commit: add repository argument to commit_graft_pos
[git] / convert.c
1 #define NO_THE_INDEX_COMPATIBILITY_MACROS
2 #include "cache.h"
3 #include "config.h"
4 #include "object-store.h"
5 #include "attr.h"
6 #include "run-command.h"
7 #include "quote.h"
8 #include "sigchain.h"
9 #include "pkt-line.h"
10 #include "sub-process.h"
11
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 crlf_to_git(const struct index_state *istate,
270                        const char *path, const char *src, size_t len,
271                        struct strbuf *buf,
272                        enum crlf_action crlf_action, int conv_flags)
273 {
274         struct text_stat stats;
275         char *dst;
276         int convert_crlf_into_lf;
277
278         if (crlf_action == CRLF_BINARY ||
279             (src && !len))
280                 return 0;
281
282         /*
283          * If we are doing a dry-run and have no source buffer, there is
284          * nothing to analyze; we must assume we would convert.
285          */
286         if (!buf && !src)
287                 return 1;
288
289         gather_stats(src, len, &stats);
290         /* Optimization: No CRLF? Nothing to convert, regardless. */
291         convert_crlf_into_lf = !!stats.crlf;
292
293         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
294                 if (convert_is_binary(len, &stats))
295                         return 0;
296                 /*
297                  * If the file in the index has any CR in it, do not
298                  * convert.  This is the new safer autocrlf handling,
299                  * unless we want to renormalize in a merge or
300                  * cherry-pick.
301                  */
302                 if ((!(conv_flags & CONV_EOL_RENORMALIZE)) &&
303                     has_crlf_in_index(istate, path))
304                         convert_crlf_into_lf = 0;
305         }
306         if (((conv_flags & CONV_EOL_RNDTRP_WARN) ||
307              ((conv_flags & CONV_EOL_RNDTRP_DIE) && len))) {
308                 struct text_stat new_stats;
309                 memcpy(&new_stats, &stats, sizeof(new_stats));
310                 /* simulate "git add" */
311                 if (convert_crlf_into_lf) {
312                         new_stats.lonelf += new_stats.crlf;
313                         new_stats.crlf = 0;
314                 }
315                 /* simulate "git checkout" */
316                 if (will_convert_lf_to_crlf(len, &new_stats, crlf_action)) {
317                         new_stats.crlf += new_stats.lonelf;
318                         new_stats.lonelf = 0;
319                 }
320                 check_global_conv_flags_eol(path, crlf_action, &stats, &new_stats, conv_flags);
321         }
322         if (!convert_crlf_into_lf)
323                 return 0;
324
325         /*
326          * At this point all of our source analysis is done, and we are sure we
327          * would convert. If we are in dry-run mode, we can give an answer.
328          */
329         if (!buf)
330                 return 1;
331
332         /* only grow if not in place */
333         if (strbuf_avail(buf) + buf->len < len)
334                 strbuf_grow(buf, len - buf->len);
335         dst = buf->buf;
336         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
337                 /*
338                  * If we guessed, we already know we rejected a file with
339                  * lone CR, and we can strip a CR without looking at what
340                  * follow it.
341                  */
342                 do {
343                         unsigned char c = *src++;
344                         if (c != '\r')
345                                 *dst++ = c;
346                 } while (--len);
347         } else {
348                 do {
349                         unsigned char c = *src++;
350                         if (! (c == '\r' && (1 < len && *src == '\n')))
351                                 *dst++ = c;
352                 } while (--len);
353         }
354         strbuf_setlen(buf, dst - buf->buf);
355         return 1;
356 }
357
358 static int crlf_to_worktree(const char *path, const char *src, size_t len,
359                             struct strbuf *buf, enum crlf_action crlf_action)
360 {
361         char *to_free = NULL;
362         struct text_stat stats;
363
364         if (!len || output_eol(crlf_action) != EOL_CRLF)
365                 return 0;
366
367         gather_stats(src, len, &stats);
368         if (!will_convert_lf_to_crlf(len, &stats, crlf_action))
369                 return 0;
370
371         /* are we "faking" in place editing ? */
372         if (src == buf->buf)
373                 to_free = strbuf_detach(buf, NULL);
374
375         strbuf_grow(buf, len + stats.lonelf);
376         for (;;) {
377                 const char *nl = memchr(src, '\n', len);
378                 if (!nl)
379                         break;
380                 if (nl > src && nl[-1] == '\r') {
381                         strbuf_add(buf, src, nl + 1 - src);
382                 } else {
383                         strbuf_add(buf, src, nl - src);
384                         strbuf_addstr(buf, "\r\n");
385                 }
386                 len -= nl + 1 - src;
387                 src  = nl + 1;
388         }
389         strbuf_add(buf, src, len);
390
391         free(to_free);
392         return 1;
393 }
394
395 struct filter_params {
396         const char *src;
397         unsigned long size;
398         int fd;
399         const char *cmd;
400         const char *path;
401 };
402
403 static int filter_buffer_or_fd(int in, int out, void *data)
404 {
405         /*
406          * Spawn cmd and feed the buffer contents through its stdin.
