[llvm-commits] [llvm] r89156 - in /llvm/trunk: include/llvm/Analysis/DebugInfo.h lib/Analysis/DebugInfo.cpp

[Jeffrey Yasskin]

On Fri, Nov 20, 2009 at 9:56 AM, Devang Patel <devang.patel at gmail.com> wrote:
> Hi,
>
> On Fri, Nov 20, 2009 at 9:45 AM, Jeffrey Yasskin <jyasskin at google.com> wrote:
>> On Fri, Nov 20, 2009 at 9:22 AM, Devang Patel <dpatel at apple.com> wrote:
>>>
>>> On Nov 20, 2009, at 8:53 AM, Jeffrey Yasskin wrote:
>>>
>>>> They weren't quite dead: I was calling them. :)
>>>
>>> oops. I didn't realize. Feel free to bring this back for short term while
>>> you update your client.
>>
>> Well, I assume the intrinsics they insert no longer actually do what
>> they used to do, and so I should replace them with something. I just
>> want to know what to replace them with. I figured out that
>> InsertStopPoint turned into
>>        IRBuilder::SetCurrentDebugLocation(
>>            DIFactory::CreateLocation(line_number, col_number,
>>                                              DISubprogram,
>>                                              DILocation(NULL)).getNode());
>> (is DILocation(NULL) really the best way to anchor the DILocation chain?)
>
> We can overload CreateLocation to accept orig location as MDNode * directly.

And then just "NULL" would terminate the chain? I'd suggest defaulting
that argument too. :)

>>
>> But I can't find what to replace InsertSubprogramStart and
>> InsertRegionEnd with. Did they just go away?
>
> Yes. Location info collected during CreateLocation() has scope info.
> The DwarfWriter scans all instruction's scope info and figures out
> where the function (or lexical scope) starts & ends.

Ah, ok, that's easier anyway. I'll change it in Unladen.

Thanks!