#92 Missing * in FieldField.C

[Erik Nijkamp]

Hi all,

I am working on updating an old solver. Namely, I am updating Pascal Beckstein's eddyCurrentFoam, which you can learn more about here. The TL;DR is that is solves for the eddy current created by an alternating magnetic field, however that is not very relevant. This code is quite old, and was written for foam-extend 4.1. I wanted to update it to 5.0, as, as far as I know, 4.1 must be complied with gcc 7, which is not readily available anymore, as well as other reasons. Before I dive into the technical element, let me preface that I am not that experienced with coding in C++, and even less knowledgeable about the foam-extend framework. So please forgive any mistakes, and tell me what you think.

While updating, I was confronted with the following error. In short, (I believe that) the list being referenced to in line 411 is a pointer to the list, rather than the list itself.

In file included from /home/ubuntu/foam/foam-extend-5.0/src/foam/lnInclude/List.H:43,
                 from /home/ubuntu/foam/foam-extend-5.0/src/foam/lnInclude/HashTable.C:30,
                 from /home/ubuntu/foam/foam-extend-5.0/src/foam/lnInclude/HashTable.H:557,
                 from eddyCurrentApp/eddyCurrentAppInclude.H:36,
                 from eddyCurrentApp/eddyCurrentApp.H:35,
                 from eddyCurrentFoam.C:35:
/home/ubuntu/foam/foam-extend-5.0/src/foam/lnInclude/FieldField.C: In instantiation of ‘void Foam::FieldField<Field, Type>::operator=(const Foam::tmp<Foam::FieldField<Field, Type> >&) [with Field = Foam::Field; Type = double]’:
include/AVLoop_AEqn.H:150:81:   required from here
/home/ubuntu/foam/foam-extend-5.0/src/foam/lnInclude/UList.H:411:36: error: request for member ‘size’ in ‘(Foam::FieldField<Foam::Field, double>*)this’, which is of pointer type ‘Foam::FieldField<Foam::Field, double>*’ (maybe you meant to use ‘->’ ?)
  411 |     for (Foam::label i=0; i<(list).size(); i++)
      |                             ~~~~~~~^~~~
/home/ubuntu/foam/foam-extend-5.0/src/foam/lnInclude/FieldField.C:321:5: note: in expansion of macro ‘forAll’
  321 |     forAll (this, i)
      |     ^~~~~~

I looked to the source, FieldField.C, (in 5.0) and I saw the (somewhat funny) section below, around line 300.

template<template<class> class Field, class Type>
void FieldField<Field, Type>::operator=(const tmp<FieldField>& tf)
{
    if (this == &(tf()))
    {
        FatalErrorInFunction
            << "attempted assignment to self"
            << abort(FatalError);
    }

    // This is dodgy stuff, don't try this at home.
    // It hurt me.  HJ, 2/May/2022 lol
    // FieldField* fieldPtr = tf.ptr();
    // PtrList<Field<Type> >::transfer(*fieldPtr);
    // delete fieldPtr;

    // Boundary field is transferred, avoiding assignment operators.
    // This disables virtual functions, which breaks the code.
    // Use normal assignment instead.  HJ, 3/May/2022
    const FieldField<Field, Type>& f = tf();

    forAll (this, i)
    {
        this->operator[](i) = f[i];
    }

    tf.clear();
}

This replaced the following dodgy code from 4.1.

template<template<class> class Field, class Type>
void FieldField<Field, Type>::operator=(const tmp<FieldField>& tf)
{
    if (this == &(tf()))
    {
        FatalErrorIn
        (
            "FieldField<Field, Type>::operator=(const tmp<FieldField>&)"
        )   << "attempted assignment to self"
            << abort(FatalError);
    }

    // This is dodgy stuff, don't try this at home.
    FieldField* fieldPtr = tf.ptr();
    PtrList<Field<Type> >::transfer(*fieldPtr);
    delete fieldPtr;
}

As mentioned, when compiling 5.0, I get a pointer problem. However, if I change the end of the 5.0 section to the following, then the program complies. Note the addition of the * saying that the input should be a pointer.

forAll (*this, i)
    {
        this->operator[](i) = f[i];
    }

    tf.clear();
}

For what it is worth, I also tried to change line 411 in UList.H to take a pointer, as the error message suggests, however that creates other errors, as I suppose that the function is used elsewhere (correctly)

So, to me, it seems that the problem was simply a missing * , however I do not have the foresight to see if this will have other changes, or if it is the best way to solve my problem.

I am curious to hear what other people have to say about this problem and solution. I assigned this to HJ because they changed the lines that I referenced here.

Best,

Erik Nijkamp

[Hrvoje Jasak]

Fixed in nextRelease, fc9d091d21dedbdfb2d5d63b4221b67b9ef493f6