2024-10-18 10:48:42 -07:00
|
|
|
#include "google/protobuf/map.h"
|
2024-06-28 06:52:26 -07:00
|
|
|
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
#include <cstddef>
|
2024-06-28 06:52:26 -07:00
|
|
|
#include <cstdint>
|
2025-09-25 10:58:07 -07:00
|
|
|
#include <memory>
|
2024-06-28 06:52:26 -07:00
|
|
|
#include <string>
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
#include <type_traits>
|
2024-07-09 04:44:21 -07:00
|
|
|
#include <utility>
|
2024-06-28 06:52:26 -07:00
|
|
|
|
2024-11-27 08:01:17 -08:00
|
|
|
#include "absl/functional/overload.h"
|
2025-09-25 10:58:07 -07:00
|
|
|
#include "absl/memory/memory.h"
|
2024-12-10 11:06:12 -08:00
|
|
|
#include "absl/strings/string_view.h"
|
2024-09-19 11:30:53 -07:00
|
|
|
#include "google/protobuf/message.h"
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
#include "google/protobuf/message_lite.h"
|
2024-06-28 06:52:26 -07:00
|
|
|
#include "rust/cpp_kernel/strings.h"
|
|
|
|
|
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
namespace google {
|
|
|
|
|
namespace protobuf {
|
|
|
|
|
namespace rust {
|
|
|
|
|
namespace {
|
|
|
|
|
|
2026-01-23 08:24:40 -08:00
|
|
|
using FfiMapValueTag = internal::UntypedMapBase::TypeKind;
|
2024-10-18 10:48:42 -07:00
|
|
|
|
2024-11-21 07:52:29 -08:00
|
|
|
// LINT.IfChange(map_ffi)
|
2026-01-23 08:24:40 -08:00
|
|
|
struct FfiMapValue {
|
|
|
|
|
FfiMapValueTag tag;
|
2024-10-18 10:48:42 -07:00
|
|
|
union {
|
|
|
|
|
bool b;
|
|
|
|
|
uint32_t u32;
|
|
|
|
|
uint64_t u64;
|
2024-11-21 07:52:29 -08:00
|
|
|
float f32;
|
|
|
|
|
double f64;
|
2024-10-18 10:48:42 -07:00
|
|
|
std::string* s;
|
|
|
|
|
google::protobuf::MessageLite* message;
|
|
|
|
|
};
|
|
|
|
|
};
|
2026-03-20 11:10:52 -07:00
|
|
|
// LINT.ThenChange(//depot/google3/third_party/protobuf/rust/cpp_kernel/map.rs:map_ffi)
|
2024-10-18 10:48:42 -07:00
|
|
|
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
template <typename T>
|
|
|
|
|
struct FromViewType {
|
|
|
|
|
using type = T;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
template <>
|
|
|
|
|
struct FromViewType<PtrAndLen> {
|
|
|
|
|
using type = std::string;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
template <typename Key>
|
|
|
|
|
using KeyMap = internal::KeyMapBase<
|
|
|
|
|
internal::KeyForBase<typename FromViewType<Key>::type>>;
|
|
|
|
|
|
2024-12-10 13:43:39 -08:00
|
|
|
template <typename T>
|
|
|
|
|
T AsViewType(T t) {
|
|
|
|
|
return t;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
absl::string_view AsViewType(PtrAndLen key) { return key.AsStringView(); }
|
|
|
|
|
|
2025-09-25 10:58:07 -07:00
|
|
|
constexpr bool kHasFullRuntime = true;
|
2024-10-18 10:48:42 -07:00
|
|
|
|
2025-09-25 10:58:07 -07:00
|
|
|
void InitializeMessageValue(void* raw_ptr,
|
|
|
|
|
std::unique_ptr<MessageLite> prototype) {
|
|
|
|
|
MessageLite* new_msg =
|
|
|
|
|
internal::RustMapHelper::PlacementNew(prototype.get(), raw_ptr);
|
2024-10-18 10:48:42 -07:00
|
|
|
// If we are working with a full (non-lite) proto, we reflectively swap the
|
|
|
|
|
// value into place. Otherwise, we have to perform a copy.
