Merge branch 'md/filter-trees'
authorJunio C Hamano <gitster@pobox.com>
Tue, 30 Oct 2018 06:43:39 +0000 (15:43 +0900)
committerJunio C Hamano <gitster@pobox.com>
Tue, 30 Oct 2018 06:43:39 +0000 (15:43 +0900)
The "rev-list --filter" feature learned to exclude all trees via
"tree:0" filter.

* md/filter-trees:
list-objects: support for skipping tree traversal
filter-trees: code clean-up of tests
list-objects-filter: implement filter tree:0
list-objects-filter-options: do not over-strbuf_init
list-objects-filter: use BUG rather than die
revision: mark non-user-given objects instead
rev-list: handle missing tree objects properly
list-objects: always parse trees gently
list-objects: refactor to process_tree_contents
list-objects: store common func args in struct

13 files changed:
Documentation/rev-list-options.txt
builtin/rev-list.c
list-objects-filter-options.c
list-objects-filter-options.h
list-objects-filter.c
list-objects-filter.h
list-objects.c
revision.c
revision.h
t/t0410-partial-clone.sh
t/t5317-pack-objects-filter-objects.sh
t/t5616-partial-clone.sh
t/t6112-rev-list-filters-objects.sh
index 7b273635de2b5bf3e3ba6ade8bcca3068a216395..5f1672913b8ab19f6b2cf20d5fa79cfe6b636356 100644 (file)
@@ -731,6 +731,11 @@ the requested refs.
 +
 The form '--filter=sparse:path=<path>' similarly uses a sparse-checkout
 specification contained in <path>.
++
+The form '--filter=tree:<depth>' omits all blobs and trees whose depth
+from the root tree is >= <depth> (minimum depth if an object is located
+at multiple depths in the commits traversed). Currently, only <depth>=0
+is supported, which omits all blobs and trees.
 
 --no-filter::
        Turn off any previous `--filter=` argument.
index cc1b70522f7bcdb77fd33ccd9d9948926b8333ad..5064d08e1b8ad04544a76d1f0496134e2c15079c 100644 (file)
@@ -6,6 +6,7 @@
 #include "list-objects.h"
 #include "list-objects-filter.h"
 #include "list-objects-filter-options.h"
+#include "object.h"
 #include "object-store.h"
 #include "pack.h"
 #include "pack-bitmap.h"
@@ -209,7 +210,8 @@ static inline void finish_object__ma(struct object *obj)
         */
        switch (arg_missing_action) {
        case MA_ERROR:
-               die("missing blob object '%s'", oid_to_hex(&obj->oid));
+               die("missing %s object '%s'",
+                   type_name(obj->type), oid_to_hex(&obj->oid));
                return;
 
        case MA_ALLOW_ANY:
@@ -222,8 +224,8 @@ static inline void finish_object__ma(struct object *obj)
        case MA_ALLOW_PROMISOR:
                if (is_promisor_object(&obj->oid))
                        return;
-               die("unexpected missing blob object '%s'",
-                   oid_to_hex(&obj->oid));
+               die("unexpected missing %s object '%s'",
+                   type_name(obj->type), oid_to_hex(&obj->oid));
                return;
 
        default:
@@ -235,7 +237,7 @@ static inline void finish_object__ma(struct object *obj)
 static int finish_object(struct object *obj, const char *name, void *cb_data)
 {
        struct rev_list_info *info = cb_data;
-       if (obj->type == OBJ_BLOB && !has_object_file(&obj->oid)) {
+       if (!has_object_file(&obj->oid)) {
                finish_object__ma(obj);
                return 1;
        }
@@ -373,6 +375,7 @@ int cmd_rev_list(int argc, const char **argv, const char *prefix)
        repo_init_revisions(the_repository, &revs, prefix);
        revs.abbrev = DEFAULT_ABBREV;
        revs.commit_format = CMIT_FMT_UNSPECIFIED;
+       revs.do_not_die_on_missing_tree = 1;
 
        /*
         * Scan the argument list before invoking setup_revisions(), so that we
index c0e2bd6a06a27b6056364c43b003d74539fd6cad..e8da2e8581b2df956411f19e796ceb94adcef903 100644 (file)
@@ -30,7 +30,6 @@ static int gently_parse_list_objects_filter(
 
        if (filter_options->choice) {
                if (errbuf) {
-                       strbuf_init(errbuf, 0);
                        strbuf_addstr(
                                errbuf,
                                _("multiple filter-specs cannot be combined"));
@@ -50,6 +49,19 @@ static int gently_parse_list_objects_filter(
                        return 0;
                }
 
