Merge branch 'jc/format-patch'
[git] / diff.c
1 /*
2  * Copyright (C) 2005 Junio C Hamano
3  */
4 #include <sys/types.h>
5 #include <sys/wait.h>
6 #include <signal.h>
7 #include "cache.h"
8 #include "quote.h"
9 #include "diff.h"
10 #include "diffcore.h"
11 #include "delta.h"
12 #include "xdiff-interface.h"
13
14 static int use_size_cache;
15
16 static int diff_detect_rename_default;
17 static int diff_rename_limit_default = -1;
18 static int diff_use_color_default;
19
20 /* "\033[1;38;5;2xx;48;5;2xxm\0" is 23 bytes */
21 static char diff_colors[][24] = {
22         "\033[m",       /* reset */
23         "",             /* normal */
24         "\033[1m",      /* bold */
25         "\033[36m",     /* cyan */
26         "\033[31m",     /* red */
27         "\033[32m",     /* green */
28         "\033[33m"      /* yellow */
29 };
30
31 static int parse_diff_color_slot(const char *var, int ofs)
32 {
33         if (!strcasecmp(var+ofs, "plain"))
34                 return DIFF_PLAIN;
35         if (!strcasecmp(var+ofs, "meta"))
36                 return DIFF_METAINFO;
37         if (!strcasecmp(var+ofs, "frag"))
38                 return DIFF_FRAGINFO;
39         if (!strcasecmp(var+ofs, "old"))
40                 return DIFF_FILE_OLD;
41         if (!strcasecmp(var+ofs, "new"))
42                 return DIFF_FILE_NEW;
43         if (!strcasecmp(var+ofs, "commit"))
44                 return DIFF_COMMIT;
45         die("bad config variable '%s'", var);
46 }
47
48 static int parse_color(const char *name, int len)
49 {
50         static const char * const color_names[] = {
51                 "normal", "black", "red", "green", "yellow",
52                 "blue", "magenta", "cyan", "white"
53         };
54         char *end;
55         int i;
56         for (i = 0; i < ARRAY_SIZE(color_names); i++) {
57                 const char *str = color_names[i];
58                 if (!strncasecmp(name, str, len) && !str[len])
59                         return i - 1;
60         }
61         i = strtol(name, &end, 10);
62         if (*name && !*end && i >= -1 && i <= 255)
63                 return i;
64         return -2;
65 }
66
67 static int parse_attr(const char *name, int len)
68 {
69         static const int attr_values[] = { 1, 2, 4, 5, 7 };
70         static const char * const attr_names[] = {
71                 "bold", "dim", "ul", "blink", "reverse"
72         };
73         int i;
74         for (i = 0; i < ARRAY_SIZE(attr_names); i++) {
75                 const char *str = attr_names[i];
76                 if (!strncasecmp(name, str, len) && !str[len])
77                         return attr_values[i];
78         }
79         return -1;
80 }
81
82 static void parse_diff_color_value(const char *value, const char *var, char *dst)
83 {
84         const char *ptr = value;
85         int attr = -1;
86         int fg = -2;
87         int bg = -2;
88
89         if (!strcasecmp(value, "reset")) {
90                 strcpy(dst, "\033[m");
91                 return;
92         }
93
94         /* [fg [bg]] [attr] */
95         while (*ptr) {
96                 const char *word = ptr;
97                 int val, len = 0;
98
99                 while (word[len] && !isspace(word[len]))
100                         len++;
101
102                 ptr = word + len;
103                 while (*ptr && isspace(*ptr))
104                         ptr++;
105
106                 val = parse_color(word, len);
107                 if (val >= -1) {
108                         if (fg == -2) {
109                                 fg = val;
110                                 continue;
111                         }
112                         if (bg == -2) {
113                                 bg = val;
114                                 continue;
115                         }
116                         goto bad;
117                 }
118                 val = parse_attr(word, len);
119                 if (val < 0 || attr != -1)
120                         goto bad;
121                 attr = val;
122         }
123
124         if (attr >= 0 || fg >= 0 || bg >= 0) {
125                 int sep = 0;
126
127                 *dst++ = '\033';
128                 *dst++ = '[';
129                 if (attr >= 0) {
130                         *dst++ = '0' + attr;
131                         sep++;
132                 }
133                 if (fg >= 0) {
134                         if (sep++)
135                                 *dst++ = ';';
136                         if (fg < 8) {
137                                 *dst++ = '3';
138                                 *dst++ = '0' + fg;
139                         } else {
140                                 dst += sprintf(dst, "38;5;%d", fg);
141                         }
142                 }
143                 if (bg >= 0) {
144                         if (sep++)
145                                 *dst++ = ';';
146                         if (bg < 8) {
147                                 *dst++ = '4';
148                                 *dst++ = '0' + bg;
149                         } else {
150                                 dst += sprintf(dst, "48;5;%d", bg);
151                         }
152                 }
153                 *dst++ = 'm';
154         }
155         *dst = 0;
156         return;
157 bad:
158         die("bad config value '%s' for variable '%s'", value, var);
159 }
160
161 /*
162  * These are to give UI layer defaults.
163  * The core-level commands such as git-diff-files should
164  * never be affected by the setting of diff.renames
165  * the user happens to have in the configuration file.
166  */
167 int git_diff_ui_config(const char *var, const char *value)
168 {
169         if (!strcmp(var, "diff.renamelimit")) {
170                 diff_rename_limit_default = git_config_int(var, value);
171                 return 0;
172         }
173         if (!strcmp(var, "diff.color")) {
174                 if (!value)
175                         diff_use_color_default = 1; /* bool */
176                 else if (!strcasecmp(value, "auto")) {
177                         diff_use_color_default = 0;
178                         if (isatty(1) || (pager_in_use && pager_use_color)) {
179                                 char *term = getenv("TERM");
180                                 if (term && strcmp(term, "dumb"))
181                                         diff_use_color_default = 1;
182                         }
183                 }
184                 else if (!strcasecmp(value, "never"))
185                         diff_use_color_default = 0;
186                 else if (!strcasecmp(value, "always"))
187                         diff_use_color_default = 1;
188                 else
189                         diff_use_color_default = git_config_bool(var, value);
190                 return 0;
191         }
192         if (!strcmp(var, "diff.renames")) {
193                 if (!value)
194                         diff_detect_rename_default = DIFF_DETECT_RENAME;
195                 else if (!strcasecmp(value, "copies") ||
196                          !strcasecmp(value, "copy"))
197                         diff_detect_rename_default = DIFF_DETECT_COPY;
198                 else if (git_config_bool(var,value))
199                         diff_detect_rename_default = DIFF_DETECT_RENAME;
200                 return 0;
201         }
202         if (!strncmp(var, "diff.color.", 11)) {
203                 int slot = parse_diff_color_slot(var, 11);
204                 parse_diff_color_value(value, var, diff_colors[slot]);
205                 return 0;
206         }
207         return git_default_config(var, value);
208 }
209
210 static char *quote_one(const char *str)
211 {
212         int needlen;
213         char *xp;
214
215         if (!str)
216                 return NULL;
217         needlen = quote_c_style(str, NULL, NULL, 0);
218         if (!needlen)
219                 return strdup(str);
220         xp = xmalloc(needlen + 1);
221         quote_c_style(str, xp, NULL, 0);
222         return xp;
223 }
224
225 static char *quote_two(const char *one, const char *two)
226 {
227         int need_one = quote_c_style(one, NULL, NULL, 1);
228         int need_two = quote_c_style(two, NULL, NULL, 1);
229         char *xp;
230
231         if (need_one + need_two) {
232                 if (!need_one) need_one = strlen(one);
233                 if (!need_two) need_one = strlen(two);
234
235                 xp = xmalloc(need_one + need_two + 3);
236                 xp[0] = '"';
237                 quote_c_style(one, xp + 1, NULL, 1);
238                 quote_c_style(two, xp + need_one + 1, NULL, 1);
239                 strcpy(xp + need_one + need_two + 1, "\"");
240                 return xp;
241         }
242         need_one = strlen(one);
243         need_two = strlen(two);
244         xp = xmalloc(need_one + need_two + 1);
245         strcpy(xp, one);
246         strcpy(xp + need_one, two);
247         return xp;
248 }
249
250 static const char *external_diff(void)
251 {
252         static const char *external_diff_cmd = NULL;
253         static int done_preparing = 0;
254
255         if (done_preparing)
256                 return external_diff_cmd;
257         external_diff_cmd = getenv("GIT_EXTERNAL_DIFF");
258         done_preparing = 1;
259         return external_diff_cmd;
260 }
261
262 #define TEMPFILE_PATH_LEN               50
263
264 static struct diff_tempfile {
265         const char *name; /* filename external diff should read from */
266         char hex[41];
267         char mode[10];
268         char tmp_path[TEMPFILE_PATH_LEN];
269 } diff_temp[2];
270
271 static int count_lines(const char *data, int size)
272 {
273         int count, ch, completely_empty = 1, nl_just_seen = 0;
274         count = 0;
275         while (0 < size--) {
276                 ch = *data++;
277                 if (ch == '\n') {
278                         count++;
279                         nl_just_seen = 1;
280                         completely_empty = 0;
281                 }
282                 else {
283                         nl_just_seen = 0;
284                         completely_empty = 0;
285                 }
286         }
287         if (completely_empty)
288                 return 0;
289         if (!nl_just_seen)
290                 count++; /* no trailing newline */
291         return count;
292 }
293
294 static void print_line_count(int count)
295 {
296         switch (count) {
297         case 0:
298                 printf("0,0");
299                 break;
300         case 1:
301                 printf("1");
302                 break;
303         default:
304                 printf("1,%d", count);
305                 break;
306         }
307 }
308
309 static void copy_file(int prefix, const char *data, int size)
310 {
311         int ch, nl_just_seen = 1;
312         while (0 < size--) {
313                 ch = *data++;
314                 if (nl_just_seen)
315                         putchar(prefix);
316                 putchar(ch);
317                 if (ch == '\n')
318                         nl_just_seen = 1;
319                 else
320                         nl_just_seen = 0;
321         }
322         if (!nl_just_seen)
323                 printf("\n\\ No newline at end of file\n");
324 }
325
326 static void emit_rewrite_diff(const char *name_a,
327                               const char *name_b,
328                               struct diff_filespec *one,
329                               struct diff_filespec *two)
330 {
331         int lc_a, lc_b;
332         diff_populate_filespec(one, 0);
333         diff_populate_filespec(two, 0);
334         lc_a = count_lines(one->data, one->size);
335         lc_b = count_lines(two->data, two->size);
336         printf("--- %s\n+++ %s\n@@ -", name_a, name_b);
337         print_line_count(lc_a);
338         printf(" +");
339         print_line_count(lc_b);
340         printf(" @@\n");
341         if (lc_a)
342                 copy_file('-', one->data, one->size);
343         if (lc_b)
344                 copy_file('+', two->data, two->size);
345 }
346
347 static int fill_mmfile(mmfile_t *mf, struct diff_filespec *one)
348 {
349         if (!DIFF_FILE_VALID(one)) {
350                 mf->ptr = (char *)""; /* does not matter */
351                 mf->size = 0;
352                 return 0;
353         }
354         else if (diff_populate_filespec(one, 0))
355                 return -1;
356         mf->ptr = one->data;
357         mf->size = one->size;
358         return 0;
359 }
360
361 struct diff_words_buffer {
362         mmfile_t text;
363         long alloc;
364         long current; /* output pointer */
365         int suppressed_newline;
366 };
367
368 static void diff_words_append(char *line, unsigned long len,
369                 struct diff_words_buffer *buffer)
370 {
371         if (buffer->text.size + len > buffer->alloc) {
372                 buffer->alloc = (buffer->text.size + len) * 3 / 2;
373                 buffer->text.ptr = xrealloc(buffer->text.ptr, buffer->alloc);
374         }
375         line++;
376         len--;
377         memcpy(buffer->text.ptr + buffer->text.size, line, len);
378         buffer->text.size += len;
379 }
380
381 struct diff_words_data {
382         struct xdiff_emit_state xm;
383         struct diff_words_buffer minus, plus;
384 };
385
386 static void print_word(struct diff_words_buffer *buffer, int len, int color,
387                 int suppress_newline)
388 {
389         const char *ptr;
390         int eol = 0;
391
392         if (len == 0)
393                 return;
394
395         ptr  = buffer->text.ptr + buffer->current;
396         buffer->current += len;
397
398         if (ptr[len - 1] == '\n') {
399                 eol = 1;
400                 len--;
401         }
402
403         fputs(diff_get_color(1, color), stdout);
404         fwrite(ptr, len, 1, stdout);
405         fputs(diff_get_color(1, DIFF_RESET), stdout);
406
407         if (eol) {
408                 if (suppress_newline)
409                         buffer->suppressed_newline = 1;
410                 else
411                         putchar('\n');
412         }
413 }
414
415 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
416 {
417         struct diff_words_data *diff_words = priv;
418
419         if (diff_words->minus.suppressed_newline) {
420                 if (line[0] != '+')
421                         putchar('\n');
422                 diff_words->minus.suppressed_newline = 0;
423         }
424
425         len--;
426         switch (line[0]) {
427                 case '-':
428                         print_word(&diff_words->minus, len, DIFF_FILE_OLD, 1);
429                         break;
430                 case '+':
431                         print_word(&diff_words->plus, len, DIFF_FILE_NEW, 0);
432                         break;
433                 case ' ':
434                         print_word(&diff_words->plus, len, DIFF_PLAIN, 0);
435                         diff_words->minus.current += len;
436                         break;
437         }
438 }
439
440 /* this executes the word diff on the accumulated buffers */
441 static void diff_words_show(struct diff_words_data *diff_words)
442 {
443         xpparam_t xpp;
444         xdemitconf_t xecfg;
445         xdemitcb_t ecb;
446         mmfile_t minus, plus;
447         int i;
448
449         minus.size = diff_words->minus.text.size;
450         minus.ptr = xmalloc(minus.size);
451         memcpy(minus.ptr, diff_words->minus.text.ptr, minus.size);
