[Haskell-cafe] real haskell difficulties (at least for me)
Duncan Coutts
duncan.coutts at worc.ox.ac.uk
Thu Jan 15 07:36:28 EST 2009
On Thu, 2009-01-15 at 13:26 +0100, Manlio Perillo wrote:
> >> In detail:
> >> - The package author submit his package, and additional build logs
> >> for each platform he have access to
> >> - Other users can submit additional build logs for their platform
> >>
> >>
> >> I think, however, that a centralized system has its advantages.
> >> As an example, a newly submitted package can be put in a 'incoming
> >> queue', and moved to the final destination only if it builds (and tests
> >> succeed) on all the supported platforms.
> >
> > What we really want is both. We want to let random users on random
> > platforms submit simple anonymous build reports (no logs).
> >
> > In addition since any client can upload reports we can set up a number
> > of dedicated clients that will build every new package in a chroot and
> > upload non-anonymous detailed build reports with build logs.
> >
>
> Since nowdays virtual machines are very affordable, it would be nice to
> have a set of standard virtual system images, preconfigured to run Cabal
> installation process.
>
> So that when I compile a package on my system, Cabal can be instructed to:
> 1) Start each virtual machine on the system
> 2) Submit the package to each machine an get the report
> 3) Stop each virtual machine
>
>
> Is this a feasible task?
Yes, and the great thing is that anyone could do that. That's the
advantage of a decentralised system. It does not require the permission
or help of the cabal/hackage hackers to do it.
> The only "problem" is with proprietary operating systems like Windows
> and Mac OS, where pre-configured systems can not be provided.
True.
The other issue is that perhaps knowing something builds on a clean
virtual machine is not quite so useful as knowing it builds on 100's of
messy machines in dozens of different configurations.
So that's not to say that using dedicated build clients running on
virtual machines is not useful, it is. But we don't need lots and lots
of those.
Duncan
More information about the Haskell-Cafe
mailing list