move to python3

[marijani101]

As Python 2 is coming to an end, wouldn't it be better to migrate to Python 3?

[zuphilip]

The hocr-tools are working with Python 3 already. We support Python 2 and 3 together.

[stweil]

Maybe some compatibility constructs can be removed from the code as soon as Python 2 is gone, but for the moment I think there is nothing to be done. @marijani101, did you notice problems with Python 3 which would require an action now? I only found that the README could be updated to mention Python 3 as well. Maybe you want to send a pull request for that?

[stweil]

The Python 2 package names in the README.md should be replaced by Python 3 package names. @marijani101, can you send a pull request?

[FriedrichFroebel]

It seems like while the setup file still advertises Python 2, https://github.com/ocropus/hocr-tools/commit/269d63a816dc801b77e549b9c3b3bde708912286 basically drops this support in the most recent time. This contradicts with the following code, as f-strings have not been available before Python 3.6: https://github.com/ocropus/hocr-tools/blob/0ad95b3606229c8a6895a3a6e782ff88d9db1d8d/setup.py#L24-L26

[stweil]

Right, thanks for reporting this. Do you want to send a pull request which removes all old entries? All Python versions before 3.7 are unsupported.

[FriedrichFroebel]

I just did some more tests regarding version support and stumbled upon some more stuff which probably needs some attention (and is more or less related to Python 3 support):

• CI fails due to beautifulsoup4 never being installed, but apparently being called by the tests (according to the code, the package is optional for regular installations):

  # ocrx_word argument
  not ok 17 - Failed: hocr-extract-images -U -p word-%03d.png -b ../testdata -e ocrx_word ../testdata/tess.hocr
  ---
  diag: |
  Traceback (most recent call last):
  File "/opt/hostedtoolcache/Python/3.7.15/x64/bin/hocr-extract-images", line 79, in <module>
  from bs4 import UnicodeDammit
  ModuleNotFoundError: No module named 'bs4'
  ...

• Current lxml versions do not work with hocr-extract-images (fixed by using doc = html.document_fromstring(content.encode('utf-8'), parser=parser) instead:

  # ocrx_word argument
  not ok 17 - Failed: hocr-extract-images -U -p word-%03d.png -b ../testdata -e ocrx_word ../testdata/tess.hocr
  ---
  diag: |
  Traceback (most recent call last):
  File "/opt/hostedtoolcache/Python/3.7.15/x64/bin/hocr-extract-images", line 83, in <module>
  doc = html.document_fromstring(content, parser=parser)
  File "/opt/hostedtoolcache/Python/3.7.15/x64/lib/python3.7/site-packages/lxml/html/__init__.py", line 759, in     document_fromstring
  value = etree.fromstring(html, parser, **kw)
  File "src/lxml/etree.pyx", line 3257, in lxml.etree.fromstring
  File "src/lxml/parser.pxi", line 1911, in lxml.etree._parseMemoryDocument
  ValueError: Unicode strings with encoding declaration are not supported. Please use bytes input or XML fragments without declaration.
  ...

• We might want to migrate CI from Travis to GitHub Actions as well.
• We might want to improve the overall package structure to declare the current scripts as console scripts entry points and move the actual implementations to some common package (making code re-usage easier). This would allow us to use the standardized stdlib unittest module for testing as well (due to the package being importable) which enables easier integration with other tools like coverage.py to actually ensure that the code is properly tested if dropping old Python 2 compatibility code for example.

[mrghosti3]

Any news on this issue?