Add an assert_not_impl macro #17

HeroicKatora · 2019-08-08T02:31:58Z

This is the polar opposite of the assert_impl macro and abuses name
resolution with trait.

A trait with type parameter A provides a simple method. The trait is
implemented on the checked type T potentially twice: with type
parameter unit A=() for all types T, and with another parameter type
-abitrarily a new Invalid- when the T implements the trait bounds.
We then try to select T as an instance of the trait, using turbo-fish
to try and access one of its items, without providing the trait type
parameter. When only the first instance is implemented the type checker
can infer A=() but when the second is implemented as well it can not.

No other ambiguity can occur, only the trait's type parameter is
ambiguous. Thus, type inference and compilation succeeds if the trait
bound is not satisified by the input type.

iliekturtles

This is exciting!

src/assert_not_impl.rs

HeroicKatora · 2019-08-08T18:44:40Z

I needed to modify the design slightly, as the method call was possibly not completely unambiguous. Now the ambiguity has moved to the type parameter of a single locally defined trait which allows us to restrict the set of possibly applicable methods to that trait alone using turbo-fish syntax. By leaving out the trait's type parameter it can only be inferred when one impl is applicable.

This also makes it somewhat simpler as only one type and trait is involved.

src/assert_not_impl.rs

nvzqz

Thank you so much for this!! I was previously convinced that this wasn't possible. I love cursed hacks like these. ❤

Regarding style, please remove whitespace between empty braces.

I'll be more than happy to merge this PR once you've made my requested changes. Also, the changes in this PR should be detailed in CHANGELOG.md under "Unreleased" in a new "Added" section.

nvzqz · 2019-08-09T14:25:32Z

src/assert_not_impl.rs

+/// [`Sync`]: https://doc.rust-lang.org/std/marker/trait.Sync.html
+/// [blanket]: https://doc.rust-lang.org/book/second-edition/ch10-02-traits.html#using-trait-bounds-to-conditionally-implement-methods
+#[macro_export(local_inner_macros)]
+macro_rules! assert_not_impl {


Please rename this to assert_not_impl_all! and add a assert_not_impl_any! counterpart. For assert_not_impl_any!, see my suggestion below.

src/assert_not_impl.rs

nvzqz · 2019-08-09T14:36:31Z

src/assert_not_impl.rs

+            trait AmbiguousIfImpl<A> { fn some_item() { } }
+
+            impl<T: ?Sized> AmbiguousIfImpl<()> for Check<T> { }
+            impl<T: ?Sized $(+ $t)*> AmbiguousIfImpl<u8> for Check<T> { }


Rather than use u8 here, add struct Invalid; under Check (or a better name if you think of one). Since it never gets constructed, make sure to add #[allow(dead_code)] to the declaration.

For assert_not_impl_any!, you should implement this bit like the following. Notice that each Invalid type is unique because of the scoping.

$({ #[allow(dead_code)] struct Invalid; impl<A: ?Sized + $t> AmbiguousIfImpl<Invalid> for Check<A> {} })+

Also, you should be matching multiple $t via + and not *.

src/assert_not_impl.rs

nvzqz · 2019-08-09T14:46:50Z

src/assert_not_impl.rs

+            impl<T: ?Sized> AmbiguousIfImpl<()> for Check<T> { }
+            impl<T: ?Sized $(+ $t)*> AmbiguousIfImpl<u8> for Check<T> { }
+
+            <Check::<$x> as AmbiguousIfImpl<_>>::some_item()


Please do let _ = <Check::<$x> as AmbiguousIfImpl<_>>::some_item;.

nvzqz · 2019-08-09T14:55:06Z

src/assert_not_impl.rs

+/// ```
+///
+/// Note that all traits given in a single invocation need to not be implemented. If you want to
+/// check that none of multiple trait are implemented you must invoke the macro several times.


Change this to refer to assert_not_impl_any! instead with a working local docs link. Also move this comment to right before "Examples".

nvzqz · 2019-08-09T14:57:09Z

src/assert_not_impl.rs

+/// # #[macro_use] extern crate static_assertions;
+/// # use static_assertions::_core::cell::Cell;
+/// # fn main() {}
+/// // Cell<u32> is not both Sync and Send


Make this normal docs text.

nvzqz · 2019-08-09T14:57:45Z

src/assert_not_impl.rs

+/// # #[macro_use] extern crate static_assertions;
+/// # use static_assertions::_core::cell::Cell;
+/// # fn main() {}
+/// // But it is Send, this fails to compile


Make this normal docs text. For visual purposes, I don't like having code blocks separated without anything in between to pad them.

nox · 2019-08-11T11:29:15Z

Out of curiosity, why do Check<T> and the type parameter on the trait exist at all?

To check that a type S doesn't implement a trait T, one can do:

trait ShouldNotImplT {}
impl ShouldNotImplT for U where U: T {}
impl ShouldNotImplT for S {}

HeroicKatora · 2019-08-11T14:04:34Z

@nox No particular reason, it turns out, only that a previous version of the implementation had a self parameter instead of the turbofish for trait resolution. Thanks for the simplification suggestion.

