Python is now Python 3 | Hacker News Reader

Python is now Python 3

archlinux.org

122 pointsenduser64 comments

I'm using Arch and the transition was smooth for all the packaged libraries including Numpy, Scipy, matplotlib, and friends, as well as pure Python packages like mpi4py and petsc4py which I install from upstream since I track development versions. The real pain is for projects that are not pure python libraries, but want to ship portable scripts. Since the "python2" link is not ubiquitous, they cannot put "/usr/bin/env python2" in the scripts (like Arch packages do, perhaps just using sed). Unfortunately there are still modern installs of Python that do not have "python2" links, so this is going to be a long process when it's not feasible to support both python-2.x and python-3 within an identical codebase. That is challenging if you have to support e.g. python2.3 which will be around for a while longer (RHEL4). Also, RHEL5 (rather, the CentOS system I have a login on) does not have a python2 link (it has python2.4) and that will be around for another four years.

I think Arch's decision was a bit premature, but waiting isn't going to make the issues much less painful if your project will always have to support an 8-year age span of distributions.

I think Arch's decision was a bit premature

This change has only been made in the 'testing' repositories.

Not really, Arch's default python right now is python3.

Nope. Python3 is now default python in the regular repos. It was in the testing for a while and has been approved.

Didn't like the move myself. pygtk broke for me and with it a small bunch of other softwares.

As a result, #python has been full of people complaining their Python installs are hosed. Just like ruby1.8 -> ruby1.9, this was a bad move, and we told them (Gentoo as well) before they did it. I can only fully support my friend and very experienced Python hacker Allen when he says:

20:20 <dash> well that confirms my impression that arch was invented by a bunch of guys who thought gentoo was too stable and easy to use

(Inflammatory, tongue in cheek, but oh-so HHOS.)

Funny... I find it exceedingly easy to use. :)

I've been using it for over two years now. Makes a great workstation OS. You get the latest releases and all their attended bug fixes, security updates, and yes.. new bugs.

But at least in the Python world, there's still virtualenv, buildout, pip, etc. I rarely install system packages for libs anyway. I don't really see how this is going to bring down the house. At the least if it does work out well and makes enough people happy, there will be more reason for people to hurry up the transition to py3 already.

(An aside: as much as I disagree with py3 forgoing practicality for purity, I wish the transition would be over already so we can get on gettin on).

I noticed the news post a day before my latest update, so I was prepared too. One should always follow package source's and general news of the distribution one uses.

Arch has the least amount of breakage of any distro I've ever used. Slack, Redhat, Gentoo, all of them would occasionally shit all over themselves, and it would sometimes be tricky to fix. The couple times that Arch did that, it was really easy to fix, because things were "normal". The Arch philosophy seems to be to defer to upstream as much as possible, which makes things easy to fix.

You can't always decide something is good or bad based on number of people running to IRC asking for help. Well let me make another "tongue in cheek" -- many o f them should be told to RTFM and read their distro's announcements". Don't support lazy folks who fail to do that.

Not a good move. Considering the large base of programs that expect Python 2.x, the safe thing to do would be to keep /usr/bin/python pointing at the 2.x binaries while using /usr/bin/python3 for the 3.x ones. At some point in the future if Python 2.x really is dead, you can just point both python and python3 at the same thing.

This seems more like somebody trying to evangelize for Python 3 and against Python 2, by purposely breaking a lot of old software, and quite antithetical to the Robustness Principle that I would expect most good software to at least attempt to shoot for.

Hopefully other distros will take a more conservative path. Python 2.x is going to be around for a long, long time.

Chances are exceedingly good that Arch package maintainers will be checking their Python applications and updating them to use /usr/bin/python2 if they don't run under Python 3. Actually, this should have already happened since the change has apparently been in testing for awhile.

I'm not really familiar with what Arch package maintainers do but aren't package maintainer mainly packager and not coders ?

I imagine that someone could easily be packaging 10 Python apps or libraries. He can't be upgrading all that code to Python 3 as it is not always a trivial task and would require a lot of work and constant patching if the developing team keeps working with Python 2.X

But overall, I think this is a good way to push developers to upgrade or at least prepare more to move to Python 3.

You're misunderstanding. Arch is still shipping Python 2.7 with the python2 package.

So package maintainers don't need to rewrite upstream code if it's not compatible with Python 3; they just need to update dependencies and set paths to point to the /usr/bin/python2 binary.

The platforms making bold enough breaking changes to progress are usually better for it. One case Apple on intel/OSX. Granted this is painful for legacy stuff but that only helps keep things up to date. If people don't start doing this Python 3 will never progress.

However, there should be no 'default' python. People should go get 2 or 3 as needed. If there is a default it should be the latest version.

