4235         }
4236
4237         if (msg) {
4238                 /*
4239                  * don't use colors when the header is intended for an
4240                  * external diff driver
4241                  */
4242                 fill_metainfo(msg, name, other, one, two, o, p,
4243                               &must_show_header,
4244                               want_color(o->use_color) && !pgm);
4245                 xfrm_msg = msg->len ? msg->buf : NULL;
4246         }
4247
4248         if (pgm) {
4249                 run_external_diff(pgm, name, other, one, two, xfrm_msg,
4250                                   complete_rewrite, o);
4251                 return;
4252         }
4253         if (one && two)
4254                 builtin_diff(name, other ? other : name,
4255                              one, two, xfrm_msg, must_show_header,
4256                              o, complete_rewrite);
4257         else
4258                 fprintf(o->file, "* Unmerged path %s\n", name);
4259 }
4260
4261 static void diff_fill_oid_info(struct diff_filespec *one, struct index_state *istate)
4262 {
4263         if (DIFF_FILE_VALID(one)) {
4264                 if (!one->oid_valid) {
4265                         struct stat st;
4266                         if (one->is_stdin) {
4267                                 oidclr(&one->oid);
4268                                 return;
4269                         }
4270                         if (lstat(one->path, &st) < 0)
4271                                 die_errno("stat '%s'", one->path);
4272                         if (index_path(istate, &one->oid, one->path, &st, 0))
4273                                 die("cannot hash %s", one->path);
4274                 }
4275         }
4276         else
4277                 oidclr(&one->oid);
4278 }
4279
4280 static void strip_prefix(int prefix_length, const char **namep, const char **otherp)
4281 {
4282         /* Strip the prefix but do not molest /dev/null and absolute paths */
4283         if (*namep && **namep != '/') {
4284                 *namep += prefix_length;
4285                 if (**namep == '/')
4286                         ++*namep;
4287         }
4288         if (*otherp && **otherp != '/') {
4289                 *otherp += prefix_length;
4290                 if (**otherp == '/')
4291                         ++*otherp;
4292         }
4293 }
4294
4295 static void run_diff(struct diff_filepair *p, struct diff_options *o)
4296 {
4297         const char *pgm = external_diff();
4298         struct strbuf msg;
4299         struct diff_filespec *one = p->one;
4300         struct diff_filespec *two = p->two;
4301         const char *name;
4302         const char *other;
4303         const char *attr_path;
4304
4305         name  = one->path;
4306         other = (strcmp(name, two->path) ? two->path : NULL);
4307         attr_path = name;
4308         if (o->prefix_length)
4309                 strip_prefix(o->prefix_length, &name, &other);
4310
4311         if (!o->flags.allow_external)
4312                 pgm = NULL;
4313
4314         if (DIFF_PAIR_UNMERGED(p)) {
4315                 run_diff_cmd(pgm, name, NULL, attr_path,
4316                              NULL, NULL, NULL, o, p);
4317                 return;
4318         }
4319
4320         diff_fill_oid_info(one, o->repo->index);
4321         diff_fill_oid_info(two, o->repo->index);
4322
4323         if (!pgm &&
4324             DIFF_FILE_VALID(one) && DIFF_FILE_VALID(two) &&
4325             (S_IFMT & one->mode) != (S_IFMT & two->mode)) {
4326                 /*
4327                  * a filepair that changes between file and symlink
4328                  * needs to be split into deletion and creation.
4329                  */
4330                 struct diff_filespec *null = alloc_filespec(two->path);
4331                 run_diff_cmd(NULL, name, other, attr_path,
4332                              one, null, &msg,
4333                              o, p);
4334                 free(null);
4335                 strbuf_release(&msg);
4336
4337                 null = alloc_filespec(one->path);
4338                 run_diff_cmd(NULL, name, other, attr_path,
4339                              null, two, &msg, o, p);
4340                 free(null);
4341         }
4342         else
4343                 run_diff_cmd(pgm, name, other, attr_path,
4344                              one, two, &msg, o, p);
4345
4346         strbuf_release(&msg);
4347 }
4348
4349 static void run_diffstat(struct diff_filepair *p, struct diff_options *o,
4350                          struct diffstat_t *diffstat)
4351 {
4352         const char *name;
4353         const char *other;
4354
4355         if (DIFF_PAIR_UNMERGED(p)) {
4356                 /* unmerged */
4357                 builtin_diffstat(p->one->path, NULL, NULL, NULL,
4358                                  diffstat, o, p);
4359                 return;
4360         }
4361
4362         name = p->one->path;
4363         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
4364
4365         if (o->prefix_length)
4366                 strip_prefix(o->prefix_length, &name, &other);
4367
4368         diff_fill_oid_info(p->one, o->repo->index);
4369         diff_fill_oid_info(p->two, o->repo->index);
4370
4371         builtin_diffstat(name, other, p->one, p->two,
4372                          diffstat, o, p);
4373 }
4374
4375 static void run_checkdiff(struct diff_filepair *p, struct diff_options *o)
4376 {
4377         const char *name;
4378         const char *other;
4379         const char *attr_path;
4380
4381         if (DIFF_PAIR_UNMERGED(p)) {
4382                 /* unmerged */
4383                 return;
4384         }
4385
4386         name = p->one->path;
4387         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
4388         attr_path = other ? other : name;
4389
4390         if (o->prefix_length)
4391                 strip_prefix(o->prefix_length, &name, &other);
4392
4393         diff_fill_oid_info(p->one, o->repo->index);
4394         diff_fill_oid_info(p->two, o->repo->index);
4395
4396         builtin_checkdiff(name, other, attr_path, p->one, p->two, o);
4397 }
4398
4399 void repo_diff_setup(struct repository *r, struct diff_options *options)
4400 {
4401         memcpy(options, &default_diff_options, sizeof(*options));
4402
4403         options->file = stdout;
4404         options->repo = r;
4405
4406         options->abbrev = DEFAULT_ABBREV;
4407         options->line_termination = '\n';
4408         options->break_opt = -1;
4409         options->rename_limit = -1;
4410         options->dirstat_permille = diff_dirstat_permille_default;
4411         options->context = diff_context_default;
4412         options->interhunkcontext = diff_interhunk_context_default;
4413         options->ws_error_highlight = ws_error_highlight_default;
4414         options->flags.rename_empty = 1;
4415         options->objfind = NULL;
4416
4417         /* pathchange left =NULL by default */
4418         options->change = diff_change;
4419         options->add_remove = diff_addremove;
4420         options->use_color = diff_use_color_default;
4421         options->detect_rename = diff_detect_rename_default;
4422         options->xdl_opts |= diff_algorithm;
4423         if (diff_indent_heuristic)
4424                 DIFF_XDL_SET(options, INDENT_HEURISTIC);
4425
4426         options->orderfile = diff_order_file_cfg;
4427
4428         if (diff_no_prefix) {
4429                 options->a_prefix = options->b_prefix = "";
4430         } else if (!diff_mnemonic_prefix) {
4431                 options->a_prefix = "a/";
4432                 options->b_prefix = "b/";
4433         }
4434
4435         options->color_moved = diff_color_moved_default;
4436         options->color_moved_ws_handling = diff_color_moved_ws_default;
4437 }
4438
4439 void diff_setup_done(struct diff_options *options)
4440 {
4441         unsigned check_mask = DIFF_FORMAT_NAME |
4442                               DIFF_FORMAT_NAME_STATUS |
4443                               DIFF_FORMAT_CHECKDIFF |
4444                               DIFF_FORMAT_NO_OUTPUT;
4445         /*
4446          * This must be signed because we're comparing against a potentially
4447          * negative value.
4448          */
4449         const int hexsz = the_hash_algo->hexsz;
4450
4451         if (options->set_default)
4452                 options->set_default(options);
4453
4454         if (HAS_MULTI_BITS(options->output_format & check_mask))
4455                 die(_("--name-only, --name-status, --check and -s are mutually exclusive"));
4456
4457         if (HAS_MULTI_BITS(options->pickaxe_opts & DIFF_PICKAXE_KINDS_MASK))
4458                 die(_("-G, -S and --find-object are mutually exclusive"));
4459
4460         /*
4461          * Most of the time we can say "there are changes"
4462          * only by checking if there are changed paths, but
4463          * --ignore-whitespace* options force us to look
4464          * inside contents.
4465          */
4466
4467         if ((options->xdl_opts & XDF_WHITESPACE_FLAGS))
4468                 options->flags.diff_from_contents = 1;
4469         else
4470                 options->flags.diff_from_contents = 0;
4471
4472         if (options->flags.find_copies_harder)
4473                 options->detect_rename = DIFF_DETECT_COPY;
4474
4475         if (!options->flags.relative_name)
4476                 options->prefix = NULL;
4477         if (options->prefix)
4478                 options->prefix_length = strlen(options->prefix);
4479         else
4480                 options->prefix_length = 0;
4481
4482         if (options->output_format & (DIFF_FORMAT_NAME |
4483                                       DIFF_FORMAT_NAME_STATUS |
4484                                       DIFF_FORMAT_CHECKDIFF |
4485                                       DIFF_FORMAT_NO_OUTPUT))
4486                 options->output_format &= ~(DIFF_FORMAT_RAW |
4487                                             DIFF_FORMAT_NUMSTAT |
4488                                             DIFF_FORMAT_DIFFSTAT |
4489                                             DIFF_FORMAT_SHORTSTAT |
4490                                             DIFF_FORMAT_DIRSTAT |
4491                                             DIFF_FORMAT_SUMMARY |
4492                                             DIFF_FORMAT_PATCH);
4493
4494         /*
4495          * These cases always need recursive; we do not drop caller-supplied
4496          * recursive bits for other formats here.
4497          */
4498         if (options->output_format & (DIFF_FORMAT_PATCH |
4499                                       DIFF_FORMAT_NUMSTAT |
4500                                       DIFF_FORMAT_DIFFSTAT |
4501                                       DIFF_FORMAT_SHORTSTAT |
4502                                       DIFF_FORMAT_DIRSTAT |
4503                                       DIFF_FORMAT_SUMMARY |
4504                                       DIFF_FORMAT_CHECKDIFF))
4505                 options->flags.recursive = 1;
4506         /*
4507          * Also pickaxe would not work very well if you do not say recursive
4508          */
4509         if (options->pickaxe_opts & DIFF_PICKAXE_KINDS_MASK)
4510                 options->flags.recursive = 1;
4511         /*
4512          * When patches are generated, submodules diffed against the work tree
4513          * must be checked for dirtiness too so it can be shown in the output
4514          */
4515         if (options->output_format & DIFF_FORMAT_PATCH)
4516                 options->flags.dirty_submodules = 1;
4517
4518         if (options->detect_rename && options->rename_limit < 0)
4519                 options->rename_limit = diff_rename_limit_default;
4520         if (hexsz < options->abbrev)
4521                 options->abbrev = hexsz; /* full */
4522
4523         /*
4524          * It does not make sense to show the first hit we happened
4525          * to have found.  It does not make sense not to return with
4526          * exit code in such a case either.