+       } else if (skip_prefix(arg, "tree:", &v0)) {
+               unsigned long depth;
+               if (!git_parse_ulong(v0, &depth) || depth != 0) {
+                       if (errbuf) {
+                               strbuf_addstr(
+                                       errbuf,
+                                       _("only 'tree:0' is supported"));
+                       }
+                       return 1;
+               }
+               filter_options->choice = LOFC_TREE_NONE;
+               return 0;
+
        } else if (skip_prefix(arg, "sparse:oid=", &v0)) {
                struct object_context oc;
                struct object_id sparse_oid;
@@ -71,10 +83,9 @@ static int gently_parse_list_objects_filter(
                return 0;
        }
 
-       if (errbuf) {
-               strbuf_init(errbuf, 0);
+       if (errbuf)
                strbuf_addf(errbuf, "invalid filter-spec '%s'", arg);
-       }
+
        memset(filter_options, 0, sizeof(*filter_options));
        return 1;
 }
index 0000a61f82d3dcc36e190b5422a13c93d47f15c2..af64e5c66fc4a242f9cc534380a9b3e8f1e67d00 100644 (file)
@@ -10,6 +10,7 @@ enum list_objects_filter_choice {
        LOFC_DISABLED = 0,
        LOFC_BLOB_NONE,
        LOFC_BLOB_LIMIT,
+       LOFC_TREE_NONE,
        LOFC_SPARSE_OID,
        LOFC_SPARSE_PATH,
        LOFC__COUNT /* must be last */
index a0ba78b20cc99bfcd2c41abd4f312b5649a2b9cd..765f3df3b067c5df7b98402ef4fc953edcd9179b 100644 (file)
@@ -44,8 +44,7 @@ static enum list_objects_filter_result filter_blobs_none(
 
        switch (filter_situation) {
        default:
-               die("unknown filter_situation");
-               return LOFR_ZERO;
+               BUG("unknown filter_situation: %d", filter_situation);
 
        case LOFS_BEGIN_TREE:
                assert(obj->type == OBJ_TREE);
@@ -80,6 +79,61 @@ static void *filter_blobs_none__init(
        return d;
 }
 
+/*
+ * A filter for list-objects to omit ALL trees and blobs from the traversal.
+ * Can OPTIONALLY collect a list of the omitted OIDs.
+ */
+struct filter_trees_none_data {
+       struct oidset *omits;
+};
+
+static enum list_objects_filter_result filter_trees_none(
+       enum list_objects_filter_situation filter_situation,
+       struct object *obj,
+       const char *pathname,
+       const char *filename,
+       void *filter_data_)
+{
+       struct filter_trees_none_data *filter_data = filter_data_;
+
+       switch (filter_situation) {
+       default:
+               BUG("unknown filter_situation: %d", filter_situation);
+
+       case LOFS_BEGIN_TREE:
+       case LOFS_BLOB:
+               if (filter_data->omits) {
+                       oidset_insert(filter_data->omits, &obj->oid);
+                       /* _MARK_SEEN but not _DO_SHOW (hard omit) */
+                       return LOFR_MARK_SEEN;
+               } else {
+                       /*
+                        * Not collecting omits so no need to to traverse tree.
+                        */
+                       return LOFR_SKIP_TREE | LOFR_MARK_SEEN;
+               }
+
+       case LOFS_END_TREE:
+               assert(obj->type == OBJ_TREE);
+               return LOFR_ZERO;
+
+       }
+}
+
+static void* filter_trees_none__init(
+       struct oidset *omitted,
+       struct list_objects_filter_options *filter_options,
+       filter_object_fn *filter_fn,
+       filter_free_fn *filter_free_fn)
+{
+       struct filter_trees_none_data *d = xcalloc(1, sizeof(*d));
+       d->omits = omitted;
+
+       *filter_fn = filter_trees_none;
+       *filter_free_fn = free;
+       return d;
+}
+
 /*
  * A filter for list-objects to omit large blobs.
  * And to OPTIONALLY collect a list of the omitted OIDs.
@@ -102,8 +156,7 @@ static enum list_objects_filter_result filter_blobs_limit(
 
        switch (filter_situation) {
        default:
-               die("unknown filter_situation");
-               return LOFR_ZERO;
+               BUG("unknown filter_situation: %d", filter_situation);
 
        case LOFS_BEGIN_TREE:
                assert(obj->type == OBJ_TREE);
@@ -208,8 +261,7 @@ static enum list_objects_filter_result filter_sparse(
 
        switch (filter_situation) {
        default:
-               die("unknown filter_situation");
-               return LOFR_ZERO;
+               BUG("unknown filter_situation: %d", filter_situation);
 
        case LOFS_BEGIN_TREE:
                assert(obj->type == OBJ_TREE);
@@ -374,6 +426,7 @@ static filter_init_fn s_filters[] = {
        NULL,
        filter_blobs_none__init,
        filter_blobs_limit__init,
+       filter_trees_none__init,
        filter_sparse_oid__init,
        filter_sparse_path__init,
 };
@@ -389,7 +442,7 @@ void *list_objects_filter__init(
        assert((sizeof(s_filters) / sizeof(s_filters[0])) == LOFC__COUNT);
 
