Report parameters with fixed = true and no binding equations

[Francesco Casella]

From Modelica 3.3 specification, Section 8.6:

If a parameter has a start-expression, does not have fixed=false,
and neither has a binding equation nor is part of a record having
a binding equation, the start-expression can be used as parameter-
expression although a diagnostic message is recommended when
initializing the model without setting the parameter value

Unfortunately we forgot to port this addition also into Modelica 3.2rev2. Anyway, this is the default behaviour in Dymola, and it is perfectly legal. OpenModelica doesn't take this rule into account, resulting in under-specified initial equation systems that often lead to division-by-zero errors when tackled by the new BLT-based initialization solver.

I would recommend to implement this feature asap, as it can fix many test cases that currently fail to initialize.

[Lennart Ochel]

I prepared a small model that covert hat issue:

model test
  parameter Real C(start=1);
  parameter Real R(start=1);

  Real u(start=5);
  Real i(start=1), u1, u2;

initial equation
  der(u1) = 0;

equation
  i = C*der(u1);
  u2 = R*i;
  u = u1 + u2;
  u = sin(time);
end test;

[Lennart Ochel]

The function BackendVariable.varFixed(var) returned only true for parameters, if either the fixed attribute was explicitly given or a binding expression was defined. I fixed that.
Also I added a warning for this special cases to the initialization.
I run the test suite and if everything is working as expected, I will commit my changes.

[Lennart Ochel]

I added the test from above here: testsuite/simulation/modelica/initialization/parameterWithoutBinding.mos

Fixed in r18913.

[Lennart Ochel]

Well, I should improve the warning a bit:
Warning: Parameter C:VARIABLE(start = 1.0 fixed = true ) = 1.0 .initializationTests.parameterWithoutBinding, .Real type: Real has no value, and is fixed during initialization (fixed=true), using available start value (start=1.0) as default value.

Better would be
Warning: Parameter C has no value, and is fixed during initialization (fixed=true), using available start value (start=1.0) as default value.

Anyway, it is good enough for now.

[Adrian Pop]

I started a new Windows nightly so Francesco can play with it.
I updated the changes to AVM tests due to new warnings.

[Francesco Casella]

I would encourage Lennart to improve the warning as he planned

[Adrian Pop]

Fixed in r18932. Included the line number info too.