Interested in helping open source friends on HP-UX?
[git/mjg.git] / convert.c
blob9a5612e93da2058f64bcec20cfd474a8b6b2d223
1 #include "cache.h"
2 #include "attr.h"
3 #include "run-command.h"
4 #include "quote.h"
5 #include "sigchain.h"
7 /*
8 * convert.c - convert a file when checking it out and checking it in.
10 * This should use the pathname to decide on whether it wants to do some
11 * more interesting conversions (automatic gzip/unzip, general format
12 * conversions etc etc), but by default it just does automatic CRLF<->LF
13 * translation when the "text" attribute or "auto_crlf" option is set.
16 enum crlf_action {
17 CRLF_GUESS = -1,
18 CRLF_BINARY = 0,
19 CRLF_TEXT,
20 CRLF_INPUT,
21 CRLF_CRLF,
22 CRLF_AUTO
25 struct text_stat {
26 /* NUL, CR, LF and CRLF counts */
27 unsigned nul, cr, lf, crlf;
29 /* These are just approximations! */
30 unsigned printable, nonprintable;
33 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
35 unsigned long i;
37 memset(stats, 0, sizeof(*stats));
39 for (i = 0; i < size; i++) {
40 unsigned char c = buf[i];
41 if (c == '\r') {
42 stats->cr++;
43 if (i+1 < size && buf[i+1] == '\n')
44 stats->crlf++;
45 continue;
47 if (c == '\n') {
48 stats->lf++;
49 continue;
51 if (c == 127)
52 /* DEL */
53 stats->nonprintable++;
54 else if (c < 32) {
55 switch (c) {
56 /* BS, HT, ESC and FF */
57 case '\b': case '\t': case '\033': case '\014':
58 stats->printable++;
59 break;
60 case 0:
61 stats->nul++;
62 /* fall through */
63 default:
64 stats->nonprintable++;
67 else
68 stats->printable++;
71 /* If file ends with EOF then don't count this EOF as non-printable. */
72 if (size >= 1 && buf[size-1] == '\032')
73 stats->nonprintable--;
77 * The same heuristics as diff.c::mmfile_is_binary()
79 static int is_binary(unsigned long size, struct text_stat *stats)
82 if (stats->nul)
83 return 1;
84 if ((stats->printable >> 7) < stats->nonprintable)
85 return 1;
87 * Other heuristics? Average line length might be relevant,
88 * as might LF vs CR vs CRLF counts..
90 * NOTE! It might be normal to have a low ratio of CRLF to LF
91 * (somebody starts with a LF-only file and edits it with an editor
92 * that adds CRLF only to lines that are added..). But do we
93 * want to support CR-only? Probably not.
95 return 0;
98 static enum eol output_eol(enum crlf_action crlf_action)
100 switch (crlf_action) {
101 case CRLF_BINARY:
102 return EOL_UNSET;
103 case CRLF_CRLF:
104 return EOL_CRLF;
105 case CRLF_INPUT:
106 return EOL_LF;
107 case CRLF_GUESS:
108 if (!auto_crlf)
109 return EOL_UNSET;
110 /* fall through */
111 case CRLF_TEXT:
112 case CRLF_AUTO:
113 if (auto_crlf == AUTO_CRLF_TRUE)
114 return EOL_CRLF;
115 else if (auto_crlf == AUTO_CRLF_INPUT)
116 return EOL_LF;
117 else if (core_eol == EOL_UNSET)
118 return EOL_NATIVE;
120 return core_eol;
123 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
124 struct text_stat *stats, enum safe_crlf checksafe)
126 if (!checksafe)
127 return;
129 if (output_eol(crlf_action) == EOL_LF) {
131 * CRLFs would not be restored by checkout:
132 * check if we'd remove CRLFs
134 if (stats->crlf) {
135 if (checksafe == SAFE_CRLF_WARN)
136 warning("CRLF will be replaced by LF in %s.\nThe file will have its original line endings in your working directory.", path);
137 else /* i.e. SAFE_CRLF_FAIL */
138 die("CRLF would be replaced by LF in %s.", path);
140 } else if (output_eol(crlf_action) == EOL_CRLF) {
142 * CRLFs would be added by checkout:
143 * check if we have "naked" LFs
145 if (stats->lf != stats->crlf) {
146 if (checksafe == SAFE_CRLF_WARN)
147 warning("LF will be replaced by CRLF in %s.\nThe file will have its original line endings in your working directory.", path);
148 else /* i.e. SAFE_CRLF_FAIL */
149 die("LF would be replaced by CRLF in %s", path);
154 static int has_cr_in_index(const char *path)
156 unsigned long sz;
157 void *data;
158 int has_cr;
160 data = read_blob_data_from_cache(path, &sz);
161 if (!data)
162 return 0;
163 has_cr = memchr(data, '\r', sz) != NULL;
164 free(data);
165 return has_cr;
168 static int crlf_to_git(const char *path, const char *src, size_t len,
169 struct strbuf *buf,
170 enum crlf_action crlf_action, enum safe_crlf checksafe)
172 struct text_stat stats;
173 char *dst;
175 if (crlf_action == CRLF_BINARY ||
176 (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE) ||
177 (src && !len))
178 return 0;
181 * If we are doing a dry-run and have no source buffer, there is
182 * nothing to analyze; we must assume we would convert.
