================
Comment at: lib/CodeGen/CGExprCXX.cpp:1730
@@ +1729,3 @@
+ I != E; ++I) {
+ if (I->Access == AS_public) { // Ignore non-public inheritance.
+ for (CXXBasePath::iterator J = I->begin(), JE = I->end(); J != JE; ++J) {
----------------
Chandler Carruth wrote:
> As Doug commented, a continue seems better here.
done.
================
Comment at: lib/CodeGen/CGExprCXX.cpp:1737-1740
@@ +1736,6 @@
+
+ // Accumulate the base class offsets.
+ const ASTRecordLayout &Layout = Context.getASTRecordLayout(J->Class);
+ Offset +=
+ Layout.getBaseClassOffset(J->Base->getType()->getAsCXXRecordDecl());
+ }
----------------
Chandler Carruth wrote:
> If we're on our second path, you can skip this computation.
done.
================
Comment at: lib/CodeGen/CGExprCXX.cpp:1717
@@ +1716,3 @@
+ CXXBasePaths Paths(/*FindAmbiguities=*/true, /*RecordPaths=*/true,
+ /*DetectVirtual=*/false);
+
----------------
Chandler Carruth wrote:
> Why not detect virtual? Then the path building can short circuit, and you can
> just detect it and return the -1?
That would fire on private virtual inheritance, which we don't want to abort on
right now. It's arguably a weird edge case but it matches GCC's behavior. This
is covered by a case in the test for this change.
http://llvm-reviews.chandlerc.com/D364
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
