Explanations about the generated code
In general, the generated code tries to be close to what the code looks like in other languages while still being ideomatic Elm code. Elm’s concept of “Only one solution to solve” a problem has several consequences here.
General
- Protobufs
message
s are product types,enum
s andoneof
s are union types. - Each
message
andenum
generatesencode[name]
anddecode[name]
functions, which integrate seamlessly with elm-protocol-buffers - Each
message
andenum
generates adefault[name]
function, which sets the defaults as seen in the table above enum
s andoneof
s generate seperate modules, to avoid naming collisions.oneof
s come in two forms, one where every constructor includes a generic type and one where all types are applied. These are needed for use inside of other messages (you will see why in the section “Module Nesting”)
Recursive Data Types
For ease of construction, protoc-gen-elm
prefers to generate type aliases instead of nominal types. Type aliases have one downside though: they cannot be recursive. Otherwise, the Elm compiler would have to do infinite work to expand the type.
So if you have a recursive type like this:
message Rec {
repeated Rec rec = 1;
}
we generate
type alias Rec = { rec : List Rec_ }
type Rec_ = Rec_ Rec
and corresponding wrapRec
and unwrapRec
functions.
gRPC
If your .proto file includes a service
declaration, an Elm module will be generated based on package
and the services name.
This file:
package some_package
service SomeService {}
will generate a Proto/SomePackage/SomeService.elm
module.
The code that needs to be generated inside is actually rather small. A gRPC call just needs
- the package name
- the method name
- the service name
- references to the en/decoder functions
The rest of the work is done by the elm-grpc
package.
It provides functions to convert the generated Grpc.Rpc
instances into Cmd
s and Task
s, as well as setting the usual Http Request fields (headers, timeout, tracker etc.)
Live Example
To run a minimal live example in your browser, follow the instructions in /example/grpc/README.md
.
For a more advanced/realistic example, look at /example/tonic_vite/README.md
.
Well-known types
If you want to use protobufs well-known-types, you need to install the pre-built package elm-protoc-types or include the paths to the proto files in the compilation.
Example: If this is your proto file test.proto
which uses the well-known type Timestamp
,
import "google/protobuf/timestamp.proto";
message TestMessage {
google.protobuf.Timestamp timestamp = 1;
}
the protoc
invocation will need to include the path to the well-known types .proto file.
protoc --elm_out=. test.proto /usr/local/include/google/protobuf/timestamp.proto
Limitations
- All limitations of
elm-protocol-buffers
apply
Development
Note: Currently, this project won’t run on Windows (WSL works) because of shell scripts/executable js files.
Execute npm install
, npm run build
and npm test
and you should be good to go.
You will need protoc
installed and on your PATH.
- The plugin logic is written in Elm itself. To be executable via node, there is a index.js wrapper. It converts the incoming bytes to base64, because there currently is no way to directly send the type
Bytes
through a port. - Main.elm essentially wires up the binding to JS: A request is received through a port, gets decoded, processed and then sent through another port.
- For decoding the protoc request, it uses “itself”, meaning that upgrading protoc versions should be done by running the plugin against the new
include
files from protoc to generate the new encoders/decoders (use theupgrade.sh
script). - A
Mapper
converts the request into a convenient internal structure - A
Generator
then uses this internal structure to build an Elm AST which is then pretty-printed to a file.
Run build.sh
to build the elm code into index.min.js
(which is imported by the entrypoint index.js
).
To analyse the protoc requests, there are debug.js
, DebugMain
and build_debug.sh
files. Run build_debug.sh
, then use debug.js
in place of index.js
when running protoc
. This should dump the deserialized request into debug.log
. You can then put this into the Elm repl for example or use it as input for tests.