307 lines
9.6 KiB
Go
Raw Normal View History

// Copyright 2018 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 impl
import (
"fmt"
"reflect"
"strconv"
"strings"
"sync"
internal/impl: add fast-path marshal implementation This is a port of the v1 table marshaler, with some substantial cleanup and refactoring. Benchstat results from the protobuf reference benchmark data comparing the v1 package with v2, with AllowPartial:true set for the new package. This is not an apples-to-apples comparison, since v1 doesn't have a way to disable required field checks. Required field checks in v2 package currently go through reflection, which performs terribly; my initial experimentation indicates that fast-path required field checks will not add a large amount of cost; these results are incomplete but not wholly inaccurate. name old time/op new time/op delta /dataset.google_message3_1.pb/Marshal-12 219ms ± 1% 232ms ± 1% +5.85% (p=0.004 n=6+5) /dataset.google_message2.pb/Marshal-12 261µs ± 3% 248µs ± 1% -5.14% (p=0.002 n=6+6) /dataset.google_message1_proto2.pb/Marshal-12 681ns ± 2% 637ns ± 3% -6.53% (p=0.002 n=6+6) /dataset.google_message1_proto3.pb/Marshal-12 1.10µs ± 8% 0.99µs ± 3% -9.63% (p=0.002 n=6+6) /dataset.google_message3_3.pb/Marshal-12 44.2ms ± 3% 35.2ms ± 1% -20.28% (p=0.004 n=6+5) /dataset.google_message4.pb/Marshal-12 91.4ms ± 2% 94.9ms ± 2% +3.78% (p=0.002 n=6+6) /dataset.google_message3_2.pb/Marshal-12 78.7ms ± 6% 80.8ms ± 4% ~ (p=0.310 n=6+6) /dataset.google_message3_4.pb/Marshal-12 10.6ms ± 3% 10.6ms ± 8% ~ (p=0.662 n=5+6) /dataset.google_message3_5.pb/Marshal-12 675ms ± 4% 510ms ± 2% -24.40% (p=0.002 n=6+6) /dataset.google_message3_1.pb/Marshal 219ms ± 1% 236ms ± 7% +8.06% (p=0.004 n=5+6) /dataset.google_message2.pb/Marshal 257µs ± 1% 250µs ± 3% ~ (p=0.052 n=5+6) /dataset.google_message1_proto2.pb/Marshal 685ns ± 1% 628ns ± 1% -8.41% (p=0.008 n=5+5) /dataset.google_message1_proto3.pb/Marshal 1.08µs ± 1% 0.98µs ± 2% -9.31% (p=0.004 n=5+6) /dataset.google_message3_3.pb/Marshal 43.7ms ± 1% 35.1ms ± 1% -19.76% (p=0.002 n=6+6) /dataset.google_message4.pb/Marshal 93.4ms ± 4% 94.9ms ± 2% ~ (p=0.180 n=6+6) /dataset.google_message3_2.pb/Marshal 105ms ± 2% 98ms ± 7% -6.81% (p=0.009 n=5+6) /dataset.google_message3_4.pb/Marshal 16.3ms ± 6% 15.7ms ± 3% -3.44% (p=0.041 n=6+6) /dataset.google_message3_5.pb/Marshal 676ms ± 4% 504ms ± 2% -25.50% (p=0.004 n=6+5) Change-Id: I72cc4597117f4cf5d236ef505777d49dd4a5f75d Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171020 Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-01 13:49:56 -07:00
"sync/atomic"
pref "google.golang.org/protobuf/reflect/protoreflect"
piface "google.golang.org/protobuf/runtime/protoiface"
)
// MessageInfo provides protobuf related functionality for a given Go type
// that represents a message. A given instance of MessageInfo is tied to
// exactly one Go type, which must be a pointer to a struct type.
type MessageInfo struct {
internal/fileinit: generate reflect data structures from raw descriptors This CL takes a significantly different approach to generating support for protobuf reflection. The previous approach involved generating a large number of Go literals to represent the reflection information. While that approach was correct, it resulted in too much binary bloat. The approach taken here initializes the reflection information from the raw descriptor proto, which is a relatively dense representation of the protobuf reflection information. In order to keep initialization cost low, several measures were taken: * At program init, the bare minimum is parsed in order to initialize naming information for enums, messages, extensions, and services declared in the file. This is done because those top-level declarations are often relevant for registration. * Only upon first are most of the other data structures for protobuf reflection actually initialized. * Instead of using proto.Unmarshal, a hand-written unmarshaler is used. This allows us to avoid a dependendency on the descriptor proto and also because the API for the descriptor proto is fundamentally non-performant since it requires an allocation for every primitive field. At a high-level, the new implementation lives in internal/fileinit. Several changes were made to other parts of the repository: * cmd/protoc-gen-go: * Stop compressing the raw descriptors. While compression does reduce the size of the descriptors by approximately 2x, it is a pre-mature optimization since the descriptors themselves are around 1% of the total binary bloat that is due to generated protobufs. * Seeding protobuf reflection from the raw descriptor significantly simplifies the generator implementation since it is no longer responsible for constructing a tree of Go literals to represent the same information. * We remove the generation of the shadow types and instead call protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation for every call to ProtoReflect since we need to allocate a tuple that wraps a pointer to the message value, and a pointer to message type. * internal/impl: * We add a MessageType.GoType field and make it required that it is set prior to first use. This is done so that we can avoid calling MessageType.init except for when it is actually needed. The allows code to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the init code will run, possibly triggering a recursive deadlock (where the init code depends on getting the Type of some dependency which may be declared within the same file). * internal/cmd/generate-types: * The code to generate reflect/prototype/protofile_list_gen.go was copied and altered to generated internal/fileinit.desc_list_gen.go. At a high-level this CL adds significant technical complexity. However, this is offset by several possible future changes: * The prototype package can be drastically simplified. We can probably reimplement internal/legacy to use internal/fileinit instead, allowing us to drop another dependency on the prototype package. As a result, we can probably delete most of the constructor types in that package. * With the prototype package significantly pruned, and the fact that generated code no longer depend on depends on that package, we can consider merging what's left of prototype into protodesc. Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715 Reviewed-on: https://go-review.googlesource.com/c/158539 Reviewed-by: Herbie Ong <herbie@google.com> Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
// GoType is the underlying message Go type and must be populated.
// Once set, this field must never be mutated.
