Merge branch 'jk/filter-branch-sha256'
[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,
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, &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
642         /* apply % substitution to cmd */
643         struct strbuf cmd = STRBUF_INIT;
644         struct strbuf path = STRBUF_INIT;
645         struct strbuf_expand_dict_entry dict[] = {
646                 { "f", NULL, },
647                 { NULL, NULL, },
648         };
649
650         /* quote the path to preserve spaces, etc. */
651         sq_quote_buf(&path, params->path);
652         dict[0].value = path.buf;
653
654         /* expand all %f with the quoted path */
655         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
656         strbuf_release(&path);
657
658         strvec_push(&child_process.args, cmd.buf);
659         child_process.use_shell = 1;
660         child_process.in = -1;
661         child_process.out = out;
662
663         if (start_command(&child_process)) {
664                 strbuf_release(&cmd);
665                 return error(_("cannot fork to run external filter '%s'"),
666                              params->cmd);
667         }
668
669         sigchain_push(SIGPIPE, SIG_IGN);
670
671         if (params->src) {
672                 write_err = (write_in_full(child_process.in,
673                                            params->src, params->size) < 0);
674                 if (errno == EPIPE)
675                         write_err = 0;
676         } else {
677                 write_err = copy_fd(params->fd, child_process.in);
678                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
679                         write_err = 0;
680         }
681
682         if (close(child_process.in))
683                 write_err = 1;
684         if (write_err)
685                 error(_("cannot feed the input to external filter '%s'"),
686                       params->cmd);
687
688         sigchain_pop(SIGPIPE);
689
690         status = finish_command(&child_process);
691         if (status)
692                 error(_("external filter '%s' failed %d"), params->cmd, status);
693
694         strbuf_release(&cmd);
695         return (write_err || status);
696 }
697
698 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
699                                     struct strbuf *dst, const char *cmd)
700 {
701         /*
702          * Create a pipeline to have the command filter the buffer's
703          * contents.
704          *
705          * (child --> cmd) --> us
706          */
707         int err = 0;
708         struct strbuf nbuf = STRBUF_INIT;
709         struct async async;
710         struct filter_params params;
711
712         memset(&async, 0, sizeof(async));
713         async.proc = filter_buffer_or_fd;
714         async.data = &params;
715         async.out = -1;
716         params.src = src;
717         params.size = len;
718         params.fd = fd;
719         params.cmd = cmd;
720         params.path = path;
721
722         fflush(NULL);
723         if (start_async(&async))
724                 return 0;       /* error was already reported */
725
726         if (strbuf_read(&nbuf, async.out, 0) < 0) {
727                 err = error(_("read from external filter '%s' failed"), cmd);
728         }
729         if (close(async.out)) {
730                 err = error(_("read from external filter '%s' failed"), cmd);
731         }
732         if (finish_async(&async)) {
733                 err = error(_("external filter '%s' failed"), cmd);
734         }
735
736         if (!err) {
737                 strbuf_swap(dst, &nbuf);
738         }
739         strbuf_release(&nbuf);
740         return !err;
741 }
742
743 #define CAP_CLEAN    (1u<<0)
744 #define CAP_SMUDGE   (1u<<1)
745 #define CAP_DELAY    (1u<<2)
746
747 struct cmd2process {
748         struct subprocess_entry subprocess; /* must be the first member! */
749         unsigned int supported_capabilities;
750 };
751
752 static int subprocess_map_initialized;
753 static struct hashmap subprocess_map;
754
755 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
756 {
757         static int versions[] = {2, 0};
758         static struct subprocess_capability capabilities[] = {
759                 { "clean",  CAP_CLEAN  },
760                 { "smudge", CAP_SMUDGE },
761                 { "delay",  CAP_DELAY  },
762                 { NULL, 0 }
763         };
764         struct cmd2process *entry = (struct cmd2process *)subprocess;
765         return subprocess_handshake(subprocess, "git-filter", versions, NULL,
766                                     capabilities,
767                                     &entry->supported_capabilities);
768 }
769
770 static void handle_filter_error(const struct strbuf *filter_status,
771                                 struct cmd2process *entry,
772                                 const unsigned int wanted_capability)
773 {
774         if (!strcmp(filter_status->buf, "error"))
775                 ; /* The filter signaled a problem with the file. */
776         else if (!strcmp(filter_status->buf, "abort") && wanted_capability) {
777                 /*
778                  * The filter signaled a permanent problem. Don't try to filter
779                  * files with the same command for the lifetime of the current
780                  * Git process.
781                  */
782                  entry->supported_capabilities &= ~wanted_capability;
783         } else {
784                 /*
785                  * Something went wrong with the protocol filter.
786                  * Force shutdown and restart if another blob requires filtering.
