Arguments allow double extended attributes

[saschanaz]

Argument ::
    ExtendedAttributeList ArgumentRest

ArgumentRest ::
    optional TypeWithExtendedAttributes ArgumentName Default
    Type Ellipsis ArgumentName

ArgumentName ::
    ArgumentNameKeyword
    identifier

Currently the syntactic item Argument is defined as above, syntactically allowing double extended attributes only for optional arguments but not for non-optional one: [ExtAttr] optional [Clamp] short argname. Is this intended?

I think this causes ambiguity before reading the token optional, where the first extended attribute typically is for the type but not when optional.

[annevk]

This seems like something we should get rid of.

[bzbarsky]

I'm not sure I understand what the problem is.

void foo([ExtAttr] optional [Clamp] short argname);
void bar([ExtAttr, Clamp] short argname);

In the former case (foo), everything is clear: [ExtAttr] applies to the argument as a whole, while [Clamp] applies to the type.

In the latter case (bar) things are a bit more complicated because there is only one space to put extended attributes. Both [ExtAttr] and [Clamp] are initially applied to the argument as a whole, but then [Clamp] is propagated through to the type by https://heycam.github.io/webidl/#idl-type-extended-attribute-associated-with step 4. So the end result is that [Clamp] applies to the type and both extended attrs apply to the argument itself. Since the argument itself does nothing with its extended attributes for the moment, that's fine. If that were to change, whatever it does would need to ignore the extended attributes matching https://heycam.github.io/webidl/#extended-attributes-applicable-to-types

The fact that you can have extended attributes on the argument itself is intentional; while Web IDL itself does not currently define any such extended attributes, consumers of Web IDL might.

[saschanaz]

@bzbarsky So you mean void bar([Clamp] optional short argname); is semantically correct, because it will be propagated? @domenic previously opened https://github.com/w3c/respec/issues/1184 so I thought it was incorrect.

[bzbarsky]

So you mean void bar([Clamp] optional short argname); is semantically correct

No, it's not. We should probably fix the propagation to only happen for non-optional arguments...

[bzbarsky]

@domenic thoughts on #691 (comment) ?

[domenic]

So that'd be modifying step 4 of https://heycam.github.io/webidl/#idl-type-extended-attribute-associated-with ? Seems reasonable to me.

[saschanaz]

Note that this also applies to dictionary members:

DictionaryMember ::
    ExtendedAttributeList DictionaryMemberRest

DictionaryMemberRest ::
    required TypeWithExtendedAttributes identifier ;
    Type identifier Default ;

like [Clamp] required short membername, which is related with step 5.

[saschanaz]

Looks like the step 4 and 5 already makes the propagation to only happen for non-optional/non-required members, as:

If type appears as part of a Type production directly within an Argument production

... a direct Type production currently means no optional or required.