From alex.mizrahi at gmail.com  Sat Dec 16 14:50:57 2006
From: alex.mizrahi at gmail.com (Alex Mizrahi)
Date: Sat, 16 Dec 2006 16:50:57 +0200
Subject: [cl-store-devel] cl-store for ABCL
Message-ID: <f571e76f0612160650k7e0d83d0xdbbb9273d4f23983@mail.gmail.com>

helo

thanks for adding support of ABCL to cl-store!

it works for simple hash-tables,
but, i have a problem with restoring nested hash-tables (hash-table
whose values are hash-tables):

CL-USER(16): (setq *ht1* (make-hash-table))
#<EQL hash table, 0 entries, 16 buckets {7E6FB2}>
CL-USER(17): (setf (gethash 1 *ht1*) (make-hash-table))
#<EQL hash table, 0 entries, 16 buckets {D71AA7}>
CL-USER(18): (setf (gethash 2 (gethash 1 *ht1*)) 1)
1
CL-USER(19): (cl-store:store *ht1* "ht1.store")
#<EQL hash table, 1 entry, 16 buckets {7E6FB2}>
CL-USER(20): (cl-store:restore "ht1.store")
Debugger invoked on condition of type RESTORE-ERROR:
  The function #<CL-STORE::REFERRER {F42466}> is undefined.
Restarts:
  0: TOP-LEVEL Return to top level.

additionally, there were problems restoring store made by CLISP. it
have written EXT:FASTHASH-EQUAL as a hash-function in CLISP, which
does not exist in ABCL, so i had to map EXT:FASTHASH-EQUAL to
CL:EQUAL, and after that hash table loaded successfully. i don't know
if cross-implementation restoring is supported, but it's very nice
that it's possible to do it :)

with best regards, Alex 'killer_storm' Mizrahi.


From alex.mizrahi at gmail.com  Sat Dec 16 15:52:55 2006
From: alex.mizrahi at gmail.com (Alex Mizrahi)
Date: Sat, 16 Dec 2006 17:52:55 +0200
Subject: [cl-store-devel] cl-store for ABCL
In-Reply-To: <1166281399.5112.5.camel@deepthought>
References: <f571e76f0612160650k7e0d83d0xdbbb9273d4f23983@mail.gmail.com>
	<1166281399.5112.5.camel@deepthought>
Message-ID: <f571e76f0612160752u13260e89xa92e9a425054b041@mail.gmail.com>

> I suspect that this is a bug in ABCL

with non-compiled cl-store i was able to get more detailed backtrace:

(MAKE-HASH-TABLE
 :TEST
 #<CL-STORE::REFERRER {DE5F11}>
 :REHASH-SIZE
 #<CL-STORE::REFERRER {184222A}>
 :REHASH-THRESHOLD
 #<CL-STORE::REFERRER {1DEF740}>
 :SIZE
 16)

so, problem is that values were not 'dereferenced'.

this is because you use eq hash-table *restored-values*. same numbers
are not guaranteed to be EQ. (small numbers are EQ in ABCL, but they
are not EQ in HT).

i've made *restored-values* EQL hash-table, and my small file got
loaded, i'm not sure it doesn't violate some semantics though.

however, my bigger file has problems, in unicode-base-string. but
unicode is not officially supported in ABCL, so it's up to me to make
a workaround for this..


From alex.mizrahi at gmail.com  Sun Dec 17 09:43:12 2006
From: alex.mizrahi at gmail.com (Alex Mizrahi)
Date: Sun, 17 Dec 2006 11:43:12 +0200
Subject: [cl-store-devel] cl-store for ABCL
In-Reply-To: <1166314541.9615.11.camel@deepthought>
References: <f571e76f0612160650k7e0d83d0xdbbb9273d4f23983@mail.gmail.com>
	<1166281399.5112.5.camel@deepthought>
	<f571e76f0612160752u13260e89xa92e9a425054b041@mail.gmail.com>
	<1166314541.9615.11.camel@deepthought>
Message-ID: <f571e76f0612170143k6052480audcbae3c6e182f414@mail.gmail.com>

> > however, my bigger file has problems, in unicode-base-string. but
> > unicode is not officially supported in ABCL, so it's up to me to make
> > a workaround for this..
> Hmm, i'm not sure I understand what the problem is with the string?
> Is the problem with detecting unicode strings or the dumping of wide
> characters?

char-code and code-char are asymmetric in ABCL -- char code can return
ucs-2 codes, but code-char will return NIL for char-codes more than
255.
i've patched ABCL's code-char to return valid charaters, and then it
loaded my file well (i didn't check correctness though -- i've passed
this file through CLISP to convert CL-STORE version, so chars code be
corrupted).
so, it's ABCL needs to be patched..


