[llvm] r214761 - Reapply "DebugInfo: Ensure that all debug location scope chains from instructions within a function, lead to the function itself."

[Reid Kleckner]

On Mon, Oct 6, 2014 at 11:39 AM, David Blaikie <dblaikie at gmail.com> wrote:

> That would be mighty unfortunate... :/ Perhaps we can find some assertions
> to put in to stabilize the failure, though I haven't a clue what they'd be
> just yet.
>

I'm staring at the .ll file diff, trying to figure out what went wrong.
There are several small differences, and I wonder if they are for things
like order of evaluation, where the only difference is a use-list order
change.

Yep, that test case doesn't produce an assertion for me. Crud. What's a
> useful way for me to debug this on a Windows machine? Am I just going to
> have to sit with you/someone - or can I get remote access somehow (from
> Linux, ideally?)?
>

Most Windows remote access solutions tend to take over the primary screen,
unless you have some server OS variant, so if you want to debug it directly
you'd want to rig up a local VM. This is doable, there are gwindows images
you can fetch, but I'll let you know why I give up over here. :)

Here's the function we assert on:

; Function Attrs: noreturn nounwind
define internal void
@"\01??$assert_cast at PAVRedeclarableTemplateDecl@clang@@@@YAPAVRedeclarableTemplateDecl at clang@@ZZ"()
#4 {
entry:
  tail call void @"\01?llvm_unreachable_internal at llvm@@YAXPBD0I at Z"(i8*
getelementptr inbounds ([16 x i8]*
@"\01??_C at _0BA@OELEAIJA at bad?5assert_cast?$AA@", i32 0, i32 0), i8*
getelementptr inbounds ([51 x i8]*
@"\01??_C at _0DD@BABLEMGH@?4?4?2tools?2clang?2lib?2Serialization@", i32
0, i32 0), i32 2108) #12, !dbg !34160
  unreachable, !dbg !34160
}

There are no diffs in the code, but here's some related debug info that is
suspiciously different:

-!1590 = metadata !{metadata !"0x2e\00assert_cast<clang::NamespaceDecl
*>\00assert_cast<clang::NamespaceDecl
*>\00\002107\001\001\000\000\00256\001\002107", metadata !13, metadata
!14, metadata !7, null, void ()*
@"\01??$assert_cast at PAVRedeclarableTemplateDecl@clang@@@@YAPAVRedeclarableTemplateDecl at clang@@ZZ",
null, null, metadata !2}
+!1590 = metadata !{metadata !"0x2e\00assert_cast<clang::NamespaceDecl
*>\00assert_cast<clang::NamespaceDecl
*>\00\002107\001\001\000\000\00256\001\002107", metadata !13, metadata
!14, metadata !7, null, null, null, null, metadata !2} ; [
DW_TAG_subprogram ] [line 2107] [local] [def]
[assert_cast<clang::NamespaceDecl *>]

Thoughts?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20141006/5a861c48/attachment.html>