e0ee2451533ae572f7f9f4ce80a39bc582cab5ae
1#include "cache.h"
2#include "attr.h"
3#include "run-command.h"
4#include "quote.h"
5/*
7 * convert.c - convert a file when checking it out and checking it in.
8 *
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.
13 */
14enum crlf_action {
16 CRLF_GUESS = -1,
17 CRLF_BINARY = 0,
18 CRLF_TEXT,
19 CRLF_INPUT,
20 CRLF_CRLF,
21 CRLF_AUTO
22};
23struct text_stat {
25 /* NUL, CR, LF and CRLF counts */
26 unsigned nul, cr, lf, crlf;
27 /* These are just approximations! */
29 unsigned printable, nonprintable;
30};
31static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
33{
34 unsigned long i;
35 memset(stats, 0, sizeof(*stats));
37 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;
45 }
46 if (c == '\n') {
47 stats->lf++;
48 continue;
49 }
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++;
64 }
65 }
66 else
67 stats->printable++;
68 }
69 /* 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--;
73}
74/*
76 * The same heuristics as diff.c::mmfile_is_binary()
77 */
78static int is_binary(unsigned long size, struct text_stat *stats)
79{
80 if (stats->nul)
82 return 1;
83 if ((stats->printable >> 7) < stats->nonprintable)
84 return 1;
85 /*
86 * Other heuristics? Average line length might be relevant,
87 * as might LF vs CR vs CRLF counts..
88 *
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.
93 */
94 return 0;
95}
96static enum eol output_eol(enum crlf_action crlf_action)
98{
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;
118 }
119 return core_eol;
120}
121static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
123 struct text_stat *stats, enum safe_crlf checksafe)
124{
125 if (!checksafe)
126 return;
127 if (output_eol(crlf_action) == EOL_LF) {
129 /*
130 * CRLFs would not be restored by checkout:
131 * check if we'd remove CRLFs
132 */
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);
138 }
139 } else if (output_eol(crlf_action) == EOL_CRLF) {
140 /*
141 * CRLFs would be added by checkout:
142 * check if we have "naked" LFs
143 */
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);
149 }
150 }
151}
152static int has_cr_in_index(const char *path)
154{
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;
161 len = strlen(path);
163 pos = index_name_pos(istate, path, len);
164 if (pos < 0) {
165 /*
166 * We might be in the middle of a merge, in which
167 * case we would read stage #2 (ours).
168 */
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;
176 }
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;
183 }
184 has_cr = memchr(data, '\r', sz) != NULL;
186 free(data);
187 return has_cr;
188}
189static 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)
193{
194 struct text_stat stats;
195 char *dst;
196 if (crlf_action == CRLF_BINARY ||
198 (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE) || !len)
199 return 0;
200 gather_stats(src, len, &stats);
202 if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
204 /*
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?
208 */
209 if (stats.cr != stats.crlf)
210 return 0;
211 /*
213 * And add some heuristics for binary vs text, of course...
214 */
215 if (is_binary(len, &stats))
216 return 0;
217 if (crlf_action == CRLF_GUESS) {
219 /*
220 * If the file in the index has any CR in it, do not convert.
221 * This is the new safer autocrlf handling.
222 */
223 if (has_cr_in_index(path))
224 return 0;
225 }
226 }
227 check_safe_crlf(path, crlf_action, &stats, checksafe);
229 /* Optimization: No CR? Nothing to convert, regardless. */
231 if (!stats.cr)
232 return 0;
233 /* 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) {
239 /*
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.
243 */
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);
255 }
256 strbuf_setlen(buf, dst - buf->buf);
257 return 1;
258}
259static int crlf_to_worktree(const char *path, const char *src, size_t len,
261 struct strbuf *buf, enum crlf_action crlf_action)
262{
263 char *to_free = NULL;
264 struct text_stat stats;
265 if (!len || output_eol(crlf_action) != EOL_CRLF)
267 return 0;
268 gather_stats(src, len, &stats);
270 /* No LF? Nothing to convert, regardless. */
272 if (!stats.lf)
273 return 0;
274 /* Was it already in CRLF format? */
276 if (stats.lf == stats.crlf)
277 return 0;
278 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;
285 }
286 /* If we have any bare CR characters, we're not going to touch it */
288 if (stats.cr != stats.crlf)
289 return 0;
290 if (is_binary(len, &stats))
292 return 0;
293 }
294 /* are we "faking" in place editing ? */
296 if (src == buf->buf)
297 to_free = strbuf_detach(buf, NULL);
298 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");
309 }
310 len -= nl + 1 - src;
311 src = nl + 1;
312 }
313 strbuf_add(buf, src, len);
314 free(to_free);
316 return 1;
317}
318struct filter_params {
320 const char *src;
321 unsigned long size;
322 const char *cmd;
323 const char *path;
324};
325static int filter_buffer(int in, int out, void *data)
327{
328 /*
329 * Spawn cmd and feed the buffer contents through its stdin.
