On Fri, Jul 15, 2011 at 8:50 AM, Mark Evenson <evenson@panix.com> wrote:

On Jul 14, 2011, at 20:42 , Erik Huelsmann wrote:
>
> My idea would be to disambiguate the two functions by making writeToString() a java-side PRINT-OBJECT, renaming it to printObject() and documenting that's what it does. Then, probably, the lisp side function %write-to-string should be renamed to %print-object too. Probably, more adjustments are required throughout the code base, but those are probably at the 'implementation detail' level.

I think that the meaning of writeToString() was that the function
should provide an implementation of serialization to a string able
to be READ with the #S notation.  It is not really performing a
PRINT-OBJECT-alike function which is a specialization of such an
implementation on Lisp STANDARD-OBJECT or STRUCTURE-OBJECT.

On the other hand, the clhs specifies that an implementation defines enough additional methods besides the STANDARD-OBJECT and STRUCTURE-OBJECT ones to make sure there will always be an applicable method. We seem to do that through a single T method which forwards to the internal writeToString() Java method.

Additionally, the clhs page for PRINT-OBJECT specifies that readable printing [for standard objects and structure objects] is done through the PRINT-OBJECT method rather than its MAKE-LOAD-FORM method. Many of our Java side functions try to behave this way by checking the values of the various *PRINT-xxx* special variables. One of the conclusions could be that our writeToString() functions are really trying to behave as PRINT-OBJECT implementations.
 
On the other hand there is some justification for not calling such
an implementation writeToString() as that will certainly confuse a
Java user used to every object either inheriting or overrriding
java.lang.Object.toString().  I might suggest then, that to clarify
that this function isn't a direct implementation of CL's PRINT-OBJECT
which actually takes an both arguments to specify the object it is
operating on and to specify the output streamthat we rename the
Java-side to writeLispObject(). The return type of this method,
java.lang.String, makes something more verbose like
writeLispObjectToString() superfluous.  I'd leave the Lisp
%write-to-string alone, as the presence of WRITE-TO-STRING in the
CL namespace makes it pretty clear that this is the primitive
implementation.  Since %write-to-string has a simple implementation,
looking up its behavior is rather easy.

In any case, I would take advantage of the architecture I started
to define in org.armedbear.lisp.protocol to create an interface
which defines this contract:

package org.armedbear.lisp.protocol;

/** The implementing object can be serialized ala WRITE. */
public interface WriteSerializable {  //  XXX not crazy about the name here
 /** Serialize the implementing object instance in a manner that may be READ via #S */
 public String writeLispObject();
}

and then make LispObject implement this interface.

This provides a suitable place for encapsulation and documentation,
being the natural place that a Java programmer will go to
figure out what this method should do as it implements the interface.


Given the purpose of PRINT-OBJECT as taken from the CLHS, I guess only the discussion about the naming remains? And of course the conclusion that if writeToString() actually implements the PRINT-OBJECT protocol, it's failing to do that horribly: generating unreadable presentations when *PRINT-READABLY* is bound to non-NIL values...

Right?


Bye,

Erik.