[llvm-dev] How to handle DWARF Missing due to function elimination?

[chuanqi.xcq via llvm-dev]

Hi,

Sorry for disturbing. I find that I made a mistake after I revert  e08f205f / D70350. I only recompile the clang instead of the lld. However, I met this problem within ThinLTO : (. Finally, the workaround to revert e08f205f/D70350 works for my case. Thanks a lot!

Thanks,
Chuanqi
------------------------------------------------------------------
From:chuanqi.xcq <yedeng.yd at linux.alibaba.com>
Send Time:2021年5月28日(星期五) 11:23
To:Jeremy Morse <jeremy.morse.llvm at gmail.com>; David Blaikie <dblaikie at gmail.com>
Cc:llvm-dev at lists.llvm.org <llvm-dev at lists.llvm.org>; aprantl <aprantl at apple.com>
Subject:Re: How to handle DWARF Missing due to function elimination?

  Hi,

Although it looks like the same problem within the link (I met this within ThinLTO and bug 48790 met in LTO), it can't help after I revert  e08f205f. (You mentioned 93592b726c7 but I believe it shouldn't be true).
I met this in a complex production code. I would try to give an minimal repro if possible.

Thanks,
Chuanqi


------------------------------------------------------------------
From:David Blaikie <dblaikie at gmail.com>
Send Time:2021年5月28日(星期五) 02:01
To:Jeremy Morse <jeremy.morse.llvm at gmail.com>
Cc:chuanqi.xcq <yedeng.yd at linux.alibaba.com>; llvm-dev at lists.llvm.org <llvm-dev at lists.llvm.org>; aprantl <aprantl at apple.com>
Subject:Re: How to handle DWARF Missing due to function elimination?

Yeah - would be good to know what the examples are - generally that is not expected (even with linker GC or other cases of symbol removal) & so is most likely something to be fixed in LLVM - but yeah, if the debugger is actually /crashing/ that's also not good, and it shouldn't do that (but lesser problem, because it also shouldn't encounter debug info like this, generally).
On Thu, May 27, 2021 at 4:47 AM Jeremy Morse <jeremy.morse.llvm at gmail.com> wrote:
Hi Chuanqi,

 On Thu, May 27, 2021 at 12:11 PM chuanqi.xcq
 <yedeng.yd at linux.alibaba.com> wrote:
 > The gdb would blame when it finds 0x0123456 doesn't refer to any functions in dwarf format.

 This is really interesting -- if you run "llvm-dwarfdump -verify" on
 the output file, does it too complain about an invalid DIE reference?

 There's a known issue with very complicated LTO builds that causes
 something similar:

https://bugs.llvm.org/show_bug.cgi?id=48790

 Reverting 93592b726c7 locally will tell you whether it's that or not.

 --
 Thanks,
 Jeremy

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20210531/fc1afb9c/attachment.html>