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