184 if (!buf && !src)
185 return 1;
187 gather_stats(src, len, &stats);
189 if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
191 * We're currently not going to even try to convert stuff
192 * that has bare CR characters. Does anybody do that crazy
193 * stuff?
195 if (stats.cr != stats.crlf)
196 return 0;
199 * And add some heuristics for binary vs text, of course...
201 if (is_binary(len, &stats))
202 return 0;
204 if (crlf_action == CRLF_GUESS) {
206 * If the file in the index has any CR in it, do not convert.
207 * This is the new safer autocrlf handling.
209 if (has_cr_in_index(path))
210 return 0;
214 check_safe_crlf(path, crlf_action, &stats, checksafe);
216 /* Optimization: No CR? Nothing to convert, regardless. */
217 if (!stats.cr)
218 return 0;
221 * At this point all of our source analysis is done, and we are sure we
222 * would convert. If we are in dry-run mode, we can give an answer.
224 if (!buf)
225 return 1;
227 /* only grow if not in place */
228 if (strbuf_avail(buf) + buf->len < len)
229 strbuf_grow(buf, len - buf->len);
230 dst = buf->buf;
231 if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
233 * If we guessed, we already know we rejected a file with
234 * lone CR, and we can strip a CR without looking at what
235 * follow it.
237 do {
238 unsigned char c = *src++;
239 if (c != '\r')
240 *dst++ = c;
241 } while (--len);
242 } else {
243 do {
244 unsigned char c = *src++;
245 if (! (c == '\r' && (1 < len && *src == '\n')))
246 *dst++ = c;
247 } while (--len);
249 strbuf_setlen(buf, dst - buf->buf);
250 return 1;
253 static int crlf_to_worktree(const char *path, const char *src, size_t len,
254 struct strbuf *buf, enum crlf_action crlf_action)
256 char *to_free = NULL;
257 struct text_stat stats;
259 if (!len || output_eol(crlf_action) != EOL_CRLF)
260 return 0;
262 gather_stats(src, len, &stats);
264 /* No LF? Nothing to convert, regardless. */
265 if (!stats.lf)
266 return 0;
268 /* Was it already in CRLF format? */
269 if (stats.lf == stats.crlf)
270 return 0;
272 if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
273 if (crlf_action == CRLF_GUESS) {
274 /* If we have any CR or CRLF line endings, we do not touch it */
275 /* This is the new safer autocrlf-handling */
276 if (stats.cr > 0 || stats.crlf > 0)
277 return 0;
280 /* If we have any bare CR characters, we're not going to touch it */
281 if (stats.cr != stats.crlf)
282 return 0;
284 if (is_binary(len, &stats))
285 return 0;
288 /* are we "faking" in place editing ? */
289 if (src == buf->buf)
290 to_free = strbuf_detach(buf, NULL);
292 strbuf_grow(buf, len + stats.lf - stats.crlf);
293 for (;;) {
294 const char *nl = memchr(src, '\n', len);
295 if (!nl)
296 break;
297 if (nl > src && nl[-1] == '\r') {
298 strbuf_add(buf, src, nl + 1 - src);
299 } else {
300 strbuf_add(buf, src, nl - src);
301 strbuf_addstr(buf, "\r\n");
303 len -= nl + 1 - src;
304 src = nl + 1;
306 strbuf_add(buf, src, len);
308 free(to_free);
309 return 1;
312 struct filter_params {
313 const char *src;
314 unsigned long size;
315 int fd;
316 const char *cmd;
317 const char *path;
320 static int filter_buffer_or_fd(int in, int out, void *data)
323 * Spawn cmd and feed the buffer contents through its stdin.
