gopls/internal/golang/semtok.go

// Copyright 2024 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 golang

// This file defines the Semantic Tokens operation for Go source.

import (
	"bytes"
	"context"
	"errors"
	"fmt"
	"go/ast"
	"go/token"
	"go/types"
	"log"
	"path/filepath"
	"regexp"
	"slices"
	"strconv"
	"strings"
	"time"

	"golang.org/x/tools/gopls/internal/cache"
	"golang.org/x/tools/gopls/internal/cache/metadata"
	"golang.org/x/tools/gopls/internal/cache/parsego"
	"golang.org/x/tools/gopls/internal/file"
	"golang.org/x/tools/gopls/internal/protocol"
	"golang.org/x/tools/gopls/internal/protocol/semtok"
	"golang.org/x/tools/gopls/internal/util/bug"
	"golang.org/x/tools/gopls/internal/util/safetoken"
	"golang.org/x/tools/internal/astutil"
	"golang.org/x/tools/internal/event"
	"golang.org/x/tools/internal/fmtstr"
)

// semDebug enables comprehensive logging of decisions
// (gopls semtok foo.go > /dev/null shows log output).
// It should never be true in checked-in code.
const semDebug = false

func SemanticTokens(ctx context.Context, snapshot *cache.Snapshot, fh file.Handle, rng *protocol.Range) (*protocol.SemanticTokens, error) {
	pkg, pgf, err := NarrowestPackageForFile(ctx, snapshot, fh.URI())
	if err != nil {
		return nil, err
	}

	// Select range.
	var start, end token.Pos
	if rng != nil {
		var err error
		start, end, err = pgf.RangePos(*rng)
		if err != nil {
			return nil, err // e.g. invalid range
		}
	} else {
		tok := pgf.Tok
		start, end = tok.Pos(0), tok.Pos(tok.Size()) // entire file
	}

	// Reject full semantic token requests for large files.
	//
	// The LSP says that errors for the semantic token requests
	// should only be returned for exceptions (a word not
	// otherwise defined). This code treats a too-large file as an
	// exception. On parse errors, the code does what it can.
	const maxFullFileSize = 100000
	if int(end-start) > maxFullFileSize {
		return nil, fmt.Errorf("semantic tokens: range %s too large (%d > %d)",
			fh.URI().Path(), end-start, maxFullFileSize)
	}

	tv := tokenVisitor{
		ctx:            ctx,
		metadataSource: snapshot,
		metadata:       pkg.Metadata(),
		info:           pkg.TypesInfo(),
		fset:           pkg.FileSet(),
		pkg:            pkg,
		pgf:            pgf,
		start:          start,
		end:            end,
	}
	tv.visit()
	return &protocol.SemanticTokens{
		Data: semtok.Encode(
			tv.tokens,
			snapshot.Options().EnabledSemanticTokenTypes(),
			snapshot.Options().EnabledSemanticTokenModifiers()),
		ResultID: time.Now().String(), // for delta requests, but we've never seen any
	}, nil
}

type tokenVisitor struct {
	// inputs
	ctx            context.Context // for event logging
	metadataSource metadata.Source // used to resolve imports
	metadata       *metadata.Package
	info           *types.Info
	fset           *token.FileSet
	pkg            *cache.Package
	pgf            *parsego.File
	start, end     token.Pos // range of interest

	// working state
	stack  []ast.Node     // path from root of the syntax tree
	tokens []semtok.Token // computed sequence of semantic tokens
}

func (tv *tokenVisitor) visit() {
	f := tv.pgf.File
	// may not be in range, but harmless
	tv.token(f.Package, len("package"), semtok.TokKeyword)
	if f.Name != nil {
		tv.token(f.Name.NamePos, len(f.Name.Name), semtok.TokNamespace)
	}
	for _, decl := range f.Decls {
		// Only look at the decls that overlap the range.
		if decl.End() <= tv.start || decl.Pos() >= tv.end {
			continue
		}
		ast.Inspect(decl, tv.inspect)
	}

