Fuzzer: Segfault on genvar out of scope

[veripoolbot]

---

Author Name: Eric Rippey
Original Redmine Issue: 1589 from https://www.veripool.org

Original Assignee: Wilson Snyder (@wsnyder)

---

Running the attached testcase with:

verilator_bin --lint-only 7.sv

On version:

Verilator 4.020 devel rev v4.020-56-gbcb766b

Produces:

%Error: 7.sv:16: Unsupported: Member call on non-enum object 'VARREF 'j'' which is a 'BASICDTYPE 'integer''
                : ... In instance t
  j.e(0),
     ^
Segmentation fault (core dumped)

I think there are two different problems here:

1. Shouldn't the "j" be inaccessible at all because it should only exist in the scope of the generate block in which it's defined?
2. Even if "j" is switched to be declared within the module as "integer j;" the segfault still persists.

[veripoolbot]

---

Original Redmine Comment
Author Name: Wilson Snyder (@wsnyder)
Original Date: 2019-11-05T02:16:18Z

---

Generate dosn't make a scope, begin/end does. Basically generate was a Verilog 2001 thing now not needed.

Cleaned up warnings & real issue.

Fixed in git towards 4.022.

[veripoolbot]

---

Original Redmine Comment
Author Name: Eric Rippey
Original Date: 2019-11-06T00:35:26Z

---

Regarding the scope: that's interesting. I think you're right but it took me longer to confirm that than I expected. I didn't find anything definitive either way in the standard, and so I tried running:

module a;
  initial $display("a=",a);
  generate
     genvar a;
  endgenerate
endmodule

with ModelSim and it choked because it didn't want to let the genvar be accessible there. Then I tried iverilog and it choked because it couldn't handle printing a genvar at all. And then I tried:

module a;
  initial $display("a=",a);
  generate
     integer a;
  endgenerate
endmodule

And they both accepted it. Meanwhile, Verilator seems to treat these two cases the same, which seems like the correct thing to do.