`--toml` crashes with `AttributeError: 'list' object has no attribute 'items'` on a `pyproject.toml` with an array-of-tables directly under `[tool]` (e.g. `[[tool.dynamic-metadata]]`)

[zklaus]

Issue

codespell --toml pyproject.toml ... raises

AttributeError: 'list' object has no attribute 'items'

when the pyproject.toml contains a TOML array of tables that is a direct child of [tool], such as [[tool.dynamic-metadata]].

The cause is in parse_options, which loads the entire [tool] table and hands all of it to configparser.read_dict:

data = tomllib.load(f).get("tool", {})
if "codespell" in data:
    data["codespell"] = _toml_to_parseconfig(data["codespell"])
config.read_dict(data)

read_dict expects a {section: {key: value}} mapping, but a [[tool.dynamic-metadata]] entry makes tool.dynamic-metadata a list, so read_dict fails when it does keys.items() on that value. codespell should only read the [tool.codespell] section, not its sibling tables.

This is distinct from #3852: there the user passed --config (INI-only) instead of --toml, and the array was [[tool.uv.index]], which is nested under tool.uv (a dict) and so does not trip read_dict. Here the correct --toml flag is used and the array sits directly under [tool].

Affected versions

2.4.1, 2.4.2 (latest), and current main — the config.read_dict(data) line in parse_options is unchanged across all of them.

Steps to reproduce

pyproject.toml:

[tool.codespell]
skip = "*.lock"

[[tool.dynamic-metadata]]
provider = "scikit_build_core.metadata.version"

hello.txt:

hte cat

Run:

$ codespell --toml pyproject.toml hello.txt

(codespell also auto-reads ./pyproject.toml, so plain codespell hello.txt from the project directory crashes the same way, even without --toml.)

Actual behavior

Traceback (most recent call last):
  ...
  File ".../codespell_lib/_codespell.py", line 664, in parse_options
    config.read_dict(data)
  File ".../configparser.py", line 763, in read_dict
    for key, value in keys.items():
AttributeError: 'list' object has no attribute 'items'

Expected behavior

codespell reads the [tool.codespell] settings and reports the misspelling (hte -> the), ignoring the unrelated [[tool.dynamic-metadata]] table.

Why this matters now

[[tool.dynamic-metadata]] is emitted by every scikit-build-core >= 1.0 project (the standard dynamic-metadata table). Any such project that runs codespell against its pyproject.toml — the common case — hits this. We ran into it while migrating PyTorch's build system to scikit-build-core 1.0.

Suggested fix

In parse_options, pass only the codespell section to read_dict rather than the whole [tool] table:

data = tomllib.load(f).get("tool", {})
if "codespell" in data:
    config.read_dict({"codespell": _toml_to_parseconfig(data["codespell"])})

so sibling tables (arrays of tables or otherwise) are never fed to configparser.

[DimitriPapadopoulos]

Thanks for the fix. Would you like to provide a PR?

It might be worth to exit gracefully even with [[tool.codespell]], just in case...