371-non-magical-eq: more graceful migration

[int-index]

While implementing this proposal, I noticed that there are quite a few packages (just to name some: primitive, containers, Cabal, haddock, exceptions) that use ~ without enabling TypeOperators.

The proposal already includes a compatibility fallback, but does not cover this particular problem. This amendment extends the proposed compatibility fallback by making the usage of ~ without TypeOperators a warning instead of an error. In due course, the special case will be removed, but maintainers won’t have to modify their code immediately.

[phadej]

See haskell/core-libraries-committee#12 (comment) and following discussion

In fact from our perspective it would be better not to have the warning stage, it's just extra churn. We'll fix it when it breaks.

[int-index]

Yes, the discussion seems relevant (but not conclusive). When @simonmar says “We'll fix it when it breaks”, the hidden assumption is that they have complete control over their code and can just fix it. But what about external dependencies?

What if the maintainer takes a while to respond? Should I just postpone the migration to a newer GHC version until the transitive closure of all my dependencies get updated, or should I create dozens of forks to use in the meantime?

The warning allows me to fix my code without waiting for others to fix their code.

[int-index]

Another perspective here: https://gitlab.haskell.org/ghc/ghc/-/issues/18862#note_312815

Writing a ~ b does not require enabling -XTypeOperators.
^ this was my only potential gripe. I see no problem with this if we don't have to enable it.

[goldfirere]

I'm in support of this direction. Would you mind naming the warning in the proposal?

[int-index]

Would you mind naming the warning in the proposal?

Named them -Wtype-equality-out-of-scope and -Wtype-equality-requires-operators. Better suggestions are welcome. The latter one probably doesn’t need a flag at all, since enabling TypeOperators and disabling the warning is the same amount of effort, but I’m not sure if it’s a good idea to introduce a warning without a flag.

[int-index]

Let’s submit, hi @nomeata

[goldfirere]

I have recommended acceptance.

/remind me in a week to accept if there is no dissent from the committee.

[reminders-prs]

@goldfirere set a reminder for Dec 24th 2021

[reminders-prs]

👋 @goldfirere, accept if there is no dissent from the committee.