Merge branch 'bb/unicode-12.1-reiwa' into maint
[git] / convert.c
1 #include "cache.h"
2 #include "config.h"
3 #include "object-store.h"
4 #include "attr.h"
5 #include "run-command.h"
6 #include "quote.h"
7 #include "sigchain.h"
8 #include "pkt-line.h"
9 #include "sub-process.h"
10 #include "utf8.h"
11
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(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(&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"), (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(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(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, "| \033[2m%2i:\033[0m %2x \033[2m%c\033[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         size_t 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                 size_t 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         size_t 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(&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(&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 *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(&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'"),
674                              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'"),
694                       params->cmd);
695
696         sigchain_pop(SIGPIPE);
697
698         status = finish_command(&child_process);
699         if (status)
700                 error(_("external filter '%s' failed %d"), params->cmd, status);
701
702         strbuf_release(&cmd);
703         return (write_err || status);
704 }
705
706 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
707                                     struct strbuf *dst, const char *cmd)
708 {
709         /*
710          * Create a pipeline to have the command filter the buffer's
711          * contents.
712          *
713          * (child --> cmd) --> us
714          */
715         int err = 0;
716         struct strbuf nbuf = STRBUF_INIT;
717         struct async async;
718         struct filter_params params;
719
720         memset(&async, 0, sizeof(async));
721         async.proc = filter_buffer_or_fd;
722         async.data = &params;
723         async.out = -1;
724         params.src = src;
725         params.size = len;
726         params.fd = fd;
727         params.cmd = cmd;
728         params.path = path;
729
730         fflush(NULL);
731         if (start_async(&async))
732                 return 0;       /* error was already reported */
733
734         if (strbuf_read(&nbuf, async.out, 0) < 0) {
735                 err = error(_("read from external filter '%s' failed"), cmd);
736         }
737         if (close(async.out)) {
738                 err = error(_("read from external filter '%s' failed"), cmd);
739         }
740         if (finish_async(&async)) {
741                 err = error(_("external filter '%s' failed"), cmd);
742         }
743
744         if (!err) {
745                 strbuf_swap(dst, &nbuf);
746         }
747         strbuf_release(&nbuf);
748         return !err;
749 }
750
751 #define CAP_CLEAN    (1u<<0)
752 #define CAP_SMUDGE   (1u<<1)
753 #define CAP_DELAY    (1u<<2)
754
755 struct cmd2process {
756         struct subprocess_entry subprocess; /* must be the first member! */
757         unsigned int supported_capabilities;
758 };
759
760 static int subprocess_map_initialized;
761 static struct hashmap subprocess_map;
762
763 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
764 {
765         static int versions[] = {2, 0};
766         static struct subprocess_capability capabilities[] = {
767                 { "clean",  CAP_CLEAN  },
768                 { "smudge", CAP_SMUDGE },
769                 { "delay",  CAP_DELAY  },
770                 { NULL, 0 }
771         };
772         struct cmd2process *entry = (struct cmd2process *)subprocess;
773         return subprocess_handshake(subprocess, "git-filter", versions, NULL,
774                                     capabilities,
775                                     &entry->supported_capabilities);
776 }
777
778 static void handle_filter_error(const struct strbuf *filter_status,
779                                 struct cmd2process *entry,
780                                 const unsigned int wanted_capability)
781 {
782         if (!strcmp(filter_status->buf, "error"))
783                 ; /* The filter signaled a problem with the file. */
784         else if (!strcmp(filter_status->buf, "abort") && wanted_capability) {
785                 /*
786                  * The filter signaled a permanent problem. Don't try to filter
787                  * files with the same command for the lifetime of the current
788                  * Git process.
789                  */
790                  entry->supported_capabilities &= ~wanted_capability;
791         } else {
792                 /*
793                  * Something went wrong with the protocol filter.
794                  * Force shutdown and restart if another blob requires filtering.
