[vhdl-200x] LRM bugs

During the development of ghdl (http://ghdl.free.fr), I think I have found
severals errors in the VHDL LRM. Here is a priliminary list with some
propositions to fix the bugs.
I hope these bugs could be clarified or fixed for the next LRM version.

1) The return type of a function can be a file type.
   However, it is of course not possible to write a correct body for such
   a declaration.

   Proposition: The return type of a function must denote neither a file type
   nor a protected type.

2) An alias of a physical type is not possible, although described in the LRM.
   Indeed, a physical type is always an anonymous type. Only physical
   subtypes can be aliased.

   Proposition: Are non-object aliases really used ? Non-object aliases were
   added to be able to re-export declarations, but they cannot be used for
   this purpose.

3) Character literal can be decorated through an attribute specification,
   but the 'decoration' cannot be referenced (because the prefix of an
   attribute cannot be a character).

4) It is possible to write pure functions that do not return the same value:
   type string_acc is access string;

   pure function alloc (str : string) return string_acc is
   begin
      return new string'(str);
   end alloc;

   Proposition: forbid allocators in pure functions (directly or indirectly)
   as well as calls to the implicit deallocate procedures.

5) Some trivial expressions are not valid:
      constant c : integer := -3;
   The expression is ambiguous, since the "-" can either that of the integer
   type or that of the universal integer type.
   For sure, in this example the result is the same, but what about:
     function "-" (v : integer) return integer is begin return 0; end;
     constant c : integer := -3;

   Proposition: Do like Ada: preference for universal operators.

6) Does a generate statement have a declarative part ?
   No, according to grammar from 9.7, but some of the block_declarative_item
   are specifications which are valid only inside declarative part...

   Propostion: Replace "{ block_declarative_item }" with
   "block_declarative_part".

7) What is the meaning of:
   type t is array (boolean) of bit;
   signal s : t;
   ...
   wait on s (time < 10 ns);

   Suppose the wait statement starts to execute at 9 ns, and there is an
   event at 11 ns on s (true). Does the wait statement finish ?
   This bug is due to the fact that globally static expression are not static.

   Proposition: why the time function is pure ?

8) What is the meaning of 'signature' in 'INSTANCE_NAME and 'PATH_NAME ?
   What is the signature of function f return natural ? [return natural] or
   [return integer]. At least, the use of these attributes is nonportable.

9) The default entity aspect is ambiguous, because rules a) and c) or b) and c)
   of 5.2.2 can results in two visible entity declaration.
   Proposition: Remove rules a) and b), that are neither intuitive
   nor understandable.

10) Constant parameter of access types are implicitly allowed:
      type line is access string;
    seems to implicitly define:
      function "=" (constant <anon> : line; constant <anon> : line)
        return boolean;
    These functions (if they really exist) are of course not overloadable.
    The restriction (constant parameter cannot be of access type) was not
    well considered (remember the endline function of std.textio).
    Although this is not explained, I suppose the reason why this restriction
    exists is because it prevents from defining pure function that returns
    different values when called with the same parameters.

    Proposition:
     1) allow constant parameter of access types for procedures and impure
        functions.
     2) Allow constant parameter of access types for pure functions but forbid
        to dereference such parameters in pure functions.

Extra bugs:
11a) There is a missing ';' in example of E'PATH_NAME after the generic clause
    of component BComp declared inside architecture top of top.

11b) There is a missing ':' before gbottom'path_name.

12) Names of protected type examples in 3.5.1 and 3.5.2 do not match:
  VariableSizedBitArray vs VariableSizeBitArray.

Tristan.
Received on Sun Nov 28 10:15:02 2004