[Nix-dev] Move to GHC 7.8.2

[Gergely Risko]

Hi,

I think all this ideas worth going for.  Having outdated different happy
and alex linked to the ghc763 packages was annoying since GHC 7.8 build
system started requiring the new versions.

Personally I always prefer to just use the newest versions and then help
the community fix the breakage.  Therefore I like your plan to always
use the newest packages, and have a different attribute (called
whatever) for "standard", but outdated version collections.

I also agree with 3, the latest from every major version should be
enough.

Mateusz mentioned that it may worth waiting for 7.8.3 and I'd definitely
like to have that release when it's out, but I guess it's not a big of
deal to switch, so maybe we shouldn't wait.

Gergely

On Tue, 06 May 2014 19:07:19 +0200, Peter Simons <simons at cryp.to> writes:

> Hi guys,
>
> I would like to get some feedback on how you feel about moving to GHC 7.8.2
> as the default compiler of choice?
>
> What are the most significant road blocks right now in your opinion?
>
> Also, I'd like to use the switch as an occasion to remove most of the Haskell
> Platform related magic that's currently performed in haskell-defaults.nix,
> i.e. I would like to abandon the notion that certain compilers prefer older
> versions of HP packages because these versions were "mandated" at the time the
> compiler was released. Instead, I'd use the latest version possible of every
> package for every compiler.
>
> Here are some ideas in no particular order:
>
>  1) We provide haskellPlatform_X_Y_Z attributes in all-packages.nix that
>     install a particular version of HP -- including the corresponding GHC
>     --, but the library choices in those packages have no effect on any of
>     the haskellPackages_ghcXYZ.foobar packages.
>
>  2) The haskellPackages_ghcXYZ.haskellPlatform attribute becomes a
>     collection of packages that corresponds roughly to the set of packages
>     that the official HP offers, and you can expect those packages to build
>     and interact nicely with each other, but the versions won't necessarily
>     conform to any HP standard. For example, that 'haskellPlatform'
>     attribute will always contain the *latest version* of Cabal and
>     cabal-install, regardless of what the HP standard says. That attribute
>     can be thought of as "recommended popular packages in stable versions",
>     but it doesn't try to conform to any particular version of the HP. The
>     packages from (1) do that. Maybe we shouldn't even call it
>     'haskellPlatform'?
>
>  3) I would like to drop a whole lot of obsolete GHC versions. As a rule of
>     thumb, I think we should keep the latest version of every major release,
>     i.e. 6.10.4, 6.12.3, 7.0.4, 7.2.2, 7.4.2, 7.6.2, and 7.8.2 -- but none
>     of the intermediate versions. It's intellectually satisfying to have
>     support for GHC 6.12.2 still around, but honestly ... who needs it?
>
> Share your thoughts, please!
>
> Best regards,
> Peter