TSDuck v3.37-3645
MPEG Transport Stream Toolkit
No Matches
Testing TSDuck

Testing overview

TSDuck is highly flexible, allowing an unlimited number of configurations. Testing it is consequently challenging. Moreover, some use cases are difficult to automate or require specific hardware. Testing live TS reception using all possible DVB tuners or Dektec devices requires a lot of material, human and time resources which are far beyond the capabilities of a free open-source project.

From a strict industrial standpoint, TSDuck is consequently not fully tested.

However, on a pragmatic standpoint, test suites have been setup to extract the best of what could be done with limited resources.

Thus, although it is impossible to guarantee that a given release of TSDuck is bug-free, we may be relatively confident that no major regression has been introduced.

Organization of the tests

The code of TSDuck is divided in two parts, a large C++ library (tsduck.dll or libtsduck.so) and a collection of small command line tools and plugins.

Similarly, the tests for TSDuck are divided in two parts.

  • The TSDuck library has its own unitary test suite based on a custom framework named "TSUnit". This test suite is part of the main tsduck repository for TSDuck in directory src/utest.
  • The tools and plugins are less easy to test. They work on large transport stream files which would clutter the tsduck repository. The repository tsduck-test contains those tests and the relevant scripts and data files.

The two test suites are fully automated.

The TSDuck library test suite

In the main TSDuck repository, the directory src/utest contains the source file for one single program named utest. This program is divided in many source files (or test suites). Each source file contains many unitary tests. The test infrastructure is based on a custom framework named "TSUnit".

The utest executable is built twice, once using the TSDuck shared library and once using the static library. On UNIX systems, both versions of the test suite are built and run using make test as illustrated below

$ make -j10 test
[CXX] utestXML.cpp
[LD] /home/tsduck/bin/release-x86_64-vmubuntu/utest
[LD] /home/tsduck/bin/release-x86_64-vmubuntu/utest_static
TSPLUGINS_PATH=/home/tsduck/bin/release-x86_64-vmubuntu LD_LIBRARY_PATH=/home/tsduck/bin/release-x86_64-vmubuntu /home/tsduck/bin/release-x86_64-vmubuntu/utest
OK (619 tests, 28730 assertions)
TSPLUGINS_PATH=/home/tsduck/bin/release-x86_64-vmubuntu LD_LIBRARY_PATH=/home/tsduck/bin/release-x86_64-vmubuntu /home/tsduck/bin/release-x86_64-vmubuntu/utest_static
OK (616 tests, 28707 assertions)

Note that the statically linked version contains slightly less tests. The missing tests are related to plugin activations and they can run only in a shared library environment.

On Windows, the Visual Studio project builds two executables named utests-tsduckdll.exe and utests-tsducklib.exe. They can be run manually or from Vidual Studio:

D:\tsduck> bin\Release-x64\utests-tsduckdll.exe
OK (619 tests, 28747 assertions)
D:\tsduck> bin\Release-x64\utests-tsducklib.exe
OK (616 tests, 28724 assertions)

By default, the test program runs all tests and reports failures only. But utest also accepts a few command options which make it appropriate to debug individual features.

The available command line options are:

  • -d : Debug messages are output on standard error.
  • -l : List all tests but do not execute them.
  • -t name : Run only one test or test suite.

First, the option -l is used to list all available tests:

$ utest -l

The left-most names are test suite names. They represent one source file in src/utest. It is possible to run all tests in a test suite or one specific test.

$ utest -t ArgsTest
$ utest -t XMLTest::testValidation

The option -d is used to produce debug message (see examples in the test source files).

Thus, utest alone can be used as an automated fairly complete non-regression test suite for the TSDuck library or as a debug environment for a given feature under development (in all good "test-driven development" approaches, the code is written at the same time as its unitary test).

The TSDuck tools and plugins test suite

The Git repository tsduck-test contains some tests for the TSDuck tools and plugins. This test suite is currently less exhaustive that the TSDuck library test suite and it should be enriched in the future.

This test suite is fully automated and works on test files only. No specific hardware (DVB tuner or Dektec device), no live stream is tested. This is a known limitation.

The principle is simple. Reference input files are stored in the repository, mainly transport stream files which were once captured from real live sources. Test scripts contain deterministic commands which should always produce the same outputs (data, logs, etc). These reference outputs are also stored in the repository.

The test suite runs using a given version of TSDuck and its outputs are compared with the reference outputs. If differences are detected, there is a potential problem.

Since different versions of TSDuck may produce slightly different outputs, a given version of the test suite formally applies to one version of TSDuck only. Git tags are aligned in both repositories (or should be...) to indicate the target version.

Structure of the test suite

In short, execute the script run-all-tests.sh to run the complete test suite.

The repository contains the following subdirectories:

  • tests : Contains one script per test or set of tests. Each test script can be executed individually. All tests are executed using the script run-all-tests.sh.
  • common : Contains utilities and common script.
  • input : Contains input data files for the tests.
  • reference : Contains reference output files for the various tests.
  • tmp : Contains output files which are created by the execution of the tests. These files are typically compared against reference output files in reference. These files are temporary by definition. The subdirectory tmp is present on test machines only and is excluded from the Git repository.

Adding new tests

To add a new test:

  • Add input files in subdirectory input.
  • Create the script test-NNN.sh in subdirectory tests. Use other existing test scripts as templates.
  • Run the command tests/test-NNN.sh --init. If the test is properly written, this creates the reference output files in subdirectory reference. Manually check the created files, verify that they are correct. Be careful with this step since these files will be used as references.
  • Run the same command without the --init option. This time, the output files are created in tmp and are compared with files in reference. Verify that all tests pass. Errors may appear if the test script is not properly written or if the output files contain unique, non-deterministic, time-dependent, system-dependent or file-system-dependent information. Make sure the output files are totally reproduceable in all environments. At worst, add code in the test script to remove any information from the output files which is known to be non-reproduceable.

Sometimes, TSDuck is modified in such in a way that an output file is modified on purpose. Usually, this starts with a failed test. When analysing the test failure, it appears that the modification of the output is intentional. In that case, re-run the command tests/test-NNN.sh --init to update the reference output files. Do not forget to manually validate them since they will act as the new reference.

Testing a development version

By default, the test suite uses the TSDuck command from the system path. Typically, it will use the installed version.

To test a development version, the two Git repositories tsduck and tsduck-test shall be checked out at the same level, side by side in the same directory. First, TSDuck shall be rebuilt in its repository.

Then, when the option --dev is specified to a test script or to run-all-tests.sh, the test suite automatically uses the TSDuck executables from the development repository.

Large files

The tsduck-test repository contains large files, typically transport stream files.

Initially, these files were not stored inside the regular GitHub repository. Instead, they used the Git Large File Storage (LFS) feature of GitHub. However, using LFS on GitHub happended to be a pain, as experienced by others and explained in this article.

As a consequence, the transport stream files were re-integrated into the Git repository. But we now limit their size to 20 MB.