Segmentation fault at zero mass flow rate

[Michael Wetter]

The following example gives a segmentation fault:

Run

export OPENMODELICALIBRARY=`pwd`:/usr/lib/omlibrary
omc +d=nogen,initialization,backenddaeinfo,discreteinfo,stateselection openmod.mos

with openmod.mos containing

loadModel(Modelica, {"3.2.1"});
getErrorString();
loadModel(Buildings); getErrorString();

simulate(Buildings.Fluid.HeatExchangers.Examples.DryEffectivenessNTU); getErrorString();

The output is

record SimulationResult
    resultFile = "",
    simulationOptions = "startTime = 0.0, stopTime = 360.0, numberOfIntervals = 500, tolerance = 1e-06, method = 'dassl', fileNamePrefix = 'Buildings.Fluid.HeatExchangers.Examples.DryEffectivenessNTU', options = '', outputFormat = 'mat', variableFilter = '.*', cflags = '', simflags = ''",
    messages = "Simulation execution failed for model: Buildings.Fluid.HeatExchangers.Examples.DryEffectivenessNTU
assert            | debug   | division by zero at time 0.72, (a=0) / (b=0), where divisor b expression is: hexPar.m2_flow
assert            | debug   | division by zero at time 0, (a=0) / (b=0), where divisor b expression is: hexPar.m2_flow

Limited backtrace at point of segmentation fault
/lib/x86_64-linux-gnu/libpthread.so.0(+0x10340)[0x7f4c7d71c340]
/lib/x86_64-linux-gnu/libc.so.6(_longjmp+0x10)[0x7f4c7d37db40]
/lib/x86_64-linux-gnu/libpthread.so.0(+0xed69)[0x7f4c7d71ad69]
/usr/lib/x86_64-linux-gnu/omc/libSimulationRuntimeC.so(throwStreamPrintWithEquationIndexes+0xfc)[0x7f4c7e9d39fc]
/usr/lib/x86_64-linux-gnu/omc/libSimulationRuntimeC.so(+0x29a77)[0x7f4c7e9d3a77]
/home/mwetter/proj/ldrd/bie/modeling/github/lbl-srg/modelica-buildings/Buildings.Fluid.HeatExchangers.Examples.DryEffectivenessNTU(Buildings_Fluid_HeatExchangers_Examples_DryEffectivenessNTU_eqFunction_1139+0x18d)[0x4782bd]
/home/mwetter/proj/ldrd/bie/modeling/github/lbl-srg/modelica-buildings/Buildings.Fluid.HeatExchangers.Examples.DryEffectivenessNTU(Buildings_Fluid_HeatExchangers_Examples_DryEffectivenessNTU_functionDAE+0x8e2)[0x486ab2]
/usr/lib/x86_64-linux-gnu/omc/libSimulationRuntimeC.so(updateDiscreteSystem+0x8d)[0x7f4c7ea0cbdd]
/home/mwetter/proj/ldrd/bie/modeling/github/lbl-srg/modelica-buildings/Buildings.Fluid.HeatExchangers.Examples.DryEffectivenessNTU[0x46a744]
/home/mwetter/proj/ldrd/bie/modeling/github/lbl-srg/modelica-buildings/Buildings.Fluid.HeatExchangers.Examples.DryEffectivenessNTU(Buildings_Fluid_HeatExchangers_Examples_DryEffectivenessNTU_performSimulation+0x574)[0x469c24]
/usr/lib/x86_64-linux-gnu/omc/libSimulationRuntimeC.so(solver_main+0x17c)[0x7f4c7ea1433c]
/usr/lib/x86_64-linux-gnu/omc/libSimulationRuntimeC.so(callSolver+0x2c9)[0x7f4c7ea36f49]
/usr/lib/x86_64-linux-gnu/omc/libSimulationRuntimeC.so(startNonInteractiveSimulation+0x5c8)[0x7f4c7ea359c8]
/usr/lib/x86_64-linux-gnu/omc/libSimulationRuntimeC.so(_main_SimulationRuntime+0x70)[0x7f4c7ea37fa0]
/home/mwetter/proj/ldrd/bie/modeling/github/lbl-srg/modelica-buildings/Buildings.Fluid.HeatExchangers.Examples.DryEffectivenessNTU(main+0x1f7)[0x487667]
Segmentation fault (core dumped)
... [omitted]

Setting POut.k=101225 avoids zero mass flow rate, in which case there is no segmentation fault.

The model is from the latest version of the master branch of Buildings, e.g., ​https://github.com/lbl-srg/modelica-buildings/commit/d0841740172f2812891be5693a9cc4450e1327b8

[anonymous]

How about

 if noEvent(hexPar.m2_flow > 0) then a/b else k_start

[Michael Wetter]

Replying to anonymous:

How about

 if noEvent(hexPar.m2_flow > 0) then a/b else k_start

Can you be a bit more specific? I see no k_start. Do you suggest this to be in the model, or in the generated code, and if in the model, where did you find k_start?

[Francesco Casella]

As of OMCompiler v1.12.0-dev.195+gd2bbe7a the following outcome is reported:

assert | division by zero at time 0.07199999999999999, 
(a=0) / (b=0), where divisor b expression is: hexCroUnm.m2_flow

I don't know if this is the expected result, but at least there is no longer a segfault.

[Michael Wetter]

I believe this comes from the code below

/*
 equation index: 1935
 type: SIMPLE_ASSIGN
 hexPar._port_b2._h_outflow = sou_2.ports[5].h_outflow + DIVISION(hexPar.Q2_flow, hexPar.m2_flow)
 */
void Buildings_Fluid_HeatExchangers_Examples_DryEffectivenessNTU_eqFunction_1935(DATA *data, threadData_t *threadData)
{
  TRACE_PUSH
  const int equationIndexes[2] = {1,1935};
  data->localData[0]->realVars[301] /* hexPar._port_b2._h_outflow variable */ = data->localData[0]->realVars[353] /* sou_2._ports[5]._h_outflow variable */ + DIVISION_SIM(data->localData[0]->realVars[262] /* hexPar._Q2_flow variable */,data->localData[0]->realVars[295] /* hexPar._m2_flow variable */,"hexPar.m2_flow",equationIndexes);
  TRACE_POP
}

Both arguments of the DIVISION macro are zero. It would be good if OpenModelica could propagate the specific enthalpy if there is zero flow. Zero flow rate is common for initialization of fluid flow models in building applications.