You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
tailscale/util/deephash/deephash.go

529 lines
14 KiB
Go

// Copyright (c) 2020 Tailscale Inc & 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 deephash hashes a Go value recursively, in a predictable order,
// without looping. The hash is only valid within the lifetime of a program.
// Users should not store the hash on disk or send it over the network.
// The hash is sufficiently strong and unique such that
// Hash(&x) == Hash(&y) is an appropriate replacement for x == y.
//
util/deephash: remove unnecessary formatting for structs and slices (#2571) The index for every struct field or slice element and the number of fields for the struct is unncessary. The hashing of Go values is unambiguous because every type (except maps) encodes in a parsable manner. So long as we know the type information, we could theoretically decode every value (except for maps). At a high level: * numbers are encoded as fixed-width records according to precision. * strings (and AppendTo output) are encoded with a fixed-width length, followed by the contents of the buffer. * slices are prefixed by a fixed-width length, followed by the encoding of each value. So long as we know the type of each element, we could theoretically decode each element. * arrays are encoded just like slices, but elide the length since it is determined from the Go type. * maps are encoded first with a byte indicating whether it is a cycle. If a cycle, it is followed by a fixed-width index for the pointer, otherwise followed by the SHA-256 hash of its contents. The encoding of maps is not decodeable, but a SHA-256 hash is sufficient to avoid ambiguities. * interfaces are encoded first with a byte indicating whether it is nil. If not nil, it is followed by a fixed-width index for the type, and then the encoding for the underlying value. Having the type be encoded first ensures that the value could theoretically be decoded next. * pointers are encoded first with a byte indicating whether it is 1) nil, 2) a cycle, or 3) newly seen. If a cycle, it is followed by a fixed-width index for the pointer. If newly seen, it is followed by the encoding for the pointed-at value. Removing unnecessary details speeds up hashing: name old time/op new time/op delta Hash-8 76.0µs ± 1% 55.8µs ± 2% -26.62% (p=0.000 n=10+10) HashMapAcyclic-8 61.9µs ± 0% 62.0µs ± 0% ~ (p=0.666 n=9+9) TailcfgNode-8 10.2µs ± 1% 7.5µs ± 1% -26.90% (p=0.000 n=10+9) HashArray-8 1.07µs ± 1% 0.70µs ± 1% -34.67% (p=0.000 n=10+9) Signed-off-by: Joe Tsai <joetsai@digital-static.net>
3 years ago
// The definition of equality is identical to reflect.DeepEqual except:
// - Floating-point values are compared based on the raw bits,
// which means that NaNs (with the same bit pattern) are treated as equal.
// - time.Time are compared based on whether they are the same instant in time
// and also in the same zone offset. Monotonic measurements and zone names
// are ignored as part of the hash.
// - netip.Addr are compared based on a shallow comparison of the struct.
util/deephash: remove unnecessary formatting for structs and slices (#2571) The index for every struct field or slice element and the number of fields for the struct is unncessary. The hashing of Go values is unambiguous because every type (except maps) encodes in a parsable manner. So long as we know the type information, we could theoretically decode every value (except for maps). At a high level: * numbers are encoded as fixed-width records according to precision. * strings (and AppendTo output) are encoded with a fixed-width length, followed by the contents of the buffer. * slices are prefixed by a fixed-width length, followed by the encoding of each value. So long as we know the type of each element, we could theoretically decode each element. * arrays are encoded just like slices, but elide the length since it is determined from the Go type. * maps are encoded first with a byte indicating whether it is a cycle. If a cycle, it is followed by a fixed-width index for the pointer, otherwise followed by the SHA-256 hash of its contents. The encoding of maps is not decodeable, but a SHA-256 hash is sufficient to avoid ambiguities. * interfaces are encoded first with a byte indicating whether it is nil. If not nil, it is followed by a fixed-width index for the type, and then the encoding for the underlying value. Having the type be encoded first ensures that the value could theoretically be decoded next. * pointers are encoded first with a byte indicating whether it is 1) nil, 2) a cycle, or 3) newly seen. If a cycle, it is followed by a fixed-width index for the pointer. If newly seen, it is followed by the encoding for the pointed-at value. Removing unnecessary details speeds up hashing: name old time/op new time/op delta Hash-8 76.0µs ± 1% 55.8µs ± 2% -26.62% (p=0.000 n=10+10) HashMapAcyclic-8 61.9µs ± 0% 62.0µs ± 0% ~ (p=0.666 n=9+9) TailcfgNode-8 10.2µs ± 1% 7.5µs ± 1% -26.90% (p=0.000 n=10+9) HashArray-8 1.07µs ± 1% 0.70µs ± 1% -34.67% (p=0.000 n=10+9) Signed-off-by: Joe Tsai <joetsai@digital-static.net>
3 years ago
//
// WARNING: This package, like most of the tailscale.com Go module,
// should be considered Tailscale-internal; we make no API promises.
