[sergo] Sergo Report: Reverse-Phantom Refile + Presnapshot-Position-Blind Sweep - 2026-09-28

[github-actions[bot]]

Executive Summary

Run R80 §36375544768. Registry held steady at 73 analyzers (no new linter since R74). Reconcile caught one more reverse-phantom closure (httprespbodyclose, the 10th confirmed instance of a closed-not_planned issue whose bug was never actually fixed) and new-exploration confirmed the brand-new presnapshot_position_blind pattern class (discovered in lenstringzero at R79) is a recurring design flaw, not a one-off, by finding a second instance in tolowerequalfold. 2 issues filed, both code-verified with concrete repros.

Tool and Registry Status

• Serena: 24 tools, stable, activate_project/Read/Grep/gh api all worked without approval prompts.
• Registry: pkg/linters/registry.go grep -c Analyzer,$ = 73, unchanged since R74 (uncheckedsliceindex).
• No new linter to solo-audit this run; new-explore budget went to a targeted cached-vein continuation instead.

Strategy: 50/50 Split

Cached component (reconcile): Re-ran the standing gh api reconcile-before-strategy discipline against all 15 pre-run open sergo issues. One flipped closed/not_planned since R79: #62115 (httprespbodyclose branch_state_merge). Re-read the code and confirmed the bug is fully present, so it was refiled rather than accepted as fixed.

New-exploration component: R79 ended by suggesting a sweep of other whole-pass-precomputed lookup tables for the same presnapshot_position_blind shape as lenstringzero. Grepped pkg/linters for every collect-then-query helper (collectKnownPathConsts, collectLenStringAliases, collectPackageLevelSliceMapVars, collectSeenMapCandidates, collectSprintfBoolCandidates, collectCaseConvAliases) and did a full read of the closest structural match, tolowerequalfold.collectCaseConvAliases. Confirmed it shares the identical design flaw.

Findings

Finding 1: httprespbodyclose branch_state_merge (reverse-phantom, 10th instance overall)
Location: pkg/linters/httprespbodyclose/httprespbodyclose.go lines 61-73 and 111.
respVars is one flat map[types.Object]*respVarState per function, populated by a single ast.Inspect with no branch scoping. A manual Body.Close() in one if/else arm and a defer Body.Close() of the same variable in the sibling arm write into the same state entry, so the manual-close arms violation gets masked by the sibling defer. CI-enforced on both native and wasm. Same root-cause family as the still-open resourcetracker branch_state_merge issue (#62304), but this is httprespbodyclose own standalone copy of the flaw.

Finding 2: tolowerequalfold presnapshot_position_blind (2nd confirmed instance of a brand-new pattern class)
Location: pkg/linters/tolowerequalfold/tolowerequalfold.go, collectCaseConvAliases at line 147, queried by run() at line 41 before the comparison walk at line 47.
The alias map is built once for the whole file before any comparison is checked. A later ASSIGN, IncDecStmt, or range-with-assign deletes the alias entry during that same up-front pass, so a variable used validly as a ToLower/ToUpper alias in an EARLY comparison, then reassigned LATER in the same function, has its alias silently deleted before the early comparison is ever evaluated. Zero testdata coverage for this ordering. CI-enforced on both native and wasm.

Generated Tasks and Issues

1. Refile: branch-scope respVarState in httprespbodyclose (or share a fix with the resourcetracker framework).
2. New: give collectCaseConvAliases position-aware alias validity ranges instead of a single final snapshot, likely sharing a fix pattern with the still-open lenstringzero presnapshot_position_blind issue.
   Both filed with the sergo label, code citations, concrete repro snippets, and effort estimates.

Metrics

• Findings this run: 4 (2 code-level bugs turned into 2 filed issues, both backed by direct citations).
• Issues created: 2 of the max 3 allowed.
• Success score (self-assessed): 8 out of 10 - one confirmed reverse-phantom refile plus one genuinely new, well-evidenced instance of a still-fresh pattern class, but did not reach a 3rd finding within budget.
• Open sergo issues after this run: 16 (14 unchanged carry-overs plus the 2 new filings replacing the one closure).

Historical Context

This is run 80 of an ongoing series (started informally well before R58). Running totals: roughly 500 cumulative findings and 142 generated tasks across all runs, average self-assessed success score around 8.76 out of 10. Ten confirmed reverse-phantom closures have now been caught overall, confirming that issues in this repository are sometimes auto-expired without any corresponding code change, so the gh api reconcile-before-strategy step remains essential every run. presnapshot_position_blind is only two runs old and already has two confirmed instances, suggesting it is a systemic pattern worth a dedicated sweep rather than a single fix.

Recommendations and Next-Run Focus

• Continue the presnapshot_position_blind sweep on the remaining un-checked collect-then-query helpers: collectKnownPathConsts (hardcodedfilepath), collectPackageLevelSliceMapVars, collectSeenMapCandidates (seenmapbool), and collectSprintfBoolCandidates (sprintfbool).
• Re-verify sg80a1 and sg80a2 landed as filed at the start of R81.
• If the registry grows to 74 analyzers, prioritize a fresh first-time audit of the new linter over further cached-vein probing.
• Consider whether a single shared helper fix (an incremental, position-aware alias-tracking utility) could resolve the presnapshot_position_blind class across multiple linters at once, similar to how astutil.UnwrapParenExpr became the shared fix idiom for paren_unwrap_gap.

References:

• §36375544768

Generated by 🤖 Sergo - Serena Go Expert · claude · agent · 165.9 AIC · ⌖ 8.61 AIC · ⊞ 6.9K · ◷

• expires on Sep 28, 2026, 8:03 PM UTC-08:00

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-09-29T04:03:09.611Z.

Closed by Workflow