list-objects: always parse trees gently
[git] / list-objects.c
1 #include "cache.h"
2 #include "tag.h"
3 #include "commit.h"
4 #include "tree.h"
5 #include "blob.h"
6 #include "diff.h"
7 #include "tree-walk.h"
8 #include "revision.h"
9 #include "list-objects.h"
10 #include "list-objects-filter.h"
11 #include "list-objects-filter-options.h"
12 #include "packfile.h"
13 #include "object-store.h"
14
15 struct traversal_context {
16         struct rev_info *revs;
17         show_object_fn show_object;
18         show_commit_fn show_commit;
19         void *show_data;
20         filter_object_fn filter_fn;
21         void *filter_data;
22 };
23
24 static void process_blob(struct traversal_context *ctx,
25                          struct blob *blob,
26                          struct strbuf *path,
27                          const char *name)
28 {
29         struct object *obj = &blob->object;
30         size_t pathlen;
31         enum list_objects_filter_result r = LOFR_MARK_SEEN | LOFR_DO_SHOW;
32
33         if (!ctx->revs->blob_objects)
34                 return;
35         if (!obj)
36                 die("bad blob object");
37         if (obj->flags & (UNINTERESTING | SEEN))
38                 return;
39
40         /*
41          * Pre-filter known-missing objects when explicitly requested.
42          * Otherwise, a missing object error message may be reported
43          * later (depending on other filtering criteria).
44          *
45          * Note that this "--exclude-promisor-objects" pre-filtering
46          * may cause the actual filter to report an incomplete list
47          * of missing objects.
48          */
49         if (ctx->revs->exclude_promisor_objects &&
50             !has_object_file(&obj->oid) &&
51             is_promisor_object(&obj->oid))
52                 return;
53
54         pathlen = path->len;
55         strbuf_addstr(path, name);
56         if (!(obj->flags & USER_GIVEN) && ctx->filter_fn)
57                 r = ctx->filter_fn(LOFS_BLOB, obj,
58                                    path->buf, &path->buf[pathlen],
59                                    ctx->filter_data);
60         if (r & LOFR_MARK_SEEN)
61                 obj->flags |= SEEN;
62         if (r & LOFR_DO_SHOW)
63                 ctx->show_object(obj, path->buf, ctx->show_data);
64         strbuf_setlen(path, pathlen);
65 }
66
67 /*
68  * Processing a gitlink entry currently does nothing, since
69  * we do not recurse into the subproject.
70  *
71  * We *could* eventually add a flag that actually does that,
72  * which would involve:
73  *  - is the subproject actually checked out?
74  *  - if so, see if the subproject has already been added
75  *    to the alternates list, and add it if not.
76  *  - process the commit (or tag) the gitlink points to
77  *    recursively.
78  *
79  * However, it's unclear whether there is really ever any
80  * reason to see superprojects and subprojects as such a
81  * "unified" object pool (potentially resulting in a totally
82  * humongous pack - avoiding which was the whole point of
83  * having gitlinks in the first place!).
84  *
85  * So for now, there is just a note that we *could* follow
86  * the link, and how to do it. Whether it necessarily makes
87  * any sense what-so-ever to ever do that is another issue.
88  */
89 static void process_gitlink(struct traversal_context *ctx,
90                             const unsigned char *sha1,
91                             struct strbuf *path,
92                             const char *name)
93 {
94         /* Nothing to do */
95 }
96
97 static void process_tree(struct traversal_context *ctx,
98                          struct tree *tree,
99                          struct strbuf *base,
100                          const char *name);
101
102 static void process_tree_contents(struct traversal_context *ctx,
103                                   struct tree *tree,
104                                   struct strbuf *base)
105 {
106         struct tree_desc desc;
107         struct name_entry entry;
108         enum interesting match = ctx->revs->diffopt.pathspec.nr == 0 ?
