The GADT debate

[wren romano]

Hi all,

There's been some discussion about whether to consider including GADTs
in the new report, but it's been mixed up with other stuff in the
thread on incorporating extensions wholesale, which has unfortunately
preempted/preceded the discussion about how to go about having such
discussions(!).

My position on the debate is that we should avoid having the debate,
just yet. (Which I intended but failed to get across in the email
which unintentionally started this all off.) I think we have many much
lower-hanging fruit and it'd be a better use of our time to try and
get those squared away first. Doing so will help us figure out and
debug the process for having such debates, which should help the GADT
debate itself actually be fruitful. As well as making progress on
other fronts, so we don't get mired down first thing.

Whenever the debate occurs, here's a summary of the relevant emails so
that they are not lost, buried in the emails of time:


* Andres Löh puts forth criteria on how to judge whether extensions
should be included. Mentions GADTs as an example of something that, if
we decide they're a good idea in principle (criterion #1), then we
should work towards things which will help them be easier to
feasibly/sensibly specify (criterion #2)

    https://mail.haskell.org/pipermail/haskell-prime/2016-May/004104.html

* wren agrees with the criteria, loves GADTs, thinks they do not fully
satisfy criterion #2

    https://mail.haskell.org/pipermail/haskell-prime/2016-May/004115.html

* Simon Peyton Jones says type inference for GADTs has, in fact,
stabilized. Albeit, not entirely satisfactory because it's "a bit
operational"

    https://mail.haskell.org/pipermail/haskell-prime/2016-May/004119.html

* Henrik Nilsson thinks GADTs are one of the most important extensions
to Haskell (fwiw, wren agrees). Also worries about being able to
specify type inference declaratively.

    https://mail.haskell.org/pipermail/haskell-prime/2016-May/004120.html

* Dominique Devriese suggests including MonoLocalBinds, finds "let
should not be generalized" convincing

    https://mail.haskell.org/pipermail/haskell-prime/2016-May/004117.html

* wren finds "let should not be generalized" unconvincing

    https://mail.haskell.org/pipermail/haskell-prime/2016-May/004145.html

-- 
Live well,
~wren