Merge branch 'jk/maint-fast-import-doc-reorder'
[git] / diff.c
1 /*
2  * Copyright (C) 2005 Junio C Hamano
3  */
4 #include "cache.h"
5 #include "quote.h"
6 #include "diff.h"
7 #include "diffcore.h"
8 #include "delta.h"
9 #include "xdiff-interface.h"
10 #include "color.h"
11 #include "attr.h"
12 #include "run-command.h"
13 #include "utf8.h"
14 #include "userdiff.h"
15 #include "sigchain.h"
16 #include "submodule.h"
17 #include "ll-merge.h"
18 #include "string-list.h"
19
20 #ifdef NO_FAST_WORKING_DIRECTORY
21 #define FAST_WORKING_DIRECTORY 0
22 #else
23 #define FAST_WORKING_DIRECTORY 1
24 #endif
25
26 static int diff_detect_rename_default;
27 static int diff_rename_limit_default = 400;
28 static int diff_suppress_blank_empty;
29 static int diff_use_color_default = -1;
30 static int diff_context_default = 3;
31 static const char *diff_word_regex_cfg;
32 static const char *external_diff_cmd_cfg;
33 int diff_auto_refresh_index = 1;
34 static int diff_mnemonic_prefix;
35 static int diff_no_prefix;
36 static int diff_stat_graph_width;
37 static int diff_dirstat_permille_default = 30;
38 static struct diff_options default_diff_options;
39
40 static char diff_colors[][COLOR_MAXLEN] = {
41         GIT_COLOR_RESET,
42         GIT_COLOR_NORMAL,       /* PLAIN */
43         GIT_COLOR_BOLD,         /* METAINFO */
44         GIT_COLOR_CYAN,         /* FRAGINFO */
45         GIT_COLOR_RED,          /* OLD */
46         GIT_COLOR_GREEN,        /* NEW */
47         GIT_COLOR_YELLOW,       /* COMMIT */
48         GIT_COLOR_BG_RED,       /* WHITESPACE */
49         GIT_COLOR_NORMAL,       /* FUNCINFO */
50 };
51
52 static int parse_diff_color_slot(const char *var, int ofs)
53 {
54         if (!strcasecmp(var+ofs, "plain"))
55                 return DIFF_PLAIN;
56         if (!strcasecmp(var+ofs, "meta"))
57                 return DIFF_METAINFO;
58         if (!strcasecmp(var+ofs, "frag"))
59                 return DIFF_FRAGINFO;
60         if (!strcasecmp(var+ofs, "old"))
61                 return DIFF_FILE_OLD;
62         if (!strcasecmp(var+ofs, "new"))
63                 return DIFF_FILE_NEW;
64         if (!strcasecmp(var+ofs, "commit"))
65                 return DIFF_COMMIT;
66         if (!strcasecmp(var+ofs, "whitespace"))
67                 return DIFF_WHITESPACE;
68         if (!strcasecmp(var+ofs, "func"))
69                 return DIFF_FUNCINFO;
70         return -1;
71 }
72
73 static int parse_dirstat_params(struct diff_options *options, const char *params_string,
74                                 struct strbuf *errmsg)
75 {
76         char *params_copy = xstrdup(params_string);
77         struct string_list params = STRING_LIST_INIT_NODUP;
78         int ret = 0;
79         int i;
80
81         if (*params_copy)
82                 string_list_split_in_place(&params, params_copy, ',', -1);
83         for (i = 0; i < params.nr; i++) {
84                 const char *p = params.items[i].string;
85                 if (!strcmp(p, "changes")) {
86                         DIFF_OPT_CLR(options, DIRSTAT_BY_LINE);
87                         DIFF_OPT_CLR(options, DIRSTAT_BY_FILE);
88                 } else if (!strcmp(p, "lines")) {
89                         DIFF_OPT_SET(options, DIRSTAT_BY_LINE);
90                         DIFF_OPT_CLR(options, DIRSTAT_BY_FILE);
91                 } else if (!strcmp(p, "files")) {
92                         DIFF_OPT_CLR(options, DIRSTAT_BY_LINE);
93                         DIFF_OPT_SET(options, DIRSTAT_BY_FILE);
94                 } else if (!strcmp(p, "noncumulative")) {
95                         DIFF_OPT_CLR(options, DIRSTAT_CUMULATIVE);
96                 } else if (!strcmp(p, "cumulative")) {
97                         DIFF_OPT_SET(options, DIRSTAT_CUMULATIVE);
98                 } else if (isdigit(*p)) {
99                         char *end;
100                         int permille = strtoul(p, &end, 10) * 10;
101                         if (*end == '.' && isdigit(*++end)) {
102                                 /* only use first digit */
103                                 permille += *end - '0';
104                                 /* .. and ignore any further digits */
105                                 while (isdigit(*++end))
106                                         ; /* nothing */
107                         }
108                         if (!*end)
109                                 options->dirstat_permille = permille;
110                         else {
111                                 strbuf_addf(errmsg, _("  Failed to parse dirstat cut-off percentage '%s'\n"),
112                                             p);
113                                 ret++;
114                         }
115                 } else {
116                         strbuf_addf(errmsg, _("  Unknown dirstat parameter '%s'\n"), p);
117                         ret++;
118                 }
119
120         }
121         string_list_clear(&params, 0);
122         free(params_copy);
123         return ret;
124 }
125
126 static int parse_submodule_params(struct diff_options *options, const char *value)
127 {
128         if (!strcmp(value, "log"))
129                 DIFF_OPT_SET(options, SUBMODULE_LOG);
130         else if (!strcmp(value, "short"))
131                 DIFF_OPT_CLR(options, SUBMODULE_LOG);
132         else
133                 return -1;
134         return 0;
135 }
136
137 static int git_config_rename(const char *var, const char *value)
138 {
139         if (!value)
140                 return DIFF_DETECT_RENAME;
141         if (!strcasecmp(value, "copies") || !strcasecmp(value, "copy"))
142                 return  DIFF_DETECT_COPY;
143         return git_config_bool(var,value) ? DIFF_DETECT_RENAME : 0;
144 }
145
146 /*
147  * These are to give UI layer defaults.
148  * The core-level commands such as git-diff-files should
149  * never be affected by the setting of diff.renames
150  * the user happens to have in the configuration file.
151  */
152 int git_diff_ui_config(const char *var, const char *value, void *cb)
153 {
154         if (!strcmp(var, "diff.color") || !strcmp(var, "color.diff")) {
155                 diff_use_color_default = git_config_colorbool(var, value);
156                 return 0;
157         }
158         if (!strcmp(var, "diff.context")) {
159                 diff_context_default = git_config_int(var, value);
160                 if (diff_context_default < 0)
161                         return -1;
162                 return 0;
163         }
164         if (!strcmp(var, "diff.renames")) {
165                 diff_detect_rename_default = git_config_rename(var, value);
166                 return 0;
167         }
168         if (!strcmp(var, "diff.autorefreshindex")) {
169                 diff_auto_refresh_index = git_config_bool(var, value);
170                 return 0;
171         }
172         if (!strcmp(var, "diff.mnemonicprefix")) {
173                 diff_mnemonic_prefix = git_config_bool(var, value);
174                 return 0;
175         }
176         if (!strcmp(var, "diff.noprefix")) {
177                 diff_no_prefix = git_config_bool(var, value);
178                 return 0;
179         }
180         if (!strcmp(var, "diff.statgraphwidth")) {
181                 diff_stat_graph_width = git_config_int(var, value);
182                 return 0;
183         }
184         if (!strcmp(var, "diff.external"))
185                 return git_config_string(&external_diff_cmd_cfg, var, value);
186         if (!strcmp(var, "diff.wordregex"))
187                 return git_config_string(&diff_word_regex_cfg, var, value);
188
189         if (!strcmp(var, "diff.ignoresubmodules"))
190                 handle_ignore_submodules_arg(&default_diff_options, value);
191
192         if (!strcmp(var, "diff.submodule")) {
193                 if (parse_submodule_params(&default_diff_options, value))
194                         warning(_("Unknown value for 'diff.submodule' config variable: '%s'"),
195                                 value);
196                 return 0;
197         }
198
199         if (git_color_config(var, value, cb) < 0)
200                 return -1;
201
202         return git_diff_basic_config(var, value, cb);
203 }
204
205 int git_diff_basic_config(const char *var, const char *value, void *cb)
206 {
207         if (!strcmp(var, "diff.renamelimit")) {
208                 diff_rename_limit_default = git_config_int(var, value);
209                 return 0;
210         }
211
212         if (userdiff_config(var, value) < 0)
213                 return -1;
214
215         if (!prefixcmp(var, "diff.color.") || !prefixcmp(var, "color.diff.")) {
216                 int slot = parse_diff_color_slot(var, 11);
217                 if (slot < 0)
218                         return 0;
219                 if (!value)
220                         return config_error_nonbool(var);
221                 color_parse(value, var, diff_colors[slot]);
222                 return 0;
223         }
224
225         /* like GNU diff's --suppress-blank-empty option  */
226         if (!strcmp(var, "diff.suppressblankempty") ||
227                         /* for backwards compatibility */
228                         !strcmp(var, "diff.suppress-blank-empty")) {
229                 diff_suppress_blank_empty = git_config_bool(var, value);
230                 return 0;
231         }
232
233         if (!strcmp(var, "diff.dirstat")) {
234                 struct strbuf errmsg = STRBUF_INIT;
235                 default_diff_options.dirstat_permille = diff_dirstat_permille_default;
236                 if (parse_dirstat_params(&default_diff_options, value, &errmsg))
237                         warning(_("Found errors in 'diff.dirstat' config variable:\n%s"),
238                                 errmsg.buf);
239                 strbuf_release(&errmsg);
240                 diff_dirstat_permille_default = default_diff_options.dirstat_permille;
241                 return 0;
242         }
243
244         if (!prefixcmp(var, "submodule."))
245                 return parse_submodule_config_option(var, value);
246
247         return git_default_config(var, value, cb);
248 }
249
250 static char *quote_two(const char *one, const char *two)
251 {
252         int need_one = quote_c_style(one, NULL, NULL, 1);
253         int need_two = quote_c_style(two, NULL, NULL, 1);
254         struct strbuf res = STRBUF_INIT;
255
256         if (need_one + need_two) {
257                 strbuf_addch(&res, '"');
258                 quote_c_style(one, &res, NULL, 1);
259                 quote_c_style(two, &res, NULL, 1);
260                 strbuf_addch(&res, '"');
261         } else {
262                 strbuf_addstr(&res, one);
263                 strbuf_addstr(&res, two);
264         }
265         return strbuf_detach(&res, NULL);
266 }
267
268 static const char *external_diff(void)
269 {
270         static const char *external_diff_cmd = NULL;
271         static int done_preparing = 0;
272
273         if (done_preparing)
274                 return external_diff_cmd;
275         external_diff_cmd = getenv("GIT_EXTERNAL_DIFF");
276         if (!external_diff_cmd)
277                 external_diff_cmd = external_diff_cmd_cfg;
278         done_preparing = 1;
279         return external_diff_cmd;
280 }
281
282 static struct diff_tempfile {
283         const char *name; /* filename external diff should read from */
284         char hex[41];
285         char mode[10];
286         char tmp_path[PATH_MAX];
287 } diff_temp[2];
288
289 typedef unsigned long (*sane_truncate_fn)(char *line, unsigned long len);
290
291 struct emit_callback {
292         int color_diff;
293         unsigned ws_rule;
294         int blank_at_eof_in_preimage;
295         int blank_at_eof_in_postimage;
296         int lno_in_preimage;
297         int lno_in_postimage;
298         sane_truncate_fn truncate;
299         const char **label_path;
300         struct diff_words_data *diff_words;
301         struct diff_options *opt;
302         int *found_changesp;
303         struct strbuf *header;
304 };
305
306 static int count_lines(const char *data, int size)
307 {
308         int count, ch, completely_empty = 1, nl_just_seen = 0;
309         count = 0;
310         while (0 < size--) {
311                 ch = *data++;
312                 if (ch == '\n') {
313                         count++;
314                         nl_just_seen = 1;
315                         completely_empty = 0;
316                 }
317                 else {
318                         nl_just_seen = 0;
319                         completely_empty = 0;
320                 }
321         }
322         if (completely_empty)
323                 return 0;
324         if (!nl_just_seen)
325                 count++; /* no trailing newline */
326         return count;
327 }
328
329 static int fill_mmfile(mmfile_t *mf, struct diff_filespec *one)
330 {
331         if (!DIFF_FILE_VALID(one)) {
332                 mf->ptr = (char *)""; /* does not matter */
333                 mf->size = 0;
334                 return 0;
335         }
336         else if (diff_populate_filespec(one, 0))
337                 return -1;
338
339         mf->ptr = one->data;
340         mf->size = one->size;
341         return 0;
342 }
343
344 /* like fill_mmfile, but only for size, so we can avoid retrieving blob */
345 static unsigned long diff_filespec_size(struct diff_filespec *one)
346 {
347         if (!DIFF_FILE_VALID(one))
348                 return 0;
349         diff_populate_filespec(one, 1);
350         return one->size;
351 }
352
353 static int count_trailing_blank(mmfile_t *mf, unsigned ws_rule)
354 {
355         char *ptr = mf->ptr;
356         long size = mf->size;
357         int cnt = 0;
358
359         if (!size)
360                 return cnt;
361         ptr += size - 1; /* pointing at the very end */
362         if (*ptr != '\n')
363                 ; /* incomplete line */
364         else
365                 ptr--; /* skip the last LF */
366         while (mf->ptr < ptr) {
367                 char *prev_eol;
368                 for (prev_eol = ptr; mf->ptr <= prev_eol; prev_eol--)
369                         if (*prev_eol == '\n')
370                                 break;
371                 if (!ws_blank_line(prev_eol + 1, ptr - prev_eol, ws_rule))
372                         break;
373                 cnt++;
374                 ptr = prev_eol - 1;
375         }
376         return cnt;
377 }
378
379 static void check_blank_at_eof(mmfile_t *mf1, mmfile_t *mf2,
380                                struct emit_callback *ecbdata)
381 {
382         int l1, l2, at;
383         unsigned ws_rule = ecbdata->ws_rule;
384         l1 = count_trailing_blank(mf1, ws_rule);
385         l2 = count_trailing_blank(mf2, ws_rule);
386         if (l2 <= l1) {
387                 ecbdata->blank_at_eof_in_preimage = 0;
388                 ecbdata->blank_at_eof_in_postimage = 0;
389                 return;
390         }
391         at = count_lines(mf1->ptr, mf1->size);
392         ecbdata->blank_at_eof_in_preimage = (at - l1) + 1;
393
394         at = count_lines(mf2->ptr, mf2->size);
395         ecbdata->blank_at_eof_in_postimage = (at - l2) + 1;
396 }
397
398 static void emit_line_0(struct diff_options *o, const char *set, const char *reset,
399                         int first, const char *line, int len)
400 {
401         int has_trailing_newline, has_trailing_carriage_return;
402         int nofirst;
403         FILE *file = o->file;
404
405         if (o->output_prefix) {
406                 struct strbuf *msg = NULL;
407                 msg = o->output_prefix(o, o->output_prefix_data);
408                 assert(msg);
409                 fwrite(msg->buf, msg->len, 1, file);
410         }
411
412         if (len == 0) {
413                 has_trailing_newline = (first == '\n');
414                 has_trailing_carriage_return = (!has_trailing_newline &&
415                                                 (first == '\r'));
416                 nofirst = has_trailing_newline || has_trailing_carriage_return;
417         } else {
418                 has_trailing_newline = (len > 0 && line[len-1] == '\n');
419                 if (has_trailing_newline)
420                         len--;
421                 has_trailing_carriage_return = (len > 0 && line[len-1] == '\r');
422                 if (has_trailing_carriage_return)
423                         len--;
424                 nofirst = 0;
425         }
426
427         if (len || !nofirst) {
428                 fputs(set, file);
429                 if (!nofirst)
430                         fputc(first, file);
431                 fwrite(line, len, 1, file);
432                 fputs(reset, file);
433         }
434         if (has_trailing_carriage_return)
435                 fputc('\r', file);
436         if (has_trailing_newline)
437                 fputc('\n', file);
438 }
439
440 static void emit_line(struct diff_options *o, const char *set, const char *reset,
441                       const char *line, int len)
442 {
443         emit_line_0(o, set, reset, line[0], line+1, len-1);
444 }
445
446 static int new_blank_line_at_eof(struct emit_callback *ecbdata, const char *line, int len)
447 {
448         if (!((ecbdata->ws_rule & WS_BLANK_AT_EOF) &&
449               ecbdata->blank_at_eof_in_preimage &&
450               ecbdata->blank_at_eof_in_postimage &&
451               ecbdata->blank_at_eof_in_preimage <= ecbdata->lno_in_preimage &&
452               ecbdata->blank_at_eof_in_postimage <= ecbdata->lno_in_postimage))
453                 return 0;
454         return ws_blank_line(line, len, ecbdata->ws_rule);
455 }
456
457 static void emit_add_line(const char *reset,
458                           struct emit_callback *ecbdata,
459                           const char *line, int len)
460 {
461         const char *ws = diff_get_color(ecbdata->color_diff, DIFF_WHITESPACE);
462         const char *set = diff_get_color(ecbdata->color_diff, DIFF_FILE_NEW);
463
464         if (!*ws)
465                 emit_line_0(ecbdata->opt, set, reset, '+', line, len);
466         else if (new_blank_line_at_eof(ecbdata, line, len))
467                 /* Blank line at EOF - paint '+' as well */
468                 emit_line_0(ecbdata->opt, ws, reset, '+', line, len);
469         else {
470                 /* Emit just the prefix, then the rest. */
471                 emit_line_0(ecbdata->opt, set, reset, '+', "", 0);
472                 ws_check_emit(line, len, ecbdata->ws_rule,
473                               ecbdata->opt->file, set, reset, ws);
474         }
475 }
476
477 static void emit_hunk_header(struct emit_callback *ecbdata,
478                              const char *line, int len)
479 {
480         const char *plain = diff_get_color(ecbdata->color_diff, DIFF_PLAIN);
481         const char *frag = diff_get_color(ecbdata->color_diff, DIFF_FRAGINFO);
482         const char *func = diff_get_color(ecbdata->color_diff, DIFF_FUNCINFO);
483         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
484         static const char atat[2] = { '@', '@' };
485         const char *cp, *ep;
486         struct strbuf msgbuf = STRBUF_INIT;
487         int org_len = len;
488         int i = 1;
489
490         /*
491          * As a hunk header must begin with "@@ -<old>, +<new> @@",
492          * it always is at least 10 bytes long.
