Suppose I modify the HTML DOM on line 1. Can I be sure that line 2 of the JavaScript will be working with the DOM modifications enacted by line 1?
This is the only explanation I can come up with some buggy behavior I've been having on a form. The previous line is supposed to update the DOM, but sometimes the DOM is not updates by the time it's not on the next one. Things seem to work fine when I go slower though.
Yes, the Javascript DOM modifications will occur sequentially, unless you are waiting for an asynchronous AJAX call to return. The next instruction will not occur until the first has completed. However please show your code!
Updating a specific property on a DOM element happens right away and should persist on a subsequent read of that property.
If you are relying on that change to propagate across the the DOM, it can be tricky. For example, such as changing the size of an element and expecting the sibling element to report a new offset position as a result - the latter may may not happen until the stack unwinds. I don't actually know the exact rules, but you have to be careful - and it is sometimes browser dependent behavior. And scary yet, sometimes throwing an alert to help debug this makes the elements "realize" their new layout right away. Then you take the alert out and it goes back to buggy behavior.
So if you are positive that a DOM change hasn't had the impact right away, then sometimes the thing to do is to call "setTimeout" with a callback function and a time value of 0. When the timer callback completes, you can complete the subsequent processing. YMMV
Related
I'm debugging a userscript, which is hard to debug, because the bug happens rarely and unpredictably (yet), and the function stack gets executed repeatedly (and the bug happens approximately once every 10-20 times). I cannot pause the devtools, otherwise the page will reload and the bug won't appear (so I cannot use breakpoints). Edit: even if I stop the page from reloading I still have to make user-input repeatedly and fast to cause the bug, so not possible to use breakpoints, because that would stop the bug from appearing.
One of the ways which I can think of is putting console.log in every function and logging variables and then searching through them to find where it went wrong. But it seems very tedious. In the performance section of devtools it's possible to record the process, and see the function calls, but it's impossible to see the variables.
Is there any way to go through the function call stack post-factum (maybe a tool other than devtools), seeing what the variables were equal to in every cycle? (And not only the current call stack, because it gets updated on every user input in my script). Or is console.logging my only option in this case?
I thought that "step out of the current function" option would work, but it seems to be unpredictable (it seemed to happen at least when there is setTimeout). And the function stack gets updated on every user input (and to capture this bug the user input sequence should be fast), so there is no way to go through the call stack either.
It would be perfect to have something like a call stack tree, which would show when a certain function was last called (and how many times during a certain period of time or since the page load), what was the position of this function in the tree of all functions, and the ability to go through this tree and search for all instances of calls to a specific function, and then the ability to see what the variables were equal to in every call.
There is a lot of code, but since it was requested, here is a link to the code inside a GitHub repository (it's a Chrome Extension, and the code where the bug happens is inside the objGA object [at the 6490's line of code]). And the bug happens somewhere between the 6600's line and the 6890's (the bug is that sometimes it doesn't move chess pieces on premoves). It's a Chrome Extension to move chess pieces with a keyboard and a mouse.
https://github.com/Sentero-esp12/Lichess-Keyboard-WebRequest-/blob/master/Extension/code.js
Edit:
Luckily I was able to reproduce it in a slow pace with breakpoints (was able to find the exact moment when the bug appeared). So it turned out I was using Math.round instead of Math.floor trying to extract the first digit from a double digit number. Since only some numbers were >= *5, the bug appeared only occasionally and I couldn't notice the cause.
I'm having a problem where sometimes when my JavaScript in a Web page gets the value of window.pageYOffset it is inexplicably 0 even though I know the user is viewing the middle of the document and its value should be huge, like 650000. Note that a huge percentage of the time I get a reasonable value. But sometimes it's zero and sometimes it's a seemingly random small value, like in the 6000 range when I'm expecting 650000.
Rather than post a bunch of code, I'd like to ask some general questions to help me figure out where to begin to look.
This page is being displayed in an iOS WKWebView (though this problem can manifest in a similar context in an Android app). JavaScript methods in my app can be invoked in one of several ways:
When my app is notified that the page has finished loading (via a delegate method), it invokes a JavaScript method using evaluateJavaScript from the Objective-C code.
My app can call evaluateJavaScript at other times, not just when the page finishes loading.
