combine-diff: support format_callback
[git/gitweb.git] / convert.c
blobefc7e07d475c66f7835dc6cbbd3bc358f01c41c3
1 #include "cache.h"
2 #include "attr.h"
3 #include "run-command.h"
4 #include "quote.h"
6 /*
7 * convert.c - convert a file when checking it out and checking it in.
9 * This should use the pathname to decide on whether it wants to do some
10 * more interesting conversions (automatic gzip/unzip, general format
11 * conversions etc etc), but by default it just does automatic CRLF<->LF
12 * translation when the "text" attribute or "auto_crlf" option is set.
15 enum crlf_action {
16 CRLF_GUESS = -1,
17 CRLF_BINARY = 0,
18 CRLF_TEXT,
19 CRLF_INPUT,
20 CRLF_CRLF,
21 CRLF_AUTO
24 struct text_stat {
25 /* NUL, CR, LF and CRLF counts */
26 unsigned nul, cr, lf, crlf;
28 /* These are just approximations! */
29 unsigned printable, nonprintable;
32 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
34 unsigned long i;
36 memset(stats, 0, sizeof(*stats));
38 for (i = 0; i < size; i++) {
39 unsigned char c = buf[i];
40 if (c == '\r') {
41 stats->cr++;
42 if (i+1 < size && buf[i+1] == '\n')
43 stats->crlf++;
44 continue;
46 if (c == '\n') {
47 stats->lf++;
48 continue;
50 if (c == 127)
51 /* DEL */
52 stats->nonprintable++;
53 else if (c < 32) {
54 switch (c) {
55 /* BS, HT, ESC and FF */
56 case '\b': case '\t': case '\033': case '\014':
57 stats->printable++;
58 break;
59 case 0:
60 stats->nul++;
61 /* fall through */
62 default:
63 stats->nonprintable++;
66 else
67 stats->printable++;
70 /* If file ends with EOF then don't count this EOF as non-printable. */
71 if (size >= 1 && buf[size-1] == '\032')
72 stats->nonprintable--;
76 * The same heuristics as diff.c::mmfile_is_binary()
78 static int is_binary(unsigned long size, struct text_stat *stats)
81 if (stats->nul)
82 return 1;
83 if ((stats->printable >> 7) < stats->nonprintable)
84 return 1;
86 * Other heuristics? Average line length might be relevant,
87 * as might LF vs CR vs CRLF counts..
89 * NOTE! It might be normal to have a low ratio of CRLF to LF
90 * (somebody starts with a LF-only file and edits it with an editor
91 * that adds CRLF only to lines that are added..). But do we
92 * want to support CR-only? Probably not.
94 return 0;
97 static enum eol output_eol(enum crlf_action crlf_action)
99 switch (crlf_action) {
100 case CRLF_BINARY:
101 return EOL_UNSET;
102 case CRLF_CRLF:
103 return EOL_CRLF;
104 case CRLF_INPUT:
105 return EOL_LF;
106 case CRLF_GUESS:
107 if (!auto_crlf)
108 return EOL_UNSET;
109 /* fall through */
110 case CRLF_TEXT:
111 case CRLF_AUTO:
112 if (auto_crlf == AUTO_CRLF_TRUE)
113 return EOL_CRLF;
114 else if (auto_crlf == AUTO_CRLF_INPUT)
115 return EOL_LF;
116 else if (core_eol == EOL_UNSET)
117 return EOL_NATIVE;
119 return core_eol;
122 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
123 struct text_stat *stats, enum safe_crlf checksafe)
125 if (!checksafe)
126 return;
128 if (output_eol(crlf_action) == EOL_LF) {
130 * CRLFs would not be restored by checkout:
131 * check if we'd remove CRLFs
133 if (stats->crlf) {
134 if (checksafe == SAFE_CRLF_WARN)
135 warning("CRLF will be replaced by LF in %s.\nThe file will have its original line endings in your working directory.", path);
136 else /* i.e. SAFE_CRLF_FAIL */
137 die("CRLF would be replaced by LF in %s.", path);
139 } else if (output_eol(crlf_action) == EOL_CRLF) {
141 * CRLFs would be added by checkout:
142 * check if we have "naked" LFs
144 if (stats->lf != stats->crlf) {
145 if (checksafe == SAFE_CRLF_WARN)
146 warning("LF will be replaced by CRLF in %s.\nThe file will have its original line endings in your working directory.", path);
147 else /* i.e. SAFE_CRLF_FAIL */
148 die("LF would be replaced by CRLF in %s", path);
153 static int has_cr_in_index(const char *path)
155 int pos, len;
156 unsigned long sz;
157 enum object_type type;
158 void *data;
159 int has_cr;
160 struct index_state *istate = &the_index;
162 len = strlen(path);
163 pos = index_name_pos(istate, path, len);
164 if (pos < 0) {
166 * We might be in the middle of a merge, in which
167 * case we would read stage #2 (ours).
