Merge branch 'sb/submodule-rm-absorb' into maint
[git] / convert.c
1 #include "cache.h"
2 #include "attr.h"
3 #include "run-command.h"
4 #include "quote.h"
5 #include "sigchain.h"
6 #include "pkt-line.h"
7
8 /*
9  * convert.c - convert a file when checking it out and checking it in.
10  *
11  * This should use the pathname to decide on whether it wants to do some
12  * more interesting conversions (automatic gzip/unzip, general format
13  * conversions etc etc), but by default it just does automatic CRLF<->LF
14  * translation when the "text" attribute or "auto_crlf" option is set.
15  */
16
17 /* Stat bits: When BIN is set, the txt bits are unset */
18 #define CONVERT_STAT_BITS_TXT_LF    0x1
19 #define CONVERT_STAT_BITS_TXT_CRLF  0x2
20 #define CONVERT_STAT_BITS_BIN       0x4
21
22 enum crlf_action {
23         CRLF_UNDEFINED,
24         CRLF_BINARY,
25         CRLF_TEXT,
26         CRLF_TEXT_INPUT,
27         CRLF_TEXT_CRLF,
28         CRLF_AUTO,
29         CRLF_AUTO_INPUT,
30         CRLF_AUTO_CRLF
31 };
32
33 struct text_stat {
34         /* NUL, CR, LF and CRLF counts */
35         unsigned nul, lonecr, lonelf, crlf;
36
37         /* These are just approximations! */
38         unsigned printable, nonprintable;
39 };
40
41 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
42 {
43         unsigned long i;
44
45         memset(stats, 0, sizeof(*stats));
46
47         for (i = 0; i < size; i++) {
48                 unsigned char c = buf[i];
49                 if (c == '\r') {
50                         if (i+1 < size && buf[i+1] == '\n') {
51                                 stats->crlf++;
52                                 i++;
53                         } else
54                                 stats->lonecr++;
55                         continue;
56                 }
57                 if (c == '\n') {
58                         stats->lonelf++;
59                         continue;
60                 }
61                 if (c == 127)
62                         /* DEL */
63                         stats->nonprintable++;
64                 else if (c < 32) {
65                         switch (c) {
66                                 /* BS, HT, ESC and FF */
67                         case '\b': case '\t': case '\033': case '\014':
68                                 stats->printable++;
69                                 break;
70                         case 0:
71                                 stats->nul++;
72                                 /* fall through */
73                         default:
74                                 stats->nonprintable++;
75                         }
76                 }
77                 else
78                         stats->printable++;
79         }
80
81         /* If file ends with EOF then don't count this EOF as non-printable. */
82         if (size >= 1 && buf[size-1] == '\032')
83                 stats->nonprintable--;
84 }
85
86 /*
87  * The same heuristics as diff.c::mmfile_is_binary()
88  * We treat files with bare CR as binary
89  */
90 static int convert_is_binary(unsigned long size, const struct text_stat *stats)
91 {
92         if (stats->lonecr)
93                 return 1;
94         if (stats->nul)
95                 return 1;
96         if ((stats->printable >> 7) < stats->nonprintable)
97                 return 1;
98         return 0;
99 }
100
101 static unsigned int gather_convert_stats(const char *data, unsigned long size)
102 {
103         struct text_stat stats;
104         int ret = 0;
105         if (!data || !size)
106                 return 0;
107         gather_stats(data, size, &stats);
108         if (convert_is_binary(size, &stats))
109                 ret |= CONVERT_STAT_BITS_BIN;
110         if (stats.crlf)
111                 ret |= CONVERT_STAT_BITS_TXT_CRLF;
112         if (stats.lonelf)
113                 ret |=  CONVERT_STAT_BITS_TXT_LF;
114
115         return ret;
116 }
117
118 static const char *gather_convert_stats_ascii(const char *data, unsigned long size)
119 {
120         unsigned int convert_stats = gather_convert_stats(data, size);
121
122         if (convert_stats & CONVERT_STAT_BITS_BIN)
123                 return "-text";
124         switch (convert_stats) {
125         case CONVERT_STAT_BITS_TXT_LF:
126                 return "lf";
127         case CONVERT_STAT_BITS_TXT_CRLF:
128                 return "crlf";
129         case CONVERT_STAT_BITS_TXT_LF | CONVERT_STAT_BITS_TXT_CRLF:
130                 return "mixed";
131         default:
132                 return "none";
133         }
134 }
135
136 const char *get_cached_convert_stats_ascii(const char *path)
137 {
138         const char *ret;
139         unsigned long sz;
140         void *data = read_blob_data_from_cache(path, &sz);
141         ret = gather_convert_stats_ascii(data, sz);
142         free(data);
143         return ret;
144 }
145
146 const char *get_wt_convert_stats_ascii(const char *path)
147 {
148         const char *ret = "";
149         struct strbuf sb = STRBUF_INIT;
150         if (strbuf_read_file(&sb, path, 0) >= 0)
151                 ret = gather_convert_stats_ascii(sb.buf, sb.len);
152         strbuf_release(&sb);
153         return ret;
154 }
155
156 static int text_eol_is_crlf(void)
157 {
158         if (auto_crlf == AUTO_CRLF_TRUE)
159                 return 1;
160         else if (auto_crlf == AUTO_CRLF_INPUT)
161                 return 0;
162         if (core_eol == EOL_CRLF)
163                 return 1;
164         if (core_eol == EOL_UNSET && EOL_NATIVE == EOL_CRLF)
165                 return 1;
166         return 0;
167 }
168
169 static enum eol output_eol(enum crlf_action crlf_action)
170 {
171         switch (crlf_action) {
172         case CRLF_BINARY:
173                 return EOL_UNSET;
174         case CRLF_TEXT_CRLF:
175                 return EOL_CRLF;
176         case CRLF_TEXT_INPUT:
177                 return EOL_LF;
178         case CRLF_UNDEFINED:
179         case CRLF_AUTO_CRLF:
180                 return EOL_CRLF;
181         case CRLF_AUTO_INPUT:
182                 return EOL_LF;
183         case CRLF_TEXT:
184         case CRLF_AUTO:
185                 /* fall through */
186                 return text_eol_is_crlf() ? EOL_CRLF : EOL_LF;
187         }
188         warning("Illegal crlf_action %d\n", (int)crlf_action);
189         return core_eol;
190 }
191
192 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
193                             struct text_stat *old_stats, struct text_stat *new_stats,
194                             enum safe_crlf checksafe)
195 {
196         if (old_stats->crlf && !new_stats->crlf ) {
197                 /*
198                  * CRLFs would not be restored by checkout
199                  */
200                 if (checksafe == SAFE_CRLF_WARN)
201                         warning(_("CRLF will be replaced by LF in %s.\n"
202                                   "The file will have its original line"
203                                   " endings in your working directory."), path);
204                 else /* i.e. SAFE_CRLF_FAIL */
