`try bikeshed $ty { … }` is not pre-expansion gated (affects beta+nightly)

[fmease]

The following code compiles even though it shouldn't:

fn main() {
    #[cfg(false)]
    try bikeshed () {}
}  

Contrary to try { … } it does not even issue a pre-expansion gate warning. For context, try blocks and some other ancient features are notoriously post-expansion gated which we can't change without breaking a large amount of crates in the ecosystem (which use the #[cfg(feature = "nightly")] pattern). For new syntactic forms however we require a pre-expansion feature gate nowadays for forward compatibility reasons. try bikeshed $ty { … } is a new syntactic form and thus should be pre-expansion gated by try_blocks try_block_heterogeneous.

[scottmcm]

Oops, absolutely agreed. My bad for copying try.

Can you point me at instructions for the right way to do this? (Or if someone who knows what they're doing wants to just fix it that would be good too.)

[fmease]

No worries!

Can you point me at instructions for the right way to do this? (Or if someone who knows what they're doing wants to just fix it that would be good too.)

Edit: See #152501 (comment) instead!

Should be easy enough. It's not gonna be your typical pre-expansion feature gating because we want to gate two things behind the same feature (try_blocks) but differently (one pre-expansion, the other post-expansion (plus pre-expansion warning)). So:

1. You add a new symbol try_bikeshed_blocks
2. In the relevant part of the parser you add self.psess.gated_spans.gate(sym::try_bikeshed_blocks, /* span */);
3. In compiler/rustc_ast_passes/src/feature_gate.rs you do basically the same thing that we do for the pairing (associated_const_equality, min_generic_const_args)
   • I.e., emit a try_blocks feature gate if we have any spans marked try_bikeshed_blocks

[fmease]

Feel free to r? me

[fmease]

Resolved

Ah, now I see that it's to be gated separately anyway, namely behind try_blocks_heterogeneous. That makes it even simpler.

Oops, absolutely agreed. My bad for copying try.

No worries!

Can you point me at instructions for the right way to do this? (Or if someone who knows what they're doing wants to just fix it that would be good too.)

In compiler/rustc_ast_passes/src/feature_gate.rs you add a gate_all!(try_block_heterogeneous, "… … …"); where all the other gate_all!(…)s are. Then, you remove the branch you've added to the PostExpansionVisitor in PR #149489.