khenaidoo | 7d3c558 | 2021-08-11 18:09:44 -0400 | [diff] [blame] | 1 | // 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 | |
| 5 | package proto |
| 6 | |
| 7 | import ( |
Akash Reddy Kankanala | 92dfdf8 | 2025-03-23 22:07:09 +0530 | [diff] [blame^] | 8 | "errors" |
| 9 | "fmt" |
| 10 | |
khenaidoo | 7d3c558 | 2021-08-11 18:09:44 -0400 | [diff] [blame] | 11 | "google.golang.org/protobuf/encoding/protowire" |
| 12 | "google.golang.org/protobuf/internal/encoding/messageset" |
| 13 | "google.golang.org/protobuf/internal/order" |
| 14 | "google.golang.org/protobuf/internal/pragma" |
| 15 | "google.golang.org/protobuf/reflect/protoreflect" |
| 16 | "google.golang.org/protobuf/runtime/protoiface" |
Akash Reddy Kankanala | 92dfdf8 | 2025-03-23 22:07:09 +0530 | [diff] [blame^] | 17 | |
| 18 | protoerrors "google.golang.org/protobuf/internal/errors" |
khenaidoo | 7d3c558 | 2021-08-11 18:09:44 -0400 | [diff] [blame] | 19 | ) |
| 20 | |
| 21 | // MarshalOptions configures the marshaler. |
| 22 | // |
| 23 | // Example usage: |
Akash Reddy Kankanala | 92dfdf8 | 2025-03-23 22:07:09 +0530 | [diff] [blame^] | 24 | // |
| 25 | // b, err := MarshalOptions{Deterministic: true}.Marshal(m) |
khenaidoo | 7d3c558 | 2021-08-11 18:09:44 -0400 | [diff] [blame] | 26 | type MarshalOptions struct { |
| 27 | pragma.NoUnkeyedLiterals |
| 28 | |
| 29 | // AllowPartial allows messages that have missing required fields to marshal |
| 30 | // without returning an error. If AllowPartial is false (the default), |
| 31 | // Marshal will return an error if there are any missing required fields. |
| 32 | AllowPartial bool |
| 33 | |
| 34 | // Deterministic controls whether the same message will always be |
| 35 | // serialized to the same bytes within the same binary. |
| 36 | // |
| 37 | // Setting this option guarantees that repeated serialization of |
| 38 | // the same message will return the same bytes, and that different |
| 39 | // processes of the same binary (which may be executing on different |
| 40 | // machines) will serialize equal messages to the same bytes. |
| 41 | // It has no effect on the resulting size of the encoded message compared |
| 42 | // to a non-deterministic marshal. |
| 43 | // |
| 44 | // Note that the deterministic serialization is NOT canonical across |
| 45 | // languages. It is not guaranteed to remain stable over time. It is |
| 46 | // unstable across different builds with schema changes due to unknown |
| 47 | // fields. Users who need canonical serialization (e.g., persistent |
| 48 | // storage in a canonical form, fingerprinting, etc.) must define |
| 49 | // their own canonicalization specification and implement their own |
| 50 | // serializer rather than relying on this API. |
| 51 | // |
| 52 | // If deterministic serialization is requested, map entries will be |
| 53 | // sorted by keys in lexographical order. This is an implementation |
| 54 | // detail and subject to change. |
| 55 | Deterministic bool |
| 56 | |
| 57 | // UseCachedSize indicates that the result of a previous Size call |
| 58 | // may be reused. |
| 59 | // |
| 60 | // Setting this option asserts that: |
| 61 | // |