109                 all_entries_interesting : entry_not_interesting;
110
111         init_tree_desc(&desc, tree->buffer, tree->size);
112
113         while (tree_entry(&desc, &entry)) {
114                 if (match != all_entries_interesting) {
115                         match = tree_entry_interesting(&entry, base, 0,
116                                                        &ctx->revs->diffopt.pathspec);
117                         if (match == all_entries_not_interesting)
118                                 break;
119                         if (match == entry_not_interesting)
120                                 continue;
121                 }
122
123                 if (S_ISDIR(entry.mode))
124                         process_tree(ctx,
125                                      lookup_tree(the_repository, entry.oid),
126                                      base, entry.path);
127                 else if (S_ISGITLINK(entry.mode))
128                         process_gitlink(ctx, entry.oid->hash,
129                                         base, entry.path);
130                 else
131                         process_blob(ctx,
132                                      lookup_blob(the_repository, entry.oid),
133                                      base, entry.path);
134         }
135 }
136
137 static void process_tree(struct traversal_context *ctx,
138                          struct tree *tree,
139                          struct strbuf *base,
140                          const char *name)
141 {
142         struct object *obj = &tree->object;
143         struct rev_info *revs = ctx->revs;
144         int baselen = base->len;
145         enum list_objects_filter_result r = LOFR_MARK_SEEN | LOFR_DO_SHOW;
146
147         if (!revs->tree_objects)
148                 return;
149         if (!obj)
150                 die("bad tree object");
151         if (obj->flags & (UNINTERESTING | SEEN))
152                 return;
153         if (parse_tree_gently(tree, 1) < 0) {
154                 if (revs->ignore_missing_links)
155                         return;
156
157                 /*
158                  * Pre-filter known-missing tree objects when explicitly
159                  * requested.  This may cause the actual filter to report
160                  * an incomplete list of missing objects.
161                  */
162                 if (revs->exclude_promisor_objects &&
163                     is_promisor_object(&obj->oid))
164                         return;
165
166                 die("bad tree object %s", oid_to_hex(&obj->oid));
167         }
168
169         strbuf_addstr(base, name);
170         if (!(obj->flags & USER_GIVEN) && ctx->filter_fn)
171                 r = ctx->filter_fn(LOFS_BEGIN_TREE, obj,
172                                    base->buf, &base->buf[baselen],
173                                    ctx->filter_data);
174         if (r & LOFR_MARK_SEEN)
175                 obj->flags |= SEEN;
176         if (r & LOFR_DO_SHOW)
177                 ctx->show_object(obj, base->buf, ctx->show_data);
178         if (base->len)
179                 strbuf_addch(base, '/');
180
181         process_tree_contents(ctx, tree, base);
182
183         if (!(obj->flags & USER_GIVEN) && ctx->filter_fn) {
184                 r = ctx->filter_fn(LOFS_END_TREE, obj,
185                                    base->buf, &base->buf[baselen],
186                                    ctx->filter_data);
187                 if (r & LOFR_MARK_SEEN)
188                         obj->flags |= SEEN;
189                 if (r & LOFR_DO_SHOW)
190                         ctx->show_object(obj, base->buf, ctx->show_data);
191         }
192
193         strbuf_setlen(base, baselen);
194         free_tree_buffer(tree);
195 }
196
197 static void mark_edge_parents_uninteresting(struct commit *commit,
198                                             struct rev_info *revs,
199                                             show_edge_fn show_edge)
200 {
201         struct commit_list *parents;
202
203         for (parents = commit->parents; parents; parents = parents->next) {
204                 struct commit *parent = parents->item;
205                 if (!(parent->object.flags & UNINTERESTING))
206                         continue;
207                 mark_tree_uninteresting(get_commit_tree(parent));
208                 if (revs->edge_hint && !(parent->object.flags & SHOWN)) {
209                         parent->object.flags |= SHOWN;
210                         show_edge(parent);
211                 }
212         }
213 }
214
215 void mark_edges_uninteresting(struct rev_info *revs, show_edge_fn show_edge)
216 {
217         struct commit_list *list;
218         int i;
219
220         for (list = revs->commits; list; list = list->next) {
221                 struct commit *commit = list->item;
222
223                 if (commit->object.flags & UNINTERESTING) {
224                         mark_tree_uninteresting(get_commit_tree(commit));
225                         if (revs->edge_hint_aggressive && !(commit->object.flags & SHOWN)) {
226                                 commit->object.flags |= SHOWN;
