[bug][slang] Fix case selector huge bitwidth segmentation fault

[likeamahoney]

Hi @MikePopoloski!

There are a two examples of casex which works incorrect with slang:

// This one is segfaults on release and debug builds

module test9(
  input reg [100:0] sel_i,
  input reg trg_i,
  output reg [1:0] reg_o
);
  always @(posedge trg_i) begin
    casex (sel_i)
      -12'b00zzzzz00001,
      12'b11: reg_o = 2'b01;
    endcase
  end
endmodule

and

// This one produce the strange error on debug build:
// internal compiler error: Assertion 'noteLocation' failed
// in file /home/yan/projects/slang/source/diagnostics/Diagnostics.cpp, line 30
//  function: slang::Diagnostic& slang::Diagnostic::addNote(slang::DiagCode, slang::SourceLocation)

module test9(
  input reg [100:0] sel_i,
  input reg trg_i,
  output reg [1:0] reg_o
);
  always @(posedge trg_i) begin
    casex (sel_i)
      12'sb00xxxxx00001,
      -12'b00zzzzz00001,
      12'b11: reg_o = 2'b01;
    endcase
  end
endmodule

I think the whole point is that it is dangerous to use unions (for example - here), especially since it is impossible to check which type was actually written to the tree node. This can also happen because of different bit lengths of labels (in one case - for the first label value.getBitWidth() returns 12 (leading zeros in front), and in the other 101).

[MikePopoloski]

Thanks for the report, but this is not the right fix. The constant value should all be sized the same. There is an invariant being broken where the constant value of the expression is not sized to the width of the expression's type, due to some of the other warning analysis stuff I've added recently. #1222 is related.

[MikePopoloski]

Superseded by e9107cf