Validator for "cartesian product" (isolated constraint) problems

[istvanrath]

A nice usability improvement would be to warn the user if the cartesian product (isolated constraint) problem can be detected inside a pattern body.

E.g.

pattern MyBadPattern(X,Y) {
  Type1(X);
  Type2(Y);
}

[ujhelyiz]

Use a union-find algorithm.

Two ways to check: one that relies on every kind of constraint; and one that only regards "positive" constraints (e.g. excluding neg/check).

[okrosa]

Proposed warning messages:

• Strict case: "Warning the pattern contains fully independent parts, which can lead to excessive performance issues. The independent partitions: ..."
• Weaker case: "Warning the pattern contains parts which are only loosely connected. This may negatively impact performance, please check if it can be avoided. The weakly dependent partitions: ..."

[istvanrath]

I suggest minor refinements:

• Strict case: "The pattern body contains isolated constraints ("cartesian products") that can lead to severe performance and memory footprint issues. The independent partitions are: ..."
• Weaker case: "The pattern body contains constraints which are only loosely connected. This may negatively impact performance. The weakly dependent partitions are: ..."

[ujhelyiz]

Two cases with incorrect warnings (at least according to me and @bergmanngabor):

Case 1: The pattern body contains constraints which are only loosely connected. This may negatively impact performance. The weakly dependent partitions are: [C ][_O ][_P ]

pattern emptyClass(C : Class) {
    neg find hasOperation(C, _O);
    neg find hasProperty(C, _P);
}

Case 2: The pattern body contains constraints which are only loosely connected. This may negatively impact performance. The weakly dependent partitions are: [P C ][No ][_<0> ]

pattern propertyNumberOfClassInPackage(P: Package, C : Class, No) {
    Package.packagedElement(P, C);
    No == count find hasProperty(C, _);
}

Both cases seems related to the incorrect handling of running variables.

[ujhelyiz]

As the issue is fixed already, it would make sense to include it in 0.6.6.

[istvanrath]

@ujhelyiz please check whether this works correctly.

[okrosa]

I have seen failing a testcase (testGood8 from the cartesian tests) in my pc once more today. After restarting the eclipse the error was gone again. So I recommend that we should leave the bug open for a bit more. I'll try to make a testcase which fails or passes more deterministically.

2012/10/16 Istvan Rath notifications@github.com

@ujhelyiz https://github.com/ujhelyiz please check whether this works correctly.

— Reply to this email directly or view it on GitHubhttps://github.com/ujhelyiz/EMF-IncQuery/issues/294#issuecomment-9483890.

[okrosa]

I just found the bug in the validator, commited in, seems to be working reliably. Some minor tweak needed to handle the unnamed variables in the union-find algorithm correctly. It was undeterministical, only come out when a set started with an unnamed variable. I set the bug to resolved now.

[ujhelyiz]

This issue is already merged and seems to be working without issues. Closing this ticket.