205                         die(_("CRLF would be replaced by LF in %s."), path);
206         } else if (old_stats->lonelf && !new_stats->lonelf ) {
207                 /*
208                  * CRLFs would be added by checkout
209                  */
210                 if (checksafe == SAFE_CRLF_WARN)
211                         warning(_("LF will be replaced by CRLF in %s.\n"
212                                   "The file will have its original line"
213                                   " endings in your working directory."), path);
214                 else /* i.e. SAFE_CRLF_FAIL */
215                         die(_("LF would be replaced by CRLF in %s"), path);
216         }
217 }
218
219 static int has_cr_in_index(const char *path)
220 {
221         unsigned long sz;
222         void *data;
223         int has_cr;
224
225         data = read_blob_data_from_cache(path, &sz);
226         if (!data)
227                 return 0;
228         has_cr = memchr(data, '\r', sz) != NULL;
229         free(data);
230         return has_cr;
231 }
232
233 static int will_convert_lf_to_crlf(size_t len, struct text_stat *stats,
234                                    enum crlf_action crlf_action)
235 {
236         if (output_eol(crlf_action) != EOL_CRLF)
237                 return 0;
238         /* No "naked" LF? Nothing to convert, regardless. */
239         if (!stats->lonelf)
240                 return 0;
241
242         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
243                 /* If we have any CR or CRLF line endings, we do not touch it */
244                 /* This is the new safer autocrlf-handling */
245                 if (stats->lonecr || stats->crlf)
246                         return 0;
247
248                 if (convert_is_binary(len, stats))
249                         return 0;
250         }
251         return 1;
252
253 }
254
255 static int crlf_to_git(const char *path, const char *src, size_t len,
256                        struct strbuf *buf,
257                        enum crlf_action crlf_action, enum safe_crlf checksafe)
258 {
259         struct text_stat stats;
260         char *dst;
261         int convert_crlf_into_lf;
262
263         if (crlf_action == CRLF_BINARY ||
264             (src && !len))
265                 return 0;
266
267         /*
268          * If we are doing a dry-run and have no source buffer, there is
269          * nothing to analyze; we must assume we would convert.
270          */
271         if (!buf && !src)
272                 return 1;
273
274         gather_stats(src, len, &stats);
275         /* Optimization: No CRLF? Nothing to convert, regardless. */
276         convert_crlf_into_lf = !!stats.crlf;
277
278         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
279                 if (convert_is_binary(len, &stats))
280                         return 0;
281                 /*
282                  * If the file in the index has any CR in it, do not
283                  * convert.  This is the new safer autocrlf handling,
284                  * unless we want to renormalize in a merge or
285                  * cherry-pick.
286                  */
287                 if ((checksafe != SAFE_CRLF_RENORMALIZE) && has_cr_in_index(path))
288                         convert_crlf_into_lf = 0;
289         }
290         if ((checksafe == SAFE_CRLF_WARN ||
291             (checksafe == SAFE_CRLF_FAIL)) && len) {
292                 struct text_stat new_stats;
293                 memcpy(&new_stats, &stats, sizeof(new_stats));
294                 /* simulate "git add" */
295                 if (convert_crlf_into_lf) {
296                         new_stats.lonelf += new_stats.crlf;
297                         new_stats.crlf = 0;
298                 }
299                 /* simulate "git checkout" */
300                 if (will_convert_lf_to_crlf(len, &new_stats, crlf_action)) {
301                         new_stats.crlf += new_stats.lonelf;
302                         new_stats.lonelf = 0;
303                 }
304                 check_safe_crlf(path, crlf_action, &stats, &new_stats, checksafe);
305         }
306         if (!convert_crlf_into_lf)
307                 return 0;
308
309         /*
310          * At this point all of our source analysis is done, and we are sure we
311          * would convert. If we are in dry-run mode, we can give an answer.
312          */
313         if (!buf)
314                 return 1;
315
316         /* only grow if not in place */
317         if (strbuf_avail(buf) + buf->len < len)
318                 strbuf_grow(buf, len - buf->len);
319         dst = buf->buf;
320         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
321                 /*
322                  * If we guessed, we already know we rejected a file with
323                  * lone CR, and we can strip a CR without looking at what
324                  * follow it.
325                  */
326                 do {
327                         unsigned char c = *src++;
328                         if (c != '\r')
329                                 *dst++ = c;
330                 } while (--len);
331         } else {
332                 do {
333                         unsigned char c = *src++;
334                         if (! (c == '\r' && (1 < len && *src == '\n')))
335                                 *dst++ = c;
336                 } while (--len);
337         }
338         strbuf_setlen(buf, dst - buf->buf);
339         return 1;
340 }
341
342 static int crlf_to_worktree(const char *path, const char *src, size_t len,
343                             struct strbuf *buf, enum crlf_action crlf_action)
344 {
345         char *to_free = NULL;
346         struct text_stat stats;
347
348         if (!len || output_eol(crlf_action) != EOL_CRLF)
349                 return 0;
350
351         gather_stats(src, len, &stats);
352         if (!will_convert_lf_to_crlf(len, &stats, crlf_action))
353                 return 0;
354
355         /* are we "faking" in place editing ? */
356         if (src == buf->buf)
357                 to_free = strbuf_detach(buf, NULL);
358
359         strbuf_grow(buf, len + stats.lonelf);
360         for (;;) {
361                 const char *nl = memchr(src, '\n', len);
362                 if (!nl)
363                         break;
364                 if (nl > src && nl[-1] == '\r') {
365                         strbuf_add(buf, src, nl + 1 - src);
366                 } else {
367                         strbuf_add(buf, src, nl - src);
368                         strbuf_addstr(buf, "\r\n");
369                 }
370                 len -= nl + 1 - src;
371                 src  = nl + 1;
372         }
373         strbuf_add(buf, src, len);
374
375         free(to_free);
376         return 1;
377 }
378
379 struct filter_params {
380         const char *src;
381         unsigned long size;
382         int fd;
383         const char *cmd;
384         const char *path;
385 };
386
387 static int filter_buffer_or_fd(int in, int out, void *data)
388 {
389         /*
390          * Spawn cmd and feed the buffer contents through its stdin.
