Git 2.15.2
[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                 strbuf_release(&cmd);
428                 return error("cannot fork to run external filter '%s'", params->cmd);
429         }
430
431         sigchain_push(SIGPIPE, SIG_IGN);
432
433         if (params->src) {
434                 write_err = (write_in_full(child_process.in,
435                                            params->src, params->size) < 0);
436                 if (errno == EPIPE)
437                         write_err = 0;
438         } else {
439                 write_err = copy_fd(params->fd, child_process.in);
440                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
441                         write_err = 0;
442         }
443
444         if (close(child_process.in))
445                 write_err = 1;
446         if (write_err)
447                 error("cannot feed the input to external filter '%s'", params->cmd);
448
449         sigchain_pop(SIGPIPE);
450
451         status = finish_command(&child_process);
452         if (status)
453                 error("external filter '%s' failed %d", params->cmd, status);
454
455         strbuf_release(&cmd);
456         return (write_err || status);
457 }
458
459 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
460                         struct strbuf *dst, const char *cmd)
461 {
462         /*
463          * Create a pipeline to have the command filter the buffer's
464          * contents.
465          *
466          * (child --> cmd) --> us
467          */
468         int err = 0;
469         struct strbuf nbuf = STRBUF_INIT;
470         struct async async;
471         struct filter_params params;
472
473         memset(&async, 0, sizeof(async));
474         async.proc = filter_buffer_or_fd;
475         async.data = &params;
476         async.out = -1;
477         params.src = src;
478         params.size = len;
479         params.fd = fd;
480         params.cmd = cmd;
481         params.path = path;
482
483         fflush(NULL);
484         if (start_async(&async))
485                 return 0;       /* error was already reported */
486
487         if (strbuf_read(&nbuf, async.out, len) < 0) {
488                 err = error("read from external filter '%s' failed", cmd);
489         }
490         if (close(async.out)) {
491                 err = error("read from external filter '%s' failed", cmd);
492         }
493         if (finish_async(&async)) {
494                 err = error("external filter '%s' failed", cmd);
495         }
496
497         if (!err) {
498                 strbuf_swap(dst, &nbuf);
499         }
500         strbuf_release(&nbuf);
501         return !err;
502 }
503
504 #define CAP_CLEAN    (1u<<0)
505 #define CAP_SMUDGE   (1u<<1)
506 #define CAP_DELAY    (1u<<2)
507
508 struct cmd2process {
509         struct subprocess_entry subprocess; /* must be the first member! */
510         unsigned int supported_capabilities;
511 };
512
513 static int subprocess_map_initialized;
514 static struct hashmap subprocess_map;
515
516 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
517 {
518         static int versions[] = {2, 0};
519         static struct subprocess_capability capabilities[] = {
520                 { "clean",  CAP_CLEAN  },
521                 { "smudge", CAP_SMUDGE },
522                 { "delay",  CAP_DELAY  },
523                 { NULL, 0 }
524         };
525         struct cmd2process *entry = (struct cmd2process *)subprocess;
526         return subprocess_handshake(subprocess, "git-filter", versions, NULL,
527                                     capabilities,
528                                     &entry->supported_capabilities);
529 }
530
531 static void handle_filter_error(const struct strbuf *filter_status,
532                                 struct cmd2process *entry,
533                                 const unsigned int wanted_capability) {
534         if (!strcmp(filter_status->buf, "error"))
535                 ; /* The filter signaled a problem with the file. */
536         else if (!strcmp(filter_status->buf, "abort") && wanted_capability) {
537                 /*
538                  * The filter signaled a permanent problem. Don't try to filter
539                  * files with the same command for the lifetime of the current
540                  * Git process.
541                  */
542                  entry->supported_capabilities &= ~wanted_capability;
543         } else {
544                 /*
545                  * Something went wrong with the protocol filter.
546                  * Force shutdown and restart if another blob requires filtering.
