Type compatibility of array expressions within if-expressions

[modelica-trac-importer]

Reported by stefanv on 6 Mar 2017 18:06 UTC Section 6.6 of the 3.3 specification states:

If A is an array expression then B must also be an array expression, and ndims(A)=ndims(B). The type compatible expression an array expression with elements compatible with the elements of both A and B. If both size(A) and size(B) are known and size(A)=size(B) then this defines the size of the type compatible expression, otherwise the size of the expression is not known until the expression is about to be evaluated. In case of an if-expression the size of the type compatible expression is defined based on the branch selected, and for other cases size(A)=size(B) must hold at this point..

There are three issues with this, one that needs further clarification, and two that are just grammatical:

• I think it must be clarified that in the case where the size of the expression is defined based on the selected branch, this branch must be selected by a condition that is a parameter expression. I.e., the if-expression cannot result in different sizes at different times during the simulation of a model.

• The second sentence, "The type compatible expression an array expression with elements compatible with the elements of both A and B.", seems to be missing some word(s).

• There are two periods at the end of the paragraph.

(I have put this down for MLS 3.4 since there is not yet a Design94 milestone.)

---

Migrated-From: https://trac.modelica.org/Modelica/ticket/2165

[modelica-trac-importer]

Comment by hansolsson on 14 Mar 2017 15:08 UTC The second sentence, "The type compatible expression an array expression with elements compatible with the elements of both A and B.", seems to be missing some word(s).

I believe the minimal missing word is "is":

"The type compatible expression is an array expression with elements compatible with the elements of both A and B." -- I added this and corrected the 3rd issue in r9685 (seemed like simple typos). However, the first issue still remains.

[HansOlsson]

After thinking more, I'm not certain the first issue should be changed in the indicated way.

Obviously if you declare

model M
  Real x[:]=if ... then {1,2} else {1,2,3};
end M;

then the if-condition must be evaluated to a specific size for the array.

However, we also have arrays sent to functions, and expressions inside functions - where you could have:

function foo
  input Integer n;
  input Real a,b;
protected
  Real x[n]=if n==1 then {(a+b)/2} else linspace(a,b,n);
...
end foo;

I don't see that n being non-parametric would be a problem inside the function.

Obviously this might be discussed, and depending on decision clarified/changed - but I don't see it for 3.5.

[HansOlsson]

This is also part of #2494 - which highlights an issue with the assumed resolution.