Ask HN: start with python 2.6 or 3?

[jgamman]

if you were learning python right now as your first language and you wanted it to play well with others (django, scipy, matplotlib, GAE etc) - do you start with something stable with loads of documentation and then migrate to python 3 (heck it's 3.1 already) or do you skip the mental overhead of unlearning some things that'll get baked in by using 2.6?

[etal]

The official guideline (http://python.org/download/) is fine:

Note that both Python 2.6 and 3.1 are considered stable production releases, but if you don't know which version to use, start with Python 2.6 since more existing third party software is compatible with Python 2 than Python 3 right now.

If you're just learning to program now, you'll have an easier time if you stick to the officially sanctioned features of Python, rather picking up bad habits from deprecated back roads. There aren't many of them in Python 2 compared to some other languages, but if you want to be safe, Python 3 was meant to eliminate as many of those as possible. You'll get a similar benefit if you follow the official Python 2.6 tutorial first, before reading introductory Python books that were written before Python 2.6 was released.

To be super specific, these are the things that you should try to stay up to date on:

1. Unicode support in strings

2. "New-style" versus "old-style" classes

3. The "with" statement, and context managers

4. Generator functions and expressions, versus list comprehensions and plain ol' for loops

If you watch out for these, and read the "What's New in Python X.Y" documents for each version between the one you started on and the newest release, the transition shouldn't be too hard.

The projects you're depending on may also offer some hints about when they'll do the 3.X transition. SciPy depends on Numpy, which depends heavily on the Python 2.X C extension API and will take a long time to convert. Django is basically pure Python and can probably convert sooner. GAE is Google's infrastructure and will probably be 2.X for years.

[jgamman → etal]

thanks for that and to all the other HNers that took the time to reply - i think i'll go with 2.6 but book a recurring task to make sure i'm at least aware of the changes in moving to 3.1

[anamax]

> do you skip the mental overhead of unlearning some things that'll get baked in by using 2.6

You're thinking that this is a significant thing. It isn't, at least not if you become a somewhat competent programmer.

Programming has very little to do with programming language details.

In fact, there's a reasonable argument that the more languages you know, the better programmer you're likely to be. Unfortunately, the difference between 2.6 and 3.1 probably isn't enough to help you in this regard. It is, however, enough to help you see some issues wrt language design.

[GeneralMaximus]

If all the libraries you need are already available for Python 3, then use Python 3. Otherwise, fall back to the 2.x series. Be practical. Evaluate your own needs before you listen to somebody else. The 2.x branch is here to stay for quite a long time (I'm guessing 2 to 5 years).

<BEGIN RAMBLING> If the users don't drive the library authors/maintainers to make their code compatible with Python 3, then we'll have to wait a very very long time for the 2.x branch to go away. I'm not saying Python 2.x is bad in any way. It just takes too much effort to maintain separate codebases for separate versions of the language. Also, I'm sure the language developers could do something much more useful with their time if they focussed on one version of the CPython implementation. <END RAMBLING>

[dtf → GeneralMaximus]

This is true. I would have said 2.6, because library support for 3.0 is quite wanting. But you're absolutely right: without demand it will never get there.

[pmorici]

You can use a lot Python 3 features in 2.6 and later. This document, http://docs.python.org/3.0/whatsnew/3.0.html describes the difference between the two. So while I agree with the other posters that it it would be better to start with 2.6 I also think you could minimize your upgrade pain by knowing the differences and doing as many things as you can in the Python 3 style.

[jcsalterego]

Practically speaking, if you want to leverage a lot of those libraries/frameworks you mentioned, 2.x is the way to go. However, the major syntactical and core differences that break backwards compatibility in Python 3000 (http://docs.python.org/3.0/whatsnew/3.0.html, http://docs.python.org/3.1/whatsnew/3.1.html) should be in the back of the your mind.

After working in a bit of py3k, I've found that "small things" such as using print() as a function rather than a statement comes more naturally than in Python 2.x, mostly because the latter doesn't complain about it.

Other things like xrange()'s departure in py3k as range() becomes a true iterator -- that's something you'll have to be cognizant of.

My advice is if you want to get a large project done, with support from popular and existing frameworks, go with the 2.x branch. If you're whipping up some disposable scripts, practice with py3k. And even if you're not, fire up py3k anyway.

One of the great Google Summer of Code projects this year is the inverse to the 2to3 tool -- 3to2 -- with the forward-thinking intention of making it easier to maintain one branch of code in py3k. It isn't a magic bullet by any means, but should definitely encourage the transition from Python 2.x to 3000.

[pavelludiq]

I'd go with 2.x, simply because theres more code written in it, and more code one can read and learn from. Especially if its a first language, its good to learn from real code. In the beginning it was hard for me to read source code, but eventually i started "getting it". 2.x will eventually evolve in to 3.x with 2.7 and 2.8 so its safe to go with it for now. Plus the mental overhead is not that big, 3 is not that much different from 2.x, its still python.

[rg123]

Since it will take a while for the 3rd party libraries to be ported to 3.x, start with 2.6. The versions are not different enough for it to be a big problem either way you learn them - but there are more books out for 2.x - like the Python in a Nutshell book, which I find to be the best reference (though not the best book for beginner programmers). Then, you can refer to the "What's New in Version x" entries in the documentation on the Python site to learn the later versions.

[timwiseman]

That depends on what you need. In general, I would recommend 3.0. It is more modern and has some nice enhancements and is the way to go in the long run.

With that said, some libraries still haven't made it to Python 2.6 let alone 3.0. For instance, I make use of NumPy which was compatible with 2.54 but not 2.6 the last time I checked.

[krishgy]

You start with 2.6. When you start working with python, you are not just going to use the libraries which comes with language. you need to use plenty of projects or modules developed by other users.

Start with 2.6 but consider 3.0 changes in mind like exception base, print statement. Start practice 3.0 convention using 2.6.

I think, atleast two to three years of development for people to port all their projects to python 3.0.

Sure, you will miss many things, but you are on the way.

[CyberFonic]

If only wxPython were to be available for Python v3.1 then I would choose that. Because it isn't yet, I prefer v2.6 and use "from future import ...."

If you don't choose to use any libraries that require v2.6 or older right now, then I'd encourage the choice of v3.1. I'm hoping all libraries will run with v3.1 given enough time.

[llimllib]

I would start with 2.6. The changes are not all that drastic between the two, and there's almost no software available for python 3.

[babo]

There is no point to lament, take whatever you want, it's not a problem to switch either way.

[TweedHeads]

Go with 3.

Being a trailblazer can reap rewards in the long run.

[jey → TweedHeads]

No, if you're a newbie, make things easy on yourself and use 2.6. Once you've got some experience under your belt you'll be able to easily pick up 3.0.

[jwecker → TweedHeads]

Just understand you'll be missing some very common python libraries. You could then get involved in porting them over though and get some serious geek cred. Or you may not need them anyway...

EDIT: specifically some of the libs you mentioned. If you have time, I agree go w/ 3.0- you'll know the libs that much better as well if you help port them.