795                  */
796                 error(_("external filter '%s' failed"), entry->subprocess.cmd);
797                 subprocess_stop(&subprocess_map, &entry->subprocess);
798                 free(entry);
799         }
800 }
801
802 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
803                                    int fd, struct strbuf *dst, const char *cmd,
804                                    const unsigned int wanted_capability,
805                                    struct delayed_checkout *dco)
806 {
807         int err;
808         int can_delay = 0;
809         struct cmd2process *entry;
810         struct child_process *process;
811         struct strbuf nbuf = STRBUF_INIT;
812         struct strbuf filter_status = STRBUF_INIT;
813         const char *filter_type;
814
815         if (!subprocess_map_initialized) {
816                 subprocess_map_initialized = 1;
817                 hashmap_init(&subprocess_map, cmd2process_cmp, NULL, 0);
818                 entry = NULL;
819         } else {
820                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
821         }
822
823         fflush(NULL);
824
825         if (!entry) {
826                 entry = xmalloc(sizeof(*entry));
827                 entry->supported_capabilities = 0;
828
829                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
830                         free(entry);
831                         return 0;
832                 }
833         }
834         process = &entry->subprocess.process;
835
836         if (!(entry->supported_capabilities & wanted_capability))
837                 return 0;
838
839         if (wanted_capability & CAP_CLEAN)
840                 filter_type = "clean";
841         else if (wanted_capability & CAP_SMUDGE)
842                 filter_type = "smudge";
843         else
844                 die(_("unexpected filter type"));
845
846         sigchain_push(SIGPIPE, SIG_IGN);
847
848         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
849         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
850         if (err)
851                 goto done;
852
853         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
854         if (err) {
855                 error(_("path name too long for external filter"));
856                 goto done;
857         }
858
859         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
860         if (err)
861                 goto done;
862
863         if ((entry->supported_capabilities & CAP_DELAY) &&
864             dco && dco->state == CE_CAN_DELAY) {
865                 can_delay = 1;
866                 err = packet_write_fmt_gently(process->in, "can-delay=1\n");
867                 if (err)
868                         goto done;
869         }
870
871         err = packet_flush_gently(process->in);
872         if (err)
873                 goto done;
874
875         if (fd >= 0)
876                 err = write_packetized_from_fd(fd, process->in);
877         else
878                 err = write_packetized_from_buf(src, len, process->in);
879         if (err)
880                 goto done;
881
882         err = subprocess_read_status(process->out, &filter_status);
883         if (err)
884                 goto done;
885
886         if (can_delay && !strcmp(filter_status.buf, "delayed")) {
887                 string_list_insert(&dco->filters, cmd);
888                 string_list_insert(&dco->paths, path);
889         } else {
890                 /* The filter got the blob and wants to send us a response. */
891                 err = strcmp(filter_status.buf, "success");
892                 if (err)
893                         goto done;
894
895                 err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
896                 if (err)
897                         goto done;
898
899                 err = subprocess_read_status(process->out, &filter_status);
900                 if (err)
901                         goto done;
902
903                 err = strcmp(filter_status.buf, "success");
904         }
905
906 done:
907         sigchain_pop(SIGPIPE);
908
909         if (err)
910                 handle_filter_error(&filter_status, entry, wanted_capability);
911         else
912                 strbuf_swap(dst, &nbuf);
913         strbuf_release(&nbuf);
914         return !err;
915 }
916
917
918 int async_query_available_blobs(const char *cmd, struct string_list *available_paths)
919 {
920         int err;
921         char *line;
922         struct cmd2process *entry;
923         struct child_process *process;
924         struct strbuf filter_status = STRBUF_INIT;
925
926         assert(subprocess_map_initialized);
927         entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
928         if (!entry) {
929                 error(_("external filter '%s' is not available anymore although "
930                         "not all paths have been filtered"), cmd);
931                 return 0;
932         }
933         process = &entry->subprocess.process;
934         sigchain_push(SIGPIPE, SIG_IGN);
935
936         err = packet_write_fmt_gently(
937                 process->in, "command=list_available_blobs\n");
938         if (err)
939                 goto done;
940
941         err = packet_flush_gently(process->in);
942         if (err)
943                 goto done;
944
945         while ((line = packet_read_line(process->out, NULL))) {
946                 const char *path;
947                 if (skip_prefix(line, "pathname=", &path))
948                         string_list_insert(available_paths, xstrdup(path));
949                 else
950                         ; /* ignore unknown keys */
951         }
952
953         err = subprocess_read_status(process->out, &filter_status);
954         if (err)
955                 goto done;
956
957         err = strcmp(filter_status.buf, "success");
958
959 done:
960         sigchain_pop(SIGPIPE);
961
962         if (err)
963                 handle_filter_error(&filter_status, entry, 0);
964         return !err;
965 }
966
967 static struct convert_driver {
968         const char *name;
969         struct convert_driver *next;
970         const char *smudge;
971         const char *clean;
972         const char *process;
973         int required;
974 } *user_convert, **user_convert_tail;
975
976 static int apply_filter(const char *path, const char *src, size_t len,
977                         int fd, struct strbuf *dst, struct convert_driver *drv,
978                         const unsigned int wanted_capability,
979                         struct delayed_checkout *dco)
980 {
981         const char *cmd = NULL;
982
983         if (!drv)
984                 return 0;
985
986         if (!dst)
987                 return 1;
988
989         if ((wanted_capability & CAP_CLEAN) && !drv->process && drv->clean)
990                 cmd = drv->clean;
991         else if ((wanted_capability & CAP_SMUDGE) && !drv->process && drv->smudge)
992                 cmd = drv->smudge;
993
994         if (cmd && *cmd)
995                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
996         else if (drv->process && *drv->process)
997                 return apply_multi_file_filter(path, src, len, fd, dst,
998                         drv->process, wanted_capability, dco);
999
1000         return 0;
1001 }
1002
1003 static int read_convert_config(const char *var, const char *value, void *cb)
1004 {
1005         const char *key, *name;
1006         int namelen;
1007         struct convert_driver *drv;
1008
1009         /*
1010          * External conversion drivers are configured using
1011          * "filter.<name>.variable".
