blob: bf7f816d0e862d51574a7784f24cd5c6f48e95aa [file] [log] [blame]
khenaidoo5fc5cea2021-08-11 17:39:16 -04001// Copyright 2019 The Go Authors. All rights reserved.
2// Use of this source code is governed by a BSD-style
3// license that can be found in the LICENSE file.
4
5package proto
6
7import (
8 "google.golang.org/protobuf/encoding/protowire"
9 "google.golang.org/protobuf/internal/encoding/messageset"
10 "google.golang.org/protobuf/internal/order"
11 "google.golang.org/protobuf/internal/pragma"
12 "google.golang.org/protobuf/reflect/protoreflect"
13 "google.golang.org/protobuf/runtime/protoiface"
14)
15
16// MarshalOptions configures the marshaler.
17//
18// Example usage:
Joey Armstrongba3d9d12024-01-15 14:22:11 -050019//
20// b, err := MarshalOptions{Deterministic: true}.Marshal(m)
khenaidoo5fc5cea2021-08-11 17:39:16 -040021type MarshalOptions struct {
22 pragma.NoUnkeyedLiterals
23
24 // AllowPartial allows messages that have missing required fields to marshal
25 // without returning an error. If AllowPartial is false (the default),
26 // Marshal will return an error if there are any missing required fields.
27 AllowPartial bool
28
29 // Deterministic controls whether the same message will always be
30 // serialized to the same bytes within the same binary.
31 //
32 // Setting this option guarantees that repeated serialization of
33 // the same message will return the same bytes, and that different
34 // processes of the same binary (which may be executing on different
35 // machines) will serialize equal messages to the same bytes.
36 // It has no effect on the resulting size of the encoded message compared
37 // to a non-deterministic marshal.
38 //
39 // Note that the deterministic serialization is NOT canonical across
40 // languages. It is not guaranteed to remain stable over time. It is
41 // unstable across different builds with schema changes due to unknown
42 // fields. Users who need canonical serialization (e.g., persistent
43 // storage in a canonical form, fingerprinting, etc.) must define
44 // their own canonicalization specification and implement their own
45 // serializer rather than relying on this API.
46 //
47 // If deterministic serialization is requested, map entries will be
48 // sorted by keys in lexographical order. This is an implementation
49 // detail and subject to change.
50 Deterministic bool
51
52 // UseCachedSize indicates that the result of a previous Size call
53 // may be reused.
54 //
55 // Setting this option asserts that:
56 //
57 // 1. Size has previously been called on this message with identical
58 // options (except for UseCachedSize itself).
59 //
60 // 2. The message and all its submessages have not changed in any
61 // way since the Size call.
62 //
63 // If either of these invariants is violated,
64 // the results are undefined and may include panics or corrupted output.
65 //
66 // Implementations MAY take this option into account to provide
67 // better performance, but there is no guarantee that they will do so.
68 // There is absolutely no guarantee that Size followed by Marshal with
69 // UseCachedSize set will perform equivalently to Marshal alone.
70 UseCachedSize bool
71}
72
73// Marshal returns the wire-format encoding of m.
74func Marshal(m Message) ([]byte, error) {
75 // Treat nil message interface as an empty message; nothing to output.
76 if m == nil {
77 return nil, nil
78 }
79
80 out, err := MarshalOptions{}.marshal(nil, m.ProtoReflect())
81 if len(out.Buf) == 0 && err == nil {
82 out.Buf = emptyBytesForMessage(m)
83 }
84 return out.Buf, err
85}
86
87// Marshal returns the wire-format encoding of m.
88func (o MarshalOptions) Marshal(m Message) ([]byte, error) {
89 // Treat nil message interface as an empty message; nothing to output.
90 if m == nil {
91 return nil, nil
92 }
93
94 out, err := o.marshal(nil, m.ProtoReflect())
95 if len(out.Buf) == 0 && err == nil {
96 out.Buf = emptyBytesForMessage(m)
97 }
98 return out.Buf, err
99}
100
101// emptyBytesForMessage returns a nil buffer if and only if m is invalid,
102// otherwise it returns a non-nil empty buffer.
