summaryrefslogtreecommitdiffstats
path: root/vendor/modernc.org/cc/v3/operand.go
diff options
context:
space:
mode:
authorWim <wim@42.be>2022-01-31 00:27:37 +0100
committerWim <wim@42.be>2022-03-20 14:57:48 +0100
commite3cafeaf9292f67459ff1d186f68283bfaedf2ae (patch)
treeb69c39620aa91dba695b3b935c6651c0fb37ce75 /vendor/modernc.org/cc/v3/operand.go
parente7b193788a56ee7cdb02a87a9db0ad6724ef66d5 (diff)
downloadmatterbridge-msglm-e3cafeaf9292f67459ff1d186f68283bfaedf2ae.tar.gz
matterbridge-msglm-e3cafeaf9292f67459ff1d186f68283bfaedf2ae.tar.bz2
matterbridge-msglm-e3cafeaf9292f67459ff1d186f68283bfaedf2ae.zip
Add dependencies/vendor (whatsapp)
Diffstat (limited to 'vendor/modernc.org/cc/v3/operand.go')
-rw-r--r--vendor/modernc.org/cc/v3/operand.go1337
1 files changed, 1337 insertions, 0 deletions
diff --git a/vendor/modernc.org/cc/v3/operand.go b/vendor/modernc.org/cc/v3/operand.go
new file mode 100644
index 00000000..24c46616
--- /dev/null
+++ b/vendor/modernc.org/cc/v3/operand.go
@@ -0,0 +1,1337 @@
+// Copyright 2019 The CC 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 cc // import "modernc.org/cc/v3"
+
+import (
+ "fmt"
+ "math"
+ "math/big"
+)
+
+var (
+ _ Value = (*Float128Value)(nil)
+ _ Value = (*InitializerValue)(nil)
+ _ Value = Complex128Value(0)
+ _ Value = Complex256Value{}
+ _ Value = Complex64Value(0)
+ _ Value = Float32Value(0)
+ _ Value = Float64Value(0)
+ _ Value = Int64Value(0)
+ _ Value = StringValue(0)
+ _ Value = Uint64Value(0)
+ _ Value = WideStringValue(0)
+
+ _ Operand = (*funcDesignator)(nil)
+ _ Operand = (*lvalue)(nil)
+ _ Operand = (*operand)(nil)
+ _ Operand = noOperand
+
+ noOperand = &operand{typ: noType}
+)
+
+type Operand interface {
+ // IsAssingmentCompatible reports whether the operand can be
+ // assigned to lhs. [0], 6.5.16.1.
+ IsAssingmentCompatible(lhs Type) bool
+ ConvertTo(Type) Operand
+ Declarator() *Declarator
+ IsConst() bool
+ IsLValue() bool
+ IsNonZero() bool
+ IsZero() bool
+ Offset() uintptr // Valid only for non nil Declarator() value
+ Type() Type
+ Value() Value
+ convertFromInt(*context, Node, Type) Operand
+ convertTo(*context, Node, Type) Operand
+ convertToInt(*context, Node, Type) Operand
+ getABI() *ABI
+ integerPromotion(*context, Node) Operand
+ normalize(*context, Node) Operand
+}
+
+type Value interface {
+ IsConst() bool
+ IsNonZero() bool
+ IsZero() bool
+ add(b Value) Value
+ and(b Value) Value
+ cpl() Value
+ div(b Value) Value
+ eq(b Value) Value
+ ge(b Value) Value
+ gt(b Value) Value
+ le(b Value) Value
+ lsh(b Value) Value
+ lt(b Value) Value
+ mod(b Value) Value
+ mul(b Value) Value
+ neg() Value
+ neq(b Value) Value
+ or(b Value) Value
+ rsh(b Value) Value
+ sub(b Value) Value
+ xor(b Value) Value
+}
+
+type WideStringValue StringID
+
+func (v WideStringValue) add(b Value) Value { panic(todo("")) }
+func (v WideStringValue) and(b Value) Value { panic(todo("")) }
+func (v WideStringValue) cpl() Value { panic(todo("")) }
+func (v WideStringValue) div(b Value) Value { panic(todo("")) }
+func (v WideStringValue) eq(b Value) Value { return boolValue(v == b.(WideStringValue)) }
+func (v WideStringValue) IsConst() bool { return true }
+func (v WideStringValue) IsNonZero() bool { return true }
+func (v WideStringValue) IsZero() bool { return false }
+func (v WideStringValue) lsh(b Value) Value { panic(todo("")) }
+func (v WideStringValue) mod(b Value) Value { panic(todo("")) }
+func (v WideStringValue) mul(b Value) Value { panic(todo("")) }
+func (v WideStringValue) neg() Value { panic(todo("")) }
+func (v WideStringValue) neq(b Value) Value { return boolValue(v != b.(WideStringValue)) }
+func (v WideStringValue) or(b Value) Value { panic(todo("")) }
+func (v WideStringValue) rsh(b Value) Value { panic(todo("")) }
+func (v WideStringValue) sub(b Value) Value { panic(todo("")) }
+func (v WideStringValue) xor(b Value) Value { panic(todo("")) }
+
+func (v WideStringValue) le(b Value) Value {
+ return boolValue(StringID(v).String() <= StringID(b.(WideStringValue)).String())
+}
+
+func (v WideStringValue) ge(b Value) Value {
+ return boolValue(StringID(v).String() >= StringID(b.(WideStringValue)).String())
+}
+
+func (v WideStringValue) gt(b Value) Value {
+ return boolValue(StringID(v).String() > StringID(b.(WideStringValue)).String())
+}
+
+func (v WideStringValue) lt(b Value) Value {
+ return boolValue(StringID(v).String() < StringID(b.(WideStringValue)).String())
+}
+
+type StringValue StringID
+
+func (v StringValue) add(b Value) Value { panic(todo("")) }
+func (v StringValue) and(b Value) Value { panic(todo("")) }
+func (v StringValue) cpl() Value { panic(todo("")) }
+func (v StringValue) div(b Value) Value { panic(todo("")) }
+func (v StringValue) eq(b Value) Value { return boolValue(v == b.(StringValue)) }
+func (v StringValue) IsConst() bool { return true }
+func (v StringValue) IsNonZero() bool { return true }
+func (v StringValue) IsZero() bool { return false }
+func (v StringValue) lsh(b Value) Value { panic(todo("")) }
+func (v StringValue) mod(b Value) Value { panic(todo("")) }
+func (v StringValue) mul(b Value) Value { panic(todo("")) }
+func (v StringValue) neg() Value { panic(todo("")) }
+func (v StringValue) neq(b Value) Value { return boolValue(v != b.(StringValue)) }
+func (v StringValue) or(b Value) Value { panic(todo("")) }
+func (v StringValue) rsh(b Value) Value { panic(todo("")) }
