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
12static int cat_one_file(int opt, const char *exp_type, const char *obj_name,
13 int unknown_type)
14{
15 unsigned char sha1[20];
16 enum object_type type;
17 char *buf;
18 unsigned long size;
19 struct object_context obj_context;
20 struct object_info oi = {NULL};
21 struct strbuf sb = STRBUF_INIT;
22 unsigned flags = LOOKUP_REPLACE_OBJECT;
23
24 if (unknown_type)
25 flags |= LOOKUP_UNKNOWN_OBJECT;
26
27 if (get_sha1_with_context(obj_name, 0, sha1, &obj_context))
28 die("Not a valid object name %s", obj_name);
29
30 buf = NULL;
31 switch (opt) {
32 case 't':
33 oi.typename = &sb;
34 if (sha1_object_info_extended(sha1, &oi, flags) < 0)
35 die("git cat-file: could not get object info");
36 if (sb.len) {
37 printf("%s\n", sb.buf);
38 strbuf_release(&sb);
39 return 0;
40 }
41 break;
42
43 case 's':
44 oi.sizep = &size;
45 if (sha1_object_info_extended(sha1, &oi, flags) < 0)
46 die("git cat-file: could not get object info");
47 printf("%lu\n", size);
48 return 0;
49
50 case 'e':
51 return !has_sha1_file(sha1);
52
53 case 'c':
54 if (!obj_context.path[0])
55 die("git cat-file --textconv %s: <object> must be <sha1:path>",
56 obj_name);
57
58 if (textconv_object(obj_context.path, obj_context.mode, sha1, 1, &buf, &size))
59 break;
60
61 case 'p':
62 type = sha1_object_info(sha1, NULL);
63 if (type < 0)
64 die("Not a valid object name %s", obj_name);
65
66 /* custom pretty-print here */
67 if (type == OBJ_TREE) {
68 const char *ls_args[3] = { NULL };
69 ls_args[0] = "ls-tree";
70 ls_args[1] = obj_name;
71 return cmd_ls_tree(2, ls_args, NULL);
72 }
73
74 if (type == OBJ_BLOB)
75 return stream_blob_to_fd(1, sha1, NULL, 0);
76 buf = read_sha1_file(sha1, &type, &size);
77 if (!buf)
78 die("Cannot read object %s", obj_name);
79
80 /* otherwise just spit out the data */
81 break;
82
83 case 0:
84 if (type_from_string(exp_type) == OBJ_BLOB) {
85 unsigned char blob_sha1[20];
86 if (sha1_object_info(sha1, NULL) == OBJ_TAG) {
87 char *buffer = read_sha1_file(sha1, &type, &size);
88 const char *target;
89 if (!skip_prefix(buffer, "object ", &target) ||
90 get_sha1_hex(target, blob_sha1))
91 die("%s not a valid tag", sha1_to_hex(sha1));
92 free(buffer);
93 } else
94 hashcpy(blob_sha1, sha1);
95
96 if (sha1_object_info(blob_sha1, NULL) == OBJ_BLOB)
97 return stream_blob_to_fd(1, blob_sha1, NULL, 0);
98 /*
99 * we attempted to dereference a tag to a blob
100 * and failed; there may be new dereference
101 * mechanisms this code is not aware of.
102 * fall-back to the usual case.
103 */
104 }
105 buf = read_object_with_reference(sha1, exp_type, &size, NULL);
106 break;
107
108 default:
109 die("git cat-file: unknown option: %s", exp_type);
110 }
111
112 if (!buf)
113 die("git cat-file %s: bad file", obj_name);
114
115 write_or_die(1, buf, size);
116 return 0;
117}
118
119struct expand_data {
120 unsigned char sha1[20];
121 enum object_type type;
122 unsigned long size;
123 unsigned long disk_size;
124 const char *rest;
125 unsigned char delta_base_sha1[20];
126
127 /*
128 * If mark_query is true, we do not expand anything, but rather
129 * just mark the object_info with items we wish to query.
130 */
131 int mark_query;
132
133 /*
134 * Whether to split the input on whitespace before feeding it to
135 * get_sha1; this is decided during the mark_query phase based on
136 * whether we have a %(rest) token in our format.
137 */
138 int split_on_whitespace;
139
140 /*
141 * After a mark_query run, this object_info is set up to be
142 * passed to sha1_object_info_extended. It will point to the data
143 * elements above, so you can retrieve the response from there.
