blob: ff8dd13b6d717cac35f32b6368c5f8703011568a [file] [log] [blame]
// Copyright 2018 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.
// This file implements type parameter substitution.
package types2
import (
"bytes"
"cmd/compile/internal/syntax"
)
type substMap map[*TypeParam]Type
// makeSubstMap creates a new substitution map mapping tpars[i] to targs[i].
// If targs[i] is nil, tpars[i] is not substituted.
func makeSubstMap(tpars []*TypeParam, targs []Type) substMap {
assert(len(tpars) == len(targs))
proj := make(substMap, len(tpars))
for i, tpar := range tpars {
proj[tpar] = targs[i]
}
return proj
}
func (m substMap) empty() bool {
return len(m) == 0
}
func (m substMap) lookup(tpar *TypeParam) Type {
if t := m[tpar]; t != nil {
return t
}
return tpar
}
// subst returns the type typ with its type parameters tpars replaced by the
// corresponding type arguments targs, recursively. subst doesn't modify the
// incoming type. If a substitution took place, the result type is different
// from the incoming type.
//
// If the given typMap is non-nil, it is used in lieu of check.typMap.
func (check *Checker) subst(pos syntax.Pos, typ Type, smap substMap, typMap map[string]*Named) Type {
if smap.empty() {
return typ
}
// common cases
switch t := typ.(type) {
case *Basic:
return typ // nothing to do
case *TypeParam:
return smap.lookup(t)
}
// general case
var subst subster
subst.pos = pos
subst.smap = smap
if check != nil {
subst.check = check
if typMap == nil {
typMap = check.typMap
}
}
if typMap == nil {
// If we don't have a *Checker and its global type map,
// use a local version. Besides avoiding duplicate work,
// the type map prevents infinite recursive substitution
// for recursive types (example: type T[P any] *T[P]).
typMap = make(map[string]*Named)
}
subst.typMap = typMap
return subst.typ(typ)
}
type subster struct {
pos syntax.Pos
smap substMap
check *Checker // nil if called via Instantiate
typMap map[string]*Named
}
func (subst *subster) typ(typ Type) Type {
switch t := typ.(type) {
case nil:
// Call typOrNil if it's possible that typ is nil.
panic("nil typ")
case *Basic, *top:
// nothing to do
case *Array:
elem := subst.typOrNil(t.elem)
if elem != t.elem {
return &Array{len: t.len, elem: elem}
}
case *Slice:
elem := subst.typOrNil(t.elem)
if elem != t.elem {
return &Slice{elem: elem}
}
case *Struct:
if fields, copied := subst.varList(t.fields); copied {
return &Struct{fields: fields, tags: t.tags}
}
case *Pointer:
base := subst.typ(t.base)
if base != t.base {
return &Pointer{base: base}
}
case *Tuple:
return subst.tuple(t)
case *Signature:
// TODO(gri) rethink the recv situation with respect to methods on parameterized types
// recv := subst.var_(t.recv) // TODO(gri) this causes a stack overflow - explain
recv := t.recv
params := subst.tuple(t.params)
results := subst.tuple(t.results)
if recv != t.recv || params != t.params || results != t.results {
return &Signature{
rparams: t.rparams,
// TODO(gri) Why can't we nil out tparams here, rather than in
// instantiate above?
tparams: t.tparams,
scope: t.scope,
recv: recv,
params: params,
results: results,
variadic: t.variadic,
}
}
case *Union:
terms, copied := subst.termlist(t.terms)
if copied {
// term list substitution may introduce duplicate terms (unlikely but possible).
// This is ok; lazy type set computation will determine the actual type set
// in normal form.
return &Union{terms, nil}
}
case *Interface:
methods, mcopied := subst.funcList(t.methods)
embeddeds, ecopied := subst.typeList(t.embeddeds)
if mcopied || ecopied {
iface := &Interface{methods: methods, embeddeds: embeddeds, complete: t.complete}
return iface
}
case *Map:
key := subst.typ(t.key)
elem := subst.typ(t.elem)
if key != t.key || elem != t.elem {
return &Map{key: key, elem: elem}
}
case *Chan:
elem := subst.typ(t.elem)
if elem != t.elem {
return &Chan{dir: t.dir, elem: elem}
}
case *Named:
// dump is for debugging
dump := func(string, ...interface{}) {}
if subst.check != nil && subst.check.conf.Trace {
subst.check.indent++
defer func() {
subst.check.indent--
}()
dump = func(format string, args ...interface{}) {
subst.check.trace(subst.pos, format, args...)
