VASG Issue Number:      0154
Comment Author:         Paul Menchini
Comment Number:         PM01
Comment Date:           1992/08/12

Larry Groves, in I0154-LG01 (dated 2 June 1992), states that IR 0040 and
LCS-0007 both address the issue raised by IR 0154.  I disagree with this
conclusion.

The part of IR 0040 discussed in I0154-LG01 deals with the decoration of a
named entity declared within a primary unit from within a corresponding
secondary unit.  IR 0040 states that this should not be possible, as a logical
result of making all specifications declarative-part-based.  This decision is
reflected in P1076-1992/A v5.

I0154-LG01 goes on to state "So it is clear that IR 40 covers the issues raised
by IR 154, and I recommend that IR 154 be closed."  It is not clear to me that
this statement is true.  Consider the following (illegal) VHDL:

package body P is
   use Synth.IMPLEMENTATION_DETAILS.all;

  function F (...) return ... is
    ...
  begin
    ...
  end function F;

  attribute IMPLEMENTATION of F: function is "vendor-specific.TTL.74LS1742";
        -- this attribute specification is illegal, as attribute
        -- specifications cannot appear in package bodies

  ...

end package body P;

This seems like an eminently reasonable thing to do, a capability that
synthesis tools in particular, may wish to exploit.  However, this is not
possible either in 1076-1987 or P1076-1992/A v5.  If this were possible, it
would be true that no expressions outside of the package body can access this
attribute; however, tools that process the library unit may.  This is why I am
proposing this change.

Note that I am not proposing, either in IR 0154 or here, that attribute
specifications appearing in the package body be allowed to adorn named entities
in the package interface, the package, the package's interface, or the
package's body; attributes are still declarative-part based.  Further note that
attribute specifications are allowed in architecture declarations; these
specifications are allowed to adorn only named entities declared in the
architecture declarative part or the architecture name itself.  I am merely
proposing that a similar capability be extended to the named entities declared
in a package body declarative part.

As a final note, I0154-LG01 goes on to state "Another issue, not addressed
here, is the affect of LCS 7, which appears to reverse the above position, and
allows attribute specification (but not declaration) within a package body."  I
cannot find anything in LCS-0007(2), dated 6 February 1992.  However, the
history log entry for this date states (in part) "Changed mind on attributing
package bodies", so perhaps version 1 contained some proposal in this area.

In any case, if this is not addressed in 1076-1992, I would recommend that we
not close IR 0154, but keep it open for the ISAC to address.