Questionable parametrization of V_pulse in Spice3 library leads to numerically unreliable computations #4478
Description
While analyzing possible regressions from MSL 4.0.0 to 4.1.0 in OpenModelica, we found issues with several Spice3 model:
Modelica.Electrical.Spice3.Examples.Inverter
Modelica.Electrical.Spice3.Examples.FourInverters
Modelica.Electrical.Spice3.Examples.InvertersApartRecord
Modelica.Electrical.Spice3.Examples.InvertersExtendedModel
Modelica.Electrical.Spice3.Examples.Nor
Modelica.Electrical.Spice3.Examples.Spice3BenchmarkFourBitBinaryAdder
all failing at the beginning of the simulation with this runtime error:
division by zero at time 1.000000000300011, (a=inf) / (b=0), where divisor b expression is: v.Tfalling - v.Twidth
I checked the Inverter model more in depth. It contains an instance v of the Modelica.Electrical.Spice3.Sources.V_pulse model:
ModelicaStandardLibrary/Modelica/Electrical/Spice3.mo
Lines 3129 to 3159 in 8e7876b
The component v is instantiated as:
so both
v.PW and v.PER retain their default value of Modelica.Constants.inf. The resulting chain of parameter assignments is thus:
v.Trising := v.TR = 1e-13
v.Twidth := v.Trising + v.PW = 1e-13 + 1.7e308
v.Tfalling := v.Twidth + v.TF + v.TR = 1e-13 + 1.7e308 + 1e-13it is then obvious that the computation of the denominator v.Tfalling - v.Twidth = 1e-13 + 1.7e308 + 1e-13 - (1e-13 + 1.7e308) in the expression for v.v is numerically ill-defined. In fact, it looks quite odd to me that this issue hasn't been raised before.
I also checked the documentation of the component, which is not very clear as it states "All parameters of sources should be set explicitly"; if "set explicitly" meant "have explicit modifiers", then there should be no defaults.
In any case, if a MSL component has default values, it should work properly when those values are not changed, and this is obviously not the case with this component, so we should do something about it.