|
|
Boost Testing : |
Subject: Re: [Boost-testing] Darwin/Sandia-intel - looks very problematic
From: Belcourt, Kenneth (kbelco_at_[hidden])
Date: 2011-07-15 19:06:35
On Jul 15, 2011, at 2:51 PM, Artyom Beilis wrote:
>> From: Artyom Beilis <artyomtnk_at_[hidden]>
>
>> To: Running Boost regression tests <boost-testing_at_[hidden]>
>> Sent: Fri, July 15, 2011 11:07:09 PM
>> Subject: Re: [Boost-testing] Darwin/Sandia-intel - looks very problematic
>>> From: "Belcourt, Kenneth" <kbelco_at_[hidden]>
>>>> Same looks to happen to too many libraries.
>>>>
>>>> Also take a look for example on UUID
>>>
>>> I've attached the trace back from the crashing test, hope that helps in
>> some
>>
>>> way. Note that this was with 11.1.091 but same it's the same problem.
>>>
>>
>> Unfortunately I'm not familiar with this particular library.
>
> Also would you be so kind to provide a backtrace of the
> Boost.Locale's test_config test - it is the simplest one
> that does almost and it still fails.
It's attached.
Artyom, now that I see your code, I think I know the problem, though not how to fix it. We've experienced numerous problems with GCC with switch statements in inlined source code. In your case, I see a switch in a virtual function so it may not be inlined but this looks an awful lot like the problems we've had. If the switch statement is replaced with an if else block, I suspect the code may work.
-- Noel