mirror of
https://github.com/ceph/ceph-csi.git
synced 2025-05-22 15:26:42 +00:00
Bumps the k8s-dependencies group with 1 update: [k8s.io/kubernetes](https://github.com/kubernetes/kubernetes). Updates `k8s.io/kubernetes` from 1.32.3 to 1.33.0 - [Release notes](https://github.com/kubernetes/kubernetes/releases) - [Commits](https://github.com/kubernetes/kubernetes/compare/v1.32.3...v1.33.0) --- updated-dependencies: - dependency-name: k8s.io/kubernetes dependency-version: 1.33.0 dependency-type: direct:production update-type: version-update:semver-minor dependency-group: k8s-dependencies ... Signed-off-by: dependabot[bot] <support@github.com> Signed-off-by: Niels de Vos <ndevos@ibm.com>
683 lines
20 KiB
Go
683 lines
20 KiB
Go
/*
|
|
Copyright 2014 Google Inc. All rights reserved.
|
|
Copyright 2014 The gofuzz Authors.
|
|
Copyright 2025 The Kubernetes Authors.
|
|
|
|
Licensed under the Apache License, Version 2.0 (the "License");
|
|
you may not use this file except in compliance with the License.
|
|
You may obtain a copy of the License at
|
|
|
|
http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
Unless required by applicable law or agreed to in writing, software
|
|
distributed under the License is distributed on an "AS IS" BASIS,
|
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
See the License for the specific language governing permissions and
|
|
limitations under the License.
|
|
*/
|
|
|
|
// Package randfill is a library for populating go objects with random values.
|
|
package randfill
|
|
|
|
import (
|
|
"fmt"
|
|
"math/rand"
|
|
"reflect"
|
|
"regexp"
|
|
"sync"
|
|
"time"
|
|
"unsafe"
|
|
|
|
"strings"
|
|
|
|
"sigs.k8s.io/randfill/bytesource"
|
|
)
|
|
|
|
// funcMap is a map from a type to a function that randfills that type. The
|
|
// function is a reflect.Value because the type being filled is different for
|
|
// each func.
|
|
type funcMap map[reflect.Type]reflect.Value
|
|
|
|
// Filler knows how to fill any object with random fields.
|
|
type Filler struct {
|
|
customFuncs funcMap
|
|
defaultFuncs funcMap
|
|
r *rand.Rand
|
|
nilChance float64
|
|
minElements int
|
|
maxElements int
|
|
maxDepth int
|
|
allowUnexportedFields bool
|
|
skipFieldPatterns []*regexp.Regexp
|
|
|
|
lock sync.Mutex
|
|
}
|
|
|
|
// New returns a new Filler. Customize your Filler further by calling Funcs,
|
|
// RandSource, NilChance, or NumElements in any order.
|
|
func New() *Filler {
|
|
return NewWithSeed(time.Now().UnixNano())
|
|
}
|
|
|
|
func NewWithSeed(seed int64) *Filler {
|
|
f := &Filler{
|
|
defaultFuncs: funcMap{
|
|
reflect.TypeOf(&time.Time{}): reflect.ValueOf(randfillTime),
|
|
},
|
|
|
|
customFuncs: funcMap{},
|
|
r: rand.New(rand.NewSource(seed)),
|
|
nilChance: .2,
|
|
minElements: 1,
|
|
maxElements: 10,
|
|
maxDepth: 100,
|
|
allowUnexportedFields: false,
|
|
}
|
|
return f
|
|
}
|
|
|
|
// NewFromGoFuzz is a helper function that enables using randfill (this
|
|
// project) with go-fuzz (https://github.com/dvyukov/go-fuzz) for continuous
|
|
// fuzzing. Essentially, it enables translating the fuzzing bytes from
|
|
// go-fuzz to any Go object using this library.
|
|
//
|
|
// This implementation promises a constant translation from a given slice of
|
|
// bytes to the fuzzed objects. This promise will remain over future
|
|
// versions of Go and of this library.
|
|
//
|
|
// Note: the returned Filler should not be shared between multiple goroutines,
|
|
// as its deterministic output will no longer be available.
