Merge branch 'rs/copy-array' into maint
[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 "builtin.h"
8 #include "parse-options.h"
9 #include "userdiff.h"
10 #include "streaming.h"
11 #include "tree-walk.h"
12 #include "sha1-array.h"
13
14 struct batch_options {
15         int enabled;
16         int follow_symlinks;
17         int print_contents;
18         int buffer_output;
19         int all_objects;
20         const char *format;
21 };
22
23 static int cat_one_file(int opt, const char *exp_type, const char *obj_name,
24                         int unknown_type)
25 {
26         unsigned char sha1[20];
27         enum object_type type;
28         char *buf;
29         unsigned long size;
30         struct object_context obj_context;
31         struct object_info oi = {NULL};
32         struct strbuf sb = STRBUF_INIT;
33         unsigned flags = LOOKUP_REPLACE_OBJECT;
34
35         if (unknown_type)
36                 flags |= LOOKUP_UNKNOWN_OBJECT;
37
38         if (get_sha1_with_context(obj_name, 0, sha1, &obj_context))
39                 die("Not a valid object name %s", obj_name);
40
41         buf = NULL;
42         switch (opt) {
43         case 't':
44                 oi.typename = &sb;
45                 if (sha1_object_info_extended(sha1, &oi, flags) < 0)
46                         die("git cat-file: could not get object info");
47                 if (sb.len) {
48                         printf("%s\n", sb.buf);
49                         strbuf_release(&sb);
50                         return 0;
51                 }
52                 break;
53
54         case 's':
55                 oi.sizep = &size;
56                 if (sha1_object_info_extended(sha1, &oi, flags) < 0)
57                         die("git cat-file: could not get object info");
58                 printf("%lu\n", size);
59                 return 0;
60
61         case 'e':
62                 return !has_sha1_file(sha1);
63
64         case 'c':
65                 if (!obj_context.path[0])
66                         die("git cat-file --textconv %s: <object> must be <sha1:path>",
67                             obj_name);
68
69                 if (textconv_object(obj_context.path, obj_context.mode, sha1, 1, &buf, &size))
70                         break;
71
72         case 'p':
73                 type = sha1_object_info(sha1, NULL);
74                 if (type < 0)
75                         die("Not a valid object name %s", obj_name);
76
77                 /* custom pretty-print here */
78                 if (type == OBJ_TREE) {
79                         const char *ls_args[3] = { NULL };
80                         ls_args[0] =  "ls-tree";
81                         ls_args[1] =  obj_name;
82                         return cmd_ls_tree(2, ls_args, NULL);
83                 }
84
85                 if (type == OBJ_BLOB)
86                         return stream_blob_to_fd(1, sha1, NULL, 0);
87                 buf = read_sha1_file(sha1, &type, &size);
88                 if (!buf)
89                         die("Cannot read object %s", obj_name);
90
91                 /* otherwise just spit out the data */
92                 break;
93
94         case 0:
95                 if (type_from_string(exp_type) == OBJ_BLOB) {
96                         unsigned char blob_sha1[20];
97                         if (sha1_object_info(sha1, NULL) == OBJ_TAG) {
98                                 char *buffer = read_sha1_file(sha1, &type, &size);
99                                 const char *target;
100                                 if (!skip_prefix(buffer, "object ", &target) ||
101                                     get_sha1_hex(target, blob_sha1))
102                                         die("%s not a valid tag", sha1_to_hex(sha1));
103                                 free(buffer);
104                         } else
105                                 hashcpy(blob_sha1, sha1);
106
107                         if (sha1_object_info(blob_sha1, NULL) == OBJ_BLOB)
108                                 return stream_blob_to_fd(1, blob_sha1, NULL, 0);
109                         /*
110                          * we attempted to dereference a tag to a blob
111                          * and failed; there may be new dereference
112                          * mechanisms this code is not aware of.
113                          * fall-back to the usual case.
114                          */
115                 }
116                 buf = read_object_with_reference(sha1, exp_type, &size, NULL);
117                 break;
118
119         default:
120                 die("git cat-file: unknown option: %s", exp_type);
121         }
122
123         if (!buf)
124                 die("git cat-file %s: bad file", obj_name);
125
126         write_or_die(1, buf, size);
127         return 0;
128 }
129
130 struct expand_data {
131         unsigned char sha1[20];
132         enum object_type type;
133         unsigned long size;
134         off_t disk_size;
135         const char *rest;
136         unsigned char delta_base_sha1[20];
137
138         /*
139          * If mark_query is true, we do not expand anything, but rather
140          * just mark the object_info with items we wish to query.
141          */
142         int mark_query;
143
144         /*
145          * Whether to split the input on whitespace before feeding it to
146          * get_sha1; this is decided during the mark_query phase based on
147          * whether we have a %(rest) token in our format.