452         for (i = 0; i < minus.size; i++)
453                 if (isspace(minus.ptr[i]))
454                         minus.ptr[i] = '\n';
455         diff_words->minus.current = 0;
456
457         plus.size = diff_words->plus.text.size;
458         plus.ptr = xmalloc(plus.size);
459         memcpy(plus.ptr, diff_words->plus.text.ptr, plus.size);
460         for (i = 0; i < plus.size; i++)
461                 if (isspace(plus.ptr[i]))
462                         plus.ptr[i] = '\n';
463         diff_words->plus.current = 0;
464
465         xpp.flags = XDF_NEED_MINIMAL;
466         xecfg.ctxlen = diff_words->minus.alloc + diff_words->plus.alloc;
467         xecfg.flags = 0;
468         ecb.outf = xdiff_outf;
469         ecb.priv = diff_words;
470         diff_words->xm.consume = fn_out_diff_words_aux;
471         xdl_diff(&minus, &plus, &xpp, &xecfg, &ecb);
472
473         free(minus.ptr);
474         free(plus.ptr);
475         diff_words->minus.text.size = diff_words->plus.text.size = 0;
476
477         if (diff_words->minus.suppressed_newline) {
478                 putchar('\n');
479                 diff_words->minus.suppressed_newline = 0;
480         }
481 }
482
483 struct emit_callback {
484         struct xdiff_emit_state xm;
485         int nparents, color_diff;
486         const char **label_path;
487         struct diff_words_data *diff_words;
488 };
489
490 static void free_diff_words_data(struct emit_callback *ecbdata)
491 {
492         if (ecbdata->diff_words) {
493                 /* flush buffers */
494                 if (ecbdata->diff_words->minus.text.size ||
495                                 ecbdata->diff_words->plus.text.size)
496                         diff_words_show(ecbdata->diff_words);
497
498                 if (ecbdata->diff_words->minus.text.ptr)
499                         free (ecbdata->diff_words->minus.text.ptr);
500                 if (ecbdata->diff_words->plus.text.ptr)
501                         free (ecbdata->diff_words->plus.text.ptr);
502                 free(ecbdata->diff_words);
503                 ecbdata->diff_words = NULL;
504         }
505 }
506
507 const char *diff_get_color(int diff_use_color, enum color_diff ix)
508 {
509         if (diff_use_color)
510                 return diff_colors[ix];
511         return "";
512 }
513
514 static void fn_out_consume(void *priv, char *line, unsigned long len)
515 {
516         int i;
517         struct emit_callback *ecbdata = priv;
518         const char *set = diff_get_color(ecbdata->color_diff, DIFF_METAINFO);
519         const char *reset = diff_get_color(ecbdata->color_diff, DIFF_RESET);
520
521         if (ecbdata->label_path[0]) {
522                 printf("%s--- %s%s\n", set, ecbdata->label_path[0], reset);
523                 printf("%s+++ %s%s\n", set, ecbdata->label_path[1], reset);
524                 ecbdata->label_path[0] = ecbdata->label_path[1] = NULL;
525         }
526
527         /* This is not really necessary for now because
528          * this codepath only deals with two-way diffs.
529          */
530         for (i = 0; i < len && line[i] == '@'; i++)
531                 ;
532         if (2 <= i && i < len && line[i] == ' ') {
533                 ecbdata->nparents = i - 1;
534                 set = diff_get_color(ecbdata->color_diff, DIFF_FRAGINFO);
535         }
536         else if (len < ecbdata->nparents)
537                 set = reset;
538         else {
539                 int nparents = ecbdata->nparents;
540                 int color = DIFF_PLAIN;
541                 if (ecbdata->diff_words && nparents != 1)
542                         /* fall back to normal diff */
543                         free_diff_words_data(ecbdata);
544                 if (ecbdata->diff_words) {
545                         if (line[0] == '-') {
546                                 diff_words_append(line, len,
547                                                 &ecbdata->diff_words->minus);
548                                 return;
549                         } else if (line[0] == '+') {
550                                 diff_words_append(line, len,
551                                                 &ecbdata->diff_words->plus);
552                                 return;
553                         }
554                         if (ecbdata->diff_words->minus.text.size ||
555                                         ecbdata->diff_words->plus.text.size)
556                                 diff_words_show(ecbdata->diff_words);
557                         line++;
558                         len--;
559                 } else
560                         for (i = 0; i < nparents && len; i++) {
561                                 if (line[i] == '-')
562                                         color = DIFF_FILE_OLD;
563                                 else if (line[i] == '+')
564                                         color = DIFF_FILE_NEW;
565                         }
566                 set = diff_get_color(ecbdata->color_diff, color);
567         }
568         if (len > 0 && line[len-1] == '\n')
569                 len--;
570         fputs (set, stdout);
571         fwrite (line, len, 1, stdout);
572         puts (reset);
573 }
574
575 static char *pprint_rename(const char *a, const char *b)
576 {
577         const char *old = a;
578         const char *new = b;
579         char *name = NULL;
580         int pfx_length, sfx_length;
581         int len_a = strlen(a);
582         int len_b = strlen(b);
583
584         /* Find common prefix */
585         pfx_length = 0;
586         while (*old && *new && *old == *new) {
587                 if (*old == '/')
588                         pfx_length = old - a + 1;
589                 old++;
590                 new++;
591         }
592
593         /* Find common suffix */
594         old = a + len_a;
595         new = b + len_b;
596         sfx_length = 0;
597         while (a <= old && b <= new && *old == *new) {
598                 if (*old == '/')
599                         sfx_length = len_a - (old - a);
600                 old--;
601                 new--;
602         }
603
604         /*
605          * pfx{mid-a => mid-b}sfx
606          * {pfx-a => pfx-b}sfx
607          * pfx{sfx-a => sfx-b}
608          * name-a => name-b
609          */
610         if (pfx_length + sfx_length) {
611                 int a_midlen = len_a - pfx_length - sfx_length;
612                 int b_midlen = len_b - pfx_length - sfx_length;
613                 if (a_midlen < 0) a_midlen = 0;
614                 if (b_midlen < 0) b_midlen = 0;
615
616                 name = xmalloc(pfx_length + a_midlen + b_midlen + sfx_length + 7);
617                 sprintf(name, "%.*s{%.*s => %.*s}%s",
618                         pfx_length, a,
619                         a_midlen, a + pfx_length,
620                         b_midlen, b + pfx_length,
621                         a + len_a - sfx_length);
622         }
623         else {
624                 name = xmalloc(len_a + len_b + 5);
625                 sprintf(name, "%s => %s", a, b);
626         }
627         return name;
628 }
629
630 struct diffstat_t {
631         struct xdiff_emit_state xm;
632
633         int nr;
634         int alloc;
635         struct diffstat_file {
636                 char *name;
637                 unsigned is_unmerged:1;
638                 unsigned is_binary:1;
639                 unsigned is_renamed:1;
640                 unsigned int added, deleted;
641         } **files;
642 };
643
644 static struct diffstat_file *diffstat_add(struct diffstat_t *diffstat,
645                                           const char *name_a,
646                                           const char *name_b)
647 {
648         struct diffstat_file *x;
649         x = xcalloc(sizeof (*x), 1);
650         if (diffstat->nr == diffstat->alloc) {
651                 diffstat->alloc = alloc_nr(diffstat->alloc);
652                 diffstat->files = xrealloc(diffstat->files,
653                                 diffstat->alloc * sizeof(x));
654         }
655         diffstat->files[diffstat->nr++] = x;
656         if (name_b) {
657                 x->name = pprint_rename(name_a, name_b);
658                 x->is_renamed = 1;
659         }
660         else
661                 x->name = strdup(name_a);
662         return x;
663 }
664
665 static void diffstat_consume(void *priv, char *line, unsigned long len)
666 {
667         struct diffstat_t *diffstat = priv;
668         struct diffstat_file *x = diffstat->files[diffstat->nr - 1];
669
670         if (line[0] == '+')
671                 x->added++;
672         else if (line[0] == '-')
673                 x->deleted++;
674 }
675
676 static const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
677 static const char minuses[]= "----------------------------------------------------------------------";
678 const char mime_boundary_leader[] = "------------";
679
680 static void show_stats(struct diffstat_t* data)
681 {
682         int i, len, add, del, total, adds = 0, dels = 0;
683         int max, max_change = 0, max_len = 0;
684         int total_files = data->nr;
685
686         if (data->nr == 0)
687                 return;
688
689         for (i = 0; i < data->nr; i++) {
690                 struct diffstat_file *file = data->files[i];
691
692                 len = strlen(file->name);
693                 if (max_len < len)
694                         max_len = len;
695
696                 if (file->is_binary || file->is_unmerged)
697                         continue;
698                 if (max_change < file->added + file->deleted)
699                         max_change = file->added + file->deleted;
700         }
701
702         for (i = 0; i < data->nr; i++) {
703                 const char *prefix = "";
704                 char *name = data->files[i]->name;
705                 int added = data->files[i]->added;
706                 int deleted = data->files[i]->deleted;
707
708                 if (0 < (len = quote_c_style(name, NULL, NULL, 0))) {
709                         char *qname = xmalloc(len + 1);
710                         quote_c_style(name, qname, NULL, 0);
711                         free(name);
712                         data->files[i]->name = name = qname;
713                 }
714
715                 /*
716                  * "scale" the filename
717                  */
718                 len = strlen(name);
719                 max = max_len;
720                 if (max > 50)
721                         max = 50;
722                 if (len > max) {
723                         char *slash;
724                         prefix = "...";
725                         max -= 3;
726                         name += len - max;
727                         slash = strchr(name, '/');
728                         if (slash)
729                                 name = slash;
730                 }
731                 len = max;
732
733                 /*
734                  * scale the add/delete
735                  */
736                 max = max_change;
737                 if (max + len > 70)
738                         max = 70 - len;
739
740                 if (data->files[i]->is_binary) {
741                         printf(" %s%-*s |  Bin\n", prefix, len, name);
742                         goto free_diffstat_file;
743                 }
744                 else if (data->files[i]->is_unmerged) {
745                         printf(" %s%-*s |  Unmerged\n", prefix, len, name);
746                         goto free_diffstat_file;
747                 }
748                 else if (!data->files[i]->is_renamed &&
749                          (added + deleted == 0)) {
750                         total_files--;
751                         goto free_diffstat_file;
752                 }
753
754                 add = added;
755                 del = deleted;
756                 total = add + del;
757                 adds += add;
758                 dels += del;
759
760                 if (max_change > 0) {
761                         total = (total * max + max_change / 2) / max_change;
762                         add = (add * max + max_change / 2) / max_change;
763                         del = total - add;
764                 }
765                 printf(" %s%-*s |%5d %.*s%.*s\n", prefix,
766                                 len, name, added + deleted,
767                                 add, pluses, del, minuses);
768         free_diffstat_file:
769                 free(data->files[i]->name);
770                 free(data->files[i]);
771         }
772         free(data->files);
773         printf(" %d files changed, %d insertions(+), %d deletions(-)\n",
774                         total_files, adds, dels);
775 }
776
777 struct checkdiff_t {
778         struct xdiff_emit_state xm;
779         const char *filename;
780         int lineno;
781 };
782
783 static void checkdiff_consume(void *priv, char *line, unsigned long len)
784 {
785         struct checkdiff_t *data = priv;
786
787         if (line[0] == '+') {
788                 int i, spaces = 0;
789
790                 data->lineno++;
791
792                 /* check space before tab */
793                 for (i = 1; i < len && (line[i] == ' ' || line[i] == '\t'); i++)
794                         if (line[i] == ' ')
795                                 spaces++;
796                 if (line[i - 1] == '\t' && spaces)
797                         printf("%s:%d: space before tab:%.*s\n",
798                                 data->filename, data->lineno, (int)len, line);
799
800                 /* check white space at line end */
801                 if (line[len - 1] == '\n')
802                         len--;
803                 if (isspace(line[len - 1]))
804                         printf("%s:%d: white space at end: %.*s\n",
805                                 data->filename, data->lineno, (int)len, line);
806         } else if (line[0] == ' ')
807                 data->lineno++;
808         else if (line[0] == '@') {
809                 char *plus = strchr(line, '+');
810                 if (plus)
811                         data->lineno = strtol(plus, NULL, 10);
812                 else
813                         die("invalid diff");
814         }
815 }
816
817 static unsigned char *deflate_it(char *data,
818                                  unsigned long size,
819                                  unsigned long *result_size)
820 {
821         int bound;
822         unsigned char *deflated;
823         z_stream stream;
824
825         memset(&stream, 0, sizeof(stream));
826         deflateInit(&stream, zlib_compression_level);
827         bound = deflateBound(&stream, size);
828         deflated = xmalloc(bound);
829         stream.next_out = deflated;
830         stream.avail_out = bound;
831
832         stream.next_in = (unsigned char *)data;
833         stream.avail_in = size;
834         while (deflate(&stream, Z_FINISH) == Z_OK)
835                 ; /* nothing */
836         deflateEnd(&stream);
837         *result_size = stream.total_out;
838         return deflated;
839 }
840
841 static void emit_binary_diff(mmfile_t *one, mmfile_t *two)
842 {
843         void *cp;
844         void *delta;
845         void *deflated;
846         void *data;
847         unsigned long orig_size;
848         unsigned long delta_size;
849         unsigned long deflate_size;
850         unsigned long data_size;
851
852         printf("GIT binary patch\n");
853         /* We could do deflated delta, or we could do just deflated two,
854          * whichever is smaller.
