Second batch for 2.19 cycle
[git] / convert.c
1 #define NO_THE_INDEX_COMPATIBILITY_MACROS
2 #include "cache.h"
3 #include "config.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
12 /*
13  * convert.c - convert a file when checking it out and checking it in.
14  *
15  * This should use the pathname to decide on whether it wants to do some
16  * more interesting conversions (automatic gzip/unzip, general format
17  * conversions etc etc), but by default it just does automatic CRLF<->LF
18  * translation when the "text" attribute or "auto_crlf" option is set.
19  */
20
21 /* Stat bits: When BIN is set, the txt bits are unset */
22 #define CONVERT_STAT_BITS_TXT_LF    0x1
23 #define CONVERT_STAT_BITS_TXT_CRLF  0x2
24 #define CONVERT_STAT_BITS_BIN       0x4
25
26 enum crlf_action {
27         CRLF_UNDEFINED,
28         CRLF_BINARY,
29         CRLF_TEXT,
30         CRLF_TEXT_INPUT,
31         CRLF_TEXT_CRLF,
32         CRLF_AUTO,
33         CRLF_AUTO_INPUT,
34         CRLF_AUTO_CRLF
35 };
36
37 struct text_stat {
38         /* NUL, CR, LF and CRLF counts */
39         unsigned nul, lonecr, lonelf, crlf;
40
41         /* These are just approximations! */
42         unsigned printable, nonprintable;
43 };
44
45 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
46 {
47         unsigned long i;
48
49         memset(stats, 0, sizeof(*stats));
50
51         for (i = 0; i < size; i++) {
52                 unsigned char c = buf[i];
53                 if (c == '\r') {
54                         if (i+1 < size && buf[i+1] == '\n') {
55                                 stats->crlf++;
56                                 i++;
57                         } else
58                                 stats->lonecr++;
59                         continue;
60                 }
61                 if (c == '\n') {
62                         stats->lonelf++;
63                         continue;
64                 }
65                 if (c == 127)
66                         /* DEL */
67                         stats->nonprintable++;
68                 else if (c < 32) {
69                         switch (c) {
70                                 /* BS, HT, ESC and FF */
71                         case '\b': case '\t': case '\033': case '\014':
72                                 stats->printable++;
73                                 break;
74                         case 0:
75                                 stats->nul++;
76                                 /* fall through */
77                         default:
78                                 stats->nonprintable++;
79                         }
80                 }
81                 else
82                         stats->printable++;
83         }
84
85         /* If file ends with EOF then don't count this EOF as non-printable. */
86         if (size >= 1 && buf[size-1] == '\032')
87                 stats->nonprintable--;
88 }
89
90 /*
91  * The same heuristics as diff.c::mmfile_is_binary()
92  * We treat files with bare CR as binary
93  */
94 static int convert_is_binary(unsigned long size, const struct text_stat *stats)
95 {
96         if (stats->lonecr)
97                 return 1;
98         if (stats->nul)
99                 return 1;
100         if ((stats->printable >> 7) < stats->nonprintable)
101                 return 1;
102         return 0;
103 }
104
105 static unsigned int gather_convert_stats(const char *data, unsigned long size)
106 {
107         struct text_stat stats;
108         int ret = 0;
109         if (!data || !size)
110                 return 0;
111         gather_stats(data, size, &stats);
112         if (convert_is_binary(size, &stats))
113                 ret |= CONVERT_STAT_BITS_BIN;
114         if (stats.crlf)
115                 ret |= CONVERT_STAT_BITS_TXT_CRLF;
116         if (stats.lonelf)
117                 ret |=  CONVERT_STAT_BITS_TXT_LF;
118
119         return ret;
120 }
121
122 static const char *gather_convert_stats_ascii(const char *data, unsigned long size)
123 {
124         unsigned int convert_stats = gather_convert_stats(data, size);
125
126         if (convert_stats & CONVERT_STAT_BITS_BIN)
127                 return "-text";
128         switch (convert_stats) {
129         case CONVERT_STAT_BITS_TXT_LF:
130                 return "lf";
131         case CONVERT_STAT_BITS_TXT_CRLF:
132                 return "crlf";
133         case CONVERT_STAT_BITS_TXT_LF | CONVERT_STAT_BITS_TXT_CRLF:
134                 return "mixed";
135         default:
136                 return "none";
137         }
138 }
139
140 const char *get_cached_convert_stats_ascii(const struct index_state *istate,
141                                            const char *path)
142 {
143         const char *ret;
144         unsigned long sz;
145         void *data = read_blob_data_from_index(istate, path, &sz);
146         ret = gather_convert_stats_ascii(data, sz);
147         free(data);
148         return ret;
149 }
150
151 const char *get_wt_convert_stats_ascii(const char *path)
152 {
153         const char *ret = "";
154         struct strbuf sb = STRBUF_INIT;
155         if (strbuf_read_file(&sb, path, 0) >= 0)
156                 ret = gather_convert_stats_ascii(sb.buf, sb.len);
157         strbuf_release(&sb);
158         return ret;
159 }
160
161 static int text_eol_is_crlf(void)
162 {
163         if (auto_crlf == AUTO_CRLF_TRUE)
164                 return 1;
165         else if (auto_crlf == AUTO_CRLF_INPUT)
166                 return 0;
167         if (core_eol == EOL_CRLF)
168                 return 1;
169         if (core_eol == EOL_UNSET && EOL_NATIVE == EOL_CRLF)
170                 return 1;
171         return 0;
172 }
173
174 static enum eol output_eol(enum crlf_action crlf_action)
175 {
176         switch (crlf_action) {
177         case CRLF_BINARY:
178                 return EOL_UNSET;
179         case CRLF_TEXT_CRLF:
180                 return EOL_CRLF;
181         case CRLF_TEXT_INPUT:
182                 return EOL_LF;
183         case CRLF_UNDEFINED:
184         case CRLF_AUTO_CRLF:
185                 return EOL_CRLF;
186         case CRLF_AUTO_INPUT:
187                 return EOL_LF;
188         case CRLF_TEXT:
189         case CRLF_AUTO:
190                 /* fall through */
191                 return text_eol_is_crlf() ? EOL_CRLF : EOL_LF;
192         }
193         warning("Illegal crlf_action %d\n", (int)crlf_action);
194         return core_eol;
195 }
196
197 static void check_global_conv_flags_eol(const char *path, enum crlf_action crlf_action,
198                             struct text_stat *old_stats, struct text_stat *new_stats,
199                             int conv_flags)
200 {
201         if (old_stats->crlf && !new_stats->crlf ) {
202                 /*
203                  * CRLFs would not be restored by checkout
204                  */
205                 if (conv_flags & CONV_EOL_RNDTRP_DIE)
206                         die(_("CRLF would be replaced by LF in %s."), path);
207                 else if (conv_flags & CONV_EOL_RNDTRP_WARN)
208                         warning(_("CRLF will be replaced by LF in %s.\n"
209                                   "The file will have its original line"
210                                   " endings in your working directory."), path);
211         } else if (old_stats->lonelf && !new_stats->lonelf ) {
212                 /*
213                  * CRLFs would be added by checkout
214                  */
215                 if (conv_flags & CONV_EOL_RNDTRP_DIE)
216                         die(_("LF would be replaced by CRLF in %s"), path);
217                 else if (conv_flags & CONV_EOL_RNDTRP_WARN)
218                         warning(_("LF will be replaced by CRLF in %s.\n"
219                                   "The file will have its original line"
220                                   " endings in your working directory."), path);
221         }
222 }
223
224 static int has_crlf_in_index(const struct index_state *istate, const char *path)
225 {
226         unsigned long sz;
227         void *data;
228         const char *crp;
229         int has_crlf = 0;
230
231         data = read_blob_data_from_index(istate, path, &sz);
232         if (!data)
233                 return 0;
234
235         crp = memchr(data, '\r', sz);
236         if (crp) {
237                 unsigned int ret_stats;
238                 ret_stats = gather_convert_stats(data, sz);
239                 if (!(ret_stats & CONVERT_STAT_BITS_BIN) &&
240                     (ret_stats & CONVERT_STAT_BITS_TXT_CRLF))
241                         has_crlf = 1;
242         }
243         free(data);
244         return has_crlf;
245 }
246
247 static int will_convert_lf_to_crlf(size_t len, struct text_stat *stats,
248                                    enum crlf_action crlf_action)
249 {
250         if (output_eol(crlf_action) != EOL_CRLF)
251                 return 0;
252         /* No "naked" LF? Nothing to convert, regardless. */
253         if (!stats->lonelf)
254                 return 0;
255
256         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
257                 /* If we have any CR or CRLF line endings, we do not touch it */
258                 /* This is the new safer autocrlf-handling */
259                 if (stats->lonecr || stats->crlf)
260                         return 0;
261
262                 if (convert_is_binary(len, stats))
263                         return 0;
264         }
265         return 1;
266
267 }
268
269 static int validate_encoding(const char *path, const char *enc,
270                       const char *data, size_t len, int die_on_error)
271 {
272         /* We only check for UTF here as UTF?? can be an alias for UTF-?? */
273         if (istarts_with(enc, "UTF")) {
274                 /*
275                  * Check for detectable errors in UTF encodings
276                  */
277                 if (has_prohibited_utf_bom(enc, data, len)) {
278                         const char *error_msg = _(
279                                 "BOM is prohibited in '%s' if encoded as %s");
280                         /*
281                          * This advice is shown for UTF-??BE and UTF-??LE encodings.