|
|
//
|
|
// Example: use go-fuzz to test the function `MyFunc(int)` in the package
|
|
// `mypackage`. Add the file: "mypackage_fuzz.go" with the content:
|
|
//
|
|
// // +build gofuzz
|
|
// package mypackage
|
|
// import "sigs.k8s.io/randfill"
|
|
//
|
|
// func Fuzz(data []byte) int {
|
|
// var i int
|
|
// randfill.NewFromGoFuzz(data).Fill(&i)
|
|
// MyFunc(i)
|
|
// return 0
|
|
// }
|
|
func NewFromGoFuzz(data []byte) *Filler {
|
|
return New().RandSource(bytesource.New(data))
|
|
}
|
|
|
|
// Funcs registers custom fill functions for this Filler.
|
|
//
|
|
// Each entry in customFuncs must be a function taking two parameters.
|
|
// The first parameter must be a pointer or map. It is the variable that
|
|
// function will fill with random data. The second parameter must be a
|
|
// randfill.Continue, which will provide a source of randomness and a way
|
|
// to automatically continue filling smaller pieces of the first parameter.
|
|
//
|
|
// These functions are called sensibly, e.g., if you wanted custom string
|
|
// filling, the function `func(s *string, c randfill.Continue)` would get
|
|
// called and passed the address of strings. Maps and pointers will always
|
|
// be made/new'd for you, ignoring the NilChance option. For slices, it
|
|
// doesn't make much sense to pre-create them--Filler doesn't know how
|
|
// long you want your slice--so take a pointer to a slice, and make it
|
|
// yourself. (If you don't want your map/pointer type pre-made, take a
|
|
// pointer to it, and make it yourself.) See the examples for a range of
|
|
// custom functions.
|
|
//
|
|
// If a function is already registered for a type, and a new function is
|
|
// provided, the previous function will be replaced with the new one.
|
|
func (f *Filler) Funcs(customFuncs ...interface{}) *Filler {
|
|
for i := range customFuncs {
|
|
v := reflect.ValueOf(customFuncs[i])
|
|
if v.Kind() != reflect.Func {
|
|
panic("Filler.Funcs: all arguments must be functions")
|
|
}
|
|
t := v.Type()
|
|
if t.NumIn() != 2 || t.NumOut() != 0 {
|
|
panic("Filler.Funcs: all customFuncs must have 2 arguments and 0 returns")
|
|
}
|
|
argT := t.In(0)
|
|
switch argT.Kind() {
|
|
case reflect.Ptr, reflect.Map:
|
|
default:
|
|
panic("Filler.Funcs: customFuncs' first argument must be a pointer or map type")
|
|
}
|
|
if t.In(1) != reflect.TypeOf(Continue{}) {
|
|
panic("Filler.Funcs: customFuncs' second argument must be a randfill.Continue")
|
|
}
|
|
f.customFuncs[argT] = v
|
|
}
|
|
return f
|
|
}
|
|
|
|
// RandSource causes this Filler to get values from the given source of
|
|
// randomness. Use this if you want deterministic filling.
|
|
func (f *Filler) RandSource(s rand.Source) *Filler {
|
|
f.r = rand.New(s)
|
|
return f
|
|
}
|
|
|
|
// NilChance sets the probability of creating a nil pointer, map, or slice to
|
|
// 'p'. 'p' should be between 0 (no nils) and 1 (all nils), inclusive.
|
|
func (f *Filler) NilChance(p float64) *Filler {
|
|
if p < 0 || p > 1 {
|
|
panic("Filler.NilChance: p must be between 0 and 1, inclusive")
|
|
}
|
|
f.nilChance = p
|
|
return f
|
|
}
|
|
|
|
// NumElements sets the minimum and maximum number of elements that will be
|
|
// added to a non-nil map or slice.
|
|
func (f *Filler) NumElements(min, max int) *Filler {
|
|
if min < 0 {
|
|
panic("Filler.NumElements: min must be >= 0")
|
|
}
|
|
if min > max {
|
|
panic("Filler.NumElements: min must be <= max")
|
|
}
|
|
f.minElements = min
|
|
f.maxElements = max
|
|
return f
|
|
}
|
|
|
|
func (f *Filler) genElementCount() int {
|
|
if f.minElements == f.maxElements {
|
|
return f.minElements
|
|
}
|
|
return f.minElements + f.r.Intn(f.maxElements-f.minElements+1)
|
|
}
|
|
|
|
func (f *Filler) genShouldFill() bool {
|
|
return f.r.Float64() >= f.nilChance
|
|
}
|
|
|
|
// MaxDepth sets the maximum number of recursive fill calls that will be made
|
|
// before stopping. This includes struct members, pointers, and map and slice
|
|
// elements.
