Merge branch 'maint'
[git] / diff.c
1 /*
2  * Copyright (C) 2005 Junio C Hamano
3  */
4 #include "cache.h"
5 #include "quote.h"
6 #include "diff.h"
7 #include "diffcore.h"
8 #include "delta.h"
9 #include "xdiff-interface.h"
10 #include "color.h"
11 #include "attr.h"
12 #include "run-command.h"
13 #include "utf8.h"
14 #include "userdiff.h"
15 #include "sigchain.h"
16 #include "submodule.h"
17 #include "ll-merge.h"
18
19 #ifdef NO_FAST_WORKING_DIRECTORY
20 #define FAST_WORKING_DIRECTORY 0
21 #else
22 #define FAST_WORKING_DIRECTORY 1
23 #endif
24
25 static int diff_detect_rename_default;
26 static int diff_rename_limit_default = 400;
27 static int diff_suppress_blank_empty;
28 int diff_use_color_default = -1;
29 static const char *diff_word_regex_cfg;
30 static const char *external_diff_cmd_cfg;
31 int diff_auto_refresh_index = 1;
32 static int diff_mnemonic_prefix;
33 static int diff_no_prefix;
34 static int diff_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         if (!it)
1277                 return 0;
1278         /*
1279          * make sure that at least one '-' or '+' is printed if
1280          * there is any change to this path. The easiest way is to
1281          * scale linearly as if the alloted width is one column shorter
1282          * than it is, and then add 1 to the result.
1283          */
1284         return 1 + (it * (width - 1) / max_change);
1285 }
1286
1287 static void show_name(FILE *file,
1288                       const char *prefix, const char *name, int len)
1289 {
1290         fprintf(file, " %s%-*s |", prefix, len, name);
1291 }
1292
1293 static void show_graph(FILE *file, char ch, int cnt, const char *set, const char *reset)
1294 {
1295         if (cnt <= 0)
1296                 return;
1297         fprintf(file, "%s", set);
1298         while (cnt--)
1299                 putc(ch, file);
1300         fprintf(file, "%s", reset);
1301 }
1302
1303 static void fill_print_name(struct diffstat_file *file)
1304 {
1305         char *pname;
1306
1307         if (file->print_name)
1308                 return;
1309
1310         if (!file->is_renamed) {
1311                 struct strbuf buf = STRBUF_INIT;
1312                 if (quote_c_style(file->name, &buf, NULL, 0)) {
1313                         pname = strbuf_detach(&buf, NULL);
1314                 } else {
1315                         pname = file->name;
1316                         strbuf_release(&buf);
1317                 }
1318         } else {
1319                 pname = pprint_rename(file->from_name, file->name);
1320         }
1321         file->print_name = pname;
1322 }
1323
1324 int print_stat_summary(FILE *fp, int files, int insertions, int deletions)
1325 {
1326         struct strbuf sb = STRBUF_INIT;
1327         int ret;
1328
1329         if (!files) {
1330                 assert(insertions == 0 && deletions == 0);
1331                 return fputs(_(" 0 files changed\n"), fp);
1332         }
1333
1334         strbuf_addf(&sb,
1335                     Q_(" %d file changed", " %d files changed", files),
1336                     files);
1337
1338         /*
1339          * For binary diff, the caller may want to print "x files
1340          * changed" with insertions == 0 && deletions == 0.
1341          *
1342          * Not omitting "0 insertions(+), 0 deletions(-)" in this case
1343          * is probably less confusing (i.e skip over "2 files changed
1344          * but nothing about added/removed lines? Is this a bug in Git?").
1345          */
1346         if (insertions || deletions == 0) {
1347                 /*
1348                  * TRANSLATORS: "+" in (+) is a line addition marker;
1349                  * do not translate it.
1350                  */
1351                 strbuf_addf(&sb,
1352                             Q_(", %d insertion(+)", ", %d insertions(+)",
1353                                insertions),
1354                             insertions);
1355         }
1356
1357         if (deletions || insertions == 0) {
1358                 /*
1359                  * TRANSLATORS: "-" in (-) is a line removal marker;
1360                  * do not translate it.
1361                  */
1362                 strbuf_addf(&sb,
1363                             Q_(", %d deletion(-)", ", %d deletions(-)",
1364                                deletions),
1365                             deletions);
1366         }
1367         strbuf_addch(&sb, '\n');
1368         ret = fputs(sb.buf, fp);
1369         strbuf_release(&sb);
1370         return ret;
1371 }
1372
1373 static void show_stats(struct diffstat_t *data, struct diff_options *options)
1374 {
1375         int i, len, add, del, adds = 0, dels = 0;
1376         uintmax_t max_change = 0, max_len = 0;
1377         int total_files = data->nr;
1378         int width, name_width, count;
1379         const char *reset, *add_c, *del_c;
1380         const char *line_prefix = "";
1381         int extra_shown = 0;
1382         struct strbuf *msg = NULL;
1383
1384         if (data->nr == 0)
1385                 return;
1386
1387         if (options->output_prefix) {
1388                 msg = options->output_prefix(options, options->output_prefix_data);
1389                 line_prefix = msg->buf;
1390         }
1391
1392         width = options->stat_width ? options->stat_width : 80;
1393         name_width = options->stat_name_width ? options->stat_name_width : 50;
1394         count = options->stat_count ? options->stat_count : data->nr;
1395
1396         /* Sanity: give at least 5 columns to the graph,
1397          * but leave at least 10 columns for the name.
1398          */
1399         if (width < 25)
1400                 width = 25;
1401         if (name_width < 10)
1402                 name_width = 10;
1403         else if (width < name_width + 15)
1404                 name_width = width - 15;
1405
1406         /* Find the longest filename and max number of changes */
1407         reset = diff_get_color_opt(options, DIFF_RESET);
1408         add_c = diff_get_color_opt(options, DIFF_FILE_NEW);
1409         del_c = diff_get_color_opt(options, DIFF_FILE_OLD);
1410
1411         for (i = 0; (i < count) && (i < data->nr); i++) {
1412                 struct diffstat_file *file = data->files[i];
1413                 uintmax_t change = file->added + file->deleted;
1414                 if (!data->files[i]->is_renamed &&
1415                          (change == 0)) {
1416                         count++; /* not shown == room for one more */
1417                         continue;
1418                 }
1419                 fill_print_name(file);
1420                 len = strlen(file->print_name);
1421                 if (max_len < len)
1422                         max_len = len;
1423
1424                 if (file->is_binary || file->is_unmerged)
1425                         continue;
1426                 if (max_change < change)
1427                         max_change = change;
1428         }
1429         count = i; /* min(count, data->nr) */
1430
1431         /* Compute the width of the graph part;
1432          * 10 is for one blank at the beginning of the line plus
1433          * " | count " between the name and the graph.
1434          *
1435          * From here on, name_width is the width of the name area,
1436          * and width is the width of the graph area.
1437          */
1438         name_width = (name_width < max_len) ? name_width : max_len;
1439         if (width < (name_width + 10) + max_change)
1440                 width = width - (name_width + 10);
1441         else
1442                 width = max_change;
1443
1444         for (i = 0; i < count; i++) {
1445                 const char *prefix = "";
1446                 char *name = data->files[i]->print_name;
1447                 uintmax_t added = data->files[i]->added;
1448                 uintmax_t deleted = data->files[i]->deleted;
1449                 int name_len;
1450
1451                 if (!data->files[i]->is_renamed &&
1452                          (added + deleted == 0)) {
1453                         total_files--;
1454                         continue;
1455                 }
1456                 /*
1457                  * "scale" the filename
1458                  */
1459                 len = name_width;
1460                 name_len = strlen(name);
1461                 if (name_width < name_len) {
1462                         char *slash;
1463                         prefix = "...";
1464                         len -= 3;
1465                         name += name_len - len;
1466                         slash = strchr(name, '/');
1467                         if (slash)
1468                                 name = slash;
1469                 }
1470
1471                 if (data->files[i]->is_binary) {
1472                         fprintf(options->file, "%s", line_prefix);
1473                         show_name(options->file, prefix, name, len);
1474                         fprintf(options->file, "  Bin ");
1475                         fprintf(options->file, "%s%"PRIuMAX"%s",
1476                                 del_c, deleted, reset);
1477                         fprintf(options->file, " -> ");
1478                         fprintf(options->file, "%s%"PRIuMAX"%s",
1479                                 add_c, added, reset);
1480                         fprintf(options->file, " bytes");
1481                         fprintf(options->file, "\n");
1482                         continue;
1483                 }
1484                 else if (data->files[i]->is_unmerged) {
1485                         fprintf(options->file, "%s", line_prefix);
1486                         show_name(options->file, prefix, name, len);
1487                         fprintf(options->file, "  Unmerged\n");
1488                         continue;
1489                 }
1490
1491                 /*
1492                  * scale the add/delete
1493                  */
1494                 add = added;
1495                 del = deleted;
1496                 adds += add;
1497                 dels += del;
1498
1499                 if (width <= max_change) {
1500                         int total = add + del;
1501
1502                         total = scale_linear(add + del, width, max_change);
1503                         if (total < 2 && add && del)
1504                                 /* width >= 2 due to the sanity check */
1505                                 total = 2;
1506                         if (add < del) {
1507                                 add = scale_linear(add, width, max_change);
1508                                 del = total - add;
1509                         } else {
1510                                 del = scale_linear(del, width, max_change);
1511                                 add = total - del;
1512                         }
1513                 }
1514                 fprintf(options->file, "%s", line_prefix);
1515                 show_name(options->file, prefix, name, len);
1516                 fprintf(options->file, "%5"PRIuMAX"%s", added + deleted,
1517                                 added + deleted ? " " : "");
1518                 show_graph(options->file, '+', add, add_c, reset);
1519                 show_graph(options->file, '-', del, del_c, reset);
1520                 fprintf(options->file, "\n");
1521         }
1522         for (i = count; i < data->nr; i++) {
1523                 uintmax_t added = data->files[i]->added;
1524                 uintmax_t deleted = data->files[i]->deleted;
1525                 if (!data->files[i]->is_renamed &&
1526                          (added + deleted == 0)) {
1527                         total_files--;
1528                         continue;
1529                 }
1530                 adds += added;
1531                 dels += deleted;
1532                 if (!extra_shown)
1533                         fprintf(options->file, "%s ...\n", line_prefix);
1534                 extra_shown = 1;
1535         }
1536         fprintf(options->file, "%s", line_prefix);
1537         print_stat_summary(options->file, total_files, adds, dels);
1538 }
1539
1540 static void show_shortstats(struct diffstat_t *data, struct diff_options *options)
1541 {
1542         int i, adds = 0, dels = 0, total_files = data->nr;
1543
1544         if (data->nr == 0)
1545                 return;
1546
1547         for (i = 0; i < data->nr; i++) {
1548                 if (!data->files[i]->is_binary &&
1549                     !data->files[i]->is_unmerged) {
1550                         int added = data->files[i]->added;
1551                         int deleted= data->files[i]->deleted;
1552                         if (!data->files[i]->is_renamed &&
1553                             (added + deleted == 0)) {
1554                                 total_files--;
1555                         } else {
1556                                 adds += added;
1557                                 dels += deleted;
1558                         }
1559                 }
1560         }
1561         if (options->output_prefix) {
1562                 struct strbuf *msg = NULL;
1563                 msg = options->output_prefix(options,
1564                                 options->output_prefix_data);
1565                 fprintf(options->file, "%s", msg->buf);
1566         }
1567         print_stat_summary(options->file, total_files, adds, dels);
1568 }
1569
1570 static void show_numstat(struct diffstat_t *data, struct diff_options *options)
1571 {
1572         int i;
1573
1574         if (data->nr == 0)
1575                 return;
1576
1577         for (i = 0; i < data->nr; i++) {
1578                 struct diffstat_file *file = data->files[i];
1579
1580                 if (options->output_prefix) {
1581                         struct strbuf *msg = NULL;
1582                         msg = options->output_prefix(options,
1583                                         options->output_prefix_data);
1584                         fprintf(options->file, "%s", msg->buf);
1585                 }
1586
1587                 if (file->is_binary)
1588                         fprintf(options->file, "-\t-\t");
1589                 else
1590                         fprintf(options->file,
1591                                 "%"PRIuMAX"\t%"PRIuMAX"\t",
1592                                 file->added, file->deleted);
1593                 if (options->line_termination) {
1594                         fill_print_name(file);
1595                         if (!file->is_renamed)
1596                                 write_name_quoted(file->name, options->file,
1597                                                   options->line_termination);
1598                         else {
1599                                 fputs(file->print_name, options->file);
1600                                 putc(options->line_termination, options->file);
1601                         }
1602                 } else {
1603                         if (file->is_renamed) {
1604                                 putc('\0', options->file);
1605                                 write_name_quoted(file->from_name, options->file, '\0');
1606                         }
1607                         write_name_quoted(file->name, options->file, '\0');
1608                 }
1609         }
1610 }
1611
1612 struct dirstat_file {
1613         const char *name;
1614         unsigned long changed;
1615 };
1616
1617 struct dirstat_dir {
1618         struct dirstat_file *files;
1619         int alloc, nr, permille, cumulative;
1620 };
1621
1622 static long gather_dirstat(struct diff_options *opt, struct dirstat_dir *dir,
1623                 unsigned long changed, const char *base, int baselen)
1624 {
1625         unsigned long this_dir = 0;
1626         unsigned int sources = 0;
1627         const char *line_prefix = "";
1628         struct strbuf *msg = NULL;
1629
1630         if (opt->output_prefix) {
1631                 msg = opt->output_prefix(opt, opt->output_prefix_data);
1632                 line_prefix = msg->buf;
1633         }
1634
1635         while (dir->nr) {
1636                 struct dirstat_file *f = dir->files;
1637                 int namelen = strlen(f->name);
1638                 unsigned long this;
1639                 char *slash;
1640
1641                 if (namelen < baselen)
1642                         break;
1643                 if (memcmp(f->name, base, baselen))
1644                         break;
1645                 slash = strchr(f->name + baselen, '/');
1646                 if (slash) {
1647                         int newbaselen = slash + 1 - f->name;
1648                         this = gather_dirstat(opt, dir, changed, f->name, newbaselen);
1649                         sources++;
1650                 } else {
1651                         this = f->changed;
1652                         dir->files++;
1653                         dir->nr--;
1654                         sources += 2;
1655                 }
1656                 this_dir += this;
1657         }
1658
1659         /*
1660          * We don't report dirstat's for
1661          *  - the top level
1662          *  - or cases where everything came from a single directory
1663          *    under this directory (sources == 1).
1664          */
1665         if (baselen && sources != 1) {
1666                 if (this_dir) {
1667                         int permille = this_dir * 1000 / changed;
1668                         if (permille >= dir->permille) {
1669                                 fprintf(opt->file, "%s%4d.%01d%% %.*s\n", line_prefix,
1670                                         permille / 10, permille % 10, baselen, base);
1671                                 if (!dir->cumulative)
1672                                         return 0;
1673                         }
1674                 }
1675         }
1676         return this_dir;
1677 }
1678
1679 static int dirstat_compare(const void *_a, const void *_b)
1680 {
1681         const struct dirstat_file *a = _a;
1682         const struct dirstat_file *b = _b;
1683         return strcmp(a->name, b->name);
1684 }
1685
1686 static void show_dirstat(struct diff_options *options)
1687 {
1688         int i;
1689         unsigned long changed;
1690         struct dirstat_dir dir;
1691         struct diff_queue_struct *q = &diff_queued_diff;
1692
1693         dir.files = NULL;
1694         dir.alloc = 0;
1695         dir.nr = 0;
1696         dir.permille = options->dirstat_permille;
1697         dir.cumulative = DIFF_OPT_TST(options, DIRSTAT_CUMULATIVE);
1698
1699         changed = 0;
1700         for (i = 0; i < q->nr; i++) {
1701                 struct diff_filepair *p = q->queue[i];
1702                 const char *name;
1703                 unsigned long copied, added, damage;
1704                 int content_changed;
1705
1706                 name = p->two->path ? p->two->path : p->one->path;
1707
1708                 if (p->one->sha1_valid && p->two->sha1_valid)
1709                         content_changed = hashcmp(p->one->sha1, p->two->sha1);
1710                 else
1711                         content_changed = 1;
1712
1713                 if (!content_changed) {
1714                         /*
1715                          * The SHA1 has not changed, so pre-/post-content is
1716                          * identical. We can therefore skip looking at the
1717                          * file contents altogether.
1718                          */
1719                         damage = 0;
1720                         goto found_damage;
1721                 }
1722
1723                 if (DIFF_OPT_TST(options, DIRSTAT_BY_FILE)) {
1724                         /*
1725                          * In --dirstat-by-file mode, we don't really need to
1726                          * look at the actual file contents at all.
1727                          * The fact that the SHA1 changed is enough for us to
1728                          * add this file to the list of results
1729                          * (with each file contributing equal damage).