282                          * We cut off the last two characters of the encoding name
283                          * to generate the encoding name suitable for BOMs.
284                          */
285                         const char *advise_msg = _(
286                                 "The file '%s' contains a byte order "
287                                 "mark (BOM). Please use UTF-%s as "
288                                 "working-tree-encoding.");
289                         const char *stripped = NULL;
290                         char *upper = xstrdup_toupper(enc);
291                         upper[strlen(upper)-2] = '\0';
292                         if (!skip_prefix(upper, "UTF-", &stripped))
293                                 skip_prefix(stripped, "UTF", &stripped);
294                         advise(advise_msg, path, stripped);
295                         free(upper);
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                         const char *stripped = NULL;
311                         char *upper = xstrdup_toupper(enc);
312                         if (!skip_prefix(upper, "UTF-", &stripped))
313                                 skip_prefix(stripped, "UTF", &stripped);
314                         advise(advise_msg, path, stripped, stripped);
315                         free(upper);
316                         if (die_on_error)
317                                 die(error_msg, path, enc);
318                         else {
319                                 return error(error_msg, path, enc);
320                         }
321                 }
322
323         }
324         return 0;
325 }
326
327 static void trace_encoding(const char *context, const char *path,
328                            const char *encoding, const char *buf, size_t len)
329 {
330         static struct trace_key coe = TRACE_KEY_INIT(WORKING_TREE_ENCODING);
331         struct strbuf trace = STRBUF_INIT;
332         int i;
333
334         strbuf_addf(&trace, "%s (%s, considered %s):\n", context, path, encoding);
335         for (i = 0; i < len && buf; ++i) {
336                 strbuf_addf(
337                         &trace,"| \e[2m%2i:\e[0m %2x \e[2m%c\e[0m%c",
338                         i,
339                         (unsigned char) buf[i],
340                         (buf[i] > 32 && buf[i] < 127 ? buf[i] : ' '),
341                         ((i+1) % 8 && (i+1) < len ? ' ' : '\n')
342                 );
343         }
344         strbuf_addchars(&trace, '\n', 1);
345
346         trace_strbuf(&coe, &trace);
347         strbuf_release(&trace);
348 }
349
350 static int check_roundtrip(const char *enc_name)
351 {
352         /*
353          * check_roundtrip_encoding contains a string of comma and/or
354          * space separated encodings (eg. "UTF-16, ASCII, CP1125").
355          * Search for the given encoding in that string.
356          */
357         const char *found = strcasestr(check_roundtrip_encoding, enc_name);
358         const char *next;
359         int len;
360         if (!found)
361                 return 0;
362         next = found + strlen(enc_name);
363         len = strlen(check_roundtrip_encoding);
364         return (found && (
365                         /*
366                          * check that the found encoding is at the
367                          * beginning of check_roundtrip_encoding or
368                          * that it is prefixed with a space or comma
369                          */
370                         found == check_roundtrip_encoding || (
371                                 (isspace(found[-1]) || found[-1] == ',')
372                         )
373                 ) && (
374                         /*
375                          * check that the found encoding is at the
376                          * end of check_roundtrip_encoding or
377                          * that it is suffixed with a space or comma
378                          */
379                         next == check_roundtrip_encoding + len || (
380                                 next < check_roundtrip_encoding + len &&
381                                 (isspace(next[0]) || next[0] == ',')
382                         )
383                 ));
384 }
385
386 static const char *default_encoding = "UTF-8";
387
388 static int encode_to_git(const char *path, const char *src, size_t src_len,
389                          struct strbuf *buf, const char *enc, int conv_flags)
390 {
391         char *dst;
392         int dst_len;
393         int die_on_error = conv_flags & CONV_WRITE_OBJECT;
394
395         /*
396          * No encoding is specified or there is nothing to encode.
397          * Tell the caller that the content was not modified.
398          */
399         if (!enc || (src && !src_len))
400                 return 0;
401
402         /*
403          * Looks like we got called from "would_convert_to_git()".
404          * This means Git wants to know if it would encode (= modify!)
405          * the content. Let's answer with "yes", since an encoding was
406          * specified.
407          */
408         if (!buf && !src)
409                 return 1;
410
411         if (validate_encoding(path, enc, src, src_len, die_on_error))
412                 return 0;
413
414         trace_encoding("source", path, enc, src, src_len);
415         dst = reencode_string_len(src, src_len, default_encoding, enc,
416                                   &dst_len);
417         if (!dst) {
418                 /*
419                  * We could add the blob "as-is" to Git. However, on checkout
420                  * we would try to reencode to the original encoding. This
421                  * would fail and we would leave the user with a messed-up
422                  * working tree. Let's try to avoid this by screaming loud.
423                  */
424                 const char* msg = _("failed to encode '%s' from %s to %s");
425                 if (die_on_error)
426                         die(msg, path, enc, default_encoding);
427                 else {
428                         error(msg, path, enc, default_encoding);
429                         return 0;
430                 }
431         }
432         trace_encoding("destination", path, default_encoding, dst, dst_len);
433
434         /*
435          * UTF supports lossless conversion round tripping [1] and conversions
436          * between UTF and other encodings are mostly round trip safe as
437          * Unicode aims to be a superset of all other character encodings.
438          * However, certain encodings (e.g. SHIFT-JIS) are known to have round
439          * trip issues [2]. Check the round trip conversion for all encodings
440          * listed in core.checkRoundtripEncoding.
441          *
442          * The round trip check is only performed if content is written to Git.
443          * This ensures that no information is lost during conversion to/from
444          * the internal UTF-8 representation.
445          *
446          * Please note, the code below is not tested because I was not able to
447          * generate a faulty round trip without an iconv error. Iconv errors
448          * are already caught above.
