Skip to content
Snippets Groups Projects
After you've reviewed these contribution guidelines, you'll be all set to contribute to this project.

Contributing to steam-runtime-tools

Issue tracking

Our main bug tracking system for the whole Steam Runtime is https://github.com/ValveSoftware/steam-runtime/issues. Before reporting an issue, please take a look at the bug reporting information to make sure your issue report contains all the information we need.

The issue tracker on our Gitlab installation, gitlab.steamos.cloud, is primarily used by steam-runtime-tools developers to track issues for which we already know the technical details of what is happening.

Contributing code

At the moment our Gitlab installation, gitlab.steamos.cloud, is not set up to receive merge requests from third-party contributors. However, git is a distributed version control system, so it is possible to push a clone of the steam-runtime-tools git repository to some other git hosting platform (such as Github or gitlab.com) and send a link to a proposed branch via an issue report.

If you want to contribute code to steam-runtime-tools, please include a Signed-off-by message in your commits to indicate acceptance of the Developer's Certificate of Origin terms.

Compiling steam-runtime-tools

steam-runtime-tools uses the Meson build system. However, because Steam supports both 64-bit x86_64 (amd64) and 32-bit IA32 (x86, i386), to get a fully functional set of diagnostic tools or a fully functional version of pressure-vessel that will work in all situations, it is necessary to build steam-runtime-tools at least twice: once for 64-bit x86_64, and once for 32-bit IA32. The 32-bit build can disable pressure-vessel, but must include at least the "helpers" that are used to identify and inspect 32-bit libraries. Building for multiple architectures is outside the scope of the Meson build system, so we have to do at least two separate Meson builds and combine them.

Also, to make sure that pressure-vessel and the diagnostic tools work with all the operating systems and runtime environments we support, production builds of steam-runtime-tools need to be compiled in a Steam Runtime 1 'scout' environment.

The script build-aux/many-builds.py can be used to compile and test steam-runtime-tools in a convenient way. Please see build-aux/many-builds.md for details.

Production builds are done by Gitlab-CI, which is also run to test proposed branches. This does not use many-builds.py, because it needs to run separate parts of the build in separate containers, but the general principle is the same.

The diagnostic tools provided by steam-runtime-tools are shipped in Steam Runtime releases via a .deb package. We usually build .deb packages via the deb-build-snapshot tool, with a command-line like this:

LC_ALL=C.UTF-8 \
deb-build-snapshot \
--upstream \
--source \
--download ~/tmp/build-area \
--deb-schroot steamrt_scout_amd64 \
--install-all \
build-vm

where build-vm is a virtual machine with steam-runtime-tools' dependencies and a steamrt_scout_amd64 chroot, that can be accessed via ssh build-vm. If the dependencies and chroot are available on your development system, use localhost, which will make deb-build-snapshot do the build locally instead of using ssh.

Design notes

  • ABIs/architectures are identified by their Debian-style multiarch tuple, which is a GNU tuple (x86_64-linux-gnu, arm-linux-gnueabi) with its CPU part normalized to the oldest/most compatible in a CPU family (so i386, not i686, and arm, not armv5tel). Helper subprocesses are prefixed with the multiarch tuple, in the same style used for cross-compilers and cross-tools (x86_64-linux-gnu-inspect-library, i386-linux-gnu-wflinfo).

  • Be careful when adding dependencies to the main library. GLib 2.32 (Ubuntu 12.04 'precise', SteamRT 1 'scout') can be relied on. JSON-GLib is OK (we have a backport in SteamRT 1 'scout'). Anything else is probably bad news.

    • In particular, try not to pull in C++ ABIs, because incompatible C++ ABIs are one of the things that steam-runtime-tools should be able to detect and diagnose. We use C ABIs because they are the lowest common denominator for compatibility between distros.
  • Any checks that can reasonably crash should be in a subprocess. This means that if they crash, they don't bring down the entire Steam client with them.

  • Any checks that need to look at more than one ABI (32- and 64-bit) must be in a subprocess, so that the same process can run both.

  • The inspect-library helper should not have non-libc dependencies - not even GLib. Other helpers may have library dependencies if it is useful to do so.

  • For system information and checks, new functionality should be accessed through SrtSystemInfo so that its cache lifetime can be managed. Separate access points like srt_check_library_presence() will be kept until the next ABI break, but don't add new ones.

Automated tests

New code should have test coverage where feasible. It's OK to have some "design for test" in the APIs to facilitate this, such as srt_system_info_set_environ().

If a helper can't be unit-tested (for example because it needs working OpenGL), we'll understand. The code that calls into the helper should still be unit-tested, by using a mock implementation of the helper.

Most automated tests run in two phases:

  • Build-time tests (meson test, ninja test, dh_auto_test): steam-runtime-tools has just been built, and probably isn't installed. The environment variables in tests/meson.build are set. The code under test should look for data, helper subprocesses, etc. relative to those environment variables. We can add more environment variables like GI_TYPELIB_PATH if they become necessary.

  • Installed-tests (autopkgtest, debian/tests): steam-runtime-tools has been built and installed system-wide. The code under test should look for data, helper subprocesses, etc. in their real installed locations. For things that get installed next to the tests, such as mock implementations of helpers and mock ABI data, it should look in the directory containing the executable.

