download-rustc fails to detect when profiler settings differ from local settings

[RalfJung]

See #133499 (comment) for context: the x86_64-gnu-aux runner builds rustc without profiling support. The CI build that is downloaded via download-rustc does have profiling support. This means if a test relies on profiling support, it will incorrectly pass on x86_64-gnu-aux if download-rustc is used, only to fail later on an unrelated PR when download-rustc cannot be used.

bootstrap tries to ensure that all the settings are the same between the local configuration and the prebuilt rustc, but that does not seem to fully work.

Cc @onur-ozkan

[RalfJung]

@onur-ozkan wrote

That PR doesn't change any contig at all? I guess we are on different points, I am talking about config.toml here.

The PR doesn't change the config indeed. x86_64-gnu-aux locally changes its config in CI (if I understood correctly). The download-rustc logic should detect that and never use the prebuilt rustc on that runner, since it was built with a different configuration than what is being locally set up.

[onur-ozkan]

Probably, this happens because we are not checking "--set" flags yet.

[onur-ozkan]

"profiler" option is not part of the "rust" section in config.toml, which is why it was missed in the config compatibility check logic (it was designed to check "rust" section only). #133705 makes it to be handled properly.