4527          */
4528         if (options->flags.quick) {
4529                 options->output_format = DIFF_FORMAT_NO_OUTPUT;
4530                 options->flags.exit_with_status = 1;
4531         }
4532
4533         options->diff_path_counter = 0;
4534
4535         if (options->flags.follow_renames && options->pathspec.nr != 1)
4536                 die(_("--follow requires exactly one pathspec"));
4537
4538         if (!options->use_color || external_diff())
4539                 options->color_moved = 0;
4540 }
4541
4542 static int opt_arg(const char *arg, int arg_short, const char *arg_long, int *val)
4543 {
4544         char c, *eq;
4545         int len;
4546
4547         if (*arg != '-')
4548                 return 0;
4549         c = *++arg;
4550         if (!c)
4551                 return 0;
4552         if (c == arg_short) {
4553                 c = *++arg;
4554                 if (!c)
4555                         return 1;
4556                 if (val && isdigit(c)) {
4557                         char *end;
4558                         int n = strtoul(arg, &end, 10);
4559                         if (*end)
4560                                 return 0;
4561                         *val = n;
4562                         return 1;
4563                 }
4564                 return 0;
4565         }
4566         if (c != '-')
4567                 return 0;
4568         arg++;
4569         eq = strchrnul(arg, '=');
4570         len = eq - arg;
4571         if (!len || strncmp(arg, arg_long, len))
4572                 return 0;
4573         if (*eq) {
4574                 int n;
4575                 char *end;
4576                 if (!isdigit(*++eq))
4577                         return 0;
4578                 n = strtoul(eq, &end, 10);
4579                 if (*end)
4580                         return 0;
4581                 *val = n;
4582         }
4583         return 1;
4584 }
4585
4586 static int diff_scoreopt_parse(const char *opt);
4587
4588 static inline int short_opt(char opt, const char **argv,
4589                             const char **optarg)
4590 {
4591         const char *arg = argv[0];
4592         if (arg[0] != '-' || arg[1] != opt)
4593                 return 0;
4594         if (arg[2] != '\0') {
4595                 *optarg = arg + 2;
4596                 return 1;
4597         }
4598         if (!argv[1])
4599                 die("Option '%c' requires a value", opt);
4600         *optarg = argv[1];
4601         return 2;
4602 }
4603
4604 int parse_long_opt(const char *opt, const char **argv,
4605                    const char **optarg)
4606 {
4607         const char *arg = argv[0];
4608         if (!skip_prefix(arg, "--", &arg))
4609                 return 0;
4610         if (!skip_prefix(arg, opt, &arg))
4611                 return 0;
4612         if (*arg == '=') { /* stuck form: --option=value */
4613                 *optarg = arg + 1;
4614                 return 1;
4615         }
4616         if (*arg != '\0')
4617                 return 0;
4618         /* separate form: --option value */
4619         if (!argv[1])
4620                 die("Option '--%s' requires a value", opt);
4621         *optarg = argv[1];
4622         return 2;
4623 }
4624
4625 static int stat_opt(struct diff_options *options, const char **av)
4626 {
4627         const char *arg = av[0];
4628         char *end;
4629         int width = options->stat_width;
4630         int name_width = options->stat_name_width;
4631         int graph_width = options->stat_graph_width;
4632         int count = options->stat_count;
4633         int argcount = 1;
4634
4635         if (!skip_prefix(arg, "--stat", &arg))
4636                 BUG("stat option does not begin with --stat: %s", arg);
4637         end = (char *)arg;
4638
4639         switch (*arg) {
4640         case '-':
4641                 if (skip_prefix(arg, "-width", &arg)) {
4642                         if (*arg == '=')
4643                                 width = strtoul(arg + 1, &end, 10);
4644                         else if (!*arg && !av[1])
4645                                 die_want_option("--stat-width");
4646                         else if (!*arg) {
4647                                 width = strtoul(av[1], &end, 10);
4648                                 argcount = 2;
4649                         }
4650                 } else if (skip_prefix(arg, "-name-width", &arg)) {
4651                         if (*arg == '=')
4652                                 name_width = strtoul(arg + 1, &end, 10);
4653                         else if (!*arg && !av[1])
4654                                 die_want_option("--stat-name-width");
4655                         else if (!*arg) {
4656                                 name_width = strtoul(av[1], &end, 10);
4657                                 argcount = 2;
4658                         }
4659                 } else if (skip_prefix(arg, "-graph-width", &arg)) {
4660                         if (*arg == '=')
4661                                 graph_width = strtoul(arg + 1, &end, 10);
4662                         else if (!*arg && !av[1])
4663                                 die_want_option("--stat-graph-width");
4664                         else if (!*arg) {
4665                                 graph_width = strtoul(av[1], &end, 10);
4666                                 argcount = 2;
4667                         }
4668                 } else if (skip_prefix(arg, "-count", &arg)) {
4669                         if (*arg == '=')
4670                                 count = strtoul(arg + 1, &end, 10);
4671                         else if (!*arg && !av[1])
4672                                 die_want_option("--stat-count");
4673                         else if (!*arg) {
4674                                 count = strtoul(av[1], &end, 10);
4675                                 argcount = 2;
4676                         }
4677                 }
4678                 break;
4679         case '=':
4680                 width = strtoul(arg+1, &end, 10);
4681                 if (*end == ',')
4682                         name_width = strtoul(end+1, &end, 10);
4683                 if (*end == ',')
4684                         count = strtoul(end+1, &end, 10);
4685         }
4686
4687         /* Important! This checks all the error cases! */
4688         if (*end)
4689                 return 0;
4690         options->output_format |= DIFF_FORMAT_DIFFSTAT;
4691         options->stat_name_width = name_width;
4692         options->stat_graph_width = graph_width;
4693         options->stat_width = width;
4694         options->stat_count = count;
4695         return argcount;
4696 }
4697
4698 static int parse_dirstat_opt(struct diff_options *options, const char *params)
4699 {
4700         struct strbuf errmsg = STRBUF_INIT;
4701         if (parse_dirstat_params(options, params, &errmsg))
4702                 die(_("Failed to parse --dirstat/-X option parameter:\n%s"),
4703                     errmsg.buf);
4704         strbuf_release(&errmsg);
4705         /*
4706          * The caller knows a dirstat-related option is given from the command
4707          * line; allow it to say "return this_function();"
4708          */
4709         options->output_format |= DIFF_FORMAT_DIRSTAT;
4710         return 1;
4711 }
4712
4713 static int parse_submodule_opt(struct diff_options *options, const char *value)
4714 {
4715         if (parse_submodule_params(options, value))
4716                 die(_("Failed to parse --submodule option parameter: '%s'"),
4717                         value);
4718         return 1;
4719 }
4720
4721 static const char diff_status_letters[] = {
4722         DIFF_STATUS_ADDED,
4723         DIFF_STATUS_COPIED,
4724         DIFF_STATUS_DELETED,
4725         DIFF_STATUS_MODIFIED,
4726         DIFF_STATUS_RENAMED,
4727         DIFF_STATUS_TYPE_CHANGED,
4728         DIFF_STATUS_UNKNOWN,
4729         DIFF_STATUS_UNMERGED,
4730         DIFF_STATUS_FILTER_AON,
4731         DIFF_STATUS_FILTER_BROKEN,
4732         '\0',
4733 };
4734
4735 static unsigned int filter_bit['Z' + 1];
4736
4737 static void prepare_filter_bits(void)
4738 {
4739         int i;
4740
4741         if (!filter_bit[DIFF_STATUS_ADDED]) {
4742                 for (i = 0; diff_status_letters[i]; i++)
4743                         filter_bit[(int) diff_status_letters[i]] = (1 << i);
4744         }
4745 }
4746
4747 static unsigned filter_bit_tst(char status, const struct diff_options *opt)
4748 {
4749         return opt->filter & filter_bit[(int) status];
4750 }
4751
4752 static int parse_diff_filter_opt(const char *optarg, struct diff_options *opt)
4753 {
4754         int i, optch;
4755
4756         prepare_filter_bits();
4757
4758         /*
4759          * If there is a negation e.g. 'd' in the input, and we haven't
4760          * initialized the filter field with another --diff-filter, start
4761          * from full set of bits, except for AON.
