Way to AOSC OS Maintainer: Advanced Techniques

Sometimes, Autobuild3 may make wrong assumptions about the build system, and this would probably result a build failure. In other cases, when building projects where multiple build systems are avaliable, it may not select the optimal one (for build time or reliability).

In this case, we can manually specify which build system to use by defining ABTYPE= in the autobuild/defines file.

Currently, these build types are supported:

• self : when a autobuild/build file is provided, uses user created autobuild/build as build script.
• autotools : generally used for GNU autotools-based source trees, with an available configure script in source root, or defined $configure script.
• cmake : used for CMake-based source tree, generates and executes Makefiles, Autobuild3 detects for CMakeList.txt in the source trees.
• cmakeninja : same as above, but generates and executes Ninja build scripts.
• waf : used for Waf-based source tree, Autobuild3 detects for waf file/script in the source trees.
• plainmake : used for source trees with a written Makefile, and therefore is able to be built with make command.
• haskell : used for standard Haskell Cabal/Hackage source trees, comes with a set of scripts that provides Haskell package management functions like registering and unregistering.
• perl : used for standard CPAN source trees.
• python : used for standard PyPI source trees.
• qtproj : used for Qt projects with .pro files in the source trees.
• ruby : used for RubyGems source trees.

Autobuild3 integrates a list of optimal build parameters. However, sometimes these parameters are not entirely compatible with the software and may cause troubles. In this case, have a look at Autobuild3's default parameters, and override them accordingly. A complete list of parameters can be found at the Autobuil3 Wiki.

One problem that stands out is LTO (or Link Time Optimization). This technique can improve run-time efficiency and reduce the size of the binary, but for now enabling LTO may result build failure (the number is constantly decreasing), and consumes a lot of RAM during build-time. Autobuild3 uses LTO by default on AMD64 for its improved efficiency, but if you encounter LTO related issues, you can disable it via adding NOLTO=1 in autobuild/defines.

In some cases, the software uses a special build system (or they don't need a build system at all, like pre-built binaries). In this case, you may take control over the build process by writing build scripts in Bash.

The build script is located in autobuild/build. If this script exists, the build type will be locked to self (unless otherwise overriden if another ABTYPE= was defined), which means Autobuild3 will not try to determine the build system and execute its integrated build script, but simply execute this script.

This script should look very similar to what you would do to manually compile programs. But one key difference is that you should NOT install the compiled program to the system root directory. Instead, it should be installed in $PKGDIR, where later Autobuild3 will make the deb based on the file inside this directory. For example, if the compiled binary is called hugo in the root of the build directory, you should install it to the bin directory of the package by:

1
2

abinfo "Installing Hugo binary..."
install -Dvm755 hugo "$PKGDIR"/usr/bin/hugo

Notice that we used a simple function to print log information to the build log called abinfo(). Just call abinfo "Desired build infomation" in the script, and it will be recorded into the build log. It is considered a good practice to use abinfo() as a way to comment your build scripts, as this could be beneficial for maintainers who may come after.

Sometimes Autobuild3 handles the build process just fine, but the finished product may need some extra tweaks (i.e: wrong directory for man pages, shell completion scripts need to be copied into the $PKGDIR, and so on). In this case, we use the autobuild/beyond script, which, like autobuild/build, is executed as a plain Bash script. It will be executed after the build process.

This is an example taken from TREE/extra-web/aria2. Here, we need to install aria2c's bash_completion file, so we use the autobuild/beyond script.

1
2
3

install -dv "$PKGDIR"/usr/share/bash-completion/completions
install -vm644 "$PKGDIR"/usr/share/doc/aria2/bash_completion/aria2c \
    "$PKGDIR"/usr/share/bash-completion/completions

Sometimes the source code does not contain (or contain an inappropriate version of) some files needed for the package. In this case, we can place files in the autobuild/override directory. Notice that files need to be put in their respective directory (as though they are installed in $PKGDIR.

For example, if we are building a package called foo and it does not contain the .desktop file needed for desktop environments in the source tree, we can just write our own .desktop file and place it in:

1

autobuild/overrides/usr/share/applications/foo.desktop

We've already learned in the Basics that we can patch the source code by simply placing patches inside the autobuild/patches directory. But sometimes the patches has to be applied in a specific order in order to work.

To mitigate this issue, we introduced the autobuild/patches/series~file. This file contains an ordred list of the names of the patches (one filename per line). If this file is present, ~Autobuild3 will apply patches as specified in the list.

In some other cases, the patches will not apply if they are not on a strip level of 1 (one). Here below is an example header from a strip level 1 patch:

1
2

--- a/kernel/init.c
+++ a/kernel/init.c

But sometimes, sources may come in different strip levels, for instance, this patch with a strip level of 3:

1
2

--- dev/working/jelly/kernel/init.c
+++ dev/working/lion/kernel/init.c

In this case, you would need to write your own autobuild/patch, which is also a plain Bash script, call your own patch commands from the script.

First, we have the PISS (Project Information Storage System) ¹, which monitors package updates from upstreams and provides an API that obtains and analyzes these information.

This is not enough, however. Although the VER has been modified, the checksum defined under CHKSUM= is still for the old tarball, and since it does not match with the actual checksum for the new tarball, ACBS will refuse to process the tarball.

There's also ways to automate this process, but there's not a standard script yet. However, at least one maintainer uses the following method:

1
2
3
4
5
6
7

cd TREE/
# First, generate a temporary group.
git --no-pager diff --name-only | grep spec | sed 's/\/spec//' > groups/gnome-changes
# Use addchksum.py to generate a patch of all the checksums
for i in `cat groups/gnome-changes`; do; python3 ../abbs-meta/tools/addchksum.py $i/spec 2> dev/null\ndone > checksums.patch
# Then, apply the patch
patch -Np0 -i checksums.patch

Another (down right dirty) implementation exists, which does not involve creating any temporary files in the repository (this should be merged into our scriptlets repository:

1
2
3
4

#!/bin/sh
for i in `git status | grep modified | grep /spec | awk '{ print $2 }'`; do
    python3 ../abbs-meta/tools/addchksum.py $i
done | patch -Np0 -i -

After this, checksums should be up-to-date.

Then we can try to build the new packages. This should be as simple as:

1
2

cd ciel/ # Enter ciel root directory
ciel build -i INSTANCE groups/gnome-changes

If all packages are built successfully, we can go ahead and commit our changes. Our commit-o-matic will accomplish just that. Similar with findupd, simply download the script, put it into your PATH, invoke the script, and bob's your auntie.

Note that if any extra modification was needed, you must note the said modifications in the git log. That said, before invoking commit-o-matic, you should first remove the modified package from the temporary group, and commit it manually.

1

commit-o-matic.sh groups/gnome-changes

Finally, we can push the built packages to the main repository.

1

pushpkg LDAP_IDENTITY BRANCH

Note that IDENTITY and BRANCH are by definition users and repositories on our Community Repository. Contributors are audited before an LDAP identities are granted by our Infrustructure Work Group - we will get in touch with you via your first PR to our ABBS tree.