	// Scan all files for imported pkgs, ignore the ambiguous pkg.
	// This is to be consistent with the behavior in [go/doc]: https://pkg.go.dev/pkg/go/doc.
	importByName := make(map[string]*types.PkgName)
	for _, pgf := range tv.pkg.CompiledGoFiles() {
		for _, imp := range pgf.File.Imports {
			if obj := tv.pkg.TypesInfo().PkgNameOf(imp); obj != nil {
				if old, ok := importByName[obj.Name()]; ok {
					if old != nil && old.Imported() != obj.Imported() {
						importByName[obj.Name()] = nil // nil => ambiguous across files
					}
					continue
				}
				importByName[obj.Name()] = obj
			}
		}
	}

	for _, cg := range f.Comments {
		for _, c := range cg.List {
			// Only look at the comment that overlap the range.
			if c.End() <= tv.start || c.Pos() >= tv.end {
				continue
			}
			tv.comment(c, importByName)
		}
	}
}

// Matches (for example) "[F]", "[*p.T]", "[p.T.M]"
// unless followed by a colon (exclude url link, e.g. "[go]: https://go.dev").
// The first group is reference name. e.g. The first group of "[*p.T.M]" is "p.T.M".
var docLinkRegex = regexp.MustCompile(`\[\*?([\pL_][\pL_0-9]*(\.[\pL_][\pL_0-9]*){0,2})](?:[^:]|$)`)

// comment emits semantic tokens for a comment.
// If the comment contains doc links or "go:" directives,
// it emits a separate token for each link or directive and
// each comment portion between them.
func (tv *tokenVisitor) comment(c *ast.Comment, importByName map[string]*types.PkgName) {
	if strings.HasPrefix(c.Text, "//go:") {
		tv.godirective(c)
		return
	}

	pkgScope := tv.pkg.Types().Scope()
	// lookupObjects interprets the name in various forms
	// (X, p.T, p.T.M, etc) and return the list of symbols
	// denoted by each identifier in the dotted list.
	lookupObjects := func(name string) (objs []types.Object) {
		scope := pkgScope
		if pkg, suffix, ok := strings.Cut(name, "."); ok {
			if obj, _ := importByName[pkg]; obj != nil {
				objs = append(objs, obj)
				scope = obj.Imported().Scope()
				name = suffix
			}
		}

		if recv, method, ok := strings.Cut(name, "."); ok {
			obj, ok := scope.Lookup(recv).(*types.TypeName)
			if !ok {
				return nil
			}
			objs = append(objs, obj)
			t, ok := obj.Type().(*types.Named)
			if !ok {
				return nil
			}
			m, _, _ := types.LookupFieldOrMethod(t, true, tv.pkg.Types(), method)
			if m == nil {
				return nil
			}
			objs = append(objs, m)
			return objs
		} else {
			obj := scope.Lookup(name)
			if obj == nil {
				return nil
			}
			if _, ok := obj.(*types.PkgName); !ok && !obj.Exported() {
				return nil
			}
			objs = append(objs, obj)
			return objs

		}
	}

	pos := c.Pos()
	for line := range strings.SplitSeq(c.Text, "\n") {
		last := 0

		for _, idx := range docLinkRegex.FindAllStringSubmatchIndex(line, -1) {
			// The first group is the reference name. e.g. "X", "p.T", "p.T.M".
			name := line[idx[2]:idx[3]]
			if objs := lookupObjects(name); len(objs) > 0 {
				if last < idx[2] {
					tv.token(pos+token.Pos(last), idx[2]-last, semtok.TokComment)
				}
				offset := pos + token.Pos(idx[2])
				for i, obj := range objs {
					if i > 0 {
						tv.token(offset, len("."), semtok.TokComment)
						offset += token.Pos(len("."))
					}
					id, rest, _ := strings.Cut(name, ".")
					name = rest
					tok, mods := tv.appendObjectModifiers(nil, obj)
					tv.token(offset, len(id), tok, mods...)
					offset += token.Pos(len(id))
				}
				last = idx[3]
			}
		}
		if last != len(c.Text) {
			tv.token(pos+token.Pos(last), len(line)-last, semtok.TokComment)
		}
		pos += token.Pos(len(line) + 1)
	}
}