4762          */
4763         if (!opt->filter) {
4764                 for (i = 0; (optch = optarg[i]) != '\0'; i++) {
4765                         if (optch < 'a' || 'z' < optch)
4766                                 continue;
4767                         opt->filter = (1 << (ARRAY_SIZE(diff_status_letters) - 1)) - 1;
4768                         opt->filter &= ~filter_bit[DIFF_STATUS_FILTER_AON];
4769                         break;
4770                 }
4771         }
4772
4773         for (i = 0; (optch = optarg[i]) != '\0'; i++) {
4774                 unsigned int bit;
4775                 int negate;
4776
4777                 if ('a' <= optch && optch <= 'z') {
4778                         negate = 1;
4779                         optch = toupper(optch);
4780                 } else {
4781                         negate = 0;
4782                 }
4783
4784                 bit = (0 <= optch && optch <= 'Z') ? filter_bit[optch] : 0;
4785                 if (!bit)
4786                         return optarg[i];
4787                 if (negate)
4788                         opt->filter &= ~bit;
4789                 else
4790                         opt->filter |= bit;
4791         }
4792         return 0;
4793 }
4794
4795 static void enable_patch_output(int *fmt) {
4796         *fmt &= ~DIFF_FORMAT_NO_OUTPUT;
4797         *fmt |= DIFF_FORMAT_PATCH;
4798 }
4799
4800 static int parse_ws_error_highlight_opt(struct diff_options *opt, const char *arg)
4801 {
4802         int val = parse_ws_error_highlight(arg);
4803
4804         if (val < 0) {
4805                 error("unknown value after ws-error-highlight=%.*s",
4806                       -1 - val, arg);
4807                 return 0;
4808         }
4809         opt->ws_error_highlight = val;
4810         return 1;
4811 }
4812
4813 static int parse_objfind_opt(struct diff_options *opt, const char *arg)
4814 {
4815         struct object_id oid;
4816
4817         if (get_oid(arg, &oid))
4818                 return error("unable to resolve '%s'", arg);
4819
4820         if (!opt->objfind)
4821                 opt->objfind = xcalloc(1, sizeof(*opt->objfind));
4822
4823         opt->pickaxe_opts |= DIFF_PICKAXE_KIND_OBJFIND;
4824         opt->flags.recursive = 1;
4825         opt->flags.tree_in_recursive = 1;
4826         oidset_insert(opt->objfind, &oid);
4827         return 1;
4828 }
4829
4830 int diff_opt_parse(struct diff_options *options,
4831                    const char **av, int ac, const char *prefix)
4832 {
4833         const char *arg = av[0];
4834         const char *optarg;
4835         int argcount;
4836
4837         if (!prefix)
4838                 prefix = "";
4839
4840         /* Output format options */
4841         if (!strcmp(arg, "-p") || !strcmp(arg, "-u") || !strcmp(arg, "--patch")
4842             || opt_arg(arg, 'U', "unified", &options->context))
4843                 enable_patch_output(&options->output_format);
4844         else if (!strcmp(arg, "--raw"))
4845                 options->output_format |= DIFF_FORMAT_RAW;
4846         else if (!strcmp(arg, "--patch-with-raw")) {
4847                 enable_patch_output(&options->output_format);
4848                 options->output_format |= DIFF_FORMAT_RAW;
4849         } else if (!strcmp(arg, "--numstat"))
4850                 options->output_format |= DIFF_FORMAT_NUMSTAT;
4851         else if (!strcmp(arg, "--shortstat"))
4852                 options->output_format |= DIFF_FORMAT_SHORTSTAT;
4853         else if (skip_prefix(arg, "-X", &arg) ||
4854                  skip_to_optional_arg(arg, "--dirstat", &arg))
4855                 return parse_dirstat_opt(options, arg);
4856         else if (!strcmp(arg, "--cumulative"))
4857                 return parse_dirstat_opt(options, "cumulative");
4858         else if (skip_to_optional_arg(arg, "--dirstat-by-file", &arg)) {
4859                 parse_dirstat_opt(options, "files");
4860                 return parse_dirstat_opt(options, arg);
4861         }
4862         else if (!strcmp(arg, "--check"))
4863                 options->output_format |= DIFF_FORMAT_CHECKDIFF;
4864         else if (!strcmp(arg, "--summary"))
4865                 options->output_format |= DIFF_FORMAT_SUMMARY;
4866         else if (!strcmp(arg, "--patch-with-stat")) {
4867                 enable_patch_output(&options->output_format);
4868                 options->output_format |= DIFF_FORMAT_DIFFSTAT;
4869         } else if (!strcmp(arg, "--name-only"))
4870                 options->output_format |= DIFF_FORMAT_NAME;
4871         else if (!strcmp(arg, "--name-status"))
4872                 options->output_format |= DIFF_FORMAT_NAME_STATUS;
4873         else if (!strcmp(arg, "-s") || !strcmp(arg, "--no-patch"))
4874                 options->output_format |= DIFF_FORMAT_NO_OUTPUT;
4875         else if (starts_with(arg, "--stat"))
4876                 /* --stat, --stat-width, --stat-name-width, or --stat-count */
4877                 return stat_opt(options, av);
4878         else if (!strcmp(arg, "--compact-summary")) {
4879                  options->flags.stat_with_summary = 1;
4880                  options->output_format |= DIFF_FORMAT_DIFFSTAT;
4881         } else if (!strcmp(arg, "--no-compact-summary"))
4882                  options->flags.stat_with_summary = 0;
4883
4884         /* renames options */
4885         else if (starts_with(arg, "-B") ||
4886                  skip_to_optional_arg(arg, "--break-rewrites", NULL)) {
4887                 if ((options->break_opt = diff_scoreopt_parse(arg)) == -1)
4888                         return error("invalid argument to -B: %s", arg+2);
4889         }
4890         else if (starts_with(arg, "-M") ||
4891                  skip_to_optional_arg(arg, "--find-renames", NULL)) {
4892                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
4893                         return error("invalid argument to -M: %s", arg+2);
4894                 options->detect_rename = DIFF_DETECT_RENAME;
4895         }
4896         else if (!strcmp(arg, "-D") || !strcmp(arg, "--irreversible-delete")) {
4897                 options->irreversible_delete = 1;
4898         }
4899         else if (starts_with(arg, "-C") ||
4900                  skip_to_optional_arg(arg, "--find-copies", NULL)) {
4901                 if (options->detect_rename == DIFF_DETECT_COPY)
4902                         options->flags.find_copies_harder = 1;
4903                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
4904                         return error("invalid argument to -C: %s", arg+2);
4905                 options->detect_rename = DIFF_DETECT_COPY;
4906         }
4907         else if (!strcmp(arg, "--no-renames"))
4908                 options->detect_rename = 0;
4909         else if (!strcmp(arg, "--rename-empty"))
4910                 options->flags.rename_empty = 1;
4911         else if (!strcmp(arg, "--no-rename-empty"))
4912                 options->flags.rename_empty = 0;
4913         else if (skip_to_optional_arg_default(arg, "--relative", &arg, NULL)) {
4914                 options->flags.relative_name = 1;
4915                 if (arg)
4916                         options->prefix = arg;
4917         }
4918
4919         /* xdiff options */
4920         else if (!strcmp(arg, "--minimal"))
4921                 DIFF_XDL_SET(options, NEED_MINIMAL);
4922         else if (!strcmp(arg, "--no-minimal"))
4923                 DIFF_XDL_CLR(options, NEED_MINIMAL);
4924         else if (!strcmp(arg, "-w") || !strcmp(arg, "--ignore-all-space"))
4925                 DIFF_XDL_SET(options, IGNORE_WHITESPACE);
4926         else if (!strcmp(arg, "-b") || !strcmp(arg, "--ignore-space-change"))
4927                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_CHANGE);
4928         else if (!strcmp(arg, "--ignore-space-at-eol"))
4929                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_AT_EOL);
4930         else if (!strcmp(arg, "--ignore-cr-at-eol"))
4931                 DIFF_XDL_SET(options, IGNORE_CR_AT_EOL);
4932         else if (!strcmp(arg, "--ignore-blank-lines"))
4933                 DIFF_XDL_SET(options, IGNORE_BLANK_LINES);
4934         else if (!strcmp(arg, "--indent-heuristic"))
4935                 DIFF_XDL_SET(options, INDENT_HEURISTIC);
4936         else if (!strcmp(arg, "--no-indent-heuristic"))
4937                 DIFF_XDL_CLR(options, INDENT_HEURISTIC);
4938         else if (!strcmp(arg, "--patience")) {
4939                 int i;
4940                 options->xdl_opts = DIFF_WITH_ALG(options, PATIENCE_DIFF);
4941                 /*
4942                  * Both --patience and --anchored use PATIENCE_DIFF
4943                  * internally, so remove any anchors previously
4944                  * specified.
4945                  */
4946                 for (i = 0; i < options->anchors_nr; i++)
4947                         free(options->anchors[i]);
4948                 options->anchors_nr = 0;
4949         } else if (!strcmp(arg, "--histogram"))
4950                 options->xdl_opts = DIFF_WITH_ALG(options, HISTOGRAM_DIFF);
4951         else if ((argcount = parse_long_opt("diff-algorithm", av, &optarg))) {
4952                 long value = parse_algorithm_value(optarg);
4953                 if (value < 0)
4954                         return error("option diff-algorithm accepts \"myers\", "
4955                                      "\"minimal\", \"patience\" and \"histogram\"");
4956                 /* clear out previous settings */
4957                 DIFF_XDL_CLR(options, NEED_MINIMAL);
4958                 options->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
4959                 options->xdl_opts |= value;
4960                 return argcount;
4961         } else if (skip_prefix(arg, "--anchored=", &arg)) {
4962                 options->xdl_opts = DIFF_WITH_ALG(options, PATIENCE_DIFF);
4963                 ALLOC_GROW(options->anchors, options->anchors_nr + 1,
4964                            options->anchors_alloc);
4965                 options->anchors[options->anchors_nr++] = xstrdup(arg);
4966         }
4967
4968         /* flags options */
4969         else if (!strcmp(arg, "--binary")) {
4970                 enable_patch_output(&options->output_format);
4971                 options->flags.binary = 1;
4972         }
4973         else if (!strcmp(arg, "--full-index"))
4974                 options->flags.full_index = 1;
4975         else if (!strcmp(arg, "-a") || !strcmp(arg, "--text"))
4976                 options->flags.text = 1;
4977         else if (!strcmp(arg, "-R"))
4978                 options->flags.reverse_diff = 1;
4979         else if (!strcmp(arg, "--find-copies-harder"))
4980                 options->flags.find_copies_harder = 1;
4981         else if (!strcmp(arg, "--follow"))
4982                 options->flags.follow_renames = 1;
4983         else if (!strcmp(arg, "--no-follow")) {
4984                 options->flags.follow_renames = 0;
4985                 options->flags.default_follow_renames = 0;
4986         } else if (skip_to_optional_arg_default(arg, "--color", &arg, "always")) {
4987                 int value = git_config_colorbool(NULL, arg);
4988                 if (value < 0)
4989                         return error("option `color' expects \"always\", \"auto\", or \"never\"");
4990                 options->use_color = value;
4991         }
4992         else if (!strcmp(arg, "--no-color"))
4993                 options->use_color = 0;
4994         else if (!strcmp(arg, "--color-moved")) {
4995                 if (diff_color_moved_default)
4996                         options->color_moved = diff_color_moved_default;
4997                 if (options->color_moved == COLOR_MOVED_NO)
4998                         options->color_moved = COLOR_MOVED_DEFAULT;
4999         } else if (!strcmp(arg, "--no-color-moved"))
5000                 options->color_moved = COLOR_MOVED_NO;
5001         else if (skip_prefix(arg, "--color-moved=", &arg)) {
5002                 int cm = parse_color_moved(arg);
5003                 if (cm < 0)
5004                         die("bad --color-moved argument: %s", arg);
5005                 options->color_moved = cm;
5006         } else if (skip_prefix(arg, "--color-moved-ws=", &arg)) {
5007                 options->color_moved_ws_handling = parse_color_moved_ws(arg);
5008         } else if (skip_to_optional_arg_default(arg, "--color-words", &options->word_regex, NULL)) {
5009                 options->use_color = 1;
