[dpdk-dev] [PATCH v3] mempool: reduce rte_mempool structure size

[Wiles, Keith]

>2016-02-12 15:07, Wiles, Keith:
>> >On 02/12/2016 03:57 PM, Thomas Monjalon wrote:
>> >> 2016-02-12 13:23, Panu Matilainen:
>> >>> On 02/10/2016 11:18 PM, Keith Wiles wrote:
>> >>>>    static inline void *rte_mempool_get_priv(struct rte_mempool *mp)
>> >>>>    {
>> >>>> +#ifdef RTE_NEXT_ABI
>> >>>> +	return (char *)mp +
>> >>>> +		MEMPOOL_HEADER_SIZE(mp, mp->pg_num, mp->cache_size);
>> >>>> +#else
>> >>>>    	return (char *)mp + MEMPOOL_HEADER_SIZE(mp, mp->pg_num);
>> >>>> +#endif /* RTE_NEXT_ABI */
>> >>>>    }
>> >>>
>> >>> This is not RTE_NEXT_ABI material IMO, the added ifdef clutter is just
>> >>> too much.
>> >>
>> >> The changes are restricted to the mempool files.
>> >> I think it is not so much. However I wonder how much the feature is important
>> >> to justify the use of NEXT_ABI.
>> >
>> >Well yes, to be precise: for the benefit of this patch, the ifdef 
>> >clutter seems too much.
>> >
>> >Its not as if every change is expected to go through a NEXT_ABI phase, 
>> >based on http://dpdk.org/ml/archives/dev/2016-February/032866.html there 
>> >might be some confusion regarding that.
>> 
>> I think the NEXT_ABI is reasonable in this case as it does change a structure everyone uses and the ifdef clutter is caused by having to remove old ifdefs, which is a good thing for DPDK. The NEXT_ABI ifdefs only exist for one release and then they will disappear, which I think is more then reasonable.
>
>OK, I'm going to sum it up with new words and let the conclusion comes
>from Keith, Panu and Olivier.
>
>We agreed to allow ABI breaking if a notification was done in the
>previous release.
>Keith has sent a notification for 16.04 so the "official" ABI will be
>changed in 16.07.
>It is also encouraged to show how the ABI will be broken when sending
>a notification. It allows to give an informed opinion before ack'ing.
>The code snippet will also be useful to app developpers when preparing
>a future upgrade.
>Keith has sent the whole code change.
>This code change may be submitted in the current release without waiting
>the deprecation time if gated in the NEXT_ABI ifdefs.
>It allows to provide the feature to app developpers who don't care about
>versioning. But the price is a more complicated code to read and manage.
>
>To make it short, the rules to use NEXT_ABI are not strict and may change.
>So now you have to decide if this change can be integrated in 16.04
>as NEXT_ABI.

I would personally go ahead with the NEXT_ABI in 16.04 as it seems more reasonable for developers. I do not know if we made this patch in 16.04 without NEXT_ABI what would break in a developers project, which to me means we need to error on the side of caution by using NEXT_ABI.

I am willing to submit a v4 patch without the NEXT_API ifdefs, but that is something everyone needs to agree on.
>
>


Regards,
Keith