Merge branch 'ds/more-index-cleanups'
[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 "oid-array.h"
16 #include "packfile.h"
17 #include "object-store.h"
18 #include "promisor-remote.h"
19
20 struct batch_options {
21         int enabled;
22         int follow_symlinks;
23         int print_contents;
24         int buffer_output;
25         int all_objects;
26         int unordered;
27         int cmdmode; /* may be 'w' or 'c' for --filters or --textconv */
28         const char *format;
29 };
30
31 static const char *force_path;
32
33 static int filter_object(const char *path, unsigned mode,
34                          const struct object_id *oid,
35                          char **buf, unsigned long *size)
36 {
37         enum object_type type;
38
39         *buf = read_object_file(oid, &type, size);
40         if (!*buf)
41                 return error(_("cannot read object %s '%s'"),
42                              oid_to_hex(oid), path);
43         if ((type == OBJ_BLOB) && S_ISREG(mode)) {
44                 struct strbuf strbuf = STRBUF_INIT;
45                 struct checkout_metadata meta;
46
47                 init_checkout_metadata(&meta, NULL, NULL, oid);
48                 if (convert_to_working_tree(&the_index, path, *buf, *size, &strbuf, &meta)) {
49                         free(*buf);
50                         *size = strbuf.len;
51                         *buf = strbuf_detach(&strbuf, NULL);
52                 }
53         }
54
55         return 0;
56 }
57
58 static int stream_blob(const struct object_id *oid)
59 {
60         if (stream_blob_to_fd(1, oid, NULL, 0))
61                 die("unable to stream %s to stdout", oid_to_hex(oid));
62         return 0;
63 }
64
65 static int cat_one_file(int opt, const char *exp_type, const char *obj_name,
66                         int unknown_type)
67 {
68         struct object_id oid;
69         enum object_type type;
70         char *buf;
71         unsigned long size;
72         struct object_context obj_context;
73         struct object_info oi = OBJECT_INFO_INIT;
74         struct strbuf sb = STRBUF_INIT;
75         unsigned flags = OBJECT_INFO_LOOKUP_REPLACE;
76         const char *path = force_path;
77
78         if (unknown_type)
79                 flags |= OBJECT_INFO_ALLOW_UNKNOWN_TYPE;
80
81         if (get_oid_with_context(the_repository, obj_name,
82                                  GET_OID_RECORD_PATH,
83                                  &oid, &obj_context))
84                 die("Not a valid object name %s", obj_name);
85
86         if (!path)
87                 path = obj_context.path;
88         if (obj_context.mode == S_IFINVALID)
89                 obj_context.mode = 0100644;
90
91         buf = NULL;
92         switch (opt) {
93         case 't':
94                 oi.type_name = &sb;
95                 if (oid_object_info_extended(the_repository, &oid, &oi, flags) < 0)
96                         die("git cat-file: could not get object info");
97                 if (sb.len) {
98                         printf("%s\n", sb.buf);
99                         strbuf_release(&sb);
100                         return 0;
101                 }
102                 break;
103
104         case 's':
105                 oi.sizep = &size;
106                 if (oid_object_info_extended(the_repository, &oid, &oi, flags) < 0)
107                         die("git cat-file: could not get object info");
108                 printf("%"PRIuMAX"\n", (uintmax_t)size);
109                 return 0;
110
111         case 'e':
112                 return !has_object_file(&oid);
113
114         case 'w':
115                 if (!path)
116                         die("git cat-file --filters %s: <object> must be "
117                             "<sha1:path>", obj_name);
118
119                 if (filter_object(path, obj_context.mode,
120                                   &oid, &buf, &size))
121                         return -1;
122                 break;
123
124         case 'c':
125                 if (!path)
126                         die("git cat-file --textconv %s: <object> must be <sha1:path>",
127                             obj_name);
128
129                 if (textconv_object(the_repository, path, obj_context.mode,
130                                     &oid, 1, &buf, &size))
131                         break;
132                 /* else fallthrough */
133
134         case 'p':
135                 type = oid_object_info(the_repository, &oid, NULL);
136                 if (type < 0)
137                         die("Not a valid object name %s", obj_name);
138
139                 /* custom pretty-print here */
140                 if (type == OBJ_TREE) {
141                         const char *ls_args[3] = { NULL };
142                         ls_args[0] =  "ls-tree";
143                         ls_args[1] =  obj_name;
144                         return cmd_ls_tree(2, ls_args, NULL);
145                 }
146
147                 if (type == OBJ_BLOB)
148                         return stream_blob(&oid);
149                 buf = read_object_file(&oid, &type, &size);
150                 if (!buf)
151                         die("Cannot read object %s", obj_name);
152
153                 /* otherwise just spit out the data */
154                 break;
155
156         case 0:
157                 if (type_from_string(exp_type) == OBJ_BLOB) {
158                         struct object_id blob_oid;
159                         if (oid_object_info(the_repository, &oid, NULL) == OBJ_TAG) {
160                                 char *buffer = read_object_file(&oid, &type,
161                                                                 &size);
162                                 const char *target;
163                                 if (!skip_prefix(buffer, "object ", &target) ||
164                                     get_oid_hex(target, &blob_oid))
165                                         die("%s not a valid tag", oid_to_hex(&oid));
166                                 free(buffer);
167                         } else
168                                 oidcpy(&blob_oid, &oid);
169
170                         if (oid_object_info(the_repository, &blob_oid, NULL) == OBJ_BLOB)
171                                 return stream_blob(&blob_oid);
172                         /*
173                          * we attempted to dereference a tag to a blob
174                          * and failed; there may be new dereference
175                          * mechanisms this code is not aware of.
176                          * fall-back to the usual case.
177                          */
178                 }
179                 buf = read_object_with_reference(the_repository,
180                                                  &oid, exp_type, &size, NULL);
181                 break;
182
183         default:
184                 die("git cat-file: unknown option: %s", exp_type);
185         }
186
187         if (!buf)
188                 die("git cat-file %s: bad file", obj_name);
189
190         write_or_die(1, buf, size);
191         free(buf);
192         free(obj_context.path);
193         return 0;
194 }
195
196 struct expand_data {
197         struct object_id oid;
198         enum object_type type;
199         unsigned long size;
200         off_t disk_size;
201         const char *rest;
202         struct object_id delta_base_oid;
203
204         /*
205          * If mark_query is true, we do not expand anything, but rather
206          * just mark the object_info with items we wish to query.
207          */
208         int mark_query;
209
210         /*
211          * Whether to split the input on whitespace before feeding it to
212          * get_sha1; this is decided during the mark_query phase based on
213          * whether we have a %(rest) token in our format.
214          */
215         int split_on_whitespace;
216
217         /*
218          * After a mark_query run, this object_info is set up to be
219          * passed to oid_object_info_extended. It will point to the data
220          * elements above, so you can retrieve the response from there.
221          */
222         struct object_info info;
223
224         /*
225          * This flag will be true if the requested batch format and options
226          * don't require us to call oid_object_info, which can then be
227          * optimized out.
228          */
229         unsigned skip_object_info : 1;
230 };
231
232 static int is_atom(const char *atom, const char *s, int slen)
233 {
234         int alen = strlen(atom);
235         return alen == slen && !memcmp(atom, s, alen);
236 }
237
238 static void expand_atom(struct strbuf *sb, const char *atom, int len,
239                         void *vdata)
240 {
241         struct expand_data *data = vdata;
242
243         if (is_atom("objectname", atom, len)) {
244                 if (!data->mark_query)
245                         strbuf_addstr(sb, oid_to_hex(&data->oid));
246         } else if (is_atom("objecttype", atom, len)) {
247                 if (data->mark_query)
248                         data->info.typep = &data->type;
249                 else
250                         strbuf_addstr(sb, type_name(data->type));
251         } else if (is_atom("objectsize", atom, len)) {
252                 if (data->mark_query)
253                         data->info.sizep = &data->size;
254                 else
255                         strbuf_addf(sb, "%"PRIuMAX , (uintmax_t)data->size);
256         } else if (is_atom("objectsize:disk", atom, len)) {
257                 if (data->mark_query)
258                         data->info.disk_sizep = &data->disk_size;
259                 else
260                         strbuf_addf(sb, "%"PRIuMAX, (uintmax_t)data->disk_size);
261         } else if (is_atom("rest", atom, len)) {
262                 if (data->mark_query)
263                         data->split_on_whitespace = 1;
264                 else if (data->rest)
265                         strbuf_addstr(sb, data->rest);
266         } else if (is_atom("deltabase", atom, len)) {
267                 if (data->mark_query)
268                         data->info.delta_base_oid = &data->delta_base_oid;
269                 else
270                         strbuf_addstr(sb,
271                                       oid_to_hex(&data->delta_base_oid));
272         } else
273                 die("unknown format element: %.*s", len, atom);
274 }
275
276 static size_t expand_format(struct strbuf *sb, const char *start, void *data)
277 {
278         const char *end;
279
280         if (*start != '(')
281                 return 0;
282         end = strchr(start + 1, ')');
283         if (!end)
284                 die("format element '%s' does not end in ')'", start);
285
286         expand_atom(sb, start + 1, end - start - 1, data);
287
288         return end - start + 1;
289 }
290
291 static void batch_write(struct batch_options *opt, const void *data, int len)
292 {
293         if (opt->buffer_output) {
294                 if (fwrite(data, 1, len, stdout) != len)
295                         die_errno("unable to write to stdout");
296         } else
297                 write_or_die(1, data, len);
298 }
299
300 static void print_object_or_die(struct batch_options *opt, struct expand_data *data)
301 {
302         const struct object_id *oid = &data->oid;
303
304         assert(data->info.typep);
305
306         if (data->type == OBJ_BLOB) {
307                 if (opt->buffer_output)
308                         fflush(stdout);
309                 if (opt->cmdmode) {
310                         char *contents;
311                         unsigned long size;
312
313                         if (!data->rest)
314                                 die("missing path for '%s'", oid_to_hex(oid));
315
316                         if (opt->cmdmode == 'w') {
317                                 if (filter_object(data->rest, 0100644, oid,
318                                                   &contents, &size))
319                                         die("could not convert '%s' %s",
320                                             oid_to_hex(oid), data->rest);
321                         } else if (opt->cmdmode == 'c') {
322                                 enum object_type type;
323                                 if (!textconv_object(the_repository,
324                                                      data->rest, 0100644, oid,
325                                                      1, &contents, &size))
326                                         contents = read_object_file(oid,
327                                                                     &type,
328                                                                     &size);
329                                 if (!contents)
330                                         die("could not convert '%s' %s",
331                                             oid_to_hex(oid), data->rest);
332                         } else
333                                 BUG("invalid cmdmode: %c", opt->cmdmode);
334                         batch_write(opt, contents, size);
335                         free(contents);
336                 } else {
337                         stream_blob(oid);
338                 }
339         }
340         else {
341                 enum object_type type;
342                 unsigned long size;
343                 void *contents;
344
345                 contents = read_object_file(oid, &type, &size);
346                 if (!contents)
347                         die("object %s disappeared", oid_to_hex(oid));
348                 if (type != data->type)
349                         die("object %s changed type!?", oid_to_hex(oid));
350                 if (data->info.sizep && size != data->size)
351                         die("object %s changed size!?", oid_to_hex(oid));
352
353                 batch_write(opt, contents, size);
354                 free(contents);
