7 Programming with Unicode

Database Access Product Stack and Unicode

Oracle offers a comprehensive set of database access products that enable programs from different development environments to access Unicode data stored in the database. These products are listed in Table 7-1.

Figure 7-1 shows how the database access products can access the database.

Figure 7-1 Oracle Database Access Products

Description of "Figure 7-1 Oracle Database Access Products "

The Oracle Call Interface (OCI) is the lowest level API that the rest of the client-side database access products use. It provides a flexible way for C/C++ programs to access Unicode data stored in SQL CHAR and NCHAR datatypes. Using OCI, you can programmatically specify the character set (UTF-8, UTF-16, and others) for the data to be inserted or retrieved. It accesses the database through Oracle Net.

Oracle Pro*C/C++ enables you to embed SQL and PL/SQL in your programs. It uses OCI's Unicode capabilities to provide UTF-16 and UTF-8 data access for SQL CHAR and NCHAR datatypes.

The Oracle ODBC driver enables C/C++, Visual Basic, and VBScript programs running on Windows platforms to access Unicode data stored in SQL CHAR and NCHAR datatypes of the database. It provides UTF-16 data access by implementing the SQLWCHAR interface specified in the ODBC standard specification.

The Oracle Provider for OLE DB enables C/C++, Visual Basic, and VBScript programs running on Windows platforms to access Unicode data stored in SQL CHAR and NCHAR datatypes. It provides UTF-16 data access through wide string OLE DB datatypes.

The Oracle Data Provider for .NET enables programs running in any .NET programming environment on Windows platforms to access Unicode data stored in SQL CHAR and NCHAR datatypes. It provides UTF-16 data access through Unicode datatypes.

Oracle JDBC drivers are the primary Java programmatic interface for accessing an Oracle database. Oracle provides the following JDBC drivers:

• The JDBC OCI driver that is used by Java applications and requires the OCI library

• The JDBC thin driver, which is a pure Java driver that is primarily used by Java applets and supports the Oracle Net protocol over TCP/IP

• The JDBC server-side thin driver, a pure Java driver used inside Java stored procedures to connect to another Oracle server

• The JDBC server-side internal driver that is used inside the Oracle server to access the data in the database

All drivers support Unicode data access to SQL CHAR and NCHAR datatypes in the database.

The PL/SQL and SQL engines process PL/SQL programs and SQL statements on behalf of client-side programs such as OCI and server-side PL/SQL stored procedures. They allow PL/SQL programs to declare CHAR, VARCHAR2, NCHAR, and NVARCHAR2 variables and to access SQL CHAR and NCHAR datatypes in the database.

The following sections describe how each of the database access products supports Unicode data access to an Oracle database and offer examples for using those products:

• SQL and PL/SQL Programming with Unicode

• OCI Programming with Unicode

• Pro*C/C++ Programming with Unicode

• JDBC Programming with Unicode

• ODBC and OLE DB Programming with Unicode

SQL NCHAR Datatypes

There are three SQL NCHAR datatypes:

• The NCHAR Datatype

• The NVARCHAR2 Datatype

• The NCLOB Datatype

CREATE TABLE table1 (column1 NCHAR(30)); 

maximum number of bytes = (maximum number of characters) x (maximum number of bytes for each character)

CREATE TABLE table2 (column2 NVARCHAR2(2000)); 

Implicit Datatype Conversion Between NCHAR and Other Datatypes

Oracle supports implicit conversions between SQL NCHAR datatypes and other Oracle datatypes, such as CHAR, VARCHAR2, NUMBER, DATE, ROWID, and CLOB. Any implicit conversions for CHAR and VARCHAR2 datatypes are also supported for SQL NCHAR datatypes. You can use SQL NCHAR datatypes the same way as SQL CHAR datatypes.

Type conversions between SQL CHAR datatypes and SQL NCHAR datatypes may involve character set conversion when the database and national character sets are different. Padding with blanks may occur if the target data is either CHAR or NCHAR.

Exception Handling for Data Loss During Datatype Conversion

Data loss can occur during datatype conversion when character set conversion is necessary. If a character in the source character set is not defined in the target character set, then a replacement character is used in its place. For example, if you try to insert NCHAR data into a regular CHAR column and the character data in NCHAR (Unicode) form cannot be converted to the database character set, then the character is replaced by a replacement character defined by the database character set. The NLS_NCHAR_CONV_EXCP initialization parameter controls the behavior of data loss during character type conversion. When this parameter is set to TRUE, any SQL statements that result in data loss return an ORA-12713 error and the corresponding operation is stopped. When this parameter is set to FALSE, data loss is not reported and the unconvertible characters are replaced with replacement characters. The default value is TRUE. This parameter works for both implicit and explicit conversion.

In PL/SQL, when data loss occurs during conversion of SQL CHAR and NCHAR datatypes, the LOSSY_CHARSET_CONVERSION exception is raised for both implicit and explicit conversion.

Rules for Implicit Datatype Conversion

In some cases, conversion between datatypes is possible in only one direction. In other cases, conversion in both directions is possible. Oracle defines a set of rules for conversion between datatypes. Table 7-2 contains the rules for conversion between datatypes.

SQL Functions for Unicode Datatypes

SQL NCHAR datatypes can be converted to and from SQL CHAR datatypes and other datatypes using explicit conversion functions. The examples in this section use the table created by the following statement:

CREATE TABLE customers 
  (id NUMBER, name NVARCHAR2(50), address NVARCHAR2(200), birthdate DATE);

INSERT INTO customers VALUES (1000, 
  TO_NCHAR('John Smith'),N'500 Oracle Parkway',sysdate);

SELECT name FROM customers WHERE TO_CHAR(name) LIKE '%Sm%';

NAME
--------------------------------------
John Smith

DECLARE
ndatestring NVARCHAR2(20) := N'12-SEP-1975';
ndstr NVARCHAR2(50);
BEGIN
SELECT name INTO ndstr FROM customers
WHERE (birthdate)> TO_DATE(ndatestring, 'DD-MON-YYYY', N'NLS_DATE_LANGUAGE =
AMERICAN');
END;

Other SQL Functions

Most SQL functions can take arguments of SQL NCHAR datatypes as well as mixed character datatypes. The return datatype is based on the type of the first argument. If a non-string datatype like NUMBER or DATE is passed to these functions, then it is converted to VARCHAR2. The following examples use the customer table created in "SQL Functions for Unicode Datatypes".