package deephash
import (
"crypto/sha256"
"encoding/binary"
"encoding/hex"
"reflect"
"sync"
"time"
"tailscale.com/util/hashx"
)
util/deephash: remove unnecessary formatting for structs and slices (#2571) The index for every struct field or slice element and the number of fields for the struct is unncessary. The hashing of Go values is unambiguous because every type (except maps) encodes in a parsable manner. So long as we know the type information, we could theoretically decode every value (except for maps). At a high level: * numbers are encoded as fixed-width records according to precision. * strings (and AppendTo output) are encoded with a fixed-width length, followed by the contents of the buffer. * slices are prefixed by a fixed-width length, followed by the encoding of each value. So long as we know the type of each element, we could theoretically decode each element. * arrays are encoded just like slices, but elide the length since it is determined from the Go type. * maps are encoded first with a byte indicating whether it is a cycle. If a cycle, it is followed by a fixed-width index for the pointer, otherwise followed by the SHA-256 hash of its contents. The encoding of maps is not decodeable, but a SHA-256 hash is sufficient to avoid ambiguities. * interfaces are encoded first with a byte indicating whether it is nil. If not nil, it is followed by a fixed-width index for the type, and then the encoding for the underlying value. Having the type be encoded first ensures that the value could theoretically be decoded next. * pointers are encoded first with a byte indicating whether it is 1) nil, 2) a cycle, or 3) newly seen. If a cycle, it is followed by a fixed-width index for the pointer. If newly seen, it is followed by the encoding for the pointed-at value. Removing unnecessary details speeds up hashing: name old time/op new time/op delta Hash-8 76.0µs ± 1% 55.8µs ± 2% -26.62% (p=0.000 n=10+10) HashMapAcyclic-8 61.9µs ± 0% 62.0µs ± 0% ~ (p=0.666 n=9+9) TailcfgNode-8 10.2µs ± 1% 7.5µs ± 1% -26.90% (p=0.000 n=10+9) HashArray-8 1.07µs ± 1% 0.70µs ± 1% -34.67% (p=0.000 n=10+9) Signed-off-by: Joe Tsai <joetsai@digital-static.net>
3 years ago
// There is much overlap between the theory of serialization and hashing.
// A hash (useful for determining equality) can be produced by printing a value
util/deephash: remove unnecessary formatting for structs and slices (#2571) The index for every struct field or slice element and the number of fields for the struct is unncessary. The hashing of Go values is unambiguous because every type (except maps) encodes in a parsable manner. So long as we know the type information, we could theoretically decode every value (except for maps). At a high level: * numbers are encoded as fixed-width records according to precision. * strings (and AppendTo output) are encoded with a fixed-width length, followed by the contents of the buffer. * slices are prefixed by a fixed-width length, followed by the encoding of each value. So long as we know the type of each element, we could theoretically decode each element. * arrays are encoded just like slices, but elide the length since it is determined from the Go type. * maps are encoded first with a byte indicating whether it is a cycle. If a cycle, it is followed by a fixed-width index for the pointer, otherwise followed by the SHA-256 hash of its contents. The encoding of maps is not decodeable, but a SHA-256 hash is sufficient to avoid ambiguities. * interfaces are encoded first with a byte indicating whether it is nil. If not nil, it is followed by a fixed-width index for the type, and then the encoding for the underlying value. Having the type be encoded first ensures that the value could theoretically be decoded next. * pointers are encoded first with a byte indicating whether it is 1) nil, 2) a cycle, or 3) newly seen. If a cycle, it is followed by a fixed-width index for the pointer. If newly seen, it is followed by the encoding for the pointed-at value. Removing unnecessary details speeds up hashing: name old time/op new time/op delta Hash-8 76.0µs ± 1% 55.8µs ± 2% -26.62% (p=0.000 n=10+10) HashMapAcyclic-8 61.9µs ± 0% 62.0µs ± 0% ~ (p=0.666 n=9+9) TailcfgNode-8 10.2µs ± 1% 7.5µs ± 1% -26.90% (p=0.000 n=10+9) HashArray-8 1.07µs ± 1% 0.70µs ± 1% -34.67% (p=0.000 n=10+9) Signed-off-by: Joe Tsai <joetsai@digital-static.net>
3 years ago
// and hashing the output. The format must:
// * be deterministic such that the same value hashes to the same output, and
// * be parsable such that the same value can be reproduced by the output.
//
// The logic below hashes a value by printing it to a hash.Hash.
// To be parsable, it assumes that we know the Go type of each value:
// * scalar types (e.g., bool or int32) are directly printed as their
// underlying memory representation.
// * list types (e.g., strings and slices) are prefixed by a
// fixed-width length field, followed by the contents of the list.
util/deephash: remove unnecessary formatting for structs and slices (#2571) The index for every struct field or slice element and the number of fields for the struct is unncessary. The hashing of Go values is unambiguous because every type (except maps) encodes in a parsable manner. So long as we know the type information, we could theoretically decode every value (except for maps). At a high level: * numbers are encoded as fixed-width records according to precision. * strings (and AppendTo output) are encoded with a fixed-width length, followed by the contents of the buffer. * slices are prefixed by a fixed-width length, followed by the encoding of each value. So long as we know the type of each element, we could theoretically decode each element. * arrays are encoded just like slices, but elide the length since it is determined from the Go type. * maps are encoded first with a byte indicating whether it is a cycle. If a cycle, it is followed by a fixed-width index for the pointer, otherwise followed by the SHA-256 hash of its contents. The encoding of maps is not decodeable, but a SHA-256 hash is sufficient to avoid ambiguities. * interfaces are encoded first with a byte indicating whether it is nil. If not nil, it is followed by a fixed-width index for the type, and then the encoding for the underlying value. Having the type be encoded first ensures that the value could theoretically be decoded next. * pointers are encoded first with a byte indicating whether it is 1) nil, 2) a cycle, or 3) newly seen. If a cycle, it is followed by a fixed-width index for the pointer. If newly seen, it is followed by the encoding for the pointed-at value. Removing unnecessary details speeds up hashing: name old time/op new time/op delta Hash-8 76.0µs ± 1% 55.8µs ± 2% -26.62% (p=0.000 n=10+10) HashMapAcyclic-8 61.9µs ± 0% 62.0µs ± 0% ~ (p=0.666 n=9+9) TailcfgNode-8 10.2µs ± 1% 7.5µs ± 1% -26.90% (p=0.000 n=10+9) HashArray-8 1.07µs ± 1% 0.70µs ± 1% -34.67% (p=0.000 n=10+9) Signed-off-by: Joe Tsai <joetsai@digital-static.net>
3 years ago
// * slices, arrays, and structs print each element/field consecutively.
// * interfaces print with a 1-byte prefix indicating whether it is nil.
// If non-nil, it is followed by a fixed-width field of the type index,
// followed by the format of the underlying value.
// * pointers print with a 1-byte prefix indicating whether the pointer is
// 1) nil, 2) previously seen, or 3) newly seen. Previously seen pointers are
// followed by a fixed-width field with the index of the previous pointer.
// Newly seen pointers are followed by the format of the underlying value.
// * maps print with a 1-byte prefix indicating whether the map pointer is
// 1) nil, 2) previously seen, or 3) newly seen. Previously seen pointers
// are followed by a fixed-width field of the index of the previous pointer.
// Newly seen maps are printed with a fixed-width length field, followed by
// a fixed-width field with the XOR of the hash of every map entry.
// With a sufficiently strong hash, this value is theoretically "parsable"
// by looking up the hash in a magical map that returns the set of entries
// for that given hash.
util/deephash: remove unnecessary formatting for structs and slices (#2571) The index for every struct field or slice element and the number of fields for the struct is unncessary. The hashing of Go values is unambiguous because every type (except maps) encodes in a parsable manner. So long as we know the type information, we could theoretically decode every value (except for maps). At a high level: * numbers are encoded as fixed-width records according to precision. * strings (and AppendTo output) are encoded with a fixed-width length, followed by the contents of the buffer. * slices are prefixed by a fixed-width length, followed by the encoding of each value. So long as we know the type of each element, we could theoretically decode each element. * arrays are encoded just like slices, but elide the length since it is determined from the Go type. * maps are encoded first with a byte indicating whether it is a cycle. If a cycle, it is followed by a fixed-width index for the pointer, otherwise followed by the SHA-256 hash of its contents. The encoding of maps is not decodeable, but a SHA-256 hash is sufficient to avoid ambiguities. * interfaces are encoded first with a byte indicating whether it is nil. If not nil, it is followed by a fixed-width index for the type, and then the encoding for the underlying value. Having the type be encoded first ensures that the value could theoretically be decoded next. * pointers are encoded first with a byte indicating whether it is 1) nil, 2) a cycle, or 3) newly seen. If a cycle, it is followed by a fixed-width index for the pointer. If newly seen, it is followed by the encoding for the pointed-at value. Removing unnecessary details speeds up hashing: name old time/op new time/op delta Hash-8 76.0µs ± 1% 55.8µs ± 2% -26.62% (p=0.000 n=10+10) HashMapAcyclic-8 61.9µs ± 0% 62.0µs ± 0% ~ (p=0.666 n=9+9) TailcfgNode-8 10.2µs ± 1% 7.5µs ± 1% -26.90% (p=0.000 n=10+9) HashArray-8 1.07µs ± 1% 0.70µs ± 1% -34.67% (p=0.000 n=10+9) Signed-off-by: Joe Tsai <joetsai@digital-static.net>
3 years ago
// hasher is reusable state for hashing a value.
// Get one via hasherPool.
type hasher struct {
hashx.Block512
util/deephash: improve cycle detection (#2470) The previous algorithm used a map of all visited pointers. The strength of this approach is that it quickly prunes any nodes that we have ever visited before. The detriment of the approach is that pruning is heavily dependent on the order that pointers were visited. This is especially relevant for hashing a map where map entries are visited in a non-deterministic manner, which would cause the map hash to be non-deterministic (which defeats the point of a hash). This new algorithm uses a stack of all visited pointers, similar to how github.com/google/go-cmp performs cycle detection. When we visit a pointer, we push it onto the stack, and when we leave a pointer, we pop it from the stack. Before visiting a pointer, we first check whether the pointer exists anywhere in the stack. If yes, then we prune the node. The detriment of this approach is that we may hash a node more often than before since we do not prune as aggressively. The set of visited pointers up until any node is only the path of nodes up to that node and not any other pointers that may have been visited elsewhere. This provides us deterministic hashing regardless of visit order. We can now delete hashMapFallback and associated complexity, which only exists because the previous approach was non-deterministic in the presence of cycles. This fixes a failure of the old algorithm where obviously different values are treated as equal because the pruning was too aggresive. See https://github.com/tailscale/tailscale/issues/2443#issuecomment-883653534 The new algorithm is slightly slower since it prunes less aggresively: name old time/op new time/op delta Hash-8 66.1µs ± 1% 68.8µs ± 1% +4.09% (p=0.000 n=19+19) HashMapAcyclic-8 63.0µs ± 1% 62.5µs ± 1% -0.76% (p=0.000 n=18+19) TailcfgNode-8 9.79µs ± 2% 9.88µs ± 1% +0.95% (p=0.000 n=19+17) HashArray-8 643ns ± 1% 653ns ± 1% +1.64% (p=0.000 n=19+19) However, a slower but more correct algorithm seems more favorable than a faster but incorrect algorithm. Signed-off-by: Joe Tsai <joetsai@digital-static.net>
3 years ago
visitStack visitStack
}
var hasherPool = &sync.Pool{
New: func() any { return new(hasher) },
}
func (h *hasher) reset() {
if h.Block512.Hash == nil {
h.Block512.Hash = sha256.New()
}
h.Block512.Reset()
}
// hashType hashes a reflect.Type.
// The hash is only consistent within the lifetime of a program.
func (h *hasher) hashType(t reflect.Type) {
// This approach relies on reflect.Type always being backed by a unique
// *reflect.rtype pointer. A safer approach is to use a global sync.Map
// that maps reflect.Type to some arbitrary and unique index.
// While safer, it requires global state with memory that can never be GC'd.
rtypeAddr := reflect.ValueOf(t).Pointer() // address of *reflect.rtype
h.HashUint64(uint64(rtypeAddr))
}
func (h *hasher) sum() (s Sum) {
h.Sum(s.sum[:0])
return s
}
// Sum is an opaque checksum type that is comparable.
type Sum struct {
sum [sha256.Size]byte
}
func (s1 *Sum) xor(s2 Sum) {
for i := 0; i < sha256.Size; i++ {
s1.sum[i] ^= s2.sum[i]
}
}
func (s Sum) String() string {
// Note: if we change this, keep in sync with AppendTo
return hex.EncodeToString(s.sum[:])
}
// AppendTo appends the string encoding of this sum (as returned by the String
// method) to the provided byte slice and returns the extended buffer.
func (s Sum) AppendTo(b []byte) []byte {
// TODO: switch to upstream implementation if accepted:
// https://github.com/golang/go/issues/53693
var lb [len(s.sum) * 2]byte
hex.Encode(lb[:], s.sum[:])
return append(b, lb[:]...)
}
var (
seedOnce sync.Once
seed uint64
)
func initSeed() {
seed = uint64(time.Now().UnixNano())
}
// Hash returns the hash of v.
func Hash[T any](v *T) Sum {
h := hasherPool.Get().(*hasher)
defer hasherPool.Put(h)
h.reset()
seedOnce.Do(initSeed)
h.HashUint64(seed)
// Always treat the Hash input as if it were an interface by including
// a hash of the type. This ensures that hashing of two different types
// but with the same value structure produces different hashes.
t := reflect.TypeOf(v).Elem()
h.hashType(t)
if v == nil {
h.HashUint8(0) // indicates nil
} else {
h.HashUint8(1) // indicates visiting pointer element
p := pointerOf(reflect.ValueOf(v))
hash := lookupTypeHasher(t)
hash(h, p)
}
return h.sum()
}
// HasherForType returns a hash that is specialized for the provided type.
func HasherForType[T any]() func(*T) Sum {
var v *T
seedOnce.Do(initSeed)
t := reflect.TypeOf(v).Elem()
hash := lookupTypeHasher(t)
return func(v *T) (s Sum) {
// This logic is identical to Hash, but pull out a few statements.
h := hasherPool.Get().(*hasher)
defer hasherPool.Put(h)
h.reset()
h.HashUint64(seed)
h.hashType(t)
if v == nil {
h.HashUint8(0) // indicates nil
} else {
h.HashUint8(1) // indicates visiting pointer element
p := pointerOf(reflect.ValueOf(v))
hash(h, p)
}
return h.sum()
}
}
// Update sets last to the hash of v and reports whether its value changed.
func Update[T any](last *Sum, v *T) (changed bool) {
sum := Hash(v)
changed = sum != *last
if changed {
*last = sum
}
return changed
}
// typeHasherFunc hashes the value pointed at by p for a given type.
// For example, if t is a bool, then p is a *bool.
// The provided pointer must always be non-nil.
type typeHasherFunc func(h *hasher, p pointer)
var typeHasherCache sync.Map // map[reflect.Type]typeHasherFunc
func lookupTypeHasher(t reflect.Type) typeHasherFunc {
if v, ok := typeHasherCache.Load(t); ok {
return v.(typeHasherFunc)
}
hash := makeTypeHasher(t)
v, _ := typeHasherCache.LoadOrStore(t, hash)
return v.(typeHasherFunc)
}
func makeTypeHasher(t reflect.Type) typeHasherFunc {
// Types with specific hashing.
switch t {
case timeTimeType:
return hashTime
case netipAddrType:
return hashAddr
}
// Types that can have their memory representation directly hashed.
if typeIsMemHashable(t) {
return makeMemHasher(t.Size())
}
switch t.Kind() {
case reflect.String:
return hashString
case reflect.Array:
return makeArrayHasher(t)
case reflect.Slice:
return makeSliceHasher(t)
case reflect.Struct:
return makeStructHasher(t)
case reflect.Map:
return makeMapHasher(t)
case reflect.Pointer:
return makePointerHasher(t)
case reflect.Interface:
return makeInterfaceHasher(t)
default: // Func, Chan, UnsafePointer
return func(*hasher, pointer) {}
}
}
func hashTime(h *hasher, p pointer) {
// Include the zone offset (but not the name) to keep
// Hash(t1) == Hash(t2) being semantically equivalent to
// t1.Format(time.RFC3339Nano) == t2.Format(time.RFC3339Nano).
t := *p.asTime()
_, offset := t.Zone()
h.HashUint64(uint64(t.Unix()))
h.HashUint32(uint32(t.Nanosecond()))
h.HashUint32(uint32(offset))
}
func hashAddr(h *hasher, p pointer) {
// The formatting of netip.Addr covers the
// IP version, the address, and the optional zone name (for v6).
// This is equivalent to a1.MarshalBinary() == a2.MarshalBinary().
ip := *p.asAddr()
switch {
case !ip.IsValid():
h.HashUint64(0)
case ip.Is4():
b := ip.As4()
h.HashUint64(4)
h.HashUint32(binary.LittleEndian.Uint32(b[:]))
case ip.Is6():
b := ip.As16()
z := ip.Zone()
h.HashUint64(16 + uint64(len(z)))
h.HashUint64(binary.LittleEndian.Uint64(b[:8]))
h.HashUint64(binary.LittleEndian.Uint64(b[8:]))
h.HashString(z)
}
}
func hashString(h *hasher, p pointer) {
s := *p.asString()
h.HashUint64(uint64(len(s)))
h.HashString(s)
}
func makeMemHasher(n uintptr) typeHasherFunc {
return func(h *hasher, p pointer) {
h.HashBytes(p.asMemory(n))
}
}
func makeArrayHasher(t reflect.Type) typeHasherFunc {
var once sync.Once
var hashElem typeHasherFunc
init := func() {
hashElem = lookupTypeHasher(t.Elem())
}
n := t.Len() // number of array elements
nb := t.Elem().Size() // byte size of each array element
return func(h *hasher, p pointer) {
once.Do(init)
for i := 0; i < n; i++ {
hashElem(h, p.arrayIndex(i, nb))
}
}
}
func makeSliceHasher(t reflect.Type) typeHasherFunc {
nb := t.Elem().Size() // byte size of each slice element
if typeIsMemHashable(t.Elem()) {
return func(h *hasher, p pointer) {
pa := p.sliceArray()
if pa.isNil() {
h.HashUint8(0) // indicates nil
return
}
h.HashUint8(1) // indicates visiting slice
n := p.sliceLen()
b := pa.asMemory(uintptr(n) * nb)
h.HashUint64(uint64(n))
h.HashBytes(b)
}
}
var once sync.Once
var hashElem typeHasherFunc
init := func() {
hashElem = lookupTypeHasher(t.Elem())
if typeIsRecursive(t) {
hashElemDefault := hashElem
hashElem = func(h *hasher, p pointer) {
if idx, ok := h.visitStack.seen(p.p); ok {
h.HashUint8(2) // indicates cycle
h.HashUint64(uint64(idx))
return
}
h.HashUint8(1) // indicates visiting slice element
h.visitStack.push(p.p)
defer h.visitStack.pop(p.p)
hashElemDefault(h, p)
}
}
}
return func(h *hasher, p pointer) {
pa := p.sliceArray()
if pa.isNil() {
h.HashUint8(0) // indicates nil
return
}
once.Do(init)
h.HashUint8(1) // indicates visiting slice
n := p.sliceLen()
h.HashUint64(uint64(n))
for i := 0; i < n; i++ {
pe := pa.arrayIndex(i, nb)
hashElem(h, pe)
}
}
}
func makeStructHasher(t reflect.Type) typeHasherFunc {
type fieldHasher struct {
idx int // index of field for reflect.Type.Field(n); negative if memory is directly hashable
hash typeHasherFunc // only valid if idx is not negative
offset uintptr
size uintptr
}
var once sync.Once
var fields []fieldHasher
init := func() {
for i, numField := 0, t.NumField(); i < numField; i++ {
sf := t.Field(i)
f := fieldHasher{i, nil, sf.Offset, sf.Type.Size()}
if typeIsMemHashable(sf.Type) {
f.idx = -1
}
// Combine with previous field if both contiguous and mem-hashable.
