Re: defmethod print-object :around ((condition simple-condition) stream)

[Bruno H. <br...@cl...>]

Hi Sam,

> After all, all simple conditions are internal to clisp

No. Users can define their own subclasses of SIMPLE-CONDITION, see
  <http://www.ai.mit.edu/projects/iiip/doc/CommonLISP/HyperSpec/Body/contyp_simple-condition.html>
And they can define their own PRINT-OBJECT methods on their own SIMPLE-CONDITION
subclasses.

> defmethod print-object :around ((condition simple-condition) stream)
cf. condition.lisp line 478
> Why is it an ":around" method?

Let's look at the applicable methods:

> (define-condition simple-foo (simple-error) () (:report report-foo))
SIMPLE-FOO

> (compute-applicable-methods-using-classes #'print-object (list (find-class 'simple-foo) (find-class 't)))
(#<STANDARD-METHOD (#<STANDARD-CLASS SIMPLE-FOO> #<BUILT-IN-CLASS T>)>
 #<STANDARD-METHOD :AROUND (#<STANDARD-CLASS SIMPLE-CONDITION> #<BUILT-IN-CLASS T>)>
 #<STANDARD-METHOD (#<STANDARD-CLASS CONDITION> #<BUILT-IN-CLASS T>)>
 #<STANDARD-METHOD (#<STANDARD-CLASS STANDARD-OBJECT> #<BUILT-IN-CLASS T>)>
)

If it were a primary method instead of an around method, the result would be

> (compute-applicable-methods-using-classes #'print-object (list (find-class 'simple-foo) (find-class 't)))
(#<STANDARD-METHOD (#<STANDARD-CLASS SIMPLE-FOO> #<BUILT-IN-CLASS T>)>
 #<STANDARD-METHOD (#<STANDARD-CLASS SIMPLE-CONDITION> #<BUILT-IN-CLASS T>)>
 #<STANDARD-METHOD (#<STANDARD-CLASS CONDITION> #<BUILT-IN-CLASS T>)>
 #<STANDARD-METHOD (#<STANDARD-CLASS STANDARD-OBJECT> #<BUILT-IN-CLASS T>)>
)

> If yes, wouldn't it be better to use a regular method

Indeed, it seems that if you use a primary method, the report-function of
SIMPLE-FOO will take precedence over the printing with
  (apply #'format stream (simple-condition-format-control condition)
                         (simple-condition-format-arguments condition))

Whereas in the current state, through :AROUND, the printing with
  (apply #'format stream (simple-condition-format-control condition)
                         (simple-condition-format-arguments condition))
takes precedence over the :REPORT option of SIMPLE-FOO.

> (or even a :report argument

A :report argument is equivalent to a defmethod anyway, see
  <http://www.ai.mit.edu/projects/iiip/doc/CommonLISP/HyperSpec/Body/mac_define-condition.html>

> I think this would add robustness because
> - the code becomes cleaner

It would definitely be a change in behaviour. The comment "Instead we define a
print-object method which will be executed regardless of the condition type's CPL"
indicates that the code is meant to provide some specific behaviour, regarding
the condition type's CPL. Only I don't know if that specific behaviour is the
one CLHS wants.

> - we do make sure that a careless developer does not add a print-object
>   method which overrides our preferred one.

Compliant programs are not allowed to add a :AROUND method for
PRINT-OBJECT on SIMPLE-CONDITION, and I think clisp warns about this already.

Bruno
-- 
In memoriam Marie Trintignant <http://en.wikipedia.org/wiki/Marie_Trintignant>