[Chicken-users] CMakeLists.txt CYGWIN vs. CMAKE_COMPILER_IS_GNUCC

[Brandon J. Van Every]

Sometimes a test is given like

IF(WIN32 AND NOT CYGWIN)

when really what is warranted is

IF(WIN32 AND NOT CMAKE_COMPILER_IS_GNUCC)

MinGW is of course a GCC variant.  It has some of the same capabilities 
as Cygwin, and lacks others.  The Cygwin stuff is specific to 
cygwin-1.dll, not to the compiler itself.  For instance, the MinGW 
compiler seems to handle library names in Unix-y fashion, prepending 
lib* to the output.  So the following is desired:

# since libchicken has to be the name of the dll or .so to load
# On unix lib is the default prefix added to any library.
# For windows we change the name of the library to be libchicken.
IF(WIN32 AND NOT CMAKE_COMPILER_IS_GNUCC)
 SET(CHICKEN_LIB_NAME libchicken)
 SET(CHICKEN_UNSAFE_LIB_NAME libuchicken)
ELSE(WIN32 AND NOT CMAKE_COMPILER_IS_GNUCC)
 SET(CHICKEN_LIB_NAME chicken)
 SET(CHICKEN_UNSAFE_LIB_NAME uchicken)
ENDIF(WIN32 AND NOT CMAKE_COMPILER_IS_GNUCC)

Or at least, this fixes a problem in library output names.  Without it, 
CMake generates liblibchicken.dll, and later stuff fails because 
libchicken.dll can't be found.  I'm not entirely sure if this is due to 
MinGW's behavior, or if CMake is generating stuff erroneously, but the 
above seems reasonable.  Similarly, I will hazard a guess that we really 
want:

IF(WIN32 AND NOT CMAKE_COMPILER_IS_GNUCC)
 ADD_DEFINITIONS(-DC_DEFAULT_TARGET_STACK_SIZE=300000)
 ADD_DEFINITIONS(-DHAVE_LOADLIBRARY)
 ADD_DEFINITIONS(-DHAVE_GETPROCADDRESS)
ENDIF(WIN32 AND NOT CMAKE_COMPILER_IS_GNUCC)

but I am not actually sure of this.  Is this a VC++ specific issue, or a 
non-GCC issue?  If it's VC++ specific, then VC++ should be explicitly 
tested for.  I'm not sure what the best test variable is for this.  The 
CMake documentation is decidedly mediocre as far as documenting the test 
variables.  There's a list of stuff on the CMake wiki, kinda buried amid 
other things, and doesn't seem to be complete either.

Also, the above changes aren't fullly working solutions yet.  They're 
only a piece of the problem: fixing the MinGW behavior is creating a 
make target name collision for chicken.exe and libchicken.dll.  They 
both get a make target of "chicken."  I've got some stuff to straighten 
out; meanwhile, be aware that CYGWIN isn't the only GCC on Windows.


Cheers,
Brandon Van Every
RTFM ISA PITA