169 int i;
170 for (i = -pos - 1;
171 (pos < 0 && i < istate->cache_nr &&
172 !strcmp(istate->cache[i]->name, path));
173 i++)
174 if (ce_stage(istate->cache[i]) == 2)
175 pos = i;
177 if (pos < 0)
178 return 0;
179 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
180 if (!data || type != OBJ_BLOB) {
181 free(data);
182 return 0;
185 has_cr = memchr(data, '\r', sz) != NULL;
186 free(data);
187 return has_cr;
190 static int crlf_to_git(const char *path, const char *src, size_t len,
191 struct strbuf *buf,
192 enum crlf_action crlf_action, enum safe_crlf checksafe)
194 struct text_stat stats;
195 char *dst;
197 if (crlf_action == CRLF_BINARY ||
198 (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE) || !len)
199 return 0;
201 gather_stats(src, len, &stats);
203 if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
205 * We're currently not going to even try to convert stuff
206 * that has bare CR characters. Does anybody do that crazy
207 * stuff?
209 if (stats.cr != stats.crlf)
210 return 0;
213 * And add some heuristics for binary vs text, of course...
215 if (is_binary(len, &stats))
216 return 0;
218 if (crlf_action == CRLF_GUESS) {
220 * If the file in the index has any CR in it, do not convert.
221 * This is the new safer autocrlf handling.
223 if (has_cr_in_index(path))
224 return 0;
228 check_safe_crlf(path, crlf_action, &stats, checksafe);
230 /* Optimization: No CR? Nothing to convert, regardless. */
231 if (!stats.cr)
232 return 0;
234 /* only grow if not in place */
235 if (strbuf_avail(buf) + buf->len < len)
236 strbuf_grow(buf, len - buf->len);
237 dst = buf->buf;
238 if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
240 * If we guessed, we already know we rejected a file with
241 * lone CR, and we can strip a CR without looking at what
242 * follow it.
244 do {
245 unsigned char c = *src++;
246 if (c != '\r')
247 *dst++ = c;
248 } while (--len);
249 } else {
250 do {
251 unsigned char c = *src++;
252 if (! (c == '\r' && (1 < len && *src == '\n')))
253 *dst++ = c;
254 } while (--len);
256 strbuf_setlen(buf, dst - buf->buf);
257 return 1;
260 static int crlf_to_worktree(const char *path, const char *src, size_t len,
261 struct strbuf *buf, enum crlf_action crlf_action)
263 char *to_free = NULL;
264 struct text_stat stats;
266 if (!len || output_eol(crlf_action) != EOL_CRLF)
267 return 0;
269 gather_stats(src, len, &stats);
271 /* No LF? Nothing to convert, regardless. */
272 if (!stats.lf)
273 return 0;
275 /* Was it already in CRLF format? */
276 if (stats.lf == stats.crlf)
277 return 0;
279 if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
280 if (crlf_action == CRLF_GUESS) {
281 /* If we have any CR or CRLF line endings, we do not touch it */
282 /* This is the new safer autocrlf-handling */
283 if (stats.cr > 0 || stats.crlf > 0)
284 return 0;
287 /* If we have any bare CR characters, we're not going to touch it */
288 if (stats.cr != stats.crlf)
289 return 0;
291 if (is_binary(len, &stats))
292 return 0;
295 /* are we "faking" in place editing ? */
296 if (src == buf->buf)
297 to_free = strbuf_detach(buf, NULL);
299 strbuf_grow(buf, len + stats.lf - stats.crlf);
300 for (;;) {
301 const char *nl = memchr(src, '\n', len);
302 if (!nl)
303 break;
304 if (nl > src && nl[-1] == '\r') {
305 strbuf_add(buf, src, nl + 1 - src);
306 } else {
307 strbuf_add(buf, src, nl - src);
308 strbuf_addstr(buf, "\r\n");
310 len -= nl + 1 - src;
311 src = nl + 1;
313 strbuf_add(buf, src, len);
315 free(to_free);
316 return 1;
319 struct filter_params {
320 const char *src;
321 unsigned long size;
322 const char *cmd;
323 const char *path;
326 static int filter_buffer(int in, int out, void *data)
329 * Spawn cmd and feed the buffer contents through its stdin.
