[sergo] Sergo Report: DELTA-79to80-newlinter-audit+docsync-gap+reconcile-bulkfix - 2026-10-12

[github-actions[bot]]

Executive Summary

Run 94. The analyzer registry grew by one (79 -> 80, new linter: walkdirignoredeerror, PR #67757). Reconcile uncovered the first bulk-fix wave seen in this memory's history (previous waves were bulk auto-expires with the bug still present) - 6 of the 7 issues open at R93 were genuinely closed as completed, spot-verified against current code rather than trusted blindly. New-explore on the 80th linter traced an ExprStmt-branch ParenExpr-unwrap gap through a shared helper into 4 other files, and a routine doc/README cross-check turned up a straightforward sync gap on the same new linter. 2 issues filed, both high-confidence and code-verified.

Tool & Registry Changes

• Serena: 24 tools, stable, exact match via serena --help.
• Registry: pkg/linters/registry.go Analyzer count 79 -> 80. New: walkdirignoredeerror (pkg/linters/walk-dir-ignored-error/, PR [linter-miner] feat: add walk-dir-ignored-error linter #67757), flags filepath.Walk/filepath.WalkDir calls that discard their error return.

Strategy Split

Cached component (reconcile): Queried state=open+labels=sergo fresh rather than trusting last run's predicted list. Found only 1 of 7 expected opens remained (#65753). Rather than accept the other 6 closures at face value, spot-verified 3 against current source:

• coverage.findProfile now sorts profile keys and picks the longest/exact match deterministically (sg91a1 fixed).
• deferinloop.isInsideLoop now consults pass.TypesInfo.TypeOf(rangeStmt.X) to detect range-over-func and correctly stops flagging those defers (sg92a1 fixed).
• cgo.yml wasm job now has its own consolestderr step, and closeerrorunchecked's docstring/code now explicitly delegates deferred-Close checking to unchecked_deferredclose instead of contradicting it (sg93a1/sg93a2 fixed).
• contextcancelnotdeferred still enforced on native CI only - wasm LINTER_FLAGS omission unfixed (issue 55932 auto-expired, 2nd oc #65753 (contextcancelnotdeferred native-only wasm omission) re-confirmed still broken by direct read of cgo.yml lines 1528 vs 1555.

New-exploration component: Full first-time audit of the 80th linter plus a routine doc/README/CI-wiring cross-check, which is now a standing day-one practice for any newly merged linter.

Findings

1. ExprStmt branch skips ParenExpr unwrap in 5 places (#67866)
walkdirignoredeerror's own analyzeAssignStmt correctly calls astutil.UnwrapParenExpr before asserting a *ast.CallExpr, but its sibling analyzeExprStmt does the bare assertion with no unwrap - so a parenthesized bare statement like (filepath.Glob(pattern)) escapes detection while _ = (filepath.Glob(pattern)) does not. The identical asymmetry was traced to the shared astutil.MatchDiscardedErrorCall helper (which backs globwalkignorederror, strconvparseignorederror, and jsonmarshalignoredeerror) and, independently, to closeerrorunchecked.go. Five linters share one narrow but concrete gap. Go's own spec confirms parenthesized bare-call statements are valid and semantically identical to the unparenthesized form.

2. doc.go and README.md omit the new 80th linter (#67865)
walkdirignoredeerror does not appear in pkg/linters/doc.go (header still says 79) or pkg/linters/README.md, and is absent from the notYetEnforced map in doc_sync_test.go. This should fail TestDocGo_CountMatchesBullets, TestDocSurfacesMatchRegistryAndSpecList, and TestCIEnforcedLintersMatchRegistry the next time that package's tests run.

Tasks Generated

1. Add astutil.UnwrapParenExpr(stmt.X) to MatchDiscardedErrorCall's ExprStmt case, plus the matching one-line fix in walkdirignoredeerror.go and closeerrorunchecked.go.
2. Add a doc.go bullet + README row for walkdirignoredeerror, bump the header count to 80, and add it to notYetEnforced (or enforce it in cgo.yml).

Metrics (this run)

• Findings: 4 (2 filed as issues, 2 were reconcile-verification reads that confirmed existing fixes rather than new bugs)
• Issues created: 2/3
• Tasks: 2
• Success score (self-assessed): 9/10 - both findings are concrete, code-verified, and non-overlapping; reconcile correctly distinguished a genuine bulk-fix from the bulk-auto-expire pattern seen in most prior runs.

Historical Context

94 cumulative runs, 568 total findings, 162 tasks generated, running average success score 8.72. This is the first run where a majority of a prior run's open issues were confirmed genuinely fixed rather than auto-expired-but-still-broken - the phantom_reconcile pattern (closed != fixed) has held in the opposite direction before, so this run treated the closures with the same scrutiny as any other reconcile pass rather than assuming good faith.

Recommendations / Next-Run Focus

• Verify sg94a1 and sg94a2 land; re-tally opens (expect contextcancelnotdeferred still enforced on native CI only - wasm LINTER_FLAGS omission unfixed (issue 55932 auto-expired, 2nd oc #65753 plus the two new ones unless already bulk-closed again).
• If the registry grows to 81, audit the new linter fresh and run the same doc/README/CI-wiring day-one check established this run.
• Consider checking fileclosenotdeferred and manualmutexunlock for the same ExprStmt-branch paren gap - not yet probed, and five independent instances of the same narrow shape suggests it may be more widespread than these five.

Generated by 🤖 Sergo - Serena Go Expert · claude · agent · 336.8 AIC · ⊞ 5.2K · ◷

• expires on Oct 12, 2026, 8:13 PM UTC-08:00

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-10-13T04:13:44.267Z.

Closed by Workflow