git-pack-objects: add "--stdout" flag to write the pack file to stdout
[git] / pack-objects.c
1 #include <ctype.h>
2 #include "cache.h"
3 #include "object.h"
4 #include "delta.h"
5 #include "csum-file.h"
6
7 static const char pack_usage[] = "git-pack-objects [--window=N] [--depth=N] {--stdout | base-name} < object-list";
8
9 /*
10  * The object type is a single-character shorthand:
11  *  - 'C' for "Commit"
12  *  - 'T' for "Tree"
13  *  - 'B' for "Blob"
14  *  - 'G' for "taG"
15  *  - 'D' for "Delta"
16  */
17 struct object_entry {
18         unsigned char sha1[20];
19         unsigned long size;
20         unsigned long offset;
21         unsigned int depth;
22         unsigned int hash;
23         unsigned char type;
24         unsigned long delta_size;
25         struct object_entry *delta;
26 };
27
28 static struct object_entry **sorted_by_sha, **sorted_by_type;
29 static struct object_entry *objects = NULL;
30 static int nr_objects = 0, nr_alloc = 0;
31 static const char *base_name;
32 static unsigned char pack_file_sha1[20];
33
34 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
35 {
36         unsigned long othersize, delta_size;
37         char type[10];
38         void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
39         void *delta_buf;
40
41         if (!otherbuf)
42                 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
43         delta_buf = diff_delta(otherbuf, othersize,
44                                buf, size, &delta_size, ~0UL);
45         if (!delta_buf || delta_size != entry->delta_size)
46                 die("delta size changed");
47         free(buf);
48         free(otherbuf);
49         return delta_buf;
50 }
51
52 static unsigned long write_object(struct sha1file *f, struct object_entry *entry)
53 {
54         unsigned long size;
55         char type[10];
56         void *buf = read_sha1_file(entry->sha1, type, &size);
57         char header[25];
58         unsigned hdrlen, datalen;
59
60         if (!buf)
61                 die("unable to read %s", sha1_to_hex(entry->sha1));
62         if (size != entry->size)
63                 die("object %s size inconsistency (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
64
65         /*
66          * The object header is a byte of 'type' followed by four bytes of
67          * length, except for deltas that has the 20 bytes of delta sha
68          * instead.
69          */
70         header[0] = entry->type;
71         hdrlen = 5;
72         if (entry->delta) {
73                 header[0] = 'D';
74                 memcpy(header+5, entry->delta, 20);
75                 buf = delta_against(buf, size, entry);
76                 size = entry->delta_size;
77                 hdrlen = 25;
78         }
79         datalen = htonl(size);
80         memcpy(header+1, &datalen, 4);
81         sha1write(f, header, hdrlen);
82         datalen = sha1write_compressed(f, buf, size);
83         free(buf);
84         return hdrlen + datalen;
85 }
86
87 static void write_pack_file(void)
88 {
89         int i;
90         struct sha1file *f;
91         unsigned long offset = 0;
92         unsigned long mb;
93
94         if (!base_name)
95                 f = sha1fd(1, "<stdout>");
96         else
97                 f = sha1create("%s.%s", base_name, "pack");
98         for (i = 0; i < nr_objects; i++) {
99                 struct object_entry *entry = objects + i;
100                 entry->offset = offset;
101                 offset += write_object(f, entry);
102         }
103         sha1close(f, pack_file_sha1, 1);
104         mb = offset >> 20;
105         offset &= 0xfffff;
106 }
107
108 static void write_index_file(void)
109 {
110         int i;
111         struct sha1file *f = sha1create("%s.%s", base_name, "idx");
112         struct object_entry **list = sorted_by_sha;
113         struct object_entry **last = list + nr_objects;
114         unsigned int array[256];
115
116         /*
117          * Write the first-level table (the list is sorted,
118          * but we use a 256-entry lookup to be able to avoid
119          * having to do eight extra binary search iterations).
120          */
121         for (i = 0; i < 256; i++) {
122                 struct object_entry **next = list;
123                 while (next < last) {
124                         struct object_entry *entry = *next;
125                         if (entry->sha1[0] != i)
126                                 break;
127                         next++;
128                 }
129                 array[i] = htonl(next - sorted_by_sha);
130                 list = next;
131         }
132         sha1write(f, array, 256 * sizeof(int));
133
134         /*
135          * Write the actual SHA1 entries..
