docs: describe impact of repack on "clone -s"
[git] / bundle.c
1 #include "cache.h"
2 #include "bundle.h"
3 #include "object.h"
4 #include "commit.h"
5 #include "diff.h"
6 #include "revision.h"
7 #include "list-objects.h"
8 #include "run-command.h"
9 #include "refs.h"
10
11 static const char bundle_signature[] = "# v2 git bundle\n";
12
13 static void add_to_ref_list(const unsigned char *sha1, const char *name,
14                 struct ref_list *list)
15 {
16         if (list->nr + 1 >= list->alloc) {
17                 list->alloc = alloc_nr(list->nr + 1);
18                 list->list = xrealloc(list->list,
19                                 list->alloc * sizeof(list->list[0]));
20         }
21         memcpy(list->list[list->nr].sha1, sha1, 20);
22         list->list[list->nr].name = xstrdup(name);
23         list->nr++;
24 }
25
26 /* returns an fd */
27 int read_bundle_header(const char *path, struct bundle_header *header)
28 {
29         char buffer[1024];
30         int fd;
31         long fpos;
32         FILE *ffd = fopen(path, "rb");
33
34         if (!ffd)
35                 return error("could not open '%s'", path);
36         if (!fgets(buffer, sizeof(buffer), ffd) ||
37                         strcmp(buffer, bundle_signature)) {
38                 fclose(ffd);
39                 return error("'%s' does not look like a v2 bundle file", path);
40         }
41         while (fgets(buffer, sizeof(buffer), ffd)
42                         && buffer[0] != '\n') {
43                 int is_prereq = buffer[0] == '-';
44                 int offset = is_prereq ? 1 : 0;
45                 int len = strlen(buffer);
46                 unsigned char sha1[20];
47                 struct ref_list *list = is_prereq ? &header->prerequisites
48                         : &header->references;
49                 char delim;
50
51                 if (len && buffer[len - 1] == '\n')
52                         buffer[len - 1] = '\0';
53                 if (get_sha1_hex(buffer + offset, sha1)) {
54                         warning("unrecognized header: %s", buffer);
55                         continue;
56                 }
57                 delim = buffer[40 + offset];
58                 if (!isspace(delim) && (delim != '\0' || !is_prereq))
59                         die ("invalid header: %s", buffer);
60                 add_to_ref_list(sha1, isspace(delim) ?
61                                 buffer + 41 + offset : "", list);
62         }
63         fpos = ftell(ffd);
64         fclose(ffd);
65         fd = open(path, O_RDONLY);
66         if (fd < 0)
67                 return error("could not open '%s'", path);
68         lseek(fd, fpos, SEEK_SET);
69         return fd;
70 }
71
72 static int list_refs(struct ref_list *r, int argc, const char **argv)
73 {
74         int i;
75
76         for (i = 0; i < r->nr; i++) {
77                 if (argc > 1) {
78                         int j;
79                         for (j = 1; j < argc; j++)
80                                 if (!strcmp(r->list[i].name, argv[j]))
81                                         break;
82                         if (j == argc)
83                                 continue;
84                 }
85                 printf("%s %s\n", sha1_to_hex(r->list[i].sha1),
86                                 r->list[i].name);
87         }
88         return 0;
89 }
90
91 #define PREREQ_MARK (1u<<16)
92
93 int verify_bundle(struct bundle_header *header, int verbose)
94 {
95         /*
96          * Do fast check, then if any prereqs are missing then go line by line
97          * to be verbose about the errors
98          */
99         struct ref_list *p = &header->prerequisites;
100         struct rev_info revs;
101         const char *argv[] = {NULL, "--all", NULL};
102         struct object_array refs;
103         struct commit *commit;
104         int i, ret = 0, req_nr;
105         const char *message = "Repository lacks these prerequisite commits:";
106
107         init_revisions(&revs, NULL);
108         for (i = 0; i < p->nr; i++) {
109                 struct ref_list_entry *e = p->list + i;
110                 struct object *o = parse_object(e->sha1);
111                 if (o) {
112                         o->flags |= PREREQ_MARK;
113                         add_pending_object(&revs, o, e->name);
114                         continue;
115                 }
116                 if (++ret == 1)
117                         error("%s", message);
118                 error("%s %s", sha1_to_hex(e->sha1), e->name);
119         }
120         if (revs.pending.nr != p->nr)
121                 return ret;
122         req_nr = revs.pending.nr;
123         setup_revisions(2, argv, &revs, NULL);
124
125         memset(&refs, 0, sizeof(struct object_array));
126         for (i = 0; i < revs.pending.nr; i++) {
127                 struct object_array_entry *e = revs.pending.objects + i;
128                 add_object_array(e->item, e->name, &refs);
129         }
130
131         if (prepare_revision_walk(&revs))
132                 die("revision walk setup failed");
133
134         i = req_nr;
135         while (i && (commit = get_revision(&revs)))
136                 if (commit->object.flags & PREREQ_MARK)
137                         i--;
138
139         for (i = 0; i < req_nr; i++)
140                 if (!(refs.objects[i].item->flags & SHOWN)) {
141                         if (++ret == 1)
142                                 error("%s", message);
143                         error("%s %s", sha1_to_hex(refs.objects[i].item->sha1),
144                                 refs.objects[i].name);
145                 }
146
147         for (i = 0; i < refs.nr; i++)
148                 clear_commit_marks((struct commit *)refs.objects[i].item, -1);
149
150         if (verbose) {
151                 struct ref_list *r;
152
153                 r = &header->references;