493          */
494         if (len < 10 ||
495             memcmp(line, atat, 2) ||
496             !(ep = memmem(line + 2, len - 2, atat, 2))) {
497                 emit_line(ecbdata->opt, plain, reset, line, len);
498                 return;
499         }
500         ep += 2; /* skip over @@ */
501
502         /* The hunk header in fraginfo color */
503         strbuf_add(&msgbuf, frag, strlen(frag));
504         strbuf_add(&msgbuf, line, ep - line);
505         strbuf_add(&msgbuf, reset, strlen(reset));
506
507         /*
508          * trailing "\r\n"
509          */
510         for ( ; i < 3; i++)
511                 if (line[len - i] == '\r' || line[len - i] == '\n')
512                         len--;
513
514         /* blank before the func header */
515         for (cp = ep; ep - line < len; ep++)
516                 if (*ep != ' ' && *ep != '\t')
517                         break;
518         if (ep != cp) {
519                 strbuf_add(&msgbuf, plain, strlen(plain));
520                 strbuf_add(&msgbuf, cp, ep - cp);
521                 strbuf_add(&msgbuf, reset, strlen(reset));
522         }
523
524         if (ep < line + len) {
525                 strbuf_add(&msgbuf, func, strlen(func));
526                 strbuf_add(&msgbuf, ep, line + len - ep);
527                 strbuf_add(&msgbuf, reset, strlen(reset));
528         }
529
530         strbuf_add(&msgbuf, line + len, org_len - len);
531         emit_line(ecbdata->opt, "", "", msgbuf.buf, msgbuf.len);
532         strbuf_release(&msgbuf);
533 }
534
535 static struct diff_tempfile *claim_diff_tempfile(void) {
536         int i;
537         for (i = 0; i < ARRAY_SIZE(diff_temp); i++)
538                 if (!diff_temp[i].name)
539                         return diff_temp + i;
540         die("BUG: diff is failing to clean up its tempfiles");
541 }
542
543 static int remove_tempfile_installed;
544
545 static void remove_tempfile(void)
546 {
547         int i;
548         for (i = 0; i < ARRAY_SIZE(diff_temp); i++) {
549                 if (diff_temp[i].name == diff_temp[i].tmp_path)
550                         unlink_or_warn(diff_temp[i].name);
551                 diff_temp[i].name = NULL;
552         }
553 }
554
555 static void remove_tempfile_on_signal(int signo)
556 {
557         remove_tempfile();
558         sigchain_pop(signo);
559         raise(signo);
560 }
561
562 static void print_line_count(FILE *file, int count)
563 {
564         switch (count) {
565         case 0:
566                 fprintf(file, "0,0");
567                 break;
568         case 1:
569                 fprintf(file, "1");
570                 break;
571         default:
572                 fprintf(file, "1,%d", count);
573                 break;
574         }
575 }
576
577 static void emit_rewrite_lines(struct emit_callback *ecb,
578                                int prefix, const char *data, int size)
579 {
580         const char *endp = NULL;
581         static const char *nneof = " No newline at end of file\n";
582         const char *old = diff_get_color(ecb->color_diff, DIFF_FILE_OLD);
583         const char *reset = diff_get_color(ecb->color_diff, DIFF_RESET);
584
585         while (0 < size) {
586                 int len;
587
588                 endp = memchr(data, '\n', size);
589                 len = endp ? (endp - data + 1) : size;
590                 if (prefix != '+') {
591                         ecb->lno_in_preimage++;
592                         emit_line_0(ecb->opt, old, reset, '-',
593                                     data, len);
594                 } else {
595                         ecb->lno_in_postimage++;
596                         emit_add_line(reset, ecb, data, len);
597                 }
598                 size -= len;
599                 data += len;
600         }
601         if (!endp) {
602                 const char *plain = diff_get_color(ecb->color_diff,
603                                                    DIFF_PLAIN);
604                 putc('\n', ecb->opt->file);
605                 emit_line_0(ecb->opt, plain, reset, '\\',
606                             nneof, strlen(nneof));
607         }
608 }
609
610 static void emit_rewrite_diff(const char *name_a,
611                               const char *name_b,
612                               struct diff_filespec *one,
613                               struct diff_filespec *two,
614                               struct userdiff_driver *textconv_one,
615                               struct userdiff_driver *textconv_two,
616                               struct diff_options *o)
617 {
618         int lc_a, lc_b;
619         const char *name_a_tab, *name_b_tab;
620         const char *metainfo = diff_get_color(o->use_color, DIFF_METAINFO);
621         const char *fraginfo = diff_get_color(o->use_color, DIFF_FRAGINFO);
622         const char *reset = diff_get_color(o->use_color, DIFF_RESET);
623         static struct strbuf a_name = STRBUF_INIT, b_name = STRBUF_INIT;
624         const char *a_prefix, *b_prefix;
625         char *data_one, *data_two;
626         size_t size_one, size_two;
627         struct emit_callback ecbdata;
628         char *line_prefix = "";
629         struct strbuf *msgbuf;
630
631         if (o && o->output_prefix) {
632                 msgbuf = o->output_prefix(o, o->output_prefix_data);
633                 line_prefix = msgbuf->buf;
634         }
635
636         if (diff_mnemonic_prefix && DIFF_OPT_TST(o, REVERSE_DIFF)) {
637                 a_prefix = o->b_prefix;
638                 b_prefix = o->a_prefix;
639         } else {
640                 a_prefix = o->a_prefix;
641                 b_prefix = o->b_prefix;
642         }
643
644         name_a += (*name_a == '/');
645         name_b += (*name_b == '/');
646         name_a_tab = strchr(name_a, ' ') ? "\t" : "";
647         name_b_tab = strchr(name_b, ' ') ? "\t" : "";
648
649         strbuf_reset(&a_name);
650         strbuf_reset(&b_name);
651         quote_two_c_style(&a_name, a_prefix, name_a, 0);
652         quote_two_c_style(&b_name, b_prefix, name_b, 0);
653
654         size_one = fill_textconv(textconv_one, one, &data_one);
655         size_two = fill_textconv(textconv_two, two, &data_two);
656
657         memset(&ecbdata, 0, sizeof(ecbdata));
658         ecbdata.color_diff = want_color(o->use_color);
659         ecbdata.found_changesp = &o->found_changes;
660         ecbdata.ws_rule = whitespace_rule(name_b ? name_b : name_a);
661         ecbdata.opt = o;
662         if (ecbdata.ws_rule & WS_BLANK_AT_EOF) {
663                 mmfile_t mf1, mf2;
664                 mf1.ptr = (char *)data_one;
665                 mf2.ptr = (char *)data_two;
666                 mf1.size = size_one;
667                 mf2.size = size_two;
668                 check_blank_at_eof(&mf1, &mf2, &ecbdata);
669         }
670         ecbdata.lno_in_preimage = 1;
671         ecbdata.lno_in_postimage = 1;
672
673         lc_a = count_lines(data_one, size_one);
674         lc_b = count_lines(data_two, size_two);
675         fprintf(o->file,
676                 "%s%s--- %s%s%s\n%s%s+++ %s%s%s\n%s%s@@ -",
677                 line_prefix, metainfo, a_name.buf, name_a_tab, reset,
678                 line_prefix, metainfo, b_name.buf, name_b_tab, reset,
679                 line_prefix, fraginfo);
680         if (!o->irreversible_delete)
681                 print_line_count(o->file, lc_a);
682         else
683                 fprintf(o->file, "?,?");
684         fprintf(o->file, " +");
685         print_line_count(o->file, lc_b);
686         fprintf(o->file, " @@%s\n", reset);
687         if (lc_a && !o->irreversible_delete)
688                 emit_rewrite_lines(&ecbdata, '-', data_one, size_one);
689         if (lc_b)
690                 emit_rewrite_lines(&ecbdata, '+', data_two, size_two);
691         if (textconv_one)
692                 free((char *)data_one);
693         if (textconv_two)
694                 free((char *)data_two);
695 }
696
697 struct diff_words_buffer {
698         mmfile_t text;
699         long alloc;
700         struct diff_words_orig {
701                 const char *begin, *end;
702         } *orig;
703         int orig_nr, orig_alloc;
704 };
705
706 static void diff_words_append(char *line, unsigned long len,
707                 struct diff_words_buffer *buffer)
708 {
709         ALLOC_GROW(buffer->text.ptr, buffer->text.size + len, buffer->alloc);
710         line++;
711         len--;
712         memcpy(buffer->text.ptr + buffer->text.size, line, len);
713         buffer->text.size += len;
714         buffer->text.ptr[buffer->text.size] = '\0';
715 }
716
717 struct diff_words_style_elem {
718         const char *prefix;
719         const char *suffix;
720         const char *color; /* NULL; filled in by the setup code if
721                             * color is enabled */
722 };
723
724 struct diff_words_style {
725         enum diff_words_type type;
726         struct diff_words_style_elem new, old, ctx;
727         const char *newline;
728 };
729
730 static struct diff_words_style diff_words_styles[] = {
731         { DIFF_WORDS_PORCELAIN, {"+", "\n"}, {"-", "\n"}, {" ", "\n"}, "~\n" },
732         { DIFF_WORDS_PLAIN, {"{+", "+}"}, {"[-", "-]"}, {"", ""}, "\n" },
733         { DIFF_WORDS_COLOR, {"", ""}, {"", ""}, {"", ""}, "\n" }
734 };
735
736 struct diff_words_data {
737         struct diff_words_buffer minus, plus;
738         const char *current_plus;
739         int last_minus;
740         struct diff_options *opt;
741         regex_t *word_regex;
742         enum diff_words_type type;
743         struct diff_words_style *style;
744 };
745
746 static int fn_out_diff_words_write_helper(FILE *fp,
747                                           struct diff_words_style_elem *st_el,
748                                           const char *newline,
749                                           size_t count, const char *buf,
750                                           const char *line_prefix)
751 {
752         int print = 0;
753
754         while (count) {
755                 char *p = memchr(buf, '\n', count);
756                 if (print)
757                         fputs(line_prefix, fp);
758                 if (p != buf) {
759                         if (st_el->color && fputs(st_el->color, fp) < 0)
760                                 return -1;
761                         if (fputs(st_el->prefix, fp) < 0 ||
762                             fwrite(buf, p ? p - buf : count, 1, fp) != 1 ||
763                             fputs(st_el->suffix, fp) < 0)
764                                 return -1;
765                         if (st_el->color && *st_el->color
766                             && fputs(GIT_COLOR_RESET, fp) < 0)
767                                 return -1;
768                 }
769                 if (!p)
770                         return 0;
771                 if (fputs(newline, fp) < 0)
772                         return -1;
773                 count -= p + 1 - buf;
774                 buf = p + 1;
775                 print = 1;
776         }
777         return 0;
778 }
779
780 /*
781  * '--color-words' algorithm can be described as:
782  *
783  *   1. collect a the minus/plus lines of a diff hunk, divided into
784  *      minus-lines and plus-lines;
785  *
786  *   2. break both minus-lines and plus-lines into words and
787  *      place them into two mmfile_t with one word for each line;
788  *
789  *   3. use xdiff to run diff on the two mmfile_t to get the words level diff;
790  *
791  * And for the common parts of the both file, we output the plus side text.
792  * diff_words->current_plus is used to trace the current position of the plus file
793  * which printed. diff_words->last_minus is used to trace the last minus word
794  * printed.
795  *
796  * For '--graph' to work with '--color-words', we need to output the graph prefix
797  * on each line of color words output. Generally, there are two conditions on
798  * which we should output the prefix.
799  *
800  *   1. diff_words->last_minus == 0 &&
801  *      diff_words->current_plus == diff_words->plus.text.ptr
802  *
803  *      that is: the plus text must start as a new line, and if there is no minus
804  *      word printed, a graph prefix must be printed.
805  *
806  *   2. diff_words->current_plus > diff_words->plus.text.ptr &&
807  *      *(diff_words->current_plus - 1) == '\n'
808  *
809  *      that is: a graph prefix must be printed following a '\n'
810  */
811 static int color_words_output_graph_prefix(struct diff_words_data *diff_words)
812 {
813         if ((diff_words->last_minus == 0 &&
814                 diff_words->current_plus == diff_words->plus.text.ptr) ||
815                 (diff_words->current_plus > diff_words->plus.text.ptr &&
816                 *(diff_words->current_plus - 1) == '\n')) {
817                 return 1;
818         } else {
819                 return 0;
820         }
821 }
822
823 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
824 {
825         struct diff_words_data *diff_words = priv;
826         struct diff_words_style *style = diff_words->style;
827         int minus_first, minus_len, plus_first, plus_len;
828         const char *minus_begin, *minus_end, *plus_begin, *plus_end;
829         struct diff_options *opt = diff_words->opt;
830         struct strbuf *msgbuf;
831         char *line_prefix = "";
832
833         if (line[0] != '@' || parse_hunk_header(line, len,
834                         &minus_first, &minus_len, &plus_first, &plus_len))
835                 return;
836
837         assert(opt);
838         if (opt->output_prefix) {
839                 msgbuf = opt->output_prefix(opt, opt->output_prefix_data);
840                 line_prefix = msgbuf->buf;
841         }
842
843         /* POSIX requires that first be decremented by one if len == 0... */
844         if (minus_len) {
845                 minus_begin = diff_words->minus.orig[minus_first].begin;
846                 minus_end =
847                         diff_words->minus.orig[minus_first + minus_len - 1].end;
848         } else
849                 minus_begin = minus_end =
850                         diff_words->minus.orig[minus_first].end;
851
852         if (plus_len) {
853                 plus_begin = diff_words->plus.orig[plus_first].begin;
854                 plus_end = diff_words->plus.orig[plus_first + plus_len - 1].end;
855         } else
856                 plus_begin = plus_end = diff_words->plus.orig[plus_first].end;
857
858         if (color_words_output_graph_prefix(diff_words)) {
859                 fputs(line_prefix, diff_words->opt->file);
860         }
861         if (diff_words->current_plus != plus_begin) {
862                 fn_out_diff_words_write_helper(diff_words->opt->file,
863                                 &style->ctx, style->newline,
864                                 plus_begin - diff_words->current_plus,
865                                 diff_words->current_plus, line_prefix);
866                 if (*(plus_begin - 1) == '\n')
867                         fputs(line_prefix, diff_words->opt->file);
868         }
869         if (minus_begin != minus_end) {
870                 fn_out_diff_words_write_helper(diff_words->opt->file,
871                                 &style->old, style->newline,
872                                 minus_end - minus_begin, minus_begin,
873                                 line_prefix);
874         }
875         if (plus_begin != plus_end) {
876                 fn_out_diff_words_write_helper(diff_words->opt->file,
877                                 &style->new, style->newline,
878                                 plus_end - plus_begin, plus_begin,
879                                 line_prefix);
880         }
881
882         diff_words->current_plus = plus_end;
883         diff_words->last_minus = minus_first;
884 }
885
886 /* This function starts looking at *begin, and returns 0 iff a word was found. */
887 static int find_word_boundaries(mmfile_t *buffer, regex_t *word_regex,
888                 int *begin, int *end)
889 {
890         if (word_regex && *begin < buffer->size) {
891                 regmatch_t match[1];
892                 if (!regexec(word_regex, buffer->ptr + *begin, 1, match, 0)) {
893                         char *p = memchr(buffer->ptr + *begin + match[0].rm_so,
894                                         '\n', match[0].rm_eo - match[0].rm_so);
895                         *end = p ? p - buffer->ptr : match[0].rm_eo + *begin;
896                         *begin += match[0].rm_so;
897                         return *begin >= *end;
898                 }
899                 return -1;
900         }
901
902         /* find the next word */
903         while (*begin < buffer->size && isspace(buffer->ptr[*begin]))
904                 (*begin)++;
905         if (*begin >= buffer->size)
906                 return -1;
907
908         /* find the end of the word */
909         *end = *begin + 1;
910         while (*end < buffer->size && !isspace(buffer->ptr[*end]))
911                 (*end)++;
912
913         return 0;
914 }
915
916 /*
917  * This function splits the words in buffer->text, stores the list with
918  * newline separator into out, and saves the offsets of the original words
919  * in buffer->orig.
920  */
921 static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out,
922                 regex_t *word_regex)
923 {
924         int i, j;
925         long alloc = 0;
926
927         out->size = 0;
928         out->ptr = NULL;
929
930         /* fake an empty "0th" word */
931         ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
932         buffer->orig[0].begin = buffer->orig[0].end = buffer->text.ptr;
933         buffer->orig_nr = 1;
934
935         for (i = 0; i < buffer->text.size; i++) {
936                 if (find_word_boundaries(&buffer->text, word_regex, &i, &j))
937                         return;
938
939                 /* store original boundaries */
940                 ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
941                                 buffer->orig_alloc);
942                 buffer->orig[buffer->orig_nr].begin = buffer->text.ptr + i;
943                 buffer->orig[buffer->orig_nr].end = buffer->text.ptr + j;
944                 buffer->orig_nr++;
945
946                 /* store one word */
947                 ALLOC_GROW(out->ptr, out->size + j - i + 1, alloc);
948                 memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
949                 out->ptr[out->size + j - i] = '\n';
950                 out->size += j - i + 1;
951
952                 i = j - 1;
953         }
954 }
955
956 /* this executes the word diff on the accumulated buffers */
957 static void diff_words_show(struct diff_words_data *diff_words)
958 {
959         xpparam_t xpp;
960         xdemitconf_t xecfg;
961         mmfile_t minus, plus;
962         struct diff_words_style *style = diff_words->style;
963
964         struct diff_options *opt = diff_words->opt;
965         struct strbuf *msgbuf;
966         char *line_prefix = "";
967
968         assert(opt);
969         if (opt->output_prefix) {
970                 msgbuf = opt->output_prefix(opt, opt->output_prefix_data);
971                 line_prefix = msgbuf->buf;
972         }
973
974         /* special case: only removal */
975         if (!diff_words->plus.text.size) {
976                 fputs(line_prefix, diff_words->opt->file);
977                 fn_out_diff_words_write_helper(diff_words->opt->file,
978                         &style->old, style->newline,
979                         diff_words->minus.text.size,
980                         diff_words->minus.text.ptr, line_prefix);
981                 diff_words->minus.text.size = 0;
982                 return;
983         }
984
985         diff_words->current_plus = diff_words->plus.text.ptr;
986         diff_words->last_minus = 0;
987
988         memset(&xpp, 0, sizeof(xpp));
989         memset(&xecfg, 0, sizeof(xecfg));
990         diff_words_fill(&diff_words->minus, &minus, diff_words->word_regex);
991         diff_words_fill(&diff_words->plus, &plus, diff_words->word_regex);
992         xpp.flags = 0;
993         /* as only the hunk header will be parsed, we need a 0-context */
994         xecfg.ctxlen = 0;
995         xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
996                       &xpp, &xecfg);
997         free(minus.ptr);
998         free(plus.ptr);
999         if (diff_words->current_plus != diff_words->plus.text.ptr +
1000                         diff_words->plus.text.size) {
1001                 if (color_words_output_graph_prefix(diff_words))
1002                         fputs(line_prefix, diff_words->opt->file);
1003                 fn_out_diff_words_write_helper(diff_words->opt->file,
1004                         &style->ctx, style->newline,
1005                         diff_words->plus.text.ptr + diff_words->plus.text.size
1006                         - diff_words->current_plus, diff_words->current_plus,
1007                         line_prefix);
1008         }
1009         diff_words->minus.text.size = diff_words->plus.text.size = 0;
1010 }
1011
1012 /* In "color-words" mode, show word-diff of words accumulated in the buffer */
1013 static void diff_words_flush(struct emit_callback *ecbdata)
1014 {
1015         if (ecbdata->diff_words->minus.text.size ||
1016             ecbdata->diff_words->plus.text.size)
1017                 diff_words_show(ecbdata->diff_words);
1018 }
1019
1020 static void diff_filespec_load_driver(struct diff_filespec *one)
1021 {
1022         /* Use already-loaded driver */
1023         if (one->driver)
1024                 return;
1025
1026         if (S_ISREG(one->mode))
1027                 one->driver = userdiff_find_by_path(one->path);
1028
1029         /* Fallback to default settings */
1030         if (!one->driver)
1031                 one->driver = userdiff_find_by_name("default");
1032 }
1033
1034 static const char *userdiff_word_regex(struct diff_filespec *one)
1035 {
1036         diff_filespec_load_driver(one);
1037         return one->driver->word_regex;
1038 }
1039
1040 static void init_diff_words_data(struct emit_callback *ecbdata,
1041                                  struct diff_options *orig_opts,
1042                                  struct diff_filespec *one,
1043                                  struct diff_filespec *two)
1044 {
1045         int i;
1046         struct diff_options *o = xmalloc(sizeof(struct diff_options));
1047         memcpy(o, orig_opts, sizeof(struct diff_options));
1048
1049         ecbdata->diff_words =
1050                 xcalloc(1, sizeof(struct diff_words_data));
1051         ecbdata->diff_words->type = o->word_diff;
1052         ecbdata->diff_words->opt = o;
1053         if (!o->word_regex)
1054                 o->word_regex = userdiff_word_regex(one);
1055         if (!o->word_regex)
1056                 o->word_regex = userdiff_word_regex(two);
1057         if (!o->word_regex)
1058                 o->word_regex = diff_word_regex_cfg;
1059         if (o->word_regex) {
1060                 ecbdata->diff_words->word_regex = (regex_t *)
1061                         xmalloc(sizeof(regex_t));
1062                 if (regcomp(ecbdata->diff_words->word_regex,
1063                             o->word_regex,
1064                             REG_EXTENDED | REG_NEWLINE))
1065                         die ("Invalid regular expression: %s",
1066                              o->word_regex);
1067         }
1068         for (i = 0; i < ARRAY_SIZE(diff_words_styles); i++) {
1069                 if (o->word_diff == diff_words_styles[i].type) {
1070                         ecbdata->diff_words->style =
1071                                 &diff_words_styles[i];
1072                         break;
1073                 }
1074         }
1075         if (want_color(o->use_color)) {
1076                 struct diff_words_style *st = ecbdata->diff_words->style;
1077                 st->old.color = diff_get_color_opt(o, DIFF_FILE_OLD);
1078                 st->new.color = diff_get_color_opt(o, DIFF_FILE_NEW);
1079                 st->ctx.color = diff_get_color_opt(o, DIFF_PLAIN);
1080         }
1081 }
1082
1083 static void free_diff_words_data(struct emit_callback *ecbdata)
1084 {
1085         if (ecbdata->diff_words) {
1086                 diff_words_flush(ecbdata);
1087                 free (ecbdata->diff_words->opt);
1088                 free (ecbdata->diff_words->minus.text.ptr);
1089                 free (ecbdata->diff_words->minus.orig);
1090                 free (ecbdata->diff_words->plus.text.ptr);
1091                 free (ecbdata->diff_words->plus.orig);
1092                 if (ecbdata->diff_words->word_regex) {
1093                         regfree(ecbdata->diff_words->word_regex);
1094                         free(ecbdata->diff_words->word_regex);
1095                 }
1096                 free(ecbdata->diff_words);
1097                 ecbdata->diff_words = NULL;
1098         }
1099 }
1100
1101 const char *diff_get_color(int diff_use_color, enum color_diff ix)
1102 {
1103         if (want_color(diff_use_color))
1104                 return diff_colors[ix];
1105         return "";
1106 }
1107
1108 static unsigned long sane_truncate_line(struct emit_callback *ecb, char *line, unsigned long len)
1109 {
1110         const char *cp;
1111         unsigned long allot;
1112         size_t l = len;
1113
1114         if (ecb->truncate)
1115                 return ecb->truncate(line, len);
1116         cp = line;
1117         allot = l;
1118         while (0 < l) {
1119                 (void) utf8_width(&cp, &l);
1120                 if (!cp)
1121                         break; /* truncated in the middle? */
1122         }
1123         return allot - l;
1124 }
1125
1126 static void find_lno(const char *line, struct emit_callback *ecbdata)
1127 {
1128         const char *p;
1129         ecbdata->lno_in_preimage = 0;
1130         ecbdata->lno_in_postimage = 0;
1131         p = strchr(line, '-');
1132         if (!p)
1133                 return; /* cannot happen */
1134         ecbdata->lno_in_preimage = strtol(p + 1, NULL, 10);
1135         p = strchr(p, '+');
1136         if (!p)
1137                 return; /* cannot happen */
1138         ecbdata->lno_in_postimage = strtol(p + 1, NULL, 10);
1139 }
1140
1141 static void fn_out_consume(void *priv, char *line, unsigned long len)
1142 {
1143         struct emit_callback *ecbdata = priv;
1144         const char *meta = diff_get_color(ecbdata->color_diff, DIFF_METAINFO);
1145         const char *plain = diff_get_color(ecbdata->color_diff, DIFF_PLAIN);
1146         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
1147         struct diff_options *o = ecbdata->opt;
1148         char *line_prefix = "";
1149         struct strbuf *msgbuf;
1150
1151         if (o && o->output_prefix) {
1152                 msgbuf = o->output_prefix(o, o->output_prefix_data);
1153                 line_prefix = msgbuf->buf;
1154         }
1155
1156         if (ecbdata->header) {
1157                 fprintf(ecbdata->opt->file, "%s", ecbdata->header->buf);
1158                 strbuf_reset(ecbdata->header);
1159                 ecbdata->header = NULL;
1160         }
1161         *(ecbdata->found_changesp) = 1;
1162
1163         if (ecbdata->label_path[0]) {
1164                 const char *name_a_tab, *name_b_tab;
1165
1166                 name_a_tab = strchr(ecbdata->label_path[0], ' ') ? "\t" : "";
1167                 name_b_tab = strchr(ecbdata->label_path[1], ' ') ? "\t" : "";
1168
1169                 fprintf(ecbdata->opt->file, "%s%s--- %s%s%s\n",
1170                         line_prefix, meta, ecbdata->label_path[0], reset, name_a_tab);
1171                 fprintf(ecbdata->opt->file, "%s%s+++ %s%s%s\n",
1172                         line_prefix, meta, ecbdata->label_path[1], reset, name_b_tab);
1173                 ecbdata->label_path[0] = ecbdata->label_path[1] = NULL;
1174         }
1175
1176         if (diff_suppress_blank_empty
1177             && len == 2 && line[0] == ' ' && line[1] == '\n') {
1178                 line[0] = '\n';
1179                 len = 1;
1180         }
1181
1182         if (line[0] == '@') {
1183                 if (ecbdata->diff_words)
1184                         diff_words_flush(ecbdata);
1185                 len = sane_truncate_line(ecbdata, line, len);
1186                 find_lno(line, ecbdata);
1187                 emit_hunk_header(ecbdata, line, len);
1188                 if (line[len-1] != '\n')
1189                         putc('\n', ecbdata->opt->file);
1190                 return;
1191         }
1192
1193         if (len < 1) {
1194                 emit_line(ecbdata->opt, reset, reset, line, len);
1195                 if (ecbdata->diff_words
1196                     && ecbdata->diff_words->type == DIFF_WORDS_PORCELAIN)
1197                         fputs("~\n", ecbdata->opt->file);
1198                 return;
1199         }
1200
1201         if (ecbdata->diff_words) {
1202                 if (line[0] == '-') {
1203                         diff_words_append(line, len,
1204                                           &ecbdata->diff_words->minus);
1205                         return;
1206                 } else if (line[0] == '+') {
1207                         diff_words_append(line, len,
1208                                           &ecbdata->diff_words->plus);
1209                         return;
1210                 } else if (!prefixcmp(line, "\\ ")) {
1211                         /*
1212                          * Eat the "no newline at eof" marker as if we
1213                          * saw a "+" or "-" line with nothing on it,
1214                          * and return without diff_words_flush() to
1215                          * defer processing. If this is the end of
1216                          * preimage, more "+" lines may come after it.
1217                          */
1218                         return;
1219                 }
1220                 diff_words_flush(ecbdata);
1221                 if (ecbdata->diff_words->type == DIFF_WORDS_PORCELAIN) {
1222                         emit_line(ecbdata->opt, plain, reset, line, len);
1223                         fputs("~\n", ecbdata->opt->file);
1224                 } else {
1225                         /*
1226                          * Skip the prefix character, if any.  With
1227                          * diff_suppress_blank_empty, there may be
1228                          * none.
1229                          */
1230                         if (line[0] != '\n') {
1231                               line++;
1232                               len--;
1233                         }
1234                         emit_line(ecbdata->opt, plain, reset, line, len);
1235                 }
1236                 return;
1237         }
1238
1239         if (line[0] != '+') {
1240                 const char *color =
1241                         diff_get_color(ecbdata->color_diff,
1242                                        line[0] == '-' ? DIFF_FILE_OLD : DIFF_PLAIN);
1243                 ecbdata->lno_in_preimage++;
1244                 if (line[0] == ' ')
1245                         ecbdata->lno_in_postimage++;
1246                 emit_line(ecbdata->opt, color, reset, line, len);
1247         } else {
1248                 ecbdata->lno_in_postimage++;
1249                 emit_add_line(reset, ecbdata, line + 1, len - 1);
1250         }
1251 }
1252
1253 static char *pprint_rename(const char *a, const char *b)
1254 {
1255         const char *old = a;
1256         const char *new = b;
1257         struct strbuf name = STRBUF_INIT;
1258         int pfx_length, sfx_length;
1259         int len_a = strlen(a);
1260         int len_b = strlen(b);
1261         int a_midlen, b_midlen;
1262         int qlen_a = quote_c_style(a, NULL, NULL, 0);
1263         int qlen_b = quote_c_style(b, NULL, NULL, 0);
1264
1265         if (qlen_a || qlen_b) {
1266                 quote_c_style(a, &name, NULL, 0);
1267                 strbuf_addstr(&name, " => ");
1268                 quote_c_style(b, &name, NULL, 0);
1269                 return strbuf_detach(&name, NULL);
1270         }
1271
1272         /* Find common prefix */
1273         pfx_length = 0;
1274         while (*old && *new && *old == *new) {
1275                 if (*old == '/')
1276                         pfx_length = old - a + 1;
1277                 old++;
1278                 new++;
1279         }
1280
1281         /* Find common suffix */
1282         old = a + len_a;
1283         new = b + len_b;
1284         sfx_length = 0;
1285         while (a <= old && b <= new && *old == *new) {
1286                 if (*old == '/')
1287                         sfx_length = len_a - (old - a);
1288                 old--;
1289                 new--;
1290         }
1291
1292         /*
1293          * pfx{mid-a => mid-b}sfx
1294          * {pfx-a => pfx-b}sfx
1295          * pfx{sfx-a => sfx-b}
1296          * name-a => name-b
1297          */
1298         a_midlen = len_a - pfx_length - sfx_length;
1299         b_midlen = len_b - pfx_length - sfx_length;
1300         if (a_midlen < 0)
1301                 a_midlen = 0;
1302         if (b_midlen < 0)
1303                 b_midlen = 0;
1304
1305         strbuf_grow(&name, pfx_length + a_midlen + b_midlen + sfx_length + 7);
1306         if (pfx_length + sfx_length) {
1307                 strbuf_add(&name, a, pfx_length);
1308                 strbuf_addch(&name, '{');
1309         }
1310         strbuf_add(&name, a + pfx_length, a_midlen);
1311         strbuf_addstr(&name, " => ");
1312         strbuf_add(&name, b + pfx_length, b_midlen);
1313         if (pfx_length + sfx_length) {
1314                 strbuf_addch(&name, '}');
1315                 strbuf_add(&name, a + len_a - sfx_length, sfx_length);
1316         }
1317         return strbuf_detach(&name, NULL);
1318 }
1319
1320 struct diffstat_t {
1321         int nr;
1322         int alloc;
1323         struct diffstat_file {
1324                 char *from_name;
1325                 char *name;
1326                 char *print_name;
1327                 unsigned is_unmerged:1;
1328                 unsigned is_binary:1;
1329                 unsigned is_renamed:1;
1330                 unsigned is_interesting:1;
1331                 uintmax_t added, deleted;
1332         } **files;
1333 };
1334
1335 static struct diffstat_file *diffstat_add(struct diffstat_t *diffstat,
1336                                           const char *name_a,
1337                                           const char *name_b)
1338 {
1339         struct diffstat_file *x;
1340         x = xcalloc(sizeof (*x), 1);
1341         if (diffstat->nr == diffstat->alloc) {
1342                 diffstat->alloc = alloc_nr(diffstat->alloc);
1343                 diffstat->files = xrealloc(diffstat->files,
1344                                 diffstat->alloc * sizeof(x));
1345         }
1346         diffstat->files[diffstat->nr++] = x;
1347         if (name_b) {
1348                 x->from_name = xstrdup(name_a);
1349                 x->name = xstrdup(name_b);
1350                 x->is_renamed = 1;
1351         }
1352         else {
1353                 x->from_name = NULL;
1354                 x->name = xstrdup(name_a);
1355         }
1356         return x;
1357 }
1358
1359 static void diffstat_consume(void *priv, char *line, unsigned long len)
1360 {
1361         struct diffstat_t *diffstat = priv;
1362         struct diffstat_file *x = diffstat->files[diffstat->nr - 1];
1363
1364         if (line[0] == '+')
1365                 x->added++;
1366         else if (line[0] == '-')
1367                 x->deleted++;
1368 }
1369
1370 const char mime_boundary_leader[] = "------------";
1371
1372 static int scale_linear(int it, int width, int max_change)
1373 {
1374         if (!it)
1375                 return 0;
1376         /*
1377          * make sure that at least one '-' or '+' is printed if
1378          * there is any change to this path. The easiest way is to
1379          * scale linearly as if the alloted width is one column shorter
1380          * than it is, and then add 1 to the result.
1381          */
1382         return 1 + (it * (width - 1) / max_change);
1383 }
1384
1385 static void show_name(FILE *file,
1386                       const char *prefix, const char *name, int len)
1387 {
1388         fprintf(file, " %s%-*s |", prefix, len, name);
1389 }
1390
1391 static void show_graph(FILE *file, char ch, int cnt, const char *set, const char *reset)
1392 {
1393         if (cnt <= 0)
1394                 return;
1395         fprintf(file, "%s", set);
1396         while (cnt--)
1397                 putc(ch, file);
1398         fprintf(file, "%s", reset);
1399 }
1400
1401 static void fill_print_name(struct diffstat_file *file)
1402 {
1403         char *pname;
1404
1405         if (file->print_name)
1406                 return;
1407
1408         if (!file->is_renamed) {
1409                 struct strbuf buf = STRBUF_INIT;
1410                 if (quote_c_style(file->name, &buf, NULL, 0)) {
1411                         pname = strbuf_detach(&buf, NULL);
1412                 } else {
1413                         pname = file->name;
1414                         strbuf_release(&buf);
1415                 }
1416         } else {
1417                 pname = pprint_rename(file->from_name, file->name);
1418         }
1419         file->print_name = pname;
1420 }
1421
1422 int print_stat_summary(FILE *fp, int files, int insertions, int deletions)
1423 {
1424         struct strbuf sb = STRBUF_INIT;
1425         int ret;
1426
1427         if (!files) {
1428                 assert(insertions == 0 && deletions == 0);
1429                 return fprintf(fp, "%s\n", " 0 files changed");
1430         }
1431
1432         strbuf_addf(&sb,
1433                     (files == 1) ? " %d file changed" : " %d files changed",
1434                     files);
1435
1436         /*
1437          * For binary diff, the caller may want to print "x files
1438          * changed" with insertions == 0 && deletions == 0.
1439          *
1440          * Not omitting "0 insertions(+), 0 deletions(-)" in this case
1441          * is probably less confusing (i.e skip over "2 files changed
1442          * but nothing about added/removed lines? Is this a bug in Git?").
1443          */
1444         if (insertions || deletions == 0) {
1445                 /*
1446                  * TRANSLATORS: "+" in (+) is a line addition marker;
1447                  * do not translate it.
1448                  */
1449                 strbuf_addf(&sb,
1450                             (insertions == 1) ? ", %d insertion(+)" : ", %d insertions(+)",
1451                             insertions);
1452         }
1453
1454         if (deletions || insertions == 0) {
1455                 /*
1456                  * TRANSLATORS: "-" in (-) is a line removal marker;
1457                  * do not translate it.
1458                  */
1459                 strbuf_addf(&sb,
1460                             (deletions == 1) ? ", %d deletion(-)" : ", %d deletions(-)",
1461                             deletions);
1462         }
1463         strbuf_addch(&sb, '\n');
1464         ret = fputs(sb.buf, fp);
1465         strbuf_release(&sb);
1466         return ret;
1467 }
1468
1469 static void show_stats(struct diffstat_t *data, struct diff_options *options)
1470 {
1471         int i, len, add, del, adds = 0, dels = 0;
1472         uintmax_t max_change = 0, max_len = 0;
1473         int total_files = data->nr, count;
1474         int width, name_width, graph_width, number_width = 0, bin_width = 0;
1475         const char *reset, *add_c, *del_c;
1476         const char *line_prefix = "";
1477         int extra_shown = 0;
1478         struct strbuf *msg = NULL;
1479
1480         if (data->nr == 0)
1481                 return;
1482
1483         if (options->output_prefix) {
1484                 msg = options->output_prefix(options, options->output_prefix_data);
1485                 line_prefix = msg->buf;
1486         }
1487
1488         count = options->stat_count ? options->stat_count : data->nr;
1489
1490         reset = diff_get_color_opt(options, DIFF_RESET);
1491         add_c = diff_get_color_opt(options, DIFF_FILE_NEW);
1492         del_c = diff_get_color_opt(options, DIFF_FILE_OLD);
1493
1494         /*
1495          * Find the longest filename and max number of changes
1496          */
1497         for (i = 0; (i < count) && (i < data->nr); i++) {
1498                 struct diffstat_file *file = data->files[i];
1499                 uintmax_t change = file->added + file->deleted;
1500
1501                 if (!file->is_interesting && (change == 0)) {
1502                         count++; /* not shown == room for one more */
1503                         continue;
1504                 }
1505                 fill_print_name(file);
1506                 len = strlen(file->print_name);
1507                 if (max_len < len)
1508                         max_len = len;
1509
1510                 if (file->is_unmerged) {
1511                         /* "Unmerged" is 8 characters */
1512                         bin_width = bin_width < 8 ? 8 : bin_width;
1513                         continue;
1514                 }
1515                 if (file->is_binary) {
1516                         /* "Bin XXX -> YYY bytes" */
1517                         int w = 14 + decimal_width(file->added)
1518                                 + decimal_width(file->deleted);
1519                         bin_width = bin_width < w ? w : bin_width;
1520                         /* Display change counts aligned with "Bin" */
1521                         number_width = 3;
1522                         continue;
1523                 }
1524
1525                 if (max_change < change)
1526                         max_change = change;
1527         }
1528         count = i; /* where we can stop scanning in data->files[] */
1529
1530         /*
1531          * We have width = stat_width or term_columns() columns total.
1532          * We want a maximum of min(max_len, stat_name_width) for the name part.
1533          * We want a maximum of min(max_change, stat_graph_width) for the +- part.
1534          * We also need 1 for " " and 4 + decimal_width(max_change)
1535          * for " | NNNN " and one the empty column at the end, altogether
1536          * 6 + decimal_width(max_change).
1537          *
1538          * If there's not enough space, we will use the smaller of
1539          * stat_name_width (if set) and 5/8*width for the filename,
1540          * and the rest for constant elements + graph part, but no more
1541          * than stat_graph_width for the graph part.
1542          * (5/8 gives 50 for filename and 30 for the constant parts + graph
1543          * for the standard terminal size).
1544          *
1545          * In other words: stat_width limits the maximum width, and
1546          * stat_name_width fixes the maximum width of the filename,
1547          * and is also used to divide available columns if there
1548          * aren't enough.
1549          *
1550          * Binary files are displayed with "Bin XXX -> YYY bytes"
1551          * instead of the change count and graph. This part is treated
1552          * similarly to the graph part, except that it is not
1553          * "scaled". If total width is too small to accomodate the
1554          * guaranteed minimum width of the filename part and the
1555          * separators and this message, this message will "overflow"
1556          * making the line longer than the maximum width.
1557          */
1558
1559         if (options->stat_width == -1)
1560                 width = term_columns() - options->output_prefix_length;
1561         else
1562                 width = options->stat_width ? options->stat_width : 80;
1563         number_width = decimal_width(max_change) > number_width ?
1564                 decimal_width(max_change) : number_width;
1565
1566         if (options->stat_graph_width == -1)
1567                 options->stat_graph_width = diff_stat_graph_width;
1568
1569         /*
1570          * Guarantee 3/8*16==6 for the graph part
1571          * and 5/8*16==10 for the filename part
1572          */
1573         if (width < 16 + 6 + number_width)
1574                 width = 16 + 6 + number_width;
1575
1576         /*
1577          * First assign sizes that are wanted, ignoring available width.
1578          * strlen("Bin XXX -> YYY bytes") == bin_width, and the part
1579          * starting from "XXX" should fit in graph_width.
1580          */
1581         graph_width = max_change + 4 > bin_width ? max_change : bin_width - 4;
1582         if (options->stat_graph_width &&
1583             options->stat_graph_width < graph_width)
1584                 graph_width = options->stat_graph_width;
1585
1586         name_width = (options->stat_name_width > 0 &&
1587                       options->stat_name_width < max_len) ?
1588                 options->stat_name_width : max_len;
1589
1590         /*
1591          * Adjust adjustable widths not to exceed maximum width
1592          */
1593         if (name_width + number_width + 6 + graph_width > width) {
1594                 if (graph_width > width * 3/8 - number_width - 6) {
1595                         graph_width = width * 3/8 - number_width - 6;
1596                         if (graph_width < 6)
1597                                 graph_width = 6;
1598                 }
1599
1600                 if (options->stat_graph_width &&
1601                     graph_width > options->stat_graph_width)
1602                         graph_width = options->stat_graph_width;
1603                 if (name_width > width - number_width - 6 - graph_width)
1604                         name_width = width - number_width - 6 - graph_width;
1605                 else
1606                         graph_width = width - number_width - 6 - name_width;
1607         }
1608
1609         /*
1610          * From here name_width is the width of the name area,
1611          * and graph_width is the width of the graph area.
1612          * max_change is used to scale graph properly.
1613          */
1614         for (i = 0; i < count; i++) {
1615                 const char *prefix = "";
1616                 struct diffstat_file *file = data->files[i];
1617                 char *name = file->print_name;
1618                 uintmax_t added = file->added;
1619                 uintmax_t deleted = file->deleted;
1620                 int name_len;
1621
1622                 if (!file->is_interesting && (added + deleted == 0))
1623                         continue;
1624
1625                 /*
1626                  * "scale" the filename
1627                  */
1628                 len = name_width;
1629                 name_len = strlen(name);
1630                 if (name_width < name_len) {
1631                         char *slash;
1632                         prefix = "...";
1633                         len -= 3;
1634                         name += name_len - len;
1635                         slash = strchr(name, '/');
1636                         if (slash)
1637                                 name = slash;
1638                 }
1639
1640                 if (file->is_binary) {
1641                         fprintf(options->file, "%s", line_prefix);
1642                         show_name(options->file, prefix, name, len);
1643                         fprintf(options->file, " %*s", number_width, "Bin");
1644                         if (!added && !deleted) {
1645                                 putc('\n', options->file);
1646                                 continue;
1647                         }
1648                         fprintf(options->file, " %s%"PRIuMAX"%s",
1649                                 del_c, deleted, reset);
1650                         fprintf(options->file, " -> ");
1651                         fprintf(options->file, "%s%"PRIuMAX"%s",
1652                                 add_c, added, reset);
1653                         fprintf(options->file, " bytes");
1654                         fprintf(options->file, "\n");
1655                         continue;
1656                 }
1657                 else if (file->is_unmerged) {
1658                         fprintf(options->file, "%s", line_prefix);
1659                         show_name(options->file, prefix, name, len);
1660                         fprintf(options->file, " Unmerged\n");
1661                         continue;
1662                 }
1663
1664                 /*
1665                  * scale the add/delete
1666                  */
1667                 add = added;
1668                 del = deleted;
1669
1670                 if (graph_width <= max_change) {
1671                         int total = add + del;
1672
1673                         total = scale_linear(add + del, graph_width, max_change);
1674                         if (total < 2 && add && del)
1675                                 /* width >= 2 due to the sanity check */
1676                                 total = 2;
1677                         if (add < del) {
1678                                 add = scale_linear(add, graph_width, max_change);
1679                                 del = total - add;
1680                         } else {
1681                                 del = scale_linear(del, graph_width, max_change);
1682                                 add = total - del;
1683                         }
1684                 }
1685                 fprintf(options->file, "%s", line_prefix);
1686                 show_name(options->file, prefix, name, len);
1687                 fprintf(options->file, " %*"PRIuMAX"%s",
1688                         number_width, added + deleted,
1689                         added + deleted ? " " : "");
1690                 show_graph(options->file, '+', add, add_c, reset);
1691                 show_graph(options->file, '-', del, del_c, reset);
1692                 fprintf(options->file, "\n");
1693         }
1694
1695         for (i = 0; i < data->nr; i++) {
1696                 struct diffstat_file *file = data->files[i];
1697                 uintmax_t added = file->added;
1698                 uintmax_t deleted = file->deleted;
1699
1700                 if (file->is_unmerged ||
1701                     (!file->is_interesting && (added + deleted == 0))) {
1702                         total_files--;
1703                         continue;
1704                 }
1705
1706                 if (!file->is_binary) {
1707                         adds += added;
1708                         dels += deleted;
1709                 }
1710                 if (i < count)
1711                         continue;
1712                 if (!extra_shown)
1713                         fprintf(options->file, "%s ...\n", line_prefix);
1714                 extra_shown = 1;
1715         }
1716         fprintf(options->file, "%s", line_prefix);
1717         print_stat_summary(options->file, total_files, adds, dels);
1718 }
1719
1720 static void show_shortstats(struct diffstat_t *data, struct diff_options *options)
1721 {
1722         int i, adds = 0, dels = 0, total_files = data->nr;
1723
1724         if (data->nr == 0)
1725                 return;
1726
1727         for (i = 0; i < data->nr; i++) {
1728                 int added = data->files[i]->added;
1729                 int deleted= data->files[i]->deleted;
1730
1731                 if (data->files[i]->is_unmerged ||
1732                     (!data->files[i]->is_interesting && (added + deleted == 0))) {
1733                         total_files--;
1734                 } else if (!data->files[i]->is_binary) { /* don't count bytes */
1735                         adds += added;
1736                         dels += deleted;
1737                 }
1738         }
1739         if (options->output_prefix) {
1740                 struct strbuf *msg = NULL;
1741                 msg = options->output_prefix(options,
1742                                 options->output_prefix_data);
1743                 fprintf(options->file, "%s", msg->buf);
1744         }
1745         print_stat_summary(options->file, total_files, adds, dels);
1746 }
1747
1748 static void show_numstat(struct diffstat_t *data, struct diff_options *options)
1749 {
1750         int i;
1751
1752         if (data->nr == 0)
1753                 return;
1754
1755         for (i = 0; i < data->nr; i++) {
1756                 struct diffstat_file *file = data->files[i];
1757
1758                 if (options->output_prefix) {
1759                         struct strbuf *msg = NULL;
1760                         msg = options->output_prefix(options,
1761                                         options->output_prefix_data);
1762                         fprintf(options->file, "%s", msg->buf);
1763                 }
1764
1765                 if (file->is_binary)
1766                         fprintf(options->file, "-\t-\t");
1767                 else
1768                         fprintf(options->file,
1769                                 "%"PRIuMAX"\t%"PRIuMAX"\t",
1770                                 file->added, file->deleted);
1771                 if (options->line_termination) {
1772                         fill_print_name(file);
1773                         if (!file->is_renamed)
1774                                 write_name_quoted(file->name, options->file,
1775                                                   options->line_termination);
1776                         else {
1777                                 fputs(file->print_name, options->file);
1778                                 putc(options->line_termination, options->file);
1779                         }
1780                 } else {
1781                         if (file->is_renamed) {
1782                                 putc('\0', options->file);
1783                                 write_name_quoted(file->from_name, options->file, '\0');
1784                         }
1785                         write_name_quoted(file->name, options->file, '\0');
1786                 }
1787         }
1788 }
1789
1790 struct dirstat_file {
1791         const char *name;
1792         unsigned long changed;
1793 };
1794
1795 struct dirstat_dir {
1796         struct dirstat_file *files;
1797         int alloc, nr, permille, cumulative;
1798 };
1799
1800 static long gather_dirstat(struct diff_options *opt, struct dirstat_dir *dir,
1801                 unsigned long changed, const char *base, int baselen)
1802 {
1803         unsigned long this_dir = 0;
1804         unsigned int sources = 0;
1805         const char *line_prefix = "";
1806         struct strbuf *msg = NULL;
1807
1808         if (opt->output_prefix) {
1809                 msg = opt->output_prefix(opt, opt->output_prefix_data);
1810                 line_prefix = msg->buf;
1811         }
1812
1813         while (dir->nr) {
1814                 struct dirstat_file *f = dir->files;
1815                 int namelen = strlen(f->name);
1816                 unsigned long this;
1817                 char *slash;
1818
1819                 if (namelen < baselen)
1820                         break;
1821                 if (memcmp(f->name, base, baselen))
1822                         break;
1823                 slash = strchr(f->name + baselen, '/');
1824                 if (slash) {
1825                         int newbaselen = slash + 1 - f->name;
1826                         this = gather_dirstat(opt, dir, changed, f->name, newbaselen);
1827                         sources++;
1828                 } else {
1829                         this = f->changed;
1830                         dir->files++;
1831                         dir->nr--;
1832                         sources += 2;
1833                 }
1834                 this_dir += this;
1835         }
1836
1837         /*
1838          * We don't report dirstat's for
1839          *  - the top level
1840          *  - or cases where everything came from a single directory
1841          *    under this directory (sources == 1).
1842          */
1843         if (baselen && sources != 1) {
1844                 if (this_dir) {
1845                         int permille = this_dir * 1000 / changed;
1846                         if (permille >= dir->permille) {
1847                                 fprintf(opt->file, "%s%4d.%01d%% %.*s\n", line_prefix,
1848                                         permille / 10, permille % 10, baselen, base);
1849                                 if (!dir->cumulative)
1850                                         return 0;
1851                         }
1852                 }
1853         }
1854         return this_dir;
1855 }
1856
1857 static int dirstat_compare(const void *_a, const void *_b)
1858 {
1859         const struct dirstat_file *a = _a;
1860         const struct dirstat_file *b = _b;
1861         return strcmp(a->name, b->name);
1862 }
1863
1864 static void show_dirstat(struct diff_options *options)
1865 {
1866         int i;
1867         unsigned long changed;
1868         struct dirstat_dir dir;
1869         struct diff_queue_struct *q = &diff_queued_diff;
1870
1871         dir.files = NULL;
1872         dir.alloc = 0;
1873         dir.nr = 0;
1874         dir.permille = options->dirstat_permille;
1875         dir.cumulative = DIFF_OPT_TST(options, DIRSTAT_CUMULATIVE);
1876
1877         changed = 0;
1878         for (i = 0; i < q->nr; i++) {
1879                 struct diff_filepair *p = q->queue[i];
1880                 const char *name;
1881                 unsigned long copied, added, damage;
1882                 int content_changed;
1883
1884                 name = p->two->path ? p->two->path : p->one->path;
1885
1886                 if (p->one->sha1_valid && p->two->sha1_valid)
1887                         content_changed = hashcmp(p->one->sha1, p->two->sha1);
1888                 else
1889                         content_changed = 1;
1890
1891                 if (!content_changed) {
1892                         /*
1893                          * The SHA1 has not changed, so pre-/post-content is
1894                          * identical. We can therefore skip looking at the
1895                          * file contents altogether.
1896                          */
1897                         damage = 0;
1898                         goto found_damage;
1899                 }
1900
1901                 if (DIFF_OPT_TST(options, DIRSTAT_BY_FILE)) {
1902                         /*
1903                          * In --dirstat-by-file mode, we don't really need to
1904                          * look at the actual file contents at all.
1905                          * The fact that the SHA1 changed is enough for us to
1906                          * add this file to the list of results
1907                          * (with each file contributing equal damage).
1908                          */
1909                         damage = 1;
1910                         goto found_damage;
1911                 }
1912
1913                 if (DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
1914                         diff_populate_filespec(p->one, 0);
1915                         diff_populate_filespec(p->two, 0);
1916                         diffcore_count_changes(p->one, p->two, NULL, NULL, 0,
1917                                                &copied, &added);
1918                         diff_free_filespec_data(p->one);
1919                         diff_free_filespec_data(p->two);
1920                 } else if (DIFF_FILE_VALID(p->one)) {
1921                         diff_populate_filespec(p->one, 1);
1922                         copied = added = 0;
1923                         diff_free_filespec_data(p->one);
1924                 } else if (DIFF_FILE_VALID(p->two)) {
1925                         diff_populate_filespec(p->two, 1);
1926                         copied = 0;
1927                         added = p->two->size;
1928                         diff_free_filespec_data(p->two);
1929                 } else
1930                         continue;
1931
1932                 /*
1933                  * Original minus copied is the removed material,
1934                  * added is the new material.  They are both damages
1935                  * made to the preimage.
1936                  * If the resulting damage is zero, we know that
1937                  * diffcore_count_changes() considers the two entries to
1938                  * be identical, but since content_changed is true, we
1939                  * know that there must have been _some_ kind of change,
1940                  * so we force all entries to have damage > 0.
1941                  */
1942                 damage = (p->one->size - copied) + added;
1943                 if (!damage)
1944                         damage = 1;
1945
1946 found_damage:
1947                 ALLOC_GROW(dir.files, dir.nr + 1, dir.alloc);
1948                 dir.files[dir.nr].name = name;
1949                 dir.files[dir.nr].changed = damage;
1950                 changed += damage;
1951                 dir.nr++;
1952         }
1953
1954         /* This can happen even with many files, if everything was renames */
1955         if (!changed)
1956                 return;
1957
1958         /* Show all directories with more than x% of the changes */
1959         qsort(dir.files, dir.nr, sizeof(dir.files[0]), dirstat_compare);
1960         gather_dirstat(options, &dir, changed, "", 0);
1961 }
1962
1963 static void show_dirstat_by_line(struct diffstat_t *data, struct diff_options *options)
1964 {
1965         int i;
1966         unsigned long changed;
1967         struct dirstat_dir dir;
1968
1969         if (data->nr == 0)
1970                 return;
1971
1972         dir.files = NULL;
1973         dir.alloc = 0;
1974         dir.nr = 0;
1975         dir.permille = options->dirstat_permille;
1976         dir.cumulative = DIFF_OPT_TST(options, DIRSTAT_CUMULATIVE);
1977
1978         changed = 0;
1979         for (i = 0; i < data->nr; i++) {
1980                 struct diffstat_file *file = data->files[i];
1981                 unsigned long damage = file->added + file->deleted;
1982                 if (file->is_binary)
1983                         /*
1984                          * binary files counts bytes, not lines. Must find some
1985                          * way to normalize binary bytes vs. textual lines.
1986                          * The following heuristic assumes that there are 64
1987                          * bytes per "line".
1988                          * This is stupid and ugly, but very cheap...
1989                          */
1990                         damage = (damage + 63) / 64;
1991                 ALLOC_GROW(dir.files, dir.nr + 1, dir.alloc);
1992                 dir.files[dir.nr].name = file->name;
1993                 dir.files[dir.nr].changed = damage;
1994                 changed += damage;
1995                 dir.nr++;
1996         }
1997
1998         /* This can happen even with many files, if everything was renames */
1999         if (!changed)
2000                 return;
2001
2002         /* Show all directories with more than x% of the changes */
2003         qsort(dir.files, dir.nr, sizeof(dir.files[0]), dirstat_compare);
2004         gather_dirstat(options, &dir, changed, "", 0);
2005 }
2006
2007 static void free_diffstat_info(struct diffstat_t *diffstat)
2008 {
2009         int i;
2010         for (i = 0; i < diffstat->nr; i++) {
2011                 struct diffstat_file *f = diffstat->files[i];
2012                 if (f->name != f->print_name)
2013                         free(f->print_name);
2014                 free(f->name);
2015                 free(f->from_name);
2016                 free(f);
2017         }
2018         free(diffstat->files);
2019 }
2020
2021 struct checkdiff_t {
2022         const char *filename;
2023         int lineno;
2024         int conflict_marker_size;
2025         struct diff_options *o;
2026         unsigned ws_rule;
2027         unsigned status;
2028 };
2029
2030 static int is_conflict_marker(const char *line, int marker_size, unsigned long len)
2031 {
2032         char firstchar;
2033         int cnt;
2034
2035         if (len < marker_size + 1)
2036                 return 0;
2037         firstchar = line[0];
2038         switch (firstchar) {
2039         case '=': case '>': case '<': case '|':
2040                 break;
2041         default:
2042                 return 0;
2043         }
2044         for (cnt = 1; cnt < marker_size; cnt++)
2045                 if (line[cnt] != firstchar)
2046                         return 0;
2047         /* line[1] thru line[marker_size-1] are same as firstchar */
2048         if (len < marker_size + 1 || !isspace(line[marker_size]))
2049                 return 0;
2050         return 1;
2051 }
2052
2053 static void checkdiff_consume(void *priv, char *line, unsigned long len)
2054 {
2055         struct checkdiff_t *data = priv;
2056         int marker_size = data->conflict_marker_size;
2057         const char *ws = diff_get_color(data->o->use_color, DIFF_WHITESPACE);
2058         const char *reset = diff_get_color(data->o->use_color, DIFF_RESET);
2059         const char *set = diff_get_color(data->o->use_color, DIFF_FILE_NEW);
2060         char *err;
2061         char *line_prefix = "";
2062         struct strbuf *msgbuf;
2063
2064         assert(data->o);
2065         if (data->o->output_prefix) {
2066                 msgbuf = data->o->output_prefix(data->o,
2067                         data->o->output_prefix_data);
2068                 line_prefix = msgbuf->buf;
2069         }
2070
2071         if (line[0] == '+') {
2072                 unsigned bad;
2073                 data->lineno++;
2074                 if (is_conflict_marker(line + 1, marker_size, len - 1)) {
2075                         data->status |= 1;
2076                         fprintf(data->o->file,
2077                                 "%s%s:%d: leftover conflict marker\n",
2078                                 line_prefix, data->filename, data->lineno);
2079                 }
2080                 bad = ws_check(line + 1, len - 1, data->ws_rule);
2081                 if (!bad)
2082                         return;
2083                 data->status |= bad;
2084                 err = whitespace_error_string(bad);
2085                 fprintf(data->o->file, "%s%s:%d: %s.\n",
2086                         line_prefix, data->filename, data->lineno, err);
2087                 free(err);
2088                 emit_line(data->o, set, reset, line, 1);
2089                 ws_check_emit(line + 1, len - 1, data->ws_rule,
2090                               data->o->file, set, reset, ws);
2091         } else if (line[0] == ' ') {
2092                 data->lineno++;
2093         } else if (line[0] == '@') {
2094                 char *plus = strchr(line, '+');
2095                 if (plus)
2096                         data->lineno = strtol(plus, NULL, 10) - 1;
2097                 else
2098                         die("invalid diff");
2099         }
2100 }
2101
2102 static unsigned char *deflate_it(char *data,
2103                                  unsigned long size,
2104                                  unsigned long *result_size)
2105 {
2106         int bound;
2107         unsigned char *deflated;
2108         git_zstream stream;
2109
2110         memset(&stream, 0, sizeof(stream));
2111         git_deflate_init(&stream, zlib_compression_level);
2112         bound = git_deflate_bound(&stream, size);
2113         deflated = xmalloc(bound);
2114         stream.next_out = deflated;
2115         stream.avail_out = bound;
2116
2117         stream.next_in = (unsigned char *)data;
2118         stream.avail_in = size;
2119         while (git_deflate(&stream, Z_FINISH) == Z_OK)
2120                 ; /* nothing */
2121         git_deflate_end(&stream);
2122         *result_size = stream.total_out;
2123         return deflated;
2124 }
2125
2126 static void emit_binary_diff_body(FILE *file, mmfile_t *one, mmfile_t *two, char *prefix)
2127 {
2128         void *cp;
2129         void *delta;
2130         void *deflated;
2131         void *data;
2132         unsigned long orig_size;
2133         unsigned long delta_size;
2134         unsigned long deflate_size;
2135         unsigned long data_size;
2136
2137         /* We could do deflated delta, or we could do just deflated two,
2138          * whichever is smaller.
2139          */
2140         delta = NULL;
2141         deflated = deflate_it(two->ptr, two->size, &deflate_size);
2142         if (one->size && two->size) {
2143                 delta = diff_delta(one->ptr, one->size,
2144                                    two->ptr, two->size,
2145                                    &delta_size, deflate_size);
2146                 if (delta) {
2147                         void *to_free = delta;
2148                         orig_size = delta_size;
2149                         delta = deflate_it(delta, delta_size, &delta_size);
2150                         free(to_free);
2151                 }
2152         }
2153
2154         if (delta && delta_size < deflate_size) {
2155                 fprintf(file, "%sdelta %lu\n", prefix, orig_size);
2156                 free(deflated);
2157                 data = delta;
2158                 data_size = delta_size;
2159         }
2160         else {
2161                 fprintf(file, "%sliteral %lu\n", prefix, two->size);
2162                 free(delta);
2163                 data = deflated;
2164                 data_size = deflate_size;
2165         }
2166
2167         /* emit data encoded in base85 */
2168         cp = data;
2169         while (data_size) {
2170                 int bytes = (52 < data_size) ? 52 : data_size;
2171                 char line[70];
2172                 data_size -= bytes;
2173                 if (bytes <= 26)
2174                         line[0] = bytes + 'A' - 1;
2175                 else
2176                         line[0] = bytes - 26 + 'a' - 1;
2177                 encode_85(line + 1, cp, bytes);
2178                 cp = (char *) cp + bytes;
2179                 fprintf(file, "%s", prefix);
2180                 fputs(line, file);
2181                 fputc('\n', file);
2182         }
2183         fprintf(file, "%s\n", prefix);
2184         free(data);
2185 }
2186
2187 static void emit_binary_diff(FILE *file, mmfile_t *one, mmfile_t *two, char *prefix)
2188 {
2189         fprintf(file, "%sGIT binary patch\n", prefix);
2190         emit_binary_diff_body(file, one, two, prefix);
2191         emit_binary_diff_body(file, two, one, prefix);
2192 }
2193
2194 int diff_filespec_is_binary(struct diff_filespec *one)
2195 {
2196         if (one->is_binary == -1) {
2197                 diff_filespec_load_driver(one);
2198                 if (one->driver->binary != -1)
2199                         one->is_binary = one->driver->binary;
2200                 else {
2201                         if (!one->data && DIFF_FILE_VALID(one))
2202                                 diff_populate_filespec(one, 0);
2203                         if (one->data)
2204                                 one->is_binary = buffer_is_binary(one->data,
2205                                                 one->size);
2206                         if (one->is_binary == -1)
2207                                 one->is_binary = 0;
2208                 }
2209         }
2210         return one->is_binary;
2211 }
2212
2213 static const struct userdiff_funcname *diff_funcname_pattern(struct diff_filespec *one)
2214 {
2215         diff_filespec_load_driver(one);
2216         return one->driver->funcname.pattern ? &one->driver->funcname : NULL;
2217 }
2218
2219 void diff_set_mnemonic_prefix(struct diff_options *options, const char *a, const char *b)
2220 {
2221         if (!options->a_prefix)
2222                 options->a_prefix = a;
2223         if (!options->b_prefix)
2224                 options->b_prefix = b;
2225 }
2226
2227 struct userdiff_driver *get_textconv(struct diff_filespec *one)
2228 {
2229         if (!DIFF_FILE_VALID(one))
2230                 return NULL;
2231
2232         diff_filespec_load_driver(one);
2233         return userdiff_get_textconv(one->driver);
2234 }
2235
2236 static void builtin_diff(const char *name_a,
2237                          const char *name_b,
2238                          struct diff_filespec *one,
2239                          struct diff_filespec *two,
2240                          const char *xfrm_msg,
2241                          int must_show_header,
2242                          struct diff_options *o,
2243                          int complete_rewrite)
2244 {
2245         mmfile_t mf1, mf2;
2246         const char *lbl[2];
2247         char *a_one, *b_two;
2248         const char *meta = diff_get_color_opt(o, DIFF_METAINFO);
2249         const char *reset = diff_get_color_opt(o, DIFF_RESET);
2250         const char *a_prefix, *b_prefix;
2251         struct userdiff_driver *textconv_one = NULL;
2252         struct userdiff_driver *textconv_two = NULL;
2253         struct strbuf header = STRBUF_INIT;
2254         struct strbuf *msgbuf;
2255         char *line_prefix = "";
2256
2257         if (o->output_prefix) {
2258                 msgbuf = o->output_prefix(o, o->output_prefix_data);
2259                 line_prefix = msgbuf->buf;
2260         }
2261
2262         if (DIFF_OPT_TST(o, SUBMODULE_LOG) &&
2263                         (!one->mode || S_ISGITLINK(one->mode)) &&
2264                         (!two->mode || S_ISGITLINK(two->mode))) {
2265                 const char *del = diff_get_color_opt(o, DIFF_FILE_OLD);
2266                 const char *add = diff_get_color_opt(o, DIFF_FILE_NEW);
2267                 show_submodule_summary(o->file, one ? one->path : two->path,
2268                                 one->sha1, two->sha1, two->dirty_submodule,
2269                                 meta, del, add, reset);
2270                 return;
2271         }
2272
2273         if (DIFF_OPT_TST(o, ALLOW_TEXTCONV)) {
2274                 textconv_one = get_textconv(one);
2275                 textconv_two = get_textconv(two);
2276         }
2277
2278         diff_set_mnemonic_prefix(o, "a/", "b/");
2279         if (DIFF_OPT_TST(o, REVERSE_DIFF)) {
2280                 a_prefix = o->b_prefix;
2281                 b_prefix = o->a_prefix;
2282         } else {
2283                 a_prefix = o->a_prefix;
2284                 b_prefix = o->b_prefix;
2285         }
2286
2287         /* Never use a non-valid filename anywhere if at all possible */
2288         name_a = DIFF_FILE_VALID(one) ? name_a : name_b;
2289         name_b = DIFF_FILE_VALID(two) ? name_b : name_a;
2290
2291         a_one = quote_two(a_prefix, name_a + (*name_a == '/'));
2292         b_two = quote_two(b_prefix, name_b + (*name_b == '/'));
2293         lbl[0] = DIFF_FILE_VALID(one) ? a_one : "/dev/null";
2294         lbl[1] = DIFF_FILE_VALID(two) ? b_two : "/dev/null";
2295         strbuf_addf(&header, "%s%sdiff --git %s %s%s\n", line_prefix, meta, a_one, b_two, reset);
2296         if (lbl[0][0] == '/') {
2297                 /* /dev/null */
2298                 strbuf_addf(&header, "%s%snew file mode %06o%s\n", line_prefix, meta, two->mode, reset);
2299                 if (xfrm_msg)
2300                         strbuf_addstr(&header, xfrm_msg);
2301                 must_show_header = 1;
2302         }
2303         else if (lbl[1][0] == '/') {
2304                 strbuf_addf(&header, "%s%sdeleted file mode %06o%s\n", line_prefix, meta, one->mode, reset);
2305                 if (xfrm_msg)
2306                         strbuf_addstr(&header, xfrm_msg);
2307                 must_show_header = 1;
2308         }
2309         else {
2310                 if (one->mode != two->mode) {
2311                         strbuf_addf(&header, "%s%sold mode %06o%s\n", line_prefix, meta, one->mode, reset);
2312                         strbuf_addf(&header, "%s%snew mode %06o%s\n", line_prefix, meta, two->mode, reset);
2313                         must_show_header = 1;
2314                 }
2315                 if (xfrm_msg)
2316                         strbuf_addstr(&header, xfrm_msg);
2317
2318                 /*
2319                  * we do not run diff between different kind
2320                  * of objects.
2321                  */
2322                 if ((one->mode ^ two->mode) & S_IFMT)
2323                         goto free_ab_and_return;
2324                 if (complete_rewrite &&
2325                     (textconv_one || !diff_filespec_is_binary(one)) &&
2326                     (textconv_two || !diff_filespec_is_binary(two))) {
2327                         fprintf(o->file, "%s", header.buf);
2328                         strbuf_reset(&header);
2329                         emit_rewrite_diff(name_a, name_b, one, two,
2330                                                 textconv_one, textconv_two, o);
2331                         o->found_changes = 1;
2332                         goto free_ab_and_return;
2333                 }
2334         }
2335
2336         if (o->irreversible_delete && lbl[1][0] == '/') {
2337                 fprintf(o->file, "%s", header.buf);
2338                 strbuf_reset(&header);
2339                 goto free_ab_and_return;
2340         } else if (!DIFF_OPT_TST(o, TEXT) &&
2341             ( (!textconv_one && diff_filespec_is_binary(one)) ||
2342               (!textconv_two && diff_filespec_is_binary(two)) )) {
2343                 if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
2344                         die("unable to read files to diff");
2345                 /* Quite common confusing case */
2346                 if (mf1.size == mf2.size &&
2347                     !memcmp(mf1.ptr, mf2.ptr, mf1.size)) {
2348                         if (must_show_header)
2349                                 fprintf(o->file, "%s", header.buf);
2350                         goto free_ab_and_return;
2351                 }
2352                 fprintf(o->file, "%s", header.buf);
2353                 strbuf_reset(&header);
2354                 if (DIFF_OPT_TST(o, BINARY))
2355                         emit_binary_diff(o->file, &mf1, &mf2, line_prefix);
2356                 else
2357                         fprintf(o->file, "%sBinary files %s and %s differ\n",
2358                                 line_prefix, lbl[0], lbl[1]);
2359                 o->found_changes = 1;
2360         } else {
2361                 /* Crazy xdl interfaces.. */
2362                 const char *diffopts = getenv("GIT_DIFF_OPTS");
2363                 xpparam_t xpp;
2364                 xdemitconf_t xecfg;
2365                 struct emit_callback ecbdata;
2366                 const struct userdiff_funcname *pe;
2367
2368                 if (must_show_header) {
2369                         fprintf(o->file, "%s", header.buf);
2370                         strbuf_reset(&header);
2371                 }
2372
2373                 mf1.size = fill_textconv(textconv_one, one, &mf1.ptr);
2374                 mf2.size = fill_textconv(textconv_two, two, &mf2.ptr);
2375
2376                 pe = diff_funcname_pattern(one);
2377                 if (!pe)
2378                         pe = diff_funcname_pattern(two);
2379
2380                 memset(&xpp, 0, sizeof(xpp));
2381                 memset(&xecfg, 0, sizeof(xecfg));
2382                 memset(&ecbdata, 0, sizeof(ecbdata));
2383                 ecbdata.label_path = lbl;
2384                 ecbdata.color_diff = want_color(o->use_color);
2385                 ecbdata.found_changesp = &o->found_changes;
2386                 ecbdata.ws_rule = whitespace_rule(name_b ? name_b : name_a);
2387                 if (ecbdata.ws_rule & WS_BLANK_AT_EOF)
2388                         check_blank_at_eof(&mf1, &mf2, &ecbdata);
2389                 ecbdata.opt = o;
2390                 ecbdata.header = header.len ? &header : NULL;
2391                 xpp.flags = o->xdl_opts;
2392                 xecfg.ctxlen = o->context;
2393                 xecfg.interhunkctxlen = o->interhunkcontext;
2394                 xecfg.flags = XDL_EMIT_FUNCNAMES;
2395                 if (DIFF_OPT_TST(o, FUNCCONTEXT))
2396                         xecfg.flags |= XDL_EMIT_FUNCCONTEXT;
2397                 if (pe)
2398                         xdiff_set_find_func(&xecfg, pe->pattern, pe->cflags);
2399                 if (!diffopts)
2400                         ;
2401                 else if (!prefixcmp(diffopts, "--unified="))
2402                         xecfg.ctxlen = strtoul(diffopts + 10, NULL, 10);
2403                 else if (!prefixcmp(diffopts, "-u"))
2404                         xecfg.ctxlen = strtoul(diffopts + 2, NULL, 10);
2405                 if (o->word_diff)
2406                         init_diff_words_data(&ecbdata, o, one, two);
2407                 xdi_diff_outf(&mf1, &mf2, fn_out_consume, &ecbdata,
2408                               &xpp, &xecfg);
2409                 if (o->word_diff)
2410                         free_diff_words_data(&ecbdata);
2411                 if (textconv_one)
2412                         free(mf1.ptr);
2413                 if (textconv_two)
2414                         free(mf2.ptr);
2415                 xdiff_clear_find_func(&xecfg);
2416         }
2417
2418  free_ab_and_return:
2419         strbuf_release(&header);
2420         diff_free_filespec_data(one);
2421         diff_free_filespec_data(two);
2422         free(a_one);
2423         free(b_two);
2424         return;
2425 }
2426
2427 static void builtin_diffstat(const char *name_a, const char *name_b,
2428                              struct diff_filespec *one,
2429                              struct diff_filespec *two,
2430                              struct diffstat_t *diffstat,
2431                              struct diff_options *o,
2432                              struct diff_filepair *p)
2433 {
2434         mmfile_t mf1, mf2;
2435         struct diffstat_file *data;
2436         int same_contents;
2437         int complete_rewrite = 0;
2438
2439         if (!DIFF_PAIR_UNMERGED(p)) {
2440                 if (p->status == DIFF_STATUS_MODIFIED && p->score)
2441                         complete_rewrite = 1;
2442         }
2443
2444         data = diffstat_add(diffstat, name_a, name_b);
2445         data->is_interesting = p->status != DIFF_STATUS_UNKNOWN;
2446
2447         if (!one || !two) {
2448                 data->is_unmerged = 1;
2449                 return;
2450         }
2451
2452         same_contents = !hashcmp(one->sha1, two->sha1);
2453
2454         if (diff_filespec_is_binary(one) || diff_filespec_is_binary(two)) {
2455                 data->is_binary = 1;
2456                 if (same_contents) {
2457                         data->added = 0;
2458                         data->deleted = 0;
2459                 } else {
2460                         data->added = diff_filespec_size(two);
2461                         data->deleted = diff_filespec_size(one);
2462                 }
2463         }
2464
2465         else if (complete_rewrite) {
2466                 diff_populate_filespec(one, 0);
2467                 diff_populate_filespec(two, 0);
2468                 data->deleted = count_lines(one->data, one->size);
2469                 data->added = count_lines(two->data, two->size);
2470         }
2471
2472         else if (!same_contents) {
2473                 /* Crazy xdl interfaces.. */
2474                 xpparam_t xpp;
2475                 xdemitconf_t xecfg;
2476
2477                 if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
2478                         die("unable to read files to diff");
2479
2480                 memset(&xpp, 0, sizeof(xpp));
2481                 memset(&xecfg, 0, sizeof(xecfg));
2482                 xpp.flags = o->xdl_opts;
2483                 xecfg.ctxlen = o->context;
2484                 xecfg.interhunkctxlen = o->interhunkcontext;
2485                 xdi_diff_outf(&mf1, &mf2, diffstat_consume, diffstat,
2486                               &xpp, &xecfg);
2487         }
2488
2489         diff_free_filespec_data(one);
2490         diff_free_filespec_data(two);
2491 }
2492
2493 static void builtin_checkdiff(const char *name_a, const char *name_b,
2494                               const char *attr_path,
2495                               struct diff_filespec *one,
2496                               struct diff_filespec *two,
2497                               struct diff_options *o)
2498 {
2499         mmfile_t mf1, mf2;
2500         struct checkdiff_t data;
2501
2502         if (!two)
2503                 return;
2504
2505         memset(&data, 0, sizeof(data));
2506         data.filename = name_b ? name_b : name_a;
2507         data.lineno = 0;
2508         data.o = o;
2509         data.ws_rule = whitespace_rule(attr_path);
2510         data.conflict_marker_size = ll_merge_marker_size(attr_path);
2511
2512         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
2513                 die("unable to read files to diff");
2514
2515         /*
2516          * All the other codepaths check both sides, but not checking
2517          * the "old" side here is deliberate.  We are checking the newly
2518          * introduced changes, and as long as the "new" side is text, we
2519          * can and should check what it introduces.
