diff-highlight: add clean target to Makefile
[git] / convert.c
1 #define NO_THE_INDEX_COMPATIBILITY_MACROS
2 #include "cache.h"
3 #include "config.h"
4 #include "attr.h"
5 #include "run-command.h"
6 #include "quote.h"
7 #include "sigchain.h"
8 #include "pkt-line.h"
9 #include "sub-process.h"
10
11 /*
12  * convert.c - convert a file when checking it out and checking it in.
13  *
14  * This should use the pathname to decide on whether it wants to do some
15  * more interesting conversions (automatic gzip/unzip, general format
16  * conversions etc etc), but by default it just does automatic CRLF<->LF
17  * translation when the "text" attribute or "auto_crlf" option is set.
18  */
19
20 /* Stat bits: When BIN is set, the txt bits are unset */
21 #define CONVERT_STAT_BITS_TXT_LF    0x1
22 #define CONVERT_STAT_BITS_TXT_CRLF  0x2
23 #define CONVERT_STAT_BITS_BIN       0x4
24
25 enum crlf_action {
26         CRLF_UNDEFINED,
27         CRLF_BINARY,
28         CRLF_TEXT,
29         CRLF_TEXT_INPUT,
30         CRLF_TEXT_CRLF,
31         CRLF_AUTO,
32         CRLF_AUTO_INPUT,
33         CRLF_AUTO_CRLF
34 };
35
36 struct text_stat {
37         /* NUL, CR, LF and CRLF counts */
38         unsigned nul, lonecr, lonelf, crlf;
39
40         /* These are just approximations! */
41         unsigned printable, nonprintable;
42 };
43
44 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
45 {
46         unsigned long i;
47
48         memset(stats, 0, sizeof(*stats));
49
50         for (i = 0; i < size; i++) {
51                 unsigned char c = buf[i];
52                 if (c == '\r') {
53                         if (i+1 < size && buf[i+1] == '\n') {
54                                 stats->crlf++;
55                                 i++;
56                         } else
57                                 stats->lonecr++;
58                         continue;
59                 }
60                 if (c == '\n') {
61                         stats->lonelf++;
62                         continue;
63                 }
64                 if (c == 127)
65                         /* DEL */
66                         stats->nonprintable++;
67                 else if (c < 32) {
68                         switch (c) {
69                                 /* BS, HT, ESC and FF */
70                         case '\b': case '\t': case '\033': case '\014':
71                                 stats->printable++;
72                                 break;
73                         case 0:
74                                 stats->nul++;
75                                 /* fall through */
76                         default:
77                                 stats->nonprintable++;
78                         }
79                 }
80                 else
81                         stats->printable++;
82         }
83
84         /* If file ends with EOF then don't count this EOF as non-printable. */
85         if (size >= 1 && buf[size-1] == '\032')
86                 stats->nonprintable--;
87 }
88
89 /*
90  * The same heuristics as diff.c::mmfile_is_binary()
91  * We treat files with bare CR as binary
92  */
93 static int convert_is_binary(unsigned long size, const struct text_stat *stats)
94 {
95         if (stats->lonecr)
96                 return 1;
97         if (stats->nul)
98                 return 1;
99         if ((stats->printable >> 7) < stats->nonprintable)
100                 return 1;
101         return 0;
102 }
103
104 static unsigned int gather_convert_stats(const char *data, unsigned long size)
105 {
106         struct text_stat stats;
107         int ret = 0;
108         if (!data || !size)
109                 return 0;
110         gather_stats(data, size, &stats);
111         if (convert_is_binary(size, &stats))
112                 ret |= CONVERT_STAT_BITS_BIN;
113         if (stats.crlf)
114                 ret |= CONVERT_STAT_BITS_TXT_CRLF;
115         if (stats.lonelf)
116                 ret |=  CONVERT_STAT_BITS_TXT_LF;
117
118         return ret;
119 }
120
121 static const char *gather_convert_stats_ascii(const char *data, unsigned long size)
122 {
123         unsigned int convert_stats = gather_convert_stats(data, size);
124
125         if (convert_stats & CONVERT_STAT_BITS_BIN)
126                 return "-text";
127         switch (convert_stats) {
128         case CONVERT_STAT_BITS_TXT_LF:
129                 return "lf";
130         case CONVERT_STAT_BITS_TXT_CRLF:
131                 return "crlf";
132         case CONVERT_STAT_BITS_TXT_LF | CONVERT_STAT_BITS_TXT_CRLF:
133                 return "mixed";
134         default:
135                 return "none";
136         }
137 }
138
139 const char *get_cached_convert_stats_ascii(const struct index_state *istate,
140                                            const char *path)
141 {
142         const char *ret;
143         unsigned long sz;
144         void *data = read_blob_data_from_index(istate, path, &sz);
145         ret = gather_convert_stats_ascii(data, sz);
146         free(data);
147         return ret;
148 }
149
150 const char *get_wt_convert_stats_ascii(const char *path)
151 {
152         const char *ret = "";
153         struct strbuf sb = STRBUF_INIT;
154         if (strbuf_read_file(&sb, path, 0) >= 0)
155                 ret = gather_convert_stats_ascii(sb.buf, sb.len);
156         strbuf_release(&sb);
157         return ret;
158 }
159
160 static int text_eol_is_crlf(void)
161 {
162         if (auto_crlf == AUTO_CRLF_TRUE)
163                 return 1;
164         else if (auto_crlf == AUTO_CRLF_INPUT)
165                 return 0;
166         if (core_eol == EOL_CRLF)
167                 return 1;
168         if (core_eol == EOL_UNSET && EOL_NATIVE == EOL_CRLF)
169                 return 1;
170         return 0;
171 }
172
173 static enum eol output_eol(enum crlf_action crlf_action)
174 {
175         switch (crlf_action) {
176         case CRLF_BINARY:
177                 return EOL_UNSET;
178         case CRLF_TEXT_CRLF:
179                 return EOL_CRLF;
180         case CRLF_TEXT_INPUT:
181                 return EOL_LF;
182         case CRLF_UNDEFINED:
183         case CRLF_AUTO_CRLF:
184                 return EOL_CRLF;
185         case CRLF_AUTO_INPUT:
186                 return EOL_LF;
187         case CRLF_TEXT:
188         case CRLF_AUTO:
189                 /* fall through */
190                 return text_eol_is_crlf() ? EOL_CRLF : EOL_LF;
191         }
192         warning("Illegal crlf_action %d\n", (int)crlf_action);
193         return core_eol;
194 }
195
196 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
197                             struct text_stat *old_stats, struct text_stat *new_stats,
198                             enum safe_crlf checksafe)
199 {
200         if (old_stats->crlf && !new_stats->crlf ) {
201                 /*
202                  * CRLFs would not be restored by checkout
203                  */
204                 if (checksafe == SAFE_CRLF_WARN)
205                         warning(_("CRLF will be replaced by LF in %s.\n"
206                                   "The file will have its original line"
207                                   " endings in your working directory."), path);
208                 else /* i.e. SAFE_CRLF_FAIL */
