Merge branch 'ds/push-sparse-tree-walk'
[git] / builtin / cat-file.c
1 /*
2  * GIT - The information manager from hell
3  *
4  * Copyright (C) Linus Torvalds, 2005
5  */
6 #define USE_THE_INDEX_COMPATIBILITY_MACROS
7 #include "cache.h"
8 #include "config.h"
9 #include "builtin.h"
10 #include "diff.h"
11 #include "parse-options.h"
12 #include "userdiff.h"
13 #include "streaming.h"
14 #include "tree-walk.h"
15 #include "sha1-array.h"
16 #include "packfile.h"
17 #include "object-store.h"
18
19 struct batch_options {
20         int enabled;
21         int follow_symlinks;
22         int print_contents;
23         int buffer_output;
24         int all_objects;
25         int unordered;
26         int cmdmode; /* may be 'w' or 'c' for --filters or --textconv */
27         const char *format;
28 };
29
30 static const char *force_path;
31
32 static int filter_object(const char *path, unsigned mode,
33                          const struct object_id *oid,
34                          char **buf, unsigned long *size)
35 {
36         enum object_type type;
37
38         *buf = read_object_file(oid, &type, size);
39         if (!*buf)
40                 return error(_("cannot read object %s '%s'"),
41                              oid_to_hex(oid), path);
42         if ((type == OBJ_BLOB) && S_ISREG(mode)) {
43                 struct strbuf strbuf = STRBUF_INIT;
44                 if (convert_to_working_tree(&the_index, path, *buf, *size, &strbuf)) {
45                         free(*buf);
46                         *size = strbuf.len;
47                         *buf = strbuf_detach(&strbuf, NULL);
48                 }
49         }
50
51         return 0;
52 }
53
54 static int stream_blob(const struct object_id *oid)
55 {
56         if (stream_blob_to_fd(1, oid, NULL, 0))
57                 die("unable to stream %s to stdout", oid_to_hex(oid));
58         return 0;
59 }
60
61 static int cat_one_file(int opt, const char *exp_type, const char *obj_name,
62                         int unknown_type)
63 {
64         struct object_id oid;
65         enum object_type type;
66         char *buf;
67         unsigned long size;
68         struct object_context obj_context;
69         struct object_info oi = OBJECT_INFO_INIT;
70         struct strbuf sb = STRBUF_INIT;
71         unsigned flags = OBJECT_INFO_LOOKUP_REPLACE;
72         const char *path = force_path;
73
74         if (unknown_type)
75                 flags |= OBJECT_INFO_ALLOW_UNKNOWN_TYPE;
76
77         if (get_oid_with_context(the_repository, obj_name,
78                                  GET_OID_RECORD_PATH,
79                                  &oid, &obj_context))
80                 die("Not a valid object name %s", obj_name);
81
82         if (!path)
83                 path = obj_context.path;
84         if (obj_context.mode == S_IFINVALID)
85                 obj_context.mode = 0100644;
86
87         buf = NULL;
88         switch (opt) {
89         case 't':
90                 oi.type_name = &sb;
91                 if (oid_object_info_extended(the_repository, &oid, &oi, flags) < 0)
92                         die("git cat-file: could not get object info");
93                 if (sb.len) {
94                         printf("%s\n", sb.buf);
95                         strbuf_release(&sb);
96                         return 0;
97                 }
98                 break;
99
100         case 's':
101                 oi.sizep = &size;
102                 if (oid_object_info_extended(the_repository, &oid, &oi, flags) < 0)
103                         die("git cat-file: could not get object info");
104                 printf("%"PRIuMAX"\n", (uintmax_t)size);
105                 return 0;
106
107         case 'e':
108                 return !has_object_file(&oid);
109
110         case 'w':
111                 if (!path)
112                         die("git cat-file --filters %s: <object> must be "
113                             "<sha1:path>", obj_name);
114
115                 if (filter_object(path, obj_context.mode,
116                                   &oid, &buf, &size))
117                         return -1;
118                 break;
119
120         case 'c':
121                 if (!path)
122                         die("git cat-file --textconv %s: <object> must be <sha1:path>",
