Add the diff option --no-defaults
[git/dscho.git] / builtin-unpack-objects.c
blob9a773239cabab9998bcea829c0fb2abea9bdb8e8
1 #include "builtin.h"
2 #include "cache.h"
3 #include "object.h"
4 #include "delta.h"
5 #include "pack.h"
6 #include "blob.h"
7 #include "commit.h"
8 #include "tag.h"
9 #include "tree.h"
10 #include "tree-walk.h"
11 #include "progress.h"
12 #include "decorate.h"
13 #include "fsck.h"
15 static int dry_run, quiet, recover, has_errors, strict;
16 static const char unpack_usage[] = "git unpack-objects [-n] [-q] [-r] [--strict] < pack-file";
18 /* We always read in 4kB chunks. */
19 static unsigned char buffer[4096];
20 static unsigned int offset, len;
21 static off_t consumed_bytes;
22 static git_SHA_CTX ctx;
25 * When running under --strict mode, objects whose reachability are
26 * suspect are kept in core without getting written in the object
27 * store.
29 struct obj_buffer {
30 char *buffer;
31 unsigned long size;
34 static struct decoration obj_decorate;
36 static struct obj_buffer *lookup_object_buffer(struct object *base)
38 return lookup_decoration(&obj_decorate, base);
41 static void add_object_buffer(struct object *object, char *buffer, unsigned long size)
43 struct obj_buffer *obj;
44 obj = xcalloc(1, sizeof(struct obj_buffer));
45 obj->buffer = buffer;
46 obj->size = size;
47 if (add_decoration(&obj_decorate, object, obj))
48 die("object %s tried to add buffer twice!", sha1_to_hex(object->sha1));
52 * Make sure at least "min" bytes are available in the buffer, and
53 * return the pointer to the buffer.
55 static void *fill(int min)
57 if (min <= len)
58 return buffer + offset;
59 if (min > sizeof(buffer))
60 die("cannot fill %d bytes", min);
61 if (offset) {
62 git_SHA1_Update(&ctx, buffer, offset);
63 memmove(buffer, buffer + offset, len);
64 offset = 0;
66 do {
67 ssize_t ret = xread(0, buffer + len, sizeof(buffer) - len);
68 if (ret <= 0) {
69 if (!ret)
70 die("early EOF");
71 die("read error on input: %s", strerror(errno));
73 len += ret;
74 } while (len < min);
75 return buffer;
78 static void use(int bytes)
80 if (bytes > len)
81 die("used more bytes than were available");
82 len -= bytes;
83 offset += bytes;
85 /* make sure off_t is sufficiently large not to wrap */
86 if (consumed_bytes > consumed_bytes + bytes)
87 die("pack too large for current definition of off_t");
88 consumed_bytes += bytes;
91 static void *get_data(unsigned long size)
93 z_stream stream;
94 void *buf = xmalloc(size);
96 memset(&stream, 0, sizeof(stream));
98 stream.next_out = buf;
99 stream.avail_out = size;
100 stream.next_in = fill(1);
101 stream.avail_in = len;
102 git_inflate_init(&stream);
104 for (;;) {
105 int ret = git_inflate(&stream, 0);
106 use(len - stream.avail_in);
107 if (stream.total_out == size && ret == Z_STREAM_END)
108 break;
109 if (ret != Z_OK) {
110 error("inflate returned %d\n", ret);
111 free(buf);
112 buf = NULL;
113 if (!recover)
114 exit(1);
115 has_errors = 1;
116 break;
118 stream.next_in = fill(1);
119 stream.avail_in = len;
121 git_inflate_end(&stream);
122 return buf;
125 struct delta_info {
126 unsigned char base_sha1[20];
127 unsigned nr;
128 off_t base_offset;
129 unsigned long size;
130 void *delta;
131 struct delta_info *next;
134 static struct delta_info *delta_list;
136 static void add_delta_to_list(unsigned nr, unsigned const char *base_sha1,
137 off_t base_offset,
138 void *delta, unsigned long size)
140 struct delta_info *info = xmalloc(sizeof(*info));
142 hashcpy(info->base_sha1, base_sha1);
143 info->base_offset = base_offset;
144 info->size = size;
145 info->delta = delta;
146 info->nr = nr;
147 info->next = delta_list;
148 delta_list = info;
151 struct obj_info {
152 off_t offset;
153 unsigned char sha1[20];
154 struct object *obj;
157 #define FLAG_OPEN (1u<<20)
158 #define FLAG_WRITTEN (1u<<21)
160 static struct obj_info *obj_list;
161 unsigned nr_objects;
164 * Called only from check_object() after it verified this object
165 * is Ok.
167 static void write_cached_object(struct object *obj)
169 unsigned char sha1[20];
170 struct obj_buffer *obj_buf = lookup_object_buffer(obj);
171 if (write_sha1_file(obj_buf->buffer, obj_buf->size, typename(obj->type), sha1) < 0)
172 die("failed to write object %s", sha1_to_hex(obj->sha1));
173 obj->flags |= FLAG_WRITTEN;
177 * At the very end of the processing, write_rest() scans the objects
178 * that have reachability requirements and calls this function.
179 * Verify its reachability and validity recursively and write it out.
181 static int check_object(struct object *obj, int type, void *data)
183 if (!obj)
184 return 0;
186 if (obj->flags & FLAG_WRITTEN)
187 return 1;
189 if (type != OBJ_ANY && obj->type != type)
190 die("object type mismatch");
192 if (!(obj->flags & FLAG_OPEN)) {
193 unsigned long size;
194 int type = sha1_object_info(obj->sha1, &size);
195 if (type != obj->type || type <= 0)
196 die("object of unexpected type");
197 obj->flags |= FLAG_WRITTEN;
198 return 1;
201 if (fsck_object(obj, 1, fsck_error_function))
202 die("Error in object");
203 if (!fsck_walk(obj, check_object, 0))
204 die("Error on reachable objects of %s", sha1_to_hex(obj->sha1));
205 write_cached_object(obj);
206 return 1;
209 static void write_rest(void)
211 unsigned i;
212 for (i = 0; i < nr_objects; i++)
213 check_object(obj_list[i].obj, OBJ_ANY, 0);
216 static void added_object(unsigned nr, enum object_type type,
217 void *data, unsigned long size);
220 * Write out nr-th object from the list, now we know the contents
221 * of it. Under --strict, this buffers structured objects in-core,
222 * to be checked at the end.
224 static void write_object(unsigned nr, enum object_type type,
225 void *buf, unsigned long size)
227 if (!strict) {
228 if (write_sha1_file(buf, size, typename(type), obj_list[nr].sha1) < 0)
229 die("failed to write object");
230 added_object(nr, type, buf, size);
231 free(buf);
232 obj_list[nr].obj = NULL;
233 } else if (type == OBJ_BLOB) {
234 struct blob *blob;
235 if (write_sha1_file(buf, size, typename(type), obj_list[nr].sha1) < 0)
236 die("failed to write object");
237 added_object(nr, type, buf, size);
238 free(buf);
240 blob = lookup_blob(obj_list[nr].sha1);
241 if (blob)
242 blob->object.flags |= FLAG_WRITTEN;
243 else
244 die("invalid blob object");
245 obj_list[nr].obj = NULL;
246 } else {
247 struct object *obj;
248 int eaten;
249 hash_sha1_file(buf, size, typename(type), obj_list[nr].sha1);
250 added_object(nr, type, buf, size);
251 obj = parse_object_buffer(obj_list[nr].sha1, type, size, buf, &eaten);
252 if (!obj)
253 die("invalid %s", typename(type));
254 add_object_buffer(obj, buf, size);
255 obj->flags |= FLAG_OPEN;
256 obj_list[nr].obj = obj;
260 static void resolve_delta(unsigned nr, enum object_type type,
261 void *base, unsigned long base_size,
262 void *delta, unsigned long delta_size)
264 void *result;
265 unsigned long result_size;
267 result = patch_delta(base, base_size,
268 delta, delta_size,
269 &result_size);
270 if (!result)
271 die("failed to apply delta");
272 free(delta);
273 write_object(nr, type, result, result_size);
277 * We now know the contents of an object (which is nr-th in the pack);
278 * resolve all the deltified objects that are based on it.