}
}
if t.TParams().Len() == 0 {
dump(">>> %s is not parameterized", t)
return t // type is not parameterized
}
var newTArgs []Type
assert(t.targs.Len() == t.TParams().Len())
// already instantiated
dump(">>> %s already instantiated", t)
// For each (existing) type argument targ, determine if it needs
// to be substituted; i.e., if it is or contains a type parameter
// that has a type argument for it.
for i, targ := range t.targs.list() {
dump(">>> %d targ = %s", i, targ)
new_targ := subst.typ(targ)
if new_targ != targ {
dump(">>> substituted %d targ %s => %s", i, targ, new_targ)
if newTArgs == nil {
newTArgs = make([]Type, t.TParams().Len())
copy(newTArgs, t.targs.list())
}
newTArgs[i] = new_targ
}
}
if newTArgs == nil {
dump(">>> nothing to substitute in %s", t)
return t // nothing to substitute
}
// before creating a new named type, check if we have this one already
h := instantiatedHash(t, newTArgs)
dump(">>> new type hash: %s", h)
if named, found := subst.typMap[h]; found {
dump(">>> found %s", named)
return named
}
// Create a new named type and populate typMap to avoid endless recursion.
// The position used here is irrelevant because validation only occurs on t
// (we don't call validType on named), but we use subst.pos to help with
// debugging.
tname := NewTypeName(subst.pos, t.obj.pkg, t.obj.name, nil)
t.load()
// It's ok to provide a nil *Checker because the newly created type
// doesn't need to be (lazily) expanded; it's expanded below.
named := (*Checker)(nil).newNamed(tname, t.orig, nil, t.tparams, t.methods) // t is loaded, so tparams and methods are available
named.targs = NewTypeList(newTArgs)
subst.typMap[h] = named
t.expand(subst.typMap) // must happen after typMap update to avoid infinite recursion
// do the substitution
dump(">>> subst %s with %s (new: %s)", t.underlying, subst.smap, newTArgs)
named.underlying = subst.typOrNil(t.underlying)
dump(">>> underlying: %v", named.underlying)
assert(named.underlying != nil)
named.fromRHS = named.underlying // for consistency, though no cycle detection is necessary
return named
case *TypeParam:
return subst.smap.lookup(t)
default:
unimplemented()
}
return typ
}
var instanceHashing = 0
func instantiatedHash(typ *Named, targs []Type) string {
assert(instanceHashing == 0)
instanceHashing++
var buf bytes.Buffer
writeTypeName(&buf, typ.obj, nil)
buf.WriteByte('[')
writeTypeList(&buf, targs, nil, nil)
buf.WriteByte(']')
instanceHashing--
// With respect to the represented type, whether a
// type is fully expanded or stored as instance
// does not matter - they are the same types.
// Remove the instanceMarkers printed for instances.
res := buf.Bytes()
i := 0
for _, b := range res {
if b != instanceMarker {
res[i] = b
i++
}
}
return string(res[:i])
}
func typeListString(list []Type) string {
var buf bytes.Buffer
writeTypeList(&buf, list, nil, nil)
return buf.String()
}
// typOrNil is like typ but if the argument is nil it is replaced with Typ[Invalid].
// A nil type may appear in pathological cases such as type T[P any] []func(_ T([]_))
// where an array/slice element is accessed before it is set up.
func (subst *subster) typOrNil(typ Type) Type {
if typ == nil {
return Typ[Invalid]
}
return subst.typ(typ)
}
func (subst *subster) var_(v *Var) *Var {
if v != nil {
if typ := subst.typ(v.typ); typ != v.typ {
copy := *v
copy.typ = typ
return &copy
}
}
return v
}
func (subst *subster) tuple(t *Tuple) *Tuple {
if t != nil {
if vars, copied := subst.varList(t.vars); copied {
return &Tuple{vars: vars}
}
}
return t
}
func (subst *subster) varList(in []*Var) (out []*Var, copied bool) {
out = in
for i, v := range in {
if w := subst.var_(v); w != v {
if !copied {
// first variable that got substituted => allocate new out slice
// and copy all variables
new := make([]*Var, len(in))
copy(new, out)
out = new
copied = true
}
out[i] = w
}
}
return
}
func (subst *subster) func_(f *Func) *Func {
if f != nil {
if typ := subst.typ(f.typ); typ != f.typ {
copy := *f
copy.typ = typ
return &copy
}
}
return f
}
func (subst *subster) funcList(in []*Func) (out []*Func, copied bool) {
out = in
for i, f := range in {
if g := subst.func_(f); g != f {
if !copied {
// first function that got substituted => allocate new out slice
// and copy all functions
new := make([]*Func, len(in))
copy(new, out)
out = new
copied = true
}
out[i] = g
}
}
return
}
func (subst *subster) typeList(in []Type) (out []Type, copied bool) {
out = in
for i, t := range in {
if u := subst.typ(t); u != t {
if !copied {
// first function that got substituted => allocate new out slice
// and copy all functions
new := make([]Type, len(in))
copy(new, out)
out = new
copied = true
}
out[i] = u
}
}
return
}
func (subst *subster) termlist(in []*Term) (out []*Term, copied bool) {
out = in
for i, t := range in {
if u := subst.typ(t.typ); u != t.typ {
if !copied {
// first function that got substituted => allocate new out slice
// and copy all functions
new := make([]*Term, len(in))
copy(new, out)
out = new
copied = true
}
out[i] = NewTerm(t.tilde, u)
}
}
return
}