Eli Zaretskii writes:
> Out of curiosity: why are you declaring a paragraph consisting
> entirely of L2R text as R2L?  is there some real-life use case behind
> this?

As you probably guessed this is a radical simplification of the actual
problem that I have.

> (If you replace "abc" with "אבג", the display will be as you expect.)

Interesting.  It seems that I have simplyfied it a bit too much.
Although the version that I created still indicates a problem, of
course.

I experimented a bit and I can't quite exactly say what is needed to
cause the problem.  The same file/URL has it and than again not when I
try again.  It seems that "g" usually fixes it, while "G RET" usually
reproduces it, if it has happend with that particular file/URL before.
It also seems that the length of the URL has an impact, probably because
it is shown at the top of the frame and during loading of the URL (as in
"Loading http[...]").  It seems to require that the URL is longer than
the frame width.

I have no sure recipe using a file, but I have something based on a
simple 10-line HTTP server look-alike, see attachments.  Put the files
in some directory, make the script executable, execute it, and call EWW
with the URL
http://127.0.0.1:1234/abc/abc/abc/abc/abc/abc/abc/abc/abc/abc/abc/abc/.
If it looks good on the first call, try "G RET", than it reliably
exhibits the problem for me.  I even get it with "emacs -Q -nw".  Note
that the script depends on netcat and different versions of netcat use
different options to run a server, so the script may not work as-is for
you.