2520          */
2521         if (diff_filespec_is_binary(two))
2522                 goto free_and_return;
2523         else {
2524                 /* Crazy xdl interfaces.. */
2525                 xpparam_t xpp;
2526                 xdemitconf_t xecfg;
2527
2528                 memset(&xpp, 0, sizeof(xpp));
2529                 memset(&xecfg, 0, sizeof(xecfg));
2530                 xecfg.ctxlen = 1; /* at least one context line */
2531                 xpp.flags = 0;
2532                 xdi_diff_outf(&mf1, &mf2, checkdiff_consume, &data,
2533                               &xpp, &xecfg);
2534
2535                 if (data.ws_rule & WS_BLANK_AT_EOF) {
2536                         struct emit_callback ecbdata;
2537                         int blank_at_eof;
2538
2539                         ecbdata.ws_rule = data.ws_rule;
2540                         check_blank_at_eof(&mf1, &mf2, &ecbdata);
2541                         blank_at_eof = ecbdata.blank_at_eof_in_postimage;
2542
2543                         if (blank_at_eof) {
2544                                 static char *err;
2545                                 if (!err)
2546                                         err = whitespace_error_string(WS_BLANK_AT_EOF);
2547                                 fprintf(o->file, "%s:%d: %s.\n",
2548                                         data.filename, blank_at_eof, err);
2549                                 data.status = 1; /* report errors */
2550                         }
2551                 }
2552         }
2553  free_and_return:
2554         diff_free_filespec_data(one);
2555         diff_free_filespec_data(two);
2556         if (data.status)
2557                 DIFF_OPT_SET(o, CHECK_FAILED);
2558 }
2559
2560 struct diff_filespec *alloc_filespec(const char *path)
2561 {
2562         int namelen = strlen(path);
2563         struct diff_filespec *spec = xmalloc(sizeof(*spec) + namelen + 1);
2564
2565         memset(spec, 0, sizeof(*spec));
2566         spec->path = (char *)(spec + 1);
2567         memcpy(spec->path, path, namelen+1);
2568         spec->count = 1;
2569         spec->is_binary = -1;
2570         return spec;
2571 }
2572
2573 void free_filespec(struct diff_filespec *spec)
2574 {
2575         if (!--spec->count) {
2576                 diff_free_filespec_data(spec);
2577                 free(spec);
2578         }
2579 }
2580
2581 void fill_filespec(struct diff_filespec *spec, const unsigned char *sha1,
2582                    int sha1_valid, unsigned short mode)
2583 {
2584         if (mode) {
2585                 spec->mode = canon_mode(mode);
2586                 hashcpy(spec->sha1, sha1);
2587                 spec->sha1_valid = sha1_valid;
2588         }
2589 }
2590
2591 /*
2592  * Given a name and sha1 pair, if the index tells us the file in
2593  * the work tree has that object contents, return true, so that
2594  * prepare_temp_file() does not have to inflate and extract.
2595  */
2596 static int reuse_worktree_file(const char *name, const unsigned char *sha1, int want_file)
2597 {
2598         struct cache_entry *ce;
2599         struct stat st;
2600         int pos, len;
2601
2602         /*
2603          * We do not read the cache ourselves here, because the
2604          * benchmark with my previous version that always reads cache
2605          * shows that it makes things worse for diff-tree comparing
2606          * two linux-2.6 kernel trees in an already checked out work
2607          * tree.  This is because most diff-tree comparisons deal with
2608          * only a small number of files, while reading the cache is
2609          * expensive for a large project, and its cost outweighs the
2610          * savings we get by not inflating the object to a temporary
2611          * file.  Practically, this code only helps when we are used
2612          * by diff-cache --cached, which does read the cache before
2613          * calling us.
2614          */
2615         if (!active_cache)
2616                 return 0;
2617
2618         /* We want to avoid the working directory if our caller
2619          * doesn't need the data in a normal file, this system
2620          * is rather slow with its stat/open/mmap/close syscalls,
2621          * and the object is contained in a pack file.  The pack
2622          * is probably already open and will be faster to obtain
2623          * the data through than the working directory.  Loose
2624          * objects however would tend to be slower as they need
2625          * to be individually opened and inflated.
2626          */
2627         if (!FAST_WORKING_DIRECTORY && !want_file && has_sha1_pack(sha1))
2628                 return 0;
2629
2630         len = strlen(name);
2631         pos = cache_name_pos(name, len);
2632         if (pos < 0)
2633                 return 0;
2634         ce = active_cache[pos];
2635
2636         /*
2637          * This is not the sha1 we are looking for, or
2638          * unreusable because it is not a regular file.
2639          */
2640         if (hashcmp(sha1, ce->sha1) || !S_ISREG(ce->ce_mode))
2641                 return 0;
2642
2643         /*
2644          * If ce is marked as "assume unchanged", there is no
2645          * guarantee that work tree matches what we are looking for.
2646          */
2647         if ((ce->ce_flags & CE_VALID) || ce_skip_worktree(ce))
2648                 return 0;
2649
2650         /*
2651          * If ce matches the file in the work tree, we can reuse it.
2652          */
2653         if (ce_uptodate(ce) ||
2654             (!lstat(name, &st) && !ce_match_stat(ce, &st, 0)))
2655                 return 1;
2656
2657         return 0;
2658 }
2659
2660 static int diff_populate_gitlink(struct diff_filespec *s, int size_only)
2661 {
2662         int len;
2663         char *data = xmalloc(100), *dirty = "";
2664
2665         /* Are we looking at the work tree? */
2666         if (s->dirty_submodule)
2667                 dirty = "-dirty";
2668
2669         len = snprintf(data, 100,
2670                        "Subproject commit %s%s\n", sha1_to_hex(s->sha1), dirty);
2671         s->data = data;
2672         s->size = len;
2673         s->should_free = 1;
2674         if (size_only) {
2675                 s->data = NULL;
2676                 free(data);
2677         }
2678         return 0;
2679 }
2680
2681 /*
2682  * While doing rename detection and pickaxe operation, we may need to
2683  * grab the data for the blob (or file) for our own in-core comparison.
2684  * diff_filespec has data and size fields for this purpose.
2685  */
2686 int diff_populate_filespec(struct diff_filespec *s, int size_only)
2687 {
2688         int err = 0;
2689         if (!DIFF_FILE_VALID(s))
2690                 die("internal error: asking to populate invalid file.");
2691         if (S_ISDIR(s->mode))
2692                 return -1;
2693
2694         if (s->data)
2695                 return 0;
2696
2697         if (size_only && 0 < s->size)
2698                 return 0;
2699
2700         if (S_ISGITLINK(s->mode))
2701                 return diff_populate_gitlink(s, size_only);
2702
2703         if (!s->sha1_valid ||
2704             reuse_worktree_file(s->path, s->sha1, 0)) {
2705                 struct strbuf buf = STRBUF_INIT;
2706                 struct stat st;
2707                 int fd;
2708
2709                 if (lstat(s->path, &st) < 0) {
2710                         if (errno == ENOENT) {
2711                         err_empty:
2712                                 err = -1;
2713                         empty:
2714                                 s->data = (char *)"";
2715                                 s->size = 0;
2716                                 return err;
2717                         }
2718                 }
2719                 s->size = xsize_t(st.st_size);
2720                 if (!s->size)
2721                         goto empty;
2722                 if (S_ISLNK(st.st_mode)) {
2723                         struct strbuf sb = STRBUF_INIT;
2724
2725                         if (strbuf_readlink(&sb, s->path, s->size))
2726                                 goto err_empty;
2727                         s->size = sb.len;
2728                         s->data = strbuf_detach(&sb, NULL);
2729                         s->should_free = 1;
2730                         return 0;
2731                 }
2732                 if (size_only)
2733                         return 0;
2734                 fd = open(s->path, O_RDONLY);
2735                 if (fd < 0)
2736                         goto err_empty;
2737                 s->data = xmmap(NULL, s->size, PROT_READ, MAP_PRIVATE, fd, 0);
2738                 close(fd);
2739                 s->should_munmap = 1;
2740
2741                 /*
2742                  * Convert from working tree format to canonical git format
2743                  */
2744                 if (convert_to_git(s->path, s->data, s->size, &buf, safe_crlf)) {
2745                         size_t size = 0;
2746                         munmap(s->data, s->size);
2747                         s->should_munmap = 0;
2748                         s->data = strbuf_detach(&buf, &size);
2749                         s->size = size;
2750                         s->should_free = 1;
2751                 }
2752         }
2753         else {
2754                 enum object_type type;
2755                 if (size_only) {
2756                         type = sha1_object_info(s->sha1, &s->size);
2757                         if (type < 0)
2758                                 die("unable to read %s", sha1_to_hex(s->sha1));
2759                 } else {
2760                         s->data = read_sha1_file(s->sha1, &type, &s->size);
2761                         if (!s->data)
2762                                 die("unable to read %s", sha1_to_hex(s->sha1));
2763                         s->should_free = 1;
2764                 }
2765         }
2766         return 0;
2767 }
2768
2769 void diff_free_filespec_blob(struct diff_filespec *s)
2770 {
2771         if (s->should_free)
2772                 free(s->data);
2773         else if (s->should_munmap)
2774                 munmap(s->data, s->size);
2775
2776         if (s->should_free || s->should_munmap) {
2777                 s->should_free = s->should_munmap = 0;
2778                 s->data = NULL;
2779         }
2780 }
2781
2782 void diff_free_filespec_data(struct diff_filespec *s)
2783 {
2784         diff_free_filespec_blob(s);
2785         free(s->cnt_data);
2786         s->cnt_data = NULL;
2787 }
2788
2789 static void prep_temp_blob(const char *path, struct diff_tempfile *temp,
2790                            void *blob,
2791                            unsigned long size,
2792                            const unsigned char *sha1,
2793                            int mode)
2794 {
2795         int fd;
2796         struct strbuf buf = STRBUF_INIT;
2797         struct strbuf template = STRBUF_INIT;
2798         char *path_dup = xstrdup(path);
2799         const char *base = basename(path_dup);
2800
2801         /* Generate "XXXXXX_basename.ext" */
2802         strbuf_addstr(&template, "XXXXXX_");
2803         strbuf_addstr(&template, base);
2804
2805         fd = git_mkstemps(temp->tmp_path, PATH_MAX, template.buf,
2806                         strlen(base) + 1);
2807         if (fd < 0)
2808                 die_errno("unable to create temp-file");
2809         if (convert_to_working_tree(path,
2810                         (const char *)blob, (size_t)size, &buf)) {
2811                 blob = buf.buf;
2812                 size = buf.len;
2813         }
2814         if (write_in_full(fd, blob, size) != size)
2815                 die_errno("unable to write temp-file");
2816         close(fd);
2817         temp->name = temp->tmp_path;
2818         strcpy(temp->hex, sha1_to_hex(sha1));
2819         temp->hex[40] = 0;
2820         sprintf(temp->mode, "%06o", mode);
2821         strbuf_release(&buf);
2822         strbuf_release(&template);
2823         free(path_dup);
2824 }
2825
2826 static struct diff_tempfile *prepare_temp_file(const char *name,
2827                 struct diff_filespec *one)
2828 {
2829         struct diff_tempfile *temp = claim_diff_tempfile();
2830
2831         if (!DIFF_FILE_VALID(one)) {
2832         not_a_valid_file:
2833                 /* A '-' entry produces this for file-2, and
2834                  * a '+' entry produces this for file-1.
2835                  */
2836                 temp->name = "/dev/null";
2837                 strcpy(temp->hex, ".");
2838                 strcpy(temp->mode, ".");
2839                 return temp;
2840         }
2841
2842         if (!remove_tempfile_installed) {
2843                 atexit(remove_tempfile);
2844                 sigchain_push_common(remove_tempfile_on_signal);
2845                 remove_tempfile_installed = 1;
2846         }
2847
2848         if (!one->sha1_valid ||
2849             reuse_worktree_file(name, one->sha1, 1)) {
2850                 struct stat st;
2851                 if (lstat(name, &st) < 0) {
2852                         if (errno == ENOENT)
2853                                 goto not_a_valid_file;
2854                         die_errno("stat(%s)", name);
2855                 }
2856                 if (S_ISLNK(st.st_mode)) {
2857                         struct strbuf sb = STRBUF_INIT;
2858                         if (strbuf_readlink(&sb, name, st.st_size) < 0)
2859                                 die_errno("readlink(%s)", name);
2860                         prep_temp_blob(name, temp, sb.buf, sb.len,
2861                                        (one->sha1_valid ?
2862                                         one->sha1 : null_sha1),
2863                                        (one->sha1_valid ?
2864                                         one->mode : S_IFLNK));
2865                         strbuf_release(&sb);
2866                 }
2867                 else {
2868                         /* we can borrow from the file in the work tree */
2869                         temp->name = name;
2870                         if (!one->sha1_valid)
2871                                 strcpy(temp->hex, sha1_to_hex(null_sha1));
2872                         else
2873                                 strcpy(temp->hex, sha1_to_hex(one->sha1));
2874                         /* Even though we may sometimes borrow the
2875                          * contents from the work tree, we always want
2876                          * one->mode.  mode is trustworthy even when
2877                          * !(one->sha1_valid), as long as
2878                          * DIFF_FILE_VALID(one).
2879                          */
2880                         sprintf(temp->mode, "%06o", one->mode);
2881                 }
2882                 return temp;
2883         }
2884         else {
2885                 if (diff_populate_filespec(one, 0))
2886                         die("cannot read data blob for %s", one->path);
2887                 prep_temp_blob(name, temp, one->data, one->size,
2888                                one->sha1, one->mode);
2889         }
2890         return temp;
2891 }
2892
2893 /* An external diff command takes:
2894  *
2895  * diff-cmd name infile1 infile1-sha1 infile1-mode \
2896  *               infile2 infile2-sha1 infile2-mode [ rename-to ]
2897  *
2898  */
2899 static void run_external_diff(const char *pgm,
2900                               const char *name,
2901                               const char *other,
2902                               struct diff_filespec *one,
2903                               struct diff_filespec *two,
2904                               const char *xfrm_msg,
2905                               int complete_rewrite)
2906 {
2907         const char *spawn_arg[10];
2908         int retval;
2909         const char **arg = &spawn_arg[0];
2910
2911         if (one && two) {
2912                 struct diff_tempfile *temp_one, *temp_two;
2913                 const char *othername = (other ? other : name);
2914                 temp_one = prepare_temp_file(name, one);
2915                 temp_two = prepare_temp_file(othername, two);
2916                 *arg++ = pgm;
2917                 *arg++ = name;
2918                 *arg++ = temp_one->name;
2919                 *arg++ = temp_one->hex;
2920                 *arg++ = temp_one->mode;
2921                 *arg++ = temp_two->name;
2922                 *arg++ = temp_two->hex;
2923                 *arg++ = temp_two->mode;
2924                 if (other) {
2925                         *arg++ = other;
2926                         *arg++ = xfrm_msg;
2927                 }
2928         } else {
2929                 *arg++ = pgm;
2930                 *arg++ = name;
2931         }
2932         *arg = NULL;
2933         fflush(NULL);
2934         retval = run_command_v_opt(spawn_arg, RUN_USING_SHELL);
2935         remove_tempfile();
2936         if (retval) {
2937                 fprintf(stderr, "external diff died, stopping at %s.\n", name);
2938                 exit(1);
2939         }
2940 }
2941
2942 static int similarity_index(struct diff_filepair *p)
2943 {
2944         return p->score * 100 / MAX_SCORE;
2945 }
2946
2947 static void fill_metainfo(struct strbuf *msg,
2948                           const char *name,
2949                           const char *other,
2950                           struct diff_filespec *one,
2951                           struct diff_filespec *two,
2952                           struct diff_options *o,
2953                           struct diff_filepair *p,
2954                           int *must_show_header,
2955                           int use_color)
2956 {
2957         const char *set = diff_get_color(use_color, DIFF_METAINFO);
2958         const char *reset = diff_get_color(use_color, DIFF_RESET);
2959         struct strbuf *msgbuf;
2960         char *line_prefix = "";
2961
2962         *must_show_header = 1;
2963         if (o->output_prefix) {
2964                 msgbuf = o->output_prefix(o, o->output_prefix_data);
2965                 line_prefix = msgbuf->buf;
2966         }
2967         strbuf_init(msg, PATH_MAX * 2 + 300);
2968         switch (p->status) {
2969         case DIFF_STATUS_COPIED:
2970                 strbuf_addf(msg, "%s%ssimilarity index %d%%",
2971                             line_prefix, set, similarity_index(p));
2972                 strbuf_addf(msg, "%s\n%s%scopy from ",
2973                             reset,  line_prefix, set);
2974                 quote_c_style(name, msg, NULL, 0);
2975                 strbuf_addf(msg, "%s\n%s%scopy to ", reset, line_prefix, set);
2976                 quote_c_style(other, msg, NULL, 0);
2977                 strbuf_addf(msg, "%s\n", reset);
2978                 break;
2979         case DIFF_STATUS_RENAMED:
2980                 strbuf_addf(msg, "%s%ssimilarity index %d%%",
2981                             line_prefix, set, similarity_index(p));
2982                 strbuf_addf(msg, "%s\n%s%srename from ",
2983                             reset, line_prefix, set);
2984                 quote_c_style(name, msg, NULL, 0);
2985                 strbuf_addf(msg, "%s\n%s%srename to ",
2986                             reset, line_prefix, set);
2987                 quote_c_style(other, msg, NULL, 0);
2988                 strbuf_addf(msg, "%s\n", reset);
2989                 break;
2990         case DIFF_STATUS_MODIFIED:
2991                 if (p->score) {
2992                         strbuf_addf(msg, "%s%sdissimilarity index %d%%%s\n",
2993                                     line_prefix,
2994                                     set, similarity_index(p), reset);
2995                         break;
2996                 }
2997                 /* fallthru */
2998         default:
2999                 *must_show_header = 0;
3000         }
3001         if (one && two && hashcmp(one->sha1, two->sha1)) {
3002                 int abbrev = DIFF_OPT_TST(o, FULL_INDEX) ? 40 : DEFAULT_ABBREV;
3003
3004                 if (DIFF_OPT_TST(o, BINARY)) {
3005                         mmfile_t mf;
3006                         if ((!fill_mmfile(&mf, one) && diff_filespec_is_binary(one)) ||
3007                             (!fill_mmfile(&mf, two) && diff_filespec_is_binary(two)))
3008                                 abbrev = 40;
3009                 }
3010                 strbuf_addf(msg, "%s%sindex %s..", line_prefix, set,
3011                             find_unique_abbrev(one->sha1, abbrev));
3012                 strbuf_addstr(msg, find_unique_abbrev(two->sha1, abbrev));
3013                 if (one->mode == two->mode)
3014                         strbuf_addf(msg, " %06o", one->mode);
3015                 strbuf_addf(msg, "%s\n", reset);
3016         }
3017 }
3018
3019 static void run_diff_cmd(const char *pgm,
3020                          const char *name,
3021                          const char *other,
3022                          const char *attr_path,
3023                          struct diff_filespec *one,
3024                          struct diff_filespec *two,
3025                          struct strbuf *msg,
3026                          struct diff_options *o,
3027                          struct diff_filepair *p)
3028 {
3029         const char *xfrm_msg = NULL;
3030         int complete_rewrite = (p->status == DIFF_STATUS_MODIFIED) && p->score;
3031         int must_show_header = 0;
3032
3033
3034         if (DIFF_OPT_TST(o, ALLOW_EXTERNAL)) {
3035                 struct userdiff_driver *drv = userdiff_find_by_path(attr_path);
3036                 if (drv && drv->external)
3037                         pgm = drv->external;
3038         }
3039
3040         if (msg) {
3041                 /*
3042                  * don't use colors when the header is intended for an
3043                  * external diff driver
3044                  */
3045                 fill_metainfo(msg, name, other, one, two, o, p,
3046                               &must_show_header,
3047                               want_color(o->use_color) && !pgm);
3048                 xfrm_msg = msg->len ? msg->buf : NULL;
3049         }
3050
3051         if (pgm) {
3052                 run_external_diff(pgm, name, other, one, two, xfrm_msg,
3053                                   complete_rewrite);
3054                 return;
3055         }
3056         if (one && two)
3057                 builtin_diff(name, other ? other : name,
3058                              one, two, xfrm_msg, must_show_header,
3059                              o, complete_rewrite);
3060         else
3061                 fprintf(o->file, "* Unmerged path %s\n", name);
3062 }
3063
3064 static void diff_fill_sha1_info(struct diff_filespec *one)
3065 {
3066         if (DIFF_FILE_VALID(one)) {
3067                 if (!one->sha1_valid) {
3068                         struct stat st;
3069                         if (one->is_stdin) {
3070                                 hashcpy(one->sha1, null_sha1);
3071                                 return;
3072                         }
3073                         if (lstat(one->path, &st) < 0)
3074                                 die_errno("stat '%s'", one->path);
3075                         if (index_path(one->sha1, one->path, &st, 0))
3076                                 die("cannot hash %s", one->path);
3077                 }
3078         }
3079         else
3080                 hashclr(one->sha1);
3081 }
3082
3083 static void strip_prefix(int prefix_length, const char **namep, const char **otherp)
3084 {
3085         /* Strip the prefix but do not molest /dev/null and absolute paths */
3086         if (*namep && **namep != '/') {
3087                 *namep += prefix_length;
3088                 if (**namep == '/')
3089                         ++*namep;
3090         }
3091         if (*otherp && **otherp != '/') {
3092                 *otherp += prefix_length;
3093                 if (**otherp == '/')
3094                         ++*otherp;
3095         }
3096 }
3097
3098 static void run_diff(struct diff_filepair *p, struct diff_options *o)
3099 {
3100         const char *pgm = external_diff();
3101         struct strbuf msg;
3102         struct diff_filespec *one = p->one;
3103         struct diff_filespec *two = p->two;
3104         const char *name;
3105         const char *other;
3106         const char *attr_path;
3107
3108         name  = p->one->path;
3109         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
3110         attr_path = name;
3111         if (o->prefix_length)
3112                 strip_prefix(o->prefix_length, &name, &other);
3113
3114         if (!DIFF_OPT_TST(o, ALLOW_EXTERNAL))
3115                 pgm = NULL;
3116
3117         if (DIFF_PAIR_UNMERGED(p)) {
3118                 run_diff_cmd(pgm, name, NULL, attr_path,
3119                              NULL, NULL, NULL, o, p);
3120                 return;
3121         }
3122
3123         diff_fill_sha1_info(one);
3124         diff_fill_sha1_info(two);
3125
3126         if (!pgm &&
3127             DIFF_FILE_VALID(one) && DIFF_FILE_VALID(two) &&
3128             (S_IFMT & one->mode) != (S_IFMT & two->mode)) {
3129                 /*
3130                  * a filepair that changes between file and symlink
3131                  * needs to be split into deletion and creation.
