<html><head><meta http-equiv="Content-Type" content="text/html charset=iso-8859-1"></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; "><div><div>On Mar 22, 2013, at 4:24 PM, Reid Kleckner <<a href="mailto:rnk@google.com">rnk@google.com</a>> wrote:</div><blockquote type="cite"><div dir="ltr">On Fri, Mar 22, 2013 at 4:44 PM, John McCall <span dir="ltr"><<a href="mailto:rjmccall@apple.com" target="_blank">rjmccall@apple.com</a>></span> wrote:<br><div class="gmail_extra"><div class="gmail_quote">
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="im">On Mar 22, 2013, at 11:41 AM, Reid Kleckner <<a href="mailto:rnk@google.com">rnk@google.com</a>> wrote:<br>
> - Update the test case to pass under x86_64.<br>
><br>
> Hi rjmccall,<br>
><br>
> <a href="http://llvm-reviews.chandlerc.com/D568" target="_blank">http://llvm-reviews.chandlerc.com/D568</a><br>
<br>
</div>+ // FIXME: Handling x86_64 will require changing this method to return sizes<br>
+ // in bytes instead of pointers. Some member pointer sizes are not<br>
+ // multiples of a pointer size.<br>
<br>
So, change the method to return sizes in bytes instead of pointers instead<br>
of reporting a diagnostic in a core AST routine.<br></blockquote><div><br></div><div style="">OK, I'll change the method. I assumed there'd be more than one caller. :)</div><div><br></div><div style="">What should I do for non-x86 architectures then? I could assert, since presumably we'll have errorred out much earlier if someone asks the frontend for mips code.</div></div></div></div></blockquote><div><br></div>I'd formalize it as a difference between 32-bit and 64-bit, but yes, you don't need to go completely hog-wild supporting the MS ABI on non-MS-supported targets.</div><div><br></div><div><blockquote type="cite"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div style="">There's prior art for issuing diagnostics in AST from RecordLayoutBuilder, but maybe that's different.</div></div></div></div></blockquote><div><br></div>Record layout is only done at most once. It's not *evil*, since this is an "unsupported" diagnostic, but yeah, there's no reason to not just get it right here.</div><div><br></div><div><blockquote type="cite"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
+// getNumBases() seems to only give us the number of direct bases, and not the<br>
+// total. This function tells us if we inherit from anybody that uses MI.<br>
+static bool recordHasMultipleInheritance(const CXXRecordDecl *RD) {<br>
+ while (RD->getNumBases() > 0) {<br>
+ if (RD->getNumBases() > 1)<br>
+ return true;<br>
+ assert(RD->getNumBases() == 1);<br>
+ RD = RD->bases_begin()->getType()->getAsCXXRecordDecl();<br>
+ }<br>
+ return false;<br>
+}<br>
<br>
Hmm. I bet that non-primary base class forces the "multiple inheritance" case:<br>
struct A { int x; void bar(); };<br>
struct B : A { virtual void foo(); };<br>
What's sizeof(void (B::*)())?<br></blockquote><div><br></div><div style="">Nice catch, 2 * sizeof(void*), so it's the "multiple" case. Added a test case.</div><div style=""><br></div><div style="">What do I need to look for to identify this case? A class that is polymorphic, but whose base is not?</div>
</div></div></div></blockquote><div><br></div>That would work.</div><div><br></div><div>John.</div></body></html>