appear below the point where the handler is set, unless a
    different action was set for the same condition between the first
    <literal>EXEC SQL WHENEVER</literal> and the SQL statement causing
    the condition, regardless of the flow of control in the C program.
    So neither of the two following C program excerpts will have the
    desired effect:
<programlisting>
/*
 * WRONG
 */
int main(int argc, char *argv[])
{
    ...
    if (verbose) {
        EXEC SQL WHENEVER SQLWARNING SQLPRINT;
    }
    ...
    EXEC SQL SELECT ...;
    ...
}
</programlisting>

<programlisting>
/*
 * WRONG
 */
int main(int argc, char *argv[])
{
    ...
    set_error_handler();
    ...
    EXEC SQL SELECT ...;
    ...
}

static void set_error_handler(void)
{
    EXEC SQL WHENEVER SQLERROR STOP;
}
</programlisting>
   </para>
  </sect2>

  <sect2 id="ecpg-sqlca">
   <title>sqlca</title>

   <para>
    For more powerful error handling, the embedded SQL interface
    provides a global variable with the name <varname>sqlca</varname>
    (SQL communication area)
    that has the following structure:
<programlisting>
struct
{
    char sqlcaid[8];
    long sqlabc;
    long sqlcode;
    struct
    {
        int sqlerrml;
        char sqlerrmc[SQLERRMC_LEN];
    } sqlerrm;
    char sqlerrp[8];
    long sqlerrd[6];
    char sqlwarn[8];
    char sqlstate[5];
} sqlca;
</programlisting>
    (In a multithreaded program, every thread automatically gets its
    own copy of <varname>sqlca</varname>.  This works similarly to the
    handling of the standard C global variable
    <varname>errno</varname>.)
   </para>

   <para>
    <varname>sqlca</varname> covers both warnings and errors.  If
    multiple warnings or errors occur during the execution of a
    statement, then <varname>sqlca</varname> will only contain
    information about the last one.
   </para>

   <para>
    If no error occurred in the last <acronym>SQL</acronym> statement,
    <literal>sqlca.sqlcode</literal> will be 0 and
    <literal>sqlca.sqlstate</literal> will be
    <literal>"00000"</literal>.  If a warning or error occurred, then
    <literal>sqlca.sqlcode</literal> will be negative and
    <literal>sqlca.sqlstate</literal> will be different from
    <literal>"00000"</literal>.  A positive
    <literal>sqlca.sqlcode</literal> indicates a harmless condition,
    such as that the last query returned zero rows.
    <literal>sqlcode</literal> and <literal>sqlstate</literal> are two
    different error code schemes; details appear below.
