#533 (locally (declare (optimize...))) is ignored

lisp error
closed-invalid
clisp (525)
5
2009-10-18
2009-09-27
No

(locally (declare (optimize ,,,))) has no effect on compile called inside locally - only the global optimize declarations are honored.

Discussion

  • Sam Steingold

    Sam Steingold - 2009-09-27

    not quite.
    (locally (declare (optimize (safety X))) (disassemble (lambda (a) (aref a 0) 1)))
    and
    > (locally (declare (optimize (safety 3))) (funcall (compile nil (lambda(a) (aref a 0) 1)) 3))
    *** - AREF: argument 3 is not an array
    > (locally (declare (optimize (safety 0))) (funcall (compile nil (lambda(a) (aref a 0) 1)) 3))
    1

    works fine, but

    (defparameter f (lambda (a) (aref a 0) 1))
    (disassemble (locally (declare (optimize (safety X))) (compile nil f)))

    prints the same regardless of X

     
  • Sam Steingold

    Sam Steingold - 2009-10-15
    • milestone: 100333 --> lisp error
     
  • Bruno Haible

    Bruno Haible - 2009-10-17

    It is not a bug, but rather the expected behaviour.

    Please remember to distinguish "scope" and "extent".
    http://www.cs.cmu.edu/Groups/AI/html/cltl/clm/node43.html

    The LOCALLY special form specifies declarations for code that is
    lexically contained in this form. When you write (COMPILE NIL 'F)
    the definition of F is not contained in the LOCALLY form;
    therefore the declarations in effect at the COMPILE form are
    irrelevant.

    In other words, what matters is the declarations in effect at the
    *definition* of F.

    What you appear to want is some mechanism that establishes some
    optimization settings for the code being compiled by the COMPILE
    function, with dynamic extent. Such a mechanism cannot exist as an
    override because, as mentioned above, what matters according to
    CLHS 3.3 is the declarations in effect at the *definition* of F.
    But what you can do is to change the defaults:
    (PROG2
    (DECLAIM (OPTIMIZE ...))
    (COMPILE NIL 'F)
    (DECLAIM (OPTIMIZE default-settings)))
    Note that even this will not work across implementations, because
    of the distinction between compile environment and execution
    environment that some implementations make. See CLHS:DECLAIM
    for details.

     
  • Bruno Haible

    Bruno Haible - 2009-10-17
    • status: open --> closed-invalid
     
  • Sam Steingold

    Sam Steingold - 2009-10-18

    it is not clear why declaim changes the result of compile while locally does not.

     
  • Sam Steingold

    Sam Steingold - 2009-10-18
    • assigned_to: sds --> haible
     
  • Bruno Haible

    Bruno Haible - 2009-10-18

    > it is not clear why declaim changes the result of compile while locally does not.

    DECLAIM is a macro wrapper around PROCLAIM. PROCLAIM establishes the declarations in the global environment. So DECLAIM and PROCLAIM have an effect with indefinite extent, not scoped. This is why (COMPILE 'F) may actually see these changed declarations in the global environment.

    Whether it actually does, depends on whether COMPILE takes the source code of F and considers it in the current global environment (changed by DECLAIM), or whether it takes a "frozen" form of the source code of F, ignoring the current global environment. Both approaches are allowed, see CLHS 3.2.2.3.

     
  • Bruno Haible

    Bruno Haible - 2009-10-19

    It says "Conforming programs should not be written using any additional assumptions about consistency between the run-time environment and the startup, evaluation, and compilation environments." The preceding paragraphs don't mention OPTIMIZE declarations. Therefore the COMPILE function is free to take the OPTIMIZE declarations from the evaluation environment or the compilation environment of F.

    But that's only a side issue. The central argumentation is that OPTIMIZE qualities are determined by declarations, and such declarations have lexical scoping. See CLHS 3.3.4:
    "Declarations that do not apply to bindings can only appear as free declarations."
    "the scope of a free declaration includes only the body subforms of the form at whose head it appears, and no other subforms."

     

Log in to post a comment.

Get latest updates about Open Source Projects, Conferences and News.

Sign up for the SourceForge newsletter:

JavaScript is required for this form.





No, thanks