Re: [Boost-testing] Regressions: Are all this errors real?

From: John Maddock (john_at_[hidden])
Date: 2007-01-15 06:19:07

Roland Schwarz wrote:
> John Maddock wrote:
>> I *think* this is the typid problem you helped me track down, but
>> the tests haven't been run on that machine since 2nd Jan, so I can't
>> be sure (there was a delay in my fixing this - I didn't catch all
>> the typeid usages first time around).
>
> Hmm, I just tried to run the test for borland-5.8.2 too.
> It is failing badly (access violations with manual confirmation dialog
> box). This also appears from regression.py .

Oh :-(

> The results in the status pages gives very surprising results in this
> case: unexpectedly passing!

Double Oh!

> What I found out from trying from within the IDE is, that there are
> some exceptions (which are not seen when run from the command line)
> before the crash. It is triggered from the file cpp_regex_traits.hpp:
>
>       // Borland's STLPort version returns a NULL-terminated
>       // string that has garbage at the end - each call to
>       // std::collate<wchar_t>::transform returns a different string!
>       // So as a workaround, we'll truncate the string at the first
>       NULL // which _seems_ to work....
> #if BOOST_WORKAROUND(__BORLANDC__, < 0x600)
>       result.erase(result.find(charT(0)));
> #else
>
> from result.erase
> The exception thrown is std::out_of_range .

That is indeed a bug, the version check should be for 0x580, I'll change 
that now.

However, I'm not able to reproduce the access violations?

> You have marked this as a linker problem?

This is strange: originally these tests were all failing with an internal 
error in the linker.  I'm not sure why they've started building now, I also 
encountered numerous internal compiler errors when building from the IDE 
(turning off precompiled headers seems to do the trick though).

Can you see if the change to the version check has had any effect? 
Otherwise whick test was failing with the access violation: just the 
threaded test or all of them?

Thanks!

John.

ps if you have a chance can you build the threaded test with VC6 again to 
see if I really have fixed the issues or not?  Cheers!