internal/fileinit: generate reflect data structures from raw descriptors This CL takes a significantly different approach to generating support for protobuf reflection. The previous approach involved generating a large number of Go literals to represent the reflection information. While that approach was correct, it resulted in too much binary bloat. The approach taken here initializes the reflection information from the raw descriptor proto, which is a relatively dense representation of the protobuf reflection information. In order to keep initialization cost low, several measures were taken: * At program init, the bare minimum is parsed in order to initialize naming information for enums, messages, extensions, and services declared in the file. This is done because those top-level declarations are often relevant for registration. * Only upon first are most of the other data structures for protobuf reflection actually initialized. * Instead of using proto.Unmarshal, a hand-written unmarshaler is used. This allows us to avoid a dependendency on the descriptor proto and also because the API for the descriptor proto is fundamentally non-performant since it requires an allocation for every primitive field. At a high-level, the new implementation lives in internal/fileinit. Several changes were made to other parts of the repository: * cmd/protoc-gen-go: * Stop compressing the raw descriptors. While compression does reduce the size of the descriptors by approximately 2x, it is a pre-mature optimization since the descriptors themselves are around 1% of the total binary bloat that is due to generated protobufs. * Seeding protobuf reflection from the raw descriptor significantly simplifies the generator implementation since it is no longer responsible for constructing a tree of Go literals to represent the same information. * We remove the generation of the shadow types and instead call protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation for every call to ProtoReflect since we need to allocate a tuple that wraps a pointer to the message value, and a pointer to message type. * internal/impl: * We add a MessageType.GoType field and make it required that it is set prior to first use. This is done so that we can avoid calling MessageType.init except for when it is actually needed. The allows code to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the init code will run, possibly triggering a recursive deadlock (where the init code depends on getting the Type of some dependency which may be declared within the same file). * internal/cmd/generate-types: * The code to generate reflect/prototype/protofile_list_gen.go was copied and altered to generated internal/fileinit.desc_list_gen.go. At a high-level this CL adds significant technical complexity. However, this is offset by several possible future changes: * The prototype package can be drastically simplified. We can probably reimplement internal/legacy to use internal/fileinit instead, allowing us to drop another dependency on the prototype package. As a result, we can probably delete most of the constructor types in that package. * With the prototype package significantly pruned, and the fact that generated code no longer depend on depends on that package, we can consider merging what's left of prototype into protodesc. Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715 Reviewed-on: https://go-review.googlesource.com/c/158539 Reviewed-by: Herbie Ong <herbie@google.com> Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
GoType reflect.Type // pointer to struct
internal/fileinit: generate reflect data structures from raw descriptors This CL takes a significantly different approach to generating support for protobuf reflection. The previous approach involved generating a large number of Go literals to represent the reflection information. While that approach was correct, it resulted in too much binary bloat. The approach taken here initializes the reflection information from the raw descriptor proto, which is a relatively dense representation of the protobuf reflection information. In order to keep initialization cost low, several measures were taken: * At program init, the bare minimum is parsed in order to initialize naming information for enums, messages, extensions, and services declared in the file. This is done because those top-level declarations are often relevant for registration. * Only upon first are most of the other data structures for protobuf reflection actually initialized. * Instead of using proto.Unmarshal, a hand-written unmarshaler is used. This allows us to avoid a dependendency on the descriptor proto and also because the API for the descriptor proto is fundamentally non-performant since it requires an allocation for every primitive field. At a high-level, the new implementation lives in internal/fileinit. Several changes were made to other parts of the repository: * cmd/protoc-gen-go: * Stop compressing the raw descriptors. While compression does reduce the size of the descriptors by approximately 2x, it is a pre-mature optimization since the descriptors themselves are around 1% of the total binary bloat that is due to generated protobufs. * Seeding protobuf reflection from the raw descriptor significantly simplifies the generator implementation since it is no longer responsible for constructing a tree of Go literals to represent the same information. * We remove the generation of the shadow types and instead call protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation for every call to ProtoReflect since we need to allocate a tuple that wraps a pointer to the message value, and a pointer to message type. * internal/impl: * We add a MessageType.GoType field and make it required that it is set prior to first use. This is done so that we can avoid calling MessageType.init except for when it is actually needed. The allows code to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the init code will run, possibly triggering a recursive deadlock (where the init code depends on getting the Type of some dependency which may be declared within the same file). * internal/cmd/generate-types: * The code to generate reflect/prototype/protofile_list_gen.go was copied and altered to generated internal/fileinit.desc_list_gen.go. At a high-level this CL adds significant technical complexity. However, this is offset by several possible future changes: * The prototype package can be drastically simplified. We can probably reimplement internal/legacy to use internal/fileinit instead, allowing us to drop another dependency on the prototype package. As a result, we can probably delete most of the constructor types in that package. * With the prototype package significantly pruned, and the fact that generated code no longer depend on depends on that package, we can consider merging what's left of prototype into protodesc. Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715 Reviewed-on: https://go-review.googlesource.com/c/158539 Reviewed-by: Herbie Ong <herbie@google.com> Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
// PBType is the underlying message descriptor type and must be populated.
// Once set, this field must never be mutated.
PBType pref.MessageType
// Exporter must be provided in a purego environment in order to provide
// access to unexported fields.
Exporter exporter
// OneofWrappers is list of pointers to oneof wrapper struct types.
OneofWrappers []interface{}
internal/impl: add fast-path marshal implementation This is a port of the v1 table marshaler, with some substantial cleanup and refactoring. Benchstat results from the protobuf reference benchmark data comparing the v1 package with v2, with AllowPartial:true set for the new package. This is not an apples-to-apples comparison, since v1 doesn't have a way to disable required field checks. Required field checks in v2 package currently go through reflection, which performs terribly; my initial experimentation indicates that fast-path required field checks will not add a large amount of cost; these results are incomplete but not wholly inaccurate. name old time/op new time/op delta /dataset.google_message3_1.pb/Marshal-12 219ms ± 1% 232ms ± 1% +5.85% (p=0.004 n=6+5) /dataset.google_message2.pb/Marshal-12 261µs ± 3% 248µs ± 1% -5.14% (p=0.002 n=6+6) /dataset.google_message1_proto2.pb/Marshal-12 681ns ± 2% 637ns ± 3% -6.53% (p=0.002 n=6+6) /dataset.google_message1_proto3.pb/Marshal-12 1.10µs ± 8% 0.99µs ± 3% -9.63% (p=0.002 n=6+6) /dataset.google_message3_3.pb/Marshal-12 44.2ms ± 3% 35.2ms ± 1% -20.28% (p=0.004 n=6+5) /dataset.google_message4.pb/Marshal-12 91.4ms ± 2% 94.9ms ± 2% +3.78% (p=0.002 n=6+6) /dataset.google_message3_2.pb/Marshal-12 78.7ms ± 6% 80.8ms ± 4% ~ (p=0.310 n=6+6) /dataset.google_message3_4.pb/Marshal-12 10.6ms ± 3% 10.6ms ± 8% ~ (p=0.662 n=5+6) /dataset.google_message3_5.pb/Marshal-12 675ms ± 4% 510ms ± 2% -24.40% (p=0.002 n=6+6) /dataset.google_message3_1.pb/Marshal 219ms ± 1% 236ms ± 7% +8.06% (p=0.004 n=5+6) /dataset.google_message2.pb/Marshal 257µs ± 1% 250µs ± 3% ~ (p=0.052 n=5+6) /dataset.google_message1_proto2.pb/Marshal 685ns ± 1% 628ns ± 1% -8.41% (p=0.008 n=5+5) /dataset.google_message1_proto3.pb/Marshal 1.08µs ± 1% 0.98µs ± 2% -9.31% (p=0.004 n=5+6) /dataset.google_message3_3.pb/Marshal 43.7ms ± 1% 35.1ms ± 1% -19.76% (p=0.002 n=6+6) /dataset.google_message4.pb/Marshal 93.4ms ± 4% 94.9ms ± 2% ~ (p=0.180 n=6+6) /dataset.google_message3_2.pb/Marshal 105ms ± 2% 98ms ± 7% -6.81% (p=0.009 n=5+6) /dataset.google_message3_4.pb/Marshal 16.3ms ± 6% 15.7ms ± 3% -3.44% (p=0.041 n=6+6) /dataset.google_message3_5.pb/Marshal 676ms ± 4% 504ms ± 2% -25.50% (p=0.004 n=6+5) Change-Id: I72cc4597117f4cf5d236ef505777d49dd4a5f75d Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171020 Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-01 13:49:56 -07:00
initMu sync.Mutex // protects all unexported fields
initDone uint32
reflectMessageInfo
// Information used by the fast-path methods.
