[cfe-commits] [PATCH] Parsing C++0x lambda expressions

[David Blaikie]

> I forgot to add that the latest patch adds a call to the new lambda
> parsing function, and one of the tests breaks:
> Parser/cxx0x-attributes.cpp. On line 20, attribute syntax ("[[]]") is
> used where an expression is expected.

Actually it's a bit more nuanced than that - line 20 is

     void after_const_attr () const [[]];

the [ is valid at parse time - that would be the start of declaring a
function that returns an array (which is illegal, but I assume that
it's just easier to parse this in a unified way & pick it up in
semantic analysis). Of course the thing inside the [] has to be an
expression - in this case it was yet more [], trying to be parsed as
an expression. Well now [ is a valid way to start an expression, so
the error is slightly different.

> Consequently, instead of
> generating the expected error ("expected expression"), two errors
> related to lambda parsing are generated. The first comes from the lambda
> parsing function, and the second comes after lambda parsing returns
> ExprEmpty(). How should this be handled?

Since the lambda parser succeeded at parsing this as an expression,
the parser progressed (or the semantic analysis - I didn't check where
the second message came from) & found that void wasn't a sensible
element type to try to return an array of from a function.

Should we fail to parse this entirely at the moment? (so we don't get
the second error) Eventually we could suggest fixups for this error
(simply adding {} is probably the only thing we can really guess).

[& I'm going to try adding some more (negative & positive) test cases
as per Doug's request/suggestion now]