Uh oh!
There was an error while loading.Please reload this page.
- Notifications
You must be signed in to change notification settings - Fork33.7k
Closed
Assignees
Description
There's a simple change that we can make to increase the preformance of these three methods. Right now they are defined as:
Lines 1491 to 1493 inf4b5588
| {"derive", (PyCFunction)BaseExceptionGroup_derive,METH_VARARGS}, | |
| {"split", (PyCFunction)BaseExceptionGroup_split,METH_VARARGS}, | |
| {"subgroup", (PyCFunction)BaseExceptionGroup_subgroup,METH_VARARGS}, |
However, they only ever use one argument:
Lines 883 to 885 inf4b5588
| if (!PyArg_ParseTuple(args,"O",&excs)) { | |
| returnNULL; | |
| } |
So, it would be much faster to useMETH_O instead. I did these measurements, before and after:
» pyperf timeit -s 'e = BaseExceptionGroup("Message", [ValueError(1)] * 10); i = [TypeError(2)] * 10' 'e.derive(i)'.....................Mean +- std dev: 353 ns +- 2 ns» pyperf timeit -s 'e = BaseExceptionGroup("Message", [ValueError(1)] * 10); i = [TypeError(2)] * 10' 'e.derive(i)'.....................Mean +- std dev: 319 ns +- 4 ns» pyperf timeit -s 'e = BaseExceptionGroup("Message", [ValueError(n) if n % 2 == 0 else TypeError(n) for n in range(100)]); f = lambda e: True' 'e.split(f)'.....................Mean +- std dev: 180 ns +- 1 ns» pyperf timeit -s 'e = BaseExceptionGroup("Message", [ValueError(n) if n % 2 == 0 else TypeError(n) for n in range(100)]); f = lambda e: True' 'e.split(f)'.....................Mean +- std dev: 151 ns +- 1 ns» pyperf timeit -s 'e = BaseExceptionGroup("Message", [ValueError(n) if n % 2 == 0 else TypeError(n) for n in range(100)]); f = lambda e: True' 'e.subgroup(f)'.....................Mean +- std dev: 153 ns +- 0 ns» pyperf timeit -s 'e = BaseExceptionGroup("Message", [ValueError(n) if n % 2 == 0 else TypeError(n) for n in range(100)]); f = lambda e: True' 'e.subgroup(f)'.....................Mean +- std dev: 121 ns +- 0 ns