On Sat, Oct 13, 2018 at 11:05:01AM +0200, Thomas A. Schmitz wrote:
You're right, this shouldn't happen. I tried in vain to find the culprit in lang-agr.lua and to see more with
\enabletrackers[hyphenator.visualize,hyphenator.steps,languages.patterns]
failed. Arthur is the guru here, so maybe he has a suggestion?
Ah, I was going for a title that inspired more awe, like “Emperor of Hyphenation”, but guru will do for the time being :-) The reason you can’t find any obvious culprit is because you need to look at the patterns that are missing: taking πράγματσς as an example, the matching patterns are .π4 ά1 α1 ο1 4ς. that allow a break after any vowel and prohibit breaks after the first letter and before the last letter in the word (which is relevant because \lefthyphenmin and \righthyphenmin are both set to 1). Since these are the only patterns that apply, the possible hyphenation points are thus πρά-γμα-τος Digging deeper in the pattern file, you’ll see that it takes quite many two-consonant clusters into account, starting at line 267 of the master file in the repository, with a comment “other divisable consonant combinations” (https://github.com/hyphenation/tex-hyphen/blob/82e5651/hyph-utf8/tex/generic...): 2β1γ 2β1ζ 2β1θ 2β1κ ... 2γ1θ 2γ1κ 2γ1ξ 2γ1π ... You can see that γμ is not there (nor, of course, γν, which was expected). If it was, the pattern 2γ1μ would force the break πράγ-μα, hence its absence leads me to believe that the breaks before γμ are intentional. I suggest you contact Dimitrios Filippou, the main author of the patterns, to ask if it was somehow an oversight or if he was following a different rule (email address at the top of the file linked to, in a slightly obfuscated form). Best, Arthur