Re: [Boost-testing] [QNX6] Status report - comments please

From: Jim Douglas (jim_at_[hidden])
Date: 2005-10-28 00:47:51

Beman Dawes wrote:
> "David Abrahams" <dave_at_[hidden]> wrote in message 
> news:umzkujy0g.fsf_at_boost-consulting.com...
> 
>>Jim Douglas <jim_at_[hidden]> writes:
>>
>>
>>>7. python
>>>The failures are due to a lack of cxxabi.h (in the Dinkum library) and
>>>the substitute code does not work either. More work...
>>
>>Then shouldn't we just detect QNX and act as though cxxabi.h doesn't
>>exist?  What macro do I check to see if I'm targeting QNX?
> 
> 
> defined(__QNXNTO__)  was used in the filesystem patch Jim submitted.

The situation is a little more complicated than that. Your current tests 
assume that if you are using the gcc compliler >= version 3 then 
cxxabi.h will be present i.e. you are assuming that the GNU compiler and 
library allways come as an item. Not to my party, they don't:-) Indeed, 
if I do use the GNU library and GNU compiler it works (see qcc_gpp) but 
when I break the assumed bond and use the Dinkum library with the GNU 
compiler your current detection scheme falls apart. I wonder what will 
happen when I introduce the Intel compiler?

IMHO the truly portable, long-tern solution would be to define something 
like BOOST_HAS_CXXABI_H which can be set in the appropriate 
boost/config/stdlib/*hpp file(s). That would also help to clean up some 
of the more convoluted tests in type_id.cpp.

I did briefly play around with the substitute code and I believe that 
the compiler choked on the line

namespace cxxabi=:: ;

This compination of compiler and lib does not seem very happy with the 
"::" symbol unless it has a name attached on _both_ sides.

Thanks for you help...
Jim