Build fails on Windows

[in3otd]

Trying to build Qucs on Windows 7 following the instruction on the Wiki (this is the first time I try), I get an error on qucs-filter :

g++ -DHAVE_CONFIG_H -I. -I..   -I/c/Qt/4.8.6//include -I/c/Qt/4.8.6//include/Qt
-I/c/Qt/4.8.6//include/QtGui -I/c/Qt/4.8.6//include/QtCore -I/c/Qt/4.8.6//includ
e/QtSvg -I/c/Qt/4.8.6//include/QtXml -I/c/Qt/4.8.6//include/QtScript -I/c/Qt/4.8
.6//include/Qt3Support -DQT_DEBUG -DQT3_SUPPORT -DQT_THREAD_SUPPORT -D_REENTRANT
 -mthreads -DQT_DLL -DUNICODE   -g -O2 -pipe -fno-exceptions -W -Wall -std=c++0x
 -O0 -MT eqn_filter.o -MD -MP -MF .deps/eqn_filter.Tpo -c -o eqn_filter.o eqn_fi
lter.cpp
eqn_filter.cpp: In static member function 'static QString* Equation_Filter::crea
teSchematic(tFilter*)':
eqn_filter.cpp:71:16: error: 'M_PI' was not declared in this scope
   Omega *= 2.0*M_PI;   // angular frequency
                ^
make[2]: *** [eqn_filter.o] Error 1
make[2]: Leaving directory `/c/git/qucs/qucs/qucs-filter'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/c/git/qucs/qucs'
make: *** [all] Error 2

did someone else try to build on Windows recently?

[in3otd]

oh, seems to be related to this http://ubuntuforums.org/showthread.php?t=583094 ; if I remove the -std=c++0x it compiles ok... interesting "improvement" in C99 :astonished:

[ra3xdh]

What is MinGW version? We moved project from math.h to <cmath>. See https://github.com/Qucs/qucs/issues/154 for additional information. It builds OK on Linux and gcc-4.8.2. Maybe your MinGW version supports only partially. Unfortunatelly, I have no Windows and no MinGW and I cannot test and fix it. Maybe add #ifdef directive for M_PI for Windows?

[guitorri]

M_PI is not C++ standard. Perhaps MinGW is being pick and sticking to C++. See http://www.cplusplus.com/forum/beginner/105176/ Either a #ifdef or a constant should help, perhaps we can add them to the config.h ?. M_PI_2 and other defines should be fixed as well.

[ra3xdh]

In this case we have no choice. We should put some #ifdef for Q_OS_WIN or something else in config.h.

[in3otd]

yes, it's strange, I used MinGW i686-4.8.2-release-posix-dwarf-rt_v3-rev3.7z , as suggested on the Wiki (@guitorri, do we have a choice or are we tied to it since Qt was compiled with it?) , but on linux with gcc 4.8.2 also here it compiles fine (and also clang 3.3). Note that of course M_PI and similar are used in several other places in the code (and often defined in a local header)

[guitorri]

@in3otd , the fastest way to get Qt is to install the official SDK binaries. To avoid headache it is best to use the compiler they recommend. The Qt installer even check for the (MinGW) compiler. Unless you want to build Qt from source on Windows... I prefer and also suggested the shortest route on the Wiki. I have the impression that MinGW headers are not the same as the Linux GCC.

[guitorri]

The brute force method used in a few places (diagramdialog.h, main.h, qucs-transcalc/units.h qucs-filter/qf_cauer.cpp) is to simply add the needed constants.

#ifndef M_PI
#define M_PI 3.14159265358979323846
#endif

This has the drawback of not being type checked.

Perhaps we can add the constant either to the main.[h,cpp] or config.h

const double M_PI = 3.14159265359;

or

namespace math
{
    const double M_PI = 3.14159265359;
}

When I asked on another issue about using more compile flags, perhaps the -ansi -pedantic -Wall would help us to clean up a bit the rust.

[in3otd]

I didn't have time to look much into this, I googled around and as usual there a lot of different opinions and solutions. I thought to simply add a check in configure.ac so that files needing these constant will just need to include configure.h, but then a similar solution will need to be done for CMake, and I am not familiar with it. Regarding type checking, can it really be a problem with a simple object-like macro like this??

Speaking of compiler flags (maybe I should open another issue for this), what about defining QT3_SUPPORT WARNINGS so we can more systematically remove also the small Qt3-related items ?

[guitorri]

On qucsator we have no alternative. I will use the namespace with cons [type] option to scope the constants and avoid collision with other linked modules, see #178. For consistency I suggest we do the same for the rest of the package. I suggest we use const [type] as needed. The namespaces are not yet in use in qucs-gui.

I did not know about QT3_SUPPORT WARNINGS. Up to know I changed the Qt3 header and followed the compiler error trail. Please enable this warning. It will make us all more aware of what needs to be done.

[guitorri]

I will take a look at this one.

[guitorri]

Should be fixed with merge in 6c1fb1e964c3a9199e7ba433fc27aa8fabb16025