Move Sinhala to USE

[behdad]

See #967 for some context. Sinhala shaping apparently changed between Win7 and Win10. We used to match Win7 behavior. Then hacked our shaper to be closer to Win10. Moving to USE should be a better long-term fix.

[behdad]

Jonathan and I looked into this. Pushed a few commits into branch, but hit a wall. Reported to Andrew Glass:

Sinhala uses explicit reph formation. Ie, Ra,H,ZWJ forms a reordering reph form. However, MS iskpota fonts seem to ligate both Ra,H and Ra,H,ZWJ in the rphf feature, only one of which should be reordered. Detecting this basically would introduce a special-casing hack in the USE, which is not documented in the spec.

Stalled for now.

[behdad]

https://github.com/harfbuzz/harfbuzz/commits/use-sinhala

[Richard57]

I believe U+0DCA SINHALA SIGN AL-LAKUNA needs to be added to the same new USE category as U+11046 BRAHMI VIRAMA, for it is the final component of U+0DDA SINHALA VOWEL SIGN DIGA KOMBUVA and U+0DDD SINHALA VOWEL SIGN KOMBUVA HAA DIGA AELA-PILLA.

[behdad]

Jonathan and I looked into this. Pushed a few commits into branch, but hit a wall. Reported to Andrew Glass:

Sinhala uses explicit reph formation. Ie, Ra,H,ZWJ forms a reordering reph form. However, MS iskpota fonts seem to ligate both Ra,H and Ra,H,ZWJ in the rphf feature, only one of which should be reordered. Detecting this basically would introduce a special-casing hack in the USE, which is not documented in the spec.

@xadxura can you advise? What does DWrite implementation do?

[khaledhosny]

What do we need to unblock this issue?