Merge branch 'bc/hash-algo'
[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_global_conv_flags_eol(const char *path, enum crlf_action crlf_action,
197                             struct text_stat *old_stats, struct text_stat *new_stats,
198                             int conv_flags)
199 {
200         if (old_stats->crlf && !new_stats->crlf ) {
201                 /*
202                  * CRLFs would not be restored by checkout
203                  */
204                 if (conv_flags & CONV_EOL_RNDTRP_DIE)
205                         die(_("CRLF would be replaced by LF in %s."), path);
206                 else if (conv_flags & CONV_EOL_RNDTRP_WARN)
207                         warning(_("CRLF will be replaced by LF in %s.\n"
208                                   "The file will have its original line"
209                                   " endings in your working directory."), path);
210         } else if (old_stats->lonelf && !new_stats->lonelf ) {
211                 /*
212                  * CRLFs would be added by checkout
213                  */
214                 if (conv_flags & CONV_EOL_RNDTRP_DIE)
215                         die(_("LF would be replaced by CRLF in %s"), path);
216                 else if (conv_flags & CONV_EOL_RNDTRP_WARN)
217                         warning(_("LF will be replaced by CRLF in %s.\n"
218                                   "The file will have its original line"
219                                   " endings in your working directory."), path);
220         }
221 }
222
223 static int has_crlf_in_index(const struct index_state *istate, const char *path)
224 {
225         unsigned long sz;
226         void *data;
227         const char *crp;
228         int has_crlf = 0;
229
230         data = read_blob_data_from_index(istate, path, &sz);
231         if (!data)
232                 return 0;
233
234         crp = memchr(data, '\r', sz);
235         if (crp) {
236                 unsigned int ret_stats;
237                 ret_stats = gather_convert_stats(data, sz);
238                 if (!(ret_stats & CONVERT_STAT_BITS_BIN) &&
239                     (ret_stats & CONVERT_STAT_BITS_TXT_CRLF))
240                         has_crlf = 1;
241         }
242         free(data);
243         return has_crlf;
244 }
245
246 static int will_convert_lf_to_crlf(size_t len, struct text_stat *stats,
247                                    enum crlf_action crlf_action)
248 {
249         if (output_eol(crlf_action) != EOL_CRLF)
250                 return 0;
251         /* No "naked" LF? Nothing to convert, regardless. */
252         if (!stats->lonelf)
253                 return 0;
254
255         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
256                 /* If we have any CR or CRLF line endings, we do not touch it */
257                 /* This is the new safer autocrlf-handling */
258                 if (stats->lonecr || stats->crlf)
259                         return 0;
260
261                 if (convert_is_binary(len, stats))
262                         return 0;
263         }
264         return 1;
265
266 }
267
268 static int crlf_to_git(const struct index_state *istate,
269                        const char *path, const char *src, size_t len,
270                        struct strbuf *buf,
271                        enum crlf_action crlf_action, int conv_flags)
272 {
273         struct text_stat stats;
274         char *dst;
275         int convert_crlf_into_lf;
276
277         if (crlf_action == CRLF_BINARY ||
278             (src && !len))
279                 return 0;
280
281         /*
282          * If we are doing a dry-run and have no source buffer, there is
283          * nothing to analyze; we must assume we would convert.
284          */
285         if (!buf && !src)
286                 return 1;
287
288         gather_stats(src, len, &stats);
289         /* Optimization: No CRLF? Nothing to convert, regardless. */
290         convert_crlf_into_lf = !!stats.crlf;
291
292         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
293                 if (convert_is_binary(len, &stats))
294                         return 0;
295                 /*
296                  * If the file in the index has any CR in it, do not
297                  * convert.  This is the new safer autocrlf handling,
298                  * unless we want to renormalize in a merge or
299                  * cherry-pick.
300                  */
301                 if ((!(conv_flags & CONV_EOL_RENORMALIZE)) &&
302                     has_crlf_in_index(istate, path))
303                         convert_crlf_into_lf = 0;
304         }
305         if (((conv_flags & CONV_EOL_RNDTRP_WARN) ||
306              ((conv_flags & CONV_EOL_RNDTRP_DIE) && len))) {
307                 struct text_stat new_stats;
308                 memcpy(&new_stats, &stats, sizeof(new_stats));
309                 /* simulate "git add" */
310                 if (convert_crlf_into_lf) {
311                         new_stats.lonelf += new_stats.crlf;
312                         new_stats.crlf = 0;
313                 }
314                 /* simulate "git checkout" */
315                 if (will_convert_lf_to_crlf(len, &new_stats, crlf_action)) {
316                         new_stats.crlf += new_stats.lonelf;
317                         new_stats.lonelf = 0;
318                 }
319                 check_global_conv_flags_eol(path, crlf_action, &stats, &new_stats, conv_flags);
320         }
321         if (!convert_crlf_into_lf)
322                 return 0;
323
324         /*
325          * At this point all of our source analysis is done, and we are sure we
326          * would convert. If we are in dry-run mode, we can give an answer.
327          */
328         if (!buf)
329                 return 1;
330
331         /* only grow if not in place */
332         if (strbuf_avail(buf) + buf->len < len)
333                 strbuf_grow(buf, len - buf->len);
334         dst = buf->buf;
335         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
336                 /*
337                  * If we guessed, we already know we rejected a file with
338                  * lone CR, and we can strip a CR without looking at what
339                  * follow it.
