Thai Script Shaping

Problem Definition

Thai script is composed of multiple levels of stacking characters. On the base line are the consonants and some leading or following vowels. Then, the upper or lower vowel may combine above or below the consonant. And the stack can then be finalized by tone mark or diacritic.

In summary, a Thai grapheme cluster can be one of the forms:

LV|FV
C [BV|AV] [T|AD]

where:

LV = Leading Vowel
FV = Following Vowel
C = Consonant
BV = Below Vowel
AV = Above Vowel
T = Tone Mark
AD = Above Diacritic

The first form is simple: just individual character. It's the second form that the shaping rules must deal with.

Typewriter Positioning

Primitive typesetting as used in mechanical typewriters and framebuffer consoles is to put characters in fixed vertical levels to prevent overlapping:

Base Level for C
Below-base Level for BV
Above-base Level for AV and certain AD
Top Level for T and some AD

And these are the default positions for glyphs in most Thai fonts, where bare combining without shaping still yields a readable, though suboptimal, rendered text.

Consonant Classification for Shaping

Some Thai consonants have extra ascender or descender which can overlap the combining character. Some rearrangement needs to be done to avoid this. Based on such rearrangement operations, Thai consonant can be classified into 4 classes:

Normal consonants (NC), without extra ascender/descender
Consonants with right extra ascender (AC), namely ป (PO PLA, U+0E1B), ฝ (FO FA, U+0E1D), ฟ (FO FAN, U+0E1F) and in some fonts ฬ (LO CHULA, U+0E2C)
Consonants with removable descender (RC), namely ญ (YO YING, U+0E0D) and ฐ (THO THAN, U+0E10)
Consonants with strict descender (DC), namely ฎ (DO CHADA, U+0E0E) and ฏ (TO PATAK, U+0E0F)

Shaping Rules

The general rules are:

SARA AM (U+0E33) must be decomposed into NIKHAHIT (U+0E4D) and SARA AA (U+0E32). And if a tone mark (T) is present before it, the NIKHAHIT must be reordered so it comes before the tone mark.
If tone mark (T) is present without upper vowel (AV), it must be lower down, with probably scaled up size.
Any above-base combining mark (T, AV, AD) that combines to consonant with extra ascender (AC) must be shifted left.
If below vowel (BV) combines to consonant with strict descender (DC), it must be lowered down.
Descender of consonant with removable descender (RC) must be removed when combined with below vowel (BV), and the BV needs not be shifted.

This can be summarized as following table:

base \ comb	AV^*	BV	T^**	{AV}T^**
NC	-	-	SD(c)	-
AC	SL(c)	-	SDL(c)	SL(c)
RC	-	RD(b)	SD(c)	-
DC	-	SD(c)	SD(c)	-

^* MAITAIKHU (U+0E47), NIKHAHIT (U+0E4D) and YAMAKKAN (U+0E4) are treated as AV here.

^** THANTHAKHAT (U+0E4C) is treated as T here.

where:

SD() = Shift Down
SL() = Shift Left
SDL() = Shift Down Left
RD() = Remove Descender

c in the parameter means the combining mark, and b means the base consonant.

PUA Shaping

Before OpenType, some ad hoc shaping solutions had been developed by vendors, all of which were logically the same positioning-by-substitution technique with the same extra glyph sets, but unfortunately with different code point assignments. There were two major extensions, one for Microsoft Windows, and the other for Apple MacOS. The discrimination had barred the fonts for both platforms from being interchanged, as fonts were tied to vendors' rendering engines.

To support those legacy fonts, which still dominate the market at present, rendering engines should know how to access those private glyphs and substitute them properly.

There were 2 sets of each vendor extension, one for 8-bit pre-Unicode fonts, and the other for Unicode Private Use Area (PUA). 8-bit fonts are rare nowadays, but let us mention them here for historical reference.

Glyph	8-bit Windows	Windows PUA	8-bit Mac	Mac PUA
Low MAI EK	0x8B	U+F70A	0x88	U+F88B
Low MAI THO	0x8C	U+F70B	0x89	U+F88E
Low MAI TRI	0x8D	U+F70C	0x8A	U+F891
Low MAI CHATTAWA	0x8E	U+F70D	0x8B	U+F894
Low THANTHAKHAT	0x8F	U+F70E	0x8C	U+F897
Low-left MAI EK	0x86	U+F705	0x83	U+F88C
Low-left MAI THO	0x87	U+F706	0x84	U+F88F
Low-left MAI TRI	0x88	U+F707	0x85	U+F892
Low-left MAI CHATTAWA	0x89	U+F708	0x86	U+F895
Low-left THANTHAKHAT	0x8A	U+F709	0x87	U+F898
Left MAI EK	0x9B	U+F713	0x98	U+F88A
Left MAI THO	0x9C	U+F714	0x99	U+F88D
Left MAI TRI	0x9D	U+F715	0x9A	U+F890
Left MAI CHATTAWA	0x9E	U+F716	0x9B	U+F893
Left THANTHAKHAT	0x9F	U+F717	0x9C	U+F896
Left MAI HAN-AKAT	0x98	U+F710	0x92	U+F884
Left SARA I	0x81	U+F701	0x94	U+F885
Left SARA II	0x82	U+F702	0x95	U+F886
Left SARA UE	0x83	U+F703	0x96	U+F887
Left SARA UEE	0x84	U+F704	0x97	U+F888
Left MAITAIKHU	0x9A	U+F712	0x93	U+F889
Left NIKHAHIT	0x99	U+F711	0x8F	U+F899
Low SARA U	0xFC	U+F718	0xFC	U+F89B
Low SARA UU	0xFD	U+F719	0xFD	U+F89C
Low PHINTHU	0xFE	U+F71A	0xFE	U+F89D
Desc-less YO YING	0x90	U+F70F	0x90	U+F89A
Desc-less THO THAN	0x80	U+F700	0x80	U+F89E

Notes

8-bit Windows is an extension to CP874.
8-bit Mac is based on MacOS Thai Codepage. Most fonts extend it by adding glyphs that are available in Windows.
Unicode PUAs are extracted from vendor fonts.

OpenType Shaping

Modern fonts should no longer rely on PUA, as modern software begins to support OpenType more widely.

Please refer to Spec for Thai OpenType Font Creation for the details.