Marshal C++ float to C# float accuracy problem

Question

I have a DBase IV database. Each row has a memo field with a ASCII encoded string that holds two serialized borland c++ structures. I am able to pull the data using OleDb, re-encode it to ascii using the ASCIIEncoding class, convert it to bytes using a BinaryReader, and cast it to my C# struct using Marshal.PtrToStructure. The data I get is correct but any float that is to big in the database is completely wrong when it is cast to the c#. For example, a value of 1149.00 cast into 764.9844 but a value like 64.00 cast fine. I can post some of the code and the structures but I figured I tried to keep it short at first. I know that floats are only precise up to 7 digits but I'm confused why I'm seeing this because the values are under that limit.

Edit:

struct cplusplusstruct // from the c++ code
{
  int   Number;
  float P;
  float ZP;
  float Hours;
  int   Month;
  int   Day;
  int   Year;
  int   Hour;
  int   Minute;
  int   Second;
  ULONG UPCTime;
  int   B;
  char  Name[21];
  float L;
  float H;
  float S;
}


[StructLayout(LayoutKind.Sequential, Pack = 1)]
public struct csharpstruct  //The C# struct I created
{
     public int Number;
     public float Pr;
     public float ZP;
     public float Hours;
     public int Month;
     public int Day;
     public int Year;
     public int Hour;
     public int Minute;
     public int Second;
     public UInt32 UPCTime;
     public int B;
 [MarshalAsAttribute(UnmanagedType.ByValTStr, SizeConst = 21)]
     public string Name;
     public float L;
     public float H;
     public float S;
 }


//OLE DB Connection and query ...

//Casting data to struct
ASCIIEncoding encoding = new ASCIIEncoding();
byte[] blob = encoding.GetBytes(memoString);
MemoryStream memoryStream = new MemoryStream(blob);
BinaryReader binaryReader = new BinaryReader(memoryStream);

int dataSize = Marshal.SizeOf(typeof(csharpstruct));
GCHandle handle = GCHandle.Alloc(binaryReader.ReadBytes(dataSize), GCHandleType.Pinned);
csharpstruct data = (csharpstruct) Marshal.PtrToStructure(handle.AddrOfPinnedObject(), typeof(csharpstruct));

Edit: The following is java code that read the data just fine but without any use of casting.

org.xBaseJ.DBF dbf = new org.xBaseJ.DBF(dbPath);
org.xBaseJ.DBF dbf = new org.xBaseJ.DBF(dbPath);
MemoField m = (MemoField) dbf.getField("MEMOFIELD");

Charset charset = Charset.forName("US-ASCII");
CharsetDecoder decoder = charset.newDecoder();
ByteBuffer trendBytes = ByteBuffer.wrap(m.getBytes());
trendBytes.order(ByteOrder.LITTLE_ENDIAN);
trendBytes.getInt();
trendBytes.getFloat();

Answer 1

I wasn't able to solve the issue directly. The problem seemed to come from the OLE data provider I was using. The data retreived from the database was slightly different then what xBaseJ provided. I ended up converting xBaseJ to CLI bytecode using IKVM.NET. This allowed me to replace the OLE data provider with the xBaseJ reader. The rest of my code remained unchanged.

Answer 2

You have Pack = 1 in your C# struct, but have not said whether your C++ struct is packed. Since you have an odd-sized field just before your floats (the 21-character string) that might cause trouble and mean that your floats are being read mis-aligned. Everything before there is 4 bytes long so packing is less likely to cause you problems. I would make sure that the packing matches in both C# and C++ before going any further.