[armedbear-devel] eql for java objects

[Alessio Stalla]

On Sun, Apr 25, 2010 at 9:49 PM, Tobias C. Rittweiler <tcr at freebits.de> wrote:
> Alessio Stalla <alessiostalla at gmail.com>
> writes:
>
>> On Sun, Apr 25, 2010 at 8:37 PM, Erik Huelsmann <ehuels at gmail.com> wrote:
>>> Hi all,
>>>
>>> On Sun, Apr 25, 2010 at 7:26 PM, Alan Ruttenberg
>>> <alanruttenberg at gmail.com> wrote:
>>>>
>>>> On Apr 25, 2010, at 6:55 AM, Tobias C. Rittweiler wrote:
>>>>
>>>>> I think that's past Alan's point which, from my understanding, is simply
>>>>> "Java enums should be comparable by EQL." Makes perfect sense as EQL is
>>>>> for object identity.
>>>>
>>>> Yes, and Java "objects" too.
>>>
>>> I've read this thread, the thread on c.l.l regarding EQUAL and the
>>> definitions of EQ and EQL. The following is my interpretation of it
>>> all:
>>>
>>> 1. Our Java side JavaObject class is merely a box for a Java instance
>>> value (a pointer to a Java object, if you will)
>>> 2. The definitions of EQ and EQL talk about Objects, but I interpret
>>> them to refer to the first meaning in [1], not to instances of lisp
>>> classes
>>> 3. Our JavaObjects have not been defined in the (Common) Lisp spec;
>>> they don't adhere to the object-instantiation protocol nor are they
>>> any of any one of the predefined built-in classes and therefore can't
>>> be taken to be "Lisp Objects" in that sense of the word
>>> 4. From point (1) and the definitions of EQ and EQL, I concur with
>>> Alan that "raison d'etre" of EQL should equally apply to JavaObjects
>>> 5. From ponits (2) and (3) and the discussion on #lisp and c.l.l, I
>>> conclude that we're basically free to extend the meaning of EQL here:
>>> the JavaObject values were not in the spec to begin with: they're not
>>> Lisp class instances, they're not symbols, numbers nor characters
>>>
>>> The only objection there is from both #lisp and c.l.l is that EQ, EQL
>>> and EQUAL shouldn't start to behave unpredictably regarding defined
>>> behaviour. For all the spec cares, we would have generated an error
>>> when EQ-comparing 2 java objects...
>>>
>>>
>>> Agreed?
>>
>> I agree almost completely, but I'd like to add that:
>>
>> 1. if EQL is modified, then EQUAL must be modified as well because
>> (eql x y) should always imply (equal x y)
>
> EQUAL does not internally call EQL? If that's indeed not the case,
> perhaps it should be made so?

Maybe it does, I was reasoning in the abstract, I meant: let's ensure
that EQUAL works consistently.

>> 2. if we have freedom to modify EQL and EQUAL, then I'd say that EQUAL
>> should call Object.equals(), but I know this is controversial
>
> No that would be false reasoning IMHO. The only reason why it's
> acceptable to extend EQL that way is because the Notes section in EQL
> kind of implies (albeit does not spell it out explicitly) certain
> implementational freedom in the interplay of EQ and EQL.

I don't read more implementational freedom in EQL than in EQUAL.

For EQL, the precise rules are:

"The value of eql is true of two objects, x and y, in the folowing (sic) cases:

1. If x and y are eq.
2. If x and y are both numbers of the same type and the same value.
3. If they are both characters that represent the same character.

Otherwise the value of eql is false."

The notes then say a more general thing, that "eql tells whether two
objects are conceptually the same, whereas eq tells whether two
objects are implementationally identical".

For EQUAL, instead:

"Returns true if x and y are structurally similar (isomorphic)
objects" describes it in general, as EQL is described in general in
its notes.

Then the precise rules are set:

"Objects are treated as follows by equal.
[...]
Other (Structures, hash-tables, instances, ...)

    Two other objects are equal only if they are eq."

So, both are presented with a description in natural language that
seems to concede leeway for an extension, but are also described by a
very strict set of rules. Whether those rules permit extensions or not
depends on whether non-CL objects, like Java objects, are considered
"objects" as defined by the CL standard or not. Imho, there are valid
rational arguments for both interpretations, but the second one (Java
objects are not Lisp objects) is the most practical in my opinion.

Alessio