1012          */
1013         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
1014                 return 0;
1015         for (drv = user_convert; drv; drv = drv->next)
1016                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
1017                         break;
1018         if (!drv) {
1019                 drv = xcalloc(1, sizeof(struct convert_driver));
1020                 drv->name = xmemdupz(name, namelen);
1021                 *user_convert_tail = drv;
1022                 user_convert_tail = &(drv->next);
1023         }
1024
1025         /*
1026          * filter.<name>.smudge and filter.<name>.clean specifies
1027          * the command line:
1028          *
1029          *      command-line
1030          *
1031          * The command-line will not be interpolated in any way.
1032          */
1033
1034         if (!strcmp("smudge", key))
1035                 return git_config_string(&drv->smudge, var, value);
1036
1037         if (!strcmp("clean", key))
1038                 return git_config_string(&drv->clean, var, value);
1039
1040         if (!strcmp("process", key))
1041                 return git_config_string(&drv->process, var, value);
1042
1043         if (!strcmp("required", key)) {
1044                 drv->required = git_config_bool(var, value);
1045                 return 0;
1046         }
1047
1048         return 0;
1049 }
1050
1051 static int count_ident(const char *cp, unsigned long size)
1052 {
1053         /*
1054          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
1055          */
1056         int cnt = 0;
1057         char ch;
1058
1059         while (size) {
1060                 ch = *cp++;
1061                 size--;
1062                 if (ch != '$')
1063                         continue;
1064                 if (size < 3)
1065                         break;
1066                 if (memcmp("Id", cp, 2))
1067                         continue;
1068                 ch = cp[2];
1069                 cp += 3;
1070                 size -= 3;
1071                 if (ch == '$')
1072                         cnt++; /* $Id$ */
1073                 if (ch != ':')
1074                         continue;
1075
1076                 /*
1077                  * "$Id: ... "; scan up to the closing dollar sign and discard.
1078                  */
1079                 while (size) {
1080                         ch = *cp++;
1081                         size--;
1082                         if (ch == '$') {
1083                                 cnt++;
1084                                 break;
1085                         }
1086                         if (ch == '\n')
1087                                 break;
1088                 }
1089         }
1090         return cnt;
1091 }
1092
1093 static int ident_to_git(const char *src, size_t len,
1094                         struct strbuf *buf, int ident)
1095 {
1096         char *dst, *dollar;
1097
1098         if (!ident || (src && !count_ident(src, len)))
1099                 return 0;
1100
1101         if (!buf)
1102                 return 1;
1103
1104         /* only grow if not in place */
1105         if (strbuf_avail(buf) + buf->len < len)
1106                 strbuf_grow(buf, len - buf->len);
1107         dst = buf->buf;
1108         for (;;) {
1109                 dollar = memchr(src, '$', len);
1110                 if (!dollar)
1111                         break;
1112                 memmove(dst, src, dollar + 1 - src);
1113                 dst += dollar + 1 - src;
1114                 len -= dollar + 1 - src;
1115                 src  = dollar + 1;
1116
1117                 if (len > 3 && !memcmp(src, "Id:", 3)) {
1118                         dollar = memchr(src + 3, '$', len - 3);
1119                         if (!dollar)
1120                                 break;
1121                         if (memchr(src + 3, '\n', dollar - src - 3)) {
1122                                 /* Line break before the next dollar. */
1123                                 continue;
1124                         }
1125
1126                         memcpy(dst, "Id$", 3);
1127                         dst += 3;
1128                         len -= dollar + 1 - src;
1129                         src  = dollar + 1;
1130                 }
1131         }
1132         memmove(dst, src, len);
1133         strbuf_setlen(buf, dst + len - buf->buf);
1134         return 1;
1135 }
1136
1137 static int ident_to_worktree(const char *src, size_t len,
1138                              struct strbuf *buf, int ident)
1139 {
1140         struct object_id oid;
1141         char *to_free = NULL, *dollar, *spc;
1142         int cnt;
1143
1144         if (!ident)
1145                 return 0;
1146
1147         cnt = count_ident(src, len);
1148         if (!cnt)
1149                 return 0;
1150
1151         /* are we "faking" in place editing ? */
1152         if (src == buf->buf)
1153                 to_free = strbuf_detach(buf, NULL);
1154         hash_object_file(src, len, "blob", &oid);
1155
1156         strbuf_grow(buf, len + cnt * (the_hash_algo->hexsz + 3));
1157         for (;;) {
1158                 /* step 1: run to the next '$' */
1159                 dollar = memchr(src, '$', len);
1160                 if (!dollar)
1161                         break;
1162                 strbuf_add(buf, src, dollar + 1 - src);
