tree: 1095eafbc59949c649ed452f2679ce8091d9083d [path history] [tgz]
  1. build/
  2. build_overrides/
  3. cast/
  4. discovery/
  5. docs/
  6. infra/
  7. osp/
  8. platform/
  9. test/
  10. testing/
  11. third_party/
  12. tools/
  13. util/
  14. .clang-format
  15. .gitignore
  16. .gn
  19. codereview.settings
  21. DEPS

Open Screen Library

The openscreen library implements the Open Screen Protocol and the Chromecast protocols (both control and streaming).

Information about the protocol and its specification can be found on GitHub.

Getting the code

Installing depot_tools

openscreen library dependencies are managed using gclient, from the depot_tools repo.

To get gclient, run the following command in your terminal:

    git clone

Then add the depot_tools folder to your PATH environment variable.

Note that openscreen does not use other features of depot_tools like repo or drover. However, some git-cl functions do work, like git cl try, git cl lint and git cl upload.

Checking out code

From the parent directory of where you want the openscreen checkout (e.g., ~/my_project_dir), configure gclient and check out openscreen with the following commands:

    cd ~/my_project_dir
    gclient config
    gclient sync

The first gclient command will create a default .gclient file in ~/my_project_dir that describes how to pull down the openscreen repository. The second command creates an openscreen/ subdirectory, downloads the source code, all third-party dependencies, and the toolchain needed to build things; and at their appropriate revisions.

Syncing your local checkout

To update your local checkout from the openscreen master repository, just run

   cd ~/my_project_dir/openscreen
   git pull
   gclient sync

This will rebase any local commits on the remote top-of-tree, and update any dependencies that have changed.

Build setup

The following are the main tools are required for development/builds:

  • Build file generator: gn
  • Code formatter: clang-format
  • Builder: ninja (GitHub releases)
  • Compiler/Linker: clang (installed by default) or gcc (installed by you)

All of these--except gcc as noted above--are automatically downloaded/updated for the Linux and Mac environments via gclient sync as described above. The first two are installed into buildtools/<platform>/.

Mac only: XCode must be installed on the system, to link against its frameworks.

clang-format is used for maintaining consistent coding style, but it is not a complete replacement for adhering to Chromium/Google C++ style (that's on you!). The presubmit script will sanity-check that it has been run on all new/changed code.

Linux clang

On Linux, the build will automatically download the Clang compiler from the Google storage cache, the same way that Chromium does it.

Ensure that libstdc++ 8 is installed, as clang depends on the system instance of it. On Debian flavors, you can run:

   sudo apt-get install libstdc++-8-dev

Linux gcc

Setting the gn argument “is_gcc=true” on Linux enables building using gcc instead.

  gn gen out/Default --args="is_gcc=true"

Note that g++ version 7 or newer must be installed. On Debian flavors you can run:

  sudo apt-get install gcc-7

Mac clang

On Mac OS X, the build will use the clang provided by XCode, which must be installed.

Debug build

Setting the gn argument “is_debug=true” enables debug build.

  gn gen out/Default --args="is_debug=true"

To install debug information for libstdc++ 8 on Debian flavors, you can run:

   sudo apt-get install libstdc++6-8-dbg

gn configuration

Running gn args opens an editor that allows to create a list of arguments passed to every invocation of gn gen.

  gn args out/Default

Building targets

Building the OSP demo

The following commands will build a sample executable and run it.

  mkdir out/Default
  gn gen out/Default          # Creates the build directory and necessary ninja files
  ninja -C out/Default demo   # Builds the executable with ninja
  ./out/Default/demo          # Runs the executable

The -C argument to ninja works just like it does for GNU Make: it specifies the working directory for the build. So the same could be done as follows:

  ./gn gen out/Default
  cd out/Default

After editing a file, only ninja needs to be rerun, not gn. If you have edited a file, ninja will re-run gn for you.

Unless you like to wait longer than necessary for builds to complete, run autoninja instead of ninja, which takes the same command-line arguments. This will automatically parallelize the build for your system, depending on number of processor cores, RAM, etc.

For details on running demo, see its

Building other targets

Running ninja -C out/Default gn_all will build all non-test targets in the repository.

gn ls --type=executable out/Default/ will list all of the executable targets that can be built.

If you want to customize the build further, you can run gn args out/Default to pull up an editor for build flags. gn args --list out/Default prints all of the build flags available.

Building and running unit tests

  ninja -C out/Default unittests

Building and running fuzzers

In order to build fuzzers, you need the GN arg use_libfuzzer=true. It's also recommended to build with is_asan=true to catch additional problems. Building and running then might look like:

  gn gen out/libfuzzer --args="use_libfuzzer=true is_asan=true is_debug=false"
  ninja -C out/libfuzzer some_fuzz_target
  out/libfuzzer/some_fuzz_target <args> <corpus_dir> [additional corpus dirs]

The arguments to the fuzzer binary should be whatever is listed in the GN target description (e.g. -max_len=1500). These arguments may be automatically scraped by Chromium's ClusterFuzz tool when it runs fuzzers, but they are not built into the target. You can also look at the file out/libfuzzer/some_fuzz_target.options for what arguments should be used. The corpus_dir is listed as seed_corpus in the GN definition of the fuzzer target.