787                  */
788                 error(_("external filter '%s' failed"), entry->subprocess.cmd);
789                 subprocess_stop(&subprocess_map, &entry->subprocess);
790                 free(entry);
791         }
792 }
793
794 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
795                                    int fd, struct strbuf *dst, const char *cmd,
796                                    const unsigned int wanted_capability,
797                                    const struct checkout_metadata *meta,
798                                    struct delayed_checkout *dco)
799 {
800         int err;
801         int can_delay = 0;
802         struct cmd2process *entry;
803         struct child_process *process;
804         struct strbuf nbuf = STRBUF_INIT;
805         struct strbuf filter_status = STRBUF_INIT;
806         const char *filter_type;
807
808         if (!subprocess_map_initialized) {
809                 subprocess_map_initialized = 1;
810                 hashmap_init(&subprocess_map, cmd2process_cmp, NULL, 0);
811                 entry = NULL;
812         } else {
813                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
814         }
815
816         fflush(NULL);
817
818         if (!entry) {
819                 entry = xmalloc(sizeof(*entry));
820                 entry->supported_capabilities = 0;
821
822                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
823                         free(entry);
824                         return 0;
825                 }
826         }
827         process = &entry->subprocess.process;
828
829         if (!(entry->supported_capabilities & wanted_capability))
830                 return 0;
831
832         if (wanted_capability & CAP_CLEAN)
833                 filter_type = "clean";
834         else if (wanted_capability & CAP_SMUDGE)
835                 filter_type = "smudge";
836         else
837                 die(_("unexpected filter type"));
838
839         sigchain_push(SIGPIPE, SIG_IGN);
840
841         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
842         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
843         if (err)
844                 goto done;
845
846         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
847         if (err) {
848                 error(_("path name too long for external filter"));
849                 goto done;
850         }
851
852         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
853         if (err)
854                 goto done;
855
856         if (meta && meta->refname) {
857                 err = packet_write_fmt_gently(process->in, "ref=%s\n", meta->refname);
858                 if (err)
859                         goto done;
860         }
861
862         if (meta && !is_null_oid(&meta->treeish)) {
863                 err = packet_write_fmt_gently(process->in, "treeish=%s\n", oid_to_hex(&meta->treeish));
864                 if (err)
865                         goto done;
866         }
867
868         if (meta && !is_null_oid(&meta->blob)) {
869                 err = packet_write_fmt_gently(process->in, "blob=%s\n", oid_to_hex(&meta->blob));
870                 if (err)
871                         goto done;
872         }
873
874         if ((entry->supported_capabilities & CAP_DELAY) &&
875             dco && dco->state == CE_CAN_DELAY) {
876                 can_delay = 1;
877                 err = packet_write_fmt_gently(process->in, "can-delay=1\n");
878                 if (err)
879                         goto done;
880         }
881
882         err = packet_flush_gently(process->in);
883         if (err)
884                 goto done;
885
886         if (fd >= 0)
887                 err = write_packetized_from_fd(fd, process->in);
888         else
889                 err = write_packetized_from_buf(src, len, process->in);
890         if (err)
891                 goto done;
892
893         err = subprocess_read_status(process->out, &filter_status);
894         if (err)
895                 goto done;
896
897         if (can_delay && !strcmp(filter_status.buf, "delayed")) {
898                 string_list_insert(&dco->filters, cmd);
899                 string_list_insert(&dco->paths, path);
900         } else {
901                 /* The filter got the blob and wants to send us a response. */
902                 err = strcmp(filter_status.buf, "success");
903                 if (err)
904                         goto done;
905
906                 err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
907                 if (err)
908                         goto done;
909
910                 err = subprocess_read_status(process->out, &filter_status);
911                 if (err)
912                         goto done;
913
914                 err = strcmp(filter_status.buf, "success");
915         }
916
917 done:
918         sigchain_pop(SIGPIPE);
919
920         if (err)
921                 handle_filter_error(&filter_status, entry, wanted_capability);
922         else
923                 strbuf_swap(dst, &nbuf);
924         strbuf_release(&nbuf);
925         return !err;
926 }
927
928
929 int async_query_available_blobs(const char *cmd, struct string_list *available_paths)
930 {
931         int err;
932         char *line;
933         struct cmd2process *entry;
934         struct child_process *process;
935         struct strbuf filter_status = STRBUF_INIT;
936
937         assert(subprocess_map_initialized);
938         entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
939         if (!entry) {
940                 error(_("external filter '%s' is not available anymore although "
941                         "not all paths have been filtered"), cmd);
942                 return 0;
943         }
944         process = &entry->subprocess.process;
945         sigchain_push(SIGPIPE, SIG_IGN);
946
947         err = packet_write_fmt_gently(
948                 process->in, "command=list_available_blobs\n");
949         if (err)
950                 goto done;
951
952         err = packet_flush_gently(process->in);
953         if (err)
954                 goto done;
955
956         while ((line = packet_read_line(process->out, NULL))) {
957                 const char *path;
958                 if (skip_prefix(line, "pathname=", &path))
959                         string_list_insert(available_paths, xstrdup(path));
960                 else
961                         ; /* ignore unknown keys */
962         }
963
964         err = subprocess_read_status(process->out, &filter_status);
965         if (err)
966                 goto done;
967
968         err = strcmp(filter_status.buf, "success");
969
970 done:
971         sigchain_pop(SIGPIPE);
972
973         if (err)
974                 handle_filter_error(&filter_status, entry, 0);
975         return !err;
976 }
977
978 static struct convert_driver {
979         const char *name;
980         struct convert_driver *next;
981         const char *smudge;
982         const char *clean;
983         const char *process;
984         int required;
985 } *user_convert, **user_convert_tail;
986
987 static int apply_filter(const char *path, const char *src, size_t len,
988                         int fd, struct strbuf *dst, struct convert_driver *drv,
989                         const unsigned int wanted_capability,
990                         const struct checkout_metadata *meta,
991                         struct delayed_checkout *dco)
992 {
993         const char *cmd = NULL;
994
995         if (!drv)
996                 return 0;
997
998         if (!dst)
999                 return 1;
1000
1001         if ((wanted_capability & CAP_CLEAN) && !drv->process && drv->clean)
1002                 cmd = drv->clean;
1003         else if ((wanted_capability & CAP_SMUDGE) && !drv->process && drv->smudge)
1004                 cmd = drv->smudge;
1005
1006         if (cmd && *cmd)
1007                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
1008         else if (drv->process && *drv->process)
1009                 return apply_multi_file_filter(path, src, len, fd, dst,
1010                         drv->process, wanted_capability, meta, dco);
1011
1012         return 0;
1013 }
1014
1015 static int read_convert_config(const char *var, const char *value, void *cb)
1016 {
1017         const char *key, *name;
1018         size_t namelen;
1019         struct convert_driver *drv;
1020
1021         /*
1022          * External conversion drivers are configured using
1023          * "filter.<name>.variable".
