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