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