Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
// Copyright: Ankitects Pty Ltd and contributors
|
|
|
|
// License: GNU AGPL, version 3 or later; http://www.gnu.org/licenses/agpl.html
|
|
|
|
|
|
|
|
use std::env;
|
|
|
|
use std::fmt::Write;
|
|
|
|
use std::path::PathBuf;
|
|
|
|
|
|
|
|
use anki_io::write_file_if_changed;
|
2023-06-22 01:38:05 +02:00
|
|
|
use anki_proto_gen::get_services;
|
|
|
|
use anki_proto_gen::BackendService;
|
|
|
|
use anki_proto_gen::CollectionService;
|
|
|
|
use anki_proto_gen::Method;
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
use anyhow::Context;
|
|
|
|
use anyhow::Result;
|
|
|
|
use inflections::Inflect;
|
2023-08-02 12:29:44 +02:00
|
|
|
use itertools::Itertools;
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
use prost_reflect::DescriptorPool;
|
|
|
|
|
|
|
|
pub fn write_rust_interface(pool: &DescriptorPool) -> Result<()> {
|
|
|
|
let mut buf = String::new();
|
|
|
|
buf.push_str("use crate::error::Result; use prost::Message;");
|
|
|
|
|
2023-06-22 01:38:05 +02:00
|
|
|
let (col_services, backend_services) = get_services(pool);
|
2023-08-02 12:29:44 +02:00
|
|
|
let col_services = col_services
|
|
|
|
.into_iter()
|
|
|
|
.filter(|s| s.name != "FrontendService")
|
|
|
|
.collect_vec();
|
|
|
|
let backend_services = backend_services
|
|
|
|
.into_iter()
|
|
|
|
.filter(|s| s.name != "BackendFrontendService")
|
|
|
|
.collect_vec();
|
2023-06-22 01:38:05 +02:00
|
|
|
|
|
|
|
render_collection_services(&col_services, &mut buf)?;
|
|
|
|
render_backend_services(&backend_services, &mut buf)?;
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
|
|
|
|
let buf = format_code(buf)?;
|
2023-06-22 01:38:05 +02:00
|
|
|
// println!("{}", &buf);
|
|
|
|
// panic!();
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
let out_dir = env::var("OUT_DIR").unwrap();
|
|
|
|
let path = PathBuf::from(out_dir).join("backend.rs");
|
|
|
|
write_file_if_changed(path, buf).context("write file")?;
|
|
|
|
Ok(())
|
|
|
|
}
|
|
|
|
|
2023-06-22 01:38:05 +02:00
|
|
|
fn render_collection_services(col_services: &[CollectionService], buf: &mut String) -> Result<()> {
|
|
|
|
for service in col_services {
|
|
|
|
render_collection_trait(service, buf);
|
|
|
|
render_individual_service_run_method_for_collection(buf, service);
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
}
|
2023-06-22 01:38:05 +02:00
|
|
|
render_top_level_run_method(
|
|
|
|
col_services.iter().map(|s| (s.index, s.name.as_str())),
|
|
|
|
"&mut self",
|
|
|
|
"crate::collection::Collection",
|
|
|
|
buf,
|
|
|
|
);
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
|
2023-06-22 01:38:05 +02:00
|
|
|
Ok(())
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
}
|
|
|
|
|
2023-06-22 01:38:05 +02:00
|
|
|
fn render_backend_services(backend_services: &[BackendService], buf: &mut String) -> Result<()> {
|
|
|
|
for service in backend_services {
|
|
|
|
render_backend_trait(service, buf);
|
|
|
|
render_delegating_backend_methods(service, buf);
|
|
|
|
render_individual_service_run_method_for_backend(buf, service);
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
}
|
2023-06-22 01:38:05 +02:00
|
|
|
render_top_level_run_method(
|
|
|
|
backend_services.iter().map(|s| (s.index, s.name.as_str())),
|
|
|
|
"&self",
|
|
|
|
"crate::backend::Backend",
|
|
|
|
buf,
|
|
|
|
);
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
|
2023-06-22 01:38:05 +02:00
|
|
|
Ok(())
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
fn format_code(code: String) -> Result<String> {
|
|
|
|
let syntax_tree = syn::parse_file(&code)?;
|
|
|
|
Ok(prettyplease::unparse(&syntax_tree))