methods piface.Methods
coderMessageInfo
internal/impl: add fast-path marshal implementation This is a port of the v1 table marshaler, with some substantial cleanup and refactoring. Benchstat results from the protobuf reference benchmark data comparing the v1 package with v2, with AllowPartial:true set for the new package. This is not an apples-to-apples comparison, since v1 doesn't have a way to disable required field checks. Required field checks in v2 package currently go through reflection, which performs terribly; my initial experimentation indicates that fast-path required field checks will not add a large amount of cost; these results are incomplete but not wholly inaccurate. name old time/op new time/op delta /dataset.google_message3_1.pb/Marshal-12 219ms ± 1% 232ms ± 1% +5.85% (p=0.004 n=6+5) /dataset.google_message2.pb/Marshal-12 261µs ± 3% 248µs ± 1% -5.14% (p=0.002 n=6+6) /dataset.google_message1_proto2.pb/Marshal-12 681ns ± 2% 637ns ± 3% -6.53% (p=0.002 n=6+6) /dataset.google_message1_proto3.pb/Marshal-12 1.10µs ± 8% 0.99µs ± 3% -9.63% (p=0.002 n=6+6) /dataset.google_message3_3.pb/Marshal-12 44.2ms ± 3% 35.2ms ± 1% -20.28% (p=0.004 n=6+5) /dataset.google_message4.pb/Marshal-12 91.4ms ± 2% 94.9ms ± 2% +3.78% (p=0.002 n=6+6) /dataset.google_message3_2.pb/Marshal-12 78.7ms ± 6% 80.8ms ± 4% ~ (p=0.310 n=6+6) /dataset.google_message3_4.pb/Marshal-12 10.6ms ± 3% 10.6ms ± 8% ~ (p=0.662 n=5+6) /dataset.google_message3_5.pb/Marshal-12 675ms ± 4% 510ms ± 2% -24.40% (p=0.002 n=6+6) /dataset.google_message3_1.pb/Marshal 219ms ± 1% 236ms ± 7% +8.06% (p=0.004 n=5+6) /dataset.google_message2.pb/Marshal 257µs ± 1% 250µs ± 3% ~ (p=0.052 n=5+6) /dataset.google_message1_proto2.pb/Marshal 685ns ± 1% 628ns ± 1% -8.41% (p=0.008 n=5+5) /dataset.google_message1_proto3.pb/Marshal 1.08µs ± 1% 0.98µs ± 2% -9.31% (p=0.004 n=5+6) /dataset.google_message3_3.pb/Marshal 43.7ms ± 1% 35.1ms ± 1% -19.76% (p=0.002 n=6+6) /dataset.google_message4.pb/Marshal 93.4ms ± 4% 94.9ms ± 2% ~ (p=0.180 n=6+6) /dataset.google_message3_2.pb/Marshal 105ms ± 2% 98ms ± 7% -6.81% (p=0.009 n=5+6) /dataset.google_message3_4.pb/Marshal 16.3ms ± 6% 15.7ms ± 3% -3.44% (p=0.041 n=6+6) /dataset.google_message3_5.pb/Marshal 676ms ± 4% 504ms ± 2% -25.50% (p=0.004 n=6+5) Change-Id: I72cc4597117f4cf5d236ef505777d49dd4a5f75d Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171020 Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-01 13:49:56 -07:00
extensionFieldInfosMu sync.RWMutex
extensionFieldInfos map[pref.ExtensionType]*extensionFieldInfo
internal/impl: add fast-path marshal implementation This is a port of the v1 table marshaler, with some substantial cleanup and refactoring. Benchstat results from the protobuf reference benchmark data comparing the v1 package with v2, with AllowPartial:true set for the new package. This is not an apples-to-apples comparison, since v1 doesn't have a way to disable required field checks. Required field checks in v2 package currently go through reflection, which performs terribly; my initial experimentation indicates that fast-path required field checks will not add a large amount of cost; these results are incomplete but not wholly inaccurate. name old time/op new time/op delta /dataset.google_message3_1.pb/Marshal-12 219ms ± 1% 232ms ± 1% +5.85% (p=0.004 n=6+5) /dataset.google_message2.pb/Marshal-12 261µs ± 3% 248µs ± 1% -5.14% (p=0.002 n=6+6) /dataset.google_message1_proto2.pb/Marshal-12 681ns ± 2% 637ns ± 3% -6.53% (p=0.002 n=6+6) /dataset.google_message1_proto3.pb/Marshal-12 1.10µs ± 8% 0.99µs ± 3% -9.63% (p=0.002 n=6+6) /dataset.google_message3_3.pb/Marshal-12 44.2ms ± 3% 35.2ms ± 1% -20.28% (p=0.004 n=6+5) /dataset.google_message4.pb/Marshal-12 91.4ms ± 2% 94.9ms ± 2% +3.78% (p=0.002 n=6+6) /dataset.google_message3_2.pb/Marshal-12 78.7ms ± 6% 80.8ms ± 4% ~ (p=0.310 n=6+6) /dataset.google_message3_4.pb/Marshal-12 10.6ms ± 3% 10.6ms ± 8% ~ (p=0.662 n=5+6) /dataset.google_message3_5.pb/Marshal-12 675ms ± 4% 510ms ± 2% -24.40% (p=0.002 n=6+6) /dataset.google_message3_1.pb/Marshal 219ms ± 1% 236ms ± 7% +8.06% (p=0.004 n=5+6) /dataset.google_message2.pb/Marshal 257µs ± 1% 250µs ± 3% ~ (p=0.052 n=5+6) /dataset.google_message1_proto2.pb/Marshal 685ns ± 1% 628ns ± 1% -8.41% (p=0.008 n=5+5) /dataset.google_message1_proto3.pb/Marshal 1.08µs ± 1% 0.98µs ± 2% -9.31% (p=0.004 n=5+6) /dataset.google_message3_3.pb/Marshal 43.7ms ± 1% 35.1ms ± 1% -19.76% (p=0.002 n=6+6) /dataset.google_message4.pb/Marshal 93.4ms ± 4% 94.9ms ± 2% ~ (p=0.180 n=6+6) /dataset.google_message3_2.pb/Marshal 105ms ± 2% 98ms ± 7% -6.81% (p=0.009 n=5+6) /dataset.google_message3_4.pb/Marshal 16.3ms ± 6% 15.7ms ± 3% -3.44% (p=0.041 n=6+6) /dataset.google_message3_5.pb/Marshal 676ms ± 4% 504ms ± 2% -25.50% (p=0.004 n=6+5) Change-Id: I72cc4597117f4cf5d236ef505777d49dd4a5f75d Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171020 Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-01 13:49:56 -07:00
}
type reflectMessageInfo struct {
fields map[pref.FieldNumber]*fieldInfo
oneofs map[pref.Name]*oneofInfo
getUnknown func(pointer) pref.RawFields
setUnknown func(pointer, pref.RawFields)
extensionMap func(pointer) *extensionMap
nilMessage atomicNilMessage
}
// exporter is a function that returns a reference to the ith field of v,
// where v is a pointer to a struct. It returns nil if it does not support
// exporting the requested field (e.g., already exported).
type exporter func(v interface{}, i int) interface{}
internal/impl: add fast-path marshal implementation This is a port of the v1 table marshaler, with some substantial cleanup and refactoring. Benchstat results from the protobuf reference benchmark data comparing the v1 package with v2, with AllowPartial:true set for the new package. This is not an apples-to-apples comparison, since v1 doesn't have a way to disable required field checks. Required field checks in v2 package currently go through reflection, which performs terribly; my initial experimentation indicates that fast-path required field checks will not add a large amount of cost; these results are incomplete but not wholly inaccurate. name old time/op new time/op delta /dataset.google_message3_1.pb/Marshal-12 219ms ± 1% 232ms ± 1% +5.85% (p=0.004 n=6+5) /dataset.google_message2.pb/Marshal-12 261µs ± 3% 248µs ± 1% -5.14% (p=0.002 n=6+6) /dataset.google_message1_proto2.pb/Marshal-12 681ns ± 2% 637ns ± 3% -6.53% (p=0.002 n=6+6) /dataset.google_message1_proto3.pb/Marshal-12 1.10µs ± 8% 0.99µs ± 3% -9.63% (p=0.002 n=6+6) /dataset.google_message3_3.pb/Marshal-12 44.2ms ± 3% 35.2ms ± 1% -20.28% (p=0.004 n=6+5) /dataset.google_message4.pb/Marshal-12 91.4ms ± 2% 94.9ms ± 2% +3.78% (p=0.002 n=6+6) /dataset.google_message3_2.pb/Marshal-12 78.7ms ± 6% 80.8ms ± 4% ~ (p=0.310 n=6+6) /dataset.google_message3_4.pb/Marshal-12 10.6ms ± 3% 10.6ms ± 8% ~ (p=0.662 n=5+6) /dataset.google_message3_5.pb/Marshal-12 675ms ± 4% 510ms ± 2% -24.40% (p=0.002 n=6+6) /dataset.google_message3_1.pb/Marshal 219ms ± 1% 236ms ± 7% +8.06% (p=0.004 n=5+6) /dataset.google_message2.pb/Marshal 257µs ± 1% 250µs ± 3% ~ (p=0.052 n=5+6) /dataset.google_message1_proto2.pb/Marshal 685ns ± 1% 628ns ± 1% -8.41% (p=0.008 n=5+5) /dataset.google_message1_proto3.pb/Marshal 1.08µs ± 1% 0.98µs ± 2% -9.31% (p=0.004 n=5+6) /dataset.google_message3_3.pb/Marshal 43.7ms ± 1% 35.1ms ± 1% -19.76% (p=0.002 n=6+6) /dataset.google_message4.pb/Marshal 93.4ms ± 4% 94.9ms ± 2% ~ (p=0.180 n=6+6) /dataset.google_message3_2.pb/Marshal 105ms ± 2% 98ms ± 7% -6.81% (p=0.009 n=5+6) /dataset.google_message3_4.pb/Marshal 16.3ms ± 6% 15.7ms ± 3% -3.44% (p=0.041 n=6+6) /dataset.google_message3_5.pb/Marshal 676ms ± 4% 504ms ± 2% -25.50% (p=0.004 n=6+5) Change-Id: I72cc4597117f4cf5d236ef505777d49dd4a5f75d Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171020 Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-01 13:49:56 -07:00
var prefMessageType = reflect.TypeOf((*pref.Message)(nil)).Elem()
// getMessageInfo returns the MessageInfo (if any) for a type.
