[sergo] Sergo Report: SuggestedFix Import Self-Containment + Autofix Behavior-Preservation - 2026-07-11

[github-actions[bot]]

Executive Summary

Run R61 audited the custom go/analysis linter suite (pkg/linters/) with a 50/50 strategy: cached = the proven SuggestedFix compile-correctness vein that produced sg60a1 + #44653; new = deep audit of the 44th linter bytescomparestring and an autofix behavior-preservation probe across all 12 SuggestedFix linters. Registry moved 43 → 44. Two distinct, evidence-backed issues filed; zero duplicates.

Metric	Value
Registry (analyzers in cmd/linters/main.go)	44 (was 43 at R60)
Open sergo issues pre-run	0 (gh api verified)
SuggestedFix linters audited	12
Findings	7
Issues filed	2 (#44864, #44865)
Success score	9 / 10

Tool & Registry Updates

• LSP tools stable at 23 — no Serena tool delta.
• Registry delta 43 → 44: the 44th linter is bytescomparestring (string(a)==string(b) over []byte → bytes.Equal). It is well-built — adds its bytes import in-fix and uses RunWithSuggestedFixes. Its only flaw (an inaccurate "allocates" premise) was already filed as bytescomparestring (new 44th linter): the "allocates" premise is inaccurate under gc — string(a) == string(b) does NOT allocate [Content truncated due to length] #44484 by an off-log run.
• Reconciliation note: two sergo issues (writebytestring autofix references io.WriteString but never adds the "io" import — non-compiling fix when the file doesn't alrea [Content truncated due to length] #44653 writebytestring missing io import; bytescomparestring (new 44th linter): the "allocates" premise is inaccurate under gc — string(a) == string(b) does NOT allocate [Content truncated due to length] #44484 bytescomparestring premise) were filed on 07-09/07-10 by runs that did not persist strategy memory. Both closed. writebytestring is now fully fixed in code (isExactString wrap at writebytestring.go:117 resolves sg60a1; addIOImportEdit at :213 resolves writebytestring autofix references io.WriteString but never adds the "io" import — non-compiling fix when the file doesn't alrea [Content truncated due to length] #44653).

Strategy Split

Cached (50%) — SuggestedFix import self-containment

The vein that produced sg60a1 (#44187) and #44653. Enumerated the 12 SuggestedFix linters and classified which introduce a new package in their emitted fix:

Linter	Fix introduces	Self-adds import?
bytescomparestring	bytes	✅ yes
writebytestring	io	✅ yes (retrofit after #44653)
sprintfint	strconv	❌ no
appendbytestring / tolowerequalfold / stringsindexcontains / stringreplaceminusone / fprintlnsprintf / lenstring* / execcommandwithoutcontext / ctxbackground	reuse already-imported pkg	n/a

sprintfint is the last import-introducing fix that isn't self-contained → #44864.

New (50%) — 44th-linter audit + autofix behavior-preservation probe

Beyond bytescomparestring (clean, already covered by #44484), probed every SuggestedFix body for fixes that change runtime semantics rather than just refactor. Found tolowerequalfold rewrites case-incompatible comparisons to EqualFold, flipping always-false comparisons into live ones → #44865.

Findings & Generated Tasks

#44864 — sprintfint autofix omits strconv import + orphans fmt (compile-safety)

• Where: pkg/linters/sprintfint/sprintfint.go:94-109 (buildItoaFix).
• Problem: emits strconv.Itoa(n) with only a call-replacement TextEdit. For a file whose sole fmt/strconv use is the flagged call, applying the fix yields undefined: strconv AND "fmt" imported and not used — two compile errors.
• Test masks it: the RWSF golden pre-imports strconv and keeps other fmt.Sprintf uses alive, so the import transition is never exercised.
• Fix: reuse the in-fix import-edit helper (factor out of writebytestring/bytescomparestring); remove orphaned fmt; add a single-use fixture.
• Effort: Small–Medium. Latent (not CI-enforced).

#44865 — tolowerequalfold autofix is not behavior-preserving (semantic-safety)

• Where: pkg/linters/tolowerequalfold/tolowerequalfold.go:67-69 (trigger) + :90-129 (buildEqualFoldFix).
• Problem: flags when either side is ToLower/ToUpper with no case-compatibility check on the other operand. strings.ToLower(a) == "HELLO" (always false) → strings.EqualFold(a, "HELLO") (not false). Mixed ToLower(a) == ToUpper(b) → EqualFold(a, b) — completely different semantics (a="Foo",b="foo": false → true).
• Test masks it: fixture only tests case-matched literals (:10-12); line :11 shows the author knows ToUpper pairs with an uppercase literal, but the analyzer never enforces it.
• Fix: require literal to already match the conversion's case; require both conversions to be the same function; add three negative fixtures.
• Effort: Small.

Historical Context

• 61 total runs, 415 cumulative findings, avg success 8.75.
• The SuggestedFix compile-correctness vein (sg60a1, writebytestring autofix references io.WriteString but never adds the "io" import — non-compiling fix when the file doesn't alrea [Content truncated due to length] #44653, and now sg61a1) is now exhausted for import-adding — all three import-introducing fixes are covered.
• New vein opened this run: autofix behavior-preservation (sg61a2) — fixes that compile but change semantics.

Recommendations & Next-Run Focus

1. Refactor the in-fix import-edit logic (currently duplicated in writebytestring + bytescomparestring, soon a 3rd for sprintfint) into pkg/linters/internal/astutil.
2. R62: watch for the 45th linter; extend the behavior-preservation probe to the remaining SuggestedFix linters (fprintlnsprintf, stringsindexcontains, lenstringsplit) for fixes that compile but alter semantics; verify sg61a1/sg61a2 land.
3. Persist strategy memory every run — two off-log runs (07-09/07-10) filed real issues without updating the JSONL, complicating reconciliation.

References:

• §29140058955

Generated by 🤖 Sergo - Serena Go Expert · 333.5 AIC · ⌖ 11.7 AIC · ⊞ 5.8K · ◷

• expires on Jul 11, 2026, 8:59 PM UTC-08:00