A JavaScript function may be called as the result of a timer firing.
A JavaScript function may be called as the result of a scroll event.
I have been operating under the assumption that the JavaScript code on the page is always operating in a single thread. That is, I don't have a situation where a timer firing, a scroll event happening, or even a call from the Objective-C code (using evaluateJavaScript) is interrupting anything that might be happening in the JavaScript runtime. So I shouldn't have to worry about interrupting some system-level activity that is modifying window.pageYOffset while I'm trying to access it.
So that's my first question: Am I correct that someone outside my code is invoking my JavaScript methods on a single thread and not monkeying with the DOM on another thread?
My second question is related: My code modifies the DOM, adding and removing div elements. I've been assuming that those modifications are synchronous -- if I insert an element with insertAfter or insertBefore, I expect that the child/parent/sibling pointers are accurate upon return, and I assume that I can immediately access things like the top and left values on some other element and they will have been updated to reflect the inserted/removed element. The point being that I shouldn't have to "wait" for the DOM to "stabilize" after making changes and before checking something like window.pageYOffset. Is this correct?
One more clue: To help mitigate this, I have had good luck simply testing window.pageYOffset for zero at the top of a function. If it is zero, I call myself back on a timer (with just a 1 msec delay). If I do that long enough, it will eventually be non-zero.
Perhaps after reading all this, none of the detail is relevant and you know the answer to the basic question: Why do I sometimes get an invalid value (usually 0) in window.pageYOffset when the same line of code gives a valid value at other times.
The problem turned out to be that there appears to be a period of time between when I give the WKWebView a new HTML string to render and when it tells me that it is done loading the page that the existing page is still active. During this time, timers continue to fire, but some document and window properties will not be valid.
Because of the difficulty of debugging JavaScript running in this environment, I was tricking myself into thinking "eventually pageYOffset becomes valid" when in fact what I was seeing was that the new page eventually finished loading, and it was this new page that was generating valid calls to my timer functions.
In my particular case (may not work for everyone) I am able to detect the value of window.pageYOffset at the top of my timer function and if it is 0, call myself back after a brief delay. This allows me to handle the case where, for some reason, window.pageYOffset is just not yet valid (my test will eventually pass and my timer function will continue as usual) and the case where everything is in the process of being thrown away in favor of the new page (in which case the timer will not fire because the page goes away).
I have two specific Javascript questions that are probably answered by one general answer. Please feel free to also submit the corresponding general question--I have difficulties expressing myself.
When I manipulate multiple DOM elements in a single Javascript callback, is the view possibly updated "live" with each individual manipulation, or atomically after the callback returns?
When a user clicks an HTML element twice in a short timeframe, and the corresponding click handler disables the HTML element, is there a guarantee that the handler won't be executed twice?
Preemptively, I do not have a standards citation for this. This is strictly in my experience.
I have never noticed the visible pixels update while Javascript is executing in real time. I suspect that they will not during the standard operation of the browser - it certainly is possible that debugging presents an exception. I have, however, observed synchronous reflow calculations occurring on DOM elements between the top and bottom of a single function call, but these reflow calculations never made it to the pixel buffer ( that I noticed ). These appear to occur synchronously.
function foo() {
$('#myElement').width(); // 100
$('#myElement').parent().width(); // 150
$('#myElement').css('width', 200);
$('#myElement').width(); // 200
$('#myElement').parent().width(); // 250
}
Regarding multiple clicks on an element that is disabled within the click handler, I suspect that the second click will not fire. I believe when the operating system receives a click event it passes it to the browser and it is placed in a queue. This queue is serviced by the same thread that executes Javascript. The OS click event will remain in the queue until Javascript completes execution at which time it will be removed, wrapped as a browser click event, and bubbled through the DOM. At this point the button will already be disabled and the click event will not activate it.
I'm guessing the pixel buffer is painted on-screen as another operation of this same thread though I may be mistaken.
This is based on my vague recollection of standards that I have seen quoted and referenced elsewhere. I don't have any links.
All script executions happen within the same thread. Therefore you can never have simultaneous actions and don't have to worry about concurrent modification of elements. This also means you don't need to worry about a click handler being fired while one is currently executed. However, this doesn't mean they cant immediately fire it again when your script is finished. The execution may be so fast that its indistinguishable.
