Add _value API for number literals in proc-macro

[GuillaumeGomez]

Part of #136652.

This PR adds the *_value for numbers (integers and floats). However, f16 and f128 are voluntarily left out as they're still unstable. Adding support for them is just a matter of uncommenting two lines, so should be fine.

Setting same reviewer as last time.

r? @Urgau

[GuillaumeGomez]

My favourite kind of failure: works locally but not in CI. T_T

Updated the value in the tests to see which one exactly is failing.

[GuillaumeGomez]

Bug comes from an issue with GCC proc-macro ABI. Ignoring this test on GCC backend for now.

[GuillaumeGomez]

CI fixed. :)

[Urgau]

A libs (or maybe even libs-api) reviewer seems more appropriate to review this PR.

r? Amanieu (since you reviewed #136355)

[Urgau]

I missed it already has an accepted ACP. The PR can be reviewed by libs reviewer.

r? libs

[Mark-Simulacrum]

Is this exactly in sync with what the rustc parser permits?

I guess that would be a mix of https://doc.rust-lang.org/nightly/reference/tokens.html#grammar-INTEGER_LITERAL and whatever we then tokenize from it?

Perhaps this should go over the bridge and ask rustc to parse the string for us?

[GuillaumeGomez]

There is no code in rustc currently which would allow this conversion without including a lot of other things. My code is a simplified version of integer_lit (from rustc_ast/src/util/literal.rs) which parses both the number, its base and its suffix (which we already have) and returns a LitKind, which is also unnecessary here since we already know the end type.

[Mark-Simulacrum]

I think my question is not about specific code in rustc but rather whether the implementation here parses exactly the same literals that rustc would from the given token(s). Is that the case? Essentially, can we document these functions as being bug-for-bug compatible with what rustc would do, or do we need a separate grammar for this?

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[GuillaumeGomez]

Added the method for f16, renamed the function only_digits into strip_underscores and improve the code of parse_number a bit to remove allocations.

[GuillaumeGomez]

Forgot to change the status.

@rustbot ready

cc @Mark-Simulacrum