13 Feb
2007
13 Feb
'07
3:05 p.m.
Hi,
I have a problem with infinite recursion in Magritte. This
problem was discussed on this list before. The problem is
having circular dependencies in descriptions.
I tried to utilize the hints from Lukas to use
reference: (MADynamicObject on: (MyObject
descriptionsWithoutReferences))
Having a descriptionsWithoutReferences introduces an infinite
loop by itself. So I changed it to getDescriptionsWithoutReferences.
This prevents the infinite recursion but doesn't work really
well with the ActiveRecord implementation from Ramon (at least
in my case).
So I decided to change Magritte not to do infinite recursion. I
changed
MADescriptionBuilder>>for: anObject
^ cache at: anObject ifAbsent: [
cache at: anObject put: true.
cache at: anObject put: (self build: anObject).
].
I know this looks very ugly but I don't have an idea how to do
this more elegant. My question is if this change introduces
problems elsewhere in Magritte? I don't like to fix one problem
by creating another one :)
thanks,
Norbert
13 Feb
13 Feb
10:24 p.m.
So I decided to change Magritte not to do infinite
recursion. I
changed
MADescriptionBuilder>>for: anObject
^ cache at: anObject ifAbsent: [
cache at: anObject put: true.
cache at: anObject put: (self build: anObject).
].
Looks like an interesting trick, but don't you then end up with the
value true in some of your meta-models? Wouldn't it make more sense
to use nil instead?
Cheers,
Lukas
--
Lukas Renggli
http://www.lukas-renggli.ch
10:42 p.m.
Better put a placeholder in the cache and then do a oneWayBecome
afterwards.
cache at: key ifAbsent: [
cache at: key put: (placeholder := Object new).
cache at: key put: (newValue := (self build: anObject)).
placeholder oneWayBecome: newValue ].
And if you want it bullet proof, you use Meta's MessageRecorder :)
cheers,
-- Adrian
On 13 Feb 2007, at 22:24 , Lukas Renggli wrote:
So I decided
to change Magritte not to do infinite recursion. I
changed
MADescriptionBuilder>>for: anObject
^ cache at: anObject ifAbsent: [
cache at: anObject put: true.
cache at: anObject put: (self build: anObject).
].
Looks like an interesting trick, but don't you then end up with the
value true in some of your meta-models? Wouldn't it make more sense
to use nil instead?
Cheers,
Lukas
--
Lukas Renggli
http://www.lukas-renggli.ch
_______________________________________________
SmallWiki, Magritte, Pier and Related Tools ...
https://www.iam.unibe.ch/mailman/listinfo/smallwiki
14 Feb
14 Feb
12:36 a.m.
On Tue, 2007-02-13 at 22:24 +0100, Lukas Renggli wrote:
Hmmm, I don't know if I got your point. The value at: anObject
gets discarded after self build returns (If I understand
IdentityDictionary the right way). So there is no difference
between true and nil. I just didn't spend a second thinking about
what to put as a placeholder. Now I can see the problem if there
is a second invocation with the same object before the placeholder
gets exchanged. Do you mean this?
Norbert
So I decided
to change Magritte not to do infinite recursion. I
changed
MADescriptionBuilder>>for: anObject
^ cache at: anObject ifAbsent: [
cache at: anObject put: true.
cache at: anObject put: (self build: anObject).
].
Looks like an interesting trick, but don't you then end up with the
value true in some of your meta-models? Wouldn't it make more sense
to use nil instead?
12:41 a.m.
On Tue, 2007-02-13 at 22:42 +0100, Adrian Kuhn wrote:
Better put a placeholder in the cache and then do a
oneWayBecome
afterwards.
cache at: key ifAbsent: [
cache at: key put: (placeholder := Object new).
cache at: key put: (newValue := (self build: anObject)).
placeholder oneWayBecome: newValue ].
And if you want it bullet proof, you use Meta's MessageRecorder :)
Hmmm, I don't have a oneWayBecome: in my image. But I like the
idea if I understand you correct. But wouldn't
cache at: key ifAbsent: [
cache at: key put: (placeholder := Object new).
placeholder oneWayBecome: self build: anObject ].
perfectly sufficient? (Ok, if I had a oneWayBecome ;) ). Is this
something totally different to protoobjects become: ???
very curious,
Norbert
12:51 a.m.
oneWayBecome is a VW method as it seems, but I dont know its
Squeakish name.
Meta is a meta-description framework for VW, and there I implemented
a placeholder that records everything that happens to it, and later
on, replays all that on the actual object.
-- Adrian
On 14 Feb 2007, at 0:42 , Norbert Hartl wrote:
And if you want it bullet proof, you use
Meta's MessageRecorder :)
Oh, btw. _what_ is Meta's MessageRecorder?
Norbert
_______________________________________________
SmallWiki, Magritte, Pier and Related Tools ...
https://www.iam.unibe.ch/mailman/listinfo/smallwiki
7:39 a.m.
oneWayBecome is a VW method as it seems, but I dont
know its
Squeakish name.
#becomeForward:
Meta is a meta-description framework for VW, and there
I
implemented a placeholder that records everything that happens to
it, and later on, replays all that on the actual object.
Due to its use of #doesNotUnderstand: you are screwed up as soon as
such an instance ends up in a debugger, database, tires to receive a
special selector ...
In my opinion #doesNotUnderstand: simply does not scale for most of
such things, as I explained when describing MADynamicObject (what is
somehow similar, but less sophisticated).