449          *
450          * [1] http://unicode.org/faq/utf_bom.html#gen2
451          * [2] https://support.microsoft.com/en-us/help/170559/prb-conversion-problem-between-shift-jis-and-unicode
452          */
453         if (die_on_error && check_roundtrip(enc)) {
454                 char *re_src;
455                 int re_src_len;
456
457                 re_src = reencode_string_len(dst, dst_len,
458                                              enc, default_encoding,
459                                              &re_src_len);
460
461                 trace_printf("Checking roundtrip encoding for %s...\n", enc);
462                 trace_encoding("reencoded source", path, enc,
463                                re_src, re_src_len);
464
465                 if (!re_src || src_len != re_src_len ||
466                     memcmp(src, re_src, src_len)) {
467                         const char* msg = _("encoding '%s' from %s to %s and "
468                                             "back is not the same");
469                         die(msg, path, enc, default_encoding);
470                 }
471
472                 free(re_src);
473         }
474
475         strbuf_attach(buf, dst, dst_len, dst_len + 1);
476         return 1;
477 }
478
479 static int encode_to_worktree(const char *path, const char *src, size_t src_len,
480                               struct strbuf *buf, const char *enc)
481 {
482         char *dst;
483         int dst_len;
484
485         /*
486          * No encoding is specified or there is nothing to encode.
487          * Tell the caller that the content was not modified.
488          */
489         if (!enc || (src && !src_len))
490                 return 0;
491
492         dst = reencode_string_len(src, src_len, enc, default_encoding,
493                                   &dst_len);
494         if (!dst) {
495                 error("failed to encode '%s' from %s to %s",
496                         path, default_encoding, enc);
497                 return 0;
498         }
499
500         strbuf_attach(buf, dst, dst_len, dst_len + 1);
501         return 1;
502 }
503
504 static int crlf_to_git(const struct index_state *istate,
505                        const char *path, const char *src, size_t len,
506                        struct strbuf *buf,
507                        enum crlf_action crlf_action, int conv_flags)
508 {
509         struct text_stat stats;
510         char *dst;
511         int convert_crlf_into_lf;
512
513         if (crlf_action == CRLF_BINARY ||
514             (src && !len))
515                 return 0;
516
517         /*
518          * If we are doing a dry-run and have no source buffer, there is
519          * nothing to analyze; we must assume we would convert.
520          */
521         if (!buf && !src)
522                 return 1;
523
524         gather_stats(src, len, &stats);
525         /* Optimization: No CRLF? Nothing to convert, regardless. */
526         convert_crlf_into_lf = !!stats.crlf;
527
528         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
529                 if (convert_is_binary(len, &stats))
530                         return 0;
531                 /*
532                  * If the file in the index has any CR in it, do not
533                  * convert.  This is the new safer autocrlf handling,
534                  * unless we want to renormalize in a merge or
535                  * cherry-pick.
536                  */
537                 if ((!(conv_flags & CONV_EOL_RENORMALIZE)) &&
538                     has_crlf_in_index(istate, path))
539                         convert_crlf_into_lf = 0;
540         }
541         if (((conv_flags & CONV_EOL_RNDTRP_WARN) ||
542              ((conv_flags & CONV_EOL_RNDTRP_DIE) && len))) {
543                 struct text_stat new_stats;
544                 memcpy(&new_stats, &stats, sizeof(new_stats));
545                 /* simulate "git add" */
546                 if (convert_crlf_into_lf) {
547                         new_stats.lonelf += new_stats.crlf;
548                         new_stats.crlf = 0;
549                 }
550                 /* simulate "git checkout" */
551                 if (will_convert_lf_to_crlf(len, &new_stats, crlf_action)) {
552                         new_stats.crlf += new_stats.lonelf;
553                         new_stats.lonelf = 0;
554                 }
555                 check_global_conv_flags_eol(path, crlf_action, &stats, &new_stats, conv_flags);
556         }
557         if (!convert_crlf_into_lf)
558                 return 0;
559
560         /*
561          * At this point all of our source analysis is done, and we are sure we
562          * would convert. If we are in dry-run mode, we can give an answer.
563          */
564         if (!buf)
565                 return 1;
566
567         /* only grow if not in place */
568         if (strbuf_avail(buf) + buf->len < len)
569                 strbuf_grow(buf, len - buf->len);
570         dst = buf->buf;
571         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
572                 /*
573                  * If we guessed, we already know we rejected a file with
574                  * lone CR, and we can strip a CR without looking at what
575                  * follow it.
576                  */
577                 do {
578                         unsigned char c = *src++;
579                         if (c != '\r')
580                                 *dst++ = c;
581                 } while (--len);
582         } else {
583                 do {
584                         unsigned char c = *src++;
585                         if (! (c == '\r' && (1 < len && *src == '\n')))
586                                 *dst++ = c;
587                 } while (--len);
588         }
589         strbuf_setlen(buf, dst - buf->buf);
590         return 1;
591 }
592
593 static int crlf_to_worktree(const char *path, const char *src, size_t len,
594                             struct strbuf *buf, enum crlf_action crlf_action)
595 {
596         char *to_free = NULL;
597         struct text_stat stats;
598
599         if (!len || output_eol(crlf_action) != EOL_CRLF)
600                 return 0;
601
602         gather_stats(src, len, &stats);
603         if (!will_convert_lf_to_crlf(len, &stats, crlf_action))
604                 return 0;
605
606         /* are we "faking" in place editing ? */
607         if (src == buf->buf)
608                 to_free = strbuf_detach(buf, NULL);
609
610         strbuf_grow(buf, len + stats.lonelf);
611         for (;;) {
612                 const char *nl = memchr(src, '\n', len);
613                 if (!nl)
614                         break;
615                 if (nl > src && nl[-1] == '\r') {
616                         strbuf_add(buf, src, nl + 1 - src);
617                 } else {
618                         strbuf_add(buf, src, nl - src);
619                         strbuf_addstr(buf, "\r\n");
620                 }
621                 len -= nl + 1 - src;
622                 src  = nl + 1;
623         }
624         strbuf_add(buf, src, len);
625
626         free(to_free);
627         return 1;
628 }
629
630 struct filter_params {
631         const char *src;
632         unsigned long size;
633         int fd;
634         const char *cmd;
635         const char *path;
636 };
637
638 static int filter_buffer_or_fd(int in, int out, void *data)
639 {
640         /*
641          * Spawn cmd and feed the buffer contents through its stdin.
642          */
643         struct child_process child_process = CHILD_PROCESS_INIT;
644         struct filter_params *params = (struct filter_params *)data;
645         int write_err, status;
646         const char *argv[] = { NULL, NULL };
647
648         /* apply % substitution to cmd */
649         struct strbuf cmd = STRBUF_INIT;
650         struct strbuf path = STRBUF_INIT;
651         struct strbuf_expand_dict_entry dict[] = {
652                 { "f", NULL, },
653                 { NULL, NULL, },
654         };
655
656         /* quote the path to preserve spaces, etc. */
657         sq_quote_buf(&path, params->path);
658         dict[0].value = path.buf;
659
660         /* expand all %f with the quoted path */
661         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
662         strbuf_release(&path);
663
664         argv[0] = cmd.buf;
665
666         child_process.argv = argv;
667         child_process.use_shell = 1;
668         child_process.in = -1;
669         child_process.out = out;
670
671         if (start_command(&child_process)) {
672                 strbuf_release(&cmd);
673                 return error("cannot fork to run external filter '%s'", params->cmd);
674         }
675
676         sigchain_push(SIGPIPE, SIG_IGN);
677
678         if (params->src) {
679                 write_err = (write_in_full(child_process.in,
680                                            params->src, params->size) < 0);
681                 if (errno == EPIPE)
682                         write_err = 0;
683         } else {
684                 write_err = copy_fd(params->fd, child_process.in);
685                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
686                         write_err = 0;
687         }
688
689         if (close(child_process.in))
690                 write_err = 1;
691         if (write_err)
692                 error("cannot feed the input to external filter '%s'", params->cmd);
693
694         sigchain_pop(SIGPIPE);
695
696         status = finish_command(&child_process);
697         if (status)
698                 error("external filter '%s' failed %d", params->cmd, status);
699
700         strbuf_release(&cmd);
701         return (write_err || status);
702 }
703
704 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
705                         struct strbuf *dst, const char *cmd)
706 {
707         /*
708          * Create a pipeline to have the command filter the buffer's
709          * contents.
