Fix ICE from invalid delimeter in attrs parsing

[chenyukang]

Fixes #146808

I tried hard to get the correct span, but didn't come out with a great result. I think maybe it's better to report an error when expand macro failing for contract:

rust/compiler/rustc_builtin_macros/src/contracts.rs

Line 50 in f524236

annotated: TokenStream,

[rustbot]

Some changes occurred in compiler/rustc_attr_parsing

cc @jdonszelmann

[rustbot]

r? @nnethercote

rustbot has assigned @nnethercote.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[jdonszelmann]

r? jdonszelmann

[jdonszelmann]

could you modify the tests to use #![core::prelude::v1::test]? the same ICE seems to happen on any macro applied to the crate root, and test is stable unlike contracts. Seems to have nothing to do with contract attributes themselves.

[jdonszelmann]

maybe add a comment about that too in the test

[jdonszelmann]

well, actually, with test it seems to be a different ICE? I'm not convinced the root cause of the ICE is the fix you applied 🤔

[jdonszelmann]

okay, I think I just found a secondary ICE, let me look at that in isolation, though the root cause might be the same as this one

[jdonszelmann]

okay, the other ICE is unrelated and handled here: #147841

[jdonszelmann]

I still think this fixes the wrong thing. Sure, the ICE is gone, but the error points to the wrong location and I'm not even 100% sure why it happens in the first place. Something seems to go fundamentally wrong in which brackets it chooses to match with?

[chenyukang]

Something seems to go fundamentally wrong in which brackets it chooses to match with
yes.

seems the span will contain the whole file when parser gets in the the scenario of expecting #![some_attribute(....)], but got the wrong code with #![some_attribute], which is missing the arguments.

I remember for this case the span of here is not right already:

rust/compiler/rustc_builtin_macros/src/contracts.rs

Line 50 in f524236

annotated: TokenStream,

as I wrote in the description, I tried to fix the span, but haven't got a completed fix, I may continue to debug it when get time.

another similar issue is here: #146834

[chenyukang]

it's only happen for #![some_attribute], but not for #[some_attribute]

[jdonszelmann]

Closing, in favor of #147849. See my explanation here: #146808 (comment)