fix: a negative --precision exits with a traceback - #2261

A negative --precision exits with a raw traceback instead of a configuration error.

$ coverage report --precision=-2
Traceback (most recent call last):
  ...
  File "coverage/results.py", line 422, in display_covered
    return f"{pc:.{precision}f}"
ValueError: Format specifier missing precision

After:

precision=-2 is invalid. Must not be negative.

The same applies to coverage html --precision=-2 and to precision = -1 in a .coveragerc, so it is reachable from a config file and not just a typo on the command line.

should_fail_under, in this same file, already raises ConfigError for an out-of-range fail_under:

if not (0 <= precision < 10):
    raise ConfigError(f"Invalid precision: {precision}")

so this is the missing parallel guard rather than a new convention. ConfigError is caught by cmdline.main() and printed as a plain message.

Normal use is unaffected: --precision=2 still gives TOTAL 3 0 100.00%.

Testing

test_display_covered_negative_precision in tests/test_results.py, parametrised over -1 and -2.

Reverting only the condition, precision < 0 to precision < -99, so the test still imports cleanly and fails as a test rather than an error:

FAILED tests/test_results.py::NumbersTest::test_display_covered_negative_precision[-1]
E   ValueError: Format specifier missing precision

pytest tests/test_results.py is 47 passed. mypy coverage/results.py is Success: no issues found. ruff-format is clean.

Two pre-existing items I left alone, both verified identical on a stashed tree: a RUF015 finding at results.py:44, and 8 failures in test_report.py that need the C extension and a subprocess PATH this environment does not provide.

Disclosure: written with AI assistance (Claude Code). I reproduced the traceback, checked all three entry points (CLI report, CLI html, config file), and ran the revert check myself.