330 */
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 };
335 /* 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, },
342 };
343 /* quote the path to preserve spaces, etc. */
345 sq_quote_buf(&path, params->path);
346 dict[0].value = path.buf;
347 /* expand all %f with the quoted path */
349 strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
350 strbuf_release(&path);
351 argv[0] = cmd.buf;
353 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;
359 if (start_command(&child_process))
361 return error("cannot fork to run external filter %s", params->cmd);
362 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);
368 status = finish_command(&child_process);
370 if (status)
371 error("external filter %s failed %d", params->cmd, status);
372 strbuf_release(&cmd);
374 return (write_err || status);
375}
376static int apply_filter(const char *path, const char *src, size_t len,
378 struct strbuf *dst, const char *cmd)
379{
380 /*
381 * Create a pipeline to have the command filter the buffer's
382 * contents.
383 *
384 * (child --> cmd) --> us
385 */
386 int ret = 1;
387 struct strbuf nbuf = STRBUF_INIT;
388 struct async async;
389 struct filter_params params;
390 if (!cmd)
392 return 0;
393 memset(&async, 0, sizeof(async));
395 async.proc = filter_buffer;
396 async.data = ¶ms;
397 async.out = -1;
398 params.src = src;
399 params.size = len;
400 params.cmd = cmd;
401 params.path = path;
402 fflush(NULL);
404 if (start_async(&async))
405 return 0; /* error was already reported */
406 if (strbuf_read(&nbuf, async.out, len) < 0) {
408 error("read from external filter %s failed", cmd);
409 ret = 0;
410 }
411 if (close(async.out)) {
412 error("read from external filter %s failed", cmd);
413 ret = 0;
414 }
415 if (finish_async(&async)) {
416 error("external filter %s failed", cmd);
417 ret = 0;
418 }
419 if (ret) {
421 strbuf_swap(dst, &nbuf);
422 }
423 strbuf_release(&nbuf);
424 return ret;
425}
426static 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;
433static int read_convert_config(const char *var, const char *value, void *cb)
435{
436 const char *ep, *name;
437 int namelen;
438 struct convert_driver *drv;
439 /*
441 * External conversion drivers are configured using
442 * "filter.<name>.variable".
443 */
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);
456 }
457 ep++;
459 /*
461 * filter.<name>.smudge and filter.<name>.clean specifies
462 * the command line:
463 *
464 * command-line
465 *
466 * The command-line will not be interpolated in any way.
467 */
468 if (!strcmp("smudge", ep))
470 return git_config_string(&drv->smudge, var, value);
471 if (!strcmp("clean", ep))
473 return git_config_string(&drv->clean, var, value);
474 return 0;
476}
477static void setup_convert_check(struct git_attr_check *check)
479{
480 static struct git_attr *attr_text;
481 static struct git_attr *attr_crlf;
482 static struct git_attr *attr_eol;
483 static struct git_attr *attr_ident;
484 static struct git_attr *attr_filter;
485 if (!attr_text) {
487 attr_text = git_attr("text");
488 attr_crlf = git_attr("crlf");
489 attr_eol = git_attr("eol");
490 attr_ident = git_attr("ident");
491 attr_filter = git_attr("filter");
492 user_convert_tail = &user_convert;
493 git_config(read_convert_config, NULL);
494 }
495 check[0].attr = attr_crlf;
496 check[1].attr = attr_ident;
497 check[2].attr = attr_filter;
498 check[3].attr = attr_eol;
499 check[4].attr = attr_text;
500}
501static int count_ident(const char *cp, unsigned long size)
503{
504 /*
505 * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
506 */
507 int cnt = 0;
508 char ch;
509 while (size) {
511 ch = *cp++;
512 size--;
513 if (ch != '$')
514 continue;
515 if (size < 3)
516 break;
517 if (memcmp("Id", cp, 2))
518 continue;
519 ch = cp[2];
520 cp += 3;
521 size -= 3;
522 if (ch == '$')
523 cnt++; /* $Id$ */
524 if (ch != ':')
525 continue;
526 /*
528 * "$Id: ... "; scan up to the closing dollar sign and discard.