internal/impl: add fast-path marshal implementation This is a port of the v1 table marshaler, with some substantial cleanup and refactoring. Benchstat results from the protobuf reference benchmark data comparing the v1 package with v2, with AllowPartial:true set for the new package. This is not an apples-to-apples comparison, since v1 doesn't have a way to disable required field checks. Required field checks in v2 package currently go through reflection, which performs terribly; my initial experimentation indicates that fast-path required field checks will not add a large amount of cost; these results are incomplete but not wholly inaccurate. name old time/op new time/op delta /dataset.google_message3_1.pb/Marshal-12 219ms ± 1% 232ms ± 1% +5.85% (p=0.004 n=6+5) /dataset.google_message2.pb/Marshal-12 261µs ± 3% 248µs ± 1% -5.14% (p=0.002 n=6+6) /dataset.google_message1_proto2.pb/Marshal-12 681ns ± 2% 637ns ± 3% -6.53% (p=0.002 n=6+6) /dataset.google_message1_proto3.pb/Marshal-12 1.10µs ± 8% 0.99µs ± 3% -9.63% (p=0.002 n=6+6) /dataset.google_message3_3.pb/Marshal-12 44.2ms ± 3% 35.2ms ± 1% -20.28% (p=0.004 n=6+5) /dataset.google_message4.pb/Marshal-12 91.4ms ± 2% 94.9ms ± 2% +3.78% (p=0.002 n=6+6) /dataset.google_message3_2.pb/Marshal-12 78.7ms ± 6% 80.8ms ± 4% ~ (p=0.310 n=6+6) /dataset.google_message3_4.pb/Marshal-12 10.6ms ± 3% 10.6ms ± 8% ~ (p=0.662 n=5+6) /dataset.google_message3_5.pb/Marshal-12 675ms ± 4% 510ms ± 2% -24.40% (p=0.002 n=6+6) /dataset.google_message3_1.pb/Marshal 219ms ± 1% 236ms ± 7% +8.06% (p=0.004 n=5+6) /dataset.google_message2.pb/Marshal 257µs ± 1% 250µs ± 3% ~ (p=0.052 n=5+6) /dataset.google_message1_proto2.pb/Marshal 685ns ± 1% 628ns ± 1% -8.41% (p=0.008 n=5+5) /dataset.google_message1_proto3.pb/Marshal 1.08µs ± 1% 0.98µs ± 2% -9.31% (p=0.004 n=5+6) /dataset.google_message3_3.pb/Marshal 43.7ms ± 1% 35.1ms ± 1% -19.76% (p=0.002 n=6+6) /dataset.google_message4.pb/Marshal 93.4ms ± 4% 94.9ms ± 2% ~ (p=0.180 n=6+6) /dataset.google_message3_2.pb/Marshal 105ms ± 2% 98ms ± 7% -6.81% (p=0.009 n=5+6) /dataset.google_message3_4.pb/Marshal 16.3ms ± 6% 15.7ms ± 3% -3.44% (p=0.041 n=6+6) /dataset.google_message3_5.pb/Marshal 676ms ± 4% 504ms ± 2% -25.50% (p=0.004 n=6+5) Change-Id: I72cc4597117f4cf5d236ef505777d49dd4a5f75d Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171020 Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-01 13:49:56 -07:00
//
// We find the MessageInfo by calling the ProtoReflect method on the type's
internal/impl: add fast-path marshal implementation This is a port of the v1 table marshaler, with some substantial cleanup and refactoring. Benchstat results from the protobuf reference benchmark data comparing the v1 package with v2, with AllowPartial:true set for the new package. This is not an apples-to-apples comparison, since v1 doesn't have a way to disable required field checks. Required field checks in v2 package currently go through reflection, which performs terribly; my initial experimentation indicates that fast-path required field checks will not add a large amount of cost; these results are incomplete but not wholly inaccurate. name old time/op new time/op delta /dataset.google_message3_1.pb/Marshal-12 219ms ± 1% 232ms ± 1% +5.85% (p=0.004 n=6+5) /dataset.google_message2.pb/Marshal-12 261µs ± 3% 248µs ± 1% -5.14% (p=0.002 n=6+6) /dataset.google_message1_proto2.pb/Marshal-12 681ns ± 2% 637ns ± 3% -6.53% (p=0.002 n=6+6) /dataset.google_message1_proto3.pb/Marshal-12 1.10µs ± 8% 0.99µs ± 3% -9.63% (p=0.002 n=6+6) /dataset.google_message3_3.pb/Marshal-12 44.2ms ± 3% 35.2ms ± 1% -20.28% (p=0.004 n=6+5) /dataset.google_message4.pb/Marshal-12 91.4ms ± 2% 94.9ms ± 2% +3.78% (p=0.002 n=6+6) /dataset.google_message3_2.pb/Marshal-12 78.7ms ± 6% 80.8ms ± 4% ~ (p=0.310 n=6+6) /dataset.google_message3_4.pb/Marshal-12 10.6ms ± 3% 10.6ms ± 8% ~ (p=0.662 n=5+6) /dataset.google_message3_5.pb/Marshal-12 675ms ± 4% 510ms ± 2% -24.40% (p=0.002 n=6+6) /dataset.google_message3_1.pb/Marshal 219ms ± 1% 236ms ± 7% +8.06% (p=0.004 n=5+6) /dataset.google_message2.pb/Marshal 257µs ± 1% 250µs ± 3% ~ (p=0.052 n=5+6) /dataset.google_message1_proto2.pb/Marshal 685ns ± 1% 628ns ± 1% -8.41% (p=0.008 n=5+5) /dataset.google_message1_proto3.pb/Marshal 1.08µs ± 1% 0.98µs ± 2% -9.31% (p=0.004 n=5+6) /dataset.google_message3_3.pb/Marshal 43.7ms ± 1% 35.1ms ± 1% -19.76% (p=0.002 n=6+6) /dataset.google_message4.pb/Marshal 93.4ms ± 4% 94.9ms ± 2% ~ (p=0.180 n=6+6) /dataset.google_message3_2.pb/Marshal 105ms ± 2% 98ms ± 7% -6.81% (p=0.009 n=5+6) /dataset.google_message3_4.pb/Marshal 16.3ms ± 6% 15.7ms ± 3% -3.44% (p=0.041 n=6+6) /dataset.google_message3_5.pb/Marshal 676ms ± 4% 504ms ± 2% -25.50% (p=0.004 n=6+5) Change-Id: I72cc4597117f4cf5d236ef505777d49dd4a5f75d Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171020 Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-01 13:49:56 -07:00
// zero value and looking at the returned type to see if it is a
// messageReflectWrapper. Note that the MessageInfo may still be uninitialized