5010                 options->word_diff = DIFF_WORDS_COLOR;
5011         }
5012         else if (!strcmp(arg, "--word-diff")) {
5013                 if (options->word_diff == DIFF_WORDS_NONE)
5014                         options->word_diff = DIFF_WORDS_PLAIN;
5015         }
5016         else if (skip_prefix(arg, "--word-diff=", &arg)) {
5017                 if (!strcmp(arg, "plain"))
5018                         options->word_diff = DIFF_WORDS_PLAIN;
5019                 else if (!strcmp(arg, "color")) {
5020                         options->use_color = 1;
5021                         options->word_diff = DIFF_WORDS_COLOR;
5022                 }
5023                 else if (!strcmp(arg, "porcelain"))
5024                         options->word_diff = DIFF_WORDS_PORCELAIN;
5025                 else if (!strcmp(arg, "none"))
5026                         options->word_diff = DIFF_WORDS_NONE;
5027                 else
5028                         die("bad --word-diff argument: %s", arg);
5029         }
5030         else if ((argcount = parse_long_opt("word-diff-regex", av, &optarg))) {
5031                 if (options->word_diff == DIFF_WORDS_NONE)
5032                         options->word_diff = DIFF_WORDS_PLAIN;
5033                 options->word_regex = optarg;
5034                 return argcount;
5035         }
5036         else if (!strcmp(arg, "--exit-code"))
5037                 options->flags.exit_with_status = 1;
5038         else if (!strcmp(arg, "--quiet"))
5039                 options->flags.quick = 1;
5040         else if (!strcmp(arg, "--ext-diff"))
5041                 options->flags.allow_external = 1;
5042         else if (!strcmp(arg, "--no-ext-diff"))
5043                 options->flags.allow_external = 0;
5044         else if (!strcmp(arg, "--textconv")) {
5045                 options->flags.allow_textconv = 1;
5046                 options->flags.textconv_set_via_cmdline = 1;
5047         } else if (!strcmp(arg, "--no-textconv"))
5048                 options->flags.allow_textconv = 0;
5049         else if (skip_to_optional_arg_default(arg, "--ignore-submodules", &arg, "all")) {
5050                 options->flags.override_submodule_config = 1;
5051                 handle_ignore_submodules_arg(options, arg);
5052         } else if (skip_to_optional_arg_default(arg, "--submodule", &arg, "log"))
5053                 return parse_submodule_opt(options, arg);
5054         else if (skip_prefix(arg, "--ws-error-highlight=", &arg))
5055                 return parse_ws_error_highlight_opt(options, arg);
5056         else if (!strcmp(arg, "--ita-invisible-in-index"))
5057                 options->ita_invisible_in_index = 1;
5058         else if (!strcmp(arg, "--ita-visible-in-index"))
5059                 options->ita_invisible_in_index = 0;
5060
5061         /* misc options */
5062         else if (!strcmp(arg, "-z"))
5063                 options->line_termination = 0;
5064         else if ((argcount = short_opt('l', av, &optarg))) {
5065                 options->rename_limit = strtoul(optarg, NULL, 10);
5066                 return argcount;
5067         }
5068         else if ((argcount = short_opt('S', av, &optarg))) {
5069                 options->pickaxe = optarg;
5070                 options->pickaxe_opts |= DIFF_PICKAXE_KIND_S;
5071                 return argcount;
5072         } else if ((argcount = short_opt('G', av, &optarg))) {
5073                 options->pickaxe = optarg;
5074                 options->pickaxe_opts |= DIFF_PICKAXE_KIND_G;
5075                 return argcount;
5076         }
5077         else if (!strcmp(arg, "--pickaxe-all"))
5078                 options->pickaxe_opts |= DIFF_PICKAXE_ALL;
5079         else if (!strcmp(arg, "--pickaxe-regex"))
5080                 options->pickaxe_opts |= DIFF_PICKAXE_REGEX;
5081         else if ((argcount = short_opt('O', av, &optarg))) {
5082                 options->orderfile = prefix_filename(prefix, optarg);
5083                 return argcount;
5084         } else if (skip_prefix(arg, "--find-object=", &arg))
5085                 return parse_objfind_opt(options, arg);
5086         else if ((argcount = parse_long_opt("diff-filter", av, &optarg))) {
5087                 int offending = parse_diff_filter_opt(optarg, options);
5088                 if (offending)
5089                         die("unknown change class '%c' in --diff-filter=%s",
5090                             offending, optarg);
5091                 return argcount;
5092         }
5093         else if (!strcmp(arg, "--no-abbrev"))
5094                 options->abbrev = 0;
5095         else if (!strcmp(arg, "--abbrev"))
5096                 options->abbrev = DEFAULT_ABBREV;
5097         else if (skip_prefix(arg, "--abbrev=", &arg)) {
5098                 options->abbrev = strtoul(arg, NULL, 10);
5099                 if (options->abbrev < MINIMUM_ABBREV)
5100                         options->abbrev = MINIMUM_ABBREV;
5101                 else if (the_hash_algo->hexsz < options->abbrev)
5102                         options->abbrev = the_hash_algo->hexsz;
5103         }
5104         else if ((argcount = parse_long_opt("src-prefix", av, &optarg))) {
5105                 options->a_prefix = optarg;
5106                 return argcount;
5107         }
5108         else if ((argcount = parse_long_opt("line-prefix", av, &optarg))) {
5109                 options->line_prefix = optarg;
5110                 options->line_prefix_length = strlen(options->line_prefix);
5111                 graph_setup_line_prefix(options);
5112                 return argcount;
5113         }
5114         else if ((argcount = parse_long_opt("dst-prefix", av, &optarg))) {
5115                 options->b_prefix = optarg;
5116                 return argcount;
5117         }
5118         else if (!strcmp(arg, "--no-prefix"))
5119                 options->a_prefix = options->b_prefix = "";
5120         else if (opt_arg(arg, '\0', "inter-hunk-context",
5121                          &options->interhunkcontext))
5122                 ;
5123         else if (!strcmp(arg, "-W"))
5124                 options->flags.funccontext = 1;
5125         else if (!strcmp(arg, "--function-context"))
5126                 options->flags.funccontext = 1;
5127         else if (!strcmp(arg, "--no-function-context"))
5128                 options->flags.funccontext = 0;
5129         else if ((argcount = parse_long_opt("output", av, &optarg))) {
5130                 char *path = prefix_filename(prefix, optarg);
5131                 options->file = xfopen(path, "w");
5132                 options->close_file = 1;
5133                 if (options->use_color != GIT_COLOR_ALWAYS)
5134                         options->use_color = GIT_COLOR_NEVER;
5135                 free(path);
5136                 return argcount;
5137         } else
5138                 return 0;
5139         return 1;
5140 }
5141
5142 int parse_rename_score(const char **cp_p)
5143 {
5144         unsigned long num, scale;
5145         int ch, dot;
5146         const char *cp = *cp_p;
5147
5148         num = 0;
5149         scale = 1;
5150         dot = 0;
5151         for (;;) {
5152                 ch = *cp;
5153                 if ( !dot && ch == '.' ) {
5154                         scale = 1;
5155                         dot = 1;
5156                 } else if ( ch == '%' ) {
5157                         scale = dot ? scale*100 : 100;
5158                         cp++;   /* % is always at the end */
5159                         break;
5160                 } else if ( ch >= '0' && ch <= '9' ) {
5161                         if ( scale < 100000 ) {
5162                                 scale *= 10;
5163                                 num = (num*10) + (ch-'0');
5164                         }
5165                 } else {
5166                         break;
5167                 }
5168                 cp++;
5169         }
5170         *cp_p = cp;
5171
5172         /* user says num divided by scale and we say internally that
5173          * is MAX_SCORE * num / scale.
5174          */
5175         return (int)((num >= scale) ? MAX_SCORE : (MAX_SCORE * num / scale));
5176 }
5177
5178 static int diff_scoreopt_parse(const char *opt)
5179 {
5180         int opt1, opt2, cmd;
5181
5182         if (*opt++ != '-')
5183                 return -1;
5184         cmd = *opt++;
5185         if (cmd == '-') {
5186                 /* convert the long-form arguments into short-form versions */
5187                 if (skip_prefix(opt, "break-rewrites", &opt)) {
5188                         if (*opt == 0 || *opt++ == '=')
5189                                 cmd = 'B';
5190                 } else if (skip_prefix(opt, "find-copies", &opt)) {
5191                         if (*opt == 0 || *opt++ == '=')
5192                                 cmd = 'C';
5193                 } else if (skip_prefix(opt, "find-renames", &opt)) {
5194                         if (*opt == 0 || *opt++ == '=')
5195                                 cmd = 'M';
5196                 }
5197         }
5198         if (cmd != 'M' && cmd != 'C' && cmd != 'B')
5199                 return -1; /* that is not a -M, -C, or -B option */
5200
5201         opt1 = parse_rename_score(&opt);
5202         if (cmd != 'B')
5203                 opt2 = 0;
5204         else {
5205                 if (*opt == 0)
5206                         opt2 = 0;
5207                 else if (*opt != '/')
5208                         return -1; /* we expect -B80/99 or -B80 */
5209                 else {
5210                         opt++;
5211                         opt2 = parse_rename_score(&opt);
5212                 }
5213         }
5214         if (*opt != 0)
5215                 return -1;
5216         return opt1 | (opt2 << 16);
5217 }
5218
5219 struct diff_queue_struct diff_queued_diff;
5220
5221 void diff_q(struct diff_queue_struct *queue, struct diff_filepair *dp)
5222 {
5223         ALLOC_GROW(queue->queue, queue->nr + 1, queue->alloc);
5224         queue->queue[queue->nr++] = dp;
5225 }
5226
5227 struct diff_filepair *diff_queue(struct diff_queue_struct *queue,
5228                                  struct diff_filespec *one,
5229                                  struct diff_filespec *two)
5230 {
5231         struct diff_filepair *dp = xcalloc(1, sizeof(*dp));
5232         dp->one = one;
5233         dp->two = two;
5234         if (queue)
5235                 diff_q(queue, dp);
5236         return dp;
5237 }
5238
5239 void diff_free_filepair(struct diff_filepair *p)
5240 {
5241         free_filespec(p->one);
5242         free_filespec(p->two);
5243         free(p);
5244 }
5245
5246 const char *diff_aligned_abbrev(const struct object_id *oid, int len)
5247 {
5248         int abblen;
5249         const char *abbrev;
5250
5251         /* Do we want all 40 hex characters? */
5252         if (len == the_hash_algo->hexsz)
5253                 return oid_to_hex(oid);
5254
5255         /* An abbreviated value is fine, possibly followed by an ellipsis. */
5256         abbrev = diff_abbrev_oid(oid, len);
5257
5258         if (!print_sha1_ellipsis())
5259                 return abbrev;
5260
5261         abblen = strlen(abbrev);
5262
5263         /*
5264          * In well-behaved cases, where the abbreviated result is the
5265          * same as the requested length, append three dots after the
5266          * abbreviation (hence the whole logic is limited to the case
5267          * where abblen < 37); when the actual abbreviated result is a
5268          * bit longer than the requested length, we reduce the number
5269          * of dots so that they match the well-behaved ones.  However,
5270          * if the actual abbreviation is longer than the requested
5271          * length by more than three, we give up on aligning, and add
5272          * three dots anyway, to indicate that the output is not the
5273          * full object name.  Yes, this may be suboptimal, but this
5274          * appears only in "diff --raw --abbrev" output and it is not
5275          * worth the effort to change it now.  Note that this would
5276          * likely to work fine when the automatic sizing of default
5277          * abbreviation length is used--we would be fed -1 in "len" in
5278          * that case, and will end up always appending three-dots, but
5279          * the automatic sizing is supposed to give abblen that ensures
5280          * uniqueness across all objects (statistically speaking).
