build/docs/build.md

The Pygos Build System

The Pygos build system creates a number of binary packages from a set of source packages using a cross toolchain, installs them to a compressed file system image and neatly packages it with an install script for the target board.

The Pygos system can be built by running the mk.sh shell script in the root of the git tree, with the desired product configuration as argument.

The shell script can be run from anywhere on the file system. All configuration files and scripts are accessed relative to the source location of the script and all generated files are accessed relative to the current working directory.

It is strongly encouraged to run the build system from outside the git tree to have the generated files cleanly separated from the build system.

A second script named check_update.sh is provided to automatically query all upstream package sources to check if newer versions are available.

The mk.sh creates a download and a src directory. In the former it stores downloaded source tar balls, in the later it extracts the tar balls and applies patches.

For all other files and directories, a sub directory named after the product configuration is created, referred to as build root.

Inside the build root the directories log, repo and toolchain are created. The compiled binary packages are stored in repo, the cross toolchain is stored in toolchain. Outputs and diagnostic messages of the build processes are stored in log and are each compressed after successfully building a package.

If you are in a real hurry in building the system, you may wish to store the input git tree and output build directory on an SSD and create the build root directory ahead of time with a tmpfs mounted to it.

Packages and Dependcies

The build system distinguishes between binary packages and source packages.

A binary package is an archive containing files and meta data, such as dependency information. Installing a binary package means extracting its contents (and recursively that of its dependencies) to a target location.

A source package is at its minimum a shell script that is run by the build system to produce binary packages. A source package can produce more than one binary package (e.g. a program, its utility libraries and development headers for the libraries could all be packaged separately).

Running a build script may require development headers and libraries of other packages to be installed to an intermediate staging sysroot used by the cross toolchain. Thus, a source package can itself depend on binary packages that have to be built first and are installed to the staging sysroot before the build process begins. The resulting binary packages can have a completely different set of dependencies (e.g. they don't need the library headers).

For simplicity, the cross toolchain, rootfs image and packaging are also implemented as source packages and the build system takes care of building everything in the right order.

Package Build Scripts

The directory pkg contains a sub directory for each source package. Each package directory is expected to contain a shell script named build.

The build script is expected to set the following variables:

• VERSION containing a package version number.
• URL containing a URL from which to download a source tar ball.
• TARBALL containing the name of the source tar ball. This is appended to the URL to download the package.
• SHA256SUM containing the SHA-256 check sum of the source tar ball.
• SRCDIR containing the name of the source directory unpacked from the tar ball.
• DEPENDS containing a space separated list of packages that have to be built first and installed to the cross toolchains sysroot.
• SUBPKG containing a space sperated list of binary packages produced. If left empty, the build system assumes one binary package with the same name as the source package.

The build script is also expected to implement the following functions:

• prepare is run after unpacking the source tar ball. The current working directory is set to the source directory. The path to the package directory is passed as first argument, so the function can easily access patch files stored in the package directory.
• build is run to compile the package. The current working directory is a temporary directory inside the build root directory. The source directory is passed as first argument.
• deploy is run after compilation to install the build output to a staging directory. Arguments and working directory are the same as for build. The function is expected to generate a *.files and a *.desc file for each sub package, so the build system can automatically package it.
• check_update is only used by the check_update.sh script. It is supposed to find out if the package has a newer version available, and if so, echo it to stdout.

Directory Variables

A number of directories exist that can be accessed through global variables from package build scripts.

The following shell variables are globally visible and identify special directories that build scripts might be interested in:

• SCRIPTDIR points to the git tree containging the build system.
• PKGDOWNLOADDIR points to the directory to which source tar balls are downloaded.
• PKGSRCDIR points to the directory into which source tar balls are unpacked.
• BUILDROOT points to the build root directory.
• PKGLOGDIR points to the directory where log files are written to stored.
• REPODIR points to the directory where binary packages are stored.
• TCDIR points to the cross toolchain directory.

While building a package, additional staging directories are temporarily created inside the build root directory:

• PKGBUILDDIR points to a temporary directory inside the build root that is used as working directory for the build and deploy functions.
• PKGDEPLOYDIR points to another such temporary directory that the deploy function is expected to install binaries to.

Additional Variables

The following variables describe the target system and the build environment:

• PRODUCT contains the product name specified on the command line
• LAYERCONF contains path to the list of active configuration layers for the target product
• TARGET specifies the host triplet of the target board
• OS_NAME is statically set to Pygos
• OS_RELEASE holds a version string generated using git-describe
• NUMJOBS contains the number of processors available for parallel builds
• HOSTTUPLE contains the host triplet of the machine that the build system is running on for compiling toolchain packages.
• CMAKETCFILE contains the absolute path to a CMake toolchain file that can be used for compiling CMake based packages with the cross toolchain.
• PACKAGELIST, DEPENDSLIST, PROVIDESLIST hold data used internally for dependency management.

The cross toolchain directory containing the executable prefixed with $TARGET- is also prepended to PATH.