SELECT INSTR(name, N'Sm', 1, 1) FROM customers;

SELECT CONCAT(name,id) FROM customers;

SELECT RPAD(name,100,' ') FROM customers;

RPAD(NAME,100,'')
------------------------------------------
John Smith

Unicode String Literals

You can input Unicode string literals in SQL and PL/SQL as follows:

• Put a prefix N before a string literal that is enclosed with single quote marks. This explicitly indicates that the following string literal is an NCHAR string literal. For example, N'résumé' is an NCHAR string literal. For information about limitations of this method, see NCHAR String Literal Replacement.

• Use the NCHR(n) SQL function, which returns a unit of character code in the national character set, which is AL16UTF16 or UTF8. The result of concatenating several NCHR(n) functions is NVARCHAR2 data. In this way, you can bypass the client and server character set conversions and create an NVARCHAR2 string directly. For example, NCHR(32) represents a blank character.

  Because NCHR(n) is associated with the national character set, portability of the resulting value is limited to applications that run with the same national character set. If this is a concern, then use the UNISTR function to remove portability limitations.

• Use the UNISTR('string') SQL function. UNISTR('string') converts a string to the national character set. To ensure portability and to preserve data, include only ASCII characters and Unicode encoding in the following form: \xxxx, where xxxx is the hexadecimal value of a character code value in UTF-16 encoding format. For example, UNISTR('G\0061ry') represents 'Gary'. The ASCII characters are converted to the database character set and then to the national character set. The Unicode encoding is converted directly to the national character set.

The last two methods can be used to encode any Unicode string literals.

NCHAR String Literal Replacement

This section provides information on how to avoid data loss when performing NCHAR string literal replacement.

Being part of a SQL or PL/SQL statement, the text of any literal, with or without the prefix N, is encoded in the same character set as the rest of the statement. On the client side, the statement is in the client character set, which is determined by the client character set defined in NLS_LANG, or specified in the OCIEnvNlsCreate() call, or predefined as UTF-16 in JDBC. On the server side the statement is in the database character set.

• When the SQL or PL/SQL statement is transferred from client to the database server, its character set is converted accordingly. It is important to note that if the database character set does not contain all characters used in the text literals, then the data is lost in this conversion. This problem affects NCHAR string literals more than the CHAR text literals. This is because the N' literals are designed to be independent of the database charactser set, and should be able to provide any data that the client character set supports.

  To avoid data loss in conversion to an incompatible database character set, you can activate the NCHAR literal replacement functionality. The functionality transparently replaces the N' literals on the client side with an internal format. The database server then decodes this to Unicode when the statement is executed.

• The sections "Handling SQL NCHAR String Literals in OCI" and "Using SQL NCHAR String Literals in JDBC" show how to switch on the replacement functionality in OCI and JDBC, respectively. Because many applications, for example, SQL*Plus, use OCI to connect to a database, and they do not control NCHAR literal replacement explicitly, you can set the client environment variable ORA_NCHAR_LITERAL_REPLACE to TRUE to control the functionality for them. By default, the functionality is switched off to maintain backward compatibility.

Using the UTL_FILE Package with NCHAR Data

The UTL_FILE package handles Unicode national character set data. The functions and procedures include the following:

• FOPEN_NCHAR

  This function opens a file in Unicode for input or output, with the maximum line size specified. With this function, you can read or write a text file in Unicode instead of in the database character set.

• GET_LINE_NCHAR

  This procedure reads text from the open file identified by the file handle and places the text in the output buffer parameter. With this procedure, you can read a text file in Unicode instead of in the database character set.

• PUT_NCHAR

  This procedure writes the text string stored in the buffer parameter to the open file identified by the file handle. With this procedure, you can write a text file in Unicode instead of in the database character set.

• PUT_LINE_NCHAR

  This procedure writes the text string stored in the buffer parameter to the open file identified by the file handle. With this procedure, you can write a text file in Unicode instead of in the database character set.

• PUTF_NCHAR

  This procedure is a formatted PUT_NCHAR procedure. With this procedure, you can write a text file in Unicode instead of in the database character set.

  See Also:

  PL/SQL Packages and Types Reference for more information about the UTL_FILE package

OCIEnvNlsCreate() Function for Unicode Programming

The OCIEnvNlsCreate() function is used to specify a SQL CHAR character set and a SQL NCHAR character set when the OCI environment is created. It is an enhanced version of the OCIEnvCreate() function and has extended arguments for two character set IDs. The OCI_UTF16ID UTF-16 character set ID replaces the Unicode mode introduced in Oracle9i release 1 (9.0.1). For example:

OCIEnv *envhp;
status = OCIEnvNlsCreate((OCIEnv **)&envhp,
(ub4)0,
(void *)0,
(void *(*) ()) 0,
(void *(*) ()) 0,
(void(*) ()) 0,
(size_t) 0,
(void **)0,
(ub2)OCI_UTF16ID, /* Metadata and SQL CHAR character set */
(ub2)OCI_UTF16ID /* SQL NCHAR character set */);

The Unicode mode, in which the OCI_UTF16 flag is used with the OCIEnvCreate() function, is deprecated.

When OCI_UTF16ID is specified for both SQL CHAR and SQL NCHAR character sets, all metadata and bound and defined data are encoded in UTF-16. Metadata includes SQL statements, user names, error messages, and column names. Thus, all inherited operations are independent of the NLS_LANG setting, and all metatext data parameters (text*) are assumed to be Unicode text datatypes (utext*) in UTF-16 encoding.