+func (v StringValue) sub(b Value) Value { panic(todo("")) }
+func (v StringValue) xor(b Value) Value { panic(todo("")) }
+
+func (v StringValue) le(b Value) Value {
+ return boolValue(StringID(v).String() <= StringID(b.(StringValue)).String())
+}
+
+func (v StringValue) ge(b Value) Value {
+ return boolValue(StringID(v).String() >= StringID(b.(StringValue)).String())
+}
+
+func (v StringValue) gt(b Value) Value {
+ return boolValue(StringID(v).String() > StringID(b.(StringValue)).String())
+}
+
+func (v StringValue) lt(b Value) Value {
+ return boolValue(StringID(v).String() < StringID(b.(StringValue)).String())
+}
+
+type Int64Value int64
+
+func (v Int64Value) add(b Value) Value { return v + b.(Int64Value) }
+func (v Int64Value) and(b Value) Value { return v & b.(Int64Value) }
+func (v Int64Value) cpl() Value { return ^v }
+func (v Int64Value) eq(b Value) Value { return boolValue(v == b.(Int64Value)) }
+func (v Int64Value) ge(b Value) Value { return boolValue(v >= b.(Int64Value)) }
+func (v Int64Value) gt(b Value) Value { return boolValue(v > b.(Int64Value)) }
+func (v Int64Value) IsConst() bool { return true }
+func (v Int64Value) IsNonZero() bool { return v != 0 }
+func (v Int64Value) IsZero() bool { return v == 0 }
+func (v Int64Value) le(b Value) Value { return boolValue(v <= b.(Int64Value)) }
+func (v Int64Value) lt(b Value) Value { return boolValue(v < b.(Int64Value)) }
+func (v Int64Value) mul(b Value) Value { return v * b.(Int64Value) }
+func (v Int64Value) neg() Value { return -v }
+func (v Int64Value) neq(b Value) Value { return boolValue(v != b.(Int64Value)) }
+func (v Int64Value) or(b Value) Value { return v | b.(Int64Value) }
+func (v Int64Value) sub(b Value) Value { return v - b.(Int64Value) }
+func (v Int64Value) xor(b Value) Value { return v ^ b.(Int64Value) }
+
+func (v Int64Value) div(b Value) Value {
+ if b.IsZero() {
+ return nil
+ }
+
+ return v / b.(Int64Value)
+}
+
+func (v Int64Value) lsh(b Value) Value {
+ switch y := b.(type) {
+ case Int64Value:
+ return v << uint64(y)
+ case Uint64Value:
+ return v << y
+ default:
+ panic(todo(""))
+ }
+}
+
+func (v Int64Value) rsh(b Value) Value {
+ switch y := b.(type) {
+ case Int64Value:
+ return v >> uint64(y)
+ case Uint64Value:
+ return v >> y
+ default:
+ panic(todo(""))
+ }
+}
+
+func (v Int64Value) mod(b Value) Value {
+ if b.IsZero() {
+ return nil
+ }
+
+ return v % b.(Int64Value)
+}
+
+type Uint64Value uint64
+
+func (v Uint64Value) add(b Value) Value { return v + b.(Uint64Value) }
+func (v Uint64Value) and(b Value) Value { return v & b.(Uint64Value) }
+func (v Uint64Value) cpl() Value { return ^v }
+func (v Uint64Value) eq(b Value) Value { return boolValue(v == b.(Uint64Value)) }
+func (v Uint64Value) ge(b Value) Value { return boolValue(v >= b.(Uint64Value)) }
+func (v Uint64Value) gt(b Value) Value { return boolValue(v > b.(Uint64Value)) }
+func (v Uint64Value) IsConst() bool { return true }
+func (v Uint64Value) IsNonZero() bool { return v != 0 }
+func (v Uint64Value) IsZero() bool { return v == 0 }
+func (v Uint64Value) le(b Value) Value { return boolValue(v <= b.(Uint64Value)) }
+func (v Uint64Value) lt(b Value) Value { return boolValue(v < b.(Uint64Value)) }
+func (v Uint64Value) mul(b Value) Value { return v * b.(Uint64Value) }
+func (v Uint64Value) neg() Value { return -v }
+func (v Uint64Value) neq(b Value) Value { return boolValue(v != b.(Uint64Value)) }
+func (v Uint64Value) or(b Value) Value { return v | b.(Uint64Value) }
+func (v Uint64Value) sub(b Value) Value { return v - b.(Uint64Value) }
+func (v Uint64Value) xor(b Value) Value { return v ^ b.(Uint64Value) }
+
+func (v Uint64Value) div(b Value) Value {
+ if b.IsZero() {
+ return nil
+ }
+
+ return v / b.(Uint64Value)
+}
+
+func (v Uint64Value) lsh(b Value) Value {
+ switch y := b.(type) {
+ case Int64Value:
+ return v << uint64(y)
+ case Uint64Value:
+ return v << y
+ default:
+ panic(todo(""))
+ }
+}
+
+func (v Uint64Value) rsh(b Value) Value {
+ switch y := b.(type) {
+ case Int64Value:
+ return v >> uint64(y)
+ case Uint64Value:
+ return v >> y
+ default:
+ panic(todo(""))
+ }
+}
+
+func (v Uint64Value) mod(b Value) Value {
+ if b.IsZero() {
+ return nil
+ }
+
+ return v % b.(Uint64Value)
+}
+
+type Float32Value float32
+
+func (v Float32Value) add(b Value) Value { return v + b.(Float32Value) }
+func (v Float32Value) and(b Value) Value { panic(todo("")) }
+func (v Float32Value) cpl() Value { panic(todo("")) }
+func (v Float32Value) div(b Value) Value { return v / b.(Float32Value) }
+func (v Float32Value) eq(b Value) Value { return boolValue(v == b.(Float32Value)) }
+func (v Float32Value) ge(b Value) Value { return boolValue(v >= b.(Float32Value)) }
+func (v Float32Value) gt(b Value) Value { return boolValue(v > b.(Float32Value)) }
+func (v Float32Value) IsConst() bool { return true }
+func (v Float32Value) IsNonZero() bool { return v != 0 }
+func (v Float32Value) IsZero() bool { return !math.Signbit(float64(v)) && v == 0 }
+func (v Float32Value) le(b Value) Value { return boolValue(v <= b.(Float32Value)) }
+func (v Float32Value) lsh(b Value) Value { panic(todo("")) }
+func (v Float32Value) lt(b Value) Value { return boolValue(v < b.(Float32Value)) }
+func (v Float32Value) mod(b Value) Value { panic(todo("")) }
+func (v Float32Value) mul(b Value) Value { return v * b.(Float32Value) }
+func (v Float32Value) neg() Value { return -v }
