P4-16 draft section 5 should mention deparsers about p4-spec

[vgurevich]

Hello Ben,

In P4-16 deparsers are nothing more than control flows. So, yes, this is an
important concept and, usually, a standalone block from the switch
architecture perspective, but there are no specialized abstractions for
deparsers in P4-16, at least not currently. That's why deparsers were not
mentioned in Section 5, but they are mentioned in other places.

We had several discussions about this. The main point is that this is
simple imperative code, not different from any other control flow (unlike
parsers, where the control flow is expressed via state transitions instead
of "traditional" if() statements). The arguments to a deparser would
typically include a packet abstraction and the code is allowed to use
primitives, such as emit(), but one can argue that depending on the target
and architecture there is nothing special about that and these might be
allowed inside other control blocks.

Happy hacking,
Vladimir

On Mon, Aug 15, 2016 at 2:12 PM, Ben Pfaff [email protected] wrote:

The P4-16 draft spec, section 5, lists the core abstractions of P4. I
suggest that it should mention deparsers.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#31, or mute the thread
https://github.com/notifications/unsubscribe-auth/AMPb84AAUSzIUbd5ezsDuzbsiFWSEM1fks5qgNZUgaJpZM4Jkxs9
.

[blp]

I see.

It seems that adding a sentence to the definition of "Control flow" would be helpful to readers. For example, "Some targets support deparser control flow blocks that reassemble packets following pipeline processing."

[vgurevich]

Thanks, Ben!

I think that does make sense. Also, please, check Section 16 ("Deparsing").
There is a paragpraph in there with the same meaning, but I think it
wouldn't hurt to mention that control flows can also be used for deparsing
in Section 5.

Happy hacking,
Vladimir

On Tue, Aug 16, 2016 at 11:52 AM, Ben Pfaff [email protected]
wrote:

I see.

It seems that adding a sentence to the definition of "Control flow" would
be helpful to readers. For example, "Some targets support deparser
control flow blocks that reassemble packets following pipeline processing."

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#31 (comment), or mute
the thread
https://github.com/notifications/unsubscribe-auth/AMPb8xzcO7_IFpq933-FS2La0Dd4xmg7ks5qggbwgaJpZM4Jkxs9
.

[chkim4142]

@blp Could you please close this by making necessary changes to the draft with Mihai? Or, has this been already taken in the latest draft?

[blp]

@chkim4142 This has been fixed in the draft.