HDF4 type sizes not safe

[Kurt Schwehr]

The sizes of char, short, int and GBigInt are not a specific bit size, so this tries to make it a little bit more predictable for oddball build envs by using GByte, GInt16/GUInt16, GInt32/GUInt32. It also adds 2 static_asserts.

Thoughts?

hdf4dataset.cpp

@@ -298 +298 @@
 
 double HDF4Dataset::AnyTypeToDouble( int32 iNumType, void *pData )
 {
+    CPL_STATIC_ASSERT(sizeof(GIntBig) == 8);
     switch ( iNumType )
     {
         case DFNT_INT8:
             return static_cast<double>(*reinterpret_cast<char *>(pData));
         case DFNT_UINT8:
-            return static_cast<double>
-                (*reinterpret_cast<unsigned char *>(pData));
+            return static_cast<double>(*reinterpret_cast<GByte *>(pData));
         case DFNT_INT16:
-            return static_cast<double>(*reinterpret_cast<short *>(pData));
+            return static_cast<double>(*reinterpret_cast<GInt16 *>(pData));
         case DFNT_UINT16:
-            return static_cast<double>
-                (*reinterpret_cast<unsigned short *>(pData));
+            return static_cast<double>(*reinterpret_cast<GUInt16 *>(pData));
         case DFNT_INT32:
-            return static_cast<double>(*reinterpret_cast<int *>(pData));
+            return static_cast<double>(*reinterpret_cast<GInt32 *>(pData));
         case DFNT_UINT32:
-            return static_cast<double>
-                (*reinterpret_cast<unsigned int *>(pData));
+            return static_cast<double>(*reinterpret_cast<GUInt32 *>(pData));
         case DFNT_INT64:
             return static_cast<double>(*reinterpret_cast<GIntBig *>(pData));
         case DFNT_UINT64:
 

[Even Rouault]

"sizeof(char) == 1" is guaranteed by the C/C++ standards as far as I recall.

As far as sizeof(GIntBig) == 8, I'm pretty sure there are a lot of places where this assumption would need to be true, so I'm not sure it's worth putting it in that particular place.

OK for other changes. (But GInt16 is just typedef short GInt16)

[Kurt Schwehr]

r34324