PyPy 1.4: Ouroboros in practice¶
We’re pleased to announce the 1.4 release of PyPy. This is a major breakthrough in our long journey, as PyPy 1.4 is the first PyPy release that can translate itself faster than CPython. Starting today, we are using PyPy more for our every-day development. So may you :) You can download it here:
What is PyPy¶
PyPy is a very compliant Python interpreter, almost a drop-in replacement for CPython. It’s fast (pypy 1.4 and cpython 2.6 comparison)
Among its new features, this release includes numerous performance improvements (which made fast self-hosting possible), a 64-bit JIT backend, as well as serious stabilization. As of now, we can consider the 32-bit and 64-bit linux versions of PyPy stable enough to run in production.
Numerous speed achievements are described on our blog. Normalized speed charts comparing pypy 1.4 and pypy 1.3 as well as pypy 1.4 and cpython 2.6 are available on benchmark website. For the impatient: yes, we got a lot faster!
- PyPy’s built-in Just-in-Time compiler is fully transparent and automatically generated; it now also has very reasonable memory requirements. The total memory used by a very complex and long-running process (translating PyPy itself) is within 1.5x to at most 2x the memory needed by CPython, for a speed-up of 2x.
- More compact instances. All instances are as compact as if they had __slots__. This can give programs a big gain in memory. (In the example of translation above, we already have carefully placed __slots__, so there is no extra win.)
- Virtualenv support: now PyPy is fully compatible with virtualenv: note that to use it, you need a recent version of virtualenv (>= 1.5).
- Faster (and JITted) regular expressions - huge boost in speeding up the re module.
- Other speed improvements, like JITted calls to functions like map().
Carl Friedrich Bolz, Antonio Cuni, Maciej Fijalkowski, Amaury Forgeot d’Arc, Armin Rigo and the PyPy team