340                  */
341                 do {
342                         unsigned char c = *src++;
343                         if (c != '\r')
344                                 *dst++ = c;
345                 } while (--len);
346         } else {
347                 do {
348                         unsigned char c = *src++;
349                         if (! (c == '\r' && (1 < len && *src == '\n')))
350                                 *dst++ = c;
351                 } while (--len);
352         }
353         strbuf_setlen(buf, dst - buf->buf);
354         return 1;
355 }
356
357 static int crlf_to_worktree(const char *path, const char *src, size_t len,
358                             struct strbuf *buf, enum crlf_action crlf_action)
359 {
360         char *to_free = NULL;
361         struct text_stat stats;
362
363         if (!len || output_eol(crlf_action) != EOL_CRLF)
364                 return 0;
365
366         gather_stats(src, len, &stats);
367         if (!will_convert_lf_to_crlf(len, &stats, crlf_action))
368                 return 0;
369
370         /* are we "faking" in place editing ? */
371         if (src == buf->buf)
372                 to_free = strbuf_detach(buf, NULL);
373
374         strbuf_grow(buf, len + stats.lonelf);
375         for (;;) {
376                 const char *nl = memchr(src, '\n', len);
377                 if (!nl)
378                         break;
379                 if (nl > src && nl[-1] == '\r') {
380                         strbuf_add(buf, src, nl + 1 - src);
381                 } else {
382                         strbuf_add(buf, src, nl - src);
383                         strbuf_addstr(buf, "\r\n");
384                 }
385                 len -= nl + 1 - src;
386                 src  = nl + 1;
387         }
388         strbuf_add(buf, src, len);
389
390         free(to_free);
391         return 1;
392 }
393
394 struct filter_params {
395         const char *src;
396         unsigned long size;
397         int fd;
398         const char *cmd;
399         const char *path;
400 };
401
402 static int filter_buffer_or_fd(int in, int out, void *data)
403 {
404         /*
405          * Spawn cmd and feed the buffer contents through its stdin.
406          */
407         struct child_process child_process = CHILD_PROCESS_INIT;
408         struct filter_params *params = (struct filter_params *)data;
409         int write_err, status;
410         const char *argv[] = { NULL, NULL };
411
412         /* apply % substitution to cmd */
413         struct strbuf cmd = STRBUF_INIT;
414         struct strbuf path = STRBUF_INIT;
415         struct strbuf_expand_dict_entry dict[] = {
416                 { "f", NULL, },
417                 { NULL, NULL, },
418         };
419
420         /* quote the path to preserve spaces, etc. */
421         sq_quote_buf(&path, params->path);
422         dict[0].value = path.buf;
423
424         /* expand all %f with the quoted path */
425         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
426         strbuf_release(&path);
427
428         argv[0] = cmd.buf;
429
430         child_process.argv = argv;
431         child_process.use_shell = 1;
432         child_process.in = -1;
433         child_process.out = out;
434
435         if (start_command(&child_process)) {
436                 strbuf_release(&cmd);
437                 return error("cannot fork to run external filter '%s'", params->cmd);
438         }
439
440         sigchain_push(SIGPIPE, SIG_IGN);
441
442         if (params->src) {
443                 write_err = (write_in_full(child_process.in,
444                                            params->src, params->size) < 0);
445                 if (errno == EPIPE)
446                         write_err = 0;
447         } else {
448                 write_err = copy_fd(params->fd, child_process.in);
449                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
450                         write_err = 0;
451         }
452
453         if (close(child_process.in))
454                 write_err = 1;
455         if (write_err)
456                 error("cannot feed the input to external filter '%s'", params->cmd);
457
458         sigchain_pop(SIGPIPE);
459
460         status = finish_command(&child_process);
461         if (status)
462                 error("external filter '%s' failed %d", params->cmd, status);
463
464         strbuf_release(&cmd);
465         return (write_err || status);
466 }
467
468 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
469                         struct strbuf *dst, const char *cmd)
470 {
471         /*
472          * Create a pipeline to have the command filter the buffer's
473          * contents.
474          *
475          * (child --> cmd) --> us
476          */
477         int err = 0;
478         struct strbuf nbuf = STRBUF_INIT;
479         struct async async;
480         struct filter_params params;
481
482         memset(&async, 0, sizeof(async));
483         async.proc = filter_buffer_or_fd;
484         async.data = &params;
485         async.out = -1;
486         params.src = src;
487         params.size = len;
488         params.fd = fd;
489         params.cmd = cmd;
490         params.path = path;
491
492         fflush(NULL);
493         if (start_async(&async))
494                 return 0;       /* error was already reported */
495
496         if (strbuf_read(&nbuf, async.out, len) < 0) {
497                 err = error("read from external filter '%s' failed", cmd);
498         }
499         if (close(async.out)) {
500                 err = error("read from external filter '%s' failed", cmd);
501         }
502         if (finish_async(&async)) {
503                 err = error("external filter '%s' failed", cmd);
504         }
505
506         if (!err) {
507                 strbuf_swap(dst, &nbuf);
508         }
509         strbuf_release(&nbuf);
510         return !err;
511 }
512
513 #define CAP_CLEAN    (1u<<0)
514 #define CAP_SMUDGE   (1u<<1)
515 #define CAP_DELAY    (1u<<2)
516
517 struct cmd2process {
518         struct subprocess_entry subprocess; /* must be the first member! */
519         unsigned int supported_capabilities;
520 };
521
522 static int subprocess_map_initialized;
523 static struct hashmap subprocess_map;
524
525 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
526 {
527         static int versions[] = {2, 0};
528         static struct subprocess_capability capabilities[] = {
529                 { "clean",  CAP_CLEAN  },
530                 { "smudge", CAP_SMUDGE },
531                 { "delay",  CAP_DELAY  },
532                 { NULL, 0 }
533         };
534         struct cmd2process *entry = (struct cmd2process *)subprocess;
535         return subprocess_handshake(subprocess, "git-filter", versions, NULL,
536                                     capabilities,
537                                     &entry->supported_capabilities);
538 }
539
540 static void handle_filter_error(const struct strbuf *filter_status,
541                                 struct cmd2process *entry,
542                                 const unsigned int wanted_capability) {
543         if (!strcmp(filter_status->buf, "error"))
544                 ; /* The filter signaled a problem with the file. */
545         else if (!strcmp(filter_status->buf, "abort") && wanted_capability) {
546                 /*
547                  * The filter signaled a permanent problem. Don't try to filter
548                  * files with the same command for the lifetime of the current
549                  * Git process.