407          */
408         struct child_process child_process = CHILD_PROCESS_INIT;
409         struct filter_params *params = (struct filter_params *)data;
410         int write_err, status;
411         const char *argv[] = { NULL, NULL };
412
413         /* apply % substitution to cmd */
414         struct strbuf cmd = STRBUF_INIT;
415         struct strbuf path = STRBUF_INIT;
416         struct strbuf_expand_dict_entry dict[] = {
417                 { "f", NULL, },
418                 { NULL, NULL, },
419         };
420
421         /* quote the path to preserve spaces, etc. */
422         sq_quote_buf(&path, params->path);
423         dict[0].value = path.buf;
424
425         /* expand all %f with the quoted path */
426         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
427         strbuf_release(&path);
428
429         argv[0] = cmd.buf;
430
431         child_process.argv = argv;
432         child_process.use_shell = 1;
433         child_process.in = -1;
434         child_process.out = out;
435
436         if (start_command(&child_process)) {
437                 strbuf_release(&cmd);
438                 return error("cannot fork to run external filter '%s'", params->cmd);
439         }
440
441         sigchain_push(SIGPIPE, SIG_IGN);
442
443         if (params->src) {
444                 write_err = (write_in_full(child_process.in,
445                                            params->src, params->size) < 0);
446                 if (errno == EPIPE)
447                         write_err = 0;
448         } else {
449                 write_err = copy_fd(params->fd, child_process.in);
450                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
451                         write_err = 0;
452         }
453
454         if (close(child_process.in))
455                 write_err = 1;
456         if (write_err)
457                 error("cannot feed the input to external filter '%s'", params->cmd);
458
459         sigchain_pop(SIGPIPE);
460
461         status = finish_command(&child_process);
462         if (status)
463                 error("external filter '%s' failed %d", params->cmd, status);
464
465         strbuf_release(&cmd);
466         return (write_err || status);
467 }
468
469 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
470                         struct strbuf *dst, const char *cmd)
471 {
472         /*
473          * Create a pipeline to have the command filter the buffer's
474          * contents.
475          *
476          * (child --> cmd) --> us
477          */
478         int err = 0;
479         struct strbuf nbuf = STRBUF_INIT;
480         struct async async;
481         struct filter_params params;
482
483         memset(&async, 0, sizeof(async));
484         async.proc = filter_buffer_or_fd;
485         async.data = &params;
486         async.out = -1;
487         params.src = src;
488         params.size = len;
489         params.fd = fd;
490         params.cmd = cmd;
491         params.path = path;
492
493         fflush(NULL);
494         if (start_async(&async))
495                 return 0;       /* error was already reported */
496
497         if (strbuf_read(&nbuf, async.out, len) < 0) {
498                 err = error("read from external filter '%s' failed", cmd);
499         }
500         if (close(async.out)) {
501                 err = error("read from external filter '%s' failed", cmd);
502         }
503         if (finish_async(&async)) {
504                 err = error("external filter '%s' failed", cmd);
505         }
506
507         if (!err) {
508                 strbuf_swap(dst, &nbuf);
509         }
510         strbuf_release(&nbuf);
511         return !err;
512 }
513
514 #define CAP_CLEAN    (1u<<0)
515 #define CAP_SMUDGE   (1u<<1)
516 #define CAP_DELAY    (1u<<2)
517
518 struct cmd2process {
519         struct subprocess_entry subprocess; /* must be the first member! */
520         unsigned int supported_capabilities;
521 };
522
523 static int subprocess_map_initialized;
524 static struct hashmap subprocess_map;
525
526 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
527 {
528         static int versions[] = {2, 0};
529         static struct subprocess_capability capabilities[] = {
530                 { "clean",  CAP_CLEAN  },
531                 { "smudge", CAP_SMUDGE },
532                 { "delay",  CAP_DELAY  },
533                 { NULL, 0 }
534         };
535         struct cmd2process *entry = (struct cmd2process *)subprocess;
536         return subprocess_handshake(subprocess, "git-filter", versions, NULL,
537                                     capabilities,
538                                     &entry->supported_capabilities);
539 }
540
541 static void handle_filter_error(const struct strbuf *filter_status,
542                                 struct cmd2process *entry,
543                                 const unsigned int wanted_capability) {
544         if (!strcmp(filter_status->buf, "error"))
545                 ; /* The filter signaled a problem with the file. */
546         else if (!strcmp(filter_status->buf, "abort") && wanted_capability) {
547                 /*
548                  * The filter signaled a permanent problem. Don't try to filter
549                  * files with the same command for the lifetime of the current
550                  * Git process.
551                  */
552                  entry->supported_capabilities &= ~wanted_capability;
553         } else {
554                 /*
555                  * Something went wrong with the protocol filter.
556                  * Force shutdown and restart if another blob requires filtering.
557                  */
558                 error("external filter '%s' failed", entry->subprocess.cmd);
559                 subprocess_stop(&subprocess_map, &entry->subprocess);
560                 free(entry);
561         }
562 }
563
564 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
565                                    int fd, struct strbuf *dst, const char *cmd,
566                                    const unsigned int wanted_capability,
567                                    struct delayed_checkout *dco)
568 {
569         int err;
570         int can_delay = 0;
571         struct cmd2process *entry;
572         struct child_process *process;
573         struct strbuf nbuf = STRBUF_INIT;
574         struct strbuf filter_status = STRBUF_INIT;
575         const char *filter_type;
576
577         if (!subprocess_map_initialized) {
578                 subprocess_map_initialized = 1;
579                 hashmap_init(&subprocess_map, cmd2process_cmp, NULL, 0);
580                 entry = NULL;
581         } else {
582                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
583         }
584
585         fflush(NULL);
586
587         if (!entry) {
588                 entry = xmalloc(sizeof(*entry));
589                 entry->supported_capabilities = 0;
590
591                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
592                         free(entry);
593                         return 0;
594                 }
595         }
596         process = &entry->subprocess.process;
597
598         if (!(entry->supported_capabilities & wanted_capability))
599                 return 0;
600
601         if (wanted_capability & CAP_CLEAN)
602                 filter_type = "clean";
603         else if (wanted_capability & CAP_SMUDGE)
604                 filter_type = "smudge";
605         else
606                 die("unexpected filter type");
607
608         sigchain_push(SIGPIPE, SIG_IGN);
609
610         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
611         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
612         if (err)
613                 goto done;
614
615         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
616         if (err) {
617                 error("path name too long for external filter");
618                 goto done;
619         }
620
621         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
622         if (err)
623                 goto done;
624
625         if ((entry->supported_capabilities & CAP_DELAY) &&
626             dco && dco->state == CE_CAN_DELAY) {
627                 can_delay = 1;
628                 err = packet_write_fmt_gently(process->in, "can-delay=1\n");
629                 if (err)
630                         goto done;
631         }
632
633         err = packet_flush_gently(process->in);
634         if (err)
635                 goto done;
636
637         if (fd >= 0)
638                 err = write_packetized_from_fd(fd, process->in);
639         else
640                 err = write_packetized_from_buf(src, len, process->in);
641         if (err)
642                 goto done;
643
644         err = subprocess_read_status(process->out, &filter_status);
645         if (err)
646                 goto done;
647
648         if (can_delay && !strcmp(filter_status.buf, "delayed")) {
649                 string_list_insert(&dco->filters, cmd);
650                 string_list_insert(&dco->paths, path);
651         } else {
652                 /* The filter got the blob and wants to send us a response. */
653                 err = strcmp(filter_status.buf, "success");
654                 if (err)
655                         goto done;
656
657                 err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
658                 if (err)
659                         goto done;
660
661                 err = subprocess_read_status(process->out, &filter_status);
662                 if (err)
663                         goto done;
664
665                 err = strcmp(filter_status.buf, "success");
666         }
667
668 done:
669         sigchain_pop(SIGPIPE);
670
671         if (err)
672                 handle_filter_error(&filter_status, entry, wanted_capability);
673         else
674                 strbuf_swap(dst, &nbuf);
675         strbuf_release(&nbuf);
676         return !err;
677 }
678
679
680 int async_query_available_blobs(const char *cmd, struct string_list *available_paths)
681 {
682         int err;
683         char *line;
684         struct cmd2process *entry;
685         struct child_process *process;
686         struct strbuf filter_status = STRBUF_INIT;
687
688         assert(subprocess_map_initialized);
689         entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
690         if (!entry) {
691                 error("external filter '%s' is not available anymore although "
692                       "not all paths have been filtered", cmd);
693                 return 0;
694         }
695         process = &entry->subprocess.process;
696         sigchain_push(SIGPIPE, SIG_IGN);
697
698         err = packet_write_fmt_gently(
699                 process->in, "command=list_available_blobs\n");
700         if (err)
701                 goto done;
702
703         err = packet_flush_gently(process->in);
704         if (err)
705                 goto done;
706
707         while ((line = packet_read_line(process->out, NULL))) {
708                 const char *path;
709                 if (skip_prefix(line, "pathname=", &path))
710                         string_list_insert(available_paths, xstrdup(path));
711                 else
712                         ; /* ignore unknown keys */
713         }
714
715         err = subprocess_read_status(process->out, &filter_status);
716         if (err)
717                 goto done;
718
719         err = strcmp(filter_status.buf, "success");
720
721 done:
722         sigchain_pop(SIGPIPE);
723
724         if (err)
725                 handle_filter_error(&filter_status, entry, 0);
726         return !err;
727 }
728
729 static struct convert_driver {
730         const char *name;
731         struct convert_driver *next;
732         const char *smudge;
733         const char *clean;
734         const char *process;
735         int required;
736 } *user_convert, **user_convert_tail;
737
738 static int apply_filter(const char *path, const char *src, size_t len,
739                         int fd, struct strbuf *dst, struct convert_driver *drv,
740                         const unsigned int wanted_capability,
741                         struct delayed_checkout *dco)
742 {
743         const char *cmd = NULL;
744
745         if (!drv)
746                 return 0;
747
748         if (!dst)
749                 return 1;
750
751         if ((wanted_capability & CAP_CLEAN) && !drv->process && drv->clean)
752                 cmd = drv->clean;
753         else if ((wanted_capability & CAP_SMUDGE) && !drv->process && drv->smudge)
754                 cmd = drv->smudge;
755
756         if (cmd && *cmd)
757                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
758         else if (drv->process && *drv->process)
759                 return apply_multi_file_filter(path, src, len, fd, dst,
760                         drv->process, wanted_capability, dco);
761
762         return 0;
763 }
764
765 static int read_convert_config(const char *var, const char *value, void *cb)
766 {
767         const char *key, *name;
768         int namelen;
769         struct convert_driver *drv;
770
771         /*
772          * External conversion drivers are configured using
773          * "filter.<name>.variable".
774          */
775         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
776                 return 0;
777         for (drv = user_convert; drv; drv = drv->next)
778                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
779                         break;
780         if (!drv) {
781                 drv = xcalloc(1, sizeof(struct convert_driver));
782                 drv->name = xmemdupz(name, namelen);
783                 *user_convert_tail = drv;
784                 user_convert_tail = &(drv->next);
785         }
786
787         /*
788          * filter.<name>.smudge and filter.<name>.clean specifies
789          * the command line:
790          *
791          *      command-line
792          *
793          * The command-line will not be interpolated in any way.