1024          */
1025         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
1026                 return 0;
1027         for (drv = user_convert; drv; drv = drv->next)
1028                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
1029                         break;
1030         if (!drv) {
1031                 CALLOC_ARRAY(drv, 1);
1032                 drv->name = xmemdupz(name, namelen);
1033                 *user_convert_tail = drv;
1034                 user_convert_tail = &(drv->next);
1035         }
1036
1037         /*
1038          * filter.<name>.smudge and filter.<name>.clean specifies
1039          * the command line:
1040          *
1041          *      command-line
1042          *
1043          * The command-line will not be interpolated in any way.
1044          */
1045
1046         if (!strcmp("smudge", key))
1047                 return git_config_string(&drv->smudge, var, value);
1048
1049         if (!strcmp("clean", key))
1050                 return git_config_string(&drv->clean, var, value);
1051
1052         if (!strcmp("process", key))
1053                 return git_config_string(&drv->process, var, value);
1054
1055         if (!strcmp("required", key)) {
1056                 drv->required = git_config_bool(var, value);
1057                 return 0;
1058         }
1059
1060         return 0;
1061 }
1062
1063 static int count_ident(const char *cp, unsigned long size)
1064 {
1065         /*
1066          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
1067          */
1068         int cnt = 0;
1069         char ch;
1070
1071         while (size) {
1072                 ch = *cp++;
1073                 size--;
1074                 if (ch != '$')
1075                         continue;
1076                 if (size < 3)
1077                         break;
1078                 if (memcmp("Id", cp, 2))
1079                         continue;
1080                 ch = cp[2];
1081                 cp += 3;
1082                 size -= 3;
1083                 if (ch == '$')
1084                         cnt++; /* $Id$ */
1085                 if (ch != ':')
1086                         continue;
1087
1088                 /*
1089                  * "$Id: ... "; scan up to the closing dollar sign and discard.
1090                  */
1091                 while (size) {
1092                         ch = *cp++;
1093                         size--;
1094                         if (ch == '$') {
1095                                 cnt++;
1096                                 break;
1097                         }
1098                         if (ch == '\n')
1099                                 break;
1100                 }
1101         }
1102         return cnt;
1103 }
1104
1105 static int ident_to_git(const char *src, size_t len,
1106                         struct strbuf *buf, int ident)
1107 {
1108         char *dst, *dollar;
1109
1110         if (!ident || (src && !count_ident(src, len)))
1111                 return 0;
1112
1113         if (!buf)
1114                 return 1;
1115
1116         /* only grow if not in place */
1117         if (strbuf_avail(buf) + buf->len < len)
1118                 strbuf_grow(buf, len - buf->len);
1119         dst = buf->buf;
1120         for (;;) {
1121                 dollar = memchr(src, '$', len);
1122                 if (!dollar)
1123                         break;
1124                 memmove(dst, src, dollar + 1 - src);
1125                 dst += dollar + 1 - src;
1126                 len -= dollar + 1 - src;
1127                 src  = dollar + 1;
1128
1129                 if (len > 3 && !memcmp(src, "Id:", 3)) {
1130                         dollar = memchr(src + 3, '$', len - 3);
1131                         if (!dollar)
1132                                 break;
1133                         if (memchr(src + 3, '\n', dollar - src - 3)) {
1134                                 /* Line break before the next dollar. */
1135                                 continue;
1136                         }
1137
1138                         memcpy(dst, "Id$", 3);
1139                         dst += 3;
1140                         len -= dollar + 1 - src;
1141                         src  = dollar + 1;
1142                 }
1143         }
1144         memmove(dst, src, len);
1145         strbuf_setlen(buf, dst + len - buf->buf);
1146         return 1;
1147 }
1148
1149 static int ident_to_worktree(const char *src, size_t len,
1150                              struct strbuf *buf, int ident)
1151 {
1152         struct object_id oid;
1153         char *to_free = NULL, *dollar, *spc;
1154         int cnt;
1155
1156         if (!ident)
1157                 return 0;
1158
1159         cnt = count_ident(src, len);
1160         if (!cnt)
1161                 return 0;
1162
1163         /* are we "faking" in place editing ? */
1164         if (src == buf->buf)
1165                 to_free = strbuf_detach(buf, NULL);
1166         hash_object_file(the_hash_algo, src, len, "blob", &oid);
1167
1168         strbuf_grow(buf, len + cnt * (the_hash_algo->hexsz + 3));
1169         for (;;) {
1170                 /* step 1: run to the next '$' */
1171                 dollar = memchr(src, '$', len);
1172                 if (!dollar)
1173                         break;
1174                 strbuf_add(buf, src, dollar + 1 - src);
1175                 len -= dollar + 1 - src;
1176                 src  = dollar + 1;
1177
1178                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
1179                 if (len < 3 || memcmp("Id", src, 2))
1180                         continue;
1181
1182                 /* step 3: skip over Id$ or Id:xxxxx$ */
1183                 if (src[2] == '$') {
1184                         src += 3;
1185                         len -= 3;
1186                 } else if (src[2] == ':') {
1187                         /*
1188                          * It's possible that an expanded Id has crept its way into the
1189                          * repository, we cope with that by stripping the expansion out.