325 struct child_process child_process = CHILD_PROCESS_INIT;
326 struct filter_params *params = (struct filter_params *)data;
327 int write_err, status;
328 const char *argv[] = { NULL, NULL };
330 /* apply % substitution to cmd */
331 struct strbuf cmd = STRBUF_INIT;
332 struct strbuf path = STRBUF_INIT;
333 struct strbuf_expand_dict_entry dict[] = {
334 { "f", NULL, },
335 { NULL, NULL, },
338 /* quote the path to preserve spaces, etc. */
339 sq_quote_buf(&path, params->path);
340 dict[0].value = path.buf;
342 /* expand all %f with the quoted path */
343 strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
344 strbuf_release(&path);
346 argv[0] = cmd.buf;
348 child_process.argv = argv;
349 child_process.use_shell = 1;
350 child_process.in = -1;
351 child_process.out = out;
353 if (start_command(&child_process))
354 return error("cannot fork to run external filter %s", params->cmd);
356 sigchain_push(SIGPIPE, SIG_IGN);
358 if (params->src) {
359 write_err = (write_in_full(child_process.in, params->src, params->size) < 0);
360 } else {
361 write_err = copy_fd(params->fd, child_process.in);
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 sigchain_pop(SIGPIPE);
371 status = finish_command(&child_process);
372 if (status)
373 error("external filter %s failed %d", params->cmd, status);
375 strbuf_release(&cmd);
376 return (write_err || status);
379 static int apply_filter(const char *path, const char *src, size_t len, int fd,
380 struct strbuf *dst, const char *cmd)
383 * Create a pipeline to have the command filter the buffer's
384 * contents.
386 * (child --> cmd) --> us
388 int ret = 1;
389 struct strbuf nbuf = STRBUF_INIT;
390 struct async async;
391 struct filter_params params;
393 if (!cmd)
394 return 0;
396 if (!dst)
397 return 1;
399 memset(&async, 0, sizeof(async));
400 async.proc = filter_buffer_or_fd;
401 async.data = &params;
402 async.out = -1;
403 params.src = src;
404 params.size = len;
405 params.fd = fd;
406 params.cmd = cmd;
407 params.path = path;
409 fflush(NULL);
410 if (start_async(&async))
411 return 0; /* error was already reported */
413 if (strbuf_read(&nbuf, async.out, len) < 0) {
414 error("read from external filter %s failed", cmd);
415 ret = 0;
417 if (close(async.out)) {
418 error("read from external filter %s failed", cmd);
419 ret = 0;
421 if (finish_async(&async)) {
422 error("external filter %s failed", cmd);
423 ret = 0;
426 if (ret) {
427 strbuf_swap(dst, &nbuf);
429 strbuf_release(&nbuf);
430 return ret;
433 static struct convert_driver {
434 const char *name;
435 struct convert_driver *next;
436 const char *smudge;
437 const char *clean;
438 int required;
439 } *user_convert, **user_convert_tail;
441 static int read_convert_config(const char *var, const char *value, void *cb)
443 const char *key, *name;
444 int namelen;
445 struct convert_driver *drv;
448 * External conversion drivers are configured using
449 * "filter.<name>.variable".
451 if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
452 return 0;
453 for (drv = user_convert; drv; drv = drv->next)
454 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
455 break;
456 if (!drv) {
457 drv = xcalloc(1, sizeof(struct convert_driver));
458 drv->name = xmemdupz(name, namelen);
459 *user_convert_tail = drv;
460 user_convert_tail = &(drv->next);
464 * filter.<name>.smudge and filter.<name>.clean specifies
465 * the command line:
467 * command-line
469 * The command-line will not be interpolated in any way.
472 if (!strcmp("smudge", key))
473 return git_config_string(&drv->smudge, var, value);
475 if (!strcmp("clean", key))
476 return git_config_string(&drv->clean, var, value);
478 if (!strcmp("required", key)) {
479 drv->required = git_config_bool(var, value);
480 return 0;
483 return 0;
486 static int count_ident(const char *cp, unsigned long size)
489 * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
491 int cnt = 0;
492 char ch;
494 while (size) {
495 ch = *cp++;
496 size--;
497 if (ch != '$')
498 continue;
499 if (size < 3)
500 break;
501 if (memcmp("Id", cp, 2))
502 continue;
503 ch = cp[2];
504 cp += 3;
505 size -= 3;
506 if (ch == '$')
507 cnt++; /* $Id$ */
508 if (ch != ':')
509 continue;
512 * "$Id: ... "; scan up to the closing dollar sign and discard.