794          */
795
796         if (!strcmp("smudge", key))
797                 return git_config_string(&drv->smudge, var, value);
798
799         if (!strcmp("clean", key))
800                 return git_config_string(&drv->clean, var, value);
801
802         if (!strcmp("process", key))
803                 return git_config_string(&drv->process, var, value);
804
805         if (!strcmp("required", key)) {
806                 drv->required = git_config_bool(var, value);
807                 return 0;
808         }
809
810         return 0;
811 }
812
813 static int count_ident(const char *cp, unsigned long size)
814 {
815         /*
816          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
817          */
818         int cnt = 0;
819         char ch;
820
821         while (size) {
822                 ch = *cp++;
823                 size--;
824                 if (ch != '$')
825                         continue;
826                 if (size < 3)
827                         break;
828                 if (memcmp("Id", cp, 2))
829                         continue;
830                 ch = cp[2];
831                 cp += 3;
832                 size -= 3;
833                 if (ch == '$')
834                         cnt++; /* $Id$ */
835                 if (ch != ':')
836                         continue;
837
838                 /*
839                  * "$Id: ... "; scan up to the closing dollar sign and discard.
840                  */
841                 while (size) {
842                         ch = *cp++;
843                         size--;
844                         if (ch == '$') {
845                                 cnt++;
846                                 break;
847                         }
848                         if (ch == '\n')
849                                 break;
850                 }
851         }
852         return cnt;
853 }
854
855 static int ident_to_git(const char *path, const char *src, size_t len,
856                         struct strbuf *buf, int ident)
857 {
858         char *dst, *dollar;
859
860         if (!ident || (src && !count_ident(src, len)))
861                 return 0;
862
863         if (!buf)
864                 return 1;
865
866         /* only grow if not in place */
867         if (strbuf_avail(buf) + buf->len < len)
868                 strbuf_grow(buf, len - buf->len);
869         dst = buf->buf;
870         for (;;) {
871                 dollar = memchr(src, '$', len);
872                 if (!dollar)
873                         break;
874                 memmove(dst, src, dollar + 1 - src);
875                 dst += dollar + 1 - src;
876                 len -= dollar + 1 - src;
877                 src  = dollar + 1;
878
879                 if (len > 3 && !memcmp(src, "Id:", 3)) {
880                         dollar = memchr(src + 3, '$', len - 3);
881                         if (!dollar)
882                                 break;
883                         if (memchr(src + 3, '\n', dollar - src - 3)) {
884                                 /* Line break before the next dollar. */
885                                 continue;
886                         }
887
888                         memcpy(dst, "Id$", 3);
889                         dst += 3;
890                         len -= dollar + 1 - src;
891                         src  = dollar + 1;
892                 }
893         }
894         memmove(dst, src, len);
895         strbuf_setlen(buf, dst + len - buf->buf);
896         return 1;
897 }
898
899 static int ident_to_worktree(const char *path, const char *src, size_t len,
900                              struct strbuf *buf, int ident)
901 {
902         struct object_id oid;
903         char *to_free = NULL, *dollar, *spc;
904         int cnt;
905
906         if (!ident)
907                 return 0;
908
909         cnt = count_ident(src, len);
910         if (!cnt)
911                 return 0;
912
913         /* are we "faking" in place editing ? */
914         if (src == buf->buf)
915                 to_free = strbuf_detach(buf, NULL);
916         hash_object_file(src, len, "blob", &oid);
917
918         strbuf_grow(buf, len + cnt * (the_hash_algo->hexsz + 3));
919         for (;;) {
920                 /* step 1: run to the next '$' */
921                 dollar = memchr(src, '$', len);
922                 if (!dollar)
923                         break;
924                 strbuf_add(buf, src, dollar + 1 - src);
925                 len -= dollar + 1 - src;
926                 src  = dollar + 1;
927
928                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
929                 if (len < 3 || memcmp("Id", src, 2))
930                         continue;
931
932                 /* step 3: skip over Id$ or Id:xxxxx$ */
933                 if (src[2] == '$') {
934                         src += 3;
935                         len -= 3;
936                 } else if (src[2] == ':') {
937                         /*
938                          * It's possible that an expanded Id has crept its way into the
939                          * repository, we cope with that by stripping the expansion out.
940                          * This is probably not a good idea, since it will cause changes
941                          * on checkout, which won't go away by stash, but let's keep it
942                          * for git-style ids.
943                          */
944                         dollar = memchr(src + 3, '$', len - 3);
945                         if (!dollar) {
946                                 /* incomplete keyword, no more '$', so just quit the loop */
947                                 break;
948                         }
949
950                         if (memchr(src + 3, '\n', dollar - src - 3)) {
951                                 /* Line break before the next dollar. */
952                                 continue;
953                         }
954
955                         spc = memchr(src + 4, ' ', dollar - src - 4);
956                         if (spc && spc < dollar-1) {
957                                 /* There are spaces in unexpected places.
958                                  * This is probably an id from some other
959                                  * versioning system. Keep it for now.