355         }
356 }
357
358 static void batch_object_write(const char *obj_name,
359                                struct strbuf *scratch,
360                                struct batch_options *opt,
361                                struct expand_data *data)
362 {
363         if (!data->skip_object_info &&
364             oid_object_info_extended(the_repository, &data->oid, &data->info,
365                                      OBJECT_INFO_LOOKUP_REPLACE) < 0) {
366                 printf("%s missing\n",
367                        obj_name ? obj_name : oid_to_hex(&data->oid));
368                 fflush(stdout);
369                 return;
370         }
371
372         strbuf_reset(scratch);
373         strbuf_expand(scratch, opt->format, expand_format, data);
374         strbuf_addch(scratch, '\n');
375         batch_write(opt, scratch->buf, scratch->len);
376
377         if (opt->print_contents) {
378                 print_object_or_die(opt, data);
379                 batch_write(opt, "\n", 1);
380         }
381 }
382
383 static void batch_one_object(const char *obj_name,
384                              struct strbuf *scratch,
385                              struct batch_options *opt,
386                              struct expand_data *data)
387 {
388         struct object_context ctx;
389         int flags = opt->follow_symlinks ? GET_OID_FOLLOW_SYMLINKS : 0;
390         enum get_oid_result result;
391
392         result = get_oid_with_context(the_repository, obj_name,
393                                       flags, &data->oid, &ctx);
394         if (result != FOUND) {
395                 switch (result) {
396                 case MISSING_OBJECT:
397                         printf("%s missing\n", obj_name);
398                         break;
399                 case SHORT_NAME_AMBIGUOUS:
400                         printf("%s ambiguous\n", obj_name);
401                         break;
402                 case DANGLING_SYMLINK:
403                         printf("dangling %"PRIuMAX"\n%s\n",
404                                (uintmax_t)strlen(obj_name), obj_name);
405                         break;
406                 case SYMLINK_LOOP:
407                         printf("loop %"PRIuMAX"\n%s\n",
408                                (uintmax_t)strlen(obj_name), obj_name);
409                         break;
410                 case NOT_DIR:
411                         printf("notdir %"PRIuMAX"\n%s\n",
412                                (uintmax_t)strlen(obj_name), obj_name);
413                         break;
414                 default:
415                         BUG("unknown get_sha1_with_context result %d\n",
416                                result);
417                         break;
418                 }
419                 fflush(stdout);
420                 return;
421         }
422
423         if (ctx.mode == 0) {
424                 printf("symlink %"PRIuMAX"\n%s\n",
425                        (uintmax_t)ctx.symlink_path.len,
426                        ctx.symlink_path.buf);
427                 fflush(stdout);
428                 return;
429         }
430
431         batch_object_write(obj_name, scratch, opt, data);
432 }
433
434 struct object_cb_data {
435         struct batch_options *opt;
436         struct expand_data *expand;
437         struct oidset *seen;
438         struct strbuf *scratch;
439 };
440
441 static int batch_object_cb(const struct object_id *oid, void *vdata)
442 {
443         struct object_cb_data *data = vdata;
444         oidcpy(&data->expand->oid, oid);
445         batch_object_write(NULL, data->scratch, data->opt, data->expand);
446         return 0;
447 }
448
449 static int collect_loose_object(const struct object_id *oid,
450                                 const char *path,
451                                 void *data)
452 {
453         oid_array_append(data, oid);
454         return 0;
455 }
456
457 static int collect_packed_object(const struct object_id *oid,
458                                  struct packed_git *pack,
459                                  uint32_t pos,
460                                  void *data)
461 {
462         oid_array_append(data, oid);
463         return 0;
464 }
465
466 static int batch_unordered_object(const struct object_id *oid, void *vdata)
467 {
468         struct object_cb_data *data = vdata;
469
470         if (oidset_insert(data->seen, oid))
471                 return 0;
472
473         return batch_object_cb(oid, data);
474 }
475
476 static int batch_unordered_loose(const struct object_id *oid,
477                                  const char *path,
478                                  void *data)
479 {
480         return batch_unordered_object(oid, data);
481 }
482
483 static int batch_unordered_packed(const struct object_id *oid,
484                                   struct packed_git *pack,
485                                   uint32_t pos,
486                                   void *data)
487 {
488         return batch_unordered_object(oid, data);
489 }
490
491 static int batch_objects(struct batch_options *opt)
492 {
493         struct strbuf input = STRBUF_INIT;
494         struct strbuf output = STRBUF_INIT;
495         struct expand_data data;
496         int save_warning;
497         int retval = 0;
498
499         if (!opt->format)
500                 opt->format = "%(objectname) %(objecttype) %(objectsize)";
501
502         /*
503          * Expand once with our special mark_query flag, which will prime the
504          * object_info to be handed to oid_object_info_extended for each
505          * object.
