Merge branch 'sb/perf-without-installed-git' 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         unsigned long 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 static int is_atom(const char *atom, const char *s, int slen)
160 {
161         int alen = strlen(atom);
162         return alen == slen && !memcmp(atom, s, alen);
163 }
164
165 static void expand_atom(struct strbuf *sb, const char *atom, int len,
166                         void *vdata)
167 {
168         struct expand_data *data = vdata;
169
170         if (is_atom("objectname", atom, len)) {
171                 if (!data->mark_query)
172                         strbuf_addstr(sb, sha1_to_hex(data->sha1));
173         } else if (is_atom("objecttype", atom, len)) {
174                 if (data->mark_query)
175                         data->info.typep = &data->type;
176                 else
177                         strbuf_addstr(sb, typename(data->type));
178         } else if (is_atom("objectsize", atom, len)) {
179                 if (data->mark_query)
180                         data->info.sizep = &data->size;
181                 else
182                         strbuf_addf(sb, "%lu", data->size);
183         } else if (is_atom("objectsize:disk", atom, len)) {
184                 if (data->mark_query)
185                         data->info.disk_sizep = &data->disk_size;
186                 else
187                         strbuf_addf(sb, "%lu", data->disk_size);
188         } else if (is_atom("rest", atom, len)) {
189                 if (data->mark_query)
190                         data->split_on_whitespace = 1;
191                 else if (data->rest)
192                         strbuf_addstr(sb, data->rest);
193         } else if (is_atom("deltabase", atom, len)) {
194                 if (data->mark_query)
195                         data->info.delta_base_sha1 = data->delta_base_sha1;
196                 else
197                         strbuf_addstr(sb, sha1_to_hex(data->delta_base_sha1));
198         } else
199                 die("unknown format element: %.*s", len, atom);
200 }
201
202 static size_t expand_format(struct strbuf *sb, const char *start, void *data)
203 {
204         const char *end;
205
206         if (*start != '(')
207                 return 0;
208         end = strchr(start + 1, ')');
209         if (!end)
210                 die("format element '%s' does not end in ')'", start);
211
212         expand_atom(sb, start + 1, end - start - 1, data);
213
214         return end - start + 1;
215 }
216
217 static void batch_write(struct batch_options *opt, const void *data, int len)
218 {
219         if (opt->buffer_output) {
220                 if (fwrite(data, 1, len, stdout) != len)
221                         die_errno("unable to write to stdout");
222         } else
223                 write_or_die(1, data, len);
224 }
225
226 static void print_object_or_die(struct batch_options *opt, struct expand_data *data)
227 {
228         const unsigned char *sha1 = data->sha1;
229
230         assert(data->info.typep);
231
232         if (data->type == OBJ_BLOB) {
233                 if (opt->buffer_output)
234                         fflush(stdout);
235                 if (stream_blob_to_fd(1, sha1, NULL, 0) < 0)
236                         die("unable to stream %s to stdout", sha1_to_hex(sha1));
237         }
238         else {
239                 enum object_type type;
240                 unsigned long size;
241                 void *contents;
242
243                 contents = read_sha1_file(sha1, &type, &size);
244                 if (!contents)
245                         die("object %s disappeared", sha1_to_hex(sha1));
246                 if (type != data->type)
247                         die("object %s changed type!?", sha1_to_hex(sha1));
248                 if (data->info.sizep && size != data->size)
249                         die("object %s changed size!?", sha1_to_hex(sha1));
250
251                 batch_write(opt, contents, size);
252                 free(contents);
253         }
254 }
255
256 static void batch_object_write(const char *obj_name, struct batch_options *opt,
257                                struct expand_data *data)
258 {
259         struct strbuf buf = STRBUF_INIT;
260
261         if (sha1_object_info_extended(data->sha1, &data->info, LOOKUP_REPLACE_OBJECT) < 0) {
262                 printf("%s missing\n", obj_name ? obj_name : sha1_to_hex(data->sha1));
263                 fflush(stdout);
264                 return;
265         }
266
267         strbuf_expand(&buf, opt->format, expand_format, data);
268         strbuf_addch(&buf, '\n');
269         batch_write(opt, buf.buf, buf.len);
270         strbuf_release(&buf);
271
272         if (opt->print_contents) {
273                 print_object_or_die(opt, data);
274                 batch_write(opt, "\n", 1);
275         }
276 }
277
278 static void batch_one_object(const char *obj_name, struct batch_options *opt,
279                              struct expand_data *data)
280 {
