Comment by ramanujan | Hacker News Reader

Comment by ramanujan | Hacker News Reader

Parent Full thread

Unfortunately, Python 3 is the new Perl 6. It offers reduced performance[1], changes some of the hardest-to-debug parts of the language (like Unicode), and capriciously introduces backwards incompatibilities. For no obvious gain!

As just one example, could Guido not have made the print statement a function through the simple mechanism of:

  from __future__ import printfn

Then new code could use printfn and old code could use print, and all would be well. But instead, literally billions of dollars will be wasted changing every single print keyword statement to a print function invocation. Sure, you can use 2to3. But the tool is that worst of all worlds, a compiler that doesn't really work 100% and actually reduces the performance of your code in the compilation process.

For those who do not want to waste their time on this port, Python 3 is going to accelerate the move towards full stack JS just like Perl 6 accelerated the move towards Python/Ruby. This JS stack will be MongoDB + node + Backbone/coffeescript or some similar permutation.

After all, if you are going to literally have to recompile your language with 2to3 and deal with all the edge cases, why not go whole hog with something like emscripten[2] or a souped-up pyjamas[3] to convert your legacy Python 2 code to JS? Then people just need to start porting C/C++ numerical libraries over to V8 through something like v8-juice[4] and we will have all the missing pieces.

[1] http://docs.python.org/release/3.0.1/whatsnew/3.0.html#perfo...

[2] https://github.com/kripken/emscripten/wiki

[3] http://pyjs.org/

[4] http://code.google.com/p/v8-juice/

> http://docs.python.org/release/3.0.1/whatsnew/3.0.html#perfo...

The thread is about 3.3, so you link to a 3.0.1 release note? That is three years old.

> literally billions of dollars will be wasted changing every single print keyword statement to a print function

I'd love to see the math on this, especially because I did it in a matter of minutes across all of my code.

> 3 years old

I'm pretty sure they'd have announced it if Python 3 ever got faster than Python 2. But this is not mentioned in the 3.1, 3.2, or 3.3 notes on performance:

  http://docs.python.org/dev/whatsnew/3.1.html#optimizations
  http://docs.python.org/dev/whatsnew/3.2.html#optimizations
  http://docs.python.org/dev/whatsnew/3.3.html#optimizations

> see the math

That one is easy. The Python 3 conversion is a serious project for core devs at Django and Numpy. Check out their mailing lists about what this takes/took.

http://blog.jarrodmillman.com/2009/01/when-will-numpy-and-sc...

  While the above procedure of using the 2to3 tool works 
  relatively for pure Python code, there is no automatic 
  conversion tool for extension code. NumPy is mostly 
  written in C and makes extensive use of the Python C-API. 
  So converting NumPy will require much more than running 
  the 2to3 tool.

https://docs.djangoproject.com/en/dev/faq/install/#can-i-use...

  ... it will be a while before most Python software catches 
  up and is able to run on Python 3.0. For larger Python-
  based software like Django, the transition is expected to 
  take at least a year or two (since it involves dropping 
  support for older Python releases and so must be done 
  gradually).

A "year or two" of the core Django team's time is easily several million bucks. Multiply by the Python codebases of every company on this list [1], and everyone on github, and every egg out there, and we are talking billions for the conversion.

[1] http://wiki.python.org/moin/OrganizationsUsingPython

> I'm pretty sure they'd have announced it if Python 3 ever got faster than Python 2.

According to this blog post, 3.2 performance now exceeds 2.6 and 2.7, using the same pystone benchmark that the 3.0 release notes refer to:

http://www.levigross.com/post/2340736877/pystone-benchmark-o...

From the full article, "Python 3.2 [beta] looks promising coming out just a little faster than Python 2.7.". This same article has PyPy coming in at 3 times as fast as Python 3.2 and Python 2.7 on the pystone benchmark, but if you have used PyPy it is not fast.

Looks like we need another benchmark. Does anyone have the Python 2 code from the Shootout? It seems to have been removed so that you can no longer compare Python 3 to Python 2 directly.

http://shootout.alioth.debian.org/u64q/benchmark.php?test=al...

Supporting your statement:

I am positive that Python 3 could be a lot faster than Python 2. If people are interested in what's possible, they should follow the corresponding mailing list (python-dev).

> A "year or two" of the core Django team's time is easily several million bucks.

Ahhhahaha, ahhhahahahahahahahahaha! Good lord I wish this was true!

https://docs.djangoproject.com/en/dev/internals/committers/

Jacob, there look like about 20 or more programmers on that page. At even $100k per programmer, isn't that about $2.4 million? And isn't the number much more than that given that you are among the most skilled engineers?

* We have about 30 people with the commit bit; at any given time I'd guess about a dozen are active.

* "Even 100k" implies that you think a six figure salary is reasonable for programmers. This is true in some rarefied echo chambers likethe Bay Area where programmer salaries are overinflated, but it's far from a universal The majority of our committers don't live near San Francisco. Actually, I think most live outside the US these days. I don't believe the median salary of a Django committers is 100k -- though for my fellow committers' sake I hope I'm wrong :)

* I only wish "being a core Django committer" was some magic "Get a Great Job Paying Six Figures Card". No good hiring manager assumes that "has a Django commit bit" translates to "among the most skilled engineers"; I only wish I could pull down "much more than" 100k based solely on my Django status. It's certainly a good time to be a Django dev, but while being contributor might help you get your foot in the door, it won't land you a salary increase over your peers.

* None of us get paid to work on Django anywhere close to full time. We get paid to do other things; we're lucky if we get a few hours a week to hack on Django as a side project. This again shows that our value in the salary marketplace isn't linked to our willingness to volunteer time to Django.

* Finally, this whole silly exercise assumes it would take "a year or two" of ALL of our time to get Django onto Python 3. That's laughable: it's mostly done already with perhaps a couple of man-months of activity. There's nowhere near 20 programmer-years worth of work; suggesting such is simply insane. Toss me that 100k you're bandying about so easily and I'll have it done by May :)

> Toss me that 100k you're bandying about so easily and I'll have it done by May :)

Considering merge with trunk will probably be done shortly after 1.4 I don't think he'll be getting his money's worth ;)

You seem to be making the assumption that many of those people are working on Django full time. It's pretty clear from their descriptions that most of them have other day jobs.

Also, in the original "year or two" mention, that's wall-clock-time, not applied-effort-time. While waiting for the world to adjust at a gentle pace, plenty of other productive work will be underway.

> I'm pretty sure they'd have announced it if Python 3 ever got faster than Python 2. But this is not mentioned in the 3.1, 3.2, or 3.3 notes on performance:

Do these documents really need to come out and spell it out for you? If you were to read what the document is telling you, you could probably figure it out.

> That one is easy.

Your original comment was about the print statement changes costing so much time and money, but now they are about the entirety of Django and NumPy. Yeah, it certainly costs money for a big project like Django to port (and people have completed it in their free-time), but print function/statement changes are trivial and do not cost on the order of billions of dollars.

Reply on news.ycombinator.com