514 while (size) {
515 ch = *cp++;
516 size--;
517 if (ch == '$') {
518 cnt++;
519 break;
521 if (ch == '\n')
522 break;
525 return cnt;
528 static int ident_to_git(const char *path, const char *src, size_t len,
529 struct strbuf *buf, int ident)
531 char *dst, *dollar;
533 if (!ident || (src && !count_ident(src, len)))
534 return 0;
536 if (!buf)
537 return 1;
539 /* only grow if not in place */
540 if (strbuf_avail(buf) + buf->len < len)
541 strbuf_grow(buf, len - buf->len);
542 dst = buf->buf;
543 for (;;) {
544 dollar = memchr(src, '$', len);
545 if (!dollar)
546 break;
547 memmove(dst, src, dollar + 1 - src);
548 dst += dollar + 1 - src;
549 len -= dollar + 1 - src;
550 src = dollar + 1;
552 if (len > 3 && !memcmp(src, "Id:", 3)) {
553 dollar = memchr(src + 3, '$', len - 3);
554 if (!dollar)
555 break;
556 if (memchr(src + 3, '\n', dollar - src - 3)) {
557 /* Line break before the next dollar. */
558 continue;
561 memcpy(dst, "Id$", 3);
562 dst += 3;
563 len -= dollar + 1 - src;
564 src = dollar + 1;
567 memmove(dst, src, len);
568 strbuf_setlen(buf, dst + len - buf->buf);
569 return 1;
572 static int ident_to_worktree(const char *path, const char *src, size_t len,
573 struct strbuf *buf, int ident)
575 unsigned char sha1[20];
576 char *to_free = NULL, *dollar, *spc;
577 int cnt;
579 if (!ident)
580 return 0;
582 cnt = count_ident(src, len);
583 if (!cnt)
584 return 0;
586 /* are we "faking" in place editing ? */
587 if (src == buf->buf)
588 to_free = strbuf_detach(buf, NULL);
589 hash_sha1_file(src, len, "blob", sha1);
591 strbuf_grow(buf, len + cnt * 43);
592 for (;;) {
593 /* step 1: run to the next '$' */
594 dollar = memchr(src, '$', len);
595 if (!dollar)
596 break;
597 strbuf_add(buf, src, dollar + 1 - src);
598 len -= dollar + 1 - src;
599 src = dollar + 1;
601 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
602 if (len < 3 || memcmp("Id", src, 2))
603 continue;
605 /* step 3: skip over Id$ or Id:xxxxx$ */
606 if (src[2] == '$') {
607 src += 3;
608 len -= 3;
609 } else if (src[2] == ':') {
611 * It's possible that an expanded Id has crept its way into the
612 * repository, we cope with that by stripping the expansion out.
613 * This is probably not a good idea, since it will cause changes
614 * on checkout, which won't go away by stash, but let's keep it
615 * for git-style ids.
617 dollar = memchr(src + 3, '$', len - 3);
618 if (!dollar) {
619 /* incomplete keyword, no more '$', so just quit the loop */
620 break;
623 if (memchr(src + 3, '\n', dollar - src - 3)) {
624 /* Line break before the next dollar. */
625 continue;
628 spc = memchr(src + 4, ' ', dollar - src - 4);
629 if (spc && spc < dollar-1) {
630 /* There are spaces in unexpected places.
631 * This is probably an id from some other
632 * versioning system. Keep it for now.