855          */
856         delta = NULL;
857         deflated = deflate_it(two->ptr, two->size, &deflate_size);
858         if (one->size && two->size) {
859                 delta = diff_delta(one->ptr, one->size,
860                                    two->ptr, two->size,
861                                    &delta_size, deflate_size);
862                 if (delta) {
863                         void *to_free = delta;
864                         orig_size = delta_size;
865                         delta = deflate_it(delta, delta_size, &delta_size);
866                         free(to_free);
867                 }
868         }
869
870         if (delta && delta_size < deflate_size) {
871                 printf("delta %lu\n", orig_size);
872                 free(deflated);
873                 data = delta;
874                 data_size = delta_size;
875         }
876         else {
877                 printf("literal %lu\n", two->size);
878                 free(delta);
879                 data = deflated;
880                 data_size = deflate_size;
881         }
882
883         /* emit data encoded in base85 */
884         cp = data;
885         while (data_size) {
886                 int bytes = (52 < data_size) ? 52 : data_size;
887                 char line[70];
888                 data_size -= bytes;
889                 if (bytes <= 26)
890                         line[0] = bytes + 'A' - 1;
891                 else
892                         line[0] = bytes - 26 + 'a' - 1;
893                 encode_85(line + 1, cp, bytes);
894                 cp = (char *) cp + bytes;
895                 puts(line);
896         }
897         printf("\n");
898         free(data);
899 }
900
901 #define FIRST_FEW_BYTES 8000
902 static int mmfile_is_binary(mmfile_t *mf)
903 {
904         long sz = mf->size;
905         if (FIRST_FEW_BYTES < sz)
906                 sz = FIRST_FEW_BYTES;
907         return !!memchr(mf->ptr, 0, sz);
908 }
909
910 static void builtin_diff(const char *name_a,
911                          const char *name_b,
912                          struct diff_filespec *one,
913                          struct diff_filespec *two,
914                          const char *xfrm_msg,
915                          struct diff_options *o,
916                          int complete_rewrite)
917 {
918         mmfile_t mf1, mf2;
919         const char *lbl[2];
920         char *a_one, *b_two;
921         const char *set = diff_get_color(o->color_diff, DIFF_METAINFO);
922         const char *reset = diff_get_color(o->color_diff, DIFF_RESET);
923
924         a_one = quote_two("a/", name_a);
925         b_two = quote_two("b/", name_b);
926         lbl[0] = DIFF_FILE_VALID(one) ? a_one : "/dev/null";
927         lbl[1] = DIFF_FILE_VALID(two) ? b_two : "/dev/null";
928         printf("%sdiff --git %s %s%s\n", set, a_one, b_two, reset);
929         if (lbl[0][0] == '/') {
930                 /* /dev/null */
931                 printf("%snew file mode %06o%s\n", set, two->mode, reset);
932                 if (xfrm_msg && xfrm_msg[0])
933                         printf("%s%s%s\n", set, xfrm_msg, reset);
934         }
935         else if (lbl[1][0] == '/') {
936                 printf("%sdeleted file mode %06o%s\n", set, one->mode, reset);
937                 if (xfrm_msg && xfrm_msg[0])
938                         printf("%s%s%s\n", set, xfrm_msg, reset);
939         }
940         else {
941                 if (one->mode != two->mode) {
942                         printf("%sold mode %06o%s\n", set, one->mode, reset);
943                         printf("%snew mode %06o%s\n", set, two->mode, reset);
944                 }
945                 if (xfrm_msg && xfrm_msg[0])
946                         printf("%s%s%s\n", set, xfrm_msg, reset);
947                 /*
948                  * we do not run diff between different kind
949                  * of objects.
950                  */
951                 if ((one->mode ^ two->mode) & S_IFMT)
952                         goto free_ab_and_return;
953                 if (complete_rewrite) {
954                         emit_rewrite_diff(name_a, name_b, one, two);
955                         goto free_ab_and_return;
956                 }
957         }
958
959         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
960                 die("unable to read files to diff");
961
962         if (!o->text && (mmfile_is_binary(&mf1) || mmfile_is_binary(&mf2))) {
963                 /* Quite common confusing case */
964                 if (mf1.size == mf2.size &&
965                     !memcmp(mf1.ptr, mf2.ptr, mf1.size))
966                         goto free_ab_and_return;
967                 if (o->binary)
968                         emit_binary_diff(&mf1, &mf2);
969                 else
970                         printf("Binary files %s and %s differ\n",
971                                lbl[0], lbl[1]);
972         }
973         else {
974                 /* Crazy xdl interfaces.. */
975                 const char *diffopts = getenv("GIT_DIFF_OPTS");
976                 xpparam_t xpp;
977                 xdemitconf_t xecfg;
978                 xdemitcb_t ecb;
979                 struct emit_callback ecbdata;
980
981                 memset(&ecbdata, 0, sizeof(ecbdata));
982                 ecbdata.label_path = lbl;
983                 ecbdata.color_diff = o->color_diff;
984                 xpp.flags = XDF_NEED_MINIMAL | o->xdl_opts;
985                 xecfg.ctxlen = o->context;
986                 xecfg.flags = XDL_EMIT_FUNCNAMES;
987                 if (!diffopts)
988                         ;
989                 else if (!strncmp(diffopts, "--unified=", 10))
990                         xecfg.ctxlen = strtoul(diffopts + 10, NULL, 10);
991                 else if (!strncmp(diffopts, "-u", 2))
992                         xecfg.ctxlen = strtoul(diffopts + 2, NULL, 10);
993                 ecb.outf = xdiff_outf;
994                 ecb.priv = &ecbdata;
995                 ecbdata.xm.consume = fn_out_consume;
996                 if (o->color_diff_words)
997                         ecbdata.diff_words =
998                                 xcalloc(1, sizeof(struct diff_words_data));
999                 xdl_diff(&mf1, &mf2, &xpp, &xecfg, &ecb);
1000                 if (o->color_diff_words)
1001                         free_diff_words_data(&ecbdata);
1002         }
1003
1004  free_ab_and_return:
1005         free(a_one);
1006         free(b_two);
1007         return;
1008 }
1009
1010 static void builtin_diffstat(const char *name_a, const char *name_b,
1011                              struct diff_filespec *one,
1012                              struct diff_filespec *two,
1013                              struct diffstat_t *diffstat,
1014                              struct diff_options *o,
1015                              int complete_rewrite)
1016 {
1017         mmfile_t mf1, mf2;
1018         struct diffstat_file *data;
1019
1020         data = diffstat_add(diffstat, name_a, name_b);
1021
1022         if (!one || !two) {
1023                 data->is_unmerged = 1;
1024                 return;
1025         }
1026         if (complete_rewrite) {
1027                 diff_populate_filespec(one, 0);
1028                 diff_populate_filespec(two, 0);
1029                 data->deleted = count_lines(one->data, one->size);
1030                 data->added = count_lines(two->data, two->size);
1031                 return;
1032         }
1033         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
1034                 die("unable to read files to diff");
1035
1036         if (mmfile_is_binary(&mf1) || mmfile_is_binary(&mf2))
1037                 data->is_binary = 1;
1038         else {
1039                 /* Crazy xdl interfaces.. */
1040                 xpparam_t xpp;
1041                 xdemitconf_t xecfg;
1042                 xdemitcb_t ecb;
1043
1044                 xpp.flags = XDF_NEED_MINIMAL | o->xdl_opts;
1045                 xecfg.ctxlen = 0;
1046                 xecfg.flags = 0;
1047                 ecb.outf = xdiff_outf;
1048                 ecb.priv = diffstat;
1049                 xdl_diff(&mf1, &mf2, &xpp, &xecfg, &ecb);
1050         }
1051 }
1052
1053 static void builtin_checkdiff(const char *name_a, const char *name_b,
1054                              struct diff_filespec *one,
1055                              struct diff_filespec *two)
1056 {
1057         mmfile_t mf1, mf2;
1058         struct checkdiff_t data;
1059
1060         if (!two)
1061                 return;
1062
1063         memset(&data, 0, sizeof(data));
1064         data.xm.consume = checkdiff_consume;
1065         data.filename = name_b ? name_b : name_a;
1066         data.lineno = 0;
1067
1068         if (fill_mmfile(&mf1, one) < 0 || fill_mmfile(&mf2, two) < 0)
1069                 die("unable to read files to diff");
1070
1071         if (mmfile_is_binary(&mf2))
1072                 return;
1073         else {
1074                 /* Crazy xdl interfaces.. */
1075                 xpparam_t xpp;
1076                 xdemitconf_t xecfg;
1077                 xdemitcb_t ecb;
1078
1079                 xpp.flags = XDF_NEED_MINIMAL;
1080                 xecfg.ctxlen = 0;
1081                 xecfg.flags = 0;
1082                 ecb.outf = xdiff_outf;
1083                 ecb.priv = &data;
1084                 xdl_diff(&mf1, &mf2, &xpp, &xecfg, &ecb);
1085         }
1086 }
1087
1088 struct diff_filespec *alloc_filespec(const char *path)
1089 {
1090         int namelen = strlen(path);
1091         struct diff_filespec *spec = xmalloc(sizeof(*spec) + namelen + 1);
1092
1093         memset(spec, 0, sizeof(*spec));
1094         spec->path = (char *)(spec + 1);
1095         memcpy(spec->path, path, namelen+1);
1096         return spec;
1097 }
1098
1099 void fill_filespec(struct diff_filespec *spec, const unsigned char *sha1,
1100                    unsigned short mode)
1101 {
1102         if (mode) {
1103                 spec->mode = canon_mode(mode);
1104                 memcpy(spec->sha1, sha1, 20);
1105                 spec->sha1_valid = !is_null_sha1(sha1);
1106         }
1107 }
1108
1109 /*
1110  * Given a name and sha1 pair, if the dircache tells us the file in
1111  * the work tree has that object contents, return true, so that
1112  * prepare_temp_file() does not have to inflate and extract.