        if (filter_options->choice >= LOFC__COUNT)
-               die("invalid list-objects filter choice: %d",
+               BUG("invalid list-objects filter choice: %d",
                    filter_options->choice);
 
        init_fn = s_filters[filter_options->choice];
index a6f6b4990b43c8f4c8cb94ba5b890370287ec868..52b4a84da9418497a4fc37e50cb8e0dcbc5b70bd 100644 (file)
@@ -24,6 +24,11 @@ struct oidset;
  *              In general, objects should only be shown once, but
  *              this result DOES NOT imply that we mark it SEEN.
  *
+ * _SKIP_TREE : Used in LOFS_BEGIN_TREE situation - indicates that
+ *              the tree's children should not be iterated over. This
+ *              is used as an optimization when all children will
+ *              definitely be ignored.
+ *
  * Most of the time, you want the combination (_MARK_SEEN | _DO_SHOW)
  * but they can be used independently, such as when sparse-checkout
  * pattern matching is being applied.
@@ -45,6 +50,7 @@ enum list_objects_filter_result {
        LOFR_ZERO      = 0,
        LOFR_MARK_SEEN = 1<<0,
        LOFR_DO_SHOW   = 1<<1,
+       LOFR_SKIP_TREE = 1<<2,
 };
 
 enum list_objects_filter_situation {
index 0c2989d5ca7cc4df523376dce94829b1b6be9110..c41cc80db5bc86279bcf560109f599a9783b2755 100644 (file)
 #include "list-objects-filter-options.h"
 #include "packfile.h"
 #include "object-store.h"
+#include "trace.h"
 
-static void process_blob(struct rev_info *revs,
+struct traversal_context {
+       struct rev_info *revs;
+       show_object_fn show_object;
+       show_commit_fn show_commit;
+       void *show_data;
+       filter_object_fn filter_fn;
+       void *filter_data;
+};
+
+static void process_blob(struct traversal_context *ctx,
                         struct blob *blob,
-                        show_object_fn show,
                         struct strbuf *path,
-                        const char *name,
-                        void *cb_data,
-                        filter_object_fn filter_fn,
-                        void *filter_data)
+                        const char *name)
 {
        struct object *obj = &blob->object;
        size_t pathlen;
        enum list_objects_filter_result r = LOFR_MARK_SEEN | LOFR_DO_SHOW;
 
-       if (!revs->blob_objects)
+       if (!ctx->revs->blob_objects)
                return;
        if (!obj)
                die("bad blob object");
@@ -41,21 +47,21 @@ static void process_blob(struct rev_info *revs,
         * may cause the actual filter to report an incomplete list
         * of missing objects.
         */
-       if (revs->exclude_promisor_objects &&
+       if (ctx->revs->exclude_promisor_objects &&
            !has_object_file(&obj->oid) &&
            is_promisor_object(&obj->oid))
                return;
 
        pathlen = path->len;
        strbuf_addstr(path, name);
-       if (!(obj->flags & USER_GIVEN) && filter_fn)
-               r = filter_fn(LOFS_BLOB, obj,
-                             path->buf, &path->buf[pathlen],
-                             filter_data);
+       if ((obj->flags & NOT_USER_GIVEN) && ctx->filter_fn)
+               r = ctx->filter_fn(LOFS_BLOB, obj,
+                                  path->buf, &path->buf[pathlen],
+                                  ctx->filter_data);
        if (r & LOFR_MARK_SEEN)
                obj->flags |= SEEN;
        if (r & LOFR_DO_SHOW)
-               show(obj, path->buf, cb_data);
+               ctx->show_object(obj, path->buf, ctx->show_data);
        strbuf_setlen(path, pathlen);
 }
 
@@ -81,34 +87,66 @@ static void process_blob(struct rev_info *revs,
  * the link, and how to do it. Whether it necessarily makes
  * any sense what-so-ever to ever do that is another issue.
  */
-static void process_gitlink(struct rev_info *revs,
+static void process_gitlink(struct traversal_context *ctx,
                            const unsigned char *sha1,
-                           show_object_fn show,
                            struct strbuf *path,
-                           const char *name,
-                           void *cb_data)
+                           const char *name)
 {
        /* Nothing to do */
 }
 