209                         die(_("CRLF would be replaced by LF in %s."), path);
210         } else if (old_stats->lonelf && !new_stats->lonelf ) {
211                 /*
212                  * CRLFs would be added by checkout
213                  */
214                 if (checksafe == SAFE_CRLF_WARN)
215                         warning(_("LF will be replaced by CRLF in %s.\n"
216                                   "The file will have its original line"
217                                   " endings in your working directory."), path);
218                 else /* i.e. SAFE_CRLF_FAIL */
219                         die(_("LF would be replaced by CRLF in %s"), path);
220         }
221 }
222
223 static int has_cr_in_index(const struct index_state *istate, const char *path)
224 {
225         unsigned long sz;
226         void *data;
227         int has_cr;
228
229         data = read_blob_data_from_index(istate, path, &sz);
230         if (!data)
231                 return 0;
232         has_cr = memchr(data, '\r', sz) != NULL;
233         free(data);
234         return has_cr;
235 }
236
237 static int will_convert_lf_to_crlf(size_t len, struct text_stat *stats,
238                                    enum crlf_action crlf_action)
239 {
240         if (output_eol(crlf_action) != EOL_CRLF)
241                 return 0;
242         /* No "naked" LF? Nothing to convert, regardless. */
243         if (!stats->lonelf)
244                 return 0;
245
246         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
247                 /* If we have any CR or CRLF line endings, we do not touch it */
248                 /* This is the new safer autocrlf-handling */
249                 if (stats->lonecr || stats->crlf)
250                         return 0;
251
252                 if (convert_is_binary(len, stats))
253                         return 0;
254         }
255         return 1;
256
257 }
258
259 static int crlf_to_git(const struct index_state *istate,
260                        const char *path, const char *src, size_t len,
261                        struct strbuf *buf,
262                        enum crlf_action crlf_action, enum safe_crlf checksafe)
263 {
264         struct text_stat stats;
265         char *dst;
266         int convert_crlf_into_lf;
267
268         if (crlf_action == CRLF_BINARY ||
269             (src && !len))
270                 return 0;
271
272         /*
273          * If we are doing a dry-run and have no source buffer, there is
274          * nothing to analyze; we must assume we would convert.
275          */
276         if (!buf && !src)
277                 return 1;
278
279         gather_stats(src, len, &stats);
280         /* Optimization: No CRLF? Nothing to convert, regardless. */
281         convert_crlf_into_lf = !!stats.crlf;
282
283         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
284                 if (convert_is_binary(len, &stats))
285                         return 0;
286                 /*
287                  * If the file in the index has any CR in it, do not
288                  * convert.  This is the new safer autocrlf handling,
289                  * unless we want to renormalize in a merge or
290                  * cherry-pick.
291                  */
292                 if ((checksafe != SAFE_CRLF_RENORMALIZE) &&
293                     has_cr_in_index(istate, path))
294                         convert_crlf_into_lf = 0;
295         }
296         if ((checksafe == SAFE_CRLF_WARN ||
297             (checksafe == SAFE_CRLF_FAIL)) && len) {
298                 struct text_stat new_stats;
299                 memcpy(&new_stats, &stats, sizeof(new_stats));
300                 /* simulate "git add" */
301                 if (convert_crlf_into_lf) {
302                         new_stats.lonelf += new_stats.crlf;
303                         new_stats.crlf = 0;
304                 }
305                 /* simulate "git checkout" */
306                 if (will_convert_lf_to_crlf(len, &new_stats, crlf_action)) {
307                         new_stats.crlf += new_stats.lonelf;
308                         new_stats.lonelf = 0;
309                 }
310                 check_safe_crlf(path, crlf_action, &stats, &new_stats, checksafe);
311         }
312         if (!convert_crlf_into_lf)
313                 return 0;
314
315         /*
316          * At this point all of our source analysis is done, and we are sure we
317          * would convert. If we are in dry-run mode, we can give an answer.
318          */
319         if (!buf)
320                 return 1;
321
322         /* only grow if not in place */
323         if (strbuf_avail(buf) + buf->len < len)
324                 strbuf_grow(buf, len - buf->len);
325         dst = buf->buf;
326         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
327                 /*
328                  * If we guessed, we already know we rejected a file with
329                  * lone CR, and we can strip a CR without looking at what
330                  * follow it.
331                  */
332                 do {
333                         unsigned char c = *src++;
334                         if (c != '\r')
335                                 *dst++ = c;
336                 } while (--len);
337         } else {
338                 do {
339                         unsigned char c = *src++;
340                         if (! (c == '\r' && (1 < len && *src == '\n')))
341                                 *dst++ = c;
342                 } while (--len);
343         }
344         strbuf_setlen(buf, dst - buf->buf);
345         return 1;
346 }
347
348 static int crlf_to_worktree(const char *path, const char *src, size_t len,
349                             struct strbuf *buf, enum crlf_action crlf_action)
350 {
351         char *to_free = NULL;
352         struct text_stat stats;
353
354         if (!len || output_eol(crlf_action) != EOL_CRLF)
355                 return 0;
356
357         gather_stats(src, len, &stats);
358         if (!will_convert_lf_to_crlf(len, &stats, crlf_action))
359                 return 0;
360
361         /* are we "faking" in place editing ? */
362         if (src == buf->buf)
363                 to_free = strbuf_detach(buf, NULL);
364
365         strbuf_grow(buf, len + stats.lonelf);
366         for (;;) {
367                 const char *nl = memchr(src, '\n', len);
368                 if (!nl)
369                         break;
370                 if (nl > src && nl[-1] == '\r') {
371                         strbuf_add(buf, src, nl + 1 - src);
372                 } else {
373                         strbuf_add(buf, src, nl - src);
374                         strbuf_addstr(buf, "\r\n");
375                 }
376                 len -= nl + 1 - src;
377                 src  = nl + 1;
378         }
379         strbuf_add(buf, src, len);
380
381         free(to_free);
382         return 1;
383 }
384
385 struct filter_params {
386         const char *src;
387         unsigned long size;
388         int fd;
389         const char *cmd;
390         const char *path;
391 };
392
393 static int filter_buffer_or_fd(int in, int out, void *data)
394 {
395         /*
396          * Spawn cmd and feed the buffer contents through its stdin.