550                  */
551                  entry->supported_capabilities &= ~wanted_capability;
552         } else {
553                 /*
554                  * Something went wrong with the protocol filter.
555                  * Force shutdown and restart if another blob requires filtering.
556                  */
557                 error("external filter '%s' failed", entry->subprocess.cmd);
558                 subprocess_stop(&subprocess_map, &entry->subprocess);
559                 free(entry);
560         }
561 }
562
563 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
564                                    int fd, struct strbuf *dst, const char *cmd,
565                                    const unsigned int wanted_capability,
566                                    struct delayed_checkout *dco)
567 {
568         int err;
569         int can_delay = 0;
570         struct cmd2process *entry;
571         struct child_process *process;
572         struct strbuf nbuf = STRBUF_INIT;
573         struct strbuf filter_status = STRBUF_INIT;
574         const char *filter_type;
575
576         if (!subprocess_map_initialized) {
577                 subprocess_map_initialized = 1;
578                 hashmap_init(&subprocess_map, cmd2process_cmp, NULL, 0);
579                 entry = NULL;
580         } else {
581                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
582         }
583
584         fflush(NULL);
585
586         if (!entry) {
587                 entry = xmalloc(sizeof(*entry));
588                 entry->supported_capabilities = 0;
589
590                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
591                         free(entry);
592                         return 0;
593                 }
594         }
595         process = &entry->subprocess.process;
596
597         if (!(entry->supported_capabilities & wanted_capability))
598                 return 0;
599
600         if (wanted_capability & CAP_CLEAN)
601                 filter_type = "clean";
602         else if (wanted_capability & CAP_SMUDGE)
603                 filter_type = "smudge";
604         else
605                 die("unexpected filter type");
606
607         sigchain_push(SIGPIPE, SIG_IGN);
608
609         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
610         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
611         if (err)
612                 goto done;
613
614         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
615         if (err) {
616                 error("path name too long for external filter");
617                 goto done;
618         }
619
620         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
621         if (err)
622                 goto done;
623
624         if ((entry->supported_capabilities & CAP_DELAY) &&
625             dco && dco->state == CE_CAN_DELAY) {
626                 can_delay = 1;
627                 err = packet_write_fmt_gently(process->in, "can-delay=1\n");
628                 if (err)
629                         goto done;
630         }
631
632         err = packet_flush_gently(process->in);
633         if (err)
634                 goto done;
635
636         if (fd >= 0)
637                 err = write_packetized_from_fd(fd, process->in);
638         else
639                 err = write_packetized_from_buf(src, len, process->in);
640         if (err)
641                 goto done;
642
643         err = subprocess_read_status(process->out, &filter_status);
644         if (err)
645                 goto done;
646
647         if (can_delay && !strcmp(filter_status.buf, "delayed")) {
648                 string_list_insert(&dco->filters, cmd);
649                 string_list_insert(&dco->paths, path);
650         } else {
651                 /* The filter got the blob and wants to send us a response. */
652                 err = strcmp(filter_status.buf, "success");
653                 if (err)
654                         goto done;
655
656                 err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
657                 if (err)
658                         goto done;
659
660                 err = subprocess_read_status(process->out, &filter_status);
661                 if (err)
662                         goto done;
663
664                 err = strcmp(filter_status.buf, "success");
665         }
666
667 done:
668         sigchain_pop(SIGPIPE);
669
670         if (err)
671                 handle_filter_error(&filter_status, entry, wanted_capability);
672         else
673                 strbuf_swap(dst, &nbuf);
674         strbuf_release(&nbuf);
675         return !err;
676 }
677
678
679 int async_query_available_blobs(const char *cmd, struct string_list *available_paths)
680 {
681         int err;
682         char *line;
683         struct cmd2process *entry;
684         struct child_process *process;
685         struct strbuf filter_status = STRBUF_INIT;
686
687         assert(subprocess_map_initialized);
688         entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
689         if (!entry) {
690                 error("external filter '%s' is not available anymore although "
691                       "not all paths have been filtered", cmd);
692                 return 0;
693         }
694         process = &entry->subprocess.process;
695         sigchain_push(SIGPIPE, SIG_IGN);
696
697         err = packet_write_fmt_gently(
698                 process->in, "command=list_available_blobs\n");
699         if (err)
700                 goto done;
701
702         err = packet_flush_gently(process->in);
703         if (err)
704                 goto done;
705
706         while ((line = packet_read_line(process->out, NULL))) {
707                 const char *path;
708                 if (skip_prefix(line, "pathname=", &path))
709                         string_list_insert(available_paths, xstrdup(path));
710                 else
711                         ; /* ignore unknown keys */
712         }
713
714         err = subprocess_read_status(process->out, &filter_status);
715         if (err)
716                 goto done;
717
718         err = strcmp(filter_status.buf, "success");
719
720 done:
721         sigchain_pop(SIGPIPE);
722
723         if (err)
724                 handle_filter_error(&filter_status, entry, 0);
725         return !err;
726 }
727
728 static struct convert_driver {
729         const char *name;
730         struct convert_driver *next;
731         const char *smudge;
732         const char *clean;
733         const char *process;
734         int required;
735 } *user_convert, **user_convert_tail;
736
737 static int apply_filter(const char *path, const char *src, size_t len,
738                         int fd, struct strbuf *dst, struct convert_driver *drv,
739                         const unsigned int wanted_capability,
740                         struct delayed_checkout *dco)
741 {
742         const char *cmd = NULL;
743
744         if (!drv)
745                 return 0;
746
747         if (!dst)
748                 return 1;
749
750         if ((wanted_capability & CAP_CLEAN) && !drv->process && drv->clean)
751                 cmd = drv->clean;
752         else if ((wanted_capability & CAP_SMUDGE) && !drv->process && drv->smudge)
753                 cmd = drv->smudge;
754
755         if (cmd && *cmd)
756                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
757         else if (drv->process && *drv->process)
758                 return apply_multi_file_filter(path, src, len, fd, dst,
759                         drv->process, wanted_capability, dco);
760
761         return 0;
762 }
763
764 static int read_convert_config(const char *var, const char *value, void *cb)
765 {
766         const char *key, *name;
767         int namelen;
768         struct convert_driver *drv;
769
770         /*
771          * External conversion drivers are configured using
772          * "filter.<name>.variable".