5281          */
5282         if (abblen < the_hash_algo->hexsz - 3) {
5283                 static char hex[GIT_MAX_HEXSZ + 1];
5284                 if (len < abblen && abblen <= len + 2)
5285                         xsnprintf(hex, sizeof(hex), "%s%.*s", abbrev, len+3-abblen, "..");
5286                 else
5287                         xsnprintf(hex, sizeof(hex), "%s...", abbrev);
5288                 return hex;
5289         }
5290
5291         return oid_to_hex(oid);
5292 }
5293
5294 static void diff_flush_raw(struct diff_filepair *p, struct diff_options *opt)
5295 {
5296         int line_termination = opt->line_termination;
5297         int inter_name_termination = line_termination ? '\t' : '\0';
5298
5299         fprintf(opt->file, "%s", diff_line_prefix(opt));
5300         if (!(opt->output_format & DIFF_FORMAT_NAME_STATUS)) {
5301                 fprintf(opt->file, ":%06o %06o %s ", p->one->mode, p->two->mode,
5302                         diff_aligned_abbrev(&p->one->oid, opt->abbrev));
5303                 fprintf(opt->file, "%s ",
5304                         diff_aligned_abbrev(&p->two->oid, opt->abbrev));
5305         }
5306         if (p->score) {
5307                 fprintf(opt->file, "%c%03d%c", p->status, similarity_index(p),
5308                         inter_name_termination);
5309         } else {
5310                 fprintf(opt->file, "%c%c", p->status, inter_name_termination);
5311         }
5312
5313         if (p->status == DIFF_STATUS_COPIED ||
5314             p->status == DIFF_STATUS_RENAMED) {
5315                 const char *name_a, *name_b;
5316                 name_a = p->one->path;
5317                 name_b = p->two->path;
5318                 strip_prefix(opt->prefix_length, &name_a, &name_b);
5319                 write_name_quoted(name_a, opt->file, inter_name_termination);
5320                 write_name_quoted(name_b, opt->file, line_termination);
5321         } else {
5322                 const char *name_a, *name_b;
5323                 name_a = p->one->mode ? p->one->path : p->two->path;
5324                 name_b = NULL;
5325                 strip_prefix(opt->prefix_length, &name_a, &name_b);
5326                 write_name_quoted(name_a, opt->file, line_termination);
5327         }
5328 }
5329
5330 int diff_unmodified_pair(struct diff_filepair *p)
5331 {
5332         /* This function is written stricter than necessary to support
5333          * the currently implemented transformers, but the idea is to
5334          * let transformers to produce diff_filepairs any way they want,
5335          * and filter and clean them up here before producing the output.
5336          */
5337         struct diff_filespec *one = p->one, *two = p->two;
5338
5339         if (DIFF_PAIR_UNMERGED(p))
5340                 return 0; /* unmerged is interesting */
5341
5342         /* deletion, addition, mode or type change
5343          * and rename are all interesting.
5344          */
5345         if (DIFF_FILE_VALID(one) != DIFF_FILE_VALID(two) ||
5346             DIFF_PAIR_MODE_CHANGED(p) ||
5347             strcmp(one->path, two->path))
5348                 return 0;
5349
5350         /* both are valid and point at the same path.  that is, we are
5351          * dealing with a change.
5352          */
5353         if (one->oid_valid && two->oid_valid &&
5354             !oidcmp(&one->oid, &two->oid) &&
5355             !one->dirty_submodule && !two->dirty_submodule)
5356                 return 1; /* no change */
5357         if (!one->oid_valid && !two->oid_valid)
5358                 return 1; /* both look at the same file on the filesystem. */
5359         return 0;
5360 }
5361
5362 static void diff_flush_patch(struct diff_filepair *p, struct diff_options *o)
5363 {
5364         if (diff_unmodified_pair(p))
5365                 return;
5366
5367         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
5368             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
5369                 return; /* no tree diffs in patch format */
5370
5371         run_diff(p, o);
5372 }
5373
5374 static void diff_flush_stat(struct diff_filepair *p, struct diff_options *o,
5375                             struct diffstat_t *diffstat)
5376 {
5377         if (diff_unmodified_pair(p))
5378                 return;
5379
5380         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
5381             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
5382                 return; /* no useful stat for tree diffs */
5383
5384         run_diffstat(p, o, diffstat);
5385 }
5386
5387 static void diff_flush_checkdiff(struct diff_filepair *p,
5388                 struct diff_options *o)
5389 {
5390         if (diff_unmodified_pair(p))
5391                 return;
5392
5393         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
5394             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
5395                 return; /* nothing to check in tree diffs */
5396
5397         run_checkdiff(p, o);
5398 }
5399
5400 int diff_queue_is_empty(void)
5401 {
5402         struct diff_queue_struct *q = &diff_queued_diff;
5403         int i;
5404         for (i = 0; i < q->nr; i++)
5405                 if (!diff_unmodified_pair(q->queue[i]))
5406                         return 0;
5407         return 1;
5408 }
5409
5410 #if DIFF_DEBUG
5411 void diff_debug_filespec(struct diff_filespec *s, int x, const char *one)
5412 {
5413         fprintf(stderr, "queue[%d] %s (%s) %s %06o %s\n",
5414                 x, one ? one : "",
5415                 s->path,
5416                 DIFF_FILE_VALID(s) ? "valid" : "invalid",
5417                 s->mode,
5418                 s->oid_valid ? oid_to_hex(&s->oid) : "");
5419         fprintf(stderr, "queue[%d] %s size %lu\n",
5420                 x, one ? one : "",
5421                 s->size);
5422 }
5423
5424 void diff_debug_filepair(const struct diff_filepair *p, int i)
5425 {
5426         diff_debug_filespec(p->one, i, "one");
5427         diff_debug_filespec(p->two, i, "two");
5428         fprintf(stderr, "score %d, status %c rename_used %d broken %d\n",
5429                 p->score, p->status ? p->status : '?',
5430                 p->one->rename_used, p->broken_pair);
5431 }
5432
5433 void diff_debug_queue(const char *msg, struct diff_queue_struct *q)
5434 {
5435         int i;
5436         if (msg)
5437                 fprintf(stderr, "%s\n", msg);
5438         fprintf(stderr, "q->nr = %d\n", q->nr);
5439         for (i = 0; i < q->nr; i++) {
5440                 struct diff_filepair *p = q->queue[i];
5441                 diff_debug_filepair(p, i);
5442         }
5443 }
5444 #endif
5445
5446 static void diff_resolve_rename_copy(void)
5447 {
5448         int i;
5449         struct diff_filepair *p;
5450         struct diff_queue_struct *q = &diff_queued_diff;
5451
5452         diff_debug_queue("resolve-rename-copy", q);
5453
5454         for (i = 0; i < q->nr; i++) {
5455                 p = q->queue[i];
5456                 p->status = 0; /* undecided */
5457                 if (DIFF_PAIR_UNMERGED(p))
5458                         p->status = DIFF_STATUS_UNMERGED;
5459                 else if (!DIFF_FILE_VALID(p->one))
5460                         p->status = DIFF_STATUS_ADDED;
5461                 else if (!DIFF_FILE_VALID(p->two))
5462                         p->status = DIFF_STATUS_DELETED;
5463                 else if (DIFF_PAIR_TYPE_CHANGED(p))
5464                         p->status = DIFF_STATUS_TYPE_CHANGED;
5465
5466                 /* from this point on, we are dealing with a pair
5467                  * whose both sides are valid and of the same type, i.e.
5468                  * either in-place edit or rename/copy edit.
5469                  */
5470                 else if (DIFF_PAIR_RENAME(p)) {
5471                         /*
5472                          * A rename might have re-connected a broken
5473                          * pair up, causing the pathnames to be the
5474                          * same again. If so, that's not a rename at
5475                          * all, just a modification..
5476                          *
5477                          * Otherwise, see if this source was used for
5478                          * multiple renames, in which case we decrement
5479                          * the count, and call it a copy.
5480                          */
5481                         if (!strcmp(p->one->path, p->two->path))
5482                                 p->status = DIFF_STATUS_MODIFIED;
5483                         else if (--p->one->rename_used > 0)
5484                                 p->status = DIFF_STATUS_COPIED;
5485                         else
5486                                 p->status = DIFF_STATUS_RENAMED;
5487                 }
5488                 else if (oidcmp(&p->one->oid, &p->two->oid) ||
5489                          p->one->mode != p->two->mode ||
5490                          p->one->dirty_submodule ||
5491                          p->two->dirty_submodule ||
5492                          is_null_oid(&p->one->oid))
5493                         p->status = DIFF_STATUS_MODIFIED;
5494                 else {
5495                         /* This is a "no-change" entry and should not
5496                          * happen anymore, but prepare for broken callers.