1730                          */
1731                         damage = 1;
1732                         goto found_damage;
1733                 }
1734
1735                 if (DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) {
1736                         diff_populate_filespec(p->one, 0);
1737                         diff_populate_filespec(p->two, 0);
1738                         diffcore_count_changes(p->one, p->two, NULL, NULL, 0,
1739                                                &copied, &added);
1740                         diff_free_filespec_data(p->one);
1741                         diff_free_filespec_data(p->two);
1742                 } else if (DIFF_FILE_VALID(p->one)) {
1743                         diff_populate_filespec(p->one, 1);
1744                         copied = added = 0;
1745                         diff_free_filespec_data(p->one);
1746                 } else if (DIFF_FILE_VALID(p->two)) {
1747                         diff_populate_filespec(p->two, 1);
1748                         copied = 0;
1749                         added = p->two->size;
1750                         diff_free_filespec_data(p->two);
1751                 } else
1752                         continue;
1753
1754                 /*
1755                  * Original minus copied is the removed material,
1756                  * added is the new material.  They are both damages
1757                  * made to the preimage.
1758                  * If the resulting damage is zero, we know that
1759                  * diffcore_count_changes() considers the two entries to
1760                  * be identical, but since content_changed is true, we
1761                  * know that there must have been _some_ kind of change,
1762                  * so we force all entries to have damage > 0.
1763                  */
1764                 damage = (p->one->size - copied) + added;
1765                 if (!damage)
1766                         damage = 1;
1767
1768 found_damage:
1769                 ALLOC_GROW(dir.files, dir.nr + 1, dir.alloc);
1770                 dir.files[dir.nr].name = name;
1771                 dir.files[dir.nr].changed = damage;
1772                 changed += damage;
1773                 dir.nr++;
1774         }
1775
1776         /* This can happen even with many files, if everything was renames */
1777         if (!changed)
1778                 return;
1779
1780         /* Show all directories with more than x% of the changes */
1781         qsort(dir.files, dir.nr, sizeof(dir.files[0]), dirstat_compare);
1782         gather_dirstat(options, &dir, changed, "", 0);
1783 }
1784
1785 static void show_dirstat_by_line(struct diffstat_t *data, struct diff_options *options)
1786 {
1787         int i;
1788         unsigned long changed;
1789         struct dirstat_dir dir;
1790
1791         if (data->nr == 0)
1792                 return;
1793
1794         dir.files = NULL;
1795         dir.alloc = 0;
1796         dir.nr = 0;
1797         dir.permille = options->dirstat_permille;
1798         dir.cumulative = DIFF_OPT_TST(options, DIRSTAT_CUMULATIVE);
1799
1800         changed = 0;
1801         for (i = 0; i < data->nr; i++) {
1802                 struct diffstat_file *file = data->files[i];
1803                 unsigned long damage = file->added + file->deleted;
1804                 if (file->is_binary)
1805                         /*
1806                          * binary files counts bytes, not lines. Must find some
1807                          * way to normalize binary bytes vs. textual lines.
1808                          * The following heuristic assumes that there are 64
1809                          * bytes per "line".
1810                          * This is stupid and ugly, but very cheap...
1811                          */
1812                         damage = (damage + 63) / 64;
1813                 ALLOC_GROW(dir.files, dir.nr + 1, dir.alloc);
1814                 dir.files[dir.nr].name = file->name;
1815                 dir.files[dir.nr].changed = damage;
1816                 changed += damage;
1817                 dir.nr++;
1818         }
1819
1820         /* This can happen even with many files, if everything was renames */
1821         if (!changed)
1822                 return;
1823
1824         /* Show all directories with more than x% of the changes */
1825         qsort(dir.files, dir.nr, sizeof(dir.files[0]), dirstat_compare);
1826         gather_dirstat(options, &dir, changed, "", 0);
1827 }
1828
1829 static void free_diffstat_info(struct diffstat_t *diffstat)
1830 {
1831         int i;
1832         for (i = 0; i < diffstat->nr; i++) {
1833                 struct diffstat_file *f = diffstat->files[i];
1834                 if (f->name != f->print_name)
1835                         free(f->print_name);
1836                 free(f->name);
1837                 free(f->from_name);
1838                 free(f);
1839         }
1840         free(diffstat->files);
1841 }
1842
1843 struct checkdiff_t {
1844         const char *filename;
1845         int lineno;
1846         int conflict_marker_size;
1847         struct diff_options *o;
1848         unsigned ws_rule;
1849         unsigned status;
1850 };
1851
1852 static int is_conflict_marker(const char *line, int marker_size, unsigned long len)
1853 {
1854         char firstchar;
1855         int cnt;
1856
1857         if (len < marker_size + 1)
1858                 return 0;
1859         firstchar = line[0];
1860         switch (firstchar) {
1861         case '=': case '>': case '<': case '|':
1862                 break;
1863         default:
1864                 return 0;
1865         }
1866         for (cnt = 1; cnt < marker_size; cnt++)
1867                 if (line[cnt] != firstchar)
1868                         return 0;
1869         /* line[1] thru line[marker_size-1] are same as firstchar */
1870         if (len < marker_size + 1 || !isspace(line[marker_size]))
1871                 return 0;
1872         return 1;
1873 }
1874
1875 static void checkdiff_consume(void *priv, char *line, unsigned long len)
1876 {
1877         struct checkdiff_t *data = priv;
1878         int marker_size = data->conflict_marker_size;
1879         const char *ws = diff_get_color(data->o->use_color, DIFF_WHITESPACE);
1880         const char *reset = diff_get_color(data->o->use_color, DIFF_RESET);
1881         const char *set = diff_get_color(data->o->use_color, DIFF_FILE_NEW);
1882         char *err;
1883         char *line_prefix = "";
1884         struct strbuf *msgbuf;
1885
1886         assert(data->o);
1887         if (data->o->output_prefix) {
1888                 msgbuf = data->o->output_prefix(data->o,
1889                         data->o->output_prefix_data);
1890                 line_prefix = msgbuf->buf;
1891         }
1892
1893         if (line[0] == '+') {
1894                 unsigned bad;
1895                 data->lineno++;
1896                 if (is_conflict_marker(line + 1, marker_size, len - 1)) {
1897                         data->status |= 1;
1898                         fprintf(data->o->file,
1899                                 "%s%s:%d: leftover conflict marker\n",
1900                                 line_prefix, data->filename, data->lineno);
1901                 }
1902                 bad = ws_check(line + 1, len - 1, data->ws_rule);
1903                 if (!bad)
1904                         return;
1905                 data->status |= bad;
1906                 err = whitespace_error_string(bad);
1907                 fprintf(data->o->file, "%s%s:%d: %s.\n",
1908                         line_prefix, data->filename, data->lineno, err);
1909                 free(err);
1910                 emit_line(data->o, set, reset, line, 1);
1911                 ws_check_emit(line + 1, len - 1, data->ws_rule,
1912                               data->o->file, set, reset, ws);
1913         } else if (line[0] == ' ') {
1914                 data->lineno++;
1915         } else if (line[0] == '@') {
1916                 char *plus = strchr(line, '+');
1917                 if (plus)
1918                         data->lineno = strtol(plus, NULL, 10) - 1;
1919                 else
1920                         die("invalid diff");
1921         }
1922 }
1923
1924 static unsigned char *deflate_it(char *data,
1925                                  unsigned long size,
1926                                  unsigned long *result_size)
1927 {
1928         int bound;
1929         unsigned char *deflated;
1930         git_zstream stream;
1931
1932         memset(&stream, 0, sizeof(stream));
1933         git_deflate_init(&stream, zlib_compression_level);
1934         bound = git_deflate_bound(&stream, size);
1935         deflated = xmalloc(bound);
1936         stream.next_out = deflated;
1937         stream.avail_out = bound;
1938
1939         stream.next_in = (unsigned char *)data;
1940         stream.avail_in = size;
1941         while (git_deflate(&stream, Z_FINISH) == Z_OK)
1942                 ; /* nothing */
1943         git_deflate_end(&stream);
1944         *result_size = stream.total_out;
1945         return deflated;
1946 }
1947
1948 static void emit_binary_diff_body(FILE *file, mmfile_t *one, mmfile_t *two, char *prefix)
1949 {
1950         void *cp;
1951         void *delta;
1952         void *deflated;
1953         void *data;
1954         unsigned long orig_size;
1955         unsigned long delta_size;
1956         unsigned long deflate_size;
1957         unsigned long data_size;
1958
1959         /* We could do deflated delta, or we could do just deflated two,
1960          * whichever is smaller.
1961          */
1962         delta = NULL;
1963         deflated = deflate_it(two->ptr, two->size, &deflate_size);
1964         if (one->size && two->size) {
1965                 delta = diff_delta(one->ptr, one->size,
1966                                    two->ptr, two->size,
1967                                    &delta_size, deflate_size);
1968                 if (delta) {
1969                         void *to_free = delta;
1970                         orig_size = delta_size;
1971                         delta = deflate_it(delta, delta_size, &delta_size);
1972                         free(to_free);
1973                 }
1974         }
1975
1976         if (delta && delta_size < deflate_size) {
1977                 fprintf(file, "%sdelta %lu\n", prefix, orig_size);
1978                 free(deflated);
1979                 data = delta;
1980                 data_size = delta_size;
1981         }
1982         else {
1983                 fprintf(file, "%sliteral %lu\n", prefix, two->size);
1984                 free(delta);
1985                 data = deflated;
1986                 data_size = deflate_size;
1987         }
1988
1989         /* emit data encoded in base85 */
1990         cp = data;
1991         while (data_size) {
1992                 int bytes = (52 < data_size) ? 52 : data_size;
1993                 char line[70];
1994                 data_size -= bytes;
1995                 if (bytes <= 26)
1996                         line[0] = bytes + 'A' - 1;
1997                 else
1998                         line[0] = bytes - 26 + 'a' - 1;
1999                 encode_85(line + 1, cp, bytes);
2000                 cp = (char *) cp + bytes;
2001                 fprintf(file, "%s", prefix);
2002                 fputs(line, file);
2003                 fputc('\n', file);
2004         }
2005         fprintf(file, "%s\n", prefix);
2006         free(data);
2007 }
2008
2009 static void emit_binary_diff(FILE *file, mmfile_t *one, mmfile_t *two, char *prefix)
2010 {
2011         fprintf(file, "%sGIT binary patch\n", prefix);
2012         emit_binary_diff_body(file, one, two, prefix);
2013         emit_binary_diff_body(file, two, one, prefix);
2014 }
2015
2016 static void diff_filespec_load_driver(struct diff_filespec *one)
2017 {
2018         /* Use already-loaded driver */
2019         if (one->driver)
2020                 return;
2021
2022         if (S_ISREG(one->mode))
2023                 one->driver = userdiff_find_by_path(one->path);
2024
2025         /* Fallback to default settings */
2026         if (!one->driver)
2027                 one->driver = userdiff_find_by_name("default");
2028 }
2029
2030 int diff_filespec_is_binary(struct diff_filespec *one)
2031 {
2032         if (one->is_binary == -1) {
2033                 diff_filespec_load_driver(one);
2034                 if (one->driver->binary != -1)
2035                         one->is_binary = one->driver->binary;
2036                 else {
2037                         if (!one->data && DIFF_FILE_VALID(one))
2038                                 diff_populate_filespec(one, 0);
2039                         if (one->data)
2040                                 one->is_binary = buffer_is_binary(one->data,
2041                                                 one->size);
2042                         if (one->is_binary == -1)
2043                                 one->is_binary = 0;
2044                 }
2045         }
2046         return one->is_binary;
2047 }
2048
2049 static const struct userdiff_funcname *diff_funcname_pattern(struct diff_filespec *one)
2050 {
2051         diff_filespec_load_driver(one);
2052         return one->driver->funcname.pattern ? &one->driver->funcname : NULL;
2053 }
2054
2055 static const char *userdiff_word_regex(struct diff_filespec *one)
2056 {
2057         diff_filespec_load_driver(one);
2058         return one->driver->word_regex;
2059 }
2060
2061 void diff_set_mnemonic_prefix(struct diff_options *options, const char *a, const char *b)
2062 {
2063         if (!options->a_prefix)
2064                 options->a_prefix = a;
2065         if (!options->b_prefix)
2066                 options->b_prefix = b;
2067 }
2068
2069 struct userdiff_driver *get_textconv(struct diff_filespec *one)
2070 {
2071         if (!DIFF_FILE_VALID(one))
2072                 return NULL;
2073
2074         diff_filespec_load_driver(one);
2075         return userdiff_get_textconv(one->driver);
2076 }
2077
2078 static void builtin_diff(const char *name_a,
2079                          const char *name_b,
2080                          struct diff_filespec *one,
2081                          struct diff_filespec *two,
2082                          const char *xfrm_msg,
2083                          int must_show_header,
2084                          struct diff_options *o,
2085                          int complete_rewrite)
2086 {
2087         mmfile_t mf1, mf2;
2088         const char *lbl[2];
2089         char *a_one, *b_two;
2090         const char *set = diff_get_color_opt(o, DIFF_METAINFO);
2091         const char *reset = diff_get_color_opt(o, DIFF_RESET);
2092         const char *a_prefix, *b_prefix;
2093         struct userdiff_driver *textconv_one = NULL;
2094         struct userdiff_driver *textconv_two = NULL;
2095         struct strbuf header = STRBUF_INIT;
2096         struct strbuf *msgbuf;
2097         char *line_prefix = "";
2098
2099         if (o->output_prefix) {
2100                 msgbuf = o->output_prefix(o, o->output_prefix_data);
2101                 line_prefix = msgbuf->buf;
2102         }
2103
2104         if (DIFF_OPT_TST(o, SUBMODULE_LOG) &&
2105                         (!one->mode || S_ISGITLINK(one->mode)) &&
2106                         (!two->mode || S_ISGITLINK(two->mode))) {
2107                 const char *del = diff_get_color_opt(o, DIFF_FILE_OLD);
2108                 const char *add = diff_get_color_opt(o, DIFF_FILE_NEW);
2109                 show_submodule_summary(o->file, one ? one->path : two->path,
2110                                 one->sha1, two->sha1, two->dirty_submodule,
2111                                 del, add, reset);
2112                 return;
2113         }
2114
2115         if (DIFF_OPT_TST(o, ALLOW_TEXTCONV)) {
2116                 textconv_one = get_textconv(one);
2117                 textconv_two = get_textconv(two);
2118         }
2119
2120         diff_set_mnemonic_prefix(o, "a/", "b/");
2121         if (DIFF_OPT_TST(o, REVERSE_DIFF)) {
2122                 a_prefix = o->b_prefix;
2123                 b_prefix = o->a_prefix;
2124         } else {
2125                 a_prefix = o->a_prefix;
2126                 b_prefix = o->b_prefix;
2127         }
2128
2129         /* Never use a non-valid filename anywhere if at all possible */
2130         name_a = DIFF_FILE_VALID(one) ? name_a : name_b;
2131         name_b = DIFF_FILE_VALID(two) ? name_b : name_a;
2132
2133         a_one = quote_two(a_prefix, name_a + (*name_a == '/'));
2134         b_two = quote_two(b_prefix, name_b + (*name_b == '/'));
2135         lbl[0] = DIFF_FILE_VALID(one) ? a_one : "/dev/null";
2136         lbl[1] = DIFF_FILE_VALID(two) ? b_two : "/dev/null";
2137         strbuf_addf(&header, "%s%sdiff --git %s %s%s\n", line_prefix, set, a_one, b_two, reset);
2138         if (lbl[0][0] == '/') {
2139                 /* /dev/null */
2140                 strbuf_addf(&header, "%s%snew file mode %06o%s\n", line_prefix, set, two->mode, reset);
2141                 if (xfrm_msg)
2142                         strbuf_addstr(&header, xfrm_msg);
2143                 must_show_header = 1;
2144         }
2145         else if (lbl[1][0] == '/') {
2146                 strbuf_addf(&header, "%s%sdeleted file mode %06o%s\n", line_prefix, set, one->mode, reset);
2147                 if (xfrm_msg)
2148                         strbuf_addstr(&header, xfrm_msg);
2149                 must_show_header = 1;
2150         }
2151         else {
2152                 if (one->mode != two->mode) {
2153                         strbuf_addf(&header, "%s%sold mode %06o%s\n", line_prefix, set, one->mode, reset);
2154                         strbuf_addf(&header, "%s%snew mode %06o%s\n", line_prefix, set, two->mode, reset);
2155                         must_show_header = 1;
2156                 }
2157                 if (xfrm_msg)
2158                         strbuf_addstr(&header, xfrm_msg);
2159
2160                 /*
2161                  * we do not run diff between different kind
2162                  * of objects.