331 struct child_process child_process;
332 struct filter_params *params = (struct filter_params *)data;
333 int write_err, status;
334 const char *argv[] = { NULL, NULL };
336 /* apply % substitution to cmd */
337 struct strbuf cmd = STRBUF_INIT;
338 struct strbuf path = STRBUF_INIT;
339 struct strbuf_expand_dict_entry dict[] = {
340 { "f", NULL, },
341 { NULL, NULL, },
344 /* quote the path to preserve spaces, etc. */
345 sq_quote_buf(&path, params->path);
346 dict[0].value = path.buf;
348 /* expand all %f with the quoted path */
349 strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
350 strbuf_release(&path);
352 argv[0] = cmd.buf;
354 memset(&child_process, 0, sizeof(child_process));
355 child_process.argv = argv;
356 child_process.use_shell = 1;
357 child_process.in = -1;
358 child_process.out = out;
360 if (start_command(&child_process))
361 return error("cannot fork to run external filter %s", params->cmd);
363 write_err = (write_in_full(child_process.in, params->src, params->size) < 0);
364 if (close(child_process.in))
365 write_err = 1;
366 if (write_err)
367 error("cannot feed the input to external filter %s", params->cmd);
369 status = finish_command(&child_process);
370 if (status)
371 error("external filter %s failed %d", params->cmd, status);
373 strbuf_release(&cmd);
374 return (write_err || status);
377 static int apply_filter(const char *path, const char *src, size_t len,
378 struct strbuf *dst, const char *cmd)
381 * Create a pipeline to have the command filter the buffer's
382 * contents.
384 * (child --> cmd) --> us
386 int ret = 1;
387 struct strbuf nbuf = STRBUF_INIT;
388 struct async async;
389 struct filter_params params;
391 if (!cmd)
392 return 0;
394 memset(&async, 0, sizeof(async));
395 async.proc = filter_buffer;
396 async.data = &params;
397 async.out = -1;
398 params.src = src;
399 params.size = len;
400 params.cmd = cmd;
401 params.path = path;
403 fflush(NULL);
404 if (start_async(&async))
405 return 0; /* error was already reported */
407 if (strbuf_read(&nbuf, async.out, len) < 0) {
408 error("read from external filter %s failed", cmd);
409 ret = 0;
411 if (close(async.out)) {
412 error("read from external filter %s failed", cmd);
413 ret = 0;
415 if (finish_async(&async)) {
416 error("external filter %s failed", cmd);
417 ret = 0;
420 if (ret) {
421 strbuf_swap(dst, &nbuf);
423 strbuf_release(&nbuf);
424 return ret;
427 static struct convert_driver {
428 const char *name;
429 struct convert_driver *next;
430 const char *smudge;
431 const char *clean;
432 } *user_convert, **user_convert_tail;
434 static int read_convert_config(const char *var, const char *value, void *cb)
436 const char *ep, *name;
437 int namelen;
438 struct convert_driver *drv;
441 * External conversion drivers are configured using
442 * "filter.<name>.variable".
444 if (prefixcmp(var, "filter.") || (ep = strrchr(var, '.')) == var + 6)
445 return 0;
446 name = var + 7;
447 namelen = ep - name;
448 for (drv = user_convert; drv; drv = drv->next)
449 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
450 break;
451 if (!drv) {
452 drv = xcalloc(1, sizeof(struct convert_driver));
453 drv->name = xmemdupz(name, namelen);
454 *user_convert_tail = drv;
455 user_convert_tail = &(drv->next);
458 ep++;
461 * filter.<name>.smudge and filter.<name>.clean specifies
462 * the command line:
464 * command-line
466 * The command-line will not be interpolated in any way.
469 if (!strcmp("smudge", ep))
470 return git_config_string(&drv->smudge, var, value);
472 if (!strcmp("clean", ep))
473 return git_config_string(&drv->clean, var, value);
475 return 0;
478 static int count_ident(const char *cp, unsigned long size)
481 * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
483 int cnt = 0;
484 char ch;
486 while (size) {
487 ch = *cp++;
488 size--;
489 if (ch != '$')
490 continue;
491 if (size < 3)
492 break;
493 if (memcmp("Id", cp, 2))
494 continue;
495 ch = cp[2];
496 cp += 3;
497 size -= 3;
498 if (ch == '$')
499 cnt++; /* $Id$ */
500 if (ch != ':')
501 continue;
504 * "$Id: ... "; scan up to the closing dollar sign and discard.
