Always omit zero disambiguators on crate-roots #69
Conversation
michaelwoerister
force-pushed
the
dont-display-zero-disambiguators
branch
from
db7e4c7
to
5449fcf
Compare
| #[test] | ||
| fn demangle_crate_with_zero_disambiguator() { | ||
| t!("_RC4f128", "f128"); | ||
| t_nohash!("_RC4f128", "f128"); |
There was a problem hiding this comment.
Do these tests fail without the change? When I was trying to work on this it seemed like these simple tests wouldn't emit [0] even as-is, and I was struggling to write a test that replicates the behavior from rust-lang/rust#123816 (comment). Hence my messy branch at https://github.com/tgross35/rustc-demangle/blob/b29eead5e3337030843b978736748a3f1c142416/src/v0.rs#L1284-L1295
There was a problem hiding this comment.
The t! line should fail, the other one should already pass.
There was a problem hiding this comment.
Verified again that it fails without the change:
thread 'v0::tests::demangle_crate_with_zero_disambiguator' panicked at src/v0.rs:1286:9:
assertion `left == right` failed
left: "f128[0]"
right: "f128"
5449fcf
to
458f9b1
Compare
|
Looks like there's some misconfiguring, the required "test" isn't one of the the CI checks in this repo. EDIT: I can't even manually merge, and don't have access to settings I don't think. |
|
It seems to have worked for me. This is the setup, so everyone on the compiler and contributors should be able to merge PRs: https://github.com/rust-lang/team/blob/master/repos/rust-lang/rustc-demangle.toml Thanks for the review, @eddyb! |
|
The merging bug was an issue with branch protection by the way https://rust-lang.zulipchat.com/#narrow/stream/242791-t-infra/topic/.22test.22.20job.20not.20completing We will need a new release for this to use it in rust-lang/rust#123816, @Mark-Simulacrum it looks like maybe you handle this? |
This PR makes rustc-demangle always omit zero-valued disambiguators of
crate-rootpaths. There is an accompanying PR to the symbol mangling documentation, which more explicitly recommends this behavior and gives some rationale for it. However, the behavior is already in line with existing documentation. E.g. the original RFC states that zero-disambiguators can always be omitted, and the current documentation also omits the disambiguators forcrate-root.Being able to rely on
C3footo be rendered asfoo(i.e. without explicit disambiguator value) rather than asfoo[0]allows the compiler to encode things like new basic types in a backward compatible way. This idea has been originally proposed by @eddyb in rust-lang/rust#122106. It is a generally useful mechanism for supporting a certain class of new elements in the v0 mangling scheme in a backward compatible way (whether as a temporary workaround until downstream tooling has picked up grammar changes or as a permanent encoding).cc @tgross35