An interesting post on the PyPy (Python written in Python) blog explains that PyPy has finally, on carefully selected tests, benchmarked ever so slightly faster than CPython (1% faster on one test, 11% faster on the other).
It's also interesting to note that a 20% speedup is achieved by method caching, which avoids the need to create a new bound method every time a method is referenced relative to an instance. Apparently the same technique would be usable in CPython were it not for certain C modules directly modifying the __dict__ of new-style classes. I wonder if CPython could obtain an equivalent speed-up, and if so whether it would be worthwhile making the guilty C modules restrict themselves like PyPy does.
I have high hopes that within a year we will regularly be seeing benchmarks that show PyPy outperforming CPython.
If you can't wait for that, and can live with heavy restrictions on your use of the language, try RPython (another product of the fertile PyPy project, and one in which Stackless Python's Christian Tismer had a hand) which on occasion benchmarks faster than C!
1 comment:
An anonymous comment which I accidentally rejected said "I had the impression that 2.6 does method caching, no? There is the clear_method_cache() in 2.6 NEWS."
However the string "method_cache" does not appear in Misc/NEWS from the trunk or from 2.6a1, so I am unsure where this information comes from and unable to query the comment's author for more information.
Post a Comment