[llvm] r212203 - Don't try to construct debug LexicalScopes hierarchy for functions that do not have top level debug information.

[David Blaikie]

On Mon, Jul 7, 2014 at 10:07 PM, Duncan P. N. Exon Smith
<dexonsmith at apple.com> wrote:
>
>> On 2014 Jul 7, at 20:22, David Blaikie <dblaikie at gmail.com> wrote:
>>
>> On Thu, Jul 3, 2014 at 6:01 PM, Aditya Nandakumar
>> <aditya_nandakumar at apple.com> wrote:
>>> Thanks Duncan
>>> Bug point reduced it from 27MB to about 3.7. I ran it through opt and
>>> brought it down to 2.4MB and am unable to reduce further. Hopefully this is
>>> not too big.
>>
>> I'll have to take a look, though I'm a bit concerned that a reduced IR
>> is going to be tricky to actually pinpoint the original source of the
>> problem as the IR may already contain the bad debug info metadata at
>> this point... and I won't be able to discover where that was
>> introduced (by which optimization, etc).
>>
>> But I'll take a look, see if I can create something by hand, etc...
>> and see how I go.
>>
>> Thanks,
>> - David
>
> Heh... doesn't sound fun.

One saving grace - I have got code I can enable in the debug info
verifier that will help me pinpoint the optimization pass that
introduces the problem. (though reducing LTO input is a bit of a pain,
but not impossible with a few hints from such investigation)

> FWIW, the source file (27MB) is from pre-LTO
> optimizations -- i.e., it's the merged bitcode at the beginning of LTO.
> If you think that would be more helpful than the bugpoint-reduced
> version, Aditya might still have it around.

Could be handy, if someone wants to put it somewhere I can access it.
(google drive, for example)