397          */
398         struct child_process child_process = CHILD_PROCESS_INIT;
399         struct filter_params *params = (struct filter_params *)data;
400         int write_err, status;
401         const char *argv[] = { NULL, NULL };
402
403         /* apply % substitution to cmd */
404         struct strbuf cmd = STRBUF_INIT;
405         struct strbuf path = STRBUF_INIT;
406         struct strbuf_expand_dict_entry dict[] = {
407                 { "f", NULL, },
408                 { NULL, NULL, },
409         };
410
411         /* quote the path to preserve spaces, etc. */
412         sq_quote_buf(&path, params->path);
413         dict[0].value = path.buf;
414
415         /* expand all %f with the quoted path */
416         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
417         strbuf_release(&path);
418
419         argv[0] = cmd.buf;
420
421         child_process.argv = argv;
422         child_process.use_shell = 1;
423         child_process.in = -1;
424         child_process.out = out;
425
426         if (start_command(&child_process))
427                 return error("cannot fork to run external filter '%s'", params->cmd);
428
429         sigchain_push(SIGPIPE, SIG_IGN);
430
431         if (params->src) {
432                 write_err = (write_in_full(child_process.in,
433                                            params->src, params->size) < 0);
434                 if (errno == EPIPE)
435                         write_err = 0;
436         } else {
437                 write_err = copy_fd(params->fd, child_process.in);
438                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
439                         write_err = 0;
440         }
441
442         if (close(child_process.in))
443                 write_err = 1;
444         if (write_err)
445                 error("cannot feed the input to external filter '%s'", params->cmd);
446
447         sigchain_pop(SIGPIPE);
448
449         status = finish_command(&child_process);
450         if (status)
451                 error("external filter '%s' failed %d", params->cmd, status);
452
453         strbuf_release(&cmd);
454         return (write_err || status);
455 }
456
457 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
458                         struct strbuf *dst, const char *cmd)
459 {
460         /*
461          * Create a pipeline to have the command filter the buffer's
462          * contents.
463          *
464          * (child --> cmd) --> us
465          */
466         int err = 0;
467         struct strbuf nbuf = STRBUF_INIT;
468         struct async async;
469         struct filter_params params;
470
471         memset(&async, 0, sizeof(async));
472         async.proc = filter_buffer_or_fd;
473         async.data = &params;
474         async.out = -1;
475         params.src = src;
476         params.size = len;
477         params.fd = fd;
478         params.cmd = cmd;
479         params.path = path;
480
481         fflush(NULL);
482         if (start_async(&async))
483                 return 0;       /* error was already reported */
484
485         if (strbuf_read(&nbuf, async.out, len) < 0) {
486                 err = error("read from external filter '%s' failed", cmd);
487         }
488         if (close(async.out)) {
489                 err = error("read from external filter '%s' failed", cmd);
490         }
491         if (finish_async(&async)) {
492                 err = error("external filter '%s' failed", cmd);
493         }
494
495         if (!err) {
496                 strbuf_swap(dst, &nbuf);
497         }
498         strbuf_release(&nbuf);
499         return !err;
500 }
501
502 #define CAP_CLEAN    (1u<<0)
503 #define CAP_SMUDGE   (1u<<1)
504 #define CAP_DELAY    (1u<<2)
505
506 struct cmd2process {
507         struct subprocess_entry subprocess; /* must be the first member! */
508         unsigned int supported_capabilities;
509 };
510
511 static int subprocess_map_initialized;
512 static struct hashmap subprocess_map;
513
514 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
515 {
516         static int versions[] = {2, 0};
517         static struct subprocess_capability capabilities[] = {
518                 { "clean",  CAP_CLEAN  },
519                 { "smudge", CAP_SMUDGE },
520                 { "delay",  CAP_DELAY  },
521                 { NULL, 0 }
522         };
523         struct cmd2process *entry = (struct cmd2process *)subprocess;
524         return subprocess_handshake(subprocess, "git-filter", versions, NULL,
525                                     capabilities,
526                                     &entry->supported_capabilities);
527 }
528
529 static void handle_filter_error(const struct strbuf *filter_status,
530                                 struct cmd2process *entry,
531                                 const unsigned int wanted_capability) {
532         if (!strcmp(filter_status->buf, "error"))
533                 ; /* The filter signaled a problem with the file. */
534         else if (!strcmp(filter_status->buf, "abort") && wanted_capability) {
535                 /*
536                  * The filter signaled a permanent problem. Don't try to filter
537                  * files with the same command for the lifetime of the current
538                  * Git process.
539                  */
540                  entry->supported_capabilities &= ~wanted_capability;
541         } else {
542                 /*
543                  * Something went wrong with the protocol filter.
544                  * Force shutdown and restart if another blob requires filtering.
