2023-10-03 11:14:36 +08:00
|
|
|
|
"use strict";
|
|
|
|
|
|
|
|
|
|
// Use the fastest means possible to execute a task in its own turn, with
|
|
|
|
|
// priority over other events including IO, animation, reflow, and redraw
|
|
|
|
|
// events in browsers.
|
|
|
|
|
//
|
|
|
|
|
// An exception thrown by a task will permanently interrupt the processing of
|
|
|
|
|
// subsequent tasks. The higher level `asap` function ensures that if an
|
|
|
|
|
// exception is thrown by a task, that the task queue will continue flushing as
|
|
|
|
|
// soon as possible, but if you use `rawAsap` directly, you are responsible to
|
|
|
|
|
// either ensure that no exceptions are thrown from your task, or to manually
|
|
|
|
|
// call `rawAsap.requestFlush` if an exception is thrown.
|
|
|
|
|
module.exports = rawAsap;
|
|
|
|
|
function rawAsap(task) {
|
|
|
|
|
if (!queue.length) {
|
|
|
|
|
requestFlush();
|
|
|
|
|
flushing = true;
|
|
|
|
|
}
|
|
|
|
|
// Equivalent to push, but avoids a function call.
|
|
|
|
|
queue[queue.length] = task;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
var queue = [];
|
|
|
|
|
// Once a flush has been requested, no further calls to `requestFlush` are
|
|
|
|
|
// necessary until the next `flush` completes.
|
|
|
|
|
var flushing = false;
|
|
|
|
|
// `requestFlush` is an implementation-specific method that attempts to kick
|
|
|
|
|
// off a `flush` event as quickly as possible. `flush` will attempt to exhaust
|
|
|
|
|
// the event queue before yielding to the browser's own event loop.
|
|
|
|
|
var requestFlush;
|
|
|
|
|
// The position of the next task to execute in the task queue. This is
|
|
|
|
|
// preserved between calls to `flush` so that it can be resumed if
|
|
|
|
|
// a task throws an exception.
|
|
|
|
|
var index = 0;
|
|
|
|
|
// If a task schedules additional tasks recursively, the task queue can grow
|
|
|
|
|
// unbounded. To prevent memory exhaustion, the task queue will periodically
|
|
|
|
|
// truncate already-completed tasks.
|
|
|
|
|
var capacity = 1024;
|
|
|
|
|
|
|
|
|
|
// The flush function processes all tasks that have been scheduled with
|
|
|
|
|
// `rawAsap` unless and until one of those tasks throws an exception.
|
|
|
|
|
// If a task throws an exception, `flush` ensures that its state will remain
|
|
|
|
|
// consistent and will resume where it left off when called again.
|
|
|
|
|
// However, `flush` does not make any arrangements to be called again if an
|
|
|
|
|
// exception is thrown.
|
|
|
|
|
function flush() {
|
|
|
|
|
while (index < queue.length) {
|
|
|
|
|
var currentIndex = index;
|
|
|
|
|
// Advance the index before calling the task. This ensures that we will
|
|
|
|
|
// begin flushing on the next task the task throws an error.
|
|
|
|
|
index = index + 1;
|
|
|
|
|
queue[currentIndex].call();
|
|
|
|
|
// Prevent leaking memory for long chains of recursive calls to `asap`.
|
|
|
|
|
// If we call `asap` within tasks scheduled by `asap`, the queue will
|
|
|
|
|
// grow, but to avoid an O(n) walk for every task we execute, we don't
|
|
|
|
|
// shift tasks off the queue after they have been executed.
|
|
|
|
|
// Instead, we periodically shift 1024 tasks off the queue.
|
|
|
|
|
if (index > capacity) {
|
|
|
|
|
// Manually shift all values starting at the index back to the
|
|
|
|
|
// beginning of the queue.
|
|
|
|
|
for (var scan = 0, newLength = queue.length - index; scan < newLength; scan++) {
|
|
|
|
|
queue[scan] = queue[scan + index];
|
|
|
|
|
}
|
|
|
|
|
queue.length -= index;
|
|
|
|
|
index = 0;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
queue.length = 0;
|
|
|
|
|
index = 0;
|
|
|
|
|
flushing = false;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// `requestFlush` is implemented using a strategy based on data collected from
|
|
|
|
|
// every available SauceLabs Selenium web driver worker at time of writing.