773          */
774         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
775                 return 0;
776         for (drv = user_convert; drv; drv = drv->next)
777                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
778                         break;
779         if (!drv) {
780                 drv = xcalloc(1, sizeof(struct convert_driver));
781                 drv->name = xmemdupz(name, namelen);
782                 *user_convert_tail = drv;
783                 user_convert_tail = &(drv->next);
784         }
785
786         /*
787          * filter.<name>.smudge and filter.<name>.clean specifies
788          * the command line:
789          *
790          *      command-line
791          *
792          * The command-line will not be interpolated in any way.
793          */
794
795         if (!strcmp("smudge", key))
796                 return git_config_string(&drv->smudge, var, value);
797
798         if (!strcmp("clean", key))
799                 return git_config_string(&drv->clean, var, value);
800
801         if (!strcmp("process", key))
802                 return git_config_string(&drv->process, var, value);
803
804         if (!strcmp("required", key)) {
805                 drv->required = git_config_bool(var, value);
806                 return 0;
807         }
808
809         return 0;
810 }
811
812 static int count_ident(const char *cp, unsigned long size)
813 {
814         /*
815          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
816          */
817         int cnt = 0;
818         char ch;
819
820         while (size) {
821                 ch = *cp++;
822                 size--;
823                 if (ch != '$')
824                         continue;
825                 if (size < 3)
826                         break;
827                 if (memcmp("Id", cp, 2))
828                         continue;
829                 ch = cp[2];
830                 cp += 3;
831                 size -= 3;
832                 if (ch == '$')
833                         cnt++; /* $Id$ */
834                 if (ch != ':')
835                         continue;
836
837                 /*
838                  * "$Id: ... "; scan up to the closing dollar sign and discard.
839                  */
840                 while (size) {
841                         ch = *cp++;
842                         size--;
843                         if (ch == '$') {
844                                 cnt++;
845                                 break;
846                         }
847                         if (ch == '\n')
848                                 break;
849                 }
850         }
851         return cnt;
852 }
853
854 static int ident_to_git(const char *path, const char *src, size_t len,
855                         struct strbuf *buf, int ident)
856 {
857         char *dst, *dollar;
858
859         if (!ident || (src && !count_ident(src, len)))
860                 return 0;
861
862         if (!buf)
863                 return 1;
864
865         /* only grow if not in place */
866         if (strbuf_avail(buf) + buf->len < len)
867                 strbuf_grow(buf, len - buf->len);
868         dst = buf->buf;
869         for (;;) {
870                 dollar = memchr(src, '$', len);
871                 if (!dollar)
872                         break;
873                 memmove(dst, src, dollar + 1 - src);
874                 dst += dollar + 1 - src;
875                 len -= dollar + 1 - src;
876                 src  = dollar + 1;
877
878                 if (len > 3 && !memcmp(src, "Id:", 3)) {
879                         dollar = memchr(src + 3, '$', len - 3);
880                         if (!dollar)
881                                 break;
882                         if (memchr(src + 3, '\n', dollar - src - 3)) {
883                                 /* Line break before the next dollar. */
884                                 continue;
885                         }
886
887                         memcpy(dst, "Id$", 3);
888                         dst += 3;
889                         len -= dollar + 1 - src;
890                         src  = dollar + 1;
891                 }
892         }
893         memmove(dst, src, len);
894         strbuf_setlen(buf, dst + len - buf->buf);
895         return 1;
896 }
897
898 static int ident_to_worktree(const char *path, const char *src, size_t len,
899                              struct strbuf *buf, int ident)
900 {
901         struct object_id oid;
902         char *to_free = NULL, *dollar, *spc;
903         int cnt;
904
905         if (!ident)
906                 return 0;
907
908         cnt = count_ident(src, len);
909         if (!cnt)
910                 return 0;
911
912         /* are we "faking" in place editing ? */
913         if (src == buf->buf)
914                 to_free = strbuf_detach(buf, NULL);
915         hash_object_file(src, len, "blob", &oid);
916
917         strbuf_grow(buf, len + cnt * 43);
918         for (;;) {
919                 /* step 1: run to the next '$' */
920                 dollar = memchr(src, '$', len);
921                 if (!dollar)
922                         break;
923                 strbuf_add(buf, src, dollar + 1 - src);
924                 len -= dollar + 1 - src;
925                 src  = dollar + 1;
926
927                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
928                 if (len < 3 || memcmp("Id", src, 2))
929                         continue;
930
931                 /* step 3: skip over Id$ or Id:xxxxx$ */
932                 if (src[2] == '$') {
933                         src += 3;
934                         len -= 3;
935                 } else if (src[2] == ':') {
936                         /*
937                          * It's possible that an expanded Id has crept its way into the
938                          * repository, we cope with that by stripping the expansion out.
939                          * This is probably not a good idea, since it will cause changes
940                          * on checkout, which won't go away by stash, but let's keep it
941                          * for git-style ids.