To prepare the SQL statement when the OCIEnv() function is initialized with the OCI_UTF16ID character set ID, call the OCIStmtPrepare() function with a (utext*) string. The following example runs on the Windows platform only. You may need to change wchar_t datatypes for other platforms.

const wchar_t sqlstr[] = L"SELECT * FROM ENAME=:ename";
...
OCIStmt* stmthp;
sts = OCIHandleAlloc(envh, (void **)&stmthp, OCI_HTYPE_STMT, 0,
NULL);
status = OCIStmtPrepare(stmthp, errhp,(const text*)sqlstr,
wcslen(sqlstr), OCI_NTV_SYNTAX, OCI_DEFAULT);

To bind and define data, you do not have to set the OCI_ATTR_CHARSET_ID attribute because the OCIEnv() function has already been initialized with UTF-16 character set IDs. The bind variable names also must be UTF-16 strings.

/* Inserting Unicode data */
OCIBindByName(stmthp1, &bnd1p, errhp, (const text*)L":ename",
(sb4)wcslen(L":ename"),
              (void *) ename, sizeof(ename), SQLT_STR, (void
*)&insname_ind,
              (ub2 *) 0, (ub2 *) 0, (ub4) 0, (ub4 *)0,
OCI_DEFAULT);
OCIAttrSet((void *) bnd1p, (ub4) OCI_HTYPE_BIND, (void *)
&ename_col_len,
           (ub4) 0, (ub4)OCI_ATTR_MAXDATA_SIZE, errhp);
...
/* Retrieving Unicode data */
OCIDefineByPos (stmthp2, &dfn1p, errhp, (ub4)1, (void *)ename,
               (sb4)sizeof(ename), SQLT_STR, (void *)0, (ub2 *)0,
(ub2*)0, (ub4)OCI_DEFAULT);

The OCIExecute() function performs the operation.

OCI Unicode Code Conversion

Unicode character set conversions take place between an OCI client and the database server if the client and server character sets are different. The conversion occurs on either the client or the server depending on the circumstances, but usually on the client side.

ub2 csid = OCI_UTF16ID;
oratext *selstmt = "SELECT ename FROM emp";
counter = 1;
... 
OCIStmtPrepare(stmthp, errhp, selstmt, (ub4)strlen((char*)selstmt),
               OCI_NTV_SYNTAX, OCI_DEFAULT);
OCIStmtExecute ( svchp, stmthp, errhp, (ub4)0, (ub4)0,
                 (CONST OCISnapshot*)0, (OCISnapshot*)0,
                 OCI_DESCRIBE_ONLY);
OCIParamGet(stmthp, OCI_HTYPE_STMT, errhp, &myparam, (ub4)counter);
OCIAttrGet((void*)myparam, (ub4)OCI_DTYPE_PARAM, (void*)&col_width,
           (ub4*)0, (ub4)OCI_ATTR_DATA_SIZE, errhp);
... 
maxenamelen = (col_width + 1) * sizeof(utext);
cbuf = (utext*)malloc(maxenamelen);
...
OCIDefineByPos(stmthp, &dfnp, errhp, (ub4)1, (void *)cbuf,
                (sb4)maxenamelen, SQLT_STR, (void *)0, (ub2 *)0,
                (ub2*)0, (ub4)OCI_DEFAULT);
OCIAttrSet((void *) dfnp, (ub4) OCI_HTYPE_DEFINE, (void *) &csid,
           (ub4) 0, (ub4)OCI_ATTR_CHARSET_ID, errhp);
OCIStmtFetch(stmthp, errhp, 1, OCI_FETCH_NEXT, OCI_DEFAULT);
...

Setting UTF-8 to the NLS_LANG Character Set in OCI

For OCI client applications that support Unicode UTF-8 encoding, use AL32UTF8 to specify the NLS_LANG character set, unless the database character set is UTF8. Use UTF8 if the database character set is UTF8.

Do not set NLS_LANG to AL16UTF16, because AL16UTF16 is the national character set for the server. If you need to use UTF-16, then you should specify the client character set to OCI_UTF16ID, using the OCIAttrSet() function when binding or defining data

Binding and Defining SQL CHAR Datatypes in OCI

To specify a Unicode character set for binding and defining data with SQL CHAR datatypes, you may need to call the OCIAttrSet() function to set the appropriate character set ID after OCIBind() or OCIDefine() APIs. There are two typical cases:

• Call OCIBind() or OCIDefine() followed by OCIAttrSet() to specify UTF-16 Unicode character set encoding. For example:

  ...
  ub2 csid = OCI_UTF16ID;
  utext ename[100]; /* enough buffer for ENAME */
  ... 
  /* Inserting Unicode data */ 
  OCIBindByName(stmthp1, &bnd1p, errhp, (oratext*)":ENAME",
               (sb4)strlen((char *)":ENAME"), (void *) ename, sizeof(ename),
               SQLT_STR, (void *)&insname_ind, (ub2 *) 0, (ub2 *) 0, (ub4) 0,
               (ub4 *)0, OCI_DEFAULT); 
  OCIAttrSet((void *) bnd1p, (ub4) OCI_HTYPE_BIND, (void *) &csid,
             (ub4) 0, (ub4)OCI_ATTR_CHARSET_ID, errhp); 
  OCIAttrSet((void *) bnd1p, (ub4) OCI_HTYPE_BIND, (void *) &ename_col_len,
             (ub4) 0, (ub4)OCI_ATTR_MAXDATA_SIZE, errhp); 
  ... 
  /* Retrieving Unicode data */
  OCIDefineByPos (stmthp2, &dfn1p, errhp, (ub4)1, (void *)ename, 
                  (sb4)sizeof(ename), SQLT_STR, (void *)0, (ub2 *)0,
                  (ub2*)0, (ub4)OCI_DEFAULT); 
  OCIAttrSet((void *) dfn1p, (ub4) OCI_HTYPE_DEFINE, (void *) &csid,
             (ub4) 0, (ub4)OCI_ATTR_CHARSET_ID, errhp); 
  ...
  

