Please take another look
================
Comment at: include/clang/Basic/ABI.h:111
@@ +110,3 @@
+ /// adjustment, if needed.
+ union VirtualAdjustment {
+ // Itanium ABI
----------------
Reid Kleckner wrote:
> I'm still not really totally comfortable with the unionification of this
> stuff. We should ask John what he thinks the right design is for this stuff.
> This is consistent with what's already been committed, but we should
> probably figure out the right way to do this going forward.
During offline discussion, we've agreed to use unions for now* to significantly
improve the support of virtual calls in this ABI and explicitly ask John for a
post-commit review of the overall ABI.h change.
* - Mostly since we're highly unlikely to change anything for the Itanium ABI
soon and just unlikely to need any changes for the Microsoft ABI.
================
Comment at: lib/AST/MicrosoftMangle.cpp:1849
@@ -1848,2 +1848,3 @@
MicrosoftCXXNameMangler &Mangler,
raw_ostream &Out) {
+ if (!Adjustment.Virtual.isEmpty()) {
----------------
Reid Kleckner wrote:
> This is complicated enough that it deserves it's own BNF comment with
> portions of the mangleFunctionClass() BNF diagram. The G, O, W and A, I, Q
> access codes clearly correspond to it.
Agreed. I'm not a BNF master, so formatting/content suggestions are welcome!
It has actually reminded me we might sometimes have far thunks... I'm not sure
when are they needed (hey, why would you need more than +-2^31 to adjust this
inside a record type?) so I've added a FIXME. Anyways, the mangleFunctionClass
didn't ever handle "far" functions either.
http://llvm-reviews.chandlerc.com/D2079
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
