Different behaviors of glyph connections when typesetting mixed languages: comparing latin+Arabic and latin+Mongolian

[mongol-surah]

Hello, the SILE community!

I was trying to typeset some Mongolian texts using SILE. When a Mongolian word is preceded by a Latin character like this: \font[family="Mongolian White"]{Aᠳᠤᠷ᠎ᠠA}, the connectivity between Mongolian characters is broken: . This is fair, since that's what harfbuzz will output, if the string "Aᠳᠤᠷ᠎ᠠA" as a whole is sent to harfbuzz.

However, the breaking of connectivity does not happen for the latin+arabic mixture. The instruction \font[family=Arial,size=16pt,direction=rtl]{AفـانيA} produces . The characters are connected very well. Harfbuzz doesn't produce connected glyphs (by design) if the whole string "AفـانيA" is fed in. Harfbuzz outputs this: where the characters are not connected to each other.

My question is, does SILE do some itemizations for Arabic input before sending it to harfbuzz, while doesn't do itemization for Mongolian input? Why does SILE show different behaviors for the connectivities of Arabic and Mongolian inputs? Thanks!!

[mongol-surah]

I rerun the examples with -d tokenizer. Below are the sources for the Arabic document and the Mongolian document, and their debug information.
Arabic:

\begin[papersize=a5]{document}
\font[family=Arial]{AفـانيA}
\end{document}

Debug info:

SILE v0.14.8 (Lua 5.4)
<arabic.sil> as sil

[tokenizer] Token: A
[tokenizer] Token: فـاني
[tokenizer] Token: A[1] 
! Font family 'Gentium Plus' not available, falling back to 'Arial' at in \par

[tokenizer] Token: 1

Mongolian:

\begin[papersize=a5]{document}
\font[family="Mongolian White"]{Aᠳᠤᠷ᠎ᠠA}
\end{document}

Debug info:

SILE v0.14.8 (Lua 5.4)
<mongolian.sil> as sil

[tokenizer] Token: Aᠳᠤᠷ᠎ᠠA[1] 
! Font family 'Gentium Plus' not available, falling back to 'Arial' at in \par


[tokenizer] Token: 1

It is apparent that the tokenization mechanism in SILE treats Mongolian and Arabic inputs differently, as mentioned in #1726.