391          */
392         struct child_process child_process = CHILD_PROCESS_INIT;
393         struct filter_params *params = (struct filter_params *)data;
394         int write_err, status;
395         const char *argv[] = { NULL, NULL };
396
397         /* apply % substitution to cmd */
398         struct strbuf cmd = STRBUF_INIT;
399         struct strbuf path = STRBUF_INIT;
400         struct strbuf_expand_dict_entry dict[] = {
401                 { "f", NULL, },
402                 { NULL, NULL, },
403         };
404
405         /* quote the path to preserve spaces, etc. */
406         sq_quote_buf(&path, params->path);
407         dict[0].value = path.buf;
408
409         /* expand all %f with the quoted path */
410         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
411         strbuf_release(&path);
412
413         argv[0] = cmd.buf;
414
415         child_process.argv = argv;
416         child_process.use_shell = 1;
417         child_process.in = -1;
418         child_process.out = out;
419
420         if (start_command(&child_process))
421                 return error("cannot fork to run external filter '%s'", params->cmd);
422
423         sigchain_push(SIGPIPE, SIG_IGN);
424
425         if (params->src) {
426                 write_err = (write_in_full(child_process.in,
427                                            params->src, params->size) < 0);
428                 if (errno == EPIPE)
429                         write_err = 0;
430         } else {
431                 write_err = copy_fd(params->fd, child_process.in);
432                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
433                         write_err = 0;
434         }
435
436         if (close(child_process.in))
437                 write_err = 1;
438         if (write_err)
439                 error("cannot feed the input to external filter '%s'", params->cmd);
440
441         sigchain_pop(SIGPIPE);
442
443         status = finish_command(&child_process);
444         if (status)
445                 error("external filter '%s' failed %d", params->cmd, status);
446
447         strbuf_release(&cmd);
448         return (write_err || status);
449 }
450
451 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
452                         struct strbuf *dst, const char *cmd)
453 {
454         /*
455          * Create a pipeline to have the command filter the buffer's
456          * contents.
457          *
458          * (child --> cmd) --> us
459          */
460         int err = 0;
461         struct strbuf nbuf = STRBUF_INIT;
462         struct async async;
463         struct filter_params params;
464
465         memset(&async, 0, sizeof(async));
466         async.proc = filter_buffer_or_fd;
467         async.data = &params;
468         async.out = -1;
469         params.src = src;
470         params.size = len;
471         params.fd = fd;
472         params.cmd = cmd;
473         params.path = path;
474
475         fflush(NULL);
476         if (start_async(&async))
477                 return 0;       /* error was already reported */
478
479         if (strbuf_read(&nbuf, async.out, len) < 0) {
480                 err = error("read from external filter '%s' failed", cmd);
481         }
482         if (close(async.out)) {
483                 err = error("read from external filter '%s' failed", cmd);
484         }
485         if (finish_async(&async)) {
486                 err = error("external filter '%s' failed", cmd);
487         }
488
489         if (!err) {
490                 strbuf_swap(dst, &nbuf);
491         }
492         strbuf_release(&nbuf);
493         return !err;
494 }
495
496 #define CAP_CLEAN    (1u<<0)
497 #define CAP_SMUDGE   (1u<<1)
498
499 struct cmd2process {
500         struct hashmap_entry ent; /* must be the first member! */
501         unsigned int supported_capabilities;
502         const char *cmd;
503         struct child_process process;
504 };
505
506 static int cmd_process_map_initialized;
507 static struct hashmap cmd_process_map;
508
509 static int cmd2process_cmp(const struct cmd2process *e1,
510                            const struct cmd2process *e2,
511                            const void *unused)
512 {
513         return strcmp(e1->cmd, e2->cmd);
514 }
515
516 static struct cmd2process *find_multi_file_filter_entry(struct hashmap *hashmap, const char *cmd)
517 {
518         struct cmd2process key;
519         hashmap_entry_init(&key, strhash(cmd));
520         key.cmd = cmd;
521         return hashmap_get(hashmap, &key, NULL);
522 }
523
524 static int packet_write_list(int fd, const char *line, ...)
525 {
526         va_list args;
527         int err;
528         va_start(args, line);
529         for (;;) {
530                 if (!line)
531                         break;
532                 if (strlen(line) > LARGE_PACKET_DATA_MAX)
533                         return -1;
534                 err = packet_write_fmt_gently(fd, "%s\n", line);
535                 if (err)
536                         return err;
537                 line = va_arg(args, const char*);
538         }
539         va_end(args);
540         return packet_flush_gently(fd);
541 }
542
543 static void read_multi_file_filter_status(int fd, struct strbuf *status)
544 {
545         struct strbuf **pair;
546         char *line;
547         for (;;) {
548                 line = packet_read_line(fd, NULL);
549                 if (!line)
550                         break;
551                 pair = strbuf_split_str(line, '=', 2);
552                 if (pair[0] && pair[0]->len && pair[1]) {
553                         /* the last "status=<foo>" line wins */
554                         if (!strcmp(pair[0]->buf, "status=")) {
555                                 strbuf_reset(status);
556                                 strbuf_addbuf(status, pair[1]);
557                         }
558                 }
559                 strbuf_list_free(pair);
560         }
561 }
562
563 static void kill_multi_file_filter(struct hashmap *hashmap, struct cmd2process *entry)
564 {
565         if (!entry)
566                 return;
567
568         entry->process.clean_on_exit = 0;
569         kill(entry->process.pid, SIGTERM);
570         finish_command(&entry->process);
571
572         hashmap_remove(hashmap, entry, NULL);
573         free(entry);
574 }
575
576 static void stop_multi_file_filter(struct child_process *process)
577 {
578         sigchain_push(SIGPIPE, SIG_IGN);
579         /* Closing the pipe signals the filter to initiate a shutdown. */
580         close(process->in);
581         close(process->out);
582         sigchain_pop(SIGPIPE);
583         /* Finish command will wait until the shutdown is complete. */
584         finish_command(process);
585 }
586
587 static struct cmd2process *start_multi_file_filter(struct hashmap *hashmap, const char *cmd)
588 {
589         int err;
590         struct cmd2process *entry;
591         struct child_process *process;
592         const char *argv[] = { cmd, NULL };
593         struct string_list cap_list = STRING_LIST_INIT_NODUP;
594         char *cap_buf;
595         const char *cap_name;
596
597         entry = xmalloc(sizeof(*entry));
598         entry->cmd = cmd;
599         entry->supported_capabilities = 0;
600         process = &entry->process;
601
602         child_process_init(process);
603         process->argv = argv;
604         process->use_shell = 1;
605         process->in = -1;
606         process->out = -1;
607         process->clean_on_exit = 1;
608         process->clean_on_exit_handler = stop_multi_file_filter;
609
610         if (start_command(process)) {
611                 error("cannot fork to run external filter '%s'", cmd);
612                 return NULL;
613         }
614
615         hashmap_entry_init(entry, strhash(cmd));
616
617         sigchain_push(SIGPIPE, SIG_IGN);
618
619         err = packet_write_list(process->in, "git-filter-client", "version=2", NULL);
620         if (err)
621                 goto done;
622
623         err = strcmp(packet_read_line(process->out, NULL), "git-filter-server");
624         if (err) {
625                 error("external filter '%s' does not support filter protocol version 2", cmd);
