[PATCH 0/3] Make update-refs more atomic

[Ronnie Sahlberg <sahlberg@google.com>]

refs.c:update_refs() intermingles doing updates and checks with actually
applying changes to the refs in loops that abort on error.
This is done one ref at a time and means that if an error is detected that
will fail the operation after only some of the ref operations have been
been updated on the disk.

These patches change the update and delete functions to use a three
call pattern of

1, lock
2, update, or flag for deletion
3, apply on disk

In the final patch I change update_refs to perform these actions in three
separate loops where the final loop to 'apply on disk' all the changes will
only be performed if there were no error conditions detected during any of
previous loops.

This should make the changes of refs in update_refs slightly more atomic.


This may overlap with other current patch series for making refs updates
more atomic which may mean these patches become obsolete, but I would still
like some review and feedback on these changes.



Ronnie Sahlberg (3):
  Split writing and commiting a ref into two separate functions.
  Split delete_ref_loose() into a separate flag-for-deletion and commit
    phase
  Change update_refs to run a single commit loop once all work is
    finished.

 branch.c               | 10 ++++--
 builtin/commit.c       |  5 +++
 builtin/fetch.c        |  7 +++-
 builtin/receive-pack.c |  4 +++
 builtin/replace.c      |  6 +++-
 builtin/tag.c          |  6 +++-
 fast-import.c          |  7 +++-
 refs.c                 | 90 +++++++++++++++++++++++++++++++++-----------------
 refs.h                 |  6 ++++
 sequencer.c            |  4 +++
 walker.c               |  4 +++
 11 files changed, 112 insertions(+), 37 deletions(-)

-- 
1.9.1.477.g7668a0d.dirty