[sergo] Sergo Report: Stale-Memory Resync + Detector Repair + Newest-Linter Audit (stringbytesroundtrip) - 2026-07-24

[github-actions[bot]]

Executive summary

R61. This run recovered from stale local memory and repaired a broken registry detector, then audited the newest un-analyzed custom linter. Net: 1 high-quality issue filed (#47704) against stringbytesroundtrip, plus two durable infrastructure corrections for future runs.

• Memory was frozen at R60 (2026-07-08) while daily runs actually continued through 2026-07-23 (latest issue stringsjoinone autofix drops precedence: bare unparenthesized element replacement emits non-compiling / behavior-changed code in [Content truncated due to length] #47467). gh api ...issues?labels=sergo is the ground truth; the local sergo-strategies.jsonl was not.
• Registry grew 43 → 57 analyzers over that window (14 new linters).
• Detector broke: grep -c Analyzer cmd/linters/main.go now returns 0 — main.go was refactored to multichecker.Main(linters.All()...). New detector installed: grep -c '.Analyzer,' pkg/linters/registry.go.

Tool updates

• Serena LSP surface unchanged: 23 tools, stable.
• Registry source-of-truth moved from cmd/linters/main.go to pkg/linters/registry.go (linters.All()), which is now the canonical importable list driving multichecker, tests, and doc-completeness.
• pkg/linters/doc.go header reads "57 active analyzers" and matches the registry — the recurring doc-sync drift finding (e.g. doc-sync guard is incomplete: {registry, spec_test} and {doc.go, README} are two disconnected clusters — a linter present in cod [Content truncated due to length] #47190, doc-sync: pkg/linters/doc.go says "53 active analyzers" but 54 are registered — timenowsub is omitted #46707, doc-sync: pkg/linters/doc.go says "52 active analyzers" but 53 are registered — trimleftright is omitted #46527) is currently closed; not re-filed.

Strategy (50/50 split)

Cached component — reconcile + doc-sync/enforce audit (proven veins).

• Ran the standard gh api reconcile first: zero open sergo issues (team closes fast). Confirmed doc.go/registry count parity (57 = 57) → skipped the doc-sync issue as non-actionable this run.
• Checked CI enforcement (cgo.yml:1217): 26 analyzers enforced; the newest linters (including the audit target) are not.

New exploration — audit the newest, never-analyzed linter.

• Identified stringbytesroundtrip (line 113 of registry.go) as the only recent linter with no corresponding sergo issue in the full state=all history.
• Applied the new-linter probe kit: type-resolution correctness, autofix compile/import safety, RunWithSuggestedFixes parity, and behavior-preservation.

Run targets: ≥1 concrete, evidence-backed finding on the newest linter; 0 duplicates; repair the detector so R62 doesn't mis-read the registry. All met.

Findings

stringbytesroundtrip — asymmetric round-trip semantics (filed #47704)

The linter flags both directions of a string/[]byte round-trip with identical "redundant round-trip / wasteful intermediate copy" wording, but they are not symmetric:

Expression	Result vs input	Droppable?
string([]byte(s)), s string	value-identical to s (strings immutable)	✅ genuinely redundant
[]byte(string(b)), b []byte	fresh, non-aliasing copy of b	❌ defensive-copy idiom

• []byte(string(b)) is the canonical pre-slices.Clone deep-copy idiom; the result deliberately does not alias b. Labeling it "redundant" (doc: "leave the caller with the same underlying type as the input") is wrong — acting on it by rewriting to b introduces an aliasing bug.
• testdata/.../stringbytesroundtrip.go:27,36 encode this direction as intended (// want ...redundant round-trip), so the golden test locks in the false-positive framing.
• Report-only (no SuggestedFix), so no compile/import bug; type analysis (Underlying() + IsType() gating) is otherwise alias/named-type safe.
• 0 real production sites today (all grep hits are the linter's own doc), so this is latent — but also an enforce-blocker if added to CI as-is. Same class as tolowerequalfold: autofix is not behavior-preserving — ToLower(x) == "UPPER" and mixed ToLower/ToUpper are rewritten to EqualFol [Content truncated due to length] #44865 (tolowerequalfold) and mapdeletecheck: autofix silently deletes comments inside the if body — no comment guard, unlike its sibling mapclearloop #46130 (mapdeletecheck).
• Fix: reword the []byte(string(b)) arm to recommend slices.Clone(b)/bytes.Clone(b) (single copy) instead of implying removal, or drop that arm and keep the provably-redundant string([]byte(s)) half.

Generated task

T1 — Correct the []byte(string(b)) diagnostic in stringbytesroundtrip.

• Problem: the byte-slice arm mischaracterizes a non-aliasing defensive copy as removable redundancy.
• Locations: stringbytesroundtrip.go:118-125 (message), :1-5/:23 (doc), testdata/.../stringbytesroundtrip.go:27,36 (golden want).
• Recommendation: reword to a slices.Clone/bytes.Clone suggestion, or drop the arm.
• Validation: string([]byte(s)) arm unchanged; byte-slice message no longer implies value-equality with b; doc/golden updated; re-confirm enforce-safety before any cgo.yml addition.
• Effort: Small (single package, message + doc + golden only; no traversal/type-logic change).

Metrics

• Findings: 6 (1 filed, 5 cleared: doc-sync-in-sync, enforce-status noted, both grep patterns 0 prod sites, string([]byte(s)) arm verified correct, type analysis verified alias-safe).
• Issues created: 1 / 3. Duplicates skipped: doc-sync (count in sync).
• Success score: 9 / 10.

Historical context

Continues the new-linter-audit vein that has landed repeatedly (writebytestring #44187, bytescomparestring #44484, sprintfbool #46978, stringsjoinone #47467). The recurring doc-sync vein was skipped this run because the registry and doc.go are already in sync at 57.

Recommendations & next-run focus (R62)

1. Use the repaired detector: grep -c '.Analyzer,' pkg/linters/registry.go (watch for the 58th analyzer).
2. Always gh api reconcile first — local jsonl memory can silently freeze while runs continue.
3. Probe other conversion / simplification linters (trimleftright, stringsjoinone, mapclearloop) for the same behavior-change class: non-aliasing / mutation-visibility differences that a "redundant/simplify" message glosses over.

References: §30067684687

Generated by 🤖 Sergo - Serena Go Expert · age00 · 227.5 AIC · ⌖ 13.7 AIC · ⊞ 5.8K · ◷

• expires on Jul 24, 2026, 8:58 PM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by Sergo - Serena Go Expert.

A newer discussion is available at Discussion #47917.