Interview with Stefano Sabatini, FFmpeg project
FFmpeg - набор свободных библиотек с открытым исходным кодом, которые позволяют записывать, конвертировать и передавать цифровые аудио- и видеозаписи в различных форматах.
Разработчик FFmpeg согласился ответить на мои вопросы о разработке и тестировании в этом проекте.
Please introduce yourself.
My name is Stefano Sabatini, and I am an animal belonging to the homo sapiens species. I have a M.S. in computer science/informatics, and I’ve been working on several areas of FFmpeg (documentation, API design, ffprobe, libavfilter) since 2007.
How does development process of FFmpeg look like?
There are many formal and informal rules which apply to the FFmpeg development process. A contributor usually sends a patch to the mailing-list (email@example.com), then the patch is discussed and is eventually applied or rejected. Some contributors act as maintainers and have the last word on their maintainance area. Also most core developers have commit rights and can push patches on the main git repository. Trivial patches or hot-fixes can be pushed with no review. Michael Niedermayer is the general maintainer, and he has the last word on design decisions.
Discussions also happens informally on the IRC development channel, and sometimes in real-life meetings (typically VDD and Linuxtage), but only a few developers attend these meetings.
How many developers involved in development?
There is a core group of about 10 people working almost daily on the project, and ~ 200 occasional contributors per year. Complete updated statistics can be seen here: https://www.openhub.net/p/ffmpeg
How does testing process look? Do you have automated regular testing?
The contributor is supposed to test a new feature/defect fix/change manually, and to run the complete testing suite on her machine.
We have a testing framework called FATE (FFmpeg Automated Test Environment), see: http://fate.ffmpeg.org/.
What is the process for testing new features or significant changes in FFmpeg?
Features are tested locally. The developers is requested to run the complete FATE test suite before pushing significant features/changes.
There is also a coverity instance which sends updates about found critical issues.
What kind of testing do you use (performance, functional, stability, unit etc)? Do you use some kind of manual testing?
Manual testing is required to test features which are not covered by the FATE tests. FATE performs functional and stability testing, ensuring that the issued output is not erroneously changed. Most tests are related to a single module, so they can be considered as unit tests. Performace testing is not currently automated, although we have performance statistics shown in the tests.
Do you use negative testing techniques (like fuzz testing, fault injection etc)?
Fuzz testing is performed by third parties to detect security issues. AFAIK fault injection is not performed at the moment.
What tools, tests and testing frameworks do you use? Do you use Continuous Integration in development?
We have our custom FATE environment, which is an FFmpeg-custom CI environment. FATE clients update their FFmpeg source code and run the suite of tests, and record regressions and failures which are then notified to the FATE server. FATE also comprises also several valgrind tests.
To detect regressions we compare the reference output hash with the newly computed hash for each single test, for each new build.
Do you measure code coverage? How often do you measure it?
We have this: http://coverage.ffmpeg.org/
I’m not sure about how often the statistics are updated.
What tool do you use for tracking open bugs?
We use trac since 2011: https://trac.ffmpeg.org/
Who is responsible for releasing of new version? What are the release criterias?
Michael Niedermayer (general maintainer) is also the maintainer of all the supported releases. We have no specific criteria, we just cut a new release from the master development repository from time to time at semi-regular intervals, which span from 3 to 6 months. Then critical and security fixes are back-ported to the supported releases.
Typically we only actively maintain a few releases, for details see: https://ffmpeg.org/download.html
Do you make security audit of code? Do you have a process to fix security vulnerability?
There are some third parties contributors (e.g. from Google) who perform security audits and notify in case of security issues. See also: https://ffmpeg.org/security.html
Interviewer’s note: Google told about it in their blog.
Do you use static or dynamic code analysis?
Coverity and valgrind (the latter being integrated in FATE).
Do you involve volunteers to testing? I mean things like ‘crowd sourcing’.
We have no active crowd sourcing campaign at the moment. Most testers are active developers or contributors running/using FFmpeg for their own purposes.
What was the most interesting bug in your practice?
Can’t really say, also I’m more a documentation/API geek so I’m not the most active on interesting bug fixes resolution.
Compatibility bugs are interesting: Ticket #2672
Also integer overflows are subtle and cool:
commit 64be0d1edad630f5bc0f287022f5880de07915b2 Author: Stefano Sabatini <firstname.lastname@example.org> Date: Wed May 18 00:43:25 2011 +0200 id3v2: prevent unsigned integer overflow in ff_id3v2_parse() In ff_id3v2_parse(), prevent unsigned integer overflow if data length indicator is skipped and tlen is < 4. Fix crash decoding file Allaby_cut.mp3, fix trac issue #182.
commit 4a745b41770893116405c22f832192510f9bcc9b Author: Stefano Sabatini <email@example.com> Date: Sun May 8 00:47:14 2011 +0200 pnmdec: add support for mono images with non-space-separated pixel digits When the file to decode contains a sequence of binary values like "1101110...", decode_frame() was reading the sequence of digits like a unique integer value, which was resulting in integer overflows. The change add support for parsing non-space-separated pixel digits for mono formats, in particular fix decoding of file battrace.pbm, and fix trac issue #154.
Thanks for answers!