My grandma using an abacus can evaluate Python code faster than CPython

> On average, Python in GraalVM is 8.92x faster than CPython

That's great, but on which benchmarks?

I would prefer a presentation like https://speed.pypy.org/ (according to which PyPy seems to be only half as fast; but this conclusion can change significantly depending on which benchmarks are used).

I think you're misreading. PyPy isn't half as fast, it takes half the time.

They point out that the geometric mean is 0.22, so PyPy is 4.5x faster.

> so PyPy is 4.5x faster.

Yes, than CPython, but I was talking about Graal Python; I would like to know what benchmarks they used and how they calculated the speedup factor.

GraalVM is awesome.

In what way, besides speed? I'm genuinely curious. When I tried it the developer experience was mind-bogglingly terrible - 12GB and upwards of 20 minutes required to build "Hello, World" on an i7.

what is the catch?

"after warmup" and "The Python runtime is yet experimental in GraalVM".

https://www.graalvm.org/python/quickstart/

"There are not enough standard libraries to run the standard package installers for many packages with GraalVM Python runtime yet."

They are running on the commercial version of GraalVM. There is a giant technological gap between Python and Truffle (the framework behind Python on GraalVM CPython is a primitive interpreter while the other is a generating a hybrid JIT interpreter for both Python and CPython C extensions using Futamura projections.

It's made by Oracle

Stop the FUD please.

Why is it FUD?

Oracle has a long history of suing customers etc. It's a pretty litigious company. It also tends to include nuanced details in their license agreements e.g. https://www.linkedin.com/pulse/oracle-java-release-17-now-fr...

I'm not anti-Oracle. I worked there (worked at Sun which was acquired), so I understand the mentality. I use GraalVM and it's pretty nice. But whenever I pick an Oracle product I always make sure I can dump it in a second if they decide they want to monetize me aggressively.

OpenJDK is free.

so if you're successful with it, they'll eventually sue you.

My guess is that it is in a similar situation as Ruby on GraalVM, C-extensions.

C extension compatibility?

Supposedly there, this isn't Jython.

https://www.graalvm.org/python/quickstart/

"the possibility to create native platform binaries using the Native Image, a managed execution mode to run, for example, NumPy extensions in a safe manner, and many more."

Lots of couching language about "experimental" and "possibility" though. So it seems new.

I think there is sulong for that (https://github.com/oracle/graal/tree/master/sulong)

I'd imagine you have to recompile every C/C++ library used by your python code

Wonder if the techniques can be upstreamed to upcoming JIT in CPython