Wow, I haven't heard of pathlib [1] before now, but it looks awesome! In particular, using the slash operator to join paths is brilliant.

    >>> p = PurePath('/etc')
    >>> p / 'init.d' / 'apache2'
    PurePosixPath('/etc/init.d/apache2')

[1]: http://docs.python.org/3.4/library/pathlib.html

Scala does something similarly neat with using #| for piping system commands.

Slash for Scala is not used by default though for path joining, but it can be by adding an implicit function:

  implicit def slash(str: String) = new { def /(joinStr: String) = str + java.io.File.separatorChar + joinStr} 

  val joinedStr = "abc"/"def"/"ghi"
  > joinedStr : java.lang.String = abc/def/ghi

Shell friendly syntax can be used for stdin/stdout in Scala as well (already in the stdlib with no implicits required):

#< Redirect STDIN

#> Redirect STDOUT

#>> Append to STDOUT

#&& Create an AND list

#!! Create an OR list

http://stackoverflow.com/questions/12772605/scala-shell-comm...

http://alvinalexander.com/scala/scala-execute-exec-external-...

There are Plumbum shell combinators in Python http://plumbum.readthedocs.org/en/latest/

There are a couple of modules that do this kind of thing but Plumbum looks interesting, particularly

http://plumbum.readthedocs.org/en/latest/local_commands.html...

>from plumbum.cmd import cat

>>cat('tmp.txt') #invokes cat tmp.txt on the system

I hadn't heard of it either, although I have played with Twisted's FilePath[1] class previously.

I didn't really understand the attraction of an OO filesystem API until I actually used it in anger and realised how handy it was that all the relevant functions were on the object I was messing with, instead of arbitrarily divided between the "os" and "os.path" modules.

Some (hopefully constructive) suggestions for the pathlib API:

- Python 3 dicts ditched .iterkeys(), .itervalues() and .iteritems(); why does pathlib use the clumsy .iterdir() when it could just be called .children?

- Like many Python programmers (I assume), I'm more familiar with POSIX command-line tools than I am with POSIX APIs. Command-line tools generally have an option to control whether they dereference symlinks (cp -L, find -L, etc) rather than having a separate. parallel tool that does the same thing. I think it would be neater if .lstat() and .lchmod() were removed in favour of a "follow_symlinks" parameter to .stat() and .chmod() (and .is_dir() and .touch() and .owner(), and, and, and..)

- If you're going to have convenience functions that pick out interesting bits from the .stat() return value, I would find .size() far more practical than .is_socket()

[1]: http://twistedmatrix.com/documents/current/api/twisted.pytho...

One of the many things I like about Python is their pragmatic approach to operator overloading. Well, their pragmatic approach to everything, really.

FWIW, the boost::filesystem C++ package uses the / operator in the same way

No one objected to using "pure" instead of the more common "abstract" (which is the actual antonym of "concrete") ?? Why do people do this?

Concrete paths inherit from pure paths - this would be quite strange using antonyms. This isn't like an "abstract" base class. It's a full class on its own, which is extended by another.

Anyway, I think it's meant to be "pure" like pure functions, generally having no side effects or external inputs.

https://en.wikipedia.org/wiki/Pure_function

i think so too. like it says in the docs: "Regardless of the system you’re running on, you can instantiate all of these classes, since they don’t provide any operation that does system calls." http://docs.python.org/3.4/library/pathlib.html#pure-paths

The precursor path library [1] by J. Orendorff had the slash operator overloading trick a decade (or more) ago.

[1] https://pypi.python.org/pypi/path.py/1.2

I thought operator overloading like that was considered bad practise. It seems to me that the + operator should have been overloaded, not the / operator.

Been using python for 7+ years now. Does everyone really think people are willing to adopt 3.x anything...? The reality of it is every place I have worked (and currently working at). Is bound by 3rd party tools and what ever interpreter they built against. Mostly that 2.6,2.7 right now. And it does not look like anyone will be switching anytime soon.

Features look great, cant wait to "play" with them. But that;s about all. Play.

I certainly can't speak to big enterprise companies, but we've used Py3 in production at the last 2 startups I've worked at, and I use it extensively in my own side projects.

