Open pat-rogers opened 3 years ago
Good question, but not something we can resolve now. Deferred to next time! ;-)
Certainly no one considered that. Given the very strict restrictions on extending language-defined attributes, I would argue that there is no problem - these attributes are not considered impl-def (especially since they are based either on previous or likely future standards). The wording of 13.12.1(2/2) is rather ambiguous, though, one could read it to mean "attributes with implementation-defined results". That would knock out many language-defined attributes (even Size and Alignment), so it doesn't seem likely that was intended.
So I don't think there is a problem here, but I'm OK with deferring the issue so that it can be considered more carefully (and perhaps the 13.12.1 wording should be cleaned up).
We say "An implementation may extend the definition of a language-defined attribute ... in the following cases:", but 13.12.1 does not talk about how this interacts with the No_Implementation_Attributes restriction. Was this intended?
SteveB