154                 printf("The bundle contains %d ref%s\n",
155                        r->nr, (1 < r->nr) ? "s" : "");
156                 list_refs(r, 0, NULL);
157                 r = &header->prerequisites;
158                 printf("The bundle requires these %d ref%s\n",
159                        r->nr, (1 < r->nr) ? "s" : "");
160                 list_refs(r, 0, NULL);
161         }
162         return ret;
163 }
164
165 int list_bundle_refs(struct bundle_header *header, int argc, const char **argv)
166 {
167         return list_refs(&header->references, argc, argv);
168 }
169
170 static int is_tag_in_date_range(struct object *tag, struct rev_info *revs)
171 {
172         unsigned long size;
173         enum object_type type;
174         char *buf, *line, *lineend;
175         unsigned long date;
176
177         if (revs->max_age == -1 && revs->min_age == -1)
178                 return 1;
179
180         buf = read_sha1_file(tag->sha1, &type, &size);
181         if (!buf)
182                 return 1;
183         line = memmem(buf, size, "\ntagger ", 8);
184         if (!line++)
185                 return 1;
186         lineend = memchr(line, buf + size - line, '\n');
187         line = memchr(line, lineend ? lineend - line : buf + size - line, '>');
188         if (!line++)
189                 return 1;
190         date = strtoul(line, NULL, 10);
191         free(buf);
192         return (revs->max_age == -1 || revs->max_age < date) &&
193                 (revs->min_age == -1 || revs->min_age > date);
194 }
195
196 int create_bundle(struct bundle_header *header, const char *path,
197                 int argc, const char **argv)
198 {
199         static struct lock_file lock;
200         int bundle_fd = -1;
201         int bundle_to_stdout;
202         const char **argv_boundary = xmalloc((argc + 4) * sizeof(const char *));
203         const char **argv_pack = xmalloc(5 * sizeof(const char *));
204         int i, ref_count = 0;
205         char buffer[1024];
206         struct rev_info revs;
207         int read_from_stdin = 0;
208         struct child_process rls;
209         FILE *rls_fout;
210
211         bundle_to_stdout = !strcmp(path, "-");
212         if (bundle_to_stdout)
213                 bundle_fd = 1;
214         else
215                 bundle_fd = hold_lock_file_for_update(&lock, path,
216                                                       LOCK_DIE_ON_ERROR);
217
218         /* write signature */
219         write_or_die(bundle_fd, bundle_signature, strlen(bundle_signature));
220
221         /* init revs to list objects for pack-objects later */
222         save_commit_buffer = 0;
223         init_revisions(&revs, NULL);
224
225         /* write prerequisites */
226         memcpy(argv_boundary + 3, argv + 1, argc * sizeof(const char *));
227         argv_boundary[0] = "rev-list";
228         argv_boundary[1] = "--boundary";
229         argv_boundary[2] = "--pretty=oneline";
230         argv_boundary[argc + 2] = NULL;
231         memset(&rls, 0, sizeof(rls));
232         rls.argv = argv_boundary;
233         rls.out = -1;
234         rls.git_cmd = 1;
235         if (start_command(&rls))
236                 return -1;
237         rls_fout = fdopen(rls.out, "r");
238         while (fgets(buffer, sizeof(buffer), rls_fout)) {
239                 unsigned char sha1[20];
240                 if (buffer[0] == '-') {
241                         write_or_die(bundle_fd, buffer, strlen(buffer));
242                         if (!get_sha1_hex(buffer + 1, sha1)) {
243                                 struct object *object = parse_object(sha1);
244                                 object->flags |= UNINTERESTING;
245                                 add_pending_object(&revs, object, buffer);
246                         }
247                 } else if (!get_sha1_hex(buffer, sha1)) {
248                         struct object *object = parse_object(sha1);
249                         object->flags |= SHOWN;
250                 }
251         }
252         fclose(rls_fout);
253         if (finish_command(&rls))
254                 return error("rev-list died");
255
256         /* write references */
257         argc = setup_revisions(argc, argv, &revs, NULL);
258
259         for (i = 1; i < argc; i++) {
260                 if (!strcmp(argv[i], "--stdin")) {
261                         if (read_from_stdin++)
262                                 die("--stdin given twice?");
263                         read_revisions_from_stdin(&revs);
264                         continue;
265                 }
266                 return error("unrecognized argument: %s'", argv[i]);
267         }
268
269         object_array_remove_duplicates(&revs.pending);
270
271         for (i = 0; i < revs.pending.nr; i++) {
272                 struct object_array_entry *e = revs.pending.objects + i;
273                 unsigned char sha1[20];
274                 char *ref;
275                 const char *display_ref;
276                 int flag;
277
278                 if (e->item->flags & UNINTERESTING)
279                         continue;
280                 if (dwim_ref(e->name, strlen(e->name), sha1, &ref) != 1)
281                         continue;
282                 if (!resolve_ref(e->name, sha1, 1, &flag))
283                         flag = 0;
284                 display_ref = (flag & REF_ISSYMREF) ? e->name : ref;
285
286                 if (e->item->type == OBJ_TAG &&
287                                 !is_tag_in_date_range(e->item, &revs)) {
288                         e->item->flags |= UNINTERESTING;
289                         continue;
290                 }
291
292                 /*
293                  * Make sure the refs we wrote out is correct; --max-count and
294                  * other limiting options could have prevented all the tips
295                  * from getting output.
