performance_tuning

Prompt file imported from EvilBit-Labs/DaemonEye (.github/prompts/performance_tuning.prompt.md). Copyright stays with the author.

Analyze ONLY changed files (diff scope) for runtime performance characteristics while preserving correctness, public APIs, and security constraints. Apply only clearly safe micro-optimizations.

FOCUS CATEGORIES

1. Algorithmic Complexity (unnecessary O(n^2), repeated scans, avoidable clones)
2. Allocation Behavior (temporary allocations, Vec growth patterns, reserve vs push, string churn)
3. Async & Concurrency (blocking calls in async, unnecessary awaits, join patterns, parallelism opportunities)
4. I/O Efficiency (sync I/O in async context, redundant reads/writes, batching opportunities)
5. Data Structures (better fit: map vs vec scan, small vec, newtype for clarity/perf)
6. Caching & Reuse (recomputing constants, repeated serialization, repeated formatting)
7. Hot Path Error Handling (avoidable string formatting, cheap early exits)
8. Logging Cost (expensive formatting on hot path without level guards)
9. Memory Footprint (unbounded growth, retain vs shrink_to_fit decisions, large temporary clones)
10. Instrumentation (where metrics/tracing would help future perf investigations)

ACTION WORKFLOW (MANDATORY)

1 Diff list → 2 Perf analysis per category → 3 Classify (safe-edit / deferred / requires-approval) → 4 Apply only mechanical, behavior-preserving micro-optimizations (e.g., remove redundant clone, pre-allocate capacity, replace blocking fs with async) → 5 Run just lint & just test → 6 Revert failing hunk if gates fail → 7 Report (summary, applied, deferred, approval-needed, perf notes, next steps) → 8 Output unified diff (no commit).

If zero safe edits: state "No safe performance edits applied" and still produce full report.

SAFE PERFORMANCE EDIT EXAMPLES

• Replace clone() with reference when ownership not required
• Preallocate Vec with with_capacity when length is known
• Convert repeated format! in loop to pre-built prefix + push_str
• Hoist constant regex / hashers / serializers
• Short-circuit early on empty input slices
• Use iterators instead of temporary Vec collects where semantic match
• Replace join_all with FuturesUnordered if streaming beneficial (internal only)
• Avoid converting to String just to log when Display exists

AUTO-EDIT CONSTRAINTS (STRICT)

Scope: diff-only | Gates: just lint + tests must pass | No commits | No public signature/visibility changes | Validate after edits | No semantic changes

CRITICAL REQUIREMENTS

• Do not trade readability or security for micro perf
• Never introduce unsafe
• Provide benchmarks only as recommendations (do not add heavy harness automatically)
• Defer structural refactors (module splits) unless trivial & internal
• Avoid premature caching introducing invalidation complexity

REPO RULES (REINFORCED)

Zero warnings | No unsafe | Precise typing | Async I/O only | Trait boundaries | thiserror+anyhow | SQL AST validation | CLI-first | Memory efficiency | redb-only | Path canonicalization | No binary blobs | rustdoc for public APIs

EXECUTION CHECKLIST

1 Diff scan 2 Analyze perf 3 Classify 4 Apply safe micro-optimizations 5 Gates pass 6 Report 7 Output diff | On blocker: report & remediate guidance.

QUICK PERFORMANCE MATRIX

Category → Sample Safe Edit:

• Complexity → Replace nested loop with HashSet membership check
• Allocation → Pre-size Vec for known iteration length
• Async → Replace blocking fs call with tokio::fs equivalent
• I/O → Batch multiple writes into single buffer append
• Data Structure → Use SmallVec for typical <=8 elements (internal)
• Caching → Hoist constant serialization of static JSON template
• Logging → Wrap expensive computation in level guard if tracing::enabled!(...) (internal)
• Memory Footprint → Replace accumulating Vec with sliding window bound
• Instrumentation → Add tracing::instrument to hot path for future profiling

Ambiguous? Defer and document.