1163                 len -= dollar + 1 - src;
1164                 src  = dollar + 1;
1165
1166                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
1167                 if (len < 3 || memcmp("Id", src, 2))
1168                         continue;
1169
1170                 /* step 3: skip over Id$ or Id:xxxxx$ */
1171                 if (src[2] == '$') {
1172                         src += 3;
1173                         len -= 3;
1174                 } else if (src[2] == ':') {
1175                         /*
1176                          * It's possible that an expanded Id has crept its way into the
1177                          * repository, we cope with that by stripping the expansion out.
1178                          * This is probably not a good idea, since it will cause changes
1179                          * on checkout, which won't go away by stash, but let's keep it
1180                          * for git-style ids.
1181                          */
1182                         dollar = memchr(src + 3, '$', len - 3);
1183                         if (!dollar) {
1184                                 /* incomplete keyword, no more '$', so just quit the loop */
1185                                 break;
1186                         }
1187
1188                         if (memchr(src + 3, '\n', dollar - src - 3)) {
1189                                 /* Line break before the next dollar. */
1190                                 continue;
1191                         }
1192
1193                         spc = memchr(src + 4, ' ', dollar - src - 4);
1194                         if (spc && spc < dollar-1) {
1195                                 /* There are spaces in unexpected places.
1196                                  * This is probably an id from some other
1197                                  * versioning system. Keep it for now.
1198                                  */
1199                                 continue;
1200                         }
1201
1202                         len -= dollar + 1 - src;
1203                         src  = dollar + 1;
1204                 } else {
1205                         /* it wasn't a "Id$" or "Id:xxxx$" */
1206                         continue;
1207                 }
1208
1209                 /* step 4: substitute */
1210                 strbuf_addstr(buf, "Id: ");
1211                 strbuf_addstr(buf, oid_to_hex(&oid));
1212                 strbuf_addstr(buf, " $");
1213         }
1214         strbuf_add(buf, src, len);
1215
1216         free(to_free);
1217         return 1;
1218 }
1219
1220 static const char *git_path_check_encoding(struct attr_check_item *check)
1221 {
1222         const char *value = check->value;
1223
1224         if (ATTR_UNSET(value) || !strlen(value))
1225                 return NULL;
1226
1227         if (ATTR_TRUE(value) || ATTR_FALSE(value)) {
1228                 die(_("true/false are no valid working-tree-encodings"));
1229         }
1230
1231         /* Don't encode to the default encoding */
1232         if (same_encoding(value, default_encoding))
1233                 return NULL;
1234
1235         return value;
1236 }
1237
1238 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
1239 {
1240         const char *value = check->value;
1241
1242         if (ATTR_TRUE(value))
1243                 return CRLF_TEXT;
1244         else if (ATTR_FALSE(value))
1245                 return CRLF_BINARY;
1246         else if (ATTR_UNSET(value))
1247                 ;
1248         else if (!strcmp(value, "input"))
1249                 return CRLF_TEXT_INPUT;
1250         else if (!strcmp(value, "auto"))
1251                 return CRLF_AUTO;
1252         return CRLF_UNDEFINED;
1253 }
1254
1255 static enum eol git_path_check_eol(struct attr_check_item *check)
1256 {
1257         const char *value = check->value;
1258
1259         if (ATTR_UNSET(value))
1260                 ;
1261         else if (!strcmp(value, "lf"))
1262                 return EOL_LF;
1263         else if (!strcmp(value, "crlf"))
1264                 return EOL_CRLF;
1265         return EOL_UNSET;
1266 }
1267
1268 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
1269 {
1270         const char *value = check->value;
1271         struct convert_driver *drv;
1272
1273         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
1274                 return NULL;
1275         for (drv = user_convert; drv; drv = drv->next)
1276                 if (!strcmp(value, drv->name))
1277                         return drv;
1278         return NULL;
1279 }
1280
1281 static int git_path_check_ident(struct attr_check_item *check)
1282 {
1283         const char *value = check->value;
1284
1285         return !!ATTR_TRUE(value);
1286 }
1287
1288 struct conv_attrs {
1289         struct convert_driver *drv;
1290         enum crlf_action attr_action; /* What attr says */
1291         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
1292         int ident;
1293         const char *working_tree_encoding; /* Supported encoding or default encoding if NULL */
1294 };
1295
1296 static void convert_attrs(const struct index_state *istate,
1297                           struct conv_attrs *ca, const char *path)
1298 {
1299         static struct attr_check *check;
1300         struct attr_check_item *ccheck = NULL;
1301