This is the polar opposite of the assert_impl macro and abuses name resolution with trait. A trait with type parameter `A` provides a simple method. The trait is implemented on the checked type `T` potentially twice: with type parameter unit `A=()` for all types `T`, and with another parameter type -abitrarily a new `Invalid`- when the `T` implements the trait bounds. We then try to select `T` as an instance of the trait, using turbo-fish to try and access one of its items, without providing the trait type parameter. When only the first instance is implemented the type checker can infer `A=()` but when the second is implemented as well it can not. No other ambiguity can occur, only the trait's type parameter is ambiguous. Thus, type inference and compilation succeeds if the trait bound is not satisified by the input type.

Makes use of the same trick as assert_not_impl_all but adding multiple possible implementations of the trait instead of at most two.

nvzqz · 2019-08-11T20:16:41Z

Awesome, thank you very much!

WorldSEnder · 2024-01-29T17:59:20Z

Coming back to this a few years later, the self parameter can be handy if you have a value of a type you can't name, e.g. as happening with futures.

async fn some_lib_feature() {
    // Do some async work. To future proof this, and not cause unintentional semver breaks,
    // we might want to assert this is !Send
}
// We can use a very similar trait to the one in the PR
trait AmbiguousIfImpl<A> { fn some_item(&self) {} }
impl<T: ?Sized> AmbiguousIfImpl<()> for T {}
struct Invalid;
impl<T: ?Sized + Send> AmbiguousIfImpl<Invalid> for T {}
// And use it to test the future is not Send
let fut = some_lib_feature(); // We can construct a value of the type, but can't name it
let () = <_ as AmbiguousIfImpl<_>>::some_item(&fut);
//        ^ type is deduced to the type of the future, although we can not name it. *

* This comment might become a bit outdated with the stabilization of type_alias_impl_trait in rust-lang/rust#63063.

iliekturtles reviewed Aug 8, 2019

View reviewed changes

src/assert_not_impl.rs Show resolved Hide resolved

HeroicKatora commented Aug 8, 2019

View reviewed changes

src/assert_not_impl.rs Outdated Show resolved Hide resolved

HeroicKatora mentioned this pull request Aug 9, 2019

Assert negative trait bounds #8

Closed

nvzqz requested changes Aug 9, 2019

View reviewed changes

HeroicKatora added 2 commits August 11, 2019 16:44

Add assert_not_impl_any

261e73a

Makes use of the same trick as assert_not_impl_all but adding multiple possible implementations of the trait instead of at most two.

nvzqz approved these changes Aug 11, 2019

View reviewed changes

nvzqz merged commit d9f73f8 into nvzqz:master Aug 11, 2019

taiki-e mentioned this pull request Sep 25, 2019

Provide a way to guarantee the type is !Unpin taiki-e/pin-project#108

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add an assert_not_impl macro #17

Add an assert_not_impl macro #17

HeroicKatora commented Aug 8, 2019 •

edited

Loading

iliekturtles left a comment

HeroicKatora commented Aug 8, 2019 •

edited

Loading

nvzqz left a comment

nvzqz Aug 9, 2019 •

edited

Loading

nvzqz Aug 9, 2019

nvzqz Aug 9, 2019

nvzqz Aug 9, 2019

nvzqz Aug 9, 2019

nvzqz Aug 9, 2019

nox commented Aug 11, 2019

HeroicKatora commented Aug 11, 2019

nvzqz commented Aug 11, 2019

WorldSEnder commented Jan 29, 2024 •

edited

Loading

Add an assert_not_impl macro #17

Add an assert_not_impl macro #17

Conversation

HeroicKatora commented Aug 8, 2019 • edited Loading

iliekturtles left a comment

Choose a reason for hiding this comment

HeroicKatora commented Aug 8, 2019 • edited Loading

nvzqz left a comment

Choose a reason for hiding this comment

nvzqz Aug 9, 2019 • edited Loading

Choose a reason for hiding this comment

nvzqz Aug 9, 2019

Choose a reason for hiding this comment

nvzqz Aug 9, 2019

Choose a reason for hiding this comment

nvzqz Aug 9, 2019

Choose a reason for hiding this comment

nvzqz Aug 9, 2019

Choose a reason for hiding this comment

nvzqz Aug 9, 2019

Choose a reason for hiding this comment

nox commented Aug 11, 2019

HeroicKatora commented Aug 11, 2019

nvzqz commented Aug 11, 2019

WorldSEnder commented Jan 29, 2024 • edited Loading

HeroicKatora commented Aug 8, 2019 •

edited

Loading

HeroicKatora commented Aug 8, 2019 •

edited

Loading

nvzqz Aug 9, 2019 •

edited

Loading

WorldSEnder commented Jan 29, 2024 •

edited

Loading