Principled code generation from OpenAPI specifications. https://guardrail.dev/
you know how I always aim to please, Shanan, I'm sorry to let you down here.
oh yeah, we've definitely just been excluding it in other systems
<arg>-Xlint:-unused,_</arg>
The -Wconf:src=... option would be preferred, as it would only exclude unused in generated sources
what guardrail may be able to do is automatically add an explicit
-Wconf:src, since we actually know exactly which directories we generated into
that would certainly improve usability
certainly. I can't look at this right now, as my time is allocated towards the modularization proposal, ramping up to the 1.x release, but if you want to test swapping that
-Xlint:-unused,_ for -Wconf:src:..., that would be a step in the right direction
oh, please disregard then, I didn't realize it was so new
it may not end up being worth it at all anyway, as a lot of warnings were removed in dotty (much to my displeasure)
it'll be interesting to see how things shake out as people start moving production systems to dotty
I should start just saying 3.0
There are only two other forks of Scala I know of, the typelevel fork as well as a paid super optimized one with some kind of other really cool features
the fragmentation all comes from how many different major/minor revisions people are using 😕
kelnos: Barring any major objections to guardrail-dev/guardrail#1197 I think I'll merge as-is -- I was thinking it would make sense to also re-package definitions and modules and whatnot, but that'll be a real PITA for folks who maintain their own forks of guardrail to do all at once.
Once that's in, I'll cut some smaller PRs to move individual classes to raise visibility and make each individual changeset tractable, instead of doing 52 card pickup with the package structure.
The nice thing about wrapping this splitout up with
guardrail_${scala.binary.version} artifact as root is that it's a no-op for the build tools at the moment, we can gradually implement the tooling improvements when there's time
once all the pieces are in place, I think the last big thing is to remove the
tracing stuff, as akka-http, http4s, dropwizard, and spring-mvc all have a better way of doing that, either by mapRoute or by mapping over the entire server itself, and for the Java servers my understanding is there's deep introspection modules that can be registered to provide the route-level tracing required
if I'm wrong about that I'm happy to revise, but I don't believe the tracing stuff has ever been broadly consumable other than the bespoke stuff in Twilio's scala-service overrides.
also, hah, it has been a while since trying to use metals from vim
Understood, just wanted to make sure you were broadly OK with the direction
agree on the tracing stuff, i think it probably should have been out of scope for guardrail in the beginning, but yeah, was mostly there for the twilio bits (that we don't use anymore anyway)
might want to write up a migration guide, though, in case anyone's using it
Yeah, as well as a class/package remapping list as things progress in that direction.
The splitout kind of lends itself towards moving stuff like ClientGenerator and ServerGenerator over in dev.guardrail.generators next to ProtocolGenerator, as well as to clean up all the cross-references between those things (like how akka-http uses the response generators from http4s)
i think i would go for something like
dev.guardrail.generators.{protocol, server, client} with perhaps also a helpers package
What do you think about
guardrail-core being reflected in dev.guardrail.core.{...}?
or would
dev.guardrail.{...} be the de-facto core?
hrm, i don't think i see the need for a
core package, but i'm not really against it either
yeah, that's where I am. We already have some dev.guardrail.core package members
so it's a weird position to be in.
I'll sort it though, no worries.
blast_hardcheese: regarding my PR comment about the slf4j-simple good. what do you think about separating out the CLI into its own module (which can then depend on sfl4j-simple)?
s/good/goo/
Wrestling with that now, actually, as part of the repackaging effort
I'll keep that as an option. I like it because it means the build tools don't need to contend with the CLI at all, as far as name collisions or anything
the downside is that the gradle plugin "uses" guardrail by just constructing CLI args and running the CLI directly 🥴
(or in the meantime the gradle plugin can just depend on the cli artifact)
I really like the fact that coursier is able to just dynamically link and emit a standalone CLI artifact, btw. No need to publish the assembly like I used to do