Let annotations take an argumentList rather than expressionList | keyValueList

[SantiagoBautista]

Currently in the P4-16 grammar, annotations can take either an expression list or a keyValuePair List for their arguments and this allows arguments to annotations to be named:

annotation
    : '@' name
    | '@' name '(' expressionList ')'
    | '@' name '(' keyValueList ')'
    ;

But this makes the grammar ambiguous, because there is no way of telling whether the empty list is an expressionList or a keyValueList.

There is a non-ambiguous and nicer way of achieving exactly the same behavior:
there is a non-terminal in P4-16 grammar called argumentList , and each argument can either be an expression or a name '=' expression (i.e. a keyValuePair).

So if annotations took an argumentList, instead of either an expressionList or a keyValueList, we would have exactly the same behavior, but without introducing ambiguity to the grammar.

I will open a Pull Request implementing this solution.

[jnfoster]

I guess the one difference is "_". But generally, yes, this seems like a nice simplification.

…

On Thu, May 31, 2018 at 4:40 PM, Santiago Bautista ***@***.*** > wrote: Currently in the P4-16 grammar, annotations can take either an expression list or a keyValuePair List for their arguments and this allows arguments to annotations to be named: annotation : '@' name | '@' name '(' expressionList ')' | '@' name '(' keyValueList ')' ; But this makes the grammar ambiguous, because there is no way of telling whether the empty list is an expressionList or a keyValueList. There is a non-ambiguous and nicer way of achieving exactly the same behavior: there is a non-terminal in P4-16 grammar called argumentList , and each argument can either be an expression or a name '=' expression (i.e. a keyValuePair). So if annotations took an argumentList, instead of either an expressionList or a keyValueList, we would have exactly the same behavior, but without introducing ambiguity to the grammar. I will open a Pull Request about this. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#633>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABwi0j3bQZIHZHwYKEi5qNqNArmxKABgks5t4FU2gaJpZM4UVs10> .

[SantiagoBautista]

To be precise, there is another difference: if annotations take argumentList, then one single annotation can take both key-value pairs and expressions; whereas with the current grammar one annotation can either only take expressions or only take key-value pairs.
So the simplification I propose would allow a more flexible behavior (both because of '_' and the possibility of mixing expressions and key-value pairs), that strictly contains the current behavior.

But this extra flexibility can, I guess, be useful, and AFAIK it doesn't introduce any problems.

[SantiagoBautista]

For information:
This issue is related to #605 and #625 .
So @mbudiu-vmw @ChrisDodd and @chrispsommers could be interested.

[chrispsommers]

@SantiagoBautista I like this, good catch!

[SantiagoBautista]

Thank you @chrispsommers !

[mihaibudiu]

This is no longer relevant, since we now have free-form annotations.