Re: [PATCH 1/3] MdePkg/Misc: Move ARM* BaseMemoryLibStm to MdePkg

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

On 2 September 2016 at 19:05, Leif Lindholm <leif.lindholm@linaro.org> wrote:
> On Fri, Sep 02, 2016 at 05:07:48PM +0100, Ard Biesheuvel wrote:
>> > This isn't a proposed alternative to your patchset, this is an
>> > alternative to reverting all of the other BaseMemoryLib changes.
>> > I'm happy to throw it away next week if we can get something better in
>> > by then.
>> >
>> >> The AArch64 optdxe changes i proposed are arguably as harmless,
>> >
>> > Yes, but that leaves ARM broken.
>>
>> Not necessarily. ARM can use the generic BaseMemoryLib in MdePkg
>
> OK, but that's still yet another change.
>
>> >> since they don't affect any other arch either. And the stm aarch64
>> >> versions are plain c to begin with
>> >
>> > And I want to discuss that next week. Against the backdrop of a
>> > working master branch.
>>
>> My concern is that having an 'ARM' flavor of BaseMemoryLib in MdePkg/
>> will make it more difficult rather than easier to simply add ARM
>> support to the existing non-sse/mmx ones
>
> I don't share that concern. I am entirely serious about nuking it next
> week if we come to an agreement on a better solution.
> And I'm happy to take on the drudgery of fixing up all the platforms.
>
> (As a side note, this is an excellent example of why we need to
> create reusable config fragments - fixing all ARM platforms to use the
> correct library should be at most a two-liner in one place in edk2.)
>
>> Added the fact that the Stm is not in great shape, I would really
>> prefer to get rid of it rather than 'promote' it to the standard ARM
>> implementation. Note that we will need another round of updates to the
>> platform .DSCs when we remove the Stm version again.
>
> My main concern is leaving the master branch unusable for ARM during
> the Labor Day weekend. If we don't resolve this tonight, with the best
> will in the world we won't be back to a functioning master before
> Tuesday evening.
>

I agree that fixing the branch now would be nice. I just don't
understand why fixing BaseMemoryLibStm in place is not a better
solution, especially if we are nuking it anyway next week. That way,
we have to change all the platforms only a single time.