136          */
137         list = sorted_by_sha;
138         for (i = 0; i < nr_objects; i++) {
139                 struct object_entry *entry = *list++;
140                 unsigned int offset = htonl(entry->offset);
141                 sha1write(f, &offset, 4);
142                 sha1write(f, entry->sha1, 20);
143         }
144         sha1write(f, pack_file_sha1, 20);
145         sha1close(f, NULL, 1);
146 }
147
148 static void add_object_entry(unsigned char *sha1, unsigned int hash)
149 {
150         unsigned int idx = nr_objects;
151         struct object_entry *entry;
152
153         if (idx >= nr_alloc) {
154                 unsigned int needed = (idx + 1024) * 3 / 2;
155                 objects = xrealloc(objects, needed * sizeof(*entry));
156                 nr_alloc = needed;
157         }
158         entry = objects + idx;
159         memset(entry, 0, sizeof(*entry));
160         memcpy(entry->sha1, sha1, 20);
161         entry->hash = hash;
162         nr_objects = idx+1;
163 }
164
165 static void check_object(struct object_entry *entry)
166 {
167         char type[20];
168
169         if (!sha1_object_info(entry->sha1, type, &entry->size)) {
170                 if (!strcmp(type, "commit")) {
171                         entry->type = 'C';
172                 } else if (!strcmp(type, "tree")) {
173                         entry->type = 'T';
174                 } else if (!strcmp(type, "blob")) {
175                         entry->type = 'B';
176                 } else if (!strcmp(type, "tag")) {
177                         entry->type = 'G';
178                 } else
179                         die("unable to pack object %s of type %s",
180                             sha1_to_hex(entry->sha1), type);
181         }
182         else
183                 die("unable to get type of object %s",
184                     sha1_to_hex(entry->sha1));
185 }
186
187 static void get_object_details(void)
188 {
189         int i;
190         struct object_entry *entry = objects;
191
192         for (i = 0; i < nr_objects; i++)
193                 check_object(entry++);
194 }
195
196 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
197
198 static entry_sort_t current_sort;
199
200 static int sort_comparator(const void *_a, const void *_b)
201 {
202         struct object_entry *a = *(struct object_entry **)_a;
203         struct object_entry *b = *(struct object_entry **)_b;
204         return current_sort(a,b);
205 }
206
207 static struct object_entry **create_sorted_list(entry_sort_t sort)
208 {
209         struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
210         int i;
211
212         for (i = 0; i < nr_objects; i++)
213                 list[i] = objects + i;
214         current_sort = sort;
215         qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
216         return list;
217 }
218
219 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
220 {
221         return memcmp(a->sha1, b->sha1, 20);
222 }
223
224 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
225 {
226         if (a->type < b->type)
227                 return -1;
228         if (a->type > b->type)
229                 return 1;
230         if (a->hash < b->hash)
231                 return -1;
232         if (a->hash > b->hash)
233                 return 1;
234         if (a->size < b->size)
235                 return -1;
236         if (a->size > b->size)
237                 return 1;
238         return a < b ? -1 : (a > b);
239 }
240
241 struct unpacked {
242         struct object_entry *entry;
243         void *data;
244 };
245
246 /*
247  * We search for deltas _backwards_ in a list sorted by type and
248  * by size, so that we see progressively smaller and smaller files.
249  * That's because we prefer deltas to be from the bigger file
250  * to the smaller - deletes are potentially cheaper, but perhaps
251  * more importantly, the bigger file is likely the more recent
252  * one.
253  */
254 static int try_delta(struct unpacked *cur, struct unpacked *old, unsigned max_depth)
255 {
256         struct object_entry *cur_entry = cur->entry;
257         struct object_entry *old_entry = old->entry;
258         unsigned long size, oldsize, delta_size, sizediff;
259         long max_size;
260         void *delta_buf;
261
262         /* Don't bother doing diffs between different types */
263         if (cur_entry->type != old_entry->type)
264                 return -1;
265
266         size = cur_entry->size;
267         if (size < 50)
268                 return -1;
269         oldsize = old_entry->size;
270         sizediff = oldsize > size ? oldsize - size : size - oldsize;
271         if (sizediff > size / 8)
272                 return -1;
273         if (old_entry->depth >= max_depth)
274                 return 0;
275
276         /*
277          * NOTE!