942                          */
943                         dollar = memchr(src + 3, '$', len - 3);
944                         if (!dollar) {
945                                 /* incomplete keyword, no more '$', so just quit the loop */
946                                 break;
947                         }
948
949                         if (memchr(src + 3, '\n', dollar - src - 3)) {
950                                 /* Line break before the next dollar. */
951                                 continue;
952                         }
953
954                         spc = memchr(src + 4, ' ', dollar - src - 4);
955                         if (spc && spc < dollar-1) {
956                                 /* There are spaces in unexpected places.
957                                  * This is probably an id from some other
958                                  * versioning system. Keep it for now.
959                                  */
960                                 continue;
961                         }
962
963                         len -= dollar + 1 - src;
964                         src  = dollar + 1;
965                 } else {
966                         /* it wasn't a "Id$" or "Id:xxxx$" */
967                         continue;
968                 }
969
970                 /* step 4: substitute */
971                 strbuf_addstr(buf, "Id: ");
972                 strbuf_addstr(buf, oid_to_hex(&oid));
973                 strbuf_addstr(buf, " $");
974         }
975         strbuf_add(buf, src, len);
976
977         free(to_free);
978         return 1;
979 }
980
981 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
982 {
983         const char *value = check->value;
984
985         if (ATTR_TRUE(value))
986                 return CRLF_TEXT;
987         else if (ATTR_FALSE(value))
988                 return CRLF_BINARY;
989         else if (ATTR_UNSET(value))
990                 ;
991         else if (!strcmp(value, "input"))
992                 return CRLF_TEXT_INPUT;
993         else if (!strcmp(value, "auto"))
994                 return CRLF_AUTO;
995         return CRLF_UNDEFINED;
996 }
997
998 static enum eol git_path_check_eol(struct attr_check_item *check)
999 {
1000         const char *value = check->value;
1001
1002         if (ATTR_UNSET(value))
1003                 ;
1004         else if (!strcmp(value, "lf"))
1005                 return EOL_LF;
1006         else if (!strcmp(value, "crlf"))
1007                 return EOL_CRLF;
1008         return EOL_UNSET;
1009 }
1010
1011 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
1012 {
1013         const char *value = check->value;
1014         struct convert_driver *drv;
1015
1016         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
1017                 return NULL;
1018         for (drv = user_convert; drv; drv = drv->next)
1019                 if (!strcmp(value, drv->name))
1020                         return drv;
1021         return NULL;
1022 }
1023
1024 static int git_path_check_ident(struct attr_check_item *check)
1025 {
1026         const char *value = check->value;
1027
1028         return !!ATTR_TRUE(value);
1029 }
1030
1031 struct conv_attrs {
1032         struct convert_driver *drv;
1033         enum crlf_action attr_action; /* What attr says */
1034         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
1035         int ident;
1036 };
1037
1038 static void convert_attrs(struct conv_attrs *ca, const char *path)
1039 {
1040         static struct attr_check *check;
1041
1042         if (!check) {
1043                 check = attr_check_initl("crlf", "ident", "filter",
1044                                          "eol", "text", NULL);
1045                 user_convert_tail = &user_convert;
1046                 git_config(read_convert_config, NULL);
1047         }
1048
1049         if (!git_check_attr(path, check)) {
1050                 struct attr_check_item *ccheck = check->items;
1051                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
1052                 if (ca->crlf_action == CRLF_UNDEFINED)
1053                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
1054                 ca->ident = git_path_check_ident(ccheck + 1);
1055                 ca->drv = git_path_check_convert(ccheck + 2);
1056                 if (ca->crlf_action != CRLF_BINARY) {
1057                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
1058                         if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1059                                 ca->crlf_action = CRLF_AUTO_INPUT;
1060                         else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1061                                 ca->crlf_action = CRLF_AUTO_CRLF;
1062                         else if (eol_attr == EOL_LF)
1063                                 ca->crlf_action = CRLF_TEXT_INPUT;
1064                         else if (eol_attr == EOL_CRLF)
1065                                 ca->crlf_action = CRLF_TEXT_CRLF;
1066                 }
1067         } else {
1068                 ca->drv = NULL;
1069                 ca->crlf_action = CRLF_UNDEFINED;
1070                 ca->ident = 0;
1071         }
1072
1073         /* Save attr and make a decision for action */
1074         ca->attr_action = ca->crlf_action;
1075         if (ca->crlf_action == CRLF_TEXT)
1076                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1077         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1078                 ca->crlf_action = CRLF_BINARY;
1079         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1080                 ca->crlf_action = CRLF_AUTO_CRLF;
1081         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1082                 ca->crlf_action = CRLF_AUTO_INPUT;
1083 }
1084
1085 int would_convert_to_git_filter_fd(const char *path)
1086 {
1087         struct conv_attrs ca;
1088
1089         convert_attrs(&ca, path);
1090         if (!ca.drv)
1091                 return 0;
1092
1093         /*
1094          * Apply a filter to an fd only if the filter is required to succeed.
1095          * We must die if the filter fails, because the original data before
1096          * filtering is not available.