634 continue;
637 len -= dollar + 1 - src;
638 src = dollar + 1;
639 } else {
640 /* it wasn't a "Id$" or "Id:xxxx$" */
641 continue;
644 /* step 4: substitute */
645 strbuf_addstr(buf, "Id: ");
646 strbuf_add(buf, sha1_to_hex(sha1), 40);
647 strbuf_addstr(buf, " $");
649 strbuf_add(buf, src, len);
651 free(to_free);
652 return 1;
655 static enum crlf_action git_path_check_crlf(const char *path, struct git_attr_check *check)
657 const char *value = check->value;
659 if (ATTR_TRUE(value))
660 return CRLF_TEXT;
661 else if (ATTR_FALSE(value))
662 return CRLF_BINARY;
663 else if (ATTR_UNSET(value))
665 else if (!strcmp(value, "input"))
666 return CRLF_INPUT;
667 else if (!strcmp(value, "auto"))
668 return CRLF_AUTO;
669 return CRLF_GUESS;
672 static enum eol git_path_check_eol(const char *path, struct git_attr_check *check)
674 const char *value = check->value;
676 if (ATTR_UNSET(value))
678 else if (!strcmp(value, "lf"))
679 return EOL_LF;
680 else if (!strcmp(value, "crlf"))
681 return EOL_CRLF;
682 return EOL_UNSET;
685 static struct convert_driver *git_path_check_convert(const char *path,
686 struct git_attr_check *check)
688 const char *value = check->value;
689 struct convert_driver *drv;
691 if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
692 return NULL;
693 for (drv = user_convert; drv; drv = drv->next)
694 if (!strcmp(value, drv->name))
695 return drv;
696 return NULL;
699 static int git_path_check_ident(const char *path, struct git_attr_check *check)
701 const char *value = check->value;
703 return !!ATTR_TRUE(value);
706 static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
708 if (text_attr == CRLF_BINARY)
709 return CRLF_BINARY;
710 if (eol_attr == EOL_LF)
711 return CRLF_INPUT;
712 if (eol_attr == EOL_CRLF)
713 return CRLF_CRLF;
714 return text_attr;
717 struct conv_attrs {
718 struct convert_driver *drv;
719 enum crlf_action crlf_action;
720 enum eol eol_attr;
721 int ident;
724 static const char *conv_attr_name[] = {
725 "crlf", "ident", "filter", "eol", "text",
727 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
729 static void convert_attrs(struct conv_attrs *ca, const char *path)
731 int i;
732 static struct git_attr_check ccheck[NUM_CONV_ATTRS];
734 if (!ccheck[0].attr) {
735 for (i = 0; i < NUM_CONV_ATTRS; i++)
736 ccheck[i].attr = git_attr(conv_attr_name[i]);
737 user_convert_tail = &user_convert;
738 git_config(read_convert_config, NULL);
741 if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
742 ca->crlf_action = git_path_check_crlf(path, ccheck + 4);
743 if (ca->crlf_action == CRLF_GUESS)
744 ca->crlf_action = git_path_check_crlf(path, ccheck + 0);
745 ca->ident = git_path_check_ident(path, ccheck + 1);
746 ca->drv = git_path_check_convert(path, ccheck + 2);
747 ca->eol_attr = git_path_check_eol(path, ccheck + 3);
748 } else {
749 ca->drv = NULL;
750 ca->crlf_action = CRLF_GUESS;
751 ca->eol_attr = EOL_UNSET;
752 ca->ident = 0;
756 int would_convert_to_git_filter_fd(const char *path)
758 struct conv_attrs ca;
760 convert_attrs(&ca, path);
761 if (!ca.drv)
762 return 0;
765 * Apply a filter to an fd only if the filter is required to succeed.
766 * We must die if the filter fails, because the original data before
767 * filtering is not available.
769 if (!ca.drv->required)
770 return 0;
772 return apply_filter(path, NULL, 0, -1, NULL, ca.drv->clean);
775 int convert_to_git(const char *path, const char *src, size_t len,
776 struct strbuf *dst, enum safe_crlf checksafe)
778 int ret = 0;
779 const char *filter = NULL;
780 int required = 0;
781 struct conv_attrs ca;
783 convert_attrs(&ca, path);
784 if (ca.drv) {
785 filter = ca.drv->clean;
786 required = ca.drv->required;
789 ret |= apply_filter(path, src, len, -1, dst, filter);
790 if (!ret && required)
791 die("%s: clean filter '%s' failed", path, ca.drv->name);
793 if (ret && dst) {
794 src = dst->buf;
795 len = dst->len;
797 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
798 ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
799 if (ret && dst) {
800 src = dst->buf;
801 len = dst->len;
803 return ret | ident_to_git(path, src, len, dst, ca.ident);
806 void convert_to_git_filter_fd(const char *path, int fd, struct strbuf *dst,
807 enum safe_crlf checksafe)
809 struct conv_attrs ca;
810 convert_attrs(&ca, path);
812 assert(ca.drv);
813 assert(ca.drv->clean);
815 if (!apply_filter(path, NULL, 0, fd, dst, ca.drv->clean))
816 die("%s: clean filter '%s' failed", path, ca.drv->name);
818 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
819 crlf_to_git(path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
820 ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
823 static int convert_to_working_tree_internal(const char *path, const char *src,
824 size_t len, struct strbuf *dst,
825 int normalizing)
827 int ret = 0, ret_filter = 0;
828 const char *filter = NULL;
829 int required = 0;
830 struct conv_attrs ca;
832 convert_attrs(&ca, path);
833 if (ca.drv) {
834 filter = ca.drv->smudge;
835 required = ca.drv->required;
838 ret |= ident_to_worktree(path, src, len, dst, ca.ident);
839 if (ret) {
840 src = dst->buf;
841 len = dst->len;
844 * CRLF conversion can be skipped if normalizing, unless there
845 * is a smudge filter. The filter might expect CRLFs.
