[PATCH] [lld] ELF: Support detection of relocation errors during processing

[Benoit Belley]

On 2015-01-15 13:31, "Philip Reames" <listmail at philipreames.com> wrote:


>
>On 01/15/2015 10:21 AM, Benoit Belley wrote:
>> I completely agree that my concern is not specific to this patch!
>>
>> I would simply add my vote to having a alternative non-exiting way of
>>reporting internal errors whenever possible. This is especially
>>important for users of the MCJIT engine. We are using it in a very large
>>application. Our QA and management does not have much tolerance and
>>understanding for these brute force exitsŠ Thankfully, we haven¹t
>>encountered too many of them.
>If you find one that's not a bug in your frontend, please speak up. In
>practice, I haven't really seen these happen due to missing cases in the
>optimizer or anything like that.  (Well, not that didn't get fixed
>immediately.)

I did speak up! Here are 2 pending code reviews waiting for approval:

[PATCH] MCJIT/ELF now support MSVC C++ mangled symbol
http://reviews.llvm.org/D6855


[PATCH] Bug 18582 - Offset overflow on calling __chkstk and __alloca on x64
http://reviews.llvm.org/D6856


Funny fact : Our management tends to blame LLVM for instabilities when QA
is logging issues for « LLVM ERROR: Š » even when it¹s a bug in our
front-end! It¹s hard to get the story straight sometimes.

>>
>> I feel that don¹t have enough experience yet with the LLVM source code
>>to propose an alternative solution of llvm::report_fatal_report() but I
>>am willing to help out if someone else can lead some of the effort.
>This is going to be a case of "patches welcome" I think.  No current
>contributor has cared enough.
>>
>> Benoit

Ok, I¹ll considerer what I can do if becomes too problematic.

Thanks for the discussion,
Benoit