prettyBlackBox: Correctly escape [ and ]

[bgamari]

The pretty-printer previously failed to round-trip, resulting in invalid primitives due to the logic in Clash.Netlist.BlackBox.mkFunInput.

I have also included some clean-ups and error message improvements that I found useful while debugging this.

Fixes #2988.
Fixes #2809.

TODO

• Add changelog entry
• Add test

[bgamari]

Sadly it appears that we now emit [\...\] in the produced Verilog, e.g.:

  // map begin
  genvar i;
  generate
  for (i=0; i < 3; i = i + 1) begin : map
    wire [43:0] map_in;
    assign map_in = a1[i*44+:44];
    wire [31:0] map_out;
    assign map_out = map_in[\0+:32\];


    assign c$ds2_app_arg_0[i*32+:32] = map_out;
  end
  endgenerate
  // map end

Unfortunately this doesn't manifest in the minimal reproducer above.

[bgamari]

This program is sufficient to reproduce the bad behavior:

{-# LANGUAGE DataKinds #-}

module Hi where

import Clash.Prelude
import Prelude ()

topEntity
    :: Signal System (Vec 4 (BitVector 40))
    -> Signal System (Vec 4 (BitVector 32))
topEntity = f

f
    :: Signal System (Vec n (BitVector 40))
    -> Signal System (Vec n (BitVector 32))
f x = fmap (fmap truncateB) x
{-# OPAQUE f #-}

{-# ANN topEntity
  (Synthesize
    { t_name   = "repro"
    , t_inputs = [ PortName "inp" ]
    , t_output = PortName "outp"
    }) #-}

Resulting in:

module Hi_topEntity_f
    ( // Inputs
      input wire [159:0] x
      // Outputs
    , output wire [127:0] result
    );

  // map begin
  genvar i;
  generate
  for (i=0; i < 4; i = i + 1) begin : map
    wire [39:0] map_in;
    assign map_in = x[i*40+:40];
    wire [31:0] map_out;
    assign map_out = map_in[\0+:32\];
    assign result[i*32+:32] = map_out;
  end
  endgenerate
  // map end
endmodule

[bgamari]

@christiaanb it would be great to hear your thoughts on the intended behavior here. I'm struggling to see how mkFunInput fits together and I'm still struggling to determine how the current patch is resulting in the errant escaping leaking into HDL output.

[bgamari]

I have shamelessly stolen (with a bit of cleanup) the approach taken in https://github.com/clash-lang/clash-compiler/tree/fix-map-head. This indeed appears to solve the issue, as demonstrated by the added tests.

[martijnbastiaan]

@kloonbot run_ci 88e6bb7

[martijnbastiaan]

Thanks @bgamari, this LGTM. I'll add a changelog entry later and merge.

[martijnbastiaan]

@kloonbot run_ci f2b20fd

I've added a changelog and a regression test.