Did Apple really make a 'breaking change' with the switch to Intel? PPC apps ran with a PPC emulator. As far as I know they didn't break, they were just deprecated.

There's a big difference between breaking and deprecation.

Python2 is also still around, if you need it.

It's not around as `python2` on other distros though, which is the problem. Either all other distros have to change to conform to arch, or arch has to change it back to how it was.

My Ubuntu has python2.

I agree with that generally, but in the case of Python 3, there are real roadblocks at this point, not just intertia. Central pieces are missing. There is not a single postgres adapter that works reliably with Python 2 and 3. It's impossible to write robust web apps using Python 3 because of the WSGI conundrum.

Actually, the situation is very damaging for Python, because Python has momentum _now_ and many new users are confused and put off by the fact that they cannot use the latest Python version for their projects.

Granted, these issues affect mainly web development, which may not be the main concern of a Linux distribution.

I failed to find a reference, but I seem to remember the Python team deciding at some point that they intended to keep the name "python" for the Python 2.X binaries perpetually, and require Python 3.X to be invoked as "python3". Arch might be alone in making this change, and inconsistent with other Python distributions.

EDIT: I can't find a conclusive decision but here is one discussion on the subject: http://mail.python.org/pipermail/python-3000/2008-February/0...

Arch has done this before. They only provide Ruby 1.9, even though 1.8 was (and remains today) the norm. I tried pointing this out and they told me to use a community package, which is unreliable and a huge pain. Their package philosophy is bleeding edge to an absurd degree, as it simply doesn't work. I stopped experimenting with Arch because of it.

On the other hand, making your software work with Ruby 1.9 is something you'll have to do anyway, sooner or later. As long as there are no release-critical bugs in Ruby-1.9 ... which I guess aren't, because otherwise I don't think Arch would provide them.

Python3 is almost 2 years old this decision isn't about the bleeding edge, it's about damn time.

yeah.. in Python world it's still pretty bleeding edge. There's a reason they kept going with 2.7 in parallel with 3

Would you consider Linux 2.6 'bleeding edge'? I mean, they have kept the 2.4 series going. Just because some development occurs in parallel, doesn't mean the new version is 'bleeding edge'.

The Python 3 migration process, for actual Python software in the wild, is and has always been meant to be a multi-year thing. As such, Python 3 is still very much "bleeding edge" in the sense that most projects are still working on their migration, and likely will be for some time.

I don't understand why more people don't get that. Guido has always said it will take a few years for it to all get over to P3. The Django folks are just now deciding when to get Django over to P3. If I have been watching the release notes for P3 versions as they come out and it still seems they are "cleaning up".

You're correct, but you misunderstood me:

I'm not saying it's bleeding edge because they're developing in parallel, I'm saying they're developing in parallel because it's bleeding edge.

It's bleeding edge because the developers explicitly refer to it as "the shiny new thing", etc.

This is one of those special case situations, since 1.9 has been the stable branch of Ruby since January of '09. That said, you're right, a sizable number of projects have stuck with 1.8.x, but you also can't blame the guys from Arch, either.

1.9.2 should change this.

Ruby 1.8 is god awful and should be taken out back and shot. Several times. In no way should anything be done to support 1.8 as "the norm", especially after 1.9.2.

Also, it's not like anyone should be using the system Ruby anyway. That's really bad practice in my opinion and there's no excuse when rvm exists.

I don't use RVM, and I've got a few excuses:

RVM doesn't work with fish. Fish is the biggest innovation in command-line shells in so long, and I'd rather drop RVM than drop fish.

Also, RVM hijacked my 'cd' command, so that 'cd' wouldn't take me back to my home directory any more.

I don't like things messing with workhorse commands that I depend on - 'cd' has to work.

'cd' hijacking is definitely a bug. I can't reproduce that behavior. How long ago did that happen? Overall I've found rvm to be very solid software.

Not a bug - it's part of RVM's design.

RVM reads .rvmrc files in each directory you change into. How does it do this? By redefining 'cd' to be a function wrapping the real CD command, with some extra RVM stuff in between the changing of directory and return.

Check it out: http://github.com/wayneeseguin/rvm/blob/master/scripts/cd

My problems ('cd' with no arguments didn't take me back to my home directory any more) occurred about a month ago (I was using the fish interop mode - that might have been the problem), and it's probably been fixed since. But the whole thing scared me out of continuing with RVM.

If there's a bug in RVM, I sure as hell don't want it to manifest in my 'cd' command. I need to be able to change directory, and my scripts rely on it to always work.

True, but those of us who have jobs working with ruby work with 1.8.7. We can't all rewrite our code from scratch you know.