281         struct object_context ctx;
282         int flags = opt->follow_symlinks ? GET_SHA1_FOLLOW_SYMLINKS : 0;
283         enum follow_symlinks_result result;
284
285         result = get_sha1_with_context(obj_name, flags, data->sha1, &ctx);
286         if (result != FOUND) {
287                 switch (result) {
288                 case MISSING_OBJECT:
289                         printf("%s missing\n", obj_name);
290                         break;
291                 case DANGLING_SYMLINK:
292                         printf("dangling %"PRIuMAX"\n%s\n",
293                                (uintmax_t)strlen(obj_name), obj_name);
294                         break;
295                 case SYMLINK_LOOP:
296                         printf("loop %"PRIuMAX"\n%s\n",
297                                (uintmax_t)strlen(obj_name), obj_name);
298                         break;
299                 case NOT_DIR:
300                         printf("notdir %"PRIuMAX"\n%s\n",
301                                (uintmax_t)strlen(obj_name), obj_name);
302                         break;
303                 default:
304                         die("BUG: unknown get_sha1_with_context result %d\n",
305                                result);
306                         break;
307                 }
308                 fflush(stdout);
309                 return;
310         }
311
312         if (ctx.mode == 0) {
313                 printf("symlink %"PRIuMAX"\n%s\n",
314                        (uintmax_t)ctx.symlink_path.len,
315                        ctx.symlink_path.buf);
316                 fflush(stdout);
317                 return;
318         }
319
320         batch_object_write(obj_name, opt, data);
321 }
322
323 struct object_cb_data {
324         struct batch_options *opt;
325         struct expand_data *expand;
326 };
327
328 static void batch_object_cb(const unsigned char sha1[20], void *vdata)
329 {
330         struct object_cb_data *data = vdata;
331         hashcpy(data->expand->sha1, sha1);
332         batch_object_write(NULL, data->opt, data->expand);
333 }
334
335 static int batch_loose_object(const unsigned char *sha1,
336                               const char *path,
337                               void *data)
338 {
339         sha1_array_append(data, sha1);
340         return 0;
341 }
342
343 static int batch_packed_object(const unsigned char *sha1,
344                                struct packed_git *pack,
345                                uint32_t pos,
346                                void *data)
347 {
348         sha1_array_append(data, sha1);
349         return 0;
350 }
351
352 static int batch_objects(struct batch_options *opt)
353 {
354         struct strbuf buf = STRBUF_INIT;
355         struct expand_data data;
356         int save_warning;
357         int retval = 0;
358
359         if (!opt->format)
360                 opt->format = "%(objectname) %(objecttype) %(objectsize)";
361
362         /*
363          * Expand once with our special mark_query flag, which will prime the
364          * object_info to be handed to sha1_object_info_extended for each
365          * object.
366          */
367         memset(&data, 0, sizeof(data));
368         data.mark_query = 1;
369         strbuf_expand(&buf, opt->format, expand_format, &data);
370         data.mark_query = 0;
371
372         /*
373          * If we are printing out the object, then always fill in the type,
374          * since we will want to decide whether or not to stream.
375          */
376         if (opt->print_contents)
377                 data.info.typep = &data.type;
378
379         if (opt->all_objects) {
380                 struct sha1_array sa = SHA1_ARRAY_INIT;
381                 struct object_cb_data cb;
382
383                 for_each_loose_object(batch_loose_object, &sa, 0);
384                 for_each_packed_object(batch_packed_object, &sa, 0);
385
386                 cb.opt = opt;
387                 cb.expand = &data;
388                 sha1_array_for_each_unique(&sa, batch_object_cb, &cb);
389
390                 sha1_array_clear(&sa);
391                 return 0;
392         }
393
394         /*
395          * We are going to call get_sha1 on a potentially very large number of
396          * objects. In most large cases, these will be actual object sha1s. The
397          * cost to double-check that each one is not also a ref (just so we can
398          * warn) ends up dwarfing the actual cost of the object lookups
399          * themselves. We can work around it by just turning off the warning.
400          */
401         save_warning = warn_on_object_refname_ambiguity;
402         warn_on_object_refname_ambiguity = 0;
403
404         while (strbuf_getline(&buf, stdin, '\n') != EOF) {
405                 if (data.split_on_whitespace) {
406                         /*
407                          * Split at first whitespace, tying off the beginning
408                          * of the string and saving the remainder (or NULL) in
409                          * data.rest.
