Hello!
It looks like that either the documentation is misleading or git is not working as promised.
Documentation/repository-layout.txt says:
info/refs::
This file is to help dumb transports to discover what
refs are available in this repository. Whenever you
create/delete a new branch or a new tag, `git
update-server-info` should be run to keep this file
up-to-date if the repository is published for dumb
transports. The `git-receive-pack` command, which is
run on a remote repository when you `git push` into it,
runs `hooks/update` hook to help you achieve this.My understanding of the above text is that git-update-server-info only needs to be run if a new branch or tag is created, not on every git-push. Just because hooks/update is run on every push, it doesn't mean that it needs to be done every time.
What happens in reality is that git-pull refuses to pull new revisions from the repository if git-update-server-info is not run after a simple git-push (from another repository, of course) that didn't create any branch or tag. However, git-clone would load all revisions. I'm accessing the repository over http.
The problem it, there is more than one way to get the branch revisions, one is from info/refs and the other is from individual files under refs/heads and refs/tags. Using database terms, the git repository in not normalized, so we get an update anomaly.
I understand that accessing one file could be faster over http, so I tend to think that it's the documentation needs to be fixed. If that's the case, maybe hooks/update should be created executable by default?
Alternatively, git-update-server-info functionality could be embedded into git-receive-pack to simplify server setup.
-- Regards, Pavel Roskin