|
|
|
|
|
// https://docs.google.com/spreadsheets/d/1mG-5UYGup5qxGdEMWkhP6BWCz053NUb2E1QoUTU16uA/edit#gid=783724593
|
|
|
|
|
|
|
|
|
|
// Safari 6 and 6.1 for desktop, iPad, and iPhone are the only browsers that
|
|
|
|
|
// have WebKitMutationObserver but not un-prefixed MutationObserver.
|
|
|
|
|
// Must use `global` or `self` instead of `window` to work in both frames and web
|
|
|
|
|
// workers. `global` is a provision of Browserify, Mr, Mrs, or Mop.
|
|
|
|
|
|
|
|
|
|
/* globals self */
|
|
|
|
|
var scope = typeof global !== "undefined" ? global : self;
|
|
|
|
|
var BrowserMutationObserver = scope.MutationObserver || scope.WebKitMutationObserver;
|
|
|
|
|
|
|
|
|
|
// MutationObservers are desirable because they have high priority and work
|
|
|
|
|
// reliably everywhere they are implemented.
|
|
|
|
|
// They are implemented in all modern browsers.
|
|
|
|
|
//
|
|
|
|
|
// - Android 4-4.3
|
|
|
|
|
// - Chrome 26-34
|
|
|
|
|
// - Firefox 14-29
|
|
|
|
|
// - Internet Explorer 11
|
|
|
|
|
// - iPad Safari 6-7.1
|
|
|
|
|
// - iPhone Safari 7-7.1
|
|
|
|
|
// - Safari 6-7
|
|
|
|
|
if (typeof BrowserMutationObserver === "function") {
|
|
|
|
|
requestFlush = makeRequestCallFromMutationObserver(flush);
|
|
|
|
|
|
|
|
|
|
// MessageChannels are desirable because they give direct access to the HTML
|
|
|
|
|
// task queue, are implemented in Internet Explorer 10, Safari 5.0-1, and Opera
|
|
|
|
|
// 11-12, and in web workers in many engines.
|
|
|
|
|
// Although message channels yield to any queued rendering and IO tasks, they
|
|
|
|
|
// would be better than imposing the 4ms delay of timers.
|
|
|
|
|
// However, they do not work reliably in Internet Explorer or Safari.
|
|
|
|
|
|
|
|
|
|
// Internet Explorer 10 is the only browser that has setImmediate but does
|
|
|
|
|
// not have MutationObservers.
|
|
|
|
|
// Although setImmediate yields to the browser's renderer, it would be
|
|
|
|
|
// preferrable to falling back to setTimeout since it does not have
|
|
|
|
|
// the minimum 4ms penalty.
|
|
|
|
|
// Unfortunately there appears to be a bug in Internet Explorer 10 Mobile (and
|
|
|
|
|
// Desktop to a lesser extent) that renders both setImmediate and
|
|
|
|
|
// MessageChannel useless for the purposes of ASAP.
|
|
|
|
|
// https://github.com/kriskowal/q/issues/396
|
|
|
|
|
|
|
|
|
|
// Timers are implemented universally.
|
|
|
|
|
// We fall back to timers in workers in most engines, and in foreground
|
|
|
|
|
// contexts in the following browsers.
|
|
|
|
|
// However, note that even this simple case requires nuances to operate in a
|
|
|
|
|
// broad spectrum of browsers.
|
|
|
|
|
//
|
|
|
|
|
// - Firefox 3-13
|
|
|
|
|
// - Internet Explorer 6-9
|
|
|
|
|
// - iPad Safari 4.3
|
|
|
|
|
// - Lynx 2.8.7
|
|
|
|
|
} else {
|
|
|
|
|
requestFlush = makeRequestCallFromTimer(flush);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// `requestFlush` requests that the high priority event queue be flushed as
|
|
|
|
|
// soon as possible.
|
|
|
|
|
// This is useful to prevent an error thrown in a task from stalling the event
|
|
|
|
|
// queue if the exception handled by Node.js’s
|
|
|
|
|
// `process.on("uncaughtException")` or by a domain.
|
|
|
|
|
rawAsap.requestFlush = requestFlush;
|
|
|
|
|
|
|
|
|
|
// To request a high priority event, we induce a mutation observer by toggling
|
|
|
|
|
// the text of a text node between "1" and "-1".
|
|
|
|
|
function makeRequestCallFromMutationObserver(callback) {
|
|
|
|
|
var toggle = 1;
|
|
|
|
|
var observer = new BrowserMutationObserver(callback);
|
|
|
|
|
var node = document.createTextNode("");
|
|
|
|
|
observer.observe(node, {characterData: true});
|
|
|
|
|
return function requestCall() {
|
|
|
|
|
toggle = -toggle;
|
|
|
|
|
node.data = toggle;
|
|
|
|
|
};
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// The message channel technique was discovered by Malte Ubl and was the
|
|
|
|
|
// original foundation for this library.
