Pierre Ricadat

Software Architect | Scala Lover | Creator of Caliban and Shardcake

[Pierre Ricadat @ghostdogpr]

Do you like Caliban? Or maybe Shardcake? Enjoyed OSS contributions I’ve done over the years? Have I helped you on Discord?

I’ve just setup a GitHub Sponsor page so if you’d like to give a little back or encourage future contributions, head up to https://t.co/htxLQ6iHwH 🙏

[Pierre Ricadat @ghostdogpr]

@hepin1989 All backends in Proteus are based on grpc-java, but akka-grpc/pekko-grpc don't use it. They have their own grpc implementation on top of akka-http.

[Pierre Ricadat @ghostdogpr]

Proteus 0.5.0 comes with a new gRPC backend, for Kyo!

Release notes: https://t.co/gzDaIoxYZ6

[Pierre Ricadat @ghostdogpr]

@matej_cerny The "safe" mode released in 3.8.3 and built on top of capture checking is very interesting though. It allows the compiler to guarantee you're not side-effecting in some parts of the code, quite nice with https://t.co/wlwMlrQinx or https://t.co/CmMrZc6Yhb

[Pierre Ricadat @ghostdogpr]

@kubukoz Thanks. JS is slightly more annoying because of missing crypto APIs, I might try Native first hoping it doesn’t destroy CI times.

[Pierre Ricadat @ghostdogpr]

Anyone interested in having Caliban server work on Scala.js/Native?

[Pierre Ricadat @ghostdogpr]

@adamhearn “ALWAYS use braceless syntax” 😩

[Pierre Ricadat @ghostdogpr]

@fwbrasil I’d love to read about your whole AI usage workflow 😄

[Pierre Ricadat @ghostdogpr]

Been a while since I spent time on Shardcake, but I am cooking some nice improvements.

[Pierre Ricadat @ghostdogpr]

Released as part of Proteus 0.4.0: https://t.co/qvifT3QGY2

[Pierre Ricadat @ghostdogpr]

Reimplemented the zio/fs2 grpc backends in Proteus without depending on zio-grpc/fs2-grpc. That gave a nice performance boost to the zio one!

Ox is still the fastest but not by far. I'm hoping to use kyo-compat later if it gets support for streams.

https://t.co/Q5KQzba81D

[Pierre Ricadat @ghostdogpr]

@hepin1989 For now Proteus grpc implementation is tied with grpc-java. Pekko and Akka are faster because they don't rely on it.

[Pierre Ricadat @ghostdogpr]

Reimplemented the zio/fs2 grpc backends in Proteus without depending on zio-grpc/fs2-grpc. That gave a nice performance boost to the zio one!

Ox is still the fastest but not by far. I'm hoping to use kyo-compat later if it gets support for streams.

https://t.co/Q5KQzba81D

[Pierre Ricadat @ghostdogpr]

Video from my talk at Scalar 2026 is online 👀

https://t.co/UMPLPOF7LT

[Pierre Ricadat @ghostdogpr]

@jdegoes can you do something to remove the bounties on zio? look at all the PRs made in the last 24h, it's going to make the few remaining maintainers run away 😅 if we can't delete bounties, we could maybe close these issues and recreate them

[Pierre Ricadat @ghostdogpr]

@Krever01 @adamwarski I think I’m gonna try to split the code to put as much as possible in a pure core then make a CE runtime on top with all the concurrency. Hopefully that layer won’t be too large so that other runtimes could be added later if needed.

[Pierre Ricadat @ghostdogpr]

Trying to build a new lib with support for ZIO/CE/Ox by abstracting over the effect type but it's proving difficult as I need fork, queues, resources, promises, etc. Abstracting over Ox is also tricky due to its eager nature (sometimes needs () => F[A]).

Should I: (1/2)

[Pierre Ricadat @ghostdogpr]

@fwbrasil @Krever01 @adamwarski Yeah, so I feel I’m reinventing cats-effect by adding all these abstractions, and some don’t translate very well to Ox. Which is why I’m thinking of changing the approach completely and keep the core fixed to a single one.

[Pierre Ricadat @ghostdogpr]

@fwbrasil @Krever01 @adamwarski But stuff like Scope, Queue, Promise, fork all require Kyo Sync/Async, no? If I make them abstract, it is similar in complexity to what I have currently.

[Pierre Ricadat @ghostdogpr]

@Krever01 this is similar to questions you had recently, I think you went for CE core.
@adamwarski you managed this quite incredibly in sttp/tapir, curious if you have a take on this.

[Pierre Ricadat @ghostdogpr]

- core using Java concurrency primitives + integrations on the edges (using unsafe run for effects)
- CE core + drop Ox (ZIO ok via interop)
- ZIO core + drop the other 2
- effect-free core and duplicate all effectful parts x3
- keep going despite the code complexity

[Pierre Ricadat @ghostdogpr]

@adamhearn Does it work with /btw?