506          */
507         memset(&data, 0, sizeof(data));
508         data.mark_query = 1;
509         strbuf_expand(&output, opt->format, expand_format, &data);
510         data.mark_query = 0;
511         strbuf_release(&output);
512         if (opt->cmdmode)
513                 data.split_on_whitespace = 1;
514
515         if (opt->all_objects) {
516                 struct object_info empty = OBJECT_INFO_INIT;
517                 if (!memcmp(&data.info, &empty, sizeof(empty)))
518                         data.skip_object_info = 1;
519         }
520
521         /*
522          * If we are printing out the object, then always fill in the type,
523          * since we will want to decide whether or not to stream.
524          */
525         if (opt->print_contents)
526                 data.info.typep = &data.type;
527
528         if (opt->all_objects) {
529                 struct object_cb_data cb;
530
531                 if (has_promisor_remote())
532                         warning("This repository uses promisor remotes. Some objects may not be loaded.");
533
534                 cb.opt = opt;
535                 cb.expand = &data;
536                 cb.scratch = &output;
537
538                 if (opt->unordered) {
539                         struct oidset seen = OIDSET_INIT;
540
541                         cb.seen = &seen;
542
543                         for_each_loose_object(batch_unordered_loose, &cb, 0);
544                         for_each_packed_object(batch_unordered_packed, &cb,
545                                                FOR_EACH_OBJECT_PACK_ORDER);
546
547                         oidset_clear(&seen);
548                 } else {
549                         struct oid_array sa = OID_ARRAY_INIT;
550
551                         for_each_loose_object(collect_loose_object, &sa, 0);
552                         for_each_packed_object(collect_packed_object, &sa, 0);
553
554                         oid_array_for_each_unique(&sa, batch_object_cb, &cb);
555
556                         oid_array_clear(&sa);
557                 }
558
559                 strbuf_release(&output);
560                 return 0;
561         }
562
563         /*
564          * We are going to call get_sha1 on a potentially very large number of
565          * objects. In most large cases, these will be actual object sha1s. The
566          * cost to double-check that each one is not also a ref (just so we can
567          * warn) ends up dwarfing the actual cost of the object lookups
568          * themselves. We can work around it by just turning off the warning.
569          */
570         save_warning = warn_on_object_refname_ambiguity;
571         warn_on_object_refname_ambiguity = 0;
572
573         while (strbuf_getline(&input, stdin) != EOF) {
574                 if (data.split_on_whitespace) {
575                         /*
576                          * Split at first whitespace, tying off the beginning
577                          * of the string and saving the remainder (or NULL) in
578                          * data.rest.
579                          */
580                         char *p = strpbrk(input.buf, " \t");
581                         if (p) {
582                                 while (*p && strchr(" \t", *p))
583                                         *p++ = '\0';
584                         }
585                         data.rest = p;
586                 }
587
588                 batch_one_object(input.buf, &output, opt, &data);
589         }
590
591         strbuf_release(&input);
592         strbuf_release(&output);
593         warn_on_object_refname_ambiguity = save_warning;
594         return retval;
595 }
596
597 static const char * const cat_file_usage[] = {
598         N_("git cat-file (-t [--allow-unknown-type] | -s [--allow-unknown-type] | -e | -p | <type> | --textconv | --filters) [--path=<path>] <object>"),
599         N_("git cat-file (--batch[=<format>] | --batch-check[=<format>]) [--follow-symlinks] [--textconv | --filters]"),
600         NULL
601 };
602
603 static int git_cat_file_config(const char *var, const char *value, void *cb)
604 {
605         if (userdiff_config(var, value) < 0)
606                 return -1;
607
608         return git_default_config(var, value, cb);
609 }
610
611 static int batch_option_callback(const struct option *opt,
612                                  const char *arg,
613                                  int unset)
614 {
615         struct batch_options *bo = opt->value;
616
617         BUG_ON_OPT_NEG(unset);
618
619         if (bo->enabled) {
620                 return error(_("only one batch option may be specified"));
621         }
622
623         bo->enabled = 1;
624         bo->print_contents = !strcmp(opt->long_name, "batch");
