Le 16/01/2017 à 22:09, Jean-Marc Lasgouttes a écrit :
Le 16/01/2017 à 21:29, Richard Heck a écrit :
Bisect blames:

24648404b3c85015584b1ca127e257cbecf3342d is the first bad commit
commit 24648404b3c85015584b1ca127e257cbecf3342d
Author: Jean-Marc Lasgouttes <lasgout...@lyx.org>
Date:   Sun Oct 23 20:52:01 2016 +0200

    Work around issues with Qt5 and Arabic text

Thanks Richard, I'll have a look. Guillaume, any intuition why master is
different from stable in this respect?

I can confirm that using QFontMetrics::width instead of the width returned by QTextLayout does the trick, but that makes me a bit nervous :)

What happens basically is that \not has 0 width, but QTextLayout does not seem to respect that. Several workarounds exist
- have a different FontMetrics::width for maths
- in mathed_string_dim, use width(char_type) repeatedly instead of width(docstring)
- or do a shortcut for cases where the string only has one character.

All these things would work, but they are fragile as long as we do not know:
- why they work
- why is master immune to the problem

Guillaume, could it just be that the definition of \not needs to be adjusted like it has been in master?

JMarc

Reply via email to