I dived into the jQuery source, and solved it myself looking at the code of the offset() routine. The comments below relate to the code there.

In IE8, the self.pageYOffset, body.scrollTop and clientTop vars always remain 0, leaving the box.top variable (which varies with scrolling) as the only var left. In both box-sizing models IE8 has a value for docElem.scrollTop, so there really is no purpose for the boxModel check there.

Just removing the jQuery.boxModel check solves the problem, and now the offset() method returns always correct offset values in both box-sizing models for IE8 & FF.

Now I don't know if that breaks Opera, or older IE versions, but it seems a perfect solution for now.

The core lines of the offset() routine now read:

top  = box.top  + (self.pageYOffset || docElem.scrollTop  || body.scrollTop ) - clientTop,
left = box.left + (self.pageXOffset || docElem.scrollLeft || body.scrollLeft) - clientLeft;

Regards,

Abraham Guyt.

reduced test case. Milestone to 1.4.5 since the user has provided a potential fix.

It would be nice to have my proposed solution included in an upcoming new jQuery release, as now I have to re-inject my fix manually in every release again to solve this bug. Given the fact it was reported a year ago, that would be too much to ask I guess :-)

Replying to [comment:5 dmethvin]:

yes ... could you please implement the suggested fix in the official release ?

Versions 1.6.2 still contains the mentioned bug for strict IE8, changing the code lines nr 8641 + 8642 to the following solves the problem (just remove the boxmodel check):

scrollTop  = win.pageYOffset || docElem.scrollTop  || body.scrollTop,
scrollLeft = win.pageXOffset || docElem.scrollLeft || body.scrollLeft,

Could it please be included in the 1.7 release ?

I need this too ! 1.7 ?

Sorry but I'm just coming up to speed with this patch. The test case appears to have been fixed in jQuery 1.6.2, can you confirm?

Works in 1.6.2: http://jsfiddle.net/ESGnf/9/

Fails in 1.5.2: http://jsfiddle.net/ESGnf/8/

If not, what's the difference between your real-life situation and this test case? Can you create a fiddle where it fails?

This seems to be related to #7986 where our boxModel feature detect is fooled by global css monkey business. I am not sure why we don't use a direct document.compatMode === "BackCompat" there.

Finally, everything still needs to work properly in IE6/7 where box-model is not available but standards/quirks mode exists. How is the patch taking that into account?

Hi, after testing with your fiddle examples and my code I found what caused the malfunction in the old and existing 1.6.2 code.

Your correct fiddle example has the 1.6.2 jQuery lib include <script> line BEFORE the stylesheet section and results in correct behaviour.

If you move that same lib include <script> line to just AFTER the stylesheet section you're getting the wrong behaviour (with IE8). I can't give you that example with the fiddle interface as it automatically places the lib include line, but if you take the source and move the line manually you'll see the mentioned behaviour.

So somehow the positioning of the jQuery lib inclusion line relative to the stylesheets influences its behaviour in this manner - is this supposed to work like that ?

It's then important devs know it works like that - I wouldn't have found out without your samples - thanks for that.

Oh, that kinda makes sense. The jQuery.support.boxModel (née jQuery.boxModel) variable is calculated during the support tests, and if the page monkeys with the CSS after that it wouldn't affect the calculation. But that doesn't mean the calculation would be right in general.

The way you've changed the value is for all elements, but if someone applied the changes to only certain elements then the assumption that jQuery.support.boxModel makes ("There's only one box model and it applies to all elements and never changes") is wrong. It seems like we'd need to check the element's box-model each time we return the value to be certain we're getting it right.

It certainly can happen that a single page uses and mixes both border-box models, so it's important the offset values are correct for those cases in IE8 as well. At least we now know the lib inclusion position is crucial to do right :-)

This is a tricky problem because IE8 supports box-sizing, but has an incorrect default. Previously with the box model test the solution was to set all the properties to the default. Which works because all browsers have the same defaults for padding, border, overflow and display.

It seems the best solution now is to use iframe to create a pristine environment to test against. This fiddle demonstrates it:

http://jsfiddle.net/GESZS/25/

I'm not sure if this will resolve the original issue, but I hope it helps.

Moving to 1.8 for a comprehensive solution.

Duplicate of #7986.

Fixed in 1.7.2, along with #7986.