5497                          */
5498                         error("feeding unmodified %s to diffcore",
5499                               p->one->path);
5500                         p->status = DIFF_STATUS_UNKNOWN;
5501                 }
5502         }
5503         diff_debug_queue("resolve-rename-copy done", q);
5504 }
5505
5506 static int check_pair_status(struct diff_filepair *p)
5507 {
5508         switch (p->status) {
5509         case DIFF_STATUS_UNKNOWN:
5510                 return 0;
5511         case 0:
5512                 die("internal error in diff-resolve-rename-copy");
5513         default:
5514                 return 1;
5515         }
5516 }
5517
5518 static void flush_one_pair(struct diff_filepair *p, struct diff_options *opt)
5519 {
5520         int fmt = opt->output_format;
5521
5522         if (fmt & DIFF_FORMAT_CHECKDIFF)
5523                 diff_flush_checkdiff(p, opt);
5524         else if (fmt & (DIFF_FORMAT_RAW | DIFF_FORMAT_NAME_STATUS))
5525                 diff_flush_raw(p, opt);
5526         else if (fmt & DIFF_FORMAT_NAME) {
5527                 const char *name_a, *name_b;
5528                 name_a = p->two->path;
5529                 name_b = NULL;
5530                 strip_prefix(opt->prefix_length, &name_a, &name_b);
5531                 fprintf(opt->file, "%s", diff_line_prefix(opt));
5532                 write_name_quoted(name_a, opt->file, opt->line_termination);
5533         }
5534 }
5535
5536 static void show_file_mode_name(struct diff_options *opt, const char *newdelete, struct diff_filespec *fs)
5537 {
5538         struct strbuf sb = STRBUF_INIT;
5539         if (fs->mode)
5540                 strbuf_addf(&sb, " %s mode %06o ", newdelete, fs->mode);
5541         else
5542                 strbuf_addf(&sb, " %s ", newdelete);
5543
5544         quote_c_style(fs->path, &sb, NULL, 0);
5545         strbuf_addch(&sb, '\n');
5546         emit_diff_symbol(opt, DIFF_SYMBOL_SUMMARY,
5547                          sb.buf, sb.len, 0);
5548         strbuf_release(&sb);
5549 }
5550
5551 static void show_mode_change(struct diff_options *opt, struct diff_filepair *p,
5552                 int show_name)
5553 {
5554         if (p->one->mode && p->two->mode && p->one->mode != p->two->mode) {
5555                 struct strbuf sb = STRBUF_INIT;
5556                 strbuf_addf(&sb, " mode change %06o => %06o",
5557                             p->one->mode, p->two->mode);
5558                 if (show_name) {
5559                         strbuf_addch(&sb, ' ');
5560                         quote_c_style(p->two->path, &sb, NULL, 0);
5561                 }
5562                 strbuf_addch(&sb, '\n');
5563                 emit_diff_symbol(opt, DIFF_SYMBOL_SUMMARY,
5564                                  sb.buf, sb.len, 0);
5565                 strbuf_release(&sb);
5566         }
5567 }
5568
5569 static void show_rename_copy(struct diff_options *opt, const char *renamecopy,
5570                 struct diff_filepair *p)
5571 {
5572         struct strbuf sb = STRBUF_INIT;
5573         struct strbuf names = STRBUF_INIT;
5574
5575         pprint_rename(&names, p->one->path, p->two->path);
5576         strbuf_addf(&sb, " %s %s (%d%%)\n",
5577                     renamecopy, names.buf, similarity_index(p));
5578         strbuf_release(&names);
5579         emit_diff_symbol(opt, DIFF_SYMBOL_SUMMARY,
5580                                  sb.buf, sb.len, 0);
5581         show_mode_change(opt, p, 0);
5582         strbuf_release(&sb);
5583 }
5584
5585 static void diff_summary(struct diff_options *opt, struct diff_filepair *p)
5586 {
5587         switch(p->status) {
5588         case DIFF_STATUS_DELETED:
5589                 show_file_mode_name(opt, "delete", p->one);
5590                 break;
5591         case DIFF_STATUS_ADDED:
5592                 show_file_mode_name(opt, "create", p->two);
5593                 break;
5594         case DIFF_STATUS_COPIED:
5595                 show_rename_copy(opt, "copy", p);
5596                 break;
5597         case DIFF_STATUS_RENAMED:
5598                 show_rename_copy(opt, "rename", p);
5599                 break;
5600         default:
5601                 if (p->score) {
5602                         struct strbuf sb = STRBUF_INIT;
5603                         strbuf_addstr(&sb, " rewrite ");
5604                         quote_c_style(p->two->path, &sb, NULL, 0);
5605                         strbuf_addf(&sb, " (%d%%)\n", similarity_index(p));
5606                         emit_diff_symbol(opt, DIFF_SYMBOL_SUMMARY,
5607                                          sb.buf, sb.len, 0);
5608                         strbuf_release(&sb);
5609                 }
5610                 show_mode_change(opt, p, !p->score);
5611                 break;
5612         }
5613 }
5614
5615 struct patch_id_t {
5616         git_SHA_CTX *ctx;
5617         int patchlen;
5618 };
5619
5620 static int remove_space(char *line, int len)
5621 {
5622         int i;
5623         char *dst = line;
5624         unsigned char c;
5625
5626         for (i = 0; i < len; i++)
5627                 if (!isspace((c = line[i])))
5628                         *dst++ = c;
5629
5630         return dst - line;
5631 }
5632
5633 static void patch_id_consume(void *priv, char *line, unsigned long len)
5634 {
5635         struct patch_id_t *data = priv;
5636         int new_len;
5637
5638         /* Ignore line numbers when computing the SHA1 of the patch */
5639         if (starts_with(line, "@@ -"))
5640                 return;
5641
5642         new_len = remove_space(line, len);
5643
5644         git_SHA1_Update(data->ctx, line, new_len);
5645         data->patchlen += new_len;
5646 }
5647
5648 static void patch_id_add_string(git_SHA_CTX *ctx, const char *str)
5649 {
5650         git_SHA1_Update(ctx, str, strlen(str));
5651 }
5652
5653 static void patch_id_add_mode(git_SHA_CTX *ctx, unsigned mode)
5654 {
5655         /* large enough for 2^32 in octal */
5656         char buf[12];
5657         int len = xsnprintf(buf, sizeof(buf), "%06o", mode);
5658         git_SHA1_Update(ctx, buf, len);
5659 }
5660
5661 /* returns 0 upon success, and writes result into sha1 */
5662 static int diff_get_patch_id(struct diff_options *options, struct object_id *oid, int diff_header_only)
5663 {
5664         struct diff_queue_struct *q = &diff_queued_diff;
5665         int i;
5666         git_SHA_CTX ctx;
5667         struct patch_id_t data;
5668
5669         git_SHA1_Init(&ctx);
5670         memset(&data, 0, sizeof(struct patch_id_t));
5671         data.ctx = &ctx;
5672
5673         for (i = 0; i < q->nr; i++) {
5674                 xpparam_t xpp;
5675                 xdemitconf_t xecfg;
5676                 mmfile_t mf1, mf2;
5677                 struct diff_filepair *p = q->queue[i];
5678                 int len1, len2;
5679
5680                 memset(&xpp, 0, sizeof(xpp));
5681                 memset(&xecfg, 0, sizeof(xecfg));
5682                 if (p->status == 0)
5683                         return error("internal diff status error");
5684                 if (p->status == DIFF_STATUS_UNKNOWN)
5685                         continue;
5686                 if (diff_unmodified_pair(p))
5687                         continue;
5688                 if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
5689                     (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
5690                         continue;
5691                 if (DIFF_PAIR_UNMERGED(p))
5692                         continue;
5693
5694                 diff_fill_oid_info(p->one, options->repo->index);
5695                 diff_fill_oid_info(p->two, options->repo->index);
5696
5697                 len1 = remove_space(p->one->path, strlen(p->one->path));
5698                 len2 = remove_space(p->two->path, strlen(p->two->path));
5699                 patch_id_add_string(&ctx, "diff--git");
5700                 patch_id_add_string(&ctx, "a/");
5701                 git_SHA1_Update(&ctx, p->one->path, len1);
5702                 patch_id_add_string(&ctx, "b/");
5703                 git_SHA1_Update(&ctx, p->two->path, len2);
5704
5705                 if (p->one->mode == 0) {
5706                         patch_id_add_string(&ctx, "newfilemode");
5707                         patch_id_add_mode(&ctx, p->two->mode);
5708                         patch_id_add_string(&ctx, "---/dev/null");
5709                         patch_id_add_string(&ctx, "+++b/");
5710                         git_SHA1_Update(&ctx, p->two->path, len2);
5711                 } else if (p->two->mode == 0) {
5712                         patch_id_add_string(&ctx, "deletedfilemode");
5713                         patch_id_add_mode(&ctx, p->one->mode);
5714                         patch_id_add_string(&ctx, "---a/");
5715                         git_SHA1_Update(&ctx, p->one->path, len1);
5716                         patch_id_add_string(&ctx, "+++/dev/null");
5717                 } else {
5718                         patch_id_add_string(&ctx, "---a/");
5719                         git_SHA1_Update(&ctx, p->one->path, len1);
5720                         patch_id_add_string(&ctx, "+++b/");
5721                         git_SHA1_Update(&ctx, p->two->path, len2);
5722                 }
5723
5724                 if (diff_header_only)
5725                         continue;
5726
5727                 if (fill_mmfile(options->repo, &mf1, p->one) < 0 ||
5728                     fill_mmfile(options->repo, &mf2, p->two) < 0)
5729                         return error("unable to read files to diff");
5730
5731                 if (diff_filespec_is_binary(options->repo, p->one) ||
5732                     diff_filespec_is_binary(options->repo, p->two)) {
5733                         git_SHA1_Update(&ctx, oid_to_hex(&p->one->oid),
5734                                         GIT_SHA1_HEXSZ);
5735                         git_SHA1_Update(&ctx, oid_to_hex(&p->two->oid),
5736                                         GIT_SHA1_HEXSZ);
5737                         continue;
5738                 }
5739
5740                 xpp.flags = 0;
5741                 xecfg.ctxlen = 3;
5742                 xecfg.flags = 0;
5743                 if (xdi_diff_outf(&mf1, &mf2, patch_id_consume, &data,
5744                                   &xpp, &xecfg))
5745                         return error("unable to generate patch-id diff for %s",
5746                                      p->one->path);
5747         }
5748
5749         git_SHA1_Final(oid->hash, &ctx);
5750         return 0;
5751 }
5752
5753 int diff_flush_patch_id(struct diff_options *options, struct object_id *oid, int diff_header_only)
5754 {
5755         struct diff_queue_struct *q = &diff_queued_diff;
5756         int i;
5757         int result = diff_get_patch_id(options, oid, diff_header_only);
5758
5759         for (i = 0; i < q->nr; i++)
5760                 diff_free_filepair(q->queue[i]);
5761
5762         free(q->queue);
5763         DIFF_QUEUE_CLEAR(q);
5764
5765         return result;
5766 }
5767
5768 static int is_summary_empty(const struct diff_queue_struct *q)
5769 {
5770         int i;
5771
5772         for (i = 0; i < q->nr; i++) {
5773                 const struct diff_filepair *p = q->queue[i];
5774
5775                 switch (p->status) {
5776                 case DIFF_STATUS_DELETED:
5777                 case DIFF_STATUS_ADDED:
5778                 case DIFF_STATUS_COPIED:
5779                 case DIFF_STATUS_RENAMED:
5780                         return 0;
5781                 default:
5782                         if (p->score)
5783                                 return 0;
5784                         if (p->one->mode && p->two->mode &&
5785                             p->one->mode != p->two->mode)
5786                                 return 0;
5787                         break;
5788                 }
5789         }
5790         return 1;
5791 }
5792
5793 static const char rename_limit_warning[] =
5794 N_("inexact rename detection was skipped due to too many files.");
5795
5796 static const char degrade_cc_to_c_warning[] =
5797 N_("only found copies from modified paths due to too many files.");
5798
5799 static const char rename_limit_advice[] =
5800 N_("you may want to set your %s variable to at least "
5801    "%d and retry the command.");
5802
5803 void diff_warn_rename_limit(const char *varname, int needed, int degraded_cc)
5804 {
5805         fflush(stdout);
5806         if (degraded_cc)
5807                 warning(_(degrade_cc_to_c_warning));
