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