1302         if (!check) {
1303                 check = attr_check_initl("crlf", "ident", "filter",
1304                                          "eol", "text", "working-tree-encoding",
1305                                          NULL);
1306                 user_convert_tail = &user_convert;
1307                 git_config(read_convert_config, NULL);
1308         }
1309
1310         git_check_attr(istate, path, check);
1311         ccheck = check->items;
1312         ca->crlf_action = git_path_check_crlf(ccheck + 4);
1313         if (ca->crlf_action == CRLF_UNDEFINED)
1314                 ca->crlf_action = git_path_check_crlf(ccheck + 0);
1315         ca->ident = git_path_check_ident(ccheck + 1);
1316         ca->drv = git_path_check_convert(ccheck + 2);
1317         if (ca->crlf_action != CRLF_BINARY) {
1318                 enum eol eol_attr = git_path_check_eol(ccheck + 3);
1319                 if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1320                         ca->crlf_action = CRLF_AUTO_INPUT;
1321                 else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1322                         ca->crlf_action = CRLF_AUTO_CRLF;
1323                 else if (eol_attr == EOL_LF)
1324                         ca->crlf_action = CRLF_TEXT_INPUT;
1325                 else if (eol_attr == EOL_CRLF)
1326                         ca->crlf_action = CRLF_TEXT_CRLF;
1327         }
1328         ca->working_tree_encoding = git_path_check_encoding(ccheck + 5);
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 struct index_state *istate, const char *path)
1343 {
1344         struct conv_attrs ca;
1345
1346         convert_attrs(istate, &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 struct index_state *istate, const char *path)
1362 {
1363         struct conv_attrs ca;
1364
1365         convert_attrs(istate, &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(istate, &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(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(istate, &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(dst->buf, dst->len, dst, ca.ident);
1437 }
1438
1439 static int convert_to_working_tree_internal(const struct index_state *istate,
1440                                             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(istate, &ca, path);
1448
1449         ret |= ident_to_worktree(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(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 struct index_state *istate,
1482                                   const char *path, const char *src,
1483                                   size_t len, struct strbuf *dst,
1484                                   void *dco)
1485 {
1486         return convert_to_working_tree_internal(istate, path, src, len, dst, 0, dco);
1487 }
1488
1489 int convert_to_working_tree(const struct index_state *istate,
1490                             const char *path, const char *src,
1491                             size_t len, struct strbuf *dst)
1492 {
1493         return convert_to_working_tree_internal(istate, path, src, len, dst, 0, NULL);
1494 }
1495
1496 int renormalize_buffer(const struct index_state *istate, const char *path,
1497                        const char *src, size_t len, struct strbuf *dst)
1498 {
1499         int ret = convert_to_working_tree_internal(istate, path, src, len, dst, 1, NULL);
1500         if (ret) {
1501                 src = dst->buf;
1502                 len = dst->len;
1503         }
1504         return ret | convert_to_git(istate, path, src, len, dst, CONV_EOL_RENORMALIZE);
1505 }
1506
1507 /*****************************************************************
1508  *
1509  * Streaming conversion support
1510  *
1511  *****************************************************************/
1512
1513 typedef int (*filter_fn)(struct stream_filter *,
1514                          const char *input, size_t *isize_p,
1515                          char *output, size_t *osize_p);
1516 typedef void (*free_fn)(struct stream_filter *);
1517
1518 struct stream_filter_vtbl {
1519         filter_fn filter;
1520         free_fn free;
1521 };
1522
1523 struct stream_filter {
1524         struct stream_filter_vtbl *vtbl;
1525 };
1526
1527 static int null_filter_fn(struct stream_filter *filter,
1528                           const char *input, size_t *isize_p,
1529                           char *output, size_t *osize_p)
1530 {
1531         size_t count;
1532
1533         if (!input)
1534                 return 0; /* we do not keep any states */
1535         count = *isize_p;
1536         if (*osize_p < count)
1537                 count = *osize_p;
1538         if (count) {
1539                 memmove(output, input, count);
1540                 *isize_p -= count;
1541                 *osize_p -= count;
1542         }
1543         return 0;
1544 }
1545
1546 static void null_free_fn(struct stream_filter *filter)
1547 {
1548         ; /* nothing -- null instances are shared */
1549 }
1550
1551 static struct stream_filter_vtbl null_vtbl = {
1552         null_filter_fn,
1553         null_free_fn,
1554 };
1555
1556 static struct stream_filter null_filter_singleton = {
1557         &null_vtbl,
1558 };
1559
1560 int is_null_stream_filter(struct stream_filter *filter)
1561 {
1562         return filter == &null_filter_singleton;
1563 }