960                                  */
961                                 continue;
962                         }
963
964                         len -= dollar + 1 - src;
965                         src  = dollar + 1;
966                 } else {
967                         /* it wasn't a "Id$" or "Id:xxxx$" */
968                         continue;
969                 }
970
971                 /* step 4: substitute */
972                 strbuf_addstr(buf, "Id: ");
973                 strbuf_addstr(buf, oid_to_hex(&oid));
974                 strbuf_addstr(buf, " $");
975         }
976         strbuf_add(buf, src, len);
977
978         free(to_free);
979         return 1;
980 }
981
982 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
983 {
984         const char *value = check->value;
985
986         if (ATTR_TRUE(value))
987                 return CRLF_TEXT;
988         else if (ATTR_FALSE(value))
989                 return CRLF_BINARY;
990         else if (ATTR_UNSET(value))
991                 ;
992         else if (!strcmp(value, "input"))
993                 return CRLF_TEXT_INPUT;
994         else if (!strcmp(value, "auto"))
995                 return CRLF_AUTO;
996         return CRLF_UNDEFINED;
997 }
998
999 static enum eol git_path_check_eol(struct attr_check_item *check)
1000 {
1001         const char *value = check->value;
1002
1003         if (ATTR_UNSET(value))
1004                 ;
1005         else if (!strcmp(value, "lf"))
1006                 return EOL_LF;
1007         else if (!strcmp(value, "crlf"))
1008                 return EOL_CRLF;
1009         return EOL_UNSET;
1010 }
1011
1012 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
1013 {
1014         const char *value = check->value;
1015         struct convert_driver *drv;
1016
1017         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
1018                 return NULL;
1019         for (drv = user_convert; drv; drv = drv->next)
1020                 if (!strcmp(value, drv->name))
1021                         return drv;
1022         return NULL;
1023 }
1024
1025 static int git_path_check_ident(struct attr_check_item *check)
1026 {
1027         const char *value = check->value;
1028
1029         return !!ATTR_TRUE(value);
1030 }
1031
1032 struct conv_attrs {
1033         struct convert_driver *drv;
1034         enum crlf_action attr_action; /* What attr says */
1035         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
1036         int ident;
1037 };
1038
1039 static void convert_attrs(struct conv_attrs *ca, const char *path)
1040 {
1041         static struct attr_check *check;
1042
1043         if (!check) {
1044                 check = attr_check_initl("crlf", "ident", "filter",
1045                                          "eol", "text", NULL);
1046                 user_convert_tail = &user_convert;
1047                 git_config(read_convert_config, NULL);
1048         }
1049
1050         if (!git_check_attr(path, check)) {
1051                 struct attr_check_item *ccheck = check->items;
1052                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
1053                 if (ca->crlf_action == CRLF_UNDEFINED)
1054                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
1055                 ca->ident = git_path_check_ident(ccheck + 1);
1056                 ca->drv = git_path_check_convert(ccheck + 2);
1057                 if (ca->crlf_action != CRLF_BINARY) {
1058                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
1059                         if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1060                                 ca->crlf_action = CRLF_AUTO_INPUT;
1061                         else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1062                                 ca->crlf_action = CRLF_AUTO_CRLF;
1063                         else if (eol_attr == EOL_LF)
1064                                 ca->crlf_action = CRLF_TEXT_INPUT;
1065                         else if (eol_attr == EOL_CRLF)
1066                                 ca->crlf_action = CRLF_TEXT_CRLF;
1067                 }
1068         } else {
1069                 ca->drv = NULL;
1070                 ca->crlf_action = CRLF_UNDEFINED;
1071                 ca->ident = 0;
1072         }
1073
1074         /* Save attr and make a decision for action */
1075         ca->attr_action = ca->crlf_action;
1076         if (ca->crlf_action == CRLF_TEXT)
1077                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1078         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1079                 ca->crlf_action = CRLF_BINARY;
1080         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1081                 ca->crlf_action = CRLF_AUTO_CRLF;
1082         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1083                 ca->crlf_action = CRLF_AUTO_INPUT;
1084 }
1085
1086 int would_convert_to_git_filter_fd(const char *path)
1087 {
1088         struct conv_attrs ca;
1089
1090         convert_attrs(&ca, path);
1091         if (!ca.drv)
1092                 return 0;
1093
1094         /*
1095          * Apply a filter to an fd only if the filter is required to succeed.
1096          * We must die if the filter fails, because the original data before
1097          * filtering is not available.
1098          */
1099         if (!ca.drv->required)
1100                 return 0;
1101
1102         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN, NULL);
1103 }
1104
1105 const char *get_convert_attr_ascii(const char *path)
1106 {
1107         struct conv_attrs ca;
1108
1109         convert_attrs(&ca, path);
1110         switch (ca.attr_action) {
1111         case CRLF_UNDEFINED:
1112                 return "";
1113         case CRLF_BINARY:
1114                 return "-text";
1115         case CRLF_TEXT:
1116                 return "text";
1117         case CRLF_TEXT_INPUT:
1118                 return "text eol=lf";
1119         case CRLF_TEXT_CRLF:
1120                 return "text eol=crlf";
1121         case CRLF_AUTO:
1122                 return "text=auto";
1123         case CRLF_AUTO_CRLF:
1124                 return "text=auto eol=crlf";
1125         case CRLF_AUTO_INPUT:
1126                 return "text=auto eol=lf";
1127         }
1128         return "";
1129 }
1130
1131 int convert_to_git(const struct index_state *istate,
1132                    const char *path, const char *src, size_t len,
1133                    struct strbuf *dst, int conv_flags)
1134 {
1135         int ret = 0;
1136         struct conv_attrs ca;
1137
1138         convert_attrs(&ca, path);
1139
1140         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN, NULL);
1141         if (!ret && ca.drv && ca.drv->required)
1142                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1143
1144         if (ret && dst) {
1145                 src = dst->buf;
1146                 len = dst->len;
1147         }
1148         if (!(conv_flags & CONV_EOL_KEEP_CRLF)) {
1149                 ret |= crlf_to_git(istate, path, src, len, dst, ca.crlf_action, conv_flags);
1150                 if (ret && dst) {
1151                         src = dst->buf;
1152                         len = dst->len;
1153                 }
1154         }
1155         return ret | ident_to_git(path, src, len, dst, ca.ident);
1156 }
1157
1158 void convert_to_git_filter_fd(const struct index_state *istate,
1159                               const char *path, int fd, struct strbuf *dst,
1160                               int conv_flags)
1161 {
1162         struct conv_attrs ca;
1163         convert_attrs(&ca, path);
1164
1165         assert(ca.drv);
1166         assert(ca.drv->clean || ca.drv->process);
1167
1168         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN, NULL))
1169                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1170
1171         crlf_to_git(istate, path, dst->buf, dst->len, dst, ca.crlf_action, conv_flags);
1172         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
1173 }
1174
1175 static int convert_to_working_tree_internal(const char *path, const char *src,
1176                                             size_t len, struct strbuf *dst,
1177                                             int normalizing, struct delayed_checkout *dco)
1178 {
1179         int ret = 0, ret_filter = 0;
1180         struct conv_attrs ca;
1181
1182         convert_attrs(&ca, path);
1183
1184         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
1185         if (ret) {
1186                 src = dst->buf;
1187                 len = dst->len;
1188         }
1189         /*
1190          * CRLF conversion can be skipped if normalizing, unless there
1191          * is a smudge or process filter (even if the process filter doesn't
1192          * support smudge).  The filters might expect CRLFs.
