Re: Bidi, HTML5 and CSS3, test bidi-html5-019

Eric, excellent question!

Fantasai, your answer raises a question for me: do you think that the CSS
spec as it stands actually forbids a user agent to "reopen" the embeddings
started by explicit formatting characters (as opposed to element tags), or
is it just something that is left unspecifed? In other words, would a user
agent that renders Eric's case with D in left-to-right-override be in
violation of the spec? (I ask because this has direct bearing on the CSS
tests that we are starting to write.)

Aharon

On Wed, Mar 14, 2012 at 1:12 AM, fantasai <fantasai.lists@inkedblade.net>wrote:

> On 03/13/2012 03:30 PM, Eric Muller wrote:
>
>> On 3/10/2012 10:16 AM, Aharon (Vladimir) Lanin wrote:
>>
>>> > What does that mean in terms of implementation?
>>>
>>> I am not sure what you mean.
>>>
>>>
>> I think this case will answer my question unambiguously:
>>
>> <div direction='ltr'> A <span style="unicode-bidi:override;
>> direction:rtl;"> B &#x202d; C<br/> D &#x202c; E </span> F </div>
>>
>> (where A...F stand for arbitrary strings, not for characters; spaces
>> added for readability, not part of the text)
>>
>> Is the explicit LRO ("&#x202d;") reopened on the other side of the <br>?
>> or asked another way, is the type of the characters
>> in the fragment D overridden to L or to R?
>>
>
> CSS's bidi behavior only inserts codes corresponding to CSS-specified
> bidi controls; it leaves the plaintext control codes intact and does
> not try to interpret or augment them. The resulting stream is then
> bidi-processed. So in your example, the LRO is not reopened on the
> other side of the <br>; it is terminated, as in plaintext, at the
> forced break. Furthermore, the PDF after D closes the RLO opened by
> the <span>.
>
> This is why we recommend not mixing markup and plaintext bidi controls.
> It can create quite a mess!
>
> ~fantasai
>
>