2163                  */
2164                 if ((one->mode ^ two->mode) & S_IFMT)
2165                         goto free_ab_and_return;
2166                 if (complete_rewrite &&
2167                     (textconv_one || !diff_filespec_is_binary(one)) &&
2168                     (textconv_two || !diff_filespec_is_binary(two))) {
2169                         fprintf(o->file, "%s", header.buf);
2170                         strbuf_reset(&header);
2171                         emit_rewrite_diff(name_a, name_b, one, two,
2172                                                 textconv_one, textconv_two, o);
2173                         o->found_changes = 1;
2174                         goto free_ab_and_return;
2175                 }
2176         }
2177
2178         if (o->irreversible_delete && lbl[1][0] == '/') {
2179                 fprintf(o->file, "%s", header.buf);
2180                 strbuf_reset(&header);
2181                 goto free_ab_and_return;
2182         } else if (!DIFF_OPT_TST(o, TEXT) &&
2183             ( (!textconv_one && diff_filespec_is_binary(one)) ||
2184               (!textconv_two && diff_filespec_is_binary(two)) )) {
2185                 if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
2186                         die("unable to read files to diff");
2187                 /* Quite common confusing case */
2188                 if (mf1.size == mf2.size &&
2189                     !memcmp(mf1.ptr, mf2.ptr, mf1.size)) {
2190                         if (must_show_header)
2191                                 fprintf(o->file, "%s", header.buf);
2192                         goto free_ab_and_return;
2193                 }
2194                 fprintf(o->file, "%s", header.buf);
2195                 strbuf_reset(&header);
2196                 if (DIFF_OPT_TST(o, BINARY))
2197                         emit_binary_diff(o->file, &mf1, &mf2, line_prefix);
2198                 else
2199                         fprintf(o->file, "%sBinary files %s and %s differ\n",
2200                                 line_prefix, lbl[0], lbl[1]);
2201                 o->found_changes = 1;
2202         } else {
2203                 /* Crazy xdl interfaces.. */
2204                 const char *diffopts = getenv("GIT_DIFF_OPTS");
2205                 xpparam_t xpp;
2206                 xdemitconf_t xecfg;
2207                 struct emit_callback ecbdata;
2208                 const struct userdiff_funcname *pe;
2209
2210                 if (!DIFF_XDL_TST(o, WHITESPACE_FLAGS) || must_show_header) {
2211                         fprintf(o->file, "%s", header.buf);
2212                         strbuf_reset(&header);
2213                 }
2214
2215                 mf1.size = fill_textconv(textconv_one, one, &mf1.ptr);
2216                 mf2.size = fill_textconv(textconv_two, two, &mf2.ptr);
2217
2218                 pe = diff_funcname_pattern(one);
2219                 if (!pe)
2220                         pe = diff_funcname_pattern(two);
2221
2222                 memset(&xpp, 0, sizeof(xpp));
2223                 memset(&xecfg, 0, sizeof(xecfg));
2224                 memset(&ecbdata, 0, sizeof(ecbdata));
2225                 ecbdata.label_path = lbl;
2226                 ecbdata.color_diff = want_color(o->use_color);
2227                 ecbdata.found_changesp = &o->found_changes;
2228                 ecbdata.ws_rule = whitespace_rule(name_b ? name_b : name_a);
2229                 if (ecbdata.ws_rule & WS_BLANK_AT_EOF)
2230                         check_blank_at_eof(&mf1, &mf2, &ecbdata);
2231                 ecbdata.opt = o;
2232                 ecbdata.header = header.len ? &header : NULL;
2233                 xpp.flags = o->xdl_opts;
2234                 xecfg.ctxlen = o->context;
2235                 xecfg.interhunkctxlen = o->interhunkcontext;
2236                 xecfg.flags = XDL_EMIT_FUNCNAMES;
2237                 if (DIFF_OPT_TST(o, FUNCCONTEXT))
2238                         xecfg.flags |= XDL_EMIT_FUNCCONTEXT;
2239                 if (pe)
2240                         xdiff_set_find_func(&xecfg, pe->pattern, pe->cflags);
2241                 if (!diffopts)
2242                         ;
2243                 else if (!prefixcmp(diffopts, "--unified="))
2244                         xecfg.ctxlen = strtoul(diffopts + 10, NULL, 10);
2245                 else if (!prefixcmp(diffopts, "-u"))
2246                         xecfg.ctxlen = strtoul(diffopts + 2, NULL, 10);
2247                 if (o->word_diff) {
2248                         int i;
2249
2250                         ecbdata.diff_words =
2251                                 xcalloc(1, sizeof(struct diff_words_data));
2252                         ecbdata.diff_words->type = o->word_diff;
2253                         ecbdata.diff_words->opt = o;
2254                         if (!o->word_regex)
2255                                 o->word_regex = userdiff_word_regex(one);
2256                         if (!o->word_regex)
2257                                 o->word_regex = userdiff_word_regex(two);
2258                         if (!o->word_regex)
2259                                 o->word_regex = diff_word_regex_cfg;
2260                         if (o->word_regex) {
2261                                 ecbdata.diff_words->word_regex = (regex_t *)
2262                                         xmalloc(sizeof(regex_t));
2263                                 if (regcomp(ecbdata.diff_words->word_regex,
2264                                                 o->word_regex,
2265                                                 REG_EXTENDED | REG_NEWLINE))
2266                                         die ("Invalid regular expression: %s",
2267                                                         o->word_regex);
2268                         }
2269                         for (i = 0; i < ARRAY_SIZE(diff_words_styles); i++) {
2270                                 if (o->word_diff == diff_words_styles[i].type) {
2271                                         ecbdata.diff_words->style =
2272                                                 &diff_words_styles[i];
2273                                         break;
2274                                 }
2275                         }
2276                         if (want_color(o->use_color)) {
2277                                 struct diff_words_style *st = ecbdata.diff_words->style;
2278                                 st->old.color = diff_get_color_opt(o, DIFF_FILE_OLD);
2279                                 st->new.color = diff_get_color_opt(o, DIFF_FILE_NEW);
2280                                 st->ctx.color = diff_get_color_opt(o, DIFF_PLAIN);
2281                         }
2282                 }
2283                 xdi_diff_outf(&mf1, &mf2, fn_out_consume, &ecbdata,
2284                               &xpp, &xecfg);
2285                 if (o->word_diff)
2286                         free_diff_words_data(&ecbdata);
2287                 if (textconv_one)
2288                         free(mf1.ptr);
2289                 if (textconv_two)
2290                         free(mf2.ptr);
2291                 xdiff_clear_find_func(&xecfg);
2292         }
2293
2294  free_ab_and_return:
2295         strbuf_release(&header);
2296         diff_free_filespec_data(one);
2297         diff_free_filespec_data(two);
2298         free(a_one);
2299         free(b_two);
2300         return;
2301 }
2302
2303 static void builtin_diffstat(const char *name_a, const char *name_b,
2304                              struct diff_filespec *one,
2305                              struct diff_filespec *two,
2306                              struct diffstat_t *diffstat,
2307                              struct diff_options *o,
2308                              int complete_rewrite)
2309 {
2310         mmfile_t mf1, mf2;
2311         struct diffstat_file *data;
2312
2313         data = diffstat_add(diffstat, name_a, name_b);
2314
2315         if (!one || !two) {
2316                 data->is_unmerged = 1;
2317                 return;
2318         }
2319
2320         if (diff_filespec_is_binary(one) || diff_filespec_is_binary(two)) {
2321                 data->is_binary = 1;
2322                 data->added = diff_filespec_size(two);
2323                 data->deleted = diff_filespec_size(one);
2324         }
2325
2326         else if (complete_rewrite) {
2327                 diff_populate_filespec(one, 0);
2328                 diff_populate_filespec(two, 0);
2329                 data->deleted = count_lines(one->data, one->size);
2330                 data->added = count_lines(two->data, two->size);
2331         }
2332
2333         else {
2334                 /* Crazy xdl interfaces.. */
2335                 xpparam_t xpp;
2336                 xdemitconf_t xecfg;
2337
2338                 if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
2339                         die("unable to read files to diff");
2340
2341                 memset(&xpp, 0, sizeof(xpp));
2342                 memset(&xecfg, 0, sizeof(xecfg));
2343                 xpp.flags = o->xdl_opts;
2344                 xecfg.ctxlen = o->context;
2345                 xecfg.interhunkctxlen = o->interhunkcontext;
2346                 xdi_diff_outf(&mf1, &mf2, diffstat_consume, diffstat,
2347                               &xpp, &xecfg);
2348         }
2349
2350         diff_free_filespec_data(one);
2351         diff_free_filespec_data(two);
2352 }
2353
2354 static void builtin_checkdiff(const char *name_a, const char *name_b,
2355                               const char *attr_path,
2356                               struct diff_filespec *one,
2357                               struct diff_filespec *two,
2358                               struct diff_options *o)
2359 {
2360         mmfile_t mf1, mf2;
2361         struct checkdiff_t data;
2362
2363         if (!two)
2364                 return;
2365
2366         memset(&data, 0, sizeof(data));
2367         data.filename = name_b ? name_b : name_a;
2368         data.lineno = 0;
2369         data.o = o;
2370         data.ws_rule = whitespace_rule(attr_path);
2371         data.conflict_marker_size = ll_merge_marker_size(attr_path);
2372
2373         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
2374                 die("unable to read files to diff");
2375
2376         /*
2377          * All the other codepaths check both sides, but not checking
2378          * the "old" side here is deliberate.  We are checking the newly
2379          * introduced changes, and as long as the "new" side is text, we
2380          * can and should check what it introduces.
2381          */
2382         if (diff_filespec_is_binary(two))
2383                 goto free_and_return;
2384         else {
2385                 /* Crazy xdl interfaces.. */
2386                 xpparam_t xpp;
2387                 xdemitconf_t xecfg;
2388
2389                 memset(&xpp, 0, sizeof(xpp));
2390                 memset(&xecfg, 0, sizeof(xecfg));
2391                 xecfg.ctxlen = 1; /* at least one context line */
2392                 xpp.flags = 0;
2393                 xdi_diff_outf(&mf1, &mf2, checkdiff_consume, &data,
2394                               &xpp, &xecfg);
2395
2396                 if (data.ws_rule & WS_BLANK_AT_EOF) {
2397                         struct emit_callback ecbdata;
2398                         int blank_at_eof;
2399
2400                         ecbdata.ws_rule = data.ws_rule;
2401                         check_blank_at_eof(&mf1, &mf2, &ecbdata);
2402                         blank_at_eof = ecbdata.blank_at_eof_in_postimage;
2403
2404                         if (blank_at_eof) {
2405                                 static char *err;
2406                                 if (!err)
2407                                         err = whitespace_error_string(WS_BLANK_AT_EOF);
2408                                 fprintf(o->file, "%s:%d: %s.\n",
2409                                         data.filename, blank_at_eof, err);
2410                                 data.status = 1; /* report errors */
2411                         }
2412                 }
2413         }
2414  free_and_return:
2415         diff_free_filespec_data(one);
2416         diff_free_filespec_data(two);
2417         if (data.status)
2418                 DIFF_OPT_SET(o, CHECK_FAILED);
2419 }
2420
2421 struct diff_filespec *alloc_filespec(const char *path)
2422 {
2423         int namelen = strlen(path);
2424         struct diff_filespec *spec = xmalloc(sizeof(*spec) + namelen + 1);
2425
2426         memset(spec, 0, sizeof(*spec));
2427         spec->path = (char *)(spec + 1);
2428         memcpy(spec->path, path, namelen+1);
2429         spec->count = 1;
2430         spec->is_binary = -1;
2431         return spec;
2432 }
2433
2434 void free_filespec(struct diff_filespec *spec)
2435 {
2436         if (!--spec->count) {
2437                 diff_free_filespec_data(spec);
2438                 free(spec);
2439         }
2440 }
2441
2442 void fill_filespec(struct diff_filespec *spec, const unsigned char *sha1,
2443                    unsigned short mode)
2444 {
2445         if (mode) {
2446                 spec->mode = canon_mode(mode);
2447                 hashcpy(spec->sha1, sha1);
2448                 spec->sha1_valid = !is_null_sha1(sha1);
2449         }
2450 }
2451
2452 /*
2453  * Given a name and sha1 pair, if the index tells us the file in
2454  * the work tree has that object contents, return true, so that
2455  * prepare_temp_file() does not have to inflate and extract.
2456  */
2457 static int reuse_worktree_file(const char *name, const unsigned char *sha1, int want_file)
2458 {
2459         struct cache_entry *ce;
2460         struct stat st;
2461         int pos, len;
2462
2463         /*
2464          * We do not read the cache ourselves here, because the
2465          * benchmark with my previous version that always reads cache
2466          * shows that it makes things worse for diff-tree comparing
2467          * two linux-2.6 kernel trees in an already checked out work
2468          * tree.  This is because most diff-tree comparisons deal with
2469          * only a small number of files, while reading the cache is
2470          * expensive for a large project, and its cost outweighs the
2471          * savings we get by not inflating the object to a temporary
2472          * file.  Practically, this code only helps when we are used
2473          * by diff-cache --cached, which does read the cache before
2474          * calling us.
2475          */
2476         if (!active_cache)
2477                 return 0;
2478
2479         /* We want to avoid the working directory if our caller
2480          * doesn't need the data in a normal file, this system
2481          * is rather slow with its stat/open/mmap/close syscalls,
2482          * and the object is contained in a pack file.  The pack
2483          * is probably already open and will be faster to obtain
2484          * the data through than the working directory.  Loose
2485          * objects however would tend to be slower as they need
2486          * to be individually opened and inflated.
2487          */
2488         if (!FAST_WORKING_DIRECTORY && !want_file && has_sha1_pack(sha1))
2489                 return 0;
2490
2491         len = strlen(name);
2492         pos = cache_name_pos(name, len);
2493         if (pos < 0)
2494                 return 0;
2495         ce = active_cache[pos];
2496
2497         /*
2498          * This is not the sha1 we are looking for, or
2499          * unreusable because it is not a regular file.
2500          */
2501         if (hashcmp(sha1, ce->sha1) || !S_ISREG(ce->ce_mode))
2502                 return 0;
2503
2504         /*
2505          * If ce is marked as "assume unchanged", there is no
2506          * guarantee that work tree matches what we are looking for.
2507          */
2508         if ((ce->ce_flags & CE_VALID) || ce_skip_worktree(ce))
2509                 return 0;
2510
2511         /*
2512          * If ce matches the file in the work tree, we can reuse it.
2513          */
2514         if (ce_uptodate(ce) ||
2515             (!lstat(name, &st) && !ce_match_stat(ce, &st, 0)))
2516                 return 1;
2517
2518         return 0;
2519 }
2520
2521 static int populate_from_stdin(struct diff_filespec *s)
2522 {
2523         struct strbuf buf = STRBUF_INIT;
2524         size_t size = 0;
2525
2526         if (strbuf_read(&buf, 0, 0) < 0)
2527                 return error("error while reading from stdin %s",
2528                                      strerror(errno));
2529
2530         s->should_munmap = 0;
2531         s->data = strbuf_detach(&buf, &size);
2532         s->size = size;
2533         s->should_free = 1;
2534         return 0;
2535 }
2536
2537 static int diff_populate_gitlink(struct diff_filespec *s, int size_only)
2538 {
2539         int len;
2540         char *data = xmalloc(100), *dirty = "";
2541
2542         /* Are we looking at the work tree? */
2543         if (s->dirty_submodule)
2544                 dirty = "-dirty";
2545
2546         len = snprintf(data, 100,
2547                        "Subproject commit %s%s\n", sha1_to_hex(s->sha1), dirty);
2548         s->data = data;
2549         s->size = len;
2550         s->should_free = 1;
2551         if (size_only) {
2552                 s->data = NULL;
2553                 free(data);
2554         }
2555         return 0;
2556 }
2557
2558 /*
2559  * While doing rename detection and pickaxe operation, we may need to
2560  * grab the data for the blob (or file) for our own in-core comparison.
2561  * diff_filespec has data and size fields for this purpose.
