Skip to content

Skip clearing reference-free ReadOnlyCollectionBuilder slots - #4

Draft
tedd wants to merge 6 commits into
mainfrom
codex/readonlycollectionbuilder-clear-reference-free
Draft

tedd wants to merge 6 commits into
mainfrom
codex/readonlycollectionbuilder-clear-reference-free

Conversation

@tedd

@tedd tedd commented Nov 15, 2026 •

Copy link
Copy Markdown
Owner

For batches of 16 collections with 65,536 items each, mean Clear() time fell by 99.43-99.45% for int and 99.98% for a 64-byte value type in two BenchmarkDotNet comparisons.

This follows dotnet#136605.

Clear() now sets the count to zero without clearing the array when T contains no object references. RemoveAt() also skips clearing the unused last slot. This matches List<T>. New items overwrite these slots before use. Types that contain references, including structs, are still cleared.

Performance

Selected ClearBench<T>.ClearOnly results for 16 collections per batch. Setup and validation are outside timing. Times are microseconds (us) per batch, shown as mean +/- sample standard deviation for each process.

Type Items Launch Baseline us Patched us Mean time reduction
int 65,536 1 74.981 +/- 11.536 0.414 +/- 0.086 99.45%
int 65,536 2 80.034 +/- 21.972 0.459 +/- 0.129 99.43%
Wide 65,536 1 2722.796 +/- 112.946 0.430 +/- 0.145 99.98%
Wide 65,536 2 2730.597 +/- 118.649 0.458 +/- 0.084 99.98%

BenchmarkDotNet 0.16.0-preview.2, InProcessEmitToolchain, verified High process priority. Each version ran in two fresh processes. Clear-only used 10 warmups and 100 measurements. Clear-and-refill and removal used 8 warmups and 30 measurements targeting 500 ms each.

Windows 11 Pro x64 (build 26200), Ryzen 9 5950X, Release .NET 12.0.0-dev. ReadyToRun and tiered compilation were disabled; concurrent workstation GC was enabled.

Clear-only iterations were shorter than BenchmarkDotNet recommends, and some results had multiple timing clusters. All samples are included. These comparisons do not establish statistical significance. The int removal benchmark was slower in the first process pair and faster in the second. Some reference-containing Clear-only runs were also slower.

Validation

Full benchmark results and setup

Wide contains eight long fields (64 bytes), compares all fields, and hashes the first. RefValue contains a long and an object reference. Input values are created once and reused.

Baseline 8ac60f59b95; PR head 9827bf6eff4. Only System.Linq.Expressions.dll differs between the baseline and patched runtime images. Patched binaries were built from 7861f1220f2 and its test-only correction f60b088cf78. The changed implementation matches this PR. Native runtime files are identical.

Baseline/patched order was randomized with seed 136605. No outliers were removed. Clear-only uses one invocation per measurement and checks that all collections are empty afterward.

CycleBench<T>.ClearAndRefill clears and refills one collection per invocation. Refill is timed; existing storage and input values are reused. Each case recorded at least 10 seconds of measured work. Cleanup checks the final count.

RemoveBench<T>.RemoveLastAndRefill removes every item from the end, then refills the collection. Both steps are timed. Removal cases recorded 9.53 to 20.98 seconds of measured work per case in each process.

Removal results, in microseconds per complete remove-and-refill cycle:

Type Items Launch Baseline us Patched us Mean time reduction
int 1,024 1 3.528 +/- 0.071 4.594 +/- 1.049 -30.23%
int 1,024 2 3.488 +/- 0.015 3.185 +/- 0.014 8.69%
Wide 1,024 1 8.658 +/- 0.038 5.546 +/- 0.120 35.95%
Wide 1,024 2 9.223 +/- 0.788 5.747 +/- 0.038 37.69%
RefValue 1,024 1 5.876 +/- 0.073 5.598 +/- 0.102 4.73%
RefValue 1,024 2 8.123 +/- 0.886 5.546 +/- 0.032 31.72%
int 65,536 1 224.152 +/- 0.972 251.725 +/- 14.016 -12.30%
int 65,536 2 223.367 +/- 0.891 204.279 +/- 0.982 8.55%
Wide 65,536 1 856.835 +/- 3.586 809.213 +/- 14.471 5.56%
Wide 65,536 2 1608.254 +/- 66.416 809.928 +/- 3.775 49.64%
RefValue 65,536 1 376.766 +/- 2.088 371.036 +/- 21.252 1.52%
RefValue 65,536 2 405.098 +/- 50.477 356.371 +/- 1.813 12.03%

Clear-only batches of sixteen collections, in microseconds.

