Pending:
[ 1479611 ] speed up function calls TP
[ 876193 ] reorganize, extend function call optimizations TP
[ 1107887 ] Speed up function calls/can add more introspection info TP
[ 1087418 ] long int bitwise ops speedup TP
[ 1366311 ] SRE engine do not release the GIL
[ 1346214 ] Better dead code elimination for the AST compiler GB
- New patch posted, ready for review
[ 1243654 ] Faster output if message already has a boundary SH
Addressed:
[ 1145039 ] Remove some invariant conditions and assert in ceval TP
[ 1442927 ] PyLong_FromString optimization
[ 1335972 ] combo speedup and bugfix for string->int conversion
[ 876206 ] scary frame speed hacks [SUCCESS]
[ 1281707 ] Speed up gzip.readline (~40%) [SUCCESS]
[ 1337051 ] remove 4 ints from PyFrameObject
[ 1243730 ] Big speedup in email message parsing SH no speedup observed
[ 921466 ] Reduce number of open calls on startup GB
[ 1359618 ] Speed charmap encoder [patch approved, MAL to apply]
Deferred:
[ 813436 ] Scalable zipfile extension
- Addressed by SoC student
[ 738094 ] for i in range(N) optimization
Related to [ 1472639 ] make range be xrange
- Topic was discussed endlessly for some times.
[ 936813 ] fast modular exponentiation
- Going to need a lot of attention, not suitable for the sprint.
[ 923643 ] long <-> byte-string conversion
- This is a feature request.
[ 1353872 ] a faster Modulefinder (Related to py2exe runtime performance)
- No one interested in it.
[ 1492828 ] Improvements to ceval.c --> RaymondHettinger
[ 1346238 ] A constant folding optimization pass for the AST