Flattenning of Modelica.Mechanics.MultiBody does not work for all components

[Jan Brugård]

The following models do not work to check, because flattening fails:

Check of Modelica.Mechanics.MultiBody.Frames.Quaternions.multipleResolve1 failed
Check of Modelica.Mechanics.MultiBody.Frames.Quaternions.multipleResolve2 failed
Check of Modelica.Mechanics.MultiBody.Frames.Quaternions.planarRotation failed
Check of Modelica.Mechanics.MultiBody.Frames.TransformationMatrices.multipleResolve1 failed
Check of Modelica.Mechanics.MultiBody.Frames.TransformationMatrices.multipleResolve2 failed

Check of Modelica.Mechanics.MultiBody.Interfaces.PartialLineForce failed

Check of Modelica.Mechanics.MultiBody.Types.RotationSequence failed

[Jan Brugård]

MathCore ticket: ​http://intranet/trac/mathmodelica/ticket/2957

[Adrian Pop]

Thanks to the fixes already made for bug #1273 [BUG:1273] and #1274 [BUG:1274], everything except for RotationSequence now works.

[Per Östlund]

RotationSequence fixed in r6134.
checkModel now works on all MultiBody models, see test case libraries/msl32/Modelica.Mechanics.MultiBody.

[dhedberg]

Run the mos file and check the results. OpenModelica produces the following for me:

true
instantiateDaeFunction failed for Math.sin in scope: Modelica.Mechanics.MultiBody.Frames.Quaternions.planarRotation
instantiateDaeFunction failed for Math.sin in scope: Modelica.Mechanics.MultiBody.Frames.Quaternions.planarRotation
"Error: No matching function found for Math.sin(angle / 2.0) of type function(x:Real) => Real in component <NO COMPONENT>, candidates are function(x:Real) => Real
Error: Class Math.sin not found in scope Modelica.Mechanics.MultiBody.Frames.Quaternions.planarRotation.
Error: Wrong type or wrong number of arguments to vector([e * Math.sin(angle / 2);Math.cos(angle / 2)])'.
 (in component <NO COMPONENT>)
Error: No matching function found for Math.sin(angle / 2.0) of type function(x:Real) => Real in component <NO COMPONENT>, candidates are function(x:Real) => Real
Error: Class Math.sin not found in scope Modelica.Mechanics.MultiBody.Frames.Quaternions.planarRotation.
Error: Wrong type or wrong number of arguments to vector([e * Math.sin(angle / 2);Math.cos(angle / 2)])'.
 (in component <NO COMPONENT>)
"
""

So it does not seem to work as suggested, hence reopening this ticket.

[dhedberg]

Run the mos file and check the results. OpenModelica produces the following for me:

true
instantiateDaeFunction failed for Math.sin in scope: Modelica.Mechanics.MultiBody.Frames.Quaternions.planarRotation
instantiateDaeFunction failed for Math.sin in scope: Modelica.Mechanics.MultiBody.Frames.Quaternions.planarRotation
"Error: No matching function found for Math.sin(angle / 2.0) of type function(x:Real) => Real in component <NO COMPONENT>, candidates are function(x:Real) => Real
Error: Class Math.sin not found in scope Modelica.Mechanics.MultiBody.Frames.Quaternions.planarRotation.
Error: Wrong type or wrong number of arguments to vector([e * Math.sin(angle / 2);Math.cos(angle / 2)])'.
 (in component <NO COMPONENT>)
Error: No matching function found for Math.sin(angle / 2.0) of type function(x:Real) => Real in component <NO COMPONENT>, candidates are function(x:Real) => Real
Error: Class Math.sin not found in scope Modelica.Mechanics.MultiBody.Frames.Quaternions.planarRotation.
Error: Wrong type or wrong number of arguments to vector([e * Math.sin(angle / 2);Math.cos(angle / 2)])'.
 (in component <NO COMPONENT>)
"
""

So it does not seem to work as suggested, hence reopening this ticket.

[dhedberg]

See attachment, planarRotation still seems to produce errors. I tested this with an OpenModelica compiled for mingw some days old.

[Per Östlund]

I am unable to reproduce this. checkModel of planarRotation succeeds without any errors with the latest trunk, and has done so since I fixed this bug and added the test case.

[Adrian Pop]

Your attached model is not TOTAL (not complete).
In package Math you're missing the sin function,
i.e. Math.sin and this is the error you're getting.
Fix your dependency analysis!

[dhedberg]

OpenModelica suffers from the same problem. The only reason you do not notice it is because you do not do a dependency analysis when you save a total model. You collect and save everything. Anyways, I fixed the problem which was in OpenModelica Absyn.traverseExp, see commit/revision 6387. Closing this ticket.

[Adrian Pop]

We know that OpenModelica has the same problem.
But as we never used the dependency analysis the problem doesn't exist.
I would really suggest you do also a checking if is really faster do do the dependency
instead of directly instantiating the class. For me it seems *really really* weird that
is faster to do dependency+smaller instantiation rather than directly the full instantiation.
The problem is not the strategy, which is sound, the problem is with the implementation
which does instantiation during dependency, so:

 Dependency analysis includes instantiation
 O(dependency) = O(smaller instantiation) + O(traverse)
 The real question is:
 O(dependency) + O(smaller instantiation) > O(full instantiation)
 or
 O(dependency) + O(smaller instantiation) < O(full instantiation)
 ??

I don't know the answer to this question, but you could run some checks.
From what tests I ran, doing the dependency analysis is *always* slower than
doing full instantiation directly, but I might be wrong.
BTW, to help you a bit I added a testsuite/dependency directory to check
a bit if the dependency works fine, so if we break it at least we know when
(what revision).
I will add the instantiation of planarRotation to the dependency directory
so we make sure we don't break it in the future.

[Adrian Pop]

I added more tests to testsuite/dependency.
Note that Modelica.Mechanics.MultiBody.Interfaces.PartialLineForce
does not work yet if +d=usedep is used, so there are still problems
with the dependency analysis.