Some backend stubs are being lost and never closed on exit

[seblund]

When running the GUI, often many engine processes are left running when the application closes. When debugging through the closeAllBackends calls you notice that the BackendDriver.openBackendConnections and QueryHandler.connections do not have as many BackendConnections as the GUI has started (it should be the same as they are only ever closed on exit or backend change).

This is clear if you explicitly keep a list of all BackendConnections started in BackendDriver.tryStartNewBackendConnection, and close them at exit, no processes are left hanging and all is good. While this may be an easy fix, it doesn't solve the underlying issue of BackendConnections disappearing. If they disappear we can't reuse them.

I can't really figure out where in the code the BackendConnections are lost since the connections move around a lot, but i haven't done a lot of debugging.

How to reproduce:

• Either add a debug variable in BackendDriver.tryStartNewBackendConnection or use Task Manager to count the number of started processes/connections.
• Add a breakpoint in setOnCloseRequest
• Run the gui with a debugger attached.
• Open the university example
• Click the run all queries button to start a bunch of connections.
• Close the GUI
• Use debugger to check the number of processes/connections backendDriver and queryHandler have compared to the actual number of processes.

[Nielswps]

This issue was also encountered during the SW5 multi-project (and reported here #111 ), however, the issue never seems to occur on my machine:laughing: I will take a look at it first thing next week. I think you are right, that the underlying issue that should be fixed is the disappearance of the connections.