|
|
|
|
}
|
|
|
|
|
2023-06-22 01:38:05 +02:00
|
|
|
fn render_collection_trait(service: &CollectionService, buf: &mut String) {
|
|
|
|
let name = &service.name;
|
|
|
|
writeln!(buf, "pub trait {name} {{").unwrap();
|
|
|
|
for method in &service.trait_methods {
|
|
|
|
render_trait_method(method, "&mut self", buf);
|
|
|
|
}
|
|
|
|
buf.push('}');
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
}
|
|
|
|
|
2023-06-22 01:38:05 +02:00
|
|
|
fn render_trait_method(method: &Method, self_kind: &str, buf: &mut String) {
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
let method_name = &method.name;
|
|
|
|
let input_with_label = method.get_input_arg_with_label();
|
|
|
|
let output_type = method.get_output_type();
|
|
|
|
writeln!(
|
|
|
|
buf,
|
2023-06-22 01:38:05 +02:00
|
|
|
"fn {method_name}({self_kind}, {input_with_label}) -> Result<{output_type}>;"
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
)
|
|
|
|
.unwrap();
|
|
|
|
}
|
|
|
|
|
2023-06-22 01:38:05 +02:00
|
|
|
fn render_backend_trait(service: &BackendService, buf: &mut String) {
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
let name = &service.name;
|
|
|
|
writeln!(buf, "pub trait {name} {{").unwrap();
|
2023-06-22 01:38:05 +02:00
|
|
|
for method in &service.trait_methods {
|
|
|
|
render_trait_method(method, "&self", buf);
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
}
|
|
|
|
buf.push('}');
|
|
|
|
}
|
|
|
|
|
2023-06-22 01:38:05 +02:00
|
|
|
fn render_delegating_backend_methods(service: &BackendService, buf: &mut String) {
|
|
|
|
buf.push_str("impl crate::backend::Backend {");
|
|
|
|
for method in &service.delegating_methods {
|
|
|
|
render_delegating_backend_method(method, service.name.trim_start_matches("Backend"), buf);
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
}
|
|
|
|
buf.push('}');
|
|
|
|
}
|
|
|
|
|
2023-06-22 01:38:05 +02:00
|
|
|
fn render_delegating_backend_method(method: &Method, method_qualifier: &str, buf: &mut String) {
|
|
|
|
let method_name = &method.name;
|
|
|
|
let input_with_label = method.get_input_arg_with_label();
|
|
|
|
let input = method.text_if_input_not_empty(|_| "input".into());
|
|
|
|
let output_type = method.get_output_type();
|
|
|
|
writeln!(
|
|
|
|
buf,
|
|
|
|
"fn {method_name}(&self, {input_with_label}) -> Result<{output_type}> {{
|
|
|
|
self.with_col(|col| {method_qualifier}::{method_name}(col, {input})) }}",
|
|
|
|
)
|
|
|
|
.unwrap();
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
// Matches all service types and delegates to the revelant self.run_foo_method()
|
2023-06-22 01:38:05 +02:00
|
|
|
fn render_top_level_run_method<'a>(
|
|
|
|
// (index, name)
|
|
|
|
services: impl Iterator<Item = (usize, &'a str)>,
|
|
|
|
self_kind: &str,
|
|
|
|
struct_name: &str,
|
|
|
|
buf: &mut String,
|
|
|
|
) {
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
writeln!(buf,
|
2023-06-22 01:38:05 +02:00
|
|
|
r#" impl {struct_name} {{
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
pub fn run_service_method({self_kind}, service: u32, method: u32, input: &[u8]) -> Result<Vec<u8>, Vec<u8>> {{
|
|
|
|
match service {{
|
|
|
|
"#,
|
|
|
|
).unwrap();
|
2023-06-22 01:38:05 +02:00
|
|
|
for (idx, service) in services {
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
writeln!(
|
|
|
|
buf,
|
|
|
|
"{idx} => self.run_{service}_method(method, input),",
|
2023-06-22 01:38:05 +02:00
|
|
|
service = service.to_snake_case()
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
)
|
|
|
|
.unwrap();
|
|
|
|
}
|
|
|
|
buf.push_str(
|
|
|
|
r#"
|
|
|
|
_ => Err(crate::error::AnkiError::InvalidServiceIndex),
