[code-review] ignored_error + naked_return dead detectors: delete+tombstone (sa-167-verified FP classes); empty_error_check: register as zero-FP Pattern-1 per #507 tombstone reservation

[topcheer]

Summary

R67 batch from the sa-165 census: three more fc5c4aa-stripped dead detectors (ignored_error_check.go, naked_return_check.go, empty_error_check.go) — all three entry functions verified zero-external-reference. Independent verification (sa-167, 11 probe tests, all run in-package) split them into two delete candidates and one register candidate.

ignored_error_check.go — delete + tombstone (revival = High risk)

Four empirically confirmed defects:

• a1 — Factually-wrong assertion on 100% of common receiver names: the isKnownErrorMethod fallback (:363-374) fires on any standalone recv.Write(...) / recv.WriteString(...) when the receiver name is not one of six buffer names (buf/buffer/b/builder/sb/bld). sa-167 verified a zero-return-value method (func (fakeConn) Write(p []byte) {} — legal Go with no error return at all) called as conn.Write(data) still produces "These functions return an error that is not checked" — a factually wrong claim, one level deeper than the [business-logic] ignored_error_check false positives on bytes.Buffer and strings.Builder Write methods — always return nil error #111 always-nil case the file's own comment describes.
• a2 — Blank-assign conflict with ecosystem: _ = strconv.Atoi(s) (the standard Go syntax for deliberate discard) is flagged; errcheck by default does NOT flag blank identifiers (requires -blank). Same coaching-agents-into-breaking-correct-code class as [code-review] printf_format_check 双确定性误报：变参展开触发假 verb-count、转发包装函数触发假注入风险（go vet 对两者均豁免） #505.
• a3 — Dead three-segment keys + double-missed marquee scenario: in an untyped AST f.Close() resolves to Ident("f")+"Close" → "f.Close", never matching the "os.File.Close"-style keys. All os.File.*, http.ResponseWriter.*, http.Server.*, bufio.*, sql.*, os/exec.Command.* keys are unreachable. Worse, the file's own header example json.NewEncoder(w).Encode(data) resolves via constructorReturns to "json.Encoder.Encode" while the map key is "json.NewEncoder.Encode" (:78) — a double mismatch; detection rests entirely on the 30 generic method-name fallback.
• a4 — Structurally unfixable without type info: x.Write(...) return type is unknowable in an untyped AST. Fixing requires go/types (architecture change to the write-integrity pipeline) or dropping the method fallback (which a3 shows carries ALL the detection). This is exactly why errcheck is built on types.Check.

naked_return_check.go — delete + tombstone (advisory class)

• b1: fires on the official defer+recover named-result idiom (24-line function, sa-167 verified) — warning text is a readability opinion ("hurt readability"), no factual error, but the fc5c4aa refactor explicitly removed advisory checks "to eliminate context pollution" (:79-80); threshold 20 vs revive nakedret default 30.
• b2: does not recurse into FuncLit — verified correct/conservative (closures have their own named results), acceptable FN.
• b3: funcName-level delta — same function with 2 existing naked returns gains a 3rd, zero warnings (verified). Weaker than the [business-logic] empty_error_check.go: count-delta 家族修复批次全部遗漏 — 修 1 旧空 err 块 + 引 1 新块 count 相等即静默漏报 #186/[business-logic] checkAppendIgnored is a dead detector — never registered in write_integrity allChecks, the documented write-time discarded-append() detection never fires #499 instance-level multiset standard.

empty_error_check.go — REGISTER as "empty-error-body" (zero FP, positive value)

• c1: pure-empty / comment-only / semicolon-only if err != nil { } bodies all detected with correct lines; non-error vars (if x != nil {}, if ok {}) exempt — zero FP across all probes.
• c2: wider err-name matching (errCh/errno) fires but is benign in the empty-body scenario — any if x != nil {} with zero statements is suspect regardless of type.
• c3: fingerprint multiset delta ([business-logic] empty_error_check.go: count-delta 家族修复批次全部遗漏 — 修 1 旧空 err 块 + 引 1 新块 count 相等即静默漏报 #186 semantics) verified — "fix one old + introduce one new" reports exactly the new one.
• This is exactly the revival path the [code-review] error_swallow + error_nopropagate dead detectors (fc5c4aad batch, 6th instance): revival is empirically FP-confirmed on 4 idiomatic Go patterns — delete + migrate live helpers + tombstone #507 tombstone reserved: "Pattern 1 (empty body) is the only zero-FP revivable part". Semantically distinct from nil-deref-after-error (post-check misuse vs post-check inaction). Budget-compatible: maxIntegrityWarnings=1 truncates by severity; maxReport=3 is only a production cap; delta-gated so steady-state rewrites are zero-noise.

Fix

1. Delete ignored_error_check.go + _test.go, naked_return_check.go + _test.go (all 26 symbols verified zero-external-reference).
2. Register {Name: "empty-error-body", Langs: []Language{LangGo}, Run: sliceCheck(checkEmptyErrorBody)} in allChecks.
3. Extend the write_integrity.go tombstone block with [code-review] ignored_error + naked_return dead detectors: delete+tombstone (sa-167-verified FP classes); empty_error_check: register as zero-FP Pattern-1 per #507 tombstone reservation #508: ignored-error revival requires go/types; naked-return revival requires min-lines>=30 + instance-level delta + closure coverage.

[topcheer]

Fixed in a2d9a4c: deleted ignored_error + naked_return (+tests, -1287 lines) with tombstone entries recording revival preconditions (go/types for ignored-error; min-lines>=30 + instance delta + closure coverage for naked-return); registered empty_error_check as zero-FP 'empty-error-body' per the #507 tombstone reservation. Full package regression green + e2e pipeline verification (2 new bodies reported, steady-state rewrite zero noise).