DEFSETF and packages

[Carlos]

Hello,

I had to add some missing functionality, so I did

(in-package :my-utils)

(defun set-file-write-date (file date) ... )

(defsetf file-write-date set-file-write-date)

and... that was it! I didn't have to export anything, not even
SET-FILE-WRITE-DATE. I can use (setf file-write-date) from
anywhere.

That's... odd, isn't it?

--

[Zach Beane]

The behavior is undefined per http://l1sp.org/cl/11.1.2.1.2 clause 13,
sorry. An implementation that locks the CL package (like SBCL and
others) will reject your code.

Aside from that, the reference from (setf file-write-date) to your
internal symbol is not explicit in calls to setf, so you don't have to
do anything special to use it.

Zach

[Carlos]

On Mon, 13 Aug 2012 10:09:46 -0400
Zach Beane <xa...@xach.com> wrote:

> Carlos <an...@quovadis.com.ar> writes:
>
> > Hello,
> >
> > I had to add some missing functionality, so I did
> >
> > (in-package :my-utils)
> >
> > (defun set-file-write-date (file date) ... )
> >
> > (defsetf file-write-date set-file-write-date)
> >
> > and... that was it! I didn't have to export anything, not even
> > SET-FILE-WRITE-DATE. I can use (setf file-write-date) from
> > anywhere.
> >
> > That's... odd, isn't it?
>
> The behavior is undefined per http://l1sp.org/cl/11.1.2.1.2 clause 13,
> sorry. An implementation that locks the CL package (like SBCL and
> others) will reject your code.

Oh, I see. Anyway, this happening when one package isn't CL doesn't
seem *so* odd, but nevertheless is a little baffling.

> Aside from that, the reference from (setf file-write-date) to your
> internal symbol is not explicit in calls to setf, so you don't have to
> do anything special to use it.

It seems strange for a package to be able to add a setf function for
another's package symbol and for that function to "piggyback" the
exported status of the original function's symbol. Intuitively, I would
expect to be required to do something like "(:export '(setf xxxx))".

But I understand why it happens at the technical level. Thanks.

--

[Zach Beane]

Carlos <an...@quovadis.com.ar> writes:

>> Aside from that, the reference from (setf file-write-date) to your
>> internal symbol is not explicit in calls to setf, so you don't have to
>> do anything special to use it.
>
> It seems strange for a package to be able to add a setf function for
> another's package symbol and for that function to "piggyback" the
> exported status of the original function's symbol. Intuitively, I would
> expect to be required to do something like "(:export '(setf xxxx))".

It helps to understand that packages are for managing *only* symbols,
and not any of the concepts or features attached to the symbols. You
aren't exporting *functions*, you're only exporting *names*. And the
packages don't know what the names they manage actually name.

Zach

[Carlos]

Yes, I understand. But it somewhat clashes with the concept that a
function can be named "(setf xxxx)". We can really only export atomic
names... :-/

--