unstable event system

[Lennart Ochel]

The OpenModelica event system seems to be unstable for PNlib.Examples.Test10 (see testsuite/simulation/libraries/3rdParty/PNlib r20361).
The example Test10 works fine with NumberOfIntervals=499 and NumberOfIntervals=501. It seems that there is a problem if the event point (time=1.5) is also a time-grid point. This is the case for example if NumberOfIntervals is set to 498 or 500. In both cases the event system is not able to find the following event:

{PNlib.Types.ArcType.normal_arc, PNlib.Types.ArcType.inhibitor_arc}[2] == PNlib.Types.ArcType.inhibitor_arc and not {P1.t_, P2.t_}[2] < {-1.0, IA1.testValue}[2]

.

[Martin Sjölund]

Is that the generated equation? Why is it not simplified to:

PNlib.Types.ArcType.inhibitor_arc == PNlib.Types.ArcType.inhibitor_arc and not P2.t_ <  IA1.testValue

[Lennart Ochel]

That is the condition of the untriggered event. It is part of an for loop within an algorithm:

algorithm
  active := true;
  weaklyInputActiveVec := fill(false, nIn);
  weaklyOutputActiveVec := fill(false, nOut);
  //check input places
  for i in 1:nIn loop
    if arcType[i] == PNlib.Types.ArcType.normal_arc then
      if tIn[i] <= minTokens[i] and not fed[i] then
        active := false;
      elseif tIn[i] <= minTokens[i] and fed[i] then
        weaklyInputActiveVec[i] := true;
      end if;
    elseif arcType[i] == PNlib.Types.ArcType.inhibitor_arc and not tIn[i] < testValue[i] then
      active := false;
    end if;
  end for;
  // normal arc
  // inhibitor arc
  // invalid
  //output places
  for i in 1:nOut loop
    if tOut[i] >= maxTokens[i] and not emptied[i] then
      active := false;
    elseif tOut[i] >= maxTokens[i] and emptied[i] then
      weaklyOutputActiveVec[i] := true;
    end if;
  end for;

[Per Östlund]

I have disabled the Test10 test case, please enable it again when this is fixed.

[Lennart Ochel]

Events get sometimes triggered a bit too early. This happens for example in Test10 for P2.t_ <= 0. It get triggered at time=1.499999... with P2.t_=9.998990524749211e-11.

[Lennart Ochel]

Replying to perost:

I have disabled the Test10 test case, please enable it again when this is fixed.

Well, I fixed it for the PNlib tests. Unfortunately my changes break a lot of other test cases (178 of 2562 failed). Probably I should deactivate all of them until someone fixes them. ;-)

[Per Östlund]

Replying to lochel:

Replying to perost:

I have disabled the Test10 test case, please enable it again when this is fixed.

Well, I fixed it for the PNlib tests. Unfortunately my changes break a lot of other test cases (178 of 2562 failed). Probably I should deactivate all of them until someone fixes them. ;-)

The difference is that Test10 broke because PNlib was updated, i.e. because of an existing compiler issue. If 178 test cases break with your changes then either all those tests are wrong, or your fix is wrong. Guess which one is more likely ;)

[Lennart Ochel]

Here are two more model, which pinpoint the Problem of the event system:

model test
  Boolean a;
equation
  a = exp(-time) > 0.0;
  annotation (experiment(StartTime=0, StopTime=25));
end test;
// LOG_EVENTS        | info    | state event at time 23.0258509298
// |                 | |       | | [0] exp(-time) > 0.0

model test2
  Boolean a;
equation
  a = 1/(time*time*time*time) > 0.0;
  annotation (experiment(StartTime=1, StopTime=1000));
end test2;
// LOG_EVENTS        | info    | state event at time 316.227766009
// |                 | |       | | [0] time ^ -4.0 > 0.0

Both model shouldn’t trigger any event.

[Lennart Ochel]

It seems that somewhere is a magic border of 1e-10:

316.227766009^(-4)  = 1.000000000*10^(-10)
exp(-23.0258509298) = 9.999999999*10^(-11)

[Lennart Ochel]

Well, it seems that my omc was not clean of changes. Both models trigger no events.

[Lennart Ochel]

Next try:

The tests work the other way around:

model test2
  Boolean a;
equation
  a = 1/(time*time*time*time) > 1e-10;
  annotation (experiment(StartTime=1, StopTime=1000));
end test2;

This case should trigger an event at time=100*sqrt(10) (which is something like 316.2278), but OpenModelica is not able to find it. If the equation for a get replaced by a = 1/(time*time*time*time) > 2e-10; even OpenModelica is able to find an event, but it get triggered too late. The reason is still the magic boarder of 1e-10, because the event is triggered after the zero crossing 1/(time*time*time*time) - 2e-10 becomes lower then -1e-10.

[Martin Sjölund]

This ticket was not closed for 1.9.1, which has now been released. It was batch modified for milestone 1.9.2 (but maybe an empty milestone was more appropriate; feel free to change it).

[Martin Sjölund]

Milestone changed to 1.9.3 since 1.9.2 was released.

[Martin Sjölund]

Moved to new milestone 1.9.4

[Willi Braun]

Is this really a blocker, since we moved it already for 3 releases, and in my option it is a corner case.

[Martin Sjölund]

Milestone pushed to 1.9.5

[Martin Sjölund]

Milestone renamed

[Lennart Ochel]

It seems that all examples in this ticket work as expected now.

[Martin Sjölund]

Milestone deleted