$ PATH= /usr/bin/which python
$ PATH=: /usr/bin/which python
./python
$ PATH=/usr: /usr/bin/which python
./python

>>> shutil.which('python', path='')
'/usr/bin/python'
>>> shutil.which('python', path=':')
'python'
>>> shutil.which('python', path='/usr:')
'python'

First, I propose interpret path='' as an empty path, not as a default path (we have None for this). However the interpreting of an empty directory in non-empty PATH can be platform-depending.

I'm not sure reproducing the quirks of /usr/bin/which is a good idea. shutil.which() is meant to be useful and easy to understand, not to be 100% bash-compatible.
And, anyway, what would be the point of passing an empty path, if the return value is guaranteed to be None?

/usr/bin/which is not a Bash. ;)

The path can be unexpectedly empty. If we got None then we can detect the error, but if we got something out of the path then we can miss our fault.

What I think it is suppose to do (the user expects it to do) is find the program that would be run if the command were typed at the command prompt.

rdmurray@hey:~>which python
/usr/bin/python
rdmurray@hey:~>export PATH=
rdmurray@hey:~>which python
python not found
rdmurray@hey:~>python
zsh: command not found: python

As Serhiy noted, this result may be platform dependent.  Which is unfortunate.

No, I noted that result of PATH=: or PATH=$PATH: can be platform dependent (I'm not sure).

I was speaking in general of 'which program would be executed if the command is typed at the prompt' as being system dependent, which it demonstrably is since the behavior on unix and windows differs with regards to the current directory.

And no, what I wrote wasn't clear :)

FWIW, the POSIX standard gives some guidance on how PATH is to be interpreted for conforming systems, including:

"A zero-length prefix is a legacy feature that indicates the current working directory. It appears as two adjacent <colon> characters ( "::" ), as an initial <colon> preceding the rest of the list, or as a trailing <colon> following the rest of the list. A strictly conforming application shall use an actual pathname (such as .) to represent the current working directory in PATH."

http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap08.html

Thank you, Ned, for information.

Here is a patch which remove the first difference (processing an empty path).

The second difference is not semantically significant and I'm not sure whether we need to get rid of it.

Ping.

The result of PATH= is also platform dependent. Testing on OS X which has a BSD heritage rather a Linux one:

$ PATH= /usr/bin/which python
./python

# without patch
$ PATH= ./python -c 'import shutil; print(shutil.which("python"))'
python
$ ./python -c 'import shutil; print(shutil.which("python", path=""))'
/usr/bin/python

# with the patch:
$ PATH= ./python -c 'import shutil; print(shutil.which("python"))'
None
$ ./python -c 'import shutil; print(shutil.which("python", path=""))'
None

So, for OS X, shutil.which doesn't match /usr/bin/which behavior for the PATH= case either with or without the patch.  FreeBSD (8.2) /usr/bin/which is the same.  The other cases are the same as Linux.

I suppose the patched behavior is preferable, though.

In any case, the shutil.which docs also need to be updated.

From a documentation standpoint, path='' is not the same as "When no path is specified", so indeed it should return None when path=''.  Serhiy's patch looks good to me.

Serhiy, I'd say go ahead and commit it.  +1 from me.

New changeset eb8c575fa781 by Barry Warsaw in branch '3.3':
- Issue #17012: shutil.which() no longer fallbacks to the PATH environment
http://hg.python.org/cpython/rev/eb8c575fa781

New changeset 8f5b37f8f964 by Barry Warsaw in branch 'default':
- Issue #17012: shutil.which() no longer fallbacks to the PATH environment
http://hg.python.org/cpython/rev/8f5b37f8f964

I couldn't wait. :)