PyPy 5.0 Brings Faster Performance, Lower Memory Use

[phoronix]

Phoronix: PyPy 5.0 Brings Faster Performance, Lower Memory Use

PyPy 5.0 has been released today as the alternative Python interpreter and JIT compiler focused on performance and efficiency...

http://www.phoronix.com/scan.php?page=news_item&px=PyPy-5.0-Released

[tessio]

If PyPy is faster and leaner, why the python community don't drop CPython?

[beast]

Because CPython is THE reference implementation of Python. Clean but not fast, easy to understand and port but not fast and never deprecated.

[nanonyme]

If PyPy is faster and leaner, why the python community don't drop CPython?

Guido doesn't want to

[schmidtbag]

If PyPy is faster and leaner, why the python community don't drop CPython?

To my knowledge, pypy isn't compatible with all libraries or all hardware platforms. Also, it's not necessarily a better choice for all applications. If you have a pretty straight-forward script, cpython is better. If you have some number crunching to do or an infinite/indefinite loop, pypy is better.

[peppercats]

s/profoiling/profiling

If PyPy is faster and leaner, why the python community don't drop CPython?

pypy has poor python3 support for starters

[Michael_S]

s/profoiling/profiling


pypy has poor python3 support for starters

Their website claims compatibility with Python 3.2.5. This page goes over the differences: http://pypy.readthedocs.org/en/lates...fferences.html

I wasn't aware of any major incompatibility.

[tegs]

I hope that they bring the Python 3 support to at least 3.5 level considering that 3.2 has been dropped by pip: https://github.com/pypa/pip/commit/b...9d8ae4cf25282f

[grigi]

I have to add that PyPy is fast getting rid of the last few issues I had with it. Note that regular PyPy is 2.7.10 equivalent, and PyPy3 is 3.2.5 equivalent. I do wish they update PyPy3 to be atleast Python 3.3 compliant, as it was the first version of Py3 that was worth it to upgrade to.

Note that for Python 3.6 they are focusing a bit on performance as well, apparently you should expect a 5-20% average speedup there, which is always welcome :-)

There is 2 different performance projects that I saw, one that works on the VM, and another that implements an AST rewriting framework (finally, yay) which fixes common mistakes humans make that isn't incorrect just not optimal for performance reasons. Surprised that some of the performance improvements seem to come from MicroPython (which is a totally different implementation for microcontrollers)