|
2025-09-25 10:58:07 -07:00
|
|
|
if constexpr (kHasFullRuntime) {
|
|
|
|
|
auto* full_msg = DynamicCastMessage<Message>(new_msg);
|
|
|
|
|
if (full_msg != nullptr) {
|
|
|
|
|
full_msg->GetReflection()->Swap(
|
|
|
|
|
full_msg, DynamicCastMessage<Message>(prototype.get()));
|
|
|
|
|
} else {
|
|
|
|
|
new_msg->CheckTypeAndMergeFrom(*prototype);
|
|
|
|
|
}
|
2024-10-18 10:48:42 -07:00
|
|
|
} else {
|
2025-09-25 10:58:07 -07:00
|
|
|
new_msg->CheckTypeAndMergeFrom(*prototype);
|
2024-10-18 10:48:42 -07:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
template <typename Key>
|
2026-01-23 08:24:40 -08:00
|
|
|
bool Insert(internal::UntypedMapBase* m, Key key, FfiMapValue value) {
|
2024-11-27 08:01:17 -08:00
|
|
|
internal::NodeBase* node = internal::RustMapHelper::AllocNode(m);
|
2026-06-14 20:24:46 -07:00
|
|
|
if constexpr (std::is_same_v<Key, PtrAndLen>) {
|
2024-12-10 11:06:12 -08:00
|
|
|
key.PlacementNewString(node->GetVoidKey());
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
} else {
|
|
|
|
|
*static_cast<Key*>(node->GetVoidKey()) = key;
|
|
|
|
|
}
|
2024-09-19 11:30:53 -07:00
|
|
|
|
2024-11-27 08:01:17 -08:00
|
|
|
m->VisitValue(node, absl::Overload{
|
|
|
|
|
[&](bool* v) { *v = value.b; },
|
|
|
|
|
[&](uint32_t* v) { *v = value.u32; },
|
|
|
|
|
[&](uint64_t* v) { *v = value.u64; },
|
|
|
|
|
[&](float* v) { *v = value.f32; },
|
|
|
|
|
[&](double* v) { *v = value.f64; },
|
|
|
|
|
[&](std::string* str) {
|
|
|
|
|
new (str) std::string(std::move(*value.s));
|
|
|
|
|
delete value.s;
|
|
|
|
|
},
|
|
|
|
|
[&](MessageLite* msg) {
|
2025-09-25 10:58:07 -07:00
|
|
|
InitializeMessageValue(
|
|
|
|
|
msg, absl::WrapUnique(value.message));
|
2024-11-27 08:01:17 -08:00
|
|
|
},
|
|
|
|
|
});
|
2024-09-19 11:30:53 -07:00
|
|
|
|
2024-12-10 13:43:39 -08:00
|
|
|
return internal::RustMapHelper::InsertOrReplaceNode(
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
static_cast<KeyMap<Key>*>(m), node);
|
|
|
|
|
}
|
|
|
|
|
|
2024-11-27 08:01:17 -08:00
|
|
|
void PopulateMapValue(const internal::UntypedMapBase& map,
|
2026-01-23 08:24:40 -08:00
|
|
|
internal::NodeBase* node, FfiMapValue& output) {
|
2024-11-27 08:01:17 -08:00
|
|
|
map.VisitValue(node, absl::Overload{
|
|
|
|
|
[&](const bool* v) {
|
2026-01-23 08:24:40 -08:00
|
|
|
output.tag = FfiMapValueTag::kBool;
|
2024-11-27 08:01:17 -08:00
|
|
|
output.b = *v;
|
|
|
|
|
},
|
|
|
|
|
[&](const uint32_t* v) {
|
2026-01-23 08:24:40 -08:00
|
|
|
output.tag = FfiMapValueTag::kU32;
|
2024-11-27 08:01:17 -08:00
|
|
|
output.u32 = *v;
|
|
|
|
|
},
|
|
|
|
|
[&](const uint64_t* v) {
|
2026-01-23 08:24:40 -08:00
|
|
|
output.tag = FfiMapValueTag::kU64;
|
2024-11-27 08:01:17 -08:00
|
|
|
output.u64 = *v;
|
|
|
|
|
},
|
|
|
|
|
[&](const float* v) {
|
2026-01-23 08:24:40 -08:00
|
|
|
output.tag = FfiMapValueTag::kFloat;
|
2024-11-27 08:01:17 -08:00
|
|
|
output.f32 = *v;
|
|
|
|
|
},
|
|
|
|
|
[&](const double* v) {
|
2026-01-23 08:24:40 -08:00
|
|
|
output.tag = FfiMapValueTag::kDouble;
|
2024-11-27 08:01:17 -08:00
|
|
|
output.f64 = *v;
|
|
|
|
|
},
|
|
|
|
|
[&](std::string* str) {
|
2026-01-23 08:24:40 -08:00
|
|
|
output.tag = FfiMapValueTag::kString;
|
2024-11-27 08:01:17 -08:00
|
|
|
output.s = str;
|
|
|
|
|
},
|
|
|
|
|
[&](MessageLite* msg) {
|
2026-01-23 08:24:40 -08:00
|
|
|
output.tag = FfiMapValueTag::kMessage;
|
2024-11-27 08:01:17 -08:00
|
|
|
output.message = msg;
|
|
|
|
|
},
|
|
|
|
|
});
|
2024-10-18 10:48:42 -07:00
|
|
|
}
|
|
|
|
|
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
template <typename Key>
|
2026-01-23 08:24:40 -08:00
|
|
|
bool Get(internal::UntypedMapBase* m, Key key, FfiMapValue* value) {
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
auto* map_base = static_cast<KeyMap<Key>*>(m);
|
2024-12-10 13:43:39 -08:00
|
|
|
auto result = internal::RustMapHelper::FindHelper(map_base, AsViewType(key));
