[darcs-users] What's wrong with darcs on Solaris-based buildbots?

Karel Gardas karel.gardas at centrum.cz
Wed Oct 8 19:52:09 UTC 2008

Eric Kow wrote:
> On Wed, Oct 08, 2008 at 19:55:28 +0200, Karel Gardas wrote:
>> it seems the issue described below in the previous report is still valid
>> and happens. So it's not been resolved as I mistakenly reported few days
>> ago.
> I'm not sure about the 100% CPU, but one problem I notice is that we do
> not cope very well with the fact that tests that play with permissions
> can fail.  For example, you now cannot fetch darcs, because buildbot
> cannot delete the directory it was working it (permissions) because the
> test did not complete.
> I'm not sure what to do about this.
> It would be nice if our test suite had some sort of try/catch/finally
> mechanism (I guess that was one advantage of the Perl tests).
> Otherwise, we could have the shell_harness recursively chmod u+w the
> directory it's working in after running tests (just in case).  Or we
> could have a fake test "zzz.sh" which does the same sort of post-test
> cleanup.  A fourth option would be to modify buildbot so that it gives
> itself permissions to delete the directory.

The problem here is that buildbot tries to delete directory with 444
permissions. That's going to fail on both Solaris and OpenBSD. My
assumption is that it runs well on Linux, but on Unix you probably need
writing permission for it. Perhaps test suite might be fixed when
creating and testing with 444 perms directory to chmod it later to
writing mode? Or that's my feeling about this issue so far w/o any
further investigation except looking into the problematic directory.


More information about the darcs-users mailing list