[tbb-dev] Moving non-mobile platforms over to the rapid release train

[Matthew Finkel]

On Fri, Oct 23, 2020 at 08:05:57PM +0000, Georg Koppen wrote:
> Hello everyone!
> 
> We have been working the last few months on getting mobile Tor Browser
> hooked into the rapid release train, away from the ESR series. It's been
> exciting all around. We are close to releasing a first stable mobile
> release and are getting used to the process of frequently updating
> toolchains, different patch sets, and auditing new code. It's time for
> looking forward and thinking about desktop Tor Browser.
> 
> Getting desktop To Browser onto the rapid release train at the same time
> while maintaining the mobile part + having additional sponsor work on
> our plate is very likely too much for our 2 1/2 current browser devs.
> So, we originally thought about postponing that until we have more
> capacity. But maybe we can be smarter.
> 
> It's not desktop vs. mobile but more that we have 4 different platforms
> we support and the first one, Android, made it onto the rapid release
> train. What if we moved the others platform by platform to test how hard
> it is and just move forward if we are confident we can handle the workload?
> 
> I'd propose we start with Linux on the non-ESR train next and soon. That
> means switching Linux nightlies, or if we are confident, even Linux
> alphas, too, off ESR.

I agree this isn't a very crazy idea. We should have the goal that every
platform we move onto rapid release has a sufficient "test plan" (and is
sufficiently tested) before we begin moving the next platform. We should
finish adding Fenix into our testsuite [0] and we should create a plan
for automatic rebasing and monitoring [1] for failures. I don't want to
transition more platforms until we have a solid foundation. When this
structure is in place, and the additional maintenance cost is low, I
think this plan of moving one-platform-at-a-time is a good idea and I
think we can move Nightly Linux onto FF85 or 86 within the next few
months. (Linux does have the advantage that it already has some test
coverage, so the delay is really due to getting Fenix up-to-speed, but
I'd like to see regular results from running the Firefox tests [2], as
well.)

[0] https://gitlab.torproject.org/tpo/applications/tor-browser/-/issues/33660
[1] https://gitlab.torproject.org/tpo/applications/tor-browser/-/issues/33656
[2] https://gitlab.torproject.org/tpo/applications/tor-browser/-/issues/33791