1190                          * This is probably not a good idea, since it will cause changes
1191                          * on checkout, which won't go away by stash, but let's keep it
1192                          * for git-style ids.
1193                          */
1194                         dollar = memchr(src + 3, '$', len - 3);
1195                         if (!dollar) {
1196                                 /* incomplete keyword, no more '$', so just quit the loop */
1197                                 break;
1198                         }
1199
1200                         if (memchr(src + 3, '\n', dollar - src - 3)) {
1201                                 /* Line break before the next dollar. */
1202                                 continue;
1203                         }
1204
1205                         spc = memchr(src + 4, ' ', dollar - src - 4);
1206                         if (spc && spc < dollar-1) {
1207                                 /* There are spaces in unexpected places.
1208                                  * This is probably an id from some other
1209                                  * versioning system. Keep it for now.
1210                                  */
1211                                 continue;
1212                         }
1213
1214                         len -= dollar + 1 - src;
1215                         src  = dollar + 1;
1216                 } else {
1217                         /* it wasn't a "Id$" or "Id:xxxx$" */
1218                         continue;
1219                 }
1220
1221                 /* step 4: substitute */
1222                 strbuf_addstr(buf, "Id: ");
1223                 strbuf_addstr(buf, oid_to_hex(&oid));
1224                 strbuf_addstr(buf, " $");
1225         }
1226         strbuf_add(buf, src, len);
1227
1228         free(to_free);
1229         return 1;
1230 }
1231
1232 static const char *git_path_check_encoding(struct attr_check_item *check)
1233 {
1234         const char *value = check->value;
1235
1236         if (ATTR_UNSET(value) || !strlen(value))
1237                 return NULL;
1238
1239         if (ATTR_TRUE(value) || ATTR_FALSE(value)) {
1240                 die(_("true/false are no valid working-tree-encodings"));
1241         }
1242
1243         /* Don't encode to the default encoding */
1244         if (same_encoding(value, default_encoding))
1245                 return NULL;
1246
1247         return value;
1248 }
1249
1250 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
1251 {
1252         const char *value = check->value;
1253
1254         if (ATTR_TRUE(value))
1255                 return CRLF_TEXT;
1256         else if (ATTR_FALSE(value))
1257                 return CRLF_BINARY;
1258         else if (ATTR_UNSET(value))
1259                 ;
1260         else if (!strcmp(value, "input"))
1261                 return CRLF_TEXT_INPUT;
1262         else if (!strcmp(value, "auto"))
1263                 return CRLF_AUTO;
1264         return CRLF_UNDEFINED;
1265 }
1266
1267 static enum eol git_path_check_eol(struct attr_check_item *check)
1268 {
1269         const char *value = check->value;
1270
1271         if (ATTR_UNSET(value))
1272                 ;
1273         else if (!strcmp(value, "lf"))
1274                 return EOL_LF;
1275         else if (!strcmp(value, "crlf"))
1276                 return EOL_CRLF;
1277         return EOL_UNSET;
1278 }
1279
1280 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
1281 {
1282         const char *value = check->value;
1283         struct convert_driver *drv;
1284
1285         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
1286                 return NULL;
1287         for (drv = user_convert; drv; drv = drv->next)
1288                 if (!strcmp(value, drv->name))
1289                         return drv;
1290         return NULL;
1291 }
1292
1293 static int git_path_check_ident(struct attr_check_item *check)
1294 {
1295         const char *value = check->value;
1296
1297         return !!ATTR_TRUE(value);
1298 }
1299
1300 struct conv_attrs {
1301         struct convert_driver *drv;
1302         enum crlf_action attr_action; /* What attr says */
1303         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
1304         int ident;
1305         const char *working_tree_encoding; /* Supported encoding or default encoding if NULL */
1306 };
1307
1308 static struct attr_check *check;
1309
1310 static void convert_attrs(const struct index_state *istate,
1311                           struct conv_attrs *ca, const char *path)
1312 {
1313         struct attr_check_item *ccheck = NULL;
1314
1315         if (!check) {
1316                 check = attr_check_initl("crlf", "ident", "filter",
1317                                          "eol", "text", "working-tree-encoding",
1318                                          NULL);
1319                 user_convert_tail = &user_convert;
1320                 git_config(read_convert_config, NULL);
1321         }
1322
1323         git_check_attr(istate, path, check);
1324         ccheck = check->items;
1325         ca->crlf_action = git_path_check_crlf(ccheck + 4);
1326         if (ca->crlf_action == CRLF_UNDEFINED)
1327                 ca->crlf_action = git_path_check_crlf(ccheck + 0);
1328         ca->ident = git_path_check_ident(ccheck + 1);
1329         ca->drv = git_path_check_convert(ccheck + 2);
1330         if (ca->crlf_action != CRLF_BINARY) {
1331                 enum eol eol_attr = git_path_check_eol(ccheck + 3);
1332                 if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1333                         ca->crlf_action = CRLF_AUTO_INPUT;
1334                 else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1335                         ca->crlf_action = CRLF_AUTO_CRLF;
1336                 else if (eol_attr == EOL_LF)
1337                         ca->crlf_action = CRLF_TEXT_INPUT;
1338                 else if (eol_attr == EOL_CRLF)
1339                         ca->crlf_action = CRLF_TEXT_CRLF;
1340         }
1341         ca->working_tree_encoding = git_path_check_encoding(ccheck + 5);
1342
1343         /* Save attr and make a decision for action */
1344         ca->attr_action = ca->crlf_action;
1345         if (ca->crlf_action == CRLF_TEXT)
1346                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1347         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1348                 ca->crlf_action = CRLF_BINARY;
1349         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1350                 ca->crlf_action = CRLF_AUTO_CRLF;
1351         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1352                 ca->crlf_action = CRLF_AUTO_INPUT;
1353 }
1354
1355 void reset_parsed_attributes(void)
1356 {
1357         struct convert_driver *drv, *next;
1358
1359         attr_check_free(check);
1360         check = NULL;
1361         reset_merge_attributes();
1362
1363         for (drv = user_convert; drv; drv = next) {
1364                 next = drv->next;
1365                 free((void *)drv->name);
1366                 free(drv);
1367         }
1368         user_convert = NULL;
1369         user_convert_tail = NULL;
1370 }
1371
1372 int would_convert_to_git_filter_fd(const struct index_state *istate, const char *path)
1373 {
1374         struct conv_attrs ca;
1375
1376         convert_attrs(istate, &ca, path);
1377         if (!ca.drv)
1378                 return 0;
1379
1380         /*
1381          * Apply a filter to an fd only if the filter is required to succeed.