internal/impl: add fast-path marshal implementation This is a port of the v1 table marshaler, with some substantial cleanup and refactoring. Benchstat results from the protobuf reference benchmark data comparing the v1 package with v2, with AllowPartial:true set for the new package. This is not an apples-to-apples comparison, since v1 doesn't have a way to disable required field checks. Required field checks in v2 package currently go through reflection, which performs terribly; my initial experimentation indicates that fast-path required field checks will not add a large amount of cost; these results are incomplete but not wholly inaccurate. name old time/op new time/op delta /dataset.google_message3_1.pb/Marshal-12 219ms ± 1% 232ms ± 1% +5.85% (p=0.004 n=6+5) /dataset.google_message2.pb/Marshal-12 261µs ± 3% 248µs ± 1% -5.14% (p=0.002 n=6+6) /dataset.google_message1_proto2.pb/Marshal-12 681ns ± 2% 637ns ± 3% -6.53% (p=0.002 n=6+6) /dataset.google_message1_proto3.pb/Marshal-12 1.10µs ± 8% 0.99µs ± 3% -9.63% (p=0.002 n=6+6) /dataset.google_message3_3.pb/Marshal-12 44.2ms ± 3% 35.2ms ± 1% -20.28% (p=0.004 n=6+5) /dataset.google_message4.pb/Marshal-12 91.4ms ± 2% 94.9ms ± 2% +3.78% (p=0.002 n=6+6) /dataset.google_message3_2.pb/Marshal-12 78.7ms ± 6% 80.8ms ± 4% ~ (p=0.310 n=6+6) /dataset.google_message3_4.pb/Marshal-12 10.6ms ± 3% 10.6ms ± 8% ~ (p=0.662 n=5+6) /dataset.google_message3_5.pb/Marshal-12 675ms ± 4% 510ms ± 2% -24.40% (p=0.002 n=6+6) /dataset.google_message3_1.pb/Marshal 219ms ± 1% 236ms ± 7% +8.06% (p=0.004 n=5+6) /dataset.google_message2.pb/Marshal 257µs ± 1% 250µs ± 3% ~ (p=0.052 n=5+6) /dataset.google_message1_proto2.pb/Marshal 685ns ± 1% 628ns ± 1% -8.41% (p=0.008 n=5+5) /dataset.google_message1_proto3.pb/Marshal 1.08µs ± 1% 0.98µs ± 2% -9.31% (p=0.004 n=5+6) /dataset.google_message3_3.pb/Marshal 43.7ms ± 1% 35.1ms ± 1% -19.76% (p=0.002 n=6+6) /dataset.google_message4.pb/Marshal 93.4ms ± 4% 94.9ms ± 2% ~ (p=0.180 n=6+6) /dataset.google_message3_2.pb/Marshal 105ms ± 2% 98ms ± 7% -6.81% (p=0.009 n=5+6) /dataset.google_message3_4.pb/Marshal 16.3ms ± 6% 15.7ms ± 3% -3.44% (p=0.041 n=6+6) /dataset.google_message3_5.pb/Marshal 676ms ± 4% 504ms ± 2% -25.50% (p=0.004 n=6+5) Change-Id: I72cc4597117f4cf5d236ef505777d49dd4a5f75d Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171020 Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-01 13:49:56 -07:00
// at this point.
func getMessageInfo(mt reflect.Type) (mi *MessageInfo, ok bool) {
internal/impl: add fast-path marshal implementation This is a port of the v1 table marshaler, with some substantial cleanup and refactoring. Benchstat results from the protobuf reference benchmark data comparing the v1 package with v2, with AllowPartial:true set for the new package. This is not an apples-to-apples comparison, since v1 doesn't have a way to disable required field checks. Required field checks in v2 package currently go through reflection, which performs terribly; my initial experimentation indicates that fast-path required field checks will not add a large amount of cost; these results are incomplete but not wholly inaccurate. name old time/op new time/op delta /dataset.google_message3_1.pb/Marshal-12 219ms ± 1% 232ms ± 1% +5.85% (p=0.004 n=6+5) /dataset.google_message2.pb/Marshal-12 261µs ± 3% 248µs ± 1% -5.14% (p=0.002 n=6+6) /dataset.google_message1_proto2.pb/Marshal-12 681ns ± 2% 637ns ± 3% -6.53% (p=0.002 n=6+6) /dataset.google_message1_proto3.pb/Marshal-12 1.10µs ± 8% 0.99µs ± 3% -9.63% (p=0.002 n=6+6) /dataset.google_message3_3.pb/Marshal-12 44.2ms ± 3% 35.2ms ± 1% -20.28% (p=0.004 n=6+5) /dataset.google_message4.pb/Marshal-12 91.4ms ± 2% 94.9ms ± 2% +3.78% (p=0.002 n=6+6) /dataset.google_message3_2.pb/Marshal-12 78.7ms ± 6% 80.8ms ± 4% ~ (p=0.310 n=6+6) /dataset.google_message3_4.pb/Marshal-12 10.6ms ± 3% 10.6ms ± 8% ~ (p=0.662 n=5+6) /dataset.google_message3_5.pb/Marshal-12 675ms ± 4% 510ms ± 2% -24.40% (p=0.002 n=6+6) /dataset.google_message3_1.pb/Marshal 219ms ± 1% 236ms ± 7% +8.06% (p=0.004 n=5+6) /dataset.google_message2.pb/Marshal 257µs ± 1% 250µs ± 3% ~ (p=0.052 n=5+6) /dataset.google_message1_proto2.pb/Marshal 685ns ± 1% 628ns ± 1% -8.41% (p=0.008 n=5+5) /dataset.google_message1_proto3.pb/Marshal 1.08µs ± 1% 0.98µs ± 2% -9.31% (p=0.004 n=5+6) /dataset.google_message3_3.pb/Marshal 43.7ms ± 1% 35.1ms ± 1% -19.76% (p=0.002 n=6+6) /dataset.google_message4.pb/Marshal 93.4ms ± 4% 94.9ms ± 2% ~ (p=0.180 n=6+6) /dataset.google_message3_2.pb/Marshal 105ms ± 2% 98ms ± 7% -6.81% (p=0.009 n=5+6) /dataset.google_message3_4.pb/Marshal 16.3ms ± 6% 15.7ms ± 3% -3.44% (p=0.041 n=6+6) /dataset.google_message3_5.pb/Marshal 676ms ± 4% 504ms ± 2% -25.50% (p=0.004 n=6+5) Change-Id: I72cc4597117f4cf5d236ef505777d49dd4a5f75d Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171020 Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-01 13:49:56 -07:00
method, ok := mt.MethodByName("ProtoReflect")
if !ok {
return nil, false
}
if method.Type.NumIn() != 1 || method.Type.NumOut() != 1 || method.Type.Out(0) != prefMessageType {
return nil, false
}
ret := reflect.Zero(mt).Method(method.Index).Call(nil)
m, ok := ret[0].Elem().Interface().(*messageReflectWrapper)
if !ok {
return nil, ok
}
return m.mi, true
}
func (mi *MessageInfo) init() {
internal/impl: add fast-path marshal implementation This is a port of the v1 table marshaler, with some substantial cleanup and refactoring. Benchstat results from the protobuf reference benchmark data comparing the v1 package with v2, with AllowPartial:true set for the new package. This is not an apples-to-apples comparison, since v1 doesn't have a way to disable required field checks. Required field checks in v2 package currently go through reflection, which performs terribly; my initial experimentation indicates that fast-path required field checks will not add a large amount of cost; these results are incomplete but not wholly inaccurate. name old time/op new time/op delta /dataset.google_message3_1.pb/Marshal-12 219ms ± 1% 232ms ± 1% +5.85% (p=0.004 n=6+5) /dataset.google_message2.pb/Marshal-12 261µs ± 3% 248µs ± 1% -5.14% (p=0.002 n=6+6) /dataset.google_message1_proto2.pb/Marshal-12 681ns ± 2% 637ns ± 3% -6.53% (p=0.002 n=6+6) /dataset.google_message1_proto3.pb/Marshal-12 1.10µs ± 8% 0.99µs ± 3% -9.63% (p=0.002 n=6+6) /dataset.google_message3_3.pb/Marshal-12 44.2ms ± 3% 35.2ms ± 1% -20.28% (p=0.004 n=6+5) /dataset.google_message4.pb/Marshal-12 91.4ms ± 2% 94.9ms ± 2% +3.78% (p=0.002 n=6+6) /dataset.google_message3_2.pb/Marshal-12 78.7ms ± 6% 