|
|
func (f *Filler) MaxDepth(d int) *Filler {
|
|
f.maxDepth = d
|
|
return f
|
|
}
|
|
|
|
// AllowUnexportedFields defines whether to fill unexported fields.
|
|
func (f *Filler) AllowUnexportedFields(flag bool) *Filler {
|
|
f.allowUnexportedFields = flag
|
|
return f
|
|
}
|
|
|
|
// SkipFieldsWithPattern tells this Filler to skip any field whose name matches
|
|
// the supplied pattern. Call this multiple times if needed. This is useful to
|
|
// skip XXX_ fields generated by protobuf.
|
|
func (f *Filler) SkipFieldsWithPattern(pattern *regexp.Regexp) *Filler {
|
|
f.skipFieldPatterns = append(f.skipFieldPatterns, pattern)
|
|
return f
|
|
}
|
|
|
|
// SimpleSelfFiller represents an object that knows how to randfill itself.
|
|
//
|
|
// Unlike NativeSelfFiller, this interface does not cause the type in question
|
|
// to depend on the randfill package. This is most useful for simple types. For
|
|
// more complex types, consider using NativeSelfFiller.
|
|
type SimpleSelfFiller interface {
|
|
// RandFill fills the current object with random data.
|
|
RandFill(r *rand.Rand)
|
|
}
|
|
|
|
// NativeSelfFiller represents an object that knows how to randfill itself.
|
|
//
|
|
// Unlike SimpleSelfFiller, this interface allows for recursive filling of
|
|
// child objects with the same rules as the parent Filler.
|
|
type NativeSelfFiller interface {
|
|
// RandFill fills the current object with random data.
|
|
RandFill(c Continue)
|
|
}
|
|
|
|
// Fill recursively fills all of obj's fields with something random. First
|
|
// this tries to find a custom fill function (see Funcs). If there is no
|
|
// custom function, this tests whether the object implements SimpleSelfFiller
|
|
// or NativeSelfFiller and if so, calls RandFill on it to fill itself. If that
|
|
// fails, this will see if there is a default fill function provided by this
|
|
// package. If all of that fails, this will generate random values for all
|
|
// primitive fields and then recurse for all non-primitives.
|
|
//
|
|
// This is safe for cyclic or tree-like structs, up to a limit. Use the
|
|
// MaxDepth method to adjust how deep you need it to recurse.
|
|
//
|
|
// obj must be a pointer. Exported (public) fields can always be set, and if
|
|
// the AllowUnexportedFields() modifier was called it can try to set unexported
|
|
// (private) fields, too.
|
|
//
|
|
// This is intended for tests, so will panic on bad input or unimplemented
|
|
// types. This method takes a lock for the whole Filler, so it is not
|
|
// reentrant. See Continue.
|
|
func (f *Filler) Fill(obj interface{}) {
|
|
f.lock.Lock()
|
|
defer f.lock.Unlock()
|
|
|
|
v := reflect.ValueOf(obj)
|
|
if v.Kind() != reflect.Ptr {
|
|
panic("Filler.Fill: obj must be a pointer")
|
|
}
|
|
v = v.Elem()
|
|
f.fillWithContext(v, 0)
|
|
}
|
|
|
|
// FillNoCustom is just like Fill, except that any custom fill function for
|
|
// obj's type will not be called and obj will not be tested for
|
|
// SimpleSelfFiller or NativeSelfFiller. This applies only to obj and not other
|
|
// instances of obj's type or to obj's child fields.
|
|
//
|
|
// obj must be a pointer. Exported (public) fields can always be set, and if
|
|
// the AllowUnexportedFields() modifier was called it can try to set unexported
|
|
// (private) fields, too.
|
|
//
|
|
// This is intended for tests, so will panic on bad input or unimplemented
|
|
// types. This method takes a lock for the whole Filler, so it is not
|
|
// reentrant. See Continue.
