[cells-devel] Cells on ABCL

[Ramarren]

Hello,

On Tue, Feb 9, 2010 at 10:58 AM, Alessio Stalla <alessiostalla at gmail.com> wrote:
> I'm interested on having Cells running on ABCL. I have downloaded the
> latest snapshot from the Ramarren branch and it appears that only a
> few read conditionals are missing (for the MOP package) to make Cells
> compile and load fine; attached you find the patch.

I have applied the patch to my repository.

> However, the tests fail due to an issue I had already found the first
> time I tried Cells on ABCL (see
> http://www.mail-archive.com/cells-devel@common-lisp.net/msg00582.html).
> The fix for that is apparently simple - replace delete with remove in
> c-unlink-used (link.lisp) - but His Kennyness himself said that this
> naive fix can have a bad impact on performance, so I have not included
> it in my patch. To me, one sensible option would be to enable the
> "fix" only on ABCL (it's better to have less than optimal performance
> than to have no Cells at all!), but it's a decision that should be
> taken by the Cells developers.

Unfortunately, I am not really a Cells developer, I just cleaned some
things up for cells-gtk to run. So I don't quite understand what the
code is doing and the context. Which is why I have no idea whether
cd-useds list can contain duplicates. If it cannot, the code doesn't
really make sense, since as far as I can tell doing this:

diff --git a/link.lisp b/link.lisp
index 047783f..7407876 100644
--- a/link.lisp
+++ b/link.lisp
@@ -78,7 +78,7 @@ See the Lisp Lesser GNU Public License for more details.
                               (progn
                                 (count-it :unlink-unused)
                                 (trc nil "c-unlink-unused" c
:dropping-used (car useds))
-                                (c-unlink-caller (car useds) c)
+                                (caller-drop (car useds) c)
                                 (rplaca useds nil))
                             (progn
                               ;; moved into record-caller 060604
(caller-ensure (car useds) c)

would be strictly equivalent when the function is done and eliminate
the inner delete entirely, which should fix ABCL problem. Unless I
don't see something. I am not sure what would happen if the cd-useds
list contained duplicates, and if it did what would happen when
caller-drop is called multiple time.

At least, if I am getting the control flow correctly. As far as I can
tell the reversed recursion doesn't care the the list is not being
destructively modified on the unwind, but I am not sure how the caller
unlinking interacts with this whole thing.

The tests pass, anyway, and if I put (assert (equal (cd-useds c)
(remove-duplicates (cd-useds c)))) before nail-unused in
c-unlink-unused it doesn't signal an error from cells-gtk demo
application. I am not sure how and if it affects performance.

But that does not preclude the possibility that there are cases when
there could be duplicates, although I am fairly sure that the only
place where this list is being added to is record-caller, which
doesn't allow duplication.

It would be best if someone who knows how this works could comment on
that fix, or at least test it.

Regards,
Jakub Higersberger