missing predefined functions returns universal_real

[hiyuh]

VHDL-1993 has following predefined functions returns universal_real, but they are missing ATM.

-- function "*" (anonymous: universal_real; anonymous: universal_integer)
-- return universal_real;
-- function "*" (anonymous: universal_integer; anonymous: universal_real)
-- return universal_real;
-- function "/" (anonymous: universal_real; anonymous: universal_integer)
-- return universal_real;

w/ proposed issueY,

entity issueY is
begin
    assert (1.0 * 1 = 1.0);
    assert (1 * 1.0 = 1.0);
    assert (1.0 / 1 = 1.0);
end entity issueY;

architecture a of issueY is
begin
end architecture a;

then i got,

$ NVC_LIBPATH=./lib/std:./lib/ieee:./lib/nvc NVC_CYG_LIB=./src ./src/nvc.exe -a test/regress/issueY.vhd
** Error: no suitable overload for operator "*"(universal real, universal
          integer) return REAL
        File test/regress/issueY.vhd, Line 3
     assert (1.0 * 1 = 1.0);
             ^^^^^^^
** Error: no suitable overload for operator "*"(universal integer, universal
          real) return REAL
        File test/regress/issueY.vhd, Line 4
     assert (1 * 1.0 = 1.0);
             ^^^^^^^
** Error: no suitable overload for operator "/"(universal real, universal
          integer) return REAL
        File test/regress/issueY.vhd, Line 5
     assert (1.0 / 1 = 1.0);
             ^^^^^^^

[hiyuh]

it just works.
could you mind to check my issue46 branch?

[nickg]

The changes above look technically fine, but the problem is they allow things like:

variable r : real;
r := r * 5;

Section 7.5 in the LRM is a bit awkward in that those operators apply only in universal expressions. At the moment NVC has no way of distinguishing universal operators from regular ones, so I've attempted to add this in commit b7dc9bc.

There's now a function sem_declare_universal which declares them and a special case in the overload resolution function that ensures universal parameter types only match arguments with universal integer/real types.

There isn't any code generation for these operators as they should always be constant-folded.

[hiyuh]

ACK, thanks your precise correction!