On Wed, Oct 31, 2018 at 4:03 PM luigi scarso <luigi.scarso@gmail.com> wrote:

On Wed, Oct 31, 2018 at 4:01 PM Hans Hagen <j.hagen@xs4all.nl> wrote:
On 10/31/2018 3:38 PM, luigi scarso wrote:

> On Tue, Oct 30, 2018 at 11:22 AM luigi scarso <luigi.scarso@gmail.com
> <mailto:luigi.scarso@gmail.com>> wrote:
>
>
>
> On Tue, Oct 30, 2018 at 10:18 AM Ulrike Fischer <news3@nililand.de
> <mailto:news3@nililand.de>> wrote:
>
> According to the xelatex output in the following example both
> variants are not correct. What would be needed would be the first
> two glyphs from variant 1 (knd2 script) and the third from variant 2
> (knda script).
>
>
> dunno, but to start with at least I visually see what's going on:
>
> \showotfcomposition{file:notosanskannada-regular.ttf*kannada-testA
> at 48pt}{-1}{ಕ್ರ}
>
>
> \showotfcomposition{file:notosanskannada-regular.ttf*kannada-testB
> at 48pt}{-1}{ಕ್ರ}
>
> support for knd2 is still in progress,
> and in knda
> glyph ಕ (U+00C95) + glyph ್ (U+00CCD) + glyph ರ (U+00CB0)
> also is not ok ( at least for what I understand until now).
> I see where it fails ( and where hb does it right), let's see if I am
> able to find a patch.

Well..not so easy (at least for me , never seen devanagari before )

glyph ಕ (U+00C95) + glyph ್ (U+00CCD) + glyph ರ (U+00CB0)

are

KA + VIRAMA + RA

and it is quite complex case (well, really it could be a moderately complex case ...)

For Kannada script we can start by considering

http://brahmi.sourceforge.net/docs/KannadaComputing.html

it's an old doc but

1) it shows the rules for kannada

2) its says that
"""

2.5 Rendering Rules

(Based on Microsoft Uniscribe-OpenType implementation of the UNICODE Rendering Rules)

"""

So Uniscribe is the reference for the rendering rules of this doc, which in turn is the base for unicode.

Today the reference unicode is

http://www.unicode.org/versions/Unicode11.0.0/ch12.pdf

The rules are a bit different from http://brahmi.sourceforge.net/docs/KannadaComputing.html

but at page 500 we see

U+0C95 ಕ ka + U+0CCD ್ halant + U+0CB0 ರ ra → ಕ್ರ kra

which is our case (unicode says that the preferred name for VIRAMA is halant, so halant is also frequently used )

What happens is explained at page 464 for devanagari:

we should consider KA + VIRAMA + RA as

KAn + VIRAMAn + RAi, so

We have KAn + VIRAMAn , hence

KAn + VIRAMAn → KAd

R6:

We have KAd + RAi, hence

KAd + RAi → KAn + RAsub

and then

R13

KAn + RAsub → K.RAn

and indeed ಕ್ರ is K.RAn

All these rules are re-explained at

https://docs.microsoft.com/en-us/typography/script-development/kannada#introduction

and font-osd.lua is based exactly on these rules.

It makes sense, given that, as seen, these are the reference.

The functions for knda are those ending in *_one and for what I have seen

after that the base syllable is correctly isolated, reorder_one doesn't implement R6 and R13 .

I am quite sure that spending a bit more time I can isolate the point,

but, as Hans has said, there is the plan to review these scripts.

luigi

PS:

Take these notes cum grano salis: devanagari is not exactly equal to kannada,

and the ms site uses its own terms, not always one finds a match with unicode.