-static void process_tree(struct rev_info *revs,
+static void process_tree(struct traversal_context *ctx,
                         struct tree *tree,
-                        show_object_fn show,
                         struct strbuf *base,
-                        const char *name,
-                        void *cb_data,
-                        filter_object_fn filter_fn,
-                        void *filter_data)
+                        const char *name);
+
+static void process_tree_contents(struct traversal_context *ctx,
+                                 struct tree *tree,
+                                 struct strbuf *base)
 {
-       struct object *obj = &tree->object;
        struct tree_desc desc;
        struct name_entry entry;
-       enum interesting match = revs->diffopt.pathspec.nr == 0 ?
-               all_entries_interesting: entry_not_interesting;
+       enum interesting match = ctx->revs->diffopt.pathspec.nr == 0 ?
+               all_entries_interesting : entry_not_interesting;
+
+       init_tree_desc(&desc, tree->buffer, tree->size);
+
+       while (tree_entry(&desc, &entry)) {
+               if (match != all_entries_interesting) {
+                       match = tree_entry_interesting(&entry, base, 0,
+                                                      &ctx->revs->diffopt.pathspec);
+                       if (match == all_entries_not_interesting)
+                               break;
+                       if (match == entry_not_interesting)
+                               continue;
+               }
+
+               if (S_ISDIR(entry.mode)) {
+                       struct tree *t = lookup_tree(the_repository, entry.oid);
+                       t->object.flags |= NOT_USER_GIVEN;
+                       process_tree(ctx, t, base, entry.path);
+               }
+               else if (S_ISGITLINK(entry.mode))
+                       process_gitlink(ctx, entry.oid->hash,
+                                       base, entry.path);
+               else {
+                       struct blob *b = lookup_blob(the_repository, entry.oid);
+                       b->object.flags |= NOT_USER_GIVEN;
+                       process_blob(ctx, b, base, entry.path);
+               }
+       }
+}
+
+static void process_tree(struct traversal_context *ctx,
+                        struct tree *tree,
+                        struct strbuf *base,
+                        const char *name)
+{
+       struct object *obj = &tree->object;
+       struct rev_info *revs = ctx->revs;
        int baselen = base->len;
        enum list_objects_filter_result r = LOFR_MARK_SEEN | LOFR_DO_SHOW;
-       int gently = revs->ignore_missing_links ||
-                    revs->exclude_promisor_objects;
+       int failed_parse;
 
        if (!revs->tree_objects)
                return;
@@ -116,7 +154,9 @@ static void process_tree(struct rev_info *revs,
                die("bad tree object");
        if (obj->flags & (UNINTERESTING | SEEN))
                return;
-       if (parse_tree_gently(tree, gently) < 0) {
+
+       failed_parse = parse_tree_gently(tree, 1);
+       if (failed_parse) {
                if (revs->ignore_missing_links)
                        return;
 
@@ -129,57 +169,35 @@ static void process_tree(struct rev_info *revs,
                    is_promisor_object(&obj->oid))
                        return;
 
-               die("bad tree object %s", oid_to_hex(&obj->oid));
+               if (!revs->do_not_die_on_missing_tree)
+                       die("bad tree object %s", oid_to_hex(&obj->oid));
        }
 
        strbuf_addstr(base, name);
-       if (!(obj->flags & USER_GIVEN) && filter_fn)
-               r = filter_fn(LOFS_BEGIN_TREE, obj,
-                             base->buf, &base->buf[baselen],
-                             filter_data);
+       if ((obj->flags & NOT_USER_GIVEN) && ctx->filter_fn)
+               r = ctx->filter_fn(LOFS_BEGIN_TREE, obj,
+                                  base->buf, &base->buf[baselen],
+                                  ctx->filter_data);
        if (r & LOFR_MARK_SEEN)
                obj->flags |= SEEN;
        if (r & LOFR_DO_SHOW)
-               show(obj, base->buf, cb_data);
+               ctx->show_object(obj, base->buf, ctx->show_data);
        if (base->len)
                strbuf_addch(base, '/');
 
-       init_tree_desc(&desc, tree->buffer, tree->size);
+       if (r & LOFR_SKIP_TREE)
+               trace_printf("Skipping contents of tree %s...\n", base->buf);
+       else if (!failed_parse)
+               process_tree_contents(ctx, tree, base);
 
