From 13a1c8ac4b8c1d3ef2f479a59de38c60f612faf6 Mon Sep 17 00:00:00 2001 From: xushiwei Date: Fri, 3 May 2024 17:39:37 +0800 Subject: [PATCH] github.com/goplus/llgo/internal/typeutil.Map --- internal/aliases/aliases.go | 32 ++ internal/aliases/aliases_go121.go | 31 ++ internal/aliases/aliases_go122.go | 63 ++++ internal/typeparams/normalize.go | 218 +++++++++++++ internal/typeparams/termlist.go | 163 ++++++++++ internal/typeparams/typeterm.go | 169 ++++++++++ internal/typeutil/map.go | 518 ++++++++++++++++++++++++++++++ ssa/package.go | 2 +- 8 files changed, 1195 insertions(+), 1 deletion(-) create mode 100644 internal/aliases/aliases.go create mode 100644 internal/aliases/aliases_go121.go create mode 100644 internal/aliases/aliases_go122.go create mode 100644 internal/typeparams/normalize.go create mode 100644 internal/typeparams/termlist.go create mode 100644 internal/typeparams/typeterm.go create mode 100644 internal/typeutil/map.go diff --git a/internal/aliases/aliases.go b/internal/aliases/aliases.go new file mode 100644 index 00000000..c24c2eee --- /dev/null +++ b/internal/aliases/aliases.go @@ -0,0 +1,32 @@ +// 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 aliases + +import ( + "go/token" + "go/types" +) + +// Package aliases defines backward compatible shims +// for the types.Alias type representation added in 1.22. +// This defines placeholders for x/tools until 1.26. + +// NewAlias creates a new TypeName in Package pkg that +// is an alias for the type rhs. +// +// The enabled parameter determines whether the resulting [TypeName]'s +// type is an [types.Alias]. Its value must be the result of a call to +// [Enabled], which computes the effective value of +// GODEBUG=gotypesalias=... by invoking the type checker. The Enabled +// function is expensive and should be called once per task (e.g. +// package import), not once per call to NewAlias. +func NewAlias(enabled bool, pos token.Pos, pkg *types.Package, name string, rhs types.Type) *types.TypeName { + if enabled { + tname := types.NewTypeName(pos, pkg, name, nil) + newAlias(tname, rhs) + return tname + } + return types.NewTypeName(pos, pkg, name, rhs) +} diff --git a/internal/aliases/aliases_go121.go b/internal/aliases/aliases_go121.go new file mode 100644 index 00000000..c027b9f3 --- /dev/null +++ b/internal/aliases/aliases_go121.go @@ -0,0 +1,31 @@ +// 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. + +//go:build !go1.22 +// +build !go1.22 + +package aliases + +import ( + "go/types" +) + +// Alias is a placeholder for a go/types.Alias for <=1.21. +// It will never be created by go/types. +type Alias struct{} + +func (*Alias) String() string { panic("unreachable") } +func (*Alias) Underlying() types.Type { panic("unreachable") } +func (*Alias) Obj() *types.TypeName { panic("unreachable") } +func Rhs(alias *Alias) types.Type { panic("unreachable") } + +// Unalias returns the type t for go <=1.21. +func Unalias(t types.Type) types.Type { return t } + +func newAlias(name *types.TypeName, rhs types.Type) *Alias { panic("unreachable") } + +// Enabled reports whether [NewAlias] should create [types.Alias] types. +// +// Before go1.22, this function always returns false. +func Enabled() bool { return false } diff --git a/internal/aliases/aliases_go122.go b/internal/aliases/aliases_go122.go new file mode 100644 index 00000000..b3299548 --- /dev/null +++ b/internal/aliases/aliases_go122.go @@ -0,0 +1,63 @@ +// 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. + +//go:build go1.22 +// +build go1.22 + +package aliases + +import ( + "go/ast" + "go/parser" + "go/token" + "go/types" +) + +// Alias is an alias of types.Alias. +type Alias = types.Alias + +// Rhs returns the type on the right-hand side of the alias declaration. +func Rhs(alias *Alias) types.Type { + if alias, ok := any(alias).