Re: [sigc] ANNOUNCE: libsigc++ 1.9.11

[Roel Vanhout <roel riks nl>]

Martin Schulze wrote:
> What is your opinion: would it be a problem to keep the MSVC compiler
> generating these warning for the tests:
> - warning C4267: 'argument' : conversion from 'size_t' to 'unsigned 
> int', possible loss of data
> - warning C4244: 'argument' : conversion from 'const int' to 'const 
> float', possible loss of data

I think it would not be a problem. An option would be to turn the 
warnings off using #pragma directives. I haven't checked if there is a 
preprocessor directive somewhere to detect the compiler? If there is, 
something like this could be done:

#ifdef _MSVC_NET_2003
#pragma warning(disable:4267)
#pragma warning(disable:4244)
#endif

<libsigc code here>

#ifdef _MSVC_NET_2003
#pragma warning(enable:4267)
#pragma warning(enable:4244)
#endif

>   (Can there really be a data loss in a conversion from int to float?

I don't know, not for all I know, though I'd be interested if someone 
could explain to me why there could be.

>    Do you know whether a conversion from int to double also triggers
>    a warning?)

I just checked, it doesn't.

> Could you try to compile something like:
>
>   #include <iostream>
>
>   struct foo {
>     template <class T1>
>     void test(T1 p) { std::cout << p << std::endl; }
>   };
>
>   int main() {
>     foo f;
>     f.test<unsigned int>(3);  // variant 1
>     // f.template test<unsigned int>(3);  // variant 2
>     // f.template void test<unsigned int>(3);  // variant 3
>     // f.void test<unsigned int>(3);  // variant 4
>     return 0;
>   }
>
> and pass me the result?

Variant 1: compiles ok, without a warning.
Variant 2: compiles ok, without a warning.
Variant 3: doesn't compile, error:

c:\test\test\test\test.cpp(12) : error C2059: syntax error : 'template'
c:\test\test\test\test.cpp(12) : error C2951: template declarations are 
only permitted at global or namespace scope

Variant 4: doesn't compile, error:
c:\test\test\test\test.cpp(13) : error C2061: syntax error : identifier 
'test'

> Also of interest would be if MSVC can handle arithmetic in the
> template argument list like we use for the bind adaptor:
>
>   #include <iostream>
>
>   template <int i>
>   struct foo {
>     void test() { std::cout << i << std::endl; }
>   };
>
>   template <int j>
>   void bar() {
>     foo<j+1> f; // variant 1
>     // foo<j?0:1> f; // variant 2
>     // foo<j?0:j+1> f; // variant 3
>     // foo<(j?0:j+1)> f; // variant 4
>     f.test();
>   }
>
>   int main() {
>     bar<0>();
>     bar<1>();
>     return 0;
>   }

Variant 1: compiles ok, no warnings.
Variant 2: compiles ok, no warnings.
Variant 3: compiles ok, no warnings.
Variant 4: compiles ok, no warnings.

> Thanks!

You're welcome. I think it would be really cool if the final release of 
libsigc2 would work with the visual studio compiler, I still use 1.2 for 
now but I can imagine I would want to upgrade in the future.

I've tracked down a few other errors. In some places (like in 
test_bind_return.cc) there is an error in a construct like
  std::cout << str << std::endl;
The error is about no operator being available which takes a right-hand 
argument 'std::string'. The solution is to add
#include <string>
to test_bind_return.cc.

Another problem was in bind_return.h on line 21. It reads

    { functor_.template operator()<typename type_trait<T_arg1>::pass>

Changing that to

    { functor_. operator()<typename type_trait<T_arg1>::pass>

makes it work.

test_bind_return.cc compiles after those two changes.

Another thing I found is on line 516 of slot.h There is a workaround for 
gcc there (at least that's what the comments say) and the msvc compiles 
chokes on that line. Maybe that part should be surrounded with 
compiler-ifdef's.

That's all for now, I don't have time to go any further. Please let me 
know if you want me to do more tests.

cheers,

roel


>   Martin
>
>
> Am 2003.10.29 12:48 schrieb(en) Roel Vanhout:
>
> > Martin Schulze wrote:
> >
> > > Generally, people are wanted to test libsigc++-1.9.11 and
> > > report their experiences on the list, especially those
> > > working with compilers other than gcc-3.2.
> >
> >
> > I cannot help you with the Forte compiler, but I have made project 
> > files for Visual Studio .Net 2003 for libsigc+++-1.9.11 and the test 
> > files. I have made it so that a .lib file is generated and there is a 
> > project file for every test. The directory structure is as follows:
> >
> > libsigc++ root
> >     |- docs
> >     |- examples
> >     |- etc...
> >     |- visualstudio
> >         |- sigc
> >         |    |- sigc.vcproj
> >         |- tests
> >         |    |- test_accumulated
> >         |    |    |- test_accumulated.vcproj
> >         |    |- etc...
> >         |- libsigc.sln
> >
> > This is how I would do it; if someone wants it in a different way, I 
> > can change it.
> >
> > Furthermore, I've tried to build it ofcourse. Most of the tests 
> > failed. You can find a buildlog on 
> > http://www.stack.be/~roel/sigc2-vs2003-buildlog.txt. Basically, the 
> > library went ok, and 5 of the tests (test_accumulated, 
> > test_deduce_result_type, test_disconnect, test_mem_fun and 
> > test_ptr_fun) were ok too.
> > You can have a look at the buildlog and maybe someone can deduce what 
> > the problem is with the rest; I have not enough knowledge of libsigc 
> > to do myself, unfortunately.
> > For the brave soul who wants to try it, I've put a zip file with the 
> > project files on http://www.stack.be/~roel/visualstudio.zip. Copy it 
> > to your libsigc++ root dir, and unzip there - that will create the 
> > directory structure as described above.
> > Hope this helps.
> >
> >
> > cheers,
> >
> > roel


--
Roel Vanhout - Programmer/Systems Administrator
RIKS BV
Papenstraat 8 - 6211 LG Maastricht, The Netherlands
Tel +31 (0)43 388 3708 - Fax +31 (0)43 325 3155