Invalid pointer dereference in igtlNDArrayMessage.h

[mikhirev]

Hi!

This issue is related to Fedora FTBFS bug that is also reproducible with OpenIGTLink 2.0 release. When using gcc 7, compilation fails with the following error:

In file included from /builddir/build/BUILD/OpenIGTLink-ccb24383e490ea1ddc07dd2d854e0d08efe80abc/Source/igtlNDArrayMessage.cxx:15:0:
/builddir/build/BUILD/OpenIGTLink-ccb24383e490ea1ddc07dd2d854e0d08efe80abc/Source/igtlNDArrayMessage.h: In member function 'int igtl::Array<T>::SetValue(igtl::ArrayBase::IndexType, T)':
/builddir/build/BUILD/OpenIGTLink-ccb24383e490ea1ddc07dd2d854e0d08efe80abc/Source/igtlNDArrayMessage.h:94:62: error: 'void*' is not a pointer-to-object type
       * (T *) this->m_ByteArray[Get1DIndex(index) * sizeof(T)] = value;
                                                              ^
/builddir/build/BUILD/OpenIGTLink-ccb24383e490ea1ddc07dd2d854e0d08efe80abc/Source/igtlNDArrayMessage.h: In member function 'int igtl::Array<T>::GetValue(igtl::ArrayBase::IndexType, T&)':
/builddir/build/BUILD/OpenIGTLink-ccb24383e490ea1ddc07dd2d854e0d08efe80abc/Source/igtlNDArrayMessage.h:105:70: error: 'void*' is not a pointer-to-object type
       value = * (T *) this->m_ByteArray[Get1DIndex(index) * sizeof(T)];
                                                                      ^

m_ByteArray is void *, so it seems that there's a superfluous dereference here.

[mikhirev]

BTW igtl::Array<T>::GetValue() and igtl::Array<T>::SetValue() are unusable anyway. They cannot be instantiated because igtl::ArrayBase::m_ByteArray is private.

[tokjun]

Thanks for reporting the issue. We will work on this soon.

[leochan2009]

@mikhirev , Thanks for your report.
The GCC 7 is not released yet, there seems to be huge number of improvements.
i will make a local build of GCC 7 and make sure that OpenIGTLink is compatible with it.

[mikhirev]

@leochan2009, I think that building GCC 7 is not required. You may try to write a simple unit test for igtl::Array<T>::GetValue() and igtl::Array<T>::SetValue() and see that they are broken. That's what I did, and it failed to compile with GCC 5.

I'm going to remove these methods in Fedora package.

[leochan2009]

@mikhirev
Thanks a lot. Now I see the point. It is definitive a bug, and was not detected by compiler previously.
Now the GCC-7 is able to detect the bug during the build.

[leochan2009]

The pointer deference bug should be fixed now