-       while (tree_entry(&desc, &entry)) {
-               if (match != all_entries_interesting) {
-                       match = tree_entry_interesting(&entry, base, 0,
-                                                      &revs->diffopt.pathspec);
-                       if (match == all_entries_not_interesting)
-                               break;
-                       if (match == entry_not_interesting)
-                               continue;
-               }
-
-               if (S_ISDIR(entry.mode))
-                       process_tree(revs,
-                                    lookup_tree(the_repository, entry.oid),
-                                    show, base, entry.path,
-                                    cb_data, filter_fn, filter_data);
-               else if (S_ISGITLINK(entry.mode))
-                       process_gitlink(revs, entry.oid->hash,
-                                       show, base, entry.path,
-                                       cb_data);
-               else
-                       process_blob(revs,
-                                    lookup_blob(the_repository, entry.oid),
-                                    show, base, entry.path,
-                                    cb_data, filter_fn, filter_data);
-       }
-
-       if (!(obj->flags & USER_GIVEN) && filter_fn) {
-               r = filter_fn(LOFS_END_TREE, obj,
-                             base->buf, &base->buf[baselen],
-                             filter_data);
+       if ((obj->flags & NOT_USER_GIVEN) && ctx->filter_fn) {
+               r = ctx->filter_fn(LOFS_END_TREE, obj,
+                                  base->buf, &base->buf[baselen],
+                                  ctx->filter_data);
                if (r & LOFR_MARK_SEEN)
                        obj->flags |= SEEN;
                if (r & LOFR_DO_SHOW)
-                       show(obj, base->buf, cb_data);
+                       ctx->show_object(obj, base->buf, ctx->show_data);
        }
 
        strbuf_setlen(base, baselen);
@@ -244,19 +262,15 @@ static void add_pending_tree(struct rev_info *revs, struct tree *tree)
        add_pending_object(revs, &tree->object, "");
 }
 
-static void traverse_trees_and_blobs(struct rev_info *revs,
-                                    struct strbuf *base,
-                                    show_object_fn show_object,
-                                    void *show_data,
-                                    filter_object_fn filter_fn,
-                                    void *filter_data)
+static void traverse_trees_and_blobs(struct traversal_context *ctx,
+                                    struct strbuf *base)
 {
        int i;
 
        assert(base->len == 0);
 
-       for (i = 0; i < revs->pending.nr; i++) {
-               struct object_array_entry *pending = revs->pending.objects + i;
+       for (i = 0; i < ctx->revs->pending.nr; i++) {
+               struct object_array_entry *pending = ctx->revs->pending.objects + i;
                struct object *obj = pending->item;
                const char *name = pending->name;
                const char *path = pending->path;
@@ -264,62 +278,52 @@ static void traverse_trees_and_blobs(struct rev_info *revs,
                        continue;
                if (obj->type == OBJ_TAG) {
                        obj->flags |= SEEN;
-                       show_object(obj, name, show_data);
+                       ctx->show_object(obj, name, ctx->show_data);
                        continue;
                }
                if (!path)
                        path = "";
                if (obj->type == OBJ_TREE) {
-                       process_tree(revs, (struct tree *)obj, show_object,
-                                    base, path, show_data,
-                                    filter_fn, filter_data);
+                       process_tree(ctx, (struct tree *)obj, base, path);
                        continue;
                }
                if (obj->type == OBJ_BLOB) {
-                       process_blob(revs, (struct blob *)obj, show_object,
-                                    base, path, show_data,
-                                    filter_fn, filter_data);
+                       process_blob(ctx, (struct blob *)obj, base, path);
                        continue;
                }
                die("unknown pending object %s (%s)",
                    oid_to_hex(&obj->oid), name);
        }
-       object_array_clear(&revs->pending);
+       object_array_clear(&ctx->revs->pending);
 }
 
-static void do_traverse(struct rev_info *revs,
-                       show_commit_fn show_commit,
-                       show_object_fn show_object,
-                       void *show_data,
-                       filter_object_fn filter_fn,
-                       void *filter_data)
+static void do_traverse(struct traversal_context *ctx)
 {
        struct commit *commit;
        struct strbuf csp; /* callee's scratch pad */
        strbuf_init(&csp, PATH_MAX);
 
-       while ((commit = get_revision(revs)) != NULL) {
+       while ((commit = get_revision(ctx->revs)) != NULL) {
                /*
                 * an uninteresting boundary commit may not have its tree
                 * parsed yet, but we are not going to show them anyway
                 */
-               if (get_commit_tree(commit))
-                       add_pending_tree(revs, get_commit_tree(commit));
-               show_commit(commit, show_data);
+               if (get_commit_tree(commit)) {
+                       struct tree *tree = get_commit_tree(commit);
+                       tree->object.flags |= NOT_USER_GIVEN;
+                       add_pending_tree(ctx->revs, tree);
+               }
+               ctx->show_commit(commit, ctx->show_data);
 
-               if (revs->tree_blobs_in_commit_order)
+               if (ctx->revs->tree_blobs_in_commit_order)
                        /*
                         * NEEDSWORK: Adding the tree and then flushing it here
                         * needs a reallocation for each commit. Can we pass the
                         * tree directory without allocation churn?
                         */
-                       traverse_trees_and_blobs(revs, &csp,
-                                                show_object, show_data,
-                                                filter_fn, filter_data);
+                       traverse_trees_and_blobs(ctx, &csp);
        }
-       traverse_trees_and_blobs(revs, &csp,
-                                show_object, show_data,
-                                filter_fn, filter_data);
+       traverse_trees_and_blobs(ctx, &csp);
        strbuf_release(&csp);
 }
 