278          *
279          * We always delta from the bigger to the smaller, since that's
280          * more space-efficient (deletes don't have to say _what_ they
281          * delete).
282          */
283         max_size = size / 2 - 20;
284         if (cur_entry->delta)
285                 max_size = cur_entry->delta_size-1;
286         if (sizediff >= max_size)
287                 return -1;
288         delta_buf = diff_delta(old->data, oldsize,
289                                cur->data, size, &delta_size, max_size);
290         if (!delta_buf)
291                 return 0;
292         cur_entry->delta = old_entry;
293         cur_entry->delta_size = delta_size;
294         cur_entry->depth = old_entry->depth + 1;
295         free(delta_buf);
296         return 0;
297 }
298
299 static void find_deltas(struct object_entry **list, int window, int depth)
300 {
301         int i, idx;
302         unsigned int array_size = window * sizeof(struct unpacked);
303         struct unpacked *array = xmalloc(array_size);
304
305         memset(array, 0, array_size);
306         i = nr_objects;
307         idx = 0;
308         while (--i >= 0) {
309                 struct object_entry *entry = list[i];
310                 struct unpacked *n = array + idx;
311                 unsigned long size;
312                 char type[10];
313                 int j;
314
315                 free(n->data);
316                 n->entry = entry;
317                 n->data = read_sha1_file(entry->sha1, type, &size);
318                 if (size != entry->size)
319                         die("object %s inconsistent object length (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
320                 j = window;
321                 while (--j > 0) {
322                         unsigned int other_idx = idx + j;
323                         struct unpacked *m;
324                         if (other_idx >= window)
325                                 other_idx -= window;
326                         m = array + other_idx;
327                         if (!m->entry)
328                                 break;
329                         if (try_delta(n, m, depth) < 0)
330                                 break;
331                 }
332                 idx++;
333                 if (idx >= window)
334                         idx = 0;
335         }
336 }
337
338 int main(int argc, char **argv)
339 {
340         char line[PATH_MAX + 20];
341         int window = 10, depth = 10, pack_to_stdout = 0;
342         int i;
343
344         for (i = 1; i < argc; i++) {
345                 const char *arg = argv[i];
346
347                 if (*arg == '-') {
348                         if (!strncmp("--window=", arg, 9)) {
349                                 char *end;
350                                 window = strtoul(arg+9, &end, 0);
351                                 if (!arg[9] || *end)
352                                         usage(pack_usage);
353                                 continue;
354                         }
355                         if (!strncmp("--depth=", arg, 8)) {
356                                 char *end;
357                                 depth = strtoul(arg+8, &end, 0);
358                                 if (!arg[8] || *end)
359                                         usage(pack_usage);
360                                 continue;
361                         }
362                         if (!strcmp("--stdout", arg)) {
363                                 pack_to_stdout = 1;
364                                 continue;
365                         }
366                         usage(pack_usage);
367                 }
368                 if (base_name)
369                         usage(pack_usage);
370                 base_name = arg;
371         }
372
373         if (pack_to_stdout != !base_name)
374                 usage(pack_usage);
375
376         while (fgets(line, sizeof(line), stdin) != NULL) {
377                 unsigned int hash;
378                 char *p;
379                 unsigned char sha1[20];
380
381                 if (get_sha1_hex(line, sha1))
382                         die("expected sha1, got garbage");
383                 hash = 0;
384                 p = line+40;
385                 while (*p) {
386                         unsigned char c = *p++;
387                         if (isspace(c))
388                                 continue;
389                         hash = hash * 11 + c;
390                 }
391                 add_object_entry(sha1, hash);
392         }
393         get_object_details();
394
395         fprintf(stderr, "Packing %d objects\n", nr_objects);
396
397         sorted_by_sha = create_sorted_list(sha1_sort);
398         sorted_by_type = create_sorted_list(type_size_sort);
399         if (window && depth)
400                 find_deltas(sorted_by_type, window+1, depth);
401
402         write_pack_file();
403         if (!pack_to_stdout)
404                 write_index_file();
405         return 0;
406 }