Issue in preOpt evalFunc involving functions with record outputs containing Real and Integer variables

[Francesco Casella]

Please check ​HelmholtzMedia.Examples.ConvergenceTest.SinglePhase_setState_b. The back-end fails during the preOpt evalFunc phase with a stack overflow error that seems to point to an infinite loop, since the sequence

[bt] #534 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateConstantFunction)
[bt] #535 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateConstantFunctionWrapper)
[bt] #536 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(boxptr_EvaluateFunctions_evaluateConstantFunctionWrapper)
[bt] #537 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(Expression.traverseExpTopDown)
[bt] #538 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateFunctions_updateStatement)
[bt] #539 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateFunctions_updateAllStatements)
[bt] #540 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.predictIfOutput)
[bt] #541 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateFunctions_updateStatement)
[bt] #542 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateFunctions_updateAllStatements)
[bt] #543 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.predictIfOutput)
[bt] #544 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateFunctions_updateStatement)
[bt] #545 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateIfStatement)
[bt] #546 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateFunctions_updateStatement)
[bt] #547 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateFunctions_updateAlgElements)
[bt] #548 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(boxptr_EvaluateFunctions_evaluateFunctions__updateAlgElements)
[bt] #549 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(List.mapFold3)
[bt] #550 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateConstantFunction)
[bt] #551 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateConstantFunctionWrapper)
[bt] #552 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(boxptr_EvaluateFunctions_evaluateConstantFunctionWrapper)
[bt] #553 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(Expression.traverseExpTopDown)
[bt] #554 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateFunctions_updateStatement)
[bt] #555 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateFunctions_updateAllStatements)
[bt] #556 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.predictIfOutput)
[bt] #557 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateFunctions_updateStatement)
[bt] #558 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(EvaluateFunctions.evaluateFunctions_updateAlgElements)
[bt] #559 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(boxptr_EvaluateFunctions_evaluateFunctions__updateAlgElements)
[bt] #560 /var/lib/hudson/slave/workspace/OpenModelica_TEST_LIBS/OpenModelica/OMCompiler/build/bin/../lib/x86_64-linux-gnu/omc/libOpenModelicaCompiler.so(List.mapFold3)

repeats endlessly.

The following test model:

model TestHelmholtzMedium_ph
  package Medium = HelmholtzMedia.HelmholtzFluids.Carbondioxide;
  Medium.AbsolutePressure p;
  Medium.SpecificEnthalpy h;
  Medium.ThermodynamicState state;
equation
  p = 1e5*(1+time);
  h = 1e5+50*time;
  state = Medium.setState_ph(p,h);
end TestHelmholtzMedium_ph;

reproduces the problem, while this other model

model TestHelmholtzMedium_pT
  package Medium = HelmholtzMedia.HelmholtzFluids.Carbondioxide;
  Medium.AbsolutePressure p;
  Medium.Temperature T;  
  Medium.ThermodynamicState state;
equation
  p = 1e5*(1+time);
  T = 300+50*time;
  state = Medium.setState_pT(p,T);
end TestHelmholtzMedium_pT;

passes the back-end successfully. It then fails due to #5167, but that's a separate issue. Please note you need -d=newInst to run both models, as the old front-end fails because of look-up issues.

When processing TestHelmholtzMedium_pT with -d=optdaedump, the following output is produced:

########################################
pre-optimization module resolveLoops (simulation)
########################################

