Eli Zaretskii <eliz@gnu.org> writes:

> It should be possible, in principle, to write something similar to
> posn-at-point that would not depend on the text being visible, or
> rewrite posn-at-point to free it from this limitation, but we don't
> have such a function at this time.  Maybe we should.

Of course would that be useful, not only for this issue.  But I guess it
would be a lot of work.

I try now what you suggested, subtracting one more column width when
calculating `shr-internal-width'.  I.e., I subtract one more
`frame-char-width' now.  Looks good so far.  I'll use it for some time
and tell you what I think.

So, here is what I'm testing, based on your first patch: