[XeTeX] Kerning flaw with MinionPro & XeTex

[Jonathan Kew]

Adam,

Thanks for the detailed analysis and testcase.

It looks like this is an instance of http://bugs.icu-project.org/trac/ticket/7753.

JK

On 7 Feb 2011, at 01:31, Adam Twardoch (List) wrote:

> I have identified this issue as a serious bug in XeTeX.
> 
> In MinionPro-Regular.otf (version 2.068 is the one I have, but it's
> likely that it applies to all versions), the relevant kerning definition
> excerpt looks like the following:
> 
> feature kern {
>  lookup kern1 {
>      pos uni0423 uni0431.ital -53;
>    subtable;
>      pos [uni040E uni0423 uni0474] [uni0435 uni043E uni0441
>      uni0444 uni0451 uni0454 uni0473 uni0431.ital] -118;
>   } kern1;
> } kern;
> 
> This means that there is an individual kern value between uni0423
> uni0431.ital with the value of -53, then a subtable break, then a
> class-based value for the same pair of -118.
> 
> Adobe InDesign CS5 correctly identifies the pair in the first subtable
> (-53) as the one that it should apply, and ignores the second value. So
> the kern comes out as -53.
> 
> It very much looks like XeTeX *incorrectly* parses *both* subtables, and
> applies a sum of both the individual and the class value (-53-118 =
> -171), which results in a very deep kern, and therefore an overlap.
> 
> This seems to be a serious, systematic bug in the way XeTeX (or its
> underlying ICU Layout library) processes the GPOS table.
> 
> I've made an isolated test case that confirmes it.
> 
> I've created a test OpenType font. It defines a kerning pair A B -300, a
> subtable break, and again the same pair. InDesign correctly applies only
> the first value (-300), which is shown in the
> SubtableKernTest-InDesign.pdf file. XeTeX applies both values
> consecutively, which is shown in SubtableKernTest-XeTeX.pdf
> 
> All the test files (.ttf, .fea feature definitions, .tex and two PDFs)
> are available at
> http://www.twardoch.com/tmp/SubtableKernTest.zip
> 
> I've also sent the test files offline to Jonathan Kew.
> 
> Many thanks to Alessandro Ceschini for raising the issue.
> 
> Regards,
> Adam
> 
> 
> 
> --------------------------------------------------
> Subscriptions, Archive, and List information, etc.:
>  http://tug.org/mailman/listinfo/xetex