Utility Functions

Some utility functions are provided for common package build tasks:

• apply_patches can be used inside the prepare function to automatically apply patches stored in the package directory to the source tree.
• strip_files takes a list of files as argument and runs the cross toolchain strip program on those that are valid ELF binaries. If a directory is encountered, the function recursively processes the sub directory. Usually you don't need to use this. The mk.sh script uses this function after the deploy step to process the bin and lib directories.
• unfuck_libtool may have to be used before running make install on packages that build shared libraries with libtool. GNU libtool is an utter piece of garbage from hell. This function removes the global /lib search path from the *.la files, so libtool doesn't crap itself during its stupid relink phase, trying to link against libraries from the host system, after already successfully cross compiling the libraries.
• verson_find_greatest can be used in check_update to find the largest version number from a list. The list of version numbers is read from stdin. Version numbers can have up to four dot separated numbers or characters.
• run_configure can be used to run autoconf generated configure scripts with all the required options set for cross compilation. Extra options can be added to the options passed to configure.

Configuration Files

The configuration for the build system is organized in layers, stored in the layer directory in the git tree.

The configuration on how to build an image for a specific target is a file in the product sub directory that specifies, what configuration layers to use and how to stack them on top of each other. Layers that are further down in the file override the ones before them.

From the layer configuartion, the build system itself merges (in layer precedence order) and processes the following configuration files:

• ROOTFS contains a list of packages that should be built and installed to the root filesystem.
• TOOLCHAIN contains shell variables for the cross compiler toolchain. See below for more detail.
• LDPATH contains a list of directories where the loader should look for dynamic libraries.
• INIT contains shell variables configuring the init system. See below for more detail.

Utility Functions

For working with configuration files, the following utility functions can be used:

• file_path_override takes a file name and looks for the last layer that contains it. The absolute path of the first found file is echoed.
• cat_file_override looks for the last layer that contains a file and prints it to standard output.
• cat_file_merge prints the content of a file to standard output, for every layer that contains the file, in layer precedence order.
• include_override includes a file using the source builtin from the last layer that contains the file.
• include_merge includes a file using the source builtin from every layer that contains the file, in layer precedence order.

Toolchain Configuration

The toolchain configuration file contains a list of shell variables for configuring the cross toolchain packages, as well as some other packages that need to know information about the target system.

Currently, the following variables are used:

• RELEASEPKG contains the name of the release package to build to trigger a build of the entire system. Typically this package depends on the rootfs package, which in turn pulls all configured packages as dependencies. It gets built last and packages the root filesystem image and boot loader files in some device specific way, so they can be installed easily on the target hardware.
• LINUXPKG contains the name of the kernel package. There is a default package called 'linux' that builds a standard, main line LTS kernel. Other packages can be specified for building vendor kernels.
• TARGET specifies the target triplet for the cross toolchain, which is also the host triplet for packages cross compiled with autotools.
• GCC_CPU specifies the target processor for GCC.
• GCC_EXTRACFG extra configure arguments passed to GCC. For instance, this may contain FPU configuration for ARM targets.
• BINUTILS_EXTRACFG extra configure arguments passed to binutils.
• LINUX_TGT contains the space seperated make targets for the generic, main line, LTS kernel package.
• CPU_IS_64BIT is set to yes for 64 bit CPUs. This is needed for some packages like nginx that need a little help for cross compiling.
• TC_HARDENING is set to yes to build user space binaries position independent, with read only relocation, immediate binding and with GCCs stack protector enabled for all functions.

Init System Configuration

The INIT configuration file contains a list of shell variables for configuring the init system.

Currently, the following variables are used:

• GETTY_TTY contains a space separated list of ttys on which to start agetty on system boot.
• HWCLOCK is set to yes if the system has a hardware clock that the time should be synchronized with during system boot and shutdown. If set to anything else, the init system is configured to keep track of time using ntpdate and a file on persistent storage.
• DHCP_PORTS contains a space separated list of network interfaces on which to operate a DHCP client for network auto configuration.
• SERVICES contains a space separated list of raw service names to enable.
• MODULES contains a space seperated list of kernel modules that should be loaded during system boot.

For configuring network interfaces, a file ifrename exists that assigns persistent, predictable names to network interfaces.

The default naming scheme of the Pygos system is to rename the Ethernet interfaces installed on the board to port where X is an index starting with 0.

For each network interface, addresses, mtu, offloading, etc can be configured in a file interfaces/<name>, where name is the interface name after renaming.

If the files nftables.rules or sysctl.conf are found, they are copied to the target system image and the coresponding services are enabled.

For more details, please refer to the not yet existing network documentation.

Package Specific Configuration Files

Additional configuration files may be present that are used by various packages.

The following files are currently used (with default override behavior):

• linux.config contains the kernel build configuration. The same name is currently used by both the main line and the board specific vendor kernels.
• dnsmasq.conf is installed to /etc by the dnsmasq package.
• unbound.conf is installed to /etc by the unbound package.
• dhcpcd.conf is installed to /etc by the dhcpcd package.
• nginx.conf is installed to /etc/nginx by the nginx package.