Fixup merge master into Chinese Word Segmentation

[CrazySteve0605]

Link to issue number:

Summary of the issue:

Description of user facing changes:

Description of developer facing changes:

Description of development approach:

Testing strategy:

Known issues with pull request:

Code Review Checklist:

• Documentation:
  • Change log entry
  • User Documentation
  • Developer / Technical Documentation
  • Context sensitive help for GUI changes
• Testing:
  • Unit tests
  • System (end to end) tests
  • Manual testing
• UX of all users considered:
  • Speech
  • Braille
  • Low Vision
  • Different web browsers
  • Localization in other languages / culture than English
• API is compatible with existing add-ons.
• Security precautions taken.

[cary-rowen]

I believe this is an unexpected change. Could you please confirm it?

[CrazySteve0605]

I mistakenly applied the changes suggested by the AI, and need to restore the previous state.

[CrazySteve0605]

It seems from commit b34a049 and I've not been familiar with how it works.

[cary-rowen]

I believe this is an unexpected change. Could you please confirm it?

[cary-rowen]

Was this an unexpected change?

[cary-rowen]

Hi @CrazySteve0605, I reviewed this PR against the Copilot comments from #19166. Most of the explicit comments are handled correctly, but I think two issues still need attention.

First, the braille offset-converter issue is still unresolved. braille.py still keeps only one converter. When Chinese word-segmentation spacing is applied first and Unicode normalization is also enabled, the segmentation converter is overwritten by the normalization converter. As a result, brailleToRawPos / rawToBraillePos are only mapped back through the last converter, so cursor routing / selection mapping can still be wrong. This probably needs converter composition, or applying the transformations while keeping a combined mapping back to the original raw text.

Second, _initCppJieba now calls cls._lib.initJieba(dictDir), but it does not check the returned boolean. Since the C++ side now returns false on initialization failure, the Python side should probably treat that as failure too, e.g. set cls._lib = None and log/debugWarn. Otherwise NVDA may think cppjieba is available even though it was not initialized successful.

The rest of the fixes look broadly in the right direction to me.

[cary-rowen]

I will test the actual experience with Focus 80 later. The above is just a response to some points that Copilot raised that may need to be considered.

[cary-rowen]

Hi @CrazySteve0605

While testing Chinese braille word segmentation, I found a regression: some NVDA built-in braille state abbreviations are being split by the word segmentation logic. For example, the checked state for a checkbox should remain ⣏⣿⣹, but with this feature enabled it becomes ⣏ ⣿ ⣹, with unexpected spaces inserted between the braille cells.

This seems to happen because WordSegWithSeparatorOffsetConverter is applied to the entire rawText in braille.Region.update(). That text contains not only user-facing content, but also NVDA-generated braille state abbreviations. The current segmentation logic avoids inserting spaces next to punctuation, but Braille Pattern characters are Unicode symbols, so they are not protected by that rule.

I think this should be fixed in this PR. A reasonable approach would be to avoid inserting word-segmentation separators between Braille Pattern characters, or more generally avoid inserting separators across symbol boundaries. It would also be good to add regression tests to ensure ⣏⣿⣹ remains unchanged, while normal Chinese text is still segmented as expected.

Btw, please remember to remove irrelevant .md file changes from the changes.

Thanks

[cary-rowen]

Regarding Braille status abbreviations, please refer to this section in the user guide.