  If bound buffers are of the utext datatype, then you should add a cast (text*) when OCIBind() or OCIDefine() is called. The value of the OCI_ATTR_MAXDATA_SIZE attribute is usually determined by the column size of the server character set because this size is only used to allocate temporary buffer space for conversion on the server when you perform binding operations.

• Call OCIBind() or OCIDefine() with the NLS_LANG character set specified as UTF8 or AL32UTF8.

  UTF8 or AL32UTF8 can be set in the NLS_LANG environment variable. You call OCIBind() and OCIDefine() in exactly the same manner as when you are not using Unicode. Set the NLS_LANG environment variable to UTF8 or AL32UTF8 and run the following OCI program:

  ...
  oratext ename[100]; /* enough buffer size for ENAME */
  ... 
  /* Inserting Unicode data */ 
  OCIBindByName(stmthp1, &bnd1p, errhp, (oratext*)":ENAME",
                (sb4)strlen((char *)":ENAME"), (void *) ename, sizeof(ename),
                SQLT_STR, (void *)&insname_ind, (ub2 *) 0, (ub2 *) 0,
                (ub4) 0, (ub4 *)0, OCI_DEFAULT); 
  OCIAttrSet((void *) bnd1p, (ub4) OCI_HTYPE_BIND, (void *) &ename_col_len,
             (ub4) 0, (ub4)OCI_ATTR_MAXDATA_SIZE, errhp); 
  ... 
  /* Retrieving Unicode data */
  OCIDefineByPos (stmthp2, &dfn1p, errhp, (ub4)1, (void *)ename,
                  (sb4)sizeof(ename), SQLT_STR, (void *)0, (ub2 *)0, (ub2*)0,
                  (ub4)OCI_DEFAULT); 
  ...
  

Binding and Defining SQL NCHAR Datatypes in OCI

Oracle recommends that you access SQL NCHAR datatypes using UTF-16 binding or defining when using OCI. Beginning with Oracle9i, SQL NCHAR datatypes are Unicode datatypes with an encoding of either UTF8 or AL16UTF16. To access data in SQL NCHAR datatypes, set the OCI_ATTR_CHARSET_FORM attribute to SQLCS_NCHAR between binding or defining and execution so that it performs an appropriate data conversion without data loss. The length of data in SQL NCHAR datatypes is always in the number of Unicode code units.

The following program is a typical example of inserting and fetching data against an NCHAR data column:

...
ub2 csid = OCI_UTF16ID;
ub1 cform = SQLCS_NCHAR;
utext ename[100]; /* enough buffer for ENAME */
... 
/* Inserting Unicode data */ 
OCIBindByName(stmthp1, &bnd1p, errhp, (oratext*)":ENAME",
              (sb4)strlen((char *)":ENAME"), (void *) ename,
              sizeof(ename), SQLT_STR, (void *)&insname_ind, (ub2 *) 0,
              (ub2 *) 0, (ub4) 0, (ub4 *)0, OCI_DEFAULT); 
OCIAttrSet((void *) bnd1p, (ub4) OCI_HTYPE_BIND, (void *) &cform, (ub4) 0,
           (ub4)OCI_ATTR_CHARSET_FORM, errhp); 
OCIAttrSet((void *) bnd1p, (ub4) OCI_HTYPE_BIND, (void *) &csid, (ub4) 0,
           (ub4)OCI_ATTR_CHARSET_ID, errhp);
OCIAttrSet((void *) bnd1p, (ub4) OCI_HTYPE_BIND, (void *) &ename_col_len,
           (ub4) 0, (ub4)OCI_ATTR_MAXDATA_SIZE, errhp); 
... 
/* Retrieving Unicode data */
OCIDefineByPos (stmthp2, &dfn1p, errhp, (ub4)1, (void *)ename,
                (sb4)sizeof(ename), SQLT_STR, (void *)0, (ub2 *)0, (ub2*)0,
                (ub4)OCI_DEFAULT); 
OCIAttrSet((void *) dfn1p, (ub4) OCI_HTYPE_DEFINE, (void *) &csid, (ub4) 0,
           (ub4)OCI_ATTR_CHARSET_ID, errhp); 
OCIAttrSet((void *) dfn1p, (ub4) OCI_HTYPE_DEFINE, (void *) &cform, (ub4) 0,
           (ub4)OCI_ATTR_CHARSET_FORM, errhp); 
...

Handling SQL NCHAR String Literals in OCI

By default, the NCHAR literal replacement is not performed in OCI. (Refer to "NCHAR String Literal Replacement".)

You can switch it on by setting the environment variable ORA_NCHAR_LITERAL_REPLACE to TRUE. You can also achieve this behavior programmatically by using the OCI_NCHAR_LITERAL_REPLACE_ON and OCI_NCHAR_LITERAL_REPLACE_OFF modes in OCIEnvCreate() and OCIEnvNlsCreate(). So, for example, OCIEnvCreate(OCI_NCHAR_LITERAL_REPLACE_ON) turns on NCHAR literal replacement, while OCIEnvCreate(OCI_NCHAR_LITERAL_REPLACE_OFF) turns it off.

As an example, consider the following statement:

int main(argc, argv)
{
   OCIEnv *envhp;
if (OCIEnvCreate((OCIEnv **) &envhp,
   (ub4)OCI_THREADED|OCI_NCHAR_LITERAL_REPLACE_ON,
   (dvoid *)0, (dvoid * (*)(dvoid *, size_t)) 0,
   (dvoid * (*)(dvoid *, dvoid *, size_t))0,
   (void (*)(dvoid *, dvoid *)) 0,
   (size_t) 0, (dvoid **) 0))
{
  printf("FAILED: OCIEnvCreate()\n";
  return 1;
}
...
}

Note that, when the NCHAR literal replacement is turned on, OCIStmtPrepare and OCIStmtPrepare2 will transform N' literals with U' literals in the SQL text and store the resulting SQL text in the statement handle. Thus, if the application uses OCI_ATTR_STATEMENT to retrieve the SQL text from the OCI statement handle, the SQL text will return U' instead of N' as specified in the original text.

See the Oracle Database SQL Reference for information regarding environment variables.

Binding and Defining CLOB and NCLOB Unicode Data in OCI

In order to write (bind) and read (define) UTF-16 data for CLOB or NCLOB columns, the UTF-16 character set ID must be specified as OCILobWrite() and OCILobRead(). When you write UTF-16 data into a CLOB column, call OCILobWrite() as follows:

...
ub2 csid = OCI_UTF16ID;
err = OCILobWrite (ctx->svchp, ctx->errhp, lobp, &amtp, offset, (void *) buf,
                   (ub4) BUFSIZE, OCI_ONE_PIECE, (void *)0,
                   (sb4 (*)()) 0, (ub2) csid, (ub1) SQLCS_IMPLICIT); 

The amtp parameter is the data length in number of Unicode code units. The offset parameter indicates the offset of data from the beginning of the data column. The csid parameter must be set for UTF-16 data.

To read UTF-16 data from CLOB columns, call OCILobRead() as follows:

...
ub2 csid = OCI_UTF16ID;
err = OCILobRead(ctx->svchp, ctx->errhp, lobp, &amtp, offset, (void *) buf,
                 (ub4)BUFSIZE , (void *) 0, (sb4 (*)()) 0, (ub2)csid,
                 (ub1) SQLCS_IMPLICIT);

The data length is always represented in the number of Unicode code units. Note one Unicode supplementary character is counted as two code units, because the encoding is UTF-16. After binding or defining a LOB column, you can measure the data length stored in the LOB column using OCILobGetLength(). The returning value is the data length in the number of code units if you bind or define as UTF-16.

err = OCILobGetLength(ctx->svchp, ctx->errhp, lobp, &lenp);

If you are using an NCLOB, then you must set OCI_ATTR_CHARSET_FORM to SQLCS_NCHAR.

Pro*C/C++ Data Conversion in Unicode

Data conversion occurs in the OCI layer, but it is the Pro*C/C++ preprocessor that instructs OCI which conversion path should be taken based on the datatypes used in a Pro*C/C++ program. Table 7-4 illustrates the conversion paths:

Using the VARCHAR Datatype in Pro*C/C++

The Pro*C/C++ VARCHAR datatype is preprocessed to a struct with a length field and text buffer field. The following example uses the C/C++ text native datatype and the VARCHAR Pro*C/C++ datatypes to bind and define table columns.

#include <sqlca.h> 
main() 
{ 
   ... 
   /* Change to STRING datatype:    */ 
   EXEC ORACLE OPTION (CHAR_MAP=STRING) ; 
   text ename[20] ;                  /* unsigned short type */ 
   varchar address[50] ;             /* Pro*C/C++ varchar type */ 

   EXEC SQL SELECT ename, address INTO :ename, :address FROM emp; 
   /* ename is NULL-terminated */ 
   printf(L"ENAME = %s, ADDRESS = %.*s\n", ename, address.len, address.arr); 
   ... 
} 

When you use the VARCHAR datatype or native text datatype in a Pro*C/C++ program, the preprocessor assumes that the program intends to access columns of SQL CHAR datatypes instead of SQL NCHAR datatypes in the database. The preprocessor generates C/C++ code to reflect this fact by doing a bind or define using the SQLCS_IMPLICIT value for the OCI_ATTR_CHARSET_FORM attribute. As a result, if a bind or define variable is bound to a column of SQL NCHAR datatypes in the database, then implicit conversion occurs in the database server to convert the data from the database character set to the national database character set and vice versa. During the conversion, data loss occurs when the database character set is a smaller set than the national character set.

Using the NVARCHAR Datatype in Pro*C/C++

The Pro*C/C++ NVARCHAR datatype is similar to the Pro*C/C++ VARCHAR datatype. It should be used to access SQL NCHAR datatypes in the database. It tells Pro*C/C++ preprocessor to bind or define a text buffer to the column of SQL NCHAR datatypes. The preprocessor specifies the SQLCS_NCHAR value for the OCI_ATTR_CHARSET_FORM attribute of the bind or define variable. As a result, no implicit conversion occurs in the database.

If the NVARCHAR buffer is bound against columns of SQL CHAR datatypes, then the data in the NVARCHAR buffer (encoded in the NLS_LANG character set) is converted to or from the national character set in OCI, and the data is then converted to the database character set in the database server. Data can be lost when the NLS_LANG character set is a larger set than the database character set.

Using the UVARCHAR Datatype in Pro*C/C++

The UVARCHAR datatype is preprocessed to a struct with a length field and utext buffer field. The following example code contains two host variables, ename and address. The ename host variable is declared as a utext buffer containing 20 Unicode characters. The address host variable is declared as a uvarchar buffer containing 50 Unicode characters. The len and arr fields are accessible as fields of a struct.

#include <sqlca.h> 
#include <sqlucs2.h> 

main() 
{ 
   ... 
   /* Change to STRING datatype:    */ 
   EXEC ORACLE OPTION (CHAR_MAP=STRING) ; 
   utext ename[20] ;                  /* unsigned short type */ 
 uvarchar address[50] ;               /* Pro*C/C++ uvarchar type */ 

   EXEC SQL SELECT ename, address INTO :ename, :address FROM emp; 
   /* ename is NULL-terminated */ 
wprintf(L"ENAME = %s, ADDRESS = %.*s\n", ename, address.len, 
address.arr); 
... 
} 

When you use the UVARCHAR datatype or native utext datatype in Pro*C/C++ programs, the preprocessor assumes that the program intends to access SQL NCHAR datatypes. The preprocessor generates C/C++ code by binding or defining using the SQLCS_NCHAR value for OCI_ATTR_CHARSET_FORM attribute. As a result, if a bind or define variable is bound to a column of a SQL NCHAR datatype, then an implicit conversion of the data from the national character set occurs in the database server. However, there is no data lost in this scenario because the national character set is always a larger set than the database character set.

Binding and Defining Java Strings to SQL CHAR Datatypes

Oracle JDBC drivers allow you to access SQL CHAR datatypes in the database using Java string bind or define variables. The following code illustrates how to bind a Java string to a CHAR column.

int employee_id = 12345;
String last_name = "Joe";
PreparedStatement pstmt = conn.prepareStatement("INSERT INTO" +
    "employees (last_name, employee_id) VALUES (?, ?)");
pstmt.setString(1, last_name);
pstmt.setInt(2, employee_id);
pstmt.execute();                 /* execute to insert into first row */
employee_id += 1;                      /* next employee number */
last_name = "\uFF2A\uFF4F\uFF45";    /* Unicode characters in name */
pstmt.setString(1, last_name);
pstmt.setInt(2, employee_id);
pstmt.execute();                 /* execute to insert into second row */

You can define the target SQL columns by specifying their datatypes and lengths. When you define a SQL CHAR column with the datatype and the length, JDBC uses this information to optimize the performance of fetching SQL CHAR data from the column. The following is an example of defining a SQL CHAR column.

OraclePreparedStatement pstmt = (OraclePreparedStatement)
     conn.prepareStatement("SELECT ename, empno from emp");
pstmt.defineColumnType(1,Types.VARCHAR, 3);
pstmt.defineColumnType(2,Types.INTEGER);
ResultSet rest = pstmt.executeQuery();
String name = rset.getString(1);
int id = reset.getInt(2);

You need to cast PreparedStatement to OraclePreparedStatement to call defineColumnType(). The second parameter of defineColumnType() is the datatype of the target SQL column. The third parameter is the length in number of characters.

Binding and Defining Java Strings to SQL NCHAR Datatypes

For binding or defining Java string variables to SQL NCHAR datatypes, Oracle provides an extended PreparedStatement which has the setFormOfUse() method through which you can explicitly specify the target column of a bind variable to be a SQL NCHAR datatype. The following code illustrates how to bind a Java string to an NCHAR column.

int employee_id = 12345;
String last_name = "Joe"
oracle.jdbc.OraclePreparedStatement pstmt =
    (oracle.jdbc.OraclePreparedStatement)
    conn.prepareStatement("INSERT INTO employees (last_name, employee_id) 
    VALUES    (?, ?)");
pstmt.setFormOfUse(1, oracle.jdbc.OraclePreparedStatement.FORM_NCHAR);
pstmt.setString(1, last_name);
pstmt.setInt(2, employee_id);
pstmt.execute();                 /* execute to insert into first row */
employee_id += 1;                      /* next employee number */
last_name = "\uFF2A\uFF4F\uFF45";    /* Unicode characters in name */
pstmt.setString(1, last_name);
pstmt.setInt(2, employee_id);
pstmt.execute();                 /* execute to insert into second row */

You can define the target SQL NCHAR columns by specifying their datatypes, forms of use, and lengths. JDBC uses this information to optimize the performance of fetching SQL NCHAR data from these columns. The following is an example of defining a SQL NCHAR column.

OraclePreparedStatement pstmt = (OraclePreparedStatement)
     conn.prepareStatement("SELECT ename, empno from emp");
   pstmt.defineColumnType(1,Types.VARCHAR, 3, 
OraclePreparedStatement.FORM_NCHAR);
   pstmt.defineColumnType(2,Types.INTEGER);
   ResultSet rest = pstmt.executeQuery();
   String name = rset.getString(1);
   int id = reset.getInt(2);

To define a SQL NCHAR column, you need to specify the datatype that is equivalent to a SQL CHAR column in the first argument, the length in number of characters in the second argument, and the form of use in the fourth argument of defineColumnType().

You can bind or define a Java string against an NCHAR column without explicitly specifying the form of use argument. This implies the following:

• If you do not specify the argument in the setString() method, then JDBC assumes that the bind or define variable is for the SQL CHAR column. As a result, it tries to convert them to the database character set. When the data gets to the database, the database implicitly converts the data in the database character set to the national character set. During this conversion, data can be lost when the database character set is a subset of the national character set. Because the national character set is either UTF8 or AL16UTF16, data loss would happen if the database character set is not UTF8 or AL32UTF8.

• Because implicit conversion from SQL CHAR to SQL NCHAR datatypes happens in the database, database performance is degraded.

In addition, if you bind or define a Java string for a column of SQL CHAR datatypes but specify the form of use argument, then performance of the database is degraded. However, data should not be lost because the national character set is always a larger set than the database character set.

Using the SQL NCHAR Datatypes Without Changing the Code

A Java system property has been introduced in the Oracle JDBC drivers for customers to tell whether the form of use argument should be specified by default in a Java application. This property has the following purposes:

• Existing applications accessing the SQL CHAR datatypes can be migrated to support the SQL NCHAR datatypes for worldwide deployment without changing a line of code.

• Applications do not need to call the setFormOfUse() method when binding and defining a SQL NCHAR column. The application code can be made neutral and independent of the datatypes being used in the backend database. With this property set, applications can be easily switched from using SQL CHAR or SQL NCHAR.

The Java system property is specified in the command line that invokes the Java application. The syntax of specifying this flag is as follows:

java -Doracle.jdbc.defaultNChar=true <application class>

With this property specified, the Oracle JDBC drivers assume the presence of the form of use argument for all bind and define operations in the application.

If you have a database schema that consists of both the SQL CHAR and SQL NCHAR columns, then using this flag may have some performance impact when accessing the SQL CHAR columns because of implicit conversion done in the database server.

Using SQL NCHAR String Literals in JDBC

When using NCHAR string literals in JDBC, there is a potential for data loss because characters are converted to the database character set before processing. See NCHAR String Literal Replacement for more details.

The desired behavior for preserving the NCHAR string literals can be achieved by enabling the property set oracle.jdbc.convertNcharLiterals. If the value is true, then this option is enabled; otherwise, it is disabled. The default setting is false. It can be enabled in two ways: a) as a Java system property or b) as a connection property. Once enabled, conversion is performed on all SQL in the VM (system property) or in the connection (connection property). For example, the property can be set as a Java system property as follows:

java -Doracle.jdbc.convertNcharLiterals="true" ...

