Linter precision (regression): the syntactic stdlib-package-match migration never landed — 5 holdouts remain, 3 CI-enforced (#40 [Content truncated due to length]

[github-actions]

Summary

Five custom linters still identify standard-library calls by the syntactic identifier name (ident.Name == "<pkg>" / != "<pkg>") instead of by package identity via pass.TypesInfo. The canonical helper astutil.IsPkgSelector(pass, sel, pkgPath) (pkg/linters/internal/astutil/astutil.go:161-178) already exists and is proven in production (rawloginlib.go:58, plus regexpcompileinfunction, sortslice, tolowerequalfold, ctxbackground all migrated).

Two earlier issues tracked exactly these holdouts — #40243 (os/fmt/strings) and #40435 (os/context) — but both were closed via tracker auto-expiry with zero migrations applied. Verified at current HEAD (branch main): all five call sites are still syntactic, and three of them are CI-enforced (.github/workflows/cgo.yml:1208 LINTER_FLAGS), where a false positive is a build-blocker.

The remaining holdouts (verified at HEAD)

Linter	Location	Current code	CI-enforced?
errstringmatch	pkg/linters/errstringmatch/errstringmatch.go:100	if ident.Name != "strings" {	✅ yes
fprintlnsprintf	pkg/linters/fprintlnsprintf/fprintlnsprintf.go:132	return ident.Name == "fmt" && sel.Sel.Name == name	✅ yes
osexitinlibrary	pkg/linters/osexitinlibrary/osexitinlibrary.go:59	if ident.Name == "os" && sel.Sel.Name == "Exit" {	✅ yes
fileclosenotdeferred	pkg/linters/fileclosenotdeferred/fileclosenotdeferred.go:174	`if !ok
contextcancelnotdeferred	pkg/linters/contextcancelnotdeferred/contextcancelnotdeferred.go:140	`if !ok

Why syntactic matching is wrong (two failure modes)

1. Alias-import false negative. import str "strings" then str.Contains(err.Error(), "x") — ident.Name is "str", so the call escapes the linter. Same for import xos "os" / import xfmt "fmt" / import ctx "context". For the three enforced linters this silently lets the targeted anti-pattern ship past CI.
2. Shadowing false positive (build-blocker for the 3 enforced ones). A local variable/parameter/field named os/fmt/strings/context with a matching method or field name matches by name even though it is not the stdlib package — a spurious diagnostic that fails the build.

Current impact

Latent today — pkg/ currently has no aliased os/fmt/strings/context imports and does not shadow those identifiers, so no live FP/FN exists right now. But this is a correctness gap in enforced tooling that will bite the first time someone adds an alias import or a shadowing local. The fix is cheap, the helper is proven, and the sibling migration is already accepted — the two tracking issues simply expired before the work was done.

Correct approach (already used by the migrated linters)

astutil.IsPkgSelector resolves the selector base to a *types.PkgName and compares the imported path:

func IsPkgSelector(pass *analysis.Pass, sel *ast.SelectorExpr, pkgPath string) bool {
    pkgIdent, ok := sel.X.(*ast.Ident)
    if !ok { return false }
    obj := pass.TypesInfo.ObjectOf(pkgIdent)
    pkgName, ok := obj.(*types.PkgName)
    if !ok || pkgName.Imported() == nil { return false }
    return pkgName.Imported().Path() == pkgPath
}

Reference migration — rawloginlib.go:58: if !astutil.IsPkgSelector(pass, sel, "log") {

Recommendation

Replace each syntactic check with the existing helper (each is a one-line change against the already-extracted *ast.SelectorExpr):

• errstringmatch.go:100 → if !astutil.IsPkgSelector(pass, sel, "strings") {
• fprintlnsprintf.go:132 → return astutil.IsPkgSelector(pass, sel, "fmt") && sel.Sel.Name == name
• osexitinlibrary.go:59 → if astutil.IsPkgSelector(pass, sel, "os") && sel.Sel.Name == "Exit" {
• fileclosenotdeferred.go:174 → if !astutil.IsPkgSelector(pass, sel, "os") {
• contextcancelnotdeferred.go:140 → resolve sel via astutil.IsPkgSelector(pass, sel, "context") (this file constructs sel/pkgIdent inline in isContextWithCancelCall; pass the *ast.SelectorExpr to the helper).

Validation checklist

• Per linter: add a testdata case with an aliased import (e.g. str "strings") — confirm it is now flagged (FN closed) for the 3 enforced linters, and detected for the 2 not-enforced.
• Per enforced linter: add a testdata case with a shadowing local named os/fmt/strings — confirm it is NOT flagged (FP / build-blocker closed).
• Existing testdata expectations for the normal strings.Contains / fmt.Fprintln / os.Exit / os.Open+.Close / context.WithCancel cases unchanged.
• go test ./pkg/linters/... green; make golint-custom over pkg/ produces no new diagnostics.

Effort

Small — one-line change per linter plus two testdata cases each; the helper already exists. Single PR. This finishes the migration that #40243 and #40435 both started and that already landed for rawloginlib/regexpcompileinfunction/sortslice/tolowerequalfold/ctxbackground.

Related

#40243 (os/fmt/strings holdouts — expired unfixed), #40435 (os/context holdouts — expired unfixed), #39981 (rawloginlib, landed), #39733 (regexpcompileinfunction), #38789 (ctxbackground). Same syntactic-package-identity class.

References:

• §28842644261

Generated by 🤖 Sergo - Serena Go Expert · 386 AIC · ⌖ 14.1 AIC · ⊞ 5.9K · ◷

• expires on Jul 13, 2026, 9:14 PM UTC-08:00

[github-actions]

🍪 Issue Monster selected this for Copilot

I have identified this issue as a good candidate for automated resolution and requested assignment to the Copilot coding agent.

If assignment succeeds, the Copilot coding agent will analyze the issue and create a pull request with the fix.

Om nom nom! 🍪

🍪 Om nom nom by Issue Monster · 2.03 AIC · ⌖ 8.49 AIC · ⊞ 6.7K · ◷