496527 - Extra space at intervals in long string with small letter-spacing

Status: NEW

(Core :: Graphics, defect)

Description

[Amir Aharoni]

User-Agent:       Mozilla/5.0 (Windows; U; Windows NT 5.1; he; rv:1.9.1b4) Gecko/20090423 Firefox/3.5b4 GTB5 (.NET CLR 3.5.30729)
Build Identifier: Mozilla/5.0 (Windows; U; Windows NT 5.1; he; rv:1.9.1b4) Gecko/20090423 Firefox/3.5b4 GTB5 (.NET CLR 3.5.30729)

It looks like Firefox adds some extra pixels to a long Hebrew string with niqqud (vowel points) in certain style setup. I noticed it in the default style of Wikipedia, which many users may see.

I entered a long string consisting of the letter waw (ו) with a dagesh after it (ּ) repeated over a hundred times; combined they are called "shuruq" and look like this וּ. Every 53 shuruqs there is an extra pixel.

It also happens with other letters and vowel signs - every 53 letters there's an extra pixel. (Here, "53 letters" means "53 letters, not counting the vowel points".)

It doesn't happen in IE and Chrome.

Such long strings with niqqud are rare, but may happen.

Reproducible: Always

Steps to Reproduce:
1. Enter a long string of Hebrew characters with vowel points (niqqud) into a Wikipedia page.
2. Preview or save the page.
Actual Results:  
There's an extra pixel every 53 letters.

Expected Results:  
The characters should be evenly distributed.

Comment 1

[Simon Montagu :smontagu]

When you say "extra pixel", do you mean extra blank pixel, i.e. extra space between characters? That's what I'm seeing in the URL on Linux (Ubuntu 9.04), but not on OSX. I'm guessing it's some kind of rounding error with glyph positioning.

Comment 2

[Amir Aharoni]

Attachment: appearance of extra column of blank pixels on a Wikipedia page in Windows XP

Simon, there's an extra column of blank pixels. I am attaching an image.

Comment 3

[Simon Montagu :smontagu]

Attachment: Minimized testcase

I am seeing the same issue without niqqud and even in Latin script. Reducing the css shows that |letter-spacing: 0.001em;| seems to be causing the problem.

Comment 4

[Amir Aharoni]

(In reply to comment #3)
> I am seeing the same issue without niqqud and even in Latin script. Reducing
> the css shows that |letter-spacing: 0.001em;| seems to be causing the problem.

Silly me. I am so used to problems with niqqud in Mozilla and other programs that i was sure that niqqud causes it. Feel free to rename the bug accordingly.

Comment 5

[Simon Montagu :smontagu]

This looks like the issue described in bug 368107 comment 0:

"Gecko inserts the space if the accumulation spaces are over a device pixel"

Comment 6

[Robert O'Callahan (:roc) (email my personal email if necessary)]

This should only happen on Windows. Fixing bug 368107, rounding letter-spacing on Windows only, would fix this.

I'm not sure why the author is using letter-spacing:0.001em though. To disable ligatures?

Comment 7

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Hmm, but Simon saw this on Ubuntu. I thought Pango supported horizontal subpixel positioning?

Comment 8

[Simon Montagu :smontagu]

(In reply to comment #6)
> I'm not sure why the author is using letter-spacing:0.001em though. To disable
> ligatures?

It comes from http://he.wikipedia.org/w/index.php?title=%D7%9E%D7%93%D7%99%D7%94_%D7%95%D7%99%D7%A7%D7%99:Common.css. There's a comment in Hebrew saying "reduce space between letters in order to (partially) solve the diacritics problem", which doesn't make sense to me.

Comment 9

[Amir Aharoni]

It is probably related to this:

* http://he.wikipedia.org/wiki/%D7%95%D7%99%D7%A7%D7%99%D7%A4%D7%93%D7%99%D7%94:%D7%A0%D7%99%D7%A7%D7%95%D7%93

* http://he.wikipedia.org/wiki/%D7%95%D7%99%D7%A7%D7%99%D7%A4%D7%93%D7%99%D7%94:%D7%A0%D7%99%D7%A7%D7%95%D7%93/%D7%91%D7%A2%D7%99%D7%95%D7%AA

The place to ask for more details is, of course, http://he.wikipedia.org/wiki/MediaWiki_talk:Common.css

Comment 10

[Tomer Cohen :tomer]

Attachment: screenshot (gimp xcf format)

I was able to reproduce similar issue on the Noami Shemer page on Wikipedia http://he.wikipedia.org/wiki/%D7%A0%D7%A2%D7%9E%D7%99_%D7%A9%D7%9E%D7%A8.

Attached a two-layers image in order to show the differences.

Comment 11

[Aharon]

This issue was brought up on Chromium's bug tracking list and was fixed. Perhaps there's something to be learned from their fix? http://code.google.com/p/chromium/issues/detail?id=31902

Comment 12

[Aharon]

I created a test page showing screenshots of this bug across a number of browser/OS configurations: http://aharon.varady.net/browser-test-for-hebrew-diacrits.html

Comment 13

[Aharon]

I'm almost certain that Wikipedia and affiliated wikimedia sites won't display this problem, or at least not consistently, since they use NFC (Normalization Form C) -- Canonical Decomposition, followed by Canonical Composition. In this case, that means consonant followed by a dagesh decomposed to a combo consonant+dagesh character in the Unicode.

Comment 14

[Aharon]

On further analysis, the problem appears to stem from certain changes made to the font after being processed from ttf to woff by font squirrel. For some reason, Chrome could cope with those changes and FF could not.

Comment 15

[Simon Montagu :smontagu]

Can you open a new bug for comments 11-14? It's not the same issue that the original bug report is about.