547                  */
548                 error("external filter '%s' failed", entry->subprocess.cmd);
549                 subprocess_stop(&subprocess_map, &entry->subprocess);
550                 free(entry);
551         }
552 }
553
554 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
555                                    int fd, struct strbuf *dst, const char *cmd,
556                                    const unsigned int wanted_capability,
557                                    struct delayed_checkout *dco)
558 {
559         int err;
560         int can_delay = 0;
561         struct cmd2process *entry;
562         struct child_process *process;
563         struct strbuf nbuf = STRBUF_INIT;
564         struct strbuf filter_status = STRBUF_INIT;
565         const char *filter_type;
566
567         if (!subprocess_map_initialized) {
568                 subprocess_map_initialized = 1;
569                 hashmap_init(&subprocess_map, cmd2process_cmp, NULL, 0);
570                 entry = NULL;
571         } else {
572                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
573         }
574
575         fflush(NULL);
576
577         if (!entry) {
578                 entry = xmalloc(sizeof(*entry));
579                 entry->supported_capabilities = 0;
580
581                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
582                         free(entry);
583                         return 0;
584                 }
585         }
586         process = &entry->subprocess.process;
587
588         if (!(entry->supported_capabilities & wanted_capability))
589                 return 0;
590
591         if (wanted_capability & CAP_CLEAN)
592                 filter_type = "clean";
593         else if (wanted_capability & CAP_SMUDGE)
594                 filter_type = "smudge";
595         else
596                 die("unexpected filter type");
597
598         sigchain_push(SIGPIPE, SIG_IGN);
599
600         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
601         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
602         if (err)
603                 goto done;
604
605         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
606         if (err) {
607                 error("path name too long for external filter");
608                 goto done;
609         }
610
611         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
612         if (err)
613                 goto done;
614
615         if ((entry->supported_capabilities & CAP_DELAY) &&
616             dco && dco->state == CE_CAN_DELAY) {
617                 can_delay = 1;
618                 err = packet_write_fmt_gently(process->in, "can-delay=1\n");
619                 if (err)
620                         goto done;
621         }
622
623         err = packet_flush_gently(process->in);
624         if (err)
625                 goto done;
626
627         if (fd >= 0)
628                 err = write_packetized_from_fd(fd, process->in);
629         else
630                 err = write_packetized_from_buf(src, len, process->in);
631         if (err)
632                 goto done;
633
634         err = subprocess_read_status(process->out, &filter_status);
635         if (err)
636                 goto done;
637
638         if (can_delay && !strcmp(filter_status.buf, "delayed")) {
639                 string_list_insert(&dco->filters, cmd);
640                 string_list_insert(&dco->paths, path);
641         } else {
642                 /* The filter got the blob and wants to send us a response. */
643                 err = strcmp(filter_status.buf, "success");
644                 if (err)
645                         goto done;
646
647                 err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
648                 if (err)
649                         goto done;
650
651                 err = subprocess_read_status(process->out, &filter_status);
652                 if (err)
653                         goto done;
654
655                 err = strcmp(filter_status.buf, "success");
656         }
657
658 done:
659         sigchain_pop(SIGPIPE);
660
661         if (err)
662                 handle_filter_error(&filter_status, entry, wanted_capability);
663         else
664                 strbuf_swap(dst, &nbuf);
665         strbuf_release(&nbuf);
666         return !err;
667 }
668
669
670 int async_query_available_blobs(const char *cmd, struct string_list *available_paths)
671 {
672         int err;
673         char *line;
674         struct cmd2process *entry;
675         struct child_process *process;
676         struct strbuf filter_status = STRBUF_INIT;
677
678         assert(subprocess_map_initialized);
679         entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
680         if (!entry) {
681                 error("external filter '%s' is not available anymore although "
682                       "not all paths have been filtered", cmd);
683                 return 0;
684         }
685         process = &entry->subprocess.process;
686         sigchain_push(SIGPIPE, SIG_IGN);
687
688         err = packet_write_fmt_gently(
689                 process->in, "command=list_available_blobs\n");
690         if (err)
691                 goto done;
692
693         err = packet_flush_gently(process->in);
694         if (err)
695                 goto done;
696
697         while ((line = packet_read_line(process->out, NULL))) {
698                 const char *path;
699                 if (skip_prefix(line, "pathname=", &path))
700                         string_list_insert(available_paths, xstrdup(path));
701                 else
702                         ; /* ignore unknown keys */
703         }
704
705         err = subprocess_read_status(process->out, &filter_status);
706         if (err)
707                 goto done;
708
709         err = strcmp(filter_status.buf, "success");
710
711 done:
712         sigchain_pop(SIGPIPE);
713
714         if (err)
715                 handle_filter_error(&filter_status, entry, 0);
716         return !err;
717 }
718
719 static struct convert_driver {
720         const char *name;
721         struct convert_driver *next;
722         const char *smudge;
723         const char *clean;
724         const char *process;
725         int required;
726 } *user_convert, **user_convert_tail;
727
728 static int apply_filter(const char *path, const char *src, size_t len,
729                         int fd, struct strbuf *dst, struct convert_driver *drv,
730                         const unsigned int wanted_capability,
731                         struct delayed_checkout *dco)
732 {
733         const char *cmd = NULL;
734
735         if (!drv)
736                 return 0;
737
738         if (!dst)
739                 return 1;
740
741         if ((wanted_capability & CAP_CLEAN) && !drv->process && drv->clean)
742                 cmd = drv->clean;
743         else if ((wanted_capability & CAP_SMUDGE) && !drv->process && drv->smudge)
744                 cmd = drv->smudge;
745
746         if (cmd && *cmd)
747                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
748         else if (drv->process && *drv->process)
749                 return apply_multi_file_filter(path, src, len, fd, dst,
750                         drv->process, wanted_capability, dco);
751
752         return 0;
753 }
754
755 static int read_convert_config(const char *var, const char *value, void *cb)
756 {
757         const char *key, *name;
758         int namelen;
759         struct convert_driver *drv;
760
761         /*
762          * External conversion drivers are configured using
763          * "filter.<name>.variable".
