Support static pointers

[edsko]

We can add support for static pointers in a fully backwards compatible way; we do not even need to bump the major version number on the package. All we are doing is adding an additional constructor to StaticLabel. Once all the work on polystatic (however it gets resolved eventually) and the new structure of the SPT is implemented in GHC, we probably want to modify this implementation, but until that time we can make use of the new static pointers extension right now. Being able to avoid remotable and co makes life a lot easier. I've tested this in a Cloud Haskell application we're developing and it seems to work very well. It's certainly cleaned out code up nicely.

Note that even if we change the internal implementation, the user facing API that this PR provides is the right one:

staticPtr :: forall a. Typeable a => StaticPtr a -> Static a

This will not need to change even when the underlying mechanisms change.

[edsko]

This relies on a very minor extension to the rank1dynamic package: haskell-distributed/rank1dynamic#13 .

Note that I have not bumped version numbers on either package as part of the PR.

[edsko]

(By the way, it doesn't mean we don't need remote tables anymore, of course, because CH internationally still relies on them, as does, incidentally, distributed-static itself. It does however mean we don't need any additional remote tables are calls to remotable in applications that don't mind insisting on ghc >= 7.10).

[edsko]

Whoops, missing some CPP somewhere for older CPP. Will fix.

[dcoutts]

And add a travis config for 7.10

[edsko]

Ok, added 7.10 to the travis and fixed build on < 7.10.

[mboes]

What's polystatic?

[edsko]

@mboes It's proposed at https://ghc.haskell.org/trac/ghc/wiki/TypeableT ; I have to say I'm not convinced by it myself so far.

[facundominguez]

Don't bother with .travis.yml, or consider using this commit eb0cdfc

Wait, actually. I'll merge that right now.
Done.

[edsko]

Ok, removed the change to the travis config.

[facundominguez]

LGTM.

I don't like much having to encode and decode the typerep on every use of a static pointer, but I see no better way to avoid crashes when the user gets the type index wrong.

[facundominguez]

This can be imported as GHC.Exts(Any) and it would allow to drop the dependency on ghc-prim, I think.

[edsko]

Ok, changed import to GHC.Exts and removed ghc-prim from dependency list in the cabal file.