NO_TRANSLATION_FOR

[kleinreact]

The following topEntity

topEntity :: Unsigned 8 -> Unsigned 8
topEntity = fromInteger . toInteger . g . fromInteger . toInteger
 where
  g :: Natural -> Natural
  g x = x + 1

produces

** GENERATED BY CLASH 1.9.0. DO NOT MODIFY.
*/
`default_nettype none
`timescale 100fs/100fs
module topEntity
    ( // Inputs
      input wire [7:0] x

      // Outputs
    , output wire [7:0] result
    );
  wire signed [63:0] c$app_arg;
  wire signed [63:0] c$app_arg_0;

  assign c$app_arg = NO_TRANSLATION_FOR:GHC.Num.Integer.integerFromNatural;

  assign c$app_arg_0 = $unsigned({{(64-8) {1'b0}},x});

  assign result = $unsigned(c$app_arg[0+:8]);

endmodule

and a similar NO_TRANSLATION_FOR, if targeting VHDL.

I would expect a dubious primitive warning here, as for Integer, or an error, but not Clash silently producing invalid HDL.

[christiaanb]

Ah, this is an unfortunate interaction between

1. a fall-through case I added a long time ago for netlist-generation not breaking on arguments that would be unused: 0c6c63d#diff-edc71942970da8c1935b90eadabf16b3fe0e29dd87f42e261c5279c89cfbb539R195
2. A GHC 9 upgrade https://github.com/clash-lang/clash-compiler/blame/d9e2a53fe81eb23616d9fbd2e7944845c5c6f6c4/clash-ghc/src-ghc/Clash/GHC/Evaluator/Primitive.hs#L1892-L1898
3. A primitive being added that was then only tested when applied to a constant so we would get compile-time reduction 211f8f3#diff-f9de82a65e58cf9f5d5b4806db49fb6e31eb292b9adfb0538441ac502c673a04R10-R14

I guess we should add a BlackBox that produces error for a negative integer, and bitcoerces otherwise.