Thomas Schwinge has written a shell script for building a complete cross-build environment for GNU/Hurd systems.

Find it in the incubator, cross-gnu/master branch.


Read through it. Understand it. Only then use it by following the next steps.


/!\ Please note that these cross toolchains does not yet encompass all of the functionality that native toolchains provide. For example, there is only support for C and C++ so far, but not for other languages. A bunch of fixes / enhancements of glibc are missing. We're working towards minimizing these differences, as well as towards pushing all patches upstream.

Supported Versions of Source Packages

/!\ This is outdated. Contact tschwinge.

The following ones are known to work. Others may work as well, but no guarantee is given. Always the preferred version is listed first.

  • src/binutils

    • CVS binutils-2_20-branch

      $ mkdir binutils-2_20-branch
      $ cd binutils-2_20-branch/
      $ cvs ↩
          co -r binutils-2_20-branch binutils

      The sources are rooted in binutils-2_20-branch/src/. Also use the above commands for updating, instead of the usual cvs update.

    • Release 2.22 or later from should also be fine.

  • src/gcc

  • src/gnumach

    • Git master branch

      $ git clone ↩
          git:// gnumach


      $ ( cd gnumach/ && autoreconf -vi )
  • src/mig

    • Git master branch

      $ git clone ↩
          git:// mig


      $ ( cd mig/ && autoreconf -vi )
  • src/hurd

    • Git master branch

      $ git clone ↩
          git:// hurd
  • src/libpthread

    • Git tschwinge/Peter_Herbolzheimer branch

      $ git clone --no-checkout ↩
          git:// libpthread
      $ cd libpthread/
      $ git checkout origin/tschwinge/Peter_Herbolzheimer


      $ ( cd libpthread/ && autoreconf -vi )
  • src/glibc

    • Git tschwinge/Roger_Whittaker branch

      $ git clone --no-checkout ↩
          git:// glibc
      $ cd glibc/
      $ git checkout origin/tschwinge/Roger_Whittaker


The raw source code trees are about 1 GiB.

Unpack the tarballs if you downloaded any.

Create a directory where the cross build shall be rooted in, and a src subdirectory in there. Then create symbolic links for every of the above packages: from src/PACKAGE to where you stored or unpacked it. If you don't intend to build several cross compilers or use the source trees otherwise, you can also directly store the source trees in src/. The source trees can be shared between multiple cross build trees since the packages' build systems are supposed not to modify the files in the source trees. Not all packages adhere to that, but they fail to do so only for pre-processed documentation, etc.

Either make sure that cross-gnu-env and cross-gnu are found in $PATH (~/bin/, for example) or alternatively remember to use their full paths in the following.

The system you're running the script on (the build system) needs to have basic development tools installed, that is, a C compiler with libraries, make, and several more packages. If anything is missing, the cross-gnu build will abort, and you have to install the missing dependencies and resume the cross-gnu build.

Setting Up the Environment

Do this every time you intend to use the cross compiler:

$ ROOT=to/the/cross/build/root
$ . cross-gnu-env

This will set several environment variables, which are later used by (a) the cross-gnu script and (b) by you, the user of the cross compiler. $TARGET will be set by the script, $PATH will be adjusted, etc. See the cross-gnu-env file for all environment variables that are set, as well as their default values. $ROOT will be made an absolute path if it isn't already.

Later, you'll be able to do things like ../configure --host="$TARGET", and the cross compiler will be found automatically.

Creating the Cross Build Environment

This will need an additional 2 GiB.

After setting up the environment, just run cross-gnu and watch the messages flow by. In the end you should see a message: [...]/cross-gnu: Everything should be in place now.

Staying Up-To-Date

You can re-run cross-gnu to rebuild the parts of the sources that have changed since the last run. This will save a lot of time compared to starting from scratch again. Also, it is especially useful if you aren't working with unpacked tarballs, but on CVS's / SVN's / Git's branches or want to quickly get a new toolchain with patches you applied to the source trees. However: do not use this technique when doing major changes to the source trees, like switching from GCC 4.4 to GCC 4.5.