385327 - Text not rendered correctly if lang is missing or incorrect

Status: RESOLVED INVALID

(Core :: Graphics, defect)

Description

[Marco Pesenti Gritti]

This is a regression from Firefox 2.0. I'm going to add 3 testcases to demonstrate the issue.

Comment 1

[Marco Pesenti Gritti]

Attachment: Urdu page with correct lang property

Comment 2

[Marco Pesenti Gritti]

Attachment: Urdu page with incorrect lang property

Comment 3

[Marco Pesenti Gritti]

Attachment: Urdu page with no lang property

Comment 4

[Marco Pesenti Gritti]

Firefox 2.0.0.4 (from Fedora 7) renders all the testcases correctly.

Self compiled firefox trunk (on Fedora 7) renders attachment 269196 [details] but screw up all the fonts for attachment 269189 [details] [diff] [review] and 269199.

Comment 5

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

Sounds like a font selection problem. We need to port Stuart's Windows font selection code to Linux (and Mac too).

Comment 6

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

The cause is same as bug 367884.

Comment 7

[Behdad Esfahbod]

Can someone comment on how Stuart's font selection code works?

And do you see any way to fix this soon?  One way would be to call pango_itemize() and let that choose the font.

Comment 8

[Behdad Esfahbod]

Current plan is for Vlad to come to Toronto early August so we can fix this together.

Comment 9

[Jeff Muizelaar [:jrmuizel]]

Jonathan, is this fixed now?

Comment 10

[Jonathan Kew [:jfkthame]]

(In reply to Jeff Muizelaar [:jrmuizel] from comment #9)
> Jonathan, is this fixed now?

It's dependent on the font that fontconfig gives us. On my Ubuntu box, the results are horrible for the page that's correctly tagged with lang="ur"; for lang="en" or no language, they're somewhat better but still broken.

In my case, fontconfig makes a particularly bad choice:
  $ fc-match :lang=ur
  LucidaSansRegular.ttf: "Lucida Sans" "Regular"
I suppose it picks this because LucidaSans includes the Unicode characters needed; however, looking into the OpenType tables, I see that they only provide the OT features needed for Devanagari, not the Arabic shaping features. So the result is that we get unshaped text - no joining forms, etc. Basically unreadable.

This isn't really a Firefox bug, though, it's an inadequate font (why include the Arabic-script characters if you're not going to provide the shaping features to support them?) combined with a poor fontconfig configuration. No doubt it could be improved by hacking a ~/.fonts.conf file or something like that.

On the pages that *aren't* tagged with lang="ur", I get somewhat better results, because
  $ fc-match :lang=ar
  DejaVuSans.ttf: "DejaVu Sans" "Book"
and DejaVu has Arabic shaping support, though it's not quite perfect for Urdu.

Comment 11

[Jeff Muizelaar [:jrmuizel]]

This is a distro bug that I don't think we should try to work around.