From rosssd at gmail.com  Sun Dec 17 12:17:44 2006
From: rosssd at gmail.com (rosssd at gmail.com)
Date: Sun, 17 Dec 2006 12:17:44 +0000
Subject: [Fwd: Re: [cl-store-devel] cl-store for ABCL]
Message-ID: <1166357864.16068.1.camel@deepthought>

-------- Forwarded Message --------
From: rosssd at gmail.com <rosssd at gmail.com>
Reply-To: rosssd at gmail.com
To: Alex Mizrahi <alex.mizrahi at gmail.com>
Cc: cl-store-devel at common-lisp.net
Subject: Re: [cl-store-devel] cl-store for ABCL
Date: Sat, 16 Dec 2006 15:03:20 +0000

On Sat, 2006-12-16 at 16:50 +0200, Alex Mizrahi wrote:
> helo
> 
> thanks for adding support of ABCL to cl-store!

Thanks for the report.
I wont be able to get to this problem right now but I'll find some 
time to investigate this later today.

I suspect that this is a bug in ABCL since hash table's seem to
serialise fine with other implementations, but I'll double check
this. 

Cheers,
 Sean.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <https://mailman.common-lisp.net/pipermail/cl-store-devel/attachments/20061217/d5c54e2d/attachment.sig>

From rosssd at gmail.com  Sun Dec 17 12:18:02 2006
From: rosssd at gmail.com (rosssd at gmail.com)
Date: Sun, 17 Dec 2006 12:18:02 +0000
Subject: [Fwd: Re: [cl-store-devel] cl-store for ABCL]
Message-ID: <1166357882.16068.3.camel@deepthought>

-------- Forwarded Message --------
From: rosssd at gmail.com <rosssd at gmail.com>
Reply-To: rosssd at gmail.com
To: Alex Mizrahi <alex.mizrahi at gmail.com>
Cc: cl-store-devel at common-lisp.net
Subject: Re: [cl-store-devel] cl-store for ABCL
Date: Sun, 17 Dec 2006 00:15:42 +0000

On Sat, 2006-12-16 at 17:52 +0200, Alex Mizrahi wrote:
> i've made *restored-values* EQL hash-table, and my small file got
> loaded, i'm not sure it doesn't violate some semantics though.

Thanks, fix applied.

> however, my bigger file has problems, in unicode-base-string. but
> unicode is not officially supported in ABCL, so it's up to me to make
> a workaround for this..
Hmm, i'm not sure I understand what the problem is with the string?
Is the problem with detecting unicode strings or the dumping of wide
characters?

Cheers,
 Sean.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <https://mailman.common-lisp.net/pipermail/cl-store-devel/attachments/20061217/feac4b59/attachment.sig>

From rosssd at gmail.com  Sun Dec 17 12:18:14 2006
From: rosssd at gmail.com (rosssd at gmail.com)
Date: Sun, 17 Dec 2006 12:18:14 +0000
Subject: [Fwd: Re: [cl-store-devel] cl-store for ABCL]
Message-ID: <1166357894.16068.5.camel@deepthought>

-------- Forwarded Message --------
From: rosssd at gmail.com <rosssd at gmail.com>
Reply-To: rosssd at gmail.com
To: Alex Mizrahi <alex.mizrahi at gmail.com>
Cc: cl-store-devel at common-lisp.net
Subject: Re: [cl-store-devel] cl-store for ABCL
Date: Sat, 16 Dec 2006 23:24:23 +0000

On Sat, 2006-12-16 at 17:52 +0200, Alex Mizrahi wrote:
> so, problem is that values were not 'dereferenced'.
> 
> this is because you use eq hash-table *restored-values*. same numbers
> are not guaranteed to be EQ. (small numbers are EQ in ABCL, but they
> are not EQ in HT).

cough, cough, remind me not to jump to conclusions....
Thanks for this report and for identifying the problem, I've been
naively assuming that fixnums are eq, a problem with relying on
implementations and not specifications.

I don't think that this has any other effects and barring any 
problems I'll commit a fix to this effect tomorrow.

Thanks again.

Sean.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <https://mailman.common-lisp.net/pipermail/cl-store-devel/attachments/20061217/838a67f9/attachment.sig>