zig cc -mcpu=? is pass-thru to clang driver and lists cpu model names incompatible with zig targets

[iacore]

Zig Version

0.9.0-dev.1433+4a76523b9

Steps to Reproduce

zig cc -o ./pony -O3 -march=x86-64 pony.o

Any object file will work.

Expected Behavior

The program should compile. clang does work with the same arguments.

Actual Behavior

> zig cc -o ./pony -O3 -march=x86-64 pony.o
info: Available CPUs for architecture 'x86_64':
...
 x86_64
 x86_64_v2
 x86_64_v3
 x86_64_v4
...

error: Unknown CPU: 'x86'

[iacore]

Using zig cc "-mcpu=?" does yield the "dash versions" of CPU architectures

        x86-64
        x86-64-v2
        x86-64-v3
        x86-64-v4

This bug seems to be of command line parsing

[mikdusan]

use -march=x86_64 instead

zig cannot use "hyphenation" in the cpu name because we allow fine-grained feature flags appended after model name using + or -. eg: -march=x86_64-sse2 disables sse2

Unfortunately, the special case -mcpu=? is pass-thru to clang driver, and clang is listing the cpu models according to its own naming. Ideally, we'd want to intercept, and emit zig's cpu model list, ie. the same output as shown by zig cc -c foo.c -mcpu=bogus

[Deecellar]

use -march=x86_64 instead

zig cannot use "hyphenation" in the cpu name because we allow fine-grained feature flags appended after model name using + or -. eg: -march=x86_64-sse2 disables sse2

Unfortunately, the special case -mcpu=? is pass-thru to clang driver, and clang is listing the cpu models according to its own naming. Ideally, we'd want to intercept, and emit zig's cpu model list, ie. the same output as shown by zig cc -c foo.c -mcpu=bogus

This is a non solution for already existing codebases like meson based build. where is not valid to pass as an option x86_64

[andrewrk]

Duplicate of #4911