Re: bug, 0.9.2 race in open_mailbox

[Peter Williams <peterw helixcode com>]

> I have just cleared a point of confusion I had with the people on
> gnome-devel... Calls to gtk_signal_emit called from a thread also need to
> take the lock first. This is probably where we are going wrong.
> 

Unfortunately this isn't good enough... the signaling system uses
unprotected
global variables, so threadsafe signaling is not guaranteed, meaning
that
eventually it will start crashing for you.... :-( I resolved this in 
Evolution/Camel by writing a new object system :-) There's probably a
less
dramatic way to do this though. Basically you want to make sure that
anything
not in the main thread never touches GTK. My suggestion would be to copy
the pertinent data into some kind of 'struct thread_operation_send' that
you
knew wasn't being monkeyed around with.