Re: [PATCH net] tools/net/ynl: fix cli.py --subscribe feature

From: Jakub Kicinski
Date: Tue Sep 03 2024 - 18:00:34 EST


On Tue, 3 Sep 2024 22:08:54 +0100 Donald Hunter wrote:
> > On Mon, 02 Sep 2024 10:51:13 +0100 Donald Hunter wrote:
> > > Reviewed-by: Donald Hunter <donald.hunter@xxxxxxxxx>
> >
> > Any preference on passing self.rsp_by_value, vs .decode() accessing
> > ynl.rsp_by_value on its own?
>
> .decode() accessing ynl.rsp_by_value would be cleaner, but I am
> working on some notification fixes that might benefit from the map
> being passed as a parameter. The netlink-raw families use a msg id
> scheme that is neither unified nor directional. It's more like a mix
> of both where req and rsp use different values but notifications reuse
> the req values. I suspect that to fix that we'd need to introduce a
> dict for ntf_by_value and then the parameter would be context
> specific. OVS reuses req/rsp values for notifications as well, but it
> uses a unified scheme, and that's mostly a problem for ynl-gen-c.

I was worried you'd say it's ID reuse related. That is tricky business.

> We could choose the cleaner approach just now and revisit it as part of
> fixing notifications for netlink-raw?

That's my intuition; there's a non-zero chance that priorities will
change or we'll head in a different direction, and the extra arg will
stick around confusing readers.