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

[Reid Kleckner]

Yeah, let me try a self-host with that patch, assuming it reverts cleanly
at TOT.

On Sun, Oct 5, 2014 at 9:09 AM, David Blaikie <dblaikie at gmail.com> wrote:

> Hey Reid - I seem to recall last time we spoke it turned out you'd sent
> the wrong .cpp file and now only had the IR for the reproduction?
>
> Any way we could get the original source reproduction for this (reducing
> debug test cases in IR is difficult)? I'd really like to move forward with
> this change as it introduces some useful invariants (and violations of
> those invariants most likely represent pretty clear bugs in LLVM's debug
> info for inlined functions).
>
> On Sat, Aug 9, 2014 at 8:59 AM, Reid Kleckner <rnk at google.com> wrote:
>
>> On Fri, Aug 8, 2014 at 2:40 PM, David Blaikie <dblaikie at gmail.com> wrote:
>>
>>> On Wed, Aug 6, 2014 at 3:54 PM, Reid Kleckner <rnk at google.com> wrote:
>>> > Here's an O0 .ll file that breaks llc if you run opt -O2 on it, but
>>> passes
>>> > if you run llc:
>>> > $ llc t.ll  # succeeds
>>> > $ opt -O2 t.ll -o - | llc -o t.s
>>> > Assertion failed: DISubprogram(Scope).describes(MF->getFunction()),
>>> file
>>> > ..\lib\CodeGen\LexicalScopes.cpp, line 179
>>>
>>> Haven't had any luck getting this to fail... any ideas? I can
>>> side-by-side debug it with you next week, if it comes to that...
>>
>>
>> Hm, try llc -mtriple=i686-windows-msvc?
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20141006/ce30d2ee/attachment.html>