|
|
|
|
}
|
|
|
|
.map_err(|err| {
|
|
|
|
let backend_err = err.into_protobuf(&self.tr);
|
|
|
|
let mut bytes = Vec::new();
|
|
|
|
backend_err.encode(&mut bytes).unwrap();
|
|
|
|
bytes
|
|
|
|
})
|
|
|
|
} }"#,
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
2023-06-22 01:38:05 +02:00
|
|
|
fn render_individual_service_run_method_for_collection(
|
|
|
|
buf: &mut String,
|
|
|
|
service: &CollectionService,
|
|
|
|
) {
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
let service_name = &service.name.to_snake_case();
|
|
|
|
writeln!(
|
|
|
|
buf,
|
|
|
|
"#[allow(unused_variables, clippy::match_single_binding)]
|
2023-06-22 01:38:05 +02:00
|
|
|
impl crate::collection::Collection {{ pub(crate) fn run_{service_name}_method(&mut self,
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
method: u32, input: &[u8]) -> Result<Vec<u8>> {{
|
|
|
|
match method {{",
|
|
|
|
)
|
|
|
|
.unwrap();
|
2023-06-22 01:38:05 +02:00
|
|
|
for method in &service.trait_methods {
|
|
|
|
render_method_in_match_expression(method, &service.name, buf);
|
|
|
|
}
|
|
|
|
buf.push_str(
|
|
|
|
r#"
|
|
|
|
_ => Err(crate::error::AnkiError::InvalidMethodIndex),
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
}
|
2023-06-22 01:38:05 +02:00
|
|
|
} }
|
|
|
|
"#,
|
|
|
|
);
|
|
|
|
}
|
|
|
|
|
|
|
|
fn render_individual_service_run_method_for_backend(buf: &mut String, service: &BackendService) {
|
|
|
|
let service_name = &service.name.to_snake_case();
|
|
|
|
writeln!(
|
|
|
|
buf,
|
|
|
|
"#[allow(unused_variables, clippy::match_single_binding)]
|
|
|
|
impl crate::backend::Backend {{ pub(crate) fn run_{service_name}_method(&self,
|
|
|
|
method: u32, input: &[u8]) -> Result<Vec<u8>> {{
|
|
|
|
match method {{",
|
|
|
|
)
|
|
|
|
.unwrap();
|
|
|
|
for method in &service.trait_methods {
|
|
|
|
render_method_in_match_expression(method, &service.name, buf);
|
|
|
|
}
|
|
|
|
for method in &service.delegating_methods {
|
|
|
|
render_method_in_match_expression(method, "crate::backend::Backend", buf);
|
Refactor service generation (#2552)
* Automatically elide empty inputs and outputs to backend methods
* Refactor service generation
Despite the fact that the majority of our Protobuf service methods require
an open collection, they were not accessible with just a Collection
object. To access the methods (e.g. because we haven't gotten around to
exposing the correct API in Collection yet), you had to wrap the collection
in a Backend object, and pay a mutex-acquisition cost for each call, even
if you have exclusive access to the object.
This commit migrates the majority of service methods to the Collection, so
they can now be used directly, and improves the ergonomics a bit at the
same time.
The approach taken:
- The service generation now happens in rslib instead of anki_proto, which
avoids the need for trait constraints and associated types.
- Service methods are assumed to be collection-based by default. Instead of
implementing the service on Backend, we now implement it on Collection, which
means our methods no longer need to use self.with_col(...).
- We automatically generate methods in Backend which use self.with_col() to
delegate to the Collection method.
- For methods that are only appropriate for the backend, we add a flag in
the .proto file. The codegen uses this flag to write the method into a
BackendFooService instead of FooService, which the backend implements.
- The flag can also allows us to define separate implementations for collection
and backend, so we can e.g. skip the collection mutex in the i18n service
while also providing the service on a collection.
