Re: [css3-writing-modes] [CSS21] bidi of replaced inline elements

> With regards to replaced elements and their bidi behavior, there are
> two cases to consider:
>  - what happens when the image falls back to its alt text
>  - what happens if the image is intended to represent a character
>    (that is not otherwise encoded or representable)

> In the first case, the element's effect on its surrounding text
> should be no different whether it is rendered as replaced or as
> text. In this case, we should be treating it as you say -- by
> handling it with the appropriate embedding codes.

I would argue that in this case, whether rendered or replaced, the best
effect to have on the surrounding is no effect at all, i.e. the effect of a
neutral character (or if you prefer to think of it that way,
unicode-bidi:isolate).

> In the second case, the author might want to control the element's
> effective directionality. In which case that approach won't work,
> and we need to replace the character with a pretend character as
> currently specified in the text.

I think it's a rare case, which can be handled by either putting the element
in a span with the desired direction or surrounding it with either LRMs or
RLMs.

So our options here are
 a) Take approach 1 (change the specs to use embedding codes)
 b) Take approach 2 (no change)
 c) Take approach 1 for 'embed' and approach 2 for 'bidi-override'

I want d): non-textual elements are always treated as a neutral character
regardless of their unicode-bidi value.

As I said in another post, this is the browsers' current behavior. Changing
it would create backward compatibility problems, since authors may have
assigned a dir to non-textual elements in order to have their alt or title
text displayed correctly, and this dir could now affect the ordering outside
the element.