[luatex] Problem with babel-greek (harfbuzz?)

[Nikos Platis]

Στις Τρί 17 Οκτ 2023 στις 4:16 μ.μ., ο/η Nikos Platis
<nplatis at gmail.com> έγραψε:
>
>
>
> Στις Τρί 17 Οκτ 2023 στις 12:20 μ.μ., ο/η Ulrike Fischer <luatex at nililand.de> έγραψε:
>>
>> Am Tue, 17 Oct 2023 09:51:55 +0300 schrieb Nikos Platis:
>>
>> > Dear all,
>> >
>> > I came across a rather strange issue while using lualatex with babel-greek.
>> > I am posting here because I believe it is mostly related to lua(la)tex and
>> > less to babel-greek.
>> >
>> > Here is an MWE:
>> >
>> > ----------------------------
>> > % !TeX program = lualatex
>> >
>> > \documentclass{book}
>> >
>> > \usepackage[sf,bf]{titlesec}
>> >
>> > \usepackage[greek]{babel}
>> > \babelfont[greek]{rm}{Times New Roman}
>> > \babelfont[greek]{sf}{Arial}
>> >
>> > % \usepackage{polyglossia}
>> > % \setmainlanguage{greek}
>> > % \newfontfamily\greekfont[Script=Greek]{Times New Roman}
>> > % \newfontfamily\greekfontsf[Script=Greek]{Arial}
>> >
>> > % \usepackage[no-math]{fontspec}
>> > % \defaultfontfeatures{Ligatures=TeX}
>> > % \setmainfont{Times New Roman}
>> > % \setsansfont{Arial}
>> > % \usepackage{xgreek}
>> >
>> > \begin{document}
>> >
>> > \tableofcontents
>> >
>> > \chapter{Πρώτο Κεφάλαιο}
>> > Καλησπέρα \TeX. Πρώτο Κεφάλαιο.
>> >
>> > \end{document}
>> > ----------------------------
>> >
>> > I was compiling this on fully updated Arch Linux with lualatex from texlive
>> > (not Arch Linux packages). The result is the attached pdf file. You can see
>> > the strange accents in "Περιεχόμενα" and "Κεφάλαιο". You can also see that,
>> > for example, "Κεφάλαιο" appears correctly in the text.
>> >
>> > After a lot of testing, I found out that this happens only for lualatex
>> > (not xelatex) with babel-greek (not polyglossia or xgreek) (only in the
>> > "stock" strings, "Περιεχόμενα", "Κεφάλαιο") and lualatex from texlive (not
>> > with lualatex from the Arch Linux package).
>> >
>> > [....]
>> > The crucial difference is probably in harfbuzz. But how can this impact the
>> > output in the ways described above?
>>
>> The harfbuzz version is not relevant.
>>
>> In the faulty words the ά is not U+03AC (alpha with tonos) but U+3B1
>> (alpha) with a following U+0301 (combining acute accent) (and
>> similar for the other char with accent).
>>
>> If the greek hyphenation patterns are used an discretionary is
>> inserted between char and accent and that disturbs the placement:
>>
>> \documentclass{book}
>>
>> \usepackage{babel}
>> \usepackage{fontspec}
>> \setmainfont{Times New Roman}
>> \setsansfont{Arial}
>> \textwidth=1pt
>> \begin{document}
>> \selectlanguage{greek}
>> \showoutput
>>
>> Κεφάλαιο
>> Κεφάλαιο
>>
>> \sffamily
>> Κεφάλαιο
>> Κεφάλαιο
>>
>> \end{document}
>>
>> basically it means that the greek hyphenation patterns needs a
>> correction, see https://tex.stackexchange.com/a/340164/2388 and
>> https://tex.stackexchange.com/q/330180/2388 for a similar problem
>> with russian.
>>
>> You can avoid the problem by using the harfbuzz renderer, e.g.
>>     \babelfont[greek]{sf}[Renderer=harfbuzz]{Arial}
>> It typically suppress some problematic hyphenation points as it
>> builds clusters earlier. (polyglossia uses harfbuzz by default,
>> that's why you don't see the problem there).
>>
>> Another option is to redefine \contentsname and \chaptername and to
>> remove the accent introduced by babel-greek here:
>>
>> \addto\captionsgreek{\renewcommand\chaptername{Κεφάλαιο}%
>> \renewcommand\contentsname{Περιεχόμενα}}
>>
>>
>> The real fix is naturally to correct the hyphenation patterns.
>>
>>
>>
>> --
>> Ulrike Fischer
>> http://www.troubleshooting-tex.de/
>>
>
> Thank you for your prompt answer.
> I remember that in the past I had witnessed similar problems when copying texts from a Mac which, I think, uses combining accents.
>
> I will write to the tex-hyphen list about it.
>
> It is still puzzling, though, why the problem was not present earlier (around June, for example, I was working with the same document that triggered this report, without any issues) or it does not show using the Arch Linux lualatex (which is packaging a texlive version around March). Maybe something has changed with babel-greek in the meantime.

As an answer to my last question: this is also a regression in babel-greek. See
https://codeberg.org/milde/greek-tex/issues/3#issuecomment-1283703
for some details.