+func (v Float32Value) neq(b Value) Value { return boolValue(v != b.(Float32Value)) }
+func (v Float32Value) or(b Value) Value { panic(todo("")) }
+func (v Float32Value) rsh(b Value) Value { panic(todo("")) }
+func (v Float32Value) sub(b Value) Value { return v - b.(Float32Value) }
+func (v Float32Value) xor(b Value) Value { panic(todo("")) }
+
+type Float64Value float64
+
+func (v Float64Value) add(b Value) Value { return v + b.(Float64Value) }
+func (v Float64Value) and(b Value) Value { panic(todo("")) }
+func (v Float64Value) cpl() Value { panic(todo("")) }
+func (v Float64Value) div(b Value) Value { return v / b.(Float64Value) }
+func (v Float64Value) eq(b Value) Value { return boolValue(v == b.(Float64Value)) }
+func (v Float64Value) ge(b Value) Value { return boolValue(v >= b.(Float64Value)) }
+func (v Float64Value) gt(b Value) Value { return boolValue(v > b.(Float64Value)) }
+func (v Float64Value) IsConst() bool { return true }
+func (v Float64Value) IsNonZero() bool { return v != 0 }
+func (v Float64Value) IsZero() bool { return !math.Signbit(float64(v)) && v == 0 }
+func (v Float64Value) le(b Value) Value { return boolValue(v <= b.(Float64Value)) }
+func (v Float64Value) lsh(b Value) Value { panic(todo("")) }
+func (v Float64Value) lt(b Value) Value { return boolValue(v < b.(Float64Value)) }
+func (v Float64Value) mod(b Value) Value { panic(todo("")) }
+func (v Float64Value) mul(b Value) Value { return v * b.(Float64Value) }
+func (v Float64Value) neg() Value { return -v }
+func (v Float64Value) neq(b Value) Value { return boolValue(v != b.(Float64Value)) }
+func (v Float64Value) or(b Value) Value { panic(todo("")) }
+func (v Float64Value) rsh(b Value) Value { panic(todo("")) }
+func (v Float64Value) sub(b Value) Value { return v - b.(Float64Value) }
+func (v Float64Value) xor(b Value) Value { panic(todo("")) }
+
+var float128Zero = &Float128Value{N: big.NewFloat(0)}
+
+type Float128Value struct {
+ N *big.Float
+ NaN bool
+}
+
+func (v *Float128Value) add(b Value) Value { return v.safe(b, func(x, y *big.Float) { x.Add(x, y) }) }
+func (v *Float128Value) and(b Value) Value { panic(todo("")) }
+func (v *Float128Value) cpl() Value { panic(todo("")) }
+func (v *Float128Value) div(b Value) Value { return v.safe(b, func(x, y *big.Float) { x.Quo(x, y) }) }
+func (v *Float128Value) eq(b Value) Value { panic(todo("")) }
+func (v *Float128Value) ge(b Value) Value { panic(todo("")) }
+func (v *Float128Value) gt(b Value) Value { return boolValue(v.cmp(b, -1, 0)) }
+func (v *Float128Value) IsNonZero() bool { panic(todo("")) }
+func (v *Float128Value) IsConst() bool { return true }
+func (v *Float128Value) IsZero() bool { return !v.NaN && !v.N.Signbit() && v.cmp(float128Zero, 0) }
+func (v *Float128Value) le(b Value) Value { panic(todo("")) }
+func (v *Float128Value) lsh(b Value) Value { panic(todo("")) }
+func (v *Float128Value) lt(b Value) Value { panic(todo("")) }
+func (v *Float128Value) mod(b Value) Value { panic(todo("")) }
+func (v *Float128Value) mul(b Value) Value { return v.safe(b, func(x, y *big.Float) { x.Mul(x, y) }) }
+func (v *Float128Value) neg() Value { return v.safe(nil, func(x, y *big.Float) { x.Neg(x) }) }
+func (v *Float128Value) neq(b Value) Value { panic(todo("")) }
+func (v *Float128Value) or(b Value) Value { panic(todo("")) }
+func (v *Float128Value) rsh(b Value) Value { panic(todo("")) }
+func (v *Float128Value) sub(b Value) Value { return v.safe(b, func(x, y *big.Float) { x.Sub(x, y) }) }
+func (v *Float128Value) xor(b Value) Value { panic(todo("")) }
+
+func (v *Float128Value) cmp(b Value, accept ...int) bool {
+ w := b.(*Float128Value)
+ if v.NaN || w.NaN {
+ return false
+ }
+
+ x := v.N.Cmp(w.N)
+ for _, v := range accept {
+ if v == x {
+ return true
+ }
+ }
+ return false
+}
+
+func (v *Float128Value) String() string {
+ switch {
+ case v == nil:
+ return "<nil>"
+ case v.NaN:
+ return "NaN"
+ default:
+ return fmt.Sprint(v.N)
+ }
+}
+
+func (v *Float128Value) safe(b Value, f func(*big.Float, *big.Float)) (ret Value) {
+ var w *Float128Value
+ if b != nil {
+ w = b.(*Float128Value)
+ }
+ if v.NaN || w != nil && w.NaN {
+ return &Float128Value{NaN: true}
+ }
+
+ r := &Float128Value{}
+
+ defer func() {
+ switch x := recover().(type) {
+ case big.ErrNaN:
+ r.N = nil
+ r.NaN = true
+ ret = r
+ case nil:
+ // ok
+ default:
+ panic(x)
+ }
+ }()
+
+ r.N = big.NewFloat(0).SetPrec(0).Set(v.N)
+ var wn *big.Float
+ if w != nil {
+ wn = w.N
+ }
+ f(r.N, wn)
+ return r
+}
+
+type Complex64Value complex64
+
+func (v Complex64Value) add(b Value) Value { return v + b.(Complex64Value) }
+func (v Complex64Value) and(b Value) Value { panic(todo("")) }
+func (v Complex64Value) cpl() Value { panic(todo("")) }
+func (v Complex64Value) div(b Value) Value { return v / b.(Complex64Value) }
+func (v Complex64Value) eq(b Value) Value { return boolValue(v == b.(Complex64Value)) }
+func (v Complex64Value) ge(b Value) Value { panic(todo("")) }
+func (v Complex64Value) gt(b Value) Value { panic(todo("")) }
+func (v Complex64Value) IsConst() bool { return true }
+func (v Complex64Value) IsNonZero() bool { return v != 0 }
+func (v Complex64Value) IsZero() bool { return v == 0 }
+func (v Complex64Value) le(b Value) Value { panic(todo("")) }
+func (v Complex64Value) lsh(b Value) Value { panic(todo("")) }
+func (v Complex64Value) lt(b Value) Value { panic(todo("")) }