764          */
765         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
766                 return 0;
767         for (drv = user_convert; drv; drv = drv->next)
768                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
769                         break;
770         if (!drv) {
771                 drv = xcalloc(1, sizeof(struct convert_driver));
772                 drv->name = xmemdupz(name, namelen);
773                 *user_convert_tail = drv;
774                 user_convert_tail = &(drv->next);
775         }
776
777         /*
778          * filter.<name>.smudge and filter.<name>.clean specifies
779          * the command line:
780          *
781          *      command-line
782          *
783          * The command-line will not be interpolated in any way.
784          */
785
786         if (!strcmp("smudge", key))
787                 return git_config_string(&drv->smudge, var, value);
788
789         if (!strcmp("clean", key))
790                 return git_config_string(&drv->clean, var, value);
791
792         if (!strcmp("process", key))
793                 return git_config_string(&drv->process, var, value);
794
795         if (!strcmp("required", key)) {
796                 drv->required = git_config_bool(var, value);
797                 return 0;
798         }
799
800         return 0;
801 }
802
803 static int count_ident(const char *cp, unsigned long size)
804 {
805         /*
806          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
807          */
808         int cnt = 0;
809         char ch;
810
811         while (size) {
812                 ch = *cp++;
813                 size--;
814                 if (ch != '$')
815                         continue;
816                 if (size < 3)
817                         break;
818                 if (memcmp("Id", cp, 2))
819                         continue;
820                 ch = cp[2];
821                 cp += 3;
822                 size -= 3;
823                 if (ch == '$')
824                         cnt++; /* $Id$ */
825                 if (ch != ':')
826                         continue;
827
828                 /*
829                  * "$Id: ... "; scan up to the closing dollar sign and discard.
830                  */
831                 while (size) {
832                         ch = *cp++;
833                         size--;
834                         if (ch == '$') {
835                                 cnt++;
836                                 break;
837                         }
838                         if (ch == '\n')
839                                 break;
840                 }
841         }
842         return cnt;
843 }
844
845 static int ident_to_git(const char *path, const char *src, size_t len,
846                         struct strbuf *buf, int ident)
847 {
848         char *dst, *dollar;
849
850         if (!ident || (src && !count_ident(src, len)))
851                 return 0;
852
853         if (!buf)
854                 return 1;
855
856         /* only grow if not in place */
857         if (strbuf_avail(buf) + buf->len < len)
858                 strbuf_grow(buf, len - buf->len);
859         dst = buf->buf;
860         for (;;) {
861                 dollar = memchr(src, '$', len);
862                 if (!dollar)
863                         break;
864                 memmove(dst, src, dollar + 1 - src);
865                 dst += dollar + 1 - src;
866                 len -= dollar + 1 - src;
867                 src  = dollar + 1;
868
869                 if (len > 3 && !memcmp(src, "Id:", 3)) {
870                         dollar = memchr(src + 3, '$', len - 3);
871                         if (!dollar)
872                                 break;
873                         if (memchr(src + 3, '\n', dollar - src - 3)) {
874                                 /* Line break before the next dollar. */
875                                 continue;
876                         }
877
878                         memcpy(dst, "Id$", 3);
879                         dst += 3;
880                         len -= dollar + 1 - src;
881                         src  = dollar + 1;
882                 }
883         }
884         memmove(dst, src, len);
885         strbuf_setlen(buf, dst + len - buf->buf);
886         return 1;
887 }
888
889 static int ident_to_worktree(const char *path, const char *src, size_t len,
890                              struct strbuf *buf, int ident)
891 {
892         unsigned char sha1[20];
893         char *to_free = NULL, *dollar, *spc;
894         int cnt;
895
896         if (!ident)
897                 return 0;
898
899         cnt = count_ident(src, len);
900         if (!cnt)
901                 return 0;
902
903         /* are we "faking" in place editing ? */
904         if (src == buf->buf)
905                 to_free = strbuf_detach(buf, NULL);
906         hash_sha1_file(src, len, "blob", sha1);
907
908         strbuf_grow(buf, len + cnt * 43);
909         for (;;) {
910                 /* step 1: run to the next '$' */
911                 dollar = memchr(src, '$', len);
912                 if (!dollar)
913                         break;
914                 strbuf_add(buf, src, dollar + 1 - src);
915                 len -= dollar + 1 - src;
916                 src  = dollar + 1;
917
918                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
919                 if (len < 3 || memcmp("Id", src, 2))
920                         continue;
921
922                 /* step 3: skip over Id$ or Id:xxxxx$ */
923                 if (src[2] == '$') {
924                         src += 3;
925                         len -= 3;
926                 } else if (src[2] == ':') {
927                         /*
928                          * It's possible that an expanded Id has crept its way into the
929                          * repository, we cope with that by stripping the expansion out.