1382          * We must die if the filter fails, because the original data before
1383          * filtering is not available.
1384          */
1385         if (!ca.drv->required)
1386                 return 0;
1387
1388         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN, NULL, NULL);
1389 }
1390
1391 const char *get_convert_attr_ascii(const struct index_state *istate, const char *path)
1392 {
1393         struct conv_attrs ca;
1394
1395         convert_attrs(istate, &ca, path);
1396         switch (ca.attr_action) {
1397         case CRLF_UNDEFINED:
1398                 return "";
1399         case CRLF_BINARY:
1400                 return "-text";
1401         case CRLF_TEXT:
1402                 return "text";
1403         case CRLF_TEXT_INPUT:
1404                 return "text eol=lf";
1405         case CRLF_TEXT_CRLF:
1406                 return "text eol=crlf";
1407         case CRLF_AUTO:
1408                 return "text=auto";
1409         case CRLF_AUTO_CRLF:
1410                 return "text=auto eol=crlf";
1411         case CRLF_AUTO_INPUT:
1412                 return "text=auto eol=lf";
1413         }
1414         return "";
1415 }
1416
1417 int convert_to_git(const struct index_state *istate,
1418                    const char *path, const char *src, size_t len,
1419                    struct strbuf *dst, int conv_flags)
1420 {
1421         int ret = 0;
1422         struct conv_attrs ca;
1423
1424         convert_attrs(istate, &ca, path);
1425
1426         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN, NULL, NULL);
1427         if (!ret && ca.drv && ca.drv->required)
1428                 die(_("%s: clean filter '%s' failed"), path, ca.drv->name);
1429
1430         if (ret && dst) {
1431                 src = dst->buf;
1432                 len = dst->len;
1433         }
1434
1435         ret |= encode_to_git(path, src, len, dst, ca.working_tree_encoding, conv_flags);
1436         if (ret && dst) {
1437                 src = dst->buf;
1438                 len = dst->len;
1439         }
1440
1441         if (!(conv_flags & CONV_EOL_KEEP_CRLF)) {
1442                 ret |= crlf_to_git(istate, path, src, len, dst, ca.crlf_action, conv_flags);
1443                 if (ret && dst) {
1444                         src = dst->buf;
1445                         len = dst->len;
1446                 }
1447         }
1448         return ret | ident_to_git(src, len, dst, ca.ident);
1449 }
1450
1451 void convert_to_git_filter_fd(const struct index_state *istate,
1452                               const char *path, int fd, struct strbuf *dst,
1453                               int conv_flags)
1454 {
1455         struct conv_attrs ca;
1456         convert_attrs(istate, &ca, path);
1457
1458         assert(ca.drv);
1459
1460         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN, NULL, NULL))
1461                 die(_("%s: clean filter '%s' failed"), path, ca.drv->name);
1462
1463         encode_to_git(path, dst->buf, dst->len, dst, ca.working_tree_encoding, conv_flags);
1464         crlf_to_git(istate, path, dst->buf, dst->len, dst, ca.crlf_action, conv_flags);
1465         ident_to_git(dst->buf, dst->len, dst, ca.ident);
1466 }
1467
1468 static int convert_to_working_tree_internal(const struct index_state *istate,
1469                                             const char *path, const char *src,
1470                                             size_t len, struct strbuf *dst,
1471                                             int normalizing,
1472                                             const struct checkout_metadata *meta,
1473                                             struct delayed_checkout *dco)
1474 {
1475         int ret = 0, ret_filter = 0;
1476         struct conv_attrs ca;
1477
1478         convert_attrs(istate, &ca, path);
1479
1480         ret |= ident_to_worktree(src, len, dst, ca.ident);
1481         if (ret) {
1482                 src = dst->buf;
1483                 len = dst->len;
1484         }
1485         /*
1486          * CRLF conversion can be skipped if normalizing, unless there
1487          * is a smudge or process filter (even if the process filter doesn't
1488          * support smudge).  The filters might expect CRLFs.