123                             obj_name);
124
125                 if (textconv_object(the_repository, path, obj_context.mode,
126                                     &oid, 1, &buf, &size))
127                         break;
128                 /* else fallthrough */
129
130         case 'p':
131                 type = oid_object_info(the_repository, &oid, NULL);
132                 if (type < 0)
133                         die("Not a valid object name %s", obj_name);
134
135                 /* custom pretty-print here */
136                 if (type == OBJ_TREE) {
137                         const char *ls_args[3] = { NULL };
138                         ls_args[0] =  "ls-tree";
139                         ls_args[1] =  obj_name;
140                         return cmd_ls_tree(2, ls_args, NULL);
141                 }
142
143                 if (type == OBJ_BLOB)
144                         return stream_blob(&oid);
145                 buf = read_object_file(&oid, &type, &size);
146                 if (!buf)
147                         die("Cannot read object %s", obj_name);
148
149                 /* otherwise just spit out the data */
150                 break;
151
152         case 0:
153                 if (type_from_string(exp_type) == OBJ_BLOB) {
154                         struct object_id blob_oid;
155                         if (oid_object_info(the_repository, &oid, NULL) == OBJ_TAG) {
156                                 char *buffer = read_object_file(&oid, &type,
157                                                                 &size);
158                                 const char *target;
159                                 if (!skip_prefix(buffer, "object ", &target) ||
160                                     get_oid_hex(target, &blob_oid))
161                                         die("%s not a valid tag", oid_to_hex(&oid));
162                                 free(buffer);
163                         } else
164                                 oidcpy(&blob_oid, &oid);
165
166                         if (oid_object_info(the_repository, &blob_oid, NULL) == OBJ_BLOB)
167                                 return stream_blob(&blob_oid);
168                         /*
169                          * we attempted to dereference a tag to a blob
170                          * and failed; there may be new dereference
171                          * mechanisms this code is not aware of.
172                          * fall-back to the usual case.
173                          */
174                 }
175                 buf = read_object_with_reference(&oid, exp_type, &size, NULL);
176                 break;
177
178         default:
179                 die("git cat-file: unknown option: %s", exp_type);
180         }
181
182         if (!buf)
183                 die("git cat-file %s: bad file", obj_name);
184
185         write_or_die(1, buf, size);
186         free(buf);
187         free(obj_context.path);
188         return 0;
189 }
190
191 struct expand_data {
192         struct object_id oid;
193         enum object_type type;
194         unsigned long size;
195         off_t disk_size;
196         const char *rest;
197         struct object_id delta_base_oid;
198
199         /*
200          * If mark_query is true, we do not expand anything, but rather
201          * just mark the object_info with items we wish to query.
202          */
203         int mark_query;
204
205         /*
206          * Whether to split the input on whitespace before feeding it to
207          * get_sha1; this is decided during the mark_query phase based on
208          * whether we have a %(rest) token in our format.
209          */
210         int split_on_whitespace;
211
212         /*
213          * After a mark_query run, this object_info is set up to be
214          * passed to sha1_object_info_extended. It will point to the data
215          * elements above, so you can retrieve the response from there.
216          */
217         struct object_info info;
218
219         /*
220          * This flag will be true if the requested batch format and options
221          * don't require us to call sha1_object_info, which can then be
222          * optimized out.
