// Copyright 2013 The Go Authors. All rights reserved. // Use of this source code is governed by a BSD-style // license that can be found in the LICENSE file. package types import ( "cmp" "fmt" "go/ast" "go/constant" "go/token" . "internal/types/errors" "slices" "strconv" "strings" "unicode" ) // A declInfo describes a package-level const, type, var, or func declaration. type declInfo struct { file *Scope // scope of file containing this declaration version goVersion // Go version of file containing this declaration lhs []*Var // lhs of n:1 variable declarations, or nil vtyp ast.Expr // type, or nil (for const and var declarations only) init ast.Expr // init/orig expression, or nil (for const and var declarations only) inherited bool // if set, the init expression is inherited from a previous constant declaration tdecl *ast.TypeSpec // type declaration, or nil fdecl *ast.FuncDecl // func declaration, or nil // The deps field tracks initialization expression dependencies. deps map[Object]bool // lazily initialized } // hasInitializer reports whether the declared object has an initialization // expression or function body. func (d *declInfo) hasInitializer() bool { return d.init != nil || d.fdecl != nil && d.fdecl.Body != nil } // addDep adds obj to the set of objects d's init expression depends on. func (d *declInfo) addDep(obj Object) { m := d.deps if m == nil { m = make(map[Object]bool) d.deps = m } m[obj] = true } // arityMatch checks that the lhs and rhs of a const or var decl // have the appropriate number of names and init exprs. For const // decls, init is the value spec providing the init exprs; for // var decls, init is nil (the init exprs are in s in this case). func (check *Checker) arityMatch(s, init *ast.ValueSpec) { l := len(s.Names) r := len(s.Values) if init != nil { r = len(init.Values) } const code = WrongAssignCount switch { case init == nil && r == 0: // var decl w/o init expr if s.Type == nil { check.error(s, code, "missing type or init expr") } case l < r: if l < len(s.Values) { // init exprs from s n := s.Values[l] check.errorf(n, code, "extra init expr %s", n) // TODO(gri) avoid declared and not used error here } else { // init exprs "inherited" check.errorf(s, code, "extra init expr at %s", check.fset.Position(init.Pos())) // TODO(gri) avoid declared and not used error here } case l > r && (init != nil || r != 1): n := s.Names[r] check.errorf(n, code, "missing init expr for %s", n) } } func validatedImportPath(path string) (string, error) { s, err := strconv.Unquote(path) if err != nil { return "", err } if s == "" { return "", fmt.Errorf("empty string") } const illegalChars = `!"#$%&'()*,:;<=>?[\]^{|}` + "`\uFFFD" for _, r := range s { if !unicode.IsGraphic(r) || unicode.IsSpace(r) || strings.ContainsRune(illegalChars, r) { return s, fmt.Errorf("invalid character %#U", r) } } return s, nil } // declarePkgObj declares obj in the package scope, records its ident -> obj mapping, // and updates check.objMap. The object must not be a function or method. func (check *Checker) declarePkgObj(ident *ast.Ident, obj Object, d *declInfo) { assert(ident.Name == obj.Name()) // spec: "A package-scope or file-scope identifier with name init // may only be declared to be a function with this (func()) signature." if ident.Name == "init" { check.error(ident, InvalidInitDecl, "cannot declare init - must be func") return } // spec: "The main package must have package name main and declare // a function main that takes no arguments and returns no value." if ident.Name == "main" && check.pkg.name == "main" { check.error(ident, InvalidMainDecl, "cannot declare main - must be func") return } check.declare(check.pkg.scope, ident, obj, nopos) check.objMap[obj] = d obj.setOrder(uint32(len(check.objMap))) } // filename returns a filename suitable for debugging output. func (check *Checker) filename(fileNo int) string { file := check.files[fileNo] if pos := file.Pos(); pos.IsValid() { return check.fset.File(pos).Name() } return fmt.Sprintf("file[%d]", fileNo) } func (check *Checker) importPackage(at positioner, path, dir string) *Package { // If we already have a package for the given (path, dir) // pair, use it instead of doing a full import. // Checker.impMap