When we have the systray icon enabled, any window visibility event seems
to have WINDOW_STATE_ICONIFIED bit on. By checking both _ICONIFIED and
_WITHDRAWN the behaviour appears nicer.
The main HUD process can hang due to unhandled exceptions, which
occurred in two separate situations:
1. Table window is closed and HUD instance killed before auto-importer
knows about it
2. Sometimes the threading can jam
These changes attempt to counter the effect of race-conditions. The
dictionary key (table name) is properly tested at the beginning of
update/create block, *but* there are two short round-trips to database
before the key is used. While these occur, the HUD instance can vanish
and thus get its key removed from the dictionary.
Also, when Tables.Table() is created, it will be populated on-demand,
and have child attributes only when such are found from the system. The
new table code pulls in data from actual windows. Again, there is a
query involved and while it runs, the table may have vanished. This
ended up as an error in this call:
foo = gtk.gdk.window_foreign_new(table.number)
The object 'table' is valid (not None) but it has been populated only
after actual table window was killed. Therefore it may not have .number
attribute, which raised an AttributeError. Now the presence of
table.number attribute is tested before the object can be sent to
create_HUD().
The return value is for Glib/GTK event loop, so it should be after
try-finally sequence. The value needs to be returned every time and we
really like to have threads_enter()/threads_leave() to be invoked in
pairs.
Table attributes are pulled from database but the table window itself
may have disappeared. The search parameters no longer match because
there is no window title to match against, so some attributes are not
set at all.