847 if (filter || !normalizing) {
848 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
849 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
850 if (ret) {
851 src = dst->buf;
852 len = dst->len;
856 ret_filter = apply_filter(path, src, len, -1, dst, filter);
857 if (!ret_filter && required)
858 die("%s: smudge filter %s failed", path, ca.drv->name);
860 return ret | ret_filter;
863 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
865 return convert_to_working_tree_internal(path, src, len, dst, 0);
868 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
870 int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
871 if (ret) {
872 src = dst->buf;
873 len = dst->len;
875 return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_FALSE);
878 /*****************************************************************
880 * Streaming conversion support
882 *****************************************************************/
884 typedef int (*filter_fn)(struct stream_filter *,
885 const char *input, size_t *isize_p,
886 char *output, size_t *osize_p);
887 typedef void (*free_fn)(struct stream_filter *);
889 struct stream_filter_vtbl {
890 filter_fn filter;
891 free_fn free;
894 struct stream_filter {
895 struct stream_filter_vtbl *vtbl;
898 static int null_filter_fn(struct stream_filter *filter,
899 const char *input, size_t *isize_p,
900 char *output, size_t *osize_p)
902 size_t count;
904 if (!input)
905 return 0; /* we do not keep any states */
906 count = *isize_p;
907 if (*osize_p < count)
908 count = *osize_p;
909 if (count) {
910 memmove(output, input, count);
911 *isize_p -= count;
912 *osize_p -= count;
914 return 0;
917 static void null_free_fn(struct stream_filter *filter)
919 ; /* nothing -- null instances are shared */
922 static struct stream_filter_vtbl null_vtbl = {
923 null_filter_fn,
924 null_free_fn,
927 static struct stream_filter null_filter_singleton = {
928 &null_vtbl,
931 int is_null_stream_filter(struct stream_filter *filter)
933 return filter == &null_filter_singleton;
938 * LF-to-CRLF filter
941 struct lf_to_crlf_filter {
942 struct stream_filter filter;
943 unsigned has_held:1;
944 char held;
947 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
948 const char *input, size_t *isize_p,
949 char *output, size_t *osize_p)
951 size_t count, o = 0;
952 struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
955 * We may be holding onto the CR to see if it is followed by a
956 * LF, in which case we would need to go to the main loop.
957 * Otherwise, just emit it to the output stream.
959 if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
960 output[o++] = lf_to_crlf->held;
961 lf_to_crlf->has_held = 0;
964 /* We are told to drain */
965 if (!input) {
966 *osize_p -= o;
967 return 0;
970 count = *isize_p;
971 if (count || lf_to_crlf->has_held) {
972 size_t i;
973 int was_cr = 0;
975 if (lf_to_crlf->has_held) {
976 was_cr = 1;
977 lf_to_crlf->has_held = 0;
980 for (i = 0; o < *osize_p && i < count; i++) {
981 char ch = input[i];
983 if (ch == '\n') {
984 output[o++] = '\r';
985 } else if (was_cr) {
987 * Previous round saw CR and it is not followed
988 * by a LF; emit the CR before processing the
989 * current character.
991 output[o++] = '\r';
995 * We may have consumed the last output slot,
996 * in which case we need to break out of this
997 * loop; hold the current character before
998 * returning.