930                          * This is probably not a good idea, since it will cause changes
931                          * on checkout, which won't go away by stash, but let's keep it
932                          * for git-style ids.
933                          */
934                         dollar = memchr(src + 3, '$', len - 3);
935                         if (!dollar) {
936                                 /* incomplete keyword, no more '$', so just quit the loop */
937                                 break;
938                         }
939
940                         if (memchr(src + 3, '\n', dollar - src - 3)) {
941                                 /* Line break before the next dollar. */
942                                 continue;
943                         }
944
945                         spc = memchr(src + 4, ' ', dollar - src - 4);
946                         if (spc && spc < dollar-1) {
947                                 /* There are spaces in unexpected places.
948                                  * This is probably an id from some other
949                                  * versioning system. Keep it for now.
950                                  */
951                                 continue;
952                         }
953
954                         len -= dollar + 1 - src;
955                         src  = dollar + 1;
956                 } else {
957                         /* it wasn't a "Id$" or "Id:xxxx$" */
958                         continue;
959                 }
960
961                 /* step 4: substitute */
962                 strbuf_addstr(buf, "Id: ");
963                 strbuf_add(buf, sha1_to_hex(sha1), 40);
964                 strbuf_addstr(buf, " $");
965         }
966         strbuf_add(buf, src, len);
967
968         free(to_free);
969         return 1;
970 }
971
972 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
973 {
974         const char *value = check->value;
975
976         if (ATTR_TRUE(value))
977                 return CRLF_TEXT;
978         else if (ATTR_FALSE(value))
979                 return CRLF_BINARY;
980         else if (ATTR_UNSET(value))
981                 ;
982         else if (!strcmp(value, "input"))
983                 return CRLF_TEXT_INPUT;
984         else if (!strcmp(value, "auto"))
985                 return CRLF_AUTO;
986         return CRLF_UNDEFINED;
987 }
988
989 static enum eol git_path_check_eol(struct attr_check_item *check)
990 {
991         const char *value = check->value;
992
993         if (ATTR_UNSET(value))
994                 ;
995         else if (!strcmp(value, "lf"))
996                 return EOL_LF;
997         else if (!strcmp(value, "crlf"))
998                 return EOL_CRLF;
999         return EOL_UNSET;
1000 }
1001
1002 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
1003 {
1004         const char *value = check->value;
1005         struct convert_driver *drv;
1006
1007         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
1008                 return NULL;
1009         for (drv = user_convert; drv; drv = drv->next)
1010                 if (!strcmp(value, drv->name))
1011                         return drv;
1012         return NULL;
1013 }
1014
1015 static int git_path_check_ident(struct attr_check_item *check)
1016 {
1017         const char *value = check->value;
1018
1019         return !!ATTR_TRUE(value);
1020 }
1021
1022 struct conv_attrs {
1023         struct convert_driver *drv;
1024         enum crlf_action attr_action; /* What attr says */
1025         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
1026         int ident;
1027 };
1028
1029 static void convert_attrs(struct conv_attrs *ca, const char *path)
1030 {
1031         static struct attr_check *check;
1032
1033         if (!check) {
1034                 check = attr_check_initl("crlf", "ident", "filter",
1035                                          "eol", "text", NULL);
1036                 user_convert_tail = &user_convert;
1037                 git_config(read_convert_config, NULL);
1038         }
1039
1040         if (!git_check_attr(path, check)) {
1041                 struct attr_check_item *ccheck = check->items;
1042                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
1043                 if (ca->crlf_action == CRLF_UNDEFINED)
1044                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
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         } else {
1059                 ca->drv = NULL;
1060                 ca->crlf_action = CRLF_UNDEFINED;
1061                 ca->ident = 0;
1062         }
1063
1064         /* Save attr and make a decision for action */
1065         ca->attr_action = ca->crlf_action;
1066         if (ca->crlf_action == CRLF_TEXT)
1067                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1068         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1069                 ca->crlf_action = CRLF_BINARY;
1070         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1071                 ca->crlf_action = CRLF_AUTO_CRLF;
1072         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1073                 ca->crlf_action = CRLF_AUTO_INPUT;
1074 }
1075
1076 int would_convert_to_git_filter_fd(const char *path)
1077 {
1078         struct conv_attrs ca;
1079
1080         convert_attrs(&ca, path);
1081         if (!ca.drv)
1082                 return 0;
1083
1084         /*
1085          * Apply a filter to an fd only if the filter is required to succeed.
1086          * We must die if the filter fails, because the original data before
1087          * filtering is not available.
1088          */
1089         if (!ca.drv->required)
1090                 return 0;
1091
1092         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN, NULL);
1093 }
1094
1095 const char *get_convert_attr_ascii(const char *path)
1096 {
1097         struct conv_attrs ca;
1098
1099         convert_attrs(&ca, path);
1100         switch (ca.attr_action) {
1101         case CRLF_UNDEFINED:
1102                 return "";
1103         case CRLF_BINARY:
1104                 return "-text";
1105         case CRLF_TEXT:
1106                 return "text";
1107         case CRLF_TEXT_INPUT:
1108                 return "text eol=lf";
1109         case CRLF_TEXT_CRLF:
1110                 return "text eol=crlf";
1111         case CRLF_AUTO:
1112                 return "text=auto";
1113         case CRLF_AUTO_CRLF:
1114                 return "text=auto eol=crlf";
1115         case CRLF_AUTO_INPUT:
1116                 return "text=auto eol=lf";
1117         }
1118         return "";
1119 }
1120
1121 int convert_to_git(const struct index_state *istate,
1122                    const char *path, const char *src, size_t len,
1123                    struct strbuf *dst, enum safe_crlf checksafe)
1124 {
1125         int ret = 0;
1126         struct conv_attrs ca;
1127
1128         convert_attrs(&ca, path);
1129
1130         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN, NULL);
1131         if (!ret && ca.drv && ca.drv->required)
1132                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1133
1134         if (ret && dst) {
1135                 src = dst->buf;
1136                 len = dst->len;
1137         }
1138         if (checksafe != SAFE_CRLF_KEEP_CRLF) {
1139                 ret |= crlf_to_git(istate, path, src, len, dst, ca.crlf_action, checksafe);
1140                 if (ret && dst) {
1141                         src = dst->buf;
1142                         len = dst->len;
1143                 }
1144         }
1145         return ret | ident_to_git(path, src, len, dst, ca.ident);
1146 }
1147
1148 void convert_to_git_filter_fd(const struct index_state *istate,
1149                               const char *path, int fd, struct strbuf *dst,
1150                               enum safe_crlf checksafe)
1151 {
1152         struct conv_attrs ca;
1153         convert_attrs(&ca, path);
1154
1155         assert(ca.drv);
1156         assert(ca.drv->clean || ca.drv->process);
1157
1158         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN, NULL))
1159                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1160
1161         crlf_to_git(istate, path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
1162         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
1163 }
1164
1165 static int convert_to_working_tree_internal(const char *path, const char *src,
1166                                             size_t len, struct strbuf *dst,
1167                                             int normalizing, struct delayed_checkout *dco)
1168 {
1169         int ret = 0, ret_filter = 0;
1170         struct conv_attrs ca;
1171
1172         convert_attrs(&ca, path);
1173
1174         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
1175         if (ret) {
1176                 src = dst->buf;
1177                 len = dst->len;
1178         }
1179         /*
1180          * CRLF conversion can be skipped if normalizing, unless there
1181          * is a smudge or process filter (even if the process filter doesn't
1182          * support smudge).  The filters might expect CRLFs.
