glib/README.win32

General
=======

For prebuilt binaries (DLLs and EXEs) and developer packages (headers,
import libraries) of GLib, GTK+, GIMP etc for Windows, surf to
http://www.gimp.org/win32/ . They are for "native" Windows meaning
they use the Win32 API and Microsoft C runtime library only, no POSIX
(Unix) emulation layer (like Cygwin). 

To build GLib on Win32, you can use either gcc or the Microsoft
compiler and tools. Both the compiler from MSVC 5.0 and from MSVC 6.0
have been used successfully.

But note that to just *use* GLib on Windows, there is no need to build
it yourself. Prepackaged runtime and developer packages are available
from the webiste above. On Unix, it is quite normal that system admins
build and install libraries like GLib themselves. But on Windows
setting up a correct build environment can be quite a task, especially
if you are used to on Unix just type ./configure; make, and expect
things to work as smoothly on Windows.

The following preprocessor macros can be used for conditional
compilation related to Win32 in GLib-using code:

- G_OS_WIN32 is defined when compiling for native Win32, without
  any POSIX emulation, other than to the extent provided by the
  bundled Microsoft C library (msvcrt.dll).

- G_WITH_CYGWIN is defined if compiling for the Cygwin
  environment. Note that G_OS_WIN32 is *not* defined in that case, as
  Cygwin is supposed to behave like Unix. G_OS_UNIX *is* defined when
  compiling for Cygwin.

- G_PLATFORM_WIN32 is defined when either G_OS_WIN32 or G_WITH_CYGWIN
  is defined.

These macros are defined in glibconfig.h, and are thus (indirectly)
available in all source files that include <glib.h> or GTK+ headers.

Additionally, there are the compiler-specific macros:
- __GNUC__ is defined when using gcc
- _MSC_VER is defined when using the Microsoft compiler

G_OS_WIN32 implies using the Microsoft C runtime MSVCRT.DLL. GLib is
not known to work with the older CRTDLL.DLL runtime, or the static
Microsoft C runtime libraries LIBC.LIB and LIBCMT.LIB. It apparently
does work with the debugging version of MSVCRT.DLL, MSVCRTD.DLL.

Building software that use GLib or GTK+
=======================================

Even building software that just *uses* GLib or GTK+ also require to
have the right compiler set up the right way, so if you intend to use
gcc, follow the relevant instructions below in that case, too.

I use gcc with the -fnative-struct flag, which means that in order to
use the prebuilt DLLs (especially of GTK+), you *must* also use that
flag. (This flag means that the struct layout rules are identical to
those used by MSVC. This is essential if the same DLLs are to be
usable both from gcc- and MSVC-compiled code. This definitely is
something one wants.)

Building GLib
=============

Again, first decide whether you really want to do this.

Before building GLib you must also have the libiconv library, either
from the same website mentioned above, or from it's homepage at
http://clisp.cons.org/~haible/packages-libiconv.html. Libiconv has
makefiles for building with MS Visual C only, but as it is one source
file only, building it "by hand" with gcc isn't hard.

You must also have the "intl" library from GNU tettext 0.10.40 (or
later). Get a prebuilt version from the website mentioned above.

Edit the correct paths to those libraries in build/win32/module.defs
as appropriate.

Where are the makefiles?
========================

If you are building from a CVS snapshot, you will not have any
makefile.mingw or makefile.msc files. You should copy the
corresponding makefile.mingw.in or makefile.msc.in file to that name,
and replace any @...@ strings with the correct value.

This is done automatically when an official GLib source distribution
package is built, so if you get GLib from a source distribution
package, there should be makefile.mingw and makefile.msc files ready
to use (after some editing).

Building GLib with gcc
======================

I use gcc-2.95.3. Version 2.95.2 will most probably also work.

You can either use gcc running on Cygwin, or the "pure" mingw
gcc. Using the latter might work better, or at least did at some
point.

Fetch the latest version of gcc for mingw and the msvcrt runtime, from
www.mingw.org.

Set up your PATH so that the gcc from the bin directory that got
created above is the one that gets used. Even if you run the mingw
gcc, you still want to have Cygwin to run make in.

Then run make -f makefile.mingw. Install the resulting DLLs somewhere
in your PATH. You can either keep the headers and import libraries
where they are, or install them somewhere else. There are no rules in
the makefile.mingws for installing, it is up to you where to put them.

Autoconfiscated build
=====================

It is also possible to use the auto*, ./configure and libtool
mechanism when building with gcc. You should be running Cygwin, or
maybe cross-compiling from real Unix, for the configure script to
work, obviously. (It might also be possible to use "MSYS", but I
haven't checked.) You most probably should have very new auto* and
libtool. Personally, I invoke configure using:

CC='gcc -mpentium -fnative-struct'
  CPPFLAGS='-I/src/libiconv-1.7/include -I/target/include'
  LDFLAGS='-L/src/libiconv-1.7/lib -L/target/lib' ./configure
  --with-libiconv --disable-static --prefix=/target
  --host=i386-pc-mingw32 --enable-maintainer-mode

(on a single line)

But please note that the ./configure mechanism should not blindly be
used to build a GLib to be distributed to other developers because it
produces a compiler-dependent glibconfig.h (and config.h, but that
shouldn't matter, as it isn't seen by GLib-using applications). For
instance, the typedef for gint64 is long long with gcc, but __int64
with MSVC.

Except for this and a few other minor issues, there really shouldn't
be any reason to distribute separate GLib headers and DLLs for gcc and
MSVC users, as both compilers generate code that uses the same C
runtime library. Thus one either has to manually edit glibconfig.h
afterwards, or use the supplied config.h.win32 and
glibconfig.h.win32. These have been produced by running configure
twice, once using gcc and once using MSVC, and merging the resulting
files with diff -D.

There might be other hickups when using auto* and configure to build
with gcc. Lately I have used auto*/configure/libtool exclusively when
building GLib, GTK+, GIMP etc on Win32, and it seems to work well
(with some patches applied to the current CVS libtool...).

The hand-written makefile.{mingw,msc} files, and the stuff in the
"build" subdirectory, produce DLLs and import libraries that match
what Makefile.am and libtool produces. For GLib, the DLL is called
libglib-1.3-15.dll (at GLib 1.3.15), and the import libraries
libglib-1.3.dll.a and glib-1.3.lib. Note that the "1.3" is part of the
"basename" of the library, it is not something that libtool have
tucked on. The -15 suffix is the value of "LT_CURRENT - LT_AGE". The
15 is *not* simply the micro version number of GLib, although, for
GLib 1.3.15, it happens to be the same. For the gory details, see
configure.in and libtool documentation.

If you want to run the Cygwin-hosted gcc, and still want to produce
code that does not use Cygwin, but the msvcrt runtime, in theory it
should work to use the -no-cygwin flag, but I haven't tested that
lately.

If you would want to use the Cygwin tools to generate a GLib that
*does* use the Cygwin runtime, the normal Unix configuration method
should work as if on Unix. Note that successfully producing shared
libraries (DLLs) for Cygwin most probably requires you to have a very
new libtool. (And a new libtool probably requires rather new autoconf
and automake.) I haven't personally tested this in a while.

Building with MSVC
==================

If using the Microsoft toolchain, build with `nmake -f
makefile.msc`.

--Tor Lillqvist <tml@iki.fi>