[PATCH] Subject: [PATCH] Add some documentation.
[git/fastimport.git] / pack-objects.c
blob3d622787cc554eb2936995441e4591ec1986bded
1 #include <ctype.h>
2 #include "cache.h"
3 #include "object.h"
4 #include "delta.h"
5 #include "pack.h"
6 #include "csum-file.h"
8 static const char pack_usage[] = "git-pack-objects [--incremental] [--window=N] [--depth=N] {--stdout | base-name} < object-list";
10 struct object_entry {
11 unsigned char sha1[20];
12 unsigned long size;
13 unsigned long offset;
14 unsigned int depth;
15 unsigned int hash;
16 enum object_type type;
17 unsigned long delta_size;
18 struct object_entry *delta;
21 static unsigned char object_list_sha1[20];
22 static int non_empty = 0;
23 static int incremental = 0;
24 static struct object_entry **sorted_by_sha, **sorted_by_type;
25 static struct object_entry *objects = NULL;
26 static int nr_objects = 0, nr_alloc = 0;
27 static const char *base_name;
28 static unsigned char pack_file_sha1[20];
30 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
32 unsigned long othersize, delta_size;
33 char type[10];
34 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
35 void *delta_buf;
37 if (!otherbuf)
38 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
39 delta_buf = diff_delta(otherbuf, othersize,
40 buf, size, &delta_size, 0);
41 if (!delta_buf || delta_size != entry->delta_size)
42 die("delta size changed");
43 free(buf);
44 free(otherbuf);
45 return delta_buf;
49 * The per-object header is a pretty dense thing, which is
50 * - first byte: low four bits are "size", then three bits of "type",
51 * and the high bit is "size continues".
52 * - each byte afterwards: low seven bits are size continuation,
53 * with the high bit being "size continues"
55 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
57 int n = 1;
58 unsigned char c;
60 if (type < OBJ_COMMIT || type > OBJ_DELTA)
61 die("bad type %d", type);
63 c = (type << 4) | (size & 15);
64 size >>= 4;
65 while (size) {
66 *hdr++ = c | 0x80;
67 c = size & 0x7f;
68 size >>= 7;
69 n++;
71 *hdr = c;
72 return n;
75 static unsigned long write_object(struct sha1file *f, struct object_entry *entry)
77 unsigned long size;
78 char type[10];
79 void *buf = read_sha1_file(entry->sha1, type, &size);
80 unsigned char header[10];
81 unsigned hdrlen, datalen;
82 enum object_type obj_type;
84 if (!buf)
85 die("unable to read %s", sha1_to_hex(entry->sha1));
86 if (size != entry->size)
87 die("object %s size inconsistency (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
90 * The object header is a byte of 'type' followed by zero or
91 * more bytes of length. For deltas, the 20 bytes of delta sha1
92 * follows that.
94 obj_type = entry->type;
95 if (entry->delta) {
96 buf = delta_against(buf, size, entry);
97 size = entry->delta_size;
98 obj_type = OBJ_DELTA;
100 hdrlen = encode_header(obj_type, size, header);
101 sha1write(f, header, hdrlen);
102 if (entry->delta) {
103 sha1write(f, entry->delta, 20);
104 hdrlen += 20;
106 datalen = sha1write_compressed(f, buf, size);
107 free(buf);
108 return hdrlen + datalen;
111 static unsigned long write_one(struct sha1file *f,
112 struct object_entry *e,
113 unsigned long offset)
115 if (e->offset)
116 /* offset starts from header size and cannot be zero
117 * if it is written already.
119 return offset;
120 e->offset = offset;
121 offset += write_object(f, e);
122 /* if we are delitified, write out its base object. */
123 if (e->delta)
124 offset = write_one(f, e->delta, offset);
125 return offset;
128 static void write_pack_file(void)
130 int i;
131 struct sha1file *f;
132 unsigned long offset;
133 unsigned long mb;
134 struct pack_header hdr;
136 if (!base_name)
137 f = sha1fd(1, "<stdout>");
138 else
139 f = sha1create("%s-%s.%s", base_name, sha1_to_hex(object_list_sha1), "pack");
140 hdr.hdr_signature = htonl(PACK_SIGNATURE);
141 hdr.hdr_version = htonl(PACK_VERSION);
142 hdr.hdr_entries = htonl(nr_objects);
143 sha1write(f, &hdr, sizeof(hdr));
144 offset = sizeof(hdr);
145 for (i = 0; i < nr_objects; i++)
146 offset = write_one(f, objects + i, offset);
148 sha1close(f, pack_file_sha1, 1);
149 mb = offset >> 20;
150 offset &= 0xfffff;
153 static void write_index_file(void)
155 int i;
156 struct sha1file *f = sha1create("%s-%s.%s", base_name, sha1_to_hex(object_list_sha1), "idx");
157 struct object_entry **list = sorted_by_sha;
158 struct object_entry **last = list + nr_objects;
159 unsigned int array[256];
162 * Write the first-level table (the list is sorted,
163 * but we use a 256-entry lookup to be able to avoid
164 * having to do eight extra binary search iterations).
