// 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.
|
|
|
|
package main
|
|
|
|
import (
|
|
"bytes"
|
|
"fmt"
|
|
exact "go/constant"
|
|
"go/token"
|
|
"go/types"
|
|
"io"
|
|
"math/big"
|
|
)
|
|
|
|
// TODO(gri) use tabwriter for alignment?
|
|
|
|
func print(w io.Writer, pkg *types.Package, filter func(types.Object) bool) {
|
|
var p printer
|
|
p.pkg = pkg
|
|
p.printPackage(pkg, filter)
|
|
p.printGccgoExtra(pkg)
|
|
io.Copy(w, &p.buf)
|
|
}
|
|
|
|
type printer struct {
|
|
pkg *types.Package
|
|
buf bytes.Buffer
|
|
indent int // current indentation level
|
|
last byte // last byte written
|
|
}
|
|
|
|
func (p *printer) print(s string) {
|
|
// Write the string one byte at a time. We care about the presence of
|
|
// newlines for indentation which we will see even in the presence of
|
|
// (non-corrupted) Unicode; no need to read one rune at a time.
|
|
for i := 0; i < len(s); i++ {
|
|
ch := s[i]
|
|
if ch != '\n' && p.last == '\n' {
|
|
// Note: This could lead to a range overflow for very large
|
|
// indentations, but it's extremely unlikely to happen for
|
|
// non-pathological code.
|
|
p.buf.WriteString("\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t"[:p.indent])
|
|
}
|
|
p.buf.WriteByte(ch)
|
|
p.last = ch
|
|
}
|
|
}
|
|
|
|
func (p *printer) printf(format string, args ...interface{}) {
|
|
p.print(fmt.Sprintf(format, args...))
|
|
}
|
|
|
|
// methodsFor returns the named type and corresponding methods if the type
|
|
// denoted by obj is not an interface and has methods. Otherwise it returns
|
|
// the zero value.
|
|
func methodsFor(obj *types.TypeName) (*types.Named, []*types.Selection) {
|
|
named, _ := obj.Type().(*types.Named)
|
|
if named == nil {
|
|
// A type name's type can also be the
|
|
// exported basic type unsafe.Pointer.
|
|
return nil, nil
|
|
}
|
|
if _, ok := named.Underlying().(*types.Interface); ok {
|
|
// ignore interfaces
|
|
return nil, nil
|
|
}
|
|
methods := combinedMethodSet(named)
|
|
if len(methods) == 0 {
|
|
return nil, nil
|
|
}
|
|
return named, methods
|
|
}
|
|
|
|
func (p *printer) printPackage(pkg *types.Package, filter func(types.Object) bool) {
|
|
// collect objects by kind
|
|
var (
|
|
consts []*types.Const
|
|
typem []*types.Named // non-interface types with methods
|
|
typez []*types.TypeName // interfaces or types without methods
|
|
vars []*types.Var
|
|
funcs []*types.Func
|
|
builtins []*types.Builtin
|
|
methods = make(map[*types.Named][]*types.Selection) // method sets for named types
|
|
)
|
|
scope := pkg.Scope()
|
|
for _, name := range scope.Names() {
|
|
obj := scope.Lookup(name)
|
|
if obj.Exported() {
|
|
// collect top-level exported and possibly filtered objects
|
|
if filter == nil || filter(obj) {
|
|
switch obj := obj.(type) {
|
|
case *types.Const:
|
|
consts = append(consts, obj)
|
|
case *types.TypeName:
|
|
// group into types with methods and types without
|
|
if named, m := methodsFor(obj); named != nil {
|
|
typem = append(typem, named)
|
|
methods[named] = m
|
|
} else {
|
|
typez = append(typez, obj)
|
|
}
|
|
case *types.Var:
|
|
vars = append(vars, obj)
|
|
case *types.Func:
|
|
funcs = append(funcs, obj)
|
|
case *types.Builtin:
|
|
// for unsafe.Sizeof, etc.
