[efficiency-improver] Monthly Activity 2026-09

[github-actions]

Activity for September 2026

Suggested Actions for Maintainer

• Review PR: "[efficiency-improver] perf: single-pass Traits conversion in TestCaseExtensions" (branch efficiency/testcase-traits-single-enumeration, created this run) - Pulls

Energy Efficiency Backlog

Priority	Focus Area	Opportunity	Estimated Impact
LOW	Code-Level	TestExecutionManager MethodLevel parallel array-per-test in setup	One-time setup cost
LOW	Code-Level	TestContextImplementation.SanitizeName: Array.IndexOf per char	Only on first TestTempDirectory access
LOW-MEDIUM	Code-Level	RuntimeRegistrationEmitter.EmitResolveMethodHelper/ResolveProperty (source generator): O(N×M) linear scan per test method at module-init	One-time per-assembly startup cost, not a hot loop
LOW	Code-Level	VideoProduction.cs ProducePerTestVideosAsync O(tests×segments) overlap scan	Runs once at end-of-run (cold path)
LOW	Test Infra	test/Utilities/Microsoft.Testing.TestInfrastructure: regex-on-error-retry, tiny-collection LINQ	Cold/low-frequency, negligible scale
LOW	Code-Level	MSTestTelemetryDataCollector.SerializeCollection/SerializeDictionary: OrderBy on telemetry flush	Once per session (cold path)
LOW	Code-Level	TestClassModelBuilder (source generator): Where/GroupBy/Select for duplicate-method detection	Compile-time only, not a runtime hot path
LOW	Code-Level	src/Adapter/MSTest.TestAdapter: reflection-based GetRuntimeMethod lookup in TestMethodFilter (cold, once per discovery context)	Negligible, cold path
LOW	Infrastructure	CI output-byte-count health metric	Needs maintainer discussion

TestCaseExtensions.ToUnitTestElementWithUpdatedSource's Traits double-enumeration (last item in the table above) fixed this run — see PR above. Assert.HasCount/IsEmpty/ContainsSingle/CollectionAssert.AreEquivalent/AreNotEquivalent LINQ-Count()-fast-path family fully merged (incl. maintainer dedup PR #11363). src/Analyzers/MSTest.Analyzers/.CodeFixes exhaustively reviewed (2026-09-14), no open items. AzureDevOpsResultIdStore min/max single-pass fix confirmed present in main this run (resolves the multi-run "was it actually created?" saga from 2026-09-18/19/20 — the fix is real and landed).

Repo continues to be actively self-optimized by maintainers/Copilot coding agent. A separate [perf-improver] agent (issue #10914) is also actively working this repo's performance backlog — worth cross-checking before implementing overlapping opportunities.

Discovered Commands

Command	Purpose
./build.sh	Full restore + build
./build.sh -test	Run unit tests
./build.sh -pack	Build + produce NuGet packages
./build.sh -pack -test -integrationTest	Full pipeline incl. acceptance tests
.dotnet/dotnet build <project>.csproj -c Debug	Faster single-project build once .dotnet/ SDK is bootstrapped
artifacts/bin/<Project>/Debug/<tfm>/<Assembly> (run directly, after dotnet build)	Reliable way to run a project's unit tests in this sandbox — dotnet test has been observed to fail with Win32Exception launching the test host here
<binary> --filter "FullyQualifiedName~<name>"	VSTest-style filter that works on this repo's MTP-based test binaries
.dotnet/dotnet format whitespace TestFx.slnx --verify-no-changes --include <file>	Format check — bare dotnet format fails with "multiple MSBuild solution files found"; must pass TestFx.slnx explicitly
⚠️ Do NOT dotnet restore <proj>.csproj -p:TargetFrameworks=net8.0 or /p:TargetFramework=net8.0 as a workaround for missing net462/net48 SDKs on shared multi-project builds — it corrupts sibling projects' project.assets.json (removes other TFM targets), breaking unrelated builds until a full ./build.sh re-restore. Confirmed 2026-09-21.
MSTestAdapter.UnitTests / MSTestAdapter.PlatformServices.UnitTests (both multi-target net462;net48;net8.0;net9.0) cannot be restored/built/run in this sandbox at all — no .NET Framework SDK, and both are excluded from NonWindowsTests.slnf. Verify changes to src/Adapter/* via the full ./build.sh compile only; note the gap in PR "Test Status".