529 */
530 while (size) {
531 ch = *cp++;
532 size--;
533 if (ch == '$') {
534 cnt++;
535 break;
536 }
537 if (ch == '\n')
538 break;
539 }
540 }
541 return cnt;
542}
543static int ident_to_git(const char *path, const char *src, size_t len,
545 struct strbuf *buf, int ident)
546{
547 char *dst, *dollar;
548 if (!ident || !count_ident(src, len))
550 return 0;
551 /* only grow if not in place */
553 if (strbuf_avail(buf) + buf->len < len)
554 strbuf_grow(buf, len - buf->len);
555 dst = buf->buf;
556 for (;;) {
557 dollar = memchr(src, '$', len);
558 if (!dollar)
559 break;
560 memcpy(dst, src, dollar + 1 - src);
561 dst += dollar + 1 - src;
562 len -= dollar + 1 - src;
563 src = dollar + 1;
564 if (len > 3 && !memcmp(src, "Id:", 3)) {
566 dollar = memchr(src + 3, '$', len - 3);
567 if (!dollar)
568 break;
569 if (memchr(src + 3, '\n', dollar - src - 3)) {
570 /* Line break before the next dollar. */
571 continue;
572 }
573 memcpy(dst, "Id$", 3);
575 dst += 3;
576 len -= dollar + 1 - src;
577 src = dollar + 1;
578 }
579 }
580 memcpy(dst, src, len);
581 strbuf_setlen(buf, dst + len - buf->buf);
582 return 1;
583}
584static int ident_to_worktree(const char *path, const char *src, size_t len,
586 struct strbuf *buf, int ident)
587{
588 unsigned char sha1[20];
589 char *to_free = NULL, *dollar, *spc;
590 int cnt;
591 if (!ident)
593 return 0;
594 cnt = count_ident(src, len);
596 if (!cnt)
597 return 0;
598 /* are we "faking" in place editing ? */
600 if (src == buf->buf)
601 to_free = strbuf_detach(buf, NULL);
602 hash_sha1_file(src, len, "blob", sha1);
603 strbuf_grow(buf, len + cnt * 43);
605 for (;;) {
606 /* step 1: run to the next '$' */
607 dollar = memchr(src, '$', len);
608 if (!dollar)
609 break;
610 strbuf_add(buf, src, dollar + 1 - src);
611 len -= dollar + 1 - src;
612 src = dollar + 1;
613 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
615 if (len < 3 || memcmp("Id", src, 2))
616 continue;
617 /* step 3: skip over Id$ or Id:xxxxx$ */
619 if (src[2] == '$') {
620 src += 3;
621 len -= 3;
622 } else if (src[2] == ':') {
623 /*
624 * It's possible that an expanded Id has crept its way into the
625 * repository, we cope with that by stripping the expansion out.
626 * This is probably not a good idea, since it will cause changes
627 * on checkout, which won't go away by stash, but let's keep it
628 * for git-style ids.
629 */
630 dollar = memchr(src + 3, '$', len - 3);
631 if (!dollar) {
632 /* incomplete keyword, no more '$', so just quit the loop */
633 break;
634 }
635 if (memchr(src + 3, '\n', dollar - src - 3)) {
637 /* Line break before the next dollar. */
638 continue;
639 }
640 spc = memchr(src + 4, ' ', dollar - src - 4);
642 if (spc && spc < dollar-1) {
643 /* There are spaces in unexpected places.
644 * This is probably an id from some other
645 * versioning system. Keep it for now.
