fix #2434 - wrong calculation of offset

[rrelmy]

Fixes the bug described in #2434

Offsets are rounded correctly now

[rrelmy]

Does something block this commit to be merged?

[ibolmo]

Are you using sub pixels when setting the left and top?

[ibolmo]

Update the PR to use String.toFloat instead.

[rrelmy]

The problem mostly occurs on slideshows if you use an element centered with auto margins within an element of odd width or using percents in width/padding.

The pull request has been updated.

[ibolmo]

This is going to be on hold, until we get a spec to check for this. You're welcome to add one yourself, or wait about a week to get a hold of the pending PR that reduces the complexity to test MooTools.

[rrelmy]

I have rebased the previous commit and written a test spec.

PhantomJS version 1.9.8 used by the karma runner is based on a webkit version released around 2011 before subpixel alignment was added to webkit.
EDIT: I broke the builds again, tested locally against phantomjs v2 …

Is a test spec needed for this change? If the browser itself does not support subpixel alignment it is just fine, but I can't think of a way to test this usefully without breaking the PhantomJS build …

If something is missing please let me know.

[kentaromiura]

LGTM but travis failed

[SergioCrisostomo]

Travis fails on PhantomJS tests. We could add a bypass to phantom.js in the specs for this.

[rrelmy]

The test case has been modified to compare the result of getPosition with the output of getBoundingClientRect.
All tests pass now without testing for PhantomJS