Well if you're not living on the bleeding edge of Ruby why would you live on the bleeding edge of Linux with Arch? Any job that requires Ruby 1.8 shouldn't be using Arch.

Tangential, however I have found rvm to be incredible helpful in such situations. The ability to run multiple versions of Ruby with various sets of gems makes developing Ruby 1.8/1.9 apps painless. So much so that this is my default mechanism of running ruby anywhere (dev/prod).

http://rvm.beginrescueend.com

"bleeding edge" doesn't usually refer to released software.

Even Slackware only provides 1.9.x now.

Use RVM if you still need 1.8.x for something.

I haven't found that their package philosophy "simply doesn't work." It's worked pretty great for me. Maybe we're doing different things, though.

I'm grateful for arch and its users, like gentoo users before them, for being a giant expendable human wave of beta-testing to make my life with foss easier.

I absolutely agree, as an Arch user, with everything you say, except for the "expendable" part.

Seriously, some distro had to be first, and which one is better positioned than Arch both technically and philosophically to take that hit?

All the other distros can watch and learn.

(And I can finally learn all the stuff I have to do to make my scripts work with python3.)

I agree. I use Arch and I think it is the best Linux distro I have ever installed.

Or, to quote Allan McRae:

"Arch is bleeding edge. We do things first. We experience the pain before others. That what makes us full of awesome."

This has been in testing for some time in Arch. The Python maintainers decided to hold off on releasing 2.7 until all packages depending on Python could be rebuilt to make the switch from UCS-2 to UCS-4. Now Python 3 and Python 2 are both UCS-4, and the 'python2' package is 2.7.

I updated Arch on my workstation and server today and can confirm that everything "works for me" except for some python packages installed manually from AUR (e.g. offlineimap).

Make sure to check your AUR packages.

I think the real point is how many third party libs and packages support Python 3 yet. http://dev.pocoo.org/~gbrandl/py3.html

Most Python programmers only use a handful of significant third party libs. I use fewer than a dozen. What is more significant is WSGI, which still has not been finalized for Python 3. There are two competing, unimplemented PEPS: 444 and 3333.

I have also heard some comments about the Python 3 standard library leaving some things to be desired. Perhaps someone more knowledgeable can elaborate.

for web (framework). Google App Engine still depends on Python 2, Django has some hesitates for the Python 3 port[1], the thread has some discuss about Python 3.2, peps 333(3) too. and have a Python 3 version itself is not the same thing it can be put into production as it's tried-and-true Python 2 version.

[1] http://www.gossamer-threads.com/lists/python/dev/862066#8620...

Good. Someone had to step up and do this eventually.

Indeed. Python is moving forward, and 3.1.2 should be the default. For backwards compatibility, install 2.4.ancient1.

I'm not using Arch, but choices like this really improve its standing in my eyes.

Interesting discussions on it at: http://tinyurl.com/2cepr72 and http://tinyurl.com/25pgqql

More in-depth post at: http://allanmcrae.com/2010/10/big-python-transition-in-arch-...

Seems there's not a lot of uproar or cheering on it (though it depends on who's doing the posting.)

As I see it, the change is minor and doesn't have any practical consequences, except for:

(1) the extra work for Arch packagers, who will need to change, in all packages, every Python2.x code that expects /usr/bin/python to be a Python2.x interpreter.

(2) making Python 3.x feel like the de facto standard in the distribution sphere.

This update broke the python-cheetah package which I was relying on.

For now, I have to use a self-compiled 2.6 python package. I have locked the python and python2 packages from further updates for now. I hope this situation get better soon.

I hope you submitted a bug against python-cheetah...?

(This kind of thing is what I signed up for when I installed Arch.)

I recently started a new project but had to go for 2.7 instead of 3.x due to the lack of 3.x support in Django. I would have loved to use 3.x. Oh well.

Use Buildout and get a python shell for any version you want in your project directory. Then add a few things to your .gitignore and distribute/deploy.

Call me when a distro that implements package signing has shipped Python 3.

I know it's a little OT, but I can't take anyone seriously who ships unsigned packages and calls them a distribution.

http://fedoraproject.org/wiki/Features/Python3F13 (Fedora 13 was released on the 25th of May).

Ubuntu uses signed packages, and has been shipping Python 3 for some time. Debian probably includes it in their unstable repository.

Based on the tenor of your comments and OT nature of it, I also cannot take anyone seriously who talks like this, without appropriate context.

Do you know why Arch does not sign its packages?

Why not simply start using /usr/bin/python3 always for Python 3?

Reminds me of Zeds rant 'Making Debian Responsible For Its Actions'.

http://news.ycombinator.com/item?id=1734936

Too many third party libs and packages still don't support PY3. Not a good idea.

This was very foolish.

Reply on news.ycombinator.com