Use compiler-native IDs instead of our id/new-id?

[aartaka]

Our ID indexes objects we create, but indexing requires either:

• Adding an id slot to an object.
• Or storing the ID-object pair somewhere else.

Both are cumbersome. What if we had some Lisp-native way to generate object IDs without any work on our side? Like...

• print-unreadable-object's :identity option
• or sxhash!

Most implementations use some kind of implementation-internal object ID fetching for unreadable objects, and most have Lisp APIs for getting this ID reliably. Here's what I've been able to hack up for SBCL, CCL, ECL, and ABCL:

(defun id (object)
  #+sbcl (sb-kernel::get-lisp-obj-address object)
  #+ccl (ccl::%address-of object)
  #+ecl (si:pointer object)
  #+abcl (system::identity-hash-code object)
  #+clisp (system::address-of object)
  #+gcl (system:address object)
  #+allegro (excl:lispval-to-address object)
  #-(or sbcl ccl ecl abcl clisp gcl allegro) (sxhash object))

For all the other implementations there's always sxhash which has nice properties (equal objects have the same sxhash), but IDs are even better and come from the reliable compiler internals and are guaranteed to never collide for different objects (which most hash functions occasionally do).

So, shall we go down that road?

EDIT: Update for CLISP, GCL, and Allegro.
EDIT: Fix a typo.

[aadcg]

Sounds interesting to me.

[Ambrevar]

Very smart, I like it! :)

[aartaka]

Updated function for some more implementations I've tested:

(defun id (object)
  #+sbcl (sb-kernel::get-lisp-obj-address object)
  #+ccl (ccl::%address-of object)
  #+ecl (si:pointer object)
  #+abcl (system::identity-hash-code object)
  #+clisp (system::address-of object)
  #+gcl (system:address object)
  #-(or sbcl ccl ecl abcl clisp gcl) (sxhash object))

[Ambrevar]

Nonguix has AllegroCL if you feel like testing it :)

[aartaka]

Nonguix has AllegroCL if you feel like testing it :)

The license has expired and I've sent a patch to Nonguix to fix it :P

[Ambrevar]

Done!

[jmercouris]

I don't see the benefit here. In fact if we serialize an object where we have the id in a slot, that id will be persisted. With this method, it would not, no?

[aartaka]

I don't see the benefit here. In fact if we serialize an object where we have the id in a slot, that id will be persisted. With this method, it would not, no?

Yes, but we can compile the data into .FASL-s to preserve the object itself, for example.

EDIT: we