[cfe-commits] Handle missing 'typename' in function template definitions

[Douglas Gregor]

On May 15, 2012, at 6:12 PM, Richard Smith <richard at metafoo.co.uk> wrote:

> Hi,
> 
> The attached patch allows Clang to recover better from a missing 'typename' keyword in a function template definition:
> 
> <stdin>:1:29: error: missing 'typename' prior to dependent type name 'T::type'
> template<typename T> void f(T::type) {}
>                             ^~~~~~~
>                             typename 
> 
> With this patch, we carry on disambiguating past such a situation. If we can't disambiguate the declaration in some other way, the absence of 'typename' is used to infer that we have a variable declaration. I've also extended the disambiguation code to inspect the token immediately after the parameter-declaration-clause to disambiguate (in the above case, the { token is used to disambiguate, and more generally we will also look for an exception-specification, ref-qualifier, cv-qualifier, etc., which cannot appear after a parenthesized initializer in a variable declaration).
> 
> Does this seem like a reasonable approach?

I like this approach. Just one trivial comment:

+    else if (NextToken().is(tok::amp) || NextToken().is(tok::ampamp) ||
+             NextToken().is(tok::kw_const) ||
+             NextToken().is(tok::kw_volatile) ||
+             NextToken().is(tok::kw_throw) ||
+             NextToken().is(tok::kw_noexcept) ||
+             NextToken().is(tok::l_square) ||
+             isCXX0XVirtSpecifier(NextToken()) ||
+             NextToken().is(tok::l_brace) || NextToken().is(tok::kw_try) ||
+             NextToken().is(tok::equal))

Please cache NextToken() here.

	- Doug

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20120516/8c2165d7/attachment.html>