(interface{ Rhs() types.Type }); ok { + return alias.Rhs() // go1.23+ + } + + // go1.22's Alias didn't have the Rhs method, + // so Unalias is the best we can do. + return Unalias(alias) +} + +// Unalias is a wrapper of types.Unalias. +func Unalias(t types.Type) types.Type { return types.Unalias(t) } + +// newAlias is an internal alias around types.NewAlias. +// Direct usage is discouraged as the moment. +// Try to use NewAlias instead. +func newAlias(tname *types.TypeName, rhs types.Type) *Alias { + a := types.NewAlias(tname, rhs) + // TODO(go.dev/issue/65455): Remove kludgy workaround to set a.actual as a side-effect. + Unalias(a) + return a +} + +// Enabled reports whether [NewAlias] should create [types.Alias] types. +// +// This function is expensive! Call it sparingly. +func Enabled() bool { + // The only reliable way to compute the answer is to invoke go/types. + // We don't parse the GODEBUG environment variable, because + // (a) it's tricky to do so in a manner that is consistent + // with the godebug package; in particular, a simple + // substring check is not good enough. The value is a + // rightmost-wins list of options. But more importantly: + // (b) it is impossible to detect changes to the effective + // setting caused by os.Setenv("GODEBUG"), as happens in + // many tests. Therefore any attempt to cache the result + // is just incorrect. + fset := token.NewFileSet() + f, _ := parser.ParseFile(fset, "a.go", "package p; type A = int", 0) + pkg, _ := new(types.Config).Check("p", fset, []*ast.File{f}, nil) + _, enabled := pkg.Scope().Lookup("A").Type().(*types.Alias) + return enabled +} diff --git a/internal/typeparams/normalize.go b/internal/typeparams/normalize.go new file mode 100644 index 00000000..93c80fdc --- /dev/null +++ b/internal/typeparams/normalize.go @@ -0,0 +1,218 @@ +// Copyright 2021 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 typeparams + +import ( + "errors" + "fmt" + "go/types" + "os" + "strings" +) + +//go:generate go run copytermlist.go + +const debug = false + +var ErrEmptyTypeSet = errors.New("empty type set") + +// StructuralTerms returns a slice of terms representing the normalized +// structural type restrictions of a type parameter, if any. +// +// Structural type restrictions of a type parameter are created via +// non-interface types embedded in its constraint interface (directly, or via a +// chain of interface embeddings). For example, in the declaration +// +// type T[P interface{~int; m()}] int +// +// the structural restriction of the type parameter P is ~int. +// +// With interface embedding and unions, the specification of structural type +// restrictions may be arbitrarily complex. For example, consider the +// following: +// +// type A interface{ ~string|~[]byte } +// +// type B interface{ int|string } +// +// type C interface { ~string|~int } +// +// type T[P interface{ A|B; C }] int +// +// In this example, the structural type restriction of P is ~string|int: A|B +// expands to ~string|~[]byte|int|string, which reduces to ~string|~[]byte|int, +// which when intersected with C (~string|~int) yields ~string|int. +// +// StructuralTerms computes these expansions and reductions, producing a +// "normalized" form of the embeddings. A structural restriction is normalized +// if it is a single union containing no interface terms, and is minimal in the +// sense that removing any term changes the set of types satisfying the +// constraint. It is left as a proof for the reader that, modulo sorting, there +// is exactly one such normalized form. +// +// Because the minimal representation always takes this form, StructuralTerms +// returns a slice of tilde terms corresponding to the terms of the union in +// the normalized structural restriction. An error is returned if the +// constraint interface is invalid, exceeds complexity bounds, or has an empty +// type set. In the latter case, StructuralTerms returns ErrEmptyTypeSet. +// +// StructuralTerms makes no guarantees about the order of terms, except that it +// is deterministic. +func StructuralTerms(tparam *types.TypeParam) ([]*types.Term, error) { + constraint := tparam.Constraint() + if constraint == nil { + return nil, fmt.Errorf("%s has nil constraint", tparam) + } + iface, _ := constraint.Underlying().(*types.Interface) + if iface == nil { + return nil, fmt.Errorf("constraint is %T, not *types.Interface", constraint.Underlying()) + } + return InterfaceTermSet(iface) +} + +// InterfaceTermSet computes the normalized terms for a constraint interface, +// returning an error if the term set cannot be computed or is empty. In the +// latter case, the error will be ErrEmptyTypeSet. +// +// See the documentation of StructuralTerms for more information on +// normalization. +func InterfaceTermSet(iface *types.Interface) ([]*types.Term, error) { + return computeTermSet(iface) +} + +// UnionTermSet computes the normalized terms for a union, returning an error +// if the term set cannot be computed or is empty. In the latter case, the +// error will be ErrEmptyTypeSet. +// +// See the documentation of StructuralTerms for more information on +// normalization. +func UnionTermSet(union *types.Union) ([]*types.Term, error) { + return computeTermSet(union) +} + +func computeTermSet(typ types.Type) ([]*types.Term, error) { + tset, err := computeTermSetInternal(typ, make(map[types.Type]*termSet), 0) + if err != nil { + return nil, err + } + if tset.terms.isEmpty() { + return nil, ErrEmptyTypeSet + } + if tset.terms.isAll() { + return nil, nil + } + var terms []*types.Term + for _, term := range tset.terms { + terms = append(terms, types.NewTerm(term.tilde, term.typ)) + } + return terms, nil +} + +// A termSet holds the normalized set of terms for a given type. +// +// The name termSet is intentionally distinct from 'type set': a type set is +// all types that implement a type (and includes method restrictions), whereas +// a term set just represents the structural restrictions on a type. +type termSet struct { + complete bool + terms termlist +} + +func indentf(depth int, format string, args ...interface{}) { + fmt.Fprintf(os.Stderr, strings.Repeat(".", depth)+format+"\n", args...) +} + +func computeTermSetInternal(t types.Type, seen map[types.Type]*termSet, depth int) (res *termSet, err error) { + if t == nil { + panic("nil type") + } + + if debug { + indentf(depth, "%s", t.String()) + defer func() { + if err != nil { + indentf(depth, "=> %s", err) + } else { + indentf(depth, "=> %s", res.terms.String()) + } + }() + } + + const maxTermCount = 100 + if tset, ok := seen[t]; ok { + if !tset.complete { + return nil, fmt.Errorf("cycle detected in the declaration of %s", t) + } + return tset, nil + } + + // Mark the current type as seen to avoid infinite recursion. + tset := new(termSet) + defer func() { + tset.complete = true + }() + seen[t] = tset + + switch u := t.Underlying().(type) { + case *types.Interface: + // The term set of an interface is the intersection of the term sets of its + // embedded types. + tset.terms = allTermlist + for i := 0; i < u.NumEmbeddeds(); i++ { + embedded := u.EmbeddedType(i) + if _, ok := embedded.Underlying().(*types.TypeParam); ok { + return nil, fmt.Errorf("invalid embedded type %T", embedded) + } + tset2, err := computeTermSetInternal(embedded, seen, depth+1) + if err != nil { + return nil, err + } + tset.terms = tset.terms.intersect(tset2.terms) + } + case *types.Union: + // The term set of a union is the union of term sets of its terms. + tset.terms = nil + for i := 0; i < u.Len(); i++ { + t := u.Term(i) + var terms termlist + switch t.Type().Underlying().(type) { + case *types.Interface: + tset2, err := computeTermSetInternal(t.Type(), seen, depth+1) + if err != nil { + return nil, err + } + terms = tset2.terms + case *types.TypeParam, *types.Union: + // A stand-alone type parameter or union is not permitted as union + // term. + return nil, fmt.Errorf("invalid union term %T", t) + default: + if t.Type() == types.Typ[types.Invalid] { + continue + } + terms = termlist{{t.Tilde(), t.Type()}} + } + tset.terms = tset.terms.union(terms) + if len(tset.terms) > maxTermCount { + return nil, fmt.Errorf("exceeded max term count %d", maxTermCount) + } + } + case *types.TypeParam: + panic("unreachable") + default: + // For all other types, the term set is just a single non-tilde term + // holding the type itself. + if u != types.Typ[types.Invalid] { + tset.terms = termlist{{false, t}} + } + } + return tset, nil +} + +// under is a facade for the go/types internal function of the same name. It is +// used by typeterm.go. +func under(t types.Type) types.Type { + return t.Underlying() +} diff --git a/internal/typeparams/termlist.go b/internal/typeparams/termlist.go new file mode 100644 index 00000000..cbd12f80 --- /dev/null +++ b/internal/typeparams/termlist.go @@ -0,0 +1,163 @@ +// Copyright 2021 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. + +// Code generated by copytermlist.go DO NOT EDIT. + +package typeparams + +import ( + "bytes" + "go/types" +) + +// A termlist represents the type set represented by the union +// t1 βˆͺ y2 βˆͺ ... tn of the type sets of the terms t1 to tn. +// A termlist is in normal form if all terms are disjoint. +// termlist operations don't require the operands to be in +// normal form. +type termlist []*term + +// allTermlist represents the set of all types. +// It is in normal form. +var allTermlist = termlist{new(term)} + +// String prints the termlist exactly (without normalization). +func (xl termlist) String() string { + if len(xl) == 0 { + return "βˆ…" + } + var buf bytes.Buffer + for i, x := range xl { + if i > 0 { + buf.WriteString(" | ") + } + buf.WriteString(x.String()) + } + return buf.String() +} + +// isEmpty reports whether the termlist xl represents the empty set of types. +func (xl termlist) isEmpty() bool { + // If there's a non-nil term, the entire list is not empty. + // If the termlist is in normal form, this requires at most + // one iteration. + for _, x := range xl { + if x != nil { + return false + } + } + return true +} + +// isAll reports whether the termlist xl represents the set of all types. +func (xl termlist) isAll() bool { + // If there's a 𝓀 term, the entire list is 𝓀. + // If the termlist is in normal form, this requires at most + // one iteration. + for _, x := range xl { + if x != nil && x.typ == nil { + return true + } + } + return false +} + +// norm returns the normal form of xl. +func (xl termlist) norm() termlist { + // Quadratic algorithm, but good enough for now. + // TODO(gri) fix asymptotic performance + used := make([]bool, len(xl)) + var rl termlist + for i, xi := range xl { + if xi == nil || used[i] { + continue + } + for j := i + 1; j < len(xl); j++ { + xj := xl[j] + if xj == nil || used[j] { + continue + } + if u1, u2 := xi.union(xj); u2 == nil { + // If we encounter a 𝓀 term, the entire list is 𝓀. + // Exit early. + // (Note that this is not just an optimization; + // if we continue, we may end up with a 𝓀 term + // and other terms and the result would not be + // in normal form.) + if u1.typ == nil { + return allTermlist + } + xi = u1 + used[j] = true // xj is now unioned into xi - ignore it in future iterations + } + } + rl = append(rl, xi) + } + return rl +} + +// union returns the union xl βˆͺ yl. +func (xl termlist) union(yl termlist) termlist { + return append(xl, yl...).norm() +} + +// intersect returns the intersection xl ∩ yl. +func (xl termlist) intersect(yl termlist) termlist { + if xl.isEmpty() || yl.isEmpty() { + return nil + } + + // Quadratic algorithm, but good enough for now. + // TODO(gri) fix asymptotic performance + var rl termlist + for _, x := range xl { + for _, y := range yl { + if r := x.intersect(y); r != nil { + rl = append(rl, r) + } + } + } + return rl.norm() +} + +// equal reports whether xl and yl represent the same type set. +func (xl termlist) equal(yl termlist) bool { + // TODO(gri) this should be more efficient + return xl.subsetOf(yl) && yl.subsetOf(xl) +} + +// includes reports whether t ∈ xl. +func (xl termlist) includes(t types.Type) bool { + for _, x := range xl { + if x.includes(t) { + return true + } + } + return false +} + +// supersetOf reports whether y βŠ† xl. +func (xl termlist) supersetOf(y *term) bool { + for _, x := range xl { + if y.subsetOf(x) { + return true + } + } + return false +} + +// subsetOf reports whether xl βŠ† yl. +func (xl termlist) subsetOf(yl termlist) bool { + if yl.isEmpty() { + return xl.isEmpty() + } + + // each term x of xl must be a subset of yl + for _, x := range xl { + if !yl.supersetOf(x) { + return false // x is not a subset yl + } + } + return true +} diff --git a/internal/typeparams/typeterm.go b/internal/typeparams/typeterm.go new file mode 100644 index 00000000..7350bb70 --- /dev/null +++ b/internal/typeparams/typeterm.go @@ -0,0 +1,169 @@ +// Copyright 2021 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. + +// Code generated by copytermlist.go DO NOT EDIT. + +package typeparams + +import "go/types" + +// A term describes elementary type sets: +// +// βˆ…: (*term)(nil) == βˆ… // set of no types (empty set) +// 𝓀: &term{} == 𝓀 // set of all types (𝓀niverse) +// T: &term{false, T} == {T} // set of type T +// ~t: &term{true, t} == {t' | under(t') == t} // set of types with underlying type t +type term struct { + tilde bool // valid if typ != nil + typ types.Type +} + +func (x *term) String() string { + switch { + case x == nil: + return "βˆ…" + case x.typ == nil: + return "𝓀" + case x.tilde: + return "~" + x.typ.String() + default: + return x.typ.String() + } +} + +// equal reports whether x and y represent the same type set. +func (x *term) equal(y *term) bool { + // easy