1113  */
1114 static int work_tree_matches(const char *name, const unsigned char *sha1)
1115 {
1116         struct cache_entry *ce;
1117         struct stat st;
1118         int pos, len;
1119
1120         /* We do not read the cache ourselves here, because the
1121          * benchmark with my previous version that always reads cache
1122          * shows that it makes things worse for diff-tree comparing
1123          * two linux-2.6 kernel trees in an already checked out work
1124          * tree.  This is because most diff-tree comparisons deal with
1125          * only a small number of files, while reading the cache is
1126          * expensive for a large project, and its cost outweighs the
1127          * savings we get by not inflating the object to a temporary
1128          * file.  Practically, this code only helps when we are used
1129          * by diff-cache --cached, which does read the cache before
1130          * calling us.
1131          */
1132         if (!active_cache)
1133                 return 0;
1134
1135         len = strlen(name);
1136         pos = cache_name_pos(name, len);
1137         if (pos < 0)
1138                 return 0;
1139         ce = active_cache[pos];
1140         if ((lstat(name, &st) < 0) ||
1141             !S_ISREG(st.st_mode) || /* careful! */
1142             ce_match_stat(ce, &st, 0) ||
1143             memcmp(sha1, ce->sha1, 20))
1144                 return 0;
1145         /* we return 1 only when we can stat, it is a regular file,
1146          * stat information matches, and sha1 recorded in the cache
1147          * matches.  I.e. we know the file in the work tree really is
1148          * the same as the <name, sha1> pair.
1149          */
1150         return 1;
1151 }
1152
1153 static struct sha1_size_cache {
1154         unsigned char sha1[20];
1155         unsigned long size;
1156 } **sha1_size_cache;
1157 static int sha1_size_cache_nr, sha1_size_cache_alloc;
1158
1159 static struct sha1_size_cache *locate_size_cache(unsigned char *sha1,
1160                                                  int find_only,
1161                                                  unsigned long size)
1162 {
1163         int first, last;
1164         struct sha1_size_cache *e;
1165
1166         first = 0;
1167         last = sha1_size_cache_nr;
1168         while (last > first) {
1169                 int cmp, next = (last + first) >> 1;
1170                 e = sha1_size_cache[next];
1171                 cmp = memcmp(e->sha1, sha1, 20);
1172                 if (!cmp)
1173                         return e;
1174                 if (cmp < 0) {
1175                         last = next;
1176                         continue;
1177                 }
1178                 first = next+1;
1179         }
1180         /* not found */
1181         if (find_only)
1182                 return NULL;
1183         /* insert to make it at "first" */
1184         if (sha1_size_cache_alloc <= sha1_size_cache_nr) {
1185                 sha1_size_cache_alloc = alloc_nr(sha1_size_cache_alloc);
1186                 sha1_size_cache = xrealloc(sha1_size_cache,
1187                                            sha1_size_cache_alloc *
1188                                            sizeof(*sha1_size_cache));
1189         }
1190         sha1_size_cache_nr++;
1191         if (first < sha1_size_cache_nr)
1192                 memmove(sha1_size_cache + first + 1, sha1_size_cache + first,
1193                         (sha1_size_cache_nr - first - 1) *
1194                         sizeof(*sha1_size_cache));
1195         e = xmalloc(sizeof(struct sha1_size_cache));
1196         sha1_size_cache[first] = e;
1197         memcpy(e->sha1, sha1, 20);
1198         e->size = size;
1199         return e;
1200 }
1201
1202 /*
1203  * While doing rename detection and pickaxe operation, we may need to
1204  * grab the data for the blob (or file) for our own in-core comparison.
1205  * diff_filespec has data and size fields for this purpose.
1206  */
1207 int diff_populate_filespec(struct diff_filespec *s, int size_only)
1208 {
1209         int err = 0;
1210         if (!DIFF_FILE_VALID(s))
1211                 die("internal error: asking to populate invalid file.");
1212         if (S_ISDIR(s->mode))
1213                 return -1;
1214
1215         if (!use_size_cache)
1216                 size_only = 0;
1217
1218         if (s->data)
1219                 return err;
1220         if (!s->sha1_valid ||
1221             work_tree_matches(s->path, s->sha1)) {
1222                 struct stat st;
1223                 int fd;
1224                 if (lstat(s->path, &st) < 0) {
1225                         if (errno == ENOENT) {
1226                         err_empty:
1227                                 err = -1;
1228                         empty:
1229                                 s->data = (char *)"";
1230                                 s->size = 0;
1231                                 return err;
1232                         }
1233                 }
1234                 s->size = st.st_size;
1235                 if (!s->size)
1236                         goto empty;
1237                 if (size_only)
1238                         return 0;
1239                 if (S_ISLNK(st.st_mode)) {
1240                         int ret;
1241                         s->data = xmalloc(s->size);
1242                         s->should_free = 1;
1243                         ret = readlink(s->path, s->data, s->size);
1244                         if (ret < 0) {
1245                                 free(s->data);
1246                                 goto err_empty;
1247                         }
1248                         return 0;
1249                 }
1250                 fd = open(s->path, O_RDONLY);
1251                 if (fd < 0)
1252                         goto err_empty;
1253                 s->data = mmap(NULL, s->size, PROT_READ, MAP_PRIVATE, fd, 0);
1254                 close(fd);
1255                 if (s->data == MAP_FAILED)
1256                         goto err_empty;
1257                 s->should_munmap = 1;
1258         }
1259         else {
1260                 char type[20];
1261                 struct sha1_size_cache *e;
1262
1263                 if (size_only) {
1264                         e = locate_size_cache(s->sha1, 1, 0);
1265                         if (e) {
1266                                 s->size = e->size;
1267                                 return 0;
1268                         }
1269                         if (!sha1_object_info(s->sha1, type, &s->size))
1270                                 locate_size_cache(s->sha1, 0, s->size);
1271                 }
1272                 else {
1273                         s->data = read_sha1_file(s->sha1, type, &s->size);
1274                         s->should_free = 1;
1275                 }
1276         }
1277         return 0;
1278 }
1279
1280 void diff_free_filespec_data(struct diff_filespec *s)
1281 {
1282         if (s->should_free)
1283                 free(s->data);
1284         else if (s->should_munmap)
1285                 munmap(s->data, s->size);
1286         s->should_free = s->should_munmap = 0;
1287         s->data = NULL;
1288         free(s->cnt_data);
1289         s->cnt_data = NULL;
1290 }
1291
1292 static void prep_temp_blob(struct diff_tempfile *temp,
1293                            void *blob,
1294                            unsigned long size,
1295                            const unsigned char *sha1,
1296                            int mode)
1297 {
1298         int fd;
1299
1300         fd = git_mkstemp(temp->tmp_path, TEMPFILE_PATH_LEN, ".diff_XXXXXX");
1301         if (fd < 0)
1302                 die("unable to create temp-file");
1303         if (write(fd, blob, size) != size)
1304                 die("unable to write temp-file");
1305         close(fd);
1306         temp->name = temp->tmp_path;
1307         strcpy(temp->hex, sha1_to_hex(sha1));
1308         temp->hex[40] = 0;
1309         sprintf(temp->mode, "%06o", mode);
1310 }
1311
1312 static void prepare_temp_file(const char *name,
1313                               struct diff_tempfile *temp,
1314                               struct diff_filespec *one)
1315 {
1316         if (!DIFF_FILE_VALID(one)) {
1317         not_a_valid_file:
1318                 /* A '-' entry produces this for file-2, and
1319                  * a '+' entry produces this for file-1.
1320                  */
1321                 temp->name = "/dev/null";
1322                 strcpy(temp->hex, ".");
1323                 strcpy(temp->mode, ".");
1324                 return;
1325         }
1326
1327         if (!one->sha1_valid ||
1328             work_tree_matches(name, one->sha1)) {
1329                 struct stat st;
1330                 if (lstat(name, &st) < 0) {
1331                         if (errno == ENOENT)
1332                                 goto not_a_valid_file;
1333                         die("stat(%s): %s", name, strerror(errno));
1334                 }
1335                 if (S_ISLNK(st.st_mode)) {
1336                         int ret;
1337                         char buf[PATH_MAX + 1]; /* ought to be SYMLINK_MAX */
1338                         if (sizeof(buf) <= st.st_size)
1339                                 die("symlink too long: %s", name);
1340                         ret = readlink(name, buf, st.st_size);
1341                         if (ret < 0)
1342                                 die("readlink(%s)", name);
1343                         prep_temp_blob(temp, buf, st.st_size,
1344                                        (one->sha1_valid ?
1345                                         one->sha1 : null_sha1),
1346                                        (one->sha1_valid ?
1347                                         one->mode : S_IFLNK));
1348                 }
1349                 else {
1350                         /* we can borrow from the file in the work tree */
1351                         temp->name = name;
1352                         if (!one->sha1_valid)
1353                                 strcpy(temp->hex, sha1_to_hex(null_sha1));
1354                         else
1355                                 strcpy(temp->hex, sha1_to_hex(one->sha1));
1356                         /* Even though we may sometimes borrow the
1357                          * contents from the work tree, we always want
1358                          * one->mode.  mode is trustworthy even when
1359                          * !(one->sha1_valid), as long as
1360                          * DIFF_FILE_VALID(one).
1361                          */
1362                         sprintf(temp->mode, "%06o", one->mode);
1363                 }
1364                 return;
1365         }
1366         else {
1367                 if (diff_populate_filespec(one, 0))
1368                         die("cannot read data blob for %s", one->path);
1369                 prep_temp_blob(temp, one->data, one->size,
1370                                one->sha1, one->mode);
1371         }
1372 }
1373
1374 static void remove_tempfile(void)
1375 {
1376         int i;
1377
1378         for (i = 0; i < 2; i++)
1379                 if (diff_temp[i].name == diff_temp[i].tmp_path) {
1380                         unlink(diff_temp[i].name);
1381                         diff_temp[i].name = NULL;
1382                 }
1383 }
1384
1385 static void remove_tempfile_on_signal(int signo)
1386 {
1387         remove_tempfile();
1388         signal(SIGINT, SIG_DFL);
1389         raise(signo);
1390 }
1391
1392 static int spawn_prog(const char *pgm, const char **arg)
1393 {
1394         pid_t pid;
1395         int status;
1396
1397         fflush(NULL);
1398         pid = fork();
1399         if (pid < 0)
1400                 die("unable to fork");
1401         if (!pid) {
1402                 execvp(pgm, (char *const*) arg);
1403                 exit(255);
1404         }
1405
1406         while (waitpid(pid, &status, 0) < 0) {
1407                 if (errno == EINTR)
1408                         continue;
1409                 return -1;
1410         }
1411
1412         /* Earlier we did not check the exit status because
1413          * diff exits non-zero if files are different, and
1414          * we are not interested in knowing that.  It was a
1415          * mistake which made it harder to quit a diff-*
1416          * session that uses the git-apply-patch-script as
1417          * the GIT_EXTERNAL_DIFF.  A custom GIT_EXTERNAL_DIFF
1418          * should also exit non-zero only when it wants to
1419          * abort the entire diff-* session.