@@ -328,7 +332,14 @@ void traverse_commit_list(struct rev_info *revs,
                          show_object_fn show_object,
                          void *show_data)
 {
-       do_traverse(revs, show_commit, show_object, show_data, NULL, NULL);
+       struct traversal_context ctx;
+       ctx.revs = revs;
+       ctx.show_commit = show_commit;
+       ctx.show_object = show_object;
+       ctx.show_data = show_data;
+       ctx.filter_fn = NULL;
+       ctx.filter_data = NULL;
+       do_traverse(&ctx);
 }
 
 void traverse_commit_list_filtered(
@@ -339,14 +350,18 @@ void traverse_commit_list_filtered(
        void *show_data,
        struct oidset *omitted)
 {
-       filter_object_fn filter_fn = NULL;
+       struct traversal_context ctx;
        filter_free_fn filter_free_fn = NULL;
-       void *filter_data = NULL;
-
-       filter_data = list_objects_filter__init(omitted, filter_options,
-                                               &filter_fn, &filter_free_fn);
-       do_traverse(revs, show_commit, show_object, show_data,
-                   filter_fn, filter_data);
-       if (filter_data && filter_free_fn)
-               filter_free_fn(filter_data);
+
+       ctx.revs = revs;
+       ctx.show_object = show_object;
+       ctx.show_commit = show_commit;
+       ctx.show_data = show_data;
+       ctx.filter_fn = NULL;
+
+       ctx.filter_data = list_objects_filter__init(omitted, filter_options,
+                                                   &ctx.filter_fn, &filter_free_fn);
+       do_traverse(&ctx);
+       if (ctx.filter_data && filter_free_fn)
+               filter_free_fn(ctx.filter_data);
 }
index b5108b75abc9ff0ef0438f4cba2e932e935ff94f..a1ddb9e11cbe3a52bb8d3eee3785524db4055854 100644 (file)
@@ -177,7 +177,6 @@ static void add_pending_object_with_path(struct rev_info *revs,
                strbuf_release(&buf);
                return; /* do not add the commit itself */
        }
-       obj->flags |= USER_GIVEN;
        add_object_array_with_path(obj, name, &revs->pending, mode, path);
 }
 
index bc30a3023e20542480b7d771dd1e4001c518f30b..1cd0c4b200887e6b73d6a2473712bb5832f7e079 100644 (file)
 #define SYMMETRIC_LEFT (1u<<8)
 #define PATCHSAME      (1u<<9)
 #define BOTTOM         (1u<<10)
-#define USER_GIVEN     (1u<<25) /* given directly by the user */
+/*
+ * Indicates object was reached by traversal. i.e. not given by user on
+ * command-line or stdin.
+ * NEEDSWORK: NOT_USER_GIVEN doesn't apply to commits because we only support
+ * filtering trees and blobs, but it may be useful to support filtering commits
+ * in the future.
+ */
+#define NOT_USER_GIVEN (1u<<25)
 #define TRACK_LINEAR   (1u<<26)
-#define ALL_REV_FLAGS  (((1u<<11)-1) | USER_GIVEN | TRACK_LINEAR)
+#define ALL_REV_FLAGS  (((1u<<11)-1) | NOT_USER_GIVEN | TRACK_LINEAR)
 
 #define DECORATE_SHORT_REFS    1
 #define DECORATE_FULL_REFS     2
@@ -133,6 +140,21 @@ struct rev_info {
                        line_level_traverse:1,
                        tree_blobs_in_commit_order:1,
 
+                       /*
+                        * Blobs are shown without regard for their existence.
+                        * But not so for trees: unless exclude_promisor_objects
+                        * is set and the tree in question is a promisor object;
+                        * OR ignore_missing_links is set, the revision walker
+                        * dies with a "bad tree object HASH" message when
+                        * encountering a missing tree. For callers that can
+                        * handle missing trees and want them to be filterable
+                        * and showable, set this to true. The revision walker
+                        * will filter and show such a missing tree as usual,
+                        * but will not attempt to recurse into this tree
+                        * object.
+                        */
+                       do_not_die_on_missing_tree:1,
+
                        /* for internal use only */
                        exclude_promisor_objects:1;
 
index c521d7d6c61f3e3c76d5c4d51440ae114004a362..ba3887f178b03a71b1ee1d2149429e2ee76814a6 100755 (executable)
@@ -239,6 +239,51 @@ test_expect_success 'rev-list stops traversal at missing and promised commit' '
        ! grep $FOO out
 '
 
