Merge branch 'jk/asan-build-fix'
[git] / convert.c
1 #include "cache.h"
2 #include "config.h"
3 #include "object-store.h"
4 #include "attr.h"
5 #include "run-command.h"
6 #include "quote.h"
7 #include "sigchain.h"
8 #include "pkt-line.h"
9 #include "sub-process.h"
10 #include "utf8.h"
11 #include "ll-merge.h"
12
13 /*
14  * convert.c - convert a file when checking it out and checking it in.
15  *
16  * This should use the pathname to decide on whether it wants to do some
17  * more interesting conversions (automatic gzip/unzip, general format
18  * conversions etc etc), but by default it just does automatic CRLF<->LF
19  * translation when the "text" attribute or "auto_crlf" option is set.
20  */
21
22 /* Stat bits: When BIN is set, the txt bits are unset */
23 #define CONVERT_STAT_BITS_TXT_LF    0x1
24 #define CONVERT_STAT_BITS_TXT_CRLF  0x2
25 #define CONVERT_STAT_BITS_BIN       0x4
26
27 enum crlf_action {
28         CRLF_UNDEFINED,
29         CRLF_BINARY,
30         CRLF_TEXT,
31         CRLF_TEXT_INPUT,
32         CRLF_TEXT_CRLF,
33         CRLF_AUTO,
34         CRLF_AUTO_INPUT,
35         CRLF_AUTO_CRLF
36 };
37
38 struct text_stat {
39         /* NUL, CR, LF and CRLF counts */
40         unsigned nul, lonecr, lonelf, crlf;
41
42         /* These are just approximations! */
43         unsigned printable, nonprintable;
44 };
45
46 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
47 {
48         unsigned long i;
49
50         memset(stats, 0, sizeof(*stats));
51
52         for (i = 0; i < size; i++) {
53                 unsigned char c = buf[i];
54                 if (c == '\r') {
55                         if (i+1 < size && buf[i+1] == '\n') {
56                                 stats->crlf++;
57                                 i++;
58                         } else
59                                 stats->lonecr++;
60                         continue;
61                 }
62                 if (c == '\n') {
63                         stats->lonelf++;
64                         continue;
65                 }
66                 if (c == 127)
67                         /* DEL */
68                         stats->nonprintable++;
69                 else if (c < 32) {
70                         switch (c) {
71                                 /* BS, HT, ESC and FF */
72                         case '\b': case '\t': case '\033': case '\014':
73                                 stats->printable++;
74                                 break;
75                         case 0:
76                                 stats->nul++;
77                                 /* fall through */
78                         default:
79                                 stats->nonprintable++;
80                         }
81                 }
82                 else
83                         stats->printable++;
84         }
85
86         /* If file ends with EOF then don't count this EOF as non-printable. */
87         if (size >= 1 && buf[size-1] == '\032')
88                 stats->nonprintable--;
89 }
90
91 /*
92  * The same heuristics as diff.c::mmfile_is_binary()
93  * We treat files with bare CR as binary
94  */
95 static int convert_is_binary(const struct text_stat *stats)
96 {
97         if (stats->lonecr)
98                 return 1;
99         if (stats->nul)
100                 return 1;
101         if ((stats->printable >> 7) < stats->nonprintable)
102                 return 1;
103         return 0;
104 }
105
106 static unsigned int gather_convert_stats(const char *data, unsigned long size)
107 {
108         struct text_stat stats;
109         int ret = 0;
110         if (!data || !size)
111                 return 0;
112         gather_stats(data, size, &stats);
113         if (convert_is_binary(&stats))
114                 ret |= CONVERT_STAT_BITS_BIN;
115         if (stats.crlf)
116                 ret |= CONVERT_STAT_BITS_TXT_CRLF;
117         if (stats.lonelf)
118                 ret |=  CONVERT_STAT_BITS_TXT_LF;
119
120         return ret;
121 }
122
123 static const char *gather_convert_stats_ascii(const char *data, unsigned long size)
124 {
125         unsigned int convert_stats = gather_convert_stats(data, size);
126
127         if (convert_stats & CONVERT_STAT_BITS_BIN)
128                 return "-text";
129         switch (convert_stats) {
130         case CONVERT_STAT_BITS_TXT_LF:
131                 return "lf";
132         case CONVERT_STAT_BITS_TXT_CRLF:
133                 return "crlf";
134         case CONVERT_STAT_BITS_TXT_LF | CONVERT_STAT_BITS_TXT_CRLF:
135                 return "mixed";
136         default:
137                 return "none";
138         }
139 }
140
141 const char *get_cached_convert_stats_ascii(const struct index_state *istate,
142                                            const char *path)
143 {
144         const char *ret;
145         unsigned long sz;
146         void *data = read_blob_data_from_index(istate, path, &sz);
147         ret = gather_convert_stats_ascii(data, sz);
148         free(data);
149         return ret;
150 }
151
152 const char *get_wt_convert_stats_ascii(const char *path)
153 {
154         const char *ret = "";
155         struct strbuf sb = STRBUF_INIT;
156         if (strbuf_read_file(&sb, path, 0) >= 0)
157                 ret = gather_convert_stats_ascii(sb.buf, sb.len);
158         strbuf_release(&sb);
159         return ret;
160 }
161
162 static int text_eol_is_crlf(void)
163 {
164         if (auto_crlf == AUTO_CRLF_TRUE)
165                 return 1;
166         else if (auto_crlf == AUTO_CRLF_INPUT)
167                 return 0;
168         if (core_eol == EOL_CRLF)
169                 return 1;
170         if (core_eol == EOL_UNSET && EOL_NATIVE == EOL_CRLF)
171                 return 1;
172         return 0;
173 }
174
175 static enum eol output_eol(enum crlf_action crlf_action)
176 {
177         switch (crlf_action) {
178         case CRLF_BINARY:
179                 return EOL_UNSET;
180         case CRLF_TEXT_CRLF:
181                 return EOL_CRLF;
182         case CRLF_TEXT_INPUT:
183                 return EOL_LF;
184         case CRLF_UNDEFINED:
185         case CRLF_AUTO_CRLF:
186                 return EOL_CRLF;
187         case CRLF_AUTO_INPUT:
188                 return EOL_LF;
189         case CRLF_TEXT:
190         case CRLF_AUTO:
191                 /* fall through */
192                 return text_eol_is_crlf() ? EOL_CRLF : EOL_LF;
193         }
194         warning(_("illegal crlf_action %d"), (int)crlf_action);
195         return core_eol;
196 }
197
198 static void check_global_conv_flags_eol(const char *path, enum crlf_action crlf_action,
199                             struct text_stat *old_stats, struct text_stat *new_stats,
200                             int conv_flags)
201 {
202         if (old_stats->crlf && !new_stats->crlf ) {
203                 /*
204                  * CRLFs would not be restored by checkout
205                  */
206                 if (conv_flags & CONV_EOL_RNDTRP_DIE)
207                         die(_("CRLF would be replaced by LF in %s"), path);
208                 else if (conv_flags & CONV_EOL_RNDTRP_WARN)
209                         warning(_("CRLF will be replaced by LF in %s.\n"
210                                   "The file will have its original line"
211                                   " endings in your working directory"), path);
212         } else if (old_stats->lonelf && !new_stats->lonelf ) {
213                 /*
214                  * CRLFs would be added by checkout
215                  */
216                 if (conv_flags & CONV_EOL_RNDTRP_DIE)
217                         die(_("LF would be replaced by CRLF in %s"), path);
218                 else if (conv_flags & CONV_EOL_RNDTRP_WARN)
219                         warning(_("LF will be replaced by CRLF in %s.\n"
220                                   "The file will have its original line"
221                                   " endings in your working directory"), path);
222         }
223 }
224
225 static int has_crlf_in_index(const struct index_state *istate, const char *path)
226 {
227         unsigned long sz;
228         void *data;
229         const char *crp;
230         int has_crlf = 0;
231
232         data = read_blob_data_from_index(istate, path, &sz);
233         if (!data)
234                 return 0;
235
236         crp = memchr(data, '\r', sz);
237         if (crp) {
238                 unsigned int ret_stats;
239                 ret_stats = gather_convert_stats(data, sz);
240                 if (!(ret_stats & CONVERT_STAT_BITS_BIN) &&
241                     (ret_stats & CONVERT_STAT_BITS_TXT_CRLF))
242                         has_crlf = 1;
243         }
244         free(data);
245         return has_crlf;
246 }
247
248 static int will_convert_lf_to_crlf(struct text_stat *stats,
249                                    enum crlf_action crlf_action)
250 {
251         if (output_eol(crlf_action) != EOL_CRLF)
252                 return 0;
253         /* No "naked" LF? Nothing to convert, regardless. */
254         if (!stats->lonelf)
255                 return 0;
256
257         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
258                 /* If we have any CR or CRLF line endings, we do not touch it */
259                 /* This is the new safer autocrlf-handling */
260                 if (stats->lonecr || stats->crlf)
261                         return 0;
262
263                 if (convert_is_binary(stats))
264                         return 0;
265         }
266         return 1;
267
268 }
269
270 static int validate_encoding(const char *path, const char *enc,
271                       const char *data, size_t len, int die_on_error)
272 {
273         const char *stripped;
274
275         /* We only check for UTF here as UTF?? can be an alias for UTF-?? */
276         if (skip_iprefix(enc, "UTF", &stripped)) {
277                 skip_prefix(stripped, "-", &stripped);
278
279                 /*
280                  * Check for detectable errors in UTF encodings
281                  */
282                 if (has_prohibited_utf_bom(enc, data, len)) {
283                         const char *error_msg = _(
284                                 "BOM is prohibited in '%s' if encoded as %s");
285                         /*
286                          * This advice is shown for UTF-??BE and UTF-??LE encodings.
