[ghc-steering-committee] Please review #319: NoFallibleDo proposal, Shepherd: Eric Seidel

Fri Dec 17 13:35:22 UTC 2021

To recapitulate, this proposal remarks that GHC does a
pre-pattern-matching-exhaustiveness check during the renaming phase to
figure out how whether it needs to insert fail in do-notation statements of
the form C x1… xn <- u. It notes that this precheck is heuristic, and
decides whether a MonadFail constraint will be required during type
checking. It argues that this is bad because you may end up being asked for
a MonadFail constraint when you were careful to write an exhaustive pattern
dependent on a GADT’s argument. And that it is a fundamental limitation of
the current desugaring scheme.

All these observations look correct (and were validated, in the PR’s
discussion by Sebastian Graf who is quite well-versed in the relevant parts
of the code). The argument that it is a bad thing is relatable (heuristic
in type checking are generally undesirable).

The proposal’s solution is to add an extension which would switch the
desugaring of C x1… xn <- u statements to never use fail (instead throw an
imprecise exception like regular case expressions). And have
incomplete-pattern-matching warnings. Basically treating <- like a let in
this context. I think.

The problem I see with this solution is that, while it can be argued that
it would have been desirable to do just that at the inception of Haskell,
it is quite unlikely that we can ever make this the default (considering
how much code exists that rely on the current fail-based desugaring, and
how hard it would be to track). So such an extension would remain as a
switch forever. And I don’t find that it’s particularly worth it.
------------------------------

John Ericsson, who is one of the co-author of the proposal, also wrote a
companion proposal https://github.com/ghc-proposals/ghc-proposals/pull/327
In this second proposal, he introduces a new statement form for the do
notation: case C x1 … xn <- u of { alts }, such that

do
  { case C x1 … xn <- u of { alts }
  ; stmts }

desugars to

do
  case u of
  { C x1 … xn -> do { stmts }
  ; alts }

The proposal is still marked as WIP, but the general idea is reasonable.
This is intended as a replacement of the fail desugaring, by inserting fail
(or whatever you see fit) manually: case C x1 … xn <- u of { _ -> fail }.

But, from my point of view, this proposal could just as well serve as a
replacement of the NoFailibleDo proposal being discussed in this thread.
Since you could force a fail-free pattern-matching as case C x1 … xn <- u
of {}.

I personally find this direction much more compelling.

/Arnaud

On Fri, Dec 17, 2021 at 10:48 AM Alejandro Serrano Mena <trupill at gmail.com>
wrote:

> Dear all,
> I missed it back then, but the authors of the “NoFallibleDo” proposal have
> re-submitted to the Committee.
>
> It seems though that we are still in some form of impasse, without leaning
> towards acceptance or rejection. From the discussion, I think that the
> feeling is that this is a desirable feature, but there are different
> opinions about whether this should be per-module or per-block. It would be
> great if all of us would discuss this matter (either here or in the GitHub
> thread) and try to come to a conclusion (or ultimately cast a vote to
> decide).
>
> The proposal itself is about being able to tweak whether an incomplete
> pattern match in a ‘do’ block generates a call to ‘fail’ — as it does now,
> leading to an additional MonadFail constraint — or works as any other
> pattern match — leading to a PatternMatchFail exception when a non-matching
> value comes there.
>
> Once again, I would love to hear your opinions :)
>
> Regards,
> Alejandro
>
>
> El 23 jul 2021 13:40:26, Alejandro Serrano Mena <trupill at gmail.com>
> escribió:
>
>> I’ve been made aware that the “NoFallibleDo” proposal has been
>> re-submitted to the Committee. My current recommendation is “reject”, as
>> outlined in the following comment
>> https://github.com/ghc-proposals/ghc-proposals/pull/319#issuecomment-885580010 (TL;DR,
>> you’d often like to enable this for a particular “do” block, not for an
>> entire file).
>>
>> Regards,
>> Alejandro
>>
>>
>> El 28 jul 2020 11:33:02, Alejandro Serrano Mena <trupill at gmail.com>
>> escribió:
>>
>>> Done. Once again, sorry for the confusion.
>>>
>>> Alejandro
>>>
>>> El mar., 28 jul. 2020 a las 11:30, Simon Peyton Jones (<
>>> simonpj at microsoft.com>) escribió:
>>>
>>>> OK, so to summarise
>>>>
>>>>    - We are waiting for the author
>>>>    - You are encouraging us to comment anyway
>>>>
>>>>
>>>> Correct?  Does the author know this?   Why encourage only us?   Maybe
>>>> post on Github to clarify the status, and encourage everyone to contribute.
>>>>
>>>>
>>>>
>>>> S
>>>>
>>>>
>>>>
>>>> *From:* Alejandro Serrano Mena <trupill at gmail.com>
>>>> *Sent:* 28 July 2020 10:25
>>>> *To:* Simon Peyton Jones <simonpj at microsoft.com>
>>>> *Cc:* ghc-steering-committee at haskell.org
>>>> *Subject:* Re: [ghc-steering-committee] Please review #319:
>>>> NoFallibleDo proposal, Shepherd: Eric Seidel
>>>>
>>>>
>>>>
>>>> I mean the last status, push back to the author for revision.
>>>>
>>>>
>>>>
>>>> Alejandro
>>>>
>>>>
>>>>
>>>> El mar., 28 jul. 2020 a las 11:24, Simon Peyton Jones (<
>>>> simonpj at microsoft.com>) escribió:
>>>>
>>>> So I’m still confused.  “We went back to GIthub”… does that mean that
>>>> we invited the author to revise and resubmit?  I don’t know what else “back
>>>> to github” means.
>>>>
>>>>
>>>>
>>>> If it’s in committee-decision status, then our process says  should
>>>> either accept, reject, or push back to the author for revision, in a timely
>>>> way (guided by the shepherd)
>>>>
>>>>
>>>>
>>>> Simon
>>>>
>>>>
>>>>
>>>> *From:* Alejandro Serrano Mena <trupill at gmail.com>
>>>> *Sent:* 28 July 2020 10:22
>>>> *To:* Simon Peyton Jones <simonpj at microsoft.com>
>>>> *Cc:* ghc-steering-committee at haskell.org
>>>> *Subject:* Re: [ghc-steering-committee] Please review #319:
>>>> NoFallibleDo proposal, Shepherd: Eric Seidel
>>>>
>>>>
>>>>
>>>> Eric was initially in charge, but I took over his duties. He thought
>>>> that a bit more discussion was needed, something I agree with, so we went
>>>> back to GitHub.
>>>>
>>>>
>>>>
>>>> Sorry about the stale status, I feel that my back-and-forth was not
>>>> very clear.
>>>>
>>>>
>>>>
>>>> Alejandro
>>>>
>>>>
>>>>
>>>> El mar., 28 jul. 2020 a las 11:17, Simon Peyton Jones (<
>>>> simonpj at microsoft.com>) escribió:
>>>>
>>>> Alejandro, this one hasn’t been on my radar.
>>>>
>>>>
>>>>
>>>> Are you the shepherd?   Have you made a recommendation?  Is the
>>>> proposal in its final form  -- ie having absorbed all discussion etc?
>>>>
>>>>
>>>>
>>>> Simon
>>>>
>>>>
>>>>
>>>> *From:* ghc-steering-committee <
>>>> ghc-steering-committee-bounces at haskell.org> *On Behalf Of *Alejandro
>>>> Serrano Mena
>>>> *Sent:* 28 July 2020 09:22
>>>> *To:* Joachim Breitner <mail at joachim-breitner.de>
>>>> *Cc:* ghc-steering-committee <ghc-steering-committee at haskell.org>
>>>> *Subject:* Re: [ghc-steering-committee] Please review #319:
>>>> NoFallibleDo proposal, Shepherd: Eric Seidel
>>>>
>>>>
>>>>
>>>> Dear Committee,
>>>>
>>>> I would like to kindly ask for your input in the NoFallibleDo proposal
>>>> -> https://github.com/ghc-proposals/ghc-proposals/pull/319
>>>> <https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fghc-proposals%2Fghc-proposals%2Fpull%2F319&data=02%7C01%7Csimonpj%40microsoft.com%7C250fdb186a4a41772fe908d832d82b6c%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637315251305340206&sdata=DHvjx8qWwjNaa9jn1mdfyBBOqMLJAXdozi3otokRKbk%3D&reserved=0>
>>>>
>>>> This was submitted, then there was some discussion, but the
>>>> conversation has stalled.
>>>>
>>>>
>>>>
>>>> Regards,
>>>>
>>>> Alejandro
>>>>
>>>>
>>>>
>>>> El jue., 14 may. 2020 a las 17:30, Alejandro Serrano Mena (<
>>>> trupill at gmail.com>) escribió:
>>>>
>>>> @Eric congratulations! enjoy! :)
>>>>
>>>>
>>>>
>>>> @Joachim I can take care of this, I think the direction Eric was
>>>> pushing this is a good one.
>>>>
>>>>
>>>>
>>>> El jue., 14 may. 2020 a las 12:16, Joachim Breitner (<
>>>> mail at joachim-breitner.de>) escribió:
>>>>
>>>> Hi,
>>>>
>>>> Am Mittwoch, den 13.05.2020, 15:19 -0500 schrieb Eric Seidel:
>>>> > My wife and I just checked into the hospital to have our second child
>>>>
>>>> Congrats, and all the best!
>>>>
>>>> > , so I’m going to be short on time for committee duties for a few
>>>> > weeks. I think it would be best to reassign this proposal so we don’t
>>>> > keep the authors waiting.
>>>>
>>>> Any volunteers?
>>>>
>>>> Cheers,
>>>> Joachim
>>>> --
>>>> Joachim Breitner
>>>>   mail at joachim-breitner.de
>>>>   http://www.joachim-breitner.de/
>>>> <https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.joachim-breitner.de%2F&data=02%7C01%7Csimonpj%40microsoft.com%7C250fdb186a4a41772fe908d832d82b6c%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637315251305340206&sdata=Xf0HrXmVOPpftJaGUUXQo2ztLCjz1sRhN4nJPP%2Fbshg%3D&reserved=0>
>>>>
>>>>
>>>> _______________________________________________
>>>> ghc-steering-committee mailing list
>>>> ghc-steering-committee at haskell.org
>>>> https://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-steering-committee
>>>> <https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fmail.haskell.org%2Fcgi-bin%2Fmailman%2Flistinfo%2Fghc-steering-committee&data=02%7C01%7Csimonpj%40microsoft.com%7C250fdb186a4a41772fe908d832d82b6c%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637315251305350196&sdata=P2ISFU6yTiEJkwsF0HmSiGnKUKZCJ0CpLOhGZkDLBpI%3D&reserved=0>
>>>>
>>>> _______________________________________________
> ghc-steering-committee mailing list
> ghc-steering-committee at haskell.org
> https://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-steering-committee
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-steering-committee/attachments/20211217/e64fe396/attachment-0001.html>