166 for (i = 0; i < 256; i++) {
167 struct object_entry **next = list;
168 while (next < last) {
169 struct object_entry *entry = *next;
170 if (entry->sha1[0] != i)
171 break;
172 next++;
174 array[i] = htonl(next - sorted_by_sha);
175 list = next;
177 sha1write(f, array, 256 * sizeof(int));
180 * Write the actual SHA1 entries..
182 list = sorted_by_sha;
183 for (i = 0; i < nr_objects; i++) {
184 struct object_entry *entry = *list++;
185 unsigned int offset = htonl(entry->offset);
186 sha1write(f, &offset, 4);
187 sha1write(f, entry->sha1, 20);
189 sha1write(f, pack_file_sha1, 20);
190 sha1close(f, NULL, 1);
193 static int add_object_entry(unsigned char *sha1, unsigned int hash)
195 unsigned int idx = nr_objects;
196 struct object_entry *entry;
198 if (incremental && has_sha1_pack(sha1))
199 return 0;
201 if (idx >= nr_alloc) {
202 unsigned int needed = (idx + 1024) * 3 / 2;
203 objects = xrealloc(objects, needed * sizeof(*entry));
204 nr_alloc = needed;
206 entry = objects + idx;
207 memset(entry, 0, sizeof(*entry));
208 memcpy(entry->sha1, sha1, 20);
209 entry->hash = hash;
210 nr_objects = idx+1;
211 return 1;
214 static void check_object(struct object_entry *entry)
216 char type[20];
218 if (!sha1_object_info(entry->sha1, type, &entry->size)) {
219 if (!strcmp(type, "commit")) {
220 entry->type = OBJ_COMMIT;
221 } else if (!strcmp(type, "tree")) {
222 entry->type = OBJ_TREE;
223 } else if (!strcmp(type, "blob")) {
224 entry->type = OBJ_BLOB;
225 } else if (!strcmp(type, "tag")) {
226 entry->type = OBJ_TAG;
227 } else
228 die("unable to pack object %s of type %s",
229 sha1_to_hex(entry->sha1), type);
231 else
232 die("unable to get type of object %s",
233 sha1_to_hex(entry->sha1));
236 static void get_object_details(void)
238 int i;
239 struct object_entry *entry = objects;
241 for (i = 0; i < nr_objects; i++)
242 check_object(entry++);
245 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
247 static entry_sort_t current_sort;
249 static int sort_comparator(const void *_a, const void *_b)
251 struct object_entry *a = *(struct object_entry **)_a;
252 struct object_entry *b = *(struct object_entry **)_b;
253 return current_sort(a,b);
256 static struct object_entry **create_sorted_list(entry_sort_t sort)
258 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
259 int i;
261 for (i = 0; i < nr_objects; i++)
262 list[i] = objects + i;
263 current_sort = sort;
264 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
265 return list;
268 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
270 return memcmp(a->sha1, b->sha1, 20);
273 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
275 if (a->type < b->type)
276 return -1;
277 if (a->type > b->type)
278 return 1;
279 if (a->hash < b->hash)
280 return -1;
281 if (a->hash > b->hash)
282 return 1;
283 if (a->size < b->size)
284 return -1;
285 if (a->size > b->size)
286 return 1;
287 return a < b ? -1 : (a > b);
290 struct unpacked {
291 struct object_entry *entry;
292 void *data;
296 * We search for deltas _backwards_ in a list sorted by type and
297 * by size, so that we see progressively smaller and smaller files.
298 * That's because we prefer deltas to be from the bigger file
299 * to the smaller - deletes are potentially cheaper, but perhaps
300 * more importantly, the bigger file is likely the more recent
301 * one.
303 static int try_delta(struct unpacked *cur, struct unpacked *old, unsigned max_depth)
305 struct object_entry *cur_entry = cur->entry;
306 struct object_entry *old_entry = old->entry;
307 unsigned long size, oldsize, delta_size, sizediff;
308 long max_size;
309 void *delta_buf;
311 /* Don't bother doing diffs between different types */
312 if (cur_entry->type != old_entry->type)
313 return -1;
315 size = cur_entry->size;
316 if (size < 50)
317 return -1;
318 oldsize = old_entry->size;
319 sizediff = oldsize > size ? oldsize - size : size - oldsize;
320 if (sizediff > size / 8)
321 return -1;
322 if (old_entry->depth >= max_depth)
323 return 0;
326 * NOTE!