+func (v Complex64Value) mod(b Value) Value { panic(todo("")) }
+func (v Complex64Value) mul(b Value) Value { return v * b.(Complex64Value) }
+func (v Complex64Value) neg() Value { return -v }
+func (v Complex64Value) neq(b Value) Value { return boolValue(v != b.(Complex64Value)) }
+func (v Complex64Value) or(b Value) Value { panic(todo("")) }
+func (v Complex64Value) rsh(b Value) Value { panic(todo("")) }
+func (v Complex64Value) sub(b Value) Value { return v - b.(Complex64Value) }
+func (v Complex64Value) xor(b Value) Value { panic(todo("")) }
+
+type Complex128Value complex128
+
+func (v Complex128Value) add(b Value) Value { return v + b.(Complex128Value) }
+func (v Complex128Value) and(b Value) Value { panic(todo("")) }
+func (v Complex128Value) cpl() Value { panic(todo("")) }
+func (v Complex128Value) div(b Value) Value { return v / b.(Complex128Value) }
+func (v Complex128Value) eq(b Value) Value { return boolValue(v == b.(Complex128Value)) }
+func (v Complex128Value) ge(b Value) Value { panic(todo("")) }
+func (v Complex128Value) gt(b Value) Value { panic(todo("")) }
+func (v Complex128Value) IsConst() bool { return true }
+func (v Complex128Value) IsNonZero() bool { return v != 0 }
+func (v Complex128Value) IsZero() bool { return v == 0 }
+func (v Complex128Value) le(b Value) Value { panic(todo("")) }
+func (v Complex128Value) lsh(b Value) Value { panic(todo("")) }
+func (v Complex128Value) lt(b Value) Value { panic(todo("")) }
+func (v Complex128Value) mod(b Value) Value { panic(todo("")) }
+func (v Complex128Value) mul(b Value) Value { return v * b.(Complex128Value) }
+func (v Complex128Value) neg() Value { return -v }
+func (v Complex128Value) neq(b Value) Value { return boolValue(v != b.(Complex128Value)) }
+func (v Complex128Value) or(b Value) Value { panic(todo("")) }
+func (v Complex128Value) rsh(b Value) Value { panic(todo("")) }
+func (v Complex128Value) sub(b Value) Value { return v - b.(Complex128Value) }
+func (v Complex128Value) xor(b Value) Value { panic(todo("")) }
+
+type Complex256Value struct {
+ Re, Im *Float128Value
+}
+
+func (v Complex256Value) add(b Value) Value {
+ w := b.(Complex256Value)
+ return Complex256Value{v.Re.add(w.Re).(*Float128Value), v.Im.add(w.Im).(*Float128Value)}
+}
+
+func (v Complex256Value) and(b Value) Value { panic(todo("")) }
+func (v Complex256Value) cpl() Value { panic(todo("")) }
+func (v Complex256Value) div(b Value) Value { panic(todo("")) }
+func (v Complex256Value) eq(b Value) Value { panic(todo("")) }
+func (v Complex256Value) ge(b Value) Value { panic(todo("")) }
+func (v Complex256Value) gt(b Value) Value { panic(todo("")) }
+func (v Complex256Value) IsConst() bool { return true }
+func (v Complex256Value) IsNonZero() bool { panic(todo("")) }
+func (v Complex256Value) IsZero() bool { return v.Re.IsZero() && v.Im.IsZero() }
+func (v Complex256Value) le(b Value) Value { panic(todo("")) }
+func (v Complex256Value) lsh(b Value) Value { panic(todo("")) }
+func (v Complex256Value) lt(b Value) Value { panic(todo("")) }
+func (v Complex256Value) mod(b Value) Value { panic(todo("")) }
+func (v Complex256Value) mul(b Value) Value { panic(todo("")) }
+func (v Complex256Value) neg() Value { panic(todo("")) }
+func (v Complex256Value) neq(b Value) Value { panic(todo("")) }
+func (v Complex256Value) or(b Value) Value { panic(todo("")) }
+func (v Complex256Value) rsh(b Value) Value { panic(todo("")) }
+func (v Complex256Value) sub(b Value) Value { panic(todo("")) }
+func (v Complex256Value) xor(b Value) Value { panic(todo("")) }
+
+type lvalue struct {
+ Operand
+ declarator *Declarator
+}
+
+func (o *lvalue) ConvertTo(to Type) (r Operand) { return o.convertTo(nil, nil, to) }
+func (o *lvalue) Declarator() *Declarator { return o.declarator }
+func (o *lvalue) IsLValue() bool { return true }
+
+func (o *lvalue) IsConst() bool {
+ if v := o.Value(); v != nil {
+ return v.IsConst()
+ }
+
+ d := o.Declarator()
+ return d != nil && (d.Linkage != None || d.IsStatic())
+}
+
+func (o *lvalue) convertTo(ctx *context, n Node, to Type) (r Operand) {
+ return &lvalue{Operand: o.Operand.convertTo(ctx, n, to), declarator: o.declarator}
+}
+
+type funcDesignator struct {
+ Operand
+ declarator *Declarator
+}
+
+func (o *funcDesignator) ConvertTo(to Type) (r Operand) { return o.convertTo(nil, nil, to) }
+func (o *funcDesignator) Declarator() *Declarator { return o.declarator }
+func (o *funcDesignator) IsLValue() bool { return false }
+func (o *funcDesignator) IsConst() bool { return true }
+
+func (o *funcDesignator) convertTo(ctx *context, n Node, to Type) (r Operand) {
+ return &lvalue{Operand: o.Operand.convertTo(ctx, n, to), declarator: o.declarator}
+}
+
+type operand struct {
+ abi *ABI
+ typ Type
+ value Value
+ offset uintptr
+}
+
+func (o *operand) ConvertTo(to Type) (r Operand) { return o.convertTo(nil, nil, to) }
+func (o *operand) Declarator() *Declarator { return nil }
+func (o *operand) Offset() uintptr { return o.offset }
+func (o *operand) IsLValue() bool { return false }
+func (o *operand) IsNonZero() bool { return o.value != nil && o.value.IsNonZero() }
+func (o *operand) IsZero() bool { return o.value != nil && o.value.IsZero() }
+func (o *operand) Type() Type { return o.typ }
+func (o *operand) Value() Value { return o.value }
+func (o *operand) getABI() *ABI { return o.abi }
+
+// IsAssingmentCompatible implements Operand.