2562  */
2563 int diff_populate_filespec(struct diff_filespec *s, int size_only)
2564 {
2565         int err = 0;
2566         if (!DIFF_FILE_VALID(s))
2567                 die("internal error: asking to populate invalid file.");
2568         if (S_ISDIR(s->mode))
2569                 return -1;
2570
2571         if (s->data)
2572                 return 0;
2573
2574         if (size_only && 0 < s->size)
2575                 return 0;
2576
2577         if (S_ISGITLINK(s->mode))
2578                 return diff_populate_gitlink(s, size_only);
2579
2580         if (!s->sha1_valid ||
2581             reuse_worktree_file(s->path, s->sha1, 0)) {
2582                 struct strbuf buf = STRBUF_INIT;
2583                 struct stat st;
2584                 int fd;
2585
2586                 if (!strcmp(s->path, "-"))
2587                         return populate_from_stdin(s);
2588
2589                 if (lstat(s->path, &st) < 0) {
2590                         if (errno == ENOENT) {
2591                         err_empty:
2592                                 err = -1;
2593                         empty:
2594                                 s->data = (char *)"";
2595                                 s->size = 0;
2596                                 return err;
2597                         }
2598                 }
2599                 s->size = xsize_t(st.st_size);
2600                 if (!s->size)
2601                         goto empty;
2602                 if (S_ISLNK(st.st_mode)) {
2603                         struct strbuf sb = STRBUF_INIT;
2604
2605                         if (strbuf_readlink(&sb, s->path, s->size))
2606                                 goto err_empty;
2607                         s->size = sb.len;
2608                         s->data = strbuf_detach(&sb, NULL);
2609                         s->should_free = 1;
2610                         return 0;
2611                 }
2612                 if (size_only)
2613                         return 0;
2614                 fd = open(s->path, O_RDONLY);
2615                 if (fd < 0)
2616                         goto err_empty;
2617                 s->data = xmmap(NULL, s->size, PROT_READ, MAP_PRIVATE, fd, 0);
2618                 close(fd);
2619                 s->should_munmap = 1;
2620
2621                 /*
2622                  * Convert from working tree format to canonical git format
2623                  */
2624                 if (convert_to_git(s->path, s->data, s->size, &buf, safe_crlf)) {
2625                         size_t size = 0;
2626                         munmap(s->data, s->size);
2627                         s->should_munmap = 0;
2628                         s->data = strbuf_detach(&buf, &size);
2629                         s->size = size;
2630                         s->should_free = 1;
2631                 }
2632         }
2633         else {
2634                 enum object_type type;
2635                 if (size_only) {
2636                         type = sha1_object_info(s->sha1, &s->size);
2637                         if (type < 0)
2638                                 die("unable to read %s", sha1_to_hex(s->sha1));
2639                 } else {
2640                         s->data = read_sha1_file(s->sha1, &type, &s->size);
2641                         if (!s->data)
2642                                 die("unable to read %s", sha1_to_hex(s->sha1));
2643                         s->should_free = 1;
2644                 }
2645         }
2646         return 0;
2647 }
2648
2649 void diff_free_filespec_blob(struct diff_filespec *s)
2650 {
2651         if (s->should_free)
2652                 free(s->data);
2653         else if (s->should_munmap)
2654                 munmap(s->data, s->size);
2655
2656         if (s->should_free || s->should_munmap) {
2657                 s->should_free = s->should_munmap = 0;
2658                 s->data = NULL;
2659         }
2660 }
2661
2662 void diff_free_filespec_data(struct diff_filespec *s)
2663 {
2664         diff_free_filespec_blob(s);
2665         free(s->cnt_data);
2666         s->cnt_data = NULL;
2667 }
2668
2669 static void prep_temp_blob(const char *path, struct diff_tempfile *temp,
2670                            void *blob,
2671                            unsigned long size,
2672                            const unsigned char *sha1,
2673                            int mode)
2674 {
2675         int fd;
2676         struct strbuf buf = STRBUF_INIT;
2677         struct strbuf template = STRBUF_INIT;
2678         char *path_dup = xstrdup(path);
2679         const char *base = basename(path_dup);
2680
2681         /* Generate "XXXXXX_basename.ext" */
2682         strbuf_addstr(&template, "XXXXXX_");
2683         strbuf_addstr(&template, base);
2684
2685         fd = git_mkstemps(temp->tmp_path, PATH_MAX, template.buf,
2686                         strlen(base) + 1);
2687         if (fd < 0)
2688                 die_errno("unable to create temp-file");
2689         if (convert_to_working_tree(path,
2690                         (const char *)blob, (size_t)size, &buf)) {
2691                 blob = buf.buf;
2692                 size = buf.len;
2693         }
2694         if (write_in_full(fd, blob, size) != size)
2695                 die_errno("unable to write temp-file");
2696         close(fd);
2697         temp->name = temp->tmp_path;
2698         strcpy(temp->hex, sha1_to_hex(sha1));
2699         temp->hex[40] = 0;
2700         sprintf(temp->mode, "%06o", mode);
2701         strbuf_release(&buf);
2702         strbuf_release(&template);
2703         free(path_dup);
2704 }
2705
2706 static struct diff_tempfile *prepare_temp_file(const char *name,
2707                 struct diff_filespec *one)
2708 {
2709         struct diff_tempfile *temp = claim_diff_tempfile();
2710
2711         if (!DIFF_FILE_VALID(one)) {
2712         not_a_valid_file:
2713                 /* A '-' entry produces this for file-2, and
2714                  * a '+' entry produces this for file-1.
2715                  */
2716                 temp->name = "/dev/null";
2717                 strcpy(temp->hex, ".");
2718                 strcpy(temp->mode, ".");
2719                 return temp;
2720         }
2721
2722         if (!remove_tempfile_installed) {
2723                 atexit(remove_tempfile);
2724                 sigchain_push_common(remove_tempfile_on_signal);
2725                 remove_tempfile_installed = 1;
2726         }
2727
2728         if (!one->sha1_valid ||
2729             reuse_worktree_file(name, one->sha1, 1)) {
2730                 struct stat st;
2731                 if (lstat(name, &st) < 0) {
2732                         if (errno == ENOENT)
2733                                 goto not_a_valid_file;
2734                         die_errno("stat(%s)", name);
2735                 }
2736                 if (S_ISLNK(st.st_mode)) {
2737                         struct strbuf sb = STRBUF_INIT;
2738                         if (strbuf_readlink(&sb, name, st.st_size) < 0)
2739                                 die_errno("readlink(%s)", name);
2740                         prep_temp_blob(name, temp, sb.buf, sb.len,
2741                                        (one->sha1_valid ?
2742                                         one->sha1 : null_sha1),
2743                                        (one->sha1_valid ?
2744                                         one->mode : S_IFLNK));
2745                         strbuf_release(&sb);
2746                 }
2747                 else {
2748                         /* we can borrow from the file in the work tree */
2749                         temp->name = name;
2750                         if (!one->sha1_valid)
2751                                 strcpy(temp->hex, sha1_to_hex(null_sha1));
2752                         else
2753                                 strcpy(temp->hex, sha1_to_hex(one->sha1));
2754                         /* Even though we may sometimes borrow the
2755                          * contents from the work tree, we always want
2756                          * one->mode.  mode is trustworthy even when
2757                          * !(one->sha1_valid), as long as
2758                          * DIFF_FILE_VALID(one).
2759                          */
2760                         sprintf(temp->mode, "%06o", one->mode);
2761                 }
2762                 return temp;
2763         }
2764         else {
2765                 if (diff_populate_filespec(one, 0))
2766                         die("cannot read data blob for %s", one->path);
2767                 prep_temp_blob(name, temp, one->data, one->size,
2768                                one->sha1, one->mode);
2769         }
2770         return temp;
2771 }
2772
2773 /* An external diff command takes:
2774  *
2775  * diff-cmd name infile1 infile1-sha1 infile1-mode \
2776  *               infile2 infile2-sha1 infile2-mode [ rename-to ]
2777  *
2778  */
2779 static void run_external_diff(const char *pgm,
2780                               const char *name,
2781                               const char *other,
2782                               struct diff_filespec *one,
2783                               struct diff_filespec *two,
2784                               const char *xfrm_msg,
2785                               int complete_rewrite)
2786 {
2787         const char *spawn_arg[10];
2788         int retval;
2789         const char **arg = &spawn_arg[0];
2790
2791         if (one && two) {
2792                 struct diff_tempfile *temp_one, *temp_two;
2793                 const char *othername = (other ? other : name);
2794                 temp_one = prepare_temp_file(name, one);
2795                 temp_two = prepare_temp_file(othername, two);
2796                 *arg++ = pgm;
2797                 *arg++ = name;
2798                 *arg++ = temp_one->name;
2799                 *arg++ = temp_one->hex;
2800                 *arg++ = temp_one->mode;
2801                 *arg++ = temp_two->name;
2802                 *arg++ = temp_two->hex;
2803                 *arg++ = temp_two->mode;
2804                 if (other) {
2805                         *arg++ = other;
2806                         *arg++ = xfrm_msg;
2807                 }
2808         } else {
2809                 *arg++ = pgm;
2810                 *arg++ = name;
2811         }
2812         *arg = NULL;
2813         fflush(NULL);
2814         retval = run_command_v_opt(spawn_arg, RUN_USING_SHELL);
2815         remove_tempfile();
2816         if (retval) {
2817                 fprintf(stderr, "external diff died, stopping at %s.\n", name);
2818                 exit(1);
2819         }
2820 }
2821
2822 static int similarity_index(struct diff_filepair *p)
2823 {
2824         return p->score * 100 / MAX_SCORE;
2825 }
2826
2827 static void fill_metainfo(struct strbuf *msg,
2828                           const char *name,
2829                           const char *other,
2830                           struct diff_filespec *one,
2831                           struct diff_filespec *two,
2832                           struct diff_options *o,
2833                           struct diff_filepair *p,
2834                           int *must_show_header,
2835                           int use_color)
2836 {
2837         const char *set = diff_get_color(use_color, DIFF_METAINFO);
2838         const char *reset = diff_get_color(use_color, DIFF_RESET);
2839         struct strbuf *msgbuf;
2840         char *line_prefix = "";
2841
2842         *must_show_header = 1;
2843         if (o->output_prefix) {
2844                 msgbuf = o->output_prefix(o, o->output_prefix_data);
2845                 line_prefix = msgbuf->buf;
2846         }
2847         strbuf_init(msg, PATH_MAX * 2 + 300);
2848         switch (p->status) {
2849         case DIFF_STATUS_COPIED:
2850                 strbuf_addf(msg, "%s%ssimilarity index %d%%",
2851                             line_prefix, set, similarity_index(p));
2852                 strbuf_addf(msg, "%s\n%s%scopy from ",
2853                             reset,  line_prefix, set);
2854                 quote_c_style(name, msg, NULL, 0);
2855                 strbuf_addf(msg, "%s\n%s%scopy to ", reset, line_prefix, set);
2856                 quote_c_style(other, msg, NULL, 0);
2857                 strbuf_addf(msg, "%s\n", reset);
2858                 break;
2859         case DIFF_STATUS_RENAMED:
2860                 strbuf_addf(msg, "%s%ssimilarity index %d%%",
2861                             line_prefix, set, similarity_index(p));
2862                 strbuf_addf(msg, "%s\n%s%srename from ",
2863                             reset, line_prefix, set);
2864                 quote_c_style(name, msg, NULL, 0);
2865                 strbuf_addf(msg, "%s\n%s%srename to ",
2866                             reset, line_prefix, set);
2867                 quote_c_style(other, msg, NULL, 0);
2868                 strbuf_addf(msg, "%s\n", reset);
2869                 break;
2870         case DIFF_STATUS_MODIFIED:
2871                 if (p->score) {
2872                         strbuf_addf(msg, "%s%sdissimilarity index %d%%%s\n",
2873                                     line_prefix,
2874                                     set, similarity_index(p), reset);
2875                         break;
2876                 }
2877                 /* fallthru */
2878         default:
2879                 *must_show_header = 0;
2880         }
2881         if (one && two && hashcmp(one->sha1, two->sha1)) {
2882                 int abbrev = DIFF_OPT_TST(o, FULL_INDEX) ? 40 : DEFAULT_ABBREV;
2883
2884                 if (DIFF_OPT_TST(o, BINARY)) {
2885                         mmfile_t mf;
2886                         if ((!fill_mmfile(&mf, one) && diff_filespec_is_binary(one)) ||
2887                             (!fill_mmfile(&mf, two) && diff_filespec_is_binary(two)))
2888                                 abbrev = 40;
2889                 }
2890                 strbuf_addf(msg, "%s%sindex %s..", line_prefix, set,
2891                             find_unique_abbrev(one->sha1, abbrev));
2892                 strbuf_addstr(msg, find_unique_abbrev(two->sha1, abbrev));
2893                 if (one->mode == two->mode)
2894                         strbuf_addf(msg, " %06o", one->mode);
2895                 strbuf_addf(msg, "%s\n", reset);
2896         }
2897 }
2898
2899 static void run_diff_cmd(const char *pgm,
2900                          const char *name,
2901                          const char *other,
2902                          const char *attr_path,
2903                          struct diff_filespec *one,
2904                          struct diff_filespec *two,
2905                          struct strbuf *msg,
2906                          struct diff_options *o,
2907                          struct diff_filepair *p)
2908 {
2909         const char *xfrm_msg = NULL;
2910         int complete_rewrite = (p->status == DIFF_STATUS_MODIFIED) && p->score;
2911         int must_show_header = 0;
2912
2913         if (!DIFF_OPT_TST(o, ALLOW_EXTERNAL))
2914                 pgm = NULL;
2915         else {
2916                 struct userdiff_driver *drv = userdiff_find_by_path(attr_path);
2917                 if (drv && drv->external)
2918                         pgm = drv->external;
2919         }
2920
2921         if (msg) {
2922                 /*
2923                  * don't use colors when the header is intended for an
2924                  * external diff driver
2925                  */
2926                 fill_metainfo(msg, name, other, one, two, o, p,
2927                               &must_show_header,
2928                               want_color(o->use_color) && !pgm);
2929                 xfrm_msg = msg->len ? msg->buf : NULL;
2930         }
2931
2932         if (pgm) {
2933                 run_external_diff(pgm, name, other, one, two, xfrm_msg,
2934                                   complete_rewrite);
2935                 return;
2936         }
2937         if (one && two)
2938                 builtin_diff(name, other ? other : name,
2939                              one, two, xfrm_msg, must_show_header,
2940                              o, complete_rewrite);
2941         else
2942                 fprintf(o->file, "* Unmerged path %s\n", name);
2943 }
2944
2945 static void diff_fill_sha1_info(struct diff_filespec *one)
2946 {
2947         if (DIFF_FILE_VALID(one)) {
2948                 if (!one->sha1_valid) {
2949                         struct stat st;
2950                         if (!strcmp(one->path, "-")) {
2951                                 hashcpy(one->sha1, null_sha1);
2952                                 return;
2953                         }
2954                         if (lstat(one->path, &st) < 0)
2955                                 die_errno("stat '%s'", one->path);
2956                         if (index_path(one->sha1, one->path, &st, 0))
2957                                 die("cannot hash %s", one->path);
2958                 }
2959         }
2960         else
2961                 hashclr(one->sha1);
2962 }
2963
2964 static void strip_prefix(int prefix_length, const char **namep, const char **otherp)
2965 {
2966         /* Strip the prefix but do not molest /dev/null and absolute paths */
2967         if (*namep && **namep != '/') {
2968                 *namep += prefix_length;
2969                 if (**namep == '/')
2970                         ++*namep;
2971         }
2972         if (*otherp && **otherp != '/') {
2973                 *otherp += prefix_length;
2974                 if (**otherp == '/')
2975                         ++*otherp;
2976         }
2977 }
2978
2979 static void run_diff(struct diff_filepair *p, struct diff_options *o)
2980 {
2981         const char *pgm = external_diff();
2982         struct strbuf msg;
2983         struct diff_filespec *one = p->one;
2984         struct diff_filespec *two = p->two;
2985         const char *name;
2986         const char *other;
2987         const char *attr_path;
2988
2989         name  = p->one->path;
2990         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
2991         attr_path = name;
2992         if (o->prefix_length)
2993                 strip_prefix(o->prefix_length, &name, &other);
2994
2995         if (DIFF_PAIR_UNMERGED(p)) {
2996                 run_diff_cmd(pgm, name, NULL, attr_path,
2997                              NULL, NULL, NULL, o, p);
2998                 return;
2999         }
3000
3001         diff_fill_sha1_info(one);
3002         diff_fill_sha1_info(two);
3003
3004         if (!pgm &&
3005             DIFF_FILE_VALID(one) && DIFF_FILE_VALID(two) &&
3006             (S_IFMT & one->mode) != (S_IFMT & two->mode)) {
3007                 /*
3008                  * a filepair that changes between file and symlink
3009                  * needs to be split into deletion and creation.