Automated testing for pressure-vessel

Testing a new build of pressure-vessel is relatively complicated, because several things have to be pulled together:

  • We need at least one Steam Linux Runtime container to use for testing, but for full test coverage, we need several containers. These can be downloaded by using pressure-vessel/populate-depot.py, which is a copy of the same script that is used to build official Steam Linux Runtime releases. build-aux/many-builds.py setup automates this, placing a suitable set of containers in _build/containers by default.

  • steam-runtime-launcher-service and pressure-vessel-adverb need to be compiled in a way that will work both on the host system and inside the test containers. The most reliable way to provide this is to build them for Steam Linux Runtime 1 'scout'. build-aux/many-builds.py install puts a complete relocatable installation of pressure-vessel in _build/containers/pressure-vessel by default.

  • pressure-vessel-wrap needs to be compiled in a way that is compatible with the host system. In production builds, we use a binary that was built on scout for maximum compatibility, but when debugging a problem with pressure-vessel, it can be useful to build it with a newer compiler to get better warnings, or to build it with debug instrumentation such as AddressSanitizer or code-coverage instrumentation. build-aux/many-builds.py compiles _build/host with AddressSanitizer by default. It also sets up _build/coverage to be built with code coverage instrumentation, but does not build that copy by default.

  • The pressure-vessel integration test needs to be told where to find the test containers and the relocatable build of pressure-vessel, using the test_containers_dir Meson option. build-aux/many-builds.py does this for the _build/host build by default.

After putting all that together, we can run tests/pressure-vessel/containers.py as an ordinary automated test. build-aux/many-builds.py test will do that for you.

Manual testing for pressure-vessel

To reproduce issues involving pressure-vessel, it's often necessary to run a real game in Steam.

build-aux/many-builds.py install puts a complete relocatable installation of pressure-vessel in _build/containers/pressure-vessel by default. This can be copied to a Steam installation for manual testing, by deleting steamapps/common/SteamLinuxRuntime_soldier/pressure-vessel and replacing it with a copy of _build/containers/pressure-vessel.

If a branch under test has been built on our Gitlab-CI, the artifacts from the relocatable-install:production job contain a complete relocatable installation of pressure-vessel at _build/production/pressure-vessel-bin.tar.gz, and a similar installation with source code included at _build/production/pressure-vessel-bin+src.tar.gz. These can be downloaded from the web interface:

  • pipeline or merge request
  • relocatable-install:production
  • Job artifacts
  • Browse
  • navigate into _build/production
  • click on one of the filenames to view the associated artifact
  • Download
  • unpack the downloaded archive as a replacement for steamapps/common/SteamLinuxRuntime_soldier/pressure-vessel

Developer-oriented information for testing various configurations is available in the guide for game developers.

Convenient choices for a reasonably small free-to-play game on various engines include:

  • Floating Point (game ID 302380)
    • very small/quick to install, works OK on weak hardware
    • UnityPlayer 4.3.4f1
  • Life is Strange, episode 1 (game ID 319630)
    • Unreal Engine 3
  • Team Fortress 2 (game ID 440)
    • Source engine
  • Unturned (game ID 304930)
    • UnityPlayer 5.5.3f1

Release procedure

  • The version number is EPOCH.YYYYMMDD.MICRO where:

    • EPOCH increases on major (incompatible) changes, or if you change the version-numbering scheme
    • YYYYMMDD is today's date if you are doing a feature release, or the date of the version you are branching from if you are applying "stable branch" hotfixes to an older version
    • MICRO starts at 0, and increases if you need to apply "stable branch" hotfixes to an old version (or if you need to do two feature releases on the same day)
  • Look at the Debian package's build log from deb-build-snapshot or Gitlab-CI. If new library ABI has been added, you will see warnings like these:

    dpkg-gensymbols: warning: debian/libsteam-runtime-tools-0-0/DEBIAN/symbols doesn't match completely debian/libsteam-runtime-tools-0-0.symbols
    --- debian/libsteam-runtime-tools-0-0.symbols (libsteam-runtime-tools-0-0_0.20190806.0+14+g7bda756-0~snapshot_amd64)
    +++ dpkg-gensymbolsWIRbYG	2019-08-16 10:34:42.915782703 +0000
    @@ -13,9 +13,20 @@
      srt_library_get_type@Base 0.20190801.0
      srt_library_issues_get_type@Base 0.20190801.0
      srt_library_symbols_format_get_type@Base 0.20190801.0
    + srt_runtime_issues_get_type@Base 0.20190806.0+14+g7bda756-0~snapshot
    + srt_steam_issues_get_type@Base 0.20190806.0+14+g7bda756-0~snapshot
      srt_system_info_can_run@Base 0.20190801.0

    Update debian/libsteam-runtime-tools-0-0.symbols, adding the symbols that you have added with the version number that you plan to release, for example srt_runtime_issues_get_type@Base 0.20190816.0.

  • Update debian/changelog: gbp dch, edit to clarify the changelog if desired, dch -r, set the version number.

  • Update the version number and ABI minor/micro version in meson.build.

  • Commit everything.

  • Add an annotated git tag vVERSION.

  • Do a final release build, for example with:

    deb-build-snapshot -d ~/tmp/build-area --source-only --release localhost
  • Upload the resulting .changes file to your OBS branch to be built. If it succeeds, submit it to the main OBS project to be included in the next scout release. If it fails, fix it and prepare a new release (with the next micro version number).