148          */
149         int split_on_whitespace;
150
151         /*
152          * After a mark_query run, this object_info is set up to be
153          * passed to sha1_object_info_extended. It will point to the data
154          * elements above, so you can retrieve the response from there.
155          */
156         struct object_info info;
157
158         /*
159          * This flag will be true if the requested batch format and options
160          * don't require us to call sha1_object_info, which can then be
161          * optimized out.
162          */
163         unsigned skip_object_info : 1;
164 };
165
166 static int is_atom(const char *atom, const char *s, int slen)
167 {
168         int alen = strlen(atom);
169         return alen == slen && !memcmp(atom, s, alen);
170 }
171
172 static void expand_atom(struct strbuf *sb, const char *atom, int len,
173                         void *vdata)
174 {
175         struct expand_data *data = vdata;
176
177         if (is_atom("objectname", atom, len)) {
178                 if (!data->mark_query)
179                         strbuf_addstr(sb, sha1_to_hex(data->sha1));
180         } else if (is_atom("objecttype", atom, len)) {
181                 if (data->mark_query)
182                         data->info.typep = &data->type;
183                 else
184                         strbuf_addstr(sb, typename(data->type));
185         } else if (is_atom("objectsize", atom, len)) {
186                 if (data->mark_query)
187                         data->info.sizep = &data->size;
188                 else
189                         strbuf_addf(sb, "%lu", data->size);
190         } else if (is_atom("objectsize:disk", atom, len)) {
191                 if (data->mark_query)
192                         data->info.disk_sizep = &data->disk_size;
193                 else
194                         strbuf_addf(sb, "%"PRIuMAX, (uintmax_t)data->disk_size);
195         } else if (is_atom("rest", atom, len)) {
196                 if (data->mark_query)
197                         data->split_on_whitespace = 1;
198                 else if (data->rest)
199                         strbuf_addstr(sb, data->rest);
200         } else if (is_atom("deltabase", atom, len)) {
201                 if (data->mark_query)
202                         data->info.delta_base_sha1 = data->delta_base_sha1;
203                 else
204                         strbuf_addstr(sb, sha1_to_hex(data->delta_base_sha1));
205         } else
206                 die("unknown format element: %.*s", len, atom);
207 }
208
209 static size_t expand_format(struct strbuf *sb, const char *start, void *data)
210 {
211         const char *end;
212
213         if (*start != '(')
214                 return 0;
215         end = strchr(start + 1, ')');
216         if (!end)
217                 die("format element '%s' does not end in ')'", start);
218
219         expand_atom(sb, start + 1, end - start - 1, data);
220
221         return end - start + 1;
222 }
223
224 static void batch_write(struct batch_options *opt, const void *data, int len)
225 {
226         if (opt->buffer_output) {
227                 if (fwrite(data, 1, len, stdout) != len)
228                         die_errno("unable to write to stdout");
229         } else
230                 write_or_die(1, data, len);
231 }
232
233 static void print_object_or_die(struct batch_options *opt, struct expand_data *data)
234 {
235         const unsigned char *sha1 = data->sha1;
236
237         assert(data->info.typep);
238
239         if (data->type == OBJ_BLOB) {
240                 if (opt->buffer_output)
241                         fflush(stdout);
242                 if (stream_blob_to_fd(1, sha1, NULL, 0) < 0)
243                         die("unable to stream %s to stdout", sha1_to_hex(sha1));
244         }
245         else {
246                 enum object_type type;
247                 unsigned long size;
248                 void *contents;
249
250                 contents = read_sha1_file(sha1, &type, &size);
251                 if (!contents)
252                         die("object %s disappeared", sha1_to_hex(sha1));
253                 if (type != data->type)
254                         die("object %s changed type!?", sha1_to_hex(sha1));
255                 if (data->info.sizep && size != data->size)
256                         die("object %s changed size!?", sha1_to_hex(sha1));
257
258                 batch_write(opt, contents, size);
259                 free(contents);
260         }
261 }
262
263 static void batch_object_write(const char *obj_name, struct batch_options *opt,
264                                struct expand_data *data)
265 {
266         struct strbuf buf = STRBUF_INIT;
267
268         if (!data->skip_object_info &&
269             sha1_object_info_extended(data->sha1, &data->info, LOOKUP_REPLACE_OBJECT) < 0) {
270                 printf("%s missing\n", obj_name ? obj_name : sha1_to_hex(data->sha1));
271                 fflush(stdout);
272                 return;
273         }
274
275         strbuf_expand(&buf, opt->format, expand_format, data);
276         strbuf_addch(&buf, '\n');
277         batch_write(opt, buf.buf, buf.len);
278         strbuf_release(&buf);
279
280         if (opt->print_contents) {
281                 print_object_or_die(opt, data);
282                 batch_write(opt, "\n", 1);
283         }
284 }
285