1193          */
1194         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1195                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
1196                 if (ret) {
1197                         src = dst->buf;
1198                         len = dst->len;
1199                 }
1200         }
1201
1202         ret_filter = apply_filter(
1203                 path, src, len, -1, dst, ca.drv, CAP_SMUDGE, dco);
1204         if (!ret_filter && ca.drv && ca.drv->required)
1205                 die("%s: smudge filter %s failed", path, ca.drv->name);
1206
1207         return ret | ret_filter;
1208 }
1209
1210 int async_convert_to_working_tree(const char *path, const char *src,
1211                                   size_t len, struct strbuf *dst,
1212                                   void *dco)
1213 {
1214         return convert_to_working_tree_internal(path, src, len, dst, 0, dco);
1215 }
1216
1217 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
1218 {
1219         return convert_to_working_tree_internal(path, src, len, dst, 0, NULL);
1220 }
1221
1222 int renormalize_buffer(const struct index_state *istate, const char *path,
1223                        const char *src, size_t len, struct strbuf *dst)
1224 {
1225         int ret = convert_to_working_tree_internal(path, src, len, dst, 1, NULL);
1226         if (ret) {
1227                 src = dst->buf;
1228                 len = dst->len;
1229         }
1230         return ret | convert_to_git(istate, path, src, len, dst, CONV_EOL_RENORMALIZE);
1231 }
1232
1233 /*****************************************************************
1234  *
1235  * Streaming conversion support
1236  *
1237  *****************************************************************/
1238
1239 typedef int (*filter_fn)(struct stream_filter *,
1240                          const char *input, size_t *isize_p,
1241                          char *output, size_t *osize_p);
1242 typedef void (*free_fn)(struct stream_filter *);
1243
1244 struct stream_filter_vtbl {
1245         filter_fn filter;
1246         free_fn free;
1247 };
1248
1249 struct stream_filter {
1250         struct stream_filter_vtbl *vtbl;
1251 };
1252
1253 static int null_filter_fn(struct stream_filter *filter,
1254                           const char *input, size_t *isize_p,
1255                           char *output, size_t *osize_p)
1256 {
1257         size_t count;
1258
1259         if (!input)
1260                 return 0; /* we do not keep any states */
1261         count = *isize_p;
1262         if (*osize_p < count)
1263                 count = *osize_p;
1264         if (count) {
1265                 memmove(output, input, count);
1266                 *isize_p -= count;
1267                 *osize_p -= count;
1268         }
1269         return 0;
1270 }
1271
1272 static void null_free_fn(struct stream_filter *filter)
1273 {
1274         ; /* nothing -- null instances are shared */
1275 }
1276
1277 static struct stream_filter_vtbl null_vtbl = {
1278         null_filter_fn,
1279         null_free_fn,
1280 };
1281
1282 static struct stream_filter null_filter_singleton = {
1283         &null_vtbl,
1284 };
1285
1286 int is_null_stream_filter(struct stream_filter *filter)
1287 {
1288         return filter == &null_filter_singleton;
1289 }
1290
1291
1292 /*
1293  * LF-to-CRLF filter
1294  */
1295
1296 struct lf_to_crlf_filter {
1297         struct stream_filter filter;
1298         unsigned has_held:1;
1299         char held;
1300 };
1301
1302 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1303                                 const char *input, size_t *isize_p,
1304                                 char *output, size_t *osize_p)
1305 {
1306         size_t count, o = 0;
1307         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1308
1309         /*
1310          * We may be holding onto the CR to see if it is followed by a
1311          * LF, in which case we would need to go to the main loop.
1312          * Otherwise, just emit it to the output stream.
1313          */
1314         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1315                 output[o++] = lf_to_crlf->held;
1316                 lf_to_crlf->has_held = 0;
1317         }
1318
1319         /* We are told to drain */
1320         if (!input) {
1321                 *osize_p -= o;
1322                 return 0;
1323         }
1324
1325         count = *isize_p;
1326         if (count || lf_to_crlf->has_held) {
1327                 size_t i;
1328                 int was_cr = 0;
1329
1330                 if (lf_to_crlf->has_held) {
1331                         was_cr = 1;
1332                         lf_to_crlf->has_held = 0;
1333                 }
1334
1335                 for (i = 0; o < *osize_p && i < count; i++) {
1336                         char ch = input[i];
1337
1338                         if (ch == '\n') {
1339                                 output[o++] = '\r';
1340                         } else if (was_cr) {
1341                                 /*
1342                                  * Previous round saw CR and it is not followed
1343                                  * by a LF; emit the CR before processing the
1344                                  * current character.