626                 goto done;
627         }
628         err = strcmp(packet_read_line(process->out, NULL), "version=2");
629         if (err)
630                 goto done;
631         err = packet_read_line(process->out, NULL) != NULL;
632         if (err)
633                 goto done;
634
635         err = packet_write_list(process->in, "capability=clean", "capability=smudge", NULL);
636
637         for (;;) {
638                 cap_buf = packet_read_line(process->out, NULL);
639                 if (!cap_buf)
640                         break;
641                 string_list_split_in_place(&cap_list, cap_buf, '=', 1);
642
643                 if (cap_list.nr != 2 || strcmp(cap_list.items[0].string, "capability"))
644                         continue;
645
646                 cap_name = cap_list.items[1].string;
647                 if (!strcmp(cap_name, "clean")) {
648                         entry->supported_capabilities |= CAP_CLEAN;
649                 } else if (!strcmp(cap_name, "smudge")) {
650                         entry->supported_capabilities |= CAP_SMUDGE;
651                 } else {
652                         warning(
653                                 "external filter '%s' requested unsupported filter capability '%s'",
654                                 cmd, cap_name
655                         );
656                 }
657
658                 string_list_clear(&cap_list, 0);
659         }
660
661 done:
662         sigchain_pop(SIGPIPE);
663
664         if (err || errno == EPIPE) {
665                 error("initialization for external filter '%s' failed", cmd);
666                 kill_multi_file_filter(hashmap, entry);
667                 return NULL;
668         }
669
670         hashmap_add(hashmap, entry);
671         return entry;
672 }
673
674 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
675                                    int fd, struct strbuf *dst, const char *cmd,
676                                    const unsigned int wanted_capability)
677 {
678         int err;
679         struct cmd2process *entry;
680         struct child_process *process;
681         struct strbuf nbuf = STRBUF_INIT;
682         struct strbuf filter_status = STRBUF_INIT;
683         const char *filter_type;
684
685         if (!cmd_process_map_initialized) {
686                 cmd_process_map_initialized = 1;
687                 hashmap_init(&cmd_process_map, (hashmap_cmp_fn) cmd2process_cmp, 0);
688                 entry = NULL;
689         } else {
690                 entry = find_multi_file_filter_entry(&cmd_process_map, cmd);
691         }
692
693         fflush(NULL);
694
695         if (!entry) {
696                 entry = start_multi_file_filter(&cmd_process_map, cmd);
697                 if (!entry)
698                         return 0;
699         }
700         process = &entry->process;
701
702         if (!(wanted_capability & entry->supported_capabilities))
703                 return 0;
704
705         if (CAP_CLEAN & wanted_capability)
706                 filter_type = "clean";
707         else if (CAP_SMUDGE & wanted_capability)
708                 filter_type = "smudge";
709         else
710                 die("unexpected filter type");
711
712         sigchain_push(SIGPIPE, SIG_IGN);
713
714         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
715         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
716         if (err)
717                 goto done;
718
719         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
720         if (err) {
721                 error("path name too long for external filter");
722                 goto done;
723         }
724
725         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
726         if (err)
727                 goto done;
728
729         err = packet_flush_gently(process->in);
730         if (err)
731                 goto done;
732
733         if (fd >= 0)
734                 err = write_packetized_from_fd(fd, process->in);
735         else
736                 err = write_packetized_from_buf(src, len, process->in);
737         if (err)
738                 goto done;
739
740         read_multi_file_filter_status(process->out, &filter_status);
741         err = strcmp(filter_status.buf, "success");
742         if (err)
743                 goto done;
744
745         err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
746         if (err)
747                 goto done;
748
749         read_multi_file_filter_status(process->out, &filter_status);
750         err = strcmp(filter_status.buf, "success");
751
752 done:
753         sigchain_pop(SIGPIPE);
754
755         if (err || errno == EPIPE) {
756                 if (!strcmp(filter_status.buf, "error")) {
757                         /* The filter signaled a problem with the file. */
758                 } else if (!strcmp(filter_status.buf, "abort")) {
759                         /*
760                          * The filter signaled a permanent problem. Don't try to filter
761                          * files with the same command for the lifetime of the current
762                          * Git process.
763                          */
764                          entry->supported_capabilities &= ~wanted_capability;
765                 } else {
766                         /*
767                          * Something went wrong with the protocol filter.
768                          * Force shutdown and restart if another blob requires filtering.
769                          */
770                         error("external filter '%s' failed", cmd);
771                         kill_multi_file_filter(&cmd_process_map, entry);
772                 }
773         } else {
774                 strbuf_swap(dst, &nbuf);
775         }
776         strbuf_release(&nbuf);
777         return !err;
778 }
779
780 static struct convert_driver {
781         const char *name;
782         struct convert_driver *next;
783         const char *smudge;
784         const char *clean;
785         const char *process;
786         int required;
787 } *user_convert, **user_convert_tail;
788
789 static int apply_filter(const char *path, const char *src, size_t len,
790                         int fd, struct strbuf *dst, struct convert_driver *drv,
791                         const unsigned int wanted_capability)
792 {
793         const char *cmd = NULL;
794
795         if (!drv)
796                 return 0;
797
798         if (!dst)
799                 return 1;
800
801         if ((CAP_CLEAN & wanted_capability) && !drv->process && drv->clean)
802                 cmd = drv->clean;
803         else if ((CAP_SMUDGE & wanted_capability) && !drv->process && drv->smudge)
804                 cmd = drv->smudge;
805
806         if (cmd && *cmd)
807                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
808         else if (drv->process && *drv->process)
809                 return apply_multi_file_filter(path, src, len, fd, dst, drv->process, wanted_capability);
810
811         return 0;
812 }
813
814 static int read_convert_config(const char *var, const char *value, void *cb)
815 {
816         const char *key, *name;
817         int namelen;
818         struct convert_driver *drv;
819
820         /*
821          * External conversion drivers are configured using
822          * "filter.<name>.variable".
823          */
824         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
825                 return 0;
826         for (drv = user_convert; drv; drv = drv->next)
827                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
828                         break;
829         if (!drv) {
830                 drv = xcalloc(1, sizeof(struct convert_driver));
831                 drv->name = xmemdupz(name, namelen);
832                 *user_convert_tail = drv;
833                 user_convert_tail = &(drv->next);
834         }
835
836         /*
837          * filter.<name>.smudge and filter.<name>.clean specifies
838          * the command line:
839          *
840          *      command-line
841          *
842          * The command-line will not be interpolated in any way.