625         bo->format = arg;
626
627         return 0;
628 }
629
630 int cmd_cat_file(int argc, const char **argv, const char *prefix)
631 {
632         int opt = 0;
633         const char *exp_type = NULL, *obj_name = NULL;
634         struct batch_options batch = {0};
635         int unknown_type = 0;
636
637         const struct option options[] = {
638                 OPT_GROUP(N_("<type> can be one of: blob, tree, commit, tag")),
639                 OPT_CMDMODE('t', NULL, &opt, N_("show object type"), 't'),
640                 OPT_CMDMODE('s', NULL, &opt, N_("show object size"), 's'),
641                 OPT_CMDMODE('e', NULL, &opt,
642                             N_("exit with zero when there's no error"), 'e'),
643                 OPT_CMDMODE('p', NULL, &opt, N_("pretty-print object's content"), 'p'),
644                 OPT_CMDMODE(0, "textconv", &opt,
645                             N_("for blob objects, run textconv on object's content"), 'c'),
646                 OPT_CMDMODE(0, "filters", &opt,
647                             N_("for blob objects, run filters on object's content"), 'w'),
648                 OPT_STRING(0, "path", &force_path, N_("blob"),
649                            N_("use a specific path for --textconv/--filters")),
650                 OPT_BOOL(0, "allow-unknown-type", &unknown_type,
651                           N_("allow -s and -t to work with broken/corrupt objects")),
652                 OPT_BOOL(0, "buffer", &batch.buffer_output, N_("buffer --batch output")),
653                 OPT_CALLBACK_F(0, "batch", &batch, "format",
654                         N_("show info and content of objects fed from the standard input"),
655                         PARSE_OPT_OPTARG | PARSE_OPT_NONEG,
656                         batch_option_callback),
657                 OPT_CALLBACK_F(0, "batch-check", &batch, "format",
658                         N_("show info about objects fed from the standard input"),
659                         PARSE_OPT_OPTARG | PARSE_OPT_NONEG,
660                         batch_option_callback),
661                 OPT_BOOL(0, "follow-symlinks", &batch.follow_symlinks,
662                          N_("follow in-tree symlinks (used with --batch or --batch-check)")),
663                 OPT_BOOL(0, "batch-all-objects", &batch.all_objects,
664                          N_("show all objects with --batch or --batch-check")),
665                 OPT_BOOL(0, "unordered", &batch.unordered,
666                          N_("do not order --batch-all-objects output")),
667                 OPT_END()
668         };
669
670         git_config(git_cat_file_config, NULL);
671
672         batch.buffer_output = -1;
673         argc = parse_options(argc, argv, prefix, options, cat_file_usage, 0);
674
675         if (opt) {
676                 if (batch.enabled && (opt == 'c' || opt == 'w'))
677                         batch.cmdmode = opt;
678                 else if (argc == 1)
679                         obj_name = argv[0];
680                 else
681                         usage_with_options(cat_file_usage, options);
682         }
683         if (!opt && !batch.enabled) {
684                 if (argc == 2) {
685                         exp_type = argv[0];
686                         obj_name = argv[1];
687                 } else
688                         usage_with_options(cat_file_usage, options);
689         }
690         if (batch.enabled) {
691                 if (batch.cmdmode != opt || argc)
692                         usage_with_options(cat_file_usage, options);
693                 if (batch.cmdmode && batch.all_objects)
694                         die("--batch-all-objects cannot be combined with "
695                             "--textconv nor with --filters");
696         }
697
698         if ((batch.follow_symlinks || batch.all_objects) && !batch.enabled) {
699                 usage_with_options(cat_file_usage, options);
700         }
701
702         if (force_path && opt != 'c' && opt != 'w') {
703                 error("--path=<path> needs --textconv or --filters");
704                 usage_with_options(cat_file_usage, options);
705         }
706
707         if (force_path && batch.enabled) {
708                 error("--path=<path> incompatible with --batch");
709                 usage_with_options(cat_file_usage, options);
710         }
711
712         if (batch.buffer_output < 0)
713                 batch.buffer_output = batch.all_objects;
714
715         if (batch.enabled)
716                 return batch_objects(&batch);
717
718         if (unknown_type && opt != 't' && opt != 's')
719                 die("git cat-file --allow-unknown-type: use with -s or -t");
720         return cat_one_file(opt, exp_type, obj_name, unknown_type);
721 }