pprint.isreadable() returns True for float('inf'), float('-inf'), float('nan')

[zetzschest]

Bug report

Bug description:

Description

pprint.isreadable() returns True for float('inf'), float('-inf'), and float('nan'), but their repr() values (inf, -inf, nan) are not valid Python literals — eval() raises NameError. The documented contract of isreadable is that the representation can be used to recreate the value via eval().

Reproducer

import pprint

print(pprint.isreadable(float('inf')))   # True — but eval('inf') raises NameError
print(pprint.isreadable(float('-inf')))  # True
print(pprint.isreadable(float('nan')))   # True

Expected behavior

isreadable() should return False for inf, -inf, and nan. The docstring states it determines if saferepr(object) is readable by eval(). Since isreadable and saferepr/pformat use the same internal repr, isreadable(x) being True implies eval(pformat(x)) == x should hold — but it doesn't:

import pprint
eval(pprint.pformat(float('inf')))  # NameError: name 'inf' is not defined

Actual behavior

Returns True for all three, violating the documented contract.

CPython versions tested on:

CPython main branch

Operating systems tested on:

No response

Linked PRs

• gh-145117: Fix pprint.isreadable() returning True for inf/nan #145116
• gh-145117: Fix pprint.isreadable() returning True for inf/nan #145120

[skirpichev]

I would say it's a documentation issue. It affects also complex numbers:

>>> pprint.isreadable(complex('inf'))
True

We also have "broken" eval(repr(x)) == x invariant for x=inf, unless you add missing import from the math module.