287                          * We cut off the last two characters of the encoding name
288                          * to generate the encoding name suitable for BOMs.
289                          */
290                         const char *advise_msg = _(
291                                 "The file '%s' contains a byte order "
292                                 "mark (BOM). Please use UTF-%.*s as "
293                                 "working-tree-encoding.");
294                         int stripped_len = strlen(stripped) - strlen("BE");
295                         advise(advise_msg, path, stripped_len, stripped);
296                         if (die_on_error)
297                                 die(error_msg, path, enc);
298                         else {
299                                 return error(error_msg, path, enc);
300                         }
301
302                 } else if (is_missing_required_utf_bom(enc, data, len)) {
303                         const char *error_msg = _(
304                                 "BOM is required in '%s' if encoded as %s");
305                         const char *advise_msg = _(
306                                 "The file '%s' is missing a byte order "
307                                 "mark (BOM). Please use UTF-%sBE or UTF-%sLE "
308                                 "(depending on the byte order) as "
309                                 "working-tree-encoding.");
310                         advise(advise_msg, path, stripped, stripped);
311                         if (die_on_error)
312                                 die(error_msg, path, enc);
313                         else {
314                                 return error(error_msg, path, enc);
315                         }
316                 }
317
318         }
319         return 0;
320 }
321
322 static void trace_encoding(const char *context, const char *path,
323                            const char *encoding, const char *buf, size_t len)
324 {
325         static struct trace_key coe = TRACE_KEY_INIT(WORKING_TREE_ENCODING);
326         struct strbuf trace = STRBUF_INIT;
327         int i;
328
329         strbuf_addf(&trace, "%s (%s, considered %s):\n", context, path, encoding);
330         for (i = 0; i < len && buf; ++i) {
331                 strbuf_addf(
332                         &trace, "| \033[2m%2i:\033[0m %2x \033[2m%c\033[0m%c",
333                         i,
334                         (unsigned char) buf[i],
335                         (buf[i] > 32 && buf[i] < 127 ? buf[i] : ' '),
336                         ((i+1) % 8 && (i+1) < len ? ' ' : '\n')
337                 );
338         }
339         strbuf_addchars(&trace, '\n', 1);
340
341         trace_strbuf(&coe, &trace);
342         strbuf_release(&trace);
343 }
344
345 static int check_roundtrip(const char *enc_name)
346 {
347         /*
348          * check_roundtrip_encoding contains a string of comma and/or
349          * space separated encodings (eg. "UTF-16, ASCII, CP1125").
350          * Search for the given encoding in that string.
351          */
352         const char *found = strcasestr(check_roundtrip_encoding, enc_name);
353         const char *next;
354         int len;
355         if (!found)
356                 return 0;
357         next = found + strlen(enc_name);
358         len = strlen(check_roundtrip_encoding);
359         return (found && (
360                         /*
361                          * check that the found encoding is at the
362                          * beginning of check_roundtrip_encoding or
363                          * that it is prefixed with a space or comma
364                          */
365                         found == check_roundtrip_encoding || (
366                                 (isspace(found[-1]) || found[-1] == ',')
367                         )
368                 ) && (
369                         /*
370                          * check that the found encoding is at the
371                          * end of check_roundtrip_encoding or
372                          * that it is suffixed with a space or comma
373                          */
374                         next == check_roundtrip_encoding + len || (
375                                 next < check_roundtrip_encoding + len &&
376                                 (isspace(next[0]) || next[0] == ',')
377                         )
378                 ));
379 }
380
381 static const char *default_encoding = "UTF-8";
382
383 static int encode_to_git(const char *path, const char *src, size_t src_len,
384                          struct strbuf *buf, const char *enc, int conv_flags)
385 {
386         char *dst;
387         size_t dst_len;
388         int die_on_error = conv_flags & CONV_WRITE_OBJECT;
389
390         /*
391          * No encoding is specified or there is nothing to encode.
392          * Tell the caller that the content was not modified.
393          */
394         if (!enc || (src && !src_len))
395                 return 0;
396
397         /*
398          * Looks like we got called from "would_convert_to_git()".
399          * This means Git wants to know if it would encode (= modify!)
400          * the content. Let's answer with "yes", since an encoding was
401          * specified.
402          */
403         if (!buf && !src)
404                 return 1;
405
406         if (validate_encoding(path, enc, src, src_len, die_on_error))
407                 return 0;
408
409         trace_encoding("source", path, enc, src, src_len);
410         dst = reencode_string_len(src, src_len, default_encoding, enc,
411                                   &dst_len);
412         if (!dst) {
413                 /*
414                  * We could add the blob "as-is" to Git. However, on checkout
415                  * we would try to re-encode to the original encoding. This
416                  * would fail and we would leave the user with a messed-up
417                  * working tree. Let's try to avoid this by screaming loud.
418                  */
419                 const char* msg = _("failed to encode '%s' from %s to %s");
420                 if (die_on_error)
421                         die(msg, path, enc, default_encoding);
422                 else {
423                         error(msg, path, enc, default_encoding);
424                         return 0;
425                 }
426         }
427         trace_encoding("destination", path, default_encoding, dst, dst_len);
428
429         /*
430          * UTF supports lossless conversion round tripping [1] and conversions
431          * between UTF and other encodings are mostly round trip safe as
432          * Unicode aims to be a superset of all other character encodings.
433          * However, certain encodings (e.g. SHIFT-JIS) are known to have round
434          * trip issues [2]. Check the round trip conversion for all encodings
435          * listed in core.checkRoundtripEncoding.
436          *
437          * The round trip check is only performed if content is written to Git.
438          * This ensures that no information is lost during conversion to/from
439          * the internal UTF-8 representation.
440          *
441          * Please note, the code below is not tested because I was not able to
442          * generate a faulty round trip without an iconv error. Iconv errors
443          * are already caught above.