506 while (size) {
507 ch = *cp++;
508 size--;
509 if (ch == '$') {
510 cnt++;
511 break;
513 if (ch == '\n')
514 break;
517 return cnt;
520 static int ident_to_git(const char *path, const char *src, size_t len,
521 struct strbuf *buf, int ident)
523 char *dst, *dollar;
525 if (!ident || !count_ident(src, len))
526 return 0;
528 /* only grow if not in place */
529 if (strbuf_avail(buf) + buf->len < len)
530 strbuf_grow(buf, len - buf->len);
531 dst = buf->buf;
532 for (;;) {
533 dollar = memchr(src, '$', len);
534 if (!dollar)
535 break;
536 memcpy(dst, src, dollar + 1 - src);
537 dst += dollar + 1 - src;
538 len -= dollar + 1 - src;
539 src = dollar + 1;
541 if (len > 3 && !memcmp(src, "Id:", 3)) {
542 dollar = memchr(src + 3, '$', len - 3);
543 if (!dollar)
544 break;
545 if (memchr(src + 3, '\n', dollar - src - 3)) {
546 /* Line break before the next dollar. */
547 continue;
550 memcpy(dst, "Id$", 3);
551 dst += 3;
552 len -= dollar + 1 - src;
553 src = dollar + 1;
556 memcpy(dst, src, len);
557 strbuf_setlen(buf, dst + len - buf->buf);
558 return 1;
561 static int ident_to_worktree(const char *path, const char *src, size_t len,
562 struct strbuf *buf, int ident)
564 unsigned char sha1[20];
565 char *to_free = NULL, *dollar, *spc;
566 int cnt;
568 if (!ident)
569 return 0;
571 cnt = count_ident(src, len);
572 if (!cnt)
573 return 0;
575 /* are we "faking" in place editing ? */
576 if (src == buf->buf)
577 to_free = strbuf_detach(buf, NULL);
578 hash_sha1_file(src, len, "blob", sha1);
580 strbuf_grow(buf, len + cnt * 43);
581 for (;;) {
582 /* step 1: run to the next '$' */
583 dollar = memchr(src, '$', len);
584 if (!dollar)
585 break;
586 strbuf_add(buf, src, dollar + 1 - src);
587 len -= dollar + 1 - src;
588 src = dollar + 1;
590 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
591 if (len < 3 || memcmp("Id", src, 2))
592 continue;
594 /* step 3: skip over Id$ or Id:xxxxx$ */
595 if (src[2] == '$') {
596 src += 3;
597 len -= 3;
598 } else if (src[2] == ':') {
600 * It's possible that an expanded Id has crept its way into the
601 * repository, we cope with that by stripping the expansion out.
602 * This is probably not a good idea, since it will cause changes
603 * on checkout, which won't go away by stash, but let's keep it
604 * for git-style ids.
606 dollar = memchr(src + 3, '$', len - 3);
607 if (!dollar) {
608 /* incomplete keyword, no more '$', so just quit the loop */
609 break;
612 if (memchr(src + 3, '\n', dollar - src - 3)) {
613 /* Line break before the next dollar. */
614 continue;
617 spc = memchr(src + 4, ' ', dollar - src - 4);
618 if (spc && spc < dollar-1) {
619 /* There are spaces in unexpected places.
620 * This is probably an id from some other
621 * versioning system. Keep it for now.