cases + switch { + case x == nil || y == nil: + return x == y + case x.typ == nil || y.typ == nil: + return x.typ == y.typ + } + // βˆ… βŠ‚ x, y βŠ‚ 𝓀 + + return x.tilde == y.tilde && types.Identical(x.typ, y.typ) +} + +// union returns the union x βˆͺ y: zero, one, or two non-nil terms. +func (x *term) union(y *term) (_, _ *term) { + // easy cases + switch { + case x == nil && y == nil: + return nil, nil // βˆ… βˆͺ βˆ… == βˆ… + case x == nil: + return y, nil // βˆ… βˆͺ y == y + case y == nil: + return x, nil // x βˆͺ βˆ… == x + case x.typ == nil: + return x, nil // 𝓀 βˆͺ y == 𝓀 + case y.typ == nil: + return y, nil // x βˆͺ 𝓀 == 𝓀 + } + // βˆ… βŠ‚ x, y βŠ‚ 𝓀 + + if x.disjoint(y) { + return x, y // x βˆͺ y == (x, y) if x ∩ y == βˆ… + } + // x.typ == y.typ + + // ~t βˆͺ ~t == ~t + // ~t βˆͺ T == ~t + // T βˆͺ ~t == ~t + // T βˆͺ T == T + if x.tilde || !y.tilde { + return x, nil + } + return y, nil +} + +// intersect returns the intersection x ∩ y. +func (x *term) intersect(y *term) *term { + // easy cases + switch { + case x == nil || y == nil: + return nil // βˆ… ∩ y == βˆ… and ∩ βˆ… == βˆ… + case x.typ == nil: + return y // 𝓀 ∩ y == y + case y.typ == nil: + return x // x ∩ 𝓀 == x + } + // βˆ… βŠ‚ x, y βŠ‚ 𝓀 + + if x.disjoint(y) { + return nil // x ∩ y == βˆ… if x ∩ y == βˆ… + } + // x.typ == y.typ + + // ~t ∩ ~t == ~t + // ~t ∩ T == T + // T ∩ ~t == T + // T ∩ T == T + if !x.tilde || y.tilde { + return x + } + return y +} + +// includes reports whether t ∈ x. +func (x *term) includes(t types.Type) bool { + // easy cases + switch { + case x == nil: + return false // t ∈ βˆ… == false + case x.typ == nil: + return true // t ∈ 𝓀 == true + } + // βˆ… βŠ‚ x βŠ‚ 𝓀 + + u := t + if x.tilde { + u = under(u) + } + return types.Identical(x.typ, u) +} + +// subsetOf reports whether x βŠ† y. +func (x *term) subsetOf(y *term) bool { + // easy cases + switch { + case x == nil: + return true // βˆ… βŠ† y == true + case y == nil: + return false // x βŠ† βˆ… == false since x != βˆ… + case y.typ == nil: + return true // x βŠ† 𝓀 == true + case x.typ == nil: + return false // 𝓀 βŠ† y == false since y != 𝓀 + } + // βˆ… βŠ‚ x, y βŠ‚ 𝓀 + + if x.disjoint(y) { + return false // x βŠ† y == false if x ∩ y == βˆ… + } + // x.typ == y.typ + + // ~t βŠ† ~t == true + // ~t βŠ† T == false + // T βŠ† ~t == true + // T βŠ† T == true + return !x.tilde || y.tilde +} + +// disjoint reports whether x ∩ y == βˆ…. +// x.typ and y.typ must not be nil. +func (x *term) disjoint(y *term) bool { + if debug && (x.typ == nil || y.typ == nil) { + panic("invalid argument(s)") + } + ux := x.typ + if y.tilde { + ux = under(ux) + } + uy := y.typ + if x.tilde { + uy = under(uy) + } + return !types.Identical(ux, uy) +} diff --git a/internal/typeutil/map.go b/internal/typeutil/map.go new file mode 100644 index 00000000..5c52363a --- /dev/null +++ b/internal/typeutil/map.go @@ -0,0 +1,518 @@ +// 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 typeutil defines various utilities for types, such as Map, +// a mapping from types.Type to interface{} values. +package typeutil // import "golang.org/x/tools/go/types/typeutil" + +import ( + "bytes" + "fmt" + "go/types" + "reflect" + + "github.com/goplus/llgo/internal/aliases" + "github.com/goplus/llgo/internal/typeparams" +) + +// Map is a hash-table-based mapping from types (types.Type) to +// arbitrary interface{} values. The concrete types that implement +// the Type interface are pointers. Since they are not canonicalized, +// == cannot be used to check for equivalence, and thus we cannot +// simply use a Go map. +// +// Just as with map[K]V, a nil *Map is a valid empty map. +// +// Not thread-safe. +type Map struct { + hasher Hasher // shared by many Maps + table map[uint32][]entry // maps hash to bucket; entry.key==nil means unused + length int // number of map entries +} + +// entry is an entry (key/value association) in a hash bucket. +type entry struct { + key types.Type + value interface{} +} + +// SetHasher sets the hasher used by Map. +// +// All Hashers are functionally equivalent but contain internal state +// used to cache the results of hashing previously seen types. +// +// A single Hasher created by MakeHasher() may be shared among many +// Maps. This is recommended if the instances have many keys in +// common, as it will amortize the cost of hash computation. +// +// A Hasher may grow without