444          *
445          * [1] http://unicode.org/faq/utf_bom.html#gen2
446          * [2] https://support.microsoft.com/en-us/help/170559/prb-conversion-problem-between-shift-jis-and-unicode
447          */
448         if (die_on_error && check_roundtrip(enc)) {
449                 char *re_src;
450                 size_t re_src_len;
451
452                 re_src = reencode_string_len(dst, dst_len,
453                                              enc, default_encoding,
454                                              &re_src_len);
455
456                 trace_printf("Checking roundtrip encoding for %s...\n", enc);
457                 trace_encoding("reencoded source", path, enc,
458                                re_src, re_src_len);
459
460                 if (!re_src || src_len != re_src_len ||
461                     memcmp(src, re_src, src_len)) {
462                         const char* msg = _("encoding '%s' from %s to %s and "
463                                             "back is not the same");
464                         die(msg, path, enc, default_encoding);
465                 }
466
467                 free(re_src);
468         }
469
470         strbuf_attach(buf, dst, dst_len, dst_len + 1);
471         return 1;
472 }
473
474 static int encode_to_worktree(const char *path, const char *src, size_t src_len,
475                               struct strbuf *buf, const char *enc)
476 {
477         char *dst;
478         size_t dst_len;
479
480         /*
481          * No encoding is specified or there is nothing to encode.
482          * Tell the caller that the content was not modified.
483          */
484         if (!enc || (src && !src_len))
485                 return 0;
486
487         dst = reencode_string_len(src, src_len, enc, default_encoding,
488                                   &dst_len);
489         if (!dst) {
490                 error(_("failed to encode '%s' from %s to %s"),
491                       path, default_encoding, enc);
492                 return 0;
493         }
494
495         strbuf_attach(buf, dst, dst_len, dst_len + 1);
496         return 1;
497 }
498
499 static int crlf_to_git(const struct index_state *istate,
500                        const char *path, const char *src, size_t len,
501                        struct strbuf *buf,
502                        enum crlf_action crlf_action, int conv_flags)
503 {
504         struct text_stat stats;
505         char *dst;
506         int convert_crlf_into_lf;
507
508         if (crlf_action == CRLF_BINARY ||
509             (src && !len))
510                 return 0;
511
512         /*
513          * If we are doing a dry-run and have no source buffer, there is
514          * nothing to analyze; we must assume we would convert.
515          */
516         if (!buf && !src)
517                 return 1;
518
519         gather_stats(src, len, &stats);
520         /* Optimization: No CRLF? Nothing to convert, regardless. */
521         convert_crlf_into_lf = !!stats.crlf;
522
523         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
524                 if (convert_is_binary(&stats))
525                         return 0;
526                 /*
527                  * If the file in the index has any CR in it, do not
528                  * convert.  This is the new safer autocrlf handling,
529                  * unless we want to renormalize in a merge or
530                  * cherry-pick.
531                  */
532                 if ((!(conv_flags & CONV_EOL_RENORMALIZE)) &&
533                     has_crlf_in_index(istate, path))
534                         convert_crlf_into_lf = 0;
535         }
536         if (((conv_flags & CONV_EOL_RNDTRP_WARN) ||
537              ((conv_flags & CONV_EOL_RNDTRP_DIE) && len))) {
538                 struct text_stat new_stats;
539                 memcpy(&new_stats, &stats, sizeof(new_stats));
540                 /* simulate "git add" */
541                 if (convert_crlf_into_lf) {
542                         new_stats.lonelf += new_stats.crlf;
543                         new_stats.crlf = 0;
544                 }
545                 /* simulate "git checkout" */
546                 if (will_convert_lf_to_crlf(&new_stats, crlf_action)) {
547                         new_stats.crlf += new_stats.lonelf;
548                         new_stats.lonelf = 0;
549                 }
550                 check_global_conv_flags_eol(path, crlf_action, &stats, &new_stats, conv_flags);
551         }
552         if (!convert_crlf_into_lf)
553                 return 0;
554
555         /*
556          * At this point all of our source analysis is done, and we are sure we
557          * would convert. If we are in dry-run mode, we can give an answer.
558          */
559         if (!buf)
560                 return 1;
561
562         /* only grow if not in place */
563         if (strbuf_avail(buf) + buf->len < len)
564                 strbuf_grow(buf, len - buf->len);
565         dst = buf->buf;
566         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
567                 /*
568                  * If we guessed, we already know we rejected a file with
569                  * lone CR, and we can strip a CR without looking at what
570                  * follow it.
571                  */
572                 do {
573                         unsigned char c = *src++;
574                         if (c != '\r')
575                                 *dst++ = c;
576                 } while (--len);
577         } else {
578                 do {
579                         unsigned char c = *src++;
580                         if (! (c == '\r' && (1 < len && *src == '\n')))
581                                 *dst++ = c;
582                 } while (--len);
583         }
584         strbuf_setlen(buf, dst - buf->buf);
585         return 1;
586 }
587
588 static int crlf_to_worktree(const char *src, size_t len,
589                             struct strbuf *buf, enum crlf_action crlf_action)
590 {
591         char *to_free = NULL;
592         struct text_stat stats;
593
594         if (!len || output_eol(crlf_action) != EOL_CRLF)
595                 return 0;
596
597         gather_stats(src, len, &stats);
598         if (!will_convert_lf_to_crlf(&stats, crlf_action))
599                 return 0;
600
601         /* are we "faking" in place editing ? */
602         if (src == buf->buf)
603                 to_free = strbuf_detach(buf, NULL);
604
605         strbuf_grow(buf, len + stats.lonelf);
606         for (;;) {
607                 const char *nl = memchr(src, '\n', len);
608                 if (!nl)
609                         break;
610                 if (nl > src && nl[-1] == '\r') {
611                         strbuf_add(buf, src, nl + 1 - src);
612                 } else {
613                         strbuf_add(buf, src, nl - src);
614                         strbuf_addstr(buf, "\r\n");
615                 }
616                 len -= nl + 1 - src;
617                 src  = nl + 1;
618         }
619         strbuf_add(buf, src, len);
620
621         free(to_free);
622         return 1;
623 }
624
625 struct filter_params {
626         const char *src;
627         unsigned long size;
628         int fd;
629         const char *cmd;
630         const char *path;
631 };
632
633 static int filter_buffer_or_fd(int in, int out, void *data)
634 {
635         /*
636          * Spawn cmd and feed the buffer contents through its stdin.
637          */
638         struct child_process child_process = CHILD_PROCESS_INIT;
639         struct filter_params *params = (struct filter_params *)data;
640         int write_err, status;
641         const char *argv[] = { NULL, NULL };
642
643         /* apply % substitution to cmd */
644         struct strbuf cmd = STRBUF_INIT;
645         struct strbuf path = STRBUF_INIT;
646         struct strbuf_expand_dict_entry dict[] = {
647                 { "f", NULL, },
648                 { NULL, NULL, },
649         };
650
651         /* quote the path to preserve spaces, etc. */
652         sq_quote_buf(&path, params->path);
653         dict[0].value = path.buf;
654
655         /* expand all %f with the quoted path */
656         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
657         strbuf_release(&path);
658
659         argv[0] = cmd.buf;
660
661         child_process.argv = argv;
662         child_process.use_shell = 1;
663         child_process.in = -1;
664         child_process.out = out;
665
666         if (start_command(&child_process)) {
667                 strbuf_release(&cmd);
668                 return error(_("cannot fork to run external filter '%s'"),
669                              params->cmd);
670         }
671
672         sigchain_push(SIGPIPE, SIG_IGN);
673
674         if (params->src) {
675                 write_err = (write_in_full(child_process.in,
676                                            params->src, params->size) < 0);
677                 if (errno == EPIPE)
678                         write_err = 0;
679         } else {
680                 write_err = copy_fd(params->fd, child_process.in);
681                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
682                         write_err = 0;
683         }
684
685         if (close(child_process.in))
686                 write_err = 1;
687         if (write_err)
688                 error(_("cannot feed the input to external filter '%s'"),
689                       params->cmd);
690
691         sigchain_pop(SIGPIPE);
692
693         status = finish_command(&child_process);
694         if (status)
695                 error(_("external filter '%s' failed %d"), params->cmd, status);
696
697         strbuf_release(&cmd);
698         return (write_err || status);
699 }
700
701 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
702                                     struct strbuf *dst, const char *cmd)
703 {
704         /*
705          * Create a pipeline to have the command filter the buffer's
706          * contents.
