Skip to contentSkip to conversation

Where'd my GIT tree go?

DISCUSSION Tony Luck 3 participants 4 messages Jun 15, 2016

lore
Share & tools
lore

Tony Luck

author
#0 lore raw

http://www.kernel.org/git has stopped showing my linux-2.6 tree (the "to Linus" one, my "test-2.6" tree is still there).

This is probably my fault ... but I'm not sure exactly why.

Here's what I did. Last Thursday I applied a set of patches ... and my "apply" script choked on one of them. Some casual inspection convinced me that the changes had been applied, and I pushed the tree up to kernel.org.

But I was wrong, the 2nd out od a series of six was all messed up. And I'd applied a couple more patches on top of that. Since none of this had been pulled by Linus, I thought I'd clean it up and apologise to anyone who had pulled from my tree.

So I backed HEAD up to the last good commit. Re-applied the changes with a fixed version of the script, and then expected to find some detritus from the first application in .git/objects. But "git-fsck-cache --unreachable ..." only complained about the 2.6.11 tag/tree. Odd.

Then I pushed up to master.kernel.org ... and an hour later when the mirrors did their thing, git-web stopped showing my tree.

HEAD is still a symlink to refs/heads/master. And that has the SHA1 of my most recent commit ... which is present in .git/objects.

So what's wrong???

-Tony

#1 lore raw

Ok, you asked for it:

...the GIT bucket.

jon.

... ducks for cover ...

author
#2 lore raw
Quoted 7 lines · Jon Seymour

Ok, you asked for it:

...the GIT bucket.

jon.

... ducks for cover ...

Groan ... as well you should.

My tree has re-appeared now. Thanks to whoever fixed it.

-Tony

#3 lore raw
Quoted 3 lines · Tony Luck

Groan ... as well you should.

My tree has re-appeared now. Thanks to whoever fixed it.

I noticed similar effects recently. Its related to the mirroring of master.kernel.org to the public server. At some point you have only the half of updates at the public site and obviously the git webscript refuses to show your tree when HEAD is updated, but the referenced commit blob is missing.

I guess that the mirroring is not aware of the update order (objects first, HEAD last) which would prevent such oddities.

tglx