80.8ms ± 4% ~ (p=0.310 n=6+6) /dataset.google_message3_4.pb/Marshal-12 10.6ms ± 3% 10.6ms ± 8% ~ (p=0.662 n=5+6) /dataset.google_message3_5.pb/Marshal-12 675ms ± 4% 510ms ± 2% -24.40% (p=0.002 n=6+6) /dataset.google_message3_1.pb/Marshal 219ms ± 1% 236ms ± 7% +8.06% (p=0.004 n=5+6) /dataset.google_message2.pb/Marshal 257µs ± 1% 250µs ± 3% ~ (p=0.052 n=5+6) /dataset.google_message1_proto2.pb/Marshal 685ns ± 1% 628ns ± 1% -8.41% (p=0.008 n=5+5) /dataset.google_message1_proto3.pb/Marshal 1.08µs ± 1% 0.98µs ± 2% -9.31% (p=0.004 n=5+6) /dataset.google_message3_3.pb/Marshal 43.7ms ± 1% 35.1ms ± 1% -19.76% (p=0.002 n=6+6) /dataset.google_message4.pb/Marshal 93.4ms ± 4% 94.9ms ± 2% ~ (p=0.180 n=6+6) /dataset.google_message3_2.pb/Marshal 105ms ± 2% 98ms ± 7% -6.81% (p=0.009 n=5+6) /dataset.google_message3_4.pb/Marshal 16.3ms ± 6% 15.7ms ± 3% -3.44% (p=0.041 n=6+6) /dataset.google_message3_5.pb/Marshal 676ms ± 4% 504ms ± 2% -25.50% (p=0.004 n=6+5) Change-Id: I72cc4597117f4cf5d236ef505777d49dd4a5f75d Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171020 Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-01 13:49:56 -07:00
// This function is called in the hot path. Inline the sync.Once
// logic, since allocating a closure for Once.Do is expensive.
// Keep init small to ensure that it can be inlined.
if atomic.LoadUint32(&mi.initDone) == 0 {
mi.initOnce()
internal/impl: add fast-path marshal implementation This is a port of the v1 table marshaler, with some substantial cleanup and refactoring. Benchstat results from the protobuf reference benchmark data comparing the v1 package with v2, with AllowPartial:true set for the new package. This is not an apples-to-apples comparison, since v1 doesn't have a way to disable required field checks. Required field checks in v2 package currently go through reflection, which performs terribly; my initial experimentation indicates that fast-path required field checks will not add a large amount of cost; these results are incomplete but not wholly inaccurate. name old time/op new time/op delta /dataset.google_message3_1.pb/Marshal-12 219ms ± 1% 232ms ± 1% +5.85% (p=0.004 n=6+5) /dataset.google_message2.pb/Marshal-12 261µs ± 3% 248µs ± 1% -5.14% (p=0.002 n=6+6) /dataset.google_message1_proto2.pb/Marshal-12 681ns ± 2% 637ns ± 3% -6.53% (p=0.002 n=6+6) /dataset.google_message1_proto3.pb/Marshal-12 1.10µs ± 8% 0.99µs ± 3% -9.63% (p=0.002 n=6+6) /dataset.google_message3_3.pb/Marshal-12 44.2ms ± 3% 35.2ms ± 1% -20.28% (p=0.004 n=6+5) /dataset.google_message4.pb/Marshal-12 91.4ms ± 2% 94.9ms ± 2% +3.78% (p=0.002 n=6+6) /dataset.google_message3_2.pb/Marshal-12 78.7ms ± 6% 80.8ms ± 4% ~ (p=0.310 n=6+6) /dataset.google_message3_4.pb/Marshal-12 10.6ms ± 3% 10.6ms ± 8% ~ (p=0.662 n=5+6) /dataset.google_message3_5.pb/Marshal-12 675ms ± 4% 510ms ± 2% -24.40% (p=0.002 n=6+6) /dataset.google_message3_1.pb/Marshal 219ms ± 1% 236ms ± 7% +8.06% (p=0.004 n=5+6) /dataset.google_message2.pb/Marshal 257µs ± 1% 250µs ± 3% ~ (p=0.052 n=5+6) /dataset.google_message1_proto2.pb/Marshal 685ns ± 1% 628ns ± 1% -8.41% (p=0.008 n=5+5) /dataset.google_message1_proto3.pb/Marshal 1.08µs ± 1% 0.98µs ± 2% -9.31% (p=0.004 n=5+6) /dataset.google_message3_3.pb/Marshal 43.7ms ± 1% 35.1ms ± 1% -19.76% (p=0.002 n=6+6) /dataset.google_message4.pb/Marshal 93.4ms ± 4% 94.9ms ± 2% ~ (p=0.180 n=6+6) /dataset.google_message3_2.pb/Marshal 105ms ± 2% 98ms ± 7% -6.81% (p=0.009 n=5+6) /dataset.google_message3_4.pb/Marshal 16.3ms ± 6% 15.7ms ± 3% -3.44% (p=0.041 n=6+6) /dataset.google_message3_5.pb/Marshal 676ms ± 4% 504ms ± 2% -25.50% (p=0.004 n=6+5) Change-Id: I72cc4597117f4cf5d236ef505777d49dd4a5f75d Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171020 Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-01 13:49:56 -07:00
}
}
func (mi *MessageInfo) initOnce() {
internal/impl: add fast-path marshal implementation This is a port of the v1 table marshaler, with some substantial cleanup and refactoring. Benchstat results from the protobuf reference benchmark data comparing the v1 package with v2, with AllowPartial:true set for the new package. This is not an apples-to-apples comparison, since v1 doesn't have a way to disable required field checks. Required field checks in v2 package currently go through reflection, which performs terribly; my initial experimentation indicates that fast-path required field checks will not add a large amount of cost; these results are incomplete but not wholly inaccurate. name old time/op new time/op delta /dataset.google_message3_1.pb/Marshal-12 219ms ± 1% 232ms ± 1% +5.85% (p=0.004 n=6+5) /dataset.google_message2.pb/Marshal-12 261µs ± 3% 248µs ± 1% -5.14% (p=0.002 n=6+6) /dataset.google_message1_proto2.pb/Marshal-12 681ns ± 2% 637ns ± 3% -6.53% (p=0.002 n=6+6) /dataset.google_message1_proto3.pb/Marshal-12 1.10µs ± 8% 0.99µs ± 3% -9.63% (p=0.002 n=6+6) /dataset.google_message3_3.pb/Marshal-12 44.2ms ± 3% 35.2ms ± 1% -20.28% (p=0.004 n=6+5) /dataset.google_message4.pb/Marshal-12 91.4ms ± 2% 94.9ms ± 2% +3.78% (p=0.002 n=6+6) /dataset.google_message3_2.pb/Marshal-12 78.7ms ± 6% 80.8ms ± 4% ~ (p=0.310 n=6+6) /dataset.google_message3_4.pb/Marshal-12 10.6ms ± 3% 10.6ms ± 8% ~ (p=0.662 n=5+6) /dataset.google_message3_5.pb/Marshal-12 675ms ± 4% 510ms ± 2% -24.40% (p=0.002 n=6+6) /dataset.google_message3_1.pb/Marshal 219ms ± 1% 236ms ± 7% +8.06% (p=0.004 n=5+6) /dataset.google_message2.pb/Marshal 257µs ± 1% 250µs ± 3% ~ (p=0.052 n=5+6) /dataset.google_message1_proto2.pb/Marshal 685ns ± 1% 628ns ± 1% -8.41% (p=0.008 n=5+5) /dataset.google_message1_proto3.pb/Marshal 1.08µs ± 1% 0.98µs ± 2% -9.31% (p=0.004 n=5+6) /dataset.google_message3_3.pb/Marshal 43.7ms ± 1% 35.1ms ± 1% -19.76% (p=0.002 n=6+6) /dataset.google_message4.pb/Marshal 93.4ms ± 4% 94.9ms ± 2% ~ (p=0.180 n=6+6) /dataset.google_message3_2.pb/Marshal 105ms ± 2% 98ms ± 7% -6.81% (p=0.009 n=5+6) /dataset.google_message3_4.pb/Marshal 16.3ms ± 6% 15.7ms ± 3% -3.44% (p=0.041 n=6+6) /dataset.google_message3_5.pb/Marshal 676ms ± 4% 504ms ± 2% -25.50% (p=0.004 n=6+5) Change-Id: I72cc4597117f4cf5d236ef505777d49dd4a5f75d Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171020 Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-01 13:49:56 -07:00