286 static void batch_one_object(const char *obj_name, struct batch_options *opt,
287                              struct expand_data *data)
288 {
289         struct object_context ctx;
290         int flags = opt->follow_symlinks ? GET_SHA1_FOLLOW_SYMLINKS : 0;
291         enum follow_symlinks_result result;
292
293         result = get_sha1_with_context(obj_name, flags, data->sha1, &ctx);
294         if (result != FOUND) {
295                 switch (result) {
296                 case MISSING_OBJECT:
297                         printf("%s missing\n", obj_name);
298                         break;
299                 case DANGLING_SYMLINK:
300                         printf("dangling %"PRIuMAX"\n%s\n",
301                                (uintmax_t)strlen(obj_name), obj_name);
302                         break;
303                 case SYMLINK_LOOP:
304                         printf("loop %"PRIuMAX"\n%s\n",
305                                (uintmax_t)strlen(obj_name), obj_name);
306                         break;
307                 case NOT_DIR:
308                         printf("notdir %"PRIuMAX"\n%s\n",
309                                (uintmax_t)strlen(obj_name), obj_name);
310                         break;
311                 default:
312                         die("BUG: unknown get_sha1_with_context result %d\n",
313                                result);
314                         break;
315                 }
316                 fflush(stdout);
317                 return;
318         }
319
320         if (ctx.mode == 0) {
321                 printf("symlink %"PRIuMAX"\n%s\n",
322                        (uintmax_t)ctx.symlink_path.len,
323                        ctx.symlink_path.buf);
324                 fflush(stdout);
325                 return;
326         }
327
328         batch_object_write(obj_name, opt, data);
329 }
330
331 struct object_cb_data {
332         struct batch_options *opt;
333         struct expand_data *expand;
334 };
335
336 static void batch_object_cb(const unsigned char sha1[20], void *vdata)
337 {
338         struct object_cb_data *data = vdata;
339         hashcpy(data->expand->sha1, sha1);
340         batch_object_write(NULL, data->opt, data->expand);
341 }
342
343 static int batch_loose_object(const unsigned char *sha1,
344                               const char *path,
345                               void *data)
346 {
347         sha1_array_append(data, sha1);
348         return 0;
349 }
350
351 static int batch_packed_object(const unsigned char *sha1,
352                                struct packed_git *pack,
353                                uint32_t pos,
354                                void *data)
355 {
356         sha1_array_append(data, sha1);
357         return 0;
358 }
359
360 static int batch_objects(struct batch_options *opt)
361 {
362         struct strbuf buf = STRBUF_INIT;
363         struct expand_data data;
364         int save_warning;
365         int retval = 0;
366
367         if (!opt->format)
368                 opt->format = "%(objectname) %(objecttype) %(objectsize)";
369
370         /*
371          * Expand once with our special mark_query flag, which will prime the
372          * object_info to be handed to sha1_object_info_extended for each
373          * object.
374          */
375         memset(&data, 0, sizeof(data));
376         data.mark_query = 1;
377         strbuf_expand(&buf, opt->format, expand_format, &data);
378         data.mark_query = 0;
379
380         if (opt->all_objects) {
381                 struct object_info empty;
382                 memset(&empty, 0, sizeof(empty));
383                 if (!memcmp(&data.info, &empty, sizeof(empty)))
384                         data.skip_object_info = 1;
385         }
386
387         /*
388          * If we are printing out the object, then always fill in the type,
389          * since we will want to decide whether or not to stream.
390          */
391         if (opt->print_contents)
392                 data.info.typep = &data.type;
393
394         if (opt->all_objects) {
395                 struct sha1_array sa = SHA1_ARRAY_INIT;
396                 struct object_cb_data cb;
397
398                 for_each_loose_object(batch_loose_object, &sa, 0);
399                 for_each_packed_object(batch_packed_object, &sa, 0);
400
401                 cb.opt = opt;
402                 cb.expand = &data;
403                 sha1_array_for_each_unique(&sa, batch_object_cb, &cb);
404
405                 sha1_array_clear(&sa);
406                 return 0;
407         }
408
409         /*
410          * We are going to call get_sha1 on a potentially very large number of
411          * objects. In most large cases, these will be actual object sha1s. The
412          * cost to double-check that each one is not also a ref (just so we can
413          * warn) ends up dwarfing the actual cost of the object lookups
414          * themselves. We can work around it by just turning off the warning.
415          */
416         save_warning = warn_on_object_refname_ambiguity;
417         warn_on_object_refname_ambiguity = 0;
418
419         while (strbuf_getline(&buf, stdin) != EOF) {
420                 if (data.split_on_whitespace) {
421                         /*
422                          * Split at first whitespace, tying off the beginning
423                          * of the string and saving the remainder (or NULL) in
424                          * data.rest.
425                          */
426                         char *p = strpbrk(buf.buf, " \t");
427                         if (p) {
428                                 while (*p && strchr(" \t", *p))
429                                         *p++ = '\0';
430                         }
431                         data.rest = p;
432                 }
433
434                 batch_one_object(buf.buf, opt, &data);
435         }
436
437         strbuf_release(&buf);
438         warn_on_object_refname_ambiguity = save_warning;
439         return retval;
440 }
441
442 static const char * const cat_file_usage[] = {
443         N_("git cat-file (-t [--allow-unknown-type] | -s [--allow-unknown-type] | -e | -p | <type> | --textconv) <object>"),
444         N_("git cat-file (--batch | --batch-check) [--follow-symlinks]"),
445         NULL
446 };
447
448 static int git_cat_file_config(const char *var, const char *value, void *cb)
449 {
450         if (userdiff_config(var, value) < 0)
451                 return -1;
452
453         return git_default_config(var, value, cb);
454 }
455
456 static int batch_option_callback(const struct option *opt,
457                                  const char *arg,
458                                  int unset)
459 {
460         struct batch_options *bo = opt->value;
461
462         if (bo->enabled) {
463                 return 1;
464         }
465
466         bo->enabled = 1;
467         bo->print_contents = !strcmp(opt->long_name, "batch");
468         bo->format = arg;
469
470         return 0;
471 }
472
473 int cmd_cat_file(int argc, const char **argv, const char *prefix)
474 {
475         int opt = 0;
476         const char *exp_type = NULL, *obj_name = NULL;
477         struct batch_options batch = {0};
478         int unknown_type = 0;
479
480         const struct option options[] = {
481                 OPT_GROUP(N_("<type> can be one of: blob, tree, commit, tag")),
482                 OPT_CMDMODE('t', NULL, &opt, N_("show object type"), 't'),
483                 OPT_CMDMODE('s', NULL, &opt, N_("show object size"), 's'),
484                 OPT_CMDMODE('e', NULL, &opt,
485                             N_("exit with zero when there's no error"), 'e'),
486                 OPT_CMDMODE('p', NULL, &opt, N_("pretty-print object's content"), 'p'),
487                 OPT_CMDMODE(0, "textconv", &opt,
488                             N_("for blob objects, run textconv on object's content"), 'c'),
489                 OPT_BOOL(0, "allow-unknown-type", &unknown_type,
490                           N_("allow -s and -t to work with broken/corrupt objects")),
491                 OPT_BOOL(0, "buffer", &batch.buffer_output, N_("buffer --batch output")),
492                 { OPTION_CALLBACK, 0, "batch", &batch, "format",
493                         N_("show info and content of objects fed from the standard input"),
494                         PARSE_OPT_OPTARG, batch_option_callback },
495                 { OPTION_CALLBACK, 0, "batch-check", &batch, "format",
496                         N_("show info about objects fed from the standard input"),
497                         PARSE_OPT_OPTARG, batch_option_callback },
498                 OPT_BOOL(0, "follow-symlinks", &batch.follow_symlinks,
499                          N_("follow in-tree symlinks (used with --batch or --batch-check)")),
500                 OPT_BOOL(0, "batch-all-objects", &batch.all_objects,
501                          N_("show all objects with --batch or --batch-check")),
502                 OPT_END()
503         };
504
505         git_config(git_cat_file_config, NULL);
506
507         batch.buffer_output = -1;
508         argc = parse_options(argc, argv, prefix, options, cat_file_usage, 0);
509
510         if (opt) {
511                 if (argc == 1)
512                         obj_name = argv[0];
513                 else
514                         usage_with_options(cat_file_usage, options);
515         }
516         if (!opt && !batch.enabled) {
517                 if (argc == 2) {
518                         exp_type = argv[0];
519                         obj_name = argv[1];
520                 } else
521                         usage_with_options(cat_file_usage, options);
522         }
523         if (batch.enabled && (opt || argc)) {
524                 usage_with_options(cat_file_usage, options);
525         }
526
527         if ((batch.follow_symlinks || batch.all_objects) && !batch.enabled) {
528                 usage_with_options(cat_file_usage, options);
529         }
530
531         if (batch.buffer_output < 0)
532                 batch.buffer_output = batch.all_objects;
533
534         if (batch.enabled)
535                 return batch_objects(&batch);
536
537         if (unknown_type && opt != 't' && opt != 's')
538                 die("git cat-file --allow-unknown-type: use with -s or -t");
539         return cat_one_file(opt, exp_type, obj_name, unknown_type);
540 }