223          */
224         unsigned skip_object_info : 1;
225 };
226
227 static int is_atom(const char *atom, const char *s, int slen)
228 {
229         int alen = strlen(atom);
230         return alen == slen && !memcmp(atom, s, alen);
231 }
232
233 static void expand_atom(struct strbuf *sb, const char *atom, int len,
234                         void *vdata)
235 {
236         struct expand_data *data = vdata;
237
238         if (is_atom("objectname", atom, len)) {
239                 if (!data->mark_query)
240                         strbuf_addstr(sb, oid_to_hex(&data->oid));
241         } else if (is_atom("objecttype", atom, len)) {
242                 if (data->mark_query)
243                         data->info.typep = &data->type;
244                 else
245                         strbuf_addstr(sb, type_name(data->type));
246         } else if (is_atom("objectsize", atom, len)) {
247                 if (data->mark_query)
248                         data->info.sizep = &data->size;
249                 else
250                         strbuf_addf(sb, "%"PRIuMAX , (uintmax_t)data->size);
251         } else if (is_atom("objectsize:disk", atom, len)) {
252                 if (data->mark_query)
253                         data->info.disk_sizep = &data->disk_size;
254                 else
255                         strbuf_addf(sb, "%"PRIuMAX, (uintmax_t)data->disk_size);
256         } else if (is_atom("rest", atom, len)) {
257                 if (data->mark_query)
258                         data->split_on_whitespace = 1;
259                 else if (data->rest)
260                         strbuf_addstr(sb, data->rest);
261         } else if (is_atom("deltabase", atom, len)) {
262                 if (data->mark_query)
263                         data->info.delta_base_sha1 = data->delta_base_oid.hash;
264                 else
265                         strbuf_addstr(sb,
266                                       oid_to_hex(&data->delta_base_oid));
267         } else
268                 die("unknown format element: %.*s", len, atom);
269 }
270
271 static size_t expand_format(struct strbuf *sb, const char *start, void *data)
272 {
273         const char *end;
274
275         if (*start != '(')
276                 return 0;
277         end = strchr(start + 1, ')');
278         if (!end)
279                 die("format element '%s' does not end in ')'", start);
280
281         expand_atom(sb, start + 1, end - start - 1, data);
282
283         return end - start + 1;
284 }
285
286 static void batch_write(struct batch_options *opt, const void *data, int len)
287 {
288         if (opt->buffer_output) {
289                 if (fwrite(data, 1, len, stdout) != len)
290                         die_errno("unable to write to stdout");
291         } else
292                 write_or_die(1, data, len);
293 }
294
295 static void print_object_or_die(struct batch_options *opt, struct expand_data *data)
296 {
297         const struct object_id *oid = &data->oid;
298
299         assert(data->info.typep);
300
301         if (data->type == OBJ_BLOB) {
302                 if (opt->buffer_output)
303                         fflush(stdout);
304                 if (opt->cmdmode) {
305                         char *contents;
306                         unsigned long size;
307
308                         if (!data->rest)
309                                 die("missing path for '%s'", oid_to_hex(oid));
310
311                         if (opt->cmdmode == 'w') {
312                                 if (filter_object(data->rest, 0100644, oid,
313                                                   &contents, &size))
314                                         die("could not convert '%s' %s",
315                                             oid_to_hex(oid), data->rest);
316                         } else if (opt->cmdmode == 'c') {
317                                 enum object_type type;
318                                 if (!textconv_object(the_repository,
319                                                      data->rest, 0100644, oid,
320                                                      1, &contents, &size))
321                                         contents = read_object_file(oid,
322                                                                     &type,
323                                                                     &size);
324                                 if (!contents)
325                                         die("could not convert '%s' %s",
326                                             oid_to_hex(oid), data->rest);
327                         } else
328                                 BUG("invalid cmdmode: %c", opt->cmdmode);
329                         batch_write(opt, contents, size);
330                         free(contents);
331                 } else {
332                         stream_blob(oid);
333                 }
334         }
335         else {
336                 enum object_type type;
337                 unsigned long size;
338                 void *contents;
339
340                 contents = read_object_file(oid, &type, &size);
341                 if (!contents)
342                         die("object %s disappeared", oid_to_hex(oid));
343                 if (type != data->type)
344                         die("object %s changed type!?", oid_to_hex(oid));
345                 if (data->info.sizep && size != data->size)
346                         die("object %s changed size!?", oid_to_hex(oid));
347
348                 batch_write(opt, contents, size);
349                 free(contents);
