On Mon, Dec 16, 2019 at 01:47:49PM +0100, Florian Westphal wrote:
> Pablo Neira Ayuso <pablo@xxxxxxxxxxxxx> wrote:
> > Hi Florian,
> >
> > This patchset removes the need to self invoke the parser and the
> > evaluation to fetch the datatype. Instead, the expression type and
> > the expression description are stored into the userdata area.
> >
> > This patch only supports for the payload expression, but it should be
> > relatively easy to extend it to support for other existing expressions
> > types.
> >
> > This patch could be squashed into 06/11 src: add "typeof" print support
> > of your patch series, which is actually not just adding support for
> > printing but also for building the userdata.
>
> I had considered that but found that storing netlink data
> needs more space in the udata area compared to text and it needs more/extra
> parsing for serialize/deserialize, so I abandoned this idea.
Yes, it's scratching a bit extra of the userdata area.
The listing path should be easier, since it's just parsing the TLVs
instead of invoking the nft parsing and evaluation phases.
> If you think its the way to go, then ok, I can rework it but
> I will be unable to add the extra steps for other expression types
> for some time I fear.
If you send a v3 including this work, I'll finishing the remaining
expressions.
One more thing regarding your patchset is:
integer,128
If the typeof works for all of the existing selectors, then I think
there is not need to expose this raw type, right?
Thanks.