3132                  */
3133                 struct diff_filespec *null = alloc_filespec(two->path);
3134                 run_diff_cmd(NULL, name, other, attr_path,
3135                              one, null, &msg, o, p);
3136                 free(null);
3137                 strbuf_release(&msg);
3138
3139                 null = alloc_filespec(one->path);
3140                 run_diff_cmd(NULL, name, other, attr_path,
3141                              null, two, &msg, o, p);
3142                 free(null);
3143         }
3144         else
3145                 run_diff_cmd(pgm, name, other, attr_path,
3146                              one, two, &msg, o, p);
3147
3148         strbuf_release(&msg);
3149 }
3150
3151 static void run_diffstat(struct diff_filepair *p, struct diff_options *o,
3152                          struct diffstat_t *diffstat)
3153 {
3154         const char *name;
3155         const char *other;
3156
3157         if (DIFF_PAIR_UNMERGED(p)) {
3158                 /* unmerged */
3159                 builtin_diffstat(p->one->path, NULL, NULL, NULL, diffstat, o, p);
3160                 return;
3161         }
3162
3163         name = p->one->path;
3164         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
3165
3166         if (o->prefix_length)
3167                 strip_prefix(o->prefix_length, &name, &other);
3168
3169         diff_fill_sha1_info(p->one);
3170         diff_fill_sha1_info(p->two);
3171
3172         builtin_diffstat(name, other, p->one, p->two, diffstat, o, p);
3173 }
3174
3175 static void run_checkdiff(struct diff_filepair *p, struct diff_options *o)
3176 {
3177         const char *name;
3178         const char *other;
3179         const char *attr_path;
3180
3181         if (DIFF_PAIR_UNMERGED(p)) {
3182                 /* unmerged */
3183                 return;
3184         }
3185
3186         name = p->one->path;
3187         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
3188         attr_path = other ? other : name;
3189
3190         if (o->prefix_length)
3191                 strip_prefix(o->prefix_length, &name, &other);
3192
3193         diff_fill_sha1_info(p->one);
3194         diff_fill_sha1_info(p->two);
3195
3196         builtin_checkdiff(name, other, attr_path, p->one, p->two, o);
3197 }
3198
3199 void diff_setup(struct diff_options *options)
3200 {
3201         memcpy(options, &default_diff_options, sizeof(*options));
3202
3203         options->file = stdout;
3204
3205         options->line_termination = '\n';
3206         options->break_opt = -1;
3207         options->rename_limit = -1;
3208         options->dirstat_permille = diff_dirstat_permille_default;
3209         options->context = diff_context_default;
3210         DIFF_OPT_SET(options, RENAME_EMPTY);
3211
3212         options->change = diff_change;
3213         options->add_remove = diff_addremove;
3214         options->use_color = diff_use_color_default;
3215         options->detect_rename = diff_detect_rename_default;
3216
3217         if (diff_no_prefix) {
3218                 options->a_prefix = options->b_prefix = "";
3219         } else if (!diff_mnemonic_prefix) {
3220                 options->a_prefix = "a/";
3221                 options->b_prefix = "b/";
3222         }
3223 }
3224
3225 void diff_setup_done(struct diff_options *options)
3226 {
3227         int count = 0;
3228
3229         if (options->output_format & DIFF_FORMAT_NAME)
3230                 count++;
3231         if (options->output_format & DIFF_FORMAT_NAME_STATUS)
3232                 count++;
3233         if (options->output_format & DIFF_FORMAT_CHECKDIFF)
3234                 count++;
3235         if (options->output_format & DIFF_FORMAT_NO_OUTPUT)
3236                 count++;
3237         if (count > 1)
3238                 die("--name-only, --name-status, --check and -s are mutually exclusive");
3239
3240         /*
3241          * Most of the time we can say "there are changes"
3242          * only by checking if there are changed paths, but
3243          * --ignore-whitespace* options force us to look
3244          * inside contents.
3245          */
3246
3247         if (DIFF_XDL_TST(options, IGNORE_WHITESPACE) ||
3248             DIFF_XDL_TST(options, IGNORE_WHITESPACE_CHANGE) ||
3249             DIFF_XDL_TST(options, IGNORE_WHITESPACE_AT_EOL))
3250                 DIFF_OPT_SET(options, DIFF_FROM_CONTENTS);
3251         else
3252                 DIFF_OPT_CLR(options, DIFF_FROM_CONTENTS);
3253
3254         if (DIFF_OPT_TST(options, FIND_COPIES_HARDER))
3255                 options->detect_rename = DIFF_DETECT_COPY;
3256
3257         if (!DIFF_OPT_TST(options, RELATIVE_NAME))
3258                 options->prefix = NULL;
3259         if (options->prefix)
3260                 options->prefix_length = strlen(options->prefix);
3261         else
3262                 options->prefix_length = 0;
3263
3264         if (options->output_format & (DIFF_FORMAT_NAME |
3265                                       DIFF_FORMAT_NAME_STATUS |
3266                                       DIFF_FORMAT_CHECKDIFF |
3267                                       DIFF_FORMAT_NO_OUTPUT))
3268                 options->output_format &= ~(DIFF_FORMAT_RAW |
3269                                             DIFF_FORMAT_NUMSTAT |
3270                                             DIFF_FORMAT_DIFFSTAT |
3271                                             DIFF_FORMAT_SHORTSTAT |
3272                                             DIFF_FORMAT_DIRSTAT |
3273                                             DIFF_FORMAT_SUMMARY |
3274                                             DIFF_FORMAT_PATCH);
3275
3276         /*
3277          * These cases always need recursive; we do not drop caller-supplied
3278          * recursive bits for other formats here.
3279          */
3280         if (options->output_format & (DIFF_FORMAT_PATCH |
3281                                       DIFF_FORMAT_NUMSTAT |
3282                                       DIFF_FORMAT_DIFFSTAT |
3283                                       DIFF_FORMAT_SHORTSTAT |
3284                                       DIFF_FORMAT_DIRSTAT |
3285                                       DIFF_FORMAT_SUMMARY |
3286                                       DIFF_FORMAT_CHECKDIFF))
3287                 DIFF_OPT_SET(options, RECURSIVE);
3288         /*
3289          * Also pickaxe would not work very well if you do not say recursive
3290          */
3291         if (options->pickaxe)
3292                 DIFF_OPT_SET(options, RECURSIVE);
3293         /*
3294          * When patches are generated, submodules diffed against the work tree
3295          * must be checked for dirtiness too so it can be shown in the output
3296          */
3297         if (options->output_format & DIFF_FORMAT_PATCH)
3298                 DIFF_OPT_SET(options, DIRTY_SUBMODULES);
3299
3300         if (options->detect_rename && options->rename_limit < 0)
3301                 options->rename_limit = diff_rename_limit_default;
3302         if (options->setup & DIFF_SETUP_USE_CACHE) {
3303                 if (!active_cache)
3304                         /* read-cache does not die even when it fails
3305                          * so it is safe for us to do this here.  Also
3306                          * it does not smudge active_cache or active_nr
3307                          * when it fails, so we do not have to worry about
3308                          * cleaning it up ourselves either.
3309                          */
3310                         read_cache();
3311         }
3312         if (options->abbrev <= 0 || 40 < options->abbrev)
3313                 options->abbrev = 40; /* full */
3314
3315         /*
3316          * It does not make sense to show the first hit we happened
3317          * to have found.  It does not make sense not to return with
3318          * exit code in such a case either.
3319          */
3320         if (DIFF_OPT_TST(options, QUICK)) {
3321                 options->output_format = DIFF_FORMAT_NO_OUTPUT;
3322                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
3323         }
3324 }
3325
3326 static int opt_arg(const char *arg, int arg_short, const char *arg_long, int *val)
3327 {
3328         char c, *eq;
3329         int len;
3330
3331         if (*arg != '-')
3332                 return 0;
3333         c = *++arg;
3334         if (!c)
3335                 return 0;
3336         if (c == arg_short) {
3337                 c = *++arg;
3338                 if (!c)
3339                         return 1;
3340                 if (val && isdigit(c)) {
3341                         char *end;
3342                         int n = strtoul(arg, &end, 10);
3343                         if (*end)
3344                                 return 0;
3345                         *val = n;
3346                         return 1;
3347                 }
3348                 return 0;
3349         }
3350         if (c != '-')
3351                 return 0;
3352         arg++;
3353         eq = strchr(arg, '=');
3354         if (eq)
3355                 len = eq - arg;
3356         else
3357                 len = strlen(arg);
3358         if (!len || strncmp(arg, arg_long, len))
3359                 return 0;
3360         if (eq) {
3361                 int n;
3362                 char *end;
3363                 if (!isdigit(*++eq))
3364                         return 0;
3365                 n = strtoul(eq, &end, 10);
3366                 if (*end)
3367                         return 0;
3368                 *val = n;
3369         }
3370         return 1;
3371 }
3372
3373 static int diff_scoreopt_parse(const char *opt);
3374
3375 static inline int short_opt(char opt, const char **argv,
3376                             const char **optarg)
3377 {
3378         const char *arg = argv[0];
3379         if (arg[0] != '-' || arg[1] != opt)
3380                 return 0;
3381         if (arg[2] != '\0') {
3382                 *optarg = arg + 2;
3383                 return 1;
3384         }
3385         if (!argv[1])
3386                 die("Option '%c' requires a value", opt);
3387         *optarg = argv[1];
3388         return 2;
3389 }
3390
3391 int parse_long_opt(const char *opt, const char **argv,
3392                    const char **optarg)
3393 {
3394         const char *arg = argv[0];
3395         if (arg[0] != '-' || arg[1] != '-')
3396                 return 0;
3397         arg += strlen("--");
3398         if (prefixcmp(arg, opt))
3399                 return 0;
3400         arg += strlen(opt);
3401         if (*arg == '=') { /* sticked form: --option=value */
3402                 *optarg = arg + 1;
3403                 return 1;
3404         }
3405         if (*arg != '\0')
3406                 return 0;
3407         /* separate form: --option value */
3408         if (!argv[1])
3409                 die("Option '--%s' requires a value", opt);
3410         *optarg = argv[1];
3411         return 2;
3412 }
3413
3414 static int stat_opt(struct diff_options *options, const char **av)
3415 {
3416         const char *arg = av[0];
3417         char *end;
3418         int width = options->stat_width;
3419         int name_width = options->stat_name_width;
3420         int graph_width = options->stat_graph_width;
3421         int count = options->stat_count;
3422         int argcount = 1;
3423
3424         arg += strlen("--stat");
3425         end = (char *)arg;
3426
3427         switch (*arg) {
3428         case '-':
3429                 if (!prefixcmp(arg, "-width")) {
3430                         arg += strlen("-width");
3431                         if (*arg == '=')
3432                                 width = strtoul(arg + 1, &end, 10);
3433                         else if (!*arg && !av[1])
3434                                 die("Option '--stat-width' requires a value");
3435                         else if (!*arg) {
3436                                 width = strtoul(av[1], &end, 10);
3437                                 argcount = 2;
3438                         }
3439                 } else if (!prefixcmp(arg, "-name-width")) {
3440                         arg += strlen("-name-width");
3441                         if (*arg == '=')
3442                                 name_width = strtoul(arg + 1, &end, 10);
3443                         else if (!*arg && !av[1])
3444                                 die("Option '--stat-name-width' requires a value");
3445                         else if (!*arg) {
3446                                 name_width = strtoul(av[1], &end, 10);
3447                                 argcount = 2;
3448                         }
3449                 } else if (!prefixcmp(arg, "-graph-width")) {
3450                         arg += strlen("-graph-width");
3451                         if (*arg == '=')
3452                                 graph_width = strtoul(arg + 1, &end, 10);
3453                         else if (!*arg && !av[1])
3454                                 die("Option '--stat-graph-width' requires a value");
3455                         else if (!*arg) {
3456                                 graph_width = strtoul(av[1], &end, 10);
3457                                 argcount = 2;
3458                         }
3459                 } else if (!prefixcmp(arg, "-count")) {
3460                         arg += strlen("-count");
3461                         if (*arg == '=')
3462                                 count = strtoul(arg + 1, &end, 10);
3463                         else if (!*arg && !av[1])
3464                                 die("Option '--stat-count' requires a value");
3465                         else if (!*arg) {
3466                                 count = strtoul(av[1], &end, 10);
3467                                 argcount = 2;
3468                         }
3469                 }
3470                 break;
3471         case '=':
3472                 width = strtoul(arg+1, &end, 10);
3473                 if (*end == ',')
3474                         name_width = strtoul(end+1, &end, 10);
3475                 if (*end == ',')
3476                         count = strtoul(end+1, &end, 10);
3477         }
3478
3479         /* Important! This checks all the error cases! */
3480         if (*end)
3481                 return 0;
3482         options->output_format |= DIFF_FORMAT_DIFFSTAT;
3483         options->stat_name_width = name_width;
3484         options->stat_graph_width = graph_width;
3485         options->stat_width = width;
3486         options->stat_count = count;
3487         return argcount;
3488 }
3489
3490 static int parse_dirstat_opt(struct diff_options *options, const char *params)
3491 {
3492         struct strbuf errmsg = STRBUF_INIT;
3493         if (parse_dirstat_params(options, params, &errmsg))
3494                 die(_("Failed to parse --dirstat/-X option parameter:\n%s"),
3495                     errmsg.buf);
3496         strbuf_release(&errmsg);
3497         /*
3498          * The caller knows a dirstat-related option is given from the command
3499          * line; allow it to say "return this_function();"
3500          */
3501         options->output_format |= DIFF_FORMAT_DIRSTAT;
3502         return 1;
3503 }
3504
3505 static int parse_submodule_opt(struct diff_options *options, const char *value)
3506 {
3507         if (parse_submodule_params(options, value))
3508                 die(_("Failed to parse --submodule option parameter: '%s'"),
3509                         value);
3510         return 1;
3511 }
3512
3513 int diff_opt_parse(struct diff_options *options, const char **av, int ac)
3514 {
3515         const char *arg = av[0];
3516         const char *optarg;
3517         int argcount;
3518
3519         /* Output format options */
3520         if (!strcmp(arg, "-p") || !strcmp(arg, "-u") || !strcmp(arg, "--patch"))
3521                 options->output_format |= DIFF_FORMAT_PATCH;
3522         else if (opt_arg(arg, 'U', "unified", &options->context))
3523                 options->output_format |= DIFF_FORMAT_PATCH;
3524         else if (!strcmp(arg, "--raw"))
3525                 options->output_format |= DIFF_FORMAT_RAW;
3526         else if (!strcmp(arg, "--patch-with-raw"))
3527                 options->output_format |= DIFF_FORMAT_PATCH | DIFF_FORMAT_RAW;
3528         else if (!strcmp(arg, "--numstat"))
3529                 options->output_format |= DIFF_FORMAT_NUMSTAT;
3530         else if (!strcmp(arg, "--shortstat"))
3531                 options->output_format |= DIFF_FORMAT_SHORTSTAT;
3532         else if (!strcmp(arg, "-X") || !strcmp(arg, "--dirstat"))
3533                 return parse_dirstat_opt(options, "");
3534         else if (!prefixcmp(arg, "-X"))
3535                 return parse_dirstat_opt(options, arg + 2);
3536         else if (!prefixcmp(arg, "--dirstat="))
3537                 return parse_dirstat_opt(options, arg + 10);
3538         else if (!strcmp(arg, "--cumulative"))
3539                 return parse_dirstat_opt(options, "cumulative");
3540         else if (!strcmp(arg, "--dirstat-by-file"))
3541                 return parse_dirstat_opt(options, "files");
3542         else if (!prefixcmp(arg, "--dirstat-by-file=")) {
3543                 parse_dirstat_opt(options, "files");
3544                 return parse_dirstat_opt(options, arg + 18);
3545         }
3546         else if (!strcmp(arg, "--check"))
3547                 options->output_format |= DIFF_FORMAT_CHECKDIFF;
3548         else if (!strcmp(arg, "--summary"))
3549                 options->output_format |= DIFF_FORMAT_SUMMARY;
3550         else if (!strcmp(arg, "--patch-with-stat"))
3551                 options->output_format |= DIFF_FORMAT_PATCH | DIFF_FORMAT_DIFFSTAT;
3552         else if (!strcmp(arg, "--name-only"))
3553                 options->output_format |= DIFF_FORMAT_NAME;
3554         else if (!strcmp(arg, "--name-status"))
3555                 options->output_format |= DIFF_FORMAT_NAME_STATUS;
3556         else if (!strcmp(arg, "-s"))
3557                 options->output_format |= DIFF_FORMAT_NO_OUTPUT;
3558         else if (!prefixcmp(arg, "--stat"))
3559                 /* --stat, --stat-width, --stat-name-width, or --stat-count */
3560                 return stat_opt(options, av);
3561
3562         /* renames options */
3563         else if (!prefixcmp(arg, "-B") || !prefixcmp(arg, "--break-rewrites=") ||
3564                  !strcmp(arg, "--break-rewrites")) {
3565                 if ((options->break_opt = diff_scoreopt_parse(arg)) == -1)
3566                         return error("invalid argument to -B: %s", arg+2);
3567         }
3568         else if (!prefixcmp(arg, "-M") || !prefixcmp(arg, "--find-renames=") ||
3569                  !strcmp(arg, "--find-renames")) {
3570                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
3571                         return error("invalid argument to -M: %s", arg+2);
3572                 options->detect_rename = DIFF_DETECT_RENAME;
3573         }
3574         else if (!strcmp(arg, "-D") || !strcmp(arg, "--irreversible-delete")) {
3575                 options->irreversible_delete = 1;
3576         }
3577         else if (!prefixcmp(arg, "-C") || !prefixcmp(arg, "--find-copies=") ||
3578                  !strcmp(arg, "--find-copies")) {
3579                 if (options->detect_rename == DIFF_DETECT_COPY)
3580                         DIFF_OPT_SET(options, FIND_COPIES_HARDER);
3581                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
3582                         return error("invalid argument to -C: %s", arg+2);
3583                 options->detect_rename = DIFF_DETECT_COPY;
3584         }
3585         else if (!strcmp(arg, "--no-renames"))
3586                 options->detect_rename = 0;
3587         else if (!strcmp(arg, "--rename-empty"))
3588                 DIFF_OPT_SET(options, RENAME_EMPTY);
3589         else if (!strcmp(arg, "--no-rename-empty"))
3590                 DIFF_OPT_CLR(options, RENAME_EMPTY);
3591         else if (!strcmp(arg, "--relative"))
3592                 DIFF_OPT_SET(options, RELATIVE_NAME);
3593         else if (!prefixcmp(arg, "--relative=")) {
3594                 DIFF_OPT_SET(options, RELATIVE_NAME);
3595                 options->prefix = arg + 11;
3596         }
3597
3598         /* xdiff options */
3599         else if (!strcmp(arg, "--minimal"))
3600                 DIFF_XDL_SET(options, NEED_MINIMAL);
3601         else if (!strcmp(arg, "--no-minimal"))
3602                 DIFF_XDL_CLR(options, NEED_MINIMAL);
3603         else if (!strcmp(arg, "-w") || !strcmp(arg, "--ignore-all-space"))
3604                 DIFF_XDL_SET(options, IGNORE_WHITESPACE);
3605         else if (!strcmp(arg, "-b") || !strcmp(arg, "--ignore-space-change"))
3606                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_CHANGE);
3607         else if (!strcmp(arg, "--ignore-space-at-eol"))
3608                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_AT_EOL);
3609         else if (!strcmp(arg, "--patience"))
3610                 options->xdl_opts = DIFF_WITH_ALG(options, PATIENCE_DIFF);
3611         else if (!strcmp(arg, "--histogram"))
3612                 options->xdl_opts = DIFF_WITH_ALG(options, HISTOGRAM_DIFF);
3613
3614         /* flags options */
3615         else if (!strcmp(arg, "--binary")) {
3616                 options->output_format |= DIFF_FORMAT_PATCH;
3617                 DIFF_OPT_SET(options, BINARY);
3618         }
3619         else if (!strcmp(arg, "--full-index"))
3620                 DIFF_OPT_SET(options, FULL_INDEX);
3621         else if (!strcmp(arg, "-a") || !strcmp(arg, "--text"))
3622                 DIFF_OPT_SET(options, TEXT);
3623         else if (!strcmp(arg, "-R"))
3624                 DIFF_OPT_SET(options, REVERSE_DIFF);
3625         else if (!strcmp(arg, "--find-copies-harder"))
3626                 DIFF_OPT_SET(options, FIND_COPIES_HARDER);
3627         else if (!strcmp(arg, "--follow"))
3628                 DIFF_OPT_SET(options, FOLLOW_RENAMES);
3629         else if (!strcmp(arg, "--color"))
3630                 options->use_color = 1;
3631         else if (!prefixcmp(arg, "--color=")) {
3632                 int value = git_config_colorbool(NULL, arg+8);
3633                 if (value < 0)
3634                         return error("option `color' expects \"always\", \"auto\", or \"never\"");
3635                 options->use_color = value;
3636         }
3637         else if (!strcmp(arg, "--no-color"))
3638                 options->use_color = 0;
3639         else if (!strcmp(arg, "--color-words")) {
3640                 options->use_color = 1;
3641                 options->word_diff = DIFF_WORDS_COLOR;
3642         }
3643         else if (!prefixcmp(arg, "--color-words=")) {
3644                 options->use_color = 1;
3645                 options->word_diff = DIFF_WORDS_COLOR;
3646                 options->word_regex = arg + 14;
3647         }
3648         else if (!strcmp(arg, "--word-diff")) {
3649                 if (options->word_diff == DIFF_WORDS_NONE)
3650                         options->word_diff = DIFF_WORDS_PLAIN;
3651         }
3652         else if (!prefixcmp(arg, "--word-diff=")) {
3653                 const char *type = arg + 12;
3654                 if (!strcmp(type, "plain"))
3655                         options->word_diff = DIFF_WORDS_PLAIN;
3656                 else if (!strcmp(type, "color")) {
3657                         options->use_color = 1;
3658                         options->word_diff = DIFF_WORDS_COLOR;
3659                 }
3660                 else if (!strcmp(type, "porcelain"))
3661                         options->word_diff = DIFF_WORDS_PORCELAIN;
3662                 else if (!strcmp(type, "none"))
3663                         options->word_diff = DIFF_WORDS_NONE;
3664                 else
3665                         die("bad --word-diff argument: %s", type);
3666         }
3667         else if ((argcount = parse_long_opt("word-diff-regex", av, &optarg))) {
3668                 if (options->word_diff == DIFF_WORDS_NONE)
3669                         options->word_diff = DIFF_WORDS_PLAIN;
3670                 options->word_regex = optarg;
3671                 return argcount;
3672         }
3673         else if (!strcmp(arg, "--exit-code"))
3674                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
3675         else if (!strcmp(arg, "--quiet"))
3676                 DIFF_OPT_SET(options, QUICK);
3677         else if (!strcmp(arg, "--ext-diff"))
3678                 DIFF_OPT_SET(options, ALLOW_EXTERNAL);
3679         else if (!strcmp(arg, "--no-ext-diff"))
3680                 DIFF_OPT_CLR(options, ALLOW_EXTERNAL);
3681         else if (!strcmp(arg, "--textconv"))
3682                 DIFF_OPT_SET(options, ALLOW_TEXTCONV);
3683         else if (!strcmp(arg, "--no-textconv"))
3684                 DIFF_OPT_CLR(options, ALLOW_TEXTCONV);
3685         else if (!strcmp(arg, "--ignore-submodules")) {
3686                 DIFF_OPT_SET(options, OVERRIDE_SUBMODULE_CONFIG);
3687                 handle_ignore_submodules_arg(options, "all");
3688         } else if (!prefixcmp(arg, "--ignore-submodules=")) {
3689                 DIFF_OPT_SET(options, OVERRIDE_SUBMODULE_CONFIG);
3690                 handle_ignore_submodules_arg(options, arg + 20);
3691         } else if (!strcmp(arg, "--submodule"))
3692                 DIFF_OPT_SET(options, SUBMODULE_LOG);
3693         else if (!prefixcmp(arg, "--submodule="))
3694                 return parse_submodule_opt(options, arg + 12);
3695
3696         /* misc options */
3697         else if (!strcmp(arg, "-z"))
3698                 options->line_termination = 0;
3699         else if ((argcount = short_opt('l', av, &optarg))) {
3700                 options->rename_limit = strtoul(optarg, NULL, 10);
3701                 return argcount;
3702         }
3703         else if ((argcount = short_opt('S', av, &optarg))) {
3704                 options->pickaxe = optarg;
3705                 options->pickaxe_opts |= DIFF_PICKAXE_KIND_S;
3706                 return argcount;
3707         } else if ((argcount = short_opt('G', av, &optarg))) {
3708                 options->pickaxe = optarg;
3709                 options->pickaxe_opts |= DIFF_PICKAXE_KIND_G;
3710                 return argcount;
3711         }
3712         else if (!strcmp(arg, "--pickaxe-all"))
3713                 options->pickaxe_opts |= DIFF_PICKAXE_ALL;
3714         else if (!strcmp(arg, "--pickaxe-regex"))
3715                 options->pickaxe_opts |= DIFF_PICKAXE_REGEX;
3716         else if ((argcount = short_opt('O', av, &optarg))) {
3717                 options->orderfile = optarg;
3718                 return argcount;
3719         }
3720         else if ((argcount = parse_long_opt("diff-filter", av, &optarg))) {
3721                 options->filter = optarg;
3722                 return argcount;
3723         }
3724         else if (!strcmp(arg, "--abbrev"))
3725                 options->abbrev = DEFAULT_ABBREV;
3726         else if (!prefixcmp(arg, "--abbrev=")) {
3727                 options->abbrev = strtoul(arg + 9, NULL, 10);
3728                 if (options->abbrev < MINIMUM_ABBREV)
3729                         options->abbrev = MINIMUM_ABBREV;
3730                 else if (40 < options->abbrev)
3731                         options->abbrev = 40;
3732         }
3733         else if ((argcount = parse_long_opt("src-prefix", av, &optarg))) {
3734                 options->a_prefix = optarg;
3735                 return argcount;
3736         }
3737         else if ((argcount = parse_long_opt("dst-prefix", av, &optarg))) {
3738                 options->b_prefix = optarg;
3739                 return argcount;
3740         }
3741         else if (!strcmp(arg, "--no-prefix"))
3742                 options->a_prefix = options->b_prefix = "";
3743         else if (opt_arg(arg, '\0', "inter-hunk-context",
3744                          &options->interhunkcontext))
3745                 ;
3746         else if (!strcmp(arg, "-W"))
3747                 DIFF_OPT_SET(options, FUNCCONTEXT);
3748         else if (!strcmp(arg, "--function-context"))
3749                 DIFF_OPT_SET(options, FUNCCONTEXT);
3750         else if (!strcmp(arg, "--no-function-context"))
3751                 DIFF_OPT_CLR(options, FUNCCONTEXT);
3752         else if ((argcount = parse_long_opt("output", av, &optarg))) {
3753                 options->file = fopen(optarg, "w");
3754                 if (!options->file)
3755                         die_errno("Could not open '%s'", optarg);
3756                 options->close_file = 1;
3757                 return argcount;
3758         } else
3759                 return 0;
3760         return 1;
3761 }
3762
3763 int parse_rename_score(const char **cp_p)
3764 {
3765         unsigned long num, scale;
3766         int ch, dot;
3767         const char *cp = *cp_p;
3768
3769         num = 0;
3770         scale = 1;
3771         dot = 0;
3772         for (;;) {
3773                 ch = *cp;
3774                 if ( !dot && ch == '.' ) {
3775                         scale = 1;
3776                         dot = 1;
3777                 } else if ( ch == '%' ) {
3778                         scale = dot ? scale*100 : 100;
3779                         cp++;   /* % is always at the end */
3780                         break;
3781                 } else if ( ch >= '0' && ch <= '9' ) {
3782                         if ( scale < 100000 ) {
3783                                 scale *= 10;
3784                                 num = (num*10) + (ch-'0');
3785                         }
3786                 } else {
3787                         break;
3788                 }
3789                 cp++;
3790         }
3791         *cp_p = cp;
3792
3793         /* user says num divided by scale and we say internally that
3794          * is MAX_SCORE * num / scale.
