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