1420          */
1421         if (WIFEXITED(status) && !WEXITSTATUS(status))
1422                 return 0;
1423         return -1;
1424 }
1425
1426 /* An external diff command takes:
1427  *
1428  * diff-cmd name infile1 infile1-sha1 infile1-mode \
1429  *               infile2 infile2-sha1 infile2-mode [ rename-to ]
1430  *
1431  */
1432 static void run_external_diff(const char *pgm,
1433                               const char *name,
1434                               const char *other,
1435                               struct diff_filespec *one,
1436                               struct diff_filespec *two,
1437                               const char *xfrm_msg,
1438                               int complete_rewrite)
1439 {
1440         const char *spawn_arg[10];
1441         struct diff_tempfile *temp = diff_temp;
1442         int retval;
1443         static int atexit_asked = 0;
1444         const char *othername;
1445         const char **arg = &spawn_arg[0];
1446
1447         othername = (other? other : name);
1448         if (one && two) {
1449                 prepare_temp_file(name, &temp[0], one);
1450                 prepare_temp_file(othername, &temp[1], two);
1451                 if (! atexit_asked &&
1452                     (temp[0].name == temp[0].tmp_path ||
1453                      temp[1].name == temp[1].tmp_path)) {
1454                         atexit_asked = 1;
1455                         atexit(remove_tempfile);
1456                 }
1457                 signal(SIGINT, remove_tempfile_on_signal);
1458         }
1459
1460         if (one && two) {
1461                 *arg++ = pgm;
1462                 *arg++ = name;
1463                 *arg++ = temp[0].name;
1464                 *arg++ = temp[0].hex;
1465                 *arg++ = temp[0].mode;
1466                 *arg++ = temp[1].name;
1467                 *arg++ = temp[1].hex;
1468                 *arg++ = temp[1].mode;
1469                 if (other) {
1470                         *arg++ = other;
1471                         *arg++ = xfrm_msg;
1472                 }
1473         } else {
1474                 *arg++ = pgm;
1475                 *arg++ = name;
1476         }
1477         *arg = NULL;
1478         retval = spawn_prog(pgm, spawn_arg);
1479         remove_tempfile();
1480         if (retval) {
1481                 fprintf(stderr, "external diff died, stopping at %s.\n", name);
1482                 exit(1);
1483         }
1484 }
1485
1486 static void run_diff_cmd(const char *pgm,
1487                          const char *name,
1488                          const char *other,
1489                          struct diff_filespec *one,
1490                          struct diff_filespec *two,
1491                          const char *xfrm_msg,
1492                          struct diff_options *o,
1493                          int complete_rewrite)
1494 {
1495         if (pgm) {
1496                 run_external_diff(pgm, name, other, one, two, xfrm_msg,
1497                                   complete_rewrite);
1498                 return;
1499         }
1500         if (one && two)
1501                 builtin_diff(name, other ? other : name,
1502                              one, two, xfrm_msg, o, complete_rewrite);
1503         else
1504                 printf("* Unmerged path %s\n", name);
1505 }
1506
1507 static void diff_fill_sha1_info(struct diff_filespec *one)
1508 {
1509         if (DIFF_FILE_VALID(one)) {
1510                 if (!one->sha1_valid) {
1511                         struct stat st;
1512                         if (lstat(one->path, &st) < 0)
1513                                 die("stat %s", one->path);
1514                         if (index_path(one->sha1, one->path, &st, 0))
1515                                 die("cannot hash %s\n", one->path);
1516                 }
1517         }
1518         else
1519                 memset(one->sha1, 0, 20);
1520 }
1521
1522 static void run_diff(struct diff_filepair *p, struct diff_options *o)
1523 {
1524         const char *pgm = external_diff();
1525         char msg[PATH_MAX*2+300], *xfrm_msg;
1526         struct diff_filespec *one;
1527         struct diff_filespec *two;
1528         const char *name;
1529         const char *other;
1530         char *name_munged, *other_munged;
1531         int complete_rewrite = 0;
1532         int len;
1533
1534         if (DIFF_PAIR_UNMERGED(p)) {
1535                 /* unmerged */
1536                 run_diff_cmd(pgm, p->one->path, NULL, NULL, NULL, NULL, o, 0);
1537                 return;
1538         }
1539
1540         name = p->one->path;
1541         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
1542         name_munged = quote_one(name);
1543         other_munged = quote_one(other);
1544         one = p->one; two = p->two;
1545
1546         diff_fill_sha1_info(one);
1547         diff_fill_sha1_info(two);
1548
1549         len = 0;
1550         switch (p->status) {
1551         case DIFF_STATUS_COPIED:
1552                 len += snprintf(msg + len, sizeof(msg) - len,
1553                                 "similarity index %d%%\n"
1554                                 "copy from %s\n"
1555                                 "copy to %s\n",
1556                                 (int)(0.5 + p->score * 100.0/MAX_SCORE),
1557                                 name_munged, other_munged);
1558                 break;
1559         case DIFF_STATUS_RENAMED:
1560                 len += snprintf(msg + len, sizeof(msg) - len,
1561                                 "similarity index %d%%\n"
1562                                 "rename from %s\n"
1563                                 "rename to %s\n",
1564                                 (int)(0.5 + p->score * 100.0/MAX_SCORE),
1565                                 name_munged, other_munged);
1566                 break;
1567         case DIFF_STATUS_MODIFIED:
1568                 if (p->score) {
1569                         len += snprintf(msg + len, sizeof(msg) - len,
1570                                         "dissimilarity index %d%%\n",
1571                                         (int)(0.5 + p->score *
1572                                               100.0/MAX_SCORE));
1573                         complete_rewrite = 1;
1574                         break;
1575                 }
1576                 /* fallthru */
1577         default:
1578                 /* nothing */
1579                 ;
1580         }
1581
1582         if (memcmp(one->sha1, two->sha1, 20)) {
1583                 int abbrev = o->full_index ? 40 : DEFAULT_ABBREV;
1584
1585                 len += snprintf(msg + len, sizeof(msg) - len,
1586                                 "index %.*s..%.*s",
1587                                 abbrev, sha1_to_hex(one->sha1),
1588                                 abbrev, sha1_to_hex(two->sha1));
1589                 if (one->mode == two->mode)
1590                         len += snprintf(msg + len, sizeof(msg) - len,
1591                                         " %06o", one->mode);
1592                 len += snprintf(msg + len, sizeof(msg) - len, "\n");
1593         }
1594
1595         if (len)
1596                 msg[--len] = 0;
1597         xfrm_msg = len ? msg : NULL;
1598
1599         if (!pgm &&
1600             DIFF_FILE_VALID(one) && DIFF_FILE_VALID(two) &&
1601             (S_IFMT & one->mode) != (S_IFMT & two->mode)) {
1602                 /* a filepair that changes between file and symlink
1603                  * needs to be split into deletion and creation.
1604                  */
1605                 struct diff_filespec *null = alloc_filespec(two->path);
1606                 run_diff_cmd(NULL, name, other, one, null, xfrm_msg, o, 0);
1607                 free(null);
1608                 null = alloc_filespec(one->path);
1609                 run_diff_cmd(NULL, name, other, null, two, xfrm_msg, o, 0);
1610                 free(null);
1611         }
1612         else
1613                 run_diff_cmd(pgm, name, other, one, two, xfrm_msg, o,
1614                              complete_rewrite);
1615
1616         free(name_munged);
1617         free(other_munged);
1618 }
1619
1620 static void run_diffstat(struct diff_filepair *p, struct diff_options *o,
1621                          struct diffstat_t *diffstat)
1622 {
1623         const char *name;
1624         const char *other;
1625         int complete_rewrite = 0;
1626
1627         if (DIFF_PAIR_UNMERGED(p)) {
1628                 /* unmerged */
1629                 builtin_diffstat(p->one->path, NULL, NULL, NULL, diffstat, o, 0);
1630                 return;
1631         }
1632
1633         name = p->one->path;
1634         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
1635
1636         diff_fill_sha1_info(p->one);
1637         diff_fill_sha1_info(p->two);
1638
1639         if (p->status == DIFF_STATUS_MODIFIED && p->score)
1640                 complete_rewrite = 1;
1641         builtin_diffstat(name, other, p->one, p->two, diffstat, o, complete_rewrite);
1642 }
1643
1644 static void run_checkdiff(struct diff_filepair *p, struct diff_options *o)
1645 {
1646         const char *name;
1647         const char *other;
1648
1649         if (DIFF_PAIR_UNMERGED(p)) {
1650                 /* unmerged */
1651                 return;
1652         }
1653
1654         name = p->one->path;
1655         other = (strcmp(name, p->two->path) ? p->two->path : NULL);
1656
1657         diff_fill_sha1_info(p->one);
1658         diff_fill_sha1_info(p->two);
1659
1660         builtin_checkdiff(name, other, p->one, p->two);
1661 }
1662
1663 void diff_setup(struct diff_options *options)
1664 {
1665         memset(options, 0, sizeof(*options));
1666         options->line_termination = '\n';
1667         options->break_opt = -1;
1668         options->rename_limit = -1;
1669         options->context = 3;
1670         options->msg_sep = "";
1671
1672         options->change = diff_change;
1673         options->add_remove = diff_addremove;
1674         options->color_diff = diff_use_color_default;
1675         options->detect_rename = diff_detect_rename_default;
1676 }
1677
1678 int diff_setup_done(struct diff_options *options)
1679 {
1680         int count = 0;
1681
1682         if (options->output_format & DIFF_FORMAT_NAME)
1683                 count++;
1684         if (options->output_format & DIFF_FORMAT_NAME_STATUS)
1685                 count++;
1686         if (options->output_format & DIFF_FORMAT_CHECKDIFF)
1687                 count++;
1688         if (options->output_format & DIFF_FORMAT_NO_OUTPUT)
1689                 count++;
1690         if (count > 1)
1691                 die("--name-only, --name-status, --check and -s are mutually exclusive");
1692
1693         if (options->find_copies_harder)
1694                 options->detect_rename = DIFF_DETECT_COPY;
1695
1696         if (options->output_format & (DIFF_FORMAT_NAME |
1697                                       DIFF_FORMAT_NAME_STATUS |
1698                                       DIFF_FORMAT_CHECKDIFF |
1699                                       DIFF_FORMAT_NO_OUTPUT))
1700                 options->output_format &= ~(DIFF_FORMAT_RAW |
1701                                             DIFF_FORMAT_DIFFSTAT |
1702                                             DIFF_FORMAT_SUMMARY |
1703                                             DIFF_FORMAT_PATCH);
1704
1705         /*
1706          * These cases always need recursive; we do not drop caller-supplied
1707          * recursive bits for other formats here.
1708          */
1709         if (options->output_format & (DIFF_FORMAT_PATCH |
1710                                       DIFF_FORMAT_DIFFSTAT |
1711                                       DIFF_FORMAT_CHECKDIFF))
1712                 options->recursive = 1;
1713         /*
1714          * Also pickaxe would not work very well if you do not say recursive
1715          */
1716         if (options->pickaxe)
1717                 options->recursive = 1;
1718
1719         if (options->detect_rename && options->rename_limit < 0)
1720                 options->rename_limit = diff_rename_limit_default;
1721         if (options->setup & DIFF_SETUP_USE_CACHE) {
1722                 if (!active_cache)
1723                         /* read-cache does not die even when it fails
1724                          * so it is safe for us to do this here.  Also
1725                          * it does not smudge active_cache or active_nr
1726                          * when it fails, so we do not have to worry about
1727                          * cleaning it up ourselves either.
