On Jun 25, 2012, at 7:55 PM, Richard Smith wrote: > On Mon, Jun 25, 2012 at 7:47 PM, Richard Trieu <[email protected]> wrote: > On Mon, Jun 25, 2012 at 2:09 PM, Richard Trieu <[email protected]> wrote: > On Sun, Jun 24, 2012 at 2:57 PM, Richard Smith <[email protected]> wrote: > Thanks, LGTM. > > > Incidentally, for this example: > > > template<typename ...T> struct S {}; > template<typename T> using U = S<int, char, T>; > template<typename T> using V = U<U<T>>; > int f(S<int, char, U<const double>>); > int k = f(V<double>()); > > I get: > > <stdin>:4:5: note: candidate function not viable: no known conversion from > 'S<[2 > * ...], struct S<int, char, double>>' to 'S<[2 * ...], U<const > double>>' for 1st argument; > > where I would have preferred: > > <stdin>:4:5: note: candidate function not viable: no known conversion from > 'S<[2 > * ...], S<[2 * ...], double>>' to 'S<[2 * ...], S<[2 * ...], const > double>>' for 1st argument; > > Any idea what's going on there? > > > Looking into it. It looks like Clang has 'struct S<int, char, double>' as a > SubstTemplateTypeParmType instead of a TemplateSpecializationType, causing > the template diffing to handle it as a type without going deeper. > > Fixed. > > > + static const TemplateSpecializationType * GetTemplateSpecializationType( > > Extra space after *. > > > + if (const SubstTemplateTypeParmType *STTPT = > > + Ty->getAs<SubstTemplateTypeParmType>()) > > + Ty = STTPT->getReplacementType(); > > Is this necessary? Since you next do this:
Not only is it unnecessary, but getAs<SugaredType>() is basically just an expensive dyn_cast, because the canonical-type check will never succeed. John.
_______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
