This patch fixes the following minor issues with the ipaddress module:

- Removed unused property _BaseV4._valid_mask_octets
- Removed unused methods _BaseV4._is_valid_netmask() and _BaseV4._is_hostmask()
- Replaced several calls to superclass constructors by super()

It also refactors the constructors of IPv4Network, IPv4Interface, IPv6Network, and IPv6Interface.
They all now use the new method _get_addr_prefix_tuple() to parse the argument. It now matches the following sentence of the documentation of IPv4/6Interface:
"The meaning of address is as in the constructor of IPv4Network".

Additionally they now also accept a bytes or an IPv4/6Address object representing a netmask (or a hostmask [only IPv4]) as second element of an address/netmask tuple. This makes it easier to work with C-APIs that provide netmasks only as bytes object.

This patch does not try to solve issue 27683.

Any updates?

Peter, are you able to take a look at this or indicate you're happy for someone else to take it? (I relinquished my co-maintainer role for the ipaddress module a while back, so you're the only currently listed maintainer)

@moritzs  Would you create a pull request on GitHub?
Or may I create a pull request for your patch?

New changeset 2430d532e240dea55f0082d1e9bf2e0f3d7505be by Inada Naoki in branch 'master':
bpo-27860: use cached_property (GH-12832)
https://github.com/python/cpython/commit/2430d532e240dea55f0082d1e9bf2e0f3d7505be

I am not owner of ipaddress module, so I don't know we should
support more form of masks.

On the other hand, IPv4Interface and IPv6Interface expose
netmask and hostmask attributes when address is not bytes or int.

These attributes are not documented.  I assume it was added
accidentally, when coping some code from Network classes.

But there are test for these attributes.
NetmaskTestMixin_v4.test_valid_netmask runs for IPv4Network and IPv4Interface.
It checks these attributes when constructor argument is string.
https://github.com/python/cpython/blob/3c7931e514faf509a39c218c2c9f55efb434628f/Lib/test/test_ipaddress.py#L422-L440

For safety, I added these attributes always, instead of remove them.

PR-12836 is fixing it, and extracts methods for Network/Interface constructor
(a part of PR-12774).  It doesn't add additional mask form support.

New changeset 6fa84bd12c4b83bee6a41b989363230d5c03b96c by Inada Naoki in branch 'master':
bpo-27860: ipaddress: fix Interface missed some attributes (GH-12836)
https://github.com/python/cpython/commit/6fa84bd12c4b83bee6a41b989363230d5c03b96c

I merged all cleanups.

I don't merge accepting any IP representations as mask, because I'm not expert of this module.
At least, I don't want to pass prefix/netmask by something like IPv4Network("0.0.255.255").

I was wondering why it was decided against backporting to 3.7? 6fa84bd12c4b83bee6a41b989363230d5c03b96c fixes an actual bug #35990 (string mask in tuple argument for IPv4Interfaces). Incidentally, there are also no tests for this behaviour.

Just to note: both merged PRs lacked news entry.

> I was wondering why it was decided against backporting to 3.7?

Because I treats this is just a code cleanup.

> 6fa84bd12c4b83bee6a41b989363230d5c03b96c fixes an actual bug #35990 (string mask in tuple argument for IPv4Interfaces).

I didn't thought it was a bug.
I don't know it is real bug which should be backported to 3.7 for now.

> Incidentally, there are also no tests for this behavior.

Because I thought there are no change about public (documented) behavior.

> Just to note: both merged PRs lacked news entry.

NEWS entry is not needed for code cleanup.

I for one have encountered #35990 in my work and would appreciate a backport.

OK, I thought I improved only undocumented behavior, but it was documented and previous behavior didn't follow the document.
I'll backport PR 12836.

New changeset f532fe5583d29d21e12aa22d8fca13e3bca94fb3 by Inada Naoki in branch '3.7':
bpo-27860: ipaddress: fix Interface constructor (GH-14200)
https://github.com/python/cpython/commit/f532fe5583d29d21e12aa22d8fca13e3bca94fb3