CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > Software User Forums > OpenFOAM > OpenFOAM Bugs

Wrong behaviour when appending vector to vectorField?

Register Blogs Community New Posts Updated Threads Search

Like Tree1Likes
  • 1 Post By alexeym

Reply
 
LinkBack Thread Tools Search this Thread Display Modes
Old   June 3, 2015, 03:20
Default Wrong behaviour when appending vector to vectorField?
  #1
Member
 
Johan Roenby
Join Date: May 2011
Location: Denmark
Posts: 92
Rep Power: 20
roenby will become famous soon enough
This code:

vectorField test;
vector t1(1.0,2.0,3.0);
test.append(t1);
vector& t2 = test[0];
test.append(t2);
Info << "test = " << test << endl;

results in the following output:

test = 2((1 2 3) (0 2 3))

The first component of the appended vector is not appended correctly.

The error does not occur if t2 is a copy of test[0] rather than a reference to it.

Here is the situation where I discovered the bug:

//Calculate a pointField and if it only contains a single point
//append this, so it occurs twice in the pointField
pointField pf;
someFunctionPopulatingPointField(pf);
if (pf.size()==1)
{
pf.append(pf[0]);
}

The bug was reported here:
http://www.openfoam.org/mantisbt/view.php?id=1729#c4882
It was marked as "resolved" by noting that one should work with a copy of instead of a reference to test[0]. As far as I can see this is not a fix but a work-around.

Could some C++ expert comment on whether this is a genuine bug that should be properly fixed or if am I just doing dodgy stuff in my coding?
roenby is offline   Reply With Quote

Old   June 3, 2015, 04:52
Default
  #2
Senior Member
 
Alexey Matveichev
Join Date: Aug 2011
Location: Nancy, France
Posts: 1,930
Rep Power: 38
alexeym has a spectacular aura aboutalexeym has a spectacular aura about
Send a message via Skype™ to alexeym
Hi,

I would call this "undefined behaviour". If you take a look at how append method is implemented (Field is a child of List):

Code:
template<class T>
inline void Foam::List<T>::append(const T& t)
{
    setSize(size()+1, t);
}
and

Code:
template<class T>
void Foam::List<T>::setSize(const label newSize, const T& a)
{
    label oldSize = label(this->size_);
    this->setSize(newSize);

    if (newSize > oldSize)
    {
        register label i = newSize - oldSize;
        register T* vv = &this->v_[newSize];
        while (i--) *--vv = a;
    }
}
So append first resizes list and then set last element to the new value.

Now let's look at setSize implementation (src/OpenFOAM/containers/Lists/List/List.C:318), below is just relevant part

Code:
template<class T>
void Foam::List<T>::setSize(const label newSize)
{
    ...
    if (newSize != this->size_)
    {
        if (newSize > 0)
        {
            T* nv = new T[label(newSize)];

            if (this->size_)
            {
                ...copy old elements into newly allocated memory...
            if (this->v_) delete[] this->v_;

            this->size_ = newSize;
            this->v_ = nv;
        }
        ...
    }
}
After setSize call your t2 vector points to invalid location (memory was freed). To fix this one can a) use copy instead of reference, b) check if list contains a (in setSize(const label newSize, const T& a)). Variant a is easier
roenby likes this.
alexeym is offline   Reply With Quote

Old   June 3, 2015, 05:07
Default
  #3
Member
 
Johan Roenby
Join Date: May 2011
Location: Denmark
Posts: 92
Rep Power: 20
roenby will become famous soon enough
Thanks a lot for that useful reply, alexeym!

I was not aware that appending to vectorFields and pointFields copied the whole field into new memory.

The right thing to do in my particular case would probably be to work instead with a DynamicList<vector> (with an appropriately guessed initial size) instead of a vectorField. This way I'll avoid all that copying.
roenby is offline   Reply With Quote

Old   June 3, 2015, 05:14
Default
  #4
Senior Member
 
Alexey Matveichev
Join Date: Aug 2011
Location: Nancy, France
Posts: 1,930
Rep Power: 38
alexeym has a spectacular aura aboutalexeym has a spectacular aura about
Send a message via Skype™ to alexeym
Hi,

Well, not quite sure you can avoid copying with DynamicList:

Code:
template<class T, unsigned SizeInc, unsigned SizeMult, unsigned SizeDiv>
inline void Foam::DynamicList<T, SizeInc, SizeMult, SizeDiv>::setSize
(
    const label nElem
)
{
        ...
        {
            capacity_ = max
            (
                nElem,
                label(SizeInc + capacity_ * SizeMult / SizeDiv)
            );
        }

        List<T>::setSize(capacity_);
    }
    
    // adjust addressed size
    List<T>::size(nElem);
}
Though since DynamicList grows in chunks, memory motion events happen less frequently.
alexeym is offline   Reply With Quote

Old   June 3, 2015, 05:47
Default
  #5
Member
 
Johan Roenby
Join Date: May 2011
Location: Denmark
Posts: 92
Rep Power: 20
roenby will become famous soon enough
Yes, hence my comment in brackets: "(with an appropriately guessed initial size)".

If for instance you know that your pointField/DynamicList<point> will be a subset of the points of a particular face in your mesh, just make sure to allocate memory for 'nFacePoints' points from the start to avoid copying.

Cheers,
Johan
roenby is offline   Reply With Quote

Reply


Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
problem during mpi in server: expected Scalar, found on line 0 the word 'nan' muth OpenFOAM Running, Solving & CFD 3 August 27, 2018 04:18
udf error srihari FLUENT 1 October 31, 2016 14:18
meshing of a compound volume in GMSH shawn3531 OpenFOAM 4 March 12, 2015 10:45
alpha1 wrong behaviour with LTSInterFoam Quentin OpenFOAM Running, Solving & CFD 16 January 10, 2015 09:56
Unstable behaviour after long period of stablility plunge11 FLUENT 1 April 6, 2011 09:15


All times are GMT -4. The time now is 04:51.