Make unicode-bidi:isolate the default for block elements instead of unicode-bidi:embed

HTML5 specifies that block elements  are supposed to have unicode-bidi: isolate
by default.

CSSDefaultStyleSheets.cpp:
After this change, div elements can not use the simple stylesheet anymore.

html.css is using the rules specified in
http://www.w3.org/html/wg/drafts/html/master/rendering.html#bidirectional-text

BUG=296863

[eseidel, 2013-11-14 06:38:14]

We could also change the simple default stylesheet.  If div's can't use it, I'm
not sure it provides any value. :(

[eseidel, 2013-11-14 06:39:49]

I'm confused why we have that long list of tags for which dir implies embed? 
Why does dir= not always imply embed?

[eseidel, 2013-11-14 06:41:29]

https://codereview.chromium.org/61733019/diff/1/Source/core/html/HTMLElement.cpp
File Source/core/html/HTMLElement.cpp (right):

https://codereview.chromium.org/61733019/diff/1/Source/core/html/HTMLElement....
Source/core/html/HTMLElement.cpp:160: static bool
shouldUseUnicodeBidiEmbed(const Element* element)
Is this trying to list all block elements?  Why wouldn't this detect if things
are display: block?  Why would dir imply embed for <div> but not <span
style="display: block" dir='ltr'>?  Is this from HTML5? If so, can you link to
the part of the spec?

[esprehn, 2013-11-14 06:46:54]

not lgtm, you can't remove <div> from the simple stylesheet (it defeats the
sheet). More importantly though the spec special casing elements like this
doesn't seem right at all. What about a custom element that extends <div> ? We
should push back on this change in the HTML5 spec, it should specify that
display: block implies unicode-bidi:isolate not list a huge set of tag names.

https://codereview.chromium.org/61733019/diff/1/Source/core/css/CSSDefaultSty...
File Source/core/css/CSSDefaultStyleSheets.cpp (right):

https://codereview.chromium.org/61733019/diff/1/Source/core/css/CSSDefaultSty...
Source/core/css/CSSDefaultStyleSheets.cpp:65: return isHTMLHtmlElement(e) ||
e->hasTagName(headTag) || e->hasTagName(bodyTag) ||  e->hasTagName(spanTag) ||
e->hasTagName(brTag) || isHTMLAnchorElement(e);
Removing div from the simple sheet doesn't look right. That defeats the
advantage  of the simple sheet. Every web page ever uses <div>'s.

https://codereview.chromium.org/61733019/diff/1/Source/core/html/HTMLElement.cpp
File Source/core/html/HTMLElement.cpp (right):

https://codereview.chromium.org/61733019/diff/1/Source/core/html/HTMLElement....
Source/core/html/HTMLElement.cpp:162: if (element->hasTagName(bdoTag)
This is not right.

[igoroliveira, 2013-11-14 07:49:10]

On 2013/11/14 06:41:29, eseidel wrote:
>
https://codereview.chromium.org/61733019/diff/1/Source/core/html/HTMLElement.cpp
> File Source/core/html/HTMLElement.cpp (right):
> 
>
https://codereview.chromium.org/61733019/diff/1/Source/core/html/HTMLElement....
> Source/core/html/HTMLElement.cpp:160: static bool
> shouldUseUnicodeBidiEmbed(const Element* element)
> Is this trying to list all block elements?  Why wouldn't this detect if things
> are display: block? 

It is not about display: block, it is about HTML block elements. if an element 

> Why would dir imply embed for <div> but not <span
> style="display: block" dir='ltr'>?  Is this from HTML5? If so, can you link to
> the part of the spec?

In fact, it is the opposite, span should imply the embed and div should not.

Basically for blink and webkit dir=something (except for bdo, bdi and output)
are embed.
I am maintaining this behavior because it is still being discussed
https://www.w3.org/Bugs/Public/show_bug.cgi?id=23260 and i do not want to break
the web :)
However for HTML block elements, the behavior is described here:
http://www.w3.org/html/wg/drafts/html/master/rendering.html#bidirectional-text

[igoroliveira, 2013-12-09 15:41:43]

Aharon,

Could you put your opinion. Looks like this CL is controversial because.(Look
esprehn message)

[igoroliveira, 2013-12-09 15:41:45]

Aharon,

Could you put your opinion. Looks like this CL is controversial because.(Look
esprehn message)

[eseidel, 2013-12-09 16:21:04]

Simple default stylesheet has been removed from blink.

[aharon, 2013-12-12 11:52:11]

On 2013/11/14 06:46:54, esprehn wrote:
> More importantly though the spec special casing elements like this
> doesn't seem right at all. What about a custom element that extends <div> ? We
> should push back on this change in the HTML5 spec, it should specify that
> display: block implies unicode-bidi:isolate not list a huge set of tag names.

There is a misunderstanding here.

Neither unicode-bidi:embed nor unicode-bidi:isolate has any effect whatsoever on
elements with display:block (see
http://www.w3.org/TR/2013/WD-css-writing-modes-3-20131126/#unicode-bidi). The
boundaries of an element with display:block form boundaries for the bidi
algorithm paragraph (see
http://www.w3.org/TR/2013/WD-css-writing-modes-3-20131126/#text-direction), and
these paragraphs get their overall direction from the element's direction style,
regardless of whether unicode-bidi for the element is normal, embed, or isolate.

So, why does the HTML spec list all the "block" elements and demand that they
get unicode-bidi:isolate? The idea is to make unicode-bidi:isolate the default
for all elements that people normally think of as "block" elements, so that when
a page overrides their default display:block with a rule giving them
display:inline, they continue to behave bidi-wise pretty much as they would if
they still had display:block. Without that, for example, with
<ol><li>a</li><li>ב</li><li>ג</li></ol> where the li's have been given
disdplay:inline but still have unicode-bidi:normal, the result would be to
display the ג between the a and the ב. This is not what people expect.

In HTML4, the "block" elements were supposed to have unicode-bidi:embed by
default for the same reason (and unicode-bidi:isolate wasn't available yet), but
unicode-bidi:embed does not prevent the problematic behavior in the case above;
only unicode-bidi:isolate does.

[eseidel, 2013-12-18 17:33:13]

lgtm

OK.  I'm OK with this approach.

The idea here is that tags which the default stylesheet lists as "display:
block" by default should also be listed as having "unicode-bidi: isolate".

Since it's the stylesheet applying this "blockness" in the first place, it makes
sense that it should also apply this "isolate-ness".

(Elements w/o any style specified are inlines by default.)

[eae, 2014-03-25 17:49:46]

What is the status of this patch?

[eseidel, 2014-05-28 22:32:54]

Igor has left the project.  This would need a new owner.

[ebraminio, 2015-05-01 23:03:50]

This is really helpful in my experience on sites dealing with bidi content and
now Firefox also has this. I hope it can find some new owner.

[eae, 2015-05-01 23:06:53]

On 2015/05/01 23:03:50, ebraminio wrote:
> This is really helpful in my experience on sites dealing with bidi content and
> now Firefox also has this. I hope it can find some new owner.

I'll see if I can find an owner or take this over myself.

[aharon, 2015-05-03 06:56:57]

On 2015/05/01 23:06:53, eae wrote:
> On 2015/05/01 23:03:50, ebraminio wrote:
> > This is really helpful in my experience on sites dealing with bidi content
and
> > now Firefox also has this. I hope it can find some new owner.
> 
> I'll see if I can find an owner or take this over myself.

Please sync up on this with kojii@