545                  */
546                 error("external filter '%s' failed", entry->subprocess.cmd);
547                 subprocess_stop(&subprocess_map, &entry->subprocess);
548                 free(entry);
549         }
550 }
551
552 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
553                                    int fd, struct strbuf *dst, const char *cmd,
554                                    const unsigned int wanted_capability,
555                                    struct delayed_checkout *dco)
556 {
557         int err;
558         int can_delay = 0;
559         struct cmd2process *entry;
560         struct child_process *process;
561         struct strbuf nbuf = STRBUF_INIT;
562         struct strbuf filter_status = STRBUF_INIT;
563         const char *filter_type;
564
565         if (!subprocess_map_initialized) {
566                 subprocess_map_initialized = 1;
567                 hashmap_init(&subprocess_map, (hashmap_cmp_fn) cmd2process_cmp,
568                              NULL, 0);
569                 entry = NULL;
570         } else {
571                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
572         }
573
574         fflush(NULL);
575
576         if (!entry) {
577                 entry = xmalloc(sizeof(*entry));
578                 entry->supported_capabilities = 0;
579
580                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
581                         free(entry);
582                         return 0;
583                 }
584         }
585         process = &entry->subprocess.process;
586
587         if (!(entry->supported_capabilities & wanted_capability))
588                 return 0;
589
590         if (wanted_capability & CAP_CLEAN)
591                 filter_type = "clean";
592         else if (wanted_capability & CAP_SMUDGE)
593                 filter_type = "smudge";
594         else
595                 die("unexpected filter type");
596
597         sigchain_push(SIGPIPE, SIG_IGN);
598
599         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
600         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
601         if (err)
602                 goto done;
603
604         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
605         if (err) {
606                 error("path name too long for external filter");
607                 goto done;
608         }
609
610         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
611         if (err)
612                 goto done;
613
614         if ((entry->supported_capabilities & CAP_DELAY) &&
615             dco && dco->state == CE_CAN_DELAY) {
616                 can_delay = 1;
617                 err = packet_write_fmt_gently(process->in, "can-delay=1\n");
618                 if (err)
619                         goto done;
620         }
621
622         err = packet_flush_gently(process->in);
623         if (err)
624                 goto done;
625
626         if (fd >= 0)
627                 err = write_packetized_from_fd(fd, process->in);
628         else
629                 err = write_packetized_from_buf(src, len, process->in);
630         if (err)
631                 goto done;
632
633         err = subprocess_read_status(process->out, &filter_status);
634         if (err)
635                 goto done;
636
637         if (can_delay && !strcmp(filter_status.buf, "delayed")) {
638                 string_list_insert(&dco->filters, cmd);
639                 string_list_insert(&dco->paths, path);
640         } else {
641                 /* The filter got the blob and wants to send us a response. */
642                 err = strcmp(filter_status.buf, "success");
643                 if (err)
644                         goto done;
645
646                 err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
647                 if (err)
648                         goto done;
649
650                 err = subprocess_read_status(process->out, &filter_status);
651                 if (err)
652                         goto done;
653
654                 err = strcmp(filter_status.buf, "success");
655         }
656
657 done:
658         sigchain_pop(SIGPIPE);
659
660         if (err)
661                 handle_filter_error(&filter_status, entry, wanted_capability);
662         else
663                 strbuf_swap(dst, &nbuf);
664         strbuf_release(&nbuf);
665         return !err;
666 }
667
668
669 int async_query_available_blobs(const char *cmd, struct string_list *available_paths)
670 {
671         int err;
672         char *line;
673         struct cmd2process *entry;
674         struct child_process *process;
675         struct strbuf filter_status = STRBUF_INIT;
676
677         assert(subprocess_map_initialized);
678         entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
679         if (!entry) {
680                 error("external filter '%s' is not available anymore although "
681                       "not all paths have been filtered", cmd);
682                 return 0;
683         }
684         process = &entry->subprocess.process;
685         sigchain_push(SIGPIPE, SIG_IGN);
686
687         err = packet_write_fmt_gently(
688                 process->in, "command=list_available_blobs\n");
689         if (err)
690                 goto done;
691
692         err = packet_flush_gently(process->in);
693         if (err)
694                 goto done;
695
696         while ((line = packet_read_line(process->out, NULL))) {
697                 const char *path;
698                 if (skip_prefix(line, "pathname=", &path))
699                         string_list_insert(available_paths, xstrdup(path));
700                 else
701                         ; /* ignore unknown keys */
702         }
703
704         err = subprocess_read_status(process->out, &filter_status);
705         if (err)
706                 goto done;
707
708         err = strcmp(filter_status.buf, "success");
709
710 done:
711         sigchain_pop(SIGPIPE);
712
713         if (err)
714                 handle_filter_error(&filter_status, entry, 0);
715         return !err;
716 }
717
718 static struct convert_driver {
719         const char *name;
720         struct convert_driver *next;
721         const char *smudge;
722         const char *clean;
723         const char *process;
724         int required;
725 } *user_convert, **user_convert_tail;
726
727 static int apply_filter(const char *path, const char *src, size_t len,
728                         int fd, struct strbuf *dst, struct convert_driver *drv,
729                         const unsigned int wanted_capability,
730                         struct delayed_checkout *dco)
731 {
732         const char *cmd = NULL;
733
734         if (!drv)
735                 return 0;
736
737         if (!dst)
738                 return 1;
739
740         if ((wanted_capability & CAP_CLEAN) && !drv->process && drv->clean)
741                 cmd = drv->clean;
742         else if ((wanted_capability & CAP_SMUDGE) && !drv->process && drv->smudge)
743                 cmd = drv->smudge;
744
745         if (cmd && *cmd)
746                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
747         else if (drv->process && *drv->process)
748                 return apply_multi_file_filter(path, src, len, fd, dst,
749                         drv->process, wanted_capability, dco);
750
751         return 0;
752 }
753
754 static int read_convert_config(const char *var, const char *value, void *cb)
755 {
756         const char *key, *name;
757         int namelen;
758         struct convert_driver *drv;
759
760         /*
761          * External conversion drivers are configured using
762          * "filter.<name>.variable".
763          */
764         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
765                 return 0;
766         for (drv = user_convert; drv; drv = drv->next)
767                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
768                         break;
769         if (!drv) {
770                 drv = xcalloc(1, sizeof(struct convert_driver));
771                 drv->name = xmemdupz(name, namelen);
772                 *user_convert_tail = drv;
773                 user_convert_tail = &(drv->next);
774         }
775
776         /*
777          * filter.<name>.smudge and filter.<name>.clean specifies
778          * the command line:
779          *
780          *      command-line
781          *
782          * The command-line will not be interpolated in any way.
783          */
784
785         if (!strcmp("smudge", key))
786                 return git_config_string(&drv->smudge, var, value);
787
788         if (!strcmp("clean", key))
789                 return git_config_string(&drv->clean, var, value);
790
791         if (!strcmp("process", key))
792                 return git_config_string(&drv->process, var, value);
793
794         if (!strcmp("required", key)) {
795                 drv->required = git_config_bool(var, value);
796                 return 0;
797         }
798
799         return 0;
800 }
801
802 static int count_ident(const char *cp, unsigned long size)
803 {
804         /*
805          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
806          */
807         int cnt = 0;
808         char ch;
809
810         while (size) {
811                 ch = *cp++;
812                 size--;
813                 if (ch != '$')
814                         continue;
815                 if (size < 3)
816                         break;
817                 if (memcmp("Id", cp, 2))
818                         continue;
819                 ch = cp[2];
820                 cp += 3;
821                 size -= 3;
822                 if (ch == '$')
823                         cnt++; /* $Id$ */
824                 if (ch != ':')
825                         continue;
826
827                 /*
828                  * "$Id: ... "; scan up to the closing dollar sign and discard.
