Re: [BUG RFC/PATCH] git-cvsimport

From: Michael Haggerty <mhagger@alum.mit.edu>
Date: Aug 16, 2010, 04:24 UTC
Message-ID: <4C68BD66.9060506@alum.mit.edu>
In-Reply-To: <AANLkTikxxqgNpmqfFQq+5GVd7VAW2DTaro5anRHcx6Tt@mail.gmail.com>

Ævar Arnfjörð Bjarmason wrote:

Show 27 quoted lines

> This is more of a BUG than a RFC/PATCH.
> 
> Now that Junio has applied my patch to not write test results under
> harness I can:
> 
>     sudo chown -R root t
> 
> And run the tests with --root=/dev/shm under prove, except for 3
> git-cvsimport tests. I have pending patches to these, after they'll
> get in I'll fix that.
> 
> But the reason they hang is interesting:
> 
>     rm -rf /tmp/meh;
>     git init /tmp/meh &&
>     cd /tmp/meh &&
>     sudo chown -R root /home/avar/g/git/t/t9601/cvsroot &&
>     cvsps --norc -q --cvs-direct -u -A --root
> /home/avar/g/git/t/t9601/cvsroot module
>     Initialized empty Git repository in /tmp/meh/.git/
> 
> cvsps will just hang due to the unfriendly chmod. Maybe we want
> something like the below to deal with that.
> 
> However, even then it'll still hang on something else, I haven't
> looked into what. I'm just going to fix this by having it copy the
> things it needs to the --root directory.

My guess is that cvsps is using CVS commands to access the test CVS repository, and that CVS wants to write to the file CVSROOT/history to log what is being done. The logging behavior can be turned off either:

* by using the "-R" option when invoking CVS.  This is a global option,
meaning that it has to appear before the subcommand: "cvs -R update"
rather than "cvs update -R".

* by setting the "$CVSREADONLYFS" environment variable.

Apparently -R and "$CVSREADONLYFS" were added in CVS 1.12.1, which was released in 2003.

Michael