Cabal constraint solver

[Simon Marlow]

On 03/10/2010 00:12, Brandon S Allbery KF8NH wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 10/2/10 18:07 , Christian Höner zu Siederdissen wrote:
>> cd syb-0.2.1
>> * remove base<4.3 constraint from syb.cabal
>> cabal install
>> * syb-0.2.1 is now installed and works!
>>
>> $ cabal install parsec-3.1.0
>> cabal: cannot configure syb-0.2.1. It requires base>=4.0&&  <4.3
>>
>> This is rather annoying: the constraint is solved (eg. syb is installed)
>> but we /still/ assume that we can not continue. So obviously, I would
>> like to be able to have cabal assume constraints fullfilled if the
>> package is installed.
>
> This is unsafe because the package database (which isn't maintained by
> cabal, but by ghc; see the ghc-pkg command) doesn't store the constraints.
> So cabal can't tell if that *particular* build was against the version of
> base it needs for everything else.

Sure it can:

$ ghc-pkg field haskell98 depends
depends: array-0.3.0.1-d7acfe6d19e77c2d80d16a9f0bc16697
          base-4.2.0.2-6ca19098c3531da73f1940ce836e19d8
          directory-1.0.1.1-4b73b609a785b25115a63a1f7bf6928d
          old-locale-1.0.0.2-0c1a4b40f2d3b9f6b725f54c00fe0923
          old-time-1.0.0.5-cfc209473b38d8ca51f7dbeea24c4f0f
          process-1.0.1.3-e944f15cf7e7ec567f14bf379c37cadd
          random-1.0.0.2-049b2f84143b7536808f44f17ec734b4

> (ghc can have multiple versions of base
> around; see the 6.10 series.  I *think* ghc is constrained to only
> supporting it for different major versions, but it can't tell Cabal that.)
>
> Now, it is arguable that if there is no different release of base in the
> package database then it should assume the visible release is the one used;
> but you again run into the limitations of ghc-pkg.  If I "ghc-pkg hide" a
> given package release, dependents of that package don't also get hidden
> because (again) ghc-pkg doesn't store that information.

hiding packages makes no difference to Cabal, it is only for setting the 
packages that are visible to GHCi and ghc --make.

 > So now you end up
> with the same problem except it will show up when you try to use one of the
> dependents, and (worse) because Cabal can't tell this happened you will at
> best get an unexpected type mismatch at compile time and at worst the
> program will compile but dump core (if the mismatch is in base.  It's hard
> to validate that runtimes are compatible, and base is very tightly bound to
> the runtime).

I'm afraid I'm not following - I thought we'd eliminated the possibilty 
of incorrect combinations of packages by using the ABI hash to identify 
dependencies (in GHC 6.12).  Could you describe exactly how you get into 
this situation?

Cheers,
	Simon



> - --
> brandon s. allbery     [linux,solaris,freebsd,perl]      allbery at kf8nh.com
> system administrator  [openafs,heimdal,too many hats]  allbery at ece.cmu.edu
> electrical and computer engineering, carnegie mellon university      KF8NH
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v2.0.10 (Darwin)
> Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
>
> iEYEARECAAYFAkynvHgACgkQIn7hlCsL25XttACfQ+9G4bIAOnQVwJddDBizaPba
> uU4AoJjFXPkeqcloqp8w2COQ4I0bh+nk
> =sZhw
> -----END PGP SIGNATURE-----
> _______________________________________________
> Glasgow-haskell-users mailing list
> Glasgow-haskell-users at haskell.org
> http://www.haskell.org/mailman/listinfo/glasgow-haskell-users