Re: [dpdk-dev] [PATCH v2] doc: announce ABI change for rte_eth_dev structure

[Thomas Monjalon <thomas.monjalon@6wind.com>]

2016-07-28 21:55, Jerin Jacob:
> On Thu, Jul 28, 2016 at 04:52:45PM +0200, Thomas Monjalon wrote:
> > 2016-07-28 19:29, Jerin Jacob:
> > > Above things worries me, I wouldn't have cared if the changes are not comes
> > > in fastpath and I don't think this sort of issues will never get fixed any time
> > > soon in this community.
> > > 
> > > So I given up.
> > 
> > I feel something goes wrong here but I cannot understand your sentence.
> > Please could you reword/explain Jerin?
> 
> I guess you have removed the context from the email. Never mind.
> 
> 1) IMHO, Introducing a new fast path API which has "performance impact"
> on existing other PMD should get the consensus from the other PMD maintainers.
> At least, bare minimum, send a patch much in advance with the
> implementation of ethdev API as well as PMD
> driver implementation to get feedback from other developers _before_ ABI
> change announcement rather just debating on hypothetical points.

I totally agree with you and it was my first comment in this thread:
	http://dpdk.org/ml/archives/dev/2016-July/044366.html
Unfortunately it is difficult to have a formal process so it is not
so strict currently. You are welcome to suggest how to improve the
process for the next releases.

> 2) What I can understand from the discussion is that it is the
> workaround for an HW limitation.
> At this point, I am not sure tx_prep is the only way to address it and
> do other PMD have similar
> restriction?. If yes, Can we have abstract it in a proper way the usage
> part will be very clear from PMD and application perspective?

I feel the tx_prep can be interesting to solve a current problem.
However, as you say, there are maybe other solutions to consider.
That's why I think we can keep this deprecation notice and follow up
with a patch-based discussion. We will be able to discard this change
if something better is found.
As an example, we have just removed a deprecation notice which has
never been implemented:
	http://dpdk.org/browse/dpdk/commit/?id=16695af340
I know this process is not perfect and the ethdev API is far from perfect,
so we must continue improving our process to define a good API.

Konstantin, Tomasz,
I generally prefer waiting for a consensus. For this case, I'll make an
exception and apply the deprecation notice.
Please make an effort to better explain your next patches and meet
a clear consensus. We'll review your patches very carefully and keep
the right to reject them.