mingw/windows build fixes

[pramodk]

coreneuron now build as submodule of neuron on windows and runs ring test:

[pramodk]

@nrnhines : everything works if build static library of coreneuron (which is default). With shared library if I build as:

cmake ..  -G "Unix Makefiles" -DCMAKE_INSTALL_PREFIX=`pwd`/install -DCOMPILE_LIBRARY_TYPE=SHARED

Then what produced is following and I get link error while running:

If I build as submodule then everything works because I believe Cmake of neuron has some extra settings that influence how libraries/dlls are generated.

Do you remember what we have done in NEURON to properly generate dlls?

[nrnhines]

I wonder if I'm using my own wrappers for dlopen, etc., implemented in
src/oc/cygwinprt.c. At least I see

$ nm libnrniv.dll.a |grep dlopen
0000000000000000 I __imp_dlopen_noerr
0000000000000000 T dlopen_noerr
0000000000000000 I __imp_dlopen
0000000000000000 T dlopen

and

$ grep DLFCN nrnconf.h
/* #undef HAVE_DLFCN_H */

and it is part of the liboc_la_SOURCES in nrn/src/oc/Makefile.am

[nrnhines]

And also look at how dlopen is declared in src/nrniv/nrnpy.cpp with

#ifdef MINGW
#define RTLD_NOW 0
#define RTLD_GLOBAL 0
#define RTLD_NOLOAD 0
extern "C" {
extern void* dlopen_noerr(const char* name, int mode);
#define dlopen dlopen_noerr
extern void* dlsym(void* handle, const char* name);
extern int dlclose(void* handle);
extern char* dlerror();
}
#else
//#define _GNU_SOURCE
#include <dlfcn.h>
#endif

[pramodk]

@nrnhines : note that we are not using dlopen in coreneuron. The issue is coming from the fact that the lib directory only contains .dll.a files which are helper files but actual dlls are in build/coreneuron directory which are not being copied in bin directory:

[pramodk]

Ok, got it! It's following CMake property:

set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/bin)