Junio, this is re-roll of the 'tr/http-push-ref-status' branch in 'pu'.
Jeff, please don't see this re-roll as me assuming you gave your go-
ahead for the latest discussion; I didn't, I thought squashing in the
hunks I mentioned would make things easier for you, rather than having
them floating around.
Summary:
This patch series applies on top of 'next', and deals with alerting
the user to rejected non-fast-forward pushes when using helpers
(smart).
Previously, git silently exited. This situation involves the curl
helper and the smart protocol. The non-fast-forward push is only
detected when curl executes the rpc client (git-send-pack). Now, we
detect it before telling the helper to push.
The series also remedies the scenario where a user would not be aware
of rejected non-fast-forward pushes.
It occurs when
1) there are one or more pushes that succeed and
2) there are one or more rejected non-fast-forward pushes that
involve refs that cannot be matched without explicit refspecs.
This is due to the re-marking of ref status in transport-helper.c::
push_refs() when interacting with the remote helper.
If only non-matched, non-fast-forward refs are involved (ie.
condition #2 is present without #1), then the situtation does not
occur - no 'push' commands are passed to the remote helper, no
interaction with the helper takes place, and no re-marking of ref
status takes place. The user will be alerted to the rejected
non-fast-forward push.
If both are present, re-marking occurs. Even if a ref was not part
of a 'push' command to the helper, 'ok'/'error' status reports might
be made for it, since the remote helper does ref matching on its
side. This led to refs that cannot be matched without an explicit
refspec marked as REF_STATUS_REJECT_NONFASTFORWARD to be re-marked
REF_STATUS_NONE.
(Note: helpers fail to match refs as explicit refspecs are always
available to the top-level transport mechanism, but only on a need-
to-know basis to the remote helper via a 'push' command when a ref
is to be pushed.)
Interestingly, it is possible for a user to be forever unaware of
the rejected push if the user relies on information from git push
alone. As long as the user ensures that one or more other pushes are
successful (eg. change a tracked ref fast-forwardedly) between the
execution of git push, the rejected non-fast-forward push won't be
reported.
Changes from v3:
- reworded commit message for the test in patch 2
- added comment on what the switch block is checking for in patch 3
- changed condition under which ref status reported by remote helper
is ignored in patch 5
Tay Ray Chuan (6):
t5541-http-push.sh: add tests for non-fast-forward pushes
t5541-http-push.sh: add test for unmatched, non-fast-forwarded refs
refactor ref status logic for pushing
transport.c::transport_push(): make ref status affect return value
transport-helper.c::push_refs(): ignore helper-reported status if ref
is not to be pushed
transport-helper.c::push_refs(): emit "no refs" error message
builtin-send-pack.c | 51 +++++++++++--------------------------------------
remote.c | 50 +++++++++++++++++++++++++++++++++++++++++++++++++
remote.h | 2 +
t/t5541-http-push.sh | 44 +++++++++++++++++++++++++++++++++++++++++++
transport-helper.c | 28 +++++++++++++++++++-------
transport.c | 11 ++++++++-
6 files changed, 137 insertions(+), 49 deletions(-)
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html