Issue: MACRO-SUBFORMS-TOP-LEVEL-PReferences: EVAL-WHEN, COMPILE-FILE,
all macros defined in the standard
Related issues: EVAL-WHEN-NON-TOP-LEVEL
Category: CLARIFICATION
Edit history: v1, 13 Feb 1991, Sandra Loosemore
v2, 11 Mar 1991, Sandra Loosemore
Problem description:
The compilation model defined by issue EVAL-WHEN-NON-TOP-LEVEL specifies
that expansions of macros inherit "top-level-ness" from the macro call.
This makes sense for user-defined macros, but users have no control
over what the expansions of macros defined in the standard are, since
they're provided by the Lisp implementation. The semantics of a macro
call form can depend on whether or not its subforms appear at top-level
in the expansion, so the standard ought to place some constraints
on what the standard macros can expand into.
Proposal (MACRO-SUBFORMS-TOP-LEVEL-P:ADD-CONSTRAINTS):
Clarify that no subforms of calls to macros defined in the standard
inherit "top-level-ness" from the macro call unless explicitly stated
otherwise.
Clarify that no macros now in the standard pass on "top-level-ness"
to their subforms.
Clarify that even if a macro is normally treated as a special form by
the file compiler, both its handling as a special form and its normal
macro expansion must still satisfy these constraints.
Clarify that compiler-macro functions must ensure that expansions
have the same semantics regarding inheritance of top-level-ness
as the normal function or macro definition of the form.
Rationale:
It fixes the problem.
Current Practice:
Probably many implementations currently return macro expansions that
violate these rules.
Cost to Implementors:
Implementors will have to examine the definitions of all macros they
provide to make sure that they conform to these rules. Some
"optimizations" currently performed may turn out to be invalid;
for example, (and <form>)
could expand into (let () <form>)
but not (locally <form>)
or (progn <form>)
or <form>
Cost to Users:
Probably none.
Cost of non-adoption:
User programs will behave differently in different implementations.
Performance impact:
Probably small.
Benefits:
The cost of non-adoption is avoided.
Esthetics:
The whole business of "top-level-ness" is kind of ugly.
Discussion:
It doesn't appear that any of the ~90 macros defined in the standard
require special exceptions from the rule. Certainly none of the
binding, conditional, or iteration macros ought to pass "top-level-ness"
through.