3010                  */
3011                 struct diff_filespec *null = alloc_filespec(two->path);
3012                 run_diff_cmd(NULL, name, other, attr_path,
3013                              one, null, &msg, o, p);
3014                 free(null);
3015                 strbuf_release(&msg);
3016
3017                 null = alloc_filespec(one->path);
3018                 run_diff_cmd(NULL, name, other, attr_path,
3019                              null, two, &msg, o, p);
3020                 free(null);
3021         }
3022         else
3023                 run_diff_cmd(pgm, name, other, attr_path,
3024                              one, two, &msg, o, p);
3025
3026         strbuf_release(&msg);
3027 }
3028
3029 static void run_diffstat(struct diff_filepair *p, struct diff_options *o,
3030                          struct diffstat_t *diffstat)
3031 {
3032         const char *name;
3033         const char *other;
3034         int complete_rewrite = 0;
3035
3036         if (DIFF_PAIR_UNMERGED(p)) {
3037                 /* unmerged */
3038                 builtin_diffstat(p->one->path, NULL, NULL, NULL, diffstat, o, 0);
3039                 return;
3040         }
3041
3042         name = p->one->path;
3043         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
3044
3045         if (o->prefix_length)
3046                 strip_prefix(o->prefix_length, &name, &other);
3047
3048         diff_fill_sha1_info(p->one);
3049         diff_fill_sha1_info(p->two);
3050
3051         if (p->status == DIFF_STATUS_MODIFIED && p->score)
3052                 complete_rewrite = 1;
3053         builtin_diffstat(name, other, p->one, p->two, diffstat, o, complete_rewrite);
3054 }
3055
3056 static void run_checkdiff(struct diff_filepair *p, struct diff_options *o)
3057 {
3058         const char *name;
3059         const char *other;
3060         const char *attr_path;
3061
3062         if (DIFF_PAIR_UNMERGED(p)) {
3063                 /* unmerged */
3064                 return;
3065         }
3066
3067         name = p->one->path;
3068         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
3069         attr_path = other ? other : name;
3070
3071         if (o->prefix_length)
3072                 strip_prefix(o->prefix_length, &name, &other);
3073
3074         diff_fill_sha1_info(p->one);
3075         diff_fill_sha1_info(p->two);
3076
3077         builtin_checkdiff(name, other, attr_path, p->one, p->two, o);
3078 }
3079
3080 void diff_setup(struct diff_options *options)
3081 {
3082         memcpy(options, &default_diff_options, sizeof(*options));
3083
3084         options->file = stdout;
3085
3086         options->line_termination = '\n';
3087         options->break_opt = -1;
3088         options->rename_limit = -1;
3089         options->dirstat_permille = diff_dirstat_permille_default;
3090         options->context = 3;
3091
3092         options->change = diff_change;
3093         options->add_remove = diff_addremove;
3094         options->use_color = diff_use_color_default;
3095         options->detect_rename = diff_detect_rename_default;
3096
3097         if (diff_no_prefix) {
3098                 options->a_prefix = options->b_prefix = "";
3099         } else if (!diff_mnemonic_prefix) {
3100                 options->a_prefix = "a/";
3101                 options->b_prefix = "b/";
3102         }
3103 }
3104
3105 int diff_setup_done(struct diff_options *options)
3106 {
3107         int count = 0;
3108
3109         if (options->output_format & DIFF_FORMAT_NAME)
3110                 count++;
3111         if (options->output_format & DIFF_FORMAT_NAME_STATUS)
3112                 count++;
3113         if (options->output_format & DIFF_FORMAT_CHECKDIFF)
3114                 count++;
3115         if (options->output_format & DIFF_FORMAT_NO_OUTPUT)
3116                 count++;
3117         if (count > 1)
3118                 die("--name-only, --name-status, --check and -s are mutually exclusive");
3119
3120         /*
3121          * Most of the time we can say "there are changes"
3122          * only by checking if there are changed paths, but
3123          * --ignore-whitespace* options force us to look
3124          * inside contents.
3125          */
3126
3127         if (DIFF_XDL_TST(options, IGNORE_WHITESPACE) ||
3128             DIFF_XDL_TST(options, IGNORE_WHITESPACE_CHANGE) ||
3129             DIFF_XDL_TST(options, IGNORE_WHITESPACE_AT_EOL))
3130                 DIFF_OPT_SET(options, DIFF_FROM_CONTENTS);
3131         else
3132                 DIFF_OPT_CLR(options, DIFF_FROM_CONTENTS);
3133
3134         if (DIFF_OPT_TST(options, FIND_COPIES_HARDER))
3135                 options->detect_rename = DIFF_DETECT_COPY;
3136
3137         if (!DIFF_OPT_TST(options, RELATIVE_NAME))
3138                 options->prefix = NULL;
3139         if (options->prefix)
3140                 options->prefix_length = strlen(options->prefix);
3141         else
3142                 options->prefix_length = 0;
3143
3144         if (options->output_format & (DIFF_FORMAT_NAME |
3145                                       DIFF_FORMAT_NAME_STATUS |
3146                                       DIFF_FORMAT_CHECKDIFF |
3147                                       DIFF_FORMAT_NO_OUTPUT))
3148                 options->output_format &= ~(DIFF_FORMAT_RAW |
3149                                             DIFF_FORMAT_NUMSTAT |
3150                                             DIFF_FORMAT_DIFFSTAT |
3151                                             DIFF_FORMAT_SHORTSTAT |
3152                                             DIFF_FORMAT_DIRSTAT |
3153                                             DIFF_FORMAT_SUMMARY |
3154                                             DIFF_FORMAT_PATCH);
3155
3156         /*
3157          * These cases always need recursive; we do not drop caller-supplied
3158          * recursive bits for other formats here.
3159          */
3160         if (options->output_format & (DIFF_FORMAT_PATCH |
3161                                       DIFF_FORMAT_NUMSTAT |
3162                                       DIFF_FORMAT_DIFFSTAT |
3163                                       DIFF_FORMAT_SHORTSTAT |
3164                                       DIFF_FORMAT_DIRSTAT |
3165                                       DIFF_FORMAT_SUMMARY |
3166                                       DIFF_FORMAT_CHECKDIFF))
3167                 DIFF_OPT_SET(options, RECURSIVE);
3168         /*
3169          * Also pickaxe would not work very well if you do not say recursive
3170          */
3171         if (options->pickaxe)
3172                 DIFF_OPT_SET(options, RECURSIVE);
3173         /*
3174          * When patches are generated, submodules diffed against the work tree
3175          * must be checked for dirtiness too so it can be shown in the output
3176          */
3177         if (options->output_format & DIFF_FORMAT_PATCH)
3178                 DIFF_OPT_SET(options, DIRTY_SUBMODULES);
3179
3180         if (options->detect_rename && options->rename_limit < 0)
3181                 options->rename_limit = diff_rename_limit_default;
3182         if (options->setup & DIFF_SETUP_USE_CACHE) {
3183                 if (!active_cache)
3184                         /* read-cache does not die even when it fails
3185                          * so it is safe for us to do this here.  Also
3186                          * it does not smudge active_cache or active_nr
3187                          * when it fails, so we do not have to worry about
3188                          * cleaning it up ourselves either.
3189                          */
3190                         read_cache();
3191         }
3192         if (options->abbrev <= 0 || 40 < options->abbrev)
3193                 options->abbrev = 40; /* full */
3194
3195         /*
3196          * It does not make sense to show the first hit we happened
3197          * to have found.  It does not make sense not to return with
3198          * exit code in such a case either.
3199          */
3200         if (DIFF_OPT_TST(options, QUICK)) {
3201                 options->output_format = DIFF_FORMAT_NO_OUTPUT;
3202                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
3203         }
3204
3205         return 0;
3206 }
3207
3208 static int opt_arg(const char *arg, int arg_short, const char *arg_long, int *val)
3209 {
3210         char c, *eq;
3211         int len;
3212
3213         if (*arg != '-')
3214                 return 0;
3215         c = *++arg;
3216         if (!c)
3217                 return 0;
3218         if (c == arg_short) {
3219                 c = *++arg;
3220                 if (!c)
3221                         return 1;
3222                 if (val && isdigit(c)) {
3223                         char *end;
3224                         int n = strtoul(arg, &end, 10);
3225                         if (*end)
3226                                 return 0;
3227                         *val = n;
3228                         return 1;
3229                 }
3230                 return 0;
3231         }
3232         if (c != '-')
3233                 return 0;
3234         arg++;
3235         eq = strchr(arg, '=');
3236         if (eq)
3237                 len = eq - arg;
3238         else
3239                 len = strlen(arg);
3240         if (!len || strncmp(arg, arg_long, len))
3241                 return 0;
3242         if (eq) {
3243                 int n;
3244                 char *end;
3245                 if (!isdigit(*++eq))
3246                         return 0;
3247                 n = strtoul(eq, &end, 10);
3248                 if (*end)
3249                         return 0;
3250                 *val = n;
3251         }
3252         return 1;
3253 }
3254
3255 static int diff_scoreopt_parse(const char *opt);
3256
3257 static inline int short_opt(char opt, const char **argv,
3258                             const char **optarg)
3259 {
3260         const char *arg = argv[0];
3261         if (arg[0] != '-' || arg[1] != opt)
3262                 return 0;
3263         if (arg[2] != '\0') {
3264                 *optarg = arg + 2;
3265                 return 1;
3266         }
3267         if (!argv[1])
3268                 die("Option '%c' requires a value", opt);
3269         *optarg = argv[1];
3270         return 2;
3271 }
3272
3273 int parse_long_opt(const char *opt, const char **argv,
3274                    const char **optarg)
3275 {
3276         const char *arg = argv[0];
3277         if (arg[0] != '-' || arg[1] != '-')
3278                 return 0;
3279         arg += strlen("--");
3280         if (prefixcmp(arg, opt))
3281                 return 0;
3282         arg += strlen(opt);
3283         if (*arg == '=') { /* sticked form: --option=value */
3284                 *optarg = arg + 1;
3285                 return 1;
3286         }
3287         if (*arg != '\0')
3288                 return 0;
3289         /* separate form: --option value */
3290         if (!argv[1])
3291                 die("Option '--%s' requires a value", opt);
3292         *optarg = argv[1];
3293         return 2;
3294 }
3295
3296 static int stat_opt(struct diff_options *options, const char **av)
3297 {
3298         const char *arg = av[0];
3299         char *end;
3300         int width = options->stat_width;
3301         int name_width = options->stat_name_width;
3302         int count = options->stat_count;
3303         int argcount = 1;
3304
3305         arg += strlen("--stat");
3306         end = (char *)arg;
3307
3308         switch (*arg) {
3309         case '-':
3310                 if (!prefixcmp(arg, "-width")) {
3311                         arg += strlen("-width");
3312                         if (*arg == '=')
3313                                 width = strtoul(arg + 1, &end, 10);
3314                         else if (!*arg && !av[1])
3315                                 die("Option '--stat-width' requires a value");
3316                         else if (!*arg) {
3317                                 width = strtoul(av[1], &end, 10);
3318                                 argcount = 2;
3319                         }
3320                 } else if (!prefixcmp(arg, "-name-width")) {
3321                         arg += strlen("-name-width");
3322                         if (*arg == '=')
3323                                 name_width = strtoul(arg + 1, &end, 10);
3324                         else if (!*arg && !av[1])
3325                                 die("Option '--stat-name-width' requires a value");
3326                         else if (!*arg) {
3327                                 name_width = strtoul(av[1], &end, 10);
3328                                 argcount = 2;
3329                         }
3330                 } else if (!prefixcmp(arg, "-count")) {
3331                         arg += strlen("-count");
3332                         if (*arg == '=')
3333                                 count = strtoul(arg + 1, &end, 10);
3334                         else if (!*arg && !av[1])
3335                                 die("Option '--stat-count' requires a value");
3336                         else if (!*arg) {
3337                                 count = strtoul(av[1], &end, 10);
3338                                 argcount = 2;
3339                         }
3340                 }
3341                 break;
3342         case '=':
3343                 width = strtoul(arg+1, &end, 10);
3344                 if (*end == ',')
3345                         name_width = strtoul(end+1, &end, 10);
3346                 if (*end == ',')
3347                         count = strtoul(end+1, &end, 10);
3348         }
3349
3350         /* Important! This checks all the error cases! */
3351         if (*end)
3352                 return 0;
3353         options->output_format |= DIFF_FORMAT_DIFFSTAT;
3354         options->stat_name_width = name_width;
3355         options->stat_width = width;
3356         options->stat_count = count;
3357         return argcount;
3358 }
3359
3360 static int parse_dirstat_opt(struct diff_options *options, const char *params)
3361 {
3362         struct strbuf errmsg = STRBUF_INIT;
3363         if (parse_dirstat_params(options, params, &errmsg))
3364                 die(_("Failed to parse --dirstat/-X option parameter:\n%s"),
3365                     errmsg.buf);
3366         strbuf_release(&errmsg);
3367         /*
3368          * The caller knows a dirstat-related option is given from the command
3369          * line; allow it to say "return this_function();"
3370          */
3371         options->output_format |= DIFF_FORMAT_DIRSTAT;
3372         return 1;
3373 }
3374
3375 int diff_opt_parse(struct diff_options *options, const char **av, int ac)
3376 {
3377         const char *arg = av[0];
3378         const char *optarg;
3379         int argcount;
3380
3381         /* Output format options */
3382         if (!strcmp(arg, "-p") || !strcmp(arg, "-u") || !strcmp(arg, "--patch"))
3383                 options->output_format |= DIFF_FORMAT_PATCH;
3384         else if (opt_arg(arg, 'U', "unified", &options->context))
3385                 options->output_format |= DIFF_FORMAT_PATCH;
3386         else if (!strcmp(arg, "--raw"))
3387                 options->output_format |= DIFF_FORMAT_RAW;
3388         else if (!strcmp(arg, "--patch-with-raw"))
3389                 options->output_format |= DIFF_FORMAT_PATCH | DIFF_FORMAT_RAW;
3390         else if (!strcmp(arg, "--numstat"))
3391                 options->output_format |= DIFF_FORMAT_NUMSTAT;
3392         else if (!strcmp(arg, "--shortstat"))
3393                 options->output_format |= DIFF_FORMAT_SHORTSTAT;
3394         else if (!strcmp(arg, "-X") || !strcmp(arg, "--dirstat"))
3395                 return parse_dirstat_opt(options, "");
3396         else if (!prefixcmp(arg, "-X"))
3397                 return parse_dirstat_opt(options, arg + 2);
3398         else if (!prefixcmp(arg, "--dirstat="))
3399                 return parse_dirstat_opt(options, arg + 10);
3400         else if (!strcmp(arg, "--cumulative"))
3401                 return parse_dirstat_opt(options, "cumulative");
3402         else if (!strcmp(arg, "--dirstat-by-file"))
3403                 return parse_dirstat_opt(options, "files");
3404         else if (!prefixcmp(arg, "--dirstat-by-file=")) {
3405                 parse_dirstat_opt(options, "files");
3406                 return parse_dirstat_opt(options, arg + 18);
3407         }
3408         else if (!strcmp(arg, "--check"))
3409                 options->output_format |= DIFF_FORMAT_CHECKDIFF;
3410         else if (!strcmp(arg, "--summary"))
3411                 options->output_format |= DIFF_FORMAT_SUMMARY;
3412         else if (!strcmp(arg, "--patch-with-stat"))
3413                 options->output_format |= DIFF_FORMAT_PATCH | DIFF_FORMAT_DIFFSTAT;
3414         else if (!strcmp(arg, "--name-only"))
3415                 options->output_format |= DIFF_FORMAT_NAME;
3416         else if (!strcmp(arg, "--name-status"))
3417                 options->output_format |= DIFF_FORMAT_NAME_STATUS;
3418         else if (!strcmp(arg, "-s"))
3419                 options->output_format |= DIFF_FORMAT_NO_OUTPUT;
3420         else if (!prefixcmp(arg, "--stat"))
3421                 /* --stat, --stat-width, --stat-name-width, or --stat-count */
3422                 return stat_opt(options, av);
3423
3424         /* renames options */
3425         else if (!prefixcmp(arg, "-B") || !prefixcmp(arg, "--break-rewrites=") ||
3426                  !strcmp(arg, "--break-rewrites")) {
3427                 if ((options->break_opt = diff_scoreopt_parse(arg)) == -1)
3428                         return error("invalid argument to -B: %s", arg+2);
3429         }
3430         else if (!prefixcmp(arg, "-M") || !prefixcmp(arg, "--find-renames=") ||
3431                  !strcmp(arg, "--find-renames")) {
3432                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
3433                         return error("invalid argument to -M: %s", arg+2);
3434                 options->detect_rename = DIFF_DETECT_RENAME;
3435         }
3436         else if (!strcmp(arg, "-D") || !strcmp(arg, "--irreversible-delete")) {
3437                 options->irreversible_delete = 1;
3438         }
3439         else if (!prefixcmp(arg, "-C") || !prefixcmp(arg, "--find-copies=") ||
3440                  !strcmp(arg, "--find-copies")) {
3441                 if (options->detect_rename == DIFF_DETECT_COPY)
3442                         DIFF_OPT_SET(options, FIND_COPIES_HARDER);
3443                 if ((options->rename_score = diff_scoreopt_parse(arg)) == -1)
3444                         return error("invalid argument to -C: %s", arg+2);
3445                 options->detect_rename = DIFF_DETECT_COPY;
3446         }
3447         else if (!strcmp(arg, "--no-renames"))
3448                 options->detect_rename = 0;
3449         else if (!strcmp(arg, "--relative"))
3450                 DIFF_OPT_SET(options, RELATIVE_NAME);
3451         else if (!prefixcmp(arg, "--relative=")) {
3452                 DIFF_OPT_SET(options, RELATIVE_NAME);
3453                 options->prefix = arg + 11;
3454         }
3455
3456         /* xdiff options */
3457         else if (!strcmp(arg, "--minimal"))
3458                 DIFF_XDL_SET(options, NEED_MINIMAL);
3459         else if (!strcmp(arg, "--no-minimal"))
3460                 DIFF_XDL_CLR(options, NEED_MINIMAL);
3461         else if (!strcmp(arg, "-w") || !strcmp(arg, "--ignore-all-space"))
3462                 DIFF_XDL_SET(options, IGNORE_WHITESPACE);
3463         else if (!strcmp(arg, "-b") || !strcmp(arg, "--ignore-space-change"))
3464                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_CHANGE);
3465         else if (!strcmp(arg, "--ignore-space-at-eol"))
3466                 DIFF_XDL_SET(options, IGNORE_WHITESPACE_AT_EOL);
3467         else if (!strcmp(arg, "--patience"))
3468                 DIFF_XDL_SET(options, PATIENCE_DIFF);
3469         else if (!strcmp(arg, "--histogram"))
3470                 DIFF_XDL_SET(options, HISTOGRAM_DIFF);
3471
3472         /* flags options */
3473         else if (!strcmp(arg, "--binary")) {
3474                 options->output_format |= DIFF_FORMAT_PATCH;
3475                 DIFF_OPT_SET(options, BINARY);
3476         }
3477         else if (!strcmp(arg, "--full-index"))
3478                 DIFF_OPT_SET(options, FULL_INDEX);
3479         else if (!strcmp(arg, "-a") || !strcmp(arg, "--text"))
3480                 DIFF_OPT_SET(options, TEXT);
3481         else if (!strcmp(arg, "-R"))
3482                 DIFF_OPT_SET(options, REVERSE_DIFF);
3483         else if (!strcmp(arg, "--find-copies-harder"))
3484                 DIFF_OPT_SET(options, FIND_COPIES_HARDER);
3485         else if (!strcmp(arg, "--follow"))
3486                 DIFF_OPT_SET(options, FOLLOW_RENAMES);
3487         else if (!strcmp(arg, "--color"))
3488                 options->use_color = 1;
3489         else if (!prefixcmp(arg, "--color=")) {
3490                 int value = git_config_colorbool(NULL, arg+8);
3491                 if (value < 0)
3492                         return error("option `color' expects \"always\", \"auto\", or \"never\"");
3493                 options->use_color = value;
3494         }
3495         else if (!strcmp(arg, "--no-color"))
3496                 options->use_color = 0;
3497         else if (!strcmp(arg, "--color-words")) {
3498                 options->use_color = 1;
3499                 options->word_diff = DIFF_WORDS_COLOR;
3500         }
3501         else if (!prefixcmp(arg, "--color-words=")) {
3502                 options->use_color = 1;
3503                 options->word_diff = DIFF_WORDS_COLOR;
3504                 options->word_regex = arg + 14;
3505         }
3506         else if (!strcmp(arg, "--word-diff")) {
3507                 if (options->word_diff == DIFF_WORDS_NONE)
3508                         options->word_diff = DIFF_WORDS_PLAIN;
3509         }
3510         else if (!prefixcmp(arg, "--word-diff=")) {
3511                 const char *type = arg + 12;
3512                 if (!strcmp(type, "plain"))
3513                         options->word_diff = DIFF_WORDS_PLAIN;
3514                 else if (!strcmp(type, "color")) {
3515                         options->use_color = 1;
3516                         options->word_diff = DIFF_WORDS_COLOR;
3517                 }
3518                 else if (!strcmp(type, "porcelain"))
3519                         options->word_diff = DIFF_WORDS_PORCELAIN;
3520                 else if (!strcmp(type, "none"))
3521                         options->word_diff = DIFF_WORDS_NONE;
3522                 else
3523                         die("bad --word-diff argument: %s", type);
3524         }
3525         else if ((argcount = parse_long_opt("word-diff-regex", av, &optarg))) {
3526                 if (options->word_diff == DIFF_WORDS_NONE)
3527                         options->word_diff = DIFF_WORDS_PLAIN;
3528                 options->word_regex = optarg;
3529                 return argcount;
3530         }
3531         else if (!strcmp(arg, "--exit-code"))
3532                 DIFF_OPT_SET(options, EXIT_WITH_STATUS);
3533         else if (!strcmp(arg, "--quiet"))
3534                 DIFF_OPT_SET(options, QUICK);
3535         else if (!strcmp(arg, "--ext-diff"))
3536                 DIFF_OPT_SET(options, ALLOW_EXTERNAL);
3537         else if (!strcmp(arg, "--no-ext-diff"))
3538                 DIFF_OPT_CLR(options, ALLOW_EXTERNAL);
3539         else if (!strcmp(arg, "--textconv"))
3540                 DIFF_OPT_SET(options, ALLOW_TEXTCONV);
3541         else if (!strcmp(arg, "--no-textconv"))
3542                 DIFF_OPT_CLR(options, ALLOW_TEXTCONV);
3543         else if (!strcmp(arg, "--ignore-submodules")) {
3544                 DIFF_OPT_SET(options, OVERRIDE_SUBMODULE_CONFIG);
3545                 handle_ignore_submodules_arg(options, "all");
3546         } else if (!prefixcmp(arg, "--ignore-submodules=")) {
3547                 DIFF_OPT_SET(options, OVERRIDE_SUBMODULE_CONFIG);
3548                 handle_ignore_submodules_arg(options, arg + 20);
3549         } else if (!strcmp(arg, "--submodule"))
3550                 DIFF_OPT_SET(options, SUBMODULE_LOG);
3551         else if (!prefixcmp(arg, "--submodule=")) {
3552                 if (!strcmp(arg + 12, "log"))
3553                         DIFF_OPT_SET(options, SUBMODULE_LOG);
3554         }
3555
3556         /* misc options */
3557         else if (!strcmp(arg, "-z"))
3558                 options->line_termination = 0;
3559         else if ((argcount = short_opt('l', av, &optarg))) {
3560                 options->rename_limit = strtoul(optarg, NULL, 10);
3561                 return argcount;
3562         }
3563         else if ((argcount = short_opt('S', av, &optarg))) {
3564                 options->pickaxe = optarg;
3565                 options->pickaxe_opts |= DIFF_PICKAXE_KIND_S;
3566                 return argcount;
3567         } else if ((argcount = short_opt('G', av, &optarg))) {
3568                 options->pickaxe = optarg;
3569                 options->pickaxe_opts |= DIFF_PICKAXE_KIND_G;
3570                 return argcount;
3571         }
3572         else if (!strcmp(arg, "--pickaxe-all"))
3573                 options->pickaxe_opts |= DIFF_PICKAXE_ALL;
3574         else if (!strcmp(arg, "--pickaxe-regex"))
3575                 options->pickaxe_opts |= DIFF_PICKAXE_REGEX;
3576         else if ((argcount = short_opt('O', av, &optarg))) {
3577                 options->orderfile = optarg;
3578                 return argcount;
3579         }
3580         else if ((argcount = parse_long_opt("diff-filter", av, &optarg))) {
3581                 options->filter = optarg;
3582                 return argcount;
3583         }
3584         else if (!strcmp(arg, "--abbrev"))
3585                 options->abbrev = DEFAULT_ABBREV;
3586         else if (!prefixcmp(arg, "--abbrev=")) {
3587                 options->abbrev = strtoul(arg + 9, NULL, 10);
3588                 if (options->abbrev < MINIMUM_ABBREV)
3589                         options->abbrev = MINIMUM_ABBREV;
3590                 else if (40 < options->abbrev)
3591                         options->abbrev = 40;
3592         }
3593         else if ((argcount = parse_long_opt("src-prefix", av, &optarg))) {
3594                 options->a_prefix = optarg;
3595                 return argcount;
3596         }
3597         else if ((argcount = parse_long_opt("dst-prefix", av, &optarg))) {
3598                 options->b_prefix = optarg;
3599                 return argcount;
3600         }
3601         else if (!strcmp(arg, "--no-prefix"))
3602                 options->a_prefix = options->b_prefix = "";
3603         else if (opt_arg(arg, '\0', "inter-hunk-context",
3604                          &options->interhunkcontext))
3605                 ;
3606         else if (!strcmp(arg, "-W"))
3607                 DIFF_OPT_SET(options, FUNCCONTEXT);
3608         else if (!strcmp(arg, "--function-context"))
3609                 DIFF_OPT_SET(options, FUNCCONTEXT);
3610         else if (!strcmp(arg, "--no-function-context"))
3611                 DIFF_OPT_CLR(options, FUNCCONTEXT);
3612         else if ((argcount = parse_long_opt("output", av, &optarg))) {
3613                 options->file = fopen(optarg, "w");
3614                 if (!options->file)
3615                         die_errno("Could not open '%s'", optarg);
3616                 options->close_file = 1;
3617                 return argcount;
3618         } else
3619                 return 0;
3620         return 1;
3621 }
3622
3623 int parse_rename_score(const char **cp_p)
3624 {
3625         unsigned long num, scale;
3626         int ch, dot;
3627         const char *cp = *cp_p;
3628
3629         num = 0;
3630         scale = 1;
3631         dot = 0;
3632         for (;;) {
3633                 ch = *cp;
3634                 if ( !dot && ch == '.' ) {
3635                         scale = 1;
3636                         dot = 1;
3637                 } else if ( ch == '%' ) {
3638                         scale = dot ? scale*100 : 100;
3639                         cp++;   /* % is always at the end */
3640                         break;
3641                 } else if ( ch >= '0' && ch <= '9' ) {
3642                         if ( scale < 100000 ) {
3643                                 scale *= 10;
3644                                 num = (num*10) + (ch-'0');
3645                         }
3646                 } else {
3647                         break;
3648                 }
3649                 cp++;
3650         }
3651         *cp_p = cp;
3652
3653         /* user says num divided by scale and we say internally that
3654          * is MAX_SCORE * num / scale.
3655          */
3656         return (int)((num >= scale) ? MAX_SCORE : (MAX_SCORE * num / scale));
3657 }
3658
3659 static int diff_scoreopt_parse(const char *opt)
3660 {
3661         int opt1, opt2, cmd;
3662
3663         if (*opt++ != '-')
3664                 return -1;
3665         cmd = *opt++;
3666         if (cmd == '-') {
3667                 /* convert the long-form arguments into short-form versions */
3668                 if (!prefixcmp(opt, "break-rewrites")) {
3669                         opt += strlen("break-rewrites");
3670                         if (*opt == 0 || *opt++ == '=')
3671                                 cmd = 'B';
3672                 } else if (!prefixcmp(opt, "find-copies")) {
3673                         opt += strlen("find-copies");
3674                         if (*opt == 0 || *opt++ == '=')
3675                                 cmd = 'C';
3676                 } else if (!prefixcmp(opt, "find-renames")) {
3677                         opt += strlen("find-renames");
3678                         if (*opt == 0 || *opt++ == '=')
3679                                 cmd = 'M';
3680                 }
3681         }
3682         if (cmd != 'M' && cmd != 'C' && cmd != 'B')
3683                 return -1; /* that is not a -M, -C nor -B option */
3684
3685         opt1 = parse_rename_score(&opt);
3686         if (cmd != 'B')
3687                 opt2 = 0;
3688         else {
3689                 if (*opt == 0)
3690                         opt2 = 0;
3691                 else if (*opt != '/')
3692                         return -1; /* we expect -B80/99 or -B80 */
3693                 else {
3694                         opt++;
3695                         opt2 = parse_rename_score(&opt);
3696                 }
3697         }
3698         if (*opt != 0)
3699                 return -1;
3700         return opt1 | (opt2 << 16);
3701 }
3702
3703 struct diff_queue_struct diff_queued_diff;
3704
3705 void diff_q(struct diff_queue_struct *queue, struct diff_filepair *dp)
3706 {
3707         if (queue->alloc <= queue->nr) {
3708                 queue->alloc = alloc_nr(queue->alloc);
3709                 queue->queue = xrealloc(queue->queue,
3710                                         sizeof(dp) * queue->alloc);
3711         }
3712         queue->queue[queue->nr++] = dp;
3713 }
3714
3715 struct diff_filepair *diff_queue(struct diff_queue_struct *queue,
3716                                  struct diff_filespec *one,
3717                                  struct diff_filespec *two)
3718 {
3719         struct diff_filepair *dp = xcalloc(1, sizeof(*dp));
3720         dp->one = one;
3721         dp->two = two;
3722         if (queue)
3723                 diff_q(queue, dp);
3724         return dp;
3725 }
3726
3727 void diff_free_filepair(struct diff_filepair *p)
3728 {
3729         free_filespec(p->one);
3730         free_filespec(p->two);
3731         free(p);
3732 }
3733
3734 /* This is different from find_unique_abbrev() in that
3735  * it stuffs the result with dots for alignment.
3736  */
3737 const char *diff_unique_abbrev(const unsigned char *sha1, int len)
3738 {
3739         int abblen;
3740         const char *abbrev;
3741         if (len == 40)
3742                 return sha1_to_hex(sha1);
3743
3744         abbrev = find_unique_abbrev(sha1, len);
3745         abblen = strlen(abbrev);
3746         if (abblen < 37) {
3747                 static char hex[41];
3748                 if (len < abblen && abblen <= len + 2)
3749                         sprintf(hex, "%s%.*s", abbrev, len+3-abblen, "..");
3750                 else
3751                         sprintf(hex, "%s...", abbrev);
3752                 return hex;
3753         }
3754         return sha1_to_hex(sha1);
3755 }
3756
3757 static void diff_flush_raw(struct diff_filepair *p, struct diff_options *opt)
3758 {
3759         int line_termination = opt->line_termination;
3760         int inter_name_termination = line_termination ? '\t' : '\0';
3761         if (opt->output_prefix) {
3762                 struct strbuf *msg = NULL;
3763                 msg = opt->output_prefix(opt, opt->output_prefix_data);
3764                 fprintf(opt->file, "%s", msg->buf);
3765         }
3766
3767         if (!(opt->output_format & DIFF_FORMAT_NAME_STATUS)) {
3768                 fprintf(opt->file, ":%06o %06o %s ", p->one->mode, p->two->mode,
3769                         diff_unique_abbrev(p->one->sha1, opt->abbrev));
3770                 fprintf(opt->file, "%s ", diff_unique_abbrev(p->two->sha1, opt->abbrev));
3771         }
3772         if (p->score) {
3773                 fprintf(opt->file, "%c%03d%c", p->status, similarity_index(p),
3774                         inter_name_termination);
3775         } else {
3776                 fprintf(opt->file, "%c%c", p->status, inter_name_termination);
3777         }
3778
3779         if (p->status == DIFF_STATUS_COPIED ||
3780             p->status == DIFF_STATUS_RENAMED) {
3781                 const char *name_a, *name_b;
3782                 name_a = p->one->path;
3783                 name_b = p->two->path;
3784                 strip_prefix(opt->prefix_length, &name_a, &name_b);
3785                 write_name_quoted(name_a, opt->file, inter_name_termination);
3786                 write_name_quoted(name_b, opt->file, line_termination);
3787         } else {
3788                 const char *name_a, *name_b;
3789                 name_a = p->one->mode ? p->one->path : p->two->path;
3790                 name_b = NULL;
3791                 strip_prefix(opt->prefix_length, &name_a, &name_b);
3792                 write_name_quoted(name_a, opt->file, line_termination);
3793         }
3794 }
3795
3796 int diff_unmodified_pair(struct diff_filepair *p)
3797 {
3798         /* This function is written stricter than necessary to support
3799          * the currently implemented transformers, but the idea is to
3800          * let transformers to produce diff_filepairs any way they want,
3801          * and filter and clean them up here before producing the output.
3802          */
3803         struct diff_filespec *one = p->one, *two = p->two;
3804
3805         if (DIFF_PAIR_UNMERGED(p))
3806                 return 0; /* unmerged is interesting */
3807
3808         /* deletion, addition, mode or type change
3809          * and rename are all interesting.
3810          */
3811         if (DIFF_FILE_VALID(one) != DIFF_FILE_VALID(two) ||
3812             DIFF_PAIR_MODE_CHANGED(p) ||
3813             strcmp(one->path, two->path))
3814                 return 0;
3815
3816         /* both are valid and point at the same path.  that is, we are
3817          * dealing with a change.
3818          */
3819         if (one->sha1_valid && two->sha1_valid &&
3820             !hashcmp(one->sha1, two->sha1) &&
3821             !one->dirty_submodule && !two->dirty_submodule)
3822                 return 1; /* no change */
3823         if (!one->sha1_valid && !two->sha1_valid)
3824                 return 1; /* both look at the same file on the filesystem. */
3825         return 0;
3826 }
3827
3828 static void diff_flush_patch(struct diff_filepair *p, struct diff_options *o)
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 tree diffs in patch format */
3836
3837         run_diff(p, o);
3838 }
3839
3840 static void diff_flush_stat(struct diff_filepair *p, struct diff_options *o,
3841                             struct diffstat_t *diffstat)
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; /* no useful stat for tree diffs */
3849
3850         run_diffstat(p, o, diffstat);
3851 }
3852
3853 static void diff_flush_checkdiff(struct diff_filepair *p,
3854                 struct diff_options *o)
3855 {
3856         if (diff_unmodified_pair(p))
3857                 return;
3858
3859         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
3860             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
3861                 return; /* nothing to check in tree diffs */
3862
3863         run_checkdiff(p, o);
3864 }
3865
3866 int diff_queue_is_empty(void)
3867 {
3868         struct diff_queue_struct *q = &diff_queued_diff;
3869         int i;
3870         for (i = 0; i < q->nr; i++)
3871                 if (!diff_unmodified_pair(q->queue[i]))
3872                         return 0;
3873         return 1;
3874 }
3875
3876 #if DIFF_DEBUG
3877 void diff_debug_filespec(struct diff_filespec *s, int x, const char *one)
3878 {
3879         fprintf(stderr, "queue[%d] %s (%s) %s %06o %s\n",
3880                 x, one ? one : "",
3881                 s->path,
3882                 DIFF_FILE_VALID(s) ? "valid" : "invalid",
3883                 s->mode,
3884                 s->sha1_valid ? sha1_to_hex(s->sha1) : "");
3885         fprintf(stderr, "queue[%d] %s size %lu flags %d\n",
3886                 x, one ? one : "",
3887                 s->size, s->xfrm_flags);
3888 }
3889
3890 void diff_debug_filepair(const struct diff_filepair *p, int i)
3891 {
3892         diff_debug_filespec(p->one, i, "one");
3893         diff_debug_filespec(p->two, i, "two");
3894         fprintf(stderr, "score %d, status %c rename_used %d broken %d\n",
3895                 p->score, p->status ? p->status : '?',
3896                 p->one->rename_used, p->broken_pair);
3897 }
3898
3899 void diff_debug_queue(const char *msg, struct diff_queue_struct *q)
3900 {
3901         int i;
3902         if (msg)
3903                 fprintf(stderr, "%s\n", msg);
3904         fprintf(stderr, "q->nr = %d\n", q->nr);
3905         for (i = 0; i < q->nr; i++) {
3906                 struct diff_filepair *p = q->queue[i];
3907                 diff_debug_filepair(p, i);
3908         }
3909 }
3910 #endif
3911
3912 static void diff_resolve_rename_copy(void)
3913 {
3914         int i;
3915         struct diff_filepair *p;
3916         struct diff_queue_struct *q = &diff_queued_diff;
3917
3918         diff_debug_queue("resolve-rename-copy", q);
3919
3920         for (i = 0; i < q->nr; i++) {
3921                 p = q->queue[i];
3922                 p->status = 0; /* undecided */
3923                 if (DIFF_PAIR_UNMERGED(p))
3924                         p->status = DIFF_STATUS_UNMERGED;
3925                 else if (!DIFF_FILE_VALID(p->one))
3926                         p->status = DIFF_STATUS_ADDED;
3927                 else if (!DIFF_FILE_VALID(p->two))
3928                         p->status = DIFF_STATUS_DELETED;
3929                 else if (DIFF_PAIR_TYPE_CHANGED(p))
3930                         p->status = DIFF_STATUS_TYPE_CHANGED;
3931
3932                 /* from this point on, we are dealing with a pair
3933                  * whose both sides are valid and of the same type, i.e.
3934                  * either in-place edit or rename/copy edit.
3935                  */
3936                 else if (DIFF_PAIR_RENAME(p)) {
3937                         /*
3938                          * A rename might have re-connected a broken
3939                          * pair up, causing the pathnames to be the
3940                          * same again. If so, that's not a rename at
3941                          * all, just a modification..
3942                          *
3943                          * Otherwise, see if this source was used for
3944                          * multiple renames, in which case we decrement
3945                          * the count, and call it a copy.
3946                          */
3947                         if (!strcmp(p->one->path, p->two->path))
3948                                 p->status = DIFF_STATUS_MODIFIED;
3949                         else if (--p->one->rename_used > 0)
3950                                 p->status = DIFF_STATUS_COPIED;
3951                         else
3952                                 p->status = DIFF_STATUS_RENAMED;
3953                 }
3954                 else if (hashcmp(p->one->sha1, p->two->sha1) ||
3955                          p->one->mode != p->two->mode ||
3956                          p->one->dirty_submodule ||
3957                          p->two->dirty_submodule ||
3958                          is_null_sha1(p->one->sha1))
3959                         p->status = DIFF_STATUS_MODIFIED;
3960                 else {
3961                         /* This is a "no-change" entry and should not
3962                          * happen anymore, but prepare for broken callers.
3963                          */
3964                         error("feeding unmodified %s to diffcore",
3965                               p->one->path);
3966                         p->status = DIFF_STATUS_UNKNOWN;
3967                 }
3968         }
3969         diff_debug_queue("resolve-rename-copy done", q);
3970 }
3971
3972 static int check_pair_status(struct diff_filepair *p)
3973 {
3974         switch (p->status) {
3975         case DIFF_STATUS_UNKNOWN:
3976                 return 0;
3977         case 0:
3978                 die("internal error in diff-resolve-rename-copy");
3979         default:
3980                 return 1;
3981         }
3982 }
3983
3984 static void flush_one_pair(struct diff_filepair *p, struct diff_options *opt)
3985 {
3986         int fmt = opt->output_format;
3987
3988         if (fmt & DIFF_FORMAT_CHECKDIFF)
3989                 diff_flush_checkdiff(p, opt);
3990         else if (fmt & (DIFF_FORMAT_RAW | DIFF_FORMAT_NAME_STATUS))
3991                 diff_flush_raw(p, opt);
3992         else if (fmt & DIFF_FORMAT_NAME) {
3993                 const char *name_a, *name_b;
3994                 name_a = p->two->path;
3995                 name_b = NULL;
3996                 strip_prefix(opt->prefix_length, &name_a, &name_b);
3997                 write_name_quoted(name_a, opt->file, opt->line_termination);
3998         }
3999 }
4000
4001 static void show_file_mode_name(FILE *file, const char *newdelete, struct diff_filespec *fs)
4002 {
4003         if (fs->mode)
4004                 fprintf(file, " %s mode %06o ", newdelete, fs->mode);
4005         else
4006                 fprintf(file, " %s ", newdelete);
4007         write_name_quoted(fs->path, file, '\n');
4008 }
4009
4010
4011 static void show_mode_change(FILE *file, struct diff_filepair *p, int show_name,
4012                 const char *line_prefix)
4013 {
4014         if (p->one->mode && p->two->mode && p->one->mode != p->two->mode) {
4015                 fprintf(file, "%s mode change %06o => %06o%c", line_prefix, p->one->mode,
4016                         p->two->mode, show_name ? ' ' : '\n');
4017                 if (show_name) {
4018                         write_name_quoted(p->two->path, file, '\n');
4019                 }
4020         }
4021 }
4022
4023 static void show_rename_copy(FILE *file, const char *renamecopy, struct diff_filepair *p,
4024                         const char *line_prefix)
4025 {
4026         char *names = pprint_rename(p->one->path, p->two->path);
4027
4028         fprintf(file, " %s %s (%d%%)\n", renamecopy, names, similarity_index(p));
4029         free(names);
4030         show_mode_change(file, p, 0, line_prefix);
4031 }
4032
4033 static void diff_summary(struct diff_options *opt, struct diff_filepair *p)
4034 {
4035         FILE *file = opt->file;
4036         char *line_prefix = "";
4037
4038         if (opt->output_prefix) {
4039                 struct strbuf *buf = opt->output_prefix(opt, opt->output_prefix_data);
4040                 line_prefix = buf->buf;
4041         }
4042
4043         switch(p->status) {
4044         case DIFF_STATUS_DELETED:
4045                 fputs(line_prefix, file);
4046                 show_file_mode_name(file, "delete", p->one);
4047                 break;
4048         case DIFF_STATUS_ADDED:
4049                 fputs(line_prefix, file);
4050                 show_file_mode_name(file, "create", p->two);
4051                 break;
4052         case DIFF_STATUS_COPIED:
4053                 fputs(line_prefix, file);
4054                 show_rename_copy(file, "copy", p, line_prefix);
4055                 break;
4056         case DIFF_STATUS_RENAMED:
4057                 fputs(line_prefix, file);
4058                 show_rename_copy(file, "rename", p, line_prefix);
4059                 break;
4060         default:
4061                 if (p->score) {
4062                         fprintf(file, "%s rewrite ", line_prefix);
4063                         write_name_quoted(p->two->path, file, ' ');
4064                         fprintf(file, "(%d%%)\n", similarity_index(p));
4065                 }
4066                 show_mode_change(file, p, !p->score, line_prefix);
4067                 break;
4068         }
4069 }
4070
4071 struct patch_id_t {
4072         git_SHA_CTX *ctx;
4073         int patchlen;
4074 };
4075
4076 static int remove_space(char *line, int len)
4077 {
4078         int i;
4079         char *dst = line;
4080         unsigned char c;
4081
4082         for (i = 0; i < len; i++)
4083                 if (!isspace((c = line[i])))
4084                         *dst++ = c;
4085
4086         return dst - line;
4087 }
4088
4089 static void patch_id_consume(void *priv, char *line, unsigned long len)
4090 {
4091         struct patch_id_t *data = priv;
4092         int new_len;
4093
4094         /* Ignore line numbers when computing the SHA1 of the patch */
4095         if (!prefixcmp(line, "@@ -"))
4096                 return;
4097
4098         new_len = remove_space(line, len);
4099
4100         git_SHA1_Update(data->ctx, line, new_len);
4101         data->patchlen += new_len;
4102 }
4103
4104 /* returns 0 upon success, and writes result into sha1 */
4105 static int diff_get_patch_id(struct diff_options *options, unsigned char *sha1)
4106 {
4107         struct diff_queue_struct *q = &diff_queued_diff;
4108         int i;
4109         git_SHA_CTX ctx;
4110         struct patch_id_t data;
4111         char buffer[PATH_MAX * 4 + 20];
4112
4113         git_SHA1_Init(&ctx);
4114         memset(&data, 0, sizeof(struct patch_id_t));
4115         data.ctx = &ctx;
4116
4117         for (i = 0; i < q->nr; i++) {
4118                 xpparam_t xpp;
4119                 xdemitconf_t xecfg;
4120                 mmfile_t mf1, mf2;
4121                 struct diff_filepair *p = q->queue[i];
4122                 int len1, len2;
4123
4124                 memset(&xpp, 0, sizeof(xpp));
4125                 memset(&xecfg, 0, sizeof(xecfg));
4126                 if (p->status == 0)
4127                         return error("internal diff status error");
4128                 if (p->status == DIFF_STATUS_UNKNOWN)
4129                         continue;
4130                 if (diff_unmodified_pair(p))
4131                         continue;
4132                 if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
4133                     (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
4134                         continue;
4135                 if (DIFF_PAIR_UNMERGED(p))
4136                         continue;
4137
4138                 diff_fill_sha1_info(p->one);
4139                 diff_fill_sha1_info(p->two);
4140                 if (fill_mmfile(&mf1, p->one) < 0 ||
4141                                 fill_mmfile(&mf2, p->two) < 0)
4142                         return error("unable to read files to diff");
4143
4144                 len1 = remove_space(p->one->path, strlen(p->one->path));
4145                 len2 = remove_space(p->two->path, strlen(p->two->path));
4146                 if (p->one->mode == 0)
4147                         len1 = snprintf(buffer, sizeof(buffer),
4148                                         "diff--gita/%.*sb/%.*s"
4149                                         "newfilemode%06o"
4150                                         "---/dev/null"
4151                                         "+++b/%.*s",
4152                                         len1, p->one->path,
4153                                         len2, p->two->path,
4154                                         p->two->mode,
4155                                         len2, p->two->path);
4156                 else if (p->two->mode == 0)
4157                         len1 = snprintf(buffer, sizeof(buffer),
4158                                         "diff--gita/%.*sb/%.*s"
4159                                         "deletedfilemode%06o"
4160                                         "---a/%.*s"
4161                                         "+++/dev/null",
4162                                         len1, p->one->path,
4163                                         len2, p->two->path,
4164                                         p->one->mode,
4165                                         len1, p->one->path);
4166                 else
4167                         len1 = snprintf(buffer, sizeof(buffer),
4168                                         "diff--gita/%.*sb/%.*s"
4169                                         "---a/%.*s"
4170                                         "+++b/%.*s",
4171                                         len1, p->one->path,
4172                                         len2, p->two->path,
4173                                         len1, p->one->path,
4174                                         len2, p->two->path);
4175                 git_SHA1_Update(&ctx, buffer, len1);
4176
4177                 if (diff_filespec_is_binary(p->one) ||
4178                     diff_filespec_is_binary(p->two)) {
4179                         git_SHA1_Update(&ctx, sha1_to_hex(p->one->sha1), 40);
4180                         git_SHA1_Update(&ctx, sha1_to_hex(p->two->sha1), 40);
4181                         continue;
4182                 }
4183
4184                 xpp.flags = 0;
4185                 xecfg.ctxlen = 3;
4186                 xecfg.flags = 0;
4187                 xdi_diff_outf(&mf1, &mf2, patch_id_consume, &data,
4188                               &xpp, &xecfg);
4189         }
4190
4191         git_SHA1_Final(sha1, &ctx);
4192         return 0;
4193 }
4194
4195 int diff_flush_patch_id(struct diff_options *options, unsigned char *sha1)
4196 {
4197         struct diff_queue_struct *q = &diff_queued_diff;
4198         int i;
4199         int result = diff_get_patch_id(options, sha1);
4200
4201         for (i = 0; i < q->nr; i++)
4202                 diff_free_filepair(q->queue[i]);
4203
4204         free(q->queue);
4205         DIFF_QUEUE_CLEAR(q);
4206
4207         return result;
4208 }
4209
4210 static int is_summary_empty(const struct diff_queue_struct *q)
4211 {
4212         int i;
4213
4214         for (i = 0; i < q->nr; i++) {
4215                 const struct diff_filepair *p = q->queue[i];
4216
4217                 switch (p->status) {
4218                 case DIFF_STATUS_DELETED:
4219                 case DIFF_STATUS_ADDED:
4220                 case DIFF_STATUS_COPIED:
4221                 case DIFF_STATUS_RENAMED:
4222                         return 0;
4223                 default:
4224                         if (p->score)
4225                                 return 0;
4226                         if (p->one->mode && p->two->mode &&
4227                             p->one->mode != p->two->mode)
4228                                 return 0;
4229                         break;
4230                 }
4231         }
4232         return 1;
4233 }
4234
4235 static const char rename_limit_warning[] =
4236 "inexact rename detection was skipped due to too many files.";
4237
4238 static const char degrade_cc_to_c_warning[] =
4239 "only found copies from modified paths due to too many files.";
4240
4241 static const char rename_limit_advice[] =
4242 "you may want to set your %s variable to at least "
4243 "%d and retry the command.";
4244
4245 void diff_warn_rename_limit(const char *varname, int needed, int degraded_cc)
4246 {
4247         if (degraded_cc)
4248                 warning(degrade_cc_to_c_warning);
4249         else if (needed)
4250                 warning(rename_limit_warning);
4251         else
4252                 return;
4253         if (0 < needed && needed < 32767)
4254                 warning(rename_limit_advice, varname, needed);
4255 }
4256
4257 void diff_flush(struct diff_options *options)
4258 {
4259         struct diff_queue_struct *q = &diff_queued_diff;
4260         int i, output_format = options->output_format;
4261         int separator = 0;
4262         int dirstat_by_line = 0;
4263
4264         /*
4265          * Order: raw, stat, summary, patch
4266          * or:    name/name-status/checkdiff (other bits clear)
4267          */
4268         if (!q->nr)
4269                 goto free_queue;
4270
4271         if (output_format & (DIFF_FORMAT_RAW |
4272                              DIFF_FORMAT_NAME |
4273                              DIFF_FORMAT_NAME_STATUS |
4274                              DIFF_FORMAT_CHECKDIFF)) {
4275                 for (i = 0; i < q->nr; i++) {
4276                         struct diff_filepair *p = q->queue[i];
4277                         if (check_pair_status(p))
4278                                 flush_one_pair(p, options);
4279                 }
4280                 separator++;
4281         }
4282
4283         if (output_format & DIFF_FORMAT_DIRSTAT && DIFF_OPT_TST(options, DIRSTAT_BY_LINE))
4284                 dirstat_by_line = 1;
4285
4286         if (output_format & (DIFF_FORMAT_DIFFSTAT|DIFF_FORMAT_SHORTSTAT|DIFF_FORMAT_NUMSTAT) ||
4287             dirstat_by_line) {
4288                 struct diffstat_t diffstat;
4289
4290                 memset(&diffstat, 0, sizeof(struct diffstat_t));
4291                 for (i = 0; i < q->nr; i++) {
4292                         struct diff_filepair *p = q->queue[i];
4293                         if (check_pair_status(p))
4294                                 diff_flush_stat(p, options, &diffstat);
4295                 }
4296                 if (output_format & DIFF_FORMAT_NUMSTAT)
4297                         show_numstat(&diffstat, options);
4298                 if (output_format & DIFF_FORMAT_DIFFSTAT)
4299                         show_stats(&diffstat, options);
4300                 if (output_format & DIFF_FORMAT_SHORTSTAT)
4301                         show_shortstats(&diffstat, options);
4302                 if (output_format & DIFF_FORMAT_DIRSTAT)
4303                         show_dirstat_by_line(&diffstat, options);
4304                 free_diffstat_info(&diffstat);
4305                 separator++;
4306         }
4307         if ((output_format & DIFF_FORMAT_DIRSTAT) && !dirstat_by_line)
4308                 show_dirstat(options);
4309
4310         if (output_format & DIFF_FORMAT_SUMMARY && !is_summary_empty(q)) {
4311                 for (i = 0; i < q->nr; i++) {
4312                         diff_summary(options, q->queue[i]);
4313                 }
4314                 separator++;
4315         }
4316
4317         if (output_format & DIFF_FORMAT_NO_OUTPUT &&
4318             DIFF_OPT_TST(options, EXIT_WITH_STATUS) &&
4319             DIFF_OPT_TST(options, DIFF_FROM_CONTENTS)) {
4320                 /*
4321                  * run diff_flush_patch for the exit status. setting
4322                  * options->file to /dev/null should be safe, becaue we
4323                  * aren't supposed to produce any output anyway.