328 * We always delta from the bigger to the smaller, since that's
329 * more space-efficient (deletes don't have to say _what_ they
330 * delete).
332 max_size = size / 2 - 20;
333 if (cur_entry->delta)
334 max_size = cur_entry->delta_size-1;
335 if (sizediff >= max_size)
336 return -1;
337 delta_buf = diff_delta(old->data, oldsize,
338 cur->data, size, &delta_size, max_size);
339 if (!delta_buf)
340 return 0;
341 cur_entry->delta = old_entry;
342 cur_entry->delta_size = delta_size;
343 cur_entry->depth = old_entry->depth + 1;
344 free(delta_buf);
345 return 0;
348 static void find_deltas(struct object_entry **list, int window, int depth)
350 int i, idx;
351 unsigned int array_size = window * sizeof(struct unpacked);
352 struct unpacked *array = xmalloc(array_size);
354 memset(array, 0, array_size);
355 i = nr_objects;
356 idx = 0;
357 while (--i >= 0) {
358 struct object_entry *entry = list[i];
359 struct unpacked *n = array + idx;
360 unsigned long size;
361 char type[10];
362 int j;
364 free(n->data);
365 n->entry = entry;
366 n->data = read_sha1_file(entry->sha1, type, &size);
367 if (size != entry->size)
368 die("object %s inconsistent object length (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
369 j = window;
370 while (--j > 0) {
371 unsigned int other_idx = idx + j;
372 struct unpacked *m;
373 if (other_idx >= window)
374 other_idx -= window;
375 m = array + other_idx;
376 if (!m->entry)
377 break;
378 if (try_delta(n, m, depth) < 0)
379 break;
381 idx++;
382 if (idx >= window)
383 idx = 0;
386 for (i = 0; i < window; ++i)
387 free(array[i].data);
388 free(array);
391 int main(int argc, char **argv)
393 SHA_CTX ctx;
394 char line[PATH_MAX + 20];
395 int window = 10, depth = 10, pack_to_stdout = 0;
396 int i;
398 for (i = 1; i < argc; i++) {
399 const char *arg = argv[i];
401 if (*arg == '-') {
402 if (!strcmp("--non-empty", arg)) {
403 non_empty = 1;
404 continue;
406 if (!strcmp("--incremental", arg)) {
407 incremental = 1;
408 continue;
410 if (!strncmp("--window=", arg, 9)) {
411 char *end;
412 window = strtoul(arg+9, &end, 0);
413 if (!arg[9] || *end)
414 usage(pack_usage);
415 continue;
417 if (!strncmp("--depth=", arg, 8)) {
418 char *end;
419 depth = strtoul(arg+8, &end, 0);
420 if (!arg[8] || *end)
421 usage(pack_usage);
422 continue;
424 if (!strcmp("--stdout", arg)) {
425 pack_to_stdout = 1;
426 continue;
428 usage(pack_usage);
430 if (base_name)
431 usage(pack_usage);
432 base_name = arg;
435 if (pack_to_stdout != !base_name)
436 usage(pack_usage);
438 SHA1_Init(&ctx);
439 while (fgets(line, sizeof(line), stdin) != NULL) {
440 unsigned int hash;
441 char *p;
442 unsigned char sha1[20];
444 if (get_sha1_hex(line, sha1))
445 die("expected sha1, got garbage");
446 hash = 0;
447 p = line+40;
448 while (*p) {
449 unsigned char c = *p++;
450 if (isspace(c))
451 continue;
452 hash = hash * 11 + c;
454 if (add_object_entry(sha1, hash))
455 SHA1_Update(&ctx, sha1, 20);
457 SHA1_Final(object_list_sha1, &ctx);
458 if (non_empty && !nr_objects)
459 return 0;
460 get_object_details();
462 fprintf(stderr, "Packing %d objects\n", nr_objects);
464 sorted_by_sha = create_sorted_list(sha1_sort);
465 sorted_by_type = create_sorted_list(type_size_sort);
466 if (window && depth)
467 find_deltas(sorted_by_type, window+1, depth);
469 write_pack_file();
470 if (!pack_to_stdout) {
471 write_index_file();
472 puts(sha1_to_hex(object_list_sha1));
474 return 0;