1564
1565
1566 /*
1567  * LF-to-CRLF filter
1568  */
1569
1570 struct lf_to_crlf_filter {
1571         struct stream_filter filter;
1572         unsigned has_held:1;
1573         char held;
1574 };
1575
1576 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1577                                 const char *input, size_t *isize_p,
1578                                 char *output, size_t *osize_p)
1579 {
1580         size_t count, o = 0;
1581         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1582
1583         /*
1584          * We may be holding onto the CR to see if it is followed by a
1585          * LF, in which case we would need to go to the main loop.
1586          * Otherwise, just emit it to the output stream.
1587          */
1588         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1589                 output[o++] = lf_to_crlf->held;
1590                 lf_to_crlf->has_held = 0;
1591         }
1592
1593         /* We are told to drain */
1594         if (!input) {
1595                 *osize_p -= o;
1596                 return 0;
1597         }
1598
1599         count = *isize_p;
1600         if (count || lf_to_crlf->has_held) {
1601                 size_t i;
1602                 int was_cr = 0;
1603
1604                 if (lf_to_crlf->has_held) {
1605                         was_cr = 1;
1606                         lf_to_crlf->has_held = 0;
1607                 }
1608
1609                 for (i = 0; o < *osize_p && i < count; i++) {
1610                         char ch = input[i];
1611
1612                         if (ch == '\n') {
1613                                 output[o++] = '\r';
1614                         } else if (was_cr) {
1615                                 /*
1616                                  * Previous round saw CR and it is not followed
1617                                  * by a LF; emit the CR before processing the
1618                                  * current character.
1619                                  */
1620                                 output[o++] = '\r';
1621                         }
1622
1623                         /*
1624                          * We may have consumed the last output slot,
1625                          * in which case we need to break out of this
1626                          * loop; hold the current character before
1627                          * returning.
1628                          */
1629                         if (*osize_p <= o) {
1630                                 lf_to_crlf->has_held = 1;
1631                                 lf_to_crlf->held = ch;
1632                                 continue; /* break but increment i */
1633                         }
1634
1635                         if (ch == '\r') {
1636                                 was_cr = 1;
1637                                 continue;
1638                         }
1639
1640                         was_cr = 0;
1641                         output[o++] = ch;
1642                 }
1643
1644                 *osize_p -= o;
1645                 *isize_p -= i;
1646
1647                 if (!lf_to_crlf->has_held && was_cr) {
1648                         lf_to_crlf->has_held = 1;
1649                         lf_to_crlf->held = '\r';
1650                 }
1651         }
1652         return 0;
1653 }
1654
1655 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1656 {
1657         free(filter);
1658 }
1659
1660 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1661         lf_to_crlf_filter_fn,
1662         lf_to_crlf_free_fn,
1663 };
1664
1665 static struct stream_filter *lf_to_crlf_filter(void)
1666 {
1667         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1668
1669         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1670         return (struct stream_filter *)lf_to_crlf;
1671 }
1672
1673 /*
1674  * Cascade filter
1675  */
1676 #define FILTER_BUFFER 1024
1677 struct cascade_filter {
1678         struct stream_filter filter;
1679         struct stream_filter *one;
1680         struct stream_filter *two;
1681         char buf[FILTER_BUFFER];
1682         int end, ptr;
1683 };
1684
1685 static int cascade_filter_fn(struct stream_filter *filter,
1686                              const char *input, size_t *isize_p,
1687                              char *output, size_t *osize_p)
1688 {
1689         struct cascade_filter *cas = (struct cascade_filter *) filter;
1690         size_t filled = 0;
1691         size_t sz = *osize_p;
1692         size_t to_feed, remaining;
1693
1694         /*
1695          * input -- (one) --> buf -- (two) --> output
1696          */
1697         while (filled < sz) {
1698                 remaining = sz - filled;
1699
1700                 /* do we already have something to feed two with? */
1701                 if (cas->ptr < cas->end) {
1702                         to_feed = cas->end - cas->ptr;
1703                         if (stream_filter(cas->two,
1704                                           cas->buf + cas->ptr, &to_feed,
1705                                           output + filled, &remaining))
1706                                 return -1;
1707                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1708                         filled = sz - remaining;
1709                         continue;
1710                 }
1711