1728                          */
1729                         read_cache();
1730         }
1731         if (options->setup & DIFF_SETUP_USE_SIZE_CACHE)
1732                 use_size_cache = 1;
1733         if (options->abbrev <= 0 || 40 < options->abbrev)
1734                 options->abbrev = 40; /* full */
1735
1736         return 0;
1737 }
1738
1739 static int opt_arg(const char *arg, int arg_short, const char *arg_long, int *val)
1740 {
1741         char c, *eq;
1742         int len;
1743
1744         if (*arg != '-')
1745                 return 0;
1746         c = *++arg;
1747         if (!c)
1748                 return 0;
1749         if (c == arg_short) {
1750                 c = *++arg;
1751                 if (!c)
1752                         return 1;
1753                 if (val && isdigit(c)) {
1754                         char *end;
1755                         int n = strtoul(arg, &end, 10);
1756                         if (*end)
1757                                 return 0;
1758                         *val = n;
1759                         return 1;
1760                 }
1761                 return 0;
1762         }
1763         if (c != '-')
1764                 return 0;
1765         arg++;
1766         eq = strchr(arg, '=');
1767         if (eq)
1768                 len = eq - arg;
1769         else
1770                 len = strlen(arg);
1771         if (!len || strncmp(arg, arg_long, len))
1772                 return 0;
1773         if (eq) {
1774                 int n;
1775                 char *end;
1776                 if (!isdigit(*++eq))
1777                         return 0;
1778                 n = strtoul(eq, &end, 10);
1779                 if (*end)
1780                         return 0;
1781                 *val = n;
1782         }
1783         return 1;
1784 }
1785
1786 int diff_opt_parse(struct diff_options *options, const char **av, int ac)
1787 {
1788         const char *arg = av[0];
1789         if (!strcmp(arg, "-p") || !strcmp(arg, "-u"))
1790                 options->output_format |= DIFF_FORMAT_PATCH;
1791         else if (opt_arg(arg, 'U', "unified", &options->context))
1792                 options->output_format |= DIFF_FORMAT_PATCH;
1793         else if (!strcmp(arg, "--raw"))
1794                 options->output_format |= DIFF_FORMAT_RAW;
1795         else if (!strcmp(arg, "--patch-with-raw")) {
1796                 options->output_format |= DIFF_FORMAT_PATCH | DIFF_FORMAT_RAW;
1797         }
1798         else if (!strcmp(arg, "--stat"))
1799                 options->output_format |= DIFF_FORMAT_DIFFSTAT;
1800         else if (!strcmp(arg, "--check"))
1801                 options->output_format |= DIFF_FORMAT_CHECKDIFF;
1802         else if (!strcmp(arg, "--summary"))
1803                 options->output_format |= DIFF_FORMAT_SUMMARY;
1804         else if (!strcmp(arg, "--patch-with-stat")) {
1805                 options->output_format |= DIFF_FORMAT_PATCH | DIFF_FORMAT_DIFFSTAT;
1806         }
1807         else if (!strcmp(arg, "-z"))
1808                 options->line_termination = 0;
1809         else if (!strncmp(arg, "-l", 2))
1810                 options->rename_limit = strtoul(arg+2, NULL, 10);
1811         else if (!strcmp(arg, "--full-index"))
1812                 options->full_index = 1;
1813         else if (!strcmp(arg, "--binary")) {
1814                 options->output_format |= DIFF_FORMAT_PATCH;
1815                 options->full_index = options->binary = 1;
1816         }
1817         else if (!strcmp(arg, "-a") || !strcmp(arg, "--text")) {
1818                 options->text = 1;
1819         }
1820         else if (!strcmp(arg, "--name-only"))
1821                 options->output_format |= DIFF_FORMAT_NAME;
1822         else if (!strcmp(arg, "--name-status"))
1823                 options->output_format |= DIFF_FORMAT_NAME_STATUS;
1824         else if (!strcmp(arg, "-R"))
1825                 options->reverse_diff = 1;
1826         else if (!strncmp(arg, "-S", 2))
1827                 options->pickaxe = arg + 2;
1828         else if (!strcmp(arg, "-s")) {
1829                 options->output_format |= DIFF_FORMAT_NO_OUTPUT;
1830         }
1831         else if (!strncmp(arg, "-O", 2))
1832                 options->orderfile = arg + 2;
1833         else if (!strncmp(arg, "--diff-filter=", 14))
1834                 options->filter = arg + 14;
1835         else if (!strcmp(arg, "--pickaxe-all"))
1836                 options->pickaxe_opts = DIFF_PICKAXE_ALL;
1837         else if (!strcmp(arg, "--pickaxe-regex"))
1838                 options->pickaxe_opts = DIFF_PICKAXE_REGEX;
1839         else if (!strncmp(arg, "-B", 2)) {
1840                 if ((options->break_opt =
1841                      diff_scoreopt_parse(arg)) == -1)
1842                         return -1;
1843         }
1844         else if (!strncmp(arg, "-M", 2)) {
1845                 if ((options->rename_score =
1846                      diff_scoreopt_parse(arg)) == -1)
1847                         return -1;
1848                 options->detect_rename = DIFF_DETECT_RENAME;
1849         }
1850         else if (!strncmp(arg, "-C", 2)) {
1851                 if ((options->rename_score =
1852                      diff_scoreopt_parse(arg)) == -1)
1853                         return -1;
1854                 options->detect_rename = DIFF_DETECT_COPY;
1855         }
1856         else if (!strcmp(arg, "--find-copies-harder"))
1857                 options->find_copies_harder = 1;
1858         else if (!strcmp(arg, "--abbrev"))
1859                 options->abbrev = DEFAULT_ABBREV;
1860         else if (!strncmp(arg, "--abbrev=", 9)) {
1861                 options->abbrev = strtoul(arg + 9, NULL, 10);
1862                 if (options->abbrev < MINIMUM_ABBREV)
1863                         options->abbrev = MINIMUM_ABBREV;
1864                 else if (40 < options->abbrev)
1865                         options->abbrev = 40;
1866         }
1867         else if (!strcmp(arg, "--color"))
1868                 options->color_diff = 1;
1869         else if (!strcmp(arg, "--no-color"))
1870                 options->color_diff = 0;
1871         else if (!strcmp(arg, "-w") || !strcmp(arg, "--ignore-all-space"))
1872                 options->xdl_opts |= XDF_IGNORE_WHITESPACE;
1873         else if (!strcmp(arg, "-b") || !strcmp(arg, "--ignore-space-change"))
1874                 options->xdl_opts |= XDF_IGNORE_WHITESPACE_CHANGE;
1875         else if (!strcmp(arg, "--color-words"))
1876                 options->color_diff = options->color_diff_words = 1;
1877         else if (!strcmp(arg, "--no-renames"))
1878                 options->detect_rename = 0;
1879         else
1880                 return 0;
1881         return 1;
1882 }
1883
1884 static int parse_num(const char **cp_p)
1885 {
1886         unsigned long num, scale;
1887         int ch, dot;
1888         const char *cp = *cp_p;
1889
1890         num = 0;
1891         scale = 1;
1892         dot = 0;
1893         for(;;) {
1894                 ch = *cp;
1895                 if ( !dot && ch == '.' ) {
1896                         scale = 1;
1897                         dot = 1;
1898                 } else if ( ch == '%' ) {
1899                         scale = dot ? scale*100 : 100;
1900                         cp++;   /* % is always at the end */
1901                         break;
1902                 } else if ( ch >= '0' && ch <= '9' ) {
1903                         if ( scale < 100000 ) {
1904                                 scale *= 10;
1905                                 num = (num*10) + (ch-'0');
1906                         }
1907                 } else {
1908                         break;
1909                 }
1910                 cp++;
1911         }
1912         *cp_p = cp;
1913
1914         /* user says num divided by scale and we say internally that
1915          * is MAX_SCORE * num / scale.
1916          */
1917         return (num >= scale) ? MAX_SCORE : (MAX_SCORE * num / scale);
1918 }
1919
1920 int diff_scoreopt_parse(const char *opt)
1921 {
1922         int opt1, opt2, cmd;
1923
1924         if (*opt++ != '-')
1925                 return -1;
1926         cmd = *opt++;
1927         if (cmd != 'M' && cmd != 'C' && cmd != 'B')
1928                 return -1; /* that is not a -M, -C nor -B option */
1929
1930         opt1 = parse_num(&opt);
1931         if (cmd != 'B')
1932                 opt2 = 0;
1933         else {
1934                 if (*opt == 0)
1935                         opt2 = 0;
1936                 else if (*opt != '/')
1937                         return -1; /* we expect -B80/99 or -B80 */
1938                 else {
1939                         opt++;
1940                         opt2 = parse_num(&opt);
1941                 }
1942         }
1943         if (*opt != 0)
1944                 return -1;
1945         return opt1 | (opt2 << 16);
1946 }
1947
1948 struct diff_queue_struct diff_queued_diff;
1949
1950 void diff_q(struct diff_queue_struct *queue, struct diff_filepair *dp)
1951 {
1952         if (queue->alloc <= queue->nr) {
1953                 queue->alloc = alloc_nr(queue->alloc);
1954                 queue->queue = xrealloc(queue->queue,
1955                                         sizeof(dp) * queue->alloc);
1956         }
1957         queue->queue[queue->nr++] = dp;
1958 }
1959
1960 struct diff_filepair *diff_queue(struct diff_queue_struct *queue,
1961                                  struct diff_filespec *one,
1962                                  struct diff_filespec *two)
1963 {
1964         struct diff_filepair *dp = xcalloc(1, sizeof(*dp));
1965         dp->one = one;
1966         dp->two = two;
1967         if (queue)
1968                 diff_q(queue, dp);
1969         return dp;
1970 }
1971
1972 void diff_free_filepair(struct diff_filepair *p)
1973 {
1974         diff_free_filespec_data(p->one);
1975         diff_free_filespec_data(p->two);
1976         free(p->one);
1977         free(p->two);
1978         free(p);
1979 }
1980
1981 /* This is different from find_unique_abbrev() in that
1982  * it stuffs the result with dots for alignment.
1983  */
1984 const char *diff_unique_abbrev(const unsigned char *sha1, int len)
1985 {
1986         int abblen;
1987         const char *abbrev;
1988         if (len == 40)
1989                 return sha1_to_hex(sha1);
1990
1991         abbrev = find_unique_abbrev(sha1, len);
1992         if (!abbrev)
1993                 return sha1_to_hex(sha1);
1994         abblen = strlen(abbrev);
1995         if (abblen < 37) {
1996                 static char hex[41];
1997                 if (len < abblen && abblen <= len + 2)
1998                         sprintf(hex, "%s%.*s", abbrev, len+3-abblen, "..");
1999                 else
2000                         sprintf(hex, "%s...", abbrev);
2001                 return hex;
2002         }
2003         return sha1_to_hex(sha1);
2004 }
2005
2006 static void diff_flush_raw(struct diff_filepair *p,
2007                            struct diff_options *options)
2008 {
2009         int two_paths;
2010         char status[10];
2011         int abbrev = options->abbrev;
2012         const char *path_one, *path_two;
2013         int inter_name_termination = '\t';
2014         int line_termination = options->line_termination;
2015
2016         if (!line_termination)
2017                 inter_name_termination = 0;
2018
2019         path_one = p->one->path;
2020         path_two = p->two->path;
2021         if (line_termination) {
2022                 path_one = quote_one(path_one);
2023                 path_two = quote_one(path_two);
2024         }
2025
2026         if (p->score)
2027                 sprintf(status, "%c%03d", p->status,
2028                         (int)(0.5 + p->score * 100.0/MAX_SCORE));
2029         else {
2030                 status[0] = p->status;
2031                 status[1] = 0;
2032         }
2033         switch (p->status) {
2034         case DIFF_STATUS_COPIED:
2035         case DIFF_STATUS_RENAMED:
2036                 two_paths = 1;
2037                 break;
2038         case DIFF_STATUS_ADDED:
2039         case DIFF_STATUS_DELETED:
2040                 two_paths = 0;
2041                 break;
2042         default:
2043                 two_paths = 0;
2044                 break;
2045         }
2046         if (!(options->output_format & DIFF_FORMAT_NAME_STATUS)) {
2047                 printf(":%06o %06o %s ",
2048                        p->one->mode, p->two->mode,
2049                        diff_unique_abbrev(p->one->sha1, abbrev));
2050                 printf("%s ",
2051                        diff_unique_abbrev(p->two->sha1, abbrev));
2052         }
2053         printf("%s%c%s", status, inter_name_termination, path_one);
2054         if (two_paths)
2055                 printf("%c%s", inter_name_termination, path_two);
2056         putchar(line_termination);
2057         if (path_one != p->one->path)
2058                 free((void*)path_one);
2059         if (path_two != p->two->path)
2060                 free((void*)path_two);
2061 }
2062
2063 static void diff_flush_name(struct diff_filepair *p, int line_termination)
2064 {
2065         char *path = p->two->path;
2066
2067         if (line_termination)
2068                 path = quote_one(p->two->path);
2069         printf("%s%c", path, line_termination);
2070         if (p->two->path != path)
2071                 free(path);
2072 }
2073
2074 int diff_unmodified_pair(struct diff_filepair *p)
2075 {
2076         /* This function is written stricter than necessary to support
2077          * the currently implemented transformers, but the idea is to
2078          * let transformers to produce diff_filepairs any way they want,
2079          * and filter and clean them up here before producing the output.
2080          */
2081         struct diff_filespec *one, *two;
2082
2083         if (DIFF_PAIR_UNMERGED(p))
2084                 return 0; /* unmerged is interesting */
2085
2086         one = p->one;
2087         two = p->two;
2088
2089         /* deletion, addition, mode or type change
2090          * and rename are all interesting.
