gh-125196: Use PyUnicodeWriter for JSON encoder #125249
Conversation
Replace the private _PyUnicodeWriter with the public PyUnicodeWriter.
| return NULL; | ||
| } | ||
| PyObject *result = PyTuple_Pack(1, str); |
There was a problem hiding this comment.
I think this new code is cleaner than the old code. But PyTuple_Pack is a bit slow (see #118222), so we should measure whether this change makes an impact. Since this is not in one of the inner loops, I suspect the impact is really small.
There was a problem hiding this comment.
See my comment below. It's not faster, it's not slower. The change has no significant effect on performance.
There was a problem hiding this comment.
I don't think that PyTuple_Pack() is the performance bottleneck of this function.
I chose to use PyTuple_Pack() to avoid creating an incomplete tuple and then fill it. I prefer an atomic API which creates directly valid tuple.
|
Benchmark: import pyperf
import json
runner = pyperf.Runner()
runner.bench_func('dumps [1,2,3]', json.dumps, [1, 2, 3])
runner.bench_func("dumps ('baz', None, 1.0, 2)", json.dumps, ('baz', None, 1.0, 2))
runner.bench_func('dumps {"c": 0, ...}', json.dumps, {"c": 0, "b": 0, "a": 0})Result, Python built with |
Replace the private _PyUnicodeWriter with the public PyUnicodeWriter.