Alternatively, you can set this as a connection property as follows:

Properties props = new Properties();
...
props.setProperty("oracle.jdbc.convertNcharLiterals", "true");
Connection conn = DriverManager.getConnection(url, props);

If you set this as a connection property, it overrides a system property setting.

Data Conversion in JDBC

Because Java strings are always encoded in UTF-16, JDBC drivers transparently convert data from the database character set to UTF-16 or the national character set. The conversion paths taken are different for the JDBC drivers:

• Data Conversion for the OCI Driver

• Data Conversion for Thin Drivers

• Data Conversion for the Server-Side Internal Driver

Using oracle.sql.CHAR in Oracle Object Types

JDBC drivers support Oracle object types. Oracle objects are always sent from database to client as an object represented in the database character set or national character set. That means the data conversion path in "Data Conversion in JDBC" does not apply to Oracle object access. Instead, the oracle.sql.CHAR class is used for passing SQL CHAR and SQL NCHAR data of an object type from the database to the client.

This section includes the following topics:

• oracle.sql.CHAR

• Accessing SQL CHAR and NCHAR Attributes with oracle.sql.CHAR

CREATE TYPE person_type AS OBJECT (
   name VARCHAR2(30), address NVARCHAR2(256), age NUMBER);
CREATE TABLE employees (id NUMBER, person PERSON_TYPE);

public class person implement SqlData  
{
   oracle.sql.CHAR name;
   oracle.sql.CHAR address;
   oracle.sql.NUMBER age;
   // SqlData interfaces
   getSqlType() {...}
   writeSql(SqlOutput stream) {...}
   readSql(SqlInput stream, String sqltype) {...}
}

TypeMap map = ((OracleConnection)conn).getTypeMap();
map.put("PERSON_TYPE", Class.forName("person"));
conn.setTypeMap(map);
    .       .        .
    .       .        .
ResultSet rs = stmt.executeQuery("SELECT PERSON FROM EMPLOYEES");
rs.next();
person p = (person) rs.getObject(1);
oracle.sql.CHAR sql_name = p.name;
oracle.sql.CHAR sql_address=p.address;
String java_name = sql_name.getString();
String java_address = sql_address.getString();

Restrictions on Accessing SQL CHAR Data with JDBC

This section contains the following topic:

• Character Integrity Issues in a Multibyte Database Environment

Character Integrity Issues in a Multibyte Database Environment

Oracle JDBC drivers perform character set conversions as appropriate when character data is inserted into or retrieved from the database. The drivers convert Unicode characters used by Java clients to Oracle database character set characters, and vice versa. Character data that makes a round trip from the Java Unicode character set to the database character set and back to Java can suffer some loss of information. This happens when multiple Unicode characters are mapped to a single character in the database character set. An example is the Unicode full-width tilde character (0xFF5E) and its mapping to Oracle's JA16SJIS character set. The round-trip conversion for this Unicode character results in the Unicode character 0x301C, which is a wave dash (a character commonly used in Japan to indicate range), not a tilde.

Figure 7-2 shows the round-trip conversion of the tilde character.

Figure 7-2 Character Integrity

Description of "Figure 7-2 Character Integrity"

This issue is not a bug in Oracle's JDBC. It is an unfortunate side effect of the ambiguity in character mapping specifications on different operating systems. Fortunately, this problem affects only a small number of characters in a small number of Oracle character sets such as JA16SJIS, JA16EUC, ZHT16BIG5, and KO16KS5601. The workaround is to avoid making a full round-trip with these characters.

Unicode-Enabled Drivers in ODBC and OLE DB

Oracle's ODBC driver and Oracle Provider for OLE DB can handle Unicode data properly without data loss. For example, you can run a Unicode ODBC application containing Japanese data on English Windows if you install Japanese fonts and an input method editor for entering Japanese characters.

Oracle provides ODBC and OLE DB products for Windows platforms only. For Unix platforms, contact your vendor.

OCI Dependency in Unicode

OCI Unicode binding and defining features are used by the ODBC and OLE DB drivers to handle Unicode data. OCI Unicode data binding and defining features are independent from NLS_LANG. This means Unicode data is handled properly, irrespective of the NLS_LANG setting on the platform.

ODBC and OLE DB Code Conversion in Unicode

In general, no redundant data conversion occurs unless you specify a different client datatype from that of the server. If you bind Unicode buffer SQL_C_WCHAR with a Unicode data column like NCHAR, for example, then ODBC and OLE DB drivers bypass it between the application and OCI layer.

If you do not specify datatypes before fetching, but call SQLGetData with the client datatypes instead, then the conversions in Table 7-7 occur.

You must specify the datatype for inserting and updating operations.

The datatype of the ODBC client buffer is given when you call SQLGetData but not immediately. Hence, SQLFetch does not have the information.

Because the ODBC driver guarantees data integrity, if you perform implicit bindings, then redundant conversion may result in performance degradation. Your choice is the trade-off between performance with explicit binding or usability with implicit binding.

ODBC Unicode Datatypes

In ODBC Unicode applications, use SQLWCHAR to store Unicode data. All standard Windows Unicode functions can be used for SQLWCHAR data manipulations. For example, wcslen counts the number of characters of SQLWCHAR data:

SQLWCHAR sqlStmt[] = L"select ename from emp";
len = wcslen(sqlStmt);

Microsoft's ODBC 3.5 specification defines three Unicode datatype identifiers for the SQL_C_WCHAR, SQL_C_WVARCHAR, and SQL_WLONGVARCHAR clients; and three Unicode datatype identifiers for servers SQL_WCHAR, SQL_WVARCHAR, and SQL_WLONGVARCHAR.

For binding operations, specify datatypes for both client and server using SQLBindParameter. The following is an example of Unicode binding, where the client buffer Name indicates that Unicode data (SQL_C_WCHAR) is bound to the first bind variable associated with the Unicode column (SQL_WCHAR):

SQLBindParameter(StatementHandle, 1, SQL_PARAM_INPUT, SQL_C_WCHAR,
SQL_WCHAR, NameLen, 0, (SQLPOINTER)Name, 0, &Name);

Table 7-9 represents the datatype mappings of the ODBC Unicode datatypes for the server against SQL NCHAR datatypes.