Type Items Launch Baseline us Patched us Mean time reduction
int 1,024 1 1.760 +/- 0.346 0.395 +/- 0.129 77.56%
int 1,024 2 1.857 +/- 0.372 0.465 +/- 0.124 74.96%
Wide 1,024 1 17.500 +/- 6.539 0.431 +/- 0.113 97.54%
Wide 1,024 2 17.986 +/- 5.176 0.507 +/- 0.093 97.18%
RefValue 1,024 1 17.704 +/- 53.644 7.519 +/- 3.389 57.53%
RefValue 1,024 2 7.437 +/- 1.177 13.785 +/- 14.869 -85.36%
int 65,536 1 74.981 +/- 11.536 0.414 +/- 0.086 99.45%
int 65,536 2 80.034 +/- 21.972 0.459 +/- 0.129 99.43%
Wide 65,536 1 2722.796 +/- 112.946 0.430 +/- 0.145 99.98%
Wide 65,536 2 2730.597 +/- 118.649 0.458 +/- 0.084 99.98%
RefValue 65,536 1 528.912 +/- 216.475 563.130 +/- 135.255 -6.47%
RefValue 65,536 2 531.588 +/- 114.246 692.421 +/- 262.761 -30.26%

Sustained clear-and-refill cycles, in microseconds.

Type Items Launch Baseline us Patched us Mean time reduction
int 1,024 1 1.319 +/- 0.033 1.347 +/- 0.005 -2.13%
int 1,024 2 1.311 +/- 0.007 1.272 +/- 0.006 3.00%
Wide 1,024 1 4.524 +/- 0.075 3.891 +/- 0.320 14.00%
Wide 1,024 2 4.430 +/- 0.019 4.367 +/- 0.113 1.42%
RefValue 1,024 1 3.427 +/- 0.026 3.415 +/- 0.020 0.37%
RefValue 1,024 2 3.420 +/- 0.016 3.463 +/- 0.135 -1.27%
int 65,536 1 92.447 +/- 0.433 86.184 +/- 0.288 6.77%
int 65,536 2 92.009 +/- 5.219 88.019 +/- 4.586 4.34%
Wide 65,536 1 925.085 +/- 41.936 694.393 +/- 2.844 24.94%
Wide 65,536 2 880.976 +/- 5.070 677.599 +/- 3.007 23.09%
RefValue 65,536 1 220.976 +/- 2.020 232.767 +/- 26.485 -5.34%
RefValue 65,536 2 221.192 +/- 1.632 222.354 +/- 3.323 -0.53%

Higher patched means in at least one process: RefValue at 1,024 items (Clear-only, up to 85.4%), RefValue at 65,536 items (Clear-only, up to 30.3%), int at 1,024 items (clear-and-refill, up to 2.1%), RefValue at 1,024 items (clear-and-refill, up to 1.3%), RefValue at 65,536 items (clear-and-refill, up to 5.3%), int at 1,024 items (remove-last-and-refill, up to 30.2%), int at 65,536 items (remove-last-and-refill, up to 12.3%).

Timed methods

[Benchmark] public void ClearOnly() => fixture.Clear();

[Benchmark] public void ClearAndRefill()
{
    fixture.Clear();
    fixture.Fill();
}
[Benchmark] public void RemoveLastAndRefill()
{
    fixture.RemoveAll();
    fixture.Fill();
}

Note

The implementation, tests, and pull request description were prepared with OpenAI Codex (Fable). I used Tedd.DotNetLab for the performance investigation. I have manually reviewed and checked everything.

@tedd
tedd marked this pull request as ready for review November 15, 2026 20:01
@tedd
tedd marked this pull request as draft November 16, 2026 12:10
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant