This is a leftover of bug 118000. I did not modify function ResolveBackwards to support surrogates pair. It should be fixed once 118000 get in.

simon, could you review my patch? Logic is very similar to ResolveForward.

With this patch the test case at http://www.geocities.com/i18nguy/unicode-example-plane1.html still doesn't display correctly on my system. The Etruscan characters are all displayed as ?????

simon, I only want to address ResolveBackwards in this bug. I set a breakpoint at this function and visited that page, it never stopped there. (Did you once mentioned that there are more than one options to do right-to-left rendering?) I guess the problem is because we did not group surrogate pair together when rendering those characters. I filed a new bug 122800 and assign it to you.

On second thoughts, maybe either this patch is unnecessary, or else there is a problem with the logic that calls ResolveForwards or ResolveBackwards. We call ResolveBackwards when mRightToLeftText is true, and mRightToLeftText is set when we display right-to-left characters on a platform that does its own bidi reordering, i.e. when we are not reversing the text ourselves but passing it in logical order to the platform rendering APIs. Was that the intention? There are no plane 1 or higher characters with right-to-left directionality, so until that changes, ResolveBackwards doesn't need to know how to handle them.

Do we only use ResolveBackwards with characters with right-to-left directionality? How about those Etruscan characters after they do get into plane 1? If you are sure about the answers, we can future this. Otherwise, we might want to just check it in. There is not significant negative impact after all.

Only *if* the Bidi category of the Etruscan characters in UCDB changes from L to R (which seems possible, if it is correct that Etruscan was historically written from right to left), and *if* operating systems start honoring this (which seems rather unlikely), will we need to treat them in the same way as other right-to-left scripts like Arabic and Hebrew. I think futuring this is the way to go.

Give this to simon, who know when this is needed if it is ever needed at all.

It's time to fix this: there are now RTL characters in plane 1 (Cypriot syllabary) and at least one font is available for them at http://home.att.net/~jameskass/code2001.htm

This brings the previous patch in line with the current version of ResolveForwards and adds proper surrogate handling in nsBidiPresUtils::CalculateCharType. It still needs cleaning up and porting to other platforms that implement ResolveBackwards.

I patched nsFontMetricsOS2FT::ResolveBackwards as well (for consistency with ResolveForwards, even though OS/2 doesn't display non-plane 0 characters).

Comment on attachment 169903 [details] [diff] [review] Patch ready for testing mkaply told me on IRC that the patch builds on OS/2 and doesn't break anything.

Comment on attachment 169903 [details] [diff] [review] Patch ready for testing r+sr=rbs

Patch checked in.