Race condition?

[Ian Lynagh]

On Thu, Sep 01, 2011 at 01:43:45PM +0100, Simon Marlow wrote:
> On 01/09/2011 11:32, Simon Peyton-Jones wrote:
> >In my latest validate run (from clean), I got many occurrences of this
> >warning:
> >
> >+WARNING: cache is out of date:
> >/64playpen/simonpj/builds/HEAD-2/bindisttest/install
> >dir/lib/ghc-7.3/package.conf.d/package.cache
> >
> >+ use 'ghc-pkg recache' to fix.
> >
> >mod146.hs:2:30:
> >
> >Conflicting exports for `m1':
> >
> >*** unexpected failure for mod146(normal)
> >
> >I consulted Simon M. Our best guess is that it’s caused by a race
> >condition: two tests both of which modify the package database, screwing
> >it up.
> 
> Just to clarify - the package database is not screwed up as such,
> but the cache is out of date.  Everything still works perfectly when
> this happens, but you get a warning that the cache should be
> updated, and ghc-pkg operations (and GHC) are slower than normal.
> 
> My guess would be that some of the tests under testsuite/tests/cabal
> aren't safe to run concurrently.

I don't think tests should be touching the global database at all. I'll
finish off the strace stuff I started recently which should be able to
detect this sort of thing.


Thanks
Ian