Bumping the client ABI causes CI failures

[Alan Griffiths]

This is demonstrated by https://code.launchpad.net/~alan-griffiths/mir/experimenting-with-ci2/+merge/226590

What appears to happen is that the GLMark2Test.benchmark_fullscreen_default test hangs because glmark-es2-mir was build with the previous ABI. (AFAICS That really shouldn't be a problem unless there are incompatible changes to the PRC protocol which isn't the case.)

The root cause may be lp:1293944 - but logging separately as we should ensure both problems are solved.

[Daniel van Vugt (vanvugt)]

Not relevant to rtm14 as its not getting any ABI bumps, hopefully...

[Alan Griffiths (alan-griffiths)]

I'm pretty sure this is now fixed as I recently MP's a client ABI bump and that worked. (Although I was talked out of landing the bump.)

[Daniel van Vugt (vanvugt)]

If we're not sure if/when it was fixed, best to not declare the milestone at least.

[Alan Griffiths (alan-griffiths)]

Well the original cause was fixed long ago.

We had a new cause in 0.12.x by having clients to incorrectly linking to libmircommon (which meant they wanted libmircommon3 even when using a libmirclient8 that used libmircommon4).

That was fixed in 0.13.x but while 0.12.x was still in archive it prevented a client ABI landing in the 0.14 series.

[Daniel van Vugt (vanvugt)]

Bug 1465883 also describes a possible root cause (and solutions) for this bug.

[Alan Griffiths (alan-griffiths)]

We stopped bumping client ABI a long time ago, and think we have this covered anyway

[Michał Sawicz (saviq)]

Syncing task from Mir.

[Launchpad Janitor (janitor)]

[Expired for mir (Ubuntu) because there has been no activity for 60 days.]