Merge branch 'rs/sha1-name-readdir-optim'
[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
16 struct batch_options {
17         int enabled;
18         int follow_symlinks;
19         int print_contents;
20         int buffer_output;
21         int all_objects;
22         int cmdmode; /* may be 'w' or 'c' for --filters or --textconv */
23         const char *format;
24 };
25
26 static const char *force_path;
27
28 static int filter_object(const char *path, unsigned mode,
29                          const struct object_id *oid,
30                          char **buf, unsigned long *size)
31 {
32         enum object_type type;
33
34         *buf = read_sha1_file(oid->hash, &type, size);
35         if (!*buf)
36                 return error(_("cannot read object %s '%s'"),
37                              oid_to_hex(oid), path);
38         if ((type == OBJ_BLOB) && S_ISREG(mode)) {
39                 struct strbuf strbuf = STRBUF_INIT;
40                 if (convert_to_working_tree(path, *buf, *size, &strbuf)) {
41                         free(*buf);
42                         *size = strbuf.len;
43                         *buf = strbuf_detach(&strbuf, NULL);
44                 }
45         }
46
47         return 0;
48 }
49
50 static int cat_one_file(int opt, const char *exp_type, const char *obj_name,
51                         int unknown_type)
52 {
53         struct object_id oid;
54         enum object_type type;
55         char *buf;
56         unsigned long size;
57         struct object_context obj_context;
58         struct object_info oi = OBJECT_INFO_INIT;
59         struct strbuf sb = STRBUF_INIT;
60         unsigned flags = LOOKUP_REPLACE_OBJECT;
61         const char *path = force_path;
62
63         if (unknown_type)
64                 flags |= LOOKUP_UNKNOWN_OBJECT;
65
66         if (get_sha1_with_context(obj_name, GET_SHA1_RECORD_PATH,
67                                   oid.hash, &obj_context))
68                 die("Not a valid object name %s", obj_name);
69
70         if (!path)
71                 path = obj_context.path;
72         if (obj_context.mode == S_IFINVALID)
73                 obj_context.mode = 0100644;
74
75         buf = NULL;
76         switch (opt) {
77         case 't':
78                 oi.typename = &sb;
79                 if (sha1_object_info_extended(oid.hash, &oi, flags) < 0)
80                         die("git cat-file: could not get object info");
81                 if (sb.len) {
82                         printf("%s\n", sb.buf);
83                         strbuf_release(&sb);
84                         return 0;
85                 }
86                 break;
87
88         case 's':
89                 oi.sizep = &size;
90                 if (sha1_object_info_extended(oid.hash, &oi, flags) < 0)
91                         die("git cat-file: could not get object info");
92                 printf("%lu\n", size);
93                 return 0;
94
95         case 'e':
96                 return !has_object_file(&oid);
97
98         case 'w':
99                 if (!path[0])
100                         die("git cat-file --filters %s: <object> must be "
101                             "<sha1:path>", obj_name);
102
103                 if (filter_object(path, obj_context.mode,
104                                   &oid, &buf, &size))
105                         return -1;
106                 break;
107
108         case 'c':
109                 if (!path[0])
110                         die("git cat-file --textconv %s: <object> must be <sha1:path>",
111                             obj_name);
112
113                 if (textconv_object(path, obj_context.mode, &oid, 1, &buf, &size))
114                         break;
115
116         case 'p':
117                 type = sha1_object_info(oid.hash, NULL);
118                 if (type < 0)
119                         die("Not a valid object name %s", obj_name);
120
121                 /* custom pretty-print here */
122                 if (type == OBJ_TREE) {
123                         const char *ls_args[3] = { NULL };
124                         ls_args[0] =  "ls-tree";
125                         ls_args[1] =  obj_name;
126                         return cmd_ls_tree(2, ls_args, NULL);
127                 }
128
129                 if (type == OBJ_BLOB)
130                         return stream_blob_to_fd(1, &oid, NULL, 0);
131                 buf = read_sha1_file(oid.hash, &type, &size);
132                 if (!buf)
133                         die("Cannot read object %s", obj_name);
134
135                 /* otherwise just spit out the data */
136                 break;
137
138         case 0:
139                 if (type_from_string(exp_type) == OBJ_BLOB) {
140                         struct object_id blob_oid;
141                         if (sha1_object_info(oid.hash, NULL) == OBJ_TAG) {
142                                 char *buffer = read_sha1_file(oid.hash, &type, &size);
143                                 const char *target;
144                                 if (!skip_prefix(buffer, "object ", &target) ||
145                                     get_oid_hex(target, &blob_oid))
146                                         die("%s not a valid tag", oid_to_hex(&oid));
147                                 free(buffer);
148                         } else
149                                 oidcpy(&blob_oid, &oid);
150
151                         if (sha1_object_info(blob_oid.hash, NULL) == OBJ_BLOB)
152                                 return stream_blob_to_fd(1, &blob_oid, NULL, 0);
153                         /*
154                          * we attempted to dereference a tag to a blob
155                          * and failed; there may be new dereference
156                          * mechanisms this code is not aware of.