710          *
711          * (child --> cmd) --> us
712          */
713         int err = 0;
714         struct strbuf nbuf = STRBUF_INIT;
715         struct async async;
716         struct filter_params params;
717
718         memset(&async, 0, sizeof(async));
719         async.proc = filter_buffer_or_fd;
720         async.data = &params;
721         async.out = -1;
722         params.src = src;
723         params.size = len;
724         params.fd = fd;
725         params.cmd = cmd;
726         params.path = path;
727
728         fflush(NULL);
729         if (start_async(&async))
730                 return 0;       /* error was already reported */
731
732         if (strbuf_read(&nbuf, async.out, len) < 0) {
733                 err = error("read from external filter '%s' failed", cmd);
734         }
735         if (close(async.out)) {
736                 err = error("read from external filter '%s' failed", cmd);
737         }
738         if (finish_async(&async)) {
739                 err = error("external filter '%s' failed", cmd);
740         }
741
742         if (!err) {
743                 strbuf_swap(dst, &nbuf);
744         }
745         strbuf_release(&nbuf);
746         return !err;
747 }
748
749 #define CAP_CLEAN    (1u<<0)
750 #define CAP_SMUDGE   (1u<<1)
751 #define CAP_DELAY    (1u<<2)
752
753 struct cmd2process {
754         struct subprocess_entry subprocess; /* must be the first member! */
755         unsigned int supported_capabilities;
756 };
757
758 static int subprocess_map_initialized;
759 static struct hashmap subprocess_map;
760
761 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
762 {
763         static int versions[] = {2, 0};
764         static struct subprocess_capability capabilities[] = {
765                 { "clean",  CAP_CLEAN  },
766                 { "smudge", CAP_SMUDGE },
767                 { "delay",  CAP_DELAY  },
768                 { NULL, 0 }
769         };
770         struct cmd2process *entry = (struct cmd2process *)subprocess;
771         return subprocess_handshake(subprocess, "git-filter", versions, NULL,
772                                     capabilities,
773                                     &entry->supported_capabilities);
774 }
775
776 static void handle_filter_error(const struct strbuf *filter_status,
777                                 struct cmd2process *entry,
778                                 const unsigned int wanted_capability) {
779         if (!strcmp(filter_status->buf, "error"))
780                 ; /* The filter signaled a problem with the file. */
781         else if (!strcmp(filter_status->buf, "abort") && wanted_capability) {
782                 /*
783                  * The filter signaled a permanent problem. Don't try to filter
784                  * files with the same command for the lifetime of the current
785                  * Git process.
786                  */
787                  entry->supported_capabilities &= ~wanted_capability;
788         } else {
789                 /*
790                  * Something went wrong with the protocol filter.
791                  * Force shutdown and restart if another blob requires filtering.
792                  */
793                 error("external filter '%s' failed", entry->subprocess.cmd);
794                 subprocess_stop(&subprocess_map, &entry->subprocess);
795                 free(entry);
796         }
797 }
798
799 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
800                                    int fd, struct strbuf *dst, const char *cmd,
801                                    const unsigned int wanted_capability,
802                                    struct delayed_checkout *dco)
803 {
804         int err;
805         int can_delay = 0;
806         struct cmd2process *entry;
807         struct child_process *process;
808         struct strbuf nbuf = STRBUF_INIT;
809         struct strbuf filter_status = STRBUF_INIT;
810         const char *filter_type;
811
812         if (!subprocess_map_initialized) {
813                 subprocess_map_initialized = 1;
814                 hashmap_init(&subprocess_map, cmd2process_cmp, NULL, 0);
815                 entry = NULL;
816         } else {
817                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
818         }
819
820         fflush(NULL);
821
822         if (!entry) {
823                 entry = xmalloc(sizeof(*entry));
824                 entry->supported_capabilities = 0;
825
826                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
827                         free(entry);
828                         return 0;
829                 }
830         }
831         process = &entry->subprocess.process;
832
833         if (!(entry->supported_capabilities & wanted_capability))
834                 return 0;
835
836         if (wanted_capability & CAP_CLEAN)
837                 filter_type = "clean";
838         else if (wanted_capability & CAP_SMUDGE)
839                 filter_type = "smudge";
840         else
841                 die("unexpected filter type");
842
843         sigchain_push(SIGPIPE, SIG_IGN);
844
845         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
846         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
847         if (err)
848                 goto done;
849
850         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
851         if (err) {
852                 error("path name too long for external filter");
853                 goto done;
854         }
855
856         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
857         if (err)
858                 goto done;
859
860         if ((entry->supported_capabilities & CAP_DELAY) &&
861             dco && dco->state == CE_CAN_DELAY) {
862                 can_delay = 1;
863                 err = packet_write_fmt_gently(process->in, "can-delay=1\n");
864                 if (err)
865                         goto done;
866         }
867
868         err = packet_flush_gently(process->in);
869         if (err)
870                 goto done;
871
872         if (fd >= 0)
873                 err = write_packetized_from_fd(fd, process->in);
874         else
875                 err = write_packetized_from_buf(src, len, process->in);
876         if (err)
877                 goto done;
878
879         err = subprocess_read_status(process->out, &filter_status);
880         if (err)
881                 goto done;
882
883         if (can_delay && !strcmp(filter_status.buf, "delayed")) {
884                 string_list_insert(&dco->filters, cmd);
885                 string_list_insert(&dco->paths, path);
886         } else {
887                 /* The filter got the blob and wants to send us a response. */
888                 err = strcmp(filter_status.buf, "success");
889                 if (err)
890                         goto done;
891
892                 err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
893                 if (err)
894                         goto done;
895
896                 err = subprocess_read_status(process->out, &filter_status);
897                 if (err)
898                         goto done;
899
900                 err = strcmp(filter_status.buf, "success");
901         }
902
903 done:
904         sigchain_pop(SIGPIPE);
905
906         if (err)
907                 handle_filter_error(&filter_status, entry, wanted_capability);
908         else
909                 strbuf_swap(dst, &nbuf);
910         strbuf_release(&nbuf);
911         return !err;
912 }
913
914
915 int async_query_available_blobs(const char *cmd, struct string_list *available_paths)
916 {
917         int err;
918         char *line;
919         struct cmd2process *entry;
920         struct child_process *process;
921         struct strbuf filter_status = STRBUF_INIT;
922
923         assert(subprocess_map_initialized);
924         entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
925         if (!entry) {
926                 error("external filter '%s' is not available anymore although "
927                       "not all paths have been filtered", cmd);
928                 return 0;
929         }
930         process = &entry->subprocess.process;
931         sigchain_push(SIGPIPE, SIG_IGN);
932
933         err = packet_write_fmt_gently(
934                 process->in, "command=list_available_blobs\n");
935         if (err)
936                 goto done;
937
938         err = packet_flush_gently(process->in);
939         if (err)
940                 goto done;
941
942         while ((line = packet_read_line(process->out, NULL))) {
943                 const char *path;
944                 if (skip_prefix(line, "pathname=", &path))
945                         string_list_insert(available_paths, xstrdup(path));
946                 else
947                         ; /* ignore unknown keys */
948         }
949
950         err = subprocess_read_status(process->out, &filter_status);
951         if (err)
952                 goto done;
953
954         err = strcmp(filter_status.buf, "success");
955
956 done:
957         sigchain_pop(SIGPIPE);
958
959         if (err)
960                 handle_filter_error(&filter_status, entry, 0);
961         return !err;
962 }
963
964 static struct convert_driver {
965         const char *name;
966         struct convert_driver *next;
967         const char *smudge;
968         const char *clean;
969         const char *process;
970         int required;
971 } *user_convert, **user_convert_tail;
972
973 static int apply_filter(const char *path, const char *src, size_t len,
974                         int fd, struct strbuf *dst, struct convert_driver *drv,
975                         const unsigned int wanted_capability,
976                         struct delayed_checkout *dco)
977 {
978         const char *cmd = NULL;
979
980         if (!drv)
981                 return 0;
982
983         if (!dst)
984                 return 1;
985
986         if ((wanted_capability & CAP_CLEAN) && !drv->process && drv->clean)
987                 cmd = drv->clean;
988         else if ((wanted_capability & CAP_SMUDGE) && !drv->process && drv->smudge)
989                 cmd = drv->smudge;
990
991         if (cmd && *cmd)
992                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
993         else if (drv->process && *drv->process)
994                 return apply_multi_file_filter(path, src, len, fd, dst,
995                         drv->process, wanted_capability, dco);
996
997         return 0;
998 }
999
1000 static int read_convert_config(const char *var, const char *value, void *cb)
1001 {
1002         const char *key, *name;
1003         int namelen;
1004         struct convert_driver *drv;
1005
1006         /*
1007          * External conversion drivers are configured using
1008          * "filter.<name>.variable".
