aio-libs / multidict / 38270578205

86%

push

github

Measure clear() alone over a thousand prebuilt multidicts (#1784)

<!-- Thank you for your contribution! -->

## What do these changes do?

`test_multidict_clear_str` and `test_watched_clear_str` copied a
100-item multidict inside the timed loop and cleared it, 100 times. The
measurement was mostly `copy()`, and it depended on where malloc put one
small table. Over the last 50 master runs on CodSpeed, `clear_str[cs-c]`
moved between 494 and 587 µs and `clear_str[ci-c]` between 786 and 905
µs, with the same commit landing anywhere in that range (c728ce1c read
882, 848 and 824 µs for `ci-c`). The instruction cost stayed within
0.4%; the swing came from the simulated memory access cost (up to 137%).
`popitem_str[cs-c]`, which does the same copy over far more work, stayed
within 0.4%.

Both benchmarks now use `benchmark.pedantic(setup=...)`: the setup
builds 1000 copies of multidicts with 0 to 99 items (and watches them,
for the watched variant) outside the measured region, and only the
`clear()` calls are timed. Spreading the work over many tables of every
size up to 128 slots, including the empty fast path, averages out the
layout effect.

## Are there changes in behavior for the user?

No, benchmarks only. The two benchmarks keep their names, so their
CodSpeed history shows one step at this commit.

## Is it a substantial burden for the maintainers to support this?

No.

## Related issue number

None.

## Checklist

- [x] I think the code is well written
- [ ] Unit tests for the changes exist (N/A, benchmark change)
- [ ] Documentation reflects the changes (N/A)
- [ ] If you provide code modification, please add yourself to
`CONTRIBUTORS.txt` (N/A)
- [x] Add a new news fragment into the `CHANGES/` folder
- [x] `make doc-spelling` passes and any new technical words are added
to `docs/spelling_wordlist.txt`

<details>
<summary>Agent run details (optional, for reviewers)</summary>

Python 3.11.13, C extension: `pytest -m 'expensive or capi or not
hypothes... (continued)

842 of 1684 branches covered (50.0%)

Branch coverage included in aggregate %.

13 of 15 new or added lines in 2 files covered. (86.67%)

9799 of 10709 relevant lines covered (91.5%)

1.83 hits per line