103//
104// This is to assist the edge-case where user-code does the following:
Joey Armstrongba3d9d12024-01-15 14:22:11 -0500105//
khenaidoo5fc5cea2021-08-11 17:39:16 -0400106// m1.OptionalBytes, _ = proto.Marshal(m2)
Joey Armstrongba3d9d12024-01-15 14:22:11 -0500107//
khenaidoo5fc5cea2021-08-11 17:39:16 -0400108// where they expect the proto2 "optional_bytes" field to be populated
109// if any only if m2 is a valid message.
110func emptyBytesForMessage(m Message) []byte {
111 if m == nil || !m.ProtoReflect().IsValid() {
112 return nil
113 }
114 return emptyBuf[:]
115}
116
117// MarshalAppend appends the wire-format encoding of m to b,
118// returning the result.
119func (o MarshalOptions) MarshalAppend(b []byte, m Message) ([]byte, error) {
120 // Treat nil message interface as an empty message; nothing to append.
121 if m == nil {
122 return b, nil
123 }
124
125 out, err := o.marshal(b, m.ProtoReflect())
126 return out.Buf, err
127}
128
129// MarshalState returns the wire-format encoding of a message.
130//
131// This method permits fine-grained control over the marshaler.
132// Most users should use Marshal instead.
133func (o MarshalOptions) MarshalState(in protoiface.MarshalInput) (protoiface.MarshalOutput, error) {
134 return o.marshal(in.Buf, in.Message)
135}
136
137// marshal is a centralized function that all marshal operations go through.
138// For profiling purposes, avoid changing the name of this function or
139// introducing other code paths for marshal that do not go through this.
140func (o MarshalOptions) marshal(b []byte, m protoreflect.Message) (out protoiface.MarshalOutput, err error) {
141 allowPartial := o.AllowPartial
142 o.AllowPartial = true
143 if methods := protoMethods(m); methods != nil && methods.Marshal != nil &&
144 !(o.Deterministic && methods.Flags&protoiface.SupportMarshalDeterministic == 0) {
145 in := protoiface.MarshalInput{
146 Message: m,
147 Buf: b,
148 }
149 if o.Deterministic {
150 in.Flags |= protoiface.MarshalDeterministic
151 }
152 if o.UseCachedSize {
153 in.Flags |= protoiface.MarshalUseCachedSize
154 }
155 if methods.Size != nil {
156 sout := methods.Size(protoiface.SizeInput{
157 Message: m,
158 Flags: in.Flags,
159 })
160 if cap(b) < len(b)+sout.Size {
161 in.Buf = make([]byte, len(b), growcap(cap(b), len(b)+sout.Size))
162 copy(in.Buf, b)
163 }
164 in.Flags |= protoiface.MarshalUseCachedSize
165 }
166 out, err = methods.Marshal(in)
167 } else {
168 out.Buf, err = o.marshalMessageSlow(b, m)
169 }
170 if err != nil {
171 return out, err
172 }
173 if allowPartial {
174 return out, nil
175 }
176 return out, checkInitialized(m)
177}
178
179func (o MarshalOptions) marshalMessage(b []byte, m protoreflect.Message) ([]byte, error) {
180 out, err := o.marshal(b, m)
181 return out.Buf, err
182}
183
184// growcap scales up the capacity of a slice.
185//
186// Given a slice with a current capacity of oldcap and a desired
187// capacity of wantcap, growcap returns a new capacity >= wantcap.
188//
189// The algorithm is mostly identical to the one used by append as of Go 1.14.
190func growcap(oldcap, wantcap int) (newcap int) {
191 if wantcap > oldcap*2 {
192 newcap = wantcap
193 } else if oldcap < 1024 {
194 // The Go 1.14 runtime takes this case when len(s) < 1024,
195 // not when cap(s) < 1024. The difference doesn't seem
196 // significant here.
197 newcap = oldcap * 2
198 } else {
199 newcap = oldcap
200 for 0 < newcap && newcap < wantcap {
201 newcap += newcap / 4
202 }
203 if newcap <= 0 {
204 newcap = wantcap
205 }
206 }
207 return newcap
208}
209
210func (o MarshalOptions) marshalMessageSlow(b []byte, m protoreflect.Message) ([]byte, error) {
211 if messageset.IsMessageSet(m.Descriptor()) {
212 return o.marshalMessageSet(b, m)
213 }
214 fieldOrder := order.AnyFieldOrder
215 if o.Deterministic {
216 // TODO: This should use a more natural ordering like NumberFieldOrder,
217 // but doing so breaks golden tests that make invalid assumption about
218 // output stability of this implementation.
219 fieldOrder = order.LegacyFieldOrder
220 }
221 var err error
222 order.RangeFields(m, fieldOrder, func(fd protoreflect.FieldDescriptor, v protoreflect.Value) bool {
223 b, err = o.marshalField(b, fd, v)
224 return err == nil
225 })
226 if err != nil {
227 return b, err
228 }
229 b = append(b, m.GetUnknown()...)
230 return b, nil
231}
232
233func (o MarshalOptions) marshalField(b []byte, fd protoreflect.FieldDescriptor, value protoreflect.Value) ([]byte, error) {
234 switch {
235 case fd.IsList():
236 return o.marshalList(b, fd, value.List())
237 case fd.IsMap():
238 return o.marshalMap(b, fd, value.Map())
239 default:
240 b = protowire.AppendTag(b, fd.Number(), wireTypes[fd.Kind()])
241 return o.marshalSingular(b, fd, value)
242 }
243}
244
245func (o MarshalOptions) marshalList(b []byte, fd protoreflect.FieldDescriptor, list protoreflect.List) ([]byte, error) {
246 if fd.IsPacked() && list.Len() > 0 {
247 b = protowire.AppendTag(b, fd.Number(), protowire.BytesType)
248 b, pos := appendSpeculativeLength(b)
249 for i, llen := 0, list.Len(); i < llen; i++ {
250 var err error
251 b, err = o.marshalSingular(b, fd, list.Get(i))
252 if err != nil {
253 return b, err
254 }
255 }
256 b = finishSpeculativeLength(b, pos)
257 return b, nil
258 }
259
260 kind := fd.Kind()
261 for i, llen := 0, list.Len(); i < llen; i++ {
262 var err error
263 b = protowire.AppendTag(b, fd.Number(), wireTypes[kind])
264 b, err = o.marshalSingular(b, fd, list.Get(i))
265 if err != nil {
266 return b, err
267 }
268 }
269 return b, nil
270}
271
272func (o MarshalOptions) marshalMap(b []byte, fd protoreflect.FieldDescriptor, mapv protoreflect.Map) ([]byte, error) {
273 keyf := fd.MapKey()
274 valf := fd.MapValue()
275 keyOrder := order.AnyKeyOrder
276 if o.Deterministic {
277 keyOrder = order.GenericKeyOrder
278 }
279 var err error
280 order.RangeEntries(mapv, keyOrder, func(key protoreflect.MapKey, value protoreflect.Value) bool {
281 b = protowire.AppendTag(b, fd.Number(), protowire.BytesType)
282 var pos int
283 b, pos = appendSpeculativeLength(b)
284
285 b, err = o.marshalField(b, keyf, key.Value())
286 if err != nil {
287 return false
288 }
289 b, err = o.marshalField(b, valf, value)
290 if err != nil {
291 return false
292 }
293 b = finishSpeculativeLength(b, pos)
294 return true
295 })
296 return b, err
297}
298
299// When encoding length-prefixed fields, we speculatively set aside some number of bytes
300// for the length, encode the data, and then encode the length (shifting the data if necessary
301// to make room).
302const speculativeLength = 1
303
304func appendSpeculativeLength(b []byte) ([]byte, int) {
305 pos := len(b)
306 b = append(b, "\x00\x00\x00\x00"[:speculativeLength]...)
307 return b, pos
308}
309
310func finishSpeculativeLength(b []byte, pos int) []byte {
311 mlen := len(b) - pos - speculativeLength
312 msiz := protowire.SizeVarint(uint64(mlen))
313 if msiz != speculativeLength {
314 for i := 0; i < msiz-speculativeLength; i++ {
315 b = append(b, 0)
316 }
317 copy(b[pos+msiz:], b[pos+speculativeLength:])
318 b = b[:pos+msiz+mlen]
319 }
320 protowire.AppendVarint(b[:pos], uint64(mlen))
321 return b
322}