+func (o *operand) IsAssingmentCompatible(lhs Type) bool { return lhs.isAssingmentCompatibleOperand(o) }
+
+func (o *operand) IsConst() bool {
+ if v := o.Value(); v != nil {
+ return v.IsConst()
+ }
+
+ d := o.Declarator()
+ return d != nil && (d.Linkage != None || d.IsStatic())
+}
+
+// [0]6.3.1.8
+//
+// Many operators that expect operands of arithmetic type cause conversions and
+// yield result types in a similar way. The purpose is to determine a common
+// real type for the operands and result. For the specified operands, each
+// operand is converted, without change of type domain, to a type whose
+// corresponding real type is the common real type. Unless explicitly stated
+// otherwise, the common real type is also the corresponding real type of the
+// result, whose type domain is the type domain of the operands if they are the
+// same, and complex otherwise. This pattern is called the usual arithmetic
+// conversions:
+func usualArithmeticConversions(ctx *context, n Node, a, b Operand, normalize bool) (Operand, Operand) {
+ if a.Type().Kind() == Invalid || b.Type().Kind() == Invalid {
+ return noOperand, noOperand
+ }
+
+ abi := a.getABI()
+ if !a.Type().IsArithmeticType() {
+ if ctx != nil {
+ ctx.errNode(n, "not an arithmetic type: %s", a.Type())
+ }
+ return noOperand, noOperand
+ }
+
+ if !b.Type().IsArithmeticType() {
+ if ctx != nil {
+ ctx.errNode(n, "not an arithmetic type: %s", b.Type())
+ }
+ return noOperand, noOperand
+ }
+
+ if a.Type() == nil || b.Type() == nil {
+ return a, b
+ }
+
+ if normalize {
+ a = a.normalize(ctx, n)
+ b = b.normalize(ctx, n)
+ }
+ if a == noOperand || b == noOperand {
+ return noOperand, noOperand
+ }
+
+ at := a.Type()
+ bt := b.Type()
+ cplx := at.IsComplexType() || bt.IsComplexType()
+
+ // First, if the corresponding real type of either operand is long
+ // double, the other operand is converted, without change of type
+ // domain, to a type whose corresponding real type is long double.
+ if at.Kind() == ComplexLongDouble || bt.Kind() == ComplexLongDouble || at.Kind() == LongDouble || bt.Kind() == LongDouble {
+ switch {
+ case cplx:
+ return a.convertTo(ctx, n, abi.Type(ComplexLongDouble)), b.convertTo(ctx, n, abi.Type(ComplexLongDouble))
+ default:
+ return a.convertTo(ctx, n, abi.Type(LongDouble)), b.convertTo(ctx, n, abi.Type(LongDouble))
+ }
+ }
+
+ // Otherwise, if the corresponding real type of either operand is
+ // double, the other operand is converted, without change of type
+ // domain, to a type whose corresponding real type is double.
+ if at.Kind() == ComplexDouble || bt.Kind() == ComplexDouble || at.Kind() == Double || bt.Kind() == Double {
+ switch {
+ case cplx:
+ return a.convertTo(ctx, n, abi.Type(ComplexDouble)), b.convertTo(ctx, n, abi.Type(ComplexDouble))
+ default:
+ return a.convertTo(ctx, n, abi.Type(Double)), b.convertTo(ctx, n, abi.Type(Double))
+ }
+ }
+
+ // Otherwise, if the corresponding real type of either operand is
+ // float, the other operand is converted, without change of type
+ // domain, to a type whose corresponding real type is float.
+ if at.Kind() == ComplexFloat || bt.Kind() == ComplexFloat || at.Kind() == Float || bt.Kind() == Float {
+ switch {
+ case cplx:
+ return a.convertTo(ctx, n, abi.Type(ComplexFloat)), b.convertTo(ctx, n, abi.Type(ComplexFloat))
+ default:
+ return a.convertTo(ctx, n, abi.Type(Float)), b.convertTo(ctx, n, abi.Type(Float))
+ }
+ }
+
+ if cplx {
+ panic(internalErrorf("TODO %v, %v", at, bt))
+ }
+
+ if !a.Type().IsIntegerType() || !b.Type().IsIntegerType() {
+ panic(todo(""))
+ }
+
+ // Otherwise, the integer promotions are performed on both operands.
+ a = a.integerPromotion(ctx, n)
+ b = b.integerPromotion(ctx, n)
+ at = a.Type()
+ bt = b.Type()
+
+ // Then the following rules are applied to the promoted operands:
+
+ // If both operands have the same type, then no further conversion is
+ // needed.
+ if at.Kind() == bt.Kind() {
+ return a, b
+ }
+
+ // Otherwise, if both operands have signed integer types or both have
+ // unsigned integer types, the operand with the type of lesser integer
+ // conversion rank is converted to the type of the operand with greater
+ // rank.
+ if abi.isSignedInteger(at.Kind()) == abi.isSignedInteger(bt.Kind()) {
+ t := a.Type()
+ if intConvRank[bt.Kind()] > intConvRank[at.Kind()] {
+ t = b.Type()
+ }
+ return a.convertTo(ctx, n, t), b.convertTo(ctx, n, t)
+
+ }
+
+ // Otherwise, if the operand that has unsigned integer type has rank
+ // greater or equal to the rank of the type of the other operand, then
+ // the operand with signed integer type is converted to the type of the
+ // operand with unsigned integer type.
+ switch {
+ case a.Type().IsSignedType(): // b is unsigned
+ if intConvRank[bt.Kind()] >= intConvRank[a.Type().Kind()] {
+ return a.convertTo(ctx, n, b.Type()), b
+ }
+ case b.Type().IsSignedType(): // a is unsigned
+ if intConvRank[at.Kind()] >= intConvRank[b.Type().Kind()] {
+ return a, b.convertTo(ctx, n, a.Type())
+ }
+ default:
+ panic(fmt.Errorf("TODO %v %v", a.Type(), b.Type()))
+ }
+
+ // Otherwise, if the type of the operand with signed integer type can
+ // represent all of the values of the type of the operand with unsigned
+ // integer type, then the operand with unsigned integer type is
+ // converted to the type of the operand with signed integer type.
+ var signed Type
+ switch {
+ case abi.isSignedInteger(at.Kind()): // b is unsigned
+ signed = a.Type()
+ if at.Size() > bt.Size() {
+ return a, b.convertTo(ctx, n, a.Type())
+ }
+ case abi.isSignedInteger(bt.Kind()): // a is unsigned
+ signed = b.Type()
+ if bt.Size() > at.Size() {
+ return a.convertTo(ctx, n, b.Type()), b
+ }
+
+ }
+
+ // Otherwise, both operands are converted to the unsigned integer type
+ // corresponding to the type of the operand with signed integer type.
+ var typ Type
+ switch signed.Kind() {
+ case Int:
+ //TODO if a.IsEnumConst || b.IsEnumConst {
+ //TODO return a, b
+ //TODO }
+
+ typ = abi.Type(UInt)
+ case Long:
+ typ = abi.Type(ULong)
+ case LongLong:
+ typ = abi.Type(ULongLong)
+ default:
+ panic(todo(""))
+ }
+ return a.convertTo(ctx, n, typ), b.convertTo(ctx, n, typ)
+}
+
+// [0]6.3.1.1-2
+//
+// If an int can represent all values of the original type, the value is
+// converted to an int; otherwise, it is converted to an unsigned int. These
+// are called the integer promotions. All other types are unchanged by the
+// integer promotions.
