Brett Zamir Reporter
	Description • 13 years ago

User-Agent:       Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0.1) Gecko/20100101 Firefox/4.0.1
Build Identifier: 

I'm not sure whether this is some other issue, but when I try to do an alert using a combining diacritical mark, it looks off, but when added into the page, it looks fine.


Reproducible: Always

Steps to Reproduce:
1. Add a character with a combining mark into an HTML page: À
2. Add a script into the same page to alert the same character:
<script type="text/javascript">
alert('A\u0300'); // The diacritic appears to the top left of the character instead of the top
</script>


Actual Results:  
The characters appear differently, with the alerted character having the diacritic askew.

Expected Results:  
See the combining character seamlessly integrated into the base character.

	Boris Zbarsky [:bzbarsky]
	Comment 1 • 13 years ago

This worksforme (on Mac).

Are you sure this isn't just an issue of the web page and the alert using different fonts and the font for the latter (or our handling of it) being broken in your case?

	Brett Zamir Reporter
	Comment 2 • 13 years ago

Yes, that may well be. I'm sorry I don't know how to confirm, just know that the default alert font isn't working--not sure that's something u can do anything about or not.

	Boris Zbarsky [:bzbarsky]
	Comment 3 • 13 years ago

Hm.  What hapens if you style your HTML with "font-family: sans-serif"?

	Brett Zamir Reporter
	Comment 4 • 13 years ago

HTML is still ok (and alert still not good). I forgot though that I was testing with a XUL-based HTML editor, and oddly, it is only in that environment that it appears incorrectly---in regular HTML loaded inside a regular browser tab, it shows correctly.

	Boris Zbarsky [:bzbarsky]
	Comment 5 • 13 years ago

Does the alert you get in the editor look the same as the regular HTML one?  Or is it a separate dialog instead of a tab-modal box?

	Boris Zbarsky [:bzbarsky]
	Comment 6 • 13 years ago

What happens in HTML styled with "font: menu"?

	Brett Zamir Reporter
	Comment 7 • 13 years ago

The alert in the editor says "[JavaScript application]" in the titlebar and its text is conveniently copy-pasteable as in Firefox 3.6 (unlike Firefox 4 now).

 * {font:menu} indeed makes the HTML messed up like the alert (whether applied in a tab or the editor).

	Boris Zbarsky [:bzbarsky]
	Comment 8 • 13 years ago

OK, yeah.  In the editor you're getting a modal dialog alert, not a tab-modal alert.  And they use different font styles.

Sounds like something is up with our interaction with your "menu" font.  Any idea what font that is?  ;)

	Brett Zamir Reporter
	Comment 9 • 13 years ago

Ok, the "winner" is: "Segoe UI" (Windows 7)

	Jonathan Kew [:jfkthame] Assignee
	Comment 10 • 13 years ago

Ugh... I suspect this is a regression from bug 635639. We may have to try and make the patch there more selective in its effects.

	Jonathan Kew [:jfkthame] Assignee
	Comment 11 • 13 years ago

The problem here is that the patch for bug 635639 was not quite right (although simply reverting it wouldn't fix things, as that leaves a gap where the diacritic occurs due to its non-zero advance).

In hb_ot_layout_position_finish(), it was incorrect to touch the diacritic's x_offset value, because at this stage GPOS::position_finish() has not yet been called. So all we need to do here is zero the advance width, and then let position_finish() do the proper attachment.

	John Daggett (:jtd)
	Comment 13 • 13 years ago

Comment on attachment 530268 [details] [diff] [review]
reftest for diacritic in Segoe UI

Might be nice to have a font that simply replicates the problem found in Segoe UI so that we would catch this on all platforms.  But I'll let you judge whether that's worth the effort or not.

	Jonathan Kew [:jfkthame] Assignee
	Comment 14 • 13 years ago

http://hg.mozilla.org/mozilla-central/rev/800a2d46bb4e (patch)
http://hg.mozilla.org/mozilla-central/rev/427fa3603d4c (testcase)

We could create a custom font that replicates the problem, but I don't see it as a high priority right now; as long as we're running tests on Win7, at least, we'll know if we regress this (e.g. when taking a harfbuzz update). If we happen to run across a free font that exhibits the same issue, we could add that to the testcase.