1009          */
1010         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
1011                 return 0;
1012         for (drv = user_convert; drv; drv = drv->next)
1013                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
1014                         break;
1015         if (!drv) {
1016                 drv = xcalloc(1, sizeof(struct convert_driver));
1017                 drv->name = xmemdupz(name, namelen);
1018                 *user_convert_tail = drv;
1019                 user_convert_tail = &(drv->next);
1020         }
1021
1022         /*
1023          * filter.<name>.smudge and filter.<name>.clean specifies
1024          * the command line:
1025          *
1026          *      command-line
1027          *
1028          * The command-line will not be interpolated in any way.
1029          */
1030
1031         if (!strcmp("smudge", key))
1032                 return git_config_string(&drv->smudge, var, value);
1033
1034         if (!strcmp("clean", key))
1035                 return git_config_string(&drv->clean, var, value);
1036
1037         if (!strcmp("process", key))
1038                 return git_config_string(&drv->process, var, value);
1039
1040         if (!strcmp("required", key)) {
1041                 drv->required = git_config_bool(var, value);
1042                 return 0;
1043         }
1044
1045         return 0;
1046 }
1047
1048 static int count_ident(const char *cp, unsigned long size)
1049 {
1050         /*
1051          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
1052          */
1053         int cnt = 0;
1054         char ch;
1055
1056         while (size) {
1057                 ch = *cp++;
1058                 size--;
1059                 if (ch != '$')
1060                         continue;
1061                 if (size < 3)
1062                         break;
1063                 if (memcmp("Id", cp, 2))
1064                         continue;
1065                 ch = cp[2];
1066                 cp += 3;
1067                 size -= 3;
1068                 if (ch == '$')
1069                         cnt++; /* $Id$ */
1070                 if (ch != ':')
1071                         continue;
1072
1073                 /*
1074                  * "$Id: ... "; scan up to the closing dollar sign and discard.
1075                  */
1076                 while (size) {
1077                         ch = *cp++;
1078                         size--;
1079                         if (ch == '$') {
1080                                 cnt++;
1081                                 break;
1082                         }
1083                         if (ch == '\n')
1084                                 break;
1085                 }
1086         }
1087         return cnt;
1088 }
1089
1090 static int ident_to_git(const char *path, const char *src, size_t len,
1091                         struct strbuf *buf, int ident)
1092 {
1093         char *dst, *dollar;
1094
1095         if (!ident || (src && !count_ident(src, len)))
1096                 return 0;
1097
1098         if (!buf)
1099                 return 1;
1100
1101         /* only grow if not in place */
1102         if (strbuf_avail(buf) + buf->len < len)
1103                 strbuf_grow(buf, len - buf->len);
1104         dst = buf->buf;
1105         for (;;) {
1106                 dollar = memchr(src, '$', len);
1107                 if (!dollar)
1108                         break;
1109                 memmove(dst, src, dollar + 1 - src);
1110                 dst += dollar + 1 - src;
1111                 len -= dollar + 1 - src;
1112                 src  = dollar + 1;
1113
1114                 if (len > 3 && !memcmp(src, "Id:", 3)) {
1115                         dollar = memchr(src + 3, '$', len - 3);
1116                         if (!dollar)
1117                                 break;
1118                         if (memchr(src + 3, '\n', dollar - src - 3)) {
1119                                 /* Line break before the next dollar. */
1120                                 continue;
1121                         }
1122
1123                         memcpy(dst, "Id$", 3);
1124                         dst += 3;
1125                         len -= dollar + 1 - src;
1126                         src  = dollar + 1;
1127                 }
1128         }
1129         memmove(dst, src, len);
1130         strbuf_setlen(buf, dst + len - buf->buf);
1131         return 1;
1132 }
1133
1134 static int ident_to_worktree(const char *path, const char *src, size_t len,
1135                              struct strbuf *buf, int ident)
1136 {
1137         struct object_id oid;
1138         char *to_free = NULL, *dollar, *spc;
1139         int cnt;
1140
1141         if (!ident)
1142                 return 0;
1143
1144         cnt = count_ident(src, len);
1145         if (!cnt)
1146                 return 0;
1147
1148         /* are we "faking" in place editing ? */
1149         if (src == buf->buf)
1150                 to_free = strbuf_detach(buf, NULL);
1151         hash_object_file(src, len, "blob", &oid);
1152
1153         strbuf_grow(buf, len + cnt * (the_hash_algo->hexsz + 3));
1154         for (;;) {
1155                 /* step 1: run to the next '$' */
1156                 dollar = memchr(src, '$', len);
1157                 if (!dollar)
1158                         break;
1159                 strbuf_add(buf, src, dollar + 1 - src);
1160                 len -= dollar + 1 - src;
1161                 src  = dollar + 1;
1162
1163                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
1164                 if (len < 3 || memcmp("Id", src, 2))
1165                         continue;
1166
1167                 /* step 3: skip over Id$ or Id:xxxxx$ */
1168                 if (src[2] == '$') {
1169                         src += 3;
1170                         len -= 3;
1171                 } else if (src[2] == ':') {
1172                         /*
1173                          * It's possible that an expanded Id has crept its way into the
1174                          * repository, we cope with that by stripping the expansion out.
1175                          * This is probably not a good idea, since it will cause changes
1176                          * on checkout, which won't go away by stash, but let's keep it
1177                          * for git-style ids.
1178                          */
1179                         dollar = memchr(src + 3, '$', len - 3);
1180                         if (!dollar) {
1181                                 /* incomplete keyword, no more '$', so just quit the loop */
1182                                 break;
1183                         }
1184
1185                         if (memchr(src + 3, '\n', dollar - src - 3)) {
1186                                 /* Line break before the next dollar. */
1187                                 continue;
1188                         }
1189
1190                         spc = memchr(src + 4, ' ', dollar - src - 4);
1191                         if (spc && spc < dollar-1) {
1192                                 /* There are spaces in unexpected places.
1193                                  * This is probably an id from some other
1194                                  * versioning system. Keep it for now.
