[Haskell-cafe] The 8 Most Important GSoC Projects

[Antoine Latter]

On Sat, Apr 3, 2010 at 12:45 PM, Don Stewart <dons at galois.com> wrote:
> schlepptop:
>> Don Stewart schrieb:
>>> While at ZuriHac, a few of us GSoC mentors got together to discuss what
>>> we think the most important student projects for the summer should be.
>>>
>>> Here's the list:
>>>
>>>     http://donsbot.wordpress.com/2010/04/01/the-8-most-important-haskell-org-gsoc-projects/
>>>
>>> Please consider applying to work on these tasks!
>>>
>> For me a cabal-install that can cope with dependencies reliably would be
>> the most important project. (I get frequently bug reports because
>> data-accessor-template cannot be installed by cabal-install although it
>> can be installed with plain Cabal.)
>>
>
> That's interesting. Why can't it be installed? Does
> data-accessor-template, or somethiing it depends on,  have incorrectly
> specified version ranges?
>

I'm no cabal expert, but I've had the same issue with other packages -
when it needs to build different with different versions of of
template-haskell the easiest thing to do is to have a flag in the
package description for different versions of the template-haskell
packages.

But then cabal-install thinks it can use either version in its install
plan, but it really can't as template-haskell is one of the libs that
is tied to your GHC install (even after fixing old th for GHC 6.12 and
installing it, you cannot use it).

So you either default to 6.12 compatible or 6.10 compatible, and then
loose the other compiler when your package is included one step deep
(or further) in a cabal-install install plan.

The solution I lean towards is to use the MIN_VERSION_TEMPLATE_HASKELL
cpp macro, what Henning has last tried is setting the package flag off
of the GHC version (which works because template-haskell is keyed to
your GHC install).

In summary - don't specify versions of template-haskell in your
package description.

Antoine