| 62 | // 1. Size has previously been called on this message with identical |
| 63 | // options (except for UseCachedSize itself). |
| 64 | // |
| 65 | // 2. The message and all its submessages have not changed in any |
Akash Reddy Kankanala | 92dfdf8 | 2025-03-23 22:07:09 +0530 | [diff] [blame^] | 66 | // way since the Size call. For lazily decoded messages, accessing |
| 67 | // a message results in decoding the message, which is a change. |
khenaidoo | 7d3c558 | 2021-08-11 18:09:44 -0400 | [diff] [blame] | 68 | // |
| 69 | // If either of these invariants is violated, |
| 70 | // the results are undefined and may include panics or corrupted output. |
| 71 | // |
| 72 | // Implementations MAY take this option into account to provide |
| 73 | // better performance, but there is no guarantee that they will do so. |
| 74 | // There is absolutely no guarantee that Size followed by Marshal with |
| 75 | // UseCachedSize set will perform equivalently to Marshal alone. |
| 76 | UseCachedSize bool |
| 77 | } |
| 78 | |
Akash Reddy Kankanala | 92dfdf8 | 2025-03-23 22:07:09 +0530 | [diff] [blame^] | 79 | // flags turns the specified MarshalOptions (user-facing) into |
| 80 | // protoiface.MarshalInputFlags (used internally by the marshaler). |
| 81 | // |
| 82 | // See impl.marshalOptions.Options for the inverse operation. |
| 83 | func (o MarshalOptions) flags() protoiface.MarshalInputFlags { |
| 84 | var flags protoiface.MarshalInputFlags |
| 85 | |
| 86 | // Note: o.AllowPartial is always forced to true by MarshalOptions.marshal, |
| 87 | // which is why it is not a part of MarshalInputFlags. |
| 88 | |
| 89 | if o.Deterministic { |
| 90 | flags |= protoiface.MarshalDeterministic |
| 91 | } |
| 92 | |
| 93 | if o.UseCachedSize { |
| 94 | flags |= protoiface.MarshalUseCachedSize |
| 95 | } |
| 96 | |
| 97 | return flags |
| 98 | } |
| 99 | |
khenaidoo | 7d3c558 | 2021-08-11 18:09:44 -0400 | [diff] [blame] | 100 | // Marshal returns the wire-format encoding of m. |
Akash Reddy Kankanala | 92dfdf8 | 2025-03-23 22:07:09 +0530 | [diff] [blame^] | 101 | // |
| 102 | // This is the most common entry point for encoding a Protobuf message. |
| 103 | // |
| 104 | // See the [MarshalOptions] type if you need more control. |
khenaidoo | 7d3c558 | 2021-08-11 18:09:44 -0400 | [diff] [blame] | 105 | func Marshal(m Message) ([]byte, error) { |
| 106 | // Treat nil message interface as an empty message; nothing to output. |
| 107 | if m == nil { |
| 108 | return nil, nil |
| 109 | } |
| 110 | |
| 111 | out, err := MarshalOptions{}.marshal(nil, m.ProtoReflect()) |
| 112 | if len(out.Buf) == 0 && err == nil { |
| 113 | out.Buf = emptyBytesForMessage(m) |
| 114 | } |
| 115 | return out.Buf, err |
| 116 | } |
| 117 | |
| 118 | // Marshal returns the wire-format encoding of m. |
| 119 | func (o MarshalOptions) Marshal(m Message) ([]byte, error) { |
| 120 | // Treat nil message interface as an empty message; nothing to output. |
| 121 | if m == nil { |
| 122 | return nil, nil |
| 123 | } |
| 124 | |
| 125 | out, err := o.marshal(nil, m.ProtoReflect()) |
| 126 | if len(out.Buf) == 0 && err == nil { |
| 127 | out.Buf = emptyBytesForMessage(m) |
| 128 | } |
| 129 | return out.Buf, err |
| 130 | } |