829                  */
830                 while (size) {
831                         ch = *cp++;
832                         size--;
833                         if (ch == '$') {
834                                 cnt++;
835                                 break;
836                         }
837                         if (ch == '\n')
838                                 break;
839                 }
840         }
841         return cnt;
842 }
843
844 static int ident_to_git(const char *path, const char *src, size_t len,
845                         struct strbuf *buf, int ident)
846 {
847         char *dst, *dollar;
848
849         if (!ident || (src && !count_ident(src, len)))
850                 return 0;
851
852         if (!buf)
853                 return 1;
854
855         /* only grow if not in place */
856         if (strbuf_avail(buf) + buf->len < len)
857                 strbuf_grow(buf, len - buf->len);
858         dst = buf->buf;
859         for (;;) {
860                 dollar = memchr(src, '$', len);
861                 if (!dollar)
862                         break;
863                 memmove(dst, src, dollar + 1 - src);
864                 dst += dollar + 1 - src;
865                 len -= dollar + 1 - src;
866                 src  = dollar + 1;
867
868                 if (len > 3 && !memcmp(src, "Id:", 3)) {
869                         dollar = memchr(src + 3, '$', len - 3);
870                         if (!dollar)
871                                 break;
872                         if (memchr(src + 3, '\n', dollar - src - 3)) {
873                                 /* Line break before the next dollar. */
874                                 continue;
875                         }
876
877                         memcpy(dst, "Id$", 3);
878                         dst += 3;
879                         len -= dollar + 1 - src;
880                         src  = dollar + 1;
881                 }
882         }
883         memmove(dst, src, len);
884         strbuf_setlen(buf, dst + len - buf->buf);
885         return 1;
886 }
887
888 static int ident_to_worktree(const char *path, const char *src, size_t len,
889                              struct strbuf *buf, int ident)
890 {
891         unsigned char sha1[20];
892         char *to_free = NULL, *dollar, *spc;
893         int cnt;
894
895         if (!ident)
896                 return 0;
897
898         cnt = count_ident(src, len);
899         if (!cnt)
900                 return 0;
901
902         /* are we "faking" in place editing ? */
903         if (src == buf->buf)
904                 to_free = strbuf_detach(buf, NULL);
905         hash_sha1_file(src, len, "blob", sha1);
906
907         strbuf_grow(buf, len + cnt * 43);
908         for (;;) {
909                 /* step 1: run to the next '$' */
910                 dollar = memchr(src, '$', len);
911                 if (!dollar)
912                         break;
913                 strbuf_add(buf, src, dollar + 1 - src);
914                 len -= dollar + 1 - src;
915                 src  = dollar + 1;
916
917                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
918                 if (len < 3 || memcmp("Id", src, 2))
919                         continue;
920
921                 /* step 3: skip over Id$ or Id:xxxxx$ */
922                 if (src[2] == '$') {
923                         src += 3;
924                         len -= 3;
925                 } else if (src[2] == ':') {
926                         /*
927                          * It's possible that an expanded Id has crept its way into the
928                          * repository, we cope with that by stripping the expansion out.
929                          * This is probably not a good idea, since it will cause changes
930                          * on checkout, which won't go away by stash, but let's keep it
931                          * for git-style ids.
932                          */
933                         dollar = memchr(src + 3, '$', len - 3);
934                         if (!dollar) {
935                                 /* incomplete keyword, no more '$', so just quit the loop */
936                                 break;
937                         }
938
939                         if (memchr(src + 3, '\n', dollar - src - 3)) {
940                                 /* Line break before the next dollar. */
941                                 continue;
942                         }
943
944                         spc = memchr(src + 4, ' ', dollar - src - 4);
945                         if (spc && spc < dollar-1) {
946                                 /* There are spaces in unexpected places.
947                                  * This is probably an id from some other
948                                  * versioning system. Keep it for now.
949                                  */
950                                 continue;
951                         }
952
953                         len -= dollar + 1 - src;
954                         src  = dollar + 1;
955                 } else {
956                         /* it wasn't a "Id$" or "Id:xxxx$" */
957                         continue;
958                 }
959
960                 /* step 4: substitute */
961                 strbuf_addstr(buf, "Id: ");
962                 strbuf_add(buf, sha1_to_hex(sha1), 40);
963                 strbuf_addstr(buf, " $");
964         }
965         strbuf_add(buf, src, len);
966
967         free(to_free);
968         return 1;
969 }
970
971 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
972 {
973         const char *value = check->value;
974
975         if (ATTR_TRUE(value))
976                 return CRLF_TEXT;
977         else if (ATTR_FALSE(value))
978                 return CRLF_BINARY;
979         else if (ATTR_UNSET(value))
980                 ;
981         else if (!strcmp(value, "input"))
982                 return CRLF_TEXT_INPUT;
983         else if (!strcmp(value, "auto"))
984                 return CRLF_AUTO;
985         return CRLF_UNDEFINED;
986 }
987
988 static enum eol git_path_check_eol(struct attr_check_item *check)
989 {
990         const char *value = check->value;
991
992         if (ATTR_UNSET(value))
993                 ;
994         else if (!strcmp(value, "lf"))
995                 return EOL_LF;
996         else if (!strcmp(value, "crlf"))
997                 return EOL_CRLF;
998         return EOL_UNSET;
999 }
1000
1001 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
1002 {
1003         const char *value = check->value;
1004         struct convert_driver *drv;
1005
1006         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
1007                 return NULL;
1008         for (drv = user_convert; drv; drv = drv->next)
1009                 if (!strcmp(value, drv->name))
1010                         return drv;
1011         return NULL;
1012 }
1013
1014 static int git_path_check_ident(struct attr_check_item *check)
1015 {
1016         const char *value = check->value;
1017
1018         return !!ATTR_TRUE(value);
1019 }
1020
1021 struct conv_attrs {
1022         struct convert_driver *drv;
1023         enum crlf_action attr_action; /* What attr says */
1024         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
1025         int ident;
1026 };
1027
1028 static void convert_attrs(struct conv_attrs *ca, const char *path)
1029 {
1030         static struct attr_check *check;
1031
1032         if (!check) {
1033                 check = attr_check_initl("crlf", "ident", "filter",
1034                                          "eol", "text", NULL);
1035                 user_convert_tail = &user_convert;
1036                 git_config(read_convert_config, NULL);
1037         }
1038
1039         if (!git_check_attr(path, check)) {
1040                 struct attr_check_item *ccheck = check->items;
1041                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
1042                 if (ca->crlf_action == CRLF_UNDEFINED)
1043                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
1044                 ca->attr_action = ca->crlf_action;
1045                 ca->ident = git_path_check_ident(ccheck + 1);
1046                 ca->drv = git_path_check_convert(ccheck + 2);
1047                 if (ca->crlf_action != CRLF_BINARY) {
1048                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
1049                         if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1050                                 ca->crlf_action = CRLF_AUTO_INPUT;
1051                         else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1052                                 ca->crlf_action = CRLF_AUTO_CRLF;
1053                         else if (eol_attr == EOL_LF)
1054                                 ca->crlf_action = CRLF_TEXT_INPUT;
1055                         else if (eol_attr == EOL_CRLF)
1056                                 ca->crlf_action = CRLF_TEXT_CRLF;
1057                 }
1058                 ca->attr_action = ca->crlf_action;
1059         } else {
1060                 ca->drv = NULL;
1061                 ca->crlf_action = CRLF_UNDEFINED;
1062                 ca->ident = 0;
1063         }
1064         if (ca->crlf_action == CRLF_TEXT)
1065                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1066         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1067                 ca->crlf_action = CRLF_BINARY;
1068         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1069                 ca->crlf_action = CRLF_AUTO_CRLF;
1070         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1071                 ca->crlf_action = CRLF_AUTO_INPUT;
1072 }
1073
1074 int would_convert_to_git_filter_fd(const char *path)
1075 {
1076         struct conv_attrs ca;
1077
1078         convert_attrs(&ca, path);
1079         if (!ca.drv)
1080                 return 0;
1081
1082         /*
1083          * Apply a filter to an fd only if the filter is required to succeed.
1084          * We must die if the filter fails, because the original data before
1085          * filtering is not available.
1086          */
1087         if (!ca.drv->required)
1088                 return 0;
1089
1090         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN, NULL);
1091 }
1092
1093 const char *get_convert_attr_ascii(const char *path)
1094 {
1095         struct conv_attrs ca;
1096
1097         convert_attrs(&ca, path);
1098         switch (ca.attr_action) {
1099         case CRLF_UNDEFINED:
1100                 return "";
1101         case CRLF_BINARY:
1102                 return "-text";
1103         case CRLF_TEXT:
1104                 return "text";
1105         case CRLF_TEXT_INPUT:
1106                 return "text eol=lf";
1107         case CRLF_TEXT_CRLF:
1108                 return "text eol=crlf";
1109         case CRLF_AUTO:
1110                 return "text=auto";
1111         case CRLF_AUTO_CRLF:
1112                 return "text=auto eol=crlf";
1113         case CRLF_AUTO_INPUT:
1114                 return "text=auto eol=lf";
1115         }
1116         return "";
1117 }
1118
1119 int convert_to_git(const struct index_state *istate,
1120                    const char *path, const char *src, size_t len,
1121                    struct strbuf *dst, enum safe_crlf checksafe)
1122 {
1123         int ret = 0;
1124         struct conv_attrs ca;
1125
1126         convert_attrs(&ca, path);
1127
1128         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN, NULL);
1129         if (!ret && ca.drv && ca.drv->required)
1130                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1131
1132         if (ret && dst) {
1133                 src = dst->buf;
1134                 len = dst->len;
1135         }
1136         ret |= crlf_to_git(istate, path, src, len, dst, ca.crlf_action, checksafe);
1137         if (ret && dst) {
1138                 src = dst->buf;
1139                 len = dst->len;
1140         }
1141         return ret | ident_to_git(path, src, len, dst, ca.ident);
1142 }
1143
1144 void convert_to_git_filter_fd(const struct index_state *istate,
1145                               const char *path, int fd, struct strbuf *dst,
1146                               enum safe_crlf checksafe)
1147 {
1148         struct conv_attrs ca;
1149         convert_attrs(&ca, path);
1150
1151         assert(ca.drv);
1152         assert(ca.drv->clean || ca.drv->process);
1153
1154         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN, NULL))
1155                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1156
1157         crlf_to_git(istate, path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
1158         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
1159 }
1160
1161 static int convert_to_working_tree_internal(const char *path, const char *src,
1162                                             size_t len, struct strbuf *dst,
1163                                             int normalizing, struct delayed_checkout *dco)
1164 {
1165         int ret = 0, ret_filter = 0;
1166         struct conv_attrs ca;
1167
1168         convert_attrs(&ca, path);
1169
1170         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
1171         if (ret) {
1172                 src = dst->buf;
1173                 len = dst->len;
1174         }
1175         /*
1176          * CRLF conversion can be skipped if normalizing, unless there
1177          * is a smudge or process filter (even if the process filter doesn't
1178          * support smudge).  The filters might expect CRLFs.
1179          */
1180         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1181                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
1182                 if (ret) {
1183                         src = dst->buf;
1184                         len = dst->len;
1185                 }
1186         }
1187
1188         ret_filter = apply_filter(
1189                 path, src, len, -1, dst, ca.drv, CAP_SMUDGE, dco);
1190         if (!ret_filter && ca.drv && ca.drv->required)
1191                 die("%s: smudge filter %s failed", path, ca.drv->name);
1192
1193         return ret | ret_filter;
1194 }
1195
1196 int async_convert_to_working_tree(const char *path, const char *src,
1197                                   size_t len, struct strbuf *dst,
1198                                   void *dco)
1199 {
1200         return convert_to_working_tree_internal(path, src, len, dst, 0, dco);
1201 }
1202
1203 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
1204 {
1205         return convert_to_working_tree_internal(path, src, len, dst, 0, NULL);
1206 }
1207
1208 int renormalize_buffer(const struct index_state *istate, const char *path,
1209                        const char *src, size_t len, struct strbuf *dst)
1210 {
1211         int ret = convert_to_working_tree_internal(path, src, len, dst, 1, NULL);
1212         if (ret) {
1213                 src = dst->buf;
1214                 len = dst->len;
1215         }
1216         return ret | convert_to_git(istate, path, src, len, dst, SAFE_CRLF_RENORMALIZE);
1217 }
1218
1219 /*****************************************************************
1220  *
1221  * Streaming conversion support
1222  *
1223  *****************************************************************/
1224
1225 typedef int (*filter_fn)(struct stream_filter *,
1226                          const char *input, size_t *isize_p,
1227                          char *output, size_t *osize_p);
1228 typedef void (*free_fn)(struct stream_filter *);
1229
1230 struct stream_filter_vtbl {
1231         filter_fn filter;
1232         free_fn free;
1233 };
1234
1235 struct stream_filter {
1236         struct stream_filter_vtbl *vtbl;
1237 };
1238
1239 static int null_filter_fn(struct stream_filter *filter,
1240                           const char *input, size_t *isize_p,
1241                           char *output, size_t *osize_p)
1242 {
1243         size_t count;
1244
1245         if (!input)
1246                 return 0; /* we do not keep any states */
1247         count = *isize_p;
1248         if (*osize_p < count)
1249                 count = *osize_p;
1250         if (count) {
1251                 memmove(output, input, count);
1252                 *isize_p -= count;
1253                 *osize_p -= count;
1254         }
1255         return 0;
1256 }
1257
1258 static void null_free_fn(struct stream_filter *filter)
1259 {
1260         ; /* nothing -- null instances are shared */
1261 }
1262
1263 static struct stream_filter_vtbl null_vtbl = {
1264         null_filter_fn,
1265         null_free_fn,
1266 };
1267
1268 static struct stream_filter null_filter_singleton = {
1269         &null_vtbl,
1270 };
1271
1272 int is_null_stream_filter(struct stream_filter *filter)
1273 {
1274         return filter == &null_filter_singleton;
1275 }
1276
1277
1278 /*
1279  * LF-to-CRLF filter
1280  */
1281
1282 struct lf_to_crlf_filter {
1283         struct stream_filter filter;
1284         unsigned has_held:1;
1285         char held;
1286 };
1287
1288 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1289                                 const char *input, size_t *isize_p,
1290                                 char *output, size_t *osize_p)
1291 {
1292         size_t count, o = 0;
1293         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1294
1295         /*
1296          * We may be holding onto the CR to see if it is followed by a
1297          * LF, in which case we would need to go to the main loop.