144 */
145 struct object_info info;
146};
147
148static int is_atom(const char *atom, const char *s, int slen)
149{
150 int alen = strlen(atom);
151 return alen == slen && !memcmp(atom, s, alen);
152}
153
154static void expand_atom(struct strbuf *sb, const char *atom, int len,
155 void *vdata)
156{
157 struct expand_data *data = vdata;
158
159 if (is_atom("objectname", atom, len)) {
160 if (!data->mark_query)
161 strbuf_addstr(sb, sha1_to_hex(data->sha1));
162 } else if (is_atom("objecttype", atom, len)) {
163 if (data->mark_query)
164 data->info.typep = &data->type;
165 else
166 strbuf_addstr(sb, typename(data->type));
167 } else if (is_atom("objectsize", atom, len)) {
168 if (data->mark_query)
169 data->info.sizep = &data->size;
170 else
171 strbuf_addf(sb, "%lu", data->size);
172 } else if (is_atom("objectsize:disk", atom, len)) {
173 if (data->mark_query)
174 data->info.disk_sizep = &data->disk_size;
175 else
176 strbuf_addf(sb, "%lu", data->disk_size);
177 } else if (is_atom("rest", atom, len)) {
178 if (data->mark_query)
179 data->split_on_whitespace = 1;
180 else if (data->rest)
181 strbuf_addstr(sb, data->rest);
182 } else if (is_atom("deltabase", atom, len)) {
183 if (data->mark_query)
184 data->info.delta_base_sha1 = data->delta_base_sha1;
185 else
186 strbuf_addstr(sb, sha1_to_hex(data->delta_base_sha1));
187 } else
188 die("unknown format element: %.*s", len, atom);
189}
190
191static size_t expand_format(struct strbuf *sb, const char *start, void *data)
192{
193 const char *end;
194
195 if (*start != '(')
196 return 0;
197 end = strchr(start + 1, ')');
198 if (!end)
199 die("format element '%s' does not end in ')'", start);
200
201 expand_atom(sb, start + 1, end - start - 1, data);
202
203 return end - start + 1;
204}
205
206static void print_object_or_die(int fd, struct expand_data *data)
207{
208 const unsigned char *sha1 = data->sha1;
209
210 assert(data->info.typep);
211
212 if (data->type == OBJ_BLOB) {
213 if (stream_blob_to_fd(fd, sha1, NULL, 0) < 0)
214 die("unable to stream %s to stdout", sha1_to_hex(sha1));
215 }
216 else {
217 enum object_type type;
218 unsigned long size;
219 void *contents;
220
221 contents = read_sha1_file(sha1, &type, &size);
222 if (!contents)
223 die("object %s disappeared", sha1_to_hex(sha1));
224 if (type != data->type)
225 die("object %s changed type!?", sha1_to_hex(sha1));
226 if (data->info.sizep && size != data->size)
227 die("object %s changed size!?", sha1_to_hex(sha1));
228
229 write_or_die(fd, contents, size);
230 free(contents);
231 }
232}
233
234struct batch_options {
235 int enabled;
236 int print_contents;
237 const char *format;
238};
239
240static int batch_one_object(const char *obj_name, struct batch_options *opt,
241 struct expand_data *data)
242{
243 struct strbuf buf = STRBUF_INIT;
244
245 if (!obj_name)
246 return 1;
247
248 if (get_sha1(obj_name, data->sha1)) {
249 printf("%s missing\n", obj_name);
250 fflush(stdout);
251 return 0;
252 }
253
254 if (sha1_object_info_extended(data->sha1, &data->info, LOOKUP_REPLACE_OBJECT) < 0) {
255 printf("%s missing\n", obj_name);
256 fflush(stdout);
257 return 0;
258 }
259
260 strbuf_expand(&buf, opt->format, expand_format, data);
261 strbuf_addch(&buf, '\n');
262 write_or_die(1, buf.buf, buf.len);
263 strbuf_release(&buf);
264
265 if (opt->print_contents) {
266 print_object_or_die(1, data);
267 write_or_die(1, "\n", 1);
268 }
269 return 0;
270}
271
272static int batch_objects(struct batch_options *opt)
273{
274 struct strbuf buf = STRBUF_INIT;
275 struct expand_data data;
276 int save_warning;
277 int retval = 0;
278
279 if (!opt->format)
280 opt->format = "%(objectname) %(objecttype) %(objectsize)";
281
282 /*
283 * Expand once with our special mark_query flag, which will prime the
284 * object_info to be handed to sha1_object_info_extended for each
285 * object.
286 */
287 memset(&data, 0, sizeof(data));
288 data.mark_query = 1;
289 strbuf_expand(&buf, opt->format, expand_format, &data);
290 data.mark_query = 0;
291
292 /*
293 * If we are printing out the object, then always fill in the type,
294 * since we will want to decide whether or not to stream.
