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