| 131 | |
| 132 | // emptyBytesForMessage returns a nil buffer if and only if m is invalid, |
| 133 | // otherwise it returns a non-nil empty buffer. |
| 134 | // |
| 135 | // This is to assist the edge-case where user-code does the following: |
Akash Reddy Kankanala | 92dfdf8 | 2025-03-23 22:07:09 +0530 | [diff] [blame^] | 136 | // |
khenaidoo | 7d3c558 | 2021-08-11 18:09:44 -0400 | [diff] [blame] | 137 | // m1.OptionalBytes, _ = proto.Marshal(m2) |
Akash Reddy Kankanala | 92dfdf8 | 2025-03-23 22:07:09 +0530 | [diff] [blame^] | 138 | // |
khenaidoo | 7d3c558 | 2021-08-11 18:09:44 -0400 | [diff] [blame] | 139 | // where they expect the proto2 "optional_bytes" field to be populated |
| 140 | // if any only if m2 is a valid message. |
| 141 | func emptyBytesForMessage(m Message) []byte { |
| 142 | if m == nil || !m.ProtoReflect().IsValid() { |
| 143 | return nil |
| 144 | } |
| 145 | return emptyBuf[:] |
| 146 | } |
| 147 | |
| 148 | // MarshalAppend appends the wire-format encoding of m to b, |
| 149 | // returning the result. |
Akash Reddy Kankanala | 92dfdf8 | 2025-03-23 22:07:09 +0530 | [diff] [blame^] | 150 | // |
| 151 | // This is a less common entry point than [Marshal], which is only needed if you |
| 152 | // need to supply your own buffers for performance reasons. |
khenaidoo | 7d3c558 | 2021-08-11 18:09:44 -0400 | [diff] [blame] | 153 | func (o MarshalOptions) MarshalAppend(b []byte, m Message) ([]byte, error) { |
| 154 | // Treat nil message interface as an empty message; nothing to append. |
| 155 | if m == nil { |
| 156 | return b, nil |
| 157 | } |
| 158 | |
| 159 | out, err := o.marshal(b, m.ProtoReflect()) |
| 160 | return out.Buf, err |
| 161 | } |
| 162 | |
| 163 | // MarshalState returns the wire-format encoding of a message. |
| 164 | // |
| 165 | // This method permits fine-grained control over the marshaler. |
Akash Reddy Kankanala | 92dfdf8 | 2025-03-23 22:07:09 +0530 | [diff] [blame^] | 166 | // Most users should use [Marshal] instead. |
khenaidoo | 7d3c558 | 2021-08-11 18:09:44 -0400 | [diff] [blame] | 167 | func (o MarshalOptions) MarshalState(in protoiface.MarshalInput) (protoiface.MarshalOutput, error) { |
| 168 | return o.marshal(in.Buf, in.Message) |
| 169 | } |
| 170 | |
| 171 | // marshal is a centralized function that all marshal operations go through. |
| 172 | // For profiling purposes, avoid changing the name of this function or |
| 173 | // introducing other code paths for marshal that do not go through this. |
| 174 | func (o MarshalOptions) marshal(b []byte, m protoreflect.Message) (out protoiface.MarshalOutput, err error) { |
| 175 | allowPartial := o.AllowPartial |
| 176 | o.AllowPartial = true |
| 177 | if methods := protoMethods(m); methods != nil && methods.Marshal != nil && |
| 178 | !(o.Deterministic && methods.Flags&protoiface.SupportMarshalDeterministic == 0) { |
| 179 | in := protoiface.MarshalInput{ |
| 180 | Message: m, |
| 181 | Buf: b, |
Akash Reddy Kankanala | 92dfdf8 | 2025-03-23 22:07:09 +0530 | [diff] [blame^] | 182 | Flags: o.flags(), |
khenaidoo | 7d3c558 | 2021-08-11 18:09:44 -0400 | [diff] [blame] | 183 | } |
| 184 | if methods.Size != nil { |
| 185 | sout := methods.Size(protoiface.SizeInput{ |
