[Python-modules-team] Bug#926351: python3-tesserocr: segfaults on import

[zaza]

Hi,

On Fri, 5 Apr 2019 23:26:55 +0200 =?UTF-8?Q?Bernhard_=c3=9cbelacker?= <bernhardu at mailbox.org> wrote:
> Dear Maintainer,
> just tried to help triaging this issue.

Me too.

> 
> Seems this is "expected" behaviour with LC_ALL not
> set to "C". In the end it leads to this upstream bug:
> 
>     https://github.com/sirfz/tesserocr/issues/165
> 
> It contains some workarounds and more information.
> 

Yes.

There is a simplistic solution. Set 'locale' before 'import tesserocr'
doing something like this:

>>> import locale
>>> locale.setlocale(locale.LC_ALL, 'C')
'C'
>>> import tesserocr
>>> 

Or adding some code, like here:
https://github.com/WeblateOrg/weblate/commit/6724204977e7c466a463595736cc5299c8f01829

The workaround - set locale - is known. It is in debian/rules[0] file. But the 
user should be informed about it.

[0] - https://sources.debian.org/src/python-tesserocr/2.4.0-4/debian/changelog/


BTW, CI tests never passed. And it is not a reproducible build. Two tests fail.

 - https://sources.debian.org/src/python-tesserocr/2.4.0-4/tests/test_api.py/#L227
 - https://sources.debian.org/src/python-tesserocr/2.4.0-4/tests/test_api.py/#L275


The package was recently added to Debian. Low popcon.



Regards,
Herbert