Sync with 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 "exec_cmd.h"
8 #include "tag.h"
9 #include "tree.h"
10 #include "builtin.h"
11 #include "parse-options.h"
12 #include "diff.h"
13 #include "userdiff.h"
14 #include "streaming.h"
15
16 static int cat_one_file(int opt, const char *exp_type, const char *obj_name)
17 {
18         unsigned char sha1[20];
19         enum object_type type;
20         char *buf;
21         unsigned long size;
22         struct object_context obj_context;
23
24         if (get_sha1_with_context(obj_name, 0, sha1, &obj_context))
25                 die("Not a valid object name %s", obj_name);
26
27         buf = NULL;
28         switch (opt) {
29         case 't':
30                 type = sha1_object_info(sha1, NULL);
31                 if (type > 0) {
32                         printf("%s\n", typename(type));
33                         return 0;
34                 }
35                 break;
36
37         case 's':
38                 type = sha1_object_info(sha1, &size);
39                 if (type > 0) {
40                         printf("%lu\n", size);
41                         return 0;
42                 }
43                 break;
44
45         case 'e':
46                 return !has_sha1_file(sha1);
47
48         case 'c':
49                 if (!obj_context.path[0])
50                         die("git cat-file --textconv %s: <object> must be <sha1:path>",
51                             obj_name);
52
53                 if (textconv_object(obj_context.path, obj_context.mode, sha1, 1, &buf, &size))
54                         break;
55
56         case 'p':
57                 type = sha1_object_info(sha1, NULL);
58                 if (type < 0)
59                         die("Not a valid object name %s", obj_name);
60
61                 /* custom pretty-print here */
62                 if (type == OBJ_TREE) {
63                         const char *ls_args[3] = { NULL };
64                         ls_args[0] =  "ls-tree";
65                         ls_args[1] =  obj_name;
66                         return cmd_ls_tree(2, ls_args, NULL);
67                 }
68
69                 if (type == OBJ_BLOB)
70                         return stream_blob_to_fd(1, sha1, NULL, 0);
71                 buf = read_sha1_file(sha1, &type, &size);
72                 if (!buf)
73                         die("Cannot read object %s", obj_name);
74
75                 /* otherwise just spit out the data */
76                 break;
77
78         case 0:
79                 if (type_from_string(exp_type) == OBJ_BLOB) {
80                         unsigned char blob_sha1[20];
81                         if (sha1_object_info(sha1, NULL) == OBJ_TAG) {
82                                 enum object_type type;
83                                 unsigned long size;
84                                 char *buffer = read_sha1_file(sha1, &type, &size);
85                                 if (memcmp(buffer, "object ", 7) ||
86                                     get_sha1_hex(buffer + 7, blob_sha1))
87                                         die("%s not a valid tag", sha1_to_hex(sha1));
88                                 free(buffer);
89                         } else
90                                 hashcpy(blob_sha1, sha1);
91
92                         if (sha1_object_info(blob_sha1, NULL) == OBJ_BLOB)
93                                 return stream_blob_to_fd(1, blob_sha1, NULL, 0);
94                         /*
95                          * we attempted to dereference a tag to a blob
96                          * and failed; there may be new dereference
97                          * mechanisms this code is not aware of.
98                          * fall-back to the usual case.
99                          */
100                 }
101                 buf = read_object_with_reference(sha1, exp_type, &size, NULL);
102                 break;
103
104         default:
105                 die("git cat-file: unknown option: %s", exp_type);
106         }
107
108         if (!buf)
109                 die("git cat-file %s: bad file", obj_name);
110
111         write_or_die(1, buf, size);
112         return 0;
113 }
114
115 struct expand_data {
116         unsigned char sha1[20];
117         enum object_type type;
118         unsigned long size;
119         unsigned long disk_size;
120         const char *rest;
121
122         /*
123          * If mark_query is true, we do not expand anything, but rather
124          * just mark the object_info with items we wish to query.
125          */
126         int mark_query;
127
128         /*
129          * Whether to split the input on whitespace before feeding it to
130          * get_sha1; this is decided during the mark_query phase based on
131          * whether we have a %(rest) token in our format.
132          */
133         int split_on_whitespace;
134
135         /*
136          * After a mark_query run, this object_info is set up to be
137          * passed to sha1_object_info_extended. It will point to the data
138          * elements above, so you can retrieve the response from there.
