Cannot check `rustc` from master branch on Windows 10: build failure in rustc_codegen_gcc

[AngelicosPhosphoros]

Summary

python x.py check fails since commit 27f8efb .

Previous commit 64df9e3 works fine.

Command used

python x.py check

Expected behaviour

check should have pass

Actual behaviour

python x.py check fails with message

Checking stage0 gcc (x86_64-pc-windows-msvc)
    Checking rustc_codegen_gcc v0.1.0 (C:\rust-lang\compiler\rustc_codegen_gcc)
error[E0464]: multiple candidates for `rmeta` dependency `object` found
  --> compiler\rustc_codegen_gcc\src\lib.rs:28:1
   |
28 | extern crate object;
   | ^^^^^^^^^^^^^^^^^^^^
   |
   = note: candidate #1: C:\rust-lang\build\x86_64-pc-windows-msvc\stage0-sysroot\lib\rustlib\x86_64-pc-windows-msvc\lib\libobject-4cf2b494bb70904a.rmeta
   = note: candidate #2: C:\rust-lang\build\x86_64-pc-windows-msvc\stage0-sysroot\lib\rustlib\x86_64-pc-windows-msvc\lib\libobject-fdfb702ed2e7b9f3.rmeta

Bootstrap configuration (bootstrap.toml)

# See bootstrap.example.toml for documentation of available options
#
profile = "library"  # Includes one of the default files in src/bootstrap/defaults
change-id = 119899

Operating system

Windows 10 22H2

HEAD

27f8efb

Additional context

I actually failed to build current master c683403 and then run git bisect python x.py check to find first commit that fails to build.

I also tried clear install of Win10 on virtual machine with only python, Visual Studio Build tools and it failed with same error.

[AngelicosPhosphoros]

It seems to me that the issue is that there are 2 versions of object crate linked

[AngelicosPhosphoros]

Btw, can we also address the issue that compiler failed to build on one of the Tier 1 platforms for 2 weeks before anybody noticed?

[ChrisDenton]

This is likely a duplicate of #142505

[Kobzol]

Btw, can we also address the issue that compiler failed to build on one of the Tier 1 platforms for 2 weeks before anybody noticed?

FWIW, It seems like the gcc backend doesn't build, not the compiler. Actually, we shouldn't even be checking it on anything else than x64 Linux..

[AngelicosPhosphoros]

FWIW, It seems like the gcc backend doesn't build, not the compiler.

Yes but why nothing checks that steps from beginner contributor guide works?
E.g. clone repo, run python x.py setup, run python x.py check.

[heiher]

It seems that not all object dependencies in the rustc_codegen_gcc dependency chain have been upgraded to version 0.37. and the CI has not enabled the GCC backend for the Windows target?

[Kobzol]

We definitely don't test/build GCC on Windows, or at least we shouldn't, it should be forcefully disabled on Windows even for local builds, IMO.

[mati865]

Maybe it should see some testing? Probably x86_64-pc-windows-gnu would be most fitting for that.

[Kobzol]

I'm pretty sure it doesn't work there at all, so no need to test it.. :D

[RalfJung]

This is likely a duplicate of #142505

Why do you think that? It doesn't look like that at all to me.

This is just ./x check (without further arguments) including the GCC backend by default.

[Kobzol]

Yes, we might want to disable that. #143048 working on that here.

[RalfJung]

I'd say we want to keep that default on Linux. But not on Windows.