Unicode Footguns in Python

Umm, you’re confusing things here. When you said “several of the standard grapheme clustering algorithms” that indicated the first element of confusion. There is one grapheme clustering algorithm in Unicode.¹ Ligature combinations, whether encoded in Unicode or not are independent of the concept of graphemes and are very much font-dependent, as is whether or not a grapheme cluster is displayed as multiple characters or a single character (any two consecutive regional indicators are considered a single grapheme, whether they map to a national flag or not and the set of national flags and whether those flags are even displayed or the output is just the regional indicator symbol is up to the font). Normalization does not play a role in grapheme clustering at all, which is why á and a+acute accent are both considered single graphemes.

I’m puzzled about the assertion of EBCDIC having Latin ligatures because I never encountered them in my IBM mainframe days and a casual search didn’t turn any up. The only 8-bit encoding that I was aware of that included any ligatures was Mac extended ASCII which included fi and fl (well, putting aside TeX’s 7-bit encoding which was only used by TeX and because of its using x20 to hold the acute accent was unusable by other applications which expected a blank character in that spot).

The question about dealing with ligatures for non-Latin scripts generally came down to compatibility with existing systems more than anything else. This is why, for example, the Devanagari and Thai encodings which both have vowel markings which can occur before, after or around a consonant handle the sequence of input differently. Assembly of jambo into characters in Hangul is another case where theoretically, one could have all the displayed characters handled through ligatures and not encode the syllables directly (which is, in fact, how most modern Korean is encoded as evidenced by Korean wikipedia), but because the original Korean font encoding has all the syllables encoded in it² those syllables are part of Unicode as well.

But the bottom line here, is that you seem to be confusing terminology a lot here. You can very much do grapheme clustering without paying attention to ligatures, and rules for things like normalized composition/decomposition are entirely independent of grapheme clustering (the K-forms of both, manage transitions like ¹ to 1 or fi to fi and represent a one way transition).

⸻

1. I wrote a Rust library implementing this and I’ve been following Unicode since it was a proposal from Microsoft and Apple in the 90s, so I know a little about this.

2. I think the more accurate term would be “most” of the syllables as I’ve seen additional syllables added to Unicode over time.