843          */
844
845         if (!strcmp("smudge", key))
846                 return git_config_string(&drv->smudge, var, value);
847
848         if (!strcmp("clean", key))
849                 return git_config_string(&drv->clean, var, value);
850
851         if (!strcmp("process", key))
852                 return git_config_string(&drv->process, var, value);
853
854         if (!strcmp("required", key)) {
855                 drv->required = git_config_bool(var, value);
856                 return 0;
857         }
858
859         return 0;
860 }
861
862 static int count_ident(const char *cp, unsigned long size)
863 {
864         /*
865          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
866          */
867         int cnt = 0;
868         char ch;
869
870         while (size) {
871                 ch = *cp++;
872                 size--;
873                 if (ch != '$')
874                         continue;
875                 if (size < 3)
876                         break;
877                 if (memcmp("Id", cp, 2))
878                         continue;
879                 ch = cp[2];
880                 cp += 3;
881                 size -= 3;
882                 if (ch == '$')
883                         cnt++; /* $Id$ */
884                 if (ch != ':')
885                         continue;
886
887                 /*
888                  * "$Id: ... "; scan up to the closing dollar sign and discard.
889                  */
890                 while (size) {
891                         ch = *cp++;
892                         size--;
893                         if (ch == '$') {
894                                 cnt++;
895                                 break;
896                         }
897                         if (ch == '\n')
898                                 break;
899                 }
900         }
901         return cnt;
902 }
903
904 static int ident_to_git(const char *path, const char *src, size_t len,
905                         struct strbuf *buf, int ident)
906 {
907         char *dst, *dollar;
908
909         if (!ident || (src && !count_ident(src, len)))
910                 return 0;
911
912         if (!buf)
913                 return 1;
914
915         /* only grow if not in place */
916         if (strbuf_avail(buf) + buf->len < len)
917                 strbuf_grow(buf, len - buf->len);
918         dst = buf->buf;
919         for (;;) {
920                 dollar = memchr(src, '$', len);
921                 if (!dollar)
922                         break;
923                 memmove(dst, src, dollar + 1 - src);
924                 dst += dollar + 1 - src;
925                 len -= dollar + 1 - src;
926                 src  = dollar + 1;
927
928                 if (len > 3 && !memcmp(src, "Id:", 3)) {
929                         dollar = memchr(src + 3, '$', len - 3);
930                         if (!dollar)
931                                 break;
932                         if (memchr(src + 3, '\n', dollar - src - 3)) {
933                                 /* Line break before the next dollar. */
934                                 continue;
935                         }
936
937                         memcpy(dst, "Id$", 3);
938                         dst += 3;
939                         len -= dollar + 1 - src;
940                         src  = dollar + 1;
941                 }
942         }
943         memmove(dst, src, len);
944         strbuf_setlen(buf, dst + len - buf->buf);
945         return 1;
946 }
947
948 static int ident_to_worktree(const char *path, const char *src, size_t len,
949                              struct strbuf *buf, int ident)
950 {
951         unsigned char sha1[20];
952         char *to_free = NULL, *dollar, *spc;
953         int cnt;
954
955         if (!ident)
956                 return 0;
957
958         cnt = count_ident(src, len);
959         if (!cnt)
960                 return 0;
961
962         /* are we "faking" in place editing ? */
963         if (src == buf->buf)
964                 to_free = strbuf_detach(buf, NULL);
965         hash_sha1_file(src, len, "blob", sha1);
966
967         strbuf_grow(buf, len + cnt * 43);
968         for (;;) {
969                 /* step 1: run to the next '$' */
970                 dollar = memchr(src, '$', len);
971                 if (!dollar)
972                         break;
973                 strbuf_add(buf, src, dollar + 1 - src);
974                 len -= dollar + 1 - src;
975                 src  = dollar + 1;
976
977                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
978                 if (len < 3 || memcmp("Id", src, 2))
979                         continue;
980
981                 /* step 3: skip over Id$ or Id:xxxxx$ */
982                 if (src[2] == '$') {
983                         src += 3;
984                         len -= 3;
985                 } else if (src[2] == ':') {
986                         /*
987                          * It's possible that an expanded Id has crept its way into the
988                          * repository, we cope with that by stripping the expansion out.
989                          * This is probably not a good idea, since it will cause changes
990                          * on checkout, which won't go away by stash, but let's keep it
991                          * for git-style ids.
992                          */
993                         dollar = memchr(src + 3, '$', len - 3);
994                         if (!dollar) {
995                                 /* incomplete keyword, no more '$', so just quit the loop */
996                                 break;
997                         }
998
999                         if (memchr(src + 3, '\n', dollar - src - 3)) {
1000                                 /* Line break before the next dollar. */
1001                                 continue;
1002                         }
1003
1004                         spc = memchr(src + 4, ' ', dollar - src - 4);
1005                         if (spc && spc < dollar-1) {
1006                                 /* There are spaces in unexpected places.
1007                                  * This is probably an id from some other
1008                                  * versioning system. Keep it for now.