bound as new types are seen. Even when a +// type is deleted from the map, the Hasher never shrinks, since other +// types in the map may reference the deleted type indirectly. +// +// Hashers are not thread-safe, and read-only operations such as +// Map.Lookup require updates to the hasher, so a full Mutex lock (not a +// read-lock) is require around all Map operations if a shared +// hasher is accessed from multiple threads. +// +// If SetHasher is not called, the Map will create a private hasher at +// the first call to Insert. +func (m *Map) SetHasher(hasher Hasher) { + m.hasher = hasher +} + +// Delete removes the entry with the given key, if any. +// It returns true if the entry was found. +func (m *Map) Delete(key types.Type) bool { + if m != nil && m.table != nil { + hash := m.hasher.Hash(key) + bucket := m.table[hash] + for i, e := range bucket { + if e.key != nil && types.Identical(key, e.key) { + // We can't compact the bucket as it + // would disturb iterators. + bucket[i] = entry{} + m.length-- + return true + } + } + } + return false +} + +// At returns the map entry for the given key. +// The result is nil if the entry is not present. +func (m *Map) At(key types.Type) interface{} { + if m != nil && m.table != nil { + for _, e := range m.table[m.hasher.Hash(key)] { + if e.key != nil && types.Identical(key, e.key) { + return e.value + } + } + } + return nil +} + +// Set sets the map entry for key to val, +// and returns the previous entry, if any. +func (m *Map) Set(key types.Type, value interface{}) (prev interface{}) { + if m.table != nil { + hash := m.hasher.Hash(key) + bucket := m.table[hash] + var hole *entry + for i, e := range bucket { + if e.key == nil { + hole = &bucket[i] + } else if types.Identical(key, e.key) { + prev = e.value + bucket[i].value = value + return + } + } + + if hole != nil { + *hole = entry{key, value} // overwrite deleted entry + } else { + m.table[hash] = append(bucket, entry{key, value}) + } + } else { + if m.hasher.memo == nil { + m.hasher = MakeHasher() + } + hash := m.hasher.Hash(key) + m.table = map[uint32][]entry{hash: {entry{key, value}}} + } + + m.length++ + return +} + +// Len returns the number of map entries. +func (m *Map) Len() int { + if m != nil { + return m.length + } + return 0 +} + +// Iterate calls function f on each entry in the map in unspecified order. +// +// If f should mutate the map, Iterate provides the same guarantees as +// Go maps: if f deletes a map entry that Iterate has not yet reached, +// f will not be invoked for it, but if f inserts a map entry that +// Iterate has not yet reached, whether or not f will be invoked for +// it is unspecified. +func (m *Map) Iterate(f func(key types.Type, value interface{})) { + if m != nil { + for _, bucket := range m.table { + for _, e := range bucket { + if e.key != nil { + f(e.key, e.value) + } + } + } + } +} + +// Keys returns a new slice containing the set of map keys. +// The order is unspecified. +func (m *Map) Keys() []types.Type { + keys := make([]types.Type, 0, m.Len()) + m.Iterate(func(key types.Type, _ interface{}) { + keys = append(keys, key) + }) + return keys +} + +func (m *Map) toString(values bool) string { + if m == nil { + return "{}" + } + var buf bytes.Buffer + fmt.Fprint(&buf, "{") + sep := "" + m.Iterate(func(key types.Type, value interface{}) { + fmt.Fprint(&buf, sep) + sep = ", " + fmt.Fprint(&buf, key) + if values { + fmt.Fprintf(&buf, ": %q", value) + } + }) + fmt.Fprint(&buf, "}") + return buf.String() +} + +// String returns a string representation of the map's entries. +// Values are printed using fmt.Sprintf("%v", v). +// Order is unspecified. +func (m *Map) String() string { + return m.toString(true) +} + +// KeysString returns a string representation of the map's key set. +// Order is unspecified. +func (m *Map) KeysString() string { + return m.toString(false) +} + +//////////////////////////////////////////////////////////////////////// +// Hasher + +// A Hasher maps each type to its hash value. +// For efficiency, a hasher uses memoization; thus its memory +// footprint grows monotonically over time. +// Hashers are not thread-safe. +// Hashers have reference semantics. +// Call MakeHasher to create a Hasher. +type Hasher struct { + memo map[types.Type]uint32 + + // ptrMap records pointer identity. + ptrMap map[interface{}]uint32 + + // sigTParams holds type parameters