707          *
708          * (child --> cmd) --> us
709          */
710         int err = 0;
711         struct strbuf nbuf = STRBUF_INIT;
712         struct async async;
713         struct filter_params params;
714
715         memset(&async, 0, sizeof(async));
716         async.proc = filter_buffer_or_fd;
717         async.data = &params;
718         async.out = -1;
719         params.src = src;
720         params.size = len;
721         params.fd = fd;
722         params.cmd = cmd;
723         params.path = path;
724
725         fflush(NULL);
726         if (start_async(&async))
727                 return 0;       /* error was already reported */
728
729         if (strbuf_read(&nbuf, async.out, 0) < 0) {
730                 err = error(_("read from external filter '%s' failed"), cmd);
731         }
732         if (close(async.out)) {
733                 err = error(_("read from external filter '%s' failed"), cmd);
734         }
735         if (finish_async(&async)) {
736                 err = error(_("external filter '%s' failed"), cmd);
737         }
738
739         if (!err) {
740                 strbuf_swap(dst, &nbuf);
741         }
742         strbuf_release(&nbuf);
743         return !err;
744 }
745
746 #define CAP_CLEAN    (1u<<0)
747 #define CAP_SMUDGE   (1u<<1)
748 #define CAP_DELAY    (1u<<2)
749
750 struct cmd2process {
751         struct subprocess_entry subprocess; /* must be the first member! */
752         unsigned int supported_capabilities;
753 };
754
755 static int subprocess_map_initialized;
756 static struct hashmap subprocess_map;
757
758 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
759 {
760         static int versions[] = {2, 0};
761         static struct subprocess_capability capabilities[] = {
762                 { "clean",  CAP_CLEAN  },
763                 { "smudge", CAP_SMUDGE },
764                 { "delay",  CAP_DELAY  },
765                 { NULL, 0 }
766         };
767         struct cmd2process *entry = (struct cmd2process *)subprocess;
768         return subprocess_handshake(subprocess, "git-filter", versions, NULL,
769                                     capabilities,
770                                     &entry->supported_capabilities);
771 }
772
773 static void handle_filter_error(const struct strbuf *filter_status,
774                                 struct cmd2process *entry,
775                                 const unsigned int wanted_capability)
776 {
777         if (!strcmp(filter_status->buf, "error"))
778                 ; /* The filter signaled a problem with the file. */
779         else if (!strcmp(filter_status->buf, "abort") && wanted_capability) {
780                 /*
781                  * The filter signaled a permanent problem. Don't try to filter
782                  * files with the same command for the lifetime of the current
783                  * Git process.
784                  */
785                  entry->supported_capabilities &= ~wanted_capability;
786         } else {
787                 /*
788                  * Something went wrong with the protocol filter.
789                  * Force shutdown and restart if another blob requires filtering.
790                  */
791                 error(_("external filter '%s' failed"), entry->subprocess.cmd);
792                 subprocess_stop(&subprocess_map, &entry->subprocess);
793                 free(entry);
794         }
795 }
796
797 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
798                                    int fd, struct strbuf *dst, const char *cmd,
799                                    const unsigned int wanted_capability,
800                                    struct delayed_checkout *dco)
801 {
802         int err;
803         int can_delay = 0;
804         struct cmd2process *entry;
805         struct child_process *process;
806         struct strbuf nbuf = STRBUF_INIT;
807         struct strbuf filter_status = STRBUF_INIT;
808         const char *filter_type;
809
810         if (!subprocess_map_initialized) {
811                 subprocess_map_initialized = 1;
812                 hashmap_init(&subprocess_map, cmd2process_cmp, NULL, 0);
813                 entry = NULL;
814         } else {
815                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
816         }
817
818         fflush(NULL);
819
820         if (!entry) {
821                 entry = xmalloc(sizeof(*entry));
822                 entry->supported_capabilities = 0;
823
824                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
825                         free(entry);
826                         return 0;
827                 }
828         }
829         process = &entry->subprocess.process;
830
831         if (!(entry->supported_capabilities & wanted_capability))
832                 return 0;
833
834         if (wanted_capability & CAP_CLEAN)
835                 filter_type = "clean";
836         else if (wanted_capability & CAP_SMUDGE)
837                 filter_type = "smudge";
838         else
839                 die(_("unexpected filter type"));
840
841         sigchain_push(SIGPIPE, SIG_IGN);
842
843         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
844         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
845         if (err)
846                 goto done;
847
848         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
849         if (err) {
850                 error(_("path name too long for external filter"));
851                 goto done;
852         }
853
854         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
855         if (err)
856                 goto done;
857
858         if ((entry->supported_capabilities & CAP_DELAY) &&
859             dco && dco->state == CE_CAN_DELAY) {
860                 can_delay = 1;
861                 err = packet_write_fmt_gently(process->in, "can-delay=1\n");
862                 if (err)
863                         goto done;
864         }
865
866         err = packet_flush_gently(process->in);
867         if (err)
868                 goto done;
869
870         if (fd >= 0)
871                 err = write_packetized_from_fd(fd, process->in);
872         else
873                 err = write_packetized_from_buf(src, len, process->in);
874         if (err)
875                 goto done;
876
877         err = subprocess_read_status(process->out, &filter_status);
878         if (err)
879                 goto done;
880
881         if (can_delay && !strcmp(filter_status.buf, "delayed")) {
882                 string_list_insert(&dco->filters, cmd);
883                 string_list_insert(&dco->paths, path);
884         } else {
885                 /* The filter got the blob and wants to send us a response. */
886                 err = strcmp(filter_status.buf, "success");
887                 if (err)
888                         goto done;
889
890                 err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
891                 if (err)
892                         goto done;
893
894                 err = subprocess_read_status(process->out, &filter_status);
895                 if (err)
896                         goto done;
897
898                 err = strcmp(filter_status.buf, "success");
899         }
900
901 done:
902         sigchain_pop(SIGPIPE);
903
904         if (err)
905                 handle_filter_error(&filter_status, entry, wanted_capability);
906         else
907                 strbuf_swap(dst, &nbuf);
908         strbuf_release(&nbuf);
909         return !err;
910 }
911
912
913 int async_query_available_blobs(const char *cmd, struct string_list *available_paths)
914 {
915         int err;
916         char *line;
917         struct cmd2process *entry;
918         struct child_process *process;
919         struct strbuf filter_status = STRBUF_INIT;
920
921         assert(subprocess_map_initialized);
922         entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
923         if (!entry) {
924                 error(_("external filter '%s' is not available anymore although "
925                         "not all paths have been filtered"), cmd);
926                 return 0;
927         }
928         process = &entry->subprocess.process;
929         sigchain_push(SIGPIPE, SIG_IGN);
930
931         err = packet_write_fmt_gently(
932                 process->in, "command=list_available_blobs\n");
933         if (err)
934                 goto done;
935
936         err = packet_flush_gently(process->in);
937         if (err)
938                 goto done;
939
940         while ((line = packet_read_line(process->out, NULL))) {
941                 const char *path;
942                 if (skip_prefix(line, "pathname=", &path))
943                         string_list_insert(available_paths, xstrdup(path));
944                 else
945                         ; /* ignore unknown keys */
946         }
947
948         err = subprocess_read_status(process->out, &filter_status);
949         if (err)
950                 goto done;
951
952         err = strcmp(filter_status.buf, "success");
953
954 done:
955         sigchain_pop(SIGPIPE);
956
957         if (err)
958                 handle_filter_error(&filter_status, entry, 0);
959         return !err;
960 }
961
962 static struct convert_driver {
963         const char *name;
964         struct convert_driver *next;
965         const char *smudge;
966         const char *clean;
967         const char *process;
968         int required;
969 } *user_convert, **user_convert_tail;
970
971 static int apply_filter(const char *path, const char *src, size_t len,
972                         int fd, struct strbuf *dst, struct convert_driver *drv,
973                         const unsigned int wanted_capability,
974                         struct delayed_checkout *dco)
975 {
976         const char *cmd = NULL;
977
978         if (!drv)
979                 return 0;
980
981         if (!dst)
982                 return 1;
983
984         if ((wanted_capability & CAP_CLEAN) && !drv->process && drv->clean)
985                 cmd = drv->clean;
986         else if ((wanted_capability & CAP_SMUDGE) && !drv->process && drv->smudge)
987                 cmd = drv->smudge;
988
989         if (cmd && *cmd)
990                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
991         else if (drv->process && *drv->process)
992                 return apply_multi_file_filter(path, src, len, fd, dst,
993                         drv->process, wanted_capability, dco);
994
995         return 0;
996 }
997
998 static int read_convert_config(const char *var, const char *value, void *cb)
999 {
1000         const char *key, *name;
1001         int namelen;
1002         struct convert_driver *drv;
1003
1004         /*
1005          * External conversion drivers are configured using
1006          * "filter.<name>.variable".
