Routing to Layout team.

Would you mind routing this to someone who can look into it Koji? Thanks!

I think this is by design, but I'm not native speaker of bidi scripts, so please correct me if anything looks wrong with the below.

## Test
http://output.jsbin.com/qilasaw

Gecko: Puts the trailing space before (as you expect) only when wrapping.  It looks like it collapses the whitespace after that.
Edge: Same as Gecko but doesn't collapse it.

## UBA Test
http://unicode.org/cldr/utility/bidi.jsp?a=Say+%D8%A7+%D8%A8+%D8%AC%0D%0ASay+%D8%A7+%0D%0A&p=LTR

You're right that the space goes before "ا‎" without wrapping, but bidi reordering applies after line breaks, so it should be the same as "Paragraph 2" above. It shows that the space should appear after "ا‎". Note that, although this is unicode.org site, this tool isn't updated, so may be showing old behavior.

## As a user...
Why would you want two spaces there? It looks like unnatural to me. Do you have use cases where your display order is useful?

I think, in your example without wrapping the trailing space is not to the left because it is not considered to be part of a rtl section. UBA test confirms that[1]: The WS both have resulting level 0. If you add a RLM[2] to the end the last WS is on level 1 and displayed to the left.

[1] http://unicode.org/cldr/utility/bidi.jsp?a=Say+%D8%A7+&p=LTR
[2] http://unicode.org/cldr/utility/bidi.jsp?a=Say+%D8%A7+%E2%80%8F&p=LTR

As a user, I don't know what behavior you would want here. I've never written from right to left. I'm currently working on CodeMirror's bidi support and a certain level of consistency in browser behavior would help with that. What Firefox does matches my mental model in this case.

Resetting trailing spaces to the paragraph embedding level is called L1[1], and you're right that Blink doesn't support it very well yet. This is tracked in issue 316409. We're discussing how to fix it, but it'll probably not in short term.

But my knowledge on bidi still prevents to understand this is about the lack of L1 when our result and UBA test matches, unless, as you noted, a RLM is added to the end, but I can't understand why we should add RLM to the end of line.

If you agree that this is about the lack of L1, I can merge this to issue 316409. If you have other information than just L1, I'd appreciate to know.

[1] http://unicode.org/reports/tr9/#L1

I'm somewhat convinced Blink is showing the right behavior and Firefox is wrong. I just didn't know about L1 at all. I opened https://bugzilla.mozilla.org/show_bug.cgi?id=1331501.

Thanks, there are experts in Gecko team, their opinions are helpful.

I'll keep this as Duplicate rather than WontFix. I will probably revisit L1 in Blink sometime in one of our long term projects, and will keep this to be re-tested when we do so. Apologies it's not happening sooner.