1298          * Otherwise, just emit it to the output stream.
1299          */
1300         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1301                 output[o++] = lf_to_crlf->held;
1302                 lf_to_crlf->has_held = 0;
1303         }
1304
1305         /* We are told to drain */
1306         if (!input) {
1307                 *osize_p -= o;
1308                 return 0;
1309         }
1310
1311         count = *isize_p;
1312         if (count || lf_to_crlf->has_held) {
1313                 size_t i;
1314                 int was_cr = 0;
1315
1316                 if (lf_to_crlf->has_held) {
1317                         was_cr = 1;
1318                         lf_to_crlf->has_held = 0;
1319                 }
1320
1321                 for (i = 0; o < *osize_p && i < count; i++) {
1322                         char ch = input[i];
1323
1324                         if (ch == '\n') {
1325                                 output[o++] = '\r';
1326                         } else if (was_cr) {
1327                                 /*
1328                                  * Previous round saw CR and it is not followed
1329                                  * by a LF; emit the CR before processing the
1330                                  * current character.
1331                                  */
1332                                 output[o++] = '\r';
1333                         }
1334
1335                         /*
1336                          * We may have consumed the last output slot,
1337                          * in which case we need to break out of this
1338                          * loop; hold the current character before
1339                          * returning.
1340                          */
1341                         if (*osize_p <= o) {
1342                                 lf_to_crlf->has_held = 1;
1343                                 lf_to_crlf->held = ch;
1344                                 continue; /* break but increment i */
1345                         }
1346
1347                         if (ch == '\r') {
1348                                 was_cr = 1;
1349                                 continue;
1350                         }
1351
1352                         was_cr = 0;
1353                         output[o++] = ch;
1354                 }
1355
1356                 *osize_p -= o;
1357                 *isize_p -= i;
1358
1359                 if (!lf_to_crlf->has_held && was_cr) {
1360                         lf_to_crlf->has_held = 1;
1361                         lf_to_crlf->held = '\r';
1362                 }
1363         }
1364         return 0;
1365 }
1366
1367 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1368 {
1369         free(filter);
1370 }
1371
1372 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1373         lf_to_crlf_filter_fn,
1374         lf_to_crlf_free_fn,
1375 };
1376
1377 static struct stream_filter *lf_to_crlf_filter(void)
1378 {
1379         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1380
1381         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1382         return (struct stream_filter *)lf_to_crlf;
1383 }
1384
1385 /*
1386  * Cascade filter
1387  */
1388 #define FILTER_BUFFER 1024
1389 struct cascade_filter {
1390         struct stream_filter filter;
1391         struct stream_filter *one;
1392         struct stream_filter *two;
1393         char buf[FILTER_BUFFER];
1394         int end, ptr;
1395 };
1396
1397 static int cascade_filter_fn(struct stream_filter *filter,
1398                              const char *input, size_t *isize_p,
1399                              char *output, size_t *osize_p)
1400 {
1401         struct cascade_filter *cas = (struct cascade_filter *) filter;
1402         size_t filled = 0;
1403         size_t sz = *osize_p;
1404         size_t to_feed, remaining;
1405
1406         /*
1407          * input -- (one) --> buf -- (two) --> output
1408          */
1409         while (filled < sz) {
1410                 remaining = sz - filled;
1411
1412                 /* do we already have something to feed two with? */
1413                 if (cas->ptr < cas->end) {
1414                         to_feed = cas->end - cas->ptr;
1415                         if (stream_filter(cas->two,
1416                                           cas->buf + cas->ptr, &to_feed,
1417                                           output + filled, &remaining))
1418                                 return -1;
1419                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1420                         filled = sz - remaining;
1421                         continue;
1422                 }
1423
1424                 /* feed one from upstream and have it emit into our buffer */
1425                 to_feed = input ? *isize_p : 0;
1426                 if (input && !to_feed)
1427                         break;
1428                 remaining = sizeof(cas->buf);
1429                 if (stream_filter(cas->one,
1430                                   input, &to_feed,
1431                                   cas->buf, &remaining))
1432                         return -1;
1433                 cas->end = sizeof(cas->buf) - remaining;
1434                 cas->ptr = 0;
1435                 if (input) {
1436                         size_t fed = *isize_p - to_feed;
1437                         *isize_p -= fed;
1438                         input += fed;
1439                 }
1440
1441                 /* do we know that we drained one completely? */
1442                 if (input || cas->end)
1443                         continue;
1444
1445                 /* tell two to drain; we have nothing more to give it */
1446                 to_feed = 0;
1447                 remaining = sz - filled;
1448                 if (stream_filter(cas->two,
1449                                   NULL, &to_feed,
1450                                   output + filled, &remaining))
1451                         return -1;
1452                 if (remaining == (sz - filled))
1453                         break; /* completely drained two */
1454                 filled = sz - remaining;
1455         }
1456         *osize_p -= filled;
1457         return 0;
1458 }
1459
1460 static void cascade_free_fn(struct stream_filter *filter)
1461 {
1462         struct cascade_filter *cas = (struct cascade_filter *)filter;
1463         free_stream_filter(cas->one);
1464         free_stream_filter(cas->two);
1465         free(filter);
1466 }
1467
1468 static struct stream_filter_vtbl cascade_vtbl = {
1469         cascade_filter_fn,
1470         cascade_free_fn,
1471 };
1472
1473 static struct stream_filter *cascade_filter(struct stream_filter *one,
1474                                             struct stream_filter *two)
1475 {
1476         struct cascade_filter *cascade;
1477
1478         if (!one || is_null_stream_filter(one))
1479                 return two;
1480         if (!two || is_null_stream_filter(two))
1481                 return one;
1482
1483         cascade = xmalloc(sizeof(*cascade));