295 */
296 if (opt->print_contents)
297 data.info.typep = &data.type;
298
299 /*
300 * We are going to call get_sha1 on a potentially very large number of
301 * objects. In most large cases, these will be actual object sha1s. The
302 * cost to double-check that each one is not also a ref (just so we can
303 * warn) ends up dwarfing the actual cost of the object lookups
304 * themselves. We can work around it by just turning off the warning.
305 */
306 save_warning = warn_on_object_refname_ambiguity;
307 warn_on_object_refname_ambiguity = 0;
308
309 while (strbuf_getline(&buf, stdin, '\n') != EOF) {
310 if (data.split_on_whitespace) {
311 /*
312 * Split at first whitespace, tying off the beginning
313 * of the string and saving the remainder (or NULL) in
314 * data.rest.
315 */
316 char *p = strpbrk(buf.buf, " \t");
317 if (p) {
318 while (*p && strchr(" \t", *p))
319 *p++ = '\0';
320 }
321 data.rest = p;
322 }
323
324 retval = batch_one_object(buf.buf, opt, &data);
325 if (retval)
326 break;
327 }
328
329 strbuf_release(&buf);
330 warn_on_object_refname_ambiguity = save_warning;
331 return retval;
332}
333
334static const char * const cat_file_usage[] = {
335 N_("git cat-file (-t [--allow-unknown-type]|-s [--allow-unknown-type]|-e|-p|<type>|--textconv) <object>"),
336 N_("git cat-file (--batch | --batch-check) < <list-of-objects>"),
337 NULL
338};
339
340static int git_cat_file_config(const char *var, const char *value, void *cb)
341{
342 if (userdiff_config(var, value) < 0)
343 return -1;
344
345 return git_default_config(var, value, cb);
346}
347
348static int batch_option_callback(const struct option *opt,
349 const char *arg,
350 int unset)
351{
352 struct batch_options *bo = opt->value;
353
354 if (unset) {
355 memset(bo, 0, sizeof(*bo));
356 return 0;
357 }
358
359 bo->enabled = 1;
360 bo->print_contents = !strcmp(opt->long_name, "batch");
361 bo->format = arg;
362
363 return 0;
364}
365
366int cmd_cat_file(int argc, const char **argv, const char *prefix)
367{
368 int opt = 0;
369 const char *exp_type = NULL, *obj_name = NULL;
370 struct batch_options batch = {0};
371 int unknown_type = 0;
372
373 const struct option options[] = {
374 OPT_GROUP(N_("<type> can be one of: blob, tree, commit, tag")),
375 OPT_CMDMODE('t', NULL, &opt, N_("show object type"), 't'),
376 OPT_CMDMODE('s', NULL, &opt, N_("show object size"), 's'),
377 OPT_CMDMODE('e', NULL, &opt,
378 N_("exit with zero when there's no error"), 'e'),
379 OPT_CMDMODE('p', NULL, &opt, N_("pretty-print object's content"), 'p'),
380 OPT_CMDMODE(0, "textconv", &opt,
381 N_("for blob objects, run textconv on object's content"), 'c'),
382 OPT_BOOL( 0, "allow-unknown-type", &unknown_type,
383 N_("allow -s and -t to work with broken/corrupt objects")),
384 { OPTION_CALLBACK, 0, "batch", &batch, "format",
385 N_("show info and content of objects fed from the standard input"),
386 PARSE_OPT_OPTARG, batch_option_callback },
387 { OPTION_CALLBACK, 0, "batch-check", &batch, "format",
388 N_("show info about objects fed from the standard input"),
389 PARSE_OPT_OPTARG, batch_option_callback },
390 OPT_END()
391 };
392
393 git_config(git_cat_file_config, NULL);
394
395 argc = parse_options(argc, argv, prefix, options, cat_file_usage, 0);
396
397 if (opt) {
398 if (argc == 1)
399 obj_name = argv[0];
400 else
401 usage_with_options(cat_file_usage, options);
402 }
403 if (!opt && !batch.enabled) {
404 if (argc == 2) {
405 exp_type = argv[0];
406 obj_name = argv[1];
407 } else
408 usage_with_options(cat_file_usage, options);
409 }
410 if (batch.enabled && (opt || argc)) {
411 usage_with_options(cat_file_usage, options);
412 }
413
414 if (batch.enabled)
415 return batch_objects(&batch);
416
417 if (unknown_type && opt != 't' && opt != 's')
418 die("git cat-file --allow-unknown-type: use with -s or -t");
419 return cat_one_file(opt, exp_type, obj_name, unknown_type);
420}