The back-end drops vectorized function calls coming from NF-flattened models

[casella]

Please check ​Modelica.Fluid.Examples.Explanatory.MeasuringTemperature. The flattened model is balanced, but when finally getting to matching, the back-end complains about an unbalanced system

Error: Too few equations, under-determined system. The model has 164 equation(s) and 170 variable(s).

I managed to reproduced the problem with a simplified model where I removed some of the components, please find it attached for reference.

The old FE outputs this flattened equation

  openTankHot3.portInDensities[1] = Modelica.Fluid.Vessels.OpenTank$openTankHot3.Medium.density(Modelica.Fluid.Vessels.OpenTank$openTankHot3.Medium.setState_phX(openTankHot3.vessel_ps_static[1], junctionIdeal.port_2.h_outflow, {}, 0, 0));

while the NF flattens it to

  openTankHot3.portInDensities[1] = MeasuringTemperature4.openTankHot3.Medium.density(MeasuringTemperature4.openTankHot3.Medium.setState_phX(openTankHot3.vessel_ps_static[1], junctionIdeal.port_2.h_outflow, {}, 0, 0));

The backend processes the equation flattened by the OF correctly, and in fact the original system ​simulates successfully and ​passess verification.

When confronted with the equation generated by the NF, the normalInlineFunction module of the backend instead turns it into

51/51 (1): openTankHot3.portInDensities[1] = {}   [dynamic |0|0|0|0|] 

which is obviously wrong, and eventually causes the backend to fail.

I guess there is something missing in the data structure output by the NF which is not immediately apparent from the textural output, but which causes the inlining to fail.

I only carried out the analysis for this model, but I suspect that there are many others affected in the Fluid library and in all libraries (e.g. Buildings) using it.

[perost]

Fixed in a391a807.

[casella]

We got 418 improvements and only 6 regressions, possibly combined with the other one mentioned in ​this report.

A small step for a men, a ​giant leap for OpenModelica :)

[casella]

With ​these two commits the coverage of ​ThermoPower also improved dramatically. We now have 55/103 models simulating, up from 17/103 before the commits. I'm not sure which of the two did it, but it doesn't matter much.