1183          */
1184         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1185                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
1186                 if (ret) {
1187                         src = dst->buf;
1188                         len = dst->len;
1189                 }
1190         }
1191
1192         ret_filter = apply_filter(
1193                 path, src, len, -1, dst, ca.drv, CAP_SMUDGE, dco);
1194         if (!ret_filter && ca.drv && ca.drv->required)
1195                 die("%s: smudge filter %s failed", path, ca.drv->name);
1196
1197         return ret | ret_filter;
1198 }
1199
1200 int async_convert_to_working_tree(const char *path, const char *src,
1201                                   size_t len, struct strbuf *dst,
1202                                   void *dco)
1203 {
1204         return convert_to_working_tree_internal(path, src, len, dst, 0, dco);
1205 }
1206
1207 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
1208 {
1209         return convert_to_working_tree_internal(path, src, len, dst, 0, NULL);
1210 }
1211
1212 int renormalize_buffer(const struct index_state *istate, const char *path,
1213                        const char *src, size_t len, struct strbuf *dst)
1214 {
1215         int ret = convert_to_working_tree_internal(path, src, len, dst, 1, NULL);
1216         if (ret) {
1217                 src = dst->buf;
1218                 len = dst->len;
1219         }
1220         return ret | convert_to_git(istate, path, src, len, dst, SAFE_CRLF_RENORMALIZE);
1221 }
1222
1223 /*****************************************************************
1224  *
1225  * Streaming conversion support
1226  *
1227  *****************************************************************/
1228
1229 typedef int (*filter_fn)(struct stream_filter *,
1230                          const char *input, size_t *isize_p,
1231                          char *output, size_t *osize_p);
1232 typedef void (*free_fn)(struct stream_filter *);
1233
1234 struct stream_filter_vtbl {
1235         filter_fn filter;
1236         free_fn free;
1237 };
1238
1239 struct stream_filter {
1240         struct stream_filter_vtbl *vtbl;
1241 };
1242
1243 static int null_filter_fn(struct stream_filter *filter,
1244                           const char *input, size_t *isize_p,
1245                           char *output, size_t *osize_p)
1246 {
1247         size_t count;
1248
1249         if (!input)
1250                 return 0; /* we do not keep any states */
1251         count = *isize_p;
1252         if (*osize_p < count)
1253                 count = *osize_p;
1254         if (count) {
1255                 memmove(output, input, count);
1256                 *isize_p -= count;
1257                 *osize_p -= count;
1258         }
1259         return 0;
1260 }
1261
1262 static void null_free_fn(struct stream_filter *filter)
1263 {
1264         ; /* nothing -- null instances are shared */
1265 }
1266
1267 static struct stream_filter_vtbl null_vtbl = {
1268         null_filter_fn,
1269         null_free_fn,
1270 };
1271
1272 static struct stream_filter null_filter_singleton = {
1273         &null_vtbl,
1274 };
1275
1276 int is_null_stream_filter(struct stream_filter *filter)
1277 {
1278         return filter == &null_filter_singleton;
1279 }
1280
1281
1282 /*
1283  * LF-to-CRLF filter
1284  */
1285
1286 struct lf_to_crlf_filter {
1287         struct stream_filter filter;
1288         unsigned has_held:1;
1289         char held;
1290 };
1291
1292 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1293                                 const char *input, size_t *isize_p,
1294                                 char *output, size_t *osize_p)
1295 {
1296         size_t count, o = 0;
1297         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1298
1299         /*
1300          * We may be holding onto the CR to see if it is followed by a
1301          * LF, in which case we would need to go to the main loop.
1302          * Otherwise, just emit it to the output stream.
1303          */
1304         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1305                 output[o++] = lf_to_crlf->held;
1306                 lf_to_crlf->has_held = 0;
1307         }
1308
1309         /* We are told to drain */
1310         if (!input) {
1311                 *osize_p -= o;
1312                 return 0;
1313         }
1314
1315         count = *isize_p;
1316         if (count || lf_to_crlf->has_held) {
1317                 size_t i;
1318                 int was_cr = 0;
1319
1320                 if (lf_to_crlf->has_held) {
1321                         was_cr = 1;
1322                         lf_to_crlf->has_held = 0;
1323                 }
1324
1325                 for (i = 0; o < *osize_p && i < count; i++) {
1326                         char ch = input[i];
1327
1328                         if (ch == '\n') {
1329                                 output[o++] = '\r';
1330                         } else if (was_cr) {
1331                                 /*
1332                                  * Previous round saw CR and it is not followed
1333                                  * by a LF; emit the CR before processing the
1334                                  * current character.
1335                                  */
1336                                 output[o++] = '\r';
1337                         }
1338
1339                         /*
1340                          * We may have consumed the last output slot,
1341                          * in which case we need to break out of this
1342                          * loop; hold the current character before
1343                          * returning.