|
|
func (f *Filler) FillNoCustom(obj interface{}) {
|
|
f.lock.Lock()
|
|
defer f.lock.Unlock()
|
|
|
|
v := reflect.ValueOf(obj)
|
|
if v.Kind() != reflect.Ptr {
|
|
panic("Filler.FillNoCustom: obj must be a pointer")
|
|
}
|
|
v = v.Elem()
|
|
f.fillWithContext(v, flagNoCustomFill)
|
|
}
|
|
|
|
const (
|
|
// Do not try to find a custom fill function. Does not apply recursively.
|
|
flagNoCustomFill uint64 = 1 << iota
|
|
)
|
|
|
|
func (f *Filler) fillWithContext(v reflect.Value, flags uint64) {
|
|
fc := &fillerContext{filler: f}
|
|
fc.doFill(v, flags)
|
|
}
|
|
|
|
// fillerContext carries context about a single filling run, which lets Filler
|
|
// be thread-safe.
|
|
type fillerContext struct {
|
|
filler *Filler
|
|
curDepth int
|
|
}
|
|
|
|
func (fc *fillerContext) doFill(v reflect.Value, flags uint64) {
|
|
if fc.curDepth >= fc.filler.maxDepth {
|
|
return
|
|
}
|
|
fc.curDepth++
|
|
defer func() { fc.curDepth-- }()
|
|
|
|
if !v.CanSet() {
|
|
if !fc.filler.allowUnexportedFields || !v.CanAddr() {
|
|
return
|
|
}
|
|
v = reflect.NewAt(v.Type(), unsafe.Pointer(v.UnsafeAddr())).Elem()
|
|
}
|
|
|
|
if flags&flagNoCustomFill == 0 {
|
|
// Check for both pointer and non-pointer custom functions.
|
|
if v.CanAddr() && fc.tryCustom(v.Addr()) {
|
|
return
|
|
}
|
|
if fc.tryCustom(v) {
|
|
return
|
|
}
|
|
}
|
|
|
|
if fn, ok := fillFuncMap[v.Kind()]; ok {
|
|
fn(v, fc.filler.r)
|
|
return
|
|
}
|
|
|
|
switch v.Kind() {
|
|
case reflect.Map:
|
|
if fc.filler.genShouldFill() {
|
|
v.Set(reflect.MakeMap(v.Type()))
|
|
n := fc.filler.genElementCount()
|
|
for i := 0; i < n; i++ {
|
|
key := reflect.New(v.Type().Key()).Elem()
|
|
fc.doFill(key, 0)
|
|
val := reflect.New(v.Type().Elem()).Elem()
|
|
fc.doFill(val, 0)
|
|
v.SetMapIndex(key, val)
|
|
}
|
|
return
|
|
}
|
|
v.Set(reflect.Zero(v.Type()))
|
|
case reflect.Ptr:
|
|
if fc.filler.genShouldFill() {
|
|
v.Set(reflect.New(v.Type().Elem()))
|
|
fc.doFill(v.Elem(), 0)
|
|
return
|
|
}
|
|
v.Set(reflect.Zero(v.Type()))
|
|
case reflect.Slice:
|
|
if fc.filler.genShouldFill() {
|
|
n := fc.filler.genElementCount()
|
|
v.Set(reflect.MakeSlice(v.Type(), n, n))
|
|
for i := 0; i < n; i++ {
|
|
fc.doFill(v.Index(i), 0)
|
|
}
|
|
return
|
|
}
|
|
v.Set(reflect.Zero(v.Type()))
|
|
case reflect.Array:
|
|
if fc.filler.genShouldFill() {
|
|
n := v.Len()
|
|
for i := 0; i < n; i++ {
|
|
fc.doFill(v.Index(i), 0)
|
|
}
|
|
return
|
|
}
|
|
v.Set(reflect.Zero(v.Type()))
|
|
case reflect.Struct:
|
|
for i := 0; i < v.NumField(); i++ {
|
|
skipField := false
|
|
fieldName := v.Type().Field(i).Name
|
|
for _, pattern := range fc.filler.skipFieldPatterns {
|
|
if pattern.MatchString(fieldName) {
|
|
skipField = true
|
|
break
|
|
}
|
|
}
|
|
if !skipField {
|
|
fc.doFill(v.Field(i), 0)
|
|
}
|
|
}
|
|
case reflect.Chan:
|
|
fallthrough
|
|
case reflect.Func:
|
|
fallthrough
|
|
case reflect.Interface:
|
|
fallthrough
|
|
default:
|
|
panic(fmt.Sprintf("can't fill type %v, kind %v", v.Type(), v.Kind()))
|
|
}
|
|
}
|
|
|
|
// tryCustom searches for custom handlers, and returns true iff it finds a match
|
|
// and successfully randomizes v.
