You might not be aware of everything that the run loop does in each iteration. (I wasn't before I researched this answer!) As it happens, CFRunLoop
is part of the open-source CoreFoundation package, so we can take a look at exactly what it entails. The run loop looks roughly like this:
while (true) {
Call kCFRunLoopBeforeTimers observer callbacks;
Call kCFRunLoopBeforeSources observer callbacks;
Perform blocks queued by CFRunLoopPerformBlock;
Call the callback of each version 0 CFRunLoopSource that has been signalled;
if (any version 0 source callbacks were called) {
Perform blocks newly queued by CFRunLoopPerformBlock;
}
if (I didn't drain the main queue on the last iteration
AND the main queue has any blocks waiting)
{
while (main queue has blocks) {
perform the next block on the main queue
}
} else {
Call kCFRunLoopBeforeWaiting observer callbacks;
Wait for a CFRunLoopSource to be signalled
OR for a timer to fire
OR for a block to be added to the main queue;
Call kCFRunLoopAfterWaiting observer callbacks;
if (the event was a timer) {
call CFRunLoopTimer callbacks for timers that should have fired by now
} else if (event was a block arriving on the main queue) {
while (main queue has blocks) {
perform the next block on the main queue
}
} else {
look up the version 1 CFRunLoopSource for the event
if (I found a version 1 source) {
call the source's callback
}
}
}
Perform blocks queued by CFRunLoopPerformBlock;
}
You can see that there are a variety of ways to hook into the run loop. You can create a CFRunLoopObserver
to be called for any of the “activities” you want. You can create a version 0 CFRunLoopSource
and signal it immediately. You can create a connected pair of CFMessagePorts
, wrap one in a version 1 CFRunLoopSource
, and send it a message. You can create a CFRunLoopTimer
. You can queue blocks using either dispatch_get_main_queue
or CFRunLoopPerformBlock
.
You will need to decide which of these APIs to use based on when you are scheduling the block, and when you need it to be called.
For example, touches are handled in a version 1 source, but if you handle the touch by updating the screen, that update isn't actually performed until the Core Animation transaction is committed, which happens in a kCFRunLoopBeforeWaiting
observer.
Now suppose you want to schedule the block while you're handling the touch, but you want it to be executed after the transaction is committed.
You can add your own CFRunLoopObserver
for the kCFRunLoopBeforeWaiting
activity, but this observer might run before or after Core Animation's observer, depending on the order you specify and the order Core Animation specifies. (Core Animation currently specifies an order of 2000000, but that is not documented so it could change.)
To make sure your block runs after Core Animation's observer, even if your observer runs before Core Animation's observer, don't call the block directly in your observer's callback. Instead, use dispatch_async
at that point to add the block to the main queue. Putting the block on the main queue will force the run loop to wake up from its “wait” immediately. It will run any kCFRunLoopAfterWaiting
observers, and then it will drain the main queue, at which time it will run your block.
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…