1344                          */
1345                         if (*osize_p <= o) {
1346                                 lf_to_crlf->has_held = 1;
1347                                 lf_to_crlf->held = ch;
1348                                 continue; /* break but increment i */
1349                         }
1350
1351                         if (ch == '\r') {
1352                                 was_cr = 1;
1353                                 continue;
1354                         }
1355
1356                         was_cr = 0;
1357                         output[o++] = ch;
1358                 }
1359
1360                 *osize_p -= o;
1361                 *isize_p -= i;
1362
1363                 if (!lf_to_crlf->has_held && was_cr) {
1364                         lf_to_crlf->has_held = 1;
1365                         lf_to_crlf->held = '\r';
1366                 }
1367         }
1368         return 0;
1369 }
1370
1371 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1372 {
1373         free(filter);
1374 }
1375
1376 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1377         lf_to_crlf_filter_fn,
1378         lf_to_crlf_free_fn,
1379 };
1380
1381 static struct stream_filter *lf_to_crlf_filter(void)
1382 {
1383         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1384
1385         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1386         return (struct stream_filter *)lf_to_crlf;
1387 }
1388
1389 /*
1390  * Cascade filter
1391  */
1392 #define FILTER_BUFFER 1024
1393 struct cascade_filter {
1394         struct stream_filter filter;
1395         struct stream_filter *one;
1396         struct stream_filter *two;
1397         char buf[FILTER_BUFFER];
1398         int end, ptr;
1399 };
1400
1401 static int cascade_filter_fn(struct stream_filter *filter,
1402                              const char *input, size_t *isize_p,
1403                              char *output, size_t *osize_p)
1404 {
1405         struct cascade_filter *cas = (struct cascade_filter *) filter;
1406         size_t filled = 0;
1407         size_t sz = *osize_p;
1408         size_t to_feed, remaining;
1409
1410         /*
1411          * input -- (one) --> buf -- (two) --> output
1412          */
1413         while (filled < sz) {
1414                 remaining = sz - filled;
1415
1416                 /* do we already have something to feed two with? */
1417                 if (cas->ptr < cas->end) {
1418                         to_feed = cas->end - cas->ptr;
1419                         if (stream_filter(cas->two,
1420                                           cas->buf + cas->ptr, &to_feed,
1421                                           output + filled, &remaining))
1422                                 return -1;
1423                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1424                         filled = sz - remaining;
1425                         continue;
1426                 }
1427
1428                 /* feed one from upstream and have it emit into our buffer */
1429                 to_feed = input ? *isize_p : 0;
1430                 if (input && !to_feed)
1431                         break;
1432                 remaining = sizeof(cas->buf);
1433                 if (stream_filter(cas->one,
1434                                   input, &to_feed,
1435                                   cas->buf, &remaining))
1436                         return -1;
1437                 cas->end = sizeof(cas->buf) - remaining;
1438                 cas->ptr = 0;
1439                 if (input) {
1440                         size_t fed = *isize_p - to_feed;
1441                         *isize_p -= fed;
1442                         input += fed;
1443                 }
1444
1445                 /* do we know that we drained one completely? */
1446                 if (input || cas->end)
1447                         continue;
1448
1449                 /* tell two to drain; we have nothing more to give it */
1450                 to_feed = 0;
1451                 remaining = sz - filled;
1452                 if (stream_filter(cas->two,
1453                                   NULL, &to_feed,
1454                                   output + filled, &remaining))
1455                         return -1;
1456                 if (remaining == (sz - filled))
1457                         break; /* completely drained two */
1458                 filled = sz - remaining;
1459         }
1460         *osize_p -= filled;
1461         return 0;
1462 }
1463
1464 static void cascade_free_fn(struct stream_filter *filter)
1465 {
1466         struct cascade_filter *cas = (struct cascade_filter *)filter;
1467         free_stream_filter(cas->one);
1468         free_stream_filter(cas->two);
1469         free(filter);
1470 }
1471
1472 static struct stream_filter_vtbl cascade_vtbl = {
1473         cascade_filter_fn,
1474         cascade_free_fn,
1475 };
1476
1477 static struct stream_filter *cascade_filter(struct stream_filter *one,
1478                                             struct stream_filter *two)
1479 {
1480         struct cascade_filter *cascade;
1481
1482         if (!one || is_null_stream_filter(one))
1483                 return two;
1484         if (!two || is_null_stream_filter(two))
1485                 return one;
1486
1487         cascade = xmalloc(sizeof(*cascade));
1488         cascade->one = one;
1489         cascade->two = two;
1490         cascade->end = cascade->ptr = 0;
1491         cascade->filter.vtbl = &cascade_vtbl;
1492         return (struct stream_filter *)cascade;
1493 }
1494
1495 /*
1496  * ident filter
1497  */
1498 #define IDENT_DRAINING (-1)
1499 #define IDENT_SKIPPING (-2)
1500 struct ident_filter {
1501         struct stream_filter filter;
1502         struct strbuf left;
1503         int state;
1504         char ident[45]; /* ": x40 $" */
1505 };
1506
1507 static int is_foreign_ident(const char *str)
1508 {
1509         int i;
1510
1511         if (!skip_prefix(str, "$Id: ", &str))
1512                 return 0;
1513         for (i = 0; str[i]; i++) {
1514                 if (isspace(str[i]) && str[i+1] != '$')
1515                         return 1;
1516         }
1517         return 0;
1518 }
1519
1520 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1521 {
1522         size_t to_drain = ident->left.len;
1523
1524         if (*osize_p < to_drain)
1525                 to_drain = *osize_p;
1526         if (to_drain) {
1527                 memcpy(*output_p, ident->left.buf, to_drain);
1528                 strbuf_remove(&ident->left, 0, to_drain);
1529                 *output_p += to_drain;
1530                 *osize_p -= to_drain;
1531         }
1532         if (!ident->left.len)
1533                 ident->state = 0;
1534 }
1535
1536 static int ident_filter_fn(struct stream_filter *filter,
1537                            const char *input, size_t *isize_p,
1538                            char *output, size_t *osize_p)
1539 {
1540         struct ident_filter *ident = (struct ident_filter *)filter;
1541         static const char head[] = "$Id";
1542
1543         if (!input) {
1544                 /* drain upon eof */
1545                 switch (ident->state) {
1546                 default:
1547                         strbuf_add(&ident->left, head, ident->state);
1548                         /* fallthrough */
1549                 case IDENT_SKIPPING:
1550                         /* fallthrough */
1551                 case IDENT_DRAINING:
1552                         ident_drain(ident, &output, osize_p);
1553                 }
1554                 return 0;
1555         }
1556
1557         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1558                 int ch;
1559
1560                 if (ident->state == IDENT_DRAINING) {
1561                         ident_drain(ident, &output, osize_p);
1562                         if (!*osize_p)
1563                                 break;
1564                         continue;
1565                 }
1566
1567                 ch = *(input++);
1568                 (*isize_p)--;
1569
1570                 if (ident->state == IDENT_SKIPPING) {
1571                         /*
1572                          * Skipping until '$' or LF, but keeping them
1573                          * in case it is a foreign ident.