350         }
351 }
352
353 static void batch_object_write(const char *obj_name,
354                                struct strbuf *scratch,
355                                struct batch_options *opt,
356                                struct expand_data *data)
357 {
358         if (!data->skip_object_info &&
359             oid_object_info_extended(the_repository, &data->oid, &data->info,
360                                      OBJECT_INFO_LOOKUP_REPLACE) < 0) {
361                 printf("%s missing\n",
362                        obj_name ? obj_name : oid_to_hex(&data->oid));
363                 fflush(stdout);
364                 return;
365         }
366
367         strbuf_reset(scratch);
368         strbuf_expand(scratch, opt->format, expand_format, data);
369         strbuf_addch(scratch, '\n');
370         batch_write(opt, scratch->buf, scratch->len);
371
372         if (opt->print_contents) {
373                 print_object_or_die(opt, data);
374                 batch_write(opt, "\n", 1);
375         }
376 }
377
378 static void batch_one_object(const char *obj_name,
379                              struct strbuf *scratch,
380                              struct batch_options *opt,
381                              struct expand_data *data)
382 {
383         struct object_context ctx;
384         int flags = opt->follow_symlinks ? GET_OID_FOLLOW_SYMLINKS : 0;
385         enum get_oid_result result;
386
387         result = get_oid_with_context(the_repository, obj_name,
388                                       flags, &data->oid, &ctx);
389         if (result != FOUND) {
390                 switch (result) {
391                 case MISSING_OBJECT:
392                         printf("%s missing\n", obj_name);
393                         break;
394                 case SHORT_NAME_AMBIGUOUS:
395                         printf("%s ambiguous\n", obj_name);
396                         break;
397                 case DANGLING_SYMLINK:
398                         printf("dangling %"PRIuMAX"\n%s\n",
399                                (uintmax_t)strlen(obj_name), obj_name);
400                         break;
401                 case SYMLINK_LOOP:
402                         printf("loop %"PRIuMAX"\n%s\n",
403                                (uintmax_t)strlen(obj_name), obj_name);
404                         break;
405                 case NOT_DIR:
406                         printf("notdir %"PRIuMAX"\n%s\n",
407                                (uintmax_t)strlen(obj_name), obj_name);
408                         break;
409                 default:
410                         BUG("unknown get_sha1_with_context result %d\n",
411                                result);
412                         break;
413                 }
414                 fflush(stdout);
415                 return;
416         }
417
418         if (ctx.mode == 0) {
419                 printf("symlink %"PRIuMAX"\n%s\n",
420                        (uintmax_t)ctx.symlink_path.len,
421                        ctx.symlink_path.buf);
422                 fflush(stdout);
423                 return;
424         }
425
426         batch_object_write(obj_name, scratch, opt, data);
427 }
428
429 struct object_cb_data {
430         struct batch_options *opt;
431         struct expand_data *expand;
432         struct oidset *seen;
433         struct strbuf *scratch;
434 };
435
436 static int batch_object_cb(const struct object_id *oid, void *vdata)
437 {
438         struct object_cb_data *data = vdata;
439         oidcpy(&data->expand->oid, oid);
440         batch_object_write(NULL, data->scratch, data->opt, data->expand);
441         return 0;
442 }
443
444 static int collect_loose_object(const struct object_id *oid,
445                                 const char *path,
446                                 void *data)
447 {
448         oid_array_append(data, oid);
449         return 0;
450 }
451
452 static int collect_packed_object(const struct object_id *oid,
453                                  struct packed_git *pack,
454                                  uint32_t pos,
455                                  void *data)
456 {
457         oid_array_append(data, oid);
458         return 0;
459 }
460
461 static int batch_unordered_object(const struct object_id *oid, void *vdata)
462 {
463         struct object_cb_data *data = vdata;
464
465         if (oidset_insert(data->seen, oid))
466                 return 0;
467
468         return batch_object_cb(oid, data);
469 }
470
471 static int batch_unordered_loose(const struct object_id *oid,
472                                  const char *path,
473                                  void *data)
474 {
475         return batch_unordered_object(oid, data);
476 }
477
478 static int batch_unordered_packed(const struct object_id *oid,
479                                   struct packed_git *pack,
480                                   uint32_t pos,
481                                   void *data)
482 {
483         return batch_unordered_object(oid, data);
484 }
485
486 static int batch_objects(struct batch_options *opt)
487 {
488         struct strbuf input = STRBUF_INIT;
489         struct strbuf output = STRBUF_INIT;
490         struct expand_data data;
491         int save_warning;
492         int retval = 0;
493
494         if (!opt->format)
495                 opt->format = "%(objectname) %(objecttype) %(objectsize)";
496
497         /*
498          * Expand once with our special mark_query flag, which will prime the
499          * object_info to be handed to sha1_object_info_extended for each
500          * object.