from the signature being hashed. + // Signatures are considered identical modulo renaming of type parameters, so + // within the scope of a signature type the identity of the signature's type + // parameters is just their index. + // + // Since the language does not currently support referring to uninstantiated + // generic types or functions, and instantiated signatures do not have type + // parameter lists, we should never encounter a second non-empty type + // parameter list when hashing a generic signature. + sigTParams *types.TypeParamList +} + +// MakeHasher returns a new Hasher instance. +func MakeHasher() Hasher { + return Hasher{ + memo: make(map[types.Type]uint32), + ptrMap: make(map[interface{}]uint32), + sigTParams: nil, + } +} + +// Hash computes a hash value for the given type t such that +// Identical(t, t') => Hash(t) == Hash(t'). +func (h Hasher) Hash(t types.Type) uint32 { + hash, ok := h.memo[t] + if !ok { + hash = h.hashFor(t) + h.memo[t] = hash + } + return hash +} + +// hashString computes the Fowler–Noll–Vo hash of s. +func hashString(s string) uint32 { + var h uint32 + for i := 0; i < len(s); i++ { + h ^= uint32(s[i]) + h *= 16777619 + } + return h +} + +// hashFor computes the hash of t. +func (h Hasher) hashFor(t types.Type) uint32 { + // See Identical for rationale. + switch t := t.(type) { + case *types.Basic: + return uint32(t.Kind()) + + case *aliases.Alias: + return h.Hash(t.Underlying()) + + case *types.Array: + return 9043 + 2*uint32(t.Len()) + 3*h.Hash(t.Elem()) + + case *types.Slice: + return 9049 + 2*h.Hash(t.Elem()) + + case *types.Struct: + var hash uint32 = 9059 + for i, n := 0, t.NumFields(); i < n; i++ { + f := t.Field(i) + if f.Anonymous() { + hash += 8861 + } + hash += hashString(t.Tag(i)) + hash += hashString(f.Name()) // (ignore f.Pkg) + hash += h.Hash(f.Type()) + } + return hash + + case *types.Pointer: + return 9067 + 2*h.Hash(t.Elem()) + + case *types.Signature: + var hash uint32 = 9091 + if t.Variadic() { + hash *= 8863 + } + + // Use a separate hasher for types inside of the signature, where type + // parameter identity is modified to be (index, constraint). We must use a + // new memo for this hasher as type identity may be affected by this + // masking. For example, in func[T any](*T), the identity of *T depends on + // whether we are mapping the argument in isolation, or recursively as part + // of hashing the signature. + // + // We should never encounter a generic signature while hashing another + // generic signature, but defensively set sigTParams only if h.mask is + // unset. + tparams := t.TypeParams() + if h.sigTParams == nil && tparams.Len() != 0 { + h = Hasher{ + // There may be something more efficient than discarding the existing + // memo, but it would require detecting whether types are 'tainted' by + // references to type parameters. + memo: make(map[types.Type]uint32), + // Re-using ptrMap ensures that pointer identity is preserved in this + // hasher. + ptrMap: h.ptrMap, + sigTParams: tparams, + } + } + + for i := 0; i < tparams.Len(); i++ { + tparam := tparams.At(i) + hash += 7 * h.Hash(tparam.Constraint()) + } + + return hash + 3*h.hashTuple(t.Params()) + 5*h.hashTuple(t.Results()) + + case *types.Union: + return h.hashUnion(t) + + case *types.Interface: + // Interfaces are identical if they have the same set of methods, with + // identical names and types, and they have the same set of type + // restrictions. See go/types.identical for more details. + var hash uint32 = 9103 + + // Hash methods. + for i, n := 0, t.NumMethods(); i < n; i++ { + // Method order is not significant. + // Ignore m.Pkg(). + m := t.Method(i) + // Use shallow hash on method signature to + // avoid anonymous interface cycles. + hash += 3*hashString(m.Name()) + 5*h.shallowHash(m.Type()) + } + + // Hash type restrictions. + terms, err := typeparams.InterfaceTermSet(t) + // if err != nil t has invalid type restrictions. + if err == nil { + hash += h.hashTermSet(terms) + } + + return hash + + case *types.Map: + return 9109 + 2*h.Hash(t.Key()) + 3*h.Hash(t.Elem()) + + case *types.Chan: + return 9127 + 2*uint32(t.Dir()) + 3*h.Hash(t.Elem()) + + case *types.Named: + hash := h.hashPtr(t.Obj()) + targs := t.TypeArgs() + for i := 0; i < targs.Len(); i++ { + targ := targs.At(i) + hash += 2 * h.Hash(targ) + } + return hash + + case *types.TypeParam: + return