ReteOO is the supported and stable algorithm, if it correctly fixes
ReteOO then keep it. Leaps is marked experimental, and it sounds like
its not a regression over previous Leaps relases, so its fine just to
make users aware of this issue. For now just xxxtest the unit test so
the build doesn't break, hopefully Alex will fix soon.
Mark
PS. I have CC'd the dev list.
Edson Tirelli wrote:
All,
The situation is this: I fixed the issue JBRULES-183 for ReteOO as
we discussed yesterday. After testing everything in reteoo, I ran a
full test and realized that leaps was broken. I thought it was because
of my changes and started talking to Alex. After some research, I
decided to get code back to revision 3550 (before my changes) and
tested and the error was happening before my changes.
This rule works fine in LEAPS:
rule "not rule test2"
salience 6
when not Cheese( price == 5 )
Cheese( price == 7 )
then
list.add( new Integer( 7 ) );
end
If you add a binding to cheese, it throws exception:
rule "not rule test2"
salience 6
when not Cheese( price == 5 )
$c : Cheese( price == 7 )
then
list.add( new Integer( 7 ) );
end
So, basically, both algorithms were suffering from the same problem
and my fix fixed only ReteOO problem.
It seems simple, but it might be not that simple. When NOT is used
in the first column, RuleBuilder right now assumes someone will create
the initial-fact and it will be index 0 on the tuple, so, it does not
increment offset. In leaps, according to Alex, there is no
initial-fact, and this way, the offset calculation should be different
for Rete and Leaps.
Alex sugests to do the calculation on the Rule.getTransformedLhs()
method. It may be possible, but my question is:
Mark, you said you wanted to release RC2 tonight. Should we simply
comment out the test case to avoid leaps error or do you want me to
remove my changes until we fix leaps also?
[]s
Edson
PS: Alex and I are working on fixing this anyway.
Alexander Bagerman wrote:
Edson,
I think the most logical place for the changes I propose would be
Rule.getTransformedLhs(). Both reteoo and leaps builders use it while
adding rule to the respective structures.
There are still problems with this approach - declaration and column
classes do not have setters on column(Declaration) and index(Column)
that would result in generating new Declaration and Column classes.
This in turn will break low level Java rules because declarations
created on columns are used in Consequences. It would be better if
Declaration takes Column and int in constractor but it would not solve
the problem either. I think I painted myself into the corner.....
I'm online till 3pm PST (California time) so feel free to ping me to
chat.
-Alex
