Non-finite rerun delay settings abort pytest with INTERNALERROR #390

Reproduction

On master 01fd226, Windows / CPython 3.13.15 / pytest 9.1.1:

# test_recovery.py
def test_recovery(request):
    assert request.node.execution_count == 3

python -m pytest -q --reruns 2 --reruns-delay=nan test_recovery.py
python -m pytest -q --reruns 2 --reruns-delay=inf test_recovery.py
python -m pytest -q --reruns 2 --reruns-delay-backoff-factor=nan test_recovery.py
python -m pytest -q --reruns 2 --reruns-delay-backoff-factor=inf test_recovery.py

All four commands exit with code 3 / INTERNALERROR in time.sleep. NaN produces ValueError; positive-infinite delay produces OverflowError; an infinite backoff factor with the default zero delay produces NaN on the second rerun. These runs use real sleep calls. The same non-finite values can also come from ini settings and flaky marker arguments.

Expected behavior / proposed fix

Validate that the configured delay and backoff factor are finite before using them, warn, and fall back to 0 / 1.0 respectively. This follows the existing handling of negative values. Negative values (including negative infinity) would keep their current warning and defaults. Valid finite settings would retain their current behavior.

I have a focused patch in progress with regression coverage for CLI, ini, keyword markers and positional markers.

This concerns explicit NaN/infinite configuration, which is separate from #388 / #389's overflow during exponentiation of a large finite factor. The proposed patch does not change the retry formula or duplicate that implementation.

AI assistance disclosure: an AI coding assistant performed this investigation and drafted this report with the account owner's authorization.