Contextual substitution of a mark glyph

[murasu]

I'm trying to contextually substitute an above based mark with the following GSUB code in psts:
sub ka-tamil iMatra-tamil' pa-tamil by iMatra-tamil.conn1;

The substitution works in CoreText but not in Harfbuzz. The image below will show and compare the outcomes. Green is CoreText.

I'm attaching the font as well. Is there anything I need to do different for HB to render this rule?

MarkSubTest-Light.otf.zip

[murasu]

I did some more tests. The only way I could get the substitution work on Harfbuzz is to add an empty sustitution to the chain as below:

script tml2;

lookup subst_imatra {
	sub iMatra-tamil by iMatra-tamil.conn1;
} subst_imatra;

lookup do_nothing {
} do_nothing;


sub ka-tamil' lookup do_nothing iMatra-tamil' lookup subst_imatra pa-tamil;

This works across CoreText and Harfbuzz ... but just for two glyphs in a sequence. Still puzzled :(

[khaledhosny]

It works if you put the substitution in calt, so I suspect it is because psts feature is applied per syllable. I don’t understand why adding more input makes it work across syllable boundaries, though.

[behdad]

If anything it should be the other way around. We do match backtrack/lookahead sequences across grapheme boundaries always IIRC.

cc @jfkthame

[murasu]

Yes, the substitution works if it's in calt or rclt without more inputs.

[khaledhosny]

FWIW, DirectWrite and Uniscribe seems to not apply the substitution regardless of the feature tag used.

[behdad]

Interesting. They probably match backtrack/lookahead within syllable only.

[behdad]

Can you attach the working font as well? Thanks.

[murasu]

Here's the working font.

MarkSubTest-Light 2.otf.zip

[behdad]

I see a bug...

[behdad]

I'm working on it in #4510