only caches packages that are marked Complete // or fake (dummy packages for failed imports). Incomplete but // non-fake packages do require an import to complete them. key := importKey{path, dir} imp := check.impMap[key] if imp != nil { return imp } // no package yet => import it if path == "C" && (check.conf.FakeImportC || check.conf.go115UsesCgo) { if check.conf.FakeImportC && check.conf.go115UsesCgo { check.error(at, BadImportPath, "cannot use FakeImportC and go115UsesCgo together") } imp = NewPackage("C", "C") imp.fake = true // package scope is not populated imp.cgo = check.conf.go115UsesCgo } else { // ordinary import var err error if importer := check.conf.Importer; importer == nil { err = fmt.Errorf("Config.Importer not installed") } else if importerFrom, ok := importer.(ImporterFrom); ok { imp, err = importerFrom.ImportFrom(path, dir, 0) if imp == nil && err == nil { err = fmt.Errorf("Config.Importer.ImportFrom(%s, %s, 0) returned nil but no error", path, dir) } } else { imp, err = importer.Import(path) if imp == nil && err == nil { err = fmt.Errorf("Config.Importer.Import(%s) returned nil but no error", path) } } // make sure we have a valid package name // (errors here can only happen through manipulation of packages after creation) if err == nil && imp != nil && (imp.name == "_" || imp.name == "") { err = fmt.Errorf("invalid package name: %q", imp.name) imp = nil // create fake package below } if err != nil { check.errorf(at, BrokenImport, "could not import %s (%s)", path, err) if imp == nil { // create a new fake package // come up with a sensible package name (heuristic) name := path if i := len(name); i > 0 && name[i-1] == '/' { name = name[:i-1] } if i := strings.LastIndex(name, "/"); i >= 0 { name = name[i+1:] } imp = NewPackage(path, name) } // continue to use the package as best as we can imp.fake = true // avoid follow-up lookup failures } } // package should be complete or marked fake, but be cautious if imp.complete || imp.fake { check.impMap[key] = imp // Once we've formatted an error message, keep the pkgPathMap // up-to-date on subsequent imports. It is used for package // qualification in error messages. if check.pkgPathMap != nil { check.markImports(imp) } return imp } // something went wrong (importer may have returned incomplete package without error) return nil } // collectObjects collects all file and package objects and inserts them // into their respective scopes. It also performs imports and associates // methods with receiver base type names. func (check *Checker) collectObjects() { pkg := check.pkg // pkgImports is the set of packages already imported by any package file seen // so far. Used to avoid duplicate entries in pkg.imports. Allocate and populate // it (pkg.imports may not be empty if we are checking test files incrementally). // Note that pkgImports is keyed by package (and thus package path), not by an // importKey value. Two different importKey values may map to the same package // which is why we cannot use the check.impMap here. var pkgImports = make(map[*Package]bool) for _, imp := range pkg.imports { pkgImports[imp] = true } type methodInfo struct { obj *Func // method ptr bool // true if pointer receiver recv *ast.Ident // receiver type name } var methods []methodInfo // collected methods with valid receivers and non-blank _ names fileScopes := make([]*Scope, len(check.files)) // fileScopes[i] corresponds to check.files[i] for fileNo, file := range check.files { check.version = asGoVersion(check.versions[file]) // The package identifier denotes the current package, // but there is no corresponding package object. check.recordDef(file.Name, nil) // Use the actual source file extent rather than *ast.File extent since the // latter doesn't include comments which appear at the start or end of the file. // Be conservative and use the *ast.File extent if we don't have a *token.File. pos, end := file.Pos(), file.End() if f := check.fset.File(file.Pos()); f != nil { pos, end = token.Pos(f.Base()), token.Pos(f.Base()+f.Size()) } fileScope := NewScope(pkg.scope, pos, end, check.filename(fileNo)) fileScopes[fileNo] = fileScope check.recordScope(file, fileScope) // determine file directory, necessary to resolve imports // FileName may be "" (typically for tests) in which case // we get "." as the directory which is what we would want. fileDir := dir(check.fset.Position(file.Name.Pos()).Filename) check.walkDecls(file.Decls, func(d decl) { switch d := d.