When U+20D7 (and possibly other combining characters) is separated from the preceding character by a span element, the combination renders badly in Safari 10.
In particular, Safari 10 renders this differently from other browsers (see ).
Safari 10 also renders more test cases badly than Safari 9 did. On the other hand, WebKitGTK+ 2.16.3 has no issues.
* minimal example 
* screenshots (archive) 
* original context 
Thanks in advance.
This is a duplicate of https://bugs.webkit.org/show_bug.cgi?id=6148.
*** This bug has been marked as a duplicate of bug 6148 ***
I can see the connection but it seems this issue might be different enough since
a) there's a regression in Safari 10 compared to 9
b) WebKitGTK+ 2.16.3 does not have a problem
while #6148 is much older and marked as applying to WebKit in general.
Reopening due to regression
See also: <https://github.com/mathjax/MathJax/issues/1709#issuecomment-305795001>
Created attachment 318653 [details]
Until we can do cross-element shaping, if you start an element with a combining mark, you're gonna have a bad time. https://bugs.webkit.org/show_bug.cgi?id=6148 will fix this.
Between Safari 9 and Safari 10, I fixed some bugs about how we account for the initial advance. These fixes are progressions, and should not be reverted. The best bet for us to fix this problem is just to fix the root issue, and not try to figure out which progression caused this bug to occur.