| 186 | Message: m, |
| 187 | Flags: in.Flags, |
| 188 | }) |
| 189 | if cap(b) < len(b)+sout.Size { |
| 190 | in.Buf = make([]byte, len(b), growcap(cap(b), len(b)+sout.Size)) |
| 191 | copy(in.Buf, b) |
| 192 | } |
| 193 | in.Flags |= protoiface.MarshalUseCachedSize |
| 194 | } |
| 195 | out, err = methods.Marshal(in) |
| 196 | } else { |
| 197 | out.Buf, err = o.marshalMessageSlow(b, m) |
| 198 | } |
| 199 | if err != nil { |
Akash Reddy Kankanala | 92dfdf8 | 2025-03-23 22:07:09 +0530 | [diff] [blame^] | 200 | var mismatch *protoerrors.SizeMismatchError |
| 201 | if errors.As(err, &mismatch) { |
| 202 | return out, fmt.Errorf("marshaling %s: %v", string(m.Descriptor().FullName()), err) |
| 203 | } |
khenaidoo | 7d3c558 | 2021-08-11 18:09:44 -0400 | [diff] [blame] | 204 | return out, err |
| 205 | } |
| 206 | if allowPartial { |
| 207 | return out, nil |
| 208 | } |
| 209 | return out, checkInitialized(m) |
| 210 | } |
| 211 | |
| 212 | func (o MarshalOptions) marshalMessage(b []byte, m protoreflect.Message) ([]byte, error) { |
| 213 | out, err := o.marshal(b, m) |
| 214 | return out.Buf, err |
| 215 | } |
| 216 | |
| 217 | // growcap scales up the capacity of a slice. |
| 218 | // |
| 219 | // Given a slice with a current capacity of oldcap and a desired |
| 220 | // capacity of wantcap, growcap returns a new capacity >= wantcap. |
| 221 | // |
| 222 | // The algorithm is mostly identical to the one used by append as of Go 1.14. |
| 223 | func growcap(oldcap, wantcap int) (newcap int) { |
| 224 | if wantcap > oldcap*2 { |
| 225 | newcap = wantcap |
| 226 | } else if oldcap < 1024 { |
| 227 | // The Go 1.14 runtime takes this case when len(s) < 1024, |
| 228 | // not when cap(s) < 1024. The difference doesn't seem |
| 229 | // significant here. |
| 230 | newcap = oldcap * 2 |
| 231 | } else { |
| 232 | newcap = oldcap |
| 233 | for 0 < newcap && newcap < wantcap { |
| 234 | newcap += newcap / 4 |
| 235 | } |
| 236 | if newcap <= 0 { |
| 237 | newcap = wantcap |
| 238 | } |
| 239 | } |
| 240 | return newcap |
| 241 | } |
| 242 | |
| 243 | func (o MarshalOptions) marshalMessageSlow(b []byte, m protoreflect.Message) ([]byte, error) { |
| 244 | if messageset.IsMessageSet(m.Descriptor()) { |
| 245 | return o.marshalMessageSet(b, m) |
| 246 | } |
| 247 | fieldOrder := order.AnyFieldOrder |
| 248 | if o.Deterministic { |
| 249 | // TODO: This should use a more natural ordering like NumberFieldOrder, |
| 250 | // but doing so breaks golden tests that make invalid assumption about |
| 251 | // output stability of this implementation. |
| 252 | fieldOrder = order.LegacyFieldOrder |
| 253 | } |
| 254 | var err error |
| 255 | order.RangeFields(m, fieldOrder, func(fd protoreflect.FieldDescriptor, v protoreflect.Value) bool { |
| 256 | b, err = o.marshalField(b, fd, v) |
| 257 | return err == nil |
| 258 | }) |
| 259 | if err != nil { |
| 260 | return b, err |
| 261 | } |
| 262 | b = append(b, m.GetUnknown()...) |
| 263 | return b, nil |
| 264 | } |
| 265 | |
| 266 | func (o MarshalOptions) marshalField(b []byte, fd protoreflect.FieldDescriptor, value protoreflect.Value) ([]byte, error) { |
| 267 | switch { |
| 268 | case fd.IsList(): |