296                  *
297                  * Non commit objects such as tags and blobs do not have
298                  * this issue as they are not affected by those extra
299                  * constraints.
300                  */
301                 if (!(e->item->flags & SHOWN) && e->item->type == OBJ_COMMIT) {
302                         warning("ref '%s' is excluded by the rev-list options",
303                                 e->name);
304                         free(ref);
305                         continue;
306                 }
307                 /*
308                  * If you run "git bundle create bndl v1.0..v2.0", the
309                  * name of the positive ref is "v2.0" but that is the
310                  * commit that is referenced by the tag, and not the tag
311                  * itself.
312                  */
313                 if (hashcmp(sha1, e->item->sha1)) {
314                         /*
315                          * Is this the positive end of a range expressed
316                          * in terms of a tag (e.g. v2.0 from the range
317                          * "v1.0..v2.0")?
318                          */
319                         struct commit *one = lookup_commit_reference(sha1);
320                         struct object *obj;
321
322                         if (e->item == &(one->object)) {
323                                 /*
324                                  * Need to include e->name as an
325                                  * independent ref to the pack-objects
326                                  * input, so that the tag is included
327                                  * in the output; otherwise we would
328                                  * end up triggering "empty bundle"
329                                  * error.
330                                  */
331                                 obj = parse_object(sha1);
332                                 obj->flags |= SHOWN;
333                                 add_pending_object(&revs, obj, e->name);
334                         }
335                         free(ref);
336                         continue;
337                 }
338
339                 ref_count++;
340                 write_or_die(bundle_fd, sha1_to_hex(e->item->sha1), 40);
341                 write_or_die(bundle_fd, " ", 1);
342                 write_or_die(bundle_fd, display_ref, strlen(display_ref));
343                 write_or_die(bundle_fd, "\n", 1);
344                 free(ref);
345         }
346         if (!ref_count)
347                 die ("Refusing to create empty bundle.");
348
349         /* end header */
350         write_or_die(bundle_fd, "\n", 1);
351
352         /* write pack */
353         argv_pack[0] = "pack-objects";
354         argv_pack[1] = "--all-progress";
355         argv_pack[2] = "--stdout";
356         argv_pack[3] = "--thin";
357         argv_pack[4] = NULL;
358         memset(&rls, 0, sizeof(rls));
359         rls.argv = argv_pack;
360         rls.in = -1;
361         rls.out = bundle_fd;
362         rls.git_cmd = 1;
363         if (start_command(&rls))
364                 return error("Could not spawn pack-objects");
365
366         /*
367          * start_command closed bundle_fd if it was > 1
368          * so set the lock fd to -1 so commit_lock_file()
369          * won't fail trying to close it.
370          */
371         lock.fd = -1;
372
373         for (i = 0; i < revs.pending.nr; i++) {
374                 struct object *object = revs.pending.objects[i].item;
375                 if (object->flags & UNINTERESTING)
376                         write_or_die(rls.in, "^", 1);
377                 write_or_die(rls.in, sha1_to_hex(object->sha1), 40);
378                 write_or_die(rls.in, "\n", 1);
379         }
380         close(rls.in);
381         if (finish_command(&rls))
382                 return error ("pack-objects died");
383         if (!bundle_to_stdout)
384                 commit_lock_file(&lock);
385         return 0;
386 }
387
388 int unbundle(struct bundle_header *header, int bundle_fd)
389 {
390         const char *argv_index_pack[] = {"index-pack",
391                 "--fix-thin", "--stdin", NULL};
392         struct child_process ip;
393
394         if (verify_bundle(header, 0))
395                 return -1;
396         memset(&ip, 0, sizeof(ip));
397         ip.argv = argv_index_pack;
398         ip.in = bundle_fd;
399         ip.no_stdout = 1;
400         ip.git_cmd = 1;
401         if (run_command(&ip))
402                 return error("index-pack died");
403         return 0;
404 }