646 */
647 continue;
648 }
649 len -= dollar + 1 - src;
651 src = dollar + 1;
652 } else {
653 /* it wasn't a "Id$" or "Id:xxxx$" */
654 continue;
655 }
656 /* step 4: substitute */
658 strbuf_addstr(buf, "Id: ");
659 strbuf_add(buf, sha1_to_hex(sha1), 40);
660 strbuf_addstr(buf, " $");
661 }
662 strbuf_add(buf, src, len);
663 free(to_free);
665 return 1;
666}
667static int git_path_check_crlf(const char *path, struct git_attr_check *check)
669{
670 const char *value = check->value;
671 if (ATTR_TRUE(value))
673 return CRLF_TEXT;
674 else if (ATTR_FALSE(value))
675 return CRLF_BINARY;
676 else if (ATTR_UNSET(value))
677 ;
678 else if (!strcmp(value, "input"))
679 return CRLF_INPUT;
680 else if (!strcmp(value, "auto"))
681 return CRLF_AUTO;
682 return CRLF_GUESS;
683}
684static int git_path_check_eol(const char *path, struct git_attr_check *check)
686{
687 const char *value = check->value;
688 if (ATTR_UNSET(value))
690 ;
691 else if (!strcmp(value, "lf"))
692 return EOL_LF;
693 else if (!strcmp(value, "crlf"))
694 return EOL_CRLF;
695 return EOL_UNSET;
696}
697static struct convert_driver *git_path_check_convert(const char *path,
699 struct git_attr_check *check)
700{
701 const char *value = check->value;
702 struct convert_driver *drv;
703 if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
705 return NULL;
706 for (drv = user_convert; drv; drv = drv->next)
707 if (!strcmp(value, drv->name))
708 return drv;
709 return NULL;
710}
711static int git_path_check_ident(const char *path, struct git_attr_check *check)
713{
714 const char *value = check->value;
715 return !!ATTR_TRUE(value);
717}
718static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
720{
721 if (text_attr == CRLF_BINARY)
722 return CRLF_BINARY;
723 if (eol_attr == EOL_LF)
724 return CRLF_INPUT;
725 if (eol_attr == EOL_CRLF)
726 return CRLF_CRLF;
727 return text_attr;
728}
729int convert_to_git(const char *path, const char *src, size_t len,
731 struct strbuf *dst, enum safe_crlf checksafe)
732{
733 struct git_attr_check check[5];
734 enum crlf_action crlf_action = CRLF_GUESS;
735 enum eol eol_attr = EOL_UNSET;
736 int ident = 0, ret = 0;
737 const char *filter = NULL;
738 setup_convert_check(check);
740 if (!git_checkattr(path, ARRAY_SIZE(check), check)) {
741 struct convert_driver *drv;
742 crlf_action = git_path_check_crlf(path, check + 4);
743 if (crlf_action == CRLF_GUESS)
744 crlf_action = git_path_check_crlf(path, check + 0);
745 ident = git_path_check_ident(path, check + 1);
746 drv = git_path_check_convert(path, check + 2);
747 eol_attr = git_path_check_eol(path, check + 3);
748 if (drv && drv->clean)
749 filter = drv->clean;
750 }
751 ret |= apply_filter(path, src, len, dst, filter);
753 if (ret) {
754 src = dst->buf;
755 len = dst->len;
756 }
757 crlf_action = input_crlf_action(crlf_action, eol_attr);
758 ret |= crlf_to_git(path, src, len, dst, crlf_action, checksafe);
759 if (ret) {
760 src = dst->buf;
761 len = dst->len;
762 }
763 return ret | ident_to_git(path, src, len, dst, ident);
764}
765static int convert_to_working_tree_internal(const char *path, const char *src,
767 size_t len, struct strbuf *dst,
768 int normalizing)
769{
770 struct git_attr_check check[5];
771 enum crlf_action crlf_action = CRLF_GUESS;
772 enum eol eol_attr = EOL_UNSET;
773 int ident = 0, ret = 0;
774 const char *filter = NULL;
775 setup_convert_check(check);
777 if (!git_checkattr(path, ARRAY_SIZE(check), check)) {
778 struct convert_driver *drv;
779 crlf_action = git_path_check_crlf(path, check + 4);
780 if (crlf_action == CRLF_GUESS)
781 crlf_action = git_path_check_crlf(path, check + 0);
782 ident = git_path_check_ident(path, check + 1);
783 drv = git_path_check_convert(path, check + 2);
784 eol_attr = git_path_check_eol(path, check + 3);
785 if (drv && drv->smudge)
786 filter = drv->smudge;
787 }
788 ret |= ident_to_worktree(path, src, len, dst, ident);
790 if (ret) {
791 src = dst->buf;
792 len = dst->len;
793 }
794 /*
795 * CRLF conversion can be skipped if normalizing, unless there
796 * is a smudge filter. The filter might expect CRLFs.
797 */
798 if (filter || !normalizing) {
799 crlf_action = input_crlf_action(crlf_action, eol_attr);
800 ret |= crlf_to_worktree(path, src, len, dst, crlf_action);
801 if (ret) {
802 src = dst->buf;
803 len = dst->len;
804 }
805 }
806 return ret | apply_filter(path, src, len, dst, filter);
807}
808int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
810{
811 return convert_to_working_tree_internal(path, src, len, dst, 0);
812}
813int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
815{
816 int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
817 if (ret) {
818 src = dst->buf;
819 len = dst->len;
820 }
821 return ret | convert_to_git(path, src, len, dst, 0);
822}