Am Donnerstag, den 01.10.2009, 14:32 -0600 schrieb Robert Dodier:
> > 2.
> > Cut out the inferences between the different declarations. Then a symbol
> > can declared to be a real and a noninteger at the same time.
>
> I'm in favor of this. I don't remember if cutting out the
> inference stuff in featurep breaks anything. I hope not.
>
> I believe it is useful to have featurep only for properties
> attached to symbols, and a separate system to do
> set membership inference or perhaps other kinds of inferences.
> At present featurep mixes up both (and it does a poor job
> on the inferences). So at this point let's restrict featurep to
> only symbol properties, and keep talking about a set membership
> inference system.
These are the necessary changes to cut out the inferences and to
introduce facts like %i is a noninteger:
1. Cutting out inferences in compar.lisp:
; Cut out inferences between the declarations
; (kind $integer $rational)
; (par ($rational $irrational) $real)
; (par ($real $imaginary) $complex)
2. Define properties for constants in compar.lisp:
;; Properties of constants
(kind $%i $noninteger)
(kind $%i $imaginary)
(kind $%pi $noninteger)
(kind $%pi $real)
(kind $%gamma $noninteger)
(kind $%gamma $real)
... more is possible
3. A small change to the function featurep:
Symbols are no longer by default complex.
(defmfun $featurep (e ind)
(setq e ($ratdisrep e))
(cond ((not (symbolp ind))
(merror (intl:gettext "featurep: second argument must be a
symbol; found ~M") ind))
((eq ind '$integer) (maxima-integerp e))
((eq ind '$noninteger) (nonintegerp e))
((eq ind '$even) (mevenp e))
((eq ind '$odd) (moddp e))
((eq ind '$real)
(if (atom e)
(or (numberp e) (kindp e '$real) (numberp (numer e)))
(free ($rectform e) '$%i)))
; Symbols and expressions are no longer by default complex
; ((eq ind '$complex) t)
((symbolp e) (kindp e ind))))
4. Change the definition of decl-complexp and decl-realp
to match the new behavior of kindp:
;; TRUE, if the symbol e is declared to be $complex or $imaginary.
(defmfun decl-complexp (e)
(and (symbolp e)
(or (kindp e '$complex)
(kindp e '$imaginary))))
;; TRUE, if the symbol e is declared to be
;; $integer, $rational, $real
(defmfun decl-realp (e)
(and (symbolp e)
(or (kindp e '$real)
(kindp e '$rational)
(kindp e '$integer))))
With these changes we have no problems with the testsuite. All examples
work as before. No changes are necessary.
The integral integrate(exp(-x^%i),x,0,1) no longer asks a question.
Remarks:
1. Featurep works for numbers and expressions too. Perhaps this can be
cut out in a next step.
2. The function decl-realp is not completely equivalent to the old
definition. To be complete the declarations $odd, $even,
$noninteger, $irrational have to be included. Because we have no
problems with the testsuite I think we should not extend the
definition of decl-realp further.
Comments?
Dieter Kaiser