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