2091          */
2092         if (DIFF_FILE_VALID(one) != DIFF_FILE_VALID(two) ||
2093             DIFF_PAIR_MODE_CHANGED(p) ||
2094             strcmp(one->path, two->path))
2095                 return 0;
2096
2097         /* both are valid and point at the same path.  that is, we are
2098          * dealing with a change.
2099          */
2100         if (one->sha1_valid && two->sha1_valid &&
2101             !memcmp(one->sha1, two->sha1, sizeof(one->sha1)))
2102                 return 1; /* no change */
2103         if (!one->sha1_valid && !two->sha1_valid)
2104                 return 1; /* both look at the same file on the filesystem. */
2105         return 0;
2106 }
2107
2108 static void diff_flush_patch(struct diff_filepair *p, struct diff_options *o)
2109 {
2110         if (diff_unmodified_pair(p))
2111                 return;
2112
2113         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
2114             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
2115                 return; /* no tree diffs in patch format */
2116
2117         run_diff(p, o);
2118 }
2119
2120 static void diff_flush_stat(struct diff_filepair *p, struct diff_options *o,
2121                             struct diffstat_t *diffstat)
2122 {
2123         if (diff_unmodified_pair(p))
2124                 return;
2125
2126         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
2127             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
2128                 return; /* no tree diffs in patch format */
2129
2130         run_diffstat(p, o, diffstat);
2131 }
2132
2133 static void diff_flush_checkdiff(struct diff_filepair *p,
2134                 struct diff_options *o)
2135 {
2136         if (diff_unmodified_pair(p))
2137                 return;
2138
2139         if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
2140             (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
2141                 return; /* no tree diffs in patch format */
2142
2143         run_checkdiff(p, o);
2144 }
2145
2146 int diff_queue_is_empty(void)
2147 {
2148         struct diff_queue_struct *q = &diff_queued_diff;
2149         int i;
2150         for (i = 0; i < q->nr; i++)
2151                 if (!diff_unmodified_pair(q->queue[i]))
2152                         return 0;
2153         return 1;
2154 }
2155
2156 #if DIFF_DEBUG
2157 void diff_debug_filespec(struct diff_filespec *s, int x, const char *one)
2158 {
2159         fprintf(stderr, "queue[%d] %s (%s) %s %06o %s\n",
2160                 x, one ? one : "",
2161                 s->path,
2162                 DIFF_FILE_VALID(s) ? "valid" : "invalid",
2163                 s->mode,
2164                 s->sha1_valid ? sha1_to_hex(s->sha1) : "");
2165         fprintf(stderr, "queue[%d] %s size %lu flags %d\n",
2166                 x, one ? one : "",
2167                 s->size, s->xfrm_flags);
2168 }
2169
2170 void diff_debug_filepair(const struct diff_filepair *p, int i)
2171 {
2172         diff_debug_filespec(p->one, i, "one");
2173         diff_debug_filespec(p->two, i, "two");
2174         fprintf(stderr, "score %d, status %c stays %d broken %d\n",
2175                 p->score, p->status ? p->status : '?',
2176                 p->source_stays, p->broken_pair);
2177 }
2178
2179 void diff_debug_queue(const char *msg, struct diff_queue_struct *q)
2180 {
2181         int i;
2182         if (msg)
2183                 fprintf(stderr, "%s\n", msg);
2184         fprintf(stderr, "q->nr = %d\n", q->nr);
2185         for (i = 0; i < q->nr; i++) {
2186                 struct diff_filepair *p = q->queue[i];
2187                 diff_debug_filepair(p, i);
2188         }
2189 }
2190 #endif
2191
2192 static void diff_resolve_rename_copy(void)
2193 {
2194         int i, j;
2195         struct diff_filepair *p, *pp;
2196         struct diff_queue_struct *q = &diff_queued_diff;
2197
2198         diff_debug_queue("resolve-rename-copy", q);
2199
2200         for (i = 0; i < q->nr; i++) {
2201                 p = q->queue[i];
2202                 p->status = 0; /* undecided */
2203                 if (DIFF_PAIR_UNMERGED(p))
2204                         p->status = DIFF_STATUS_UNMERGED;
2205                 else if (!DIFF_FILE_VALID(p->one))
2206                         p->status = DIFF_STATUS_ADDED;
2207                 else if (!DIFF_FILE_VALID(p->two))
2208                         p->status = DIFF_STATUS_DELETED;
2209                 else if (DIFF_PAIR_TYPE_CHANGED(p))
2210                         p->status = DIFF_STATUS_TYPE_CHANGED;
2211
2212                 /* from this point on, we are dealing with a pair
2213                  * whose both sides are valid and of the same type, i.e.
2214                  * either in-place edit or rename/copy edit.
2215                  */
2216                 else if (DIFF_PAIR_RENAME(p)) {
2217                         if (p->source_stays) {
2218                                 p->status = DIFF_STATUS_COPIED;
2219                                 continue;
2220                         }
2221                         /* See if there is some other filepair that
2222                          * copies from the same source as us.  If so
2223                          * we are a copy.  Otherwise we are either a
2224                          * copy if the path stays, or a rename if it
2225                          * does not, but we already handled "stays" case.
2226                          */
2227                         for (j = i + 1; j < q->nr; j++) {
2228                                 pp = q->queue[j];
2229                                 if (strcmp(pp->one->path, p->one->path))
2230                                         continue; /* not us */
2231                                 if (!DIFF_PAIR_RENAME(pp))
2232                                         continue; /* not a rename/copy */
2233                                 /* pp is a rename/copy from the same source */
2234                                 p->status = DIFF_STATUS_COPIED;
2235                                 break;
2236                         }
2237                         if (!p->status)
2238                                 p->status = DIFF_STATUS_RENAMED;
2239                 }
2240                 else if (memcmp(p->one->sha1, p->two->sha1, 20) ||
2241                          p->one->mode != p->two->mode)
2242                         p->status = DIFF_STATUS_MODIFIED;
2243                 else {
2244                         /* This is a "no-change" entry and should not
2245                          * happen anymore, but prepare for broken callers.
2246                          */
2247                         error("feeding unmodified %s to diffcore",
2248                               p->one->path);
2249                         p->status = DIFF_STATUS_UNKNOWN;
2250                 }
2251         }
2252         diff_debug_queue("resolve-rename-copy done", q);
2253 }
2254
2255 static int check_pair_status(struct diff_filepair *p)
2256 {
2257         switch (p->status) {
2258         case DIFF_STATUS_UNKNOWN:
2259                 return 0;
2260         case 0:
2261                 die("internal error in diff-resolve-rename-copy");
2262         default:
2263                 return 1;
2264         }
2265 }
2266
2267 static void flush_one_pair(struct diff_filepair *p, struct diff_options *opt)
2268 {
2269         int fmt = opt->output_format;
2270
2271         if (fmt & DIFF_FORMAT_CHECKDIFF)
2272                 diff_flush_checkdiff(p, opt);
2273         else if (fmt & (DIFF_FORMAT_RAW | DIFF_FORMAT_NAME_STATUS))
2274                 diff_flush_raw(p, opt);
2275         else if (fmt & DIFF_FORMAT_NAME)
2276                 diff_flush_name(p, opt->line_termination);
2277 }
2278
2279 static void show_file_mode_name(const char *newdelete, struct diff_filespec *fs)
2280 {
2281         if (fs->mode)
2282                 printf(" %s mode %06o %s\n", newdelete, fs->mode, fs->path);
2283         else
2284                 printf(" %s %s\n", newdelete, fs->path);
2285 }
2286
2287
2288 static void show_mode_change(struct diff_filepair *p, int show_name)
2289 {
2290         if (p->one->mode && p->two->mode && p->one->mode != p->two->mode) {
2291                 if (show_name)
2292                         printf(" mode change %06o => %06o %s\n",
2293                                p->one->mode, p->two->mode, p->two->path);
2294                 else
2295                         printf(" mode change %06o => %06o\n",
2296                                p->one->mode, p->two->mode);
2297         }
2298 }
2299
2300 static void show_rename_copy(const char *renamecopy, struct diff_filepair *p)
2301 {
2302         const char *old, *new;
2303
2304         /* Find common prefix */
2305         old = p->one->path;
2306         new = p->two->path;
2307         while (1) {
2308                 const char *slash_old, *slash_new;
2309                 slash_old = strchr(old, '/');
2310                 slash_new = strchr(new, '/');
2311                 if (!slash_old ||
2312                     !slash_new ||
2313                     slash_old - old != slash_new - new ||
2314                     memcmp(old, new, slash_new - new))
2315                         break;
2316                 old = slash_old + 1;
2317                 new = slash_new + 1;
2318         }
2319         /* p->one->path thru old is the common prefix, and old and new
2320          * through the end of names are renames
2321          */
2322         if (old != p->one->path)
2323                 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2324                        (int)(old - p->one->path), p->one->path,
2325                        old, new, (int)(0.5 + p->score * 100.0/MAX_SCORE));
2326         else
2327                 printf(" %s %s => %s (%d%%)\n", renamecopy,
2328                        p->one->path, p->two->path,
2329                        (int)(0.5 + p->score * 100.0/MAX_SCORE));
2330         show_mode_change(p, 0);
2331 }
2332
2333 static void diff_summary(struct diff_filepair *p)
2334 {
2335         switch(p->status) {
2336         case DIFF_STATUS_DELETED:
2337                 show_file_mode_name("delete", p->one);
2338                 break;
2339         case DIFF_STATUS_ADDED:
2340                 show_file_mode_name("create", p->two);
2341                 break;
2342         case DIFF_STATUS_COPIED:
2343                 show_rename_copy("copy", p);
2344                 break;
2345         case DIFF_STATUS_RENAMED:
2346                 show_rename_copy("rename", p);
2347                 break;
2348         default:
2349                 if (p->score) {
2350                         printf(" rewrite %s (%d%%)\n", p->two->path,
2351                                 (int)(0.5 + p->score * 100.0/MAX_SCORE));
2352                         show_mode_change(p, 0);
2353                 } else  show_mode_change(p, 1);
2354                 break;
2355         }
2356 }
2357
2358 struct patch_id_t {
2359         struct xdiff_emit_state xm;
2360         SHA_CTX *ctx;
2361         int patchlen;
2362 };
2363
2364 static int remove_space(char *line, int len)
2365 {
2366         int i;
2367         char *dst = line;
2368         unsigned char c;
2369
2370         for (i = 0; i < len; i++)
2371                 if (!isspace((c = line[i])))
2372                         *dst++ = c;
2373
2374         return dst - line;
2375 }
2376
2377 static void patch_id_consume(void *priv, char *line, unsigned long len)
2378 {
2379         struct patch_id_t *data = priv;
2380         int new_len;
2381
2382         /* Ignore line numbers when computing the SHA1 of the patch */
2383         if (!strncmp(line, "@@ -", 4))
2384                 return;
2385
2386         new_len = remove_space(line, len);
2387
2388         SHA1_Update(data->ctx, line, new_len);
2389         data->patchlen += new_len;
2390 }
2391
2392 /* returns 0 upon success, and writes result into sha1 */
2393 static int diff_get_patch_id(struct diff_options *options, unsigned char *sha1)
2394 {
2395         struct diff_queue_struct *q = &diff_queued_diff;
2396         int i;
2397         SHA_CTX ctx;
2398         struct patch_id_t data;
2399         char buffer[PATH_MAX * 4 + 20];
2400
2401         SHA1_Init(&ctx);
2402         memset(&data, 0, sizeof(struct patch_id_t));
2403         data.ctx = &ctx;
2404         data.xm.consume = patch_id_consume;
2405
2406         for (i = 0; i < q->nr; i++) {
2407                 xpparam_t xpp;
2408                 xdemitconf_t xecfg;
2409                 xdemitcb_t ecb;
2410                 mmfile_t mf1, mf2;
2411                 struct diff_filepair *p = q->queue[i];
2412                 int len1, len2;
2413
2414                 if (p->status == 0)
2415                         return error("internal diff status error");
2416                 if (p->status == DIFF_STATUS_UNKNOWN)
2417                         continue;
2418                 if (diff_unmodified_pair(p))
2419                         continue;
2420                 if ((DIFF_FILE_VALID(p->one) && S_ISDIR(p->one->mode)) ||
2421                     (DIFF_FILE_VALID(p->two) && S_ISDIR(p->two->mode)))
2422                         continue;
2423                 if (DIFF_PAIR_UNMERGED(p))
2424                         continue;
2425
2426                 diff_fill_sha1_info(p->one);
