Re: [RFC] Use .opt for the SPARC port

[Richard Sandiford <rsandifo at redhat dot com>]

Thanks for doing this!

Eric Botcazou <ebotcazou@libertysurf.fr> writes:
> This patch switches the SPARC port over to the new option
> specification style.  Nothing really worth mentioning, except that
> this port defines TARGET_* macros without defining the corresponding
> options.  Unless I'm mistaken, that's a novelty so I had to define a
> new option property, namely MaskPlaceholder.
>
> Lightly tested with crosses from Linux to the 12 targets for the time being.
>
> Richard, what do you think about the new mechanism?  TIA.
> [...]
> +@item MaskPlaceholder(@var{name})
> +This property is special: it disables the regular processing of the
> +option specification so that no option would be ultimately registered;
> +instead a mask would be allocated and a macro would be declared as if 
> +@item Mask(@var{name}) was specified.
> +
> [...]
> +mv8
> +Target MaskPlaceholder(V8)
> +Generate code for SPARC-V8

So no configuration actually accepts an option called -mv8?  You just
define it so that you can add the mask?

I wonder if would be better to have a separate RegisterMask record,
a bit like the existing Language records.  E.g.:

    RegisterMask
    V8 SPARC SPARCLITE V9 DEPRECATED_V8_INSNS

It would avoid confusion about having .opt files that don't correspond
to real options.  Also (having recently been burned by gcc.pot ;)
it would stop us adding strings to gcc.pot that don't actually need
translation.  (Although you could of course just leave the dummy
options as undocumented instead.)

FWIW, you could also add LONG_DOUBLE_128 to the list above and add
MaskExists to the long-double-switch.opt entry.  You then wouldn't
need a separate long-double-no-switch.opt file.

Richard