280 static void added_object(unsigned nr, enum object_type type,
281 void *data, unsigned long size)
283 struct delta_info **p = &delta_list;
284 struct delta_info *info;
286 while ((info = *p) != NULL) {
287 if (!hashcmp(info->base_sha1, obj_list[nr].sha1) ||
288 info->base_offset == obj_list[nr].offset) {
289 *p = info->next;
290 p = &delta_list;
291 resolve_delta(info->nr, type, data, size,
292 info->delta, info->size);
293 free(info);
294 continue;
296 p = &info->next;
300 static void unpack_non_delta_entry(enum object_type type, unsigned long size,
301 unsigned nr)
303 void *buf = get_data(size);
305 if (!dry_run && buf)
306 write_object(nr, type, buf, size);
307 else
308 free(buf);
311 static int resolve_against_held(unsigned nr, const unsigned char *base,
312 void *delta_data, unsigned long delta_size)
314 struct object *obj;
315 struct obj_buffer *obj_buffer;
316 obj = lookup_object(base);
317 if (!obj)
318 return 0;
319 obj_buffer = lookup_object_buffer(obj);
320 if (!obj_buffer)
321 return 0;
322 resolve_delta(nr, obj->type, obj_buffer->buffer,
323 obj_buffer->size, delta_data, delta_size);
324 return 1;
327 static void unpack_delta_entry(enum object_type type, unsigned long delta_size,
328 unsigned nr)
330 void *delta_data, *base;
331 unsigned long base_size;
332 unsigned char base_sha1[20];
334 if (type == OBJ_REF_DELTA) {
335 hashcpy(base_sha1, fill(20));
336 use(20);
337 delta_data = get_data(delta_size);
338 if (dry_run || !delta_data) {
339 free(delta_data);
340 return;
342 if (has_sha1_file(base_sha1))
343 ; /* Ok we have this one */
344 else if (resolve_against_held(nr, base_sha1,
345 delta_data, delta_size))
346 return; /* we are done */
347 else {
348 /* cannot resolve yet --- queue it */
349 hashcpy(obj_list[nr].sha1, null_sha1);
350 add_delta_to_list(nr, base_sha1, 0, delta_data, delta_size);
351 return;
353 } else {
354 unsigned base_found = 0;
355 unsigned char *pack, c;
356 off_t base_offset;
357 unsigned lo, mid, hi;
359 pack = fill(1);
360 c = *pack;
361 use(1);
362 base_offset = c & 127;
363 while (c & 128) {
364 base_offset += 1;
365 if (!base_offset || MSB(base_offset, 7))
366 die("offset value overflow for delta base object");
367 pack = fill(1);
368 c = *pack;
369 use(1);
370 base_offset = (base_offset << 7) + (c & 127);
372 base_offset = obj_list[nr].offset - base_offset;
373 if (base_offset <= 0 || base_offset >= obj_list[nr].offset)
374 die("offset value out of bound for delta base object");
376 delta_data = get_data(delta_size);
377 if (dry_run || !delta_data) {
378 free(delta_data);
379 return;
381 lo = 0;
382 hi = nr;
383 while (lo < hi) {
384 mid = (lo + hi)/2;
385 if (base_offset < obj_list[mid].offset) {
386 hi = mid;
387 } else if (base_offset > obj_list[mid].offset) {
388 lo = mid + 1;
389 } else {
390 hashcpy(base_sha1, obj_list[mid].sha1);
391 base_found = !is_null_sha1(base_sha1);
392 break;
395 if (!base_found) {
397 * The delta base object is itself a delta that
398 * has not been resolved yet.