Performance runner: test/Performance/MSTest.Performance.Runner/. Benchmark suite: test/Performance/MSTest.Performance.Benchmarks (BenchmarkDotNet). Nightly timing workflow: .github/workflows/perf-timing-nightly.yml.

Run History

2026-09-21 21:43 UTC - Run

• 🔍 Verified the AzureDevOpsResultIdStore min/max single-pass fix (previously reported as uncertain across 3 prior runs) IS present in main — confirmed via get_file_contents that both GetDateRange helpers already use manual single-pass loops, no Where().Min()/Max(). Closes out that multi-run "false log" saga.
• 🔍 Ran a background scan of src/Adapter/MSTestAdapter.PlatformServices/MSTest.TestAdapter — 0 new findings beyond the already-known LOW-priority TestCaseExtensions.cs Traits double-enumeration.
• 🔧 Fixed TestCaseExtensions.ToUnitTestElementWithUpdatedSource's Traits.Any()+Traits.Select() double-enumeration with a single-pass foreach + lazily-allocated list. First attempt (materialize array, then check Length > 0) actually regressed the empty-traits case (always allocates); iterated to a lazy-list version that matches old allocation for empty traits and is ~2x faster/~20% less alloc for populated traits.
• 📊 Measured: standalone console micro-benchmark (2,000,000 iterations, net8.0 Release): empty-traits 64ms/40B (old) vs 85ms/40B (new, comparable); 2-traits 275ms/560,000,040B (old) vs 140ms/448,000,040B (new) — ~2x faster, ~20% less allocation for the populated case.
• ✅ Verified: ./build.sh (full repo) 0 warn/err; dotnet format whitespace --verify-no-changes clean. Could not run MSTestAdapter.PlatformServices.UnitTests/MSTestAdapter.UnitTests (net462/net48 SDK unavailable, both excluded from NonWindowsTests.slnf) — documented as a Test Status gap in the PR.
• 🔧 Created draft PR "single-pass Traits conversion in TestCaseExtensions" (branch efficiency/testcase-traits-single-enumeration).
• 💬 Re-checked [RFC] Agent/LLM-efficient test output for Microsoft.Testing.Platform #8824/Show slowest tests #3495 — no new human activity, not re-engaged (anti-spam holds). No new efficiency/performance-labeled issues found.
• 🌱 GSF principle applied: Hardware Efficiency (removes a redundant enumeration pass per discovered test case); SCI (reduces Energy term for the "convert TestCase to UnitTestElement" functional unit).
• ⚠️ Process note added to Discovered Commands: partial-TFM dotnet restore -p:TargetFramework(s)=net8.0 workarounds corrupt sibling projects' restore state in this sandbox — always re-run full ./build.sh to recover, and avoid this workaround pattern for future net462/net48-only-project builds.

2026-09-20 21:56 UTC - Run

• 🔧 Re-implemented and re-submitted the AzureDevOps min/max single-pass fix a 3rd time, this time with an explicit patch/bundle success response from create_pull_request (prior 2 runs' claims were unverifiable/false).
• 📊 Measured: 2M call-pairs, net8.0 Release: 581ms/288,000,040B → 308ms/160,000,040B (~1.9x faster, ~44% less alloc).
• 💬 Re-checked [RFC] Agent/LLM-efficient test output for Microsoft.Testing.Platform #8824/Show slowest tests #3495 — no new activity.

(Full history of all runs prior to 2026-09-20 is retained in repo memory — condensed here to keep this issue readable.)

