On 29/07/2011 14:28, Edson Tirelli wrote:

   Yes, that is exactly what I think. Pattern matching constraints are like query parameters. They need to exist and evaluate to true in order to match. So, for this to match:

a.b.c == null

   a needs to exist and be non-null, b needs to exist and be non-null, c needs to exist and be null. So it is not just NP safe navigation... it is an existence test at the same time. So for maps

a[x].b[y].c[z] == null

   The keys x, y and z need to exist, and c[z] must have a value of null. That is what the expression above is asking for, in my understanding.

   This presents no loss of completeness to the language, as you can still test non-existence of keys if that is what you want, but the most common case you are looking for the opposite and it becomes much simpler to write rules that way.

> So, a builder option to turn this on is allright with me.

   We can probably do that and have a configuration option to turn this feature on/off.

I'm strongly against configuration options in this case, we decide on one way and stick with it. We already have too many configurations and a casual person looking at the code could introduce a bug as they weren't aware of what configuratino was on for null safety.

I think part of the problem here is we are mixing domains, between script evaluation and relational constraints. There is a reason why other rule engines don't do nested accessors :) (ignoring the technical issues too).

Mark

Mark

Edson
2011/7/29 Mark Proctor <mproctor@codehaus.org>
Lets forget that these are nested accessors and the problems they bring. Lets look at what they would be if they were real relations:

On 29/07/2011 08:55, Wolfgang Laun wrote:
Whoa! See below...

2011/7/28 Edson Tirelli <ed.tirelli@gmail.com>

   I think we need to differentiate paradigms here. When using rules, contrary to imperative code, what we are doing is pattern matching.

X( a.b.c == <value> )

   In the above case, we are looking for Xs that make that whole constraint true (i.e. match). If a or b are null, the whole expression will be false, does not matter the value of c or the value it is being compared against.

(Edson: Only if you define it so. The logical implication of c being null in an absent a.b (i.e., a.b==null) could very well be that "a.b.c does not exist", and you can't claim that a.b.c exists if a.b. doesn't!

Is there no house at some address?
    (city.street[name].house[number] == null) # true => no such house

$c : City()
$s : Street( city == $c, street = "name" )
       House( number == null)

The above is identical logic to the more convenient form of nested accessors, it's the proper relational form. In this case if there was no Street, it wouldn't match.
This test data with false when null: Vienna/TirelliStrasse/42 returns "false", hence there is such a house. But we don't have a Tirelli Street in Vienna (yet)!

Confer this to Perl's
    ! exists $city{-streets}{"Tirelli"}[42]
)

Raising a null pointer exception, IMO, brings no advantage at all to the table... on the contrary, makes writing rules more difficult.

Edson, Mark,... please do recall the times where you have had an NPE in the code in a boolean expression? How painful would it have been if Java would have returned "false", continuing to cover a coding error made elsewhere?

Why don't other languages tolerate "null" silently? (Perl, the most pragmatic of all, doesn't - it has introduced an operator I can use or not.)

I have no problem when folks want to take shortcuts and live la dolce vita, but
<em>I don't want to be led into the bog without my consent.</em>

So, a builder option to turn this on is allright with me.

   Another example we had in the past:

class Circle implements Shape

class Square implements Shape

rule X

when

   Circle() from $shapes

...

   In the above example, $shapes is a list and the rule is clearly looking for Circles. If there are Squares in there, they will just not match. Raising a ClassCastException like it would happen in an imperative language brings no advantage to the table, IMO.

This is an entirely different matter than the previous one. I see no reason whatsoever, not to define this "from" as working with an implicit filter.

-W

   So, IMO, all property navigation should be null pointer safe in the LHS of the rules.

   This is not what happens today, but I think it should be fixed.

   Edson




2011/7/28 Vincent LEGENDRE <vincent.legendre@eurodecision.com>

Hi all,

I agree with W. : NPE should be the default, and "null" cases behaviour should be planned by programmers.
But I am not sure about using a new operator in rules (and do the update in Guvnor ...).
Why not using some drools annotations on the getter specifying the behaviour of an eval on a null value returned by this getter ?
And may be these annotation could be added to an existing POJO via the declared type syntax (just like event role in fusion) ?

Vincent.

_______________________________________________
rules-users mailing list
rules-users@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users

--
Edson Tirelli
JBoss Drools Core Development
JBoss by Red Hat @ www.jboss.com

_______________________________________________
rules-users mailing list
rules-users@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
_______________________________________________
rules-users mailing list
rules-users@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
_______________________________________________
rules-users mailing list
rules-users@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
JBoss Drools Core Development
JBoss by Red Hat @ www.jboss.com
_______________________________________________
rules-users mailing list
rules-users@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users