1484         cascade->one = one;
1485         cascade->two = two;
1486         cascade->end = cascade->ptr = 0;
1487         cascade->filter.vtbl = &cascade_vtbl;
1488         return (struct stream_filter *)cascade;
1489 }
1490
1491 /*
1492  * ident filter
1493  */
1494 #define IDENT_DRAINING (-1)
1495 #define IDENT_SKIPPING (-2)
1496 struct ident_filter {
1497         struct stream_filter filter;
1498         struct strbuf left;
1499         int state;
1500         char ident[45]; /* ": x40 $" */
1501 };
1502
1503 static int is_foreign_ident(const char *str)
1504 {
1505         int i;
1506
1507         if (!skip_prefix(str, "$Id: ", &str))
1508                 return 0;
1509         for (i = 0; str[i]; i++) {
1510                 if (isspace(str[i]) && str[i+1] != '$')
1511                         return 1;
1512         }
1513         return 0;
1514 }
1515
1516 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1517 {
1518         size_t to_drain = ident->left.len;
1519
1520         if (*osize_p < to_drain)
1521                 to_drain = *osize_p;
1522         if (to_drain) {
1523                 memcpy(*output_p, ident->left.buf, to_drain);
1524                 strbuf_remove(&ident->left, 0, to_drain);
1525                 *output_p += to_drain;
1526                 *osize_p -= to_drain;
1527         }
1528         if (!ident->left.len)
1529                 ident->state = 0;
1530 }
1531
1532 static int ident_filter_fn(struct stream_filter *filter,
1533                            const char *input, size_t *isize_p,
1534                            char *output, size_t *osize_p)
1535 {
1536         struct ident_filter *ident = (struct ident_filter *)filter;
1537         static const char head[] = "$Id";
1538
1539         if (!input) {
1540                 /* drain upon eof */
1541                 switch (ident->state) {
1542                 default:
1543                         strbuf_add(&ident->left, head, ident->state);
1544                 case IDENT_SKIPPING:
1545                         /* fallthru */
1546                 case IDENT_DRAINING:
1547                         ident_drain(ident, &output, osize_p);
1548                 }
1549                 return 0;
1550         }
1551
1552         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1553                 int ch;
1554
1555                 if (ident->state == IDENT_DRAINING) {
1556                         ident_drain(ident, &output, osize_p);
1557                         if (!*osize_p)
1558                                 break;
1559                         continue;
1560                 }
1561
1562                 ch = *(input++);
1563                 (*isize_p)--;
1564
1565                 if (ident->state == IDENT_SKIPPING) {
1566                         /*
1567                          * Skipping until '$' or LF, but keeping them
1568                          * in case it is a foreign ident.
1569                          */
1570                         strbuf_addch(&ident->left, ch);
1571                         if (ch != '\n' && ch != '$')
1572                                 continue;
1573                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1574                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1575                                 strbuf_addstr(&ident->left, ident->ident);
1576                         }
1577                         ident->state = IDENT_DRAINING;
1578                         continue;
1579                 }
1580
1581                 if (ident->state < sizeof(head) &&
1582                     head[ident->state] == ch) {
1583                         ident->state++;
1584                         continue;
1585                 }
1586
1587                 if (ident->state)
1588                         strbuf_add(&ident->left, head, ident->state);
1589                 if (ident->state == sizeof(head) - 1) {
1590                         if (ch != ':' && ch != '$') {
1591                                 strbuf_addch(&ident->left, ch);
1592                                 ident->state = 0;
1593                                 continue;
1594                         }
1595
1596                         if (ch == ':') {
1597                                 strbuf_addch(&ident->left, ch);
1598                                 ident->state = IDENT_SKIPPING;
1599                         } else {
1600                                 strbuf_addstr(&ident->left, ident->ident);
1601                                 ident->state = IDENT_DRAINING;
1602                         }
1603                         continue;
1604                 }
1605
1606                 strbuf_addch(&ident->left, ch);
1607                 ident->state = IDENT_DRAINING;
1608         }
1609         return 0;
1610 }
1611
1612 static void ident_free_fn(struct stream_filter *filter)
1613 {
1614         struct ident_filter *ident = (struct ident_filter *)filter;
1615         strbuf_release(&ident->left);
1616         free(filter);
1617 }
1618
1619 static struct stream_filter_vtbl ident_vtbl = {
1620         ident_filter_fn,
1621         ident_free_fn,
1622 };
1623
1624 static struct stream_filter *ident_filter(const unsigned char *sha1)
1625 {
1626         struct ident_filter *ident = xmalloc(sizeof(*ident));
1627
1628         xsnprintf(ident->ident, sizeof(ident->ident),
1629                   ": %s $", sha1_to_hex(sha1));
1630         strbuf_init(&ident->left, 0);
1631         ident->filter.vtbl = &ident_vtbl;
1632         ident->state = 0;
1633         return (struct stream_filter *)ident;
1634 }
1635
1636 /*
1637  * Return an appropriately constructed filter for the path, or NULL if
1638  * the contents cannot be filtered without reading the whole thing
1639  * in-core.
1640  *
1641  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1642  * large binary blob you would want us not to slurp into the memory!
1643  */
1644 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1645 {
1646         struct conv_attrs ca;
1647         struct stream_filter *filter = NULL;
1648
1649         convert_attrs(&ca, path);
1650         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1651                 return NULL;
1652
1653         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1654                 return NULL;
1655
1656         if (ca.ident)
1657                 filter = ident_filter(sha1);
1658
1659         if (output_eol(ca.crlf_action) == EOL_CRLF)
1660                 filter = cascade_filter(filter, lf_to_crlf_filter());
1661         else
1662                 filter = cascade_filter(filter, &null_filter_singleton);
1663
1664         return filter;
1665 }
1666
1667 void free_stream_filter(struct stream_filter *filter)
1668 {
1669         filter->vtbl->free(filter);
1670 }
1671
1672 int stream_filter(struct stream_filter *filter,
1673                   const char *input, size_t *isize_p,
1674                   char *output, size_t *osize_p)
1675 {
1676         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1677 }