Without the change:

$ env -i ./python -c 'import distutils.spawn; print(distutils.spawn.find_executable("true"))'
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/home/vstinner/prog/python/master/Lib/distutils/spawn.py", line 176, in find_executable
    path = os.environ['PATH']
  File "/home/vstinner/prog/python/master/Lib/os.py", line 672, in __getitem__
    raise KeyError(key) from None
KeyError: 'PATH'

With the change:

$ env -i ./python -c 'import distutils.spawn; print(distutils.spawn.find_executable("true"))'
/bin/true

I thought distutils is frozen.

If this change acceptable, it needs an open issue, news entry and tests.

I thought distutils is frozen.

I'm not aware of that. Are you refering to the dead distutils2 project? distutils now accepts changes.

If this change acceptable, it needs an open issue, news entry and tests.

done

It’s not 100% frozen anymore, but still very stable. New things like wheel support happens outside of distutils and the stdlib.

For example, it would break a lot of code to use shutil.which in distutils and remove find_executable. This PR however fixes two issues but keeps the interface intact, so it is acceptable.

@serhiy-storchaka: I added a NEWS entry and a basic test. I don't want to write a long test suite, since shutil.which() is very well tested in test_shutils, and find_executable() is now really just calling which(). Would you mind to review my change?

I missed the fact that find_executable() always lookup in the current directory, whereas shutil.which() doesn't. I abandon this change in favor of PR #9049.