go/src/cmd/api/goapi.go

837 lines
20 KiB
Go
Raw Normal View History

// Copyright 2011 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.
// Binary api computes the exported API of a set of Go packages.
package main
import (
"bufio"
"bytes"
"flag"
"fmt"
"go/ast"
"go/build"
"go/parser"
"go/token"
"go/types"
"io"
"io/ioutil"
"log"
"os"
"os/exec"
"path/filepath"
"regexp"
"runtime"
"sort"
"strings"
)
// Flags
var (
checkFile = flag.String("c", "", "optional comma-separated filename(s) to check API against")
allowNew = flag.Bool("allow_new", true, "allow API additions")
exceptFile = flag.String("except", "", "optional filename of packages that are allowed to change without triggering a failure in the tool")
nextFile = flag.String("next", "", "optional filename of tentative upcoming API features for the next release. This file can be lazily maintained. It only affects the delta warnings from the -c file printed on success.")
verbose = flag.Bool("v", false, "verbose debugging")
forceCtx = flag.String("contexts", "", "optional comma-separated list of <goos>-<goarch>[-cgo] to override default contexts.")
)
// contexts are the default contexts which are scanned, unless
// overridden by the -contexts flag.
var contexts = []*build.Context{
{GOOS: "linux", GOARCH: "386", CgoEnabled: true},
{GOOS: "linux", GOARCH: "386"},
{GOOS: "linux", GOARCH: "amd64", CgoEnabled: true},
{GOOS: "linux", GOARCH: "amd64"},
{GOOS: "linux", GOARCH: "arm", CgoEnabled: true},
{GOOS: "linux", GOARCH: "arm"},
{GOOS: "darwin", GOARCH: "386", CgoEnabled: true},
{GOOS: "darwin", GOARCH: "386"},
{GOOS: "darwin", GOARCH: "amd64", CgoEnabled: true},
{GOOS: "darwin", GOARCH: "amd64"},
{GOOS: "windows", GOARCH: "amd64"},
{GOOS: "windows", GOARCH: "386"},
{GOOS: "freebsd", GOARCH: "386", CgoEnabled: true},
{GOOS: "freebsd", GOARCH: "386"},
{GOOS: "freebsd", GOARCH: "amd64", CgoEnabled: true},
{GOOS: "freebsd", GOARCH: "amd64"},
{GOOS: "freebsd", GOARCH: "arm", CgoEnabled: true},
{GOOS: "freebsd", GOARCH: "arm"},
{GOOS: "netbsd", GOARCH: "386", CgoEnabled: true},
{GOOS: "netbsd", GOARCH: "386"},
{GOOS: "netbsd", GOARCH: "amd64", CgoEnabled: true},
{GOOS: "netbsd", GOARCH: "amd64"},
{GOOS: "netbsd", GOARCH: "arm", CgoEnabled: true},
{GOOS: "netbsd", GOARCH: "arm"},
{GOOS: "openbsd", GOARCH: "386", CgoEnabled: true},
{GOOS: "openbsd", GOARCH: "386"},
{GOOS: "openbsd", GOARCH: "amd64", CgoEnabled: true},
{GOOS: "openbsd", GOARCH: "amd64"},
}
func contextName(c *build.Context) string {
s := c.GOOS + "-" + c.GOARCH
if c.CgoEnabled {
return s + "-cgo"
}
return s
}
func parseContext(c string) *build.Context {
parts := strings.Split(c, "-")
if len(parts) < 2 {
log.Fatalf("bad context: %q", c)
}
bc := &build.Context{
GOOS: parts[0],
GOARCH: parts[1],
}
if len(parts) == 3 {
if parts[2] == "cgo" {
bc.CgoEnabled = true
} else {
log.Fatalf("bad context: %q", c)
}
}
return bc
}
func setContexts() {
contexts = []*build.Context{}
for _, c := range strings.Split(*forceCtx, ",") {
contexts = append(contexts, parseContext(c))
}
}
var internalPkg = regexp.MustCompile(`(^|/)internal($|/)`)
func main() {
flag.Parse()
if !strings.Contains(runtime.Version(), "weekly") && !strings.Contains(runtime.Version(), "devel") {
if *nextFile != "" {
fmt.Printf("Go version is %q, ignoring -next %s\n", runtime.Version(), *nextFile)
*nextFile = ""
}
}
if *forceCtx != "" {
setContexts()
}
for _, c := range contexts {
c.Compiler = build.Default.Compiler
}
var pkgNames []string
if flag.NArg() > 0 {
pkgNames = flag.Args()
} else {
stds, err := exec.Command("go", "list", "std").Output()
if err != nil {
log.Fatal(err)
}
for _, pkg := range strings.Fields(string(stds)) {
if !internalPkg.MatchString(pkg) {
pkgNames = append(pkgNames, pkg)
}
}
}
var featureCtx = make(map[string]map[string]bool) // feature -> context name -> true
for _, context := range contexts {
w := NewWalker(context, filepath.Join(build.Default.GOROOT, "src"))
for _, name := range pkgNames {
// Vendored packages do not contribute to our
// public API surface.
if strings.HasPrefix(name, "vendor/") {
continue
}
// - Package "unsafe" contains special signatures requiring
// extra care when printing them - ignore since it is not
// going to change w/o a language change.
// - We don't care about the API of commands.
if name != "unsafe" && !strings.HasPrefix(name, "cmd/") {
if name == "runtime/cgo" && !context.CgoEnabled {
// w.Import(name) will return nil
continue
}
pkg, _ := w.Import(name)
w.export(pkg)
}
}
ctxName := contextName(context)
for _, f := range w.Features() {
if featureCtx[f] == nil {
featureCtx[f] = make(map[string]bool)
}
featureCtx[f][ctxName] = true
}
}
var features []string
for f, cmap := range featureCtx {
if len(cmap) == len(contexts) {
features = append(features, f)
continue
}
comma := strings.Index(f, ",")
for cname := range cmap {
f2 := fmt.Sprintf("%s (%s)%s", f[:comma], cname, f[comma:])
features = append(features, f2)
}
}
fail := false
defer func() {
if fail {
os.Exit(1)
}
}()
bw := bufio.NewWriter(os.Stdout)
defer bw.Flush()
if *checkFile == "" {
sort.Strings(features)
for _, f := range features {
fmt.Fprintln(bw, f)
}
return
}
var required []string
for _, file := range strings.Split(*checkFile, ",") {
required = append(required, fileFeatures(file)...)
}
optional := fileFeatures(*nextFile)
exception := fileFeatures(*exceptFile)
fail = !compareAPI(bw, features, required, optional, exception,
*allowNew && strings.Contains(runtime.Version(), "devel"))
}
// export emits the exported package features.
func (w *Walker) export(pkg *types.Package) {
if *verbose {
log.Println(pkg)
}
pop := w.pushScope("pkg " + pkg.Path())
w.current = pkg
scope := pkg.Scope()
for _, name := range scope.Names() {
if ast.IsExported(name) {
w.emitObj(scope.Lookup(name))
}
}
pop()
}
func set(items []string) map[string]bool {
s := make(map[string]bool)
for _, v := range items {
s[v] = true
}
return s
}
var spaceParensRx = regexp.MustCompile(` \(\S+?\)`)
func featureWithoutContext(f string) string {
if !strings.Contains(f, "(") {
return f
}
return spaceParensRx.ReplaceAllString(f, "")
}
func compareAPI(w io.Writer, features, required, optional, exception []string, allowAdd bool) (ok bool) {
ok = true
optionalSet := set(optional)
exceptionSet := set(exception)
featureSet := set(features)
sort.Strings(features)
sort.Strings(required)
take := func(sl *[]string) string {
s := (*sl)[0]
*sl = (*sl)[1:]
return s
}
for len(required) > 0 || len(features) > 0 {
switch {
case len(features) == 0 || (len(required) > 0 && required[0] < features[0]):
feature := take(&required)
if exceptionSet[feature] {
// An "unfortunate" case: the feature was once
// included in the API (e.g. go1.txt), but was
// subsequently removed. These are already
// acknowledged by being in the file
// "api/except.txt". No need to print them out
// here.
} else if featureSet[featureWithoutContext(feature)] {
// okay.
} else {
fmt.Fprintf(w, "-%s\n", feature)
ok = false // broke compatibility
}
case len(required) == 0 || (len(features) > 0 && required[0] > features[0]):
newFeature := take(&features)
if optionalSet[newFeature] {
// Known added feature to the upcoming release.
// Delete it from the map so we can detect any upcoming features
// which were never seen. (so we can clean up the nextFile)
delete(optionalSet, newFeature)
} else {
fmt.Fprintf(w, "+%s\n", newFeature)
if !allowAdd {
ok = false // we're in lock-down mode for next release
}
}
default:
take(&required)
take(&features)
}
}
// In next file, but not in API.
var missing []string
for feature := range optionalSet {
missing = append(missing, feature)
}
sort.Strings(missing)
for _, feature := range missing {
fmt.Fprintf(w, "±%s\n", feature)
}
return
}
func fileFeatures(filename string) []string {
if filename == "" {
return nil
}
bs, err := ioutil.ReadFile(filename)
if err != nil {
log.Fatalf("Error reading file %s: %v", filename, err)
}
lines := strings.Split(string(bs), "\n")
var nonblank []string
for _, line := range lines {
line = strings.TrimSpace(line)
if line != "" && !strings.HasPrefix(line, "#") {
nonblank = append(nonblank, line)
}
}
return nonblank
}
var fset = token.NewFileSet()
type Walker struct {
context *build.Context
root string
scope []string
current *types.Package
features map[string]bool // set
imported map[string]*types.Package // packages already imported
}
func NewWalker(context *build.Context, root string) *Walker {
return &Walker{
context: context,
root: root,
features: map[string]bool{},
imported: map[string]*types.Package{"unsafe": types.Unsafe},
}
}
func (w *Walker) Features() (fs []string) {
for f := range w.features {
fs = append(fs, f)
}
sort.Strings(fs)
return
}
var parsedFileCache = make(map[string]*ast.File)
func (w *Walker) parseFile(dir, file string) (*ast.File, error) {
filename := filepath.Join(dir, file)
if f := parsedFileCache[filename]; f != nil {
return f, nil
}
f, err := parser.ParseFile(fset, filename, nil, 0)
if err != nil {
return nil, err
}
parsedFileCache[filename] = f
return f, nil
}
// The package cache doesn't operate correctly in rare (so far artificial)
// circumstances (issue 8425). Disable before debugging non-obvious errors
// from the type-checker.
const usePkgCache = true
var (
pkgCache = map[string]*types.Package{} // map tagKey to package
pkgTags = map[string][]string{} // map import dir to list of relevant tags
)
// tagKey returns the tag-based key to use in the pkgCache.
// It is a comma-separated string; the first part is dir, the rest tags.
// The satisfied tags are derived from context but only those that
// matter (the ones listed in the tags argument) are used.
// The tags list, which came from go/build's Package.AllTags,
// is known to be sorted.
func tagKey(dir string, context *build.Context, tags []string) string {
ctags := map[string]bool{
context.GOOS: true,
context.GOARCH: true,
}
if context.CgoEnabled {
ctags["cgo"] = true
}
for _, tag := range context.BuildTags {
ctags[tag] = true
}
// TODO: ReleaseTags (need to load default)
key := dir
for _, tag := range tags {
if ctags[tag] {
key += "," + tag
}
}
return key
}
// Importing is a sentinel taking the place in Walker.imported
// for a package that is in the process of being imported.
var importing types.Package
func (w *Walker) Import(name string) (*types.Package, error) {
pkg := w.imported[name]
if pkg != nil {
if pkg == &importing {
log.Fatalf("cycle importing package %q", name)
}
return pkg, nil
}
w.imported[name] = &importing
root := w.root
if strings.HasPrefix(name, "golang.org/x/") {
root = filepath.Join(root, "vendor")
}
// Determine package files.
dir := filepath.Join(root, filepath.FromSlash(name))
if fi, err := os.Stat(dir); err != nil || !fi.IsDir() {
log.Fatalf("no source in tree for import %q: %v", name, err)
}
context := w.context
if context == nil {
context = &build.Default
}
// Look in cache.
// If we've already done an import with the same set
// of relevant tags, reuse the result.
var key string
if usePkgCache {
if tags, ok := pkgTags[dir]; ok {
key = tagKey(dir, context, tags)
if pkg := pkgCache[key]; pkg != nil {
w.imported[name] = pkg
return pkg, nil
}
}
}
info, err := context.ImportDir(dir, 0)
if err != nil {
if _, nogo := err.(*build.NoGoError); nogo {
return nil, nil
}
log.Fatalf("pkg %q, dir %q: ScanDir: %v", name, dir, err)
}
// Save tags list first time we see a directory.
if usePkgCache {
if _, ok := pkgTags[dir]; !ok {
pkgTags[dir] = info.AllTags
key = tagKey(dir, context, info.AllTags)
}
}
filenames := append(append([]string{}, info.GoFiles...), info.CgoFiles...)
// Parse package files.
var files []*ast.File
for _, file := range filenames {
f, err := w.parseFile(dir, file)
if err != nil {
log.Fatalf("error parsing package %s: %s", name, err)
}
files = append(files, f)
}
// Type-check package files.
conf := types.Config{
IgnoreFuncBodies: true,
FakeImportC: true,
Importer: w,
}
pkg, err = conf.Check(name, fset, files, nil)
if err != nil {
ctxt := "<no context>"
if w.context != nil {
ctxt = fmt.Sprintf("%s-%s", w.context.GOOS, w.context.GOARCH)
}
log.Fatalf("error typechecking package %s: %s (%s)", name, err, ctxt)
}
if usePkgCache {
pkgCache[key] = pkg
}
w.imported[name] = pkg
return pkg, nil
}
// pushScope enters a new scope (walking a package, type, node, etc)
// and returns a function that will leave the scope (with sanity checking
// for mismatched pushes & pops)
func (w *Walker) pushScope(name string) (popFunc func()) {
w.scope = append(w.scope, name)
return func() {
if len(w.scope) == 0 {
log.Fatalf("attempt to leave scope %q with empty scope list", name)
}
if w.scope[len(w.scope)-1] != name {
log.Fatalf("attempt to leave scope %q, but scope is currently %#v", name, w.scope)
}
w.scope = w.scope[:len(w.scope)-1]
}
}
func sortedMethodNames(typ *types.Interface) []string {
n := typ.NumMethods()
list := make([]string, n)
for i := range list {
list[i] = typ.Method(i).Name()
}
sort.Strings(list)
return list
}
func (w *Walker) writeType(buf *bytes.Buffer, typ types.Type) {
switch typ := typ.(type) {
case *types.Basic:
s := typ.Name()
switch typ.Kind() {
case types.UnsafePointer:
s = "unsafe.Pointer"
case types.UntypedBool:
s = "ideal-bool"
case types.UntypedInt:
s = "ideal-int"
case types.UntypedRune:
// "ideal-char" for compatibility with old tool
// TODO(gri) change to "ideal-rune"
s = "ideal-char"
case types.UntypedFloat:
s = "ideal-float"
case types.UntypedComplex:
s = "ideal-complex"
case types.UntypedString:
s = "ideal-string"
case types.UntypedNil:
panic("should never see untyped nil type")
default:
switch s {
case "byte":
s = "uint8"
case "rune":
s = "int32"
}
}
buf.WriteString(s)
case *types.Array:
fmt.Fprintf(buf, "[%d]", typ.Len())
w.writeType(buf, typ.Elem())
case *types.Slice:
buf.WriteString("[]")
w.writeType(buf, typ.Elem())
case *types.Struct:
buf.WriteString("struct")
case *types.Pointer:
buf.WriteByte('*')
w.writeType(buf, typ.Elem())
case *types.Tuple:
panic("should never see a tuple type")
case *types.Signature:
buf.WriteString("func")
w.writeSignature(buf, typ)
case *types.Interface:
buf.WriteString("interface{")
if typ.NumMethods() > 0 {
buf.WriteByte(' ')
buf.WriteString(strings.Join(sortedMethodNames(typ), ", "))
buf.WriteByte(' ')
}
buf.WriteString("}")
case *types.Map:
buf.WriteString("map[")
w.writeType(buf, typ.Key())
buf.WriteByte(']')
w.writeType(buf, typ.Elem())
case *types.Chan:
var s string
switch typ.Dir() {
case types.SendOnly:
s = "chan<- "
case types.RecvOnly:
s = "<-chan "
case types.SendRecv:
s = "chan "
default:
panic("unreachable")
}
buf.WriteString(s)
w.writeType(buf, typ.Elem())
case *types.Named:
obj := typ.Obj()
pkg := obj.Pkg()
if pkg != nil && pkg != w.current {
buf.WriteString(pkg.Name())
buf.WriteByte('.')
}
buf.WriteString(typ.Obj().Name())
default:
panic(fmt.Sprintf("unknown type %T", typ))
}
}
func (w *Walker) writeSignature(buf *bytes.Buffer, sig *types.Signature) {
w.writeParams(buf, sig.Params(), sig.Variadic())
switch res := sig.Results(); res.Len() {
case 0:
// nothing to do
case 1:
buf.WriteByte(' ')
w.writeType(buf, res.At(0).Type())
default:
buf.WriteByte(' ')
w.writeParams(buf, res, false)
}
}
func (w *Walker) writeParams(buf *bytes.Buffer, t *types.Tuple, variadic bool) {
buf.WriteByte('(')
for i, n := 0, t.Len(); i < n; i++ {
if i > 0 {
buf.WriteString(", ")
}
typ := t.At(i).Type()
if variadic && i+1 == n {
buf.WriteString("...")
typ = typ.(*types.Slice).Elem()
}
w.writeType(buf, typ)
}
buf.WriteByte(')')
}
func (w *Walker) typeString(typ types.Type) string {
var buf bytes.Buffer
w.writeType(&buf, typ)
return buf.String()
}
func (w *Walker) signatureString(sig *types.Signature) string {
var buf bytes.Buffer
w.writeSignature(&buf, sig)
return buf.String()
}
func (w *Walker) emitObj(obj types.Object) {
switch obj := obj.(type) {
case *types.Const:
w.emitf("const %s %s", obj.Name(), w.typeString(obj.Type()))
go/constant: switch to floating-point representation when fractions become too large Use two internal representations for Float values (similar to what is done for Int values). Transparently switch to a big.Float representation when big.Rat values become unwieldy. This is almost never needed for real-world programs but it is trivial to create test cases that cannot be handled with rational arithmetic alone. As a consequence, the go/constant API semantics changes slightly: Until now, a value could always be represented in its "smallest" form (e.g., float values that happened to be integers would be represented as integers). Now, constant Kind depends on how the value was created, rather than its actual value. (The reason why we cannot automatically "normalize" values to their smallest form anymore is because floating-point numbers are not exact in general; and thus normalization is often not possible in the first place, or would throw away precision when it is not desired.) This has repercussions as to how constant Values are used go/types and required corresponding adjustments. Details of the changes: go/constant package: - use big.Rat and big.Float values to represent floating-point values (internal change) - changed semantic of Value.Kind accordingly - String now returns a short, human-readable form of a value (this leads to better error messages in go/types) - added ToInt, ToFloat, and ToComplex conversion functions - added ExactString to obtain an exact string form of a value go/types: - adjusted and simplified implementation of representableConst - adjusted various places where Value.Kind was expected to be "smallest" by calling the respective ToInt/Float/Complex conversion functions - enabled 5 disabled tests in stdlib_test.go that now work api checker: - print all constant values in a short human-readable form (floats are printed in floating-point form), but also print an exact form if it is different from the short form - adjusted test golden file and go.1.1.text reference file Fixes #11327. Change-Id: I492b704aae5b0238e5b7cee13e18ffce61193587 Reviewed-on: https://go-review.googlesource.com/17360 Reviewed-by: Alan Donovan <adonovan@google.com> Run-TryBot: Robert Griesemer <gri@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-12-02 17:41:39 -08:00
x := obj.Val()
short := x.String()
exact := x.ExactString()
if short == exact {
w.emitf("const %s = %s", obj.Name(), short)
} else {
w.emitf("const %s = %s // %s", obj.Name(), short, exact)
}
case *types.Var:
w.emitf("var %s %s", obj.Name(), w.typeString(obj.Type()))
case *types.TypeName:
w.emitType(obj)
case *types.Func:
w.emitFunc(obj)
default:
panic("unknown object: " + obj.String())
}
}
func (w *Walker) emitType(obj *types.TypeName) {
name := obj.Name()
typ := obj.Type()
switch typ := typ.Underlying().(type) {
case *types.Struct:
w.emitStructType(name, typ)
case *types.Interface:
w.emitIfaceType(name, typ)
return // methods are handled by emitIfaceType
default:
w.emitf("type %s %s", name, w.typeString(typ.Underlying()))
}
// emit methods with value receiver
var methodNames map[string]bool
vset := types.NewMethodSet(typ)
for i, n := 0, vset.Len(); i < n; i++ {
m := vset.At(i)
if m.Obj().Exported() {
w.emitMethod(m)
if methodNames == nil {
methodNames = make(map[string]bool)
}
methodNames[m.Obj().Name()] = true
}
}
// emit methods with pointer receiver; exclude
// methods that we have emitted already
// (the method set of *T includes the methods of T)
pset := types.NewMethodSet(types.NewPointer(typ))
for i, n := 0, pset.Len(); i < n; i++ {
m := pset.At(i)
if m.Obj().Exported() && !methodNames[m.Obj().Name()] {
w.emitMethod(m)
}
}
}
func (w *Walker) emitStructType(name string, typ *types.Struct) {
typeStruct := fmt.Sprintf("type %s struct", name)
w.emitf(typeStruct)
defer w.pushScope(typeStruct)()
for i := 0; i < typ.NumFields(); i++ {
f := typ.Field(i)
if !f.Exported() {
continue
}
typ := f.Type()
if f.Anonymous() {
w.emitf("embedded %s", w.typeString(typ))
continue
}
w.emitf("%s %s", f.Name(), w.typeString(typ))
}
}
func (w *Walker) emitIfaceType(name string, typ *types.Interface) {
pop := w.pushScope("type " + name + " interface")
var methodNames []string
complete := true
mset := types.NewMethodSet(typ)
for i, n := 0, mset.Len(); i < n; i++ {
m := mset.At(i).Obj().(*types.Func)
if !m.Exported() {
complete = false
continue
}
methodNames = append(methodNames, m.Name())
w.emitf("%s%s", m.Name(), w.signatureString(m.Type().(*types.Signature)))
}
if !complete {
// The method set has unexported methods, so all the
// implementations are provided by the same package,
// so the method set can be extended. Instead of recording
// the full set of names (below), record only that there were
// unexported methods. (If the interface shrinks, we will notice
// because a method signature emitted during the last loop
// will disappear.)
w.emitf("unexported methods")
}
pop()
if !complete {
return
}
if len(methodNames) == 0 {
w.emitf("type %s interface {}", name)
return
}
sort.Strings(methodNames)
w.emitf("type %s interface { %s }", name, strings.Join(methodNames, ", "))
}
func (w *Walker) emitFunc(f *types.Func) {
sig := f.Type().(*types.Signature)
if sig.Recv() != nil {
panic("method considered a regular function: " + f.String())
}
w.emitf("func %s%s", f.Name(), w.signatureString(sig))
}
func (w *Walker) emitMethod(m *types.Selection) {
sig := m.Type().(*types.Signature)
recv := sig.Recv().Type()
// report exported methods with unexported receiver base type
if true {
base := recv
if p, _ := recv.(*types.Pointer); p != nil {
base = p.Elem()
}
if obj := base.(*types.Named).Obj(); !obj.Exported() {
log.Fatalf("exported method with unexported receiver base type: %s", m)
}
}
w.emitf("method (%s) %s%s", w.typeString(recv), m.Obj().Name(), w.signatureString(sig))
}
func (w *Walker) emitf(format string, args ...interface{}) {
f := strings.Join(w.scope, ", ") + ", " + fmt.Sprintf(format, args...)
if strings.Contains(f, "\n") {
panic("feature contains newlines: " + f)
}
if _, dup := w.features[f]; dup {
panic("duplicate feature inserted: " + f)
}
w.features[f] = true
if *verbose {
log.Printf("feature: %s", f)
}
}