Case expression with for loop variable

[Blebowski]

I ran across an interesting issue. NVC does not seem to determine the range of iteration variable i correctly.

Consider following example:

entity iterator_bounds is
end entity iterator_bounds;

architecture test of iterator_bounds is

begin

    process
        variable higher_bound : integer := 3;
    begin
        for i in 1 to higher_bound loop
            case i is
            when 1 => report "'i' is 1";
            when 2 => report "'i' is 2";
            when 3 => report "'i' is 3";
            end case;
        end loop;

        wait;
    end process;

end architecture;

i can only get values from 1 to 3. However, NVC expects full integer range:

 nvc -a iterator_bounds.vhd -e iterator_bounds -r
** Error: missing choice for element -2147483648 to 0, 4 to 2147483647 of type INTEGER
    > iterator_bounds.vhd:12
    |
 12 |             case i is
    |             ^

If I replace the higher_bound with literal constant 3, then no error occurs.
I guess NVC determines range at compile time in such case.

Obviously, adding others choice solves the issue, but I am not sure it should be needed.
In GHDL this "works" without complaining. If I bump-up the value of higher_bound to 4
without adding corresponding case choice, GHDL accepts it too, and over-flows and when
i=4, it prints the report for i=1 (which seems to me a GHDL bug).

The code works OK also in VCS. In Riviera PRO, I get:

Case expression is not of a locally static subtype. Use the choice 'others' or compile with -relax." "iterator_bounds.vhd"

If I recompile with -relax Riviera PRO takes the code. If I bump-up the higher_bound value to 4
without adding corresponding case alternative, it takes the code too, but does not print anything
in the iteration when i=4.

I am not sure what does the LRM says on this topic, so this issue might be rubbish, but I would
guess it is valid since GHDL, VCS and Riviera accept the code.

[nickg]

I'm pretty sure this error is correct. The subtype is not static (as the RiveraPRO message says), so the choices need to cover the full range since the value of higher_bound could change.

Questa reports basically the same error for this code:

Questa Intel Starter FPGA Edition-64 vcom 2021.2 Compiler 2021.04 Apr 14 2021
Start time: 23:04:42 on Jun 25,2024
vcom test.vhd 
-- Loading package STANDARD
-- Compiling entity iterator_bounds
-- Compiling architecture test of iterator_bounds
** Error (suppressible): test.vhd(12): (vcom-1339) Case statement choices cover only 3 out of 4294967296 cases.
** Note: test.vhd(22): VHDL Compiler exiting
End time: 23:04:42 on Jun 25,2024, Elapsed time: 0:00:00
Errors: 1, Warnings: 0

Because higher_bound is a variable its value could change e.g.

    process
        variable higher_bound : integer := 3;
    begin
        for i in 1 to higher_bound loop
            case i is
            when 1 => report "'i' is 1";
            when 2 => report "'i' is 2";
            when 3 => report "'i' is 3";
            end case;
        end loop;
        higher_bound := higher_bound + 1;
        wait;
    end process;

GHDL doesn't error on this either which is definitely wrong.

If you make higher_bound a constant then NVC and Questa accept it without errors.

[Blebowski]

OK, I see. Thanks for explanation.

Do you think it would be good to accept the behavior with -relax similar to Riviera ?
If not, feel free to close the issue.

[nickg]

--relaxed turns it into a warning now.