Sun, 20 Apr 2008

Cross Compiling for Legacy Win32 Systems (Part 2).

Cross compiling from Linux to Windows requires the installation of a couple of packages. On a Debian or Ubuntu system this can be done using:


  sudo apt-get install build-essential
  sudo apt-get install mingw32 mingw32-binutils mingw32-runtime wine

I'm running Ubuntu's Hardy Heron pre-release and the following is known to work with these versions:


  mingw32               4.2.1.dfsg-1ubuntu1
  mingw32-binutils      2.17.50-20070129.1-1
  mingw32-runtime       3.13-1
  wine                  0.9.59-0ubuntu5

For an example of a project which can be successfully cross-compiled, I have chosen libogg which is one of the two libraries required to encode and decode Ogg/Vorbis files. I also happen to know that the current libogg sources in the Xiph Foundation's SVN repository cross-compile from Linux to Windows correctly because I committed the patch to make it possible.

However, we need to look ahead a little. After we have cross compiled libogg we will also want to cross compile the associated libvorbis library which relies on libogg. We therefore need to configure libogg so that when we install it, it can be found by the libvorbis configure script.

For me that meant creating a MinGW32 directory in my home directory:


  mkdir $HOME/MinGW32

The next step to to grab the libogg source code from the Xiph SVN server. This can be achieved using the command:


  svn co http://svn.xiph.org/trunk/ogg libogg

Changing into the libogg directory, we are now ready to configure, test and install the library. That can be done using:


  ./autogen.sh
  ./configure --host=i586-mingw32msvc --target=i586-mingw32msvc \
      --build=i586-linux --prefix=$HOME/MinGW32
  make
  make check
  make install

The first command above, runs the auto tools to generate that configure script. The second command, configure is broken across two lines. It sets up the generated Makefiles to compile Windows binaries from a Linux host, with the install directory we set up before. The third line builds the windows version of libogg, the fourth line runs the test suite, with the windows executables being run under WINE and the final line installs everything in the MinGW32 directory created earlier.

All of the above commands should pass without errors. If they don't, check your versions of of the mingw cross compiler tools and/or WINE.

Posted at: 20:51 | Category: CodeHacking/MinGWCross | Permalink

Wed, 16 Apr 2008

Cross Compiling for Legacy Win32 Systems (Part 1).

My main two FOSS projects, libsndfile and libsamplerate have significant numbers of users that are tied to that particularly odious legacy system, Microsoft Windows. Since I don't normally use Windows myself, maintaining support for that OS has always been a huge pain in the neck.

Originally I shipped Microsoft project files for libsndfile, but that became unworkable because the different versions of the Microsoft tools (Visual C++ 5, Visual C++ 6, Visual Studio 2003, Visual Studio 2005 etc) used different and incompatible project file formats. I solved this by shipping a simple Makefile that used Microsoft's nmake and the command line compilers to build libsndfile. However, by about 2004, the Microsoft compiler's complete lack of support for the 1999 ISO C Standard made maintaining support too much trouble, so it was dropped.

Instead, I started using Cygwin and MinGW to compile libsndfile on Windows. Both of these tool-sets use a version of the GNU GCC compiler just like Linux and building libsndfile using these two tool-sets was trivial:


  ./configure
  make
  make check

Of course there were howls of protest from Windows users, but since they (with a small number of exceptions) had contributed so little, I didn't fell like I owed them anything. I also started releasing pre-compiled Windows binaries at the same time as the source code tarballs were released.

However, while the MinGW compiler was a huge improvement over the Microsoft one it was still a huge pain in the neck. I had to keep a Windows machine and keep it updated and patched against vulnerabilities. Furthermore, installing and updating MinGW was a painful manual process. Oh how I longed for a Debian/Ubuntu style apt-get command to look for and install updates. Finally, copying source code back and forth between Linux and Windows while debugging Windows issues was another pain point because version control systems like GNU Arch and bzr simply didn't work very well on Windows.