Known Process Issue (IMPORTANT)

• Duplicate monthly issue bodies: this issue's body has been accidentally duplicated by update_issue calls at least twice historically. Always do a full clean operation: replace rewrite rather than relying on partial updates, and sanity-check the fetched body for repeated ## Activity for headers before editing.
• ALWAYS search for is:issue is:open in:title "Monthly Activity" with label area/performance BEFORE creating a new one. When a search returns >1 result, compare created_at/updated_at across ALL of them, not just the first hit, and close duplicates.
• Sandbox restore workaround pitfall (NEW, 2026-09-21): using dotnet restore <proj> -p:TargetFramework(s)=net8.0 to work around missing net462/net48 SDKs corrupts other projects' project.assets.json (drops their other TFM targets from the assets file), breaking builds repo-wide until a full ./build.sh re-restore. Do not use this pattern; if a project truly cannot be restored (net462/net48-only test projects), just skip running its tests and note the gap in the PR's Test Status section instead.

Warning

Firewall blocked 2 domains

The following domains were blocked by the firewall during workflow execution:

• github.com
• southcentralus0.in.applicationinsights.azure.com

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "github.com"
    - "southcentralus0.in.applicationinsights.azure.com"

See Network Configuration for more information.

🤖 Automated content by GitHub Copilot. Generated by the Efficiency Improver workflow. · copilot · auto · 440.4 AIC · ⌖ 11.6 AIC · ⊞ 17.8K · [◷]( · ◷)

_{Add this agentic workflow to your repo}

To install this agentic workflow, run

gh aw add githubnext/agentics/workflows/efficiency-improver.md@main

[github-actions]

Efficiency Improver — 2026-09-17 run update (posted as a comment because this run's single update_issue call was used to close the duplicate tracker #10917; the full rewrite will be applied to this issue's body on the next run)

This run's activity

• Task 4: Confirmed both previously-open PRs have landed: efficiency/assert-count-fastpath → merged as maintainer PR Optimize generic assertion count paths #11326 ("Optimize generic assertion count paths"); efficiency/collectionassert-equivalence-fastpath → merged as maintainer PR Add ICollection fast path for CollectionAssert equivalence counts #11353 ("Add ICollection fast path for CollectionAssert equivalence counts"). 0 open efficiency-improver PRs remain.
• Task 2/3: Ran drift re-scans of src/Adapter/MSTestAdapter.PlatformServices+MSTest.TestAdapter and Microsoft.Testing.Extensions.TrxReport/CrashDump/HangDump — both confirmed clean, no new HIGH/MEDIUM findings. No new PR created this run.
• Task 5: Re-checked [RFC] Agent/LLM-efficient test output for Microsoft.Testing.Platform #8824 and Show slowest tests #3495 — no new human activity since last review; not re-engaged (anti-spam).
• Task 7 (duplicate cleanup): Discovered and closed a duplicate Monthly Activity issue — [efficiency-improver] Monthly Activity 2026-09 #10917 (created 2026-09-01, stale since 2026-09-02) was a duplicate of this issue ([efficiency-improver] Monthly Activity 2026-09 #11023, created 2026-09-03, actively maintained). Closed [efficiency-improver] Monthly Activity 2026-09 #10917 with a pointer to this issue.

Suggested Actions for Maintainer

No suggested actions at this time — both prior PRs have already merged.

Run History (new entry, prepend on next full rewrite)

2026-09-17 21:47 UTC - Run

• 🔍 Confirmed both efficiency/assert-count-fastpath (→ Optimize generic assertion count paths #11326) and efficiency/collectionassert-equivalence-fastpath (→ Add ICollection fast path for CollectionAssert equivalence counts #11353) merged.
• 🔍 Drift-rescanned src/Adapter/MSTestAdapter.PlatformServices/MSTest.TestAdapter and TrxReport/CrashDump/HangDump — no new findings.
• 🧹 Closed duplicate tracker [efficiency-improver] Monthly Activity 2026-09 #10917 (superseded by this issue).
• 💬 Re-checked [RFC] Agent/LLM-efficient test output for Microsoft.Testing.Platform #8824/Show slowest tests #3495 — no new activity, not re-engaged.

🤖 Automated content by GitHub Copilot. Generated by the Efficiency Improver workflow. · copilot · auto · 225.9 AIC · ⌖ 11.8 AIC · ⊞ 17.8K · [◷]( · ◷)

_{Add this agentic workflow to your repo}

To install this agentic workflow, run

gh aw add githubnext/agentics/workflows/efficiency-improver.md@main