| 269 | return o.marshalList(b, fd, value.List()) |
| 270 | case fd.IsMap(): |
| 271 | return o.marshalMap(b, fd, value.Map()) |
| 272 | default: |
| 273 | b = protowire.AppendTag(b, fd.Number(), wireTypes[fd.Kind()]) |
| 274 | return o.marshalSingular(b, fd, value) |
| 275 | } |
| 276 | } |
| 277 | |
| 278 | func (o MarshalOptions) marshalList(b []byte, fd protoreflect.FieldDescriptor, list protoreflect.List) ([]byte, error) { |
| 279 | if fd.IsPacked() && list.Len() > 0 { |
| 280 | b = protowire.AppendTag(b, fd.Number(), protowire.BytesType) |
| 281 | b, pos := appendSpeculativeLength(b) |
| 282 | for i, llen := 0, list.Len(); i < llen; i++ { |
| 283 | var err error |
| 284 | b, err = o.marshalSingular(b, fd, list.Get(i)) |
| 285 | if err != nil { |
| 286 | return b, err |
| 287 | } |
| 288 | } |
| 289 | b = finishSpeculativeLength(b, pos) |
| 290 | return b, nil |
| 291 | } |
| 292 | |
| 293 | kind := fd.Kind() |
| 294 | for i, llen := 0, list.Len(); i < llen; i++ { |
| 295 | var err error |
| 296 | b = protowire.AppendTag(b, fd.Number(), wireTypes[kind]) |
| 297 | b, err = o.marshalSingular(b, fd, list.Get(i)) |
| 298 | if err != nil { |
| 299 | return b, err |
| 300 | } |
| 301 | } |
| 302 | return b, nil |
| 303 | } |
| 304 | |
| 305 | func (o MarshalOptions) marshalMap(b []byte, fd protoreflect.FieldDescriptor, mapv protoreflect.Map) ([]byte, error) { |
| 306 | keyf := fd.MapKey() |
| 307 | valf := fd.MapValue() |
| 308 | keyOrder := order.AnyKeyOrder |
| 309 | if o.Deterministic { |
| 310 | keyOrder = order.GenericKeyOrder |
| 311 | } |
| 312 | var err error |
| 313 | order.RangeEntries(mapv, keyOrder, func(key protoreflect.MapKey, value protoreflect.Value) bool { |
| 314 | b = protowire.AppendTag(b, fd.Number(), protowire.BytesType) |
| 315 | var pos int |
| 316 | b, pos = appendSpeculativeLength(b) |
| 317 | |
| 318 | b, err = o.marshalField(b, keyf, key.Value()) |
| 319 | if err != nil { |
| 320 | return false |
| 321 | } |
| 322 | b, err = o.marshalField(b, valf, value) |
| 323 | if err != nil { |
| 324 | return false |
| 325 | } |
| 326 | b = finishSpeculativeLength(b, pos) |
| 327 | return true |
| 328 | }) |
| 329 | return b, err |
| 330 | } |
| 331 | |
| 332 | // When encoding length-prefixed fields, we speculatively set aside some number of bytes |
| 333 | // for the length, encode the data, and then encode the length (shifting the data if necessary |
| 334 | // to make room). |
| 335 | const speculativeLength = 1 |
| 336 | |
| 337 | func appendSpeculativeLength(b []byte) ([]byte, int) { |
| 338 | pos := len(b) |
| 339 | b = append(b, "\x00\x00\x00\x00"[:speculativeLength]...) |
| 340 | return b, pos |
| 341 | } |
| 342 | |
| 343 | func finishSpeculativeLength(b []byte, pos int) []byte { |
| 344 | mlen := len(b) - pos - speculativeLength |
| 345 | msiz := protowire.SizeVarint(uint64(mlen)) |
| 346 | if msiz != speculativeLength { |
| 347 | for i := 0; i < msiz-speculativeLength; i++ { |
| 348 | b = append(b, 0) |
| 349 | } |
| 350 | copy(b[pos+msiz:], b[pos+speculativeLength:]) |
| 351 | b = b[:pos+msiz+mlen] |
| 352 | } |
| 353 | protowire.AppendVarint(b[:pos], uint64(mlen)) |
| 354 | return b |
| 355 | } |