[boost] Re: [1.32 release] TODO list synchronization

From: Vladimir Prus (ghost_at_[hidden])
Date: 2004-07-14 03:14:45

Aleksey Gurtovoy wrote:

> What about Metrowerks Codewarrior? Do you plan to mark it unusable?
>>From a quick glance, the errors seem to be legitimate.

Heck, this is fresh failure intoduced yesterday (to make the library
compiler on an earlier borland version). I'll see what I can do.

>>> > 2. Finally get some results on gcc2.95-stlport. I expect that the
>>> > library
>>> >    should work there, but due to Boost.Test failure I don't have any
>>> > results.
>>
>> Not done. In fact, most of the program_option tests suddenly fail now on
>> Linux. Moreover, *lots* of other library fail too. BGL is almost
>> red/yellow, and even smart_ptr fails most tests. Seems a configuration
>> problem.
> 
> It was; fixed now. I suppose "not done" means you are still in the
> process?

Now that I finally see the results, I've marked 2.95-stlport unusable. It
ICEs and I have no idea what to look at.

>>> > 4. Check dynamic linking on Windows.
>>
>> There are problems.
>>
>> 1. VC7.1 can't find one symbol
> 
> Same for many Intel tests, it seems. Do you need any help figuring it
> out?

I'd appreciate some advice. One header file
(boost/program_options/detail/value_semantic.hpp) contains:

    extern BOOST_PROGRAM_OPTIONS_DECL std::string arg;

and libs/program_options/src/value_semantic.cpp has:

    BOOST_PROGRAM_OPTIONS_DECL std::string arg("arg");

The definitions of BOOST_PROGRAM_OPTIONS_DECL appear to be right, since all
functions link OK. Further, it links OK on borland.

I have several alternatives to try, including:

- moving the variable out of namespace
- changing the definition of variable to 
   std::string arg("arg");
- changing the definition of variable to 
   BOOST_PROGRAM_OPTIONS_DECL std::string arg;
   std::string arg("arg");
- using function instead of variable

but trying them will take 4 days (given than my machine where Platform SDK
was installed is dead now), so I'd appreciat any help.

>> 2. Intel tests seems to crash at runtime or don't start at all. I'd
>> appreciated if you let me know what exactly happens
> 
> I'll take a look at it after MPL is in the CVS.

Thanks.

- Volodya