Wierd git over http behaviour

[John Williams <john.williams@petalogix.com>]

Hi,

We have a git tree that accesses fine via the git protocol, however I have
also opened up http support for it as some of our users are behind
git-unfriendly corporate firewalls.

I know http transport is not encouraged, but not much I can do about that.

Strange thing is, when doing an http fetch of said tree, it gets no updates
to the tree later than about Jan 13 this year, even though the tree itself
has many, many commits since then (its origin is Linus's tree and we pull
regularly).  Heads created later than that date are also not visible (doing
git branch -a) in the clone pulled over http.  Clones made using git: are
fine.

Anything I can look for in the .git repo on the server, permissions or
similar that might be relevant?  It's almost like there's a single commit in
there that somehow is making the http fetch puke.  If you browse the http
tree with a web browser, it's all there as expected.

For reference here are the URLs:

git://developer.petalogix.cmo/linux-2.6-microblaze

http://developer.petalogix.com/git2/linux-2.6-microblaze

Initially the http reference was a symlink on the server to the actual git
tree, but I also tried just rsync'ing it across to /var/www/html/git2 to
make sure it wasn't the symlink breaking things - no change.

Am I misunderstanding something or is this just flakey behaviour of the http
protocol support?

Thanks,

John