fix(install): default to GNU host in Cygwin/MSYS/MinGW environments (#4421)

[cachebag]

Addresses first mentioned issue in #4421, modifying the install script to default to x86_64-pc-windows-gnu under Cygwin/MSYS when no --default-host is provided.

The second issue requires changes in the compiled rustup-init binary (src/), which is out of scope for now

[rami3l]

Thank you very much :)

However, 4221 is not an issue. Maybe you've got the number wrong?

[rami3l]

Should we consider the possibility of aarch64-pc-windows-gnullvm as well? It's a tier 2 target for now.

cc @ChrisDenton

[ChrisDenton]

Hm... for arm we should probably not touch the default for now, at least until support for gnu on the target is more mainstream. The gnullvm target was only fairly recently promoted to having host tooling and iirc requires installing llvm tools (instead of gnu ones). We could revisit this in the future.

[cachebag]

However, 4221 is not an issue. Maybe you've got the number wrong?

@rami3l
Thanks for letting me know! Lmk what else may be needed

[rami3l]

How is this different from the existing functionality below?

rustup/rustup-init.sh

Lines 425 to 427 in 7365ab7

	MINGW* | MSYS* | CYGWIN* | Windows_NT)
	_ostype=pc-windows-gnu
	;;

[cachebag]

@rami3l

How is this different from the existing functionality below?

I presume that it's essentially the safety net to this, no? The idea here is that the issue at hand was detecting x86_64-pc-windows-msvc instead of x86_64-pc-windows-gnu in Cygwin despite the existing detection. I figured that a blanket override ensures that shell script users on Windows get GNU regardless of detection edge cases.

Do we address/refactor the existing detection instead?

[rami3l]

How is this different from the existing functionality below?

I presume that it's essentially the safety net to this, no? The idea here is that the issue at hand was detecting x86_64-pc-windows-msvc instead of x86_64-pc-windows-gnu in Cygwin despite the existing detection. I figured that a blanket override ensures that shell script users on Windows get GNU regardless of detection edge cases.

Do we address/refactor the existing detection instead?

Yes, if this PR is made to address #4421 (comment). I don't have a Cygwin test environment, but the detection seems right already to me, so this PR seems unnecessary in its current state.

cc @ChrisDenton for further verification.