• NeedForSpeed
• Goals

Sprint topics following the NeedForSpeed theme

You can add additional topics below. Please discuss the specific goals and possible approaches to these tasks!

RichardJones: In revision 11, editor "66" removed the Python Package Index item from this page. Does that indicate that PyPI is not appropriate work for this sprint?

JohnBenediktsson: The focus of this sprint is on performance and the use of python in areas that might not previously have been considered. This can mean many things, but please keep that in mind when suggesting topics.

RichardJones: PyPI work was specifically suggested to me. Even though I do have a lot I'd like to do to PyPI, I'm happy to work on other things. I just need a clear call so I can focus my pre-sprint attention on appropriate things.

CPython

• Evaluate the PEPs for optimizing global and attribute lookups

  • [http://www.python.org/dev/peps/pep-0266/ PEP 266], [http://www.python.org/dev/peps/pep-0267/ PEP 267], and [http://www.python.org/dev/peps/pep-0280/ PEP 280]

• Can floating point ops be sped-up by avoiding flag/exception checks at every step? Can some floating point ops be in-lined in ceval.c?
• Implement portions of the decimal module in C
• Build-out struct module to support fast, high-volume binary conversions -- perhaps with regexp analogs such struct.compile() and struct.finditer()
• Create a string subclass that provides lazy slicing without copying
• Allow selective removal of unused features such as profiling support
• Faster parsing of strings and bytes into int, long, etc.
• Buffer for use with network I/O
• Build-out the collections module for optimized data structures:
  • ordered dictionary
  • alternate list implementation optimized for fast insertion and deletion
  • red/black tree
  • pivot tables
  • skip list

• Create a 64 bit PyInt type (for 32 bit machines)

• Optimize methods in stringobject.c
• Add itertools.imerge() and itertools.izip_longest()
• Guido has a standing request to have threading.py written in C
• Revisit Armin's zombie frame idea for reducing function call overhead. IIRC, the unsolved problem was how to save partially constructed frames without impacting the performance of recursive functions.

Pure Python Projects

• Improve [http://dada.perl.it/shootout/python.html language shootout] submissions

• Improve the API for timeit.py. Guido thinks that much of the intelligence in the command-line interface should be shifted to importable functions.

• Update PythonSpeed/PerformanceTips

Twisted

• Speed improvements to select and poll reactors
• Reactor based on /dev/epoll
• Better integration with psyco
• Improvements against twisted benchmark

Psyco

• Support for generator expressions
• Support for nested scopes
• Support for more dictionary operations
• Speedup float arithmetic
• Support for more built-ins (e.g. int(), long(), float(), etc.)
• Upgrade for python 2.5
• Better tools for profiling psyco-ness of application
• Investigate usefulness of IVM (with aim to producing a more streamlined dispatch loop)
• LLVM backend
• Virtualized longs (for long longs)

• Virtualized slots (Ability to cache getattribute() values)

PyPy

• Compiling RPython modules into CPython extension modules
• Port modules to use ctypes instead of having them written in C

• Stackless PyPy (pickling running tasklets)

• Documentation
• LLVM backend
• JIT compiler (possibly)

Py3000

• Make a wishlist for possible performance gains in Py3.0