3 releases
0.1.2 | Aug 27, 2021 |
---|---|
0.1.1 | Aug 17, 2021 |
0.1.0 | Aug 11, 2021 |
#955 in Rust patterns
70KB
2K
SLoC
Autoproto
This crate implements a custom derive macro for the prost::Message
trait, along with including
some helper traits to make automatic derivation as simple as possible - basically putting more
information in the type system instead of in the macro implementation. If the macro doesn't do
something that you need it to do, you can implement one of the "base traits" like ProtoStruct
and
ProtoOneof
and use the functions in the generic
module to implement the rest of the traits.
Limitations
(Currently) no oneof-nested-within-struct
This can be lifted eventually, but right now a major limitation of this derive macro is that we cannot support protobuf types like so:
message Foo {
string first = 1;
oneof some_oneof {
string second = 2;
string third = 3;
}
}
As the library is written, it will only support oneof
when it is the only member of a message.
No automatic generation from .proto
files
This is the biggest change from prost
, and was a deliberate choice. I consider automatically
generating the Rust files from protobuf files to be (at least in part) a misfeature of prost
,
since for many cases it leads to extremely unwieldy types in Rust. Rust has a deep and rich type
system and it isn't possible for Protobuf to nicely represent it. Eventually I want to have compile-
time checking that a Rust type is compatible with a given protobuf file, but this is currently not
implemented.
Improvements and changes from #[derive(prost::Message)]
Supporting more types as fields
This macro supports collections other than Vec
for repeated fields, bare enumerations in structs,
usize
/isize
, and maps. For repeated collections, any type that implements std::iter::Extend
and where a reference to that type can be iterated over can be used as a collection in a struct.
Currently you must manually implement Proto
and ProtoEncode
for these types, but if and when
feature(specialization)
is stablised this restriction should hopefully be lifted.
While currently we cannot automatically support any type implementing proto::Message
as a field -
because that trait is implemented for scalars and the main reason we have our own trait is so we can
have a different impl for scalars - you just need to implement the IsMessage
marker trait to allow
the type to be used as a field in a #[derive(autoproto::Message)]
type.
Also, messages no longer need to be wrapped in an Option
, as protobuf's eager decoding already
requires that all messages implement Default
. You can wrap any type in an Option
if you want to
distinguish between a field being supplied with default values or not supplied at all.
Deriving for more kinds of types
This macro allows deriving for pretty much any tagged union, and deriving for generic structures. For example:
# #![feature(generic_associated_types)]
#[derive(Copy, Clone, PartialEq, Debug, autoproto::Message)]
enum Oneof<A, B, C> {
Nothing,
One(A),
Two(A, B),
Three(A, B, C),
}
No mixed tagged-untagged structs
One change from the prost
macro is that either all fields must be tagged or no fields can be
tagged. For example, these two are ok:
# #![feature(generic_associated_types)]
#[derive(Copy, Clone, PartialEq, Default, Debug, autoproto::Message)]
struct SomeStructTagged<A, B, C, D, E> {
#[autoproto(tag = 1)]
a: A,
#[autoproto(tag = 2)]
b: B,
#[autoproto(tag = 3)]
c: C,
#[autoproto(tag = 4)]
d: D,
#[autoproto(tag = 5)]
e: E,
}
#[derive(Copy, Clone, PartialEq, Default, Debug, autoproto::Message)]
struct SomeStruct<A, B, C, D, E> {
a: A,
b: B,
c: C,
d: D,
e: E,
}
But the following is not:
# #![feature(generic_associated_types)]
#[derive(Copy, Clone, PartialEq, Default, Debug, autoproto::Message)]
struct SomeStruct<A, B, C, D, E> {
a: A,
b: B,
#[autoproto(tag = 1)]
c: C,
d: D,
e: E,
}
Dependencies
~2MB
~48K SLoC