139          */
140         struct object_info info;
141 };
142
143 static int is_atom(const char *atom, const char *s, int slen)
144 {
145         int alen = strlen(atom);
146         return alen == slen && !memcmp(atom, s, alen);
147 }
148
149 static void expand_atom(struct strbuf *sb, const char *atom, int len,
150                         void *vdata)
151 {
152         struct expand_data *data = vdata;
153
154         if (is_atom("objectname", atom, len)) {
155                 if (!data->mark_query)
156                         strbuf_addstr(sb, sha1_to_hex(data->sha1));
157         } else if (is_atom("objecttype", atom, len)) {
158                 if (data->mark_query)
159                         data->info.typep = &data->type;
160                 else
161                         strbuf_addstr(sb, typename(data->type));
162         } else if (is_atom("objectsize", atom, len)) {
163                 if (data->mark_query)
164                         data->info.sizep = &data->size;
165                 else
166                         strbuf_addf(sb, "%lu", data->size);
167         } else if (is_atom("objectsize:disk", atom, len)) {
168                 if (data->mark_query)
169                         data->info.disk_sizep = &data->disk_size;
170                 else
171                         strbuf_addf(sb, "%lu", data->disk_size);
172         } else if (is_atom("rest", atom, len)) {
173                 if (data->mark_query)
174                         data->split_on_whitespace = 1;
175                 else if (data->rest)
176                         strbuf_addstr(sb, data->rest);
177         } else
178                 die("unknown format element: %.*s", len, atom);
179 }
180
181 static size_t expand_format(struct strbuf *sb, const char *start, void *data)
182 {
183         const char *end;
184
185         if (*start != '(')
186                 return 0;
187         end = strchr(start + 1, ')');
188         if (!end)
189                 die("format element '%s' does not end in ')'", start);
190
191         expand_atom(sb, start + 1, end - start - 1, data);
192
193         return end - start + 1;
194 }
195
196 static void print_object_or_die(int fd, const unsigned char *sha1,
197                                 enum object_type type, unsigned long size)
198 {
199         if (type == OBJ_BLOB) {
200                 if (stream_blob_to_fd(fd, sha1, NULL, 0) < 0)
201                         die("unable to stream %s to stdout", sha1_to_hex(sha1));
202         }
203         else {
204                 enum object_type rtype;
205                 unsigned long rsize;
206                 void *contents;
207
208                 contents = read_sha1_file(sha1, &rtype, &rsize);
209                 if (!contents)
210                         die("object %s disappeared", sha1_to_hex(sha1));
211                 if (rtype != type)
212                         die("object %s changed type!?", sha1_to_hex(sha1));
213                 if (rsize != size)
214                         die("object %s change size!?", sha1_to_hex(sha1));
215
216                 write_or_die(fd, contents, size);
217                 free(contents);
218         }
219 }
220
221 struct batch_options {
222         int enabled;
223         int print_contents;
224         const char *format;
225 };
226
227 static int batch_one_object(const char *obj_name, struct batch_options *opt,
228                             struct expand_data *data)
229 {
230         struct strbuf buf = STRBUF_INIT;
231
232         if (!obj_name)
233            return 1;
234
235         if (get_sha1(obj_name, data->sha1)) {
236                 printf("%s missing\n", obj_name);
237                 fflush(stdout);
238                 return 0;
239         }
240
241         if (sha1_object_info_extended(data->sha1, &data->info) < 0) {
242                 printf("%s missing\n", obj_name);
243                 fflush(stdout);
244                 return 0;
245         }
246
247         strbuf_expand(&buf, opt->format, expand_format, data);
248         strbuf_addch(&buf, '\n');
249         write_or_die(1, buf.buf, buf.len);
250         strbuf_release(&buf);
251
252         if (opt->print_contents) {
253                 print_object_or_die(1, data->sha1, data->type, data->size);
254                 write_or_die(1, "\n", 1);
255         }
256         return 0;
257 }
258
259 static int batch_objects(struct batch_options *opt)
260 {
261         struct strbuf buf = STRBUF_INIT;
262         struct expand_data data;
263
264         if (!opt->format)
265                 opt->format = "%(objectname) %(objecttype) %(objectsize)";
266
267         /*
268          * Expand once with our special mark_query flag, which will prime the
269          * object_info to be handed to sha1_object_info_extended for each
270          * object.
