On 20 June 2017 at 17:44, Laszlo Ersek <[email protected]> wrote:
> On 06/20/17 14:00, Leif Lindholm wrote:
>> On Tue, Jun 20, 2017 at 01:28:38PM +0200, Ard Biesheuvel wrote:
>>> (+ Laszlo)
>>>
>>> On 20 June 2017 at 13:00, Leif Lindholm <[email protected]> wrote:
>>>> When building without LTO, gcc incorrectly resolves the hazards for
>>>> 'PciRegBase’ when inlining, leading to "may be used uninitialized"
>>>> warnings (and hence build failure with -Werror).
>>>> Eliminate this warning by explicitly initializing the variable to 0.
>>>>
>>>> Contributed-under: TianoCore Contribution Agreement 1.0
>>>> Signed-off-by: Leif Lindholm <[email protected]>
>>>
>>> Reviewed-by: Ard Biesheuvel <[email protected]>
>>
>> Thanks!
>>
>>>> ---
>>>>
>>>> For those who have heard me mentioning this before but arguing against
>>>> upstreaming this patch: I only just tweaked that this warning doesn't
>>>> go away with more recent toolchains, but simply when switching to GCC5
>>>> build profile, and hence LTO. Build failure still reproducible with
>>>> gcc 6.3.1 and GCC49.
>>>
>>> /me annoyed
>>
>> +1
>>
>>> We keep hitting this with GCC, and I profoundly dislike having to add
>>> redundant initialization sequences. Is there any other solution
>>> possible, e.g., disable this warning for certain builds?
>>
>> I would really like to not disable it.
>> When it gets it right, that's usually a pretty horrific bug caught.
>> And if we switch to LTO as the norm, I guess these will become less
>> noticeable over time.
>> But it does bug me too.
>
> Perhaps we should introduce two macros to "MdePkg/Include/Base.h",
>
> #define UNUSED_POINTER NULL
> #define UNUSED_INTEGER 0
>
> These could be used to suppress such warnings, without the risk of
> misleading programmers (as to the real necessity of the variable
> assignment at hand).
>

Or perhaps

#ifdef __GNUC__
#define USED(var)  var = (typeof(var))0
#else
#define USED(var)
#endif
_______________________________________________
edk2-devel mailing list
[email protected]
https://lists.01.org/mailman/listinfo/edk2-devel

Reply via email to