In about 2004, I tried the MinGW Linux to Windows cross compiler, a compiler that runs on Linux but generates binaries for Windows. This compiler worked, but left one rather large problem; how do I run libsndfile's rather large and comprehensive test suite? Compiling libsndfile without running the test suite is a waste of time. I did try to run the tests under WINE (the Windows emulator), but at the time tests were failing under WINE that didn't fail on Windows.

From that time on, I would try running the cross-compiled test suite under WINE once or twice a year. Then, some time in the last year or so, the number of problems with the test suite dropped to one, which was only a FIXME message. A little hacking on the WINE sources resulted in a patch that was sent to the WINE mailing list and has since been applied to the main WINE source tree.

With that bug fixed, I can now cross compile from Linux to Windows and run the full libsndfile test suite under WINE. That means that Windows has just become that little bit less relevant that it was before.

A future post will explain how to set up the cross compiler and WINE and walk through compiling and testing of a standard FOSS project.

Posted at: 23:12 | Category: CodeHacking/MinGWCross | Permalink

Mon, 24 Mar 2008

Cross Compiling with pkg-config.

I'm currently playing with the MinGW cross compiler versions of the GNU C and C++ compilers available via apt-get on Debian and Ubuntu systems. These cross compilers generate windows binaries from a Linux host system which is potentially a much less painful way turning FOSS code into binaries for that particularly odious legacy platform.

Most of the software I'm compiling uses the GNU tools; autoconf, automake, libtool and pkg-config for configuring the software before compiling. Autoconf already has good support for cross compiling and automake and libtool just do what autoconf tells them to do. Pkg-config however is the odd one out.

Pkg-config's job is to retrieve information about installed libraries so that the compiler can find the required header files for inclusion and libraries for linking. For instance, if you wanted compile a program that uses the gconf-2.0 library you could find out the required CFLAGS to be passed to the C compiler and required libraries for linking, by doing something like the following in the Makefile.


  GCONF_CFLAGS = $(shell pkg-config --cflags gconf-2.0)
  GCONF_LIBS = $(shell pkg-config --libs gconf-2.0)

In the above example, when pkg-config is run, it looks in the directory /usr/lib/pkg-config/ and reads information from the file gconf-2.0.pc (each installed library should have one or more of these pkg-config files) which then gets printed out. While the information given by pkg-config would be correct for a native build, it is unlikely to be correct for the cross compiling case.

This issue came up as early as 2003 and there is even a wiki page which suggests some quite extensive changes to pkg-config. Unfortunately I think these suggestions are somewhat fragile and pkg-config itself (I'm using version 0.22) already has features for a better solution.

Like many Unix programs, pkg-config's behaviour can be modified by manipulating certain environment variables. The pkg-config man page explains these variables very well. The first one is PKG_CONFIG_LIBDIR which modifies the default location where pkg-config looks for its per installed library config file. Secondly, the PKG_CONFIG_PATH variable can be set to allow additional pkg-config search paths.

Overriding these two variables results in a MinGW cross pkg-config bash script which I have named i586-mingw32msvc-pkg-config and which looks like this:


  #!/bin/bash

  # This file has no copyright assigned and is placed in the Public Domain.
  # No warranty is given.

  # When using the mingw32msvc cross compiler tools, the native Linux
  # pkg-config executable works fine as long as the default PKG_CONFIG_LIBDIR
  # is overridden.
  export PKG_CONFIG_LIBDIR=/usr/i586-mingw32msvc/lib/pkgconfig

  # Also want to override the standard user defined PKG_CONFIG_PATH with
  # a mingw32msvc specific one.
  export PKG_CONFIG_PATH=$PKG_CONFIG_PATH_MINGW32MSVC

  # Now just execute pkg-config with the given command line args.
  pkg-config $@

Now autoconf generated configure scripts that realise that the i586-mingw32msvc-gcc cross compiler is being used will run the above script and get suitable information for the cross compiler rather than the native compiler.

The only downside to this solution is that a separate script is required for each cross compiler which uses pkg-config. This however is a minor price to pay and it is unlikely that people will end up with huge numbers of XXXX-pkg-config scripts like was common before the widespread use of pkg-config.

Until a better solution becomes available, this is what I will be using.

Posted at: 13:24 | Category: CodeHacking/MinGWCross | Permalink