Add a lint error to detect two names that differ only in case

[MoonbaseOtago]

Some synthesis/placement/routing tools (cough openroad cough) get confused when they see two nets whose names differ only in case - for example 'din' and 'DIN' - this can result in LVS errors that are particularly hard to track down.

System flows like librelane use verilator lint for sanity checking at the beginning of their flows that end with openroad.

Variable names that differ by case are separate variables in verilog and not an error.

I'd like to suggest that verilator lint detect these cases and by default issue a warning (something like "other tools may be confused by this"), but allow a command line option that can convert it to an error so that something like librelane can enable it as an error because they know it's a potential minefield elsewhere

thanks - Paul

[wsnyder]

Might you contribute a PR to add this style warning?

Any warning can be turned off or made an error, no code needed for that.

[MoonbaseOtago]

I'll look into it

[MoonbaseOtago]

I created PR:

#8020

I haven't worked inside verilator before, please review it aggressively :-) (it's not long)

[MoonbaseOtago]

Seems to be causing a bunch of actions to fail, all seem to be because of actual use of mixed-case variable (one is actually a test that mixed case variables are separate which is OK, the warning should be tolerated, or turned off for that test)

[MoonbaseOtago]

BTW I've been using this test just as a simple sanity check

module Z(input m, input M, output N, output n, input p, output P, input q, input r, output s, output t);

    wire a;
    wire A;

    reg b;
    wire B;

    wire c;
    reg C;

    wire D;
    reg d;

    wire e;
    wire [7:0]E;

    wire Q, R, S, T;

    wire AaBbCc;
    wire aAbBcC;

endmodule

[wsnyder]

Thanks for your first pull. For the failures please edit to cleanup the tests (other than UVM) rather than suppressing the warning. I will review shortly.

[MoonbaseOtago]

What's an appropriate way to fix this? just add a "SIMILARNAMES lint_off" in the failing tests?