1000 if (*osize_p <= o) {
1001 lf_to_crlf->has_held = 1;
1002 lf_to_crlf->held = ch;
1003 continue; /* break but increment i */
1006 if (ch == '\r') {
1007 was_cr = 1;
1008 continue;
1011 was_cr = 0;
1012 output[o++] = ch;
1015 *osize_p -= o;
1016 *isize_p -= i;
1018 if (!lf_to_crlf->has_held && was_cr) {
1019 lf_to_crlf->has_held = 1;
1020 lf_to_crlf->held = '\r';
1023 return 0;
1026 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1028 free(filter);
1031 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1032 lf_to_crlf_filter_fn,
1033 lf_to_crlf_free_fn,
1036 static struct stream_filter *lf_to_crlf_filter(void)
1038 struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1040 lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1041 return (struct stream_filter *)lf_to_crlf;
1045 * Cascade filter
1047 #define FILTER_BUFFER 1024
1048 struct cascade_filter {
1049 struct stream_filter filter;
1050 struct stream_filter *one;
1051 struct stream_filter *two;
1052 char buf[FILTER_BUFFER];
1053 int end, ptr;
1056 static int cascade_filter_fn(struct stream_filter *filter,
1057 const char *input, size_t *isize_p,
1058 char *output, size_t *osize_p)
1060 struct cascade_filter *cas = (struct cascade_filter *) filter;
1061 size_t filled = 0;
1062 size_t sz = *osize_p;
1063 size_t to_feed, remaining;
1066 * input -- (one) --> buf -- (two) --> output
1068 while (filled < sz) {
1069 remaining = sz - filled;
1071 /* do we already have something to feed two with? */
1072 if (cas->ptr < cas->end) {
1073 to_feed = cas->end - cas->ptr;
1074 if (stream_filter(cas->two,
1075 cas->buf + cas->ptr, &to_feed,
1076 output + filled, &remaining))
1077 return -1;
1078 cas->ptr += (cas->end - cas->ptr) - to_feed;
1079 filled = sz - remaining;
1080 continue;
1083 /* feed one from upstream and have it emit into our buffer */
1084 to_feed = input ? *isize_p : 0;
1085 if (input && !to_feed)
1086 break;
1087 remaining = sizeof(cas->buf);
1088 if (stream_filter(cas->one,
1089 input, &to_feed,
1090 cas->buf, &remaining))
1091 return -1;
1092 cas->end = sizeof(cas->buf) - remaining;
1093 cas->ptr = 0;
1094 if (input) {
1095 size_t fed = *isize_p - to_feed;
1096 *isize_p -= fed;
1097 input += fed;
1100 /* do we know that we drained one completely? */
1101 if (input || cas->end)
1102 continue;
1104 /* tell two to drain; we have nothing more to give it */
1105 to_feed = 0;
1106 remaining = sz - filled;
1107 if (stream_filter(cas->two,
1108 NULL, &to_feed,
1109 output + filled, &remaining))
1110 return -1;
1111 if (remaining == (sz - filled))
1112 break; /* completely drained two */
1113 filled = sz - remaining;
1115 *osize_p -= filled;
1116 return 0;
1119 static void cascade_free_fn(struct stream_filter *filter)
1121 struct cascade_filter *cas = (struct cascade_filter *)filter;
1122 free_stream_filter(cas->one);
1123 free_stream_filter(cas->two);
1124 free(filter);
1127 static struct stream_filter_vtbl cascade_vtbl = {
1128 cascade_filter_fn,
1129 cascade_free_fn,
1132 static struct stream_filter *cascade_filter(struct stream_filter *one,
1133 struct stream_filter *two)
1135 struct cascade_filter *cascade;
1137 if (!one || is_null_stream_filter(one))
1138 return two;
1139 if (!two || is_null_stream_filter(two))
1140 return one;
1142 cascade = xmalloc(sizeof(*cascade));
1143 cascade->one = one;
1144 cascade->two = two;
1145 cascade->end = cascade->ptr = 0;
1146 cascade->filter.vtbl = &cascade_vtbl;
1147 return (struct stream_filter *)cascade;
1151 * ident filter
1153 #define IDENT_DRAINING (-1)
1154 #define IDENT_SKIPPING (-2)
1155 struct ident_filter {
1156 struct stream_filter filter;
1157 struct strbuf left;
1158 int state;
1159 char ident[45]; /* ": x40 $" */
1162 static int is_foreign_ident(const char *str)
1164 int i;
1166 if (!skip_prefix(str, "$Id: ", &str))
1167 return 0;
1168 for (i = 0; str[i]; i++) {
1169 if (isspace(str[i]) && str[i+1] != '$')
1170 return 1;
1172 return 0;
1175 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1177 size_t to_drain = ident->left.len;
1179 if (*osize_p < to_drain)
1180 to_drain = *osize_p;
1181 if (to_drain) {
1182 memcpy(*output_p, ident->left.buf, to_drain);
1183 strbuf_remove(&ident->left, 0, to_drain);
1184 *output_p += to_drain;
1185 *osize_p -= to_drain;
1187 if (!ident->left.len)
1188 ident->state = 0;
1191 static int ident_filter_fn(struct stream_filter *filter,
1192 const char *input, size_t *isize_p,
1193 char *output, size_t *osize_p)
1195 struct ident_filter *ident = (struct ident_filter *)filter;
1196 static const char head[] = "$Id";
1198 if (!input) {
1199 /* drain upon eof */
1200 switch (ident->state) {
1201 default:
1202 strbuf_add(&ident->left, head, ident->state);
1203 case IDENT_SKIPPING:
1204 /* fallthru */
1205 case IDENT_DRAINING:
1206 ident_drain(ident, &output, osize_p);
1208 return 0;
1211 while (*isize_p || (ident->state == IDENT_DRAINING)) {
1212 int ch;
1214 if (ident->state == IDENT_DRAINING) {
1215 ident_drain(ident, &output, osize_p);
1216 if (!*osize_p)
1217 break;
1218 continue;
1221 ch = *(input++);
1222 (*isize_p)--;
1224 if (ident->state == IDENT_SKIPPING) {
1226 * Skipping until '$' or LF, but keeping them
1227 * in case it is a foreign ident.
