Microsoft provides some information on what (opentype) features a word processor should support by default for certain scripts.
Caveat: Just because a feature is documented and looks useful does not mean Uniscribe will use it for your script. Many Latin script fonts would like to use 'init', 'medi', 'calt', etc. but none of these features is turned on by Uniscribe for Latin.
Caveat: Just because Uniscribe supports a feature that does not mean any given application will. Uniscribe (as of 2005) supports 'liga' for latin, but neither Word nor Office does.
Caveat: Uniscribe (MS unicode text layout routines) may ignore either the GPOS or the GSUB table depending on the script, and may even refuse to use the font at all if it doesn't have the right stuff in GPOS/GSUB. A Hebrew font must have both a GPOS and a GSUB. If it doesn't the font is not used. A latin font need not have either, but if it doesn't have GSUB then GPOS won't be used. So now if one table is present and the other isn't, FontForge will generate a dummy version of the other.
Many characters are used in more than one script. The digits, marks of punctuation, etc. are said in Unicode to belong to the script "Common". OpenType does not recognize this script. The closest thing it has is the script 'DFLT'. My understanding is that characters in the common script will have the script of adjacent text assigned to them by OpenType.
Thus if a font supports latin, greek or cyrillic then digits and punctuation might find themselves in any of those three scripts and all features which apply to any such character should be present in all scripts. For example if the digit 9 kerns with digit 1, then that kerning data should be present in cyrillic and greek as well as latin.
However it is possible (and I gather common in Japanese) to use the digits from one font surrounded by Kanji characters from another font. This means the digits may be in a font which does not support Kanji. However OpenType will assign them the Kanji script. Thus no lookups will be applied. Adobe suggests that for most features in scripts should also appear in the fallback script 'DFLT'. I'm not sure that anyone else follows this convention.
Uniscribe supports the following features
There are not many special complications in latin. Latin fonts can generally fit in a single byte encoding with no (or few) font tables. There are a plethora of accented glyphs which could be built -- or you could use mark to base positioning. Kerning should be generated for some glyph combinations. A few ligatures need to be generated (the "f" ligatures: ff, fi, fl, ffi, ffl and perhaps st -- however for some languages (Turkish) the "fi" ligature should not be built).
You may want to add a set of smallcaps. Adobe has reserved a block in the private use area for latin small-caps -- this is now depreciated.
Some languages have specific requirements of their own
Uniscribe supports the following features
Greek also does not have many complications. Modern Greek fonts generally fit in a single byte encoding. For modern greek are a few accented glyphs that need to be built, while for polytonic greek there are many -- mark to base & mark to mark are options. Kerning should be generated. I am not aware of any standard ligatures for modern greek (ancient greek had ligatures and variants on some of the glyphs though).
Small caps are again an option, and I have reserved a block in the private use area for them -- again this is depreciated.
Uniscribe supports the following features
Cyrillic fonts also fit in a single byte encoding. There are a few accented glyphs. Kerning should be generated. I am not aware of any standard ligatures.
Some languages need variant glyphs (specified with a 'loca' feature)
Uniscribe supports the following features
Arabic needs a complete set of initial, medial, final and isolated forms -- Unicode has reserved space for these. Arabic also needs a vast set of ligatures -- Unicode has reserved space for many, but I'd guess that extra ligatures will be needed sometimes. Arabic also needs a set of mark (mark to base, mark to ligature) attachments to position vowels above letters. Arabic may need a glyph decomposition table.
I'm told that in good Arabic typography there need to be many more than 4 forms per glyph. I'm not sure how this should be supported.
Right to left.
Uniscribe supports the following features
Hebrew has a few final forms but no special tables are needed for these. Hebrew may need kerning. Hebrew should have a set of mark (mark to base) tables to position vowels over letters. Hebrew may need a glyph decomposition table. I am not aware of any needed ligatures.
Right to left
Uniscribe supports the following features
Indic scripts need a set of ligatures.
(they probably need a lot more but I'm not aware of what)
Uniscribe supports the following features
The Hangul script consists of a set of syllables built out of a phonetic alphabet. Generally fonts consist of a set of precomposed syllables.
Complications are introduced by the massive combinatorial explosion of all these syllables. These are eased in postscript by CID-keyed fonts.
Vertical writing and left to right writing are used, and some glyphs have a different orientation when drawn vertically (parentheses for example).
MicroSoft does not describe these scripts (that I can find).
Again a massive collection of glyphs is needed, and postscript uses CID keyed fonts to deal with this.
Vertical writing and left to right writing are used, and some glyphs have a different orientation when drawn vertically (parentheses for example).