Add #![cfg_assert]

[Amanieu]

The idea came from this thread: #![cfg_assert] allows a user to place guards on his code to fail early on incompatible configurations.

Rendered

[jonas-schievink]

Missing not(...)

[brendanzab]

I agree that it would be nice to trigger build errors, but it might also be nice to be able to trigger warnings as well (without having to create a custom lint). Not sure how this would fit in with this design.

[seanmonstar]

This looks quite similar to static_assert, which was removed: https://github.com/rust-lang/rfcs/blob/master/text/1096-remove-static-assert.md

[oli-obk]

with the static_assert crate you can write static_assert!(cfg!(target_os = "linux"))

[pnkfelix]

@Amanieu will #![cfg_assert("this will never compile in any cfg")] work? I cannot tell from the RFC whether having no cfg-inputs at all is legal.

[Amanieu]

@pnkfelix That isn't part of the RFC, however I can see it being useful in some cases.

[nikomatsakis]

We discussed this briefly in yesterday's @rust-lang/lang meeting. It seemed like, once we have even rudimentary plugins stabilized, we could make a specialized plugin that implements this functionality relatively easily. In the meantime, one could certainly achieve the same effect by having some code that fails to compile which is conditionally included:

#[cfg(...)] const error: usize = "this configuration is not supported";

Inelegant, but it works.

However, reading over the thread where this suggestion originated, we were wondering something that perhaps @Amanieu or @alexcrichton could clarify: is there some other use for this metadata beyond simply halting compilation? How frequent is the desire for this?

In other words: is there a strong reason for this to be logic that the compiler itself understands, versus something supplied externally or achieved via hackery?

[Amanieu]

I guess there isn't a very urgent need for this feature, so I'm ok with closing this RFC and waiting for compiler plugins or something similar.

[alexcrichton]

@nikomatsakis In the past I've used static assertions in C for just verifying sanity of various properties like size/alignment, but that's not covered by this definition of #[cfg_assert], so beyond that I don't personally know if there's much need beyond halting the compile with a nice user defined error message.

[nikomatsakis]

Personally, I'd be inclined to close and revisit once plugins stabilization has progressed a bit more.

[aturon]

Triaging old RFCs: based on the last few comments, I'm going to go ahead and close.