On which env are you?

Could you give a try to last stable LO version 6.0.5?

If it doesn't work, it could be interesting, just for the test, to know if the bug is still reproduceable on a daily build from master branch (see https://dev-builds.libreoffice.org/daily/master/Win-x86_64@42/current/).

Yo, Julien,

I'm cool, so I'm using Ubuntu 18.04.

I added your repo and tried 6.0.5. Same problem.

I hacked your W!nd@w$ link and the .deb package didn't install.

But, just clone or DL the repo and try it yourself, either the .doc or .odt files and see if you get the spaces after the Greek letters that look like "a" or "w"...

WRONG:
πολλῶ ν

ὑδά των
OR
ὑ δά των

RIGHT:
πολλῶν

ὑδάτων

git clone http://github.com/jessesteele/pdf-bug

You may need TexGyre installed to use real fonts for real publishing...

sudo apt install tex-gyre

On pc Debian x86-64 with master sources updated today, I git cloned your repo.

I noticed that LO pdf showed pairs of words in columns, on Calligra pdf it's not the case.
Then I opened the odt, I saw these same colums.
I exported the file on pdf, same columns too.

I didn't see:
πολλῶ ν
ὑδά των
OR
ὑ δά των

I suppose I missed something but don't know what.

The "columns" are a product of "justify text". In Pagella, each line ends with a different word.

Look on PDF page 9 (labeled page 1)

You will see the little "v" thing by itself. It's not supposed to be. It's been pushed out into the margin where it shouldn't be able to be.

The u da twv should be all one word.

Looking down at PDF pages 10-11 (labeled 2-3) the Greek letters are grouped as all one word as they should be.

You can see this in the file I put in the GitHub repo.

On "Problem example DOC - via LibreOffice.pdf", 9th page of pdf, I see:
hudatoen/ ὑ δάάτων

whereas in "Problem example DOC - via Calligra.pdf", I see:
hudatoen/ὑδάτων

Ok so Calligra seems ok, not LO.

Strangely, I've opened "Problem example.doc" with 6.0.5.2 LO Debian (testing) package and exported it, I got:
hudatoen/ ὑ δάτων (from copy paste)
but I see:
hudatoen/ὑδάτων
as if the copy paste would add some spaces.

I noticed on Evince that when highlighting "ὑδάτων", "ά" was replaced by a square only on LO PDF export.

In brief, I don't reproduce exactly what you describe but I got a pb too.

Miklos: thought you might be interested in this one since it concerns PDF export unless it's more about fonts rendering, in that case, Khaled may help here?

Yeah! Julien, you are seeing it, and the other problems.

I can explain. I studied Greek in college and I get the font...

Those accent mark things above and under the letters are rendered by fonts similarly to how "ff" might be a single character. I don't know base-language level font encoding, but UTF-8 might regard ὑ or ά as actually two separate characters, like combo letters in creating fonts.

Dealing with that "correctly" is probably where the problem begins.

Summary:

TexGyre might not do their fonts "correctly", but that Roman default font does.

Calligra got TexGyre to render correctly... BUT, Amazon's print on demand says the Calligra .pdf file is broken...

(Calligra rant: in Calligra, .doc > .pdf, it messed up the page numbers; .odt > .pdf Calligra was great, but Amazon said it was broken or someth. So, not even Calligra is perfect in this.)

That's what I know.

Cheers and kudos all!

For what it's worth, I've had problems with ghostscript processing .pdf files to get CMYK working correctly. Scribus won't import the same files it could before. Command line gs makes a .pdf blank. These don't work:

http://wiki.inkscape.org/wiki/index.php/ExportPDFCMYK
http://zeroset.mnim.org/2014/07/14/save-a-pdf-to-cmyk-with-inkscape/

...Just in case you're using ghostscript and it's giving everyone problems. :-)

NEW as Julien's confirmation was confirmed.

Created attachment 182405 [details]
PDF from LO 7.4

I can’t reproduce this. The accented Greek letters use a fallback font (which mean TeX Gyre fonts do not support them), but the spacing looks fine in exported PDF.