unspecified partition
========================================

Variables (9)
========================================
1: state.s:VARIABLE(flow=false min = 0.52129 max = 1e+060 start = 5e+059 unit = "J/(kg.K)" nominal = 1000.0 )  "Specific entropy of medium" type: Real 
2: state.h:VARIABLE(flow=false min = -100000.0 max = 1300000.0 start = 600000.0 unit = "J/kg" nominal = 1000000.0 )  "Specific enthalpy of medium" type: Real 
3: state.u:VARIABLE(flow=false min = -100000000.0 max = 100000000.0 unit = "J/kg" nominal = 1000000.0 )  "Specific inner energy of medium" type: Real 
4: state.d:VARIABLE(flow=false min = 1e-060 max = 1638.9 start = 467.6 unit = "kg/m3" nominal = 1.0 )  "Density of medium" type: Real 
5: state.p:VARIABLE(flow=false min = 0.0 max = 801000000.0 start = 101325.0 unit = "Pa" nominal = 100000.0 )  "Absolute pressure of medium" type: Real 
6: state.T:VARIABLE(flow=false min = 216.592 max = 2000.0 start = 298.15 unit = "K" nominal = 300.0 )  "Temperature of medium" type: Real 
7: state.phase:DISCRETE(flow=false min = 0 max = 2 start = 0 )  "Phase of the fluid: 1 for 1-phase, 2 for two-phase, 0 for not known, e.g., interactive use" type: Integer 
8: T:VARIABLE(flow=false min = 216.592 max = 2000.0 start = 298.15 unit = "K" nominal = 300.0 )  type: Real 
9: p:VARIABLE(flow=false min = 0.0 max = 801000000.0 start = 101325.0 unit = "Pa" nominal = 100000.0 )  type: Real 

Equations (3, 9)
========================================
1/1 (1): p = 100000.0 * (1.0 + time)   [dynamic |0|0|0|0|]
2/2 (1): T = 300.0 + 50.0 * time   [dynamic |0|0|0|0|]
3/3 (7): state = TestHelmholtzMedium_pT.Medium.setState_pTX(p, T, {}, 0)   [dynamic |0|0|0|0|]

########################################
pre-optimization module evalFunc (simulation)
########################################


unspecified partition
========================================

Variables (8)
========================================
1: state.s:VARIABLE(flow=false min = 0.52129 max = 1e+060 start = 5e+059 unit = "J/(kg.K)" nominal = 1000.0 )  "Specific entropy of medium" type: Real 
2: state.h:VARIABLE(flow=false min = -100000.0 max = 1300000.0 start = 600000.0 unit = "J/kg" nominal = 1000000.0 )  "Specific enthalpy of medium" type: Real 
3: state.u:VARIABLE(flow=false min = -100000000.0 max = 100000000.0 unit = "J/kg" nominal = 1000000.0 )  "Specific inner energy of medium" type: Real 
4: state.d:VARIABLE(flow=false min = 1e-060 max = 1638.9 start = 467.6 unit = "kg/m3" nominal = 1.0 )  "Density of medium" type: Real 
5: state.p:VARIABLE(flow=false min = 0.0 max = 801000000.0 start = 101325.0 unit = "Pa" nominal = 100000.0 )  "Absolute pressure of medium" type: Real 
6: state.T:VARIABLE(flow=false min = 216.592 max = 2000.0 start = 298.15 unit = "K" nominal = 300.0 )  "Temperature of medium" type: Real 
7: T:VARIABLE(flow=false min = 216.592 max = 2000.0 start = 298.15 unit = "K" nominal = 300.0 )  type: Real 
8: p:VARIABLE(flow=false min = 0.0 max = 801000000.0 start = 101325.0 unit = "Pa" nominal = 100000.0 )  type: Real 


Equations (3, 8)
========================================
1/1 (1): p = 100000.0 * (1.0 + time)   [dynamic |0|0|0|0|]
2/2 (1): T = 300.0 + 50.0 * time   [dynamic |0|0|0|0|]
3/3 (6): (state.T, state.p, state.d, state.u, state.h, state.s) = estHelmholtzMedium_pT.Medium.setState_pTX_eval1(p, T, {}, 0)   [dynamic |0|0|0|0|]

BackendDAEType: simulation


Known variables only depending on parameters and constants - globalKnownVars (5)
========================================
1: Medium.fluidConstants.molarMass:CONST(flow=false )  "Molar mass" type: Real  [1]
2: Medium.fluidConstants.gasConstant:CONST(flow=false )  type: Real  [1]
3: Medium.fluidConstants.criticalTemperature:CONST(flow=false )  "Critical temperature" type: Real  [1]
4: Medium.fluidConstants.criticalMolarVolume:CONST(flow=false )  "Critical molar Volume" type: Real  [1]
5: state.phase:DISCRETE(flow=false min = 0 max = 2 start = 0 )  = 1  "Phase of the fluid: 1 for 1-phase, 2 for two-phase, 0 for not known, e.g., interactive use" type: Integer 

Apparently, the back-end can figure out that the discrete variable state.phase in the setState_pT() function has a fixed value (because it only contains one assignment state.phase := 1 and factors that out, computing it as a constant. BTW, this is quite remarkable.