Continuous build and try jobs

openscreen uses LUCI builders to monitor the build and test health of the library. Current builders include:

linux64_debugx86-64Ubuntu Linux 16.04clangdebugASAN enabled
linux64_gcc_debugx86-64Ubuntu Linux 18.04gcc-7debug
linux64_tsanx86-64Ubuntu Linux 16.04clangreleaseTSAN enabled
mac_debugx86-64Mac OS X/Xcodeclangdebug
chromium_linux64_debugx86-64Ubuntu Linux 16.04clangdebugbuilt within chromium
chromium_mac_debugx86-64Mac OS X/Xcodeclangdebugbuilt within chromium
linux64_coverage_debugx86-64Ubuntu Linux 16.04clangdebugused for code coverage

You can run a patch through the try job queue (which tests it on all non-chromium builders) using git cl try, or through Gerrit (details below).

The chromium builders compile openscreen HEAD vs. chromium HEAD. They run as experimental trybots and continuous-integration FYI bots.

Submitting changes

openscreen library code should follow the Open Screen Library Style Guide.

openscreen uses Chromium Gerrit for patch management and code review (for better or worse).

The following sections contain some tips about dealing with Gerrit for code reviews, specifically when pushing patches for review, getting patches reviewed, and committing patches.

Uploading a patch for review

The git cl tool handles details of interacting with Gerrit (the Chromium code review tool) and is recommended for pushing patches for review. Once you have committed changes locally, simply run:

  git cl format
  git cl upload

The first command will will auto-format the code changes. Then, the second command runs the script to check style and, if it passes, a newcode review will be posted on

If you make additional commits to your local branch, then running git cl upload again in the same branch will merge those commits into the ongoing review as a new patchset.

It's simplest to create a local git branch for each patch you want reviewed separately. git cl keeps track of review status separately for each local branch.

Addressing merge conflicts

If conflicting commits have been landed in the repository for a patch in review, Gerrit will flag the patch as having a merge conflict. In that case, use the instructions above to rebase your commits on top-of-tree and upload a new patchset with the merge conflicts resolved.


Clicking the CQ DRY RUN button (also, confusingly, labeled COMMIT QUEUE +1) will run the current patchset through all LUCI builders and report the results. It is always a good idea get a green tryjob on a patch before sending it for review to avoid extra back-and-forth.

You can also run git cl try from the commandline to submit a tryjob.

Code reviews

Send your patch to one or more committers in the COMMITTERS file for code review. All patches must receive at least one LGTM by a committer before it can be submitted.


After your patch has received one or more LGTM commit it by clicking the SUBMIT button (or, confusingly, COMMIT QUEUE +2) in Gerrit. This will run your patch through the builders again before committing to the main openscreen repository.

Working with ARM/ARM64/the Raspberry PI

openscreen supports cross compilation for both arm32 and arm64 platforms, by using the gn args parameter target_cpu="arm" or target_cpu="arm64" respectively. Note that quotes are required around the target arch value.

Setting an arm(64) target_cpu causes GN to pull down a sysroot from openscreen's public cloud storage bucket. Google employees may update the sysroots stored by requesting access to the Open Screen pantheon project and uploading a new tar.xz to the openscreen-sysroots bucket.

NOTE: The “arm” image is taken from Chromium's debian arm image, however it has been manually patched to include support for libavcodec and libsdl2. To update this image, the new image must be manually patched to include the necessary header and library dependencies. Note that if the versions of libavcodec and libsdl2 are too out of sync from the copies in the sysroot, compilation will succeed, but you may experience issues decoding content.

To install the last known good version of the libavcodec and libsdl packages on a Raspberry Pi, you can run the following command:

sudo ./cast/standalone_receiver/

NOTE: until Issue 106 is resolved, you may experience issues streaming to a Raspberry Pi if multiple network interfaces (e.g. WiFi + Ethernet) are enabled. The workaround is to disable either the WiFi or ethernet connection.

Code Coverage

Code coverage can be checked using clang‘s source-based coverage tools. You must use the GN argument use_coverage=true. It’s recommended to do this in a separate output directory since the added instrumentation will affect performance and generate an output file every time a binary is run. You can read more about this in clang's documentation but the bare minimum steps are also outlined below. You will also need to download the pre-built clang coverage tools, which are not downloaded by default. The easiest way to do this is to set a custom variable in your .gclient file. Under the “openscreen” solution, add:

  "custom_vars": {
    "checkout_clang_coverage_tools": True,

then run gclient runhooks. You can also run the python command from the clang_coverage_tools hook in //DEPS yourself or even download the tools manually (link).

Once you have your GN directory (we'll call it out/coverage) and have downloaded the tools, do the following to generate an HTML coverage report:

third_party/llvm-build/Release+Asserts/bin/llvm-profdata merge -sparse default.profraw -o foo.profdata
third_party/llvm-build/Release+Asserts/bin/llvm-cov show out/coverage/openscreen_unittests -instr-profile=foo.profdata -format=html -output-dir=<out dir> [filter paths]

There are a few things to note here:

  • default.profraw is generated by running the instrumented code, but foo.profdata can be any path you want.
  • <out dir> should be an empty directory for placing the generated HTML files. You can view the report at <out dir>/index.html.
  • [filter paths] is a list of paths to which you want to limit the coverage report. For example, you may want to limit it to cast/ or even cast/streaming/. If this list is empty, all data will be in the report.

The same process can be used to check the coverage of a fuzzer's corpus. Just add -runs=0 to the fuzzer arguments to make sure it only runs the existing corpus then exits.