Fix crash on package definition in interface decl.

[NikLeberg]

My own instance of a fuzzer (libFuzzer based, nvc-fuzz) found a crash.

If instead of the usual package <x> is new ... in an interface a full package definition package <x> is ... end package; was given, then parsing did crash.

With the added simple checks, the crash no longer happens and the parser gives the reasonable error of unexpected end while parsing interface package declaration, expecting new.

Note: The standard is a bit moot in the definition, but I think to understand it so that generic package in interfaces may only be defined in generic interfaces.

An interface package declaration declares an interface package that appears in a generic clause. [VHDL2019, Chapter 6.5.5]

But NVC allows them everywhere. E.g. the following is analyzing, elaborating and running without error:

package gen_pkg is
    generic (N : integer);
    constant c_pkg : integer := N;
end package;

entity ent is
    generic (
        package p0 is new work.gen_pkg generic map(<>);
    );
    port (
        package p1 is new work.gen_pkg generic map(<>); -- no error?
    );
end entity;

architecture arch of ent is
    constant c0 : integer := p0.c_pkg;
    constant c1 : integer := p1.c_pkg;
begin
end architecture;

Cheers

[nickg]

Note: The standard is a bit moot in the definition, but I think to understand it so that generic package in interfaces may only be defined in generic interfaces.

I think the relevant bit is in 6.5.6.1:

A generic interface list consists entirely of interface constant declarations, interface type declarations, interface subprogram declarations, and interface package declarations. A port interface list consists entirely of interface signal declarations. A parameter interface list may contain interface constant declarations, interface signal declarations, interface variable declarations, interface file declarations, or any combination thereof.

sem_check_param_decl is missing this check, I guess because any class was valid in 1993. E.g. sem_check_port_decl does

   ...
   else if (class != C_SIGNAL)
      sem_error(t, "invalid object class %s for port %s",
                class_str(class), istr(tree_ident(t)));

[nickg]

Suggested change

	if (tree_has_type(p0))
	type_add_param(type, tree_type(p0));
	if (class_has_type(tree_class(p0)))
	type_add_param(type, tree_type(p0));
	else
	type_add_param(type, type_new(T_NONE)); // Will raise error later

[nickg]

Maybe we can merge this into the loop below to avoid duplicating the check above. E.g.

if (i == 0 && tree_has_value(p))
   tree_set_flag(decl, TREE_F_CALL_NO_ARGS);

[NikLeberg]

Thanks for the review. Implemented the suggested change and merged the special case for the first port into the loop.