|
|
builtins = append(builtins, obj)
|
|
}
|
|
}
|
|
} else if filter == nil {
|
|
// no filtering: collect top-level unexported types with methods
|
|
if obj, _ := obj.(*types.TypeName); obj != nil {
|
|
// see case *types.TypeName above
|
|
if named, m := methodsFor(obj); named != nil {
|
|
typem = append(typem, named)
|
|
methods[named] = m
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
p.printf("package %s // %q\n", pkg.Name(), pkg.Path())
|
|
|
|
p.printDecl("const", len(consts), func() {
|
|
for _, obj := range consts {
|
|
p.printObj(obj)
|
|
p.print("\n")
|
|
}
|
|
})
|
|
|
|
p.printDecl("var", len(vars), func() {
|
|
for _, obj := range vars {
|
|
p.printObj(obj)
|
|
p.print("\n")
|
|
}
|
|
})
|
|
|
|
p.printDecl("type", len(typez), func() {
|
|
for _, obj := range typez {
|
|
p.printf("%s ", obj.Name())
|
|
typ := obj.Type()
|
|
if isAlias(obj) {
|
|
p.print("= ")
|
|
p.writeType(p.pkg, typ)
|
|
} else {
|
|
p.writeType(p.pkg, typ.Underlying())
|
|
}
|
|
p.print("\n")
|
|
}
|
|
})
|
|
|
|
// non-interface types with methods
|
|
for _, named := range typem {
|
|
first := true
|
|
if obj := named.Obj(); obj.Exported() {
|
|
if first {
|
|
p.print("\n")
|
|
first = false
|
|
}
|
|
p.printf("type %s ", obj.Name())
|
|
p.writeType(p.pkg, named.Underlying())
|
|
p.print("\n")
|
|
}
|
|
for _, m := range methods[named] {
|
|
if obj := m.Obj(); obj.Exported() {
|
|
if first {
|
|
p.print("\n")
|
|
first = false
|
|
}
|
|
p.printFunc(m.Recv(), obj.(*types.Func))
|
|
p.print("\n")
|
|
}
|
|
}
|
|
}
|
|
|
|
if len(funcs) > 0 {
|
|
p.print("\n")
|
|
for _, obj := range funcs {
|
|
p.printFunc(nil, obj)
|
|
p.print("\n")
|
|
}
|
|
}
|
|
|
|
// TODO(gri) better handling of builtins (package unsafe only)
|
|
if len(builtins) > 0 {
|
|
p.print("\n")
|
|
for _, obj := range builtins {
|
|
p.printf("func %s() // builtin\n", obj.Name())
|
|
}
|
|
}
|
|
|
|
p.print("\n")
|
|
}
|
|
|
|
func (p *printer) printDecl(keyword string, n int, printGroup func()) {
|
|
switch n {
|
|
case 0:
|
|
// nothing to do
|
|
case 1:
|
|
p.printf("\n%s ", keyword)
|
|
printGroup()
|
|
default:
|
|
p.printf("\n%s (\n", keyword)
|
|
p.indent++
|
|
printGroup()
|
|
p.indent--
|
|
p.print(")\n")
|
|
}
|
|
}
|
|
|
|
// absInt returns the absolute value of v as a *big.Int.
|
|
// v must be a numeric value.
|
|
func absInt(v exact.Value) *big.Int {
|
|
// compute big-endian representation of v
|
|
b := exact.Bytes(v) // little-endian
|
|
for i, j := 0, len(b)-1; i < j; i, j = i+1, j-1 {
|
|
b[i], b[j] = b[j], b[i]
|
|
}
|
|
return new(big.Int).SetBytes(b)
|
|
}
|
|
|
|
var (
|
|
one = big.NewRat(1, 1)
|
|
ten = big.NewRat(10, 1)
|
|
)
|
|
|
|
// floatString returns the string representation for a
|
|
// numeric value v in normalized floating-point format.
|
|
func floatString(v exact.Value) string {
|
|
if exact.Sign(v) == 0 {
|
|
return "0.0"
|
|
}
|
|
// x != 0
|
|
|
|
// convert |v| into a big.Rat x
|
|
x := new(big.Rat).SetFrac(absInt(exact.Num(v)), absInt(exact.Denom(v)))
|
|
|
|
// normalize x and determine exponent e
|
|
// (This is not very efficient, but also not speed-critical.)
|
|
var e int
|
|
for x.Cmp(ten) >= 0 {
|
|
x.Quo(x, ten)
|
|
e++
|
|
}
|
|
for x.Cmp(one) < 0 {
|
|
x.Mul(x, ten)
|
|
e--
|
|
}
|
|
|
|
// TODO(gri) Values such as 1/2 are easier to read in form 0.5
|
|
// rather than 5.0e-1. Similarly, 1.0e1 is easier to read as
|
|
// 10.0. Fine-tune best exponent range for readability.
