Skip to content

How to build Gem on Linux for Windows

Jump to bottom Edit New page
umläute edited this page Jan 21, 2019 · 18 revisions

How to build Gem on Linux for Microsoft Windows using MinGW64

This document describes how to cross-compile Gem on Linux for W32 using MinGW-W64

Requirements

Tool chain

MinGW-W64

Install MinGW-W64 for your distribution.

I'm using Debian (testing/unstable), so I can simply do:

# aptitude install mingw-w64 mingw-w64-tools mingw-w64-i686-dev

This might also install a number of programs you don't need, including FORTRAN and ADA compilers. You can uninstall them if you don't like them (or if you want to save a few MByte).

Pure Data

You need the Pure Data sources to build Gem, either Pd-vanilla or Pd-Extended. The easiest way is to download a zip for Windows from http://puredata.info/downloads. Then unpack it.

I installed Pd-vanilla and unzipped it into ~/lib/W32/ with ~ being my ${HOME}-directory. So the actual pd.exe can be found as ~/lib/W32/pd/bin/pd.exe). I will use this path in the rest of the tutorial.

runtime library

msvcrt.dll from Pd/bin/ conflicts with the one used by MinGW-W64. For now i just disabled the one provided by Pd by renaming it to .../pd/bin/msvcrt.dll_

LATER check about the implications of this. expected side-effects:

  • Pd won't start without this dll.
  • using MinGW-W64's msvcrt.dll might crash Pd.

In any case, if you want to use the pd.exe in this directory, you have to restore this library before running Pd.

In practice I haven't found this a problem, as

  • when building, I use a local copy of Pd extracted to ~/lib/W32/pd/bin/pd.exe on my linux partition.
  • when running Pd/Gem, I use a Pd-installation in C:\Program Files\pd\bin\pd.exe on my W32 partition.

Since the two are not the same, disabling msvcrt.dll in the build-environment, does no harm to the runtime environment.

Gem

Obviously you will need the Gem sources, to compile it. See the other WIKI pages where to find it.

I put the sources into ~/src/puredata/externals/Gem.

cross-compiling

setting the target architecture

Since you are cross-compiling, you must specify the target architecture (the CPU/System/... the resulting binary is supposed to run on), via the --host flag.

Typical target architectures targeting Windows are:

  • i686-w64-mingw32 for a 32bit binaries
  • x86_64-w64-mingw32 for a 64bit binaries

E.g.

$ ./configure --host=i686-w64-mingw32

malloc

When cross-compiling, autotools assumes that malloc is not GNU-compatible, and does some #define trickery, which fails badly on C++. Luckily, there's a trick to force autotools into believing that malloc is fine. Just add ac_cv_func_malloc_0_nonnull=yes to the configure options, like:

$ ./configure --host=i686-w64-mingw32 ac_cv_func_malloc_0_nonnull=yes

third party libraries

Gem has a plugin system which adds lot's of functionalities depending on the installed libraries. You will need to have these libraries installed when building Gem, in order to be able to use them.

Here we briefly discuss the use of some (optional) 3rd-party libraries that change the behaviour of the entire Gem system. Also have a look at the plugins section for adding even more functionality depending on 3rd-party libraries.

FTGL

For text rendering, the FTGL library is required. If you don't need text display within Gem, you can skip this part, and add --without-ftgl to the configure flags.

Download freetype (2.5.3) and FTGL (2.1.3-rc5) and extract them (I extracted them to ~/lib/W32).

Then build them.

The freetype2 build system will tell you (when running configure) that it detected "unix" as your system. While this is not strictly true, just ignore it. I also disabled png support, as ./configure wrongly detected my linux PNG-library.

$ cd ~/lib/W32/
$ ./configure --host=i686-w64-mingw32 \
        --prefix=${HOME}/lib/W32/usr  \
        --without-png
$ make install

After installing freetype2 (into some local directory within ~/lib/W32), we can build FTGL. We have to do a few hacks in order to make FTGL recognize the W32-names of the the OpenGL libraries (which are called libopengl32 resp. libglu32, instead of the libGL and libGLU as found on un*x systems).

I hacked together a replacement for m4/gl.m4, which you can get here. Get the file and copy it into ftgl-2.1.3~rc5/m4/, replacing the existing file. Then run:

$ cd ~/src/3rdparty/ftgl-2.1.3~rc5/
$ ./autogen.sh
$ ./configure --host=i686-w64-mingw32      \
        --prefix=${HOME}/lib/W32/usr       \
		--with-gl-lib="-lglu32 -lopengl32" \
		--with-ft-prefix=${HOME}/lib/W32/usr 
$ make install

Unfortunately libfreetype2 (or is it ftgl??) handles an our --prefix argument badly, so we need to fix some things. In ${HOME}/lib/W32/usr/lib/libftgl.la, you have to replace /usr/local/lib/libfreetype.la in the dependency_libs section with ${HOME}/lib/W32/usr/lib/libfreetype.la (replace ${HOME} with your real home-directory). E.g.

sed -e "s|/usr/local/lib/|${HOME}/lib/W32/usr/lib/|" \
    -i ${HOME}/lib/W32/usr/lib/libftgl.la

Since we will do dynamic linking, we also need to put the dll's we just created into a place where W32 will find them. A good start is, the directory where the Gem-binary will live (the root of the Gem sources):

$ cp ~/lib/W32/usr/bin/libfreetype*.dll ~/src/puredata/externals/Gem/
$ cp ~/lib/W32/usr/bin/libftgl*.dll     ~/src/puredata/externals/Gem/

fribidi

In order to (properly) display text that does not have the default left-to-right flow (e.g. Hebrew and/or Arabic texts), you need the GNU FriBidi library.

