How to detect when cancel is clicked on file input?

While not a direct solution, and also bad in that it only (as far as I've tested) works with onfocus (requiring a pretty limiting event blocking) you can achieve it with the following:

document.body.onfocus = function(){ /*rock it*/ }

What's nice about this, is that you can attach/detach it in time with the file event, and it also seems to work fine with hidden inputs (a definite perk if you're using a visual workaround for the crappy default input type='file'). After that, you just need to figure out if the input value changed.

An example:

var godzilla = document.getElementById('godzilla')

godzilla.onclick = charge

function charge()
{
    document.body.onfocus = roar
    console.log('chargin')
}

function roar()
{
    if(godzilla.value.length) alert('ROAR! FILES!')
    else alert('*empty wheeze*')
    document.body.onfocus = null
    console.log('depleted')
}

See it in action: http://jsfiddle.net/Shiboe/yuK3r/6/

Sadly, it only seems to work on webkit browsers. Maybe someone else can figure out the firefox/IE solution

So I'll throw my hat into this question since I came up with a novel solution. I have a Progressive Web App which allows users to capture photos and videos and upload them. We use WebRTC when possible, but fall back to HTML5 file pickers for devices with less support *cough Safari cough*. If you're working specifically on an Android/iOS mobile web application which uses the native camera to capture photos/videos directly, then this is the best solution I have come across.

The crux of this problem is that when the page loads, the file is null, but then when the user opens the dialog and presses "Cancel", the file is still null, hence it did not "change", so no "change" event is triggered. For desktops, this isn't too bad because most desktop UI's aren't dependent on knowing when a cancel is invoked, but mobile UI's which bring up the camera to capture a photo/video are very dependent on knowing when a cancel is pressed.

I originally used the document.body.onfocus event to detect when the user returned from the file picker, and this worked for most devices, but iOS 11.3 broke it as that event is not triggered.

Concept

My solution to this is *shudder* to measure CPU timing to determine if the page is currently in the foreground or the background. On mobile devices, processing time is given to the app currently in the foreground. When a camera is visible it will steal CPU time and deprioritize the browser. All we need to do is measure how much processing time our page is given, when camera launches our available time will drop drastically. When the camera is dismissed (either cancelled or otherwise), our available time spike back up.

Implementation

We can measure CPU timing by using setTimeout() to invoke a callback in X milliseconds, and then measure how long it took to actually invoke it. The browser will never invoke it exactly after X milliseconds, but if it is reasonable close then we must be in the foreground. If the browser is very far away (over 10x slower than requested) then we must be in the background. A basic implementation of this is like so:

function waitForCameraDismiss() {
  const REQUESTED_DELAY_MS = 25;
  const ALLOWED_MARGIN_OF_ERROR_MS = 25;
  const MAX_REASONABLE_DELAY_MS =
      REQUESTED_DELAY_MS + ALLOWED_MARGIN_OF_ERROR_MS;
  const MAX_TRIALS_TO_RECORD = 10;

  const triggerDelays = [];
  let lastTriggerTime = Date.now();

  return new Promise((resolve) => {
    const evtTimer = () => {
      // Add the time since the last run
      const now = Date.now();
      triggerDelays.push(now - lastTriggerTime);
      lastTriggerTime = now;

      // Wait until we have enough trials before interpreting them.
      if (triggerDelays.length < MAX_TRIALS_TO_RECORD) {
        window.setTimeout(evtTimer, REQUESTED_DELAY_MS);
        return;
      }

      // Only maintain the last few event delays as trials so as not
      // to penalize a long time in the camera and to avoid exploding
      // memory.
      if (triggerDelays.length > MAX_TRIALS_TO_RECORD) {
        triggerDelays.shift();
      }

      // Compute the average of all trials. If it is outside the
      // acceptable margin of error, then the user must have the
      // camera open. If it is within the margin of error, then the
      // user must have dismissed the camera and returned to the page.
      const averageDelay =
          triggerDelays.reduce((l, r) => l + r) / triggerDelays.length
      if (averageDelay < MAX_REASONABLE_DELAY_MS) {
        // Beyond any reasonable doubt, the user has returned from the
        // camera
        resolve();
      } else {
        // Probably not returned from camera, run another trial.
        window.setTimeout(evtTimer, REQUESTED_DELAY_MS);
      }
    };
    window.setTimeout(evtTimer, REQUESTED_DELAY_MS);
  });
}

I tested this on recent version of iOS and Android, bringing up the native camera by setting the attributes on the <input /> element.

<input type="file" accept="image/*" capture="camera" />
<input type="file" accept="video/*" capture="camcorder" />

This works out actually a lot better than I expected. It runs 10 trials by requesting a timer to be invoked in 25 milliseconds. It then measures how long it actually took to invoke, and if the average of 10 trials is less than 50 milliseconds, we assume that we must be in the foreground and the camera is gone. If it is greater than 50 milliseconds, then we must still be in the background and should continue to wait.

Some additional details

I used setTimeout() rather than setInterval() because the latter can queue multiple invocations which execute immediately after each other. This could drastically increase the noise in our data, so I stuck with setTimeout() even though it is a little more complicated to do so.

These particular numbers worked well for me, though I have see at least once instance where the camera dismiss was detected prematurely. I believe this is because the camera may be slow to open, and the device may run 10 trials before it actually becomes backgrounded. Adding more trials or waiting some 25-50 milliseconds before starting this function may be a workaround for that.

Desktop

Unfortuantely, this doesn't really work for desktop browsers. In theory the same trick is possible as they do prioritize the current page over backgrounded pages. However many desktops have enough resources to keep the page running at full speed even when backgrounded, so this strategy doesn't really work in practice.

Alternative solutions

One alternative solution not many people mention that I did explore was mocking a FileList. We start with null in the <input /> and then if the user opens the camera and cancels they come back to null, which is not a change and no event will trigger. One solution would be to assign a dummy file to the <input /> at page start, therefore setting to null would be a change which would trigger the appropriate event.

Unfortunately, there's no way official way to create a FileList, and the <input /> element requires a FileList in particular and will not accept any other value besides null. Naturally, FileList objects cannot be directly constructed, do to some old security issue which isn't even relevant anymore apparently. The only way to get ahold of one outside of an <input /> element is to utilize a hack which copy-pastes data to fake a clipboard event which can contain a FileList object (you're basically faking a drag-and-drop-a-file-on-your-website event). This is possible in Firefox, but not for iOS Safari, so it was not viable for my particular use case.

Browsers, please...

Needless to say this is patently ridiculous. The fact that web pages are given zero notification that a critical UI element has changed is simply laughable. This is really a bug in the spec, as it was never intended for a full-screen media capture UI, and not triggering the "change" event is technically to spec.

However, can browser vendors please recognize the reality of this? This could be solved with either a new "done" event which is triggered even when no change occurs, or you could just trigger "change" anyways. Yeah, that would be against spec, but it is trivial for me to dedup a change event on the JavaScript side, yet fundamentally impossible to invent my own "done" event. Even my solution is really just heuristics, if offer no guarantees on the state of the browser.

As it stands, this API is fundamentally unusable for mobile devices, and I think a relatively simple browser change could make this infinitely easier for web developers *steps off soap box*.

You can't.

The result of the file dialog is not exposed to the browser.

When you select a file and click open/cancel, the input element should lose focus aka blur. Assuming the initial value of the input is empty, any non empty value in your blur handler would indicate an OK, and an empty value would mean a Cancel.

UPDATE: The blur is not triggered when the input is hidden. So can't use this trick with IFRAME-based uploads, unless you want to temporarily display the input.