Re: [dpdk-dev] [EXT] [PATCH v1 3/4] regexdev: add regexdev core functions

[Thomas Monjalon <thomas@monjalon.net>]

07/04/2020 07:49, Jerin Jacob:
> > >
> > > If it abstracts it properly adding vdev and PCI is a simple change.
> > > See
> > >
> > > lib/librte_eventdev/rte_eventdev_pmd_vdev.h
> > > lib/librte_eventdev/rte_eventdev_pmd_pci.h
> > >
> > > I think, the common code should take from other matured subsystem instead if
> > > writing from scratch,
> > >
> > I agree with you about the rewrite, but this is why I don't want to add this code
> > until I know what this code should do and how it should be used.
> > I  don't agree, that one subsystem is like other one by default, and that if something
> > is done for one subsystem it should be done for other.
> > Not always what was done before is the best.
> >
> > Some time back there was a long thread about ethdev and the rte device
> > where should it be released and by whom.
> > My basic thinking is that unless proven otherwise the code should be in the PMD
> > this is also why it is important for me to get this rte level API acked.
> > when starting to implement the code for the PMD it will be cleared what
> > is the shared code and how it is best to configure the system.
> > Also this is not external API so it can be changed at any time.
> > Saying that I don't think we should wait long before adding such code.
> > I think that when we will have our first PMD we know better if such
> > function is needed.
> > Also think about that maybe this PMD will be shared with the
> > net PMD so such function will also introduce more complexity.
> 
> 
> My thought process was I like this when I added the common code for eventdev.
> I have checked ethdev, cryptodev and followed a similar scheme
> wherever it applicable for eventdev.
> If we see the regexdev API, it is similar to ethdev. cryptodev and
> eventdev API. From the device
> API PoV, the framework needs to follow the same behavior to avoid
> having new behavior for regexdev,
> Especially in configure->queue_setup->start->rx_burst->tx_burst->stop->reconfigure->start
> sequence.
> 
> 
> Ethdev may be bloated by features,

No, ethdev has more features and is evolving as a better API.

> My request is to take cryptodev and
> eventdev as a base
> change to accordingly.

When writing a new API, we should learn from past lessons,
including eventdev and cryptodev, but also all the things evolving
in ethdev.

> That makes review process easy, As reviewer
> needs only think, The rationale  behind,
> Why it regexdev common code chosen a different approach. Writing from
> scratch makes the reviewer's job
> difficult.
> 
> We spend a lot of time reviewing and make mature cryptodev and
> eventdev common code, Please leverage that.

Again, leveraging doesn't mean to implement all in one patch.
Instead of asking for adding more, please focus on what must be changed
in what is proposed in this series.