4324                  */
4325                 if (options->close_file)
4326                         fclose(options->file);
4327                 options->file = fopen("/dev/null", "w");
4328                 if (!options->file)
4329                         die_errno("Could not open /dev/null");
4330                 options->close_file = 1;
4331                 for (i = 0; i < q->nr; i++) {
4332                         struct diff_filepair *p = q->queue[i];
4333                         if (check_pair_status(p))
4334                                 diff_flush_patch(p, options);
4335                         if (options->found_changes)
4336                                 break;
4337                 }
4338         }
4339
4340         if (output_format & DIFF_FORMAT_PATCH) {
4341                 if (separator) {
4342                         putc(options->line_termination, options->file);
4343                         if (options->stat_sep) {
4344                                 /* attach patch instead of inline */
4345                                 fputs(options->stat_sep, options->file);
4346                         }
4347                 }
4348
4349                 for (i = 0; i < q->nr; i++) {
4350                         struct diff_filepair *p = q->queue[i];
4351                         if (check_pair_status(p))
4352                                 diff_flush_patch(p, options);
4353                 }
4354         }
4355
4356         if (output_format & DIFF_FORMAT_CALLBACK)
4357                 options->format_callback(q, options, options->format_callback_data);
4358
4359         for (i = 0; i < q->nr; i++)
4360                 diff_free_filepair(q->queue[i]);
4361 free_queue:
4362         free(q->queue);
4363         DIFF_QUEUE_CLEAR(q);
4364         if (options->close_file)
4365                 fclose(options->file);
4366
4367         /*
4368          * Report the content-level differences with HAS_CHANGES;
4369          * diff_addremove/diff_change does not set the bit when
4370          * DIFF_FROM_CONTENTS is in effect (e.g. with -w).
4371          */
4372         if (DIFF_OPT_TST(options, DIFF_FROM_CONTENTS)) {
4373                 if (options->found_changes)
4374                         DIFF_OPT_SET(options, HAS_CHANGES);
4375                 else
4376                         DIFF_OPT_CLR(options, HAS_CHANGES);
4377         }
4378 }
4379
4380 static void diffcore_apply_filter(const char *filter)
4381 {
4382         int i;
4383         struct diff_queue_struct *q = &diff_queued_diff;
4384         struct diff_queue_struct outq;
4385         DIFF_QUEUE_CLEAR(&outq);
4386
4387         if (!filter)
4388                 return;
4389
4390         if (strchr(filter, DIFF_STATUS_FILTER_AON)) {
4391                 int found;
4392                 for (i = found = 0; !found && i < q->nr; i++) {
4393                         struct diff_filepair *p = q->queue[i];
4394                         if (((p->status == DIFF_STATUS_MODIFIED) &&
4395                              ((p->score &&
4396                                strchr(filter, DIFF_STATUS_FILTER_BROKEN)) ||
4397                               (!p->score &&
4398                                strchr(filter, DIFF_STATUS_MODIFIED)))) ||
4399                             ((p->status != DIFF_STATUS_MODIFIED) &&
4400                              strchr(filter, p->status)))
4401                                 found++;
4402                 }
4403                 if (found)
4404                         return;
4405
4406                 /* otherwise we will clear the whole queue
4407                  * by copying the empty outq at the end of this
4408                  * function, but first clear the current entries
4409                  * in the queue.
4410                  */
4411                 for (i = 0; i < q->nr; i++)
4412                         diff_free_filepair(q->queue[i]);
4413         }
4414         else {
4415                 /* Only the matching ones */
4416                 for (i = 0; i < q->nr; i++) {
4417                         struct diff_filepair *p = q->queue[i];
4418
4419                         if (((p->status == DIFF_STATUS_MODIFIED) &&
4420                              ((p->score &&
4421                                strchr(filter, DIFF_STATUS_FILTER_BROKEN)) ||
4422                               (!p->score &&
4423                                strchr(filter, DIFF_STATUS_MODIFIED)))) ||
4424                             ((p->status != DIFF_STATUS_MODIFIED) &&
4425                              strchr(filter, p->status)))
4426                                 diff_q(&outq, p);
4427                         else
4428                                 diff_free_filepair(p);
4429                 }
4430         }
4431         free(q->queue);
4432         *q = outq;
4433 }
4434
4435 /* Check whether two filespecs with the same mode and size are identical */
4436 static int diff_filespec_is_identical(struct diff_filespec *one,
4437                                       struct diff_filespec *two)
4438 {
4439         if (S_ISGITLINK(one->mode))
4440                 return 0;
4441         if (diff_populate_filespec(one, 0))
4442                 return 0;
4443         if (diff_populate_filespec(two, 0))
4444                 return 0;
4445         return !memcmp(one->data, two->data, one->size);
4446 }
4447
4448 static void diffcore_skip_stat_unmatch(struct diff_options *diffopt)
4449 {
4450         int i;
4451         struct diff_queue_struct *q = &diff_queued_diff;
4452         struct diff_queue_struct outq;
4453         DIFF_QUEUE_CLEAR(&outq);
4454
4455         for (i = 0; i < q->nr; i++) {
4456                 struct diff_filepair *p = q->queue[i];
4457
4458                 /*
4459                  * 1. Entries that come from stat info dirtiness
4460                  *    always have both sides (iow, not create/delete),
4461                  *    one side of the object name is unknown, with
4462                  *    the same mode and size.  Keep the ones that
4463                  *    do not match these criteria.  They have real
4464                  *    differences.
4465                  *
4466                  * 2. At this point, the file is known to be modified,
4467                  *    with the same mode and size, and the object
4468                  *    name of one side is unknown.  Need to inspect
4469                  *    the identical contents.
4470                  */
4471                 if (!DIFF_FILE_VALID(p->one) || /* (1) */
4472                     !DIFF_FILE_VALID(p->two) ||
4473                     (p->one->sha1_valid && p->two->sha1_valid) ||
4474                     (p->one->mode != p->two->mode) ||
4475                     diff_populate_filespec(p->one, 1) ||
4476                     diff_populate_filespec(p->two, 1) ||
4477                     (p->one->size != p->two->size) ||
4478                     !diff_filespec_is_identical(p->one, p->two)) /* (2) */
4479                         diff_q(&outq, p);
4480                 else {
4481                         /*
4482                          * The caller can subtract 1 from skip_stat_unmatch
4483                          * to determine how many paths were dirty only
4484                          * due to stat info mismatch.
4485                          */
4486                         if (!DIFF_OPT_TST(diffopt, NO_INDEX))
4487                                 diffopt->skip_stat_unmatch++;
4488                         diff_free_filepair(p);
4489                 }
4490         }
4491         free(q->queue);
4492         *q = outq;
4493 }
4494
4495 static int diffnamecmp(const void *a_, const void *b_)
4496 {
4497         const struct diff_filepair *a = *((const struct diff_filepair **)a_);
4498         const struct diff_filepair *b = *((const struct diff_filepair **)b_);
4499         const char *name_a, *name_b;
4500
4501         name_a = a->one ? a->one->path : a->two->path;
4502         name_b = b->one ? b->one->path : b->two->path;
4503         return strcmp(name_a, name_b);
4504 }
4505
4506 void diffcore_fix_diff_index(struct diff_options *options)
4507 {
4508         struct diff_queue_struct *q = &diff_queued_diff;
4509         qsort(q->queue, q->nr, sizeof(q->queue[0]), diffnamecmp);
4510 }
4511
4512 void diffcore_std(struct diff_options *options)
4513 {
4514         if (options->skip_stat_unmatch)
4515                 diffcore_skip_stat_unmatch(options);
4516         if (!options->found_follow) {
4517                 /* See try_to_follow_renames() in tree-diff.c */
4518                 if (options->break_opt != -1)
4519                         diffcore_break(options->break_opt);
4520                 if (options->detect_rename)
4521                         diffcore_rename(options);
4522                 if (options->break_opt != -1)
4523                         diffcore_merge_broken();
4524         }
4525         if (options->pickaxe)
4526                 diffcore_pickaxe(options);
4527         if (options->orderfile)
4528                 diffcore_order(options->orderfile);
4529         if (!options->found_follow)
4530                 /* See try_to_follow_renames() in tree-diff.c */
4531                 diff_resolve_rename_copy();
4532         diffcore_apply_filter(options->filter);
4533
4534         if (diff_queued_diff.nr && !DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
4535                 DIFF_OPT_SET(options, HAS_CHANGES);
4536         else
4537                 DIFF_OPT_CLR(options, HAS_CHANGES);
4538
4539         options->found_follow = 0;
4540 }
4541
4542 int diff_result_code(struct diff_options *opt, int status)
4543 {
4544         int result = 0;
4545
4546         diff_warn_rename_limit("diff.renamelimit",
4547                                opt->needed_rename_limit,
4548                                opt->degraded_cc_to_c);
4549         if (!DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
4550             !(opt->output_format & DIFF_FORMAT_CHECKDIFF))
4551                 return status;
4552         if (DIFF_OPT_TST(opt, EXIT_WITH_STATUS) &&
4553             DIFF_OPT_TST(opt, HAS_CHANGES))
4554                 result |= 01;
4555         if ((opt->output_format & DIFF_FORMAT_CHECKDIFF) &&
4556             DIFF_OPT_TST(opt, CHECK_FAILED))
4557                 result |= 02;
4558         return result;
4559 }
4560
4561 int diff_can_quit_early(struct diff_options *opt)
4562 {
4563         return (DIFF_OPT_TST(opt, QUICK) &&
4564                 !opt->filter &&
4565                 DIFF_OPT_TST(opt, HAS_CHANGES));
4566 }
4567
4568 /*
4569  * Shall changes to this submodule be ignored?
4570  *
4571  * Submodule changes can be configured to be ignored separately for each path,
4572  * but that configuration can be overridden from the command line.
4573  */
4574 static int is_submodule_ignored(const char *path, struct diff_options *options)
4575 {
4576         int ignored = 0;
4577         unsigned orig_flags = options->flags;
4578         if (!DIFF_OPT_TST(options, OVERRIDE_SUBMODULE_CONFIG))
4579                 set_diffopt_flags_from_submodule_config(options, path);
4580         if (DIFF_OPT_TST(options, IGNORE_SUBMODULES))
4581                 ignored = 1;
4582         options->flags = orig_flags;
4583         return ignored;
4584 }
4585
4586 void diff_addremove(struct diff_options *options,
4587                     int addremove, unsigned mode,
4588                     const unsigned char *sha1,
4589                     const char *concatpath, unsigned dirty_submodule)
4590 {
4591         struct diff_filespec *one, *two;
4592
4593         if (S_ISGITLINK(mode) && is_submodule_ignored(concatpath, options))
4594                 return;
4595
4596         /* This may look odd, but it is a preparation for
4597          * feeding "there are unchanged files which should
4598          * not produce diffs, but when you are doing copy
4599          * detection you would need them, so here they are"
4600          * entries to the diff-core.  They will be prefixed
4601          * with something like '=' or '*' (I haven't decided
4602          * which but should not make any difference).
4603          * Feeding the same new and old to diff_change()
4604          * also has the same effect.
4605          * Before the final output happens, they are pruned after
4606          * merged into rename/copy pairs as appropriate.
4607          */
4608         if (DIFF_OPT_TST(options, REVERSE_DIFF))
4609                 addremove = (addremove == '+' ? '-' :
4610                              addremove == '-' ? '+' : addremove);
4611
4612         if (options->prefix &&
4613             strncmp(concatpath, options->prefix, options->prefix_length))
4614                 return;
4615
4616         one = alloc_filespec(concatpath);
4617         two = alloc_filespec(concatpath);
4618
4619         if (addremove != '+')
4620                 fill_filespec(one, sha1, mode);
4621         if (addremove != '-') {
4622                 fill_filespec(two, sha1, mode);
4623                 two->dirty_submodule = dirty_submodule;
4624         }
4625
4626         diff_queue(&diff_queued_diff, one, two);
4627         if (!DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
4628                 DIFF_OPT_SET(options, HAS_CHANGES);
4629 }
4630
4631 void diff_change(struct diff_options *options,
4632                  unsigned old_mode, unsigned new_mode,
4633                  const unsigned char *old_sha1,
4634                  const unsigned char *new_sha1,
4635                  const char *concatpath,
4636                  unsigned old_dirty_submodule, unsigned new_dirty_submodule)
4637 {
4638         struct diff_filespec *one, *two;
4639
4640         if (S_ISGITLINK(old_mode) && S_ISGITLINK(new_mode) &&
4641             is_submodule_ignored(concatpath, options))
4642                 return;
4643
4644         if (DIFF_OPT_TST(options, REVERSE_DIFF)) {
4645                 unsigned tmp;
4646                 const unsigned char *tmp_c;
4647                 tmp = old_mode; old_mode = new_mode; new_mode = tmp;
4648                 tmp_c = old_sha1; old_sha1 = new_sha1; new_sha1 = tmp_c;
4649                 tmp = old_dirty_submodule; old_dirty_submodule = new_dirty_submodule;
4650                         new_dirty_submodule = tmp;
4651         }
4652
4653         if (options->prefix &&
4654             strncmp(concatpath, options->prefix, options->prefix_length))
4655                 return;
4656
4657         one = alloc_filespec(concatpath);
4658         two = alloc_filespec(concatpath);
4659         fill_filespec(one, old_sha1, old_mode);
4660         fill_filespec(two, new_sha1, new_mode);
4661         one->dirty_submodule = old_dirty_submodule;
4662         two->dirty_submodule = new_dirty_submodule;
4663
4664         diff_queue(&diff_queued_diff, one, two);
4665         if (!DIFF_OPT_TST(options, DIFF_FROM_CONTENTS))
4666                 DIFF_OPT_SET(options, HAS_CHANGES);
4667 }
4668
4669 struct diff_filepair *diff_unmerge(struct diff_options *options, const char *path)
4670 {
4671         struct diff_filepair *pair;
4672         struct diff_filespec *one, *two;
4673
4674         if (options->prefix &&
4675             strncmp(path, options->prefix, options->prefix_length))
4676                 return NULL;
4677
4678         one = alloc_filespec(path);
4679         two = alloc_filespec(path);
4680         pair = diff_queue(&diff_queued_diff, one, two);
4681         pair->is_unmerged = 1;
4682         return pair;
4683 }
4684
4685 static char *run_textconv(const char *pgm, struct diff_filespec *spec,
4686                 size_t *outsize)
4687 {
4688         struct diff_tempfile *temp;
4689         const char *argv[3];
4690         const char **arg = argv;
4691         struct child_process child;
4692         struct strbuf buf = STRBUF_INIT;
4693         int err = 0;
4694
4695         temp = prepare_temp_file(spec->path, spec);
4696         *arg++ = pgm;
4697         *arg++ = temp->name;
4698         *arg = NULL;
4699
4700         memset(&child, 0, sizeof(child));
4701         child.use_shell = 1;
4702         child.argv = argv;
4703         child.out = -1;
4704         if (start_command(&child)) {
4705                 remove_tempfile();
4706                 return NULL;
4707         }
4708
4709         if (strbuf_read(&buf, child.out, 0) < 0)
4710                 err = error("error reading from textconv command '%s'", pgm);
4711         close(child.out);
4712
4713         if (finish_command(&child) || err) {
4714                 strbuf_release(&buf);
4715                 remove_tempfile();
4716                 return NULL;
4717         }
4718         remove_tempfile();
4719
4720         return strbuf_detach(&buf, outsize);
4721 }
4722
4723 size_t fill_textconv(struct userdiff_driver *driver,
4724                      struct diff_filespec *df,
4725                      char **outbuf)
4726 {
4727         size_t size;
4728
4729         if (!driver || !driver->textconv) {
4730                 if (!DIFF_FILE_VALID(df)) {
4731                         *outbuf = "";
4732                         return 0;
4733                 }
4734                 if (diff_populate_filespec(df, 0))
4735                         die("unable to read files to diff");
4736                 *outbuf = df->data;
4737                 return df->size;
4738         }
4739
4740         if (driver->textconv_cache && df->sha1_valid) {
4741                 *outbuf = notes_cache_get(driver->textconv_cache, df->sha1,
4742                                           &size);
4743                 if (*outbuf)
4744                         return size;
4745         }
4746
4747         *outbuf = run_textconv(driver->textconv, df, &size);
4748         if (!*outbuf)
4749                 die("unable to read files to diff");
4750
4751         if (driver->textconv_cache && df->sha1_valid) {
4752                 /* ignore errors, as we might be in a readonly repository */
4753                 notes_cache_put(driver->textconv_cache, df->sha1, *outbuf,
4754                                 size);
4755                 /*
4756                  * we could save up changes and flush them all at the end,
4757                  * but we would need an extra call after all diffing is done.
4758                  * Since generating a cache entry is the slow path anyway,
4759                  * this extra overhead probably isn't a big deal.
4760                  */
4761                 notes_cache_write(driver->textconv_cache);
4762         }
4763
4764         return size;
4765 }