VHDL Issue Number:      1124
Classification:         Language Definition Problem
Language Version:       VHDL-93
Summary:                Are all actuals evaluated in incremental binding?
Related Issues:         IR1096, 1081
Relevant LRM Sections:  5.2.1 
Key Words and Phrases:  Incremental binding,
Current Status:         ISAC-Approved
1076-1987 Disposition:  N/A 
Disposition Rationale:  N/A
Superseded By:          N/A
-----------------------
Date Submitted:         1997/01/22
Author of Submission:   Chuck Swart
Author's Affiliation:   Analogy
Author's Post Address:  9205 SW Gemini Drive
                        Beaverton, OR 97008
Author's Phone Number:  503-520-2741
Author's Fax Number:    503-643-3361
Author's Net Address:   cswart@analogy.com
-----------------------
Date Analyzed:          1997/07/22 
Author of Analysis:     Greg Peterson
                        Peter Ashenden
Revision Number:        $Revision: 1.1 $
Date Last Revised:      $Date: 1997/01/24 19:12:10 $


Description of Problem
----------------------
Consider the following code fragment constructed along the lines
of the incremental binding example in LRM section 5.2.1
(impurefcn is an impure function with no arguments which returns an
integer value):

ENTITY test1 IS
    GENERIC( g1,g2,g3: integer);
END ENTITY test1;

PACKAGE my_gates IS
    COMPONENT test1 IS
        GENERIC( g1,g2,g3: integer);
    END COMPONENT test1;
END PACKAGE my_gates;

ENTITY test2 IS
END ENTITY test2;

USE work.my_gates.ALL;
ARCHITECTURE structure OF test2 IS
    FOR L1:test1 USE
        ENTITY work.test1(behavior)                --primary binding indication
            GENERIC MAP (impurefcn,OPEN,impurefcn) --for instance L1
    ...
BEGIN
    L1: test1;
    ...
END ARCHITECTURE structure;

CONFIGURATION different OF test2 IS
  FOR structure
    FOR L1: test1                               --incremental binding
      GENERIC MAP (g2=>impurefcn,g3=>impurefcn) --of L1; rebinds its generics
...
END CONFIGURATION different.

Two questions appear to be unresolved in the LRM:
(1) How many calls to impurefcn are made? In particular are the
actual expressions related to the primary binding indication evaluated
even when these values are overridden by the incremental binding?
(2) In what order are the calls to impurefcn made?


Proposed Resolution
-------------------
TBD


VASG-ISAC Analysis & Rationale
------------------------------

While it might be tempting to prohibit use of impure functions in
actuals of binding indications, the language definition does not allow
that.  The normative part of the standard does not specify that the
actual should be static.  Furthermore, 7.4.2 Note 2 lines 664-667
explictly indicate that a generic constant may be initialized with an
expression that is neither globally nor locally static.


The answer to question (1) is that the actual expression for an
overriden generic association in a primary binding indication are not
evaluated.

5.2.1 lines 180-181 state that, for a formal generic that is rebound
by an incremental binding indication, "the association given in the
primary binding indication has no effect for the given instance."
While this makes it clear that the formal is not associated with the
actual in the primary binding indication, it does not state whether
the actual expression is evaluated.

To determine whether the expression is evaluated, it is necessary to
consider the equivalent block structure for the component instance and
the elaboration rules.  9.6.1 states that a component instantiation is
equivalent to a pair of nested block statements, the outer of which
corresponds to the component instance, and the inner of which
corresponds to the bound design entity.  9.6.1 lines 432-435 specify
that the generic map aspect of the inner block is the generic map
aspect of the binding indication that binds the component instance to
the design entity.  12.4.3 specifies that elaboration of a component
instantiation statement consists of elaboration of the equivalent
block statements.  12.4.1 specifies that elaboration of a block
statement includes elaborating the block header.  12.2 specifies that
elaboration of the block header includes elaborating the generic map
aspect.  12.2.2 specifies that elaborating the generic map aspect
involves (indirectly) evaluation of the actual expression in an
element association.

The implication of these rules is that the binding indication for the
equivalent block is formed first, and the actual expressions in the
binding indication are evaluated later.  An overridden association in
the primary binding indication has no effect by virtue of being
replaced with the association in the incremental binding indication.
Hence the overridden association does not appear in the final binding
indication for the component instance, and its actual part is never
evaluated.


The answer to question (2) is that the order is undefined.  The LRM
does not define the order of elaboration of elements in an association
list.

In the issue author's example, the generic map aspects of the primary
and incremental binding indications together form a generic map aspect
for the inner equivalent block, as follows:

        generic map ( g1 => impurefcn,
                      g2 => impurefcn,
                      g3 => impurefcn )

The order of evaluation of the function calls is undefined.


VASG-ISAC Recommendation for IEEE Std 1076-1993
-----------------------------------------------

No change.


VASG-ISAC Recommendation for Future Revisions
---------------------------------------------

No change.