1345                                  */
1346                                 output[o++] = '\r';
1347                         }
1348
1349                         /*
1350                          * We may have consumed the last output slot,
1351                          * in which case we need to break out of this
1352                          * loop; hold the current character before
1353                          * returning.
1354                          */
1355                         if (*osize_p <= o) {
1356                                 lf_to_crlf->has_held = 1;
1357                                 lf_to_crlf->held = ch;
1358                                 continue; /* break but increment i */
1359                         }
1360
1361                         if (ch == '\r') {
1362                                 was_cr = 1;
1363                                 continue;
1364                         }
1365
1366                         was_cr = 0;
1367                         output[o++] = ch;
1368                 }
1369
1370                 *osize_p -= o;
1371                 *isize_p -= i;
1372
1373                 if (!lf_to_crlf->has_held && was_cr) {
1374                         lf_to_crlf->has_held = 1;
1375                         lf_to_crlf->held = '\r';
1376                 }
1377         }
1378         return 0;
1379 }
1380
1381 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1382 {
1383         free(filter);
1384 }
1385
1386 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1387         lf_to_crlf_filter_fn,
1388         lf_to_crlf_free_fn,
1389 };
1390
1391 static struct stream_filter *lf_to_crlf_filter(void)
1392 {
1393         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1394
1395         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1396         return (struct stream_filter *)lf_to_crlf;
1397 }
1398
1399 /*
1400  * Cascade filter
1401  */
1402 #define FILTER_BUFFER 1024
1403 struct cascade_filter {
1404         struct stream_filter filter;
1405         struct stream_filter *one;
1406         struct stream_filter *two;
1407         char buf[FILTER_BUFFER];
1408         int end, ptr;
1409 };
1410
1411 static int cascade_filter_fn(struct stream_filter *filter,
1412                              const char *input, size_t *isize_p,
1413                              char *output, size_t *osize_p)
1414 {
1415         struct cascade_filter *cas = (struct cascade_filter *) filter;
1416         size_t filled = 0;
1417         size_t sz = *osize_p;
1418         size_t to_feed, remaining;
1419
1420         /*
1421          * input -- (one) --> buf -- (two) --> output
1422          */
1423         while (filled < sz) {
1424                 remaining = sz - filled;
1425
1426                 /* do we already have something to feed two with? */
1427                 if (cas->ptr < cas->end) {
1428                         to_feed = cas->end - cas->ptr;
1429                         if (stream_filter(cas->two,
1430                                           cas->buf + cas->ptr, &to_feed,
1431                                           output + filled, &remaining))
1432                                 return -1;
1433                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1434                         filled = sz - remaining;
1435                         continue;
1436                 }
1437
1438                 /* feed one from upstream and have it emit into our buffer */
1439                 to_feed = input ? *isize_p : 0;
1440                 if (input && !to_feed)
1441                         break;
1442                 remaining = sizeof(cas->buf);
1443                 if (stream_filter(cas->one,
1444                                   input, &to_feed,
1445                                   cas->buf, &remaining))
1446                         return -1;
1447                 cas->end = sizeof(cas->buf) - remaining;
1448                 cas->ptr = 0;
1449                 if (input) {
1450                         size_t fed = *isize_p - to_feed;
1451                         *isize_p -= fed;
1452                         input += fed;
1453                 }
1454
1455                 /* do we know that we drained one completely? */
1456                 if (input || cas->end)
1457                         continue;
1458
1459                 /* tell two to drain; we have nothing more to give it */
1460                 to_feed = 0;
1461                 remaining = sz - filled;
1462                 if (stream_filter(cas->two,
1463                                   NULL, &to_feed,
1464                                   output + filled, &remaining))
1465                         return -1;
1466                 if (remaining == (sz - filled))
1467                         break; /* completely drained two */
1468                 filled = sz - remaining;
1469         }
1470         *osize_p -= filled;
1471         return 0;
1472 }
1473
1474 static void cascade_free_fn(struct stream_filter *filter)
1475 {
1476         struct cascade_filter *cas = (struct cascade_filter *)filter;
1477         free_stream_filter(cas->one);
1478         free_stream_filter(cas->two);
1479         free(filter);
1480 }
1481
1482 static struct stream_filter_vtbl cascade_vtbl = {
1483         cascade_filter_fn,
1484         cascade_free_fn,
1485 };
1486
1487 static struct stream_filter *cascade_filter(struct stream_filter *one,
1488                                             struct stream_filter *two)
1489 {
1490         struct cascade_filter *cascade;
1491
1492         if (!one || is_null_stream_filter(one))
1493                 return two;
1494         if (!two || is_null_stream_filter(two))
1495                 return one;
1496
1497         cascade = xmalloc(sizeof(*cascade));
1498         cascade->one = one;
1499         cascade->two = two;
1500         cascade->end = cascade->ptr = 0;
1501         cascade->filter.vtbl = &cascade_vtbl;
1502         return (struct stream_filter *)cascade;
1503 }
1504
1505 /*
1506  * ident filter
1507  */
1508 #define IDENT_DRAINING (-1)
1509 #define IDENT_SKIPPING (-2)
1510 struct ident_filter {
1511         struct stream_filter filter;
1512         struct strbuf left;
1513         int state;
1514         char ident[GIT_MAX_HEXSZ + 5]; /* ": x40 $" */
1515 };
1516
1517 static int is_foreign_ident(const char *str)
1518 {
1519         int i;
1520
1521         if (!skip_prefix(str, "$Id: ", &str))