3795          */
3796         return (int)((num >= scale) ? MAX_SCORE : (MAX_SCORE * num / scale));
3797 }
3798
3799 static int diff_scoreopt_parse(const char *opt)
3800 {
3801         int opt1, opt2, cmd;
3802
3803         if (*opt++ != '-')
3804                 return -1;
3805         cmd = *opt++;
3806         if (cmd == '-') {
3807                 /* convert the long-form arguments into short-form versions */
3808                 if (!prefixcmp(opt, "break-rewrites")) {
3809                         opt += strlen("break-rewrites");
3810                         if (*opt == 0 || *opt++ == '=')
3811                                 cmd = 'B';
3812                 } else if (!prefixcmp(opt, "find-copies")) {
3813                         opt += strlen("find-copies");
3814                         if (*opt == 0 || *opt++ == '=')
3815                                 cmd = 'C';
3816                 } else if (!prefixcmp(opt, "find-renames")) {
3817                         opt += strlen("find-renames");
3818                         if (*opt == 0 || *opt++ == '=')
3819                                 cmd = 'M';
3820                 }
3821         }
3822         if (cmd != 'M' && cmd != 'C' && cmd != 'B')
3823                 return -1; /* that is not a -M, -C nor -B option */
3824
3825         opt1 = parse_rename_score(&opt);
3826         if (cmd != 'B')
3827                 opt2 = 0;
3828         else {
3829                 if (*opt == 0)
3830                         opt2 = 0;
3831                 else if (*opt != '/')
3832                         return -1; /* we expect -B80/99 or -B80 */
3833                 else {
3834                         opt++;
3835                         opt2 = parse_rename_score(&opt);
3836                 }
3837         }
3838         if (*opt != 0)
3839                 return -1;
3840         return opt1 | (opt2 << 16);
3841 }
3842
3843 struct diff_queue_struct diff_queued_diff;
3844
3845 void diff_q(struct diff_queue_struct *queue, struct diff_filepair *dp)
3846 {
3847         if (queue->alloc <= queue->nr) {
3848                 queue->alloc = alloc_nr(queue->alloc);
3849                 queue->queue = xrealloc(queue->queue,
3850                                         sizeof(dp) * queue->alloc);
3851         }
3852         queue->queue[queue->nr++] = dp;
3853 }
3854
3855 struct diff_filepair *diff_queue(struct diff_queue_struct *queue,
3856                                  struct diff_filespec *one,
3857                                  struct diff_filespec *two)
3858 {
3859         struct diff_filepair *dp = xcalloc(1, sizeof(*dp));
3860         dp->one = one;
3861         dp->two = two;
3862         if (queue)
3863                 diff_q(queue, dp);
3864         return dp;
3865 }
3866
3867 void diff_free_filepair(struct diff_filepair *p)
3868 {
3869         free_filespec(p->one);
3870         free_filespec(p->two);
3871         free(p);
3872 }
3873
3874 /* This is different from find_unique_abbrev() in that
3875  * it stuffs the result with dots for alignment.
3876  */
3877 const char *diff_unique_abbrev(const unsigned char *sha1, int len)
3878 {
3879         int abblen;
3880         const char *abbrev;
3881         if (len == 40)
3882                 return sha1_to_hex(sha1);
3883
3884         abbrev = find_unique_abbrev(sha1, len);
3885         abblen = strlen(abbrev);
3886         if (abblen < 37) {
3887                 static char hex[41];
3888                 if (len < abblen && abblen <= len + 2)
3889                         sprintf(hex, "%s%.*s", abbrev, len+3-abblen, "..");
3890                 else
3891                         sprintf(hex, "%s...", abbrev);
3892                 return hex;
3893         }
3894         return sha1_to_hex(sha1);
3895 }
3896
3897 static void diff_flush_raw(struct diff_filepair *p, struct diff_options *opt)
3898 {
3899         int line_termination = opt->line_termination;
3900         int inter_name_termination = line_termination ? '\t' : '\0';
3901         if (opt->output_prefix) {
3902                 struct strbuf *msg = NULL;
3903                 msg = opt->output_prefix(opt, opt->output_prefix_data);
3904                 fprintf(opt->file, "%s", msg->buf);
3905         }
3906
3907         if (!(opt->output_format & DIFF_FORMAT_NAME_STATUS)) {
3908                 fprintf(opt->file, ":%06o %06o %s ", p->one->mode, p->two->mode,
3909                         diff_unique_abbrev(p->one->sha1, opt->abbrev));
3910                 fprintf(opt->file, "%s ", diff_unique_abbrev(p->two->sha1, opt->abbrev));
3911         }
3912         if (p->score) {
3913                 fprintf(opt->file, "%c%03d%c", p->status, similarity_index(p),
3914                         inter_name_termination);
3915         } else {
3916                 fprintf(opt->file, "%c%c", p->status, inter_name_termination);
3917         }
3918
3919         if (p->status == DIFF_STATUS_COPIED ||
3920             p->status == DIFF_STATUS_RENAMED) {
3921                 const char *name_a, *name_b;
3922                 name_a = p->one->path;
3923                 name_b = p->two->path;
3924                 strip_prefix(opt->prefix_length, &name_a, &name_b);
3925                 write_name_quoted(name_a, opt->file, inter_name_termination);
3926                 write_name_quoted(name_b, opt->file, line_termination);
3927         } else {
3928                 const char *name_a, *name_b;
3929                 name_a = p->one->mode ? p->one->path : p->two->path;
3930                 name_b = NULL;
3931                 strip_prefix(opt->prefix_length, &name_a, &name_b);
3932                 write_name_quoted(name_a, opt->file, line_termination);
3933         }
3934 }
3935
3936 int diff_unmodified_pair(struct diff_filepair *p)
3937 {
3938         /* This function is written stricter than necessary to support
3939          * the currently implemented transformers, but the idea is to
3940          * let transformers to produce diff_filepairs any way they want,
3941          * and filter and clean them up here before producing the output.
3942          */
3943         struct diff_filespec *one = p->one, *two = p->two;
3944
3945         if (DIFF_PAIR_UNMERGED(p))
3946                 return 0; /* unmerged is interesting */
3947
3948         /* deletion, addition, mode or type change
3949          * and rename are all interesting.
3950          */
3951         if (DIFF_FILE_VALID(one) != DIFF_FILE_VALID(two) ||
3952             DIFF_PAIR_MODE_CHANGED(p) ||
3953             strcmp(one->path, two->path))
3954                 return 0;
3955
3956         /* both are valid and point at the same path.  that is, we are
3957          * dealing with a change.
3958          */
3959         if (one->sha1_valid && two->sha1_valid &&
3960             !hashcmp(one->sha1, two->sha1) &&
3961             !one->dirty_submodule && !two->dirty_submodule)
3962                 return 1; /* no change */
3963         if (!one->sha1_valid && !two->sha1_valid)
3964                 return 1; /* both look at the same file on the filesystem. */
3965         return 0;
3966 }
3967
3968 static void diff_flush_patch(struct diff_filepair *p, struct diff_options *o)
3969 {
3970         if (diff_unmodified_pair(p))
3971                 return;
3972
3973         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
3974             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
3975                 return; /* no tree diffs in patch format */
3976
3977         run_diff(p, o);
3978 }
3979
3980 static void diff_flush_stat(struct diff_filepair *p, struct diff_options *o,
3981                             struct diffstat_t *diffstat)
3982 {
3983         if (diff_unmodified_pair(p))
3984                 return;
3985
3986         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
3987             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
3988                 return; /* no useful stat for tree diffs */
3989
3990         run_diffstat(p, o, diffstat);
3991 }
3992
3993 static void diff_flush_checkdiff(struct diff_filepair *p,
3994                 struct diff_options *o)
3995 {
3996         if (diff_unmodified_pair(p))
3997                 return;
3998
3999         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
4000             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
4001                 return; /* nothing to check in tree diffs */
4002
4003         run_checkdiff(p, o);
4004 }
4005
4006 int diff_queue_is_empty(void)
4007 {
4008         struct diff_queue_struct *q = &diff_queued_diff;
4009         int i;
4010         for (i = 0; i < q->nr; i++)
4011                 if (!diff_unmodified_pair(q->queue[i]))
4012                         return 0;
4013         return 1;
4014 }
4015
4016 #if DIFF_DEBUG
4017 void diff_debug_filespec(struct diff_filespec *s, int x, const char *one)
4018 {
4019         fprintf(stderr, "queue[%d] %s (%s) %s %06o %s\n",
4020                 x, one ? one : "",
4021                 s->path,
4022                 DIFF_FILE_VALID(s) ? "valid" : "invalid",
4023                 s->mode,
4024                 s->sha1_valid ? sha1_to_hex(s->sha1) : "");
4025         fprintf(stderr, "queue[%d] %s size %lu flags %d\n",
4026                 x, one ? one : "",
4027                 s->size, s->xfrm_flags);
4028 }
4029
4030 void diff_debug_filepair(const struct diff_filepair *p, int i)
4031 {
4032         diff_debug_filespec(p->one, i, "one");
4033         diff_debug_filespec(p->two, i, "two");
4034         fprintf(stderr, "score %d, status %c rename_used %d broken %d\n",
4035                 p->score, p->status ? p->status : '?',
4036                 p->one->rename_used, p->broken_pair);
4037 }
4038
4039 void diff_debug_queue(const char *msg, struct diff_queue_struct *q)
4040 {
4041         int i;
4042         if (msg)
4043                 fprintf(stderr, "%s\n", msg);
4044         fprintf(stderr, "q->nr = %d\n", q->nr);
4045         for (i = 0; i < q->nr; i++) {
4046                 struct diff_filepair *p = q->queue[i];
4047                 diff_debug_filepair(p, i);
4048         }
4049 }
4050 #endif
4051
4052 static void diff_resolve_rename_copy(void)
4053 {
4054         int i;
4055         struct diff_filepair *p;
4056         struct diff_queue_struct *q = &diff_queued_diff;
4057
4058         diff_debug_queue("resolve-rename-copy", q);
4059
4060         for (i = 0; i < q->nr; i++) {
4061                 p = q->queue[i];
4062                 p->status = 0; /* undecided */
4063                 if (DIFF_PAIR_UNMERGED(p))
4064                         p->status = DIFF_STATUS_UNMERGED;
4065                 else if (!DIFF_FILE_VALID(p->one))
4066                         p->status = DIFF_STATUS_ADDED;
4067                 else if (!DIFF_FILE_VALID(p->two))
4068                         p->status = DIFF_STATUS_DELETED;
4069                 else if (DIFF_PAIR_TYPE_CHANGED(p))
4070                         p->status = DIFF_STATUS_TYPE_CHANGED;
4071
4072                 /* from this point on, we are dealing with a pair
4073                  * whose both sides are valid and of the same type, i.e.
4074                  * either in-place edit or rename/copy edit.
4075                  */
4076                 else if (DIFF_PAIR_RENAME(p)) {
4077                         /*
4078                          * A rename might have re-connected a broken
4079                          * pair up, causing the pathnames to be the
4080                          * same again. If so, that's not a rename at
4081                          * all, just a modification..
4082                          *
4083                          * Otherwise, see if this source was used for
4084                          * multiple renames, in which case we decrement
4085                          * the count, and call it a copy.
4086                          */
4087                         if (!strcmp(p->one->path, p->two->path))
4088                                 p->status = DIFF_STATUS_MODIFIED;
4089                         else if (--p->one->rename_used > 0)
4090                                 p->status = DIFF_STATUS_COPIED;
4091                         else
4092                                 p->status = DIFF_STATUS_RENAMED;
4093                 }
4094                 else if (hashcmp(p->one->sha1, p->two->sha1) ||
4095                          p->one->mode != p->two->mode ||
4096                          p->one->dirty_submodule ||
4097                          p->two->dirty_submodule ||
4098                          is_null_sha1(p->one->sha1))
4099                         p->status = DIFF_STATUS_MODIFIED;
4100                 else {
4101                         /* This is a "no-change" entry and should not
4102                          * happen anymore, but prepare for broken callers.
4103                          */
4104                         error("feeding unmodified %s to diffcore",
4105                               p->one->path);
4106                         p->status = DIFF_STATUS_UNKNOWN;
4107                 }
4108         }
4109         diff_debug_queue("resolve-rename-copy done", q);
4110 }
4111
4112 static int check_pair_status(struct diff_filepair *p)
4113 {
4114         switch (p->status) {
4115         case DIFF_STATUS_UNKNOWN:
4116                 return 0;
4117         case 0:
4118                 die("internal error in diff-resolve-rename-copy");
4119         default:
4120                 return 1;
4121         }
4122 }
4123
4124 static void flush_one_pair(struct diff_filepair *p, struct diff_options *opt)
4125 {
4126         int fmt = opt->output_format;
4127
4128         if (fmt & DIFF_FORMAT_CHECKDIFF)
4129                 diff_flush_checkdiff(p, opt);
4130         else if (fmt & (DIFF_FORMAT_RAW | DIFF_FORMAT_NAME_STATUS))
4131                 diff_flush_raw(p, opt);
4132         else if (fmt & DIFF_FORMAT_NAME) {
4133                 const char *name_a, *name_b;
4134                 name_a = p->two->path;
4135                 name_b = NULL;
4136                 strip_prefix(opt->prefix_length, &name_a, &name_b);
4137                 write_name_quoted(name_a, opt->file, opt->line_termination);
4138         }
4139 }
4140
4141 static void show_file_mode_name(FILE *file, const char *newdelete, struct diff_filespec *fs)
4142 {
4143         if (fs->mode)
4144                 fprintf(file, " %s mode %06o ", newdelete, fs->mode);
4145         else
4146                 fprintf(file, " %s ", newdelete);
4147         write_name_quoted(fs->path, file, '\n');
4148 }
4149
4150
4151 static void show_mode_change(FILE *file, struct diff_filepair *p, int show_name,
4152                 const char *line_prefix)
4153 {
4154         if (p->one->mode && p->two->mode && p->one->mode != p->two->mode) {
4155                 fprintf(file, "%s mode change %06o => %06o%c", line_prefix, p->one->mode,
4156                         p->two->mode, show_name ? ' ' : '\n');
4157                 if (show_name) {
4158                         write_name_quoted(p->two->path, file, '\n');
4159                 }
4160         }
4161 }
4162
4163 static void show_rename_copy(FILE *file, const char *renamecopy, struct diff_filepair *p,
4164                         const char *line_prefix)
4165 {
4166         char *names = pprint_rename(p->one->path, p->two->path);
4167
4168         fprintf(file, " %s %s (%d%%)\n", renamecopy, names, similarity_index(p));
4169         free(names);
4170         show_mode_change(file, p, 0, line_prefix);
4171 }
4172
4173 static void diff_summary(struct diff_options *opt, struct diff_filepair *p)
4174 {
4175         FILE *file = opt->file;
4176         char *line_prefix = "";
4177
4178         if (opt->output_prefix) {
4179                 struct strbuf *buf = opt->output_prefix(opt, opt->output_prefix_data);
4180                 line_prefix = buf->buf;
4181         }
4182
4183         switch(p->status) {
4184         case DIFF_STATUS_DELETED:
4185                 fputs(line_prefix, file);
4186                 show_file_mode_name(file, "delete", p->one);
4187                 break;
4188         case DIFF_STATUS_ADDED:
4189                 fputs(line_prefix, file);
4190                 show_file_mode_name(file, "create", p->two);
4191                 break;
4192         case DIFF_STATUS_COPIED:
4193                 fputs(line_prefix, file);
4194                 show_rename_copy(file, "copy", p, line_prefix);
4195                 break;
4196         case DIFF_STATUS_RENAMED:
4197                 fputs(line_prefix, file);
4198                 show_rename_copy(file, "rename", p, line_prefix);
4199                 break;
4200         default:
4201                 if (p->score) {
4202                         fprintf(file, "%s rewrite ", line_prefix);
4203                         write_name_quoted(p->two->path, file, ' ');
4204                         fprintf(file, "(%d%%)\n", similarity_index(p));
4205                 }
4206                 show_mode_change(file, p, !p->score, line_prefix);
4207                 break;
4208         }
4209 }
4210
4211 struct patch_id_t {
4212         git_SHA_CTX *ctx;
4213         int patchlen;
4214 };
4215
4216 static int remove_space(char *line, int len)
4217 {
4218         int i;
4219         char *dst = line;
4220         unsigned char c;
4221
4222         for (i = 0; i < len; i++)
4223                 if (!isspace((c = line[i])))
4224                         *dst++ = c;
4225
4226         return dst - line;
4227 }
4228
4229 static void patch_id_consume(void *priv, char *line, unsigned long len)
4230 {
4231         struct patch_id_t *data = priv;
4232         int new_len;
4233
4234         /* Ignore line numbers when computing the SHA1 of the patch */
4235         if (!prefixcmp(line, "@@ -"))
4236                 return;
4237
4238         new_len = remove_space(line, len);
4239
4240         git_SHA1_Update(data->ctx, line, new_len);
4241         data->patchlen += new_len;
4242 }
4243
4244 /* returns 0 upon success, and writes result into sha1 */
4245 static int diff_get_patch_id(struct diff_options *options, unsigned char *sha1)
4246 {
4247         struct diff_queue_struct *q = &diff_queued_diff;
4248         int i;
4249         git_SHA_CTX ctx;
4250         struct patch_id_t data;
4251         char buffer[PATH_MAX * 4 + 20];
4252
4253         git_SHA1_Init(&ctx);
4254         memset(&data, 0, sizeof(struct patch_id_t));
4255         data.ctx = &ctx;
4256
4257         for (i = 0; i < q->nr; i++) {
4258                 xpparam_t xpp;
4259                 xdemitconf_t xecfg;
4260                 mmfile_t mf1, mf2;
4261                 struct diff_filepair *p = q->queue[i];
4262                 int len1, len2;
4263
4264                 memset(&xpp, 0, sizeof(xpp));
4265                 memset(&xecfg, 0, sizeof(xecfg));
4266                 if (p->status == 0)
4267                         return error("internal diff status error");
4268                 if (p->status == DIFF_STATUS_UNKNOWN)
4269                         continue;
4270                 if (diff_unmodified_pair(p))
4271                         continue;
4272                 if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
4273                     (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
4274                         continue;
4275                 if (DIFF_PAIR_UNMERGED(p))
4276                         continue;
4277
4278                 diff_fill_sha1_info(p->one);
4279                 diff_fill_sha1_info(p->two);
4280                 if (fill_mmfile(&mf1, p->one) < 0 ||
4281                                 fill_mmfile(&mf2, p->two) < 0)
4282                         return error("unable to read files to diff");
4283
4284                 len1 = remove_space(p->one->path, strlen(p->one->path));
4285                 len2 = remove_space(p->two->path, strlen(p->two->path));
4286                 if (p->one->mode == 0)
4287                         len1 = snprintf(buffer, sizeof(buffer),
4288                                         "diff--gita/%.*sb/%.*s"
4289                                         "newfilemode%06o"
4290                                         "---/dev/null"
4291                                         "+++b/%.*s",
4292                                         len1, p->one->path,
4293                                         len2, p->two->path,
4294                                         p->two->mode,
4295                                         len2, p->two->path);
4296                 else if (p->two->mode == 0)
4297                         len1 = snprintf(buffer, sizeof(buffer),
4298                                         "diff--gita/%.*sb/%.*s"
4299                                         "deletedfilemode%06o"
4300                                         "---a/%.*s"
4301                                         "+++/dev/null",
4302                                         len1, p->one->path,
4303                                         len2, p->two->path,
4304                                         p->one->mode,
4305                                         len1, p->one->path);
4306                 else
4307                         len1 = snprintf(buffer, sizeof(buffer),
4308                                         "diff--gita/%.*sb/%.*s"
4309                                         "---a/%.*s"
4310                                         "+++b/%.*s",
4311                                         len1, p->one->path,
4312                                         len2, p->two->path,
4313                                         len1, p->one->path,
4314                                         len2, p->two->path);
4315                 git_SHA1_Update(&ctx, buffer, len1);
4316
4317                 if (diff_filespec_is_binary(p->one) ||
4318                     diff_filespec_is_binary(p->two)) {
4319                         git_SHA1_Update(&ctx, sha1_to_hex(p->one->sha1), 40);
4320                         git_SHA1_Update(&ctx, sha1_to_hex(p->two->sha1), 40);
4321                         continue;
4322                 }
4323
4324                 xpp.flags = 0;
4325                 xecfg.ctxlen = 3;
4326                 xecfg.flags = 0;
4327                 xdi_diff_outf(&mf1, &mf2, patch_id_consume, &data,
4328                               &xpp, &xecfg);
4329         }
4330
4331         git_SHA1_Final(sha1, &ctx);
4332         return 0;
4333 }
4334
4335 int diff_flush_patch_id(struct diff_options *options, unsigned char *sha1)
4336 {
4337         struct diff_queue_struct *q = &diff_queued_diff;
4338         int i;
4339         int result = diff_get_patch_id(options, sha1);
4340
4341         for (i = 0; i < q->nr; i++)
4342                 diff_free_filepair(q->queue[i]);
4343
4344         free(q->queue);
4345         DIFF_QUEUE_CLEAR(q);
4346
4347         return result;
4348 }
4349
4350 static int is_summary_empty(const struct diff_queue_struct *q)
4351 {
4352         int i;
4353
4354         for (i = 0; i < q->nr; i++) {
4355                 const struct diff_filepair *p = q->queue[i];
4356
4357                 switch (p->status) {
4358                 case DIFF_STATUS_DELETED:
4359                 case DIFF_STATUS_ADDED:
4360                 case DIFF_STATUS_COPIED:
4361                 case DIFF_STATUS_RENAMED:
4362                         return 0;
4363                 default:
4364                         if (p->score)
4365                                 return 0;
4366                         if (p->one->mode && p->two->mode &&
4367                             p->one->mode != p->two->mode)
4368                                 return 0;
4369                         break;
4370                 }
4371         }
4372         return 1;
4373 }
4374
4375 static const char rename_limit_warning[] =
4376 "inexact rename detection was skipped due to too many files.";
4377
4378 static const char degrade_cc_to_c_warning[] =
4379 "only found copies from modified paths due to too many files.";
4380
4381 static const char rename_limit_advice[] =
4382 "you may want to set your %s variable to at least "
4383 "%d and retry the command.";
4384
4385 void diff_warn_rename_limit(const char *varname, int needed, int degraded_cc)
4386 {
4387         if (degraded_cc)
4388                 warning(degrade_cc_to_c_warning);
4389         else if (needed)
4390                 warning(rename_limit_warning);
4391         else
4392                 return;
4393         if (0 < needed && needed < 32767)
4394                 warning(rename_limit_advice, varname, needed);
4395 }
4396
4397 void diff_flush(struct diff_options *options)
4398 {
4399         struct diff_queue_struct *q = &diff_queued_diff;
4400         int i, output_format = options->output_format;
4401         int separator = 0;
4402         int dirstat_by_line = 0;
4403
4404         /*
4405          * Order: raw, stat, summary, patch
4406          * or:    name/name-status/checkdiff (other bits clear)
4407          */
4408         if (!q->nr)
4409                 goto free_queue;
4410
4411         if (output_format & (DIFF_FORMAT_RAW |
4412                              DIFF_FORMAT_NAME |
4413                              DIFF_FORMAT_NAME_STATUS |
4414                              DIFF_FORMAT_CHECKDIFF)) {
4415                 for (i = 0; i < q->nr; i++) {
4416                         struct diff_filepair *p = q->queue[i];
4417                         if (check_pair_status(p))
4418                                 flush_one_pair(p, options);
4419                 }
4420                 separator++;
4421         }
4422
4423         if (output_format & DIFF_FORMAT_DIRSTAT && DIFF_OPT_TST(options, DIRSTAT_BY_LINE))
4424                 dirstat_by_line = 1;
4425
4426         if (output_format & (DIFF_FORMAT_DIFFSTAT|DIFF_FORMAT_SHORTSTAT|DIFF_FORMAT_NUMSTAT) ||
4427             dirstat_by_line) {
4428                 struct diffstat_t diffstat;
4429
4430                 memset(&diffstat, 0, sizeof(struct diffstat_t));
4431                 for (i = 0; i < q->nr; i++) {
4432                         struct diff_filepair *p = q->queue[i];
4433                         if (check_pair_status(p))
4434                                 diff_flush_stat(p, options, &diffstat);
4435                 }
4436                 if (output_format & DIFF_FORMAT_NUMSTAT)
4437                         show_numstat(&diffstat, options);
4438                 if (output_format & DIFF_FORMAT_DIFFSTAT)
4439                         show_stats(&diffstat, options);
4440                 if (output_format & DIFF_FORMAT_SHORTSTAT)
4441                         show_shortstats(&diffstat, options);
4442                 if (output_format & DIFF_FORMAT_DIRSTAT)
4443                         show_dirstat_by_line(&diffstat, options);
4444                 free_diffstat_info(&diffstat);
4445                 separator++;
4446         }
4447         if ((output_format & DIFF_FORMAT_DIRSTAT) && !dirstat_by_line)
4448                 show_dirstat(options);
4449
4450         if (output_format & DIFF_FORMAT_SUMMARY && !is_summary_empty(q)) {
4451                 for (i = 0; i < q->nr; i++) {
4452                         diff_summary(options, q->queue[i]);
4453                 }
4454                 separator++;
4455         }
4456
4457         if (output_format & DIFF_FORMAT_NO_OUTPUT &&
4458             DIFF_OPT_TST(options, EXIT_WITH_STATUS) &&
4459             DIFF_OPT_TST(options, DIFF_FROM_CONTENTS)) {
4460                 /*
4461                  * run diff_flush_patch for the exit status. setting
4462                  * options->file to /dev/null should be safe, becaue we
4463                  * aren't supposed to produce any output anyway.
