Re: [PATCH 1/1] OvmfPkg/IoMmuDxe: shut up "unused-const-variable" gcc-6 warning in RELEASE

[Ard Biesheuvel <ard.biesheuvel@linaro.org>]

On 6 September 2017 at 18:09, Laszlo Ersek <lersek@redhat.com> wrote:
> On 09/06/17 18:56, Ard Biesheuvel wrote:
>> On 6 September 2017 at 17:48, Laszlo Ersek <lersek@redhat.com> wrote:
>>> Starting with gcc-6, a new warning option called "-Wunused-const-variable"
>>> has become available (and enabled by default under our build settings):
>>>
>>>   https://gcc.gnu.org/onlinedocs/gcc-6.4.0/gcc/Warning-Options.html
>>>
>>> We should give it the same treatment as Ard gave "unused-but-set-variable"
>>> in commit 20d00edf21d2 ("BaseTools/GCC: set -Wno-unused-but-set-variables
>>> only on RELEASE builds", 2016-03-24); i.e., we should restrict the warning
>>> to the DEBUG build target.
>>>
>>> However, because the new warning is gcc-6+ only, we cannot add
>>> "-Wno-unused-const-variable" to any GCC5 macros in
>>> "BaseTools/Conf/tools_def.template". While the GCC6 toolchain and/or the
>>> desired handling of the new warning are investigated in
>>> <https://bugzilla.tianocore.org/show_bug.cgi?id=700>, suppress the warning
>>> for gcc-6+ (in RELEASE builds) as follows:
>>>
>>> - Replace the "mBusMasterOperationName" array with the
>>>   BUS_MASTER_OPERATION_NAME(Expression, ShortName) macro that compares
>>>   Expression against EdkiiIoMmuOperationBusMaster<ShortName>, and in case
>>>   of a match, evaluates to "ShortName", stringified,
>>>
>>> - when composing the DEBUG message -- which the preprocessor might
>>>   eliminate from RELEASE builds, thereby removing its references to
>>>   variables --, build a ladder of comparisons with
>>>   BUS_MASTER_OPERATION_NAME().
>>>
>>
>> I wonder if we should put this in a header file somewhere. I really
>> hate that we need to create new GCCx versions just for small
>> deviations between versions of the compiler.
>>
>> I.e., add this to MdePkg/Include/Base.h
>>
>> #ifdef MDEPKG_NDEBUG
>>   //
>>   // Don't warn about unused but set variables: they may only be
>>   // referenced by DEBUG code.
>>   //
>>   #pragma GCC diagnostic ignored "-Wunused-but-set-variable"
>>   #if __GNUC__ >= 6
>>     #pragma GCC diagnostic ignored "-Wunused-const-variable"
>>   #endif
>> #endif
>
> The option in question is "-Wno-unused-const-variable". But, that's a
> side question.
>

No. The option is "-Wunused-const-variable", and adding the 'no-'
prefix is the same as ignoring it.

> If the above is acceptable to the MdePkg maintainers, it is certainly
> good enough for me. In that case, I will reassign TianoCore BZ#700 to
> MdePkg. (I was careful to write "desired handling of the new warning" in
> the commit message, because I expected that the introduction of the GCC6
> toolchain just for this might not be universally liked.)
>

No. I think we should have fewer rather than more. I think the current
collection in tools_def is ridiculous, but that is my personal
opinion.

> Personally I wouldn't like to submit the MdePkg patch myself, as GCC
> pragmas are another delicacy that doesn't fit on my plate right now. (If
> I happen to break something with it, I get to fix it too... Like I'm
> fixing it now within OvmfPkg.)
>
> The point of this patch is to get OVMF back to a building state ASAP.
> Once we fix BZ#700, I'm more than happy to revert the patch. I sought to
> write the patch in a way that would be tolerable both short-term and
> mid/long-term.
>

Fair enough.