Action calls from parsers

[mihaibudiu]

Can parsers directly call actions?

[mihaibudiu]

One issue to consider is whether such actions can call exit. exit is only allowed in controls.

[jafingerhut]

It seems perfectly reasonable that if actions can be called from parsers, then those actions should be limited to things that could be done within a parser, e.g. no return, no exit, no apply on tables, etc.

Even with those restrictions, such an action seems useful in some scenarios, e.g. to give a name to that function, and enable using it in multiple places without copy & paste.

[jnfoster]

As Mihai has pointed out several times, actions are P4's only function construct. So this seems natural...

…

-N

On Sat, Apr 1, 2017 at 2:13 PM, Andy Fingerhut ***@***.***> wrote: It seems perfectly reasonable that if actions can be called from parsers, then those actions should be limited to things that could be done within a parser, e.g. no return, no exit, no apply on tables, etc. Even with those restrictions, such an action seems useful in some scenarios, e.g. to give a name to that function, and enable using it in multiple places without copy & paste. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#151 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABwi0goefO29vtcLGw040WsSHkE9WHt5ks5rrr3lgaJpZM4MwSRi> .

[mihaibudiu]

This should be on the agenda for the next design meeting.

[cole-barefoot]

From a usability/code reuse perspective, it would be nice to know from the type of an action whether it can be used in a parser, rather than having to inspect the body of the action. But maybe this is a longer-term goal...

[jafingerhut]

In 2017-Apr-10 language design meeting, someone suggested keeping this issue, but changing the title to 'Add functions to P4' (some time after P4_16 first release in May). @mbudiu-vmw Do you have ability to change the title? I don't think I do. Also happy to close this one and open a new one if that is easier.

Rationale for not allowing actions to be called from within parsers - actions are intended to be part of a match-action pipeline. It seems a bit too much of a hack to allow calling them from inside parsers. It would be preferable to add functions to P4 vs. using actions in places they were not intended.

Below are what I think might be some properties of P4 functions. They have a heavy overlap with properties of actions. The main difference between actions and functions as I describe them below are that actions can have directionless parameters, but functions cannot, and functions can have return values. I realize that maybe my expectations of P4 functions are different than those of others.

• all parameters must have directions specified, i.e. no directionless parameters
• have a return type that could be of many possible types, at least bit<W> and int<W>, and perhaps most base types and even derived types like structs would be allowed as return types.
• Can be called from inside parser states, control apply blocks, actions, or other functions. As always, target-specific compilers would be responsible for rejecting programs that are too complex for the target.
• Can be defined at top level, or inside of a parser or control block top level. TBD whether it is considered important to be able to define them within parser states, control apply blocks, actions, or other functions.
• Can read header or metadata values that are in scope where they are defined, or modify them (i.e. side effects would be permitted), subject to possible in/out/inout restrictions on access to those values in the enclosing scope.

[mihaibudiu]

Adding functions would also solve issue #51

[jnfoster]

@jafingerhut I think we resolved not to take this and think about adding proper functions down the road. If you agree, can we close this? If not, let's update the spec?

[jafingerhut]

@jnfoster I am perfectly fine with either of these: (a) rename this issue as 'Add functions to P4' and mark it as post P4_16, (b) close this issue and create a new one for adding functions to P4.

[jnfoster]

Tagged as post P4_16.

[mihaibudiu]

Since we have functions I think we can close this issue.