Add warning category to WARNINGs on Data.List.{head,tail}

[adamgundry]

Recently the CLC approved proposal #87 to add WARNING pragmas to head and tail from Data.List. Users who do not wish to see such warnings are advised to use -Wno-warnings-deprecations. However this has the effect of disabling all messages from WARNING pragmas, which is not necessarily what the user wants.

Partly motivated by this, and in order to provide more flexibility to users, the GHC Streering Committee has subsequently accepted ghc-proposals/ghc-proposals#541 that permits WARNING pragmas to be annotated with a category name. In particular, it will permit the following:

{-# WARNING in "x-partial" head "This is a partial function...." #-}
{-# WARNING in "x-partial" tail "This is a partial function...." #-}

Users will then be able to disable these warnings with -Wno-x-partial (and similar flags, e.g. promoting them to an error with -Werror=x-partial). They are no longer under the control of -Wwarnings-deprecations (but there is a new warning group -Wextended-warnings that covers all messages from WARNING pragmas regardless of category).

I propose that the x-partial warning category be added to the WARNING pragmas being added to head and tail by !9290.

This is no more of a breaking change than the original #87, provided it happens in the same GHC release. (If it happens in a subsequent release, users who set -Wno-warnings-deprecations will start to see warnings again until they set -Wno-x-partial.)

An implementation of warning categories in GHC is available at !9684. It seems likely that this should be ready in time for GHC 9.8, which is also likely to be the first release that can include #87.

[Kleidukos]

Excellent suggestion!

With my Haddock hat on:

I'd be delighted to have such a mechanism that I could re-use from inside Haddock to signify the cause of warnings.
There is an adjacent thread in the Forum: https://discourse.haskell.org/t/collectings-occurences-of-incomplete-patterns-to-make-them-available-to-haddock/5494/

With my user hat on:

I support the addition of a partiality tag for the WARNING pragma, and even go further:

• Add it to the Errors index;
• Have a strict, unambiguous meaning of "partiality" in this context that users can read on the errors index page.

[nomeata]

Excellent idea to use the error index to help with consistent warning categories across libraries!

[Bodigrim]

It would be nice to have a framework of softly recommended categories lest we end up with x-partial, x-non-total and x-diverging. But I don't know how to foster such catalog. Having an entry in the error index sounds like a good idea.

Narrowing back to this ticket, can we agree on a CLC guidance for users which functions are suggested to use x-partial?

[jappeace]

It's a good idea,
but the claim this effects all warnings is an exaggeration since you can disable warnings per module.
For example adding {-# OPTIONS_GHC -Wno-warnings-deprecations #-} on the top of your module.

[adamgundry]

I've updated the issue to reflect a minor syntactic change in the design of the pragma. Please could the CLC make a decision on this?

I agree that it could be useful to define a catalogue of recommended categories, perhaps alongside the errors index, and to have guidance on when x-partial in particular is applicable. I don't believe that is necessary for this change, however.

the claim this effects all warnings is an exaggeration since you can disable warnings per module.

The text refers to "disabling all messages from WARNING pragmas" which is precisely the effect of -Wno-warnings-deprecations (prior to the introduction of categories).

[hasufell]

Sounds good. Since I'm not a friend of warning about head/tail in the first place, anything that makes it less annoying is welcome.

[adamgundry]

@Bodigrim

Please could the CLC make a decision on this?

[parsonsmatt]

+1

[Bodigrim]

I don't believe that is necessary for this change, however.

OK, let's vote on the proposal as is.

---

Dear CLC members, let's vote on the proposal to classify warnings for Data.List.{head,tail} as belonging to partial group, allowing a fine-grained control over them with -Wno-x-partial. This has been discussed originally in the course of #87, but the GHC proposal for custom warnings hasd not been approved yet. The implementation of the feature is in https://gitlab.haskell.org/ghc/ghc/-/merge_requests/9684, which I assume could be merged reasonably soon.

@tomjaguarpaw @chshersh @angerman @mixphix @hasufell

---

+1 from me.

[mixphix]

+1

[konsumlamm]

This has been discussed originally in the course of #87, but the GHC proposal for custom warnings has not been approved yet.

I assume you mean "the GHC proposal for custom warnings had not been approved yet"? It has been accepted since then.

[angerman]

More diagnostics sound good to me +1. It may break people with -Weverything -Werror, but then you are asking for it.

[konsumlamm]

More diagnostics sound good to me +1. It may break people with -Weverything -Werror, but then you are asking for it.

Note that head and tail already have these warnings (see #87), this proposal is about making it possible to control them with a new warning category. So this doesn't break anything that wasn't already broken before.

[hasufell]

+1