Building process

MinGW-W64 allows us to use the autotools.

For the following steps, I will assume that you have changed your working directory to the Gem root folder:

$ cd ~/src/puredata/externals/Gem/

autogen.sh

If you have cloned the git repository, you have to build the building tool yourself:

$ ./autogen.sh

configure

Next we need to run configure in order to detect all installed libraries and setup the build chain:

$ ./configure                      \
      --host=i686-w64-mingw32 ac_cv_func_malloc_0_nonnull=yes \
      --without-ALL                \
      --with-pd=${HOME}/lib/W32/pd \
	  --with-ftgl                  \
	  --with-ftgl-cflags="-I${HOME}/lib/W32/usr/include -I${HOME}/lib/W32/usr/include/freetype2" \
	  --with-ftgl-libs="-L${HOME}/lib/W32/usr/lib -lftgl"  \
      --with-vfw32

The --host option specifies the target architecture (i686-CPU (32bit!), Windows, using the mingw toolchain). The --with-pd option tells configure where to find Pd (headers and libraries). The --without-ALL option disables the use of all (optional) libraries (for now; when cross-compiling, some libraries get wrongly detected to be available). The --with-vfw32 option turns on video-capturing via the olde video-for-windows API. The --with-ftgl options turn on font rendering support (according to how we compiled freetype2/ftgl; see above)

Build

Finally start the build by running:

$ make

This will take a while.

If all went well, you should now have a Gem.dll in your directory, and hopefully a number of Gem-output externals and plugins:

$ ls *.dll
Gem.dll
gem_filmAVI.dll
gem_filmDS.dll
gem_imageSGI.dll
gem_videoVFW.dll
gemw32window.dll

separate sources and build

If you want (like me) to use the same source-tree for multiple configurations (e.g. for building both a linux and a w32 version of Gem), you can easily do so by running configure and make steps in a separate directory.

.../Gem$ mkdir build-w32
.../Gem$ cd build-w32
.../Gem/build-w32$ ./configure                        \
                        --host=i686-w64-mingw32 ac_cv_func_malloc_0_nonnull=yes \
			--without-ALL                \
			--with-pd=${HOME}/lib/W32/pd \
			--with-vfw32
.../Gem/build-w32$ make

Problems

At time of writing, the modelOBJ plugin is known to fail the build. For now, you can ignore this.

In order to complete the built despite this error, use

make -k

Try it

When trying out the so-created binary within Pd, you might get an error about missing dynamic libraries.

To solve this, locate the given library and copy it next to the Gem.dll.

Some libraries I needed to copy:

  • libgcc_s_sjlj-1.dll
  • libstdc++-6.dll
  • libfreetype-6.dll
  • libftgl-2.dll

Good starting places (at least on Debian) are

  • /usr/lib/gcc/i686-w64-mingw32/4.9-win32/
  • ~/lib/W32/usr/bin/

plugins

TODO

videoVLC

VideoLAN's VLC comes with an SDK. (Unfortunately it seems that only the .exe installer includes the SDK, not the .zip package). Install VLC on your W32 partition, and copy the sdk folder (as found in %ProgramFiles%/VideoLAN/VLC/sdk) as ${HOME}/lib/W32/VLC/sdk to your linux partition.

When running Gem's configure, add something like the following to the configure-flags:

    $ ./configure \
       [...] \
       --with-libvlc \
       --with-libvlc-cflags="-I${HOME}/lib/W32/VLC/sdk/include" \
       --with-libvlc-libs="-L${HOME}/lib/W32/VLC/sdk/lib -lvlc" \
    #...

After compilation, copy the following files from %ProgramFiles%/VideoLAN/VLC/ next to your Gem.dll:

  • libvlc.dll
  • libvlccore.dll

QuickTime

As of 2015, the QuickTime for Windows SDK can still be downloaded from apple's website (you need an appleid for that - iirc you need to be registered as a developer).

Install the SDK on your W32 partition, and copy it as ${HOME}/lib/W32/QuickTime to your linux partition.

$ QT_CFLAGS="-I${HOME}/lib/W32/QuickTime/CIncludes"
$ QT_LFLAGS="-I${HOME}/lib/W32/QuickTime/Libraries -lQTMLClient"
$ for i in filmQT imageQT recordQT
do
 pushd plugins/${i}
 make \
    CPPFLAGS="${QT_CFLAGS}"                                                  \
	LIBS="${QT_LFLAGS} $(egrep '^LIBS = ' Makefile | sed -e 's|^LIBS = ||')" \
	pkglib_LTLIBRARIES=gem_${i}.la                                           \
	am_gem_${i}_la_rpath="-rpath '\$(pkglibdir)'"
 popd
done
$

You will also need to install QuickTime for Windows on the W32 machine, in order to use the plugins.

Add a custom footer
Add a custom sidebar
Clone this wiki locally