501          */
502         memset(&data, 0, sizeof(data));
503         data.mark_query = 1;
504         strbuf_expand(&output, opt->format, expand_format, &data);
505         data.mark_query = 0;
506         strbuf_release(&output);
507         if (opt->cmdmode)
508                 data.split_on_whitespace = 1;
509
510         if (opt->all_objects) {
511                 struct object_info empty = OBJECT_INFO_INIT;
512                 if (!memcmp(&data.info, &empty, sizeof(empty)))
513                         data.skip_object_info = 1;
514         }
515
516         /*
517          * If we are printing out the object, then always fill in the type,
518          * since we will want to decide whether or not to stream.
519          */
520         if (opt->print_contents)
521                 data.info.typep = &data.type;
522
523         if (opt->all_objects) {
524                 struct object_cb_data cb;
525
526                 if (repository_format_partial_clone)
527                         warning("This repository has extensions.partialClone set. Some objects may not be loaded.");
528
529                 cb.opt = opt;
530                 cb.expand = &data;
531                 cb.scratch = &output;
532
533                 if (opt->unordered) {
534                         struct oidset seen = OIDSET_INIT;
535
536                         cb.seen = &seen;
537
538                         for_each_loose_object(batch_unordered_loose, &cb, 0);
539                         for_each_packed_object(batch_unordered_packed, &cb,
540                                                FOR_EACH_OBJECT_PACK_ORDER);
541
542                         oidset_clear(&seen);
543                 } else {
544                         struct oid_array sa = OID_ARRAY_INIT;
545
546                         for_each_loose_object(collect_loose_object, &sa, 0);
547                         for_each_packed_object(collect_packed_object, &sa, 0);
548
549                         oid_array_for_each_unique(&sa, batch_object_cb, &cb);
550
551                         oid_array_clear(&sa);
552                 }
553
554                 strbuf_release(&output);
555                 return 0;
556         }
557
558         /*
559          * We are going to call get_sha1 on a potentially very large number of
560          * objects. In most large cases, these will be actual object sha1s. The
561          * cost to double-check that each one is not also a ref (just so we can
562          * warn) ends up dwarfing the actual cost of the object lookups
563          * themselves. We can work around it by just turning off the warning.
564          */
565         save_warning = warn_on_object_refname_ambiguity;
566         warn_on_object_refname_ambiguity = 0;
567
568         while (strbuf_getline(&input, stdin) != EOF) {
569                 if (data.split_on_whitespace) {
570                         /*
571                          * Split at first whitespace, tying off the beginning
572                          * of the string and saving the remainder (or NULL) in
573                          * data.rest.
574                          */
575                         char *p = strpbrk(input.buf, " \t");
576                         if (p) {
577                                 while (*p && strchr(" \t", *p))
578                                         *p++ = '\0';
579                         }
580                         data.rest = p;
581                 }
582
583                 batch_one_object(input.buf, &output, opt, &data);
584         }
585
586         strbuf_release(&input);
587         strbuf_release(&output);
588         warn_on_object_refname_ambiguity = save_warning;
589         return retval;
590 }
591
592 static const char * const cat_file_usage[] = {
593         N_("git cat-file (-t [--allow-unknown-type] | -s [--allow-unknown-type] | -e | -p | <type> | --textconv | --filters) [--path=<path>] <object>"),
594         N_("git cat-file (--batch | --batch-check) [--follow-symlinks] [--textconv | --filters]"),
595         NULL
596 };
597
598 static int git_cat_file_config(const char *var, const char *value, void *cb)
599 {
600         if (userdiff_config(var, value) < 0)
601                 return -1;
602
603         return git_default_config(var, value, cb);
604 }
605
606 static int batch_option_callback(const struct option *opt,
607                                  const char *arg,
608                                  int unset)
609 {
610         struct batch_options *bo = opt->value;
611
612         BUG_ON_OPT_NEG(unset);
613
614         if (bo->enabled) {
615                 return error(_("only one batch option may be specified"));
616         }
617
618         bo->enabled = 1;
619         bo->print_contents = !strcmp(opt->long_name, "batch");
