Python socket programming tutorial | Hacker News Reader

Python socket programming tutorial

binarytides.com

88 pointssilv3r_m00n20 comments

If you want to understand sockets, there's no excuse for not reading Beej's Guide to Network Programming [1]. It's C, not Python, but it covers all of the dirty details and explains exactly what's going on when you create a socket. For instance, this article never mentions that sockets are file descriptors — in Beej's article, it's right after the introduction (Section 2 [2]).

How does this article describe a socket?

    > sockets are the fundamental "things" behind any kind of network communications done by your computer"

Not very helpful.

[1] http://beej.us/guide/bgnet/output/html/singlepage/bgnet.html

[2] http://beej.us/guide/bgnet/output/html/singlepage/bgnet.html...

I came here to say just this. Anyone who wants to do real socket programming should learn some of the low level reasons for how sockets work behind the scenes. Plus, it's a great read! (I especially like his story about IPv6)

Is there a reason to catch all those exceptions? If you're just sys.exit()ing anyways, why sacrifice all the traceback info?

You should always assume, that a best effort network fails. Therefore, you should always handle the exceptions ( even in a trivial example).

Decent tutorial, though I find the official Python HOWTO to be a more thorough and appropriate introduction to sockets for the complete newbie...

http://docs.python.org/howto/sockets.html

As a python newbie, I couldn't disagree more.

I would have saved hours while working on (and possibly completed) the recent Stripe CTF. Frankly, I started to dislike the python docs thoroughly. However, I'm very much looking forward to the much-rumored new docs--I very much dig the language.

As an example to my point, however, look at the two side-by-side as an aspiring tinkerer with the desire to send data over a socket. To me, the aforementioned tutorial is miles ahead of the python docs.

We may have an unreconcilable difference in philosophy here. I understand there is value in jumping straight to code, but I find real value in a (good) treatment of the model you're coding against before jumping in.

Both tutorials are good, but I think a Python programmer totally new to network programming, which is the audience of the OP, is better served first reading that HOWTO before following the code samples...

Understood.

For me, I just can't learn that way--that's surely where the discrepancy is. I've never been able to, say, read one of those O'Reilly programming books all the way through (though I have bought a couple with the best of intentions). I just have to start playing around. It's probably not the right way, but its served me well. I think this is why the python docs don't work for me--they don't seem suited to that type of learning.

I really highly recommend "Foundations of Python Network Programming" by Brandon Rhodes and John Goerzen. It's just one of the best technical books I've read, from the opening example onwards. http://www.apress.com/9781430230038

Shouldn't the first step be install http://www.zeromq.org/? That is some seriously amazing stuff right there. Makes network programming (something I've avoided for years) almost trivial.

More appropriately, the first step should be to install Twisted.

txZMQ is pretty cool if you are going the Twisted route. It's nice to reuse twisted's event loop instead of rolling your own.

Or, if you'd rather stay out of Callback Hell, you could go with the alternate approach of not using Twisted. Personally, I'm a fan of Eventlet:

http://eventlet.net/doc/

I also hear good things about Gevent, which is similar, though not nearly as well-documented. Or just use vanilla Python; sometimes it's all you need.

The point is (or should be) to learn how socket programming works so you know what's going on under all that abstraction and why. That's also why we catch all those exceptions: so we understand why something failed, and how to deal with it in the future. It's also really fun to write complex networking apps from scratch =)

>It's also really fun to write complex networking apps from scratch

I agree and disagree. It definitely is fun to write awesome complex code, in general, but I'm currently in the middle of developing a network application and the most frustrating part was the socket handling. Everything riding on top of these connections was fun, but actually working out the TCP logic was un-fun.

That said, diff'rent strokes for diff'rent folks, I'm sure you're not the only one to love TCP coding, and I'm glad you guys exist to write tutorials for people like me :)

I write modules to decode network protocols for fun. So yeah, maybe i'm not quite right in the head.

On the other hand, it taught me about perverted networking standards like 802.1Q which is more common than you'd expect, and helps when you're trying to figure out why tcpdump isn't returning results you expected.

If that was the point we should all break out our chemistry sets to refine elements so we can create our logic chips and build up from there.

At some point you have to admit you've climbed down enough turtles and it's time to get some shit done. I'm claiming ZeroMQ is far enough down. It's not that much abstraction. It's mostly avoiding you the annoyances and giving you framework of best practices.

Each hacker choses how far down the rabbit hole they go, don't disparage others for going deeper.

First of all, ZeroMQ is fucking tons of abstraction. It's its own wire protocol. It does crazy advanced things that it would take you forever to figure out and re-implement. It's nice.

You also don't have a clue what it's doing or why, which means you don't have a clue what your program is doing or why. In general it's thought of as a good idea to understand these things.

At the very least learn the tcp/ip stack and why all of the complex things are the way they are. Learn why building a long-distance fault-tolerant fast reliable network protocol is hard. Learn why there's many things that can cause your application to fail to just connect to a remote host. Heck, just learn the differences in different operating systems' tcp/ip stacks and why they are that way. Just using the same framework everywhere isn't going to solve all your problems for you.

You can't always choose the protocol you want. So no, the first step is not necessarily to use ZeroMQ, or HTTP, or Thrift, or whatever protocol you like best. Sometimes raw sockets are necessary.

Reply on news.ycombinator.com