Filter the benchmark job, drop netframework4.8, pin nuget sources - #377
Merged
Conversation
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #377 +/- ##
=======================================
Coverage 98.01% 98.01%
=======================================
Files 39 39
Lines 52818 52818
Branches 1105 1105
=======================================
Hits 51770 51770
Misses 795 795
Partials 253 253 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
📊 Benchmark Results
PR branch
PR base
|
twcclegg
marked this pull request as ready for review
August 5, 2026 21:50
This was referenced Aug 14, 2026
Closed
Closed
This was referenced Aug 23, 2026
Closed
Closed
This was referenced Aug 31, 2026
Open
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Four items. Three landed; the fourth was measured and rejected.
The benchmark job only runs when it could have moved
A
paths:filter onrun_performance_tests, which at ~12 minutes is the longest job in the repo andpreviously ran on every PR including documentation-only ones.
csharp/PhoneNumbers/**is in the listdeliberately:
PerformanceTestis not in the solution, so nothing else compiles it and a librarychange could otherwise break the benchmarks unnoticed.
Checked branch protection first — no required status checks,
strict: false— so a skipped runcannot block a merge and the usual always-pass-on-skip workaround is unnecessary. The list is
duplicated across the
pull_requestandpushtriggers because GitHub Actions does not support YAMLanchors; the anchored version parses locally and fails there.
netframework4.8 is gone from the test projects
It only ever built on Windows, and there are no Windows runners, so it was exercised only if a
contributor happened to run tests locally. Its
System.Text.RegularExpressionsreference andPackageVersionentry go with it.This unblocks something: the OS-conditional TFM list was exactly why those two projects were excluded
from lock files in #373 — a lock file generated on Linux omitted a framework a Windows contributor
would restore. With the list now unconditional, both have committed
packages.lock.jsonfiles andare covered by
--locked-mode.nuget.config
<clear />discards any source a machine- or user-level config would otherwise contribute, so arestore resolves from nuget.org and nowhere else. Package source mapping changes nothing with a
single source; it matters the moment a second one is added, since without it any source can serve any
package id.
Benchmark job speed
Two trees now build concurrently before benchmarking (~6%), and the redundant
PhoneNumberCount=10000case is dropped (~7%) — the 1000 and 10000 results have agreed in direction and magnitude in every
same-runner comparison, so it demonstrated linear scaling rather than testing anything.
The benchmarks themselves still run strictly one after another. Running them concurrently would have
base and branch competing for the same cores, which would undo the same-runner comparison that caught
a real 2% regression in #374.
Reducing iteration counts was tried and reverted. Measuring first showed where the time actually
goes: 12 cases at ~22–27s each regardless of how fast the operation is (BDN's fixed per-case cost),
plus ~37% in per-class project generation.
warmupCount: 3, iterationCount: 7cut ~20% but cost farmore precision than predicted — base and branch ran in the same job, so this is a controlled
comparison:
GetDisplayCountryInputDigitPerKeystrokeFindNumbers_StrictGroupingAt 2.34% relative error,
InputDigitPerKeystrokecould no longer resolve the 2% regression thisharness caught two PRs ago. Extra iterations were averaging out run-to-run instability, not just
sampling noise — which is why the degradation was 8–44× rather than the √(15/7)≈1.5× a
sampling-only model predicts. Reverted, and the error bars are confirmed back under 0.6%.
Recorded here so it isn't re-attempted without knowing the cost.