SUMMARY
The ODBC Specification describes Bufferlength for SQLGetDiagRec as follows:
BufferLength
[Input]
Length (in bytes) of the *MessageText buffer. There is no maximum length of the diagnostic message text. If the value returned in *MessageText is of a Unicode string (when calling SQLGetDiagRecW), the BufferLength argument must be an even number.
The above statement is wrong as the Bufferlength takes in the number of Unicode Characters instead when using SQLGetDiagRecW as the ODBC Spec also says:
Unicode functions that always return or take strings or length arguments are passed as count-of-characters. For functions that return length information for server data, the display size and precision are described in number of characters. When a length (transfer size of the data) could refer to string or non-string data, the length is described in octet lengths. For example, SQLGetInfoW will still take the length as count-of-bytes, but SQLExecDirectW uses count-of-characters
For example,if a typical application that expects 100 bytes (50 Unicode characters) of error message passes the Bufferlength as 100, as many characters are returned instead.
The application should carefully allocate as much memory as necessary. (Twice the value specified for the buffer length.) In the example above, the memory allocated should be 200 bytes for holding 100 Unicode characters.