Improve start-value priority by considering parameters.

And add example.
Closes modelica#3008

chapters/equations.tex

@@ -887,4 +887,31 @@ \subsection{Recommended selection of start values}\label{recommended-selection-o
 The model component hierarchy level number is used to give \lstinline!start!-attribute a confidence number, where a lower number means that the \lstinline!start!-attribute is more confident.
 Loosely, if the \lstinline!start!-attribute is set or modified on level $i$ then the confidence number is $i$.
 If a \lstinline!start!-attribute is set by a possibly hierarchical modifier at the top level, then this \lstinline!start!-attribute has the highest confidence, namely 1 irrespectively on what level, the variable itself is declared.
+In case this process is \emph{tied} and at least one of the \lstinline!start!-attributes is equal to a parameter, which may be equal to another parameter (etc), the confidence number is propagated through these bindings and used to break the tie.
+Note that this is only applied if the expression is exactly the parameter - not an expression depending on one or more parameters.
+
+\begin{example}
+\begin{lstlisting}[language=modelica]
+model M1
+  Real x(start = 4.0);
+  Real y(start = 5.0);
+equation
+  x=y;
+end M1;
+model M2
+  parameter Real xStart = 4.0;
+  parameter Real yStart = 5.0;
+  Real x(start = xStart);
+  Real y(start = yStart);
+equation
+  x=y;
+end M2;
+M1 m1(x(start = 3.0)); 
+// Using m1.x.start = 3.0 with confidence number 1
+// over m1.y.start = 5.0 with confidence number 2.
+M2 m2(xStart = 3.0); 
+// Using m2.x.start = m2.xStart = 3.0 with confidence number 2, and tie-breaker 1
+// over m2.y.start = m2.yStart = 5.0 with confidence number 2, and tie-breaker 1
+\end{lstlisting}
+\end{example}
 \end{nonnormative}