(type) { case importDecl: // import package if d.spec.Path.Value == "" { return // error reported by parser } path, err := validatedImportPath(d.spec.Path.Value) if err != nil { check.errorf(d.spec.Path, BadImportPath, "invalid import path (%s)", err) return } imp := check.importPackage(d.spec.Path, path, fileDir) if imp == nil { return } // local name overrides imported package name name := imp.name if d.spec.Name != nil { name = d.spec.Name.Name if path == "C" { // match 1.17 cmd/compile (not prescribed by spec) check.error(d.spec.Name, ImportCRenamed, `cannot rename import "C"`) return } } if name == "init" { check.error(d.spec, InvalidInitDecl, "cannot import package as init - init must be a func") return } // add package to list of explicit imports // (this functionality is provided as a convenience // for clients; it is not needed for type-checking) if !pkgImports[imp] { pkgImports[imp] = true pkg.imports = append(pkg.imports, imp) } pkgName := NewPkgName(d.spec.Pos(), pkg, name, imp) if d.spec.Name != nil { // in a dot-import, the dot represents the package check.recordDef(d.spec.Name, pkgName) } else { check.recordImplicit(d.spec, pkgName) } if imp.fake { // match 1.17 cmd/compile (not prescribed by spec) pkgName.used = true } // add import to file scope check.imports = append(check.imports, pkgName) if name == "." { // dot-import if check.dotImportMap == nil { check.dotImportMap = make(map[dotImportKey]*PkgName) } // merge imported scope with file scope for name, obj := range imp.scope.elems { // Note: Avoid eager resolve(name, obj) here, so we only // resolve dot-imported objects as needed. // A package scope may contain non-exported objects, // do not import them! if token.IsExported(name) { // declare dot-imported object // (Do not use check.declare because it modifies the object // via Object.setScopePos, which leads to a race condition; // the object may be imported into more than one file scope // concurrently. See go.dev/issue/32154.) if alt := fileScope.Lookup(name); alt != nil { err := check.newError(DuplicateDecl) err.addf(d.spec.Name, "%s redeclared in this block", alt.Name()) err.addAltDecl(alt) err.report() } else { fileScope.insert(name, obj) check.dotImportMap[dotImportKey{fileScope, name}] = pkgName } } } } else { // declare imported package object in file scope // (no need to provide s.Name since we called check.recordDef earlier) check.declare(fileScope, nil, pkgName, nopos) } case constDecl: // declare all constants for i, name := range d.spec.Names { obj := NewConst(name.Pos(), pkg, name.Name, nil, constant.MakeInt64(int64(d.iota))) var init ast.Expr if i < len(d.init) { init = d.init[i] } d := &declInfo{file: fileScope, version: check.version, vtyp: d.typ, init: init, inherited: d.inherited} check.declarePkgObj(name, obj, d) } case varDecl: lhs := make([]*Var, len(d.spec.Names)) // If there's exactly one rhs initializer, use // the same declInfo d1 for all lhs variables // so that each lhs variable depends on the same // rhs initializer (n:1 var declaration). var d1 *declInfo if len(d.spec.Values) == 1 { // The lhs elements are only set up after the for loop below, // but that's ok because declareVar only collects the declInfo // for a later phase. d1 = &declInfo{file: fileScope, version: check.version, lhs: lhs, vtyp: d.spec.Type, init: d.spec.Values[0]} } // declare all variables for i, name := range d.spec.Names { obj := NewVar(name.Pos(), pkg, name.Name, nil) lhs[i] = obj di := d1 if di == nil { // individual assignments var init ast.Expr if i < len(d.spec.Values) { init = d.spec.Values[i] } di = &declInfo{file: fileScope, version: check.version, vtyp: d.spec.Type, init: init} } check.declarePkgObj(name, obj, di) } case typeDecl: obj := NewTypeName(d.spec.Name.Pos(), pkg, d.spec.Name.Name, nil) check.declarePkgObj(d.spec.Name, obj, &declInfo{file: fileScope, version: check.version, tdecl: d.spec}) case funcDecl: name := d.decl.Name.Name obj := NewFunc(d.decl.Name.Pos(), pkg, name, nil) // signature set later hasTParamError := false // avoid duplicate type parameter errors if d.decl.Recv.NumFields() == 0 { // regular function if d.decl.Recv != nil { check.error(d.decl.Recv, BadRecv, "method has no receiver") // treat as function } if name == "init" || (name == "main" && check.pkg.name == "main") { code := InvalidInitDecl if name == "main" { code = InvalidMainDecl } if d.decl.Type.TypeParams.NumFields() != 0 { check.softErrorf(d.decl.Type.TypeParams.List[0], code, "func %s must have no type parameters", name) hasTParamError = true } if t := d.decl.Type; t.Params.NumFields() != 0 || t.Results != nil { // TODO(rFindley) Should this be a hard error? check.softErrorf(d.decl.Name, code, "func %s must have no arguments and no return values", name) } } if name == "init" { // don't declare init functions in the package scope - they are invisible obj.parent = pkg.scope check.recordDef(d.decl.Name, obj) // init functions must have a body if d.decl.Body == nil { // TODO(gri) make this error message consistent with the others above check.softErrorf(obj, MissingInitBody, "missing function body") } } else { check.declare(pkg.scope, d.decl.Name, obj, nopos) } } else { // method // TODO(rFindley) earlier versions of this code checked that methods // have no type parameters, but this is checked later // when type checking the function type. Confirm that // we don't need to check tparams here. ptr, base, _ := check.unpackRecv(d.decl.Recv.List[0].Type, false) // (Methods with invalid receiver cannot be associated to a type, and // methods with blank _ names are never found; no need to collect any // of them. They will still be type-checked with all the other functions.) if recv, _ := base.(*ast.Ident); recv != nil && name != "_" { methods = append(methods, methodInfo{obj, ptr, recv}) } check.recordDef(d.decl.Name, obj) } _ = d.decl.Type.TypeParams.NumFields() != 0 && !hasTParamError && check.verifyVersionf(d.decl.Type.TypeParams.List[0], go1_18, "type parameter") info := &declInfo{file: fileScope, version: check.version, fdecl: d.decl} // Methods are not package-level objects but we still track them in the // object map so that we can handle them like regular functions (if the // receiver is invalid); also we need their fdecl info when associating // them with their receiver base type, below. check.objMap[obj] = info obj.setOrder(uint32(len(check.objMap))) } }) } // verify that objects in package and file scopes have different names for _, scope := range fileScopes { for name, obj := range scope.elems { if alt := pkg.scope.Lookup(name); alt != nil { obj = resolve(name, obj) err := check.newError(DuplicateDecl) if pkg, ok := obj.(*PkgName); ok { err.addf(alt, "%s already declared through import of %s", alt.Name(), pkg.Imported()) err.addAltDecl(pkg) } else { err.addf(alt, "%s already declared through dot-import of %s", alt.Name(), obj.Pkg()) // TODO(gri) dot-imported objects don't have a position; addAltDecl won't print anything err.addAltDecl(obj) } err.report() } } } // Now that we have all package scope objects and all methods, // associate methods with receiver base type name where possible. // Ignore methods that have an invalid receiver. They will be // type-checked later, with regular functions. if methods == nil { return } // lookupScope returns the file scope which contains the given name, // or nil if the name is not found in any scope. The search does not // step inside blocks (function bodies). // This function is only used in conjuction with import "C", and even // then only rarely. It doesn't have to be particularly fast. lookupScope := func(name *ast.Ident) *Scope { for i, file := range check.files { found := false ast.Inspect(file, func(n ast.Node) bool { if found { return false // we're done } switch n := n.