On Posix, it is documented that setting PATH to the empty string is equivalent to not setting PATH at all, which is an exception to the rule that in a string like "/bin::/usr/bin" the empty string in the middle gets interpreted as ".".

PYTHONPATH does not have this exception: an empty PYTHONPATH is interpreted as equivalent to ".".

This difference is not documented.  This is a detail, but a possible source of confusion, so I'm reporting it here.

How to reproduce:

file x/x.py: "import z"
file z.py: "print(42)"

The following two lines behave differently (Bash syntax):

PYTHONPATH= python x/x.py
unset PYTHONPATH && python x/x.py

For comparison, if "./foo" is an executable, the following two lines behave identically (neither finds "./foo"):

PATH= foo
unset PATH && foo

Given how confusing it seems, perhaps we should change it to adopt a PATH-like behaviour.

Attached the diff, in case we want to do that.  The diff is only about non-Windows platforms, given that it follows a Posix use case about PATH.  But do we also need to change PC/getpathp.c?

Attached another simpler diff (only one "+" line instead of two...).

> Given how confusing it seems, perhaps we should change it to
> adopt a PATH-like behaviour.

Wouldn't that introduce a backward-compatibility issue?  FWIW, otherwise I agree that it makes a lot of sense to conform to the same behavior as PATH.

LGTM.

I would to see some unittest if possible.

Unit test added. pythonpath2.diff is included into issue16309.diff.

New changeset 05f8f7544f92 by Andrew Svetlov in branch 'default':
Issue #16309: Make PYTHONPATH= behavior the same as if PYTHONPATH not set at all.
http://hg.python.org/cpython/rev/05f8f7544f92

Fixed. Thanks.

New changeset 803e5a732331 by Ezio Melotti in branch 'default':
#16309: avoid using deprecated method and turn docstring in a comment.
http://hg.python.org/cpython/rev/803e5a732331

PATHONPATH=/tmp: will also be interpreted as "/tmp" and ".". Do the patches also address this case? PYTHONPATH=:/tmp and PYTHONPATH=/foo::/bar are also problematic.

It may be problematic, but it is also consistent with the way the shell works in general.  (Try the same things with PATH.)

Hmm.  Although that argues that the original behavior was in fact correct.  At least, the behavior of bash and zsh for me is that

   PATH= foo

finds foo if it is in the local directory, contradicting the claim made in the original post.

Oh wow. That is not what I remembered/assumed about sh and PATH. Mind. Blown.

Uh, confusion.  Indeed, "PATH= foo" finds foo in the current directory on bash.  I'm not sure how I ran the original example.  It seems that a default PATH is used, which includes at least "/bin" and ".".

The point I was making in the original post is still valid: "PATH= foo" appears to behave identically to "unset PATH && foo" in all cases I tried so far.  For example, for me both work with some local executable or with "ls" (which is in /bin), and neither works with "which" (which is in /usr/bin).

I don't think bin is included:

  > PATH= gzip
  bash: gzip: No such file or directory

But you are right...unset seems to be equivalent to PATH=
I could have sworn it acted different when I tried it, but I just ran it again and it it acted the same, which makes sense.

So given that I agree that the fix is good.  And further fixing is not :).

Grrr, ok, I have an "alias ls='/bin/ls'".  It seems that both "PATH=" and "unset PATH" are equivalent to "PATH=.".  This is behavior that we cannot add to PYTHONPATH, I fear, because so far "." is not implicitly included if PYTHONPATH is not set.  Or if we do it's a big change, not just an internal fix...