hard float ABI passed to arm32 android compilations, which require softfp ABI

[Beyley]

Zig Version

0.14.0-dev.1632+d83a3f174

Steps to Reproduce and Observed Behavior

Clone the 1ed2bd6b00604eceeb4cc28937eec4e962e2ea30 commit of this repo.

install the android NDK version 21

uncomment these lines
https://github.com/LittleBigRefresh/scetool/blob/1ed2bd6b00604eceeb4cc28937eec4e962e2ea30/build.zig#L72

receive this error

error: sub-compilation of libunwind failed
    /home/beyley/zig-versions/master/lib/libunwind/src/UnwindRegistersRestore.S:1:1: note: clang exited with code 1
    :1:1: note: unsupported option '-mfloat-abi=hard' for target 'armv4t-unknown-linux-android'
    :1:1: note: '-mfloat-abi=hard': selected processor lacks floating point registers

Expected Behavior

Compilation succeeds

[alexrp]

#21269 regressed this case inadvertently because std.Target.Abi.floatAbi() is not aware of Android ABIs; it's incorrectly considering these to be hard float.

However, it's a bit more complicated because Abi.android is not enough information to go on; it would be incorrect to report soft float for Abi.android for aarch64 and x86. It needs to be split into android (for aarch64, x86, and x86_64), androideabi (for arm, soft float), and androideabihf (for arm, hard float). Or, alternatively, floatAbi() needs to be changed to also take the Cpu.Arch value. I think I like the former approach more because it's in line with what we've done for other targets.

I'll fix this as part of my follow-up work on float ABI stuff.

[Beyley]

#21269 regressed this case inadvertently because std.Target.Abi.floatAbi() is not aware of Android ABIs; it's incorrectly considering these to be hard float.

However, it's a bit more complicated because Abi.android is not enough information to go on; it would be incorrect to report soft float for Abi.android for aarch64 and x86. It needs to be split into android (for aarch64, x86, and x86_64), androideabi (for arm, soft float), and androideabihf (for arm, hard float). Or, alternatively, floatAbi() needs to be changed to also take the Cpu.Arch value. I think I like the former approach more because it's in line with what we've done for other targets.

I'll fix this as part of my follow-up work on float ABI stuff.

For reference, the docs seem to say that hardfloat ABI is never used on ARM32 android