[rurban]

The arguments presented here are extremely dumb. Of course everyone knows already that the ops itself are slow. What not many people know is that all this could be easily optimized. javascript had the very same problems, but had good engineers to overcome the dynamic overhead. php7 just restructured their data, lua and luajit are extreme examples of small data and ops winning the cache race. look at v8, which was based on strongtalk. look at guile. look at any lisp. All these problems were already solved in the 80ies, and then again in the 90ies, and then again in the 2000ies.

python is similar to perl and ruby plagued with dumb engineering. python is arguably the worst. They should just look at lua or v8. How to slim down the ops to one word, how to unbox primitives (int, bool), how to represent numbers and strings. How to speed up method dispatch, how to inline functions, how to optimize internals, how to call functions and represent lexicals. How to win with optional types. Basically almost everything done in python, perl and ruby is extremely dumb. And it will not change. I'm still wondering how php7 could overcame the very same problems, and I believe it was the external threat from hhvm and some kind of inner revolution, which could convince them to rewrite it. I blame google who had the chance to improve it, after they hired Guido, and went nowhere. You still have the old guys around resisting any improvements. They didn't have an idea how to write a fast vm's in the old days, and they still don't know.

[sitkack]

The semantics of the language are extremely flexible, that flexibility add a ton of "yeah but" when trying to do obvious things. "yeah but what if they monkey patched boolean?" ... If I were starting a Python VM from scratch, objects would be closed and only deopt if they were mucked with. Most Python code in the wild is written in a dynamically typed version of Java. Yes most Python is Java w/o the types. Typed inference, escape analysis and closed classes will allow for the same speed as the JVM.

[rurban]

Yes, but then it's not python anymore. You will get a similar speedup for a fully dynamic python with easier optimizations, as done with v8 or lua. Even without types. types and attributes (like closed classes) do help, but a proper vm design is more important.

But not within this community. It needs to be a completely separate project, as cython, mypy or pypy. pypy is pretty good already, but a bit overblown.

[sitkack]

Agree, but Pythonic-Python, isn't Python the language, it is dynamically typed Java. It isn't considered good form to use the flexibility that is there.

[orf]

> javascript had the very same problems, but had good engineers to overcome the dynamic overhead

Sure, it also had Google to funnel an insane amount of money and engineering time/skill into it. If it's so simple to do then put all your ideas into a PEP and submit it.

[rurban]

A PEP? python needs to be rewritten completely. cython, pypy or mypy are good starts, but their op and data layout is still suboptimal and not best practice.

And I have no interest in python as language, only in fast dynamic VMs. e.g. potion could just add a python parser and get a 200x function call speedup, just as done for ruby or perl. The problem is the library.

[kbenson]

I think the last thing rurban needs is to jump into another VM. He's already got a few forks of Perl on GitHub[1].

1: https://github.com/rurban