Hi,
It is said in the datastore-manual:
"... you can specify that a certain slot is transient, which means that it will not be snapshotted and that its value can be changed outside of a transaction."
In my application the value of a transient slot is specified on object creation. When reloading the transaction log (server restart) the transient slot is bound and its value is correct. Is this intended behaviour or not? How is snapshotting the database different in that context?
Thanks in advance,
Hi Kamen,
are you saying that your transient slots are written to the transaction log and snapshot? This is not the intended behavior. Can you verify the bug by writing some distinctive string to such a slot, snapshot and then look in the snapshot file for that string? Do you see the the value retained when restoring or when starting your lisp afresh?
Thanks, Hans
2007/7/6, Kamen TOMOV kamen@cybuild.com:
Hi,
It is said in the datastore-manual:
"... you can specify that a certain slot is transient, which means that it will not be snapshotted and that its value can be changed outside of a transaction."
In my application the value of a transient slot is specified on object creation. When reloading the transaction log (server restart) the transient slot is bound and its value is correct. Is this intended behaviour or not? How is snapshotting the database different in that context?
Thanks in advance,
-- Kamen _______________________________________________ bknr-devel mailing list bknr-devel@common-lisp.net http://common-lisp.net/cgi-bin/mailman/listinfo/bknr-devel
Hello Hans,
Lets define a class:
(define-persistent-class a () ((s1 :update :transient t) (s2 :update)))
and make an instance of it:
(let ((obj (make-object 'a :s1 23 :s2 "ar"))) ...
and then modify its transient slot:
(setf (a-s1 obj) 3)
When I stop and start Lisp and do:
(make-instance 'mp-store....
I get:
(slot-value obj 's1) => 23 (slot-value obj 's2) => "ar"
s1 is transient and one might expect it to be unbound, but it is not, so I wondered if it is a bug or a feature, cause I might decide to use it as a feature :-)
However, I guess that if I snapshot and then restore the database the transient slot would be unbound (and that would prevent me from using it as a feature, but I probably need a persistent slot anyway). Am I correct?
Regards,
Kamen,
what you describes looks to me as if the :transient does not have any effect at all, so the question is what you'd want to use this as a feature for :) - Seriously, what you describe seems to be a bug, as :transient slots should never be logged. In fact, I am depending on this behavior in several places and I wonder how the store serializes the slots that I'm using.
I'd recommend to use :transient the way that is described in the documentation, i.e. assume that such slots will not be restored. I will eventually fix this, but it will propably take a few days until I commit.
-Hans
2007/7/6, Kamen TOMOV kamen@cybuild.com:
Hello Hans,
Lets define a class:
(define-persistent-class a () ((s1 :update :transient t) (s2 :update)))
and make an instance of it:
(let ((obj (make-object 'a :s1 23 :s2 "ar"))) ...
and then modify its transient slot:
(setf (a-s1 obj) 3)
When I stop and start Lisp and do:
(make-instance 'mp-store....
I get:
(slot-value obj 's1) => 23 (slot-value obj 's2) => "ar"
s1 is transient and one might expect it to be unbound, but it is not, so I wondered if it is a bug or a feature, cause I might decide to use it as a feature :-)
However, I guess that if I snapshot and then restore the database the transient slot would be unbound (and that would prevent me from using it as a feature, but I probably need a persistent slot anyway). Am I correct?
Regards,
-- Камен
----- Original Message ----- From: Hans Hübner Sent: 06 Юли 2007 at 13:47 Subject: Re: [bknr-devel] objects with transient slots
Hi Kamen,
are you saying that your transient slots are written to the transaction log and snapshot? This is not the intended behavior. Can you verify the bug by writing some distinctive string to such a slot, snapshot and then look in the snapshot file for that string? Do you see the the value retained when restoring or when starting your lisp afresh?
Thanks, Hans
2007/7/6, Kamen TOMOV kamen@cybuild.com:
Hi,
It is said in the datastore-manual:
"... you can specify that a certain slot is transient, which means that it will not be snapshotted and that its value can be changed outside of a transaction."
In my application the value of a transient slot is specified on object creation. When reloading the transaction log (server restart) the transient slot is bound and its value is correct. Is this intended behaviour or not? How is snapshotting the database different in that context?
