kria
kria copied to clipboard
bluemont
Clojure Protobuf
While I understand (and agree with) the rationale (in the README) for not using lein-protobuf, it's companion library clj-protobuf seems like it might be a pretty good fit to replace the code currently under /src/clojure/kria/pb, favoring a data-driven approach to constructing messages.
This pull demonstrates the usage of clojure-protobuf with the same riak-pb compiled java provided by Basho, but without any of the direct interop formerly found in the pb namspaces.
Pros
- Much easier to maintain (net of 1291 LOC removed)
- Better, schema-aware errors
- Protodefs can be queried for their schema
- All args are just clojure data (with the exception of byte-string literals), no mapping
Pair->pbover stuff.
Cons
- clojure-protobuf uses some reflection to build messages, which can slow things down slightly when encoding/decoding complex messages.
- A bit more dependency-heavy
Breaking Changes
There are some small breaking changes to the API, namely every api fn that formerly took plain clojure string args has been changed to require pb byte-strings instead. This eases the burden of coercion on subsequent calls, for instance using Links, etc, and makes the API consistent.
Parsing/Dispatch
kria.core/call no longer takes the req->bytes and bytes->resp fns, dispatching instead on the request/response keys given. This also removes the need to supply no-ops like (fn [_] true) and (fn [_] (byte-array 0)) in api functions, as these are provided when there is no definition found for the message. Clojure-protobuf has the added benefit of more human-readable exceptions when it gets a bad message to write, which should ease use considerably, with direct interop I found myself hunting down NPE's a lot.
The tests only required minor changes mostly concerning api functions that formerly took strings.
Further Work
I plan to do more investigation into the best way to declare/call protobuf definitions, you can see the way I am currently doing it in the pb namespaces.