mi.initMu.Lock()
defer mi.initMu.Unlock()
if mi.initDone == 1 {
return
}
t := mi.GoType
if t.Kind() != reflect.Ptr && t.Elem().Kind() != reflect.Struct {
panic(fmt.Sprintf("got %v, want *struct kind", t))
}
si := mi.makeStructInfo(t.Elem())
mi.makeKnownFieldsFunc(si)
mi.makeUnknownFieldsFunc(t.Elem(), si)
mi.makeExtensionFieldsFunc(t.Elem(), si)
mi.makeMethods(t.Elem(), si)
internal/impl: add fast-path marshal implementation This is a port of the v1 table marshaler, with some substantial cleanup and refactoring. Benchstat results from the protobuf reference benchmark data comparing the v1 package with v2, with AllowPartial:true set for the new package. This is not an apples-to-apples comparison, since v1 doesn't have a way to disable required field checks. Required field checks in v2 package currently go through reflection, which performs terribly; my initial experimentation indicates that fast-path required field checks will not add a large amount of cost; these results are incomplete but not wholly inaccurate. name old time/op new time/op delta /dataset.google_message3_1.pb/Marshal-12 219ms ± 1% 232ms ± 1% +5.85% (p=0.004 n=6+5) /dataset.google_message2.pb/Marshal-12 261µs ± 3% 248µs ± 1% -5.14% (p=0.002 n=6+6) /dataset.google_message1_proto2.pb/Marshal-12 681ns ± 2% 637ns ± 3% -6.53% (p=0.002 n=6+6) /dataset.google_message1_proto3.pb/Marshal-12 1.10µs ± 8% 0.99µs ± 3% -9.63% (p=0.002 n=6+6) /dataset.google_message3_3.pb/Marshal-12 44.2ms ± 3% 35.2ms ± 1% -20.28% (p=0.004 n=6+5) /dataset.google_message4.pb/Marshal-12 91.4ms ± 2% 94.9ms ± 2% +3.78% (p=0.002 n=6+6) /dataset.google_message3_2.pb/Marshal-12 78.7ms ± 6% 80.8ms ± 4% ~ (p=0.310 n=6+6) /dataset.google_message3_4.pb/Marshal-12 10.6ms ± 3% 10.6ms ± 8% ~ (p=0.662 n=5+6) /dataset.google_message3_5.pb/Marshal-12 675ms ± 4% 510ms ± 2% -24.40% (p=0.002 n=6+6) /dataset.google_message3_1.pb/Marshal 219ms ± 1% 236ms ± 7% +8.06% (p=0.004 n=5+6) /dataset.google_message2.pb/Marshal 257µs ± 1% 250µs ± 3% ~ (p=0.052 n=5+6) /dataset.google_message1_proto2.pb/Marshal 685ns ± 1% 628ns ± 1% -8.41% (p=0.008 n=5+5) /dataset.google_message1_proto3.pb/Marshal 1.08µs ± 1% 0.98µs ± 2% -9.31% (p=0.004 n=5+6) /dataset.google_message3_3.pb/Marshal 43.7ms ± 1% 35.1ms ± 1% -19.76% (p=0.002 n=6+6) /dataset.google_message4.pb/Marshal 93.4ms ± 4% 94.9ms ± 2% ~ (p=0.180 n=6+6) /dataset.google_message3_2.pb/Marshal 105ms ± 2% 98ms ± 7% -6.81% (p=0.009 n=5+6) /dataset.google_message3_4.pb/Marshal 16.3ms ± 6% 15.7ms ± 3% -3.44% (p=0.041 n=6+6) /dataset.google_message3_5.pb/Marshal 676ms ± 4% 504ms ± 2% -25.50% (p=0.004 n=6+5) Change-Id: I72cc4597117f4cf5d236ef505777d49dd4a5f75d Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171020 Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-01 13:49:56 -07:00
atomic.StoreUint32(&mi.initDone, 1)
}
type (
SizeCache = int32
WeakFields = map[int32]piface.MessageV1
UnknownFields = []byte
ExtensionFields = map[int32]ExtensionField
)
var (
sizecacheType = reflect.TypeOf(SizeCache(0))
weakFieldsType = reflect.TypeOf(WeakFields(nil))
unknownFieldsType = reflect.TypeOf(UnknownFields(nil))
extensionFieldsType = reflect.TypeOf(ExtensionFields(nil))
)
internal/impl: add fast-path marshal implementation This is a port of the v1 table marshaler, with some substantial cleanup and refactoring. Benchstat results from the protobuf reference benchmark data comparing the v1 package with v2, with AllowPartial:true set for the new package. This is not an apples-to-apples comparison, since v1 doesn't have a way to disable required field checks. Required field checks in v2 package currently go through reflection, which performs terribly; my initial experimentation indicates that fast-path required field checks will not add a large amount of cost; these results are incomplete but not wholly inaccurate. name old time/op new time/op delta /dataset.google_message3_1.pb/Marshal-12 219ms ± 1% 232ms ± 1% +5.85% (p=0.004 n=6+5) /dataset.google_message2.pb/Marshal-12 261µs ± 3% 248µs ± 1% -5.14% (p=0.002 n=6+6) /dataset.google_message1_proto2.pb/Marshal-12 681ns ± 2% 637ns ± 3% -6.53% (p=0.002 n=6+6) /dataset.google_message1_proto3.pb/Marshal-12 1.10µs ± 8% 0.99µs ± 3% -9.63% (p=0.002 n=6+6) /dataset.google_message3_3.pb/Marshal-12 44.2ms ± 3% 35.2ms ± 1% -20.28% (p=0.004 n=6+5) /dataset.google_message4.pb/Marshal-12 91.4ms ± 2% 94.9ms ± 2% +3.78% (p=0.002 n=6+6) /dataset.google_message3_2.pb/Marshal-12 78.7ms ± 6% 80.8ms ± 4% ~ (p=0.310 n=6+6) /dataset.google_message3_4.pb/Marshal-12 10.6ms ± 3% 10.6ms ± 8% ~ (p=0.662 n=5+6) /dataset.google_message3_5.pb/Marshal-12 675ms ± 4% 510ms ± 2% -24.40% (p=0.002 n=6+6) /dataset.google_message3_1.pb/Marshal 219ms ± 1% 236ms ± 7% +8.06% (p=0.004 n=5+6) /dataset.google_message2.pb/Marshal 257µs ± 1% 250µs ± 3% ~ (p=0.052 n=5+6) /dataset.google_message1_proto2.pb/Marshal 685ns ± 1% 628ns ± 1% -8.41% (p=0.008 n=5+5) /dataset.google_message1_proto3.pb/Marshal 1.08µs ± 1% 0.98µs ± 2% -9.31% (p=0.004 n=5+6) /dataset.google_message3_3.pb/Marshal 43.7ms ± 1% 35.1ms ± 1% -19.76% (p=0.002 n=6+6) /dataset.google_message4.pb/Marshal 93.4ms ± 4% 94.9ms ± 2% ~ (p=0.180 n=6+6) /dataset.google_message3_2.pb/Marshal 105ms ± 2% 98ms ± 7% -6.81% (p=0.009 n=5+6) /dataset.google_message3_4.pb/Marshal 16.3ms ± 6% 15.7ms ± 3% -3.44% (p=0.041 n=6+6) /dataset.google_message3_5.pb/Marshal 676ms ± 4% 504ms ± 2% -25.50% (p=0.004 n=6+5) Change-Id: I72cc4597117f4cf5d236ef505777d49dd4a5f75d Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171020 Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-01 13:49:56 -07:00
