Haskellator Currently Failing to Parse Amaranth Lang Output with process Statements

[ThePerfectComputer]

Here is an MVP of what causes Haskellator to fail:

attribute \generator "Amaranth"
attribute \src "/Users/yehowshuaimmanuel/git/Haskellator1/fsm.py:19"
attribute \top 1
module \top

  attribute \src "/Users/yehowshuaimmanuel/git/Haskellator1/fsm.py:21"
  process $26
    assign $11 [4:0] \ctr [4:0]
    switch $1 [0]
      case 1'1
        assign $11 [4:0] 5'10011
      case
        assign $11 [4:0] $2 [4:0]
    end
  end

end

[ThePerfectComputer]

The issues lies in the switch statement as A works but not B:

A

attribute \generator "Amaranth"
attribute \src "/Users/yehowshuaimmanuel/git/Haskellator1/fsm.py:19"
attribute \top 1
module \top

  attribute \src "/Users/yehowshuaimmanuel/git/Haskellator1/fsm.py:21"
  process $26
    assign $11 [4:0] \ctr [4:0]
  end

end

B

attribute \generator "Amaranth"
attribute \src "/Users/yehowshuaimmanuel/git/Haskellator1/fsm.py:19"
attribute \top 1
module \top

  attribute \src "/Users/yehowshuaimmanuel/git/Haskellator1/fsm.py:22"
  process $34
    assign \stb [0] 1'0
    switch $6 [0]
      case 1'1
        assign \stb [0] 1'1
      case
    end
  end

end

[ThePerfectComputer]

The minified example I gave above technically wasn't complete... I added back the full context to that example which was derived from this amaranth lang example. After adding back that full context, Haskellator failed to parse the emitted RTLIL; thus I am re-opening this issue.

This is also an indicator that we should expand the corpus with examples emitted directly from amaranth lang. Perhaps the correct approach would be to comb through the amaranth lang repo and emit RTLIL examples from it - as detailed out in issue #2.

[chathhorn]

The original example with the full context really appears to be invalid RTLIL according to the official documentation and grammar: "the body of a process consists of zero or more assignments, exactly one switch, and zero or more syncs."

<process-body> ::= <assign-stmt>* <switch>? <assign-stmt>* <sync>*

I could update the parser to accept this, but it doesn't seem like a very subtle difference and I wonder what's going on here.

Here's the original failing example for reference:

attribute \generator "Amaranth"
attribute \src "/Users/yehowshuaimmanuel/git/Haskellator1/fsm.py:19"
attribute \top 1
module \top

  attribute \src "/Users/yehowshuaimmanuel/git/Haskellator1/fsm.py:21"
  process $26
    assign $11 [4:0] \ctr [4:0]
    switch $1 [0]
      case 1'1
        assign $11 [4:0] 5'10011
      case
        assign $11 [4:0] $2 [4:0]
    end
    switch \fsm_state [2:0]
      case 3'000
        switch $3 [0]
          case 1'1
            assign $11 [4:0] 5'01010
        end
      case 3'001
      case 3'010
      case 3'011
      case 3'100
    end
    switch \rst [0]
      case 1'1
        assign $11 [4:0] 5'00000
    end
  end

end