Consider the following Windows API-level program:
#define NOMINMAX
#include <windows.h>
int main()
{
MessageBox( 0, "Blah blah...", "My Windows app!", MB_SETFOREGROUND );
}
Now let's build it using GNU toolchain (i.e. g++), no special options. Here gnuc
is just a batch file that I use for that. It only supplies options to make g++ more standard:
C:est> gnuc x.cpp
C:est> objdump -x a.exe | findstr /i "^subsystem"
Subsystem 00000003 (Windows CUI)
C:est> _
This means that the linker by default produced a console subsystem executable. The subsystem value in the file header tells Windows what services the program requires. In this case, with console system, that the program requires a console window.
This also causes the command interpreter to wait for the program to complete.
Now let's build it with GUI subsystem, which just means that the program does not require a console window:
C:est> gnuc x.cpp -mwindows
C:est> objdump -x a.exe | findstr /i "^subsystem"
Subsystem 00000002 (Windows GUI)
C:est> _
Hopefully that's OK so far, although the -mwindows
flag is just semi-documented.
Building without that semi-documented flag one would have to more specifically tell the linker which subsystem value one desires, and some Windows API import libraries will then in general have to be specified explicitly:
C:est> gnuc x.cpp -Wl,-subsystem,windows
C:est> objdump -x a.exe | findstr /i "^subsystem"
Subsystem 00000002 (Windows GUI)
C:est> _
That worked fine, with the GNU toolchain.
But what about the Microsoft toolchain, i.e. Visual C++?
Well, building as a console subsystem executable works fine:
C:est> msvc x.cpp user32.lib
x.cpp
C:est> dumpbin /headers x.exe | find /i "subsystem" | find /i "Windows"
3 subsystem (Windows CUI)
C:est> _
However, with Microsoft's toolchain building as GUI subsystem does not work by default:
C:est> msvc x.cpp user32.lib /link /subsystem:windows
x.cpp
LIBCMT.lib(wincrt0.obj) : error LNK2019: unresolved external symbol _WinMain@16 referenced in function ___tmainCRTStartu
p
x.exe : fatal error LNK1120: 1 unresolved externals
C:est> _
Technically this is because Microsoft’s linker is non-standard by default for GUI subsystem. By default, when the subsystem is GUI, then Microsoft's linker uses a runtime library entry point, the function where the machine code execution starts, called winMainCRTStartup
, that calls Microsoft's non-standard WinMain
instead of standard main
.
No big deal to fix that, though.
All you have to do is to tell Microsoft's linker which entry point to use, namely mainCRTStartup
, which calls standard main
:
C:est> msvc x.cpp user32.lib /link /subsystem:windows /entry:mainCRTStartup
x.cpp
C:est> dumpbin /headers x.exe | find /i "subsystem" | find /i "Windows"
2 subsystem (Windows GUI)
C:est> _
No problem, but very tedious. And so arcane and hidden that most Windows programmers, who mostly only use Microsoft’s non-standard-by-default tools, do not even know about it, and mistakenly think that a Windows GUI subsystem program “must” have non-standard WinMain
instead of standard main
. In passing, with C++0x Microsoft will have a problem with this, since the compiler must then advertize whether it's free-standing or hosted (when hosted it must support standard main
).
Anyway, that's the reason why g++ can complain about WinMain
missing: it's a silly non-standard startup function that Microsoft's tools require by default for GUI subsystem programs.
But as you can see above, g++ has no problem with standard main
even for a GUI subsystem program.
So what could be the problem?
Well, you are probably missing a main
. And you probably have no (proper) WinMain
either! And then g++, after having searched for main
(no such), and for Microsoft's non-standard WinMain
(no such), reports that the latter is missing.
Testing with an empty source:
C:est> type nul >y.cpp
C:est> gnuc y.cpp -mwindows
c:/program files/mingw/bin/../lib/gcc/mingw32/4.4.1/../../../libmingw32.a(main.o):main.c:(.text+0xd2): undefined referen
ce to `WinMain@16'
collect2: ld returned 1 exit status
C:est> _