1229 strbuf_addch(&ident->left, ch);
1230 if (ch != '\n' && ch != '$')
1231 continue;
1232 if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1233 strbuf_setlen(&ident->left, sizeof(head) - 1);
1234 strbuf_addstr(&ident->left, ident->ident);
1236 ident->state = IDENT_DRAINING;
1237 continue;
1240 if (ident->state < sizeof(head) &&
1241 head[ident->state] == ch) {
1242 ident->state++;
1243 continue;
1246 if (ident->state)
1247 strbuf_add(&ident->left, head, ident->state);
1248 if (ident->state == sizeof(head) - 1) {
1249 if (ch != ':' && ch != '$') {
1250 strbuf_addch(&ident->left, ch);
1251 ident->state = 0;
1252 continue;
1255 if (ch == ':') {
1256 strbuf_addch(&ident->left, ch);
1257 ident->state = IDENT_SKIPPING;
1258 } else {
1259 strbuf_addstr(&ident->left, ident->ident);
1260 ident->state = IDENT_DRAINING;
1262 continue;
1265 strbuf_addch(&ident->left, ch);
1266 ident->state = IDENT_DRAINING;
1268 return 0;
1271 static void ident_free_fn(struct stream_filter *filter)
1273 struct ident_filter *ident = (struct ident_filter *)filter;
1274 strbuf_release(&ident->left);
1275 free(filter);
1278 static struct stream_filter_vtbl ident_vtbl = {
1279 ident_filter_fn,
1280 ident_free_fn,
1283 static struct stream_filter *ident_filter(const unsigned char *sha1)
1285 struct ident_filter *ident = xmalloc(sizeof(*ident));
1287 sprintf(ident->ident, ": %s $", sha1_to_hex(sha1));
1288 strbuf_init(&ident->left, 0);
1289 ident->filter.vtbl = &ident_vtbl;
1290 ident->state = 0;
1291 return (struct stream_filter *)ident;
1295 * Return an appropriately constructed filter for the path, or NULL if
1296 * the contents cannot be filtered without reading the whole thing
1297 * in-core.
1299 * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1300 * large binary blob you would want us not to slurp into the memory!
1302 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1304 struct conv_attrs ca;
1305 enum crlf_action crlf_action;
1306 struct stream_filter *filter = NULL;
1308 convert_attrs(&ca, path);
1310 if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1311 return filter;
1313 if (ca.ident)
1314 filter = ident_filter(sha1);
1316 crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
1318 if ((crlf_action == CRLF_BINARY) || (crlf_action == CRLF_INPUT) ||
1319 (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE))
1320 filter = cascade_filter(filter, &null_filter_singleton);
1322 else if (output_eol(crlf_action) == EOL_CRLF &&
1323 !(crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS))
1324 filter = cascade_filter(filter, lf_to_crlf_filter());
1326 return filter;
1329 void free_stream_filter(struct stream_filter *filter)
1331 filter->vtbl->free(filter);
1334 int stream_filter(struct stream_filter *filter,
1335 const char *input, size_t *isize_p,
1336 char *output, size_t *osize_p)
1338 return filter->vtbl->filter(filter, input, isize_p, output, osize_p);