On 12/25/2014 12:23 PM, Maxim Uvarov wrote:
> On 12/25/2014 12:15 PM, Taras Kondratiuk wrote:
>> There is no need to make it so complex. At least until 'static'
>> before 'not static' functions is forbidden by some post-C11 standard).
>>
>> Me and Anders are making another try to separate clean API. In a
>> previous try common (clean) API headers were public. They had
>> <plat/...> include hooks to pull platform specific definitions. That
>> appeared to be not flexible enough. Current approach is to make common
>> API headers private in a sense that application won't include them
>> directly, but instead implementation will provide public headers and
>> embedded clean API headers there.
>>
>> Please check [1] for an example of this approach tried for a few
>> headers (atomic, byteorder). It seems to be working fine for now. I'm
>> reworking Keystone implementation based on this approach.
>>
>> [1] 
>> https://git.linaro.org/people/taras.kondratiuk/odp.git/shortlog/refs/heads/api/change_namespace
> 
> I see from your patch you are doing:
> 
> rename from platform/linux-generic/include/api/odp/buffer.h
> rename to include/api/odp/buffer.h
> 
> "api" looks like is not needed and it can be simple include/buffer.h

The exact path can be changed. It is just a demo of the approach.

> 
> btw,  what should I do with thread patch? Should I send updated version?

I think atomic code should be fixed first not to use inline assembler,
before you can add checker in configure.

-- 
Taras Kondratiuk

_______________________________________________
lng-odp mailing list
[email protected]
http://lists.linaro.org/mailman/listinfo/lng-odp

Reply via email to