[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH v3] osdep: Make MIN/MAX evaluate arguments only once
From: |
Richard Henderson |
Subject: |
Re: [PATCH v3] osdep: Make MIN/MAX evaluate arguments only once |
Date: |
Wed, 3 Jun 2020 09:32:43 -0700 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101 Thunderbird/68.8.0 |
On 6/2/20 7:29 PM, Eric Blake wrote:
> Because:
>
> #if MIN(...)
>
> now fails to compile (you can't have { in a preprocessor expression), and:
>
> #if MIN_CONST(...)
>
> fails to compile (__builtin_constant_p() is not a preprocessor macro, so it
> warns that it is being treated as 0). The only fix is to move the MIN() out
> of
> the #if and into the #define.
Ah, right. Thanks.
>> Is it possible to use qemu_build_not_reached?
>
> Possibly.
>
> /me goes and recompiles; touching osdep.h recompiles the world...
>
> No, it blows up hard, because qemu_build_not_reached() is not embeddable in an
> expression:
Ah, right, because without -O, qemu_build_not_reached expands to
g_assert_not_reached and not to a symbol marked with QEMU_ERROR.
>> I'd prefer we generate a compile-time error than a runtime trap (or nothing,
>> depending on compiler flags controlling __builtin_unreachable).
>
> What we have DOES produce a compile-time error. If either expression to
> MIN_CONST() is not actually const, the fact that __builtin_unreachable()
> returns void causes a compilation failure because a value is expected.
Ah! Well, that's good and certainly sufficient for my needs.
I do now wonder if it wouldn't be clearer to use "(void)0"
instead of __builtin_unreachable, and add a note to the comment just above.
r~