Created by Kevin DuBois on 2016-08-22 and last modified on 2016-08-22
Get this branch:
bzr branch lp:~kdub/mir/fix-1615703
Only Kevin DuBois can upload to this branch. If you are Kevin DuBois please log in for upload directions.

Branch merges

Related bugs

Related blueprints

Branch information

Kevin DuBois

Recent revisions

3661. By Kevin DuBois on 2016-08-22

back out rev 3650 that loads the fb module prior to the hwc module. mx4/powervr devices seem to segfault when trying to close the fb module. fixes: LP: #1615703

3660. By Kevin DuBois on 2016-08-19

platforms: remove ifdefs around mg::NativeBuffer to make it simply a forward-declared class for server code. Platforms can define what mg::NativeBuffer is to whatever works best. (eg, Android defines it to be an interface that makes it easy to convert to different android types and manage the buffer fences).

Android defines it to be an interface that makes it easy to convert to different android types and manage the buffer fences.

Mesa was hacking around the POD type it was using with GBMNativeBuffer to get the gbm_bo*. GBMNativeBuffer is just mg::NativeBuffer for that platform now.

Eglstream could use ShmFile as its native type (although I didn't make that change in this MP)

Nested will want a mg::NativeBuffer that gives access to its underlying MirBuffer*.

Approved by mir-ci-bot, Chris Halse Rogers, Alexandros Frantzis.

3659. By Chris Halse Rogers on 2016-08-19

mir_acceptance_tests: Check that we don't double-close file descriptors, then fix the places we do.

This adds a close() shim to mir_acceptance_tests, which calls the real close() and then aborts if the close() call failed with EBADF, indicating that we've passed in an already closed (or simply nonsensical) file descriptor.

Then fix the double-closes this exposes.

Approved by mir-ci-bot, Cemil Azizoglu, Alan Griffiths.

3658. By Daniel van Vugt on 2016-08-19

Document early release operation (in case it is enabled) more accurately. Fixes: https://bugs.launchpad.net/bugs/1395421.

Approved by Cemil Azizoglu, mir-ci-bot.

3657. By Daniel van Vugt on 2016-08-18

Tidy up the cow. If it's going to hang around it may as well be less

Approved by Cemil Azizoglu, Alan Griffiths, mir-ci-bot.

3656. By Alexandros Frantzis on 2016-08-18

nested: Fix race between event dispatch and DisplayBuffer destruction. Fixes: https://bugs.launchpad.net/bugs/1612012.

Approved by Cemil Azizoglu, mir-ci-bot, Alan Griffiths, Kevin DuBois.

3655. By Kevin DuBois on 2016-08-18

platform: remove read() and write() from mg::Buffer, placing it in a software-rendering specific backend.

The chief advantage here is that GBMBuffer (and other buffers that don't support linear software rendering) don't have to throw in read() and write(). This provides 'software rendering' support in the same manner that 'gl rendering' is supported.

The name choice "PixelSource" was chosen to mirror "TextureSource" from the gl-rendering system.

Approved by Alexandros Frantzis, Alan Griffiths, mir-ci-bot.

3654. By Chris Halse Rogers on 2016-08-18

Ensure a mt::Signal lives at least as long as any callback runnning on the relevant MainLoop.

This is (a) obviously more correct, and (b) *seems* to resolve a crazy deadlock where MesaDisplayConfigurationTest.reads_updated_subpixel_information blocks indefinitely waiting on a lock to be taken by std::condition_variable::notify_all().

(Probably) Fixes: https://bugs.launchpad.net/mir/+bug/1613352. Fixes: https://bugs.launchpad.net/bugs/1613352.

Approved by mir-ci-bot, Alberto Aguirre, Daniel van Vugt.

3653. By Daniel van Vugt on 2016-08-17

mirvanity: Avoid recording a spuriously low first sample due to thread
start-up races.

Although such a glitch would resolve itself and vanish at 50% test
completion, it's better to never see it at all because people will get

With this change the first sample recorded is much more likely to be
accurate and no longer an outlier that skews the results.

Approved by Alan Griffiths, mir-ci-bot.

3652. By Kevin DuBois on 2016-08-16

tests: split off the android and nested tests into their own suites (like mesa-kms and mesa-x11 have already done). make sure that the shared mtd::StubBuffer code is properly instantiating its NativeBuffer, given build conditions.


Approved by mir-ci-bot, Andreas Pokorny, Cemil Azizoglu, Alan Griffiths.

Branch metadata

Branch format:
Branch format 7
Repository format:
Bazaar repository format 2a (needs bzr 1.16 or later)
Stacked on:
This branch contains Public information 
Everyone can see this information.