Repository navigation
Improve performance of exceptions #161242
Description
Feature or enhancement
Raising exceptions in python is rather expensive:
d = {'a': 'b', 'c': 1}
def f():
try: return d['x']
except: pass
def g():
if 'x' in d:
return d['x']
%timeit f()
%timeit g()
211 ns ± 1.08 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)
31.5 ns ± 0.265 ns per loop (mean ± std. dev. of 7 runs, 10,000,000 loops each)We can improve this by using freelists (for tracebacks and exceptions), using managed dicts for exceptions and improving the performance of attribute setters.
Has this already been discussed elsewhere?
This is a minor feature, which does not need previous discussion elsewhere
Links to previous discussion of this feature:
No response
Linked PRs
Reacted by Cody Maloney
Activity
cmaloney commentedon Nov 3, 2026
👍 to adding a freelist.
I'm planning to use AC Vectorcall on the core exception hierarchy including AttributeError and KeyError which should help a number of cases although have to be careful in application. Some exceptions store the args tuple and going tuple -> Vectorcall -> tuple is slower than not doing vectorcall unless take special care.
OSErrorcurrently spends most its time re-encoding the output ofstrerror_ras astrwhich I plan to fix (just keep all the ones we've encountered as a dict of immortal strings).Can I use this issue for that work or would you rather a separate one?
pythongh-161242: Cache strerror() messages per errno
pythongh-161242: Cache strerror() messages per errno