1489          */
1490         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1491                 ret |= crlf_to_worktree(src, len, dst, ca.crlf_action);
1492                 if (ret) {
1493                         src = dst->buf;
1494                         len = dst->len;
1495                 }
1496         }
1497
1498         ret |= encode_to_worktree(path, src, len, dst, ca.working_tree_encoding);
1499         if (ret) {
1500                 src = dst->buf;
1501                 len = dst->len;
1502         }
1503
1504         ret_filter = apply_filter(
1505                 path, src, len, -1, dst, ca.drv, CAP_SMUDGE, meta, dco);
1506         if (!ret_filter && ca.drv && ca.drv->required)
1507                 die(_("%s: smudge filter %s failed"), path, ca.drv->name);
1508
1509         return ret | ret_filter;
1510 }
1511
1512 int async_convert_to_working_tree(const struct index_state *istate,
1513                                   const char *path, const char *src,
1514                                   size_t len, struct strbuf *dst,
1515                                   const struct checkout_metadata *meta,
1516                                   void *dco)
1517 {
1518         return convert_to_working_tree_internal(istate, path, src, len, dst, 0, meta, dco);
1519 }
1520
1521 int convert_to_working_tree(const struct index_state *istate,
1522                             const char *path, const char *src,
1523                             size_t len, struct strbuf *dst,
1524                             const struct checkout_metadata *meta)
1525 {
1526         return convert_to_working_tree_internal(istate, path, src, len, dst, 0, meta, NULL);
1527 }
1528
1529 int renormalize_buffer(const struct index_state *istate, const char *path,
1530                        const char *src, size_t len, struct strbuf *dst)
1531 {
1532         int ret = convert_to_working_tree_internal(istate, path, src, len, dst, 1, NULL, NULL);
1533         if (ret) {
1534                 src = dst->buf;
1535                 len = dst->len;
1536         }
1537         return ret | convert_to_git(istate, path, src, len, dst, CONV_EOL_RENORMALIZE);
1538 }
1539
1540 /*****************************************************************
1541  *
1542  * Streaming conversion support
1543  *
1544  *****************************************************************/
1545
1546 typedef int (*filter_fn)(struct stream_filter *,
1547                          const char *input, size_t *isize_p,
1548                          char *output, size_t *osize_p);
1549 typedef void (*free_fn)(struct stream_filter *);
1550
1551 struct stream_filter_vtbl {
1552         filter_fn filter;
1553         free_fn free;
1554 };
1555
1556 struct stream_filter {
1557         struct stream_filter_vtbl *vtbl;
1558 };
1559
1560 static int null_filter_fn(struct stream_filter *filter,
1561                           const char *input, size_t *isize_p,
1562                           char *output, size_t *osize_p)
1563 {
1564         size_t count;
1565
1566         if (!input)
1567                 return 0; /* we do not keep any states */
1568         count = *isize_p;
1569         if (*osize_p < count)
1570                 count = *osize_p;
1571         if (count) {
1572                 memmove(output, input, count);
1573                 *isize_p -= count;
1574                 *osize_p -= count;
1575         }
1576         return 0;
1577 }
1578
1579 static void null_free_fn(struct stream_filter *filter)
1580 {
1581         ; /* nothing -- null instances are shared */
1582 }
1583
1584 static struct stream_filter_vtbl null_vtbl = {
1585         null_filter_fn,
1586         null_free_fn,
1587 };
1588
1589 static struct stream_filter null_filter_singleton = {
1590         &null_vtbl,
1591 };
1592
1593 int is_null_stream_filter(struct stream_filter *filter)
1594 {
1595         return filter == &null_filter_singleton;
1596 }
1597
1598
1599 /*
1600  * LF-to-CRLF filter
1601  */
1602
1603 struct lf_to_crlf_filter {
1604         struct stream_filter filter;
1605         unsigned has_held:1;
1606         char held;
1607 };
1608
1609 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1610                                 const char *input, size_t *isize_p,
1611                                 char *output, size_t *osize_p)
1612 {
1613         size_t count, o = 0;
1614         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1615
1616         /*
1617          * We may be holding onto the CR to see if it is followed by a
1618          * LF, in which case we would need to go to the main loop.
1619          * Otherwise, just emit it to the output stream.
1620          */
1621         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1622                 output[o++] = lf_to_crlf->held;
1623                 lf_to_crlf->has_held = 0;
1624         }
1625
1626         /* We are told to drain */
1627         if (!input) {
1628                 *osize_p -= o;
1629                 return 0;
1630         }
1631
1632         count = *isize_p;
1633         if (count || lf_to_crlf->has_held) {
1634                 size_t i;
1635                 int was_cr = 0;
1636
1637                 if (lf_to_crlf->has_held) {
1638                         was_cr = 1;
1639                         lf_to_crlf->has_held = 0;
1640                 }
1641
1642                 for (i = 0; o < *osize_p && i < count; i++) {
1643                         char ch = input[i];
1644
1645                         if (ch == '\n') {
1646                                 output[o++] = '\r';
1647                         } else if (was_cr) {
1648                                 /*
1649                                  * Previous round saw CR and it is not followed
1650                                  * by a LF; emit the CR before processing the
1651                                  * current character.
1652                                  */
1653                                 output[o++] = '\r';
1654                         }
1655
1656                         /*
1657                          * We may have consumed the last output slot,
1658                          * in which case we need to break out of this
1659                          * loop; hold the current character before
1660                          * returning.