1007          */
1008         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
1009                 return 0;
1010         for (drv = user_convert; drv; drv = drv->next)
1011                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
1012                         break;
1013         if (!drv) {
1014                 drv = xcalloc(1, sizeof(struct convert_driver));
1015                 drv->name = xmemdupz(name, namelen);
1016                 *user_convert_tail = drv;
1017                 user_convert_tail = &(drv->next);
1018         }
1019
1020         /*
1021          * filter.<name>.smudge and filter.<name>.clean specifies
1022          * the command line:
1023          *
1024          *      command-line
1025          *
1026          * The command-line will not be interpolated in any way.
1027          */
1028
1029         if (!strcmp("smudge", key))
1030                 return git_config_string(&drv->smudge, var, value);
1031
1032         if (!strcmp("clean", key))
1033                 return git_config_string(&drv->clean, var, value);
1034
1035         if (!strcmp("process", key))
1036                 return git_config_string(&drv->process, var, value);
1037
1038         if (!strcmp("required", key)) {
1039                 drv->required = git_config_bool(var, value);
1040                 return 0;
1041         }
1042
1043         return 0;
1044 }
1045
1046 static int count_ident(const char *cp, unsigned long size)
1047 {
1048         /*
1049          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
1050          */
1051         int cnt = 0;
1052         char ch;
1053
1054         while (size) {
1055                 ch = *cp++;
1056                 size--;
1057                 if (ch != '$')
1058                         continue;
1059                 if (size < 3)
1060                         break;
1061                 if (memcmp("Id", cp, 2))
1062                         continue;
1063                 ch = cp[2];
1064                 cp += 3;
1065                 size -= 3;
1066                 if (ch == '$')
1067                         cnt++; /* $Id$ */
1068                 if (ch != ':')
1069                         continue;
1070
1071                 /*
1072                  * "$Id: ... "; scan up to the closing dollar sign and discard.
1073                  */
1074                 while (size) {
1075                         ch = *cp++;
1076                         size--;
1077                         if (ch == '$') {
1078                                 cnt++;
1079                                 break;
1080                         }
1081                         if (ch == '\n')
1082                                 break;
1083                 }
1084         }
1085         return cnt;
1086 }
1087
1088 static int ident_to_git(const char *src, size_t len,
1089                         struct strbuf *buf, int ident)
1090 {
1091         char *dst, *dollar;
1092
1093         if (!ident || (src && !count_ident(src, len)))
1094                 return 0;
1095
1096         if (!buf)
1097                 return 1;
1098
1099         /* only grow if not in place */
1100         if (strbuf_avail(buf) + buf->len < len)
1101                 strbuf_grow(buf, len - buf->len);
1102         dst = buf->buf;
1103         for (;;) {
1104                 dollar = memchr(src, '$', len);
1105                 if (!dollar)
1106                         break;
1107                 memmove(dst, src, dollar + 1 - src);
1108                 dst += dollar + 1 - src;
1109                 len -= dollar + 1 - src;
1110                 src  = dollar + 1;
1111
1112                 if (len > 3 && !memcmp(src, "Id:", 3)) {
1113                         dollar = memchr(src + 3, '$', len - 3);
1114                         if (!dollar)
1115                                 break;
1116                         if (memchr(src + 3, '\n', dollar - src - 3)) {
1117                                 /* Line break before the next dollar. */
1118                                 continue;
1119                         }
1120
1121                         memcpy(dst, "Id$", 3);
1122                         dst += 3;
1123                         len -= dollar + 1 - src;
1124                         src  = dollar + 1;
1125                 }
1126         }
1127         memmove(dst, src, len);
1128         strbuf_setlen(buf, dst + len - buf->buf);
1129         return 1;
1130 }
1131
1132 static int ident_to_worktree(const char *src, size_t len,
1133                              struct strbuf *buf, int ident)
1134 {
1135         struct object_id oid;
1136         char *to_free = NULL, *dollar, *spc;
1137         int cnt;
1138
1139         if (!ident)
1140                 return 0;
1141
1142         cnt = count_ident(src, len);
1143         if (!cnt)
1144                 return 0;
1145
1146         /* are we "faking" in place editing ? */
1147         if (src == buf->buf)
1148                 to_free = strbuf_detach(buf, NULL);
1149         hash_object_file(src, len, "blob", &oid);
1150
1151         strbuf_grow(buf, len + cnt * (the_hash_algo->hexsz + 3));
1152         for (;;) {
1153                 /* step 1: run to the next '$' */
1154                 dollar = memchr(src, '$', len);
1155                 if (!dollar)
1156                         break;
1157                 strbuf_add(buf, src, dollar + 1 - src);
1158                 len -= dollar + 1 - src;
1159                 src  = dollar + 1;
1160
1161                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
1162                 if (len < 3 || memcmp("Id", src, 2))
1163                         continue;
1164
1165                 /* step 3: skip over Id$ or Id:xxxxx$ */
1166                 if (src[2] == '$') {
1167                         src += 3;
1168                         len -= 3;
1169                 } else if (src[2] == ':') {
1170                         /*
1171                          * It's possible that an expanded Id has crept its way into the
1172                          * repository, we cope with that by stripping the expansion out.
1173                          * This is probably not a good idea, since it will cause changes
1174                          * on checkout, which won't go away by stash, but let's keep it
1175                          * for git-style ids.
1176                          */
1177                         dollar = memchr(src + 3, '$', len - 3);
1178                         if (!dollar) {
1179                                 /* incomplete keyword, no more '$', so just quit the loop */
1180                                 break;
1181                         }
1182
1183                         if (memchr(src + 3, '\n', dollar - src - 3)) {
1184                                 /* Line break before the next dollar. */
1185                                 continue;
1186                         }
1187
1188                         spc = memchr(src + 4, ' ', dollar - src - 4);
1189                         if (spc && spc < dollar-1) {
1190                                 /* There are spaces in unexpected places.
1191                                  * This is probably an id from some other
1192                                  * versioning system. Keep it for now.