type structInfo struct {
sizecacheOffset offset
weakOffset offset
unknownOffset offset
extensionOffset offset
fieldsByNumber map[pref.FieldNumber]reflect.StructField
oneofsByName map[pref.Name]reflect.StructField
oneofWrappersByType map[reflect.Type]pref.FieldNumber
oneofWrappersByNumber map[pref.FieldNumber]reflect.Type
}
func (mi *MessageInfo) makeStructInfo(t reflect.Type) structInfo {
si := structInfo{
sizecacheOffset: invalidOffset,
weakOffset: invalidOffset,
unknownOffset: invalidOffset,
extensionOffset: invalidOffset,
fieldsByNumber: map[pref.FieldNumber]reflect.StructField{},
oneofsByName: map[pref.Name]reflect.StructField{},
oneofWrappersByType: map[reflect.Type]pref.FieldNumber{},
oneofWrappersByNumber: map[pref.FieldNumber]reflect.Type{},
}
if f, _ := t.FieldByName("sizeCache"); f.Type == sizecacheType {
si.sizecacheOffset = offsetOf(f, mi.Exporter)
}
if f, _ := t.FieldByName("XXX_sizecache"); f.Type == sizecacheType {
si.sizecacheOffset = offsetOf(f, mi.Exporter)
}
if f, _ := t.FieldByName("XXX_weak"); f.Type == weakFieldsType {
si.weakOffset = offsetOf(f, mi.Exporter)
}
if f, _ := t.FieldByName("unknownFields"); f.Type == unknownFieldsType {
si.unknownOffset = offsetOf(f, mi.Exporter)
}
if f, _ := t.FieldByName("XXX_unrecognized"); f.Type == unknownFieldsType {
si.unknownOffset = offsetOf(f, mi.Exporter)
}
if f, _ := t.FieldByName("extensionFields"); f.Type == extensionFieldsType {
si.extensionOffset = offsetOf(f, mi.Exporter)
}
if f, _ := t.FieldByName("XXX_InternalExtensions"); f.Type == extensionFieldsType {
si.extensionOffset = offsetOf(f, mi.Exporter)
}
if f, _ := t.FieldByName("XXX_extensions"); f.Type == extensionFieldsType {
si.extensionOffset = offsetOf(f, mi.Exporter)
}
// Generate a mapping of field numbers and names to Go struct field or type.
fieldLoop:
for i := 0; i < t.NumField(); i++ {
f := t.Field(i)
for _, s := range strings.Split(f.Tag.Get("protobuf"), ",") {
if len(s) > 0 && strings.Trim(s, "0123456789") == "" {
n, _ := strconv.ParseUint(s, 10, 64)
si.fieldsByNumber[pref.FieldNumber(n)] = f
continue fieldLoop
}
}
if s := f.Tag.Get("protobuf_oneof"); len(s) > 0 {
si.oneofsByName[pref.Name(s)] = f
continue fieldLoop
}
}
// Derive a mapping of oneof wrappers to fields.
oneofWrappers := mi.OneofWrappers
if fn, ok := reflect.PtrTo(t).MethodByName("XXX_OneofFuncs"); ok {
oneofWrappers = fn.Func.Call([]reflect.Value{reflect.Zero(fn.Type.In(0))})[3].Interface().([]interface{})
}
if fn, ok := reflect.PtrTo(t).MethodByName("XXX_OneofWrappers"); ok {
oneofWrappers = fn.Func.Call([]reflect.Value{reflect.Zero(fn.Type.In(0))})[0].Interface().([]interface{})
}
for _, v := range oneofWrappers {
tf := reflect.TypeOf(v).Elem()
f := tf.Field(0)
for _, s := range strings.Split(f.Tag.Get("protobuf"), ",") {
if len(s) > 0 && strings.Trim(s, "0123456789") == "" {
n, _ := strconv.ParseUint(s, 10, 64)
si.oneofWrappersByType[tf] = pref.FieldNumber(n)
si.oneofWrappersByNumber[pref.FieldNumber(n)] = tf
break
}
}
}
return si
}
// makeKnownFieldsFunc generates functions for operations that can be performed
// on each protobuf message field. It takes in a reflect.Type representing the
// Go struct and matches message fields with struct fields.
//
// This code assumes that the struct is well-formed and panics if there are
// any discrepancies.
func (mi *MessageInfo) makeKnownFieldsFunc(si structInfo) {
mi.fields = map[pref.FieldNumber]*fieldInfo{}
for i := 0; i < mi.PBType.Fields().Len(); i++ {
fd := mi.PBType.Fields().Get(i)
fs := si.fieldsByNumber[fd.Number()]
var fi fieldInfo
switch {
case fd.ContainingOneof() != nil:
fi = fieldInfoForOneof(fd, si.oneofsByName[fd.ContainingOneof().Name()], mi.Exporter, si.oneofWrappersByNumber[fd.Number()])
case fd.IsMap():
fi = fieldInfoForMap(fd, fs, mi.Exporter)
case fd.IsList():
fi = fieldInfoForList(fd, fs, mi.Exporter)
case fd.IsWeak():
fi = fieldInfoForWeakMessage(fd, si.weakOffset)
case fd.Kind() == pref.MessageKind || fd.Kind() == pref.GroupKind:
fi = fieldInfoForMessage(fd, fs, mi.Exporter)
default:
fi = fieldInfoForScalar(fd, fs, mi.Exporter)
}
mi.fields[fd.Number()] = &fi
}
mi.oneofs = map[pref.Name]*oneofInfo{}
for i := 0; i < mi.PBType.Oneofs().Len(); i++ {
od := mi.PBType.Oneofs().Get(i)
mi.oneofs[od.Name()] = makeOneofInfo(od, si.oneofsByName[od.Name()], mi.Exporter, si.oneofWrappersByType)
}
}
func (mi *MessageInfo) makeUnknownFieldsFunc(t reflect.Type, si structInfo) {
mi.getUnknown = func(pointer) pref.RawFields { return nil }
mi.setUnknown = func(pointer, pref.RawFields) { return }
if si.unknownOffset.IsValid() {
mi.getUnknown = func(p pointer) pref.RawFields {
if p.IsNil() {
return nil
}
rv := p.Apply(si.unknownOffset).AsValueOf(unknownFieldsType)
return pref.RawFields(*rv.Interface().(*[]byte))
}
mi.setUnknown = func(p pointer, b pref.RawFields) {
if p.IsNil() {
panic("invalid SetUnknown on nil Message")
}
rv := p.Apply(si.unknownOffset).AsValueOf(unknownFieldsType)
*rv.Interface().(*[]byte) = []byte(b)
}
} else {
mi.getUnknown = func(pointer) pref.RawFields {
return nil
}
mi.setUnknown = func(p pointer, _ pref.RawFields) {
if p.IsNil() {
panic("invalid SetUnknown on nil Message")
}
}
}
}
func (mi *MessageInfo) makeExtensionFieldsFunc(t reflect.Type, si structInfo) {
if si.extensionOffset.IsValid() {
mi.extensionMap = func(p pointer) *extensionMap {
if p.IsNil() {
return (*extensionMap)(nil)
}
v := p.Apply(si.extensionOffset).AsValueOf(extensionFieldsType)
return (*extensionMap)(v.Interface().(*map[int32]ExtensionField))
}
} else {
mi.extensionMap = func(pointer) *extensionMap {
return (*extensionMap)(nil)
}
}
}