Michael Neale wrote:
Nice - will this be JIT like the extractors - or will it be part of
the compiler step?
I think the ASM bit is harder then you think though. If its in
compiler MVEL templating plus compiling can be used, but obviously not
if it is JIT in the core (in which case, yes, ASM is needed, but
fiddly to do and get right - will take a while to flush out all the
invalid class errors).
Yes the idea is to JIT it - and ASM is never easy :) but the end result
here would be worth while. I'm just showing the algorithm itself isn't
hard, including keeping alpha node hashing.
So the original AlphaNode network exists but we compile each
ObjectType's network into a class and attach that to the ObjectTypeNode
and propagate into that instead. If the ObjectTypeNode or any of it's
alpha nodes become dirty from a rule add/remove we drop the class for
that ObjectType and rebuild it.
On Jan 14, 2008 8:26 AM, Mark Proctor <[EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>> wrote:
I can now generate an alpha network, currently java as a string.
We can
change this to generate a class with ASM bytecode - so thats one
class
per one ObjectType. So the main thing is it shows the algorithm
works,
including node sharing and hashing via the case statement.
package org.drools.benchmark.manners
rule rule1
when
context : Context( state == Context.START_UP )
guest : Guest( name == "mark", sex != Sex.M, hobby ==
Hobby.H1)
count : Count()
then
end
rule rule2
when
context : Context( state == Context.START_UP )
guest : Guest( name == "mark", sex != Sex.M , hobby ==
Hobby.H2)
count : Count()
then
end
rule rule3
when
context : Context( state == Context.START_UP )
guest : Guest( name == "luke", sex != Sex.M, hobby ==
Hobby.H3)
count : Count()
then
end
rule rule4
when
context : Context( state == Context.START_UP )
guest : Guest( name == "mark", sex != Sex.M, hobby ==
Hobby.H4 )
count : Count()
then
end
The following is produced for above (obviously doesn't compile, but
shows the algo works):
public class Compiled[ClassObjectType
class=org.drools.benchmark.manners.Count ]AlphaNetwork implements
RightTupleSink {
RightTupleSink sink10;
RightTupleSink sink17;
RightTupleSink sink26;
RightTupleSink sink33;
public void assertObject(....) {
sink10.assertRight (handle, wm, context);
sink17.assertRight(handle, wm, context);
sink26.assertRight(handle, wm, context);
sink33.assertRight(handle, wm, context);
}
}
public class Compiled[ClassObjectType
class=org.drools.benchmark.manners.Context]AlphaNetwork implements
RightTupleSink {
AlphaNodeFieldConstraint alphaNodeConstraint2;
RightTupleSink sink3;
public void assertObject(....) {
if ( alphaNodeConstraint2.isAllowed(handle, wm) ) {
sink3.assertRight(handle, wm, context);
}
}
}
public class Compiled[ClassObjectType
class=org.drools.benchmark.manners.Guest]AlphaNetwork implements
RightTupleSink {
AlphaNodeFieldConstraint alphaNodeConstraint5;
AlphaNodeFieldConstraint alphaNodeConstraint6;
AlphaNodeFieldConstraint alphaNodeConstraint30;
RightTupleSink sink31;
AlphaNodeFieldConstraint alphaNodeConstraint7;
RightTupleSink sink8;
AlphaNodeFieldConstraint alphaNodeConstraint14;
RightTupleSink sink15;
AlphaNodeFieldConstraint alphaNodeConstraint21;
AlphaNodeFieldConstraint alphaNodeConstraint22;
AlphaNodeFieldConstraint alphaNodeConstraint23;
RightTupleSink sink24;
public void assertObject(....) {
if ( alphaNodeConstraint5.isAllowed(handle, wm) ) {
if ( alphaNodeConstraint6.isAllowed (handle, wm) ) {
HashKey key = new HashKey(handle);
swtich
((RightTupleSink)this.hashedSinkedMap(key)).getId() {
case 7: {
sink8.assertRight (handle, wm, context);
};
case 14: {
sink15.assertRight(handle, wm, context);
};
case 30: {
sink31.assertRight(handle, wm, context);
};
};
}
}
if ( alphaNodeConstraint21.isAllowed(handle, wm) ) {
if ( alphaNodeConstraint22.isAllowed (handle, wm) ) {
if ( alphaNodeConstraint23.isAllowed(handle, wm) ) {
sink24.assertRight(handle, wm, context);
}
}
}
}
}
--
Michael D Neale
home: www.michaelneale.net <http://www.michaelneale.net>
blog: michaelneale.blogspot.com <http://michaelneale.blogspot.com>
_______________________________________________
rules-dev mailing list
[email protected]
https://lists.jboss.org/mailman/listinfo/rules-dev
