[Haskell-cafe] type error when specializing lens zooms in ghc >= 9.0

[Olaf Klinke]

On Fri, 2023-05-12 at 17:13 -0400, Brandon Allbery wrote:
> My first suspicion would be simplified subsumption. Try eta-expanding them.
> 

Indeed, the eta-expanded versions do compile with GHC >= 9.
Now I remember that I've fallen into a similar trap before [1] (and
Brandon offered the right hint back then, too. Thanks, Brandon!)  
So let me re-cap: 
(1) Haskell is not a category: there are types a, b, c 
    and morphisms f :: a -> b, g :: b -> c
    such that g.f is not a morphism.
(2) Haskell is not a lambda calculus [2]: There are terms f such that 
    \x -> f x 
    belongs to the calculus but f does not.

Personally, I would already call the situation improved when the
compiler instead of Brandon could tell me the fix. From the discussion
[3]:

> The most interesting thing I've learned from the lengthy threads,
> along with deficiencies in the testing process for ecosystem
> breakage, is how much absolute confusion there is over how rank-n-
> types work, and the limitations and complexities of algorithms
> pertaining to them. Now that they are in widespread use and on by
> default in ghc, it seems like an important case where user education
> is warranted.

I'd love to be educated. And I'd be very pleased if helpful compiler
error messages were the first entry point for that, so that I can spare
the patient and helpful Haskell Cafe members some time. 

Olaf

[1] https://mail.haskell.org/pipermail/haskell-cafe/2023-February/135903.html
[2] https://github.com/ghc-proposals/ghc-proposals/blob/master/proposals/0287-simplify-subsumption.rst#costs-and-drawbacks
[3] https://github.com/ghc-proposals/ghc-proposals/pull/287

that apparently I have not managed to form an in

> On Fri, May 12, 2023 at 5:10 PM Olaf Klinke <olf at aatal-apotheke.de> wrote:
> > 
> > Dear Cafe,
> > 
> > I'm an infrequent lens user so please forgive me if the below problem
> > has a trivial solution.
> > 
> > For a long time the documentation of Control.Lens.Zoom has been
> > claiming that 'zoom' and 'magnify' can be specialized to the following
> > types.
> > 
> > {-# LANGUAGE RankNTypes #-}
> > -- Type error with GHC 9.0.2 or 9.2.7, okay with GHC 8.8.4
> > import Control.Lens.Zoom (zoom,magnify)
> > import Control.Lens.Type (Lens')
> > import Control.Monad.RWS (RWS)
> > zoomRWS ::    Monoid w => Lens' s s' -> RWS r w s' a -> RWS r w s a
> > zoomRWS = zoom
> > magnifyRWS :: Monoid w => Lens' r r' -> RWS r' w s a -> RWS r w s a
> > magnifyRWS = magnify
> > 
> > Indeed GHC 8.8.4 compiles this happily (on lens-4.18.1), but 9.0.2 (on
> > lens-5.0.1) throws nearly undecipherable errors, claiming that Lens'
> > can not be matched with some more specialized type involving Magnified
> > and LensLike.
> > The source of Control.Lens.Zoom does not differ substantially between
> > 4.18.1 and 5.0.1 so my guess is it is a type checker issue. Can anyone
> > explain? On which bug tracker should I raise this issue, if it is
> > indeed a valid one?
> > (I pulled the above versions from stackage lts-16.31, -19.5 and -20.20,
> > respectively and compiled with stack.)
> > 
> > Thanks
> > Olaf
> > 
> > _______________________________________________
> > Haskell-Cafe mailing list
> > To (un)subscribe, modify options or view archives go to:
> > http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
> > Only members subscribed via the mailman list are allowed to post.
> 
> 
>