1661                          */
1662                         if (*osize_p <= o) {
1663                                 lf_to_crlf->has_held = 1;
1664                                 lf_to_crlf->held = ch;
1665                                 continue; /* break but increment i */
1666                         }
1667
1668                         if (ch == '\r') {
1669                                 was_cr = 1;
1670                                 continue;
1671                         }
1672
1673                         was_cr = 0;
1674                         output[o++] = ch;
1675                 }
1676
1677                 *osize_p -= o;
1678                 *isize_p -= i;
1679
1680                 if (!lf_to_crlf->has_held && was_cr) {
1681                         lf_to_crlf->has_held = 1;
1682                         lf_to_crlf->held = '\r';
1683                 }
1684         }
1685         return 0;
1686 }
1687
1688 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1689 {
1690         free(filter);
1691 }
1692
1693 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1694         lf_to_crlf_filter_fn,
1695         lf_to_crlf_free_fn,
1696 };
1697
1698 static struct stream_filter *lf_to_crlf_filter(void)
1699 {
1700         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1701
1702         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1703         return (struct stream_filter *)lf_to_crlf;
1704 }
1705
1706 /*
1707  * Cascade filter
1708  */
1709 #define FILTER_BUFFER 1024
1710 struct cascade_filter {
1711         struct stream_filter filter;
1712         struct stream_filter *one;
1713         struct stream_filter *two;
1714         char buf[FILTER_BUFFER];
1715         int end, ptr;
1716 };
1717
1718 static int cascade_filter_fn(struct stream_filter *filter,
1719                              const char *input, size_t *isize_p,
1720                              char *output, size_t *osize_p)
1721 {
1722         struct cascade_filter *cas = (struct cascade_filter *) filter;
1723         size_t filled = 0;
1724         size_t sz = *osize_p;
1725         size_t to_feed, remaining;
1726
1727         /*
1728          * input -- (one) --> buf -- (two) --> output
1729          */
1730         while (filled < sz) {
1731                 remaining = sz - filled;
1732
1733                 /* do we already have something to feed two with? */
1734                 if (cas->ptr < cas->end) {
1735                         to_feed = cas->end - cas->ptr;
1736                         if (stream_filter(cas->two,
1737                                           cas->buf + cas->ptr, &to_feed,
1738                                           output + filled, &remaining))
1739                                 return -1;
1740                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1741                         filled = sz - remaining;
1742                         continue;
1743                 }
1744
1745                 /* feed one from upstream and have it emit into our buffer */
1746                 to_feed = input ? *isize_p : 0;
1747                 if (input && !to_feed)
1748                         break;
1749                 remaining = sizeof(cas->buf);
1750                 if (stream_filter(cas->one,
1751                                   input, &to_feed,
1752                                   cas->buf, &remaining))
1753                         return -1;
1754                 cas->end = sizeof(cas->buf) - remaining;
1755                 cas->ptr = 0;
1756                 if (input) {
1757                         size_t fed = *isize_p - to_feed;
1758                         *isize_p -= fed;
1759                         input += fed;
1760                 }
1761
1762                 /* do we know that we drained one completely? */
1763                 if (input || cas->end)
1764                         continue;
1765
1766                 /* tell two to drain; we have nothing more to give it */
1767                 to_feed = 0;
1768                 remaining = sz - filled;
1769                 if (stream_filter(cas->two,
1770                                   NULL, &to_feed,
1771                                   output + filled, &remaining))
1772                         return -1;
1773                 if (remaining == (sz - filled))
1774                         break; /* completely drained two */
1775                 filled = sz - remaining;
1776         }
1777         *osize_p -= filled;
1778         return 0;
1779 }
1780
1781 static void cascade_free_fn(struct stream_filter *filter)
1782 {
1783         struct cascade_filter *cas = (struct cascade_filter *)filter;
1784         free_stream_filter(cas->one);
1785         free_stream_filter(cas->two);
1786         free(filter);
1787 }
1788
1789 static struct stream_filter_vtbl cascade_vtbl = {
1790         cascade_filter_fn,
1791         cascade_free_fn,
1792 };
1793
1794 static struct stream_filter *cascade_filter(struct stream_filter *one,
1795                                             struct stream_filter *two)
1796 {
1797         struct cascade_filter *cascade;
1798
1799         if (!one || is_null_stream_filter(one))
1800                 return two;
1801         if (!two || is_null_stream_filter(two))
1802                 return one;
1803
1804         cascade = xmalloc(sizeof(*cascade));
1805         cascade->one = one;
1806         cascade->two = two;
1807         cascade->end = cascade->ptr = 0;
1808         cascade->filter.vtbl = &cascade_vtbl;
1809         return (struct stream_filter *)cascade;
1810 }
1811
1812 /*
1813  * ident filter
1814  */
1815 #define IDENT_DRAINING (-1)
1816 #define IDENT_SKIPPING (-2)
1817 struct ident_filter {
1818         struct stream_filter filter;
1819         struct strbuf left;
1820         int state;
1821         char ident[GIT_MAX_HEXSZ + 5]; /* ": x40 $" */
1822 };
1823
1824 static int is_foreign_ident(const char *str)
1825 {
1826         int i;
1827
1828         if (!skip_prefix(str, "$Id: ", &str))
1829                 return 0;
1830         for (i = 0; str[i]; i++) {
1831                 if (isspace(str[i]) && str[i+1] != '$')
1832                         return 1;
1833         }
1834         return 0;
1835 }
1836
1837 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1838 {
1839         size_t to_drain = ident->left.len;
1840
1841         if (*osize_p < to_drain)
1842                 to_drain = *osize_p;
1843         if (to_drain) {
1844                 memcpy(*output_p, ident->left.buf, to_drain);
1845                 strbuf_remove(&ident->left, 0, to_drain);
1846                 *output_p += to_drain;
1847                 *osize_p -= to_drain;
1848         }
1849         if (!ident->left.len)
1850                 ident->state = 0;
1851 }
1852
1853 static int ident_filter_fn(struct stream_filter *filter,
1854                            const char *input, size_t *isize_p,
1855                            char *output, size_t *osize_p)
1856 {
1857         struct ident_filter *ident = (struct ident_filter *)filter;
1858         static const char head[] = "$Id";
1859
1860         if (!input) {
1861                 /* drain upon eof */
1862                 switch (ident->state) {
1863                 default:
1864                         strbuf_add(&ident->left, head, ident->state);
1865                         /* fallthrough */
1866                 case IDENT_SKIPPING:
1867                         /* fallthrough */
1868                 case IDENT_DRAINING:
1869                         ident_drain(ident, &output, osize_p);
1870                 }
1871                 return 0;
1872         }
1873
1874         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1875                 int ch;
1876
1877                 if (ident->state == IDENT_DRAINING) {
1878                         ident_drain(ident, &output, osize_p);
1879                         if (!*osize_p)
1880                                 break;
1881                         continue;
1882                 }
1883
1884                 ch = *(input++);
1885                 (*isize_p)--;
1886
1887                 if (ident->state == IDENT_SKIPPING) {
1888                         /*
1889                          * Skipping until '$' or LF, but keeping them
1890                          * in case it is a foreign ident.
