Some words about template checks

Aug 31 2012

Author: Andrey Karpov

The PVS-Studio analyzer sometimes generates mysterious messages mentioning a template class. For example: V614 Instantiate ReconstructMB<PlaneY, PlaneUV, color_format, false, 1>: Uninitialized variable 'iRefFieldTop' used. On encountering such a message and reviewing the template class code you may feel sad. It all looks complicated and incomprehensible. But it's not that horrible. Let's try to parse one sample.

Consider a code fragment taken from a real application.

template <typename PlaneY, typename PlaneUV,
          Ipp32s color_format, Ipp32s is_field, Ipp32s is_weight,
          bool nv12_support = false>
class ReconstructMB
{
public:
  ...
  void CompensateMotionChromaBlock(ReconstructParams *pParams,
         Ipp32s iDir, Ipp32s iBlockNumber, Ipp32s iUniDir)
  {
    Ipp32s iRefFieldTop;
    if (is_field)
    {
      if (pParams->is_mbaff)
      {
        iRefFieldTop = pParams->is_bottom_mb ^
                       (pParams->m_iRefIndex[iDir] & 1);
        pParams->m_iRefIndex[iDir] = iRefIndex;
      }
      else
        iRefFieldTop =
          GetReferenceField(pParams->m_pSegDec->m_pFields[iDir],
                            iRefIndex);
      
      if (iRefFieldTop)
      { ..... }
    }
    .....
    interpolateInfo->pointVector.y +=
      (pParams->is_bottom_mb - iRefFieldTop) * 2; // <= V614
    .....
};

Long and obscure, do you agree? You don't want to review this code, but the analyzer urges you to pay attention to it. I took such a long sample intentionally: I want to show you that programmers shouldn't be lazy. You have to understand what exactly the analyzer doesn't like in this code.

Here is the message generated by PVS-Studio:

V614 Instantiate ReconstructMB<PlaneY, PlaneUV, color_format, false, 1>: Uninitialized variable 'iRefFieldTop' used. umc_h264_reconstruct_templates.h 227

PVS-Studio suspects that an uninitialized variable 'iRefFieldTop' is used in line 227.

If we examine the code, we'll see that the 'iRefFieldTop' variable is initialized only if the "if (is_field)" condition holds. The 'is_field' value is a template argument and has the 'Ipp32s' type.

Surely, no error will occur if the ReconstructMB class is instanced with values "is_field != 0". For example, it may look like this:

ReconstructMB<PlaneY, PlaneUV, color_format, 1, 1, false> X;

But no, the analyzer warns you not without reason. If you look attentively, you will see a funny variable declaration:

ReconstructMB<PlaneY, PlaneUV, color_format, false, 0> mb;

Something is obviously wrong here. Look what arguments we are expecting to be passed and what arguments are actually passed:

Template arguments      Arguments in the variable declaration
typename PlaneY                      PlaneY
typename PlaneUV                     PlaneUV
Ipp32s color_format                  color_format
Ipp32s is_field                      false
Ipp32s is_weight,                    0
bool nv12_support = false            (none)

It's strange that one argument of the 'Ipp32s' type is defined as being equal to '0', while the other as being equal to 'false'. A very strange programming style indeed!

Most likely, the template parameters were being added or changed during the refactoring process. In this particular fragment the developer forgot to fix the declaration of the 'mb' variable. The code compiled well and nobody noticed the bug.

Well, it's not really important how this code has appeared. What is important, there is uninitialized memory usage in the program, and the PVS-Studio analyzer has found it. It also points out in the message that the error will occur when the class is instanced in the following way: ReconstructMB<PlaneY, PlaneUV, color_format, false, 1>.

I hope I've managed to clarify what such analyzer-generated messages mean and what information they contain.

Make your code safer by using the PVS-Studio analyzer!