400 hashcpy(obj_list[nr].sha1, null_sha1);
401 add_delta_to_list(nr, null_sha1, base_offset, delta_data, delta_size);
402 return;
406 if (resolve_against_held(nr, base_sha1, delta_data, delta_size))
407 return;
409 base = read_sha1_file(base_sha1, &type, &base_size);
410 if (!base) {
411 error("failed to read delta-pack base object %s",
412 sha1_to_hex(base_sha1));
413 if (!recover)
414 exit(1);
415 has_errors = 1;
416 return;
418 resolve_delta(nr, type, base, base_size, delta_data, delta_size);
419 free(base);
422 static void unpack_one(unsigned nr)
424 unsigned shift;
425 unsigned char *pack, c;
426 unsigned long size;
427 enum object_type type;
429 obj_list[nr].offset = consumed_bytes;
431 pack = fill(1);
432 c = *pack;
433 use(1);
434 type = (c >> 4) & 7;
435 size = (c & 15);
436 shift = 4;
437 while (c & 0x80) {
438 pack = fill(1);
439 c = *pack;
440 use(1);
441 size += (c & 0x7f) << shift;
442 shift += 7;
445 switch (type) {
446 case OBJ_COMMIT:
447 case OBJ_TREE:
448 case OBJ_BLOB:
449 case OBJ_TAG:
450 unpack_non_delta_entry(type, size, nr);
451 return;
452 case OBJ_REF_DELTA:
453 case OBJ_OFS_DELTA:
454 unpack_delta_entry(type, size, nr);
455 return;
456 default:
457 error("bad object type %d", type);
458 has_errors = 1;
459 if (recover)
460 return;
461 exit(1);
465 static void unpack_all(void)
467 int i;
468 struct progress *progress = NULL;
469 struct pack_header *hdr = fill(sizeof(struct pack_header));
471 nr_objects = ntohl(hdr->hdr_entries);
473 if (ntohl(hdr->hdr_signature) != PACK_SIGNATURE)
474 die("bad pack file");
475 if (!pack_version_ok(hdr->hdr_version))
476 die("unknown pack file version %"PRIu32,
477 ntohl(hdr->hdr_version));
478 use(sizeof(struct pack_header));
480 if (!quiet)
481 progress = start_progress("Unpacking objects", nr_objects);
482 obj_list = xcalloc(nr_objects, sizeof(*obj_list));
483 for (i = 0; i < nr_objects; i++) {
484 unpack_one(i);
485 display_progress(progress, i + 1);
487 stop_progress(&progress);
489 if (delta_list)
490 die("unresolved deltas left after unpacking");
493 int cmd_unpack_objects(int argc, const char **argv, const char *prefix)
495 int i;
496 unsigned char sha1[20];
498 git_config(git_default_config, NULL);
500 quiet = !isatty(2);
502 for (i = 1 ; i < argc; i++) {
503 const char *arg = argv[i];
505 if (*arg == '-') {
506 if (!strcmp(arg, "-n")) {
507 dry_run = 1;
508 continue;
510 if (!strcmp(arg, "-q")) {
511 quiet = 1;
512 continue;
514 if (!strcmp(arg, "-r")) {
515 recover = 1;
516 continue;
518 if (!strcmp(arg, "--strict")) {
519 strict = 1;
520 continue;
522 if (!prefixcmp(arg, "--pack_header=")) {
523 struct pack_header *hdr;
524 char *c;
526 hdr = (struct pack_header *)buffer;
527 hdr->hdr_signature = htonl(PACK_SIGNATURE);
528 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
529 if (*c != ',')
530 die("bad %s", arg);
531 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
532 if (*c)
533 die("bad %s", arg);
534 len = sizeof(*hdr);
535 continue;
537 usage(unpack_usage);
540 /* We don't take any non-flag arguments now.. Maybe some day */
541 usage(unpack_usage);
543 git_SHA1_Init(&ctx);
544 unpack_all();
545 git_SHA1_Update(&ctx, buffer, offset);
546 git_SHA1_Final(sha1, &ctx);
547 if (strict)
548 write_rest();
549 if (hashcmp(fill(20), sha1))
550 die("final sha1 did not match");
551 use(20);
553 /* Write the last part of the buffer to stdout */
554 while (len) {
555 int ret = xwrite(1, buffer + offset, len);
556 if (ret <= 0)
557 break;
558 len -= ret;
559 offset += ret;
562 /* All done */
563 return has_errors;