620         bo->format = arg;
621
622         return 0;
623 }
624
625 int cmd_cat_file(int argc, const char **argv, const char *prefix)
626 {
627         int opt = 0;
628         const char *exp_type = NULL, *obj_name = NULL;
629         struct batch_options batch = {0};
630         int unknown_type = 0;
631
632         const struct option options[] = {
633                 OPT_GROUP(N_("<type> can be one of: blob, tree, commit, tag")),
634                 OPT_CMDMODE('t', NULL, &opt, N_("show object type"), 't'),
635                 OPT_CMDMODE('s', NULL, &opt, N_("show object size"), 's'),
636                 OPT_CMDMODE('e', NULL, &opt,
637                             N_("exit with zero when there's no error"), 'e'),
638                 OPT_CMDMODE('p', NULL, &opt, N_("pretty-print object's content"), 'p'),
639                 OPT_CMDMODE(0, "textconv", &opt,
640                             N_("for blob objects, run textconv on object's content"), 'c'),
641                 OPT_CMDMODE(0, "filters", &opt,
642                             N_("for blob objects, run filters on object's content"), 'w'),
643                 OPT_STRING(0, "path", &force_path, N_("blob"),
644                            N_("use a specific path for --textconv/--filters")),
645                 OPT_BOOL(0, "allow-unknown-type", &unknown_type,
646                           N_("allow -s and -t to work with broken/corrupt objects")),
647                 OPT_BOOL(0, "buffer", &batch.buffer_output, N_("buffer --batch output")),
648                 { OPTION_CALLBACK, 0, "batch", &batch, "format",
649                         N_("show info and content of objects fed from the standard input"),
650                         PARSE_OPT_OPTARG | PARSE_OPT_NONEG,
651                         batch_option_callback },
652                 { OPTION_CALLBACK, 0, "batch-check", &batch, "format",
653                         N_("show info about objects fed from the standard input"),
654                         PARSE_OPT_OPTARG | PARSE_OPT_NONEG,
655                         batch_option_callback },
656                 OPT_BOOL(0, "follow-symlinks", &batch.follow_symlinks,
657                          N_("follow in-tree symlinks (used with --batch or --batch-check)")),
658                 OPT_BOOL(0, "batch-all-objects", &batch.all_objects,
659                          N_("show all objects with --batch or --batch-check")),
660                 OPT_BOOL(0, "unordered", &batch.unordered,
661                          N_("do not order --batch-all-objects output")),
662                 OPT_END()
663         };
664
665         git_config(git_cat_file_config, NULL);
666
667         batch.buffer_output = -1;
668         argc = parse_options(argc, argv, prefix, options, cat_file_usage, 0);
669
670         if (opt) {
671                 if (batch.enabled && (opt == 'c' || opt == 'w'))
672                         batch.cmdmode = opt;
673                 else if (argc == 1)
674                         obj_name = argv[0];
675                 else
676                         usage_with_options(cat_file_usage, options);
677         }
678         if (!opt && !batch.enabled) {
679                 if (argc == 2) {
680                         exp_type = argv[0];
681                         obj_name = argv[1];
682                 } else
683                         usage_with_options(cat_file_usage, options);
684         }
685         if (batch.enabled) {
686                 if (batch.cmdmode != opt || argc)
687                         usage_with_options(cat_file_usage, options);
688                 if (batch.cmdmode && batch.all_objects)
689                         die("--batch-all-objects cannot be combined with "
690                             "--textconv nor with --filters");
691         }
692
693         if ((batch.follow_symlinks || batch.all_objects) && !batch.enabled) {
694                 usage_with_options(cat_file_usage, options);
695         }
696
697         if (force_path && opt != 'c' && opt != 'w') {
698                 error("--path=<path> needs --textconv or --filters");
699                 usage_with_options(cat_file_usage, options);
700         }
701
702         if (force_path && batch.enabled) {
703                 error("--path=<path> incompatible with --batch");
704                 usage_with_options(cat_file_usage, options);
705         }
706
707         if (batch.buffer_output < 0)
708                 batch.buffer_output = batch.all_objects;
709
710         if (batch.enabled)
711                 return batch_objects(&batch);
712
713         if (unknown_type && opt != 't' && opt != 's')
714                 die("git cat-file --allow-unknown-type: use with -s or -t");
715         return cat_one_file(opt, exp_type, obj_name, unknown_type);
716 }