In the case of TestHelmholtzMedium_ph, the state.phase output is instead depending on the values of p and h, so this trick doesn't work. The following output is produced:

########################################
pre-optimization module resolveLoops (simulation)
########################################


unspecified partition
========================================

Variables (9)
========================================
1: state.s:VARIABLE(flow=false min = 0.52129 max = 1e+060 start = 5e+059 unit = "J/(kg.K)" nominal = 1000.0 )  "Specific entropy of medium" type: Real 
2: state.h:VARIABLE(flow=false min = -100000.0 max = 1300000.0 start = 600000.0 unit = "J/kg" nominal = 1000000.0 )  "Specific enthalpy of medium" type: Real 
3: state.u:VARIABLE(flow=false min = -100000000.0 max = 100000000.0 unit = "J/kg" nominal = 1000000.0 )  "Specific inner energy of medium" type: Real 
4: state.d:VARIABLE(flow=false min = 1e-060 max = 1638.9 start = 467.6 unit = "kg/m3" nominal = 1.0 )  "Density of medium" type: Real 
5: state.p:VARIABLE(flow=false min = 0.0 max = 801000000.0 start = 101325.0 unit = "Pa" nominal = 100000.0 )  "Absolute pressure of medium" type: Real 
6: state.T:VARIABLE(flow=false min = 216.592 max = 2000.0 start = 298.15 unit = "K" nominal = 300.0 )  "Temperature of medium" type: Real 
7: state.phase:DISCRETE(flow=false min = 0 max = 2 start = 0 )  "Phase of the fluid: 1 for 1-phase, 2 for two-phase, 0 for not known, e.g., interactive use" type: Integer 
8: h:VARIABLE(flow=false min = -100000.0 max = 1300000.0 start = 600000.0 unit = "J/kg" nominal = 1000000.0 )  type: Real 
9: p:VARIABLE(flow=false min = 0.0 max = 801000000.0 start = 101325.0 unit = "Pa" nominal = 100000.0 )  type: Real 


Equations (3, 9)
========================================
1/1 (1): p = 100000.0 * (1.0 + time)   [dynamic |0|0|0|0|]
2/2 (1): h = 100000.0 + 50.0 * time   [dynamic |0|0|0|0|]
3/3 (7): state = TestHelmholtzMedium_ph.Medium.setState_phX(p, h, {}, 0)   [dynamic |0|0|0|0|]

then the execution hangs.

From what I understand, there is some issue with the mixed nature of the state output which contains both Real and Integer variables. The back-end tries to figure out the value of state.phase, and in doing so enters an infinite recursion, thus breaking the evalFunc module.

@lochel, can you please have a look at this issue, assess how easy/hard it is to fix it, and possibly re-assign it to someone else who is competent on this part of the back-end?

[Lennart Ochel]

According to the discussion in today’s OMDev meeting, no one from the backend team has time to look into it right now.

[Francesco Casella]

Too bad. Currently, 100% of the test in HelmholtzMedia pass the new front-end, which is a marked improvement compared to the 6% that passed the old front-end.

Unfortunately, 75% fail during back-end processing because of this issue, while the remaining 25% fail at compile time because of a package constant issue (#5167). I guess also those 75% will be plagued by the same problem anyway.

@m.thorade, as soon as #5167 is resolved, we can try to see if we find someone to look into this issue.

We have a commitment to achieve high coverage of open-source Modelica libraries with release 2.0.0, so if fixing this issue brings HelmholtzMedia from 25% to 100% coverage, it will be worth having a look.

[Per Östlund]

It seems I accidentally fixed the issue in e8bea859. But I'm not sure whether this is an actual issue in evalFunc that's no longer triggered, or if it was due to some issue with the flat model produced by the NF.

But since we no longer seem to have any way to reproduce the issue I'll close the ticket for now.