1009                                  */
1010                                 continue;
1011                         }
1012
1013                         len -= dollar + 1 - src;
1014                         src  = dollar + 1;
1015                 } else {
1016                         /* it wasn't a "Id$" or "Id:xxxx$" */
1017                         continue;
1018                 }
1019
1020                 /* step 4: substitute */
1021                 strbuf_addstr(buf, "Id: ");
1022                 strbuf_add(buf, sha1_to_hex(sha1), 40);
1023                 strbuf_addstr(buf, " $");
1024         }
1025         strbuf_add(buf, src, len);
1026
1027         free(to_free);
1028         return 1;
1029 }
1030
1031 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
1032 {
1033         const char *value = check->value;
1034
1035         if (ATTR_TRUE(value))
1036                 return CRLF_TEXT;
1037         else if (ATTR_FALSE(value))
1038                 return CRLF_BINARY;
1039         else if (ATTR_UNSET(value))
1040                 ;
1041         else if (!strcmp(value, "input"))
1042                 return CRLF_TEXT_INPUT;
1043         else if (!strcmp(value, "auto"))
1044                 return CRLF_AUTO;
1045         return CRLF_UNDEFINED;
1046 }
1047
1048 static enum eol git_path_check_eol(struct attr_check_item *check)
1049 {
1050         const char *value = check->value;
1051
1052         if (ATTR_UNSET(value))
1053                 ;
1054         else if (!strcmp(value, "lf"))
1055                 return EOL_LF;
1056         else if (!strcmp(value, "crlf"))
1057                 return EOL_CRLF;
1058         return EOL_UNSET;
1059 }
1060
1061 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
1062 {
1063         const char *value = check->value;
1064         struct convert_driver *drv;
1065
1066         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
1067                 return NULL;
1068         for (drv = user_convert; drv; drv = drv->next)
1069                 if (!strcmp(value, drv->name))
1070                         return drv;
1071         return NULL;
1072 }
1073
1074 static int git_path_check_ident(struct attr_check_item *check)
1075 {
1076         const char *value = check->value;
1077
1078         return !!ATTR_TRUE(value);
1079 }
1080
1081 struct conv_attrs {
1082         struct convert_driver *drv;
1083         enum crlf_action attr_action; /* What attr says */
1084         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
1085         int ident;
1086 };
1087
1088 static void convert_attrs(struct conv_attrs *ca, const char *path)
1089 {
1090         static struct attr_check *check;
1091
1092         if (!check) {
1093                 check = attr_check_initl("crlf", "ident", "filter",
1094                                          "eol", "text", NULL);
1095                 user_convert_tail = &user_convert;
1096                 git_config(read_convert_config, NULL);
1097         }
1098
1099         if (!git_check_attr(path, check)) {
1100                 struct attr_check_item *ccheck = check->items;
1101                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
1102                 if (ca->crlf_action == CRLF_UNDEFINED)
1103                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
1104                 ca->attr_action = ca->crlf_action;
1105                 ca->ident = git_path_check_ident(ccheck + 1);
1106                 ca->drv = git_path_check_convert(ccheck + 2);
1107                 if (ca->crlf_action != CRLF_BINARY) {
1108                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
1109                         if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1110                                 ca->crlf_action = CRLF_AUTO_INPUT;
1111                         else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1112                                 ca->crlf_action = CRLF_AUTO_CRLF;
1113                         else if (eol_attr == EOL_LF)
1114                                 ca->crlf_action = CRLF_TEXT_INPUT;
1115                         else if (eol_attr == EOL_CRLF)
1116                                 ca->crlf_action = CRLF_TEXT_CRLF;
1117                 }
1118                 ca->attr_action = ca->crlf_action;
1119         } else {
1120                 ca->drv = NULL;
1121                 ca->crlf_action = CRLF_UNDEFINED;
1122                 ca->ident = 0;
1123         }
1124         if (ca->crlf_action == CRLF_TEXT)
1125                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1126         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1127                 ca->crlf_action = CRLF_BINARY;
1128         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1129                 ca->crlf_action = CRLF_AUTO_CRLF;
1130         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1131                 ca->crlf_action = CRLF_AUTO_INPUT;
1132 }
1133
1134 int would_convert_to_git_filter_fd(const char *path)
1135 {
1136         struct conv_attrs ca;
1137
1138         convert_attrs(&ca, path);
1139         if (!ca.drv)
1140                 return 0;
1141
1142         /*
1143          * Apply a filter to an fd only if the filter is required to succeed.
1144          * We must die if the filter fails, because the original data before
1145          * filtering is not available.
1146          */
1147         if (!ca.drv->required)
1148                 return 0;
1149
1150         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN);
1151 }
1152
1153 const char *get_convert_attr_ascii(const char *path)
1154 {
1155         struct conv_attrs ca;
1156
1157         convert_attrs(&ca, path);
1158         switch (ca.attr_action) {
1159         case CRLF_UNDEFINED:
1160                 return "";
1161         case CRLF_BINARY:
1162                 return "-text";
1163         case CRLF_TEXT:
1164                 return "text";
1165         case CRLF_TEXT_INPUT:
1166                 return "text eol=lf";
1167         case CRLF_TEXT_CRLF:
1168                 return "text eol=crlf";
1169         case CRLF_AUTO:
1170                 return "text=auto";
1171         case CRLF_AUTO_CRLF:
1172                 return "text=auto eol=crlf";
1173         case CRLF_AUTO_INPUT:
1174                 return "text=auto eol=lf";
1175         }
1176         return "";
1177 }
1178
1179 int convert_to_git(const char *path, const char *src, size_t len,
1180                    struct strbuf *dst, enum safe_crlf checksafe)
1181 {
1182         int ret = 0;
1183         struct conv_attrs ca;
1184
1185         convert_attrs(&ca, path);
1186
1187         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN);
1188         if (!ret && ca.drv && ca.drv->required)
1189                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1190
1191         if (ret && dst) {
1192                 src = dst->buf;
1193                 len = dst->len;
1194         }
1195         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
1196         if (ret && dst) {
1197                 src = dst->buf;
1198                 len = dst->len;
1199         }
1200         return ret | ident_to_git(path, src, len, dst, ca.ident);
1201 }
1202
1203 void convert_to_git_filter_fd(const char *path, int fd, struct strbuf *dst,
1204                               enum safe_crlf checksafe)
1205 {
1206         struct conv_attrs ca;
1207         convert_attrs(&ca, path);
1208
1209         assert(ca.drv);
1210         assert(ca.drv->clean || ca.drv->process);
1211
1212         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN))
1213                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1214
1215         crlf_to_git(path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
1216         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
1217 }
1218
1219 static int convert_to_working_tree_internal(const char *path, const char *src,
1220                                             size_t len, struct strbuf *dst,
1221                                             int normalizing)
1222 {
1223         int ret = 0, ret_filter = 0;
1224         struct conv_attrs ca;
1225
1226         convert_attrs(&ca, path);
1227
1228         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
1229         if (ret) {
1230                 src = dst->buf;
1231                 len = dst->len;
1232         }
1233         /*
1234          * CRLF conversion can be skipped if normalizing, unless there
1235          * is a smudge or process filter (even if the process filter doesn't
1236          * support smudge).  The filters might expect CRLFs.