227                                 show_edge(commit);
228                         }
229                         continue;
230                 }
231                 mark_edge_parents_uninteresting(commit, revs, show_edge);
232         }
233         if (revs->edge_hint_aggressive) {
234                 for (i = 0; i < revs->cmdline.nr; i++) {
235                         struct object *obj = revs->cmdline.rev[i].item;
236                         struct commit *commit = (struct commit *)obj;
237                         if (obj->type != OBJ_COMMIT || !(obj->flags & UNINTERESTING))
238                                 continue;
239                         mark_tree_uninteresting(get_commit_tree(commit));
240                         if (!(obj->flags & SHOWN)) {
241                                 obj->flags |= SHOWN;
242                                 show_edge(commit);
243                         }
244                 }
245         }
246 }
247
248 static void add_pending_tree(struct rev_info *revs, struct tree *tree)
249 {
250         add_pending_object(revs, &tree->object, "");
251 }
252
253 static void traverse_trees_and_blobs(struct traversal_context *ctx,
254                                      struct strbuf *base)
255 {
256         int i;
257
258         assert(base->len == 0);
259
260         for (i = 0; i < ctx->revs->pending.nr; i++) {
261                 struct object_array_entry *pending = ctx->revs->pending.objects + i;
262                 struct object *obj = pending->item;
263                 const char *name = pending->name;
264                 const char *path = pending->path;
265                 if (obj->flags & (UNINTERESTING | SEEN))
266                         continue;
267                 if (obj->type == OBJ_TAG) {
268                         obj->flags |= SEEN;
269                         ctx->show_object(obj, name, ctx->show_data);
270                         continue;
271                 }
272                 if (!path)
273                         path = "";
274                 if (obj->type == OBJ_TREE) {
275                         process_tree(ctx, (struct tree *)obj, base, path);
276                         continue;
277                 }
278                 if (obj->type == OBJ_BLOB) {
279                         process_blob(ctx, (struct blob *)obj, base, path);
280                         continue;
281                 }
282                 die("unknown pending object %s (%s)",
283                     oid_to_hex(&obj->oid), name);
284         }
285         object_array_clear(&ctx->revs->pending);
286 }
287
288 static void do_traverse(struct traversal_context *ctx)
289 {
290         struct commit *commit;
291         struct strbuf csp; /* callee's scratch pad */
292         strbuf_init(&csp, PATH_MAX);
293
294         while ((commit = get_revision(ctx->revs)) != NULL) {
295                 /*
296                  * an uninteresting boundary commit may not have its tree
297                  * parsed yet, but we are not going to show them anyway
298                  */
299                 if (get_commit_tree(commit))
300                         add_pending_tree(ctx->revs, get_commit_tree(commit));
301                 ctx->show_commit(commit, ctx->show_data);
302
303                 if (ctx->revs->tree_blobs_in_commit_order)
304                         /*
305                          * NEEDSWORK: Adding the tree and then flushing it here
306                          * needs a reallocation for each commit. Can we pass the
307                          * tree directory without allocation churn?
308                          */
309                         traverse_trees_and_blobs(ctx, &csp);
310         }
311         traverse_trees_and_blobs(ctx, &csp);
312         strbuf_release(&csp);
313 }
314
315 void traverse_commit_list(struct rev_info *revs,
316                           show_commit_fn show_commit,
317                           show_object_fn show_object,
318                           void *show_data)
319 {
320         struct traversal_context ctx;
321         ctx.revs = revs;
322         ctx.show_commit = show_commit;
323         ctx.show_object = show_object;
324         ctx.show_data = show_data;
325         ctx.filter_fn = NULL;
326         ctx.filter_data = NULL;
327         do_traverse(&ctx);
328 }
329
330 void traverse_commit_list_filtered(
331         struct list_objects_filter_options *filter_options,
332         struct rev_info *revs,
333         show_commit_fn show_commit,
334         show_object_fn show_object,
335         void *show_data,
336         struct oidset *omitted)
337 {
338         struct traversal_context ctx;
339         filter_free_fn filter_free_fn = NULL;
340
341         ctx.revs = revs;
342         ctx.show_object = show_object;
343         ctx.show_commit = show_commit;
344         ctx.show_data = show_data;
345         ctx.filter_fn = NULL;
346
347         ctx.filter_data = list_objects_filter__init(omitted, filter_options,
348                                                     &ctx.filter_fn, &filter_free_fn);
349         do_traverse(&ctx);
350         if (ctx.filter_data && filter_free_fn)
351                 filter_free_fn(ctx.filter_data);
352 }