<div dir="ltr"><div>What's simplest is not obvious.</div><div><br></div><div>- Only accepting concrete `!pat` if there's a %p gives simpler error messages, but is a few lines of code extra (and you may end up with lets that are inferred linear but you can't add `let %1` on them if the inference algorithm uses a more clever definition of strict)<br></div><div>- Considering a pattern as strict when `-XStrict` or in brackets, is a few extra lines of code but maybe more intuitive.</div><div><br></div><div>In any case, I consider all these choices to be quite comparable in simplicity from a code perspective (and for the record, the least amount of work for me is to use both alternatives, because they're the version that I've already implemented). What does the user consider simplest?<br></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Sat, 13 Jan 2024 at 21:00, Adam Gundry <<a href="mailto:adam@well-typed.com">adam@well-typed.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">I haven't looked at this closely, but I'm happy to trust Arnaud and <br>
Richard's judgment and accept, especially since LinearTypes is at a very <br>
experimental stage.<br>
<br>
Adam<br>
<br>
<br>
On 12/01/2024 10:08, Arnaud Spiwack wrote:<br>
> At Richard's prompting, I've added the following alternative to the <br>
> proposal (the current proposal is the most conservative of the two, <br>
> which we can choose to stick to if we're unsure). I'm copying the <br>
> alternative here because rendering seems to be broken on Github right now.<br>
> <br>
> I'm rather agnostic on which side we choose, to be honest. Anyone with <br>
> medium-to-strong opinions on the question?<br>
> <br>
> <br>
> Restrictions of multiplicity-annotated let bindings<br>
> ---------------------------------------------------<br>
> <br>
> The proposal specifies that a multiplicity annotated non-variable let <br>
> binding ``let %p pat``<br>
> must be such that ``pat = !pat'`` even if ``p = 'Many``. It is easy to<br>
> lift this restriction on two dimension:<br>
> <br>
> - We can say, instead, that patterns not of the form ``!pat'`` emit a<br>
> ``p ~ 'Many`` constraint instead. They already do (for the sake of<br>
> inference), so this is strictly less code.<br>
> - We can generalise to more strict patterns. For instance, we don't<br>
> need to require a ``!`` if ``-XStrict`` is on, we can have patterns<br>
> of the form ``(!pat')`` (with additional parentheses). This is a few<br>
> lines of codes, inference actually already does this in my<br>
> implementation, so it's already paid for (though it does annoyingly<br>
> mostly duplicate another definition of strict pattern which I<br>
> couldn't find a way to factor as a single function, I don't like<br>
> this).<br>
> <br>
> The reason that motivated the stronger restriction is to improve error<br>
> messages, because we can then error out with “multiplicity-annotated<br>
> let-bound patterns must be of the form !pat”, instead of the more<br>
> mysterious “Couldn't unify 'Many with 'One”<br>
> (see `#23586 <<a href="https://gitlab.haskell.org/ghc/ghc/-/issues/23586" rel="noreferrer" target="_blank">https://gitlab.haskell.org/ghc/ghc/-/issues/23586</a> <br>
> <<a href="https://gitlab.haskell.org/ghc/ghc/-/issues/23586" rel="noreferrer" target="_blank">https://gitlab.haskell.org/ghc/ghc/-/issues/23586</a>>>`_).<br>
> But maybe the additional restrictions are more surprising than the<br>
> error messages are helpful.<br>
> <br>
> On Mon, 8 Jan 2024 at 10:07, Simon Peyton Jones <br>
> <<a href="mailto:simon.peytonjones@gmail.com" target="_blank">simon.peytonjones@gmail.com</a> <mailto:<a href="mailto:simon.peytonjones@gmail.com" target="_blank">simon.peytonjones@gmail.com</a>>> wrote:<br>
> <br>
> I support acceptance. Let's land this soon.<br>
> <br>
> Simon<br>
> <br>
> On Sat, 6 Jan 2024 at 04:45, Richard Eisenberg <<a href="mailto:rae@richarde.dev" target="_blank">rae@richarde.dev</a><br>
> <mailto:<a href="mailto:rae@richarde.dev" target="_blank">rae@richarde.dev</a>>> wrote:<br>
> <br>
> Hi all,<br>
> <br>
> I've reviewed Arnaud's<br>
> <a href="https://github.com/ghc-proposals/ghc-proposals/pull/624" rel="noreferrer" target="_blank">https://github.com/ghc-proposals/ghc-proposals/pull/624</a><br>
> <<a href="https://github.com/ghc-proposals/ghc-proposals/pull/624" rel="noreferrer" target="_blank">https://github.com/ghc-proposals/ghc-proposals/pull/624</a>> and<br>
> wish to recommend acceptance.<br>
> <br>
> The proposal is an amendment to the proposal for linear types,<br>
> adding support for linear let bindings.<br>
> <br>
> Today, if you have<br>
> <br>
> f :: T %1-> T<br>
> f t = let t2 = t in t2<br>
> <br>
> you'll get an error because t2 is not linear. The only way to<br>
> bind a linear variable is via a `case`, never a `let` or<br>
> `where`. This is annoying. With this proposal, the little<br>
> program above is accepted, with an inferred linearity<br>
> restriction on t2. Users can also annotated their lets like `let<br>
> %1 x = ... in ...`. Bindings in `where` clauses can also be<br>
> inferred or annotated as linear.<br>
> <br>
> There is a downside, of course: linear bindings have various<br>
> restrictions, chiefly that they must be strict bindings (because<br>
> projections are hard with linear types) and that bindings cannot<br>
> be generalized. I'm a little unsure that the choices in the<br>
> proposal (particularly around generalization) are the best for<br>
> users, but I think the best way to learn is to experiment. In my<br>
> understanding, the community knows that -XLinearTypes is subject<br>
> to revision, and so I think we should just blast ahead, revising<br>
> if and when necessary.<br>
> <br>
> Please share your thoughts!<br>
> <br>
> Thanks,<br>
> Richard<br>
<br>
<br>
-- <br>
Adam Gundry, Haskell Consultant<br>
Well-Typed LLP, <a href="https://www.well-typed.com/" rel="noreferrer" target="_blank">https://www.well-typed.com/</a><br>
<br>
Registered in England & Wales, OC335890<br>
27 Old Gloucester Street, London WC1N 3AX, England<br>
<br>
_______________________________________________<br>
ghc-steering-committee mailing list<br>
<a href="mailto:ghc-steering-committee@haskell.org" target="_blank">ghc-steering-committee@haskell.org</a><br>
<a href="https://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-steering-committee" rel="noreferrer" target="_blank">https://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-steering-committee</a><br>
</blockquote></div><br clear="all"><br><span class="gmail_signature_prefix">-- </span><br><div dir="ltr" class="gmail_signature"><div dir="ltr">Arnaud Spiwack<br>Director, Research at <a href="https://moduscreate.com" rel="noopener noreferrer" target="_blank">https://moduscreate.com</a> and <a href="https://tweag.io" rel="noopener noreferrer" target="_blank">https://tweag.io</a>.</div></div>