type checker plugin does not affect inferred type signatures

[adam vogt]

Hi Adam,

I've added a README which tries to explain things:
https://github.com/aavogt/HListPlugin

When I produce a wanted constraint from a wanted constraint, things
work as I wanted. Thanks for the suggestion!

Regards,
Adam

On Mon, Feb 16, 2015 at 4:36 AM, Adam Gundry <adam at well-typed.com> wrote:
> Hi Adam,
>
> It's great to hear that you are trying the plugins functionality, this
> is exactly the kind of experimentation it's designed for! I'm a little
> confused about what you're trying to achieve, though. Can you give some
> examples of code you'd like to be able to write?
>
> In general, GHC's type inference algorithm isn't expecting wanted
> constraints to be produced from givens; confusing things will happen if
> they are, and it's likely that *less* things will be typeable rather
> than *more*. Perhaps the plugin infrastructure should prevent you from
> doing so.
>
> It makes sense to produce givens from givens or wanteds from wanteds
> though. I'd imagine you might want to look for *wanted* constraints
> (HLength x ~ HLength y) and add an additional *wanted* (SameLength x y).
>
> One other thing to note is that plugins are called twice, once to
> simplify the givens (with empty wanteds), and once to solve the wanteds
> (https://ghc.haskell.org/trac/ghc/wiki/Plugins/TypeChecker#Callingpluginsfromthetypechecker).
>
> Hope this helps,