It turns out that Darcs never actually uses the full power of the GADTs
extension, that is, to restrict sum and products of some ADT *by
replacing some type variable by a concrete type* (see eg [1]).

Darcs' code does, however, restrict certain ADT by repeating the same
type variable:

    -- | Forward lists
    data FL a wX wZ where
        (:>:) :: a wX wY -> FL a wY wZ -> FL a wX wZ
        NilFL :: FL a wX wX

However (and to my surprise) this is handled already by the strictly
less poweful GADTSyntax extension (that exists since GHC 7.2) [2], which
now must be accompanied by ExistentialQuantification.

I haven't measured whether this change makes Darcs' compilation faster.
I hope so!

Anyway I think it is a good idea that we use the smallest language
extensions possible; when we need real GADTs we can activate them again.

[1] http://darcs.net/Internals/200801-ganesh-gadts.pdf
[2]
https://downloads.haskell.org/~ghc/latest/docs/html/users_guide/glasgow_exts.html#declaring-data-types-with-explicit-constructor-signatures


1 patch for repository http://darcs.net:

patch 4fe172b29bfc4da016ec2f78c81c20a5f093f999
Author: Guillaume Hoffmann <guillaumh@gmail.com>
Date:   Thu Feb 22 12:04:27 -03 2018
  * replace GADTs extension by GADTSyntax and ExistentialQuantification

Attachments

• replace-gadts-extension-by-gadtsyntax-and-existentialquantification.dpatch

So the explanation for why can we repeat type variables even with just
GADTSyntax is that it is enabled by the ExistentialQuantification extension.

I'm attaching and screening a patch that does the same as the previous
one for the test harness, and also disables all extensions for the darcs
executable.

Attachments

• also-use-gadtsyntax-in-test-harness_-disable-all-extensions-for-executable.dpatch

Self-accept.