Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

loop-break-value (issue #961) #1624

Merged
merged 14 commits into from
Oct 22, 2016
Merged
Changes from 12 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
290 changes: 290 additions & 0 deletions text/0000-loop-break-value.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,290 @@
- Feature Name: loop_break_value
- Start Date: 2016-05-20
- RFC PR: (leave this empty)
- Rust Issue: (leave this empty)

# Summary
[summary]: #summary

(This is a result of discussion of
[issue #961](https://github.com/rust-lang/rfcs/issues/961) and related to RFCs
[352](https://github.com/rust-lang/rfcs/pull/352) and
[955](https://github.com/rust-lang/rfcs/pull/955).)

Let a `loop { ... }` expression return a value via `break my_value;`.

# Motivation
[motivation]: #motivation

> Rust is an expression-oriented language. Currently loop constructs don't
> provide any useful value as expressions, they are run only for their
> side-effects. But there clearly is a "natural-looking", practical case,
> described in [this thread](https://github.com/rust-lang/rfcs/issues/961)
> and [this] RFC, where the loop expressions could have
> meaningful values. I feel that not allowing that case runs against the
> expression-oriented conciseness of Rust.
> [comment by golddranks](https://github.com/rust-lang/rfcs/issues/961#issuecomment-220820787)

Some examples which can be much more concisely written with this RFC:

```rust
// without loop-break-value:
let x = {
let temp_bar;
loop {
...
if ... {
temp_bar = bar;
break;
}
}
foo(temp_bar)
};
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why not the following?

let x;
loop {
    ...
    if ... {
        x = foo(bar);
        break;
    }
}


// with loop-break-value:
let x = foo(loop {
...
if ... { break bar; }
});

// without loop-break-value:
let computation = {
let result;
loop {
if let Some(r) = self.do_something() {
result = r;
break;
}
}
result.do_computation()
};
self.use(computation);
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why not the following?

let computation;
loop {
    if let Some(r) = self.do_something() {
        computation = r.do_computation();
    }
}
self.use(computation);

Or

let computation;
loop {
    if let Some(r) = self.do_something() {
        computation = r;
        break;
    }
}
let computation = computation.do_computation();
self.use(computation);


// with loop-break-value:
let computation = loop {
if let Some(r) = self.do_something() {
break r;
}
}.do_computation();
self.use(computation);
```

# Detailed design
[design]: #detailed-design

This proposal does two things: let `break` take a value, and let `loop` have a
result type other than `()`.

### Break Syntax

Four forms of `break` will be supported:

1. `break;`
2. `break 'label;`
3. `break EXPR;`
4. `break 'label EXPR;`
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is the ; after break EXPR compulsory? Currently existing forms of break do not require a semicolon after if they are in expression position.

Could we get an example of what would the grammar look like with syntax changes? See this and this.

Copy link

@vitiral vitiral Oct 7, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would say that the ; is not necessary. It should map how return works, and the following compiles:

fn test() -> bool {
    if true {
        return true
    } else {
        return false
    }
}

fn test2() -> bool {
    return false
}

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes just like today

---------------
(break 'a e): !


where `'label` is the name of a loop and `EXPR` is an expression.

### Result type of loop

Currently the result type of a 'loop' without 'break' is `!` (never returns),
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

s/'/`/

Copy link
Contributor Author

@dhardy dhardy Oct 5, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The RFC is written in English, not code. shrug. 'loop' isn't a code snippet in this context, it's a quoted keyword.

which may be coerced to any type), and the result type of a 'loop' with 'break'

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

consistency nit: "result-type" vs "result type"

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

s/'/`/

I will not nit on any more of the formatting, but the issues with it probably should be fixed across the document. Namely, plain apostrophe (') does not begin/end an inline code snippet.

is `()`. This is important since a loop may appear as
the last expression of a function:

```rust
fn f() {
loop {
do_something();
// never breaks
}
}
fn g() -> () {
loop {
do_something();
if Q() { break; }
}
}
fn h() -> ! {
loop {
do_something();
// this loop is not allowed to break due to inferred `!` type
}
}
```

This proposal changes the result type of 'loop' to `T`, where:

* if a loop is "broken" via `break;` or `break 'label;`, the loop's result type must be `()`
* if a loop is "broken" via `break EXPR;` or `break 'label EXPR;`, `EXPR` must evaluate to type `T`
* as a special case, if a loop is "broken" via `break EXPR;` or `break 'label EXPR;` where `EXPR` evaluates to type `!` (does not return), this does not place a constraint on the type of the loop

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't see the need for this special case: there's no reason to write break EXPR; instead of EXPR; when the expr is diverging.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No, and I suppose this is why it's not legal to type let x: ! = ...;.

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Well, that's because ! is not a real type (yet - there is an RFC open for that) and is only allowed in function return type notation.

Copy link
Member

@nagisa nagisa Oct 4, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Now ! is a legal type everywhere, but this special case is still useful as a note, even though it follows just fine from rules applicable to ! (i.e. ! coerces to any other type).

(EDIT: Weird, this was a reply to an existing thread on the same line; explains why it has no the reply field though…)

* if external constaint on the loop's result type exist (e.g. `let x: S = loop { ... };`), then `T` must be coercible to this type
Copy link
Member

@nagisa nagisa Oct 4, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This list still has issues with handling of coercion. For example I imagine something like this ought to be valid:

loop {
     // ...
     break (expr: SomeT);
     // ...
     break (expr: OtherT);
}: RT

where SomeT and OtherT both coerce to RT. If this discrepancy was fixed, the list could avoid having to contain the last two notes/requirements. I also dislike those two extra requirements-exceptions-rules, because they have potential to diverge from type rules elsewhere in the language.

(Why this matters? Substitute SomeT = &[u8; 42], OtherT = &[u8; 32], RT = &[u8].)

Copy link
Contributor

@Ericson2314 Ericson2314 Oct 5, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think the typing and coersions rules here can be exactly like those for match arms or if-else.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please write a diff against the RFC; I was not 100% sure how to handle coercions (same with the ! special case).


It is an error if these types do not agree or if the compiler's type deduction
rules do not yield a concrete type.

Examples of errors:

```rust
// error: loop type must be () and must be i32
let a: i32 = loop { break; };
// error: loop type must be i32 and must be &str
let b: i32 = loop { break "I am not an integer."; };
// error: loop type must be Option<_> and must be &str
let c = loop {
if Q() {
break "answer";
} else {
break None;
}
};
fn z() -> ! {
// function does not return
// error: loop may break (same behaviour as before)
loop {
if Q() { break; }
}
}
```

Examples involving `!`:

```rust
fn f() -> () {
// ! coerces to ()
loop {}
}
fn g() -> u32 {
// ! coerces to u32
loop {}
}
fn z() -> ! {
loop {
break panic!();
}
}
```

Example showing the equivalence of `break;` and `break ();`:

```rust
fn y() -> () {
loop {
if coin_flip() {
break;
} else {
break ();
}
}
}
```

### Result value

A loop only yields a value if broken via some form of `break ...;` statement,
in which case it yields the value resulting from the evaulation of the
statement's expression (`EXPR` above), or `()` if there is no `EXPR`
expression.

Examples:

```rust
assert_eq!(loop { break; }, ());
assert_eq!(loop { break 5; }, 5);
let x = 'a loop {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

syntactic nit: this should be 'a: loop { and 'b: loop { on the line below.

'b loop {
break 'a 1;
}
break 'a 2;
};
assert_eq!(x, 1);
```

# Drawbacks
[drawbacks]: #drawbacks

The proposal changes the syntax of `break` statements, requiring updates to
parsers and possibly syntax highlighters.

# Alternatives
[alternatives]: #alternatives

No alternatives to the design have been suggested. It has been suggested that
the feature itself is unnecessary, and indeed much Rust code already exists
without it, however the pattern solves some cases which are difficult to handle
otherwise and allows more flexibility in code layout.

# Unresolved questions
[unresolved]: #unresolved-questions

### Extension to for, while, while let

A frequently discussed issue is extension of this concept to allow `for`,
`while` and `while let` expressions to return values in a similar way. There is
however a complication: these expressions may also terminate "naturally" (not
via break), and no consensus has been reached on how the result value should
be determined in this case, or even the result type.

There are three options:

1. Do not adjust `for`, `while` or `while let` at this time
2. Adjust these control structures to return an `Option<T>`, returning `None`
in the default case
3. Specify the default return value via some extra syntax

#### Via `Option<T>`

Unfortunately, option (2) is not possible to implement cleanly without breaking
a lot of existing code: many functions use one of these control structures in
tail position, where the current "value" of the expression, `()`, is implicitly
used:

```rust
// function returns `()`
fn print_my_values(v: &Vec<i32>) {
for x in v {
println!("Value: {}", x);
}
// loop exits with `()` which is implicitly "returned" from the function
}
```

Two variations of option (2) are possible:

* Only adjust the control structures where they contain a `break EXPR;` or
`break 'label EXPR;` statement. This may work but would necessitate that
`break;` and `break ();` mean different things.
* As a special case, make `break ();` return `()` instead of `Some(())`,
while for other values `break x;` returns `Some(x)`.

#### Via extra syntax for the default value

Several syntaxes have been proposed for how a control structure's default value
is set. For example:

```rust
fn first<T: Copy>(list: Iterator<T>) -> Option<T> {
for x in list {
break Some(x);
} else default {
None
}
}
```

or:

```rust
let x = for thing in things default "nope" {
if thing.valid() { break "found it!"; }
}
```

There are two things to bear in mind when considering new syntax:

* It is undesirable to add a new keyword to the list of Rust's keywords
* It is strongly desirable that unbounded lookahead is required while syntax
parsing Rust code

For more discussion on this topic, see [issue #961](https://github.com/rust-lang/rfcs/issues/961).