First Bullet: The updates will be live. For example, attach the following function to an onclick handler:
function(){
var d = document.getElementById("myelement")
d.setAttribute("align", "center")
d.setAttribute("data-foo","bar")
d.setAttribute("data-bar","baz")
}
Now load this in your browser set a breakpoint on the first line. trigger the event and step through line-by-line while watching the DOM. The updates will happen live, they are not going to happen all at once.
If you want them to happen atomically, you'll want to clone the DOM element in question, make the changes on the clone, then replace the original element with the clone. The cloned element is still being updated in realtime, but the user-visible effect is atomic.
Second Bullet: If the second click event comes in after the element has been disabled, then yes, you won't get a second callback. But if there is any delay between the first click and the disable call, (for example some kind of lengthy check needs to be performed to determine if the element should be disabled) and the second click occurs in that delay, it will fire the callback a second time. The browser has no way to know that multiple click events isn't acceptable behavior in a given script.
Given this code:
$('#foo').css('height', '100px'); // or any other change to the DOM
console.log('done!');
When the 2nd statement executes, is it safe to assume that the reflow is complete?
Follow on question: if you replaced the second line with this, does it change the answer?
window.setTimeout('console.log("done");', 1);
I hope my underlying question is clear. Thanks a lot for any input.
Browsers usually queue DOM modifications that require a reflow/repaint, to avoid performing that expensive operation multiple times. There are exceptions to that, however, as you can see in this Q&A: When does reflow happen in a DOM environment?.
Considering the code you posted, and assuming the console will log the output synchronously*, the answer to your first question is no. If you just change the height of an element, the browser will typically finish running all other synchronous code before performing the reflow/repaint operation. But, as the answers on the link above says, some actions do trigger an immediate reflow, so it's not possible to answer the "or any other change to the DOM" part of your question.
Considering the same assumptions above, the answer to your second question would be yes. The string "done" will be logged to the console in the next tick of the browser's event loop, so it's safe to assume that's after the reflow.
Usually you don't have to worry about that kind of browser behavior, unless you're optimizing code for performance, and trying to avoid reflows.
* Sometimes the console outputs later than expected; unfortunately I couldn't find a good link about that.
The .css() method is not asynchronous, thus you can safely assume that any statements after it will execute as you expect.
$('#foo').css('height', '100px');
console.log($('#foo').css('height')); // will log '100px'
I have a block of JavaScript code running on my page; let's call it func1. It takes several milliseconds to run. While that code is running, the user may click, move the mouse, enter some keyboard input, etc. I have another block of code, func2, that I want to run after all of those queued-up input events have resolved. That is, I want to ensure the order:
func1
All handlers bound to input events that occurred while func1 was running
func2
My question is: Is calling setTimeout func2, 0 at the end of func1 sufficient to guarantee this ordering, across all modern browsers? What if that line came at the beginning of func1—what order should I expect in that case?
Please back up your answers with either references to the relevant specs, or test cases.
Update: It turns out that no, it's not sufficient. What I failed to realize in my original question was that input events aren't even added to the queue until the current code block has been executed. So if I write
// time-consuming loop...
setTimeout func2, 0
then only after that setTimeout is run will any input events (clicks, etc.) that occurred during the time-consuming loop be queued. (To test this, note that if you remove, say, an onclick callback immediately after the time-consuming loop, then clicks that happened during the loop won't trigger that callback.) So func2 is queued first and takes precedence.
Setting a timeout of 1 seemed to work around the issue in Chrome and Safari, but in Firefox, I saw input events resolving after timeouts as high as 80 (!). So a purely time-based approach clearly isn't going to do what I want.
Nor is it sufficient to simply wrap one setTimeout ... 0 inside of another. (I'd hoped that the first timeout would fire after the input events queued, and the second would fire after they resolved. No such luck.) Nor did adding a third, or a fourth, level of nesting suffice (see Update 2 below).
So if anyone has a way of achieving what I described (other than setting a timeout of 90+ milliseconds), I'd be very grateful. Or is this simply impossible with the current JavaScript event model?