+func (o *operand) integerPromotion(ctx *context, n Node) Operand {
+ t := o.Type()
+ if t2 := integerPromotion(o.abi, t); t2.Kind() != t.Kind() {
+ return o.convertTo(ctx, n, t2)
+ }
+
+ return o
+}
+
+// [0]6.3.1.1-2
+//
+// If an int can represent all values of the original type, the value is
+// converted to an int; otherwise, it is converted to an unsigned int. These
+// are called the integer promotions. All other types are unchanged by the
+// integer promotions.
+func integerPromotion(abi *ABI, t Type) Type {
+ // github.com/gcc-mirror/gcc/gcc/testsuite/gcc.c-torture/execute/bf-sign-2.c
+ //
+ // This test checks promotion of bitfields. Bitfields
+ // should be promoted very much like chars and shorts:
+ //
+ // Bitfields (signed or unsigned) should be promoted to
+ // signed int if their value will fit in a signed int,
+ // otherwise to an unsigned int if their value will fit
+ // in an unsigned int, otherwise we don't promote them
+ // (ANSI/ISO does not specify the behavior of bitfields
+ // larger than an unsigned int).
+ if t.IsBitFieldType() {
+ f := t.BitField()
+ intBits := int(abi.Types[Int].Size) * 8
+ switch {
+ case t.IsSignedType():
+ if f.BitFieldWidth() < intBits-1 {
+ return abi.Type(Int)
+ }
+ default:
+ if f.BitFieldWidth() < intBits {
+ return abi.Type(Int)
+ }
+ }
+ return t
+ }
+
+ switch t.Kind() {
+ case Invalid:
+ return t
+ case Char, SChar, UChar, Short, UShort:
+ return abi.Type(Int)
+ default:
+ return t
+ }
+}
+
+func (o *operand) convertTo(ctx *context, n Node, to Type) Operand {
+ if o.Type().Kind() == Invalid {
+ return o
+ }
+
+ v := o.Value()
+ r := &operand{abi: o.abi, typ: to, offset: o.offset, value: v}
+ switch v.(type) {
+ case nil, *InitializerValue:
+ return r
+ }
+
+ if o.Type().Kind() == to.Kind() {
+ return r.normalize(ctx, n)
+ }
+
+ if o.Type().IsIntegerType() {
+ return o.convertFromInt(ctx, n, to)
+ }
+
+ if to.IsIntegerType() {
+ return o.convertToInt(ctx, n, to)
+ }
+
+ switch o.Type().Kind() {
+ case Array:
+ switch to.Kind() {
+ case Ptr:
+ return r
+ default:
+ panic(todo("", n.Position()))
+ }
+ case ComplexFloat:
+ v := v.(Complex64Value)
+ switch to.Kind() {
+ case ComplexDouble:
+ r.value = Complex128Value(v)
+ case Float:
+ r.value = Float32Value(real(v))
+ case Double:
+ r.value = Float64Value(real(v))
+ case ComplexLongDouble:
+ panic(todo("", n.Position()))
+ default:
+ panic(todo("", n.Position()))
+ }
+ case ComplexDouble:
+ v := v.(Complex128Value)
+ switch to.Kind() {
+ case ComplexFloat:
+ r.value = Complex64Value(v)
+ case ComplexLongDouble:
+ //TODO panic(todo("", n.Position()))
+ r.value = nil
+ case Float:
+ r.value = Float32Value(real(v))
+ case Double:
+ r.value = Float64Value(real(v))
+ default:
+ //TODO panic(todo("", n.Position(), o.Type(), to))
+ r.value = nil
+ }
+ case Float:
+ v := v.(Float32Value)
+ switch to.Kind() {
+ case ComplexFloat:
+ r.value = Complex64Value(complex(v, 0))
+ case ComplexDouble:
+ r.value = Complex128Value(complex(v, 0))
+ case Double:
+ r.value = Float64Value(v)
+ case ComplexLongDouble:
+ panic(todo("", n.Position()))
+ case LongDouble:
+ r.value = &Float128Value{N: big.NewFloat(float64(v))}
+ case Decimal32, Decimal64, Decimal128:
+ // ok
+ default:
+ panic(todo("695 %s", to.Kind()))
+ }
+ case Double:
+ v := v.(Float64Value)
+ switch to.Kind() {
+ case ComplexFloat:
+ r.value = Complex64Value(complex(v, 0))
+ case ComplexDouble:
+ r.value = Complex128Value(complex(v, 0))
+ case LongDouble:
+ f := float64(v)
+ switch {
+ case math.IsNaN(f):
+ r.value = &Float128Value{NaN: true}
+ default:
+ r.value = &Float128Value{N: big.NewFloat(f)}
+ }
+ case Float:
+ r.value = Float32Value(v)
+ case ComplexLongDouble:
+ panic(todo("", n.Position()))
+ case Vector:
+ r.value = nil
+ case Decimal32, Decimal64, Decimal128:
+ // ok
+ default:
+ panic(todo("", to.Kind()))
+ }
+ case LongDouble:
+ v := v.(*Float128Value)
+ switch to.Kind() {
+ case Double:
+ if v.NaN {
+ r.value = Float64Value(math.NaN())
+ break
+ }
+
+ d, _ := v.N.Float64()
+ r.value = Float64Value(d)
+ case Float:
+ if v.NaN {
+ r.value = Float32Value(math.NaN())
+ break
+ }
+
+ d, _ := v.N.Float64()
+ r.value = Float32Value(d)
+ case ComplexLongDouble:
+ if v.NaN {
+ r.value = Complex256Value{v, &Float128Value{NaN: true}}
+ break
+ }
+
+ r.value = Complex256Value{v, &Float128Value{N: big.NewFloat(0)}}
+ case Decimal32, Decimal64, Decimal128:
+ // ok
+ default:
+ panic(todo("813 %v", to.Kind()))
+ }
+ case Ptr:
+ switch to.Kind() {
+ case Void:
+ return noOperand
+ default:
+ panic(internalErrorf("%v: %v y-> %v %v", n.Position(), o.Type(), to, to.Kind()))
+ }
+ default:
+ panic(internalErrorf("%v: %v -> %v %v", n.Position(), o.Type(), to, to.Kind()))
+ }
+ return r.normalize(ctx, n)
+}
+
+type signedSaturationLimit struct {
+ fmin, fmax float64
+ min, max int64
+}
+
+type unsignedSaturationLimit struct {
+ fmax float64
+ max uint64
+}
+
+var (
+ signedSaturationLimits = [...]signedSaturationLimit{
+ 1: {math.Nextafter(math.MinInt32, 0), math.Nextafter(math.MaxInt32, 0), math.MinInt32, math.MaxInt32},
+ 2: {math.Nextafter(math.MinInt32, 0), math.Nextafter(math.MaxInt32, 0), math.MinInt32, math.MaxInt32},
