1
0
mirror of https://github.com/golang/go synced 2024-11-23 20:50:04 -07:00
go/test/fixedbugs/issue9604b.go
Keith Randall c3b459d328 test: generate tests for arithmetic on narrow types
Fixes #9607
Related to #9604
Inadvertently found #9609

Change-Id: I8a8ddf84ac72d3e18986fd8e9288734459f3f174
Reviewed-on: https://go-review.googlesource.com/2962
Reviewed-by: Minux Ma <minux@golang.org>
2015-01-17 21:00:15 +00:00

178 lines
4.8 KiB
Go

// runoutput
// Copyright 2015 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 (
"fmt"
"math/big"
"unsafe"
)
var one = big.NewInt(1)
type _type struct {
name string
bits uint
signed bool
}
// testvalues returns a list of all test values for this type.
func (t *_type) testvalues() []*big.Int {
var a []*big.Int
a = append(a, big.NewInt(0))
a = append(a, big.NewInt(1))
a = append(a, big.NewInt(2))
if t.signed {
a = append(a, big.NewInt(-1))
a = append(a, big.NewInt(-2))
r := big.NewInt(1)
a = append(a, r.Lsh(r, t.bits-1).Sub(r, big.NewInt(1)))
r = big.NewInt(1)
a = append(a, r.Lsh(r, t.bits-1).Sub(r, big.NewInt(2)))
r = big.NewInt(1)
a = append(a, r.Lsh(r, t.bits-1).Neg(r))
r = big.NewInt(1)
a = append(a, r.Lsh(r, t.bits-1).Neg(r).Add(r, big.NewInt(1)))
} else {
r := big.NewInt(1)
a = append(a, r.Lsh(r, t.bits).Sub(r, big.NewInt(1)))
r = big.NewInt(1)
a = append(a, r.Lsh(r, t.bits).Sub(r, big.NewInt(2)))
}
return a
}
// trunc truncates a value to the range of the given type.
func (t *_type) trunc(x *big.Int) *big.Int {
r := new(big.Int)
m := new(big.Int)
m.Lsh(one, t.bits)
m.Sub(m, one)
r.And(x, m)
if t.signed && r.Bit(int(t.bits)-1) == 1 {
m.Neg(one)
m.Lsh(m, t.bits)
r.Or(r, m)
}
return r
}
var types = []_type{
_type{"byte", 8, false},
_type{"int8", 8, true},
_type{"uint8", 8, false},
_type{"rune", 32, true},
_type{"int16", 16, true},
_type{"uint16", 16, false},
_type{"int32", 32, true},
_type{"uint32", 32, false},
_type{"int64", 64, true},
_type{"uint64", 64, false},
_type{"int", 8 * uint(unsafe.Sizeof(int(0))), true},
_type{"uint", 8 * uint(unsafe.Sizeof(uint(0))), false},
_type{"uintptr", 8 * uint(unsafe.Sizeof((*byte)(nil))), false},
}
type binop struct {
name string
eval func(x, y *big.Int) *big.Int
}
var binops = []binop{
binop{"+", func(x, y *big.Int) *big.Int { return new(big.Int).Add(x, y) }},
binop{"-", func(x, y *big.Int) *big.Int { return new(big.Int).Sub(x, y) }},
binop{"*", func(x, y *big.Int) *big.Int { return new(big.Int).Mul(x, y) }},
binop{"/", func(x, y *big.Int) *big.Int { return new(big.Int).Quo(x, y) }},
binop{"%", func(x, y *big.Int) *big.Int { return new(big.Int).Rem(x, y) }},
binop{"&", func(x, y *big.Int) *big.Int { return new(big.Int).And(x, y) }},
binop{"|", func(x, y *big.Int) *big.Int { return new(big.Int).Or(x, y) }},
binop{"^", func(x, y *big.Int) *big.Int { return new(big.Int).Xor(x, y) }},
binop{"&^", func(x, y *big.Int) *big.Int { return new(big.Int).AndNot(x, y) }},
}
type unop struct {
name string
eval func(x *big.Int) *big.Int
}
var unops = []unop{
unop{"+", func(x *big.Int) *big.Int { return new(big.Int).Set(x) }},
unop{"-", func(x *big.Int) *big.Int { return new(big.Int).Neg(x) }},
unop{"^", func(x *big.Int) *big.Int { return new(big.Int).Not(x) }},
}
type shiftop struct {
name string
eval func(x *big.Int, i uint) *big.Int
}
var shiftops = []shiftop{
shiftop{"<<", func(x *big.Int, i uint) *big.Int { return new(big.Int).Lsh(x, i) }},
shiftop{">>", func(x *big.Int, i uint) *big.Int { return new(big.Int).Rsh(x, i) }},
}
// valname returns the name of n as can be used as part of a variable name.
func valname(n *big.Int) string {
s := fmt.Sprintf("%d", n)
if s[0] == '-' {
s = "neg" + s[1:]
}
return s
}
func main() {
fmt.Println("package main")
// We make variables to hold all the different values we'd like to use.
// We use global variables to prevent any constant folding.
for _, t := range types {
for _, n := range t.testvalues() {
fmt.Printf("var %s_%s %s = %d\n", t.name, valname(n), t.name, n)
}
}
fmt.Println("func main() {")
for _, t := range types {
// test binary ops
for _, op := range binops {
for _, x := range t.testvalues() {
for _, y := range t.testvalues() {
if (op.name == "/" || op.name == "%") && y.Sign() == 0 {
continue
}
r := t.trunc(op.eval(x, y))
eqn := fmt.Sprintf("%s_%s %s %s_%s != %d", t.name, valname(x), op.name, t.name, valname(y), r)
fmt.Printf("\tif %s { println(\"bad: %s\") }\n", eqn, eqn)
}
}
}
// test unary ops
for _, op := range unops {
for _, x := range t.testvalues() {
r := t.trunc(op.eval(x))
eqn := fmt.Sprintf("%s %s_%s != %d", op.name, t.name, valname(x), r)
fmt.Printf("\tif %s { println(\"bad: %s\") }\n", eqn, eqn)
}
}
// test shifts
for _, op := range shiftops {
for _, x := range t.testvalues() {
for _, i := range []uint{0, 1, t.bits - 2, t.bits - 1, t.bits, t.bits + 1} {
r := t.trunc(op.eval(x, i))
eqn := fmt.Sprintf("%s_%s %s %d != %d", t.name, valname(x), op.name, i, r)
fmt.Printf("\tif %s { println(\"bad: %s\") }\n", eqn, eqn)
}
}
}
}
fmt.Println("}")
}