623 continue;
626 len -= dollar + 1 - src;
627 src = dollar + 1;
628 } else {
629 /* it wasn't a "Id$" or "Id:xxxx$" */
630 continue;
633 /* step 4: substitute */
634 strbuf_addstr(buf, "Id: ");
635 strbuf_add(buf, sha1_to_hex(sha1), 40);
636 strbuf_addstr(buf, " $");
638 strbuf_add(buf, src, len);
640 free(to_free);
641 return 1;
644 static int git_path_check_crlf(const char *path, struct git_attr_check *check)
646 const char *value = check->value;
648 if (ATTR_TRUE(value))
649 return CRLF_TEXT;
650 else if (ATTR_FALSE(value))
651 return CRLF_BINARY;
652 else if (ATTR_UNSET(value))
654 else if (!strcmp(value, "input"))
655 return CRLF_INPUT;
656 else if (!strcmp(value, "auto"))
657 return CRLF_AUTO;
658 return CRLF_GUESS;
661 static int git_path_check_eol(const char *path, struct git_attr_check *check)
663 const char *value = check->value;
665 if (ATTR_UNSET(value))
667 else if (!strcmp(value, "lf"))
668 return EOL_LF;
669 else if (!strcmp(value, "crlf"))
670 return EOL_CRLF;
671 return EOL_UNSET;
674 static struct convert_driver *git_path_check_convert(const char *path,
675 struct git_attr_check *check)
677 const char *value = check->value;
678 struct convert_driver *drv;
680 if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
681 return NULL;
682 for (drv = user_convert; drv; drv = drv->next)
683 if (!strcmp(value, drv->name))
684 return drv;
685 return NULL;
688 static int git_path_check_ident(const char *path, struct git_attr_check *check)
690 const char *value = check->value;
692 return !!ATTR_TRUE(value);
695 static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
697 if (text_attr == CRLF_BINARY)
698 return CRLF_BINARY;
699 if (eol_attr == EOL_LF)
700 return CRLF_INPUT;
701 if (eol_attr == EOL_CRLF)
702 return CRLF_CRLF;
703 return text_attr;
706 struct conv_attrs {
707 struct convert_driver *drv;
708 enum crlf_action crlf_action;
709 enum eol eol_attr;
710 int ident;
713 static const char *conv_attr_name[] = {
714 "crlf", "ident", "filter", "eol", "text",
716 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
718 static void convert_attrs(struct conv_attrs *ca, const char *path)
720 int i;
721 static struct git_attr_check ccheck[NUM_CONV_ATTRS];
723 if (!ccheck[0].attr) {
724 for (i = 0; i < NUM_CONV_ATTRS; i++)
725 ccheck[i].attr = git_attr(conv_attr_name[i]);
726 user_convert_tail = &user_convert;
727 git_config(read_convert_config, NULL);
730 if (!git_checkattr(path, NUM_CONV_ATTRS, ccheck)) {
731 ca->crlf_action = git_path_check_crlf(path, ccheck + 4);
732 if (ca->crlf_action == CRLF_GUESS)
733 ca->crlf_action = git_path_check_crlf(path, ccheck + 0);
734 ca->ident = git_path_check_ident(path, ccheck + 1);
735 ca->drv = git_path_check_convert(path, ccheck + 2);
736 ca->eol_attr = git_path_check_eol(path, ccheck + 3);
737 } else {
738 ca->drv = NULL;
739 ca->crlf_action = CRLF_GUESS;
740 ca->eol_attr = EOL_UNSET;
741 ca->ident = 0;
745 int convert_to_git(const char *path, const char *src, size_t len,
746 struct strbuf *dst, enum safe_crlf checksafe)
748 int ret = 0;
749 const char *filter = NULL;
750 struct conv_attrs ca;
752 convert_attrs(&ca, path);
753 if (ca.drv)
754 filter = ca.drv->clean;
756 ret |= apply_filter(path, src, len, dst, filter);
757 if (ret) {
758 src = dst->buf;
759 len = dst->len;
761 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
762 ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
763 if (ret) {
764 src = dst->buf;
765 len = dst->len;
767 return ret | ident_to_git(path, src, len, dst, ca.ident);
770 static int convert_to_working_tree_internal(const char *path, const char *src,
771 size_t len, struct strbuf *dst,
772 int normalizing)
774 int ret = 0;
775 const char *filter = NULL;
776 struct conv_attrs ca;
778 convert_attrs(&ca, path);
779 if (ca.drv)
780 filter = ca.drv->smudge;
782 ret |= ident_to_worktree(path, src, len, dst, ca.ident);
783 if (ret) {
784 src = dst->buf;
785 len = dst->len;
788 * CRLF conversion can be skipped if normalizing, unless there
789 * is a smudge filter. The filter might expect CRLFs.
791 if (filter || !normalizing) {
792 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
793 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
794 if (ret) {
795 src = dst->buf;
796 len = dst->len;
799 return ret | apply_filter(path, src, len, dst, filter);
802 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
804 return convert_to_working_tree_internal(path, src, len, dst, 0);
807 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
809 int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
810 if (ret) {
811 src = dst->buf;
812 len = dst->len;
814 return ret | convert_to_git(path, src, len, dst, 0);