unconstrained others error not covered by --relaxed option

[weilanad]

Description
When assigning an others value to an unconstrained std_logic_vector the compiler throws an error. By using the --relaxed option this error should be suppressed.
With GHDL this error is suppressed by using -frelaxed.

How to reproduce?
In the following a minimal working example to reproduce

library IEEE;
use     IEEE.STD_LOGIC_1164.all;

entity module is
  port (
    vec : out std_logic_vector
  );
end entity;

architecture rtl of module is
begin
  vec <= (others => '1');
end architecture;

The following commands have been used for execution:

nvc --std=08 -a --relaxed test_unconstrained_others.vhdl   # NVC
ghdl -a --std=08 -frelaxed test_unconstrained_others.vhdl  # GHDL

Context

• OS: Windows + MSYS2/UCRT64
• Origin:
  • Built from sources: nvc 1.15.2 (1.15.2.r0.g2fe12d4f)

/cc @Paebbels

[bpadalino]

@weilanad Not to say it's more correct, but ...

vec <= (vec'range => '1');

... works with NVC without the relaxed rules.

For unconstrained inputs, maybe this should be the recommended coding style?

[Paebbels]

Is this a missing definition in the LRM? Because GHDL also has it behind -frelaxed, whereas Vivado and Riviera-PRO accept it as-is.

[nickg]

In 9.3 it says

The index range of an array aggregate that has an others choice shall be determinable from the context

and then

e) As the value expression in an assignment statement, where the target is a declared object (or member thereof), and either the subtype of the target is a fully constrained array subtype or the target is a slice name.

But the target here is neither fully constrained nor a slice name so none of the listed contexts apply here and the others choice is illegal. See also #641.

[Paebbels]

Is it possible, when unconstrained ports were allowed in VHDL, they forgot to modify this part?

[nickg]

I think it would be reasonable to allow this in the same contexts where the 2019 "function knows vector size" feature works, and only if the others choice is the only choice.

BTW I noticed GHDL has an odd behaviour where it accepts the example above with --std=08 -frelaxed but not -frelaxed --std=08.

$ ghdl -a --std=08 -frelaxed test.vhd
$ ghdl -a -frelaxed --std=08 test.vhd
test.vhd:12:10:error: 'others' choice not allowed for an aggregate in this context
  vec <= (others => '1');
         ^
ghdl:error: compilation error
$ ghdl --version
GHDL 5.0.1 (4.1.0.r602.g37ad91899) [Dunoon edition]

[Paebbels]

Are you going to report it?

[nickg]

Are you going to report it?

I'll leave it to you. Possibly it's intended that --std overrides -frelaxed (resets language settings?).

[Paebbels]

@tgingold can you please have a look at #1175 (comment)
Is this intended behavior?

[tgingold]

@Paebbels Yes --std=XX reset (or set) -frelaxed.

[Paebbels]

@nickg I found it also in GHDL's documentation: https://ghdl.github.io/ghdl/using/InvokingGHDL.html#cmdoption-ghdl-std

---

I should have known that fact, but maybe knowledge like this get's lost once some scripts are written (like OSVVM-Scripts) ;).

---

/cc @JimLewis

[JimLewis]

Certainly at elaboration time the port is fully constrained. Or is there a requirement that it is fully constrained at compile time.