lp:~raof/mir/wayland-input-fixes

Branch merges

Propose for merging

Merged into lp:mir

Mir CI Bot: Needs Fixing (continuous-integration) on 2017-09-26

Alan Griffiths: Approve on 2017-09-26

Diff: 680 lines (+210/-65)

4 files modified

debian/control (+1/-0)
src/protocol/wrapper_generator.cpp (+11/-1)
src/server/frontend/wayland/core_generated_interfaces.h (+60/-12)
src/server/frontend/wayland/wayland_connector.cpp (+138/-52)

api

Related source package recipes

No recipes using this branch. (?)

Create packaging recipe

Related snap packages

Related bugs

Bug #1718677: WlShellSurface instances are leaked

High

Fix Released

Link a bug report

Related blueprints

4278. By Chris Halse Rogers on 2017-09-26

Simplify button mapping iteration

4277. By Chris Halse Rogers on 2017-09-25

Hopefully work around Xenial GCC bugs

4276. By Chris Halse Rogers on 2017-09-25

Wayland: Handle WlPointer and WlTouch lifetimes in the same way as WlKeyboard

4275. By Chris Halse Rogers on 2017-09-25

Wayland: Collect WlKeyboard lifetime handling into a run_unless wrapper

4274. By Chris Halse Rogers on 2017-09-25

Merge mir::client::Event branch

4273. By Chris Halse Rogers on 2017-09-25

Wayland: Add missed locking from new InputCtx::handle_event override

4272. By Chris Halse Rogers on 2017-09-25

Wayland: Actually shrink the inputCtx::listeners vector when removing listeners

4271. By Chris Halse Rogers on 2017-09-25

Wayland: Don't double-free WlSurfaces.

The lifetime handling for wayland::* classes requires that the destructors *only* be called
via the destruction of the associated wl_resource.

4270. By Chris Halse Rogers on 2017-09-25

Wayland: Handle keyboard lifetime, take 2.

Unfortunately, keyboard.release() is not the only way that keyboards can get destroyed.
All client resources will be destroyed on client disconnect, so ~WlKeyboard can't rely
on WlKeyboard::release() having ensured that no events are queued to the keyboard on
the Wayland event loop.

Instead maintain a “destroyed” flag to short-circuit the event handler callbacks once
the WlKeyboard has been destroyed.

4269. By Chris Halse Rogers on 2017-09-25

Wayland: Lifetimes should be tied to client resources only for non-global objects.

Globals are *not* tied to client lifetime - the globals themselves persist as long as the shell
wants them to.

There *are* client resources associated on bind, but these are separate from the global resource.

Instead, properly set up non-global destructors from their constructors.