More stringy names in Clash.GHC.Evaluator.Primitive

[bgamari]

Hard-coded stringy names are problematic, especially in GHC 9.10.1 due to the ghc-internal rewrite. Some occurrences were handled in e565d26, but there are quite a few more in Clash.GHC.Evaluator.Primitive (c.f. GHC.Float.integerToFloat# and many more).

[bgamari]

There also appear to be many more such cases in the blackboxes in clash-lib/prims/common.

[DigitalBrains1]

Maybe we can replace the YAML files in clash-lib/prims with something like

import GHC.Classes (gtInt, geInt, ltInt, leInt)

let
  gtIntName = 'gtInt
  geIntName = 'geInt
  ltIntName = 'ltInt
  leIntName = 'leInt
in InlineYamlPrimitive [minBound..] [__i|
     - BlackBox:
         name: #{gtIntName}
         kind: Expression
         type: 'gtInt :: Int ->
           Int -> Bool'
         template: ~ARG[0] > ~ARG[1]
     - BlackBox:
         name: #{geIntName}
         kind: Expression
         type: 'geInt :: Int ->
           Int -> Bool'
         template: ~ARG[0] >= ~ARG[1]
     - BlackBox:
         name: #{ltIntName}
         kind: Expression
         type: 'ltInt :: Int ->
           Int -> Bool'
         template: ~ARG[0] < ~ARG[1]
     - BlackBox:
         name: #{leIntName}
         kind: Expression
         type: 'leInt :: Int ->
           Int -> Bool'
         template: ~ARG[0] <= ~ARG[1]
     |])

somewhere as part of clash-lib Haskell files. This a sketch, first of all I don't know if InlineYamlPrimitive accepts a list of BlackBox and secondly it's still not tied in anywhere anyway, but as a general idea.

Also, we might define an ADT somewhat like

data BlackBox =
  { name :: Language.Haskell.TH.Syntax.Name
  , kind :: ...
  , type :: Text
  , template :: Text
  }

where we can immediately pass a Name. I think this would be a nice way to define primitives in user designs as well, but it might be significantly more effort (not my area of expertise, that part of the code).