157                          * fall-back to the usual case.
158                          */
159                 }
160                 buf = read_object_with_reference(oid.hash, exp_type, &size, NULL);
161                 break;
162
163         default:
164                 die("git cat-file: unknown option: %s", exp_type);
165         }
166
167         if (!buf)
168                 die("git cat-file %s: bad file", obj_name);
169
170         write_or_die(1, buf, size);
171         free(buf);
172         free(obj_context.path);
173         return 0;
174 }
175
176 struct expand_data {
177         struct object_id oid;
178         enum object_type type;
179         unsigned long size;
180         off_t disk_size;
181         const char *rest;
182         struct object_id delta_base_oid;
183
184         /*
185          * If mark_query is true, we do not expand anything, but rather
186          * just mark the object_info with items we wish to query.
187          */
188         int mark_query;
189
190         /*
191          * Whether to split the input on whitespace before feeding it to
192          * get_sha1; this is decided during the mark_query phase based on
193          * whether we have a %(rest) token in our format.
194          */
195         int split_on_whitespace;
196
197         /*
198          * After a mark_query run, this object_info is set up to be
199          * passed to sha1_object_info_extended. It will point to the data
200          * elements above, so you can retrieve the response from there.
201          */
202         struct object_info info;
203
204         /*
205          * This flag will be true if the requested batch format and options
206          * don't require us to call sha1_object_info, which can then be
207          * optimized out.
208          */
209         unsigned skip_object_info : 1;
210 };
211
212 static int is_atom(const char *atom, const char *s, int slen)
213 {
214         int alen = strlen(atom);
215         return alen == slen && !memcmp(atom, s, alen);
216 }
217
218 static void expand_atom(struct strbuf *sb, const char *atom, int len,
219                         void *vdata)
220 {
221         struct expand_data *data = vdata;
222
223         if (is_atom("objectname", atom, len)) {
224                 if (!data->mark_query)
225                         strbuf_addstr(sb, oid_to_hex(&data->oid));
226         } else if (is_atom("objecttype", atom, len)) {
227                 if (data->mark_query)
228                         data->info.typep = &data->type;
229                 else
230                         strbuf_addstr(sb, typename(data->type));
231         } else if (is_atom("objectsize", atom, len)) {
232                 if (data->mark_query)
233                         data->info.sizep = &data->size;
234                 else
235                         strbuf_addf(sb, "%lu", data->size);
236         } else if (is_atom("objectsize:disk", atom, len)) {
237                 if (data->mark_query)
238                         data->info.disk_sizep = &data->disk_size;
239                 else
240                         strbuf_addf(sb, "%"PRIuMAX, (uintmax_t)data->disk_size);
241         } else if (is_atom("rest", atom, len)) {
242                 if (data->mark_query)
243                         data->split_on_whitespace = 1;
244                 else if (data->rest)
245                         strbuf_addstr(sb, data->rest);
246         } else if (is_atom("deltabase", atom, len)) {
247                 if (data->mark_query)
248                         data->info.delta_base_sha1 = data->delta_base_oid.hash;
249                 else
250                         strbuf_addstr(sb,
251                                       oid_to_hex(&data->delta_base_oid));
252         } else
253                 die("unknown format element: %.*s", len, atom);
254 }
255
256 static size_t expand_format(struct strbuf *sb, const char *start, void *data)
257 {
258         const char *end;
259
260         if (*start != '(')
261                 return 0;
262         end = strchr(start + 1, ')');
263         if (!end)
264                 die("format element '%s' does not end in ')'", start);
265
266         expand_atom(sb, start + 1, end - start - 1, data);
267
268         return end - start + 1;
269 }
270
271 static void batch_write(struct batch_options *opt, const void *data, int len)
272 {
273         if (opt->buffer_output) {
274                 if (fwrite(data, 1, len, stdout) != len)
275                         die_errno("unable to write to stdout");
276         } else
277                 write_or_die(1, data, len);