+test_expect_success 'missing tree objects with --missing=allow-promisor and --exclude-promisor-objects' '
+       rm -rf repo &&
+       test_create_repo repo &&
+       test_commit -C repo foo &&
+       test_commit -C repo bar &&
+       test_commit -C repo baz &&
+
+       promise_and_delete $(git -C repo rev-parse bar^{tree}) &&
+       promise_and_delete $(git -C repo rev-parse foo^{tree}) &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+
+       git -C repo rev-list --missing=allow-promisor --objects HEAD >objs 2>rev_list_err &&
+       test_must_be_empty rev_list_err &&
+       # 3 commits, 3 blobs, and 1 tree
+       test_line_count = 7 objs &&
+
+       # Do the same for --exclude-promisor-objects, but with all trees gone.
+       promise_and_delete $(git -C repo rev-parse baz^{tree}) &&
+       git -C repo rev-list --exclude-promisor-objects --objects HEAD >objs 2>rev_list_err &&
+       test_must_be_empty rev_list_err &&
+       # 3 commits, no blobs or trees
+       test_line_count = 3 objs
+'
+
+test_expect_success 'missing non-root tree object and rev-list' '
+       rm -rf repo &&
+       test_create_repo repo &&
+       mkdir repo/dir &&
+       echo foo >repo/dir/foo &&
+       git -C repo add dir/foo &&
+       git -C repo commit -m "commit dir/foo" &&
+
+       promise_and_delete $(git -C repo rev-parse HEAD:dir) &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+
+       git -C repo rev-list --missing=allow-any --objects HEAD >objs 2>rev_list_err &&
+       test_must_be_empty rev_list_err &&
+       # 1 commit and 1 tree
+       test_line_count = 2 objs
+'
+
 test_expect_success 'rev-list stops traversal at missing and promised tree' '
        rm -rf repo &&
        test_create_repo repo &&
index 2e718f0bdede397d399399536707743f9f7cff34..24541ea137517354a8b738bdef34728b49f2850e 100755 (executable)
@@ -67,6 +67,47 @@ test_expect_success 'verify normal and blob:none packfiles have same commits/tre
        test_cmp expected observed
 '
 
+test_expect_success 'get an error for missing tree object' '
+       git init r5 &&
+       echo foo >r5/foo &&
+       git -C r5 add foo &&
+       git -C r5 commit -m "foo" &&
+       del=$(git -C r5 rev-parse HEAD^{tree} | sed "s|..|&/|") &&
+       rm r5/.git/objects/$del &&
+       test_must_fail git -C r5 pack-objects --rev --stdout 2>bad_tree <<-EOF &&
+       HEAD
+       EOF
+       grep "bad tree object" bad_tree
+'
+
+test_expect_success 'setup for tests of tree:0' '
+       mkdir r1/subtree &&
+       echo "This is a file in a subtree" >r1/subtree/file &&
+       git -C r1 add subtree/file &&
+       git -C r1 commit -m subtree
+'
+
+test_expect_success 'verify tree:0 packfile has no blobs or trees' '
+       git -C r1 pack-objects --rev --stdout --filter=tree:0 >commitsonly.pack <<-EOF &&
+       HEAD
+       EOF
+       git -C r1 index-pack ../commitsonly.pack &&
+       git -C r1 verify-pack -v ../commitsonly.pack >objs &&
+       ! grep -E "tree|blob" objs
+'
+
+test_expect_success 'grab tree directly when using tree:0' '
+       # We should get the tree specified directly but not its blobs or subtrees.
+       git -C r1 pack-objects --rev --stdout --filter=tree:0 >commitsonly.pack <<-EOF &&
+       HEAD:
+       EOF
+       git -C r1 index-pack ../commitsonly.pack &&
+       git -C r1 verify-pack -v ../commitsonly.pack >objs &&
+       awk "/tree|blob/{print \$1}" objs >trees_and_blobs &&
+       git -C r1 rev-parse HEAD: >expected &&
+       test_cmp expected trees_and_blobs
+'
+
 # Test blob:limit=<n>[kmg] filter.
 # We boundary test around the size parameter.  The filter is strictly less than
 # the value, so size 500 and 1000 should have the same results, but 1001 should
index 6391437529ca7c4f742d2668fdefc77df1112743..336f02a41a66b48269e8eb6030d5323cffb8b686 100755 (executable)
@@ -166,6 +166,48 @@ test_expect_success 'partial clone with transfer.fsckobjects=1 uses index-pack -
        grep "git index-pack.*--fsck-objects" trace
 '
 