2023-06-19 07:33:40 +02:00
|
|
|
}
|
|
|
|
buf.push_str(
|
|
|
|
r#"
|
|
|
|
_ => Err(crate::error::AnkiError::InvalidMethodIndex),
|
|
|
|
}
|
|
|
|
} }
|
|
|
|
"#,
|
|
|
|
);
|
|
|
|
}
|
2023-06-22 01:38:05 +02:00
|
|
|
|
|
|
|
fn render_method_in_match_expression(method: &Method, method_qualifier: &str, buf: &mut String) {
|
|
|
|
let decode_input =
|
|
|
|
method.text_if_input_not_empty(|kind| format!("let input = {kind}::decode(input)?;"));
|
|
|
|
let rust_method = &method.name;
|
|
|
|
let input = method.text_if_input_not_empty(|_| "input".into());
|
|
|
|
let output_assign = method.text_if_output_not_empty(|_| "let output = ".into());
|
|
|
|
let idx = method.index;
|
|
|
|
let output = if method.output().is_none() {
|
|
|
|
"Vec::new()"
|
|
|
|
} else {
|
|
|
|
"{ let mut out_bytes = Vec::new();
|
|
|
|
output.encode(&mut out_bytes)?;
|
|
|
|
out_bytes }"
|
|
|
|
};
|
|
|
|
writeln!(
|
|
|
|
buf,
|
|
|
|
"{idx} => {{ {decode_input}
|
|
|
|
{output_assign} {method_qualifier}::{rust_method}(self, {input})?;
|
|
|
|
Ok({output}) }},",
|
|
|
|
)
|
|
|
|
.unwrap();
|
|
|
|
}
|
|
|
|
|
|
|
|
trait MethodHelpers {
|
|
|
|
fn input_type(&self) -> Option<String>;
|
|
|
|
fn output_type(&self) -> Option<String>;
|
|
|
|
fn text_if_input_not_empty(&self, text: impl Fn(&String) -> String) -> String;
|
|
|
|
fn get_input_arg_with_label(&self) -> String;
|
|
|
|
fn get_output_type(&self) -> String;
|
|
|
|
fn text_if_output_not_empty(&self, text: impl Fn(&String) -> String) -> String;
|
|
|
|
}
|
|
|
|
|
|
|
|
impl MethodHelpers for Method {
|
|
|
|
fn input_type(&self) -> Option<String> {
|
|
|
|
self.input().map(|t| rust_type(t.full_name()))
|
|
|
|
}
|
|
|
|
|
|
|
|
fn output_type(&self) -> Option<String> {
|
|
|
|
self.output().map(|t| rust_type(t.full_name()))
|
|
|
|
}
|
|
|
|
/// No text if generic::Empty
|
|
|
|
fn text_if_input_not_empty(&self, text: impl Fn(&String) -> String) -> String {
|
|
|
|
self.input_type().as_ref().map(text).unwrap_or_default()
|
|
|
|
}
|
|
|
|
|
|
|
|
/// No text if generic::Empty
|
|
|
|
fn get_input_arg_with_label(&self) -> String {
|
|
|
|
self.input_type()
|
|
|
|
.as_ref()
|
|
|
|
.map(|t| format!("input: {}", t))
|
|
|
|
.unwrap_or_default()
|
|
|
|
}
|
|
|
|
|
|
|
|
/// () if generic::Empty
|
|
|
|
fn get_output_type(&self) -> String {
|
|
|
|
self.output_type().as_deref().unwrap_or("()").into()
|
|
|
|
}
|
|
|
|
|
|
|
|
fn text_if_output_not_empty(&self, text: impl Fn(&String) -> String) -> String {
|
|
|
|
self.output_type().as_ref().map(text).unwrap_or_default()
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
fn rust_type(name: &str) -> String {
|
2023-09-02 08:13:03 +02:00
|
|
|
let Some((head, tail)) = name.rsplit_once('.') else {
|
|
|
|
panic!()
|
|
|
|
};
|
2023-06-22 01:38:05 +02:00
|
|
|
format!(
|
|
|
|
"{}::{}",
|
|
|
|
head.to_snake_case()
|
|
|
|
.replace('.', "::")
|
|
|
|
.replace("anki::", "anki_proto::"),
|
|
|
|
tail
|
|
|
|
)
|
|
|
|
}
|