Dear Taco, thanks for your reply.

> I opened the xhtml with firefox and the only problem I see
> is the ignored \displaystyle (which is why the integral looks small).
> Inline looks fine otherwise, no interchanging of limits.

Hmm, I use firefox 68.0.1 of Ubuntu 19.04 and firefox 60.8.0
of Debian/unstable.  I see the interchanging of limits with
both firefoxes.  BTW, it seems Chromium doesn't support MathML.
If any Linux users could test the xhtml with firefox (or any browsers)

I believe it would help us much.

> Well, as fine as it can be in firefox, which is pretty bad.

I'm not sure if I understand you correctly but I think MathJax
might be much better than MathML.

> ConTeXt outputs b_a^\int in display mode; the order of the objects
> inside the generated <msubsub> is wrong, which is clearly a bug.
> And it seems to do this for all large operators.

I see.

> I think this somehow triggered by the nested mode switch (\text{\math{}}) in the \vsym.
> If I change that definition to the more simple:

>   \define[1]\vsym{{\textstyle\bi #1}}

> it works ok. Whether that helps you ...

It helps me to some extent but with the original definition, we can
use \vI in both math mode (\math{\vI}) and text mode like "vector \vI".
So if ConTeXt could convert the nested mode switch correctly
it would be very nice.

Thanks for your info.
Best regards,
Atsuhito Kohda