[GHC DevOps Group] CircleCI status

Phyx lonetiger at gmail.com
Sun Dec 23 04:51:24 UTC 2018

On Sun, Dec 23, 2018, 00:31 Ben Gamari <ben at well-typed.com> wrote:

> Moritz Angermann <moritz.angermann at gmail.com> writes:
> > What prevents us in principle to cross compile ghc on Linux for
> > Windows? We’d still want to use the built artifacts and run the
> > testsuite on Windows I think, but building ghc?
> >
> We currently cannot run the full testsuite on a binary distribution due to
> #13897 (essentially we don't ship a few utilities which the testsuite would
> require; Alp is looking at fixing this in Hadrian).
> Moreover, we can't run Haddock when cross-compiling which means that we
> would need to find a way to skip the haddock performance tests.
> However, in general I think it is important that Tier 1 platforms be
> tested in a non-cross-compiled environment. Afterall, this is the
> environment in which we need to build binary distributions (for the
> above-mentioned Haddock issue, among others).

And a bootstrap is representative of real code. This catches more issues
than a bunch of small tests. The resulting binaries won't be representative
of the code we ship to users as the cross compiler won't contain the patch
(or any other optimization in master) unless you build the cross first at
which point you're increasing your Linux coverage while reducing Windows.

Secondly Windows is statically linked, this means you'll end up linking
against library code from the cross tool chain instead of the native. The
cross tool chains are often way older as they are just there to bootstrap
builds. So you have a significant reduction in code coverage.

> Cheers,
> - Ben
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-devops-group/attachments/20181223/1dc71a12/attachment-0001.html>

More information about the Ghc-devops-group mailing list