271          */
272         memset(&data, 0, sizeof(data));
273         data.mark_query = 1;
274         strbuf_expand(&buf, opt->format, expand_format, &data);
275         data.mark_query = 0;
276
277         /*
278          * We are going to call get_sha1 on a potentially very large number of
279          * objects. In most large cases, these will be actual object sha1s. The
280          * cost to double-check that each one is not also a ref (just so we can
281          * warn) ends up dwarfing the actual cost of the object lookups
282          * themselves. We can work around it by just turning off the warning.
283          */
284         warn_on_object_refname_ambiguity = 0;
285
286         while (strbuf_getline(&buf, stdin, '\n') != EOF) {
287                 int error;
288
289                 if (data.split_on_whitespace) {
290                         /*
291                          * Split at first whitespace, tying off the beginning
292                          * of the string and saving the remainder (or NULL) in
293                          * data.rest.
294                          */
295                         char *p = strpbrk(buf.buf, " \t");
296                         if (p) {
297                                 while (*p && strchr(" \t", *p))
298                                         *p++ = '\0';
299                         }
300                         data.rest = p;
301                 }
302
303                 error = batch_one_object(buf.buf, opt, &data);
304                 if (error)
305                         return error;
306         }
307
308         return 0;
309 }
310
311 static const char * const cat_file_usage[] = {
312         N_("git cat-file (-t|-s|-e|-p|<type>|--textconv) <object>"),
313         N_("git cat-file (--batch|--batch-check) < <list_of_objects>"),
314         NULL
315 };
316
317 static int git_cat_file_config(const char *var, const char *value, void *cb)
318 {
319         if (userdiff_config(var, value) < 0)
320                 return -1;
321
322         return git_default_config(var, value, cb);
323 }
324
325 static int batch_option_callback(const struct option *opt,
326                                  const char *arg,
327                                  int unset)
328 {
329         struct batch_options *bo = opt->value;
330
331         if (unset) {
332                 memset(bo, 0, sizeof(*bo));
333                 return 0;
334         }
335
336         bo->enabled = 1;
337         bo->print_contents = !strcmp(opt->long_name, "batch");
338         bo->format = arg;
339
340         return 0;
341 }
342
343 int cmd_cat_file(int argc, const char **argv, const char *prefix)
344 {
345         int opt = 0;
346         const char *exp_type = NULL, *obj_name = NULL;
347         struct batch_options batch = {0};
348
349         const struct option options[] = {
350                 OPT_GROUP(N_("<type> can be one of: blob, tree, commit, tag")),
351                 OPT_SET_INT('t', NULL, &opt, N_("show object type"), 't'),
352                 OPT_SET_INT('s', NULL, &opt, N_("show object size"), 's'),
353                 OPT_SET_INT('e', NULL, &opt,
354                             N_("exit with zero when there's no error"), 'e'),
355                 OPT_SET_INT('p', NULL, &opt, N_("pretty-print object's content"), 'p'),
356                 OPT_SET_INT(0, "textconv", &opt,
357                             N_("for blob objects, run textconv on object's content"), 'c'),
358                 { OPTION_CALLBACK, 0, "batch", &batch, "format",
359                         N_("show info and content of objects fed from the standard input"),
360                         PARSE_OPT_OPTARG, batch_option_callback },
361                 { OPTION_CALLBACK, 0, "batch-check", &batch, "format",
362                         N_("show info about objects fed from the standard input"),
363                         PARSE_OPT_OPTARG, batch_option_callback },
364                 OPT_END()
365         };
366
367         git_config(git_cat_file_config, NULL);
368
369         if (argc != 3 && argc != 2)
370                 usage_with_options(cat_file_usage, options);
371
372         argc = parse_options(argc, argv, prefix, options, cat_file_usage, 0);
373
374         if (opt) {
375                 if (argc == 1)
376                         obj_name = argv[0];
377                 else
378                         usage_with_options(cat_file_usage, options);
379         }
380         if (!opt && !batch.enabled) {
381                 if (argc == 2) {
382                         exp_type = argv[0];
383                         obj_name = argv[1];
384                 } else
385                         usage_with_options(cat_file_usage, options);
386         }
387         if (batch.enabled && (opt || argc)) {
388                 usage_with_options(cat_file_usage, options);
389         }
390
391         if (batch.enabled)
392                 return batch_objects(&batch);
393
394         return cat_one_file(opt, exp_type, obj_name);
395 }