|
|
func (fc *fillerContext) tryCustom(v reflect.Value) bool {
|
|
// First: see if we have a fill function for it.
|
|
doCustom, ok := fc.filler.customFuncs[v.Type()]
|
|
if !ok {
|
|
// Second: see if it can fill itself.
|
|
if v.CanInterface() {
|
|
intf := v.Interface()
|
|
if fillable, ok := intf.(SimpleSelfFiller); ok {
|
|
fillable.RandFill(fc.filler.r)
|
|
return true
|
|
}
|
|
if fillable, ok := intf.(NativeSelfFiller); ok {
|
|
fillable.RandFill(Continue{fc: fc, Rand: fc.filler.r})
|
|
return true
|
|
}
|
|
}
|
|
// Finally: see if there is a default fill function.
|
|
doCustom, ok = fc.filler.defaultFuncs[v.Type()]
|
|
if !ok {
|
|
return false
|
|
}
|
|
}
|
|
|
|
switch v.Kind() {
|
|
case reflect.Ptr:
|
|
if v.IsNil() {
|
|
if !v.CanSet() {
|
|
return false
|
|
}
|
|
v.Set(reflect.New(v.Type().Elem()))
|
|
}
|
|
case reflect.Map:
|
|
if v.IsNil() {
|
|
if !v.CanSet() {
|
|
return false
|
|
}
|
|
v.Set(reflect.MakeMap(v.Type()))
|
|
}
|
|
default:
|
|
return false
|
|
}
|
|
|
|
doCustom.Call([]reflect.Value{
|
|
v,
|
|
reflect.ValueOf(Continue{
|
|
fc: fc,
|
|
Rand: fc.filler.r,
|
|
}),
|
|
})
|
|
return true
|
|
}
|
|
|
|
// Continue can be passed to custom fill functions to allow them to use
|
|
// the correct source of randomness and to continue filling their members.
|
|
type Continue struct {
|
|
fc *fillerContext
|
|
|
|
// For convenience, Continue implements rand.Rand via embedding.
|
|
// Use this for generating any randomness if you want your filling
|
|
// to be repeatable for a given seed.
|
|
*rand.Rand
|
|
}
|
|
|
|
// Fill continues filling obj. obj must be a pointer or a reflect.Value of a
|
|
// pointer. See Filler.Fill.
|
|
func (c Continue) Fill(obj interface{}) {
|
|
v, ok := obj.(reflect.Value)
|
|
if !ok {
|
|
v = reflect.ValueOf(obj)
|
|
}
|
|
if v.Kind() != reflect.Ptr {
|
|
panic("Continue.Fill: obj must be a pointer")
|
|
}
|
|
v = v.Elem()
|
|
c.fc.doFill(v, 0)
|
|
}
|
|
|
|
// FillNoCustom continues filling obj, except that any custom fill function for
|
|
// obj's type will not be called and obj will not be tested for
|
|
// SimpleSelfFiller or NativeSelfFiller. See Filler.FillNoCustom.
|
|
func (c Continue) FillNoCustom(obj interface{}) {
|
|
v, ok := obj.(reflect.Value)
|
|
if !ok {
|
|
v = reflect.ValueOf(obj)
|
|
}
|
|
if v.Kind() != reflect.Ptr {
|
|
panic("Continue.FillNoCustom: obj must be a pointer")
|
|
}
|
|
v = v.Elem()
|
|
c.fc.doFill(v, flagNoCustomFill)
|
|
}
|
|
|
|
const defaultStringMaxLen = 20
|
|
|
|
// String makes a random string up to n characters long. If n is 0, the default
|
|
// size range is [0-20). The returned string may include a variety of (valid)
|
|
// UTF-8 encodings.
|
|
func (c Continue) String(n int) string {
|
|
return randString(c.Rand, n)
|
|
}
|
|
|
|
// Uint64 makes random 64 bit numbers.
|
|
// Weirdly, rand doesn't have a function that gives you 64 random bits.
|
|
func (c Continue) Uint64() uint64 {
|
|
return randUint64(c.Rand)
|
|
}
|
|
|
|
// Bool returns true or false randomly.
|
|
func (c Continue) Bool() bool {
|
|
return randBool(c.Rand)
|
|
}
|
|
|
|
func fillInt(v reflect.Value, r *rand.Rand) {
|
|
v.SetInt(int64(randUint64(r)))
|
|
}
|
|
|
|
func fillUint(v reflect.Value, r *rand.Rand) {
|
|
v.SetUint(randUint64(r))
|
|
}
|
|
|
|
func randfillTime(t *time.Time, c Continue) {
|
|
var sec, nsec int64
|
|
// Allow for about 1000 years of random time values, which keeps things
|
|
// like JSON parsing reasonably happy.
|
|
sec = c.Rand.Int63n(1000 * 365 * 24 * 60 * 60)
|
|
// Nanosecond values greater than 1Bn are technically allowed but result in
|
|
// time.Time values with invalid timezone offsets.
|
|
nsec = c.Rand.Int63n(999999999)
|
|
*t = time.Unix(sec, nsec)
|
|
}
|
|
|
|
var fillFuncMap = map[reflect.Kind]func(reflect.Value, *rand.Rand){
|
|
reflect.Bool: func(v reflect.Value, r *rand.Rand) {
|
|
v.SetBool(randBool(r))
|
|
},
|
|
reflect.Int: fillInt,
|
|
reflect.Int8: fillInt,
|
|
reflect.Int16: fillInt,
|
|
reflect.Int32: fillInt,
|
|
reflect.Int64: fillInt,
|
|
reflect.Uint: fillUint,
|
|
reflect.Uint8: fillUint,
|
|
reflect.Uint16: fillUint,
|
|
reflect.Uint32: fillUint,
|
|
reflect.Uint64: fillUint,
|
|
reflect.Uintptr: fillUint,
|
|
reflect.Float32: func(v reflect.Value, r *rand.Rand) {
|
|
v.SetFloat(float64(r.Float32()))
|
|
},
|
|
reflect.Float64: func(v reflect.Value, r *rand.Rand) {
|
|
v.SetFloat(r.Float64())
|
|
},
|
|
reflect.Complex64: func(v reflect.Value, r *rand.Rand) {
|
|
v.SetComplex(complex128(complex(r.Float32(), r.Float32())))
|
|
},
|
|
reflect.Complex128: func(v reflect.Value, r *rand.Rand) {
|
|
v.SetComplex(complex(r.Float64(), r.Float64()))
|
|
},
|
|
reflect.String: func(v reflect.Value, r *rand.Rand) {
|
|
v.SetString(randString(r, 0))
|
|
},
|
|
reflect.UnsafePointer: func(v reflect.Value, r *rand.Rand) {
|
|
panic("filling of UnsafePointers is not implemented")
|
|
},
|
|
}
|
|
|
|
// randBool returns true or false randomly.
|
|
func randBool(r *rand.Rand) bool {
|
|
return r.Int31()&(1<<30) == 0
|
|
}
|
|
|
|
type int63nPicker interface {
|
|
Int63n(int64) int64
|
|
}
|
|
|
|
// UnicodeRange describes a sequential range of unicode characters.
|
|
// Last must be numerically greater than First.
|
|
type UnicodeRange struct {
|
|
First, Last rune
|
|
}
|
|
|
|
// UnicodeRanges describes an arbitrary number of sequential ranges of unicode characters.
|
|
// To be useful, each range must have at least one character (First <= Last) and
|
|
// there must be at least one range.
|
|
type UnicodeRanges []UnicodeRange
|
|
|
|
// choose returns a random unicode character from the given range, using the
|
|
// given randomness source.