Cheers,
Lukas
--
Lukas Renggli
http://www.lukas-renggli.ch
9:28 a.m.
The latest version of Magritte fixes the reported problem without any
#becomeForward: or #doesNotUnderstand: tricks. The idea is simple, it
is a two step process:
1. The empty container description is registered in the cache.
2. The element descriptions are added to the container.
-+- -+- -+- -+- -+- -+- -+- -+- -+- -+- -+- -+- -+-
Name: Magritte-Model-lr.239
Author: lr
Time: 14 February 2007, 9:21:13 am
UUID: 51418e16-08d2-4a88-8dcb-1d2b6d786282
Ancestors: Magritte-Model-lr.238
- fixed problem with recursion in MADescriptionBuilder
-+- -+- -+- -+- -+- -+- -+- -+- -+- -+- -+- -+- -+-
Name: Magritte-Tests-lr.104
Author: lr
Time: 14 February 2007, 9:21:18 am
UUID: 06eff360-a6fe-49d6-9425-dd87f3d83499
Ancestors: Magritte-Tests-lr.103
- fixed problem with recursion in MADescriptionBuilder
-+- -+- -+- -+- -+- -+- -+- -+- -+- -+- -+- -+- -+-
The tests have been updated as well. Hope this helps for your problem?
Now the next step is to get rid of recursion when traversing (e.g.
for validation) cyclic graphs, but I'll wait with that until somebody
complains ;-)
Cheers,
Lukas
--
Lukas Renggli
http://www.lukas-renggli.ch
10:46 a.m.
On Wed, 2007-02-14 at 09:28 +0100, Lukas Renggli wrote:
The latest version of Magritte fixes the reported
problem without any
#becomeForward: or #doesNotUnderstand: tricks. The idea is simple, it
is a two step process:
1. The empty container description is registered in the cache.
2. The element descriptions are added to the container.
I cannot say I understand your changes completely but they look
good and more important they work.
The tests have been updated as well. Hope this helps
for your problem?
Yes, it does. Thanks very much.
Now the next step is to get rid of recursion when
traversing (e.g.
for validation) cyclic graphs, but I'll wait with that until somebody
complains ;-)
I will :)
regards,
Norbert
12:04 p.m.
The latest version of Magritte fixes the reported
problem without any
#becomeForward: or #doesNotUnderstand: tricks. The idea is simple, it
is a two step process:
1. The empty container description is registered in the cache.
2. The element descriptions are added to the container.
Very nice!
Meta is a
meta-description framework for VW, and there I
implemented a placeholder that records everything that happens to
it, and later on, replays all that on the actual object.
Due to its use of #doesNotUnderstand: you are screwed up as soon as
such an instance ends up in a debugger, database, tires to receive a
special selector ... In my opinion #doesNotUnderstand: simply does not
scale for most of
such things, as I explained when describing MADynamicObject (what is
somehow similar, but less sophisticated).
It scales IMHO, but has a lot of very nasty pitfall. Therefor only
skilled engineers should use DNU or become.
Cheers,
Lukas
--
Lukas Renggli
http://www.lukas-renggli.ch
_______________________________________________
SmallWiki, Magritte, Pier and Related Tools ...
https://www.iam.unibe.ch/mailman/listinfo/smallwiki
12:37 p.m.
Dear Adrian,
code that used them. The only thing I could not solve, was the case
when a spawned recorder is used a receiver of an ifThen.
There is a trick that you might be able to use in this case.
You can trap this case by implementing #mustBeBooleanIn: context
For an example of code which does this, see:
http://www.squeaksource.com/Null
Keith
___________________________________________________________
All new Yahoo! Mail "The new Interface is stunning in its simplicity and ease of
use." - PC Magazine
http://uk.docs.yahoo.com/nowyoucan.html
2:01 p.m.
Wow, that's cool stuff, Keith!
Just read the comment of the Squeak version, where can I get the
mentioned VW version? ^^
-- Adrian
On 14 Feb 2007, at 12:37 , Keith Hodges wrote:
Dear Adrian,
code that used them. The only thing I could not solve, was the case
when a spawned recorder is used a receiver of an ifThen.
There is a trick that you might be able to use in this case.
You can trap this case by implementing #mustBeBooleanIn: context
For an example of code which does this, see:
http://www.squeaksource.com/Null
Keith
___________________________________________________________
All new Yahoo! Mail "The new Interface is stunning in its
simplicity and ease of use." - PC Magazine
http://uk.docs.yahoo.com/nowyoucan.html
_______________________________________________
SmallWiki, Magritte, Pier and Related Tools ...
https://www.iam.unibe.ch/mailman/listinfo/smallwiki
15 Feb
15 Feb
2:06 a.m.
Adrian Kuhn wrote:
Wow, that's cool stuff, Keith!
Just read the comment of the Squeak version, where can I get the
mentioned VW version? ^^
-- Adrian
The only reference that I could find is a copy of the original paper by
Nevin Pratt, which contains code for VW.
http://www.diit.unict.it/~acalva/SE/slides/se2/generalized-nullobject.rtf
enjoy
Keith
___________________________________________________________
The all-new Yahoo! Mail goes wherever you go - free your email address from your Internet
provider. http://uk.docs.yahoo.com/nowyoucan.html
6448
days inactive
6450
days old
13 comments
4 participants
participants (4)
-
Adrian Kuhn -
Keith Hodges -
Lukas Renggli -
Norbert Hartl