gh-71810: fix corner case (length==0) for int.to_bytes()

[skirpichev]

>>> (0).to_bytes(0, 'big', signed=True)
b''
>>> (-1).to_bytes(0, 'big', signed=True)  # was b''
Traceback (most recent call last):
  File "<python-input-0>", line 1, in <module>
    (-1).to_bytes(0, 'big', signed=True)
    ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^
OverflowError: int too big to convert

• Issue: int.to_bytes() and int.from_bytes(): raise ValueError when bytes count is zero #71810

[vstinner]

LGTM

[vstinner]

@serhiy-storchaka: I would suggest to backport this change to Python 3.13 and 3.14. What do you think?

[serhiy-storchaka]

It seems there are no enough tests for signed=True. Please add also tests for (128).to_bytes(0, 'big', signed=True) and (-129).to_bytes(0, 'big', signed=True).

[skirpichev]

Disclaimer:
This pr implements "the minimal change necessary to remove ambiguity" (c) @mdickinson

I was thinking about (0).to_bytes(0, 'big', signed=True) too, but that means rejecting also int.from_bytes(b"", ..., signed=True). That case looks for me as a special case, that aren't special enough to warrant compatibility break.

[serhiy-storchaka]

Or better add several tests for extreme values and values just above the limit for all combination of signedness and n=0, 1, 2.

[skirpichev]

It seems there are no enough tests for signed=True. Please add also tests for (128).to_bytes(0, 'big', signed=True) and (-129).to_bytes(0, 'big', signed=True).

Done.

There are such tests, but rather for "automatic" value for length parameter (determined by result):

cpython/Lib/test/test_long.py

Line 1262 in 1ce0553

def test_to_bytes(self):

Or better add several tests for extreme values and values just above the limit for all combination of signedness and n=0, 1, 2.

It seems to be a big refactoring of tests. Maybe it does make sense as a separate pr?

[serhiy-storchaka]

Sorry, use n=1 here.

[serhiy-storchaka]

And this test is redundant. We need to test for 128.