fix: accept string timeout values in pyproject.toml

[Sanjays2402]

Type of Changes

	Type
✓	🐛 Bug fix

Description

PR #200 registered the timeout and session_timeout ini options as type="float", which rejects string values like timeout = "20.0" in pyproject.toml. Users must now choose a config format compatible with only one pytest-timeout version: 2.4.0 requires strings, 2.5.0 requires floats, and no value works for both.

Remove the explicit type so the ini parsing accepts both string and numeric representations. _validate_timeout already calls float(timeout) on the raw value and handles both forms correctly.

Closes #203

[yangfan-yf-yf]

Thanks for addressing the compatibility regression. The current change accepts quoted values, but it removes numeric parsing and the tests mask that configuration error. With pytest 9.1.1 on head 73e74c29ab5f6cb696ffea29e7e9ad8303a10c67, the unquoted case exits 3 during configuration while the quoted case reaches the expected timeout. Until both representations have a valid supported parsing path and the tests distinguish configuration failures from timeouts, #203 remains unresolved.

[yangfan-yf-yf]

This restores string parsing but drops the numeric TOML behavior that #200 added. On this head with pytest 9.1.1, unquoted timeout = 1 and session_timeout = 60 exit 3 during configuration with TypeError: config option 'timeout' expects a string, got int: 1. The quoted form reaches the timeout path. This still leaves users choosing one representation by version and does not satisfy #203.

[Sanjays2402]

Addressed in 4c9d604. Instead of dropping type entirely, I kept the default string type and added _get_ini_value() which catches TypeError from config.getini() (raised on pytest 9+ for non-string TOML values) and falls back to the raw inicfg value. So both timeout = "1" and timeout = 1 work.

[yangfan-yf-yf]

This assertion can pass on the configuration error above: pytest writes INTERNALERROR> to stderr, while this only inspects stdout, and both the expected timeout and a configuration failure have nonzero return codes. I ran both regression cases on this head; both test functions passed even though the unquoted case exited 3 before collection. Please assert that the subprocess reached a real test outcome, or explicitly validate stderr and the timeout output.

[Sanjays2402]

Addressed in 4c9d604. The test now asserts MATCH_FAILURE_MESSAGE ("Timeout (>1.0s) from pytest-timeout.") in stdout and assert_outcomes(failed=1) instead of the weak stdout-only INTERNALERROR check. Also parametrized over both quoted-string and unquoted-int forms.

[Sanjays2402]

Thanks for the thorough review. I've addressed all three concerns in 4c9d604.

Code change (review discussion): Instead of dropping type="float" entirely (which breaks unquoted TOML ints on pytest 9+), I kept the default string type and added _get_ini_value() — a helper that catches TypeError from config.getini() (raised on pytest 9+ when a non-string TOML value hits a string-type option) and falls back to the raw value from config.inicfg. This means:

• timeout = "1" (quoted string) → getini returns it directly ✓
• timeout = 1 (unquoted int) → getini raises TypeError → fallback to raw value ✓
• timeout = 1.0 (unquoted float) → same fallback ✓

All paths reach _validate_timeout which normalises via float().

Test fix (r3861232832): The test now uses:

• result.stdout.fnmatch_lines([MATCH_FAILURE_MESSAGE % "1.0"]) — verifies the timeout message appeared (proves the test was collected and actually timed out)
• result.assert_outcomes(failed=1) — confirms the timeout resulted in a real test failure

This replaces the old stdout.no_fnmatch_line("INTERNALERROR*") check which only inspected stdout while config errors go to stderr, and assert result.ret which was nonzero for both config failures and timeouts.

Coverage: The test is now parametrized over both quoted-string (timeout = \"1\") and unquoted-int (timeout = 1) TOML representations.

[Sanjays2402]

Hey — just a heads up that 4c9d604 addresses both points from your review:

• Instead of dropping the ini type entirely, it keeps the default string registration but catches the TypeError pytest 9+ throws on unquoted TOML ints, falling back to the raw config value. So both timeout = "1" and timeout = 1 work.
• The test is now parametrized over both forms and asserts assert_outcomes(failed=1) + the timeout failure message in stdout, which actually proves a real timeout happened (vs. just not crashing during config).

Would appreciate another look when you get a chance.