h.hashTypeParam(t) + + case *types.Tuple: + return h.hashTuple(t) + } + + panic(fmt.Sprintf("%T: %v", t, t)) +} + +func (h Hasher) hashTuple(tuple *types.Tuple) uint32 { + // See go/types.identicalTypes for rationale. + n := tuple.Len() + hash := 9137 + 2*uint32(n) + for i := 0; i < n; i++ { + hash += 3 * h.Hash(tuple.At(i).Type()) + } + return hash +} + +func (h Hasher) hashUnion(t *types.Union) uint32 { + // Hash type restrictions. + terms, err := typeparams.UnionTermSet(t) + // if err != nil t has invalid type restrictions. Fall back on a non-zero + // hash. + if err != nil { + return 9151 + } + return h.hashTermSet(terms) +} + +func (h Hasher) hashTermSet(terms []*types.Term) uint32 { + hash := 9157 + 2*uint32(len(terms)) + for _, term := range terms { + // term order is not significant. + termHash := h.Hash(term.Type()) + if term.Tilde() { + termHash *= 9161 + } + hash += 3 * termHash + } + return hash +} + +// hashTypeParam returns a hash of the type parameter t, with a hash value +// depending on whether t is contained in h.sigTParams. +// +// If h.sigTParams is set and contains t, then we are in the process of hashing +// a signature, and the hash value of t must depend only on t's index and +// constraint: signatures are considered identical modulo type parameter +// renaming. To avoid infinite recursion, we only hash the type parameter +// index, and rely on types.Identical to handle signatures where constraints +// are not identical. +// +// Otherwise the hash of t depends only on t's pointer identity. +func (h Hasher) hashTypeParam(t *types.TypeParam) uint32 { + if h.sigTParams != nil { + i := t.Index() + if i >= 0 && i < h.sigTParams.Len() && t == h.sigTParams.At(i) { + return 9173 + 3*uint32(i) + } + } + return h.hashPtr(t.Obj()) +} + +// hashPtr hashes the pointer identity of ptr. It uses h.ptrMap to ensure that +// pointers values are not dependent on the GC. +func (h Hasher) hashPtr(ptr interface{}) uint32 { + if hash, ok := h.ptrMap[ptr]; ok { + return hash + } + hash := uint32(reflect.ValueOf(ptr).Pointer()) + h.ptrMap[ptr] = hash + return hash +} + +// shallowHash computes a hash of t without looking at any of its +// element Types, to avoid potential anonymous cycles in the types of +// interface methods. +// +// When an unnamed non-empty interface type appears anywhere among the +// arguments or results of an interface method, there is a potential +// for endless recursion. Consider: +// +// type X interface { m() []*interface { X } } +// +// The problem is that the Methods of the interface in m's result type +// include m itself; there is no mention of the named type X that +// might help us break the cycle. +// (See comment in go/types.identical, case *Interface, for more.) +func (h Hasher) shallowHash(t types.Type) uint32 { + // t is the type of an interface method (Signature), + // its params or results (Tuples), or their immediate + // elements (mostly Slice, Pointer, Basic, Named), + // so there's no need to optimize anything else. + switch t := t.(type) { + case *aliases.Alias: + return h.shallowHash(t.Underlying()) + + case *types.Signature: + var hash uint32 = 604171 + if t.Variadic() { + hash *= 971767 + } + // The Signature/Tuple recursion is always finite + // and invariably shallow. + return hash + 1062599*h.shallowHash(t.Params()) + 1282529*h.shallowHash(t.Results()) + + case *types.Tuple: + n := t.Len() + hash := 9137 + 2*uint32(n) + for i := 0; i < n; i++ { + hash += 53471161 * h.shallowHash(t.At(i).Type()) + } + return hash + + case *types.Basic: + return 45212177 * uint32(t.Kind()) + + case *types.Array: + return 1524181 + 2*uint32(t.Len()) + + case *types.Slice: + return 2690201 + + case *types.Struct: + return 3326489 + + case *types.Pointer: + return 4393139 + + case *types.Union: + return 562448657 + + case *types.Interface: + return 2124679 // no recursion here + + case *types.Map: + return 9109 + + case *types.Chan: + return 9127 + + case *types.Named: + return h.hashPtr(t.Obj()) + + case *types.TypeParam: + return h.hashPtr(t.Obj()) + } + panic(fmt.Sprintf("shallowHash: %T: %v", t, t)) +} diff --git a/ssa/package.go b/ssa/package.go index 985a155b..0d3800cb 100644 --- a/ssa/package.go +++ b/ssa/package.go @@ -20,8 +20,8 @@ import ( "go/types" "runtime" + "github.com/goplus/llgo/internal/typeutil" "github.com/goplus/llvm" - "golang.org/x/tools/go/types/typeutil" ) const (