[ghc-steering-committee] Amend or patterns (#522) to use p1 ; p2 (#609)

[Chris Dornan]

With regards backwards compatibility I have asked the authors to clarify whether implementing the proposal is expected to break anything when the extension is not enabled.

Chris

> On 10 Sep 2023, at 18:11, Arnaud Spiwack <arnaud.spiwack at tweag.io> wrote:
> 
> I'm in favour of acceptance. The possibility to use layout, like in Chris's example in his recommendation email, is especially convincing to me: in Ocaml, many or patterns are layed out that way (despite the fact that Ocaml's grammar is insensitive to layout). Something like
> 
> foo = function
>   | A -> a
>   | B
>   | C -> b
>   | D -> d
> 
> It feels very natural there, it'll feel natural in Haskell too.
> 
> ---
> 
> Regarding backward compatibility, my understanding is the same as Moritz: in its current text, pattern synonyms are parsed differently regardless of whether -XOrPattern is turned on. This looks like a rather minor breakage, so I'm happy to follow the rest of the committee on the stance to take regarding this breakage. However, I would point out that this change to the pattern synonym syntax doesn't appear to be motivated in the text, at least it's not immediately apparent to me why it was deemed necessary. Maybe I missed something?
> 
> On Sun, 10 Sept 2023 at 07:26, Moritz Angermann <moritz.angermann at gmail.com <mailto:moritz.angermann at gmail.com>> wrote:
>> Joachim,
>> 
>> I have absolutely no issue with new extensions rejecting code that compiles fine without them. My only objections is to GHC not accepting code it accepted before without a deprecation period in between.
>> 
>> The only thing I object is GHC X compiling code just fine and GHC X+1 rejecting the exact same code.
>> 
>> Imo I think only X+3 should be permitted to reject code that X compiled. And X+1 and X+2 should warn about the upcoming change to syntax.
>> 
>> But I can compromise on X+2, with X+1 having deprecation warnings.
>> 
>> What I am unwilling to permit is X+1 rejecting code that X accepted without warning. That is sudden breakage of code.
>> 
>> This form of basic absolutely minimal backwards compatibility is absolutely essential for proper regression testing of the compiler. If I can not trivially swap out the compiler in a working codebase, we have no hope for proper regression testing or quality control.
>> 
>> Best,
>>  Moritz
>> 
>> On Sun, 10 Sep 2023 at 12:47 PM, Joachim Breitner <mail at joachim-breitner.de <mailto:mail at joachim-breitner.de>> wrote:
>>> 09.09.2023 21:17:01 Moritz Angermann <moritz.angermann at gmail.com <mailto:moritz.angermann at gmail.com>>:
>>> 
>>> Joachim,
>>> 
>>> My understand if the cost and drawbacks section[1], is that existing code breaks even without explicitly enabling the extension. If this is indeed not the case it should be called our explicitly in the section that breakage requires the extension to be enabled. Also an explanation why the patsyn test cases fail. Do we automatically enable that extension in the testsuite? Where does the regression come from?
>>> 
>>> Maybe I’m misreading that section and it just needs to be clarified that there is _no breakage without **explixitly** enabling the extension_ to existing code. 
>>> 
>>> Cheers,
>>>  Moritz
>>> 
>>> [1]: 
>>> https://github.com/ghc-proposals/ghc-proposals/blob/eb4b67c29282520b2c5c6a49c3047dbecb15dde1/proposals/0522-or-patterns.rst#costs-and-drawbacks
>>> 
>>> On Sun, 10 Sep 2023 at 12:04 PM, Joachim Breitner <mail at joachim-breitner.de <mailto:mail at joachim-breitner.de>> wrote:
>>>> Hi, 
>>>> 
>>>> since this is guarded by an extension that doesn't even exist yet, no code is broken, is there? 
>>>> 
>>>> I also don't expect this to be enabled in the future without coinciding with an intentional action by the developers - enabling this extension or switching to a future language edition that has this enabled by default (should that ever exist). Is it not sufficient if they are _then_ bothered with this change? 
>>>> 
>>>> (That said, we could say that a unparenthized type annotation on a pattern synonym is simply confusing, and thus use a warning to nudge the developers to add the parentheses now.) 
>>>> 
>>>> So not opposed to an early warning, I just don't think it's strictly necessary for this change. 
>>>> 
>>>> Cheers, 
>>>> Joachim 
>>>> 
>>>> _______________________________________________
>>>> ghc-steering-committee mailing list
>>>> ghc-steering-committee at haskell.org <mailto:ghc-steering-committee at haskell.org>
>>>> https://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-steering-committee
>>> 
>>> Hi, 
>>> 
>>> I would assume that all this only applies under {- LANGUAGE Or Patterns -}, but it seems it's not actually explicitly stated as such. 
>>> 
>>> I understand the breakage section as “what code needs to be changed when you enable the extension”. But maybe that's too optimistic? 
>>> 
>>> Chris, can you get clarification on this? 
>>> 
>>> Mortiz, assuming all changes are guarded by an extension, do you agree that no early warning would be necessary? 
>>> 
>>> Cheers, 
>>> Joachim
>> _______________________________________________
>> ghc-steering-committee mailing list
>> ghc-steering-committee at haskell.org <mailto:ghc-steering-committee at haskell.org>
>> https://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-steering-committee
> 
> 
> -- 
> Arnaud Spiwack
> Director, Research at https://moduscreate.com <https://moduscreate.com/> and https://tweag.io <https://tweag.io/>.
> _______________________________________________
> 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/20230911/556879ae/attachment.html>