RFC: or-patterns in let and if / while let expressions

[Centril]

Rendered.
Tracking issue

Enables the use of or-patterns in let, if let, and while let constructs as in:

if let A(x) | B(x) | ... = expr {
    do_stuff_with(x);
}

while let A(x) | B(x) | ... = expr {
    do_stuff_with(x);
}

let Ok(index) | Err(index) = slice.binary_search(&x);

Acknowledgements & ping

• @scottmcm, fixing typos, and helping me with the RFC in general.

[Centril]

This change was also discussed in #2144.

[petrochenkov]

Looks like a stripped-down version of #1882, which was recently postponed.

This RFC adds significantly less new stuff though, since | is still a special construction, not a part of a pattern.
Since if let and while let are desugared into match anyway and match already supports this special construction, I see no reasons to not accept this.

[Centril]

@petrochenkov I wouldn't say this and #1882 are related tho. This only makes match, if let and while let consistent by letting the latter two do what the former already can do =)

[nugend]

I just saw this RFC in TWIR and funnily enough, a related crate was nominated for crate of the week: https://docs.rs/if_chain/0.1.2/if_chain/#multiple-patterns

[Centril]

@nugend Interesting! Perhaps worthy of mention in motivation, and possibly alternatives.

[cramertj]

It'd be cool to also have this in plain-old let bindings if the match is exhaustive:

#[derive(Copy, Clone)]
enum Foo {
    Bar(i32),
    Baz(i32),
}

impl Foo {
    fn inner(&self) -> i32 {
        let Bar(x) | Baz(x) = *self;
        x
    }
}

[Centril]

@cramertj I'm not opposed to this.. But is that a common enough use case tho? Do you have some real world examples I could work into the motivation perhaps?

[cramertj]

@Centril I'm not sure how common it is-- it also only saves a few characters compared to a match. For me, the motivation would be primarily about making patterns consistent.

[Centril]

@cramertj I guess I buy the consistency argument since I use it under motivation already - but it would be even more compelling with some actual example.

If anyone finds one, please let us know =)

[cramertj]

@Centril There're some functions like the example I gave in rustc, but usually they have one or two variants that don't bind the variable and bug! instead. I think the reason it's tough to come up with examples is that usually you'd write an enum like my Foo above as a struct with a kind enum:

// Instead of this:

#[derive(Copy, Clone)]
enum Foo {
    Bar(i32),
    Baz(i32),
}

impl Foo {
    fn inner(&self) -> i32 {
        let Bar(x) | Baz(x) = *self;
        x
    }
}


// People write this:

struct Foo {
    val: i32,
    kind: FooKind,
}

enum FooKind { Bar, Baz }

[Centril]

@cramertj Right. And isn't that the recommended way of writing it?

Devil's advocate: Wouldn't we risk encouraging the "wrong" way of writing this by allowing:

let Bar(x) | Baz(x) = *self;

?

[cramertj]

@Centril Yeah, that seems like a reasonable objection.

[Centril]

@cramertj Hmm.. Perhaps this can be made into a separate RFC since it might be more controversial? - so that this RFC can be merged faster (tho we have several months to discuss it before the impl period is over...)

[petrochenkov]

@cramertj
Instead of doing #2175 (comment) I'd suggest to go full #1882 (after the end of the implementation period).
The nice point of this RFC is that it's really minimal (if let and while let desugars into match supporting |) while let doesn't desugar into match increasing the scope of changes.

[nikomatsakis]

My take is that we definitely want this in let.

I have definitely encountered cases where I want to do:

let Foo(x) | Bar(x) = value;

An example would be in Chalk, where I have this enum:

enum ParameterKind<T, L = T> {
    Ty(T),
    Lifetime(L),
}

indicating e.g. the value of a parameter in a list like Foo<'a, T>. But note the default L = T, this is because often I don't need different data for the two cases. As it happens now, I have an impl like:

impl<T> ParameterKind<T> {
    fn into_inner(self) -> T { ... }
}

there are few times that I wished I could use let syntax to destructure.

It doesn't come up a lot, but as @cramertj says, to me having the syntax of if let and let diverge is pretty confusing.

[nikomatsakis]

Nominating for discussin @rust-lang/lang meeting. Seems like one of those "tiny extension to the syntax" sorts of RFCs that are always hard to judge. Obviously not a major blocker for anyone, but maybe a useful thing to go ahead and do? This seems to have pretty minimal implementation cost, I would think, since everything winds up desugared in the compiler to a match anyway. It's just about the "user cost". (Right?)

[Centril]

@nikomatsakis

An example would be in Chalk, where I have this enum:

That's an interesting use case! And I agree that diverging would be unfortunate.
Should this be fixed in this RFC, or revisited later?

[nikomatsakis]

@Centril I say fix it in the RFC now =)

[nikomatsakis]

@rfcbot fcp merge

Subjection to the proviso that the RFC is amended to permit | patterns in let, I move that we merge it. It can be hard to judge when it makes sense to do these sorts of small extensions to the syntax. This particular extension seems worthwhile primarily because it is mostly echoes syntax already found elsewhere in the language, so it makes things slightly more expressive without really making people learn new things they don't already have to know.

[rfcbot]

Team member @nikomatsakis has proposed to merge this. The next step is review by the rest of the tagged teams:

• @aturon
• @cramertj
• @eddyb
• @nikomatsakis
• @nrc
• @pnkfelix
• @withoutboats

No concerns currently listed.

Once these reviewers reach consensus, this will enter its final comment period. If you spot a major issue that hasn't been raised at any point in this process, please speak up!

See this document for info about what commands tagged team members can give me.

[Centril]

@nikomatsakis Amended =)

[rfcbot]

🔔 This is now entering its final comment period, as per the review above. 🔔

[rfcbot]

The final comment period is now complete.

[aturon]

This RFC has been merged!

Tracking issue

[nox]

This seems quite unfortunate to me that this landed. Other forms of sugar around if let and whatnot that actually brought more expressiveness to the table were rejected with "just use match", why are or-patterns considered differently?

[eddyb]

@nox IMO it's a language design bug that pattern | pattern is not itself a pattern, that can be nested in other patterns, etc. This is a bandaid, sure, but eventually we should fix it fully.

[nox]

@eddyb Mmmh, I guess that makes sense if you frame it this way. Thanks for the answer.