1712                 /* feed one from upstream and have it emit into our buffer */
1713                 to_feed = input ? *isize_p : 0;
1714                 if (input && !to_feed)
1715                         break;
1716                 remaining = sizeof(cas->buf);
1717                 if (stream_filter(cas->one,
1718                                   input, &to_feed,
1719                                   cas->buf, &remaining))
1720                         return -1;
1721                 cas->end = sizeof(cas->buf) - remaining;
1722                 cas->ptr = 0;
1723                 if (input) {
1724                         size_t fed = *isize_p - to_feed;
1725                         *isize_p -= fed;
1726                         input += fed;
1727                 }
1728
1729                 /* do we know that we drained one completely? */
1730                 if (input || cas->end)
1731                         continue;
1732
1733                 /* tell two to drain; we have nothing more to give it */
1734                 to_feed = 0;
1735                 remaining = sz - filled;
1736                 if (stream_filter(cas->two,
1737                                   NULL, &to_feed,
1738                                   output + filled, &remaining))
1739                         return -1;
1740                 if (remaining == (sz - filled))
1741                         break; /* completely drained two */
1742                 filled = sz - remaining;
1743         }
1744         *osize_p -= filled;
1745         return 0;
1746 }
1747
1748 static void cascade_free_fn(struct stream_filter *filter)
1749 {
1750         struct cascade_filter *cas = (struct cascade_filter *)filter;
1751         free_stream_filter(cas->one);
1752         free_stream_filter(cas->two);
1753         free(filter);
1754 }
1755
1756 static struct stream_filter_vtbl cascade_vtbl = {
1757         cascade_filter_fn,
1758         cascade_free_fn,
1759 };
1760
1761 static struct stream_filter *cascade_filter(struct stream_filter *one,
1762                                             struct stream_filter *two)
1763 {
1764         struct cascade_filter *cascade;
1765
1766         if (!one || is_null_stream_filter(one))
1767                 return two;
1768         if (!two || is_null_stream_filter(two))
1769                 return one;
1770
1771         cascade = xmalloc(sizeof(*cascade));
1772         cascade->one = one;
1773         cascade->two = two;
1774         cascade->end = cascade->ptr = 0;
1775         cascade->filter.vtbl = &cascade_vtbl;
1776         return (struct stream_filter *)cascade;
1777 }
1778
1779 /*
1780  * ident filter
1781  */
1782 #define IDENT_DRAINING (-1)
1783 #define IDENT_SKIPPING (-2)
1784 struct ident_filter {
1785         struct stream_filter filter;
1786         struct strbuf left;
1787         int state;
1788         char ident[GIT_MAX_HEXSZ + 5]; /* ": x40 $" */
1789 };
1790
1791 static int is_foreign_ident(const char *str)
1792 {
1793         int i;
1794
1795         if (!skip_prefix(str, "$Id: ", &str))
1796                 return 0;
1797         for (i = 0; str[i]; i++) {
1798                 if (isspace(str[i]) && str[i+1] != '$')
1799                         return 1;
1800         }
1801         return 0;
1802 }
1803
1804 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1805 {
1806         size_t to_drain = ident->left.len;
1807
1808         if (*osize_p < to_drain)
1809                 to_drain = *osize_p;
1810         if (to_drain) {
1811                 memcpy(*output_p, ident->left.buf, to_drain);
1812                 strbuf_remove(&ident->left, 0, to_drain);
1813                 *output_p += to_drain;
1814                 *osize_p -= to_drain;
1815         }
1816         if (!ident->left.len)
1817                 ident->state = 0;
1818 }
1819
1820 static int ident_filter_fn(struct stream_filter *filter,
1821                            const char *input, size_t *isize_p,
1822                            char *output, size_t *osize_p)
1823 {
1824         struct ident_filter *ident = (struct ident_filter *)filter;
1825         static const char head[] = "$Id";
1826
1827         if (!input) {
1828                 /* drain upon eof */
1829                 switch (ident->state) {
1830                 default:
1831                         strbuf_add(&ident->left, head, ident->state);
1832                         /* fallthrough */
1833                 case IDENT_SKIPPING:
1834                         /* fallthrough */
1835                 case IDENT_DRAINING:
1836                         ident_drain(ident, &output, osize_p);
1837                 }
1838                 return 0;
1839         }
1840
1841         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1842                 int ch;
1843
1844                 if (ident->state == IDENT_DRAINING) {
1845                         ident_drain(ident, &output, osize_p);
1846                         if (!*osize_p)
1847                                 break;
1848                         continue;
1849                 }
1850
1851                 ch = *(input++);
1852                 (*isize_p)--;
1853
1854                 if (ident->state == IDENT_SKIPPING) {
1855                         /*
1856                          * Skipping until '$' or LF, but keeping them
1857                          * in case it is a foreign ident.