1237          */
1238         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1239                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
1240                 if (ret) {
1241                         src = dst->buf;
1242                         len = dst->len;
1243                 }
1244         }
1245
1246         ret_filter = apply_filter(path, src, len, -1, dst, ca.drv, CAP_SMUDGE);
1247         if (!ret_filter && ca.drv && ca.drv->required)
1248                 die("%s: smudge filter %s failed", path, ca.drv->name);
1249
1250         return ret | ret_filter;
1251 }
1252
1253 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
1254 {
1255         return convert_to_working_tree_internal(path, src, len, dst, 0);
1256 }
1257
1258 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
1259 {
1260         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
1261         if (ret) {
1262                 src = dst->buf;
1263                 len = dst->len;
1264         }
1265         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_RENORMALIZE);
1266 }
1267
1268 /*****************************************************************
1269  *
1270  * Streaming conversion support
1271  *
1272  *****************************************************************/
1273
1274 typedef int (*filter_fn)(struct stream_filter *,
1275                          const char *input, size_t *isize_p,
1276                          char *output, size_t *osize_p);
1277 typedef void (*free_fn)(struct stream_filter *);
1278
1279 struct stream_filter_vtbl {
1280         filter_fn filter;
1281         free_fn free;
1282 };
1283
1284 struct stream_filter {
1285         struct stream_filter_vtbl *vtbl;
1286 };
1287
1288 static int null_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;
1293
1294         if (!input)
1295                 return 0; /* we do not keep any states */
1296         count = *isize_p;
1297         if (*osize_p < count)
1298                 count = *osize_p;
1299         if (count) {
1300                 memmove(output, input, count);
1301                 *isize_p -= count;
1302                 *osize_p -= count;
1303         }
1304         return 0;
1305 }
1306
1307 static void null_free_fn(struct stream_filter *filter)
1308 {
1309         ; /* nothing -- null instances are shared */
1310 }
1311
1312 static struct stream_filter_vtbl null_vtbl = {
1313         null_filter_fn,
1314         null_free_fn,
1315 };
1316
1317 static struct stream_filter null_filter_singleton = {
1318         &null_vtbl,
1319 };
1320
1321 int is_null_stream_filter(struct stream_filter *filter)
1322 {
1323         return filter == &null_filter_singleton;
1324 }
1325
1326
1327 /*
1328  * LF-to-CRLF filter
1329  */
1330
1331 struct lf_to_crlf_filter {
1332         struct stream_filter filter;
1333         unsigned has_held:1;
1334         char held;
1335 };
1336
1337 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1338                                 const char *input, size_t *isize_p,
1339                                 char *output, size_t *osize_p)
1340 {
1341         size_t count, o = 0;
1342         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1343
1344         /*
1345          * We may be holding onto the CR to see if it is followed by a
1346          * LF, in which case we would need to go to the main loop.
1347          * Otherwise, just emit it to the output stream.
1348          */
1349         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1350                 output[o++] = lf_to_crlf->held;
1351                 lf_to_crlf->has_held = 0;
1352         }
1353
1354         /* We are told to drain */
1355         if (!input) {
1356                 *osize_p -= o;
1357                 return 0;
1358         }
1359
1360         count = *isize_p;
1361         if (count || lf_to_crlf->has_held) {
1362                 size_t i;
1363                 int was_cr = 0;
1364
1365                 if (lf_to_crlf->has_held) {
1366                         was_cr = 1;
1367                         lf_to_crlf->has_held = 0;
1368                 }
1369
1370                 for (i = 0; o < *osize_p && i < count; i++) {
1371                         char ch = input[i];
1372
1373                         if (ch == '\n') {
1374                                 output[o++] = '\r';
1375                         } else if (was_cr) {
1376                                 /*
1377                                  * Previous round saw CR and it is not followed
1378                                  * by a LF; emit the CR before processing the
1379                                  * current character.
1380                                  */
1381                                 output[o++] = '\r';
1382                         }
1383
1384                         /*
1385                          * We may have consumed the last output slot,
1386                          * in which case we need to break out of this
1387                          * loop; hold the current character before
1388                          * returning.
1389                          */
1390                         if (*osize_p <= o) {
1391                                 lf_to_crlf->has_held = 1;
1392                                 lf_to_crlf->held = ch;
1393                                 continue; /* break but increment i */
1394                         }
1395
1396                         if (ch == '\r') {
1397                                 was_cr = 1;
1398                                 continue;
1399                         }
1400
1401                         was_cr = 0;
1402                         output[o++] = ch;
1403                 }
1404
1405                 *osize_p -= o;
1406                 *isize_p -= i;
1407
1408                 if (!lf_to_crlf->has_held && was_cr) {
1409                         lf_to_crlf->has_held = 1;
1410                         lf_to_crlf->held = '\r';
1411                 }
1412         }
1413         return 0;
1414 }
1415
1416 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1417 {
1418         free(filter);
1419 }
1420
1421 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1422         lf_to_crlf_filter_fn,
1423         lf_to_crlf_free_fn,
1424 };
1425
1426 static struct stream_filter *lf_to_crlf_filter(void)
1427 {
1428         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1429
1430         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1431         return (struct stream_filter *)lf_to_crlf;
1432 }
1433
1434 /*
1435  * Cascade filter
1436  */
1437 #define FILTER_BUFFER 1024
1438 struct cascade_filter {
1439         struct stream_filter filter;
1440         struct stream_filter *one;
1441         struct stream_filter *two;
1442         char buf[FILTER_BUFFER];
1443         int end, ptr;
1444 };
1445
1446 static int cascade_filter_fn(struct stream_filter *filter,
1447                              const char *input, size_t *isize_p,
1448                              char *output, size_t *osize_p)
1449 {
1450         struct cascade_filter *cas = (struct cascade_filter *) filter;
1451         size_t filled = 0;
1452         size_t sz = *osize_p;
1453         size_t to_feed, remaining;
1454
1455         /*
1456          * input -- (one) --> buf -- (two) --> output
1457          */
1458         while (filled < sz) {
1459                 remaining = sz - filled;
1460
1461                 /* do we already have something to feed two with? */
1462                 if (cas->ptr < cas->end) {
1463                         to_feed = cas->end - cas->ptr;
1464                         if (stream_filter(cas->two,
1465                                           cas->buf + cas->ptr, &to_feed,
1466                                           output + filled, &remaining))
1467                                 return -1;
1468                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1469                         filled = sz - remaining;
1470                         continue;
1471                 }
1472
1473                 /* feed one from upstream and have it emit into our buffer */
1474                 to_feed = input ? *isize_p : 0;
1475                 if (input && !to_feed)
1476                         break;
1477                 remaining = sizeof(cas->buf);
1478                 if (stream_filter(cas->one,
1479                                   input, &to_feed,
1480                                   cas->buf, &remaining))
1481                         return -1;
1482                 cas->end = sizeof(cas->buf) - remaining;
1483                 cas->ptr = 0;
1484                 if (input) {
1485                         size_t fed = *isize_p - to_feed;
