Giant blob of minor changes
[dotfiles/.git] / .config / coc / extensions / coc-go-data / tools / pkg / mod / golang.org / x / tools@v0.0.0-20201028153306-37f0764111ff / internal / analysisinternal / analysis.go
diff --git a/.config/coc/extensions/coc-go-data/tools/pkg/mod/golang.org/x/tools@v0.0.0-20201028153306-37f0764111ff/internal/analysisinternal/analysis.go b/.config/coc/extensions/coc-go-data/tools/pkg/mod/golang.org/x/tools@v0.0.0-20201028153306-37f0764111ff/internal/analysisinternal/analysis.go
new file mode 100644 (file)
index 0000000..01f6e82
--- /dev/null
@@ -0,0 +1,425 @@
+// Copyright 2020 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 analysisinternal exposes internal-only fields from go/analysis.
+package analysisinternal
+
+import (
+       "bytes"
+       "fmt"
+       "go/ast"
+       "go/token"
+       "go/types"
+       "strings"
+
+       "golang.org/x/tools/go/ast/astutil"
+       "golang.org/x/tools/internal/lsp/fuzzy"
+)
+
+var (
+       GetTypeErrors func(p interface{}) []types.Error
+       SetTypeErrors func(p interface{}, errors []types.Error)
+)
+
+func TypeErrorEndPos(fset *token.FileSet, src []byte, start token.Pos) token.Pos {
+       // Get the end position for the type error.
+       offset, end := fset.PositionFor(start, false).Offset, start
+       if offset >= len(src) {
+               return end
+       }
+       if width := bytes.IndexAny(src[offset:], " \n,():;[]+-*"); width > 0 {
+               end = start + token.Pos(width)
+       }
+       return end
+}
+
+func ZeroValue(fset *token.FileSet, f *ast.File, pkg *types.Package, typ types.Type) ast.Expr {
+       under := typ
+       if n, ok := typ.(*types.Named); ok {
+               under = n.Underlying()
+       }
+       switch u := under.(type) {
+       case *types.Basic:
+               switch {
+               case u.Info()&types.IsNumeric != 0:
+                       return &ast.BasicLit{Kind: token.INT, Value: "0"}
+               case u.Info()&types.IsBoolean != 0:
+                       return &ast.Ident{Name: "false"}
+               case u.Info()&types.IsString != 0:
+                       return &ast.BasicLit{Kind: token.STRING, Value: `""`}
+               default:
+                       panic("unknown basic type")
+               }
+       case *types.Chan, *types.Interface, *types.Map, *types.Pointer, *types.Signature, *types.Slice, *types.Array:
+               return ast.NewIdent("nil")
+       case *types.Struct:
+               texpr := TypeExpr(fset, f, pkg, typ) // typ because we want the name here.
+               if texpr == nil {
+                       return nil
+               }
+               return &ast.CompositeLit{
+                       Type: texpr,
+               }
+       }
+       return nil
+}
+
+// IsZeroValue checks whether the given expression is a 'zero value' (as determined by output of
+// analysisinternal.ZeroValue)
+func IsZeroValue(expr ast.Expr) bool {
+       switch e := expr.(type) {
+       case *ast.BasicLit:
+               return e.Value == "0" || e.Value == `""`
+       case *ast.Ident:
+               return e.Name == "nil" || e.Name == "false"
+       default:
+               return false
+       }
+}
+
+func TypeExpr(fset *token.FileSet, f *ast.File, pkg *types.Package, typ types.Type) ast.Expr {
+       switch t := typ.(type) {
+       case *types.Basic:
+               switch t.Kind() {
+               case types.UnsafePointer:
+                       return &ast.SelectorExpr{X: ast.NewIdent("unsafe"), Sel: ast.NewIdent("Pointer")}
+               default:
+                       return ast.NewIdent(t.Name())
+               }
+       case *types.Pointer:
+               x := TypeExpr(fset, f, pkg, t.Elem())
+               if x == nil {
+                       return nil
+               }
+               return &ast.UnaryExpr{
+                       Op: token.MUL,
+                       X:  x,
+               }
+       case *types.Array:
+               elt := TypeExpr(fset, f, pkg, t.Elem())
+               if elt == nil {
+                       return nil
+               }
+               return &ast.ArrayType{
+                       Len: &ast.BasicLit{
+                               Kind:  token.INT,
+                               Value: fmt.Sprintf("%d", t.Len()),
+                       },
+                       Elt: elt,
+               }
+       case *types.Slice:
+               elt := TypeExpr(fset, f, pkg, t.Elem())
+               if elt == nil {
+                       return nil
+               }
+               return &ast.ArrayType{
+                       Elt: elt,
+               }
+       case *types.Map:
+               key := TypeExpr(fset, f, pkg, t.Key())
+               value := TypeExpr(fset, f, pkg, t.Elem())
+               if key == nil || value == nil {
+                       return nil
+               }
+               return &ast.MapType{
+                       Key:   key,
+                       Value: value,
+               }
+       case *types.Chan:
+               dir := ast.ChanDir(t.Dir())
+               if t.Dir() == types.SendRecv {
+                       dir = ast.SEND | ast.RECV
+               }
+               value := TypeExpr(fset, f, pkg, t.Elem())
+               if value == nil {
+                       return nil
+               }
+               return &ast.ChanType{
+                       Dir:   dir,
+                       Value: value,
+               }
+       case *types.Signature:
+               var params []*ast.Field
+               for i := 0; i < t.Params().Len(); i++ {
+                       p := TypeExpr(fset, f, pkg, t.Params().At(i).Type())
+                       if p == nil {
+                               return nil
+                       }
+                       params = append(params, &ast.Field{
+                               Type: p,
+                               Names: []*ast.Ident{
+                                       {
+                                               Name: t.Params().At(i).Name(),
+                                       },
+                               },
+                       })
+               }
+               var returns []*ast.Field
+               for i := 0; i < t.Results().Len(); i++ {
+                       r := TypeExpr(fset, f, pkg, t.Results().At(i).Type())
+                       if r == nil {
+                               return nil
+                       }
+                       returns = append(returns, &ast.Field{
+                               Type: r,
+                       })
+               }
+               return &ast.FuncType{
+                       Params: &ast.FieldList{
+                               List: params,
+                       },
+                       Results: &ast.FieldList{
+                               List: returns,
+                       },
+               }
+       case *types.Named:
+               if t.Obj().Pkg() == nil {
+                       return ast.NewIdent(t.Obj().Name())
+               }
+               if t.Obj().Pkg() == pkg {
+                       return ast.NewIdent(t.Obj().Name())
+               }
+               pkgName := t.Obj().Pkg().Name()
+               // If the file already imports the package under another name, use that.
+               for _, group := range astutil.Imports(fset, f) {
+                       for _, cand := range group {
+                               if strings.Trim(cand.Path.Value, `"`) == t.Obj().Pkg().Path() {
+                                       if cand.Name != nil && cand.Name.Name != "" {
+                                               pkgName = cand.Name.Name
+                                       }
+                               }
+                       }
+               }
+               if pkgName == "." {
+                       return ast.NewIdent(t.Obj().Name())
+               }
+               return &ast.SelectorExpr{
+                       X:   ast.NewIdent(pkgName),
+                       Sel: ast.NewIdent(t.Obj().Name()),
+               }
+       case *types.Struct:
+               return ast.NewIdent(t.String())
+       case *types.Interface:
+               return ast.NewIdent(t.String())
+       default:
+               return nil
+       }
+}
+
+type TypeErrorPass string
+
+const (
+       NoNewVars      TypeErrorPass = "nonewvars"
+       NoResultValues TypeErrorPass = "noresultvalues"
+       UndeclaredName TypeErrorPass = "undeclaredname"
+)
+
+// StmtToInsertVarBefore returns the ast.Stmt before which we can safely insert a new variable.
+// Some examples:
+//
+// Basic Example:
+// z := 1
+// y := z + x
+// If x is undeclared, then this function would return `y := z + x`, so that we
+// can insert `x := ` on the line before `y := z + x`.
+//
+// If stmt example:
+// if z == 1 {
+// } else if z == y {}
+// If y is undeclared, then this function would return `if z == 1 {`, because we cannot
+// insert a statement between an if and an else if statement. As a result, we need to find
+// the top of the if chain to insert `y := ` before.
+func StmtToInsertVarBefore(path []ast.Node) ast.Stmt {
+       enclosingIndex := -1
+       for i, p := range path {
+               if _, ok := p.(ast.Stmt); ok {
+                       enclosingIndex = i
+                       break
+               }
+       }
+       if enclosingIndex == -1 {
+               return nil
+       }
+       enclosingStmt := path[enclosingIndex]
+       switch enclosingStmt.(type) {
+       case *ast.IfStmt:
+               // The enclosingStmt is inside of the if declaration,
+               // We need to check if we are in an else-if stmt and
+               // get the base if statement.
+               return baseIfStmt(path, enclosingIndex)
+       case *ast.CaseClause:
+               // Get the enclosing switch stmt if the enclosingStmt is
+               // inside of the case statement.
+               for i := enclosingIndex + 1; i < len(path); i++ {
+                       if node, ok := path[i].(*ast.SwitchStmt); ok {
+                               return node
+                       } else if node, ok := path[i].(*ast.TypeSwitchStmt); ok {
+                               return node
+                       }
+               }
+       }
+       if len(path) <= enclosingIndex+1 {
+               return enclosingStmt.(ast.Stmt)
+       }
+       // Check if the enclosing statement is inside another node.
+       switch expr := path[enclosingIndex+1].(type) {
+       case *ast.IfStmt:
+               // Get the base if statement.
+               return baseIfStmt(path, enclosingIndex+1)
+       case *ast.ForStmt:
+               if expr.Init == enclosingStmt || expr.Post == enclosingStmt {
+                       return expr
+               }
+       }
+       return enclosingStmt.(ast.Stmt)
+}
+
+// baseIfStmt walks up the if/else-if chain until we get to
+// the top of the current if chain.
+func baseIfStmt(path []ast.Node, index int) ast.Stmt {
+       stmt := path[index]
+       for i := index + 1; i < len(path); i++ {
+               if node, ok := path[i].(*ast.IfStmt); ok && node.Else == stmt {
+                       stmt = node
+                       continue
+               }
+               break
+       }
+       return stmt.(ast.Stmt)
+}
+
+// WalkASTWithParent walks the AST rooted at n. The semantics are
+// similar to ast.Inspect except it does not call f(nil).
+func WalkASTWithParent(n ast.Node, f func(n ast.Node, parent ast.Node) bool) {
+       var ancestors []ast.Node
+       ast.Inspect(n, func(n ast.Node) (recurse bool) {
+               if n == nil {
+                       ancestors = ancestors[:len(ancestors)-1]
+                       return false
+               }
+
+               var parent ast.Node
+               if len(ancestors) > 0 {
+                       parent = ancestors[len(ancestors)-1]
+               }
+               ancestors = append(ancestors, n)
+               return f(n, parent)
+       })
+}
+
+// FindMatchingIdents finds all identifiers in 'node' that match any of the given types.
+// 'pos' represents the position at which the identifiers may be inserted. 'pos' must be within
+// the scope of each of identifier we select. Otherwise, we will insert a variable at 'pos' that
+// is unrecognized.
+func FindMatchingIdents(typs []types.Type, node ast.Node, pos token.Pos, info *types.Info, pkg *types.Package) map[types.Type][]*ast.Ident {
+       matches := map[types.Type][]*ast.Ident{}
+       // Initialize matches to contain the variable types we are searching for.