410                          */
411                         char *p = strpbrk(buf.buf, " \t");
412                         if (p) {
413                                 while (*p && strchr(" \t", *p))
414                                         *p++ = '\0';
415                         }
416                         data.rest = p;
417                 }
418
419                 batch_one_object(buf.buf, opt, &data);
420         }
421
422         strbuf_release(&buf);
423         warn_on_object_refname_ambiguity = save_warning;
424         return retval;
425 }
426
427 static const char * const cat_file_usage[] = {
428         N_("git cat-file (-t [--allow-unknown-type]|-s [--allow-unknown-type]|-e|-p|<type>|--textconv) <object>"),
429         N_("git cat-file (--batch | --batch-check) [--follow-symlinks] < <list-of-objects>"),
430         NULL
431 };
432
433 static int git_cat_file_config(const char *var, const char *value, void *cb)
434 {
435         if (userdiff_config(var, value) < 0)
436                 return -1;
437
438         return git_default_config(var, value, cb);
439 }
440
441 static int batch_option_callback(const struct option *opt,
442                                  const char *arg,
443                                  int unset)
444 {
445         struct batch_options *bo = opt->value;
446
447         if (bo->enabled) {
448                 return 1;
449         }
450
451         bo->enabled = 1;
452         bo->print_contents = !strcmp(opt->long_name, "batch");
453         bo->format = arg;
454
455         return 0;
456 }
457
458 int cmd_cat_file(int argc, const char **argv, const char *prefix)
459 {
460         int opt = 0;
461         const char *exp_type = NULL, *obj_name = NULL;
462         struct batch_options batch = {0};
463         int unknown_type = 0;
464
465         const struct option options[] = {
466                 OPT_GROUP(N_("<type> can be one of: blob, tree, commit, tag")),
467                 OPT_CMDMODE('t', NULL, &opt, N_("show object type"), 't'),
468                 OPT_CMDMODE('s', NULL, &opt, N_("show object size"), 's'),
469                 OPT_CMDMODE('e', NULL, &opt,
470                             N_("exit with zero when there's no error"), 'e'),
471                 OPT_CMDMODE('p', NULL, &opt, N_("pretty-print object's content"), 'p'),
472                 OPT_CMDMODE(0, "textconv", &opt,
473                             N_("for blob objects, run textconv on object's content"), 'c'),
474                 OPT_BOOL(0, "allow-unknown-type", &unknown_type,
475                           N_("allow -s and -t to work with broken/corrupt objects")),
476                 OPT_BOOL(0, "buffer", &batch.buffer_output, N_("buffer --batch output")),
477                 { OPTION_CALLBACK, 0, "batch", &batch, "format",
478                         N_("show info and content of objects fed from the standard input"),
479                         PARSE_OPT_OPTARG, batch_option_callback },
480                 { OPTION_CALLBACK, 0, "batch-check", &batch, "format",
481                         N_("show info about objects fed from the standard input"),
482                         PARSE_OPT_OPTARG, batch_option_callback },
483                 OPT_BOOL(0, "follow-symlinks", &batch.follow_symlinks,
484                          N_("follow in-tree symlinks (used with --batch or --batch-check)")),
485                 OPT_BOOL(0, "batch-all-objects", &batch.all_objects,
486                          N_("show all objects with --batch or --batch-check")),
487                 OPT_END()
488         };
489
490         git_config(git_cat_file_config, NULL);
491
492         argc = parse_options(argc, argv, prefix, options, cat_file_usage, 0);
493
494         if (opt) {
495                 if (argc == 1)
496                         obj_name = argv[0];
497                 else
498                         usage_with_options(cat_file_usage, options);
499         }
500         if (!opt && !batch.enabled) {
501                 if (argc == 2) {
502                         exp_type = argv[0];
503                         obj_name = argv[1];
504                 } else
505                         usage_with_options(cat_file_usage, options);
506         }
507         if (batch.enabled && (opt || argc)) {
508                 usage_with_options(cat_file_usage, options);
509         }
510
511         if ((batch.follow_symlinks || batch.all_objects) && !batch.enabled) {
512                 usage_with_options(cat_file_usage, options);
513         }
514
515         if (batch.enabled)
516                 return batch_objects(&batch);
517
518         if (unknown_type && opt != 't' && opt != 's')
519                 die("git cat-file --allow-unknown-type: use with -s or -t");
520         return cat_one_file(opt, exp_type, obj_name, unknown_type);
521 }