On Fri, Aug 02, 2002 at 02:34:05PM +0100, Tomas Frydrych wrote: > > We have got a problem in the mechanism that blinks the cursor. The > blinking is done via a timer and if we happen to be in another timer > context, the auto-update gets blocked until the timer-triggered code > finishes execution, and so the cursor does not blink. > > A current example of this is the background spell checker. To avoid > the cursor disappearing while doing the background checking, on > entry into fl_BlockLayout::checkSpelling() we flick the cursor off and > then flick it on again just before we leave. This results in erratic > blinking since the interval of the blink is dictated by the time it takes > to spellcheck the current block, which is considerable for long > blocks and virtually null for short blocks (load a reasonably long > document, say 50 pages, and watch the behaviour of the cursor > immediately after the loading of the document finishes). > > Now, I could improve the code in fl_BlockLayout::checkSpelling() so > that it does not flick the cursor more often than every 500 ms, but > that seems like treating the symptoms, not the problem. It would > seem to me that the proper solution would be to have the cursor- > blinking timer running in a separate thread, so that it does not get > blocked by other timer events in the main thead. This way we could > remove all the cursor-handling stuff from the spellchecking code, > and would not need to worry about the cursor if/when in the future > we implement other timer-driven stuff.
Sounds good. More specifically, the fv_Cursor class should have a lock and a boolean (which is protected by the lock) indicating whether the cursor is on or off. The fv_Cursor class should also have a thread with a timer; its only purpose in life is to set the boolean every 500ms. We don't strictly need an fv_Cursor class, we just need a bool 'n' lock in fv_View, along with a few methods: one for the main thread routine, and one for the timer callback. If it's in another thread, it doesn't even need a timer callback, just a sleep(500); this may be ideal. I would be willing to implement this tomorrow if you don't want to do it, tf. pat
