Typing conformance results

No changes detected ✅

Memory usage report

Summary

Significant changes

ecosystem-analyzer results

No diagnostic changes detected ✅

Full report with detailed diff (timing results)

Merging this PR will improve performance by 16.33%

⚡ 1 improved benchmark
✅ 46 untouched benchmarks
⏩ 60 skipped benchmarks¹

Performance Changes

_{Comparing charlie/bind-i (91a87cd) with main (65d768e)}

Do you know why we're running binary operator inference multiple times? Are we performing multi-inference somewhere? It seems like it would be more effective to memoize at an outer layer rather than specifically binary operators, if this is a more general problem.

Hmm, I don't think multi-inference is involved. Given:

import pandas as pd

df = pd.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6], "c": [7, 8, 9]})
df["d"] = df["a"] + df["b"] + df["c"] + 1 + (df["a"] ** 2 + df["b"] ** 2 + df["c"] ** 2)

We run try_call_bin_op(Series, Add, Series) 5 times, try_call_bin_op(Series, Pow, int) 3 times, etc. And the expression-level cache doesn't help.

Ah I see, that seems reasonable.

I do wonder if there’s a more general idea here though. I’ll explore it separately.