+       for _, typ := range typs {
+               if typ == nil {
+                       continue
+               }
+               matches[typ] = []*ast.Ident{}
+       }
+       seen := map[types.Object]struct{}{}
+       ast.Inspect(node, func(n ast.Node) bool {
+               if n == nil {
+                       return false
+               }
+               // Prevent circular definitions. If 'pos' is within an assignment statement, do not
+               // allow any identifiers in that assignment statement to be selected. Otherwise,
+               // we could do the following, where 'x' satisfies the type of 'f0':
+               //
+               // x := fakeStruct{f0: x}
+               //
+               assignment, ok := n.(*ast.AssignStmt)
+               if ok && pos > assignment.Pos() && pos <= assignment.End() {
+                       return false
+               }
+               if n.End() > pos {
+                       return n.Pos() <= pos
+               }
+               ident, ok := n.(*ast.Ident)
+               if !ok || ident.Name == "_" {
+                       return true
+               }
+               obj := info.Defs[ident]
+               if obj == nil || obj.Type() == nil {
+                       return true
+               }
+               if _, ok := obj.(*types.TypeName); ok {
+                       return true
+               }
+               // Prevent duplicates in matches' values.
+               if _, ok = seen[obj]; ok {
+                       return true
+               }
+               seen[obj] = struct{}{}
+               // Find the scope for the given position. Then, check whether the object
+               // exists within the scope.
+               innerScope := pkg.Scope().Innermost(pos)
+               if innerScope == nil {
+                       return true
+               }
+               _, foundObj := innerScope.LookupParent(ident.Name, pos)
+               if foundObj != obj {
+                       return true
+               }
+               // The object must match one of the types that we are searching for.
+               if idents, ok := matches[obj.Type()]; ok {
+                       matches[obj.Type()] = append(idents, ast.NewIdent(ident.Name))
+               }
+               // If the object type does not exactly match any of the target types, greedily
+               // find the first target type that the object type can satisfy.
+               for typ := range matches {
+                       if obj.Type() == typ {
+                               continue
+                       }
+                       if equivalentTypes(obj.Type(), typ) {
+                               matches[typ] = append(matches[typ], ast.NewIdent(ident.Name))
+                       }
+               }
+               return true
+       })
+       return matches
+}
+
+func equivalentTypes(want, got types.Type) bool {
+       if want == got || types.Identical(want, got) {
+               return true
+       }
+       // Code segment to help check for untyped equality from (golang/go#32146).
+       if rhs, ok := want.(*types.Basic); ok && rhs.Info()&types.IsUntyped > 0 {
+               if lhs, ok := got.Underlying().(*types.Basic); ok {
+                       return rhs.Info()&types.IsConstType == lhs.Info()&types.IsConstType
+               }
+       }
+       return types.AssignableTo(want, got)
+}
+
+// FindBestMatch employs fuzzy matching to evaluate the similarity of each given identifier to the
+// given pattern. We return the identifier whose name is most similar to the pattern.
+func FindBestMatch(pattern string, idents []*ast.Ident) ast.Expr {
+       fuzz := fuzzy.NewMatcher(pattern)
+       var bestFuzz ast.Expr
+       highScore := float32(0) // minimum score is 0 (no match)
+       for _, ident := range idents {
+               // TODO: Improve scoring algorithm.
+               score := fuzz.Score(ident.Name)
+               if score > highScore {
+                       highScore = score
+                       bestFuzz = ident
+               } else if score == 0 {
+                       // Order matters in the fuzzy matching algorithm. If we find no match
+                       // when matching the target to the identifier, try matching the identifier
+                       // to the target.
+                       revFuzz := fuzzy.NewMatcher(ident.Name)
+                       revScore := revFuzz.Score(pattern)
+                       if revScore > highScore {
+                               highScore = revScore
+                               bestFuzz = ident
+                       }
+               }
+       }
+       return bestFuzz
+}