// token emits a token of the specified extent and semantics.
func (tv *tokenVisitor) token(start token.Pos, length int, typ semtok.Type, modifiers ...semtok.Modifier) {
	if !start.IsValid() {
		return
	}
	if length <= 0 {
		return // vscode doesn't like 0-length Tokens
	}
	end := start + token.Pos(length)
	if start >= tv.end || end <= tv.start {
		return
	}
	// want a line and column from start (in LSP coordinates). Ignore line directives.
	rng, err := tv.pgf.PosRange(start, end)
	if err != nil {
		event.Error(tv.ctx, "failed to convert to range", err)
		return
	}
	if rng.End.Line != rng.Start.Line {
		// this happens if users are typing at the end of the file, but report nothing
		return
	}
	tv.tokens = append(tv.tokens, semtok.Token{
		Line:      rng.Start.Line,
		Start:     rng.Start.Character,
		Len:       rng.End.Character - rng.Start.Character, // (on same line)
		Type:      typ,
		Modifiers: modifiers,
	})
}

// strStack converts the stack to a string, for debugging and error messages.
func (tv *tokenVisitor) strStack() string {
	msg := []string{"["}
	for i := len(tv.stack) - 1; i >= 0; i-- {
		n := tv.stack[i]
		msg = append(msg, strings.TrimPrefix(fmt.Sprintf("%T", n), "*ast."))
	}
	if len(tv.stack) > 0 {
		pos := tv.stack[len(tv.stack)-1].Pos()
		if _, err := safetoken.Offset(tv.pgf.Tok, pos); err != nil {
			msg = append(msg, fmt.Sprintf("invalid position %v for %s", pos, tv.pgf.URI))
		} else {
			posn := safetoken.Position(tv.pgf.Tok, pos)
			msg = append(msg, fmt.Sprintf("(%s:%d,col:%d)",
				filepath.Base(posn.Filename), posn.Line, posn.Column))
		}
	}
	msg = append(msg, "]")
	return strings.Join(msg, " ")
}

// srcLine returns the source text for n (truncated at first newline).
func (tv *tokenVisitor) srcLine(n ast.Node) string {
	file := tv.pgf.Tok
	line := safetoken.Line(file, n.Pos())
	start, err := safetoken.Offset(file, file.LineStart(line))
	if err != nil {
		return ""
	}
	end := start
	for ; end < len(tv.pgf.Src) && tv.pgf.Src[end] != '\n'; end++ {

	}
	return string(tv.pgf.Src[start:end])
}

func (tv *tokenVisitor) inspect(n ast.Node) (descend bool) {
	if n == nil {
		tv.stack = tv.stack[:len(tv.stack)-1] // pop
		return true
	}
	tv.stack = append(tv.stack, n) // push
	defer func() {
		if !descend {
			tv.stack = tv.stack[:len(tv.stack)-1] // pop
		}
	}()

	switch n := n.(type) {
	case *ast.ArrayType:
	case *ast.AssignStmt:
		tv.token(n.TokPos, len(n.Tok.String()), semtok.TokOperator)
	case *ast.BasicLit:
		if n.Kind == token.STRING {
			if strings.Contains(n.Value, "\n") {
				// has to be a string.
				tv.multiline(n.Pos(), n.End(), semtok.TokString)
			} else if !tv.formatString(n) {
				// not a format string, color the whole as a TokString.
				tv.token(n.Pos(), len(n.Value), semtok.TokString)
			}
		} else {
			tv.token(n.Pos(), len(n.Value), semtok.TokNumber)
		}
	case *ast.BinaryExpr:
		tv.token(n.OpPos, len(n.Op.String()), semtok.TokOperator)
	case *ast.BlockStmt:
	case *ast.BranchStmt:
		tv.token(n.TokPos, len(n.Tok.String()), semtok.TokKeyword)
	case *ast.CallExpr:
		if n.Ellipsis.IsValid() {
			tv.token(n.Ellipsis, len("..."), semtok.TokOperator)
		}
	case *ast.CaseClause:
		iam := "case"
		if n.List == nil {
			iam = "default"
		}
		tv.token(n.Case, len(iam), semtok.TokKeyword)
	case *ast.ChanType:
		// chan | chan <- | <- chan
		switch {
		case n.Arrow == token.NoPos:
			tv.token(n.Begin, len("chan"), semtok.TokKeyword)
		case n.Arrow == n.Begin:
			tv.token(n.Arrow, 2, semtok.TokOperator)
			pos := tv.findKeyword("chan", n.Begin+2, n.Value.Pos())
			tv.token(pos, len("chan"), semtok.TokKeyword)
		case n.Arrow != n.Begin:
			tv.token(n.Begin, len("chan"), semtok.TokKeyword)
			tv.token(n.Arrow, 2, semtok.TokOperator)
		}
	case *ast.CommClause:
		length := len("case")
		if n.Comm == nil {
			length = len("default")
		}
		tv.token(n.Case, length, semtok.TokKeyword)
	case *ast.CompositeLit:
	case *ast.DeclStmt:
	case *ast.DeferStmt:
		tv.token(n.Defer, len("defer"), semtok.TokKeyword)
	case *ast.Ellipsis:
		tv.token(n.Ellipsis, len("..."), semtok.TokOperator)
	case *ast.EmptyStmt:
	case *ast.ExprStmt:
	case *ast.Field:
	case *ast.FieldList:
	case *ast.ForStmt:
		tv.token(n.For, len("for"), semtok.TokKeyword)
	case *ast.FuncDecl:
	case *ast.FuncLit:
	case *ast.FuncType:
		if n.Func != token.NoPos {
			tv.token(n.Func, len("func"), semtok.TokKeyword)
		}
	case *ast.GenDecl:
		tv.token(n.TokPos, len(n.Tok.String()), semtok.TokKeyword)
	case *ast.GoStmt:
		tv.token(n.Go, len("go"), semtok.TokKeyword)
	case *ast.Ident:
		tv.ident(n)
	case *ast.IfStmt:
		tv.token(n.If, len("if"), semtok.TokKeyword)
		if n.Else != nil {
			// x.Body.End() or x.Body.End()+1, not that it matters
			pos := tv.findKeyword("else", n.Body.End(), n.Else.Pos())
			tv.token(pos, len("else"), semtok.TokKeyword)
		}
	case *ast.ImportSpec:
		tv.importSpec(n)
		return false
	case *ast.IncDecStmt:
		tv.token(n.TokPos, len(n.Tok.String()), semtok.TokOperator)
	case *ast.IndexExpr:
	case *ast.IndexListExpr:
	case *ast.InterfaceType:
		tv.token(n.Interface, len("interface"), semtok.TokKeyword)
	case *ast.KeyValueExpr:
	case *ast.LabeledStmt:
	case *ast.MapType:
		tv.token(n.Map, len("map"), semtok.TokKeyword)
	case *ast.ParenExpr:
	case *ast.RangeStmt:
		tv.token(n.For, len("for"), semtok.TokKeyword)
		// x.TokPos == token.NoPos is legal (for range foo {})
		offset := n.TokPos
		if offset == token.NoPos {
			offset = n.For
		}
		pos := tv.findKeyword("range", offset, n.X.Pos())
		tv.token(pos, len("range"), semtok.TokKeyword)
	case *ast.ReturnStmt:
		tv.token(n.Return, len("return"), semtok.TokKeyword)
	case *ast.SelectStmt:
		tv.token(n.Select, len("select"), semtok.TokKeyword)
	case *ast.SelectorExpr:
	case *ast.SendStmt:
		tv.token(n.Arrow, len("<-"), semtok.TokOperator)
	case *ast.SliceExpr:
	case *ast.StarExpr:
		tv.token(n.Star, len("*"), semtok.TokOperator)
	case *ast.StructType:
		tv.token(n.Struct, len("struct"), semtok.TokKeyword)
	case *ast.SwitchStmt:
		tv.token(n.Switch, len("switch"), semtok.TokKeyword)
	case *ast.TypeAssertExpr:
		if n.Type == nil {
			pos := tv.findKeyword("type", n.Lparen, n.Rparen)
			tv.token(pos, len("type"), semtok.TokKeyword)
		}
	case *ast.TypeSpec:
	case *ast.TypeSwitchStmt:
		tv.token(n.Switch, len("switch"), semtok.TokKeyword)
	case *ast.UnaryExpr:
		tv.token(n.OpPos, len(n.Op.String()), semtok.TokOperator)
	case *ast.ValueSpec:
	// things only seen with parsing or type errors, so ignore them
	case *ast.BadDecl, *ast.BadExpr, *ast.BadStmt:
		return false
	// not going to see these
	case *ast.File, *ast.Package:
		tv.errorf("implement %T %s", n, safetoken.Position(tv.pgf.Tok, n.Pos()))
	// other things we knowingly ignore
	case *ast.Comment, *ast.CommentGroup:
		return false
	default:
		tv.errorf("failed to implement %T", n)
	}
	return true
}