1097          */
1098         if (!ca.drv->required)
1099                 return 0;
1100
1101         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN, NULL);
1102 }
1103
1104 const char *get_convert_attr_ascii(const char *path)
1105 {
1106         struct conv_attrs ca;
1107
1108         convert_attrs(&ca, path);
1109         switch (ca.attr_action) {
1110         case CRLF_UNDEFINED:
1111                 return "";
1112         case CRLF_BINARY:
1113                 return "-text";
1114         case CRLF_TEXT:
1115                 return "text";
1116         case CRLF_TEXT_INPUT:
1117                 return "text eol=lf";
1118         case CRLF_TEXT_CRLF:
1119                 return "text eol=crlf";
1120         case CRLF_AUTO:
1121                 return "text=auto";
1122         case CRLF_AUTO_CRLF:
1123                 return "text=auto eol=crlf";
1124         case CRLF_AUTO_INPUT:
1125                 return "text=auto eol=lf";
1126         }
1127         return "";
1128 }
1129
1130 int convert_to_git(const struct index_state *istate,
1131                    const char *path, const char *src, size_t len,
1132                    struct strbuf *dst, int conv_flags)
1133 {
1134         int ret = 0;
1135         struct conv_attrs ca;
1136
1137         convert_attrs(&ca, path);
1138
1139         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN, NULL);
1140         if (!ret && ca.drv && ca.drv->required)
1141                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1142
1143         if (ret && dst) {
1144                 src = dst->buf;
1145                 len = dst->len;
1146         }
1147         if (!(conv_flags & CONV_EOL_KEEP_CRLF)) {
1148                 ret |= crlf_to_git(istate, path, src, len, dst, ca.crlf_action, conv_flags);
1149                 if (ret && dst) {
1150                         src = dst->buf;
1151                         len = dst->len;
1152                 }
1153         }
1154         return ret | ident_to_git(path, src, len, dst, ca.ident);
1155 }
1156
1157 void convert_to_git_filter_fd(const struct index_state *istate,
1158                               const char *path, int fd, struct strbuf *dst,
1159                               int conv_flags)
1160 {
1161         struct conv_attrs ca;
1162         convert_attrs(&ca, path);
1163
1164         assert(ca.drv);
1165         assert(ca.drv->clean || ca.drv->process);
1166
1167         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN, NULL))
1168                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1169
1170         crlf_to_git(istate, path, dst->buf, dst->len, dst, ca.crlf_action, conv_flags);
1171         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
1172 }
1173
1174 static int convert_to_working_tree_internal(const char *path, const char *src,
1175                                             size_t len, struct strbuf *dst,
1176                                             int normalizing, struct delayed_checkout *dco)
1177 {
1178         int ret = 0, ret_filter = 0;
1179         struct conv_attrs ca;
1180
1181         convert_attrs(&ca, path);
1182
1183         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
1184         if (ret) {
1185                 src = dst->buf;
1186                 len = dst->len;
1187         }
1188         /*
1189          * CRLF conversion can be skipped if normalizing, unless there
1190          * is a smudge or process filter (even if the process filter doesn't
1191          * support smudge).  The filters might expect CRLFs.
1192          */
1193         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1194                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
1195                 if (ret) {
1196                         src = dst->buf;
1197                         len = dst->len;
1198                 }
1199         }
1200
1201         ret_filter = apply_filter(
1202                 path, src, len, -1, dst, ca.drv, CAP_SMUDGE, dco);
1203         if (!ret_filter && ca.drv && ca.drv->required)
1204                 die("%s: smudge filter %s failed", path, ca.drv->name);
1205
1206         return ret | ret_filter;
1207 }
1208
1209 int async_convert_to_working_tree(const char *path, const char *src,
1210                                   size_t len, struct strbuf *dst,
1211                                   void *dco)
1212 {
1213         return convert_to_working_tree_internal(path, src, len, dst, 0, dco);
1214 }
1215
1216 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
1217 {
1218         return convert_to_working_tree_internal(path, src, len, dst, 0, NULL);
1219 }
1220
1221 int renormalize_buffer(const struct index_state *istate, const char *path,
1222                        const char *src, size_t len, struct strbuf *dst)
1223 {
1224         int ret = convert_to_working_tree_internal(path, src, len, dst, 1, NULL);
1225         if (ret) {
1226                 src = dst->buf;
1227                 len = dst->len;
1228         }
1229         return ret | convert_to_git(istate, path, src, len, dst, CONV_EOL_RENORMALIZE);
1230 }
1231
1232 /*****************************************************************
1233  *
1234  * Streaming conversion support
1235  *
1236  *****************************************************************/
1237
1238 typedef int (*filter_fn)(struct stream_filter *,
1239                          const char *input, size_t *isize_p,
1240                          char *output, size_t *osize_p);
1241 typedef void (*free_fn)(struct stream_filter *);
1242
1243 struct stream_filter_vtbl {
1244         filter_fn filter;
1245         free_fn free;
1246 };
1247
1248 struct stream_filter {
1249         struct stream_filter_vtbl *vtbl;
1250 };
1251
1252 static int null_filter_fn(struct stream_filter *filter,
1253                           const char *input, size_t *isize_p,
1254                           char *output, size_t *osize_p)
1255 {
1256         size_t count;
1257
1258         if (!input)
1259                 return 0; /* we do not keep any states */
1260         count = *isize_p;
1261         if (*osize_p < count)
1262                 count = *osize_p;
1263         if (count) {
1264                 memmove(output, input, count);
1265                 *isize_p -= count;
1266                 *osize_p -= count;
1267         }
1268         return 0;
1269 }
1270
1271 static void null_free_fn(struct stream_filter *filter)
1272 {
1273         ; /* nothing -- null instances are shared */
1274 }
1275
1276 static struct stream_filter_vtbl null_vtbl = {
1277         null_filter_fn,
1278         null_free_fn,
1279 };
1280
1281 static struct stream_filter null_filter_singleton = {
1282         &null_vtbl,
1283 };
1284
1285 int is_null_stream_filter(struct stream_filter *filter)
1286 {
1287         return filter == &null_filter_singleton;
1288 }
1289
1290
1291 /*
1292  * LF-to-CRLF filter
1293  */
1294
1295 struct lf_to_crlf_filter {
1296         struct stream_filter filter;
1297         unsigned has_held:1;
1298         char held;
1299 };
1300
1301 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1302                                 const char *input, size_t *isize_p,
1303                                 char *output, size_t *osize_p)
1304 {
1305         size_t count, o = 0;
1306         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1307
1308         /*
1309          * We may be holding onto the CR to see if it is followed by a
1310          * LF, in which case we would need to go to the main loop.
1311          * Otherwise, just emit it to the output stream.
