| cblack | |
| apol |
| KWin |
Valgrind flags an error on teardown.
EventQueue has a pointer to ConnectionThread internally
Registry has a pointer to the EventQueue internally
teardown order needs to be
Registry
EventQueue
Connection
registry was explicitly deleted before connectionthread already, we just
need to put event queue in the right place.
Ran kwin_wayland nested in valgrind
| Automatic diff as part of commit; lint not applicable. |
| Automatic diff as part of commit; unit tests not applicable. |
| wayland_server.cpp | ||
|---|---|---|
| 608 | Wouldn't it be better to destroy the event queue explicitly in destroyInternalConnection()? It would be a more robust (what if other globals need to access the event queue during tear down too?) and easier solution (a "flat" tear down sequence is much easier to understand than the one where children are destroyed together with their parent implicitly), imho. [1] In fact, we got a bug or two because of "unexpected" (technically, it was okay; but conceptually it wasn't) tear down order of parents and their children. | |
| wayland_server.cpp | ||
|---|---|---|
| 608 | implicitly [1])* | |
Wouldn't it be better to destroy the event queue explicitly in destroyInternalConnection()
I don't think it'd be better or worse.
I went with this as there's a relationship between the registry and the event queue, that we want to explicitly convey. It encapsulates the lifespan logic.
I would probably have gone with that if the event queue was a member var usable by others.