Hello,

Paolo Bonzini wrote:
> 
>> Well, the %qD gives "<declaration error>", which seemed OK to me. With
>> what you suggest, we'll not give any error about the fact that a named
>> initializer is not valid here.
> 
> It's your call, it depends on how the errors before this one look like,
> and whether this one would be automatically fixed after correcting the
> errors before.
I don't have a pre-3.3 compiler around, so I cannot tell for sure, but
looking at the code, I think that we would not report the error. I think
it's not perfect because even if the user fixes the first error (X being
unknown; could be a typo for a member if we were working on an array of
structs instead of an array of ints), the code will still be erroneous.

>> So if the "<declaration error>" really is bad, I could submit another
>> patch giving the current error message when we have an IDENTIFIER_NODE,
>> and another one ("ce independent") when we have an error_mark_node.
> 
> Yes, that's a possibility.
The attached patch does this; the test case and Changelogs are unchanged.

I've successfully tested it on x86_64-apple-darwin-9. Is it better?
OK for 4.3 and for the mainline?

Thanks,
Simon