[boost] Re: old regression test

From: Daryle Walker (darylew_at_[hidden])
Date: 2002-09-04 13:06:44

On Tuesday, September 3, 2002, at 8:38 AM, bdawes_at_[hidden] wrote:

> At 06:27 AM 9/3/2002, Toon Knapen wrote:
>
>> I suppose the old regression tests do not work anymore ?
>> I get some failiures due to main not being defined (because in the 
>> test
>> program it's 'test_main') and some other symbols from the
>> unit_test_framework.
>>
>> ld: 0711-317 ERROR: Undefined symbol:
>> .boost::test_toolbox::detail::test_and_continue_impl(bool,const
>> boost::test_toolbox::detail::wrapstrstream&,const
>> char*,int,bool,boost::unit_test_framework::report_level)
>> ld: 0711-317 ERROR: Undefined symbol: .main
>>
>>
>> I'm switching to the new regression tests but at the same time I 
>> wanted to
>> do a quick comparison of my upgraded compiler using the old framework.
>
> Toon,
>
> AFAIK, You have three choices. The code in the programs to be tested 
> now
> reads:
>
>    #define BOOST_INCLUDE_MAIN
>    #include <boost/test/test_tools.hpp>
>
> 1) You can change that to use a V1 drop-in replacement:
>
>    #include <boost/detail/lightweight_test.hpp>
>
> 2) You can instead add an #include to suck in all the V2 heavyweight 
> code:
>
>    #include <boost/test/included/test_exec_monitor.hpp>
>
> 3) You can hack <boost/test/test_tools.hpp> to do (1) when
> BOOST_INCLUDE_MAIN is defined.
>
> To me, only (3) makes any sense for existing code like the Boost 
> tests, and
> similar cases where a really lightweight header only solution is
> useful.  Hopefully Gennadiy is finally convinced that (3) best meets 
> the
> requirements for some users, and will make (3) a regular part of
> Boost.Test.
>

I've haven't been following the new Test stuff that much, but your 
words sound like that any code that uses (1) will currently _break_ as 
soon as the new Test stuff is released!  I don't think that is 
acceptable.  Option (3) should be done to fix the problem.

Daryle Walker
Mac, Internet, Video Game Junkie
darylew_at_[hidden]