weird fatal on push when post-receive hook exists but is empty

Helping someone with what at first appeared to be a gitolite problem, we found something curious. Here it is, reduced to pure git, even without ssh and other complications in between:

    21:47:05 sita-lt:t $ rm -rf a b c
    21:47:10 sita-lt:t $ git init --bare b
    Initialized empty Git repository in /home/sitaram/t/b/
    21:47:23 sita-lt:t $ git clone b a
    Initialized empty Git repository in /home/sitaram/t/a/.git/
    warning: You appear to have cloned an empty repository.
    21:47:29 sita-lt:t $ cat > b/hooks/post-receive
    #!/bin/bash
    # dummy
    21:47:55 sita-lt:t $ chmod +x b/hooks/post-receive
    21:48:00 sita-lt:t $ cd a
    21:48:02 sita-lt:a (master) $ date > foo;git add foo; git commit
-m foo; git push origin master
    [master (root-commit) f55bd25] foo
     1 files changed, 1 insertions(+), 0 deletions(-)
     create mode 100644 foo
    Counting objects: 3, done.
    Writing objects: 100% (3/3), 232 bytes, done.
    Total 3 (delta 0), reused 0 (delta 0)
    Unpacking objects: 100% (3/3), done.
    fatal: The remote end hung up unexpectedly
    error: error in sideband demultiplexer
    error: .git/objects/f5/5bd251325ad1647a29869e75a8080fee1a5d97 died
of signal 13
    To /home/sitaram/t/b
     * [new branch]      master -> master
    error: failed to push some refs to '/home/sitaram/t/b'
    (rc=1)21:48:23 sita-lt:a $

Of course, this does not happen when the hook doesn't even exist...

Sitaram Chamarty venit, vidit, dixit 20.04.2010 18:21:

With git version 1.7.1.rc1.248.gcefbb and this script:

#!/bin/bash rm -rf a b c git init --bare b git clone b a cat > b/hooks/post-receive <<EOF #!/bin/bash # dummy EOF chmod +x b/hooks/post-receive cd a date > foo;git add foo; git commit -m foo; git push origin master

I get this output:
Initialized empty Git repository in /tmp/t/b/
Initialized empty Git repository in /tmp/t/a/.git/
warning: You appear to have cloned an empty repository.
[master (root-commit) 020a1fb] foo
 1 files changed, 1 insertions(+), 0 deletions(-)
 create mode 100644 foo
Counting objects: 3, done.
Writing objects: 100% (3/3), 241 bytes, done.
Total 3 (delta 0), reused 0 (delta 0)
Unpacking objects: 100% (3/3), done.
To /tmp/t/b
 * [new branch]      master -> master

What are we doing differently?

Michael

Am 4/21/2010 10:46, schrieb Michael J Gruber:

Your box has a different scheduler. I doesn't run (or exit) the script soon enough that the parent (receive-pack) can get SIGPIPE when it writes refs to the script's stdin.

-- Hannes