1486                         *isize_p -= fed;
1487                         input += fed;
1488                 }
1489
1490                 /* do we know that we drained one completely? */
1491                 if (input || cas->end)
1492                         continue;
1493
1494                 /* tell two to drain; we have nothing more to give it */
1495                 to_feed = 0;
1496                 remaining = sz - filled;
1497                 if (stream_filter(cas->two,
1498                                   NULL, &to_feed,
1499                                   output + filled, &remaining))
1500                         return -1;
1501                 if (remaining == (sz - filled))
1502                         break; /* completely drained two */
1503                 filled = sz - remaining;
1504         }
1505         *osize_p -= filled;
1506         return 0;
1507 }
1508
1509 static void cascade_free_fn(struct stream_filter *filter)
1510 {
1511         struct cascade_filter *cas = (struct cascade_filter *)filter;
1512         free_stream_filter(cas->one);
1513         free_stream_filter(cas->two);
1514         free(filter);
1515 }
1516
1517 static struct stream_filter_vtbl cascade_vtbl = {
1518         cascade_filter_fn,
1519         cascade_free_fn,
1520 };
1521
1522 static struct stream_filter *cascade_filter(struct stream_filter *one,
1523                                             struct stream_filter *two)
1524 {
1525         struct cascade_filter *cascade;
1526
1527         if (!one || is_null_stream_filter(one))
1528                 return two;
1529         if (!two || is_null_stream_filter(two))
1530                 return one;
1531
1532         cascade = xmalloc(sizeof(*cascade));
1533         cascade->one = one;
1534         cascade->two = two;
1535         cascade->end = cascade->ptr = 0;
1536         cascade->filter.vtbl = &cascade_vtbl;
1537         return (struct stream_filter *)cascade;
1538 }
1539
1540 /*
1541  * ident filter
1542  */
1543 #define IDENT_DRAINING (-1)
1544 #define IDENT_SKIPPING (-2)
1545 struct ident_filter {
1546         struct stream_filter filter;
1547         struct strbuf left;
1548         int state;
1549         char ident[45]; /* ": x40 $" */
1550 };
1551
1552 static int is_foreign_ident(const char *str)
1553 {
1554         int i;
1555
1556         if (!skip_prefix(str, "$Id: ", &str))
1557                 return 0;
1558         for (i = 0; str[i]; i++) {
1559                 if (isspace(str[i]) && str[i+1] != '$')
1560                         return 1;
1561         }
1562         return 0;
1563 }
1564
1565 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1566 {
1567         size_t to_drain = ident->left.len;
1568
1569         if (*osize_p < to_drain)
1570                 to_drain = *osize_p;
1571         if (to_drain) {
1572                 memcpy(*output_p, ident->left.buf, to_drain);
1573                 strbuf_remove(&ident->left, 0, to_drain);
1574                 *output_p += to_drain;
1575                 *osize_p -= to_drain;
1576         }
1577         if (!ident->left.len)
1578                 ident->state = 0;
1579 }
1580
1581 static int ident_filter_fn(struct stream_filter *filter,
1582                            const char *input, size_t *isize_p,
1583                            char *output, size_t *osize_p)
1584 {
1585         struct ident_filter *ident = (struct ident_filter *)filter;
1586         static const char head[] = "$Id";
1587
1588         if (!input) {
1589                 /* drain upon eof */
1590                 switch (ident->state) {
1591                 default:
1592                         strbuf_add(&ident->left, head, ident->state);
1593                 case IDENT_SKIPPING:
1594                         /* fallthru */
1595                 case IDENT_DRAINING:
1596                         ident_drain(ident, &output, osize_p);
1597                 }
1598                 return 0;
1599         }
1600
1601         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1602                 int ch;
1603
1604                 if (ident->state == IDENT_DRAINING) {
1605                         ident_drain(ident, &output, osize_p);
1606                         if (!*osize_p)
1607                                 break;
1608                         continue;
1609                 }
1610
1611                 ch = *(input++);
1612                 (*isize_p)--;
1613
1614                 if (ident->state == IDENT_SKIPPING) {
1615                         /*
1616                          * Skipping until '$' or LF, but keeping them
1617                          * in case it is a foreign ident.
1618                          */
1619                         strbuf_addch(&ident->left, ch);
1620                         if (ch != '\n' && ch != '$')
1621                                 continue;
1622                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1623                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1624                                 strbuf_addstr(&ident->left, ident->ident);
1625                         }
1626                         ident->state = IDENT_DRAINING;
1627                         continue;
1628                 }
1629
1630                 if (ident->state < sizeof(head) &&
1631                     head[ident->state] == ch) {
1632                         ident->state++;
1633                         continue;
1634                 }
1635
1636                 if (ident->state)
1637                         strbuf_add(&ident->left, head, ident->state);
1638                 if (ident->state == sizeof(head) - 1) {
1639                         if (ch != ':' && ch != '$') {
1640                                 strbuf_addch(&ident->left, ch);
1641                                 ident->state = 0;
1642                                 continue;
1643                         }
1644
1645                         if (ch == ':') {
1646                                 strbuf_addch(&ident->left, ch);
1647                                 ident->state = IDENT_SKIPPING;
1648                         } else {
1649                                 strbuf_addstr(&ident->left, ident->ident);
1650                                 ident->state = IDENT_DRAINING;
1651                         }
1652                         continue;
1653                 }
1654
1655                 strbuf_addch(&ident->left, ch);
1656                 ident->state = IDENT_DRAINING;
1657         }
1658         return 0;
1659 }
1660
1661 static void ident_free_fn(struct stream_filter *filter)
1662 {
1663         struct ident_filter *ident = (struct ident_filter *)filter;
1664         strbuf_release(&ident->left);
1665         free(filter);
1666 }
1667
1668 static struct stream_filter_vtbl ident_vtbl = {
1669         ident_filter_fn,
1670         ident_free_fn,
1671 };
1672
1673 static struct stream_filter *ident_filter(const unsigned char *sha1)
1674 {
1675         struct ident_filter *ident = xmalloc(sizeof(*ident));
1676
1677         xsnprintf(ident->ident, sizeof(ident->ident),
1678                   ": %s $", sha1_to_hex(sha1));
1679         strbuf_init(&ident->left, 0);
1680         ident->filter.vtbl = &ident_vtbl;
1681         ident->state = 0;
1682         return (struct stream_filter *)ident;
1683 }
1684
1685 /*
1686  * Return an appropriately constructed filter for the path, or NULL if
1687  * the contents cannot be filtered without reading the whole thing
1688  * in-core.
1689  *
1690  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1691  * large binary blob you would want us not to slurp into the memory!
1692  */
1693 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1694 {
1695         struct conv_attrs ca;
1696         struct stream_filter *filter = NULL;
1697
1698         convert_attrs(&ca, path);
1699         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1700                 return NULL;
1701
1702         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1703                 return NULL;
1704
1705         if (ca.ident)
1706                 filter = ident_filter(sha1);
1707
1708         if (output_eol(ca.crlf_action) == EOL_CRLF)
1709                 filter = cascade_filter(filter, lf_to_crlf_filter());
1710         else
1711                 filter = cascade_filter(filter, &null_filter_singleton);
1712
1713         return filter;
1714 }
1715
1716 void free_stream_filter(struct stream_filter *filter)
1717 {
1718         filter->vtbl->free(filter);
1719 }
1720
1721 int stream_filter(struct stream_filter *filter,
1722                   const char *input, size_t *isize_p,
1723                   char *output, size_t *osize_p)
1724 {
1725         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1726 }