1312          */
1313         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1314                 output[o++] = lf_to_crlf->held;
1315                 lf_to_crlf->has_held = 0;
1316         }
1317
1318         /* We are told to drain */
1319         if (!input) {
1320                 *osize_p -= o;
1321                 return 0;
1322         }
1323
1324         count = *isize_p;
1325         if (count || lf_to_crlf->has_held) {
1326                 size_t i;
1327                 int was_cr = 0;
1328
1329                 if (lf_to_crlf->has_held) {
1330                         was_cr = 1;
1331                         lf_to_crlf->has_held = 0;
1332                 }
1333
1334                 for (i = 0; o < *osize_p && i < count; i++) {
1335                         char ch = input[i];
1336
1337                         if (ch == '\n') {
1338                                 output[o++] = '\r';
1339                         } else if (was_cr) {
1340                                 /*
1341                                  * Previous round saw CR and it is not followed
1342                                  * by a LF; emit the CR before processing the
1343                                  * current character.
1344                                  */
1345                                 output[o++] = '\r';
1346                         }
1347
1348                         /*
1349                          * We may have consumed the last output slot,
1350                          * in which case we need to break out of this
1351                          * loop; hold the current character before
1352                          * returning.
1353                          */
1354                         if (*osize_p <= o) {
1355                                 lf_to_crlf->has_held = 1;
1356                                 lf_to_crlf->held = ch;
1357                                 continue; /* break but increment i */
1358                         }
1359
1360                         if (ch == '\r') {
1361                                 was_cr = 1;
1362                                 continue;
1363                         }
1364
1365                         was_cr = 0;
1366                         output[o++] = ch;
1367                 }
1368
1369                 *osize_p -= o;
1370                 *isize_p -= i;
1371
1372                 if (!lf_to_crlf->has_held && was_cr) {
1373                         lf_to_crlf->has_held = 1;
1374                         lf_to_crlf->held = '\r';
1375                 }
1376         }
1377         return 0;
1378 }
1379
1380 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1381 {
1382         free(filter);
1383 }
1384
1385 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1386         lf_to_crlf_filter_fn,
1387         lf_to_crlf_free_fn,
1388 };
1389
1390 static struct stream_filter *lf_to_crlf_filter(void)
1391 {
1392         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1393
1394         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1395         return (struct stream_filter *)lf_to_crlf;
1396 }
1397
1398 /*
1399  * Cascade filter
1400  */
1401 #define FILTER_BUFFER 1024
1402 struct cascade_filter {
1403         struct stream_filter filter;
1404         struct stream_filter *one;
1405         struct stream_filter *two;
1406         char buf[FILTER_BUFFER];
1407         int end, ptr;
1408 };
1409
1410 static int cascade_filter_fn(struct stream_filter *filter,
1411                              const char *input, size_t *isize_p,
1412                              char *output, size_t *osize_p)
1413 {
1414         struct cascade_filter *cas = (struct cascade_filter *) filter;
1415         size_t filled = 0;
1416         size_t sz = *osize_p;
1417         size_t to_feed, remaining;
1418
1419         /*
1420          * input -- (one) --> buf -- (two) --> output
1421          */
1422         while (filled < sz) {
1423                 remaining = sz - filled;
1424
1425                 /* do we already have something to feed two with? */
1426                 if (cas->ptr < cas->end) {
1427                         to_feed = cas->end - cas->ptr;
1428                         if (stream_filter(cas->two,
1429                                           cas->buf + cas->ptr, &to_feed,
1430                                           output + filled, &remaining))
1431                                 return -1;
1432                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1433                         filled = sz - remaining;
1434                         continue;
1435                 }
1436
1437                 /* feed one from upstream and have it emit into our buffer */
1438                 to_feed = input ? *isize_p : 0;
1439                 if (input && !to_feed)
1440                         break;
1441                 remaining = sizeof(cas->buf);
1442                 if (stream_filter(cas->one,
1443                                   input, &to_feed,
1444                                   cas->buf, &remaining))
1445                         return -1;
1446                 cas->end = sizeof(cas->buf) - remaining;
1447                 cas->ptr = 0;
1448                 if (input) {
1449                         size_t fed = *isize_p - to_feed;
1450                         *isize_p -= fed;
1451                         input += fed;
1452                 }
1453
1454                 /* do we know that we drained one completely? */
1455                 if (input || cas->end)
1456                         continue;
1457
1458                 /* tell two to drain; we have nothing more to give it */
1459                 to_feed = 0;
1460                 remaining = sz - filled;
1461                 if (stream_filter(cas->two,
1462                                   NULL, &to_feed,
1463                                   output + filled, &remaining))
1464                         return -1;
1465                 if (remaining == (sz - filled))
1466                         break; /* completely drained two */
1467                 filled = sz - remaining;
1468         }
1469         *osize_p -= filled;
1470         return 0;
1471 }
1472
1473 static void cascade_free_fn(struct stream_filter *filter)
1474 {
1475         struct cascade_filter *cas = (struct cascade_filter *)filter;
1476         free_stream_filter(cas->one);
1477         free_stream_filter(cas->two);
1478         free(filter);
1479 }
1480
1481 static struct stream_filter_vtbl cascade_vtbl = {
1482         cascade_filter_fn,
1483         cascade_free_fn,
1484 };
1485
1486 static struct stream_filter *cascade_filter(struct stream_filter *one,
1487                                             struct stream_filter *two)
1488 {
1489         struct cascade_filter *cascade;
1490
1491         if (!one || is_null_stream_filter(one))
1492                 return two;
1493         if (!two || is_null_stream_filter(two))
1494                 return one;
1495
1496         cascade = xmalloc(sizeof(*cascade));
1497         cascade->one = one;