(type) { case *ast.Ident: if n == name { found = true return false } case *ast.BlockStmt: return false // don't descend into function bodies } return true }) if found { return fileScopes[i] } } return nil } check.methods = make(map[*TypeName][]*Func) for i := range methods { m := &methods[i] // Determine the receiver base type and associate m with it. ptr, base := check.resolveBaseTypeName(m.ptr, m.recv, lookupScope) if base != nil { m.obj.hasPtrRecv_ = ptr check.methods[base] = append(check.methods[base], m.obj) } } } // unpackRecv unpacks a receiver type expression and returns its components: ptr indicates // whether rtyp is a pointer receiver, base is the receiver base type expression stripped // of its type parameters (if any), and tparams are its type parameter names, if any. The // type parameters are only unpacked if unpackParams is set. For instance, given the rtyp // // *T[A, _] // // ptr is true, base is T, and tparams is [A, _] (assuming unpackParams is set). // Note that base may not be a *ast.Ident for erroneous programs. func (check *Checker) unpackRecv(rtyp ast.Expr, unpackParams bool) (ptr bool, base ast.Expr, tparams []*ast.Ident) { // unpack receiver type base = ast.Unparen(rtyp) if t, _ := base.(*ast.StarExpr); t != nil { ptr = true base = ast.Unparen(t.X) } // unpack type parameters, if any switch base.(type) { case *ast.IndexExpr, *ast.IndexListExpr: ix := unpackIndexedExpr(base) base = ix.x if unpackParams { for _, arg := range ix.indices { var par *ast.Ident switch arg := arg.(type) { case *ast.Ident: par = arg case *ast.BadExpr: // ignore - error already reported by parser case nil: check.error(ix.orig, InvalidSyntaxTree, "parameterized receiver contains nil parameters") default: check.errorf(arg, BadDecl, "receiver type parameter %s must be an identifier", arg) } if par == nil { par = &ast.Ident{NamePos: arg.Pos(), Name: "_"} } tparams = append(tparams, par) } } } return } // resolveBaseTypeName returns the non-alias base type name for typ, and whether // there was a pointer indirection to get to it. The base type name must be declared // in package scope, and there can be at most one pointer indirection. If no such type // name exists, the returned base is nil. func (check *Checker) resolveBaseTypeName(seenPtr bool, typ ast.Expr, lookupScope func(*ast.Ident) *Scope) (ptr bool, base *TypeName) { // Algorithm: Starting from a type expression, which may be a name, // we follow that type through alias declarations until we reach a // non-alias type name. If we encounter anything but pointer types or // parentheses we're done. If we encounter more than one pointer type // we're done. ptr = seenPtr var seen map[*TypeName]bool for { // Note: this differs from types2, but is necessary. The syntax parser // strips unnecessary parens. typ = ast.Unparen(typ) // check if we have a pointer type if pexpr, _ := typ.(*ast.StarExpr); pexpr != nil { // if we've already seen a pointer, we're done if ptr { return false, nil } ptr = true typ = ast.Unparen(pexpr.X) // continue with pointer base type } // typ must be a name, or a C.name cgo selector. var name string switch typ := typ.(type) { case *ast.Ident: name = typ.Name case *ast.SelectorExpr: // C.struct_foo is a valid type name for packages using cgo. // See go.dev/issue/59944. // TODO(gri) why is it possible to associate methods with C types? // // Detect this case, and adjust name so that the correct TypeName is // resolved below. if ident, _ := typ.X.(*ast.Ident); ident != nil && ident.Name == "C" { // Check whether "C" actually resolves to an import of "C", by looking // in the appropriate file scope. obj := lookupScope(ident).Lookup(ident.Name) // the fileScope must always be found // If Config.go115UsesCgo is set, the typechecker will resolve Cgo // selectors to their cgo name. We must do the same here. if pname, _ := obj.(*PkgName); pname != nil { if pname.imported.cgo { // only set if Config.go115UsesCgo is set name = "_Ctype_" + typ.Sel.Name } } } if name == "" { return false, nil } default: return false, nil } // name must denote an object found in the current package scope // (note that dot-imported objects are not in the package scope!) obj := check.pkg.scope.Lookup(name) if obj == nil { return false, nil } // the object must be a type name... tname, _ := obj.