1522                 return 0;
1523         for (i = 0; str[i]; i++) {
1524                 if (isspace(str[i]) && str[i+1] != '$')
1525                         return 1;
1526         }
1527         return 0;
1528 }
1529
1530 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1531 {
1532         size_t to_drain = ident->left.len;
1533
1534         if (*osize_p < to_drain)
1535                 to_drain = *osize_p;
1536         if (to_drain) {
1537                 memcpy(*output_p, ident->left.buf, to_drain);
1538                 strbuf_remove(&ident->left, 0, to_drain);
1539                 *output_p += to_drain;
1540                 *osize_p -= to_drain;
1541         }
1542         if (!ident->left.len)
1543                 ident->state = 0;
1544 }
1545
1546 static int ident_filter_fn(struct stream_filter *filter,
1547                            const char *input, size_t *isize_p,
1548                            char *output, size_t *osize_p)
1549 {
1550         struct ident_filter *ident = (struct ident_filter *)filter;
1551         static const char head[] = "$Id";
1552
1553         if (!input) {
1554                 /* drain upon eof */
1555                 switch (ident->state) {
1556                 default:
1557                         strbuf_add(&ident->left, head, ident->state);
1558                         /* fallthrough */
1559                 case IDENT_SKIPPING:
1560                         /* fallthrough */
1561                 case IDENT_DRAINING:
1562                         ident_drain(ident, &output, osize_p);
1563                 }
1564                 return 0;
1565         }
1566
1567         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1568                 int ch;
1569
1570                 if (ident->state == IDENT_DRAINING) {
1571                         ident_drain(ident, &output, osize_p);
1572                         if (!*osize_p)
1573                                 break;
1574                         continue;
1575                 }
1576
1577                 ch = *(input++);
1578                 (*isize_p)--;
1579
1580                 if (ident->state == IDENT_SKIPPING) {
1581                         /*
1582                          * Skipping until '$' or LF, but keeping them
1583                          * in case it is a foreign ident.
1584                          */
1585                         strbuf_addch(&ident->left, ch);
1586                         if (ch != '\n' && ch != '$')
1587                                 continue;
1588                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1589                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1590                                 strbuf_addstr(&ident->left, ident->ident);
1591                         }
1592                         ident->state = IDENT_DRAINING;
1593                         continue;
1594                 }
1595
1596                 if (ident->state < sizeof(head) &&
1597                     head[ident->state] == ch) {
1598                         ident->state++;
1599                         continue;
1600                 }
1601
1602                 if (ident->state)
1603                         strbuf_add(&ident->left, head, ident->state);
1604                 if (ident->state == sizeof(head) - 1) {
1605                         if (ch != ':' && ch != '$') {
1606                                 strbuf_addch(&ident->left, ch);
1607                                 ident->state = 0;
1608                                 continue;
1609                         }
1610
1611                         if (ch == ':') {
1612                                 strbuf_addch(&ident->left, ch);
1613                                 ident->state = IDENT_SKIPPING;
1614                         } else {
1615                                 strbuf_addstr(&ident->left, ident->ident);
1616                                 ident->state = IDENT_DRAINING;
1617                         }
1618                         continue;
1619                 }
1620
1621                 strbuf_addch(&ident->left, ch);
1622                 ident->state = IDENT_DRAINING;
1623         }
1624         return 0;
1625 }
1626
1627 static void ident_free_fn(struct stream_filter *filter)
1628 {
1629         struct ident_filter *ident = (struct ident_filter *)filter;
1630         strbuf_release(&ident->left);
1631         free(filter);
1632 }
1633
1634 static struct stream_filter_vtbl ident_vtbl = {
1635         ident_filter_fn,
1636         ident_free_fn,
1637 };
1638
1639 static struct stream_filter *ident_filter(const struct object_id *oid)
1640 {
1641         struct ident_filter *ident = xmalloc(sizeof(*ident));
1642
1643         xsnprintf(ident->ident, sizeof(ident->ident),
1644                   ": %s $", oid_to_hex(oid));
1645         strbuf_init(&ident->left, 0);
1646         ident->filter.vtbl = &ident_vtbl;
1647         ident->state = 0;
1648         return (struct stream_filter *)ident;
1649 }
1650
1651 /*
1652  * Return an appropriately constructed filter for the path, or NULL if
1653  * the contents cannot be filtered without reading the whole thing
1654  * in-core.
1655  *
1656  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1657  * large binary blob you would want us not to slurp into the memory!
1658  */
1659 struct stream_filter *get_stream_filter(const char *path, const struct object_id *oid)
1660 {
1661         struct conv_attrs ca;
1662         struct stream_filter *filter = NULL;
1663
1664         convert_attrs(&ca, path);
1665         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1666                 return NULL;
1667
1668         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1669                 return NULL;
1670
1671         if (ca.ident)
1672                 filter = ident_filter(oid);
1673
1674         if (output_eol(ca.crlf_action) == EOL_CRLF)
1675                 filter = cascade_filter(filter, lf_to_crlf_filter());
1676         else
1677                 filter = cascade_filter(filter, &null_filter_singleton);
1678
1679         return filter;
1680 }
1681
1682 void free_stream_filter(struct stream_filter *filter)
1683 {
1684         filter->vtbl->free(filter);
1685 }
1686
1687 int stream_filter(struct stream_filter *filter,
1688                   const char *input, size_t *isize_p,
1689                   char *output, size_t *osize_p)
1690 {
1691         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1692 }