Here's my latest JSFiddle testbed: http://jsfiddle.net/EJNSu/7/
Update 2: A partial workaround is to nest func2 inside of two timeouts, removing all input event handlers in the first timeout. However, this has the unfortunate side effect of causing some—or even all—input events that occurred during func1 to fail to resolve. (Head to http://jsfiddle.net/EJNSu/10/ and try rapidly clicking the link several times to observe this behavior. How many clicks does the alert tell you that you had?) So this, again, surprises me; I wouldn't think that calling setTimeout func2, 0, where func2 sets onclick to null, could prevent that callback from being run in response to a click that happened a full second ago. I want to ensure that all input events fire, but that my function fires after them.
Update 3: I posted my answer below after playing with this testbed, which is illuminating: http://jsfiddle.net/TrevorBurnham/uJxQB/
Move the mouse over the box (triggering a 1-second blocking loop), then click multiple times. After the loop, all the clicks you performed play out: The top box's click handler flips it under the other box, which then receives the next click, and so on. The timeout triggered in the mouseenter callback does not consistently occur after the click events, and the time it takes for the click events to occur varies wildly across browsers even on the same hardware and OS. (Another odd thing this experiment turned up: I sometimes get multiple jQuery mouseenter events even when I move the mouse steadily into the box. Not sure what's going on there.)
I think you are on the wrong track with your experiments. One problem is of course that you are fighting different message loop implementations here. The other (the one you didn't recognize it seems) is different double click handling. If you click the link twice you won't get two click events in MSIE - it's rather one click event and a dblclick event (for you that looks like the second click was "swallowed"). All other browsers seem to generate two click events and a dblclick event in this scenario. So you need to handle dblclick events as well.
As message loops go, Firefox should be easiest to handle. From all I know, Firefox adds messages to the queue even when JavaScript code is running. So a simple setTimeout(..., 0) is sufficient to run code after the messages are processed. You should refrain from hiding the link after func1() is done however - at this point clicks aren't processed yet and they won't trigger event handlers on a hidden element. Note that even a zero timeout doesn't get added to the queue immediately, current Firefox versions have 4 milliseconds as the lowest possible timeout value.
MSIE is similar, only that there you need to handle dblclick events as I mentioned before. Opera seems to work like that as well but it doesn't like it if you don't call event.preventDefault() (or return false from the event handler which is essentially the same thing).
Chrome however seems to add the timeout to the queue first and only add incoming messages after that. Nesting two timeouts (with zero timeout value) seems to do the job here.
The only browser where I cannot make things work reliably is Safari (version 4.0 on Windows). The scheduling of messages seems random there, looks like timers there execute on a different thread and can push messages into the message queue at random times. In the end you probably have to accept that your code might not get interrupted on the first occasion and the user might have to wait a second longer.
Here is my adaptation of your code: http://jsfiddle.net/KBFqn/7/
If I'm understanding your question correctly, you have a long-running function but you don't want to block the UI while it is running? After the long-running function is done you then want to run another function?
If so instead of using timeouts or intervals you might want to use Web Workers instead. All modern browsers including IE9 should support Web Workers.
I threw together an example page (couldn't put it on jsfiddle since Web Workers rely on an external .js file that has to be hosted on the same origin).
If you click A, B, C or D a message will be logged on the right. When you press start a Web Worker starts processing for 3 seconds. Any clicks during those 3 seconds will be immediately logged.
The important parts of the code are here:
func1.js The code that runs inside the Web Worker
onmessage = function (e) {
var result,
data = e.data, // get the data passed in when this worker was called
// data now contains the JS literal {theData: 'to be processed by func1'}
startTime;
// wait for a second
startTime = (new Date).getTime();
while ((new Date).getTime() - startTime < 1000) {
continue;
}
result = 42;
// return our result
postMessage(result);
}
The code that invokes the Web Worker:
var worker = new Worker("func1.js");
// this is the callback which will fire when "func1.js" is done executing
worker.onmessage = function(event) {
log('Func1 finished');
func2();
};
worker.onerror = function(error) {
throw error;
};
// send some data to be processed
log('Firing Func1');
worker.postMessage({theData: 'to be processed by func1'});
At this point, I'm prepared to say that, regrettably, there is no solution to this problem that will work under all browsers, in every scenario, every time. In a nutshell: If you run a JavaScript function, there's no way to reliably distinguish between input events that the user triggered during that time and those the user triggered afterward. This has interesting implications for JS developers, especially those working with interactive canvases.