|
|
|
|
s := x.FloatString(100) // good-enough precision
|
|
|
|
// trim trailing 0's
|
|
i := len(s)
|
|
for i > 0 && s[i-1] == '0' {
|
|
i--
|
|
}
|
|
s = s[:i]
|
|
|
|
// add a 0 if the number ends in decimal point
|
|
if len(s) > 0 && s[len(s)-1] == '.' {
|
|
s += "0"
|
|
}
|
|
|
|
// add exponent and sign
|
|
if e != 0 {
|
|
s += fmt.Sprintf("e%+d", e)
|
|
}
|
|
if exact.Sign(v) < 0 {
|
|
s = "-" + s
|
|
}
|
|
|
|
// TODO(gri) If v is a "small" fraction (i.e., numerator and denominator
|
|
// are just a small number of decimal digits), add the exact fraction as
|
|
// a comment. For instance: 3.3333...e-1 /* = 1/3 */
|
|
|
|
return s
|
|
}
|
|
|
|
// valString returns the string representation for the value v.
|
|
// Setting floatFmt forces an integer value to be formatted in
|
|
// normalized floating-point format.
|
|
// TODO(gri) Move this code into package exact.
|
|
func valString(v exact.Value, floatFmt bool) string {
|
|
switch v.Kind() {
|
|
case exact.Int:
|
|
if floatFmt {
|
|
return floatString(v)
|
|
}
|
|
case exact.Float:
|
|
return floatString(v)
|
|
case exact.Complex:
|
|
re := exact.Real(v)
|
|
im := exact.Imag(v)
|
|
var s string
|
|
if exact.Sign(re) != 0 {
|
|
s = floatString(re)
|
|
if exact.Sign(im) >= 0 {
|
|
s += " + "
|
|
} else {
|
|
s += " - "
|
|
im = exact.UnaryOp(token.SUB, im, 0) // negate im
|
|
}
|
|
}
|
|
// im != 0, otherwise v would be exact.Int or exact.Float
|
|
return s + floatString(im) + "i"
|
|
}
|
|
return v.String()
|
|
}
|
|
|
|
func (p *printer) printObj(obj types.Object) {
|
|
p.print(obj.Name())
|
|
|
|
typ, basic := obj.Type().Underlying().(*types.Basic)
|
|
if basic && typ.Info()&types.IsUntyped != 0 {
|
|
// don't write untyped types
|
|
} else {
|
|
p.print(" ")
|
|
p.writeType(p.pkg, obj.Type())
|
|
}
|
|
|
|
if obj, ok := obj.(*types.Const); ok {
|
|
floatFmt := basic && typ.Info()&(types.IsFloat|types.IsComplex) != 0
|
|
p.print(" = ")
|
|
p.print(valString(obj.Val(), floatFmt))
|
|
}
|
|
}
|
|
|
|
func (p *printer) printFunc(recvType types.Type, obj *types.Func) {
|
|
p.print("func ")
|
|
sig := obj.Type().(*types.Signature)
|
|
if recvType != nil {
|
|
p.print("(")
|
|
p.writeType(p.pkg, recvType)
|
|
p.print(") ")
|
|
}
|
|
p.print(obj.Name())
|
|
p.writeSignature(p.pkg, sig)
|
|
}
|
|
|
|
// combinedMethodSet returns the method set for a named type T
|
|
// merged with all the methods of *T that have different names than
|
|
// the methods of T.
|
|
//
|
|
// combinedMethodSet is analogous to types/typeutil.IntuitiveMethodSet
|
|
// but doesn't require a MethodSetCache.
|
|
// TODO(gri) If this functionality doesn't change over time, consider
|
|
// just calling IntuitiveMethodSet eventually.
|
|
func combinedMethodSet(T *types.Named) []*types.Selection {
|
|
// method set for T
|
|
mset := types.NewMethodSet(T)
|
|
var res []*types.Selection
|
|
for i, n := 0, mset.Len(); i < n; i++ {
|
|
res = append(res, mset.At(i))
|
|
}
|
|
|
|
// add all *T methods with names different from T methods
|
|
pmset := types.NewMethodSet(types.NewPointer(T))
|
|
for i, n := 0, pmset.Len(); i < n; i++ {
|
|
pm := pmset.At(i)
|
|
if obj := pm.Obj(); mset.Lookup(obj.Pkg(), obj.Name()) == nil {
|
|
res = append(res, pm)
|
|
}
|
|
}
|
|
|
|
return res
|
|
}
|