1891                          */
1892                         strbuf_addch(&ident->left, ch);
1893                         if (ch != '\n' && ch != '$')
1894                                 continue;
1895                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1896                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1897                                 strbuf_addstr(&ident->left, ident->ident);
1898                         }
1899                         ident->state = IDENT_DRAINING;
1900                         continue;
1901                 }
1902
1903                 if (ident->state < sizeof(head) &&
1904                     head[ident->state] == ch) {
1905                         ident->state++;
1906                         continue;
1907                 }
1908
1909                 if (ident->state)
1910                         strbuf_add(&ident->left, head, ident->state);
1911                 if (ident->state == sizeof(head) - 1) {
1912                         if (ch != ':' && ch != '$') {
1913                                 strbuf_addch(&ident->left, ch);
1914                                 ident->state = 0;
1915                                 continue;
1916                         }
1917
1918                         if (ch == ':') {
1919                                 strbuf_addch(&ident->left, ch);
1920                                 ident->state = IDENT_SKIPPING;
1921                         } else {
1922                                 strbuf_addstr(&ident->left, ident->ident);
1923                                 ident->state = IDENT_DRAINING;
1924                         }
1925                         continue;
1926                 }
1927
1928                 strbuf_addch(&ident->left, ch);
1929                 ident->state = IDENT_DRAINING;
1930         }
1931         return 0;
1932 }
1933
1934 static void ident_free_fn(struct stream_filter *filter)
1935 {
1936         struct ident_filter *ident = (struct ident_filter *)filter;
1937         strbuf_release(&ident->left);
1938         free(filter);
1939 }
1940
1941 static struct stream_filter_vtbl ident_vtbl = {
1942         ident_filter_fn,
1943         ident_free_fn,
1944 };
1945
1946 static struct stream_filter *ident_filter(const struct object_id *oid)
1947 {
1948         struct ident_filter *ident = xmalloc(sizeof(*ident));
1949
1950         xsnprintf(ident->ident, sizeof(ident->ident),
1951                   ": %s $", oid_to_hex(oid));
1952         strbuf_init(&ident->left, 0);
1953         ident->filter.vtbl = &ident_vtbl;
1954         ident->state = 0;
1955         return (struct stream_filter *)ident;
1956 }
1957
1958 /*
1959  * Return an appropriately constructed filter for the path, or NULL if
1960  * the contents cannot be filtered without reading the whole thing
1961  * in-core.
1962  *
1963  * Note that you would be crazy to set CRLF, smudge/clean or ident to a
1964  * large binary blob you would want us not to slurp into the memory!
1965  */
1966 struct stream_filter *get_stream_filter(const struct index_state *istate,
1967                                         const char *path,
1968                                         const struct object_id *oid)
1969 {
1970         struct conv_attrs ca;
1971         struct stream_filter *filter = NULL;
1972
1973         convert_attrs(istate, &ca, path);
1974         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1975                 return NULL;
1976
1977         if (ca.working_tree_encoding)
1978                 return NULL;
1979
1980         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1981                 return NULL;
1982
1983         if (ca.ident)
1984                 filter = ident_filter(oid);
1985
1986         if (output_eol(ca.crlf_action) == EOL_CRLF)
1987                 filter = cascade_filter(filter, lf_to_crlf_filter());
1988         else
1989                 filter = cascade_filter(filter, &null_filter_singleton);
1990
1991         return filter;
1992 }
1993
1994 void free_stream_filter(struct stream_filter *filter)
1995 {
1996         filter->vtbl->free(filter);
1997 }
1998
1999 int stream_filter(struct stream_filter *filter,
2000                   const char *input, size_t *isize_p,
2001                   char *output, size_t *osize_p)
2002 {
2003         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
2004 }
2005
2006 void init_checkout_metadata(struct checkout_metadata *meta, const char *refname,
2007                             const struct object_id *treeish,
2008                             const struct object_id *blob)
2009 {
2010         memset(meta, 0, sizeof(*meta));
2011         if (refname)
2012                 meta->refname = refname;
2013         if (treeish)
2014                 oidcpy(&meta->treeish, treeish);
2015         if (blob)
2016                 oidcpy(&meta->blob, blob);
2017 }
2018
2019 void clone_checkout_metadata(struct checkout_metadata *dst,
2020                              const struct checkout_metadata *src,
2021                              const struct object_id *blob)
2022 {
2023         memcpy(dst, src, sizeof(*dst));
2024         if (blob)
2025                 oidcpy(&dst->blob, blob);
2026 }