2427                 diff_fill_sha1_info(p->two);
2428                 if (fill_mmfile(&mf1, p->one) < 0 ||
2429                                 fill_mmfile(&mf2, p->two) < 0)
2430                         return error("unable to read files to diff");
2431
2432                 /* Maybe hash p->two? into the patch id? */
2433                 if (mmfile_is_binary(&mf2))
2434                         continue;
2435
2436                 len1 = remove_space(p->one->path, strlen(p->one->path));
2437                 len2 = remove_space(p->two->path, strlen(p->two->path));
2438                 if (p->one->mode == 0)
2439                         len1 = snprintf(buffer, sizeof(buffer),
2440                                         "diff--gita/%.*sb/%.*s"
2441                                         "newfilemode%06o"
2442                                         "---/dev/null"
2443                                         "+++b/%.*s",
2444                                         len1, p->one->path,
2445                                         len2, p->two->path,
2446                                         p->two->mode,
2447                                         len2, p->two->path);
2448                 else if (p->two->mode == 0)
2449                         len1 = snprintf(buffer, sizeof(buffer),
2450                                         "diff--gita/%.*sb/%.*s"
2451                                         "deletedfilemode%06o"
2452                                         "---a/%.*s"
2453                                         "+++/dev/null",
2454                                         len1, p->one->path,
2455                                         len2, p->two->path,
2456                                         p->one->mode,
2457                                         len1, p->one->path);
2458                 else
2459                         len1 = snprintf(buffer, sizeof(buffer),
2460                                         "diff--gita/%.*sb/%.*s"
2461                                         "---a/%.*s"
2462                                         "+++b/%.*s",
2463                                         len1, p->one->path,
2464                                         len2, p->two->path,
2465                                         len1, p->one->path,
2466                                         len2, p->two->path);
2467                 SHA1_Update(&ctx, buffer, len1);
2468
2469                 xpp.flags = XDF_NEED_MINIMAL;
2470                 xecfg.ctxlen = 3;
2471                 xecfg.flags = XDL_EMIT_FUNCNAMES;
2472                 ecb.outf = xdiff_outf;
2473                 ecb.priv = &data;
2474                 xdl_diff(&mf1, &mf2, &xpp, &xecfg, &ecb);
2475         }
2476
2477         SHA1_Final(sha1, &ctx);
2478         return 0;
2479 }
2480
2481 int diff_flush_patch_id(struct diff_options *options, unsigned char *sha1)
2482 {
2483         struct diff_queue_struct *q = &diff_queued_diff;
2484         int i;
2485         int result = diff_get_patch_id(options, sha1);
2486
2487         for (i = 0; i < q->nr; i++)
2488                 diff_free_filepair(q->queue[i]);
2489
2490         free(q->queue);
2491         q->queue = NULL;
2492         q->nr = q->alloc = 0;
2493
2494         return result;
2495 }
2496
2497 static int is_summary_empty(const struct diff_queue_struct *q)
2498 {
2499         int i;
2500
2501         for (i = 0; i < q->nr; i++) {
2502                 const struct diff_filepair *p = q->queue[i];
2503
2504                 switch (p->status) {
2505                 case DIFF_STATUS_DELETED:
2506                 case DIFF_STATUS_ADDED:
2507                 case DIFF_STATUS_COPIED:
2508                 case DIFF_STATUS_RENAMED:
2509                         return 0;
2510                 default:
2511                         if (p->score)
2512                                 return 0;
2513                         if (p->one->mode && p->two->mode &&
2514                             p->one->mode != p->two->mode)
2515                                 return 0;
2516                         break;
2517                 }
2518         }
2519         return 1;
2520 }
2521
2522 void diff_flush(struct diff_options *options)
2523 {
2524         struct diff_queue_struct *q = &diff_queued_diff;
2525         int i, output_format = options->output_format;
2526         int separator = 0;
2527
2528         /*
2529          * Order: raw, stat, summary, patch
2530          * or:    name/name-status/checkdiff (other bits clear)
2531          */
2532         if (!q->nr)
2533                 goto free_queue;
2534
2535         if (output_format & (DIFF_FORMAT_RAW |
2536                              DIFF_FORMAT_NAME |
2537                              DIFF_FORMAT_NAME_STATUS |
2538                              DIFF_FORMAT_CHECKDIFF)) {
2539                 for (i = 0; i < q->nr; i++) {
2540                         struct diff_filepair *p = q->queue[i];
2541                         if (check_pair_status(p))
2542                                 flush_one_pair(p, options);
2543                 }
2544                 separator++;
2545         }
2546
2547         if (output_format & DIFF_FORMAT_DIFFSTAT) {
2548                 struct diffstat_t diffstat;
2549
2550                 memset(&diffstat, 0, sizeof(struct diffstat_t));
2551                 diffstat.xm.consume = diffstat_consume;
2552                 for (i = 0; i < q->nr; i++) {
2553                         struct diff_filepair *p = q->queue[i];
2554                         if (check_pair_status(p))
2555                                 diff_flush_stat(p, options, &diffstat);
2556                 }
2557                 show_stats(&diffstat);
2558                 separator++;
2559         }
2560
2561         if (output_format & DIFF_FORMAT_SUMMARY && !is_summary_empty(q)) {
2562                 for (i = 0; i < q->nr; i++)
2563                         diff_summary(q->queue[i]);
2564                 separator++;
2565         }
2566
2567         if (output_format & DIFF_FORMAT_PATCH) {
2568                 if (separator) {
2569                         if (options->stat_sep) {
2570                                 /* attach patch instead of inline */
2571                                 fputs(options->stat_sep, stdout);
2572                         } else {
2573                                 putchar(options->line_termination);
2574                         }
2575                 }
2576
2577                 for (i = 0; i < q->nr; i++) {
2578                         struct diff_filepair *p = q->queue[i];
2579                         if (check_pair_status(p))
2580                                 diff_flush_patch(p, options);
2581                 }
2582         }
2583
2584         for (i = 0; i < q->nr; i++)
2585                 diff_free_filepair(q->queue[i]);
2586 free_queue:
2587         free(q->queue);
2588         q->queue = NULL;
2589         q->nr = q->alloc = 0;
2590 }
2591
2592 static void diffcore_apply_filter(const char *filter)
2593 {
2594         int i;
2595         struct diff_queue_struct *q = &diff_queued_diff;
2596         struct diff_queue_struct outq;
2597         outq.queue = NULL;
2598         outq.nr = outq.alloc = 0;
2599
2600         if (!filter)
2601                 return;
2602
2603         if (strchr(filter, DIFF_STATUS_FILTER_AON)) {
2604                 int found;
2605                 for (i = found = 0; !found && i < q->nr; i++) {
2606                         struct diff_filepair *p = q->queue[i];
2607                         if (((p->status == DIFF_STATUS_MODIFIED) &&
2608                              ((p->score &&
2609                                strchr(filter, DIFF_STATUS_FILTER_BROKEN)) ||
2610                               (!p->score &&
2611                                strchr(filter, DIFF_STATUS_MODIFIED)))) ||
2612                             ((p->status != DIFF_STATUS_MODIFIED) &&
2613                              strchr(filter, p->status)))
2614                                 found++;
2615                 }
2616                 if (found)
2617                         return;
2618
2619                 /* otherwise we will clear the whole queue
2620                  * by copying the empty outq at the end of this
2621                  * function, but first clear the current entries
2622                  * in the queue.
2623                  */
2624                 for (i = 0; i < q->nr; i++)
2625                         diff_free_filepair(q->queue[i]);
2626         }
2627         else {
2628                 /* Only the matching ones */
2629                 for (i = 0; i < q->nr; i++) {
2630                         struct diff_filepair *p = q->queue[i];
2631
2632                         if (((p->status == DIFF_STATUS_MODIFIED) &&
2633                              ((p->score &&
2634                                strchr(filter, DIFF_STATUS_FILTER_BROKEN)) ||
2635                               (!p->score &&
2636                                strchr(filter, DIFF_STATUS_MODIFIED)))) ||
2637                             ((p->status != DIFF_STATUS_MODIFIED) &&
2638                              strchr(filter, p->status)))
2639                                 diff_q(&outq, p);
2640                         else
2641                                 diff_free_filepair(p);
2642                 }
2643         }
2644         free(q->queue);
2645         *q = outq;
2646 }
2647
2648 void diffcore_std(struct diff_options *options)
2649 {
2650         if (options->break_opt != -1)
2651                 diffcore_break(options->break_opt);
2652         if (options->detect_rename)
2653                 diffcore_rename(options);
2654         if (options->break_opt != -1)
2655                 diffcore_merge_broken();
2656         if (options->pickaxe)
2657                 diffcore_pickaxe(options->pickaxe, options->pickaxe_opts);
2658         if (options->orderfile)
2659                 diffcore_order(options->orderfile);
2660         diff_resolve_rename_copy();
2661         diffcore_apply_filter(options->filter);
2662 }
2663
2664
2665 void diffcore_std_no_resolve(struct diff_options *options)
2666 {
2667         if (options->pickaxe)
2668                 diffcore_pickaxe(options->pickaxe, options->pickaxe_opts);
2669         if (options->orderfile)
2670                 diffcore_order(options->orderfile);
2671         diffcore_apply_filter(options->filter);
2672 }
2673
2674 void diff_addremove(struct diff_options *options,
2675                     int addremove, unsigned mode,
2676                     const unsigned char *sha1,
2677                     const char *base, const char *path)
2678 {
2679         char concatpath[PATH_MAX];
2680         struct diff_filespec *one, *two;
2681
2682         /* This may look odd, but it is a preparation for
2683          * feeding "there are unchanged files which should
2684          * not produce diffs, but when you are doing copy
2685          * detection you would need them, so here they are"
2686          * entries to the diff-core.  They will be prefixed
2687          * with something like '=' or '*' (I haven't decided
2688          * which but should not make any difference).
2689          * Feeding the same new and old to diff_change() 
2690          * also has the same effect.
2691          * Before the final output happens, they are pruned after
2692          * merged into rename/copy pairs as appropriate.
2693          */
2694         if (options->reverse_diff)
2695                 addremove = (addremove == '+' ? '-' :
2696                              addremove == '-' ? '+' : addremove);
2697
2698         if (!path) path = "";
2699         sprintf(concatpath, "%s%s", base, path);
2700         one = alloc_filespec(concatpath);
2701         two = alloc_filespec(concatpath);
2702
2703         if (addremove != '+')
2704                 fill_filespec(one, sha1, mode);
2705         if (addremove != '-')
2706                 fill_filespec(two, sha1, mode);
2707
2708         diff_queue(&diff_queued_diff, one, two);
2709 }
2710
2711 void diff_change(struct diff_options *options,
2712                  unsigned old_mode, unsigned new_mode,
2713                  const unsigned char *old_sha1,
2714                  const unsigned char *new_sha1,
2715                  const char *base, const char *path) 
2716 {
2717         char concatpath[PATH_MAX];
2718         struct diff_filespec *one, *two;
2719
2720         if (options->reverse_diff) {
2721                 unsigned tmp;
2722                 const unsigned char *tmp_c;
2723                 tmp = old_mode; old_mode = new_mode; new_mode = tmp;
2724                 tmp_c = old_sha1; old_sha1 = new_sha1; new_sha1 = tmp_c;
2725         }
2726         if (!path) path = "";
2727         sprintf(concatpath, "%s%s", base, path);
2728         one = alloc_filespec(concatpath);
2729         two = alloc_filespec(concatpath);
2730         fill_filespec(one, old_sha1, old_mode);
2731         fill_filespec(two, new_sha1, new_mode);
2732
2733         diff_queue(&diff_queued_diff, one, two);
2734 }
2735
2736 void diff_unmerge(struct diff_options *options,
2737                   const char *path)
2738 {
2739         struct diff_filespec *one, *two;
2740         one = alloc_filespec(path);
2741         two = alloc_filespec(path);
2742         diff_queue(&diff_queued_diff, one, two);
2743 }