fix(skills): misleading DCE advice in setup-harness

[SuperMuel]

The skill told users to use benchmark.pedantic() in Python to prevent dead code elimination, like black_box() in Rust. Two problems with that.

benchmark.pedantic does not prevent dead code elimination. It just lets you set rounds, iterations, and a setup function for the benchmark.

CPython does not delete unused function calls. It can't know if a function has side effects, so it always runs it. Pure-Python benchmarks do not need a black_box.

The new wording only mentions compiled languages and uses the right name for each one.

[codspeed-hq]

Merging this PR will not alter performance

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

✅ 7 untouched benchmarks

---

_{Comparing fix/skill-pedantic-not-for-dce (8bf3d7d) with main (0496dcd)}

[greptile-apps]

Greptile Summary

Corrects a single misleading line in the codspeed-setup-harness skill that wrongly suggested benchmark.pedantic() prevents dead code elimination in Python. The replacement swaps it for accurate compiled-language examples — benchmark::DoNotOptimize (Google Benchmark / C++) and Blackhole.consume (JMH / Java).

• Removes the Python benchmark.pedantic() DCE claim, which was doubly wrong: CPython never elides live function calls, and pedantic controls rounds/iterations, not optimization barriers.
• Adds benchmark::DoNotOptimize (C++) and Blackhole.consume (JMH) as replacement examples; both are correct but JMH isn't covered elsewhere in the skill document.

Confidence Score: 5/5

This is a one-line documentation fix removing factually incorrect advice; it introduces no code or behavioural changes and the new text is accurate.

The change only touches documentation, removes genuinely wrong guidance, and replaces it with correct compiler-barrier references. The only minor note is that JMH appears in a skill that has no Java harness section, which could mildly confuse readers.

No files require special attention; the single changed line in skills/codspeed-setup-harness/SKILL.md is straightforward documentation.

Important Files Changed

Filename	Overview
skills/codspeed-setup-harness/SKILL.md	Removes incorrect benchmark.pedantic() Python DCE advice; replaces with accurate C++ (benchmark::DoNotOptimize) and JMH (Blackhole.consume) references. Single-line documentation fix.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Skill: Prevent DCE advice] --> B{Language?}
    B --> C[Rust]
    B --> D[C++]
    B --> E[Java / JMH]
    B --> F[Python]
    B --> G[Go]
    C --> C1["`black_box()`"]
    D --> D1["`benchmark::DoNotOptimize`"]
    E --> E1["`Blackhole.consume`"]
    F --> F2["No DCE — CPython always executes calls (no special wrapper needed)"]
    G --> G2["Assign to package-level sink (idiomatic Go workaround)"]
    C1 --> OK["✅ Mentioned in updated line"]
    D1 --> OK
    E1 --> OK2["⚠️ Mentioned but Java/JMH harness not covered in this skill"]
    F2 --> FIX["✅ Correctly removed from advice"]
    G2 --> MISS["❓ Go is covered in skill but not mentioned in DCE guidance"]

Loading

_{Reviews (1): Last reviewed commit: "fix(skills): misleading DCE advice in se..." | Re-trigger Greptile}