4464                  */
4465                 if (options->close_file)
4466                         fclose(options->file);
4467                 options->file = fopen("/dev/null", "w");
4468                 if (!options->file)
4469                         die_errno("Could not open /dev/null");
4470                 options->close_file = 1;
4471                 for (i = 0; i < q->nr; i++) {
4472                         struct diff_filepair *p = q->queue[i];
4473                         if (check_pair_status(p))
4474                                 diff_flush_patch(p, options);
4475                         if (options->found_changes)
4476                                 break;
4477                 }
4478         }
4479
4480         if (output_format & DIFF_FORMAT_PATCH) {
4481                 if (separator) {
4482                         if (options->output_prefix) {
4483                                 struct strbuf *msg = NULL;
4484                                 msg = options->output_prefix(options,
4485                                         options->output_prefix_data);
4486                                 fwrite(msg->buf, msg->len, 1, stdout);
4487                         }
4488                         putc(options->line_termination, options->file);
4489                         if (options->stat_sep) {
4490                                 /* attach patch instead of inline */
4491                                 fputs(options->stat_sep, options->file);
4492                         }
4493                 }
4494
4495                 for (i = 0; i < q->nr; i++) {
4496                         struct diff_filepair *p = q->queue[i];
4497                         if (check_pair_status(p))
4498                                 diff_flush_patch(p, options);
4499                 }
4500         }
4501
4502         if (output_format & DIFF_FORMAT_CALLBACK)
4503                 options->format_callback(q, options, options->format_callback_data);
4504
4505         for (i = 0; i < q->nr; i++)
4506                 diff_free_filepair(q->queue[i]);
4507 free_queue:
4508         free(q->queue);
4509         DIFF_QUEUE_CLEAR(q);
4510         if (options->close_file)
4511                 fclose(options->file);
4512
4513         /*
4514          * Report the content-level differences with HAS_CHANGES;
4515          * diff_addremove/diff_change does not set the bit when
4516          * DIFF_FROM_CONTENTS is in effect (e.g. with -w).
4517          */
4518         if (DIFF_OPT_TST(options, DIFF_FROM_CONTENTS)) {
4519                 if (options->found_changes)
4520                         DIFF_OPT_SET(options, HAS_CHANGES);
4521                 else
4522                         DIFF_OPT_CLR(options, HAS_CHANGES);
4523         }
4524 }
4525
4526 static void diffcore_apply_filter(const char *filter)
4527 {
4528         int i;
4529         struct diff_queue_struct *q = &diff_queued_diff;
4530         struct diff_queue_struct outq;
4531         DIFF_QUEUE_CLEAR(&outq);
4532
4533         if (!filter)
4534                 return;
4535
4536         if (strchr(filter, DIFF_STATUS_FILTER_AON)) {
4537                 int found;
4538                 for (i = found = 0; !found && i < q->nr; i++) {
4539                         struct diff_filepair *p = q->queue[i];
4540                         if (((p->status == DIFF_STATUS_MODIFIED) &&
4541                              ((p->score &&
4542                                strchr(filter, DIFF_STATUS_FILTER_BROKEN)) ||
4543                               (!p->score &&
4544                                strchr(filter, DIFF_STATUS_MODIFIED)))) ||
4545                             ((p->status != DIFF_STATUS_MODIFIED) &&
4546                              strchr(filter, p->status)))
4547                                 found++;
4548                 }
4549                 if (found)
4550                         return;
4551
4552                 /* otherwise we will clear the whole queue
4553                  * by copying the empty outq at the end of this
4554                  * function, but first clear the current entries
4555                  * in the queue.
4556                  */
4557                 for (i = 0; i < q->nr; i++)
4558                         diff_free_filepair(q->queue[i]);
4559         }
4560         else {
4561                 /* Only the matching ones */
4562                 for (i = 0; i < q->nr; i++) {
4563                         struct diff_filepair *p = q->queue[i];
4564
4565                         if (((p->status == DIFF_STATUS_MODIFIED) &&
4566                              ((p->score &&
4567                                strchr(filter, DIFF_STATUS_FILTER_BROKEN)) ||
4568                               (!p->score &&
4569                                strchr(filter, DIFF_STATUS_MODIFIED)))) ||
4570                             ((p->status != DIFF_STATUS_MODIFIED) &&
4571                              strchr(filter, p->status)))
4572                                 diff_q(&outq, p);
4573                         else
4574                                 diff_free_filepair(p);
4575                 }
4576         }
4577         free(q->queue);
4578         *q = outq;
4579 }
4580
4581 /* Check whether two filespecs with the same mode and size are identical */
4582 static int diff_filespec_is_identical(struct diff_filespec *one,
4583                                       struct diff_filespec *two)
4584 {
4585         if (S_ISGITLINK(one->mode))
4586                 return 0;
4587         if (diff_populate_filespec(one, 0))
4588                 return 0;
4589         if (diff_populate_filespec(two, 0))
4590                 return 0;
4591         return !memcmp(one->data, two->data, one->size);
4592 }
4593
4594 static void diffcore_skip_stat_unmatch(struct diff_options *diffopt)
4595 {
4596         int i;
4597         struct diff_queue_struct *q = &diff_queued_diff;
4598         struct diff_queue_struct outq;
4599         DIFF_QUEUE_CLEAR(&outq);
4600
4601         for (i = 0; i < q->nr; i++) {
4602                 struct diff_filepair *p = q->queue[i];
4603
4604                 /*
4605                  * 1. Entries that come from stat info dirtiness
4606                  *    always have both sides (iow, not create/delete),
4607                  *    one side of the object name is unknown, with
4608                  *    the same mode and size.  Keep the ones that
4609                  *    do not match these criteria.  They have real
4610                  *    differences.
4611                  *
4612                  * 2. At this point, the file is known to be modified,
4613                  *    with the same mode and size, and the object
4614                  *    name of one side is unknown.  Need to inspect
4615                  *    the identical contents.
4616                  */
4617                 if (!DIFF_FILE_VALID(p->one) || /* (1) */
4618                     !DIFF_FILE_VALID(p->two) ||
4619                     (p->one->sha1_valid && p->two->sha1_valid) ||
4620                     (p->one->mode != p->two->mode) ||
4621                     diff_populate_filespec(p->one, 1) ||
4622                     diff_populate_filespec(p->two, 1) ||
4623                     (p->one->size != p->two->size) ||
4624                     !diff_filespec_is_identical(p->one, p->two)) /* (2) */
4625                         diff_q(&outq, p);
4626                 else {
4627                         /*
4628                          * The caller can subtract 1 from skip_stat_unmatch
4629                          * to determine how many paths were dirty only
4630                          * due to stat info mismatch.
4631                          */
4632                         if (!DIFF_OPT_TST(diffopt, NO_INDEX))
4633                                 diffopt->skip_stat_unmatch++;
4634                         diff_free_filepair(p);
4635                 }
4636         }
4637         free(q->queue);
4638         *q = outq;
4639 }
4640
4641 static int diffnamecmp(const void *a_, const void *b_)
4642 {
4643         const struct diff_filepair *a = *((const struct diff_filepair **)a_);
4644         const struct diff_filepair *b = *((const struct diff_filepair **)b_);
4645         const char *name_a, *name_b;
4646
4647         name_a = a->one ? a->one->path : a->two->path;
4648         name_b = b->one ? b->one->path : b->two->path;
4649         return strcmp(name_a, name_b);
4650 }
4651
4652 void diffcore_fix_diff_index(struct diff_options *options)
4653 {
4654         struct diff_queue_struct *q = &diff_queued_diff;
4655         qsort(q->queue, q->nr, sizeof(q->queue[0]), diffnamecmp);
4656 }
4657
4658 void diffcore_std(struct diff_options *options)
4659 {
4660         if (options->skip_stat_unmatch)
4661                 diffcore_skip_stat_unmatch(options);
4662         if (!options->found_follow) {
4663                 /* See try_to_follow_renames() in tree-diff.c */
4664                 if (options->break_opt != -1)
4665                         diffcore_break(options->break_opt);
4666                 if (options->detect_rename)
4667                         diffcore_rename(options);
4668                 if (options->break_opt != -1)
4669                         diffcore_merge_broken();
4670         }
4671         if (options->pickaxe)
4672                 diffcore_pickaxe(options);
4673         if (options->orderfile)
4674                 diffcore_order(options->orderfile);
4675         if (!options->found_follow)
4676                 /* See try_to_follow_renames() in tree-diff.c */
4677                 diff_resolve_rename_copy();
4678         diffcore_apply_filter(options->filter);
4679
4680         if (diff_queued_diff.nr && !DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
4681                 DIFF_OPT_SET(options, HAS_CHANGES);
4682         else
4683                 DIFF_OPT_CLR(options, HAS_CHANGES);
4684
4685         options->found_follow = 0;
4686 }
4687
4688 int diff_result_code(struct diff_options *opt, int status)
4689 {
4690         int result = 0;
4691
4692         diff_warn_rename_limit("diff.renamelimit",
4693                                opt->needed_rename_limit,
4694                                opt->degraded_cc_to_c);
4695         if (!DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
4696             !(opt->output_format & DIFF_FORMAT_CHECKDIFF))
4697                 return status;
4698         if (DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
4699             DIFF_OPT_TST(opt, HAS_CHANGES))
4700                 result |= 01;
4701         if ((opt->output_format & DIFF_FORMAT_CHECKDIFF) &&
4702             DIFF_OPT_TST(opt, CHECK_FAILED))
4703                 result |= 02;
4704         return result;
4705 }
4706
4707 int diff_can_quit_early(struct diff_options *opt)
4708 {
4709         return (DIFF_OPT_TST(opt, QUICK) &&
4710                 !opt->filter &&
4711                 DIFF_OPT_TST(opt, HAS_CHANGES));
4712 }
4713
4714 /*
4715  * Shall changes to this submodule be ignored?
4716  *
4717  * Submodule changes can be configured to be ignored separately for each path,
4718  * but that configuration can be overridden from the command line.
4719  */
4720 static int is_submodule_ignored(const char *path, struct diff_options *options)
4721 {
4722         int ignored = 0;
4723         unsigned orig_flags = options->flags;
4724         if (!DIFF_OPT_TST(options, OVERRIDE_SUBMODULE_CONFIG))
4725                 set_diffopt_flags_from_submodule_config(options, path);
4726         if (DIFF_OPT_TST(options, IGNORE_SUBMODULES))
4727                 ignored = 1;
4728         options->flags = orig_flags;
4729         return ignored;
4730 }
4731
4732 void diff_addremove(struct diff_options *options,
4733                     int addremove, unsigned mode,
4734                     const unsigned char *sha1,
4735                     int sha1_valid,
4736                     const char *concatpath, unsigned dirty_submodule)
4737 {
4738         struct diff_filespec *one, *two;
4739
4740         if (S_ISGITLINK(mode) && is_submodule_ignored(concatpath, options))
4741                 return;
4742
4743         /* This may look odd, but it is a preparation for
4744          * feeding "there are unchanged files which should
4745          * not produce diffs, but when you are doing copy
4746          * detection you would need them, so here they are"
4747          * entries to the diff-core.  They will be prefixed
4748          * with something like '=' or '*' (I haven't decided
4749          * which but should not make any difference).
4750          * Feeding the same new and old to diff_change()
4751          * also has the same effect.
4752          * Before the final output happens, they are pruned after
4753          * merged into rename/copy pairs as appropriate.
4754          */
4755         if (DIFF_OPT_TST(options, REVERSE_DIFF))
4756                 addremove = (addremove == '+' ? '-' :
4757                              addremove == '-' ? '+' : addremove);
4758
4759         if (options->prefix &&
4760             strncmp(concatpath, options->prefix, options->prefix_length))
4761                 return;
4762
4763         one = alloc_filespec(concatpath);
4764         two = alloc_filespec(concatpath);
4765
4766         if (addremove != '+')
4767                 fill_filespec(one, sha1, sha1_valid, mode);
4768         if (addremove != '-') {
4769                 fill_filespec(two, sha1, sha1_valid, mode);
4770                 two->dirty_submodule = dirty_submodule;
4771         }
4772
4773         diff_queue(&diff_queued_diff, one, two);
4774         if (!DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
4775                 DIFF_OPT_SET(options, HAS_CHANGES);
4776 }
4777
4778 void diff_change(struct diff_options *options,
4779                  unsigned old_mode, unsigned new_mode,
4780                  const unsigned char *old_sha1,
4781                  const unsigned char *new_sha1,
4782                  int old_sha1_valid, int new_sha1_valid,
4783                  const char *concatpath,
4784                  unsigned old_dirty_submodule, unsigned new_dirty_submodule)
4785 {
4786         struct diff_filespec *one, *two;
4787
4788         if (S_ISGITLINK(old_mode) && S_ISGITLINK(new_mode) &&
4789             is_submodule_ignored(concatpath, options))
4790                 return;
4791
4792         if (DIFF_OPT_TST(options, REVERSE_DIFF)) {
4793                 unsigned tmp;
4794                 const unsigned char *tmp_c;
4795                 tmp = old_mode; old_mode = new_mode; new_mode = tmp;
4796                 tmp_c = old_sha1; old_sha1 = new_sha1; new_sha1 = tmp_c;
4797                 tmp = old_sha1_valid; old_sha1_valid = new_sha1_valid;
4798                         new_sha1_valid = tmp;
4799                 tmp = old_dirty_submodule; old_dirty_submodule = new_dirty_submodule;
4800                         new_dirty_submodule = tmp;
4801         }
4802
4803         if (options->prefix &&
4804             strncmp(concatpath, options->prefix, options->prefix_length))
4805                 return;
4806
4807         one = alloc_filespec(concatpath);
4808         two = alloc_filespec(concatpath);
4809         fill_filespec(one, old_sha1, old_sha1_valid, old_mode);
4810         fill_filespec(two, new_sha1, new_sha1_valid, new_mode);
4811         one->dirty_submodule = old_dirty_submodule;
4812         two->dirty_submodule = new_dirty_submodule;
4813
4814         diff_queue(&diff_queued_diff, one, two);
4815         if (!DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
4816                 DIFF_OPT_SET(options, HAS_CHANGES);
4817 }
4818
4819 struct diff_filepair *diff_unmerge(struct diff_options *options, const char *path)
4820 {
4821         struct diff_filepair *pair;
4822         struct diff_filespec *one, *two;
4823
4824         if (options->prefix &&
4825             strncmp(path, options->prefix, options->prefix_length))
4826                 return NULL;
4827
4828         one = alloc_filespec(path);
4829         two = alloc_filespec(path);
4830         pair = diff_queue(&diff_queued_diff, one, two);
4831         pair->is_unmerged = 1;
4832         return pair;
4833 }
4834
4835 static char *run_textconv(const char *pgm, struct diff_filespec *spec,
4836                 size_t *outsize)
4837 {
4838         struct diff_tempfile *temp;
4839         const char *argv[3];
4840         const char **arg = argv;
4841         struct child_process child;
4842         struct strbuf buf = STRBUF_INIT;
4843         int err = 0;
4844
4845         temp = prepare_temp_file(spec->path, spec);
4846         *arg++ = pgm;
4847         *arg++ = temp->name;
4848         *arg = NULL;
4849
4850         memset(&child, 0, sizeof(child));
4851         child.use_shell = 1;
4852         child.argv = argv;
4853         child.out = -1;
4854         if (start_command(&child)) {
4855                 remove_tempfile();
4856                 return NULL;
4857         }
4858
4859         if (strbuf_read(&buf, child.out, 0) < 0)
4860                 err = error("error reading from textconv command '%s'", pgm);
4861         close(child.out);
4862
4863         if (finish_command(&child) || err) {
4864                 strbuf_release(&buf);
4865                 remove_tempfile();
4866                 return NULL;
4867         }
4868         remove_tempfile();
4869
4870         return strbuf_detach(&buf, outsize);
4871 }
4872
4873 size_t fill_textconv(struct userdiff_driver *driver,
4874                      struct diff_filespec *df,
4875                      char **outbuf)
4876 {
4877         size_t size;
4878
4879         if (!driver || !driver->textconv) {
4880                 if (!DIFF_FILE_VALID(df)) {
4881                         *outbuf = "";
4882                         return 0;
4883                 }
4884                 if (diff_populate_filespec(df, 0))
4885                         die("unable to read files to diff");
4886                 *outbuf = df->data;
4887                 return df->size;
4888         }
4889
4890         if (driver->textconv_cache && df->sha1_valid) {
4891                 *outbuf = notes_cache_get(driver->textconv_cache, df->sha1,
4892                                           &size);
4893                 if (*outbuf)
4894                         return size;
4895         }
4896
4897         *outbuf = run_textconv(driver->textconv, df, &size);
4898         if (!*outbuf)
4899                 die("unable to read files to diff");
4900
4901         if (driver->textconv_cache && df->sha1_valid) {
4902                 /* ignore errors, as we might be in a readonly repository */
4903                 notes_cache_put(driver->textconv_cache, df->sha1, *outbuf,
4904                                 size);
4905                 /*
4906                  * we could save up changes and flush them all at the end,
4907                  * but we would need an extra call after all diffing is done.
4908                  * Since generating a cache entry is the slow path anyway,
4909                  * this extra overhead probably isn't a big deal.
4910                  */
4911                 notes_cache_write(driver->textconv_cache);
4912         }
4913
4914         return size;
4915 }
4916
4917 void setup_diff_pager(struct diff_options *opt)
4918 {
4919         /*
4920          * If the user asked for our exit code, then either they want --quiet
4921          * or --exit-code. We should definitely not bother with a pager in the
4922          * former case, as we will generate no output. Since we still properly
4923          * report our exit code even when a pager is run, we _could_ run a
4924          * pager with --exit-code. But since we have not done so historically,
4925          * and because it is easy to find people oneline advising "git diff
4926          * --exit-code" in hooks and other scripts, we do not do so.
4927          */
4928         if (!DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
4929             check_pager_config("diff") != 0)
4930                 setup_pager();
4931 }