Re: Lots of crashes using new locale tests

[Martin Sebor <se...@roguewave.com>]

Travis Vitek wrote:
> Sorry, somehow hit wrong key and sent that...
> 
> My question is, how should I be handling these issues? File them, ignore
> them for now, or make the test always use shared locales? 

I've already filed a few of them for those the tests that I recently
wrote have uncovered so I suppose you could just continue on the same
path. That said, I'm not really sure what the best process is for
dealing with these types of problems is. We know that tests fail but
we don't yet know that the failures aren't all due to the same set
of library bugs. By filing issues for failing tests without
understanding why they are failing we might be wasting cycles for
what might turn out to be caused by just a couple of  underlying
issues.

If we follow the suggested process of writing [thread safety] tests
first and worrying about their results later, I think it might be
more expedient to defer filing issues until we've done the analysis
of the problems. The problem with this approach is that it introduces
the potential for some (maybe a lot) of overhead if the tests crash
not because of a bug in the library but because of the same problem
in the tests themselves. The strategy of wating to use the tests to
analyze the problems they highlight also has the disadvnatage of all
the tests being less than ideally design for this analysis and needing
to be enhanced or even partially rewritten.

What's your opinion?

Martin

> 
> 
>>-----Original Message-----
>>From: Travis Vitek 
>>Sent: Friday, August 10, 2007 12:25 PM
>>To: Martin Sebor
>>Subject: Lots of crashes using new locale tests
>>
>>
>>Martin,
>>
>>I've compiled my tests on gcc/linux and they fail a lot. Things that
>>I'm seeing...
>>
>> 1. locale creation inside the test threads fails with a 
>>    runtime_exception. It looks like the locale name that
>>    is used gets corrupted during the locale construction.
>>
>> 2. locale creation succeeds, but the correct locale is
>>    not applied when formatting the data to the stream.
>>
>> 3. runtime_exception from the internal methods of the
>>    std::money_put<...> template.
>>
>>Fortunately every one of these errors vanishes when I use the flag
>>--shared-locale. 
> 
>