1195                                  */
1196                                 continue;
1197                         }
1198
1199                         len -= dollar + 1 - src;
1200                         src  = dollar + 1;
1201                 } else {
1202                         /* it wasn't a "Id$" or "Id:xxxx$" */
1203                         continue;
1204                 }
1205
1206                 /* step 4: substitute */
1207                 strbuf_addstr(buf, "Id: ");
1208                 strbuf_addstr(buf, oid_to_hex(&oid));
1209                 strbuf_addstr(buf, " $");
1210         }
1211         strbuf_add(buf, src, len);
1212
1213         free(to_free);
1214         return 1;
1215 }
1216
1217 static const char *git_path_check_encoding(struct attr_check_item *check)
1218 {
1219         const char *value = check->value;
1220
1221         if (ATTR_UNSET(value) || !strlen(value))
1222                 return NULL;
1223
1224         if (ATTR_TRUE(value) || ATTR_FALSE(value)) {
1225                 die(_("true/false are no valid working-tree-encodings"));
1226         }
1227
1228         /* Don't encode to the default encoding */
1229         if (same_encoding(value, default_encoding))
1230                 return NULL;
1231
1232         return value;
1233 }
1234
1235 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
1236 {
1237         const char *value = check->value;
1238
1239         if (ATTR_TRUE(value))
1240                 return CRLF_TEXT;
1241         else if (ATTR_FALSE(value))
1242                 return CRLF_BINARY;
1243         else if (ATTR_UNSET(value))
1244                 ;
1245         else if (!strcmp(value, "input"))
1246                 return CRLF_TEXT_INPUT;
1247         else if (!strcmp(value, "auto"))
1248                 return CRLF_AUTO;
1249         return CRLF_UNDEFINED;
1250 }
1251
1252 static enum eol git_path_check_eol(struct attr_check_item *check)
1253 {
1254         const char *value = check->value;
1255
1256         if (ATTR_UNSET(value))
1257                 ;
1258         else if (!strcmp(value, "lf"))
1259                 return EOL_LF;
1260         else if (!strcmp(value, "crlf"))
1261                 return EOL_CRLF;
1262         return EOL_UNSET;
1263 }
1264
1265 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
1266 {
1267         const char *value = check->value;
1268         struct convert_driver *drv;
1269
1270         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
1271                 return NULL;
1272         for (drv = user_convert; drv; drv = drv->next)
1273                 if (!strcmp(value, drv->name))
1274                         return drv;
1275         return NULL;
1276 }
1277
1278 static int git_path_check_ident(struct attr_check_item *check)
1279 {
1280         const char *value = check->value;
1281
1282         return !!ATTR_TRUE(value);
1283 }
1284
1285 struct conv_attrs {
1286         struct convert_driver *drv;
1287         enum crlf_action attr_action; /* What attr says */
1288         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
1289         int ident;
1290         const char *working_tree_encoding; /* Supported encoding or default encoding if NULL */
1291 };
1292
1293 static void convert_attrs(struct conv_attrs *ca, const char *path)
1294 {
1295         static struct attr_check *check;
1296
1297         if (!check) {
1298                 check = attr_check_initl("crlf", "ident", "filter",
1299                                          "eol", "text", "working-tree-encoding",
1300                                          NULL);
1301                 user_convert_tail = &user_convert;
1302                 git_config(read_convert_config, NULL);
1303         }
1304
1305         if (!git_check_attr(path, check)) {
1306                 struct attr_check_item *ccheck = check->items;
1307                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
1308                 if (ca->crlf_action == CRLF_UNDEFINED)
1309                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
1310                 ca->ident = git_path_check_ident(ccheck + 1);
1311                 ca->drv = git_path_check_convert(ccheck + 2);
1312                 if (ca->crlf_action != CRLF_BINARY) {
1313                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
1314                         if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1315                                 ca->crlf_action = CRLF_AUTO_INPUT;
1316                         else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1317                                 ca->crlf_action = CRLF_AUTO_CRLF;
1318                         else if (eol_attr == EOL_LF)
1319                                 ca->crlf_action = CRLF_TEXT_INPUT;
1320                         else if (eol_attr == EOL_CRLF)
1321                                 ca->crlf_action = CRLF_TEXT_CRLF;
1322                 }
1323                 ca->working_tree_encoding = git_path_check_encoding(ccheck + 5);
1324         } else {
1325                 ca->drv = NULL;
1326                 ca->crlf_action = CRLF_UNDEFINED;
1327                 ca->ident = 0;
1328         }
1329
1330         /* Save attr and make a decision for action */
1331         ca->attr_action = ca->crlf_action;
1332         if (ca->crlf_action == CRLF_TEXT)
1333                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1334         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1335                 ca->crlf_action = CRLF_BINARY;
1336         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1337                 ca->crlf_action = CRLF_AUTO_CRLF;
1338         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1339                 ca->crlf_action = CRLF_AUTO_INPUT;
1340 }
1341
1342 int would_convert_to_git_filter_fd(const char *path)
1343 {
1344         struct conv_attrs ca;
1345
1346         convert_attrs(&ca, path);
1347         if (!ca.drv)
1348                 return 0;
1349
1350         /*
1351          * Apply a filter to an fd only if the filter is required to succeed.
1352          * We must die if the filter fails, because the original data before
1353          * filtering is not available.
1354          */
1355         if (!ca.drv->required)
1356                 return 0;
1357
1358         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN, NULL);
1359 }
1360
1361 const char *get_convert_attr_ascii(const char *path)
1362 {
1363         struct conv_attrs ca;
1364
1365         convert_attrs(&ca, path);
1366         switch (ca.attr_action) {
1367         case CRLF_UNDEFINED:
1368                 return "";
1369         case CRLF_BINARY:
1370                 return "-text";
1371         case CRLF_TEXT:
1372                 return "text";
1373         case CRLF_TEXT_INPUT:
1374                 return "text eol=lf";
1375         case CRLF_TEXT_CRLF:
1376                 return "text eol=crlf";
1377         case CRLF_AUTO:
1378                 return "text=auto";
1379         case CRLF_AUTO_CRLF:
1380                 return "text=auto eol=crlf";
1381         case CRLF_AUTO_INPUT:
1382                 return "text=auto eol=lf";
1383         }
1384         return "";
1385 }
1386
1387 int convert_to_git(const struct index_state *istate,
1388                    const char *path, const char *src, size_t len,
1389                    struct strbuf *dst, int conv_flags)
1390 {
1391         int ret = 0;
1392         struct conv_attrs ca;
1393
1394         convert_attrs(&ca, path);
1395
1396         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN, NULL);
1397         if (!ret && ca.drv && ca.drv->required)
1398                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1399
1400         if (ret && dst) {
1401                 src = dst->buf;
1402                 len = dst->len;
1403         }
1404
1405         ret |= encode_to_git(path, src, len, dst, ca.working_tree_encoding, conv_flags);
1406         if (ret && dst) {
1407                 src = dst->buf;
1408                 len = dst->len;
1409         }
1410
1411         if (!(conv_flags & CONV_EOL_KEEP_CRLF)) {
1412                 ret |= crlf_to_git(istate, path, src, len, dst, ca.crlf_action, conv_flags);
1413                 if (ret && dst) {
1414                         src = dst->buf;
1415                         len = dst->len;
1416                 }
1417         }
1418         return ret | ident_to_git(path, src, len, dst, ca.ident);
1419 }
1420
1421 void convert_to_git_filter_fd(const struct index_state *istate,
1422                               const char *path, int fd, struct strbuf *dst,
1423                               int conv_flags)
1424 {
1425         struct conv_attrs ca;
1426         convert_attrs(&ca, path);
1427
1428         assert(ca.drv);
1429         assert(ca.drv->clean || ca.drv->process);
1430
1431         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN, NULL))
1432                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1433
1434         encode_to_git(path, dst->buf, dst->len, dst, ca.working_tree_encoding, conv_flags);
1435         crlf_to_git(istate, path, dst->buf, dst->len, dst, ca.crlf_action, conv_flags);
1436         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
1437 }
1438
1439 static int convert_to_working_tree_internal(const char *path, const char *src,
1440                                             size_t len, struct strbuf *dst,
1441                                             int normalizing, struct delayed_checkout *dco)
1442 {
1443         int ret = 0, ret_filter = 0;
1444         struct conv_attrs ca;
1445
1446         convert_attrs(&ca, path);
1447
1448         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
1449         if (ret) {
1450                 src = dst->buf;
1451                 len = dst->len;
1452         }
1453         /*
1454          * CRLF conversion can be skipped if normalizing, unless there
1455          * is a smudge or process filter (even if the process filter doesn't
1456          * support smudge).  The filters might expect CRLFs.