1858                          */
1859                         strbuf_addch(&ident->left, ch);
1860                         if (ch != '\n' && ch != '$')
1861                                 continue;
1862                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1863                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1864                                 strbuf_addstr(&ident->left, ident->ident);
1865                         }
1866                         ident->state = IDENT_DRAINING;
1867                         continue;
1868                 }
1869
1870                 if (ident->state < sizeof(head) &&
1871                     head[ident->state] == ch) {
1872                         ident->state++;
1873                         continue;
1874                 }
1875
1876                 if (ident->state)
1877                         strbuf_add(&ident->left, head, ident->state);
1878                 if (ident->state == sizeof(head) - 1) {
1879                         if (ch != ':' && ch != '$') {
1880                                 strbuf_addch(&ident->left, ch);
1881                                 ident->state = 0;
1882                                 continue;
1883                         }
1884
1885                         if (ch == ':') {
1886                                 strbuf_addch(&ident->left, ch);
1887                                 ident->state = IDENT_SKIPPING;
1888                         } else {
1889                                 strbuf_addstr(&ident->left, ident->ident);
1890                                 ident->state = IDENT_DRAINING;
1891                         }
1892                         continue;
1893                 }
1894
1895                 strbuf_addch(&ident->left, ch);
1896                 ident->state = IDENT_DRAINING;
1897         }
1898         return 0;
1899 }
1900
1901 static void ident_free_fn(struct stream_filter *filter)
1902 {
1903         struct ident_filter *ident = (struct ident_filter *)filter;
1904         strbuf_release(&ident->left);
1905         free(filter);
1906 }
1907
1908 static struct stream_filter_vtbl ident_vtbl = {
1909         ident_filter_fn,
1910         ident_free_fn,
1911 };
1912
1913 static struct stream_filter *ident_filter(const struct object_id *oid)
1914 {
1915         struct ident_filter *ident = xmalloc(sizeof(*ident));
1916
1917         xsnprintf(ident->ident, sizeof(ident->ident),
1918                   ": %s $", oid_to_hex(oid));
1919         strbuf_init(&ident->left, 0);
1920         ident->filter.vtbl = &ident_vtbl;
1921         ident->state = 0;
1922         return (struct stream_filter *)ident;
1923 }
1924
1925 /*
1926  * Return an appropriately constructed filter for the path, or NULL if
1927  * the contents cannot be filtered without reading the whole thing
1928  * in-core.
1929  *
1930  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1931  * large binary blob you would want us not to slurp into the memory!
1932  */
1933 struct stream_filter *get_stream_filter(const struct index_state *istate,
1934                                         const char *path,
1935                                         const struct object_id *oid)
1936 {
1937         struct conv_attrs ca;
1938         struct stream_filter *filter = NULL;
1939
1940         convert_attrs(istate, &ca, path);
1941         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1942                 return NULL;
1943
1944         if (ca.working_tree_encoding)
1945                 return NULL;
1946
1947         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1948                 return NULL;
1949
1950         if (ca.ident)
1951                 filter = ident_filter(oid);
1952
1953         if (output_eol(ca.crlf_action) == EOL_CRLF)
1954                 filter = cascade_filter(filter, lf_to_crlf_filter());
1955         else
1956                 filter = cascade_filter(filter, &null_filter_singleton);
1957
1958         return filter;
1959 }
1960
1961 void free_stream_filter(struct stream_filter *filter)
1962 {
1963         filter->vtbl->free(filter);
1964 }
1965
1966 int stream_filter(struct stream_filter *filter,
1967                   const char *input, size_t *isize_p,
1968                   char *output, size_t *osize_p)
1969 {
1970         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1971 }