Enhance benchmark results and optimize instrumentation performance

[Bertk]

This pull request updates the benchmark history documentation and improves the PowerShell script used to process benchmark results. The main focus is on making the benchmark table clearer by only showing user-defined parameter columns, handling different number formats, and skipping invalid or infrastructure-related data. Additionally, the script is enhanced for robustness and flexibility.

Key improvements include:

Benchmark Table and Documentation:

• The benchmark table in BenchmarkHistory.md now displays only user-configured [Params] values in the "Options" column, omitting infrastructure columns like Job, Toolchain, and others. Explanatory notes have been added to clarify this behavior.

Coverlet.core performance improvement proposals

#	Proposal	Effort	Expected Impact	Risk
2	Cache type-exclusion results	Low	High	Low
6	Pre-compile filter expressions	Low	High	Low
3	Replace LINQ hot paths with loops	Low	Medium	Low
4	Pool StringBuilder in reporters	Low	Medium	Low
1	Avoid double module read	Medium	High	Medium
5	Parallelise PrepareModules	High	Very High	High

[Bertk]

**Phase 1 results

Signal	Verdict
Mean timing	Flat — differences of 1–10 ms are within ShortRun noise (3 iterations). Not statistically significant.
Max / worst-case timing	Clear improvement in p1 — spikes up to 328 ms in p0 collapse to ≤185 ms in p1. The caches eliminate the cold-path JIT / first-iteration outliers.
Allocation	Consistent −0.24–0.25 MB per instrumentation run across all option combinations. The compiled regex cache and type-filter caches reduce per-run allocations reliably.
Report formats	No change — not on the optimised path.

The most important takeaway is that with ShortRun (3 iterations), a single slow outlier can dominate the Max and inflate the Mean. This is exactly why the switch to MediumRun (15 iterations) was applied — the p1 Max values already show the stabilising effect even within 3 iterations once the caches eliminate cold-start spikes.

[Bertk]

Signal	p0	p2	Verdict
Mean timing (instrumentation)	165–171 ms	165–177 ms	Flat — within measurement noise
Max timing (instrumentation)	176–328 ms	168–188 ms	✅ Clear improvement — worst-case reduced by up to 156 ms
Max spread across all option combos	152 ms	17 ms	✅ 9× more consistent worst-case
Instrumentation allocation	44.68 MB	44.41 MB	✅ −0.27 MB (−0.6%) per run
Report allocation (cobertura/lcov/opencover)	baseline	−3–25%	✅ Proposal 4 visible
Mean timing (workflow)	468 ms	484 ms	Flat / noise-dominated at ShortRun p0 baseline

The cumulative effect of all five proposals is primarily a stability gain (worst-case latency and iteration-to-iteration variance significantly reduced) plus a consistent allocation saving of ~0.28 MB per instrumentation run. Mean times are stable, confirming the optimisations have no regression risk.

[Copilot]

Pull request overview

This PR improves Coverlet’s performance benchmarking workflow and implements several runtime performance optimizations in coverlet.core, primarily targeting instrumentation hot paths and reporter allocations.

Changes:

• Optimize instrumentation performance by caching type include/exclude decisions, avoiding a second module read for reachability analysis, and caching compiled include/exclude filter regexes.
• Reduce reporter allocations by pooling buffers (MemoryStream / StringBuilder) for OpenCover/Cobertura/Lcov report generation.
• Update benchmark suite and documentation: refine which benchmark columns are captured, improve benchmark history update scripting, and refresh benchmark history content.

Reviewed changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 9 comments.

Show a summary per file

File	Description
test/coverlet.core.benchmark.tests/Simulator.cs	Adjust benchmark attributes (diagnoser configuration).
test/coverlet.core.benchmark.tests/Program.cs	Update BenchmarkDotNet job settings and benchmark set executed.
test/coverlet.core.benchmark.tests/PerformanceImprovementProposal.md	Add an implementation plan section for the proposals.
test/coverlet.core.benchmark.tests/InstrumenterBenchmarks.cs	Rework benchmark to instrument a clean per-iteration DLL/PDB copy and isolate setup overhead.
test/coverlet.core.benchmark.tests/InstrumentationOptionsBenchmarks.cs	Similar per-iteration clean binary setup; remove benchmarks deemed noise-only.
test/coverlet.core.benchmark.tests/HowTo.md	Add guidance to update benchmark history documentation via script.
src/coverlet.core/Reporters/OpenCoverReporter.cs	Pool MemoryStream and avoid per-call ToArray() allocation.
src/coverlet.core/Reporters/LcovReporter.cs	Switch to pooled StringBuilder to reduce allocations during report generation.
src/coverlet.core/Reporters/CoberturaReporter.cs	Pool MemoryStream and avoid per-call ToArray() allocation.
src/coverlet.core/Instrumentation/Instrumenter.cs	Add caching and hot-path loop optimizations; attempt to avoid double module reads.
src/coverlet.core/Helpers/InstrumentationHelper.cs	Cache compiled filter regex pairs for include/exclude evaluation.
scripts/Update-BenchmarkHistory.ps1	Improve parsing/normalization and add upsert/dedup + version guard logic.
Documentation/BenchmarkHistory.md	Update benchmark history format notes and add new benchmark rows.