1193                                  */
1194                                 continue;
1195                         }
1196
1197                         len -= dollar + 1 - src;
1198                         src  = dollar + 1;
1199                 } else {
1200                         /* it wasn't a "Id$" or "Id:xxxx$" */
1201                         continue;
1202                 }
1203
1204                 /* step 4: substitute */
1205                 strbuf_addstr(buf, "Id: ");
1206                 strbuf_addstr(buf, oid_to_hex(&oid));
1207                 strbuf_addstr(buf, " $");
1208         }
1209         strbuf_add(buf, src, len);
1210
1211         free(to_free);
1212         return 1;
1213 }
1214
1215 static const char *git_path_check_encoding(struct attr_check_item *check)
1216 {
1217         const char *value = check->value;
1218
1219         if (ATTR_UNSET(value) || !strlen(value))
1220                 return NULL;
1221
1222         if (ATTR_TRUE(value) || ATTR_FALSE(value)) {
1223                 die(_("true/false are no valid working-tree-encodings"));
1224         }
1225
1226         /* Don't encode to the default encoding */
1227         if (same_encoding(value, default_encoding))
1228                 return NULL;
1229
1230         return value;
1231 }
1232
1233 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
1234 {
1235         const char *value = check->value;
1236
1237         if (ATTR_TRUE(value))
1238                 return CRLF_TEXT;
1239         else if (ATTR_FALSE(value))
1240                 return CRLF_BINARY;
1241         else if (ATTR_UNSET(value))
1242                 ;
1243         else if (!strcmp(value, "input"))
1244                 return CRLF_TEXT_INPUT;
1245         else if (!strcmp(value, "auto"))
1246                 return CRLF_AUTO;
1247         return CRLF_UNDEFINED;
1248 }
1249
1250 static enum eol git_path_check_eol(struct attr_check_item *check)
1251 {
1252         const char *value = check->value;
1253
1254         if (ATTR_UNSET(value))
1255                 ;
1256         else if (!strcmp(value, "lf"))
1257                 return EOL_LF;
1258         else if (!strcmp(value, "crlf"))
1259                 return EOL_CRLF;
1260         return EOL_UNSET;
1261 }
1262
1263 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
1264 {
1265         const char *value = check->value;
1266         struct convert_driver *drv;
1267
1268         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
1269                 return NULL;
1270         for (drv = user_convert; drv; drv = drv->next)
1271                 if (!strcmp(value, drv->name))
1272                         return drv;
1273         return NULL;
1274 }
1275
1276 static int git_path_check_ident(struct attr_check_item *check)
1277 {
1278         const char *value = check->value;
1279
1280         return !!ATTR_TRUE(value);
1281 }
1282
1283 struct conv_attrs {
1284         struct convert_driver *drv;
1285         enum crlf_action attr_action; /* What attr says */
1286         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
1287         int ident;
1288         const char *working_tree_encoding; /* Supported encoding or default encoding if NULL */
1289 };
1290
1291 static struct attr_check *check;
1292
1293 static void convert_attrs(const struct index_state *istate,
1294                           struct conv_attrs *ca, const char *path)
1295 {
1296         struct attr_check_item *ccheck = NULL;
1297
1298         if (!check) {
1299                 check = attr_check_initl("crlf", "ident", "filter",
1300                                          "eol", "text", "working-tree-encoding",
1301                                          NULL);
1302                 user_convert_tail = &user_convert;
1303                 git_config(read_convert_config, NULL);
1304         }
1305
1306         git_check_attr(istate, path, check);
1307         ccheck = check->items;
1308         ca->crlf_action = git_path_check_crlf(ccheck + 4);
1309         if (ca->crlf_action == CRLF_UNDEFINED)
1310                 ca->crlf_action = git_path_check_crlf(ccheck + 0);
1311         ca->ident = git_path_check_ident(ccheck + 1);
1312         ca->drv = git_path_check_convert(ccheck + 2);
1313         if (ca->crlf_action != CRLF_BINARY) {
1314                 enum eol eol_attr = git_path_check_eol(ccheck + 3);
1315                 if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1316                         ca->crlf_action = CRLF_AUTO_INPUT;
1317                 else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1318                         ca->crlf_action = CRLF_AUTO_CRLF;
1319                 else if (eol_attr == EOL_LF)
1320                         ca->crlf_action = CRLF_TEXT_INPUT;
1321                 else if (eol_attr == EOL_CRLF)
1322                         ca->crlf_action = CRLF_TEXT_CRLF;
1323         }
1324         ca->working_tree_encoding = git_path_check_encoding(ccheck + 5);
1325
1326         /* Save attr and make a decision for action */
1327         ca->attr_action = ca->crlf_action;
1328         if (ca->crlf_action == CRLF_TEXT)
1329                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1330         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1331                 ca->crlf_action = CRLF_BINARY;
1332         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1333                 ca->crlf_action = CRLF_AUTO_CRLF;
1334         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1335                 ca->crlf_action = CRLF_AUTO_INPUT;
1336 }
1337
1338 void reset_parsed_attributes(void)
1339 {
1340         struct convert_driver *drv, *next;
1341
1342         attr_check_free(check);
1343         check = NULL;
1344         reset_merge_attributes();
1345
1346         for (drv = user_convert; drv; drv = next) {
1347                 next = drv->next;
1348                 free((void *)drv->name);
1349                 free(drv);
1350         }
1351         user_convert = NULL;
1352         user_convert_tail = NULL;
1353 }
1354
1355 int would_convert_to_git_filter_fd(const struct index_state *istate, const char *path)
1356 {
1357         struct conv_attrs ca;
1358
1359         convert_attrs(istate, &ca, path);
1360         if (!ca.drv)
1361                 return 0;
1362
1363         /*
1364          * Apply a filter to an fd only if the filter is required to succeed.
1365          * We must die if the filter fails, because the original data before
1366          * filtering is not available.
1367          */
1368         if (!ca.drv->required)
1369                 return 0;
1370
1371         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN, NULL);
1372 }
1373
1374 const char *get_convert_attr_ascii(const struct index_state *istate, const char *path)
1375 {
1376         struct conv_attrs ca;
1377
1378         convert_attrs(istate, &ca, path);
1379         switch (ca.attr_action) {
1380         case CRLF_UNDEFINED:
1381                 return "";
1382         case CRLF_BINARY:
1383                 return "-text";
1384         case CRLF_TEXT:
1385                 return "text";
1386         case CRLF_TEXT_INPUT:
1387                 return "text eol=lf";
1388         case CRLF_TEXT_CRLF:
1389                 return "text eol=crlf";
1390         case CRLF_AUTO:
1391                 return "text=auto";
1392         case CRLF_AUTO_CRLF:
1393                 return "text=auto eol=crlf";
1394         case CRLF_AUTO_INPUT:
1395                 return "text=auto eol=lf";
1396         }
1397         return "";
1398 }
1399
1400 int convert_to_git(const struct index_state *istate,
1401                    const char *path, const char *src, size_t len,
1402                    struct strbuf *dst, int conv_flags)
1403 {
1404         int ret = 0;
1405         struct conv_attrs ca;
1406
1407         convert_attrs(istate, &ca, path);
1408
1409         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN, NULL);
1410         if (!ret && ca.drv && ca.drv->required)
1411                 die(_("%s: clean filter '%s' failed"), path, ca.drv->name);
1412
1413         if (ret && dst) {
1414                 src = dst->buf;
1415                 len = dst->len;
1416         }
1417
1418         ret |= encode_to_git(path, src, len, dst, ca.working_tree_encoding, conv_flags);
1419         if (ret && dst) {
1420                 src = dst->buf;
1421                 len = dst->len;
1422         }
1423
1424         if (!(conv_flags & CONV_EOL_KEEP_CRLF)) {
1425                 ret |= crlf_to_git(istate, path, src, len, dst, ca.crlf_action, conv_flags);
1426                 if (ret && dst) {
1427                         src = dst->buf;
1428                         len = dst->len;
1429                 }
1430         }
1431         return ret | ident_to_git(src, len, dst, ca.ident);
1432 }
1433
1434 void convert_to_git_filter_fd(const struct index_state *istate,
1435                               const char *path, int fd, struct strbuf *dst,
1436                               int conv_flags)
1437 {
1438         struct conv_attrs ca;
1439         convert_attrs(istate, &ca, path);
1440
1441         assert(ca.drv);
1442         assert(ca.drv->clean || ca.drv->process);
1443
1444         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN, NULL))
1445                 die(_("%s: clean filter '%s' failed"), path, ca.drv->name);
1446
1447         encode_to_git(path, dst->buf, dst->len, dst, ca.working_tree_encoding, conv_flags);
1448         crlf_to_git(istate, path, dst->buf, dst->len, dst, ca.crlf_action, conv_flags);
1449         ident_to_git(dst->buf, dst->len, dst, ca.ident);
1450 }
1451
1452 static int convert_to_working_tree_internal(const struct index_state *istate,
1453                                             const char *path, const char *src,
1454                                             size_t len, struct strbuf *dst,
1455                                             int normalizing, struct delayed_checkout *dco)
1456 {
1457         int ret = 0, ret_filter = 0;
1458         struct conv_attrs ca;
1459
1460         convert_attrs(istate, &ca, path);
1461
1462         ret |= ident_to_worktree(src, len, dst, ca.ident);
1463         if (ret) {
1464                 src = dst->buf;
1465                 len = dst->len;
1466         }
1467         /*
1468          * CRLF conversion can be skipped if normalizing, unless there
1469          * is a smudge or process filter (even if the process filter doesn't
1470          * support smudge).  The filters might expect CRLFs.
1471          */
1472         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1473                 ret |= crlf_to_worktree(src, len, dst, ca.crlf_action);
1474                 if (ret) {
1475                         src = dst->buf;
1476                         len = dst->len;
1477                 }
1478         }
1479
1480         ret |= encode_to_worktree(path, src, len, dst, ca.working_tree_encoding);
1481         if (ret) {
1482                 src = dst->buf;
1483                 len = dst->len;
1484         }
1485
1486         ret_filter = apply_filter(
1487                 path, src, len, -1, dst, ca.drv, CAP_SMUDGE, dco);
1488         if (!ret_filter && ca.drv && ca.drv->required)
1489                 die(_("%s: smudge filter %s failed"), path, ca.drv->name);
1490
1491         return ret | ret_filter;
1492 }
1493
1494 int async_convert_to_working_tree(const struct index_state *istate,
1495                                   const char *path, const char *src,
1496                                   size_t len, struct strbuf *dst,
1497                                   void *dco)
1498 {
1499         return convert_to_working_tree_internal(istate, path, src, len, dst, 0, dco);
1500 }
1501
1502 int convert_to_working_tree(const struct index_state *istate,
1503                             const char *path, const char *src,
1504                             size_t len, struct strbuf *dst)
1505 {
1506         return convert_to_working_tree_internal(istate, path, src, len, dst, 0, NULL);
1507 }
1508
1509 int renormalize_buffer(const struct index_state *istate, const char *path,
1510                        const char *src, size_t len, struct strbuf *dst)
1511 {
1512         int ret = convert_to_working_tree_internal(istate, path, src, len, dst, 1, NULL);
1513         if (ret) {
1514                 src = dst->buf;
1515                 len = dst->len;
1516         }
1517         return ret | convert_to_git(istate, path, src, len, dst, CONV_EOL_RENORMALIZE);
1518 }
1519
1520 /*****************************************************************
1521  *
1522  * Streaming conversion support
1523  *
1524  *****************************************************************/
1525
1526 typedef int (*filter_fn)(struct stream_filter *,
1527                          const char *input, size_t *isize_p,
1528                          char *output, size_t *osize_p);
1529 typedef void (*free_fn)(struct stream_filter *);
1530
1531 struct stream_filter_vtbl {
1532         filter_fn filter;
1533         free_fn free;
1534 };
1535
1536 struct stream_filter {
1537         struct stream_filter_vtbl *vtbl;
1538 };
1539
1540 static int null_filter_fn(struct stream_filter *filter,
1541                           const char *input, size_t *isize_p,
1542                           char *output, size_t *osize_p)
1543 {
1544         size_t count;
1545
1546         if (!input)
1547                 return 0; /* we do not keep any states */
1548         count = *isize_p;
1549         if (*osize_p < count)
1550                 count = *osize_p;
1551         if (count) {
1552                 memmove(output, input, count);
1553                 *isize_p -= count;
1554                 *osize_p -= count;
1555         }
1556         return 0;
1557 }
1558
1559 static void null_free_fn(struct stream_filter *filter)
1560 {
1561         ; /* nothing -- null instances are shared */
1562 }
1563
1564 static struct stream_filter_vtbl null_vtbl = {
1565         null_filter_fn,
1566         null_free_fn,
1567 };
1568
1569 static struct stream_filter null_filter_singleton = {
1570         &null_vtbl,
1571 };
1572
1573 int is_null_stream_filter(struct stream_filter *filter)
1574 {
1575         return filter == &null_filter_singleton;
1576 }
1577
1578
1579 /*
1580  * LF-to-CRLF filter
1581  */
1582
1583 struct lf_to_crlf_filter {
1584         struct stream_filter filter;
1585         unsigned has_held:1;
1586         char held;
1587 };
1588
1589 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1590                                 const char *input, size_t *isize_p,
1591                                 char *output, size_t *osize_p)
1592 {
1593         size_t count, o = 0;
1594         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1595
1596         /*
1597          * We may be holding onto the CR to see if it is followed by a
1598          * LF, in which case we would need to go to the main loop.
1599          * Otherwise, just emit it to the output stream.
1600          */
1601         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1602                 output[o++] = lf_to_crlf->held;
1603                 lf_to_crlf->has_held = 0;
1604         }
1605
1606         /* We are told to drain */
1607         if (!input) {
1608                 *osize_p -= o;
1609                 return 0;
1610         }
1611
1612         count = *isize_p;
1613         if (count || lf_to_crlf->has_held) {
1614                 size_t i;
1615                 int was_cr = 0;
1616
1617                 if (lf_to_crlf->has_held) {
1618                         was_cr = 1;
1619                         lf_to_crlf->has_held = 0;
1620                 }
1621
1622                 for (i = 0; o < *osize_p && i < count; i++) {
1623                         char ch = input[i];
1624
1625                         if (ch == '\n') {
1626                                 output[o++] = '\r';
1627                         } else if (was_cr) {
1628                                 /*
1629                                  * Previous round saw CR and it is not followed
1630                                  * by a LF; emit the CR before processing the
1631                                  * current character.
1632                                  */
1633                                 output[o++] = '\r';
1634                         }
1635
1636                         /*
1637                          * We may have consumed the last output slot,
1638                          * in which case we need to break out of this
1639                          * loop; hold the current character before
1640                          * returning.
1641                          */
1642                         if (*osize_p <= o) {
1643                                 lf_to_crlf->has_held = 1;
1644                                 lf_to_crlf->held = ch;
1645                                 continue; /* break but increment i */
1646                         }
1647
1648                         if (ch == '\r') {
1649                                 was_cr = 1;
1650                                 continue;
1651                         }
1652
1653                         was_cr = 0;
1654                         output[o++] = ch;
1655                 }
1656
1657                 *osize_p -= o;
1658                 *isize_p -= i;
1659
1660                 if (!lf_to_crlf->has_held && was_cr) {
1661                         lf_to_crlf->has_held = 1;
1662                         lf_to_crlf->held = '\r';
1663                 }
1664         }
1665         return 0;
1666 }
1667
1668 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1669 {
1670         free(filter);
1671 }
1672
1673 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1674         lf_to_crlf_filter_fn,
1675         lf_to_crlf_free_fn,
1676 };
1677
1678 static struct stream_filter *lf_to_crlf_filter(void)
1679 {
1680         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1681
1682         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1683         return (struct stream_filter *)lf_to_crlf;
1684 }
1685
1686 /*
1687  * Cascade filter
1688  */
1689 #define FILTER_BUFFER 1024
1690 struct cascade_filter {
1691         struct stream_filter filter;
1692         struct stream_filter *one;
1693         struct stream_filter *two;
1694         char buf[FILTER_BUFFER];
1695         int end, ptr;
1696 };
1697
1698 static int cascade_filter_fn(struct stream_filter *filter,
1699                              const char *input, size_t *isize_p,
1700                              char *output, size_t *osize_p)
1701 {
1702         struct cascade_filter *cas = (struct cascade_filter *) filter;
1703         size_t filled = 0;
1704         size_t sz = *osize_p;
1705         size_t to_feed, remaining;
1706
1707         /*
1708          * input -- (one) --> buf -- (two) --> output
1709          */
1710         while (filled < sz) {
1711                 remaining = sz - filled;
1712
1713                 /* do we already have something to feed two with? */
1714                 if (cas->ptr < cas->end) {
1715                         to_feed = cas->end - cas->ptr;
1716                         if (stream_filter(cas->two,
1717                                           cas->buf + cas->ptr, &to_feed,
1718                                           output + filled, &remaining))
1719                                 return -1;
1720                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1721                         filled = sz - remaining;
1722                         continue;
1723                 }
1724
1725                 /* feed one from upstream and have it emit into our buffer */
1726                 to_feed = input ? *isize_p : 0;
1727                 if (input && !to_feed)
1728                         break;
1729                 remaining = sizeof(cas->buf);
1730                 if (stream_filter(cas->one,
1731                                   input, &to_feed,
1732                                   cas->buf, &remaining))
1733                         return -1;
1734                 cas->end = sizeof(cas->buf) - remaining;
1735                 cas->ptr = 0;
1736                 if (input) {
1737                         size_t fed = *isize_p - to_feed;
1738                         *isize_p -= fed;
1739                         input += fed;
1740                 }
1741
1742                 /* do we know that we drained one completely? */
1743                 if (input || cas->end)
1744                         continue;
1745
1746                 /* tell two to drain; we have nothing more to give it */
1747                 to_feed = 0;
1748                 remaining = sz - filled;
1749                 if (stream_filter(cas->two,
1750                                   NULL, &to_feed,
1751                                   output + filled, &remaining))
1752                         return -1;
1753                 if (remaining == (sz - filled))
1754                         break; /* completely drained two */
1755                 filled = sz - remaining;
1756         }
1757         *osize_p -= filled;
1758         return 0;
1759 }
1760
1761 static void cascade_free_fn(struct stream_filter *filter)
1762 {
1763         struct cascade_filter *cas = (struct cascade_filter *)filter;
1764         free_stream_filter(cas->one);
1765         free_stream_filter(cas->two);
1766         free(filter);
1767 }
1768
1769 static struct stream_filter_vtbl cascade_vtbl = {
1770         cascade_filter_fn,
1771         cascade_free_fn,
1772 };
1773
1774 static struct stream_filter *cascade_filter(struct stream_filter *one,
1775                                             struct stream_filter *two)
1776 {
1777         struct cascade_filter *cascade;
1778
1779         if (!one || is_null_stream_filter(one))
1780                 return two;
1781         if (!two || is_null_stream_filter(two))
1782                 return one;
1783
1784         cascade = xmalloc(sizeof(*cascade));
1785         cascade->one = one;
1786         cascade->two = two;
1787         cascade->end = cascade->ptr = 0;
1788         cascade->filter.vtbl = &cascade_vtbl;
1789         return (struct stream_filter *)cascade;
1790 }
1791
1792 /*
1793  * ident filter
1794  */
1795 #define IDENT_DRAINING (-1)
1796 #define IDENT_SKIPPING (-2)
1797 struct ident_filter {
1798         struct stream_filter filter;
1799         struct strbuf left;
1800         int state;
1801         char ident[GIT_MAX_HEXSZ + 5]; /* ": x40 $" */
1802 };
1803
1804 static int is_foreign_ident(const char *str)
1805 {
1806         int i;
1807
1808         if (!skip_prefix(str, "$Id: ", &str))
1809                 return 0;
1810         for (i = 0; str[i]; i++) {
1811                 if (isspace(str[i]) && str[i+1] != '$')
1812                         return 1;
1813         }
1814         return 0;
1815 }
1816
1817 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1818 {
1819         size_t to_drain = ident->left.len;
1820
1821         if (*osize_p < to_drain)
1822                 to_drain = *osize_p;
1823         if (to_drain) {
1824                 memcpy(*output_p, ident->left.buf, to_drain);
1825                 strbuf_remove(&ident->left, 0, to_drain);
1826                 *output_p += to_drain;
1827                 *osize_p -= to_drain;
1828         }
1829         if (!ident->left.len)
1830                 ident->state = 0;
1831 }
1832
1833 static int ident_filter_fn(struct stream_filter *filter,
1834                            const char *input, size_t *isize_p,
1835                            char *output, size_t *osize_p)
1836 {
1837         struct ident_filter *ident = (struct ident_filter *)filter;
1838         static const char head[] = "$Id";
1839
1840         if (!input) {
1841                 /* drain upon eof */
1842                 switch (ident->state) {
1843                 default:
1844                         strbuf_add(&ident->left, head, ident->state);
1845                         /* fallthrough */
1846                 case IDENT_SKIPPING:
1847                         /* fallthrough */
1848                 case IDENT_DRAINING:
1849                         ident_drain(ident, &output, osize_p);
1850                 }
1851                 return 0;
1852         }
1853
1854         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1855                 int ch;
1856
1857                 if (ident->state == IDENT_DRAINING) {
1858                         ident_drain(ident, &output, osize_p);
1859                         if (!*osize_p)
1860                                 break;
1861                         continue;
1862                 }
1863
1864                 ch = *(input++);
1865                 (*isize_p)--;
1866
1867                 if (ident->state == IDENT_SKIPPING) {
1868                         /*
1869                          * Skipping until '$' or LF, but keeping them
1870                          * in case it is a foreign ident.
1871                          */
1872                         strbuf_addch(&ident->left, ch);
1873                         if (ch != '\n' && ch != '$')
1874                                 continue;
1875                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1876                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1877                                 strbuf_addstr(&ident->left, ident->ident);
1878                         }
1879                         ident->state = IDENT_DRAINING;
1880                         continue;
1881                 }
1882
1883                 if (ident->state < sizeof(head) &&
1884                     head[ident->state] == ch) {
1885                         ident->state++;
1886                         continue;
1887                 }
1888
1889                 if (ident->state)
1890                         strbuf_add(&ident->left, head, ident->state);
1891                 if (ident->state == sizeof(head) - 1) {
1892                         if (ch != ':' && ch != '$') {
1893                                 strbuf_addch(&ident->left, ch);
1894                                 ident->state = 0;
1895                                 continue;
1896                         }
1897
1898                         if (ch == ':') {
1899                                 strbuf_addch(&ident->left, ch);
1900                                 ident->state = IDENT_SKIPPING;
1901                         } else {
1902                                 strbuf_addstr(&ident->left, ident->ident);
1903                                 ident->state = IDENT_DRAINING;
1904                         }
1905                         continue;
1906                 }
1907
1908                 strbuf_addch(&ident->left, ch);
1909                 ident->state = IDENT_DRAINING;
1910         }
1911         return 0;
1912 }
1913
1914 static void ident_free_fn(struct stream_filter *filter)
1915 {
1916         struct ident_filter *ident = (struct ident_filter *)filter;
1917         strbuf_release(&ident->left);
1918         free(filter);
1919 }
1920
1921 static struct stream_filter_vtbl ident_vtbl = {
1922         ident_filter_fn,
1923         ident_free_fn,
1924 };
1925
1926 static struct stream_filter *ident_filter(const struct object_id *oid)
1927 {
1928         struct ident_filter *ident = xmalloc(sizeof(*ident));
1929
1930         xsnprintf(ident->ident, sizeof(ident->ident),
1931                   ": %s $", oid_to_hex(oid));
1932         strbuf_init(&ident->left, 0);
1933         ident->filter.vtbl = &ident_vtbl;
1934         ident->state = 0;
1935         return (struct stream_filter *)ident;
1936 }
1937
1938 /*
1939  * Return an appropriately constructed filter for the path, or NULL if
1940  * the contents cannot be filtered without reading the whole thing
1941  * in-core.
1942  *
1943  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1944  * large binary blob you would want us not to slurp into the memory!
1945  */
1946 struct stream_filter *get_stream_filter(const struct index_state *istate,
1947                                         const char *path,
1948                                         const struct object_id *oid)
1949 {
1950         struct conv_attrs ca;
1951         struct stream_filter *filter = NULL;
1952
1953         convert_attrs(istate, &ca, path);
1954         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1955                 return NULL;
1956
1957         if (ca.working_tree_encoding)
1958                 return NULL;
1959
1960         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1961                 return NULL;
1962
1963         if (ca.ident)
1964                 filter = ident_filter(oid);
1965
1966         if (output_eol(ca.crlf_action) == EOL_CRLF)
1967                 filter = cascade_filter(filter, lf_to_crlf_filter());
1968         else
1969                 filter = cascade_filter(filter, &null_filter_singleton);
1970
1971         return filter;
1972 }
1973
1974 void free_stream_filter(struct stream_filter *filter)
1975 {
1976         filter->vtbl->free(filter);
1977 }
1978
1979 int stream_filter(struct stream_filter *filter,
1980                   const char *input, size_t *isize_p,
1981                   char *output, size_t *osize_p)
1982 {
1983         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1984 }