My mental model of how JS input events work was off the mark. I'd thought that it went
The user clicks a DOM element while code is running
If that element has a click event handler, the callback is queued
When all blocking code has executed, the callback is run
However, my experiments, and those contributed by Wladimir Palant (thanks, Wladimir) show that the correct model is
The user clicks a DOM element while code is running
The browser captures the coordinates, etc. of the click
Some time after all blocking code has executed, the browser checks which DOM element is at those coordinates, then runs the callback (if any)
I say "some time after" because different browsers seem to have very different behaviors for this—in Chrome for Mac, I can set a setTimeout func2, 0 at the end of my blocking code and expect func2 to run after the click callbacks (which run only 1-3ms after the blocking code finished); but in Firefox, the timeout always resolves first, and the click callbacks typically happen ~40ms after the blocking code finished executing. This behavior is apparently beyond the purview of any JS or DOM spec. As John Resig put it in his classic How JavaScript Timers Work:
When an asynchronous event occurs (like a mouse click, a timer firing, or an XMLHttpRequest completing) it gets queued up to be executed later (how this queueing actually occurs surely varies from browser-to-browser, so consider this to be a simplification).
(Emphasis mine.)
So what does this mean from a practical standpoint? This is a non-issue as the execution time of blocking code approaches 0. Which means that this problem is yet another reason to hew to that old advice: Break up your JS operations into small chunks to avoid blocking the thread.
Web workers, as Useless Code suggested, are even better when you can use them—but be aware that you're foregoing compatibility with Internet Explorer and all major mobile browsers.
Finally, I hope browser-makers will move forward on standardizing input events in the future. This is one of many quirks in that area. I hope Chrome will lead the way to the future: excellent thread isolation, low event latency, and relatively consistent queueing behavior. A web developer can dream, can't he?
You can use dispatchEvent with a custom event name at the end of your function. This won't work on IE, but is still possible; just use fireEvent instead.
Take a look at this:
http://jsfiddle.net/minitech/NsY9V/
Click "start the long run", and click on the textbox and type in it. Voilà!
You can make the event handlers check to see if a flag is set by func1; if so queue func2 if not already queued.
This may either be elegant or ugly depending on the specializedness of func2. (Actually it's probably just ugly.) If you choose this approach, you need some way to hook events, or alternatively your own bindEvent(event,handler,...) function which wraps the handler and binds the wrapped handler.
The correctness of this approach depends on all the events during func1 being queued at the same time. If this is not the case, you can either make func2 idempotent, or (depending on the semantics of func2) put an ugly "cannot be called again for N milliseconds" lock on it.
please describe better you scenario.
What you need do
some time ago i need do something how that was so i build an simple javascript's routine across serialize async call in one sync call. maybe you could used that added one variant
for example that let me show how that work
first register all async or sync routines
second register end callback
third register call's to routines with yours parameters
fourth thrown process
in your case it neccesary added one call routine and that routine should be UoW of user actions.
Now the main problem is not call to routine in and order of execution if not track changes done by the user
first register all async or sync routines
second register end callback
third register call's to routines with yours paramter
--register your first routine
--register BlockUi //maybe for not accept more changes in the view
--register UiWriter // UoW of change done by user
--register you last routine
fourth thrown process
in real code that is one call dummy's function
function Should_Can_Serializer_calls()
{
RegisterMethods(model);
model.Queue.BeginUnitProcess(); //clear stack of execution, y others
model.Queue.AddEndMethod(SucessfullEnd); // callback to end routine
model.AbstractCall("func1",1,"edu",15,""); //set routine how first to execute
model.AbstractCall("BlockUi"); //track changes and user's actions
model.AbstractCall("UiWork"); //track changes and user's actions
model.AbstractCall("func2","VALUE"); //set second routine for execute
model.Process(); //throw call
}
Now the methods should be async for themselves for this you could use that library http://devedge-temp.mozilla.org/toolbox/examples/2003/CCallWrapper/index_en.html
so, what do you want do?