physik wrote:

I think for the 2.1 version our goal should be to have just one <math> block and some javascript to help browsers to understrand that.

From James Craig:

It might be safer to set up a new class on a new parent element as the positioning context (container of both .mwe-math-fallback-image-inline and .mwe-math-mathml-a11y), and then absolutely position the math element to the top left of the context.
.mwe-math-context {

position: relative; /* context for descendants */
display: inline-block;
margin: 0;
padding: 0;

}
.mwe-math-context > .mwe-math-mathml-a11y {

position: absolute;
top: 0; /* explicit positions within context */
left: 0;
opacity: 0;

}

So since this bug was confusely used by the MathJax team to justify their refusal to provide MathML in the DOM by default in order to help assistive technologies, I'd like to clarify the issue:

• when the MathML is in the DOM (as currently provided by the MediaWiki MathML mode), it will be usable by any assistive technologies.

• however when the MathML is "hidden" (as currently done in the MediaWiki MathML mode for WebKit), the accessibility features that depend on the visual rendering (e.g. touch screen navigation here) will obviously not work.

Note: future version of Gecko will work with VoiceOver and will get "visible" MathML by default, so I'm updating the title to clarify that it is only for WebKit.