|
|
|
|
|
// http://www.nonblocking.io/2011/06/windownexttick.html
|
|
|
|
|
|
|
|
|
|
// Safari 6.0.5 (at least) intermittently fails to create message ports on a
|
|
|
|
|
// page's first load. Thankfully, this version of Safari supports
|
|
|
|
|
// MutationObservers, so we don't need to fall back in that case.
|
|
|
|
|
|
|
|
|
|
// function makeRequestCallFromMessageChannel(callback) {
|
|
|
|
|
// var channel = new MessageChannel();
|
|
|
|
|
// channel.port1.onmessage = callback;
|
|
|
|
|
// return function requestCall() {
|
|
|
|
|
// channel.port2.postMessage(0);
|
|
|
|
|
// };
|
|
|
|
|
// }
|
|
|
|
|
|
|
|
|
|
// For reasons explained above, we are also unable to use `setImmediate`
|
|
|
|
|
// under any circumstances.
|
|
|
|
|
// Even if we were, there is another bug in Internet Explorer 10.
|
|
|
|
|
// It is not sufficient to assign `setImmediate` to `requestFlush` because
|
|
|
|
|
// `setImmediate` must be called *by name* and therefore must be wrapped in a
|
|
|
|
|
// closure.
|
|
|
|
|
// Never forget.
|
|
|
|
|
|
|
|
|
|
// function makeRequestCallFromSetImmediate(callback) {
|
|
|
|
|
// return function requestCall() {
|
|
|
|
|
// setImmediate(callback);
|
|
|
|
|
// };
|
|
|
|
|
// }
|
|
|
|
|
|
|
|
|
|
// Safari 6.0 has a problem where timers will get lost while the user is
|
|
|
|
|
// scrolling. This problem does not impact ASAP because Safari 6.0 supports
|
|
|
|
|
// mutation observers, so that implementation is used instead.
|
|
|
|
|
// However, if we ever elect to use timers in Safari, the prevalent work-around
|
|
|
|
|
// is to add a scroll event listener that calls for a flush.
|
|
|
|
|
|
|
|
|
|
// `setTimeout` does not call the passed callback if the delay is less than
|
|
|
|
|
// approximately 7 in web workers in Firefox 8 through 18, and sometimes not
|
|
|
|
|
// even then.
|
|
|
|
|
|
|
|
|
|
function makeRequestCallFromTimer(callback) {
|
|
|
|
|
return function requestCall() {
|
|
|
|
|
// We dispatch a timeout with a specified delay of 0 for engines that
|
|
|
|
|
// can reliably accommodate that request. This will usually be snapped
|
|
|
|
|
// to a 4 milisecond delay, but once we're flushing, there's no delay
|
|
|
|
|
// between events.
|
|
|
|
|
var timeoutHandle = setTimeout(handleTimer, 0);
|
|
|
|
|
// However, since this timer gets frequently dropped in Firefox
|
|
|
|
|
// workers, we enlist an interval handle that will try to fire
|
|
|
|
|
// an event 20 times per second until it succeeds.
|
|
|
|
|
var intervalHandle = setInterval(handleTimer, 50);
|
|
|
|
|
|
|
|
|
|
function handleTimer() {
|
|
|
|
|
// Whichever timer succeeds will cancel both timers and
|
|
|
|
|
// execute the callback.
|
|
|
|
|
clearTimeout(timeoutHandle);
|
|
|
|
|
clearInterval(intervalHandle);
|
|
|
|
|
callback();
|
|
|
|
|
}
|
|
|
|
|
};
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// This is for `asap.js` only.
|
|
|
|
|
// Its name will be periodically randomized to break any code that depends on
|
|
|
|
|
// its existence.
|
|
|
|
|
rawAsap.makeRequestCallFromTimer = makeRequestCallFromTimer;
|
|
|
|
|
|
|
|
|
|
// ASAP was originally a nextTick shim included in Q. This was factored out
|
|
|
|
|
// into this ASAP package. It was later adapted to RSVP which made further
|
|
|
|
|
// amendments. These decisions, particularly to marginalize MessageChannel and
|
|
|
|
|
// to capture the MutationObserver implementation in a closure, were integrated
|
|
|
|
|
// back into ASAP proper.
|
|
|
|
|
// https://github.com/tildeio/rsvp.js/blob/cddf7232546a9cf858524b75cde6f9edf72620a7/lib/rsvp/asap.js
|