// formatString tries to report directives and string literals
// inside a (possible) printf-like call, it returns false and does nothing
// if the string is not a format string.
func (tv *tokenVisitor) formatString(lit *ast.BasicLit) bool {
	if len(tv.stack) <= 1 {
		return false
	}
	call, ok := tv.stack[len(tv.stack)-2].(*ast.CallExpr)
	if !ok {
		return false
	}
	lastNonVariadic, idx := formatStringAndIndex(tv.info, call)
	if idx == -1 || lit != lastNonVariadic {
		return false
	}
	format, err := strconv.Unquote(lit.Value)
	if err != nil {
		return false
	}
	if !strings.Contains(format, "%") {
		return false
	}
	operations, err := fmtstr.Parse(format, idx)
	if err != nil {
		return false
	}

	// It's a format string, compute interleaved sub range of directives and literals.
	// pos tracks literal substring position within the overall BasicLit.
	pos := lit.ValuePos
	for _, op := range operations {
		// Skip "%%".
		if op.Verb.Verb == '%' {
			continue
		}
		rangeStart, rangeEnd, err := astutil.RangeInStringLiteral(lit, op.Range.Start, op.Range.End)
		if err != nil {
			return false
		}
		// Report literal substring.
		tv.token(pos, int(rangeStart-pos), semtok.TokString)
		// Report formatting directive.
		tv.token(rangeStart, int(rangeEnd-rangeStart), semtok.TokString, semtok.ModFormat)
		pos = rangeEnd
	}
	// Report remaining literal substring.
	tv.token(pos, int(lit.End()-pos), semtok.TokString)
	return true
}

func (tv *tokenVisitor) appendObjectModifiers(mods []semtok.Modifier, obj types.Object) (semtok.Type, []semtok.Modifier) {
	if obj.Pkg() == nil {
		mods = append(mods, semtok.ModDefaultLibrary)
	}

	// Note: PkgName, Builtin, Label have type Invalid, which adds no modifiers.
	mods = appendTypeModifiers(mods, obj.Type())

	switch obj := obj.(type) {
	case *types.PkgName:
		return semtok.TokNamespace, mods

	case *types.Builtin:
		return semtok.TokFunction, mods

	case *types.Func:
		if obj.Signature().Recv() != nil {
			return semtok.TokMethod, mods
		} else {
			return semtok.TokFunction, mods
		}

	case *types.TypeName:
		if is[*types.TypeParam](types.Unalias(obj.Type())) {
			return semtok.TokTypeParam, mods
		}
		return semtok.TokType, mods

	case *types.Const:
		mods = append(mods, semtok.ModReadonly)
		return semtok.TokVariable, mods

	case *types.Var:
		if tv.isParam(obj.Pos()) {
			return semtok.TokParameter, mods
		} else {
			return semtok.TokVariable, mods
		}

	case *types.Label:
		return semtok.TokLabel, mods

	case *types.Nil:
		mods = append(mods, semtok.ModReadonly)
		return semtok.TokVariable, mods
	}

	panic(obj)
}

// appendTypeModifiers appends optional modifiers that describe the top-level
// type constructor of t: "pointer", "map", etc.
func appendTypeModifiers(mods []semtok.Modifier, t types.Type) []semtok.Modifier {
	// For a type parameter, don't report "interface".
	if is[*types.TypeParam](types.Unalias(t)) {
		return mods
	}

	switch t := t.Underlying().(type) {
	case *types.Interface:
		mods = append(mods, semtok.ModInterface)
	case *types.Struct:
		mods = append(mods, semtok.ModStruct)
	case *types.Signature:
		mods = append(mods, semtok.ModSignature)
	case *types.Pointer:
		mods = append(mods, semtok.ModPointer)
	case *types.Array:
		mods = append(mods, semtok.ModArray)
	case *types.Map:
		mods = append(mods, semtok.ModMap)
	case *types.Slice:
		mods = append(mods, semtok.ModSlice)
	case *types.Chan:
		mods = append(mods, semtok.ModChan)
	case *types.Basic:
		switch t.Kind() {
		case types.Invalid:
			// ignore (e.g. Builtin, PkgName, Label)
		case types.String:
			mods = append(mods, semtok.ModString)
		case types.Bool:
			mods = append(mods, semtok.ModBool)
		case types.UnsafePointer:
			mods = append(mods, semtok.ModPointer)
		default:
			if t.Info()&types.IsNumeric != 0 {
				mods = append(mods, semtok.ModNumber)
			}
		}
	}
	return mods
}

func (tv *tokenVisitor) ident(id *ast.Ident) {
	var (
		tok  semtok.Type
		mods []semtok.Modifier
		obj  types.Object
		ok   bool
	)
	if obj, _ = tv.info.Defs[id]; obj != nil {
		// definition
		mods = append(mods, semtok.ModDefinition)
		tok, mods = tv.appendObjectModifiers(mods, obj)

	} else if obj, ok = tv.info.Uses[id]; ok {
		// use
		tok, mods = tv.appendObjectModifiers(mods, obj)

	} else if tok, mods = tv.unkIdent(id); tok != "" {
		// ok

	} else {
		return
	}

	// Emit a token for the identifier's extent.
	tv.token(id.Pos(), len(id.Name), tok, mods...)

	if semDebug {
		q := "nil"
		if obj != nil {
			q = fmt.Sprintf("%T", obj.Type()) // e.g. "*types.Map"
		}
		log.Printf(" use %s/%T/%s got %s %v (%s)",
			id.Name, obj, q, tok, mods, tv.strStack())
	}
}

// isParam reports whether the position is that of a parameter name of
// an enclosing function.
func (tv *tokenVisitor) isParam(pos token.Pos) bool {
	for i := len(tv.stack) - 1; i >= 0; i-- {
		switch n := tv.stack[i].(type) {
		case *ast.FuncDecl:
			for _, f := range n.Type.Params.List {
				for _, id := range f.Names {
					if id.Pos() == pos {
						return true
					}
				}
			}
		case *ast.FuncLit:
			for _, f := range n.Type.Params.List {
				for _, id := range f.Names {
					if id.Pos() == pos {
						return true
					}
				}
			}
		}
	}
	return false
}

// unkIdent handles identifiers with no types.Object (neither use nor
// def), use the parse stack.
// A lot of these only happen when the package doesn't compile,
// but in that case it is all best-effort from the parse tree.
func (tv *tokenVisitor) unkIdent(id *ast.Ident) (semtok.Type, []semtok.Modifier) {
	def := []semtok.Modifier{semtok.ModDefinition}
	n := len(tv.stack) - 2 // parent of Ident; stack is [File ... Ident]
	if n < 0 {
		tv.errorf("no stack") // can't happen
		return "", nil
	}
	switch parent := tv.stack[n].(type) {
	case *ast.BinaryExpr, *ast.UnaryExpr, *ast.ParenExpr, *ast.StarExpr,
		*ast.IncDecStmt, *ast.SliceExpr, *ast.ExprStmt, *ast.IndexExpr,
		*ast.ReturnStmt, *ast.ChanType, *ast.SendStmt,
		*ast.ForStmt,      // possibly incomplete
		*ast.IfStmt,       /* condition */
		*ast.KeyValueExpr, // either key or value
		*ast.IndexListExpr:
		return semtok.TokVariable, nil
	case *ast.Ellipsis:
		return semtok.TokType, nil
	case *ast.CaseClause:
		if n-2 >= 0 && is[ast.TypeSwitchStmt](tv.stack[n-2]) {
			return semtok.TokType, nil
		}
		return semtok.TokVariable, nil
	case *ast.ArrayType:
		if id == parent.Len {
			// or maybe a Type Param, but we can't just from the parse tree
			return semtok.TokVariable, nil
		} else {
			return semtok.TokType, nil
		}
	case *ast.MapType:
		return semtok.TokType, nil
	case *ast.CallExpr:
		if id == parent.Fun {
			return semtok.TokFunction, nil
		}
		return semtok.TokVariable, nil
	case *ast.SwitchStmt:
		return semtok.TokVariable, nil
	case *ast.TypeAssertExpr:
		if id == parent.X {
			return semtok.TokVariable, nil
		} else if id == parent.Type {
			return semtok.TokType, nil
		}
	case *ast.ValueSpec:
		if slices.Contains(parent.Names, id) {
			return semtok.TokVariable, def
		}
		for _, p := range parent.Values {
			if p == id {
				return semtok.TokVariable, nil
			}
		}
		return semtok.TokType, nil
	case *ast.SelectorExpr: // e.ti.Selections[nd] is nil, so no help
		if n-1 >= 0 {
			if ce, ok := tv.stack[n-1].(*ast.CallExpr); ok {
				// ... CallExpr SelectorExpr Ident (_.x())
				if ce.Fun == parent && parent.Sel == id {
					return semtok.TokFunction, nil
				}
			}
		}
		return semtok.TokVariable, nil
	case *ast.AssignStmt:
		for _, p := range parent.Lhs {
			// x := ..., or x = ...
			if p == id {
				if parent.Tok != token.DEFINE {
					def = nil
				}
				return semtok.TokVariable, def // '_' in _ = ...
			}
		}
		// RHS, = x
		return semtok.TokVariable, nil
	case *ast.TypeSpec: // it's a type if it is either the Name or the Type
		if id == parent.Type {
			def = nil
		}
		return semtok.TokType, def
	case *ast.Field:
		// ident could be type in a field, or a method in an interface type, or a variable
		if id == parent.Type {
			return semtok.TokType, nil
		}
		if n > 2 &&
			is[*ast.InterfaceType](tv.stack[n-2]) &&
			is[*ast.FieldList](tv.stack[n-1]) {

			return semtok.TokMethod, def
		}
		return semtok.TokVariable, nil
	case *ast.LabeledStmt:
		if id == parent.Label {
			return semtok.TokLabel, def
		}
	case *ast.BranchStmt:
		if id == parent.Label {
			return semtok.TokLabel, nil
		}
	case *ast.CompositeLit:
		if parent.Type == id {
			return semtok.TokType, nil
		}
		return semtok.TokVariable, nil
	case *ast.RangeStmt:
		if parent.Tok != token.DEFINE {
			def = nil
		}
		return semtok.TokVariable, def
	case *ast.FuncDecl:
		return semtok.TokFunction, def
	default:
		tv.errorf("%T unexpected: %s %s%q", parent, id.Name, tv.strStack(), tv.srcLine(id))
	}
	return "", nil
}

// multiline emits a multiline token (`string` or /*comment*/).
func (tv *tokenVisitor) multiline(start, end token.Pos, tok semtok.Type) {
	// TODO(adonovan): test with non-ASCII.

	f := tv.fset.File(start)
	// the hard part is finding the lengths of lines. include the \n
	length := func(line int) int {
		n := f.LineStart(line)
		if line >= f.LineCount() {
			return f.Size() - int(n)
		}
		return int(f.LineStart(line+1) - n)
	}
	spos := safetoken.StartPosition(tv.fset, start)
	epos := safetoken.EndPosition(tv.fset, end)
	sline := spos.Line
	eline := epos.Line
	// first line is from spos.Column to end
	tv.token(start, length(sline)-spos.Column, tok) // leng(sline)-1 - (spos.Column-1)
	for i := sline + 1; i < eline; i++ {
		// intermediate lines are from 1 to end
		tv.token(f.LineStart(i), length(i)-1, tok) // avoid the newline
	}
	// last line is from 1 to epos.Column
	tv.token(f.LineStart(eline), epos.Column-1, tok) // columns are 1-based
}