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
if (result.node == nullptr) {
|
|
|
|
|
return false;
|
|
|
|
|
}
|
2024-11-27 08:01:17 -08:00
|
|
|
PopulateMapValue(*m, result.node, *value);
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
return true;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template <typename Key>
|
2024-11-27 08:01:17 -08:00
|
|
|
bool Remove(internal::UntypedMapBase* m, Key key) {
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
auto* map_base = static_cast<KeyMap<Key>*>(m);
|
2024-12-10 13:43:39 -08:00
|
|
|
return internal::RustMapHelper::EraseImpl(map_base, AsViewType(key));
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template <typename Key>
|
2024-11-27 08:01:17 -08:00
|
|
|
void IterGet(const internal::UntypedMapIterator* iter, Key* key,
|
2026-01-23 08:24:40 -08:00
|
|
|
FfiMapValue* value) {
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
internal::NodeBase* node = iter->node_;
|
2026-06-14 20:24:46 -07:00
|
|
|
if constexpr (std::is_same_v<Key, PtrAndLen>) {
|
2024-11-27 08:01:17 -08:00
|
|
|
const std::string* s = iter->m_->GetKey<std::string>(node);
|
2024-08-23 13:24:13 -07:00
|
|
|
*key = PtrAndLen{s->data(), s->size()};
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
} else {
|
2024-11-27 08:01:17 -08:00
|
|
|
*key = *iter->m_->GetKey<Key>(node);
|
2024-09-30 14:09:02 -07:00
|
|
|
}
|
2024-11-27 08:01:17 -08:00
|
|
|
PopulateMapValue(*iter->m_, node, *value);
|
2024-09-30 14:09:02 -07:00
|
|
|
}
|
|
|
|
|
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
} // namespace
|
|
|
|
|
} // namespace rust
|
|
|
|
|
} // namespace protobuf
|
|
|
|
|
} // namespace google
|
|
|
|
|
|
2024-06-28 06:52:26 -07:00
|
|
|
extern "C" {
|
|
|
|
|
|
2024-07-02 11:42:07 -07:00
|
|
|
void proto2_rust_thunk_UntypedMapIterator_increment(
|
2024-06-28 06:52:26 -07:00
|
|
|
google::protobuf::internal::UntypedMapIterator* iter) {
|
|
|
|
|
iter->PlusPlus();
|
|
|
|
|
}
|
|
|
|
|
|
2024-11-21 07:52:29 -08:00
|
|
|
google::protobuf::internal::UntypedMapBase* proto2_rust_map_new(
|
2026-01-23 08:24:40 -08:00
|
|
|
google::protobuf::rust::FfiMapValue key_prototype,
|
|
|
|
|
google::protobuf::rust::FfiMapValue value_prototype) {
|
2024-11-21 07:52:29 -08:00
|
|
|
return new google::protobuf::internal::UntypedMapBase(
|
|
|
|
|
google::protobuf::internal::UntypedMapBase::GetTypeInfoDynamic(
|
|
|
|
|
key_prototype.tag, value_prototype.tag,
|
2026-01-23 08:24:40 -08:00
|
|
|
value_prototype.tag == google::protobuf::rust::FfiMapValueTag::kMessage
|
2024-11-21 07:52:29 -08:00
|
|
|
? value_prototype.message
|
|
|
|
|
: nullptr));
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
size_t proto2_rust_map_size(google::protobuf::internal::UntypedMapBase* m) {
|
|
|
|
|
return m->size();
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
google::protobuf::internal::UntypedMapIterator proto2_rust_map_iter(
|
|
|
|
|
google::protobuf::internal::UntypedMapBase* m) {
|
|
|
|
|
return m->begin();
|
|
|
|
|
}
|
|
|
|
|
|
2024-11-22 09:35:25 -08:00
|
|
|
void proto2_rust_map_free(google::protobuf::internal::UntypedMapBase* m) {
|
2025-10-13 21:16:11 -07:00
|
|
|
m->ClearTable(m->arena(), /*reset=*/false);
|
2024-11-22 09:35:25 -08:00
|
|
|
delete m;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void proto2_rust_map_clear(google::protobuf::internal::UntypedMapBase* m) {
|
2025-10-13 21:16:11 -07:00
|
|
|
m->ClearTable(m->arena(), /*reset=*/true);
|
2024-11-22 09:35:25 -08:00
|
|
|
}
|
|
|
|
|
|
2024-11-27 08:01:17 -08:00
|
|
|
#define DEFINE_KEY_SPECIFIC_MAP_OPERATIONS(cpp_type, suffix) \
|
|
|
|
|
bool proto2_rust_map_insert_##suffix(google::protobuf::internal::UntypedMapBase* m, \
|
|
|
|
|
cpp_type key, \
|
2026-01-23 08:24:40 -08:00
|
|
|
google::protobuf::rust::FfiMapValue value) { \
|
2024-11-27 08:01:17 -08:00
|
|
|
return google::protobuf::rust::Insert(m, key, value); \
