Re: [PATCH] Format's W directive not handling shared structures correctly.

[Per Bothner <per at bothner dot com>]

On 11/23/2011 06:47 AM, Charles Turner wrote:
> ~W wasn't being treated as special case in the ObjectFormat class,
> I've changed it to communicate with the pretty printer.

Thanks!  Some comments below.

> > (define a (cons 'a 'z)
> > (set-cdr! a a)
> > (format "~W" a)
> ... OutOfMemory ...
>
> with the patch
>
> > (format "~W" a)
> #1=(a . #1#) ; as described in
> http://www.lispworks.com/documentation/HyperSpec/Body/22_cdc.htm

Actually, my reading the Common Lisp Hyperspec is that ~W does *not*
automatically set *print-circle* - it needs to be set explicitly.

However, more to the point is SRFI 48: Intermediate Format Strings
(http://srfi.schemers.org/srfi-48/srfi-48.html).  This *does*
specify that ~W is "WriteCircular", which seems to match your patch.

I fixed your patch to update TracedProcedure.java.  Don't know you
you missed that.

> One obviously unpleasing part of this patch is that I'm doing the same
> setup/finalisation in ObjectFormat#print as I did in
> write-with-shared-structure in ports.scm.

I factored out the common code in my modification.

> This patch comes with a change log and a couple of tests.

I added a couple of tests - and I uncovered a bug.  This fails:

(define circ2 (cons 'a 'z))
(define circ23 (list circ2 circ2 circ2))
(format "~W" circ23)

The result is "#1=((a . z) #1# #1#)" but IMO it should be
"(#1=(a . z) #1# #1#)".  Could you look into it?

> I noticed that the javadoc linked from the Kawa homepage hasn't been
> updated for some time (well, my documentation changes aren't in it).
> Is that just a no time thing, or am I the only one who's noticed?

It hasn't been a priority - but it would be reasonable to update.
--
	--Per Bothner
per@bothner.com   http://per.bothner.com/

Attachment: formatW2.patch
Description: Text document