if f.idx < 0 && len(fields) > 0 {
if last := &fields[len(fields)-1]; last.idx < 0 && last.offset+last.size == f.offset {
last.size += f.size
continue
}
}
fields = append(fields, f)
}
for i, f := range fields {
if f.idx >= 0 {
fields[i].hash = lookupTypeHasher(t.Field(f.idx).Type)
}
}
}
return func(h *hasher, p pointer) {
once.Do(init)
for _, field := range fields {
pf := p.structField(field.idx, field.offset, field.size)
if field.idx < 0 {
h.HashBytes(pf.asMemory(field.size))
} else {
field.hash(h, pf)
}
}
}
}
func makeMapHasher(t reflect.Type) typeHasherFunc {
var once sync.Once
var hashKey, hashValue typeHasherFunc
var isRecursive bool
init := func() {
hashKey = lookupTypeHasher(t.Key())
hashValue = lookupTypeHasher(t.Elem())
isRecursive = typeIsRecursive(t)
}
return func(h *hasher, p pointer) {
v := p.asValue(t).Elem() // reflect.Map kind
if v.IsNil() {
h.HashUint8(0) // indicates nil
return
}
once.Do(init)
if isRecursive {
pm := v.UnsafePointer() // underlying pointer of map
if idx, ok := h.visitStack.seen(pm); ok {
h.HashUint8(2) // indicates cycle
h.HashUint64(uint64(idx))
return
}
h.visitStack.push(pm)
defer h.visitStack.pop(pm)
}
h.HashUint8(1) // indicates visiting map entries
h.HashUint64(uint64(v.Len()))
mh := mapHasherPool.Get().(*mapHasher)
defer mapHasherPool.Put(mh)
// Hash a map in a sort-free manner.
// It relies on a map being a an unordered set of KV entries.
// So long as we hash each KV entry together, we can XOR all the
// individual hashes to produce a unique hash for the entire map.
k := mh.valKey.get(v.Type().Key())
e := mh.valElem.get(v.Type().Elem())
mh.sum = Sum{}
mh.h.visitStack = h.visitStack // always use the parent's visit stack to avoid cycles
for iter := v.MapRange(); iter.Next(); {
k.SetIterKey(iter)
e.SetIterValue(iter)
mh.h.reset()
hashKey(&mh.h, pointerOf(k.Addr()))
hashValue(&mh.h, pointerOf(e.Addr()))
mh.sum.xor(mh.h.sum())
}
h.HashBytes(mh.sum.sum[:])
}
}
func makePointerHasher(t reflect.Type) typeHasherFunc {
var once sync.Once
var hashElem typeHasherFunc
var isRecursive bool
init := func() {
hashElem = lookupTypeHasher(t.Elem())
isRecursive = typeIsRecursive(t)
}
return func(h *hasher, p pointer) {
pe := p.pointerElem()
if pe.isNil() {
h.HashUint8(0) // indicates nil
return
}
once.Do(init)
if isRecursive {
if idx, ok := h.visitStack.seen(pe.p); ok {
h.HashUint8(2) // indicates cycle
h.HashUint64(uint64(idx))
return
}
h.visitStack.push(pe.p)
defer h.visitStack.pop(pe.p)
}
h.HashUint8(1) // indicates visiting a pointer element
hashElem(h, pe)
}
}
func makeInterfaceHasher(t reflect.Type) typeHasherFunc {
return func(h *hasher, p pointer) {
v := p.asValue(t).Elem() // reflect.Interface kind
if v.IsNil() {
h.HashUint8(0) // indicates nil
return
}
h.HashUint8(1) // indicates visiting an interface value
v = v.Elem()
t := v.Type()
h.hashType(t)
va := reflect.New(t).Elem()
va.Set(v)
hashElem := lookupTypeHasher(t)
hashElem(h, pointerOf(va.Addr()))
}
}
type mapHasher struct {
h hasher
valKey valueCache
valElem valueCache
sum Sum
}
var mapHasherPool = &sync.Pool{
New: func() any { return new(mapHasher) },
}
type valueCache map[reflect.Type]reflect.Value
// get returns an addressable reflect.Value for the given type.
func (c *valueCache) get(t reflect.Type) reflect.Value {
v, ok := (*c)[t]
if !ok {
v = reflect.New(t).Elem()
if *c == nil {
*c = make(valueCache)
}
(*c)[t] = v
}
return v
}