+test_expect_success 'use fsck before and after manually fetching a missing subtree' '
+       # push new commit so server has a subtree
+       mkdir src/dir &&
+       echo "in dir" >src/dir/file.txt &&
+       git -C src add dir/file.txt &&
+       git -C src commit -m "file in dir" &&
+       git -C src push -u srv master &&
+       SUBTREE=$(git -C src rev-parse HEAD:dir) &&
+
+       rm -rf dst &&
+       git clone --no-checkout --filter=tree:0 "file://$(pwd)/srv.bare" dst &&
+       git -C dst fsck &&
+
+       # Make sure we only have commits, and all trees and blobs are missing.
+       git -C dst rev-list --missing=allow-any --objects master \
+               >fetched_objects &&
+       awk -f print_1.awk fetched_objects |
+       xargs -n1 git -C dst cat-file -t >fetched_types &&
+
+       sort -u fetched_types >unique_types.observed &&
+       echo commit >unique_types.expected &&
+       test_cmp unique_types.expected unique_types.observed &&
+
+       # Auto-fetch a tree with cat-file.
+       git -C dst cat-file -p $SUBTREE >tree_contents &&
+       grep file.txt tree_contents &&
+
+       # fsck still works after an auto-fetch of a tree.
+       git -C dst fsck &&
+
+       # Auto-fetch all remaining trees and blobs with --missing=error
+       git -C dst rev-list --missing=error --objects master >fetched_objects &&
+       test_line_count = 70 fetched_objects &&
+
+       awk -f print_1.awk fetched_objects |
+       xargs -n1 git -C dst cat-file -t >fetched_types &&
+
+       sort -u fetched_types >unique_types.observed &&
+       test_write_lines blob commit tree >unique_types.expected &&
+       test_cmp unique_types.expected unique_types.observed
+'
+
 test_expect_success 'partial clone fetches blobs pointed to by refs even if normally filtered out' '
        rm -rf src dst &&
        git init src &&
index 53975c572465314957aaef360dea040a5be2db22..eb32505a6ef8804f4f7762ce0332f00beeacbe99 100755 (executable)
@@ -34,6 +34,18 @@ test_expect_success 'verify blob:none omits all 5 blobs' '
        test_cmp expected observed
 '
 
+test_expect_success 'specify blob explicitly prevents filtering' '
+       file_3=$(git -C r1 ls-files -s file.3 |
+                awk -f print_2.awk) &&
+
+       file_4=$(git -C r1 ls-files -s file.4 |
+                awk -f print_2.awk) &&
+
+       git -C r1 rev-list --objects --filter=blob:none HEAD $file_3 >observed &&
+       grep "$file_3" observed &&
+       ! grep "$file_4" observed
+'
+
 test_expect_success 'verify emitted+omitted == all' '
        git -C r1 rev-list --objects HEAD >revs &&
        awk -f print_1.awk revs |
@@ -232,6 +244,56 @@ test_expect_success 'verify sparse:oid=oid-ish omits top-level files' '
        test_cmp expected observed
 '
 
+test_expect_success 'rev-list W/ --missing=print and --missing=allow-any for trees' '
+       TREE=$(git -C r3 rev-parse HEAD:dir1) &&
+
+       # Create a spare repo because we will be deleting objects from this one.
+       git clone r3 r3.b &&
+
+       rm r3.b/.git/objects/$(echo $TREE | sed "s|^..|&/|") &&
+
+       git -C r3.b rev-list --quiet --missing=print --objects HEAD \
+               >missing_objs 2>rev_list_err &&
+       echo "?$TREE" >expected &&
+       test_cmp expected missing_objs &&
+
+       # do not complain when a missing tree cannot be parsed
+       test_must_be_empty rev_list_err &&
+
+       git -C r3.b rev-list --missing=allow-any --objects HEAD \
+               >objs 2>rev_list_err &&
+       ! grep $TREE objs &&
+       test_must_be_empty rev_list_err
+'
+
+# Test tree:0 filter.
+
+test_expect_success 'verify tree:0 includes trees in "filtered" output' '
+       git -C r3 rev-list --quiet --objects --filter-print-omitted \
+               --filter=tree:0 HEAD >revs &&
+
+       awk -f print_1.awk revs |
+       sed s/~// |
+       xargs -n1 git -C r3 cat-file -t >unsorted_filtered_types &&
+
+       sort -u unsorted_filtered_types >filtered_types &&
+       test_write_lines blob tree >expected &&
+       test_cmp expected filtered_types
+'
+
+# Make sure tree:0 does not iterate through any trees.
+
+test_expect_success 'filter a GIANT tree through tree:0' '
+       GIT_TRACE=1 git -C r3 rev-list \
+               --objects --filter=tree:0 HEAD 2>filter_trace &&
+       grep "Skipping contents of tree [.][.][.]" filter_trace >actual &&
+       # One line for each commit traversed.
+       test_line_count = 2 actual &&
+
+       # Make sure no other trees were considered besides the root.
+       ! grep "Skipping contents of tree [^.]" filter_trace
+'
+
 # Delete some loose objects and use rev-list, but WITHOUT any filtering.
 # This models previously omitted objects that we did not receive.