Status of the Python 3.14 / uv modernization and what is still open.

• The translator runs on Python 3.14+, packaged with uv (pyproject.toml, uv_build). setup.py is gone.
• ANTLR 3.1.3 Python runtime vendored and ported to Python 3 in java2python/lang/antlr3/.
• Generated code targets Python 3.14+: python3 shebang, no __future__ imports, class X:, super(), ABC interfaces, int for long, 0o octal literals, Java integer division via an inlined jdiv() helper.
• The full cd test && make suite has been run against a JDK: all 107 test/*.java programs compile, translate, and run, with the Python output diffing clean against the Java output. make selectors passes (11 tests).
• CLI fixes: a directory walk now writes under the output directory instead of beside the input (path.join discards it for absolute names), each file of a walk gets its own config list instead of accumulating the previous files' configs, and unreadable or unparseable input reports an error and an exit code instead of a traceback (JavaSyntaxError from buildAST).
• A Claude Code plugin lives in plugin/; see CLAUDE.md.
• Labeled break and continue are translated. A jump a bare keyword can express still emits one; the rest raise a generated per-label exception that the labeled statement catches (java2python/compiler/visitor.py, basic.maybeLabelClasses). Covered by test/Break2-4.java, test/Continue3-5.java and test/ForEach1.java.

1. Java % semantics. Java % truncates toward zero, while Python % floors, so the results differ when an operand is negative. Consider a jmod() helper alongside jdiv() (java2python/compiler/visitor.py, java2python/mod/include/).
2. Division readability. Every / becomes jdiv(a, b), even for floats, because types aren't tracked. Possible improvements: emit plain / when an operand is a float literal or a float/double cast, or track declared local variable types.
3. Parser regeneration. The java2python/lang/Makefile flow (ANTLR 3.1.3 tool, then rewriting the imports to the vendored runtime) is untested. Regenerating would also bring back the __main__ blocks that import antlr3.main, which isn't vendored.
4. test/configs/defaults.py. It defines outputSubs, but the code reads moduleOutputSubs, so those substitutions probably never apply. Rename it and check that the tests still pass.
5. Generated-code style. Possible cleanups: emit f-strings instead of .format() for String.format and drop redundant int()/str() casts on literals (flagged by ruff --select UP on the test output). @classmethod for static methods could become @staticmethod when cls isn't used.
6. Optional dependency on zope.interface. It is only needed by test/configs/defaults.py and the zope* handlers in mod/basic.py. Consider dropping those handlers.
7. Tooling. Add ruff configuration and CI (GitHub Actions with astral-sh/setup-uv, a JDK, and make -C test). Possibly port the Makefile harness to pytest.
8. Docs. doc/usage.md, doc/translation.md, and doc/customization.md still show Python 2-era output in places, and the links still point at the upstream natural/java2python repository.