According to ODBC specifications, SQL_WCHAR, SQL_WVARCHAR, and SQL_WLONGVARCHAR are treated as Unicode data, and are therefore measured in the number of characters instead of the number of bytes.

OLE DB Unicode Datatypes

OLE DB offers the wchar_t, BSTR, and OLESTR datatypes for a Unicode C client. In practice, wchar_t is the most common datatype and the others are for specific purposes. The following example assigns a static SQL statement:

wchar_t *sqlStmt = OLESTR("SELECT ename FROM emp");

The OLESTR macro works exactly like an "L" modifier to indicate the Unicode string. If you need to allocate Unicode data buffer dynamically using OLESTR, then use the IMalloc allocator (for example, CoTaskMemAlloc). However, using OLESTR is not the normal method for variable length data; use wchar_t* instead for generic string types. BSTR is similar. It is a string with a length prefix in the memory location preceding the string. Some functions and methods can accept only BSTR Unicode datatypes. Therefore, BSTR Unicode string must be manipulated with special functions like SysAllocString for allocation and SysFreeString for freeing memory.

Unlike ODBC, OLE DB does not allow you to specify the server datatype explicitly. When you set the client datatype, the OLE DB driver automatically performs data conversion if necessary.

Table 7-10 illustrates OLE DB datatype mapping.

If DBTYPE_BSTR is specified, then it is assumed to be DBTYPE_WCHAR because both are Unicode strings.

ADO Access

ADO is a high-level API to access database with the OLE DB and ODBC drivers. Most database application developers use the ADO interface on Windows because it is easily accessible from Visual Basic, the primary scripting language for Active Server Pages (ASP) for the Internet Information Server (IIS). To OLE DB and ODBC drivers, ADO is simply an OLE DB consumer or ODBC application. ADO assumes that OLE DB and ODBC drivers are Unicode-aware components; hence, it always attempts to manipulate Unicode data.

Writing an XML File in Unicode with Java

A common mistake in reading and writing XML files is using the Reader and Writer classes for character input and output. Using Reader and Writer for XML files should be avoided because it requires character set conversion based on the default character encoding of the runtime environment.

For example, using FileWriter class is not safe because it converts the document to the default character encoding. The output file can suffer from a parsing error or data loss if the document contains characters that are not available in the default character encoding.

UTF-8 is popular for XML documents, but UTF-8 is not usually the default file encoding for Java. Thus using a Java class that assumes the default file encoding can cause problems.

The following example shows how to avoid these problems:

import java.io.*;
import oracle.xml.parser.v2.*;

public class I18nSafeXMLFileWritingSample 
{
  public static void main(String[] args) throws Exception
  {
    // create a test document
    XMLDocument         doc  = new XMLDocument();
    doc.setVersion( "1.0" );
    doc.appendChild(doc.createComment( "This is a test empty document." ));
    doc.appendChild(doc.createElement( "root" ));
    
    // create a file
    File                file = new File( "myfile.xml" );

    // create a binary output stream to write to the file just created
    FileOutputStream    fos  = new FileOutputStream( file );

    // create a Writer that converts Java character stream to UTF-8 stream
    OutputStreamWriter  osw  = new OutputStreamWriter( fos, "UTF8" );

    // buffering for efficiency
    Writer              w    = new BufferedWriter( osw );

    // create a PrintWriter to adapt to the printing method
    PrintWriter         out  = new PrintWriter( w );

    // print the document to the file through the connected objects
    doc.print( out ); 
  }
}

Reading an XML File in Unicode with Java

Do not read XML files as text input. When reading an XML document stored in a file system, use the parser to automatically detect the character encoding of the document. Avoid using a Reader class or specifying a character encoding on the input stream. Given a binary input stream with no external encoding information, the parser automatically figures out the character encoding based on the byte order mark and encoding declaration of the XML document. Any well-formed document in any supported encoding can be successfully parsed using the following sample code:

import java.io.*;
import oracle.xml.parser.v2.*;

public class I18nSafeXMLFileReadingSample 
{
  public static void main(String[] args) throws Exception
  {
    // create an instance of the xml file
    File                 file = new File( "myfile.xml" );

    // create a binary input stream
    FileInputStream      fis  = new FileInputStream( file );

    // buffering for efficiency
    BufferedInputStream  in = new BufferedInputStream( fis );

    // get an instance of the parser
    DOMParser  parser = new DOMParser();

    // parse the xml file
    parser.parse( in );
  }
}

Parsing an XML Stream in Unicode with Java

When the source of an XML document is not a file system, the encoding information is usually available before reading the document. For example, if the input document is provided in the form of a Java character stream or Reader, its encoding is evident and no detection should take place. The parser can begin parsing a Reader in Unicode without regard to the character encoding.

The following is an example of parsing a document with external encoding information:

import java.io.*;
import java.net.*;
import org.xml.sax.*;
import oracle.xml.parser.v2.*;

public class I18nSafeXMLStreamReadingSample 
{
  public static void main(String[] args) throws Exception
  {
    // create an instance of the xml file
    URL  url = new URL( "http://myhost/mydocument.xml" );

    // create a connection to the xml document    
    URLConnection  conn = url.openConnection();

    // get an input stream
    InputStream  is = conn.getInputStream();

    // buffering for efficiency
    BufferedInputStream  bis = new BufferedInputStream( is );

    /* figure out the character encoding here                              */
    /* a typical source of encoding information is the content-type header */
    /* we assume it is found to be utf-8 in this example                   */
    String  charset = "utf-8";

    // create an InputSource for UTF-8 stream
    InputSource  in = new InputSource( bis );
    in.setEncoding( charset );
    
    // get an instance of the parser
    DOMParser  parser = new DOMParser();

    // parse the xml stream
    parser.parse( in );  
  }
}