[PATCHv3 0/9] object store: oid_object_info is the next contender
- From
Stefan Beller <sbeller@google.com>
- Date
- Apr 25, 2018, 18:20 UTC
- Message-ID
- <20180425182106.162972-1-sbeller@google.com>
- In-Reply-To
- <20180424215910.22201-1-sbeller@google.com>
v3: * fixed and extended the commit message of last commit * fixed the last patch, as Jonathan Tan suggested, see interdiff:
$ git diff remotes/origin/sb/oid-object-info (which is v2)
diff --git c/sha1_file.c w/sha1_file.c
index 94123e0299..dcd6b879ac 100644
--- c/sha1_file.c
+++ w/sha1_file.c
@@ -1289,14 +1289,13 @@ int oid_object_info_extended(struct repository *r, const struct object_id *oid,
/* Check if it is a missing object */
if (fetch_if_missing && repository_format_partial_clone &&
- !already_retried) {
+ !already_retried && r == the_repository) {
/*
* TODO Investigate having fetch_object() return
* TODO error/success and stopping the music here.
- * TODO Pass a repository struct through fetch_object.
+ * TODO Pass a repository struct through fetch_object,
+ * such that arbitrary repositories work.
*/
- if (r != the_repository)
- die(_("partial clones only supported in the_repository"));
fetch_object(repository_format_partial_clone, real->hash);
already_retried = 1;
continue;
Thanks,
Stefanv2:
* fixed the sha1/oid typo * removed spurious new line * Brandon and Jonthan discovered another dependency that I missed due to cherrypicking that commit from a tree before partial clone was a thing. We error out when attempting to use fetch_object for repos that are not the_repository.
Thanks, Stefan
v1: This applies on top of origin/sb/object-store-replace and is available as https://github.com/stefanbeller/git/tree/oid_object_info
This continues the work of sb/packfiles-in-repository, extending the layer at which we have to pass in an explicit repository object to oid_object_info.
A test merge to next shows only a minor merge conflicit (adding different #include lines in one c file), so this might be a good next step for the object store series.
Notes on further object store series: I plan on converting the "parsed object store" next, which would be {alloc, object, tree, commit, tag}.c as that is a prerequisite for migrating shallow (which is intermingled with grafts) information to the object store.
There is currently work going on in allocation (mempool - Jameson Miller) and grafts (deprecate grafts - DScho), which is why I am sending this series first. I think it can go in parallel to the "parsed object store" that is coming next.
Thanks, Stefan
Jonathan Nieder (1): packfile: add repository argument to packed_object_info
Stefan Beller (8): cache.h: add repository argument to oid_object_info_extended cache.h: add repository argument to oid_object_info packfile: add repository argument to retry_bad_packed_offset packfile: add repository argument to packed_to_object_type packfile: add repository argument to read_object packfile: add repository argument to unpack_entry packfile: add repository argument to cache_or_unpack_entry cache.h: allow oid_object_info to handle arbitrary repositories
archive-tar.c | 2 +- archive-zip.c | 3 ++- blame.c | 4 ++-- builtin/blame.c | 2 +- builtin/cat-file.c | 12 ++++++------ builtin/describe.c | 2 +- builtin/fast-export.c | 2 +- builtin/fetch.c | 2 +- builtin/fsck.c | 3 ++- builtin/index-pack.c | 4 ++-- builtin/ls-tree.c | 2 +- builtin/mktree.c | 2 +- builtin/pack-objects.c | 11 +++++++---- builtin/prune.c | 3 ++- builtin/replace.c | 11 ++++++----- builtin/tag.c | 4 ++-- builtin/unpack-objects.c | 2 +- cache.h | 7 +++++-- diff.c | 3 ++- fast-import.c | 16 ++++++++++------ list-objects-filter.c | 2 +- object.c | 2 +- pack-bitmap-write.c | 3 ++- pack-check.c | 3 ++- packfile.c | 40 +++++++++++++++++++++++----------------- packfile.h | 6 ++++-- reachable.c | 2 +- refs.c | 2 +- remote.c | 2 +- sequencer.c | 3 ++- sha1_file.c | 37 +++++++++++++++++++++---------------- sha1_name.c | 12 ++++++------ streaming.c | 2 +- submodule.c | 2 +- tag.c | 2 +- 35 files changed, 124 insertions(+), 93 deletions(-)
-- 2.17.0.441.gb46fe60e1d-goog