// findKeyword returns the position of a keyword by searching within
// the specified range, for when it cannot be exactly known from the AST.
// It returns NoPos if the keyword was not present in the source due to parse error.
func (tv *tokenVisitor) findKeyword(keyword string, start, end token.Pos) token.Pos {
	// TODO(adonovan): use safetoken.Offset.
	offset := int(start) - tv.pgf.Tok.Base()
	last := int(end) - tv.pgf.Tok.Base()
	buf := tv.pgf.Src
	idx := bytes.Index(buf[offset:last], []byte(keyword))
	if idx < 0 {
		// Ill-formed code may form syntax trees without their usual tokens.
		// For example, "type _ <-<-chan int" parses as <-chan (chan int),
		// with two nested ChanTypes but only one chan keyword.
		return token.NoPos
	}
	return start + token.Pos(idx)
}

func (tv *tokenVisitor) importSpec(spec *ast.ImportSpec) {
	// a local package name or the last component of the Path
	if spec.Name != nil {
		name := spec.Name.String()
		if name != "_" && name != "." {
			tv.token(spec.Name.Pos(), len(name), semtok.TokNamespace)
		}
		return // don't mark anything for . or _
	}
	importPath := metadata.UnquoteImportPath(spec)
	if importPath == "" {
		return
	}
	// Import strings are implementation defined. Try to match with parse information.
	depID := tv.metadata.DepsByImpPath[importPath]
	if depID == "" {
		return
	}
	depMD := tv.metadataSource.Metadata(depID)
	if depMD == nil {
		// unexpected, but impact is that maybe some import is not colored
		return
	}
	// Check whether the original literal contains the package's declared name.
	j := strings.LastIndex(spec.Path.Value, string(depMD.Name))
	if j < 0 {
		// Package name does not match import path, so there is nothing to report.
		return
	}
	// Report virtual declaration at the position of the substring.
	start := spec.Path.Pos() + token.Pos(j)
	tv.token(start, len(depMD.Name), semtok.TokNamespace)
}

// errorf logs an error and reports a bug.
func (tv *tokenVisitor) errorf(format string, args ...any) {
	msg := fmt.Sprintf(format, args...)
	bug.Report(msg)
	event.Error(tv.ctx, tv.strStack(), errors.New(msg))
}

var godirectives = map[string]struct{}{
	// https://pkg.go.dev/cmd/compile
	"noescape":       {},
	"uintptrescapes": {},
	"noinline":       {},
	"norace":         {},
	"nosplit":        {},
	"linkname":       {},

	// https://pkg.go.dev/go/build
	"build":               {},
	"binary-only-package": {},
	"embed":               {},
}

// Tokenize godirective at the start of the comment c, if any, and the surrounding comment.
// If there is any failure, emits the entire comment as a TokComment token.
// Directives are highlighted as-is, even if used incorrectly. Typically there are
// dedicated analyzers that will warn about misuse.
func (tv *tokenVisitor) godirective(c *ast.Comment) {
	// First check if '//go:directive args...' is a valid directive.
	directive, args, _ := strings.Cut(c.Text, " ")
	kind, _ := stringsCutPrefix(directive, "//go:")
	if _, ok := godirectives[kind]; !ok {
		// Unknown 'go:' directive.
		tv.token(c.Pos(), len(c.Text), semtok.TokComment)
		return
	}

	// Make the 'go:directive' part stand out, the rest is comments.
	tv.token(c.Pos(), len("//"), semtok.TokComment)

	directiveStart := c.Pos() + token.Pos(len("//"))
	tv.token(directiveStart, len(directive[len("//"):]), semtok.TokNamespace)

	if len(args) > 0 {
		tailStart := c.Pos() + token.Pos(len(directive)+len(" "))
		tv.token(tailStart, len(args), semtok.TokComment)
	}
}

// Go 1.20 strings.CutPrefix.
func stringsCutPrefix(s, prefix string) (after string, found bool) {
	if !strings.HasPrefix(s, prefix) {
		return s, false
	}
	return s[len(prefix):], true
}

func is[T any](x any) bool {
	_, ok := x.(T)
	return ok
}