[core libraries] List of core libraries

[Neil Mitchell]

> Btw, I'm a bit wondering, why do you think the testsuite's failure won't
> mark the Travis-CI build as failed? :)

It's a bit hard to know for sure, but my logic is:

* https://github.com/haskell/filepath/blob/master/tests/AutoTest.hs#L38
means that the output of the QuickCheck test doesn't cause an
exception (as it originally was), but just prints a message about
failure
* The travis doesn't look at
https://github.com/haskell/filepath/blob/master/tests/filepath.stdout

So Travis doesn't check the test won't be broken by GHC, and I think
without that check it doesn't spot failures. (That's my chain of
reasoning, but it's all a little subtle, so could quite easily be
wrong!)

> I'd suggest to simply add you to the maintainer-team for filepath on
> https://github.com/haskell/filepath/ if that's ok for you (in fact, I've
> just added you in just in case)

Great.

>> * What's the integration with GHC? Does the filepath module get pulled
>> directly into GHC source trees? I see it's in the GHC test suite - is
>> that considered a good idea?
>
> it's a submodule in ghc.git, and we have an automatic mirroring job
> mirroring to git.haskell.org/packages/filepath.git
>
> And from time to time, we'd update the commit ghc.git points to inside
> filepath.git; so you can freely push to master on
> https://github.com/haskell/filepath w/o affecting GHC HEAD, only when we
> update the filepath gitlink we'd see the effect of it in GHC HEAD.

Perfect. I'm sure I'll do something incompatible with the GHC world at
some point, but just shout at me when I do (or, if there happens to be
a list of the most likely stupid things I might do, let me know).

> I think this is mostly a historic artefact, as something Travis wasn't
> around when filepath started being used by GHC.

Cabal was barely around when filepath started being used by GHC...
Cabal test was at least one incompatible iteration as well.

> Do you have the means to test filepath on all GHC tier-1
> platforms/archs? (That's the only benefit I currently see with having
> the tests as part of GHC's validation testsuite)

No, that's valuable, and certainly makes it work keeping.

Thanks, Neil