+ 4: {math.Nextafter(math.MinInt32, 0), math.Nextafter(math.MaxInt32, 0), math.MinInt32, math.MaxInt32},
+ 8: {math.Nextafter(math.MinInt64, 0), math.Nextafter(math.MaxInt64, 0), math.MinInt64, math.MaxInt64},
+ }
+
+ unsignedSaturationLimits = [...]unsignedSaturationLimit{
+ 1: {math.Nextafter(math.MaxUint32, 0), math.MaxUint32},
+ 2: {math.Nextafter(math.MaxUint32, 0), math.MaxUint32},
+ 4: {math.Nextafter(math.MaxUint32, 0), math.MaxUint32},
+ 8: {math.Nextafter(math.MaxUint64, 0), math.MaxUint64},
+ }
+)
+
+func (o *operand) convertToInt(ctx *context, n Node, to Type) (r Operand) {
+ v := o.Value()
+ switch o.Type().Kind() {
+ case Float:
+ v := float64(v.(Float32Value))
+ switch {
+ case to.IsSignedType():
+ limits := &signedSaturationLimits[to.Size()]
+ if v > limits.fmax {
+ return (&operand{abi: o.abi, typ: to, value: Int64Value(limits.max)}).normalize(ctx, n)
+ }
+
+ if v < limits.fmin {
+ return (&operand{abi: o.abi, typ: to, value: Int64Value(limits.min)}).normalize(ctx, n)
+ }
+
+ return (&operand{abi: o.abi, typ: to, value: Int64Value(v)}).normalize(ctx, n)
+ default:
+ limits := &unsignedSaturationLimits[to.Size()]
+ if v > limits.fmax {
+ return (&operand{abi: o.abi, typ: to, value: Uint64Value(limits.max)}).normalize(ctx, n)
+ }
+
+ return (&operand{abi: o.abi, typ: to, value: Uint64Value(v)}).normalize(ctx, n)
+ }
+ case Double:
+ v := float64(v.(Float64Value))
+ switch {
+ case to.IsSignedType():
+ limits := &signedSaturationLimits[to.Size()]
+ if v > limits.fmax {
+ return (&operand{abi: o.abi, typ: to, value: Int64Value(limits.max)}).normalize(ctx, n)
+ }
+
+ if v < limits.fmin {
+ return (&operand{abi: o.abi, typ: to, value: Int64Value(limits.min)}).normalize(ctx, n)
+ }
+
+ return (&operand{abi: o.abi, typ: to, value: Int64Value(v)}).normalize(ctx, n)
+ default:
+ limits := &unsignedSaturationLimits[to.Size()]
+ if v > limits.fmax {
+ return (&operand{abi: o.abi, typ: to, value: Uint64Value(limits.max)}).normalize(ctx, n)
+ }
+
+ return (&operand{abi: o.abi, typ: to, value: Uint64Value(v)}).normalize(ctx, n)
+ }
+ case LongDouble:
+ panic(todo("", n.Position()))
+ case Ptr:
+ var v uint64
+ switch x := o.Value().(type) {
+ case Int64Value:
+ v = uint64(x)
+ case Uint64Value:
+ v = uint64(x)
+ case *InitializerValue:
+ return (&operand{abi: o.abi, typ: to})
+ default:
+ panic(internalErrorf("%v: %T", n.Position(), x))
+ }
+ switch {
+ case to.IsSignedType():
+ return (&operand{abi: o.abi, typ: to, value: Int64Value(v)}).normalize(ctx, n)
+ default:
+ return (&operand{abi: o.abi, typ: to, value: Uint64Value(v)}).normalize(ctx, n)
+ }
+ case Array:
+ return &operand{abi: o.abi, typ: to}
+ case Vector:
+ if o.Type().Size() == to.Size() {
+ return &operand{abi: o.abi, typ: to}
+ }
+ }
+ if ctx != nil {
+ ctx.errNode(n, "cannot convert %s to %s", o.Type(), to)
+ }
+ return &operand{abi: o.abi, typ: to}
+}
+
+func (o *operand) convertFromInt(ctx *context, n Node, to Type) (r Operand) {
+ var v uint64
+ switch x := o.Value().(type) {
+ case Int64Value:
+ v = uint64(x)
+ case Uint64Value:
+ v = uint64(x)
+ default:
+ if ctx != nil {
+ ctx.errNode(n, "conversion to integer: invalid value")
+ }
+ return &operand{abi: o.abi, typ: to}
+ }
+
+ if to.IsIntegerType() {
+ switch {
+ case to.IsSignedType():
+ return (&operand{abi: o.abi, typ: to, value: Int64Value(v)}).normalize(ctx, n)
+ default:
+ return (&operand{abi: o.abi, typ: to, value: Uint64Value(v)}).normalize(ctx, n)
+ }
+ }
+
+ switch to.Kind() {
+ case ComplexFloat:
+ switch {
+ case o.Type().IsSignedType():
+ return (&operand{abi: o.abi, typ: to, value: Complex64Value(complex(float64(int64(v)), 0))}).normalize(ctx, n)
+ default:
+ return (&operand{abi: o.abi, typ: to, value: Complex64Value(complex(float64(v), 0))}).normalize(ctx, n)
+ }
+ case ComplexDouble:
+ switch {
+ case o.Type().IsSignedType():
+ return (&operand{abi: o.abi, typ: to, value: Complex128Value(complex(float64(int64(v)), 0))}).normalize(ctx, n)
+ default:
+ return (&operand{abi: o.abi, typ: to, value: Complex128Value(complex(float64(v), 0))}).normalize(ctx, n)
+ }
+ case Float:
+ switch {
+ case o.Type().IsSignedType():
+ return (&operand{abi: o.abi, typ: to, value: Float32Value(float64(int64(v)))}).normalize(ctx, n)
+ default:
+ return (&operand{abi: o.abi, typ: to, value: Float32Value(float64(v))}).normalize(ctx, n)
+ }
+ case ComplexLongDouble:
+ panic(todo("", n.Position()))
+ case Double:
+ switch {
+ case o.Type().IsSignedType():
+ return (&operand{abi: o.abi, typ: to, value: Float64Value(int64(v))}).normalize(ctx, n)
+ default:
+ return (&operand{abi: o.abi, typ: to, value: Float64Value(v)}).normalize(ctx, n)
+ }
+ case LongDouble:
+ switch {
+ case o.Type().IsSignedType():
+ return (&operand{abi: o.abi, typ: to, value: &Float128Value{N: big.NewFloat(0).SetInt64(int64(v))}}).normalize(ctx, n)
+ default:
+ return (&operand{abi: o.abi, typ: to, value: &Float128Value{N: big.NewFloat(0).SetUint64(v)}}).normalize(ctx, n)
+ }
+ case Ptr:
+ return (&operand{abi: o.abi, typ: to, value: Uint64Value(v)}).normalize(ctx, n)
+ case Struct, Union, Array, Void, Int128, UInt128:
+ return &operand{abi: o.abi, typ: to}
+ case Vector:
+ if o.Type().Size() == to.Size() {
+ return &operand{abi: o.abi, typ: to}
+ }
+ }
+ if ctx != nil {