1498         cascade->two = two;
1499         cascade->end = cascade->ptr = 0;
1500         cascade->filter.vtbl = &cascade_vtbl;
1501         return (struct stream_filter *)cascade;
1502 }
1503
1504 /*
1505  * ident filter
1506  */
1507 #define IDENT_DRAINING (-1)
1508 #define IDENT_SKIPPING (-2)
1509 struct ident_filter {
1510         struct stream_filter filter;
1511         struct strbuf left;
1512         int state;
1513         char ident[45]; /* ": x40 $" */
1514 };
1515
1516 static int is_foreign_ident(const char *str)
1517 {
1518         int i;
1519
1520         if (!skip_prefix(str, "$Id: ", &str))
1521                 return 0;
1522         for (i = 0; str[i]; i++) {
1523                 if (isspace(str[i]) && str[i+1] != '$')
1524                         return 1;
1525         }
1526         return 0;
1527 }
1528
1529 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1530 {
1531         size_t to_drain = ident->left.len;
1532
1533         if (*osize_p < to_drain)
1534                 to_drain = *osize_p;
1535         if (to_drain) {
1536                 memcpy(*output_p, ident->left.buf, to_drain);
1537                 strbuf_remove(&ident->left, 0, to_drain);
1538                 *output_p += to_drain;
1539                 *osize_p -= to_drain;
1540         }
1541         if (!ident->left.len)
1542                 ident->state = 0;
1543 }
1544
1545 static int ident_filter_fn(struct stream_filter *filter,
1546                            const char *input, size_t *isize_p,
1547                            char *output, size_t *osize_p)
1548 {
1549         struct ident_filter *ident = (struct ident_filter *)filter;
1550         static const char head[] = "$Id";
1551
1552         if (!input) {
1553                 /* drain upon eof */
1554                 switch (ident->state) {
1555                 default:
1556                         strbuf_add(&ident->left, head, ident->state);
1557                         /* fallthrough */
1558                 case IDENT_SKIPPING:
1559                         /* fallthrough */
1560                 case IDENT_DRAINING:
1561                         ident_drain(ident, &output, osize_p);
1562                 }
1563                 return 0;
1564         }
1565
1566         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1567                 int ch;
1568
1569                 if (ident->state == IDENT_DRAINING) {
1570                         ident_drain(ident, &output, osize_p);
1571                         if (!*osize_p)
1572                                 break;
1573                         continue;
1574                 }
1575
1576                 ch = *(input++);
1577                 (*isize_p)--;
1578
1579                 if (ident->state == IDENT_SKIPPING) {
1580                         /*
1581                          * Skipping until '$' or LF, but keeping them
1582                          * in case it is a foreign ident.
1583                          */
1584                         strbuf_addch(&ident->left, ch);
1585                         if (ch != '\n' && ch != '$')
1586                                 continue;
1587                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1588                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1589                                 strbuf_addstr(&ident->left, ident->ident);
1590                         }
1591                         ident->state = IDENT_DRAINING;
1592                         continue;
1593                 }
1594
1595                 if (ident->state < sizeof(head) &&
1596                     head[ident->state] == ch) {
1597                         ident->state++;
1598                         continue;
1599                 }
1600
1601                 if (ident->state)
1602                         strbuf_add(&ident->left, head, ident->state);
1603                 if (ident->state == sizeof(head) - 1) {
1604                         if (ch != ':' && ch != '$') {
1605                                 strbuf_addch(&ident->left, ch);
1606                                 ident->state = 0;
1607                                 continue;
1608                         }
1609
1610                         if (ch == ':') {
1611                                 strbuf_addch(&ident->left, ch);
1612                                 ident->state = IDENT_SKIPPING;
1613                         } else {
1614                                 strbuf_addstr(&ident->left, ident->ident);
1615                                 ident->state = IDENT_DRAINING;
1616                         }
1617                         continue;
1618                 }
1619
1620                 strbuf_addch(&ident->left, ch);
1621                 ident->state = IDENT_DRAINING;
1622         }
1623         return 0;
1624 }
1625
1626 static void ident_free_fn(struct stream_filter *filter)
1627 {
1628         struct ident_filter *ident = (struct ident_filter *)filter;
1629         strbuf_release(&ident->left);
1630         free(filter);
1631 }
1632
1633 static struct stream_filter_vtbl ident_vtbl = {
1634         ident_filter_fn,
1635         ident_free_fn,
1636 };
1637
1638 static struct stream_filter *ident_filter(const unsigned char *sha1)
1639 {
1640         struct ident_filter *ident = xmalloc(sizeof(*ident));
1641
1642         xsnprintf(ident->ident, sizeof(ident->ident),
1643                   ": %s $", sha1_to_hex(sha1));
1644         strbuf_init(&ident->left, 0);
1645         ident->filter.vtbl = &ident_vtbl;
1646         ident->state = 0;
1647         return (struct stream_filter *)ident;
1648 }
1649
1650 /*
1651  * Return an appropriately constructed filter for the path, or NULL if
1652  * the contents cannot be filtered without reading the whole thing
1653  * in-core.
1654  *
1655  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1656  * large binary blob you would want us not to slurp into the memory!
1657  */
1658 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1659 {
1660         struct conv_attrs ca;
1661         struct stream_filter *filter = NULL;
1662
1663         convert_attrs(&ca, path);
1664         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1665                 return NULL;
1666
1667         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1668                 return NULL;
1669
1670         if (ca.ident)
1671                 filter = ident_filter(sha1);
1672
1673         if (output_eol(ca.crlf_action) == EOL_CRLF)
1674                 filter = cascade_filter(filter, lf_to_crlf_filter());
1675         else
1676                 filter = cascade_filter(filter, &null_filter_singleton);
1677
1678         return filter;
1679 }
1680
1681 void free_stream_filter(struct stream_filter *filter)
1682 {
1683         filter->vtbl->free(filter);
1684 }
1685
1686 int stream_filter(struct stream_filter *filter,
1687                   const char *input, size_t *isize_p,
1688                   char *output, size_t *osize_p)
1689 {
1690         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1691 }