If $HOME is not set, netrc will raise an exception.

Attached patch fixes the problem by using `os.path.expanduser` instead

Thanks for the patch, Dimitri. I think this is a reasonable improvement. However, since we are changing the behavior of the netrc() class, I'm not sure this can be considered as a bug fix.

In any case, 3.3 and 3.4 are in security-fix-only mode so I'm going to remove them from the versions field.

We need two things from you to move this forward:

1. A test. It should go in Lib/test/test_netrc.py. You can use

       env = support.EnvironmentVarGuard()
       env.unset('HOME')

   to test the new behavior.

2. A CLA form. You can sign it online at https://www.python.org/psf/contrib/contrib-form/

> However, since we are changing the behavior of the netrc() class, I'm not sure this can be considered as a bug fix.

I was not sure either. The patch does change behavior in subtle ways...

I've added a new patch, and sent the CLA form.

Thanks for your time!

On Windows, HOME is both split into two variables and replaced by USERPROFILE.
C:\Users\Terry>set HOME
HOMEDRIVE=C:
HOMEPATH=\Users\Terry

C:\Users\Terry>set USERPROFILE
USERPROFILE=C:\Users\Terry

So if it make sense to run this on Windows*, I would call it a bug.
* I am guessing that Windows' FTP clients do not use .netrc.

The doc says "the file .netrc in the user’s home directory will be read." without qualification by "if $HOME is set".  I don't remember how os.path.expanduser might otherwise find the home directory on unix and don't know what unix users might reasonbly expect.

> if it make sense to run this on Windows

I found this issue while running cross-platform code. I needed to store some credentials, did not mind having them in plain-text and I thought .netrc was a good place for this. (did not need to re-invent the wheel ...)

>  don't know what unix users might reasonbly expect.

Well, I guess as a bonus of this patch we could add a link to the `os.path.expanduser` section[1] in `netrc` documentation.

The behavior when $HOME is not set is properly documented there.

[1] https://docs.python.org/3/library/os.path.html#os.path.expanduser

FWIW, some history: The $HOME lookup was in Guido's original 1998-12-22 patch:
        if not file:
            file = os.path.join(os.environ['HOME'], ".netrc")
The lookup was wrapped in try-except 4 years later to give a 'consistent error message'.

Module ntpath dates back to 1994.  Expanduser only used HOMEDRIVE and HOMEPATH.  Guido added support for HOME in 1997.  Support for USERPROFILE was added a decade later.

Since the module is NOT marked 'Unix-only', I think it a bug to not use os.path.expanduser.  Larry, Ned, do either of you have an opinion on whether the change should be made in current versions or only 3.7?

During review SilentGhost suggested that maybe a test was not essential.

In any case, I think patching documentation about the new behavior won't hurt. 

Do you want me to do that?

New changeset 8d9bb11d8fcbf10cc9b1eb0a647bcf3658a4e3dd by Berker Peksag in branch 'master':
bpo-28334: netrc() now uses expanduser() to find .netrc file (GH-4537)
https://github.com/python/cpython/commit/8d9bb11d8fcbf10cc9b1eb0a647bcf3658a4e3dd

Thanks, Dimitri.