+ ctx.errNode(n, "cannot convert %s to %s", o.Type(), to)
+ }
+ return &operand{abi: o.abi, typ: to}
+}
+
+func (o *operand) normalize(ctx *context, n Node) (r Operand) {
+ if o.Type() == nil {
+ ctx.errNode(n, "operand has unsupported, invalid or incomplete type")
+ return noOperand
+ }
+
+ if o.Type().IsIntegerType() {
+ switch {
+ case o.Type().IsSignedType():
+ if x, ok := o.value.(Uint64Value); ok {
+ o.value = Int64Value(x)
+ }
+ default:
+ if x, ok := o.value.(Int64Value); ok {
+ o.value = Uint64Value(x)
+ }
+ }
+ switch x := o.Value().(type) {
+ case Int64Value:
+ if v := convertInt64(int64(x), o.Type(), o.abi); v != int64(x) {
+ o.value = Int64Value(v)
+ }
+ case Uint64Value:
+ v := uint64(x)
+ switch o.Type().Size() {
+ case 1:
+ v &= 0xff
+ case 2:
+ v &= 0xffff
+ case 4:
+ v &= 0xffffffff
+ }
+ if v != uint64(x) {
+ o.value = Uint64Value(v)
+ }
+ case *InitializerValue, nil:
+ // ok
+ default:
+ panic(internalErrorf("%T %v", x, x))
+ }
+ return o
+ }
+
+ switch o.Type().Kind() {
+ case ComplexFloat:
+ switch o.Value().(type) {
+ case Complex64Value, nil:
+ return o
+ default:
+ panic(todo(""))
+ }
+ case ComplexDouble:
+ switch o.Value().(type) {
+ case Complex128Value, nil:
+ return o
+ default:
+ panic(todo(""))
+ }
+ case ComplexLongDouble:
+ switch o.Value().(type) {
+ case Complex256Value, nil:
+ return o
+ default:
+ panic(todo("934 %v", o.Type().Kind()))
+ }
+ case Float:
+ switch o.Value().(type) {
+ case Float32Value, *InitializerValue, nil:
+ return o
+ default:
+ panic(todo(""))
+ }
+ case Double:
+ switch x := o.Value().(type) {
+ case Float64Value, *InitializerValue, nil:
+ return o
+ default:
+ panic(internalErrorf("%T %v", x, x))
+ }
+ case LongDouble:
+ switch x := o.Value().(type) {
+ case *Float128Value, nil:
+ return o
+ default:
+ panic(internalErrorf("%T %v TODO980 %v", x, x, n.Position()))
+ }
+ case Ptr:
+ switch o.Value().(type) {
+ case Int64Value, Uint64Value, *InitializerValue, StringValue, WideStringValue, nil:
+ return o
+ default:
+ panic(todo(""))
+ }
+ case Array, Void, Function, Struct, Union, Vector, Decimal32, Decimal64, Decimal128:
+ return o
+ case ComplexChar, ComplexInt, ComplexLong, ComplexLongLong, ComplexShort, ComplexUInt, ComplexUShort:
+ //TOD
+ if ctx != nil {
+ ctx.errNode(n, "unsupported type: %s", o.Type())
+ }
+ return noOperand
+ }
+ panic(internalErrorf("%v, %v", o.Type(), o.Type().Kind()))
+}
+
+func convertInt64(n int64, t Type, abi *ABI) int64 {
+ k := t.Kind()
+ if k == Enum {
+ //TODO
+ }
+ signed := abi.isSignedInteger(k)
+ switch sz := abi.size(k); sz {
+ case 1:
+ switch {
+ case signed:
+ switch {
+ case int8(n) < 0:
+ return n | ^math.MaxUint8
+ default:
+ return n & math.MaxUint8
+ }
+ default:
+ return n & math.MaxUint8
+ }
+ case 2:
+ switch {
+ case signed:
+ switch {
+ case int16(n) < 0:
+ return n | ^math.MaxUint16
+ default:
+ return n & math.MaxUint16
+ }
+ default:
+ return n & math.MaxUint16
+ }
+ case 4:
+ switch {
+ case signed:
+ switch {
+ case int32(n) < 0:
+ return n | ^math.MaxUint32
+ default:
+ return n & math.MaxUint32
+ }
+ default:
+ return n & math.MaxUint32
+ }
+ default:
+ return n
+ }
+}
+
+func boolValue(b bool) Value {
+ if b {
+ return Int64Value(1)
+ }
+
+ return Int64Value(0)
+}
+
+type initializer interface {
+ List() []*Initializer
+ IsConst() bool
+}
+
+type InitializerValue struct {
+ typ Type
+ initializer initializer
+}
+
+func (v *InitializerValue) List() []*Initializer {
+ if v == nil || v.initializer == nil {
+ return nil
+ }
+
+ return v.initializer.List()
+}
+
+func (v *InitializerValue) IsConst() bool {
+ return v != nil && v.initializer != nil && v.initializer.IsConst()
+}
+func (v *InitializerValue) Type() Type { return v.typ }
+func (v *InitializerValue) add(b Value) Value { return nil }
+func (v *InitializerValue) and(b Value) Value { return nil }
+func (v *InitializerValue) cpl() Value { return nil }
+func (v *InitializerValue) div(b Value) Value { return nil }
+func (v *InitializerValue) eq(b Value) Value { return nil }
+func (v *InitializerValue) ge(b Value) Value { return nil }
+func (v *InitializerValue) gt(b Value) Value { return nil }
+func (v *InitializerValue) le(b Value) Value { return nil }
+func (v *InitializerValue) lsh(b Value) Value { return nil }
+func (v *InitializerValue) lt(b Value) Value { return nil }
+func (v *InitializerValue) mod(b Value) Value { return nil }
+func (v *InitializerValue) mul(b Value) Value { return nil }
+func (v *InitializerValue) neg() Value { return nil }
+func (v *InitializerValue) neq(b Value) Value { return nil }
+func (v *InitializerValue) or(b Value) Value { return nil }
+func (v *InitializerValue) rsh(b Value) Value { return nil }
+func (v *InitializerValue) sub(b Value) Value { return nil }
+func (v *InitializerValue) xor(b Value) Value { return nil }
+
+func (v *InitializerValue) IsNonZero() bool {
+ if v == nil {
+ return false
+ }
+
+ for _, v := range v.List() {
+ if !v.AssignmentExpression.Operand.IsZero() {
+ return true
+ }
+ }
+ return false
+}
+
+func (v *InitializerValue) IsZero() bool {
+ if v == nil {
+ return false
+ }
+
+ for _, v := range v.List() {
+ if !v.AssignmentExpression.Operand.IsZero() {
+ return false
+ }
+ }
+ return true
+}