fix(toml): swap uiri/toml encoder for tomli_w (issue #439 residual)

[ricaskew]

The uiri/toml encoder raises IndexError on strings containing certain characters (notably real control characters like chr(27)/ANSI escape). Issue #439 identified this class of bugs and proposed switching to tomli-w. Decode was migrated to stdlib tomllib on 3.11+ previously, but encode still routed through toml.dumps and still crashed.

Reproducer on main:

>>> benedict({"color": "\033[31m"}).to_toml()
IndexError: list index out of range

Change: TOMLSerializer.encode() now calls tomli_w.dumps(). Decode path untouched (tomllib on 3.11+, toml on 3.10). toml stays in the [toml] extra guarded by python_version < '3.11' for the 3.10 decode fallback; tomli-w is added unconditionally for encode.

Regression tests cover:

• ANSI control character (chr(27)) encode + round-trip — was crashing
• Issue Bug: Many serious TOML parsing/loading bugs caused by underlying toml library, consider switching? #439's literal-backslash examples — guard against regression
• Round-trip on 7 tricky values (control chars, tabs, unicode, quotes)
• Nested dict with embedded control chars
• Direct serializer encode/decode path

tests/serializers/test_toml_serializer.py replaces the prior TODO stubs with 5 real tests. test_io_dict_toml's "extra not installed" test patches tomli_w_installed (the encode dependency) instead of toml_installed.

API note: tomli_w.dumps kwargs differ from toml.dumps (no encoder= param; gains multiline_strings and indent). Callers of .to_toml(**kwargs) passing uiri-specific kwargs will hit TypeError and should migrate to tomli-w's kwarg surface.

Full suite: 800 tests pass (1 pre-existing skip).

Fixes the encode-side failure mode documented in #439.

---

name: Pull request
about: Submit a pull request for this project
assignees: fabiocaccamo

---

Describe your changes
?

Related issue
?

Checklist before requesting a review

• I have performed a self-review of my code.
• I have added tests for the proposed changes.
• I have run the tests and there are not errors.

[Copilot]

Pull request overview

This PR fixes TOML encoding crashes (notably on real control characters like ESC/ANSI sequences) by switching the encode path from uiri/toml to tomli-w, while keeping the existing decode behavior (tomllib on Python 3.11+, toml on 3.10).

Changes:

• Update TOMLSerializer.encode() to use tomli_w.dumps() and require tomli-w at runtime for encoding.
• Adjust TOML optional-dependency configuration to add tomli-w and gate toml to Python < 3.11.
• Replace prior TOML serializer TODO tests with concrete regression/round-trip coverage and update the “extra not installed” patch target accordingly.

Reviewed changes

Copilot reviewed 4 out of 5 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
benedict/serializers/toml.py	Swap TOML encoding implementation to tomli-w and add optional import/installed flag.
pyproject.toml	Update [project.optional-dependencies].toml to include tomli-w and conditionally include toml for Python < 3.11.
requirements.txt	Add tomli-w to the pinned dev/test requirements set.
tests/serializers/test_toml_serializer.py	Add regression and round-trip tests covering control characters and other tricky strings.
tests/dicts/io/test_io_dict_toml.py	Patch tomli_w_installed (encode dependency) for the “extra not installed” test.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Test methods in this file omit the -> None return type annotation, while the other serializer test modules consistently include it (e.g., tests/serializers/test_abstract_serializer.py:12). For consistency (and to keep type-checking expectations uniform), please add -> None to these new test methods.

[Copilot]

The docstring includes "\033[31m" inside a Python string literal, so \033 will be interpreted and the docstring will contain a real ESC control character at runtime. This can lead to invisible/control characters in test output and makes the docstring harder to read/copy. Consider escaping it as \\033[31m (or writing it as \x1b[31m) in the docstring example so the docstring remains printable text.

Suggested change

	`benedict({"color": "\033[31m"}).to_toml()` must not raise. On
	`benedict({"color": "\\033[31m"}).to_toml()` must not raise. On

[fabiocaccamo]

@copilot apply changes based on the comments in this thread

[fabiocaccamo]

@ricaskew thanks for the PR, the crash is confirmed and the fix is good.

I have one concern: we are still using uiri/toml for decode on Python 3.10, but this library is abandoned and has known bugs. Since we are already adding tomli_w (from hukkin), I think it is better to finish the migration and remove uiri/toml completely.

My proposal:

• decode -> tomli (on Python 3.10 only) + tomllib (stdlib on Python 3.11+)
• encode -> tomli_w (already in this PR)

So the dependencies would be:

• on Python 3.10: only tomli + tomli-w
• on Python 3.11+: only stdlib + tomli-w

Can you update the PR in this way? I'm happy to review again after.

---

PS. remember to run pre-commit.

[fabiocaccamo]

@copilot resolve the merge conflicts in this pull request

[ricaskew]

@fabiocaccamo it's been around 4 days since the @copilot resolve directive and the conflicts on toml.py are still standing. Happy to resolve them manually from this side and push to fix/toml-encoder-swap so you can re-run the matrix on a clean tip — would that be useful, or would you prefer to keep waiting on copilot? Either way works on my end; the local branch on a5fe910 is pre-commit clean (9/9 hooks including mypy) and 21+1 tests pass on the affected files, so once the conflict is resolved CI should run cleanly.