[chriss85]

I'm seeing an odd error where sometimes, when calling virtual tmp<scalarField> kappaEff(const label patchI) const in RASModel.H, the size of the alphat(patchI) field is zero. This causes a crash in return thermo().kappaEff(at, patchI); because it tries to calculate the sum of fields of different sizes. So far I have only noticed this problem in parallel runs and it does not occur every time.

Inside of the virtual tmp<scalarField> alphat(const label patchI) const function in the compressible turbulenceModel.H, the size of the alphat patch field is correct on each processor core, but after returning it sometimes becomes empty which leads to the crash.

I'm wondering if this is because the tmp<volScalarField> returned from alphat() goes out of scope and gets deleted, but I don't know what happens to the boundary fields when the parent volScalarField is deleted. On return there should only be a tmp<scalarField> object pointing to the patch field, but no valid pointer or tmp object which points to the parent.

Could this be a reason? If yes, then this is possibly a bug. A possible fix could be to store the volScalarField in the class instead of using a temporary object so it doesn't get invalid during runtime.

Any comments or helpful explanations? I've put quite some thought into the memory management now but I think I can't manage to grasp the whole picture at once

In short: returning the boundary patch field of a temporary volScalarField results in erroneous behavior, because the data gets invalidated. At least that's what I think is happening.
I have now tried to store a permanent tmp<volScalarField> object in the laminar turbulence model class and it does not seem to happen anymore, so I am slightly optimistic. I think this problem might also exist in more functions in this class, so if this turns out to fix it I will create a bug report.

[chriss85]

Report: http://www.openfoam.org/mantisbt/view.php?id=1686

Let me know if there is anything else about this topic.