|
|
func (ur UnicodeRange) choose(r int63nPicker) rune {
|
|
count := int64(ur.Last - ur.First + 1)
|
|
return ur.First + rune(r.Int63n(count))
|
|
}
|
|
|
|
// CustomStringFillFunc constructs a FillFunc which produces random strings.
|
|
// Each character is selected from the range ur. If there are no characters
|
|
// in the range (cr.Last < cr.First), this will panic.
|
|
func (ur UnicodeRange) CustomStringFillFunc(n int) func(s *string, c Continue) {
|
|
ur.check()
|
|
return func(s *string, c Continue) {
|
|
*s = ur.randString(c.Rand, n)
|
|
}
|
|
}
|
|
|
|
// check is a function that used to check whether the first of ur(UnicodeRange)
|
|
// is greater than the last one.
|
|
func (ur UnicodeRange) check() {
|
|
if ur.Last < ur.First {
|
|
panic("UnicodeRange.check: the last encoding must be greater than the first")
|
|
}
|
|
}
|
|
|
|
// randString of UnicodeRange makes a random string up to 20 characters long.
|
|
// Each character is selected form ur(UnicodeRange).
|
|
func (ur UnicodeRange) randString(r *rand.Rand, max int) string {
|
|
if max == 0 {
|
|
max = defaultStringMaxLen
|
|
}
|
|
n := r.Intn(max)
|
|
sb := strings.Builder{}
|
|
sb.Grow(n)
|
|
for i := 0; i < n; i++ {
|
|
sb.WriteRune(ur.choose(r))
|
|
}
|
|
return sb.String()
|
|
}
|
|
|
|
// defaultUnicodeRanges sets a default unicode range when users do not set
|
|
// CustomStringFillFunc() but want to fill strings.
|
|
var defaultUnicodeRanges = UnicodeRanges{
|
|
{' ', '~'}, // ASCII characters
|
|
{'\u00a0', '\u02af'}, // Multi-byte encoded characters
|
|
{'\u4e00', '\u9fff'}, // Common CJK (even longer encodings)
|
|
}
|
|
|
|
// CustomStringFillFunc constructs a FillFunc which produces random strings.
|
|
// Each character is selected from one of the ranges of ur(UnicodeRanges).
|
|
// Each range has an equal probability of being chosen. If there are no ranges,
|
|
// or a selected range has no characters (.Last < .First), this will panic.
|
|
// Do not modify any of the ranges in ur after calling this function.
|
|
func (ur UnicodeRanges) CustomStringFillFunc(n int) func(s *string, c Continue) {
|
|
// Check unicode ranges slice is empty.
|
|
if len(ur) == 0 {
|
|
panic("UnicodeRanges is empty")
|
|
}
|
|
// if not empty, each range should be checked.
|
|
for i := range ur {
|
|
ur[i].check()
|
|
}
|
|
return func(s *string, c Continue) {
|
|
*s = ur.randString(c.Rand, n)
|
|
}
|
|
}
|
|
|
|
// randString of UnicodeRanges makes a random string up to 20 characters long.
|
|
// Each character is selected form one of the ranges of ur(UnicodeRanges),
|
|
// and each range has an equal probability of being chosen.
|
|
func (ur UnicodeRanges) randString(r *rand.Rand, max int) string {
|
|
if max == 0 {
|
|
max = defaultStringMaxLen
|
|
}
|
|
n := r.Intn(max)
|
|
sb := strings.Builder{}
|
|
sb.Grow(n)
|
|
for i := 0; i < n; i++ {
|
|
sb.WriteRune(ur[r.Intn(len(ur))].choose(r))
|
|
}
|
|
return sb.String()
|
|
}
|
|
|
|
// randString makes a random string up to 20 characters long. The returned string
|
|
// may include a variety of (valid) UTF-8 encodings.
|
|
func randString(r *rand.Rand, max int) string {
|
|
return defaultUnicodeRanges.randString(r, max)
|
|
}
|
|
|
|
// randUint64 makes random 64 bit numbers.
|
|
// Weirdly, rand doesn't have a function that gives you 64 random bits.
|
|
func randUint64(r *rand.Rand) uint64 {
|
|
return uint64(r.Uint32())<<32 | uint64(r.Uint32())
|
|
}
|