using innerHTML can cause css display errors in Opera

[davidserduke]

There are times when setting innerHTML can cause the css display value to be incorrect in Opera. This is a rather convoluted issue and one specific case has been identified and is shown in the attached test case.

Basically, when a div structure like this exists

<style type="text/css">
  #hid { display:none; }
</style>
<div id="container" style="display:none;">
  <div>
    <div id="hid">hidden</div>
  </div>
</div>  

and you set the innerHTML (yes you set it to itself):

$("#container")[0].innerHTML = $("#container")[0].innerHTML;

it can cause the display value from .css() to give improper results on the inner div. In this case, #hid should be display:none but instead it returns display:block.

There are many things that appear to be required so far to cause the bug including the #container div must be display:none or an ancestor must be. There must be a div between the #container and the #hid. The #hid must be display:none through a cascade in the stylesheet and cannot be style="display:none;". And, of course, it only happens in Opera. There may be other criteria or cases too but this is the only one I've found so far.

Interestingly there is a way to fix it as well by doing

var elem = $("#hid")[0];
var save = elem.style.display;
elem.style.display = "block";
elem.style.display = save;

Clearly this should just set the display to block then immediately reset to the original value. It has the consequence of "fixing" the Opera problem though such that the .css() value becomes "none" again as it should be.

No one has reported this bug in a real-life instance. It was found in the test suite which is constantly reseting the innerHTML of the test area thus giving the problem a chance to surface which it did. At this time it is not clear what the best fix is so we are opening a ticket for it.

[davidserduke]

test case

[davidserduke]

patch

[davidserduke]

Fixed in [4196].

[dvessel]

Hello, I'm coming from Drupal and we are trying to get jQuery 1.2.3 in before release. I tracked a major slowdown in two instances do to this change. It's triggered when .offset() is used.

 http://drupal.org/node/210131#comment-711567

Would there be another way to fix this problem without the performance hit? This affects Opera 9.2.5 in both Mac and Windows.

[Junyor]

Alternative fix

[Junyor]

I've attached an alternative patch that fixes the reported issue and avoids the performance problems seen in Drupal 6.

[Junyor]

The fix was found by David Bloom ( http://my.opera.com/dbloom/) with help from fearphage ( http://my.opera.com/fearphage/).

[Junyor]

Note: Opera's bug is fixed in Kestrel (AKA Opera 9.5), so the work-around won't be necessary there. However, the work-around I proposed should cause no performance problems.

[john]

Fixed in SVN rev [4660].