aboutsummaryrefslogtreecommitdiff
path: root/lib/Sema/SemaCXXScopeSpec.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Sema/SemaCXXScopeSpec.cpp')
-rw-r--r--lib/Sema/SemaCXXScopeSpec.cpp6
1 files changed, 5 insertions, 1 deletions
diff --git a/lib/Sema/SemaCXXScopeSpec.cpp b/lib/Sema/SemaCXXScopeSpec.cpp
index 251ffea925..352e553edb 100644
--- a/lib/Sema/SemaCXXScopeSpec.cpp
+++ b/lib/Sema/SemaCXXScopeSpec.cpp
@@ -357,7 +357,8 @@ Sema::CXXScopeTy *Sema::ActOnCXXNestedNameSpecifier(Scope *S,
// unqualified name lookup in the given scope.
// FIXME: When we're instantiating a template, do we actually have to
- // look in the scope of the template? Seems fishy...
+ // look in the scope of the template? Both EDG and GCC do it; GCC
+ // requires the lookup to be successful, EDG doesn't.
Found = LookupName(S, &II, LookupNestedNameSpecifierName);
ObjectTypeSearchedInScope = true;
}
@@ -366,6 +367,9 @@ Sema::CXXScopeTy *Sema::ActOnCXXNestedNameSpecifier(Scope *S,
// base object type or prior nested-name-specifier, so this
// nested-name-specifier refers to an unknown specialization. Just build
// a dependent nested-name-specifier.
+ if (!Prefix)
+ return NestedNameSpecifier::Create(Context, &II);
+
return NestedNameSpecifier::Create(Context, Prefix, &II);
} else {
// Perform unqualified name lookup in the current scope.