[Mingw-users] Trying to understand how to use MinGW/msys to configure and make a GNU package

[Peter J. F. I. <pjf...@ea...>]

I'm not entirely a newbie to using a GNU configure/make process under
linux, and I *thought* I understood how to go about it using
MinGW/msys, but I am failing completely and I do not completely
understand why.  Please help cure my obviously considerable ignorance.

I have read a lot of the startup pages on the wiki and the HOWTO
pages, but I found nothing there which speaks to using GNU configure
successfully or what facilities are actually available in MinGW and
which are not.

My goal is to be able to take a GNU project -- e.g. gawk 4.1 -- and
figure out how to run the configure step to create the Makefile(s) and
then use mingw32-make to build a MinGW executable, not an msys
executable.

After downloading and unpacking a GNU tarball or doing a git clone of
the stable branch of a project, I fired up msys bash to run the
configure step (from an original CMD.EXE prompt), but configure does
not find things I think I see as supported in MinGW/msys, like
"dlfcn.h" for the dynamic linking facility.

Well, at least I think "dlfcn.h" is supported.  I saw that the
MinGW/msys version of "dlfcn.h" exists only in the msys include
directory, not in the MinGW include directory.  Does that mean that
"dlopen" functions are only supported for msys-based executables?  I
tried exporting CFLAGS and LDFLAGS definitions containing both the
MinGW include and msys include directories before running the
configure script, but "dlfcn.h" was still not found.

I guess I am just not quite getting what GNU facilities are available
for pure MinGW compiles.  Is it only what is contained in the MinGW
include directory?  So, for instance, "stdarg.h" is not available for
MinGW compiles, only for msys compiles, since "stdarg.h" lives only in
msys include?

I see in zip files pulled from ezwinports that the entire GNU
configure process is bypassed and Eli's hand-crafted Makefile's are
substituted instead (e.g., for gawk 4.x).  Is this the only way to
build MinGW versions of GNU projects?  Bypass configure entirely and
hand-craft a Makefile?   Shouldn't the configure process be able to be
modified to (1) determine that it is a MinGW environment and (2)
arrange to do all the things that Eli did to manually create the
Makefiles he supplies?  For instance, since libltdl is provided in
MinGW and "ltdl.h" is provided in MinGW include, couldn't a "dlfcn.h"
substitute be provided that translated "dlopen" calls to the
appropriate "lt_dlopen" call, along with the needed initialization and
setup calls?  Or is that a wrong path?

I am obviously missing either multiple concepts or multiple setup
steps, so let me ask the basic questions:  Should I be able to run a
GNU configure under msys bash and then use mingw32-make to build a GNU
project as a MinGW executable?  If so, is there anywhere you can point
me to where I can learn *exactly* how to do this, step-by-step?

I do understand that at least some (perhaps many?) POSIX facilities
are intentionally not available in MinGW and never will be, which
would prevent at least some GNU projects from being build or executed
without some (perhaps a lot of) coding work to replace those
facilities with Windows equivalents or other changes to the project.
I am more than willing to try to do such work, but I need to
understand how the process should work *if* all needed facilities are
in fact supported by MinGW.  And also how to determine what GNU
facilities are supported and which are not.

TIA for any and all information, advice and RTFM's you can provide.

Regards,

Peter
--