Thanks in advance,
-- Kamen _______________________________________________ bknr-devel mailing list bknr-devel@common-lisp.net http://common-lisp.net/cgi-bin/mailman/listinfo/bknr-devel
Hans,
"transient" has some effect because none of the setf-s of slots of this kind are in the transaction log, only when suppling data in the :initarg they are written there.
In the morning I built something that relies on that behaviour, but then realized that it might not be what was intended. That's why I mentioned taht this might be a feature ;) Anyway, I'll follow your advise - thanks for it!
Perhaps wherever you use transient slots in your applications you setf them instead of using :initargs on them and in result they are never in the transaction log and things are fine.
Regards,
Hi Kamen,
now I understand what you write and I consider this being a real bug. Transient slots should be initialized with their default-initargs/initforms, not with whatever has been specified at object creation time. Everything else would be confusing.in general. I will fix this and commit later on.
Thanks for reporting! Hans
2007/7/6, Kamen TOMOV kamen@cybuild.com:
Hans,
"transient" has some effect because none of the setf-s of slots of this kind are in the transaction log, only when suppling data in the :initarg they are written there.
In the morning I built something that relies on that behaviour, but then realized that it might not be what was intended. That's why I mentioned taht this might be a feature ;) Anyway, I'll follow your advise - thanks for it!
Perhaps wherever you use transient slots in your applications you setf them instead of using :initargs on them and in result they are never in the transaction log and things are fine.
Regards,
-- Камен
----- Original Message ----- From: Hans Hübner Sent: 06 Юли 2007 at 14:50 Subject: Re: [bknr-devel] objects with transient slots
Kamen,
what you describes looks to me as if the :transient does not have any effect at all, so the question is what you'd want to use this as a feature for :)
- Seriously, what you describe seems to be a bug, as :transient slots
should never be logged. In fact, I am depending on this behavior in several places and I wonder how the store serializes the slots that I'm using.
I'd recommend to use :transient the way that is described in the documentation, i.e. assume that such slots will not be restored. I will eventually fix this, but it will propably take a few days until I commit.
-Hans
2007/7/6, Kamen TOMOV kamen@cybuild.com:
Hello Hans,
Lets define a class:
(define-persistent-class a () ((s1 :update :transient t) (s2 :update)))
and make an instance of it:
(let ((obj (make-object 'a :s1 23 :s2 "ar"))) ...
and then modify its transient slot:
(setf (a-s1 obj) 3)
When I stop and start Lisp and do:
(make-instance 'mp-store....
I get:
(slot-value obj 's1) => 23 (slot-value obj 's2) => "ar"
s1 is transient and one might expect it to be unbound, but it is not, so I wondered if it is a bug or a feature, cause I might decide to use it as a feature :-)
However, I guess that if I snapshot and then restore the database the transient slot would be unbound (and that would prevent me from using it as a feature, but I probably need a persistent slot anyway). Am I correct?
Regards,
-- Камен
----- Original Message ----- From: Hans Hübner Sent: 06 Юли 2007 at 13:47 Subject: Re: [bknr-devel] objects with transient slots
Hi Kamen,
are you saying that your transient slots are written to the transaction log and snapshot? This is not the intended behavior. Can you verify the
bug
by writing some distinctive string to such a slot, snapshot and then look
in
the snapshot file for that string? Do you see the the value retained
when
restoring or when starting your lisp afresh?
Thanks, Hans
2007/7/6, Kamen TOMOV kamen@cybuild.com:
Hi,
It is said in the datastore-manual:
"... you can specify that a certain slot is transient, which means that it will not be snapshotted and that its value can be changed outside of a transaction."
In my application the value of a transient slot is specified on object creation. When reloading the transaction log (server restart) the transient slot is bound and its value is correct. Is this intended behaviour or not? How is snapshotting the database different in that context?
Thanks in advance,
-- Kamen _______________________________________________ bknr-devel mailing list bknr-devel@common-lisp.net http://common-lisp.net/cgi-bin/mailman/listinfo/bknr-devel
-
Hans Hübner -
Kamen TOMOV