5808         else if (needed)
5809                 warning(_(rename_limit_warning));
5810         else
5811                 return;
5812         if (0 < needed)
5813                 warning(_(rename_limit_advice), varname, needed);
5814 }
5815
5816 static void diff_flush_patch_all_file_pairs(struct diff_options *o)
5817 {
5818         int i;
5819         static struct emitted_diff_symbols esm = EMITTED_DIFF_SYMBOLS_INIT;
5820         struct diff_queue_struct *q = &diff_queued_diff;
5821
5822         if (WSEH_NEW & WS_RULE_MASK)
5823                 BUG("WS rules bit mask overlaps with diff symbol flags");
5824
5825         if (o->color_moved)
5826                 o->emitted_symbols = &esm;
5827
5828         for (i = 0; i < q->nr; i++) {
5829                 struct diff_filepair *p = q->queue[i];
5830                 if (check_pair_status(p))
5831                         diff_flush_patch(p, o);
5832         }
5833
5834         if (o->emitted_symbols) {
5835                 if (o->color_moved) {
5836                         struct hashmap add_lines, del_lines;
5837
5838                         if (o->color_moved_ws_handling &
5839                             COLOR_MOVED_WS_ALLOW_INDENTATION_CHANGE)
5840                                 o->color_moved_ws_handling |= XDF_IGNORE_WHITESPACE;
5841
5842                         hashmap_init(&del_lines, moved_entry_cmp, o, 0);
5843                         hashmap_init(&add_lines, moved_entry_cmp, o, 0);
5844
5845                         add_lines_to_move_detection(o, &add_lines, &del_lines);
5846                         mark_color_as_moved(o, &add_lines, &del_lines);
5847                         if (o->color_moved == COLOR_MOVED_ZEBRA_DIM)
5848                                 dim_moved_lines(o);
5849
5850                         hashmap_free(&add_lines, 0);
5851                         hashmap_free(&del_lines, 0);
5852                 }
5853
5854                 for (i = 0; i < esm.nr; i++)
5855                         emit_diff_symbol_from_struct(o, &esm.buf[i]);
5856
5857                 for (i = 0; i < esm.nr; i++)
5858                         free((void *)esm.buf[i].line);
5859         }
5860         esm.nr = 0;
5861 }
5862
5863 void diff_flush(struct diff_options *options)
5864 {
5865         struct diff_queue_struct *q = &diff_queued_diff;
5866         int i, output_format = options->output_format;
5867         int separator = 0;
5868         int dirstat_by_line = 0;
5869
5870         /*
5871          * Order: raw, stat, summary, patch
5872          * or:    name/name-status/checkdiff (other bits clear)
5873          */
5874         if (!q->nr)
5875                 goto free_queue;
5876
5877         if (output_format & (DIFF_FORMAT_RAW |
5878                              DIFF_FORMAT_NAME |
5879                              DIFF_FORMAT_NAME_STATUS |
5880                              DIFF_FORMAT_CHECKDIFF)) {
5881                 for (i = 0; i < q->nr; i++) {
5882                         struct diff_filepair *p = q->queue[i];
5883                         if (check_pair_status(p))
5884                                 flush_one_pair(p, options);
5885                 }
5886                 separator++;
5887         }
5888
5889         if (output_format & DIFF_FORMAT_DIRSTAT && options->flags.dirstat_by_line)
5890                 dirstat_by_line = 1;
5891
5892         if (output_format & (DIFF_FORMAT_DIFFSTAT|DIFF_FORMAT_SHORTSTAT|DIFF_FORMAT_NUMSTAT) ||
5893             dirstat_by_line) {
5894                 struct diffstat_t diffstat;
5895
5896                 memset(&diffstat, 0, sizeof(struct diffstat_t));
5897                 for (i = 0; i < q->nr; i++) {
5898                         struct diff_filepair *p = q->queue[i];
5899                         if (check_pair_status(p))
5900                                 diff_flush_stat(p, options, &diffstat);
5901                 }
5902                 if (output_format & DIFF_FORMAT_NUMSTAT)
5903                         show_numstat(&diffstat, options);
5904                 if (output_format & DIFF_FORMAT_DIFFSTAT)
5905                         show_stats(&diffstat, options);
5906                 if (output_format & DIFF_FORMAT_SHORTSTAT)
5907                         show_shortstats(&diffstat, options);
5908                 if (output_format & DIFF_FORMAT_DIRSTAT && dirstat_by_line)
5909                         show_dirstat_by_line(&diffstat, options);
5910                 free_diffstat_info(&diffstat);
5911                 separator++;
5912         }
5913         if ((output_format & DIFF_FORMAT_DIRSTAT) && !dirstat_by_line)
5914                 show_dirstat(options);
5915
5916         if (output_format & DIFF_FORMAT_SUMMARY && !is_summary_empty(q)) {
5917                 for (i = 0; i < q->nr; i++) {
5918                         diff_summary(options, q->queue[i]);
5919                 }
5920                 separator++;
5921         }
5922
5923         if (output_format & DIFF_FORMAT_NO_OUTPUT &&
5924             options->flags.exit_with_status &&
5925             options->flags.diff_from_contents) {
5926                 /*
5927                  * run diff_flush_patch for the exit status. setting
5928                  * options->file to /dev/null should be safe, because we
5929                  * aren't supposed to produce any output anyway.
5930                  */
5931                 if (options->close_file)
5932                         fclose(options->file);
5933                 options->file = xfopen("/dev/null", "w");
5934                 options->close_file = 1;
5935                 options->color_moved = 0;
5936                 for (i = 0; i < q->nr; i++) {
5937                         struct diff_filepair *p = q->queue[i];
5938                         if (check_pair_status(p))
5939                                 diff_flush_patch(p, options);
5940                         if (options->found_changes)
5941                                 break;
5942                 }
5943         }
5944
5945         if (output_format & DIFF_FORMAT_PATCH) {
5946                 if (separator) {
5947                         emit_diff_symbol(options, DIFF_SYMBOL_SEPARATOR, NULL, 0, 0);
5948                         if (options->stat_sep)
5949                                 /* attach patch instead of inline */
5950                                 emit_diff_symbol(options, DIFF_SYMBOL_STAT_SEP,
5951                                                  NULL, 0, 0);
5952                 }
5953
5954                 diff_flush_patch_all_file_pairs(options);
5955         }
5956
5957         if (output_format & DIFF_FORMAT_CALLBACK)
5958                 options->format_callback(q, options, options->format_callback_data);
5959
5960         for (i = 0; i < q->nr; i++)
5961                 diff_free_filepair(q->queue[i]);
5962 free_queue:
5963         free(q->queue);
5964         DIFF_QUEUE_CLEAR(q);
5965         if (options->close_file)
5966                 fclose(options->file);
5967
5968         /*
5969          * Report the content-level differences with HAS_CHANGES;
5970          * diff_addremove/diff_change does not set the bit when
5971          * DIFF_FROM_CONTENTS is in effect (e.g. with -w).
5972          */
5973         if (options->flags.diff_from_contents) {
5974                 if (options->found_changes)
5975                         options->flags.has_changes = 1;
5976                 else
5977                         options->flags.has_changes = 0;
5978         }
5979 }
5980
5981 static int match_filter(const struct diff_options *options, const struct diff_filepair *p)
5982 {
5983         return (((p->status == DIFF_STATUS_MODIFIED) &&
5984                  ((p->score &&
5985                    filter_bit_tst(DIFF_STATUS_FILTER_BROKEN, options)) ||
5986                   (!p->score &&
5987                    filter_bit_tst(DIFF_STATUS_MODIFIED, options)))) ||
5988                 ((p->status != DIFF_STATUS_MODIFIED) &&
5989                  filter_bit_tst(p->status, options)));
5990 }
5991
5992 static void diffcore_apply_filter(struct diff_options *options)
5993 {
5994         int i;
5995         struct diff_queue_struct *q = &diff_queued_diff;
5996         struct diff_queue_struct outq;
5997
5998         DIFF_QUEUE_CLEAR(&outq);
5999
6000         if (!options->filter)
6001                 return;
6002
6003         if (filter_bit_tst(DIFF_STATUS_FILTER_AON, options)) {
6004                 int found;
6005                 for (i = found = 0; !found && i < q->nr; i++) {
6006                         if (match_filter(options, q->queue[i]))
6007                                 found++;
6008                 }
6009                 if (found)
6010                         return;
6011
6012                 /* otherwise we will clear the whole queue
6013                  * by copying the empty outq at the end of this
6014                  * function, but first clear the current entries
6015                  * in the queue.
6016                  */
6017                 for (i = 0; i < q->nr; i++)
6018                         diff_free_filepair(q->queue[i]);
6019         }
6020         else {
6021                 /* Only the matching ones */
6022                 for (i = 0; i < q->nr; i++) {
6023                         struct diff_filepair *p = q->queue[i];
6024                         if (match_filter(options, p))
6025                                 diff_q(&outq, p);
6026                         else
6027                                 diff_free_filepair(p);
6028                 }
6029         }
6030         free(q->queue);
6031         *q = outq;
6032 }
6033
6034 /* Check whether two filespecs with the same mode and size are identical */
6035 static int diff_filespec_is_identical(struct repository *r,
6036                                       struct diff_filespec *one,
6037                                       struct diff_filespec *two)
6038 {
6039         if (S_ISGITLINK(one->mode))
6040                 return 0;
6041         if (diff_populate_filespec(r, one, 0))
6042                 return 0;
6043         if (diff_populate_filespec(r, two, 0))
6044                 return 0;
6045         return !memcmp(one->data, two->data, one->size);
6046 }
6047
6048 static int diff_filespec_check_stat_unmatch(struct repository *r,
6049                                             struct diff_filepair *p)
6050 {
6051         if (p->done_skip_stat_unmatch)
6052                 return p->skip_stat_unmatch_result;
6053
6054         p->done_skip_stat_unmatch = 1;
6055         p->skip_stat_unmatch_result = 0;
6056         /*
6057          * 1. Entries that come from stat info dirtiness
6058          *    always have both sides (iow, not create/delete),
6059          *    one side of the object name is unknown, with
6060          *    the same mode and size.  Keep the ones that
6061          *    do not match these criteria.  They have real
6062          *    differences.
6063          *
6064          * 2. At this point, the file is known to be modified,
6065          *    with the same mode and size, and the object
6066          *    name of one side is unknown.  Need to inspect
6067          *    the identical contents.
6068          */
6069         if (!DIFF_FILE_VALID(p->one) || /* (1) */
6070             !DIFF_FILE_VALID(p->two) ||
6071             (p->one->oid_valid && p->two->oid_valid) ||
6072             (p->one->mode != p->two->mode) ||
6073             diff_populate_filespec(r, p->one, CHECK_SIZE_ONLY) ||
6074             diff_populate_filespec(r, p->two, CHECK_SIZE_ONLY) ||
6075             (p->one->size != p->two->size) ||
6076             !diff_filespec_is_identical(r, p->one, p->two)) /* (2) */
6077                 p->skip_stat_unmatch_result = 1;
6078         return p->skip_stat_unmatch_result;
6079 }
6080
6081 static void diffcore_skip_stat_unmatch(struct diff_options *diffopt)
6082 {
6083         int i;
6084         struct diff_queue_struct *q = &diff_queued_diff;
6085         struct diff_queue_struct outq;
6086         DIFF_QUEUE_CLEAR(&outq);
6087
6088         for (i = 0; i < q->nr; i++) {
6089                 struct diff_filepair *p = q->queue[i];
6090
6091                 if (diff_filespec_check_stat_unmatch(diffopt->repo, p))
6092                         diff_q(&outq, p);
6093                 else {
6094                         /*
6095                          * The caller can subtract 1 from skip_stat_unmatch
6096                          * to determine how many paths were dirty only
6097                          * due to stat info mismatch.