1574                          */
1575                         strbuf_addch(&ident->left, ch);
1576                         if (ch != '\n' && ch != '$')
1577                                 continue;
1578                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1579                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1580                                 strbuf_addstr(&ident->left, ident->ident);
1581                         }
1582                         ident->state = IDENT_DRAINING;
1583                         continue;
1584                 }
1585
1586                 if (ident->state < sizeof(head) &&
1587                     head[ident->state] == ch) {
1588                         ident->state++;
1589                         continue;
1590                 }
1591
1592                 if (ident->state)
1593                         strbuf_add(&ident->left, head, ident->state);
1594                 if (ident->state == sizeof(head) - 1) {
1595                         if (ch != ':' && ch != '$') {
1596                                 strbuf_addch(&ident->left, ch);
1597                                 ident->state = 0;
1598                                 continue;
1599                         }
1600
1601                         if (ch == ':') {
1602                                 strbuf_addch(&ident->left, ch);
1603                                 ident->state = IDENT_SKIPPING;
1604                         } else {
1605                                 strbuf_addstr(&ident->left, ident->ident);
1606                                 ident->state = IDENT_DRAINING;
1607                         }
1608                         continue;
1609                 }
1610
1611                 strbuf_addch(&ident->left, ch);
1612                 ident->state = IDENT_DRAINING;
1613         }
1614         return 0;
1615 }
1616
1617 static void ident_free_fn(struct stream_filter *filter)
1618 {
1619         struct ident_filter *ident = (struct ident_filter *)filter;
1620         strbuf_release(&ident->left);
1621         free(filter);
1622 }
1623
1624 static struct stream_filter_vtbl ident_vtbl = {
1625         ident_filter_fn,
1626         ident_free_fn,
1627 };
1628
1629 static struct stream_filter *ident_filter(const unsigned char *sha1)
1630 {
1631         struct ident_filter *ident = xmalloc(sizeof(*ident));
1632
1633         xsnprintf(ident->ident, sizeof(ident->ident),
1634                   ": %s $", sha1_to_hex(sha1));
1635         strbuf_init(&ident->left, 0);
1636         ident->filter.vtbl = &ident_vtbl;
1637         ident->state = 0;
1638         return (struct stream_filter *)ident;
1639 }
1640
1641 /*
1642  * Return an appropriately constructed filter for the path, or NULL if
1643  * the contents cannot be filtered without reading the whole thing
1644  * in-core.
1645  *
1646  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1647  * large binary blob you would want us not to slurp into the memory!
1648  */
1649 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1650 {
1651         struct conv_attrs ca;
1652         struct stream_filter *filter = NULL;
1653
1654         convert_attrs(&ca, path);
1655         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1656                 return NULL;
1657
1658         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1659                 return NULL;
1660
1661         if (ca.ident)
1662                 filter = ident_filter(sha1);
1663
1664         if (output_eol(ca.crlf_action) == EOL_CRLF)
1665                 filter = cascade_filter(filter, lf_to_crlf_filter());
1666         else
1667                 filter = cascade_filter(filter, &null_filter_singleton);
1668
1669         return filter;
1670 }
1671
1672 void free_stream_filter(struct stream_filter *filter)
1673 {
1674         filter->vtbl->free(filter);
1675 }
1676
1677 int stream_filter(struct stream_filter *filter,
1678                   const char *input, size_t *isize_p,
1679                   char *output, size_t *osize_p)
1680 {
1681         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1682 }