[oe] [OE-core] [RFT] GCC 8.1

[Andre McCurdy]

On Thu, May 10, 2018 at 6:06 PM, Khem Raj <raj.khem at gmail.com> wrote:
> On Thu, May 10, 2018 at 6:00 PM, Andre McCurdy <armccurdy at gmail.com> wrote:
>> On Thu, May 10, 2018 at 5:55 PM, Khem Raj <raj.khem at gmail.com> wrote:
>>> On Thu, May 10, 2018 at 4:11 PM, Andre McCurdy <armccurdy at gmail.com> wrote:
>>>> On Thu, May 10, 2018 at 3:50 PM, Martin Jansa <martin.jansa at gmail.com> wrote:
>>>>> On Thu, May 10, 2018 at 03:40:53PM -0700, Andre McCurdy wrote:
>>>>>> On Thu, May 10, 2018 at 3:38 PM, Martin Jansa <martin.jansa at gmail.com> wrote:
>>>>>> > see
>>>>>> > http://lists.openembedded.org/pipermail/openembedded-core/2018-May/150654.html
>>>>>>
>>>>>> Removing -fno-omit-frame-pointer isn't the same as adding
>>>>>> -fomit-frame-pointer. Frame pointers may get enabled depending on the
>>>>>> optimisation level etc (ie not only by -fno-omit-frame-pointer).
>>>>>
>>>>> Should I send v2 adding -fomit-frame-pointer instead of removing
>>>>> -fno-omit-frame-pointer?
>>>>>
>>>>> The v1 fixes the issue for me with default config + DEBUG_BUILD.
>>>>
>>>> The v1 patch isn't wrong, it's just incomplete (the problem could come
>>>> back if someone changes optimisation level or switches gcc to clang,
>>>> etc).
>>>>
>>>> My choice would be a v2 patch which adds -fomit-frame-pointer to
>>>> CFLAGS unconditionally for all ARM builds when Thumb is enabled. That
>>>> should fix the problem for all optimisation levels etc and avoids
>>>> building the main strace binary differently depending on whether or
>>>> not ptest is enabled.
>>>
>>> explicitly adding this option is a poor choice especially for debug
>>> builds where we should
>>> let the -On level decide and not explicitly ask for either
>>> enable/disable frame-pointers
>>> that will also make it compiler proof.
>>
>> Of course, we should let the compiler decide whenever it's possible to do so.
>>
>> Unfortunately there are cases like this one where frame pointers clash
>> with inline assembler and we need to over-rule the compiler's choice.
>
> Here we are adding -fno-omit-frame-pointer via global opt flags that
> is the issue
> where we have fallouts from default O options I agree we should teach
> this to build
> system and help the compiler

Since there's NO situation where enabling frame pointers is going to
work for this code + ARM + Thumb, I don't see the advantage of leaving
anything up to chance. Just explicitly disabling frame pointers is the
safest and cleanest option.