Unladen swallow: accelerating Python [LWN.net]

Unladen swallow: accelerating Python

Posted May 7, 2009 21:35 UTC (Thu) by kune (guest, #172)
Parent article: Unladen swallow: accelerating Python

Unladen Swallow does support Python C modules. All the other implementations (IronPython, JPython and PyPy) don't and they are not decidingly faster than CPython. The developers from Google looked at JPython but the missing C module support was a major reason not to follow that road.

It's certainly super-smart to adress the compatibility issue by starting with the CPython source code. It has also the nice effect that patches are merged upstream. The next CPython releases will have improved performance based on work done in the project.

Whether the LLVM implementation will really lead to the performance targets set by the project is an open question. A concern is also that LLVM requires a C++ compiler making support on exotic platforms more difficult.

Regardless of those concerns Unladen Swallow is a project worth being started.

to post comments

Unladen swallow: accelerating Python

Posted May 8, 2009 2:40 UTC (Fri) by jamesh (guest, #1159) [Link]

It will be interesting to see how much compatibility they maintain as they move forward with trying to remove the global interpreter lock.

There are many C extensions that depend on the GIL for safe operation. Borrowing references, using it to synchronise access to their own state, etc. Much of the C API is written such that it requires the GIL to be used safely too.

I do think that removing the GIL is a worthy goal if they can achieve it without significant performance decrease, but it is one area where it will be difficult to keep compatibility.