In order to simplify the calling conventions for tests that we run on our continuous integration system, we require them to follow a simple API convention. For a given GN label //X:Y and build directory //Z in a checkout (e.g., //url:url_unittests and //out/Release), we expect there to be:
* A file `$Z/$Y.runtime_deps` containing a list of files needed to run the test (in the format produced by `gn desc //Z //X:Y runtime_deps`, which is a newline-separated list of paths relative to Z) * An executable file `$Z/bin/run_$Y` which does everything needed to set up and run the test with all of the appropriate flags. This will usually be a vpython script. * (on Windows) A file `$Z/bin/run_$Y.bat` file that will turn around and invoke the corresponding run_$ vpython script.
If you create a directory snapshot with the files listed in the .runtime_deps file, cd to $Z, and run bin/run_$Y, then the test should run to completion successfully.
The script file MUST honor the GTEST_SHARD_INDEX and GTEST_TOTAL_SHARDS environment variables as documented in the Test Executable API and SHOULD conform to the Test Executable API in other respects (i.e., honor the --isolated-script-test-filter arg and other command line flags specified in that API).
TODO(crbug.com/816629): Convert everything to the Test Executable API, and change the above SHOULD to a MUST.