(*TypeName) if tname == nil { return false, nil } // ... which we have not seen before if seen[tname] { return false, nil } // we're done if tdecl defined tname as a new type // (rather than an alias) tdecl := check.objMap[tname].tdecl // must exist for objects in package scope if !tdecl.Assign.IsValid() { return ptr, tname } // otherwise, continue resolving typ = tdecl.Type if seen == nil { seen = make(map[*TypeName]bool) } seen[tname] = true } } // packageObjects typechecks all package objects, but not function bodies. func (check *Checker) packageObjects() { // process package objects in source order for reproducible results objList := make([]Object, len(check.objMap)) i := 0 for obj := range check.objMap { objList[i] = obj i++ } slices.SortFunc(objList, func(a, b Object) int { return cmp.Compare(a.order(), b.order()) }) // add new methods to already type-checked types (from a prior Checker.Files call) for _, obj := range objList { if obj, _ := obj.(*TypeName); obj != nil && obj.typ != nil { check.collectMethods(obj) } } if false && check.conf._EnableAlias { // With Alias nodes we can process declarations in any order. // // TODO(adonovan): unfortunately, Alias nodes // (GODEBUG=gotypesalias=1) don't entirely resolve // problems with cycles. For example, in // GOROOT/test/typeparam/issue50259.go, // // type T[_ any] struct{} // type A T[B] // type B = T[A] // // TypeName A has Type Named during checking, but by // the time the unified export data is written out, // its Type is Invalid. // // Investigate and reenable this branch. for _, obj := range objList { check.objDecl(obj, nil) } } else { // Without Alias nodes, we process non-alias type declarations first, followed by // alias declarations, and then everything else. This appears to avoid most situations // where the type of an alias is needed before it is available. // There may still be cases where this is not good enough (see also go.dev/issue/25838). // In those cases Checker.ident will report an error ("invalid use of type alias"). var aliasList []*TypeName var othersList []Object // everything that's not a type // phase 1: non-alias type declarations for _, obj := range objList { if tname, _ := obj.(*TypeName); tname != nil { if check.objMap[tname].tdecl.Assign.IsValid() { aliasList = append(aliasList, tname) } else { check.objDecl(obj, nil) } } else { othersList = append(othersList, obj) } } // phase 2: alias type declarations for _, obj := range aliasList { check.objDecl(obj, nil) } // phase 3: all other declarations for _, obj := range othersList { check.objDecl(obj, nil) } } // At this point we may have a non-empty check.methods map; this means that not all // entries were deleted at the end of typeDecl because the respective receiver base // types were not found. In that case, an error was reported when declaring those // methods. We can now safely discard this map. check.methods = nil } // unusedImports checks for unused imports. func (check *Checker) unusedImports() { // If function bodies are not checked, packages' uses are likely missing - don't check. if check.conf.IgnoreFuncBodies { return } // spec: "It is illegal (...) to directly import a package without referring to // any of its exported identifiers. To import a package solely for its side-effects // (initialization), use the blank identifier as explicit package name." for _, obj := range check.imports { if !obj.used && obj.name != "_" { check.errorUnusedPkg(obj) } } } func (check *Checker) errorUnusedPkg(obj *PkgName) { // If the package was imported with a name other than the final // import path element, show it explicitly in the error message. // Note that this handles both renamed imports and imports of // packages containing unconventional package declarations. // Note that this uses / always, even on Windows, because Go import // paths always use forward slashes. path := obj.imported.path elem := path if i := strings.LastIndex(elem, "/"); i >= 0 { elem = elem[i+1:] } if obj.name == "" || obj.name == "." || obj.name == elem { check.softErrorf(obj, UnusedImport, "%q imported and not used", path) } else { check.softErrorf(obj, UnusedImport, "%q imported as %s and not used", path, obj.name) } } // dir makes a good-faith attempt to return the directory // portion of path. If path is empty, the result is ".". // (Per the go/build package dependency tests, we cannot import // path/filepath and simply use filepath.Dir.) func dir(path string) string { if i := strings.LastIndexAny(path, `/\`); i > 0 { return path[:i] } // i <= 0 return "." }