generator.go

// Copyright 2014 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.

// +build go1.5

// Stringer is a tool to automate the creation of methods that satisfy the fmt.Stringer
// interface. Given the name of a (signed or unsigned) integer type T that has constants
// defined, stringer will create a new self-contained Go source file implementing
//	func (t T) String() string
// The file is created in the same package and directory as the package that defines T.
// It has helpful defaults designed for use with go generate.
//
// Stringer works best with constants that are consecutive values such as created using iota,
// but creates good code regardless. In the future it might also provide custom support for
// constant sets that are bit patterns.
//
// For example, given this snippet,
//
//	package painkiller
//
//	type Pill int
//
//	const (
//		Placebo Pill = iota
//		Aspirin
//		Ibuprofen
//		Paracetamol
//		Acetaminophen = Paracetamol
//	)
//
// running this command
//
//	stringer -type=Pill
//
// in the same directory will create the file pill_string.go, in package painkiller,
// containing a definition of
//
//	func (Pill) String() string
//
// That method will translate the value of a Pill constant to the string representation
// of the respective constant name, so that the call fmt.Print(painkiller.Aspirin) will
// print the string "Aspirin".
//
// Typically this process would be run using go generate, like this:
//
//	//go:generate stringer -type=Pill
//
// If multiple constants have the same value, the lexically first matching name will
// be used (in the example, Acetaminophen will print as "Paracetamol").
//
// With no arguments, it processes the package in the current directory.
// Otherwise, the arguments must name a single directory holding a Go package
// or a set of Go source files that represent a single Go package.
//
// The -type flag accepts a comma-separated list of types so a single run can
// generate methods for multiple types. The default output file is t_string.go,
// where t is the lower-cased name of the first type listed. It can be overridden
// with the -output flag.
//
package main // import "github.com/twinj/opal/generator"

import (
	"bytes"
	"flag"
	"fmt"
	"go/ast"
	"go/build"
	exact "go/constant"
	"go/format"
	"go/importer"
	"go/parser"
	"go/token"
	"go/types"
	"io/ioutil"
	"log"
	"os"
	"path/filepath"
	"sort"
	"strings"
)

var (
	typeNames = flag.String("type", "", "comma-separated list of type names; must be set")
	output = flag.String("output", "", "output file name; default srcdir/<type>_string.go")
)

// Usage is a replacement usage function for the flags package.
func Usage() {
	fmt.Fprintf(os.Stderr, "Usage of %s:\n", os.Args[0])
	fmt.Fprintf(os.Stderr, "\toyster [flags] -type T [directory]\n")
	fmt.Fprintf(os.Stderr, "\toyster [flags] -type T files... # Must be a single package\n")
	fmt.Fprintf(os.Stderr, "For more information, see:\n")
	fmt.Fprintf(os.Stderr, "\thttp://godoc.org/github.com/opal/oyster\n")
	fmt.Fprintf(os.Stderr, "Flags:\n")
	flag.PrintDefaults()
}

func main() {
	log.SetFlags(0)
	log.SetPrefix("oyster: ")
	flag.Usage = Usage
	flag.Parse()
	if len(*typeNames) == 0 {
		flag.Usage()
		os.Exit(2)
	}
	types := strings.Split(*typeNames, ",")

	// We accept either one directory or a list of files. Which do we have?
	args := flag.Args()
	if len(args) == 0 {
		// Default: process whole package in current directory.
		args = []string{"."}
	}

	// Parse the package once.
	var (
		dir string
		g Generator
	)
	if len(args) == 1 && isDirectory(args[0]) {
		dir = args[0]
		g.parsePackageDir(args[0])
	} else {
		dir = filepath.Dir(args[0])
		g.parsePackageFiles(args)
	}

	// Print the header and package clause.
	g.Printf("// Code generated by \"oyster %s\"; DO NOT EDIT\n", strings.Join(os.Args[1:], " "))
	g.Printf("\n")
	g.Printf("package %s", g.pkg.name)
	g.Printf("\n")
	g.Printf("import \"fmt\"\n") // Used by all methods.

	// Run generate for each type.
	for _, typeName := range types {
		g.generate(typeName)
	}

	// Format the output.
	src := g.format()

	// Write to file.
	outputName := *output
	if outputName == "" {
		baseName := fmt.Sprintf("%s_string.go", types[0])
		outputName = filepath.Join(dir, strings.ToLower(baseName))
	}
	err := ioutil.WriteFile(outputName, src, 0644)
	if err != nil {
		log.Fatalf("writing output: %s", err)
	}
}

// isDirectory reports whether the named file is a directory.
func isDirectory(name string) bool {
	info, err := os.Stat(name)
	if err != nil {
		log.Fatal(err)
	}
	return info.IsDir()
}

// Generator holds the state of the analysis. Primarily used to buffer
// the output for format.Source.
type Generator struct {
	buf bytes.Buffer // Accumulated output.
	pkg *Package     // Package we are scanning.
}

func (g *Generator) Printf(format string, args ...interface{}) {
	fmt.Fprintf(&g.buf, format, args...)
}

// File holds a single parsed file and associated data.
type File struct {
	pkg      *Package  // Package to which this file belongs.
	file     *ast.File // Parsed AST.
			   // These fields are reset for each type being generated.
	typeName string    // Name of the constant type.
	values   []Value   // Accumulator for constant values of that type.
}

type Package struct {
	dir      string
	name     string
	defs     map[*ast.Ident]types.Object
	files    []*File
	typesPkg *types.Package
}

// parsePackageDir parses the package residing in the directory.
func (g *Generator) parsePackageDir(directory string) {
	pkg, err := build.Default.ImportDir(directory, 0)
	if err != nil {
		log.Fatalf("cannot process directory %s: %s", directory, err)
	}
	var names []string
	names = append(names, pkg.GoFiles...)
	names = append(names, pkg.CgoFiles...)
	// TODO: Need to think about constants in test files. Maybe write type_string_test.go
	// in a separate pass? For later.
	// names = append(names, pkg.TestGoFiles...) // These are also in the "foo" package.
	names = append(names, pkg.SFiles...)
	names = prefixDirectory(directory, names)
	g.parsePackage(directory, names, nil)
}

// parsePackageFiles parses the package occupying the named files.
func (g *Generator) parsePackageFiles(names []string) {
	g.parsePackage(".", names, nil)
}

// prefixDirectory places the directory name on the beginning of each name in the list.
func prefixDirectory(directory string, names []string) []string {
	if directory == "." {
		return names
	}
	ret := make([]string, len(names))
	for i, name := range names {
		ret[i] = filepath.Join(directory, name)
	}
	return ret
}

// parsePackage analyzes the single package constructed from the named files.
// If text is non-nil, it is a string to be used instead of the content of the file,
// to be used for testing. parsePackage exits if there is an error.
func (g *Generator) parsePackage(directory string, names []string, text interface{}) {
	var files []*File
	var astFiles []*ast.File
	g.pkg = new(Package)
	fs := token.NewFileSet()
	for _, name := range names {
		if !strings.HasSuffix(name, ".go") {
			continue
		}
		parsedFile, err := parser.ParseFile(fs, name, text, 0)
		if err != nil {
			log.Fatalf("parsing package: %s: %s", name, err)
		}
		astFiles = append(astFiles, parsedFile)
		files = append(files, &File{
			file: parsedFile,
			pkg:  g.pkg,
		})
	}
	if len(astFiles) == 0 {
		log.Fatalf("%s: no buildable Go files", directory)
	}
	g.pkg.name = astFiles[0].Name.Name
	g.pkg.files = files
	g.pkg.dir = directory
	// Type check the package.
	g.pkg.check(fs, astFiles)
}

// check type-checks the package. The package must be OK to proceed.
func (pkg *Package) check(fs *token.FileSet, astFiles []*ast.File) {
	pkg.defs = make(map[*ast.Ident]types.Object)
	config := types.Config{Importer: importer.Default(), FakeImportC: true}
	info := &types.Info{
		Defs: pkg.defs,
	}
	typesPkg, err := config.Check(pkg.dir, fs, astFiles, info)
	if err != nil {
		log.Fatalf("checking package: %s", err)
	}
	pkg.typesPkg = typesPkg
}

// generate produces the String method for the named type.
func (g *Generator) generate(typeName string) {
	values := make([]Value, 0, 100)
	for _, file := range g.pkg.files {
		// Set the state for this run of the walker.
		file.typeName = typeName
		file.values = nil
		if file.file != nil {
			ast.Inspect(file.file, file.genDecl)
			values = append(values, file.values...)
		}
	}

	if len(values) == 0 {
		log.Fatalf("no values defined for type %s", typeName)
	}
	runs := splitIntoRuns(values)
	// The decision of which pattern to use depends on the number of
	// runs in the numbers. If there's only one, it's easy. For more than
	// one, there's a tradeoff between complexity and size of the data
	// and code vs. the simplicity of a map. A map takes more space,
	// but so does the code. The decision here (crossover at 10) is
	// arbitrary, but considers that for large numbers of runs the cost
	// of the linear scan in the switch might become important, and
	// rather than use yet another algorithm such as binary search,
	// we punt and use a map. In any case, the likelihood of a map
	// being necessary for any realistic example other than bitmasks
	// is very low. And bitmasks probably deserve their own analysis,
	// to be done some other day.
	switch {
	case len(runs) == 1:
		g.buildOneRun(runs, typeName)
	case len(runs) <= 10:
		g.buildMultipleRuns(runs, typeName)
	default:
		g.buildMap(runs, typeName)
	}
}

// splitIntoRuns breaks the values into runs of contiguous sequences.
// For example, given 1,2,3,5,6,7 it returns {1,2,3},{5,6,7}.
// The input slice is known to be non-empty.
func splitIntoRuns(values []Value) [][]Value {
	// We use stable sort so the lexically first name is chosen for equal elements.
	sort.Stable(byValue(values))
	// Remove duplicates. Stable sort has put the one we want to print first,
	// so use that one. The String method won't care about which named constant
	// was the argument, so the first name for the given value is the only one to keep.
	// We need to do this because identical values would cause the switch or map
	// to fail to compile.
	j := 1
	for i := 1; i < len(values); i++ {
		if values[i].value != values[i - 1].value {
			values[j] = values[i]
			j++
		}
	}
	values = values[:j]
	runs := make([][]Value, 0, 10)
	for len(values) > 0 {
		// One contiguous sequence per outer loop.
		i := 1
		for i < len(values) && values[i].value == values[i - 1].value + 1 {
			i++
		}
		runs = append(runs, values[:i])
		values = values[i:]
	}
	return runs
}

// format returns the gofmt-ed contents of the Generator's buffer.
func (g *Generator) format() []byte {
	src, err := format.Source(g.buf.Bytes())
	if err != nil {
		// Should never happen, but can arise when developing this code.
		// The user can compile the output to see the error.
		log.Printf("warning: internal error: invalid Go generated: %s", err)
		log.Printf("warning: compile the package to analyze the error")
		return g.buf.Bytes()
	}
	return src
}

// Value represents a declared constant.
type Value struct {
	name   string // The name of the constant.
		      // The value is stored as a bit pattern alone. The boolean tells us
		      // whether to interpret it as an int64 or a uint64; the only place
		      // this matters is when sorting.
		      // Much of the time the str field is all we need; it is printed
		      // by Value.String.
	value  uint64 // Will be converted to int64 when needed.
	signed bool   // Whether the constant is a signed type.
	str    string // The string representation given by the "go/exact" package.
}

func (v *Value) String() string {
	return v.str
}

// byValue lets us sort the constants into increasing order.
// We take care in the Less method to sort in signed or unsigned order,
// as appropriate.
type byValue []Value

func (b byValue) Len() int {
	return len(b)
}
func (b byValue) Swap(i, j int) {
	b[i], b[j] = b[j], b[i]
}
func (b byValue) Less(i, j int) bool {
	if b[i].signed {
		return int64(b[i].value) < int64(b[j].value)
	}
	return b[i].value < b[j].value
}

// genDecl processes one declaration clause.
func (f *File) genDecl(node ast.Node) bool {
	decl, ok := node.(*ast.GenDecl)
	if !ok || decl.Tok != token.CONST {
		// We only care about const declarations.
		return true
	}
	// The name of the type of the constants we are declaring.
	// Can change if this is a multi-element declaration.
	typ := ""
	// Loop over the elements of the declaration. Each element is a ValueSpec:
	// a list of names possibly followed by a type, possibly followed by values.
	// If the type and value are both missing, we carry down the type (and value,
	// but the "go/types" package takes care of that).
	for _, spec := range decl.Specs {
		vspec := spec.(*ast.ValueSpec) // Guaranteed to succeed as this is CONST.
		if vspec.Type == nil && len(vspec.Values) > 0 {
			// "X = 1". With no type but a value, the constant is untyped.
			// Skip this vspec and reset the remembered type.
			typ = ""
			continue
		}
		if vspec.Type != nil {
			// "X T". We have a type. Remember it.
			ident, ok := vspec.Type.(*ast.Ident)
			if !ok {
				continue
			}
			typ = ident.Name
		}
		if typ != f.typeName {
			// This is not the type we're looking for.
			continue
		}
		// We now have a list of names (from one line of source code) all being
		// declared with the desired type.
		// Grab their names and actual values and store them in f.values.
		for _, name := range vspec.Names {
			if name.Name == "_" {
				continue
			}
			// This dance lets the type checker find the values for us. It's a
			// bit tricky: look up the object declared by the name, find its
			// types.Const, and extract its value.
			obj, ok := f.pkg.defs[name]
			if !ok {
				log.Fatalf("no value for constant %s", name)
			}
			info := obj.Type().Underlying().(*types.Basic).Info()
			if info & types.IsInteger == 0 {
				log.Fatalf("can't handle non-integer constant type %s", typ)
			}
			value := obj.(*types.Const).Val() // Guaranteed to succeed as this is CONST.
			if value.Kind() != exact.Int {
				log.Fatalf("can't happen: constant is not an integer %s", name)
			}
			i64, isInt := exact.Int64Val(value)
			u64, isUint := exact.Uint64Val(value)
			if !isInt && !isUint {
				log.Fatalf("internal error: value of %s is not an integer: %s", name, value.String())
			}
			if !isInt {
				u64 = uint64(i64)
			}
			v := Value{
				name:   name.Name,
				value:  u64,
				signed: info & types.IsUnsigned == 0,
				str:    value.String(),
			}
			f.values = append(f.values, v)
		}
	}
	return false
}

// Helpers

// usize returns the number of bits of the smallest unsigned integer
// type that will hold n. Used to create the smallest possible slice of
// integers to use as indexes into the concatenated strings.
func usize(n int) int {
	switch {
	case n < 1 << 8:
		return 8
	case n < 1 << 16:
		return 16
	default:
		// 2^32 is enough constants for anyone.
		return 32
	}
}

// declareIndexAndNameVars declares the index slices and concatenated names
// strings representing the runs of values.
func (g *Generator) declareIndexAndNameVars(runs [][]Value, typeName string) {
	var indexes, names []string
	for i, run := range runs {
		index, name := g.createIndexAndNameDecl(run, typeName, fmt.Sprintf("_%d", i))
		indexes = append(indexes, index)
		names = append(names, name)
	}
	g.Printf("const (\n")
	for _, name := range names {
		g.Printf("\t%s\n", name)
	}
	g.Printf(")\n\n")
	g.Printf("var (")
	for _, index := range indexes {
		g.Printf("\t%s\n", index)
	}
	g.Printf(")\n\n")
}

// declareIndexAndNameVar is the single-run version of declareIndexAndNameVars
func (g *Generator) declareIndexAndNameVar(run []Value, typeName string) {
	index, name := g.createIndexAndNameDecl(run, typeName, "")
	g.Printf("const %s\n", name)
	g.Printf("var %s\n", index)
}

// createIndexAndNameDecl returns the pair of declarations for the run. The caller will add "const" and "var".
func (g *Generator) createIndexAndNameDecl(run []Value, typeName string, suffix string) (string, string) {
	b := new(bytes.Buffer)
	indexes := make([]int, len(run))
	for i := range run {
		b.WriteString(run[i].name)
		indexes[i] = b.Len()
	}
	nameConst := fmt.Sprintf("_%s_name%s = %q", typeName, suffix, b.String())
	nameLen := b.Len()
	b.Reset()
	fmt.Fprintf(b, "_%s_index%s = [...]uint%d{0, ", typeName, suffix, usize(nameLen))
	for i, v := range indexes {
		if i > 0 {
			fmt.Fprintf(b, ", ")
		}
		fmt.Fprintf(b, "%d", v)
	}
	fmt.Fprintf(b, "}")
	return b.String(), nameConst
}

// declareNameVars declares the concatenated names string representing all the values in the runs.
func (g *Generator) declareNameVars(runs [][]Value, typeName string, suffix string) {
	g.Printf("const _%s_name%s = \"", typeName, suffix)
	for _, run := range runs {
		for i := range run {
			g.Printf("%s", run[i].name)
		}
	}
	g.Printf("\"\n")
}

// buildOneRun generates the variables and String method for a single run of contiguous values.
func (g *Generator) buildOneRun(runs [][]Value, typeName string) {
	values := runs[0]
	g.Printf("\n")
	g.declareIndexAndNameVar(values, typeName)
	// The generated code is simple enough to write as a Printf format.
	lessThanZero := ""
	if values[0].signed {
		lessThanZero = "i < 0 || "
	}
	if values[0].value == 0 {
		// Signed or unsigned, 0 is still 0.
		g.Printf(stringOneRun, typeName, usize(len(values)), lessThanZero)
	} else {
		g.Printf(stringOneRunWithOffset, typeName, values[0].String(), usize(len(values)), lessThanZero)
	}
}

// Arguments to format are:
//	[1]: type name
//	[2]: size of index element (8 for uint8 etc.)
//	[3]: less than zero check (for signed types)
const stringOneRun = `func (i %[1]s) String() string {
	if %[3]si >= %[1]s(len(_%[1]s_index)-1) {
		return fmt.Sprintf("%[1]s(%%d)", i)
	}
	return _%[1]s_name[_%[1]s_index[i]:_%[1]s_index[i+1]]
}
`

// Arguments to format are:
//	[1]: type name
//	[2]: lowest defined value for type, as a string
//	[3]: size of index element (8 for uint8 etc.)
//	[4]: less than zero check (for signed types)
/*
 */
const stringOneRunWithOffset = `func (i %[1]s) String() string {
	i -= %[2]s
	if %[4]si >= %[1]s(len(_%[1]s_index)-1) {
		return fmt.Sprintf("%[1]s(%%d)", i + %[2]s)
	}
	return _%[1]s_name[_%[1]s_index[i] : _%[1]s_index[i+1]]
}
`

// buildMultipleRuns generates the variables and String method for multiple runs of contiguous values.
// For this pattern, a single Printf format won't do.
func (g *Generator) buildMultipleRuns(runs [][]Value, typeName string) {
	g.Printf("\n")
	g.declareIndexAndNameVars(runs, typeName)
	g.Printf("func (i %s) String() string {\n", typeName)
	g.Printf("\tswitch {\n")
	for i, values := range runs {
		if len(values) == 1 {
			g.Printf("\tcase i == %s:\n", &values[0])
			g.Printf("\t\treturn _%s_name_%d\n", typeName, i)
			continue
		}
		g.Printf("\tcase %s <= i && i <= %s:\n", &values[0], &values[len(values) - 1])
		if values[0].value != 0 {
			g.Printf("\t\ti -= %s\n", &values[0])
		}
		g.Printf("\t\treturn _%s_name_%d[_%s_index_%d[i]:_%s_index_%d[i+1]]\n",
			typeName, i, typeName, i, typeName, i)
	}
	g.Printf("\tdefault:\n")
	g.Printf("\t\treturn fmt.Sprintf(\"%s(%%d)\", i)\n", typeName)
	g.Printf("\t}\n")
	g.Printf("}\n")
}

// buildMap handles the case where the space is so sparse a map is a reasonable fallback.
// It's a rare situation but has simple code.
func (g *Generator) buildMap(runs [][]Value, typeName string) {
	g.Printf("\n")
	g.declareNameVars(runs, typeName, "")
	g.Printf("\nvar _%s_map = map[%s]string{\n", typeName, typeName)
	n := 0
	for _, values := range runs {
		for _, value := range values {
			g.Printf("\t%s: _%s_name[%d:%d],\n", &value, typeName, n, n + len(value.name))
			n += len(value.name)
		}
	}
	g.Printf("}\n\n")
	g.Printf(stringMap, typeName)
}

// Argument to format is the type name.
const stringMap = `func (i %[1]s) String() string {
	if str, ok := _%[1]s_map[i]; ok {
		return str
	}
	return fmt.Sprintf("%[1]s(%%d)", i)
}
`