6098                          */
6099                         if (!diffopt->flags.no_index)
6100                                 diffopt->skip_stat_unmatch++;
6101                         diff_free_filepair(p);
6102                 }
6103         }
6104         free(q->queue);
6105         *q = outq;
6106 }
6107
6108 static int diffnamecmp(const void *a_, const void *b_)
6109 {
6110         const struct diff_filepair *a = *((const struct diff_filepair **)a_);
6111         const struct diff_filepair *b = *((const struct diff_filepair **)b_);
6112         const char *name_a, *name_b;
6113
6114         name_a = a->one ? a->one->path : a->two->path;
6115         name_b = b->one ? b->one->path : b->two->path;
6116         return strcmp(name_a, name_b);
6117 }
6118
6119 void diffcore_fix_diff_index(struct diff_options *options)
6120 {
6121         struct diff_queue_struct *q = &diff_queued_diff;
6122         QSORT(q->queue, q->nr, diffnamecmp);
6123 }
6124
6125 void diffcore_std(struct diff_options *options)
6126 {
6127         /* NOTE please keep the following in sync with diff_tree_combined() */
6128         if (options->skip_stat_unmatch)
6129                 diffcore_skip_stat_unmatch(options);
6130         if (!options->found_follow) {
6131                 /* See try_to_follow_renames() in tree-diff.c */
6132                 if (options->break_opt != -1)
6133                         diffcore_break(options->repo,
6134                                        options->break_opt);
6135                 if (options->detect_rename)
6136                         diffcore_rename(options);
6137                 if (options->break_opt != -1)
6138                         diffcore_merge_broken();
6139         }
6140         if (options->pickaxe_opts & DIFF_PICKAXE_KINDS_MASK)
6141                 diffcore_pickaxe(options);
6142         if (options->orderfile)
6143                 diffcore_order(options->orderfile);
6144         if (!options->found_follow)
6145                 /* See try_to_follow_renames() in tree-diff.c */
6146                 diff_resolve_rename_copy();
6147         diffcore_apply_filter(options);
6148
6149         if (diff_queued_diff.nr && !options->flags.diff_from_contents)
6150                 options->flags.has_changes = 1;
6151         else
6152                 options->flags.has_changes = 0;
6153
6154         options->found_follow = 0;
6155 }
6156
6157 int diff_result_code(struct diff_options *opt, int status)
6158 {
6159         int result = 0;
6160
6161         diff_warn_rename_limit("diff.renameLimit",
6162                                opt->needed_rename_limit,
6163                                opt->degraded_cc_to_c);
6164         if (!opt->flags.exit_with_status &&
6165             !(opt->output_format & DIFF_FORMAT_CHECKDIFF))
6166                 return status;
6167         if (opt->flags.exit_with_status &&
6168             opt->flags.has_changes)
6169                 result |= 01;
6170         if ((opt->output_format & DIFF_FORMAT_CHECKDIFF) &&
6171             opt->flags.check_failed)
6172                 result |= 02;
6173         return result;
6174 }
6175
6176 int diff_can_quit_early(struct diff_options *opt)
6177 {
6178         return (opt->flags.quick &&
6179                 !opt->filter &&
6180                 opt->flags.has_changes);
6181 }
6182
6183 /*
6184  * Shall changes to this submodule be ignored?
6185  *
6186  * Submodule changes can be configured to be ignored separately for each path,
6187  * but that configuration can be overridden from the command line.
6188  */
6189 static int is_submodule_ignored(const char *path, struct diff_options *options)
6190 {
6191         int ignored = 0;
6192         struct diff_flags orig_flags = options->flags;
6193         if (!options->flags.override_submodule_config)
6194                 set_diffopt_flags_from_submodule_config(options, path);
6195         if (options->flags.ignore_submodules)
6196                 ignored = 1;
6197         options->flags = orig_flags;
6198         return ignored;
6199 }
6200
6201 void diff_addremove(struct diff_options *options,
6202                     int addremove, unsigned mode,
6203                     const struct object_id *oid,
6204                     int oid_valid,
6205                     const char *concatpath, unsigned dirty_submodule)
6206 {
6207         struct diff_filespec *one, *two;
6208
6209         if (S_ISGITLINK(mode) && is_submodule_ignored(concatpath, options))
6210                 return;
6211
6212         /* This may look odd, but it is a preparation for
6213          * feeding "there are unchanged files which should
6214          * not produce diffs, but when you are doing copy
6215          * detection you would need them, so here they are"
6216          * entries to the diff-core.  They will be prefixed
6217          * with something like '=' or '*' (I haven't decided
6218          * which but should not make any difference).
6219          * Feeding the same new and old to diff_change()
6220          * also has the same effect.
6221          * Before the final output happens, they are pruned after
6222          * merged into rename/copy pairs as appropriate.
6223          */
6224         if (options->flags.reverse_diff)
6225                 addremove = (addremove == '+' ? '-' :
6226                              addremove == '-' ? '+' : addremove);
6227
6228         if (options->prefix &&
6229             strncmp(concatpath, options->prefix, options->prefix_length))
6230                 return;
6231
6232         one = alloc_filespec(concatpath);
6233         two = alloc_filespec(concatpath);
6234
6235         if (addremove != '+')
6236                 fill_filespec(one, oid, oid_valid, mode);
6237         if (addremove != '-') {
6238                 fill_filespec(two, oid, oid_valid, mode);
6239                 two->dirty_submodule = dirty_submodule;
6240         }
6241
6242         diff_queue(&diff_queued_diff, one, two);
6243         if (!options->flags.diff_from_contents)
6244                 options->flags.has_changes = 1;
6245 }
6246
6247 void diff_change(struct diff_options *options,
6248                  unsigned old_mode, unsigned new_mode,
6249                  const struct object_id *old_oid,
6250                  const struct object_id *new_oid,
6251                  int old_oid_valid, int new_oid_valid,
6252                  const char *concatpath,
6253                  unsigned old_dirty_submodule, unsigned new_dirty_submodule)
6254 {
6255         struct diff_filespec *one, *two;
6256         struct diff_filepair *p;
6257
6258         if (S_ISGITLINK(old_mode) && S_ISGITLINK(new_mode) &&
6259             is_submodule_ignored(concatpath, options))
6260                 return;
6261
6262         if (options->flags.reverse_diff) {
6263                 SWAP(old_mode, new_mode);
6264                 SWAP(old_oid, new_oid);
6265                 SWAP(old_oid_valid, new_oid_valid);
6266                 SWAP(old_dirty_submodule, new_dirty_submodule);
6267         }
6268
6269         if (options->prefix &&
6270             strncmp(concatpath, options->prefix, options->prefix_length))
6271                 return;
6272
6273         one = alloc_filespec(concatpath);
6274         two = alloc_filespec(concatpath);
6275         fill_filespec(one, old_oid, old_oid_valid, old_mode);
6276         fill_filespec(two, new_oid, new_oid_valid, new_mode);
6277         one->dirty_submodule = old_dirty_submodule;
6278         two->dirty_submodule = new_dirty_submodule;
6279         p = diff_queue(&diff_queued_diff, one, two);
6280
6281         if (options->flags.diff_from_contents)
6282                 return;
6283
6284         if (options->flags.quick && options->skip_stat_unmatch &&
6285             !diff_filespec_check_stat_unmatch(options->repo, p))
6286                 return;
6287
6288         options->flags.has_changes = 1;
6289 }
6290
6291 struct diff_filepair *diff_unmerge(struct diff_options *options, const char *path)
6292 {
6293         struct diff_filepair *pair;
6294         struct diff_filespec *one, *two;
6295
6296         if (options->prefix &&
6297             strncmp(path, options->prefix, options->prefix_length))
6298                 return NULL;
6299
6300         one = alloc_filespec(path);
6301         two = alloc_filespec(path);
6302         pair = diff_queue(&diff_queued_diff, one, two);
6303         pair->is_unmerged = 1;
6304         return pair;
6305 }
6306
6307 static char *run_textconv(struct repository *r,
6308                           const char *pgm,
6309                           struct diff_filespec *spec,
6310                           size_t *outsize)
6311 {
6312         struct diff_tempfile *temp;
6313         const char *argv[3];
6314         const char **arg = argv;
6315         struct child_process child = CHILD_PROCESS_INIT;
6316         struct strbuf buf = STRBUF_INIT;
6317         int err = 0;
6318
6319         temp = prepare_temp_file(r, spec->path, spec);
6320         *arg++ = pgm;
6321         *arg++ = temp->name;
6322         *arg = NULL;
6323
6324         child.use_shell = 1;
6325         child.argv = argv;
6326         child.out = -1;
6327         if (start_command(&child)) {
6328                 remove_tempfile();
6329                 return NULL;
6330         }
6331
6332         if (strbuf_read(&buf, child.out, 0) < 0)
6333                 err = error("error reading from textconv command '%s'", pgm);
6334         close(child.out);
6335
6336         if (finish_command(&child) || err) {
6337                 strbuf_release(&buf);
6338                 remove_tempfile();
6339                 return NULL;
6340         }
6341         remove_tempfile();
6342
6343         return strbuf_detach(&buf, outsize);
6344 }
6345
6346 size_t fill_textconv(struct repository *r,
6347                      struct userdiff_driver *driver,
6348                      struct diff_filespec *df,
6349                      char **outbuf)
6350 {
6351         size_t size;
6352
6353         if (!driver) {
6354                 if (!DIFF_FILE_VALID(df)) {
6355                         *outbuf = "";
6356                         return 0;
6357                 }
6358                 if (diff_populate_filespec(r, df, 0))
6359                         die("unable to read files to diff");
6360                 *outbuf = df->data;
6361                 return df->size;
6362         }
6363
6364         if (!driver->textconv)
6365                 BUG("fill_textconv called with non-textconv driver");
6366
6367         if (driver->textconv_cache && df->oid_valid) {
6368                 *outbuf = notes_cache_get(driver->textconv_cache,
6369                                           &df->oid,
6370                                           &size);
6371                 if (*outbuf)
6372                         return size;
6373         }
6374
6375         *outbuf = run_textconv(r, driver->textconv, df, &size);
6376         if (!*outbuf)
6377                 die("unable to read files to diff");
6378
6379         if (driver->textconv_cache && df->oid_valid) {
6380                 /* ignore errors, as we might be in a readonly repository */
6381                 notes_cache_put(driver->textconv_cache, &df->oid, *outbuf,
6382                                 size);
6383                 /*
6384                  * we could save up changes and flush them all at the end,
6385                  * but we would need an extra call after all diffing is done.
6386                  * Since generating a cache entry is the slow path anyway,
6387                  * this extra overhead probably isn't a big deal.
6388                  */
6389                 notes_cache_write(driver->textconv_cache);
6390         }
6391
6392         return size;
6393 }
6394
6395 int textconv_object(struct repository *r,
6396                     const char *path,
6397                     unsigned mode,
6398                     const struct object_id *oid,
6399                     int oid_valid,
6400                     char **buf,
6401                     unsigned long *buf_size)
6402 {
6403         struct diff_filespec *df;
6404         struct userdiff_driver *textconv;
6405
6406         df = alloc_filespec(path);
6407         fill_filespec(df, oid, oid_valid, mode);
6408         textconv = get_textconv(r->index, df);
6409         if (!textconv) {
6410                 free_filespec(df);
6411                 return 0;
6412         }
6413
6414         *buf_size = fill_textconv(r, textconv, df, buf);
6415         free_filespec(df);
6416         return 1;
6417 }
6418
6419 void setup_diff_pager(struct diff_options *opt)
6420 {
6421         /*
6422          * If the user asked for our exit code, then either they want --quiet
6423          * or --exit-code. We should definitely not bother with a pager in the
6424          * former case, as we will generate no output. Since we still properly
6425          * report our exit code even when a pager is run, we _could_ run a
6426          * pager with --exit-code. But since we have not done so historically,
6427          * and because it is easy to find people oneline advising "git diff
6428          * --exit-code" in hooks and other scripts, we do not do so.
6429          */
6430         if (!opt->flags.exit_with_status &&
6431             check_pager_config("diff") != 0)
6432                 setup_pager();
6433 }