Python 3.11 Is Much Faster, But Pyston & PyPy Still Show Advantages

[blackshard]

Mmh, I'm the only one thinking that these microbenchmarks numbers are actually pretty useless?
I mean, there is not a single real use case benchmark that shows some real improvement, just sub-seconds things that are not tangible at all...

[kieffer]

Michael, the speed up your benchmark show is just amazing ! I cannot believe this is possible with so little modification of the "original" C-python. Of course PyPy is faster, but no more that much.

Well it does not impact me that much since most of my code is anyway written in Cython.

[sdack]

Mmh, I'm the only one thinking that these microbenchmarks numbers are actually pretty useless?
I mean, there is not a single real use case benchmark that shows some real improvement, just sub-seconds things that are not tangible at all...

When you have a very specific problem then, yes, these benchmarks will not tell you about how it will effect your problem. However, when you are looking for alternatives and want to get an idea of what is available and how these roughly compare then these benchmarks do tell you about it and show you what differences can be achieved, and chances are that even a specific problem not covered by these benchmarks may also benefit from using one of the alternatives.

[sinepgib]

Cython requires special annotation to the types so that it can compile python code to efficient C code.
Otherwise it would just compile the python bytecode to some equivalent C code, which enables almost no optimization and performs the same as CPython.

Compiling to equivalent C code is already worth the comparison IMO. So yeah, of course having the special annotations would be an unfair comparison (also because, to use the same code, you'd need to do it the decorator way, which introduces a slight extra overhead for the others), but since the other runtimes do more or less the same but at runtime it's as interesting as those.

I would also have liked to see Cython in this comparison.

Would be good, yeah.

Oh, I am so sorry, did I offend the JIT?

Lol.

[bachchain]

I remember how Dropbox created Pyston, but then gave up after realizing RIIRing their core applications was easier than trying to make Python fast.

[hax0r]

I love Python even more now, as I do PHP, great to see both of these getting speed bumps, underrated old dogs that get work done.

[Vistaus]

/me *waiting for uid313's usual list of things he doesn't like about Python and wishlist that he posts in every single Python thread*

[sinepgib]

I love Python even more now, as I do PHP, great to see both of these getting speed bumps, underrated old dogs that get work done.

Python doesn't seem underrated to me. While I've seen tons of hate directed to PHP, most I see for Python is love. Doesn't it also tend to rank high on surveys?

[Terr-E]

Python doesn't seem underrated to me. While I've seen tons of hate directed to PHP, most I see for Python is love. Doesn't it also tend to rank high on surveys?

Top-6 in both "Most Popular" and "Most Loved" categories on StackOverflow.
I'd say that's a pretty good score.

[undingen]

phoronix I'm one of the Pyston developers - huge thanks for the independently generated comparison .

We released yesterday 2.3.4 which contains further speedups over 2.3.3 but is ABI compatible with it.
In addition we announced 'pyston_lite' which bundles the JIT compiler of Pyston as extension module for CPython 3.8. Which makes it as easy as 'pip install pyston_lite_autoload' to use a JIT compiler inside your python environment (supports linux and macos on amd64 and arm64) but the speedup is only about half as much as "full" Pyston.
More infos can be found here: Announcing Pyston-lite: our Python JIT as an extension module