Re: [PATCH 2/7] Skip ELF-specific tests when targeting pe-aarch64

[Jan Beulich <jbeulich@suse.com>]

On 06.01.2023 02:25, Mark Harmstone wrote:
> --- a/binutils/testsuite/binutils-all/objcopy.exp
> +++ b/binutils/testsuite/binutils-all/objcopy.exp
> @@ -1411,6 +1411,7 @@ proc objcopy_test_without_global_symbol { } {
>  # The AArch64 and ARM targets preserve mapping symbols
>  # in object files, so they will fail this test.
>  setup_xfail aarch64*-*-* arm*-*-*
> +clear_xfail aarch64*-*-pe* aarch64*-*-mingw*

This change clearly doesn't fit with the title, and hence would better be
a separate change.

> --- a/gas/testsuite/gas/aarch64/adr_1.d
> +++ b/gas/testsuite/gas/aarch64/adr_1.d
> @@ -1,5 +1,6 @@
>  #as: -mabi=lp64
>  #objdump: -dr
> +#notarget: *-*-pe* *-*-mingw*

While it's up to the arch maintainers to judge, to me there's nothing ELF-
specific in this and similar tests; all you need to do is adjust the
expectations to also accept the COFF form of the respective relocations.
That's what we do on x86, for example.

> --- a/gas/testsuite/gas/aarch64/advsimd-mov-bad.d
> +++ b/gas/testsuite/gas/aarch64/advsimd-mov-bad.d
> @@ -1,5 +1,6 @@
>  #source: advsimd-mov-bad.s
>  #readelf: -s --wide
> +#notarget: *-*-pe* *-*-mingw*

Tests using readelf, otoh, are appropriate to exclude. I wonder though
whether that wouldn't better be done in a more generic manner.

Jan