Re: [RFC 6/7] Lazily and dynamically create i386-linux target descriptions

[Pedro Alves <palves at redhat dot com>]

On 05/11/2017 04:55 PM, Yao Qi wrote:

> +#if GDB_SELF_TEST
> +  struct xml_and_mask
> +  {
> +    const char *xml_file_name;
> +    uint64_t mask;
> +  };

Minor nit: I'd instinctively find it more natural to read the
list below as { input, output }  (left to right), while you
have have { output, input }.

> +
> +  struct xml_and_mask array[] = {
> +    { "i386/i386-linux.xml", X86_XSTATE_SSE_MASK },
> +    { "i386/i386-mmx-linux.xml", X86_XSTATE_X87_MASK },
> +    { "i386/i386-avx-linux.xml", X86_XSTATE_AVX_MASK },
> +    { "i386/i386-mpx-linux.xml", X86_XSTATE_MPX_MASK },
> +    { "i386/i386-avx-mpx-linux.xml", X86_XSTATE_AVX_MPX_MASK },
> +    { "i386/i386-avx-avx512-linux.xml", X86_XSTATE_AVX_AVX512_MASK },
> +    { "i386/i386-avx-mpx-avx512-pku-linux.xml",
> +      X86_XSTATE_AVX_MPX_AVX512_PKU_MASK },

About these xml files.  Is your idea that:

 #1 - we remove these xml description files at some point, keeping only
      the description fragments which are currently xi:included by the xml
      files above?
 #2 - or, we'll still continue adding new xml files and grow this
      list here?
 #3 - or, we'll keep the existing xml files as representative / legacy,
      and use them in the unit tests going forward, just to make sure
      the machinery builds correct descriptions?

(I don't expect to answer to be #2, I just put it there for completeness.)

> +  };
> +
> +  for (auto &a : array)
> +    {
> +      auto tdesc = i386_linux_read_description (a.mask);
> +
> +      selftests::record_xml_tdesc (a.xml_file_name, tdesc);
> +    }
> +#endif /* GDB_SELF_TEST */
>  }
> 

Thanks,
Pedro Alves