2008/8/4 Judah Jacobson <[EMAIL PROTECTED]>:
> It seems like most of the recent build breakages have been with
> patches not being validated on a complete set of OSes (OS X and
> Windows, in particular).  In last week's IRC meeting, Neil Mitchell
> mentioned:
>
>> i've always wondered why there isn't a pending repo, and patches get
>> moved from pending to branch after a the buildbots kick in? might ensure
>> you never end up with validation failures
>
> It seems to me that process would be more friendly to those of us who
> run different OSes than the primary GHC developers :-)

I've done a little searching, and I found at least one continuous
integration tool that will do this (for Git though, not Darcs)
(http://hudson.gotdns.com/wiki/display/HUDSON/Git+Plugin). How it
seems to work is:

1) The GHC developers write all their bug fixes / new features on
feature branches, NOT on HEAD
2) You request the continuous integration server to try and integrate
your changes by pushing that branch to the main GHC repo
3) Anytime you push patches to your remote feature branch, the CI
server merges those changes into it's local branch of HEAD and doesn't
push the resulting compiler to the real HEAD (that everyone else can
see) unless the tests pass (i.e. in our case, it validates on all
platforms)

This requires a discipline change (use feature branches rather than
just developing like mad on HEAD), but if we decide to migrate to a
DAG based VCS like Git/Hg/Bzr this seems like the better thing to do
in any case.

Cheers,
Max

_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc

Reply via email to