I certainly understand it can be somewhat tedious to convert legacy projects -- I've had to do it a few times myself -- but for any new dev, I'd highly recommend using Python3.

It's come to the point that Python3 is the assumed default for any new project - If someone wants to use Py2 for a project, they better have a compelling reason.. Almost every library we'd want uses Py3, or there's an alternative that does (such as PIL->Pillow).

I find it rather annoying when I go back to a Python2 project and have to constantly jump through hoops to avoid what I know are solved problems :/

Admittedly, my experiences (like yours) are anecdotal - But to answer your question, from what I've seen, Python3 is used in production quite a bit.

Everyone is aware of this reality. Everyone was aware of this reality when the transition was planned. It's been years now and we're still maybe halfway through the transition and that's ok.

I've done some internal work written entirely in Python 3.3. We largely did it as a "proof of concept" to try out the new language. And yeah, we did indeed have dependencies. No complaints.

Going back to 2 is kind of a bummer when you add up all the little niceties you get with 3(.3): unicode native strings, smarter import system, yield from, keyword-only arguments... at least those were the features I used. It has some other stuff that I never got around to touching.

I don't understand why you are being downvoted for your data-point. I've not seen 3.x used anywhere in anger either. But I've seen plenty of people embarking on new 2.7 projects. That's just my data-point too.

I understand why GP is being downvoted. He presents his anecdote as a general trend, dismissing python3 and its features as play without providing proofs.

Well, some people are bound to already installed interpreters. And those come with Linux distros. Many of which will be using Python 3 as the default for their 2014 releases (Fedora, Ubuntu, Arch). Plus these distributions are helping/pushing 3rd party Python developers to port to Python 3.

So 2014 really looks like Python 3 mass adoption.

In corporate environments basically nobody uses Fedora, really nobody uses Arch, and some people use Ubuntu. RHEL7 will be new in 2014, and it will have Python 2.7. We can expect RHEL8 to come with Python 3, in 2017, and the migration to that should be almost complete by 2021. I wish I were exaggerating.

Mostly interested by Python3 because of asyncio (formely tulip). Trying to beat nodejs at its own game. I'm waiting for PyPy Python3 compatible to be released to do some comparisons. Anyway PyPy seems to already beat nodejs, Python 3 will just make working with async code easier.

I'm feeling optimistic about Python 3. For personal and work projects I've been writing 3-compatible code, and am hoping to have a production system using it by the end of the month.

I think everyone is expecting a really long tail to the transition, but we'll get there eventually.

> a new C API for implementing custom memory allocators

Wow. This'll lead to some interesting hacks, I think.

Finally all the hacks done internaly by some big Python company is released in the wild and not by them. Check out the presentation by the creator (in french) https://github.com/haypo/conf/blob/master/2013-PyconFR-Stras...

I started the development of tracemalloc at Wyplay, but I rewrote all the code for the version included in Python 3.4: complelty different API, you get different data, new implementation, etc. The old code: https://github.com/wyplay/pytracemalloc/

The PEP 445 (malloc API) also allows to plug hooks on memory allocators at runtime, instead of having to recompile Python for that.

tracemalloc is the first user of the PEP 445 (malloc API), but I wrote another toy: https://bitbucket.org/haypo/pyfailmalloc

"Debug tool for Python injecting memory allocation faults to simulate a low memory system to test how your application handles MemoryError exceptions."

isn't 2.7 the best python still because twisted?

The most exciting feature in 3.4 (at least in my opinion) is the addition of asyncio to the standard library, which aims to provide a standard event loop which libraries like Twisted can then use to make themselves interoperable with other async Python solutions:

http://www.python.org/dev/peps/pep-3156/

"The event loop is the place where most interoperability occurs. It should be easy for (Python 3.3 ports of) frameworks like Twisted, Tornado, or even gevents to either adapt the default event loop implementation to their needs using a lightweight adapter or proxy, or to replace the default event loop implementation with an adaptation of their own event loop implementation. (Some frameworks, like Twisted, have multiple event loop implementations. This should not be a problem since these all have the same interface.)"