rjmccall added a comment.
Oh, I see. The code currently tries to work with just the specialization and
the pattern. To do the instantiation, we have to find template arguments for
the context in which the pattern appears. For function temploids that aren't
defined in a friend declaration, we can just consider the semantic DC hierarchy
of the specialization, which will be the same across all redeclarations. For
function temploids that *are* defined in a friend declaration, we have to look
at the lexical DC of the declaration that was actually instantiated from the
class temploid that defined the friend function, which isn't necessarily the
specialization because it might have been redeclared after the friend function
was instantiated. So your patch is mostly just changing the find-the-pattern
lookup to remember that lexical DC when we find a pattern this way, which seems
sensible. Richard should look over the actual lookup-algorithm changes.
================
Comment at: include/clang/AST/Decl.h:2443
+ /// defined in a friend declaration, this parameter is assigned pointer to
the
+ /// class containing the friend declaration.
+ ///
----------------
"If this is non-null and the pattern is a friend function declaration, this is
set to the class containing the friend declaration."
================
Comment at: lib/AST/Decl.cpp:3389
+ if (Orig->getFriendObjectKind() != Decl::FOK_None) {
+ if (auto *RDC = dyn_cast<CXXRecordDecl>(D->getLexicalDeclContext()))
+ Host = RDC;
----------------
I don't think this can fail; a friend should always be lexically nested in a
class.
================
Comment at: lib/AST/Decl.cpp:3399
+ return Def;
+ }
+ }
----------------
It's unfortunate that this needs to walk the redeclaration chain itself looking
for definitions. Can't you just call `getDefinition` to find the definition
and then run your extra checks on that?
================
Comment at: lib/AST/Decl.cpp:3444
+ if (MFD->getFriendObjectKind() != Decl::FOK_None)
+ if (auto *DC = dyn_cast<CXXRecordDecl>(getLexicalDeclContext()))
+ if (HostForFriend)
----------------
Same observation.
Repository:
rC Clang
https://reviews.llvm.org/D21767
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