|
|
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
bool proto2_rust_map_get_##suffix(google::protobuf::internal::UntypedMapBase* m, \
|
|
|
|
|
cpp_type key, \
|
2026-01-23 08:24:40 -08:00
|
|
|
google::protobuf::rust::FfiMapValue* value) { \
|
2024-11-27 08:01:17 -08:00
|
|
|
return google::protobuf::rust::Get(m, key, value); \
|
|
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
bool proto2_rust_map_remove_##suffix(google::protobuf::internal::UntypedMapBase* m, \
|
|
|
|
|
cpp_type key) { \
|
|
|
|
|
return google::protobuf::rust::Remove(m, key); \
|
|
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
void proto2_rust_map_iter_get_##suffix( \
|
|
|
|
|
const google::protobuf::internal::UntypedMapIterator* iter, cpp_type* key, \
|
2026-01-23 08:24:40 -08:00
|
|
|
google::protobuf::rust::FfiMapValue* value) { \
|
2024-11-27 08:01:17 -08:00
|
|
|
return google::protobuf::rust::IterGet(iter, key, value); \
|
Rust: cut down on the amount of generated C++ code needed for maps
With the C++ kernel for Rust, we currently need to generate quite a few C++
thunks for operations on map fields. For each message we generate, we generate
these thunks for all possible map types that could have that message as a
value. These operations are for things such as insertion, removal, clearing,
iterating, etc.
The reason we do this is that templated types don't play well with FFI, so we
effectively need separate FFI endpoints for every possible combination of key
and value types used (or even potentially used) as a map field.
This CL fixes the problem by replacing the generated thunks with functions in
the runtime that can operate on `proto2::MessageLite*` without needing to care
about the specific message type.
The way it works is that we implement the operations using either
`UntypedMapBase` (the base class of all map types, which knows nothing about
the key and value types) or `KeyMapBase`, which knows the key type but not the
value type. I roughly followed the example of the table-driven parser, which
has a similar problem of needing to operate generically on maps without having
access to the concrete types.
I removed 54 thunks per message (that's 6 key types times 9 operations per
key), but had to add two new thunks per message:
- The `size_info` thunk looks up the `MapNodeSizeInfoT`, which is stored in a
small constant table. The important thing here is an offset indicating where
to look for the value in each map entry. This offset can be different for
every pair of key and value types, but we can safely assume that the result
does not depend on the signedness of the key. As a result we only need to
store four entries per message: one each for i32, i64, bool, and string.
- The `placement_new` thunk move-constructs a message in place. We need this
to be able to efficiently implement map insertion.
There are two big things that this CL does not address yet but which I plan to
follow up on:
- Enums still generate many map-related C++ thunks that could be replaced with
a common implementation. This should actually be much easier to handle than
messages, because every enum has the same representation as an i32.
- We still generate six `ProxiedInMapValue` implementations for every message,
but it should be possible to replace these with a blanket implementation that
works for all message types.
PiperOrigin-RevId: 657681421
2024-07-30 12:03:36 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
DEFINE_KEY_SPECIFIC_MAP_OPERATIONS(int32_t, i32)
|
|
|
|
|
DEFINE_KEY_SPECIFIC_MAP_OPERATIONS(uint32_t, u32)
|
|
|
|
|
DEFINE_KEY_SPECIFIC_MAP_OPERATIONS(int64_t, i64)
|
|
|
|
|
DEFINE_KEY_SPECIFIC_MAP_OPERATIONS(uint64_t, u64)
|
|
|
|
|
DEFINE_KEY_SPECIFIC_MAP_OPERATIONS(bool, bool)
|
|
|
|
|
DEFINE_KEY_SPECIFIC_MAP_OPERATIONS(google::protobuf::rust::PtrAndLen, ProtoString)
|
|
|
|
|
|
2024-06-28 06:52:26 -07:00
|
|
|
} // extern "C"
|