Re: [dpdk-dev] [PATCH v6 00/11] al: replace calls to rte_panic and refrain from new instances

[Arnon Warshavsky <arnon@qwilt.com>]

>
>
> What happened to this patchset?
>
> This is definitely an improvement. We must remove rte_panic from libs.
> Arnon, are you still available to rebase this patchset in preparation
> of 19.08? Or someone else?
>
> What are the required API breakages? I see one in ethdev which requires
> a deprecation notice to be sent for publishing in 19.05.
>
> Is there more rte_panic to remove?
>
>
>
Hi,
I should be able to address this for 19.08, at least the straight forward
parts.
I need to touch base again with this patchset later this week,
to see what changed since then, and see what deprecation notices are
required.
I would like to address the ones that are a direct part of the
initialization sequence,
and mostly change functions and their callers from void to return a value
that propagates upwards.
The 2nd kind under lib that I wanted to remove at the time are the ones
that live in threads and I would not like to handle them now.
Given a 3rd kind that is found inside PMDs that may panic during callbacks,
the former poses a similar challenge of managing the device state after a
panic event which is not trivial,
and tmho deserves either a separate patchset or a defeat recognition.

In this respect , In addition to removing the ones from the initialization
sequence,
I would like to revive my original proposal to add a callback registration
to the panic event.
As I do not expect all the PMD callback panics to disappear completely,
I still need to allow the running process to do some kind of orderly
tear-down to other modules when possible.

Does this sound ok for 19.08?