1457          */
1458         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1459                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
1460                 if (ret) {
1461                         src = dst->buf;
1462                         len = dst->len;
1463                 }
1464         }
1465
1466         ret |= encode_to_worktree(path, src, len, dst, ca.working_tree_encoding);
1467         if (ret) {
1468                 src = dst->buf;
1469                 len = dst->len;
1470         }
1471
1472         ret_filter = apply_filter(
1473                 path, src, len, -1, dst, ca.drv, CAP_SMUDGE, dco);
1474         if (!ret_filter && ca.drv && ca.drv->required)
1475                 die("%s: smudge filter %s failed", path, ca.drv->name);
1476
1477         return ret | ret_filter;
1478 }
1479
1480 int async_convert_to_working_tree(const char *path, const char *src,
1481                                   size_t len, struct strbuf *dst,
1482                                   void *dco)
1483 {
1484         return convert_to_working_tree_internal(path, src, len, dst, 0, dco);
1485 }
1486
1487 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
1488 {
1489         return convert_to_working_tree_internal(path, src, len, dst, 0, NULL);
1490 }
1491
1492 int renormalize_buffer(const struct index_state *istate, const char *path,
1493                        const char *src, size_t len, struct strbuf *dst)
1494 {
1495         int ret = convert_to_working_tree_internal(path, src, len, dst, 1, NULL);
1496         if (ret) {
1497                 src = dst->buf;
1498                 len = dst->len;
1499         }
1500         return ret | convert_to_git(istate, path, src, len, dst, CONV_EOL_RENORMALIZE);
1501 }
1502
1503 /*****************************************************************
1504  *
1505  * Streaming conversion support
1506  *
1507  *****************************************************************/
1508
1509 typedef int (*filter_fn)(struct stream_filter *,
1510                          const char *input, size_t *isize_p,
1511                          char *output, size_t *osize_p);
1512 typedef void (*free_fn)(struct stream_filter *);
1513
1514 struct stream_filter_vtbl {
1515         filter_fn filter;
1516         free_fn free;
1517 };
1518
1519 struct stream_filter {
1520         struct stream_filter_vtbl *vtbl;
1521 };
1522
1523 static int null_filter_fn(struct stream_filter *filter,
1524                           const char *input, size_t *isize_p,
1525                           char *output, size_t *osize_p)
1526 {
1527         size_t count;
1528
1529         if (!input)
1530                 return 0; /* we do not keep any states */
1531         count = *isize_p;
1532         if (*osize_p < count)
1533                 count = *osize_p;
1534         if (count) {
1535                 memmove(output, input, count);
1536                 *isize_p -= count;
1537                 *osize_p -= count;
1538         }
1539         return 0;
1540 }
1541
1542 static void null_free_fn(struct stream_filter *filter)
1543 {
1544         ; /* nothing -- null instances are shared */
1545 }
1546
1547 static struct stream_filter_vtbl null_vtbl = {
1548         null_filter_fn,
1549         null_free_fn,
1550 };
1551
1552 static struct stream_filter null_filter_singleton = {
1553         &null_vtbl,
1554 };
1555
1556 int is_null_stream_filter(struct stream_filter *filter)
1557 {
1558         return filter == &null_filter_singleton;
1559 }
1560
1561
1562 /*
1563  * LF-to-CRLF filter
1564  */
1565
1566 struct lf_to_crlf_filter {
1567         struct stream_filter filter;
1568         unsigned has_held:1;
1569         char held;
1570 };
1571
1572 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1573                                 const char *input, size_t *isize_p,
1574                                 char *output, size_t *osize_p)
1575 {
1576         size_t count, o = 0;
1577         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1578
1579         /*
1580          * We may be holding onto the CR to see if it is followed by a
1581          * LF, in which case we would need to go to the main loop.
1582          * Otherwise, just emit it to the output stream.
1583          */
1584         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1585                 output[o++] = lf_to_crlf->held;
1586                 lf_to_crlf->has_held = 0;
1587         }
1588
1589         /* We are told to drain */
1590         if (!input) {
1591                 *osize_p -= o;
1592                 return 0;
1593         }
1594
1595         count = *isize_p;
1596         if (count || lf_to_crlf->has_held) {
1597                 size_t i;
1598                 int was_cr = 0;
1599
1600                 if (lf_to_crlf->has_held) {
1601                         was_cr = 1;
1602                         lf_to_crlf->has_held = 0;
1603                 }
1604
1605                 for (i = 0; o < *osize_p && i < count; i++) {
1606                         char ch = input[i];
1607
1608                         if (ch == '\n') {
1609                                 output[o++] = '\r';
1610                         } else if (was_cr) {
1611                                 /*
1612                                  * Previous round saw CR and it is not followed
1613                                  * by a LF; emit the CR before processing the
1614                                  * current character.
1615                                  */
1616                                 output[o++] = '\r';
1617                         }
1618
1619                         /*
1620                          * We may have consumed the last output slot,
1621                          * in which case we need to break out of this
1622                          * loop; hold the current character before
1623                          * returning.
1624                          */
1625                         if (*osize_p <= o) {
1626                                 lf_to_crlf->has_held = 1;
1627                                 lf_to_crlf->held = ch;
1628                                 continue; /* break but increment i */
1629                         }
1630
1631                         if (ch == '\r') {
1632                                 was_cr = 1;
1633                                 continue;
1634                         }
1635
1636                         was_cr = 0;
1637                         output[o++] = ch;
1638                 }
1639
1640                 *osize_p -= o;
1641                 *isize_p -= i;
1642
1643                 if (!lf_to_crlf->has_held && was_cr) {
1644                         lf_to_crlf->has_held = 1;
1645                         lf_to_crlf->held = '\r';
1646                 }
1647         }
1648         return 0;
1649 }
1650
1651 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1652 {
1653         free(filter);
1654 }
1655
1656 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1657         lf_to_crlf_filter_fn,
1658         lf_to_crlf_free_fn,
1659 };
1660
1661 static struct stream_filter *lf_to_crlf_filter(void)
1662 {
1663         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1664
1665         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1666         return (struct stream_filter *)lf_to_crlf;
1667 }
1668
1669 /*
1670  * Cascade filter
1671  */
1672 #define FILTER_BUFFER 1024
1673 struct cascade_filter {
1674         struct stream_filter filter;
1675         struct stream_filter *one;
1676         struct stream_filter *two;
1677         char buf[FILTER_BUFFER];
1678         int end, ptr;
1679 };
1680
1681 static int cascade_filter_fn(struct stream_filter *filter,
1682                              const char *input, size_t *isize_p,
1683                              char *output, size_t *osize_p)
1684 {
1685         struct cascade_filter *cas = (struct cascade_filter *) filter;
1686         size_t filled = 0;
1687         size_t sz = *osize_p;
1688         size_t to_feed, remaining;
1689
1690         /*
1691          * input -- (one) --> buf -- (two) --> output
1692          */
1693         while (filled < sz) {
1694                 remaining = sz - filled;
1695
1696                 /* do we already have something to feed two with? */
1697                 if (cas->ptr < cas->end) {
1698                         to_feed = cas->end - cas->ptr;
1699                         if (stream_filter(cas->two,
1700                                           cas->buf + cas->ptr, &to_feed,
1701                                           output + filled, &remaining))
1702                                 return -1;
1703                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1704                         filled = sz - remaining;
1705                         continue;
1706                 }
1707
1708                 /* feed one from upstream and have it emit into our buffer */
1709                 to_feed = input ? *isize_p : 0;
1710                 if (input && !to_feed)
1711                         break;
1712                 remaining = sizeof(cas->buf);
1713                 if (stream_filter(cas->one,
1714                                   input, &to_feed,
