Narrowing Conversion of unsigned int to short unsigned int

Question

warning: narrowing conversion of '(stride * 4u)' from 'unsigned int' to 'WORD {aka short unsigned int}' inside { } is ill-formed in C++11 [-Wnarrowing]

I cannot figure out why I am getting this warning compiling the following code from MinGW:

unsigned stride = 3;

D3DVERTEXELEMENT9 NORMALELEMENT =
{ 0, stride * sizeof(gs_scalar), D3DDECLTYPE_FLOAT3, D3DDECLMETHOD_DEFAULT, D3DDECLUSAGE_NORMAL, 0 };
if (useNormals) stride += 3;

It is complaining about the stride * sizeof(gs_scalar) (gs_scalar is float) inside those braces, but I do not see how this is a narrowing conversion since sizeof returns the number of bytes. I tried changing stride's data type to WORD, DWORD, CHAR, everything, but I keep getting the same or similar warning.

Using brace initialization prevents narrowing conversions in c++11. Have you tried wrapping stride * sizeof(gs_scalar) in an explicit cast if that's your intention? — goji

gx_ gx_ · Accepted Answer · 2013-08-31T09:44:15

Look at the definition of D3DVERTEXELEMENT9:

struct D3DVERTEXELEMENT9 {
  WORD Stream;
  WORD Offset;
  BYTE Type;
  BYTE Method;
  BYTE Usage;
  BYTE UsageIndex;
};

(from http://msdn.microsoft.com/en-us/library/windows/desktop/bb172630%28v=vs.85%29.aspx but removed the typedef things).

Thus your are initializing NORMALELEMENT.Offset with stride * sizeof(gs_scalar).

The type of sizeof(gs_scalar) is std::size_t which is apparently unsigned int on your platform, and the type of stride is unsigned (i.e. unsigned int), so the type of stride * sizeof(gs_scalar) is unsigned int. But the type of NORMALELEMENT.Offset is WORD which is unsigned short.

I guess that on your platform unsigned int is 32-bits wide but unsigned short only 16-bits wide, so this is indeed a narrowing conversion (if the value of stride * sizeof(gs_scalar) can't fit in 16 bits you'll lose data).

Even if you define stride as a WORD, it is promoted to unsigned int in the multiplication with sizeof(gs_scalar), so the situation keeps the same.

If you're sure that stride * sizeof(gs_scalar) will never be more than USHRT_MAX (i.e. for you 2¹⁶−1 i.e. 65535), which seems likely (in the example it's 3 * 4 i.e. 12), then you can use a cast (as said by Troy in the comments), e.g. static_cast<WORD>(stride * sizeof(gs_scalar)).

Narrowing Conversion of unsigned int to short unsigned int

2 Answers