All Python 3 releases ?

[lgautier]

I have been in the trenches of Python 2->3 for quite some time now with rpy2, and some of the earlier Python 3 releases did not appear to gain as much traction as the later ones. For rpy2, the target is currently Python 3.5, with the possibility (but absence of support) for going back up to Python 3.3.

Would it make sense to make Python 3 version required on the Gantt chart on your web page ?

[Carreau]

That's up to you. One of the problem is that the gantt chart is starting to become big. As long as it reflect what you are planning to support (like one LTS version and one 3-only) that would be fine.

[takluyver]

I don't think most projects support the first couple of Py3 versions, and I don't think it's terribly important to show that - people looking at switching now won't be switching to 3.2. As @Carreau said, we're also looking for ways to shrink the Gantt chart, or better display that kind of information.

I think one of the reasons that we added the Gantt chart originally was to highlight that IPython 5.x will get bugfix releases for longer than usual, so although IPython 6 will require Python 3, we're not ditching Py2 users just yet.

[lgautier]

@takluyer: the idea is more to communicate around what is meant by "Python 3" and may be give a sense to users/developers about what is the version they should aim for depending of their project. I might be wrong but the purpose is to communicate that it is about time to move away from Python 2 (creating a support burden). If so, the target audience seems to me be Python 2 users, some of them in environments where there is a globally maintained Python installation, and the early notion that "Python 3" means in fact Python >= 3.3, >= 3.4, >= 3.5, or soon >= 3.6, rather than "whatever Python 3 is installed" might help with the transition. Just a thought.

[tacaswell]

👍 for providing some guidance at what 'python3' means, it will help reduce the frustration of 'I did pip3 install blah and it did not work' where pip3 is an old system python pip.

I advocate it explicitly be >=3.4.

On the other hand, the current debian stable version of py3 is 3.4 (old stable is 3.2), the ubuntu 14.04 is 3.4 (12.04 is 3.2), as near as I can tell fedora only has details back to f23 which has py3.4, and centOS does not seem to have python3 (?!) so this may be a very small issue.

[lgautier]

Looking at what main distros ship certainly makes sense.

Ubuntu 14.04 is quite widespread, but the latest LTS is 16.10. There python 3 is 3.5.

The argument against making 3.5 a starting point is that not all places will be able to move to a distro released in 2016.

Arguments for having Python 3.5 as a starting point for scientific software are:

• matrix multiplication operator @ (PEP 465)
• %-formatting for bytes and bytearray (PEP 461)
• async/await keywords language keyword is only in 3.5 (PEP 492) and they might be handy for data-ingesting code

[Carreau]

Other arguments may also be that 3.4 is EOL before 2.7 EOL. So if we are arguing earlier about EOL....

[mscuthbert]

At this point it seems unwieldy to add this information to all the signed projects unless it could be done automatically.

[lgautier]

I think that the momentum for Python 3 is now where we wanted it to be, but while some of the point raised here might no longer be relevant as I think that the wider community of users knows that there are several Python 3 releases and there exists incompatibility between them.

Unless there is a common agreement about a minimal Python release, and then the reason(s) for having one can easily be disputed, it is indeed not practical to have that information on the chart.

[lgautier]

Other arguments may also be that 3.4 is EOL before 2.7 EOL. So if we are arguing earlier about EOL....

As of today Ubuntu 14.04, with Python 3 being 3.4, is still the default ubuntu on Travis for example.
The point here is about clarifying with users that "Python 3" doesn't mean "Any Python 3".