Replace @free with @tagless, and drop the requirement of an annotation #92
Conversation
tbrown1979
changed the title
tbrown1979
changed the title
| case cls: Trait => | ||
| defaultServiceExtras(cls.name, cls) | ||
| case _ => | ||
| abort(s"$invalid. $abstractOnly") |
There was a problem hiding this comment.
Maybe a better error here since @free isn't required anymore
|
|
||
| val (algName, template) = defn match { | ||
| lazy val (algName, template) = defn match { |
There was a problem hiding this comment.
I don't think these all need to be lazy.
| val call: Term.Tuple = streamingType match { | ||
| case Some(RequestStreaming) => | ||
| q""" | ||
| def call(callType: Option[String]): Term.Tuple = { |
There was a problem hiding this comment.
Passing an Option[String] for the callType is kind of a weird way to determine which call to use - unaryMethod/unaryMethodFree.
|
|
||
| case class ServiceAlg(defn: Defn) extends RPCService { | ||
| val typeParam: Type.Param = defn match { | ||
| case c: Class => c.tparams.head |
Codecov Report
@@ Coverage Diff @@
## master #92 +/- ##
=========================================
- Coverage 85.71% 85.6% -0.12%
=========================================
Files 12 12
Lines 126 125 -1
=========================================
- Hits 108 107 -1
Misses 18 18
Continue to review full report at Codecov.
|
tbrown1979
changed the title
|
This is great. I'll look at it tomorrow, thanks @tbrown1979! |
There was a problem hiding this comment.
Excellent work @tbrown1979. This is a great step forward. RPC services didn't need to be encapsulated in @free because the Handlers they provide can already be implemented in terms of Free programs if the target M[_] is a FreeS[Op, ?] so the extra level of indirection was not buying us anything in the protocol algebra.
This PR drops the
@freeannotation and replaces it with@tagless. This PR also removes the requirement that a@servicebe annotated with@tagless. You are free to define it with your own parameterized type if you want.This resulted in some changes to the generated code because we're no longer converting from
Op ~> Manymore.