test-break case four.break (for Thai) fails on Debian
Versions:
- Pango 1.46.2
- Debian unstable (rolling release/Debian 11 alpha)
- also tried Debian 10 with selected packages upgraded to early 2020 versions that approximately match the
registry.gitlab.gnome.org/gnome/pango/fedora:v2
CI image
- also tried Debian 10 with selected packages upgraded to early 2020 versions that approximately match the
- GLib 2.66.0
- also tried 2.64.3, 2.62.5 to test a theory that the culprit was the upgrade to Unicode 13 data (it was not)
- libthai 0.1.28 (not patched by Debian)
Steps to reproduce:
- Have at least one Thai font (I used fonts-noto-core)
- Build pango as configured by the Debian packaging: libthai is enabled.
- Run tests
Expected result:
Tests pass
Actual result:
/break/four.break
in test-break
fails.
The failure mode appears to be that every time the input text contains -
, the Words
line is expected to contain bs
(is_word_boundary && is_word_start && !is_word_end
) but actually contains bse
(is_word_boundary && is_word_start && is_word_end
).
The failure mode (which I got the wrong way round in the initial report) appears to be that every time the input text contains -
, the Words
line is expected to contain bse
(is_word_boundary && is_word_start && is_word_end
) but actually contains bs
(is_word_boundary && is_word_start && !is_word_end
).
So far, I haven't been able to isolate what it is about these Debian systems that is behaving unlike the Fedora-based CI image.