1715                                   cas->buf, &remaining))
1716                         return -1;
1717                 cas->end = sizeof(cas->buf) - remaining;
1718                 cas->ptr = 0;
1719                 if (input) {
1720                         size_t fed = *isize_p - to_feed;
1721                         *isize_p -= fed;
1722                         input += fed;
1723                 }
1724
1725                 /* do we know that we drained one completely? */
1726                 if (input || cas->end)
1727                         continue;
1728
1729                 /* tell two to drain; we have nothing more to give it */
1730                 to_feed = 0;
1731                 remaining = sz - filled;
1732                 if (stream_filter(cas->two,
1733                                   NULL, &to_feed,
1734                                   output + filled, &remaining))
1735                         return -1;
1736                 if (remaining == (sz - filled))
1737                         break; /* completely drained two */
1738                 filled = sz - remaining;
1739         }
1740         *osize_p -= filled;
1741         return 0;
1742 }
1743
1744 static void cascade_free_fn(struct stream_filter *filter)
1745 {
1746         struct cascade_filter *cas = (struct cascade_filter *)filter;
1747         free_stream_filter(cas->one);
1748         free_stream_filter(cas->two);
1749         free(filter);
1750 }
1751
1752 static struct stream_filter_vtbl cascade_vtbl = {
1753         cascade_filter_fn,
1754         cascade_free_fn,
1755 };
1756
1757 static struct stream_filter *cascade_filter(struct stream_filter *one,
1758                                             struct stream_filter *two)
1759 {
1760         struct cascade_filter *cascade;
1761
1762         if (!one || is_null_stream_filter(one))
1763                 return two;
1764         if (!two || is_null_stream_filter(two))
1765                 return one;
1766
1767         cascade = xmalloc(sizeof(*cascade));
1768         cascade->one = one;
1769         cascade->two = two;
1770         cascade->end = cascade->ptr = 0;
1771         cascade->filter.vtbl = &cascade_vtbl;
1772         return (struct stream_filter *)cascade;
1773 }
1774
1775 /*
1776  * ident filter
1777  */
1778 #define IDENT_DRAINING (-1)
1779 #define IDENT_SKIPPING (-2)
1780 struct ident_filter {
1781         struct stream_filter filter;
1782         struct strbuf left;
1783         int state;
1784         char ident[GIT_MAX_HEXSZ + 5]; /* ": x40 $" */
1785 };
1786
1787 static int is_foreign_ident(const char *str)
1788 {
1789         int i;
1790
1791         if (!skip_prefix(str, "$Id: ", &str))
1792                 return 0;
1793         for (i = 0; str[i]; i++) {
1794                 if (isspace(str[i]) && str[i+1] != '$')
1795                         return 1;
1796         }
1797         return 0;
1798 }
1799
1800 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1801 {
1802         size_t to_drain = ident->left.len;
1803
1804         if (*osize_p < to_drain)
1805                 to_drain = *osize_p;
1806         if (to_drain) {
1807                 memcpy(*output_p, ident->left.buf, to_drain);
1808                 strbuf_remove(&ident->left, 0, to_drain);
1809                 *output_p += to_drain;
1810                 *osize_p -= to_drain;
1811         }
1812         if (!ident->left.len)
1813                 ident->state = 0;
1814 }
1815
1816 static int ident_filter_fn(struct stream_filter *filter,
1817                            const char *input, size_t *isize_p,
1818                            char *output, size_t *osize_p)
1819 {
1820         struct ident_filter *ident = (struct ident_filter *)filter;
1821         static const char head[] = "$Id";
1822
1823         if (!input) {
1824                 /* drain upon eof */
1825                 switch (ident->state) {
1826                 default:
1827                         strbuf_add(&ident->left, head, ident->state);
1828                         /* fallthrough */
1829                 case IDENT_SKIPPING:
1830                         /* fallthrough */
1831                 case IDENT_DRAINING:
1832                         ident_drain(ident, &output, osize_p);
1833                 }
1834                 return 0;
1835         }
1836
1837         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1838                 int ch;
1839
1840                 if (ident->state == IDENT_DRAINING) {
1841                         ident_drain(ident, &output, osize_p);
1842                         if (!*osize_p)
1843                                 break;
1844                         continue;
1845                 }
1846
1847                 ch = *(input++);
1848                 (*isize_p)--;
1849
1850                 if (ident->state == IDENT_SKIPPING) {
1851                         /*
1852                          * Skipping until '$' or LF, but keeping them
1853                          * in case it is a foreign ident.
1854                          */
1855                         strbuf_addch(&ident->left, ch);
1856                         if (ch != '\n' && ch != '$')
1857                                 continue;
1858                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1859                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1860                                 strbuf_addstr(&ident->left, ident->ident);
1861                         }
1862                         ident->state = IDENT_DRAINING;
1863                         continue;
1864                 }
1865
1866                 if (ident->state < sizeof(head) &&
1867                     head[ident->state] == ch) {
1868                         ident->state++;
1869                         continue;
1870                 }
1871
1872                 if (ident->state)
1873                         strbuf_add(&ident->left, head, ident->state);
1874                 if (ident->state == sizeof(head) - 1) {
1875                         if (ch != ':' && ch != '$') {
1876                                 strbuf_addch(&ident->left, ch);
1877                                 ident->state = 0;
1878                                 continue;
1879                         }
1880
1881                         if (ch == ':') {
1882                                 strbuf_addch(&ident->left, ch);
1883                                 ident->state = IDENT_SKIPPING;
1884                         } else {
1885                                 strbuf_addstr(&ident->left, ident->ident);
1886                                 ident->state = IDENT_DRAINING;
1887                         }
1888                         continue;
1889                 }
1890
1891                 strbuf_addch(&ident->left, ch);
1892                 ident->state = IDENT_DRAINING;
1893         }
1894         return 0;
1895 }
1896
1897 static void ident_free_fn(struct stream_filter *filter)
1898 {
1899         struct ident_filter *ident = (struct ident_filter *)filter;
1900         strbuf_release(&ident->left);
1901         free(filter);
1902 }
1903
1904 static struct stream_filter_vtbl ident_vtbl = {
1905         ident_filter_fn,
1906         ident_free_fn,
1907 };
1908
1909 static struct stream_filter *ident_filter(const struct object_id *oid)
1910 {
1911         struct ident_filter *ident = xmalloc(sizeof(*ident));
1912
1913         xsnprintf(ident->ident, sizeof(ident->ident),
1914                   ": %s $", oid_to_hex(oid));
1915         strbuf_init(&ident->left, 0);
1916         ident->filter.vtbl = &ident_vtbl;
1917         ident->state = 0;
1918         return (struct stream_filter *)ident;
1919 }
1920
1921 /*
1922  * Return an appropriately constructed filter for the path, or NULL if
1923  * the contents cannot be filtered without reading the whole thing
1924  * in-core.
1925  *
1926  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1927  * large binary blob you would want us not to slurp into the memory!
1928  */
1929 struct stream_filter *get_stream_filter(const char *path, const struct object_id *oid)
1930 {
1931         struct conv_attrs ca;
1932         struct stream_filter *filter = NULL;
1933
1934         convert_attrs(&ca, path);
1935         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1936                 return NULL;
1937
1938         if (ca.working_tree_encoding)
1939                 return NULL;
1940
1941         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1942                 return NULL;
1943
1944         if (ca.ident)
1945                 filter = ident_filter(oid);
1946
1947         if (output_eol(ca.crlf_action) == EOL_CRLF)
1948                 filter = cascade_filter(filter, lf_to_crlf_filter());
1949         else
1950                 filter = cascade_filter(filter, &null_filter_singleton);
1951
1952         return filter;
1953 }
1954
1955 void free_stream_filter(struct stream_filter *filter)
1956 {
1957         filter->vtbl->free(filter);
1958 }
1959
1960 int stream_filter(struct stream_filter *filter,
1961                   const char *input, size_t *isize_p,
1962                   char *output, size_t *osize_p)
1963 {
1964         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1965 }