278 }
279
280 static void print_object_or_die(struct batch_options *opt, struct expand_data *data)
281 {
282         const struct object_id *oid = &data->oid;
283
284         assert(data->info.typep);
285
286         if (data->type == OBJ_BLOB) {
287                 if (opt->buffer_output)
288                         fflush(stdout);
289                 if (opt->cmdmode) {
290                         char *contents;
291                         unsigned long size;
292
293                         if (!data->rest)
294                                 die("missing path for '%s'", oid_to_hex(oid));
295
296                         if (opt->cmdmode == 'w') {
297                                 if (filter_object(data->rest, 0100644, oid,
298                                                   &contents, &size))
299                                         die("could not convert '%s' %s",
300                                             oid_to_hex(oid), data->rest);
301                         } else if (opt->cmdmode == 'c') {
302                                 enum object_type type;
303                                 if (!textconv_object(data->rest, 0100644, oid,
304                                                      1, &contents, &size))
305                                         contents = read_sha1_file(oid->hash, &type,
306                                                                   &size);
307                                 if (!contents)
308                                         die("could not convert '%s' %s",
309                                             oid_to_hex(oid), data->rest);
310                         } else
311                                 die("BUG: invalid cmdmode: %c", opt->cmdmode);
312                         batch_write(opt, contents, size);
313                         free(contents);
314                 } else if (stream_blob_to_fd(1, oid, NULL, 0) < 0)
315                         die("unable to stream %s to stdout", oid_to_hex(oid));
316         }
317         else {
318                 enum object_type type;
319                 unsigned long size;
320                 void *contents;
321
322                 contents = read_sha1_file(oid->hash, &type, &size);
323                 if (!contents)
324                         die("object %s disappeared", oid_to_hex(oid));
325                 if (type != data->type)
326                         die("object %s changed type!?", oid_to_hex(oid));
327                 if (data->info.sizep && size != data->size)
328                         die("object %s changed size!?", oid_to_hex(oid));
329
330                 batch_write(opt, contents, size);
331                 free(contents);
332         }
333 }
334
335 static void batch_object_write(const char *obj_name, struct batch_options *opt,
336                                struct expand_data *data)
337 {
338         struct strbuf buf = STRBUF_INIT;
339
340         if (!data->skip_object_info &&
341             sha1_object_info_extended(data->oid.hash, &data->info, LOOKUP_REPLACE_OBJECT) < 0) {
342                 printf("%s missing\n",
343                        obj_name ? obj_name : oid_to_hex(&data->oid));
344                 fflush(stdout);
345                 return;
346         }
347
348         strbuf_expand(&buf, opt->format, expand_format, data);
349         strbuf_addch(&buf, '\n');
350         batch_write(opt, buf.buf, buf.len);
351         strbuf_release(&buf);
352
353         if (opt->print_contents) {
354                 print_object_or_die(opt, data);
355                 batch_write(opt, "\n", 1);
356         }
357 }
358
359 static void batch_one_object(const char *obj_name, struct batch_options *opt,
360                              struct expand_data *data)
361 {
362         struct object_context ctx;
363         int flags = opt->follow_symlinks ? GET_SHA1_FOLLOW_SYMLINKS : 0;
364         enum follow_symlinks_result result;
365
366         result = get_sha1_with_context(obj_name, flags, data->oid.hash, &ctx);
367         if (result != FOUND) {
368                 switch (result) {
369                 case MISSING_OBJECT:
370                         printf("%s missing\n", obj_name);
371                         break;
372                 case DANGLING_SYMLINK:
373                         printf("dangling %"PRIuMAX"\n%s\n",
374                                (uintmax_t)strlen(obj_name), obj_name);
375                         break;
376                 case SYMLINK_LOOP:
377                         printf("loop %"PRIuMAX"\n%s\n",
378                                (uintmax_t)strlen(obj_name), obj_name);
379                         break;
380                 case NOT_DIR:
381                         printf("notdir %"PRIuMAX"\n%s\n",
382                                (uintmax_t)strlen(obj_name), obj_name);
383                         break;
384                 default:
385                         die("BUG: unknown get_sha1_with_context result %d\n",
386                                result);
387                         break;
388                 }
389                 fflush(stdout);
390                 return;
391         }
392
393         if (ctx.mode == 0) {
394                 printf("symlink %"PRIuMAX"\n%s\n",
395                        (uintmax_t)ctx.symlink_path.len,
396                        ctx.symlink_path.buf);
397                 fflush(stdout);
398                 return;
399         }
400
401         batch_object_write(obj_name, opt, data);
402 }
403
404 struct object_cb_data {
405         struct batch_options *opt;
406         struct expand_data *expand;
407 };
408
409 static int batch_object_cb(const struct object_id *oid, void *vdata)
410 {
411         struct object_cb_data *data = vdata;
412         oidcpy(&data->expand->oid, oid);
413         batch_object_write(NULL, data->opt, data->expand);
414         return 0;
415 }
416
417 static int batch_loose_object(const struct object_id *oid,
418                               const char *path,
419                               void *data)
420 {
421         oid_array_append(data, oid);
422         return 0;
423 }
424
425 static int batch_packed_object(const struct object_id *oid,
426                                struct packed_git *pack,
427                                uint32_t pos,
428                                void *data)
429 {
430         oid_array_append(data, oid);
431         return 0;
432 }
433
434 static int batch_objects(struct batch_options *opt)
435 {
436         struct strbuf buf = STRBUF_INIT;
437         struct expand_data data;
438         int save_warning;
439         int retval = 0;
440
441         if (!opt->format)
442                 opt->format = "%(objectname) %(objecttype) %(objectsize)";
443
444         /*
445          * Expand once with our special mark_query flag, which will prime the
446          * object_info to be handed to sha1_object_info_extended for each
447          * object.
448          */
449         memset(&data, 0, sizeof(data));
450         data.mark_query = 1;
451         strbuf_expand(&buf, opt->format, expand_format, &data);
452         data.mark_query = 0;
453         if (opt->cmdmode)
454                 data.split_on_whitespace = 1;
455
456         if (opt->all_objects) {
457                 struct object_info empty = OBJECT_INFO_INIT;
458                 if (!memcmp(&data.info, &empty, sizeof(empty)))
459                         data.skip_object_info = 1;
460         }
461
462         /*
463          * If we are printing out the object, then always fill in the type,
464          * since we will want to decide whether or not to stream.
465          */
466         if (opt->print_contents)
467                 data.info.typep = &data.type;
468
469         if (opt->all_objects) {
470                 struct oid_array sa = OID_ARRAY_INIT;
471                 struct object_cb_data cb;
472
473                 for_each_loose_object(batch_loose_object, &sa, 0);
474                 for_each_packed_object(batch_packed_object, &sa, 0);
475
476                 cb.opt = opt;
477                 cb.expand = &data;
478                 oid_array_for_each_unique(&sa, batch_object_cb, &cb);
479
480                 oid_array_clear(&sa);
481                 return 0;
482         }
483
484         /*
485          * We are going to call get_sha1 on a potentially very large number of
486          * objects. In most large cases, these will be actual object sha1s. The
487          * cost to double-check that each one is not also a ref (just so we can
488          * warn) ends up dwarfing the actual cost of the object lookups
489          * themselves. We can work around it by just turning off the warning.
490          */
491         save_warning = warn_on_object_refname_ambiguity;
492         warn_on_object_refname_ambiguity = 0;
493
494         while (strbuf_getline(&buf, stdin) != EOF) {
495                 if (data.split_on_whitespace) {
496                         /*
497                          * Split at first whitespace, tying off the beginning
498                          * of the string and saving the remainder (or NULL) in
499                          * data.rest.
500                          */
501                         char *p = strpbrk(buf.buf, " \t");
502                         if (p) {
503                                 while (*p && strchr(" \t", *p))
504                                         *p++ = '\0';
505                         }
506                         data.rest = p;
507                 }
508
509                 batch_one_object(buf.buf, opt, &data);
510         }
511
512         strbuf_release(&buf);
513         warn_on_object_refname_ambiguity = save_warning;
514         return retval;
515 }
516
517 static const char * const cat_file_usage[] = {
518         N_("git cat-file (-t [--allow-unknown-type] | -s [--allow-unknown-type] | -e | -p | <type> | --textconv | --filters) [--path=<path>] <object>"),
519         N_("git cat-file (--batch | --batch-check) [--follow-symlinks] [--textconv | --filters]"),
520         NULL
521 };
522
523 static int git_cat_file_config(const char *var, const char *value, void *cb)
524 {
525         if (userdiff_config(var, value) < 0)
526                 return -1;
527
528         return git_default_config(var, value, cb);
529 }
530
531 static int batch_option_callback(const struct option *opt,
532                                  const char *arg,
533                                  int unset)
534 {
535         struct batch_options *bo = opt->value;
536
537         if (bo->enabled) {
538                 return 1;
539         }
540
541         bo->enabled = 1;
542         bo->print_contents = !strcmp(opt->long_name, "batch");
543         bo->format = arg;
544
545         return 0;
546 }
547
548 int cmd_cat_file(int argc, const char **argv, const char *prefix)
549 {
550         int opt = 0;
551         const char *exp_type = NULL, *obj_name = NULL;
552         struct batch_options batch = {0};
553         int unknown_type = 0;
554
555         const struct option options[] = {
556                 OPT_GROUP(N_("<type> can be one of: blob, tree, commit, tag")),
557                 OPT_CMDMODE('t', NULL, &opt, N_("show object type"), 't'),
558                 OPT_CMDMODE('s', NULL, &opt, N_("show object size"), 's'),
559                 OPT_CMDMODE('e', NULL, &opt,
560                             N_("exit with zero when there's no error"), 'e'),
561                 OPT_CMDMODE('p', NULL, &opt, N_("pretty-print object's content"), 'p'),
562                 OPT_CMDMODE(0, "textconv", &opt,
563                             N_("for blob objects, run textconv on object's content"), 'c'),
564                 OPT_CMDMODE(0, "filters", &opt,
565                             N_("for blob objects, run filters on object's content"), 'w'),
566                 OPT_STRING(0, "path", &force_path, N_("blob"),
567                            N_("use a specific path for --textconv/--filters")),
568                 OPT_BOOL(0, "allow-unknown-type", &unknown_type,
569                           N_("allow -s and -t to work with broken/corrupt objects")),
570                 OPT_BOOL(0, "buffer", &batch.buffer_output, N_("buffer --batch output")),
571                 { OPTION_CALLBACK, 0, "batch", &batch, "format",
572                         N_("show info and content of objects fed from the standard input"),
573                         PARSE_OPT_OPTARG, batch_option_callback },
574                 { OPTION_CALLBACK, 0, "batch-check", &batch, "format",
575                         N_("show info about objects fed from the standard input"),
576                         PARSE_OPT_OPTARG, batch_option_callback },
577                 OPT_BOOL(0, "follow-symlinks", &batch.follow_symlinks,
578                          N_("follow in-tree symlinks (used with --batch or --batch-check)")),
579                 OPT_BOOL(0, "batch-all-objects", &batch.all_objects,
580                          N_("show all objects with --batch or --batch-check")),
581                 OPT_END()
582         };
583
584         git_config(git_cat_file_config, NULL);
585
586         batch.buffer_output = -1;
587         argc = parse_options(argc, argv, prefix, options, cat_file_usage, 0);
588
589         if (opt) {
590                 if (batch.enabled && (opt == 'c' || opt == 'w'))
591                         batch.cmdmode = opt;
592                 else if (argc == 1)
593                         obj_name = argv[0];
594                 else
595                         usage_with_options(cat_file_usage, options);
596         }
597         if (!opt && !batch.enabled) {
598                 if (argc == 2) {
599                         exp_type = argv[0];
600                         obj_name = argv[1];
601                 } else
602                         usage_with_options(cat_file_usage, options);
603         }
604         if (batch.enabled) {
605                 if (batch.cmdmode != opt || argc)
606                         usage_with_options(cat_file_usage, options);
607                 if (batch.cmdmode && batch.all_objects)
608                         die("--batch-all-objects cannot be combined with "
609                             "--textconv nor with --filters");
610         }
611
612         if ((batch.follow_symlinks || batch.all_objects) && !batch.enabled) {
613                 usage_with_options(cat_file_usage, options);
614         }
615
616         if (force_path && opt != 'c' && opt != 'w') {
617                 error("--path=<path> needs --textconv or --filters");
618                 usage_with_options(cat_file_usage, options);
619         }
620
621         if (force_path && batch.enabled) {
622                 error("--path=<path> incompatible with --batch");
623                 usage_with_options(cat_file_usage, options);
624         }
625
626         if (batch.buffer_output < 0)
627                 batch.buffer_output = batch.all_objects;
628
629         if (batch.enabled)
630                 return batch_objects(&batch);
631
632         if (unknown_type && opt != 't' && opt != 's')
633                 die("git cat-file --allow-unknown-type: use with -s or -t");
634         return cat_one_file(opt, exp_type, obj_name, unknown_type);
635 }