[jseop-lim]

Description

Fixes #1253

LogNode had three gaps in its argument checks. The specializations that take a base tested the value with < 0, so a zero value returned -inf. The one-argument specializations took the base as an unguarded PNone, so None was treated as a missing base. The specializations for a non-number base converted it before the value was checked. CPython's math_log computes log(x) first, rejecting any x <= 0, and only then converts the second argument, whatever it is (mathmodule.c#L2269-L2273).

AS-IS

>>> math.log(0, 2)
-inf
>>> math.log(10, None)
2.302585092994046
>>> math.log(-1, object())
TypeError: must be real number, not object

TO-BE

>>> math.log(0, 2)
ValueError: math domain error
>>> math.log(10, None)
TypeError: must be real number, not NoneType
>>> math.log(-1, object())
ValueError: math domain error

(matching CPython)

Changes

• Reject a zero value with <= 0 in logPIN, logDD, logDPI, logPID, logLPI and logPIPI, as logDN already does.
• Guard log(long, PNone), logDN, logPIN and logO with isNoValue(novalue), so None goes to the specializations that convert the base.
• Check the value's domain in logOO, logLO, logDO and logPIO before converting the base, and add !isNoValue(base) to logLO, logDO and logPIO, as logOO has. Without that guard, once a None base had activated one of them, a later math.log(10) was dispatched there and raised TypeError.
• Add tests/test_math_log.py.

Testing

mx python-jvm on darwin-aarch64, then mx graalpytest test_math_log.py (3 passed) and test_math.py (61 passed), and test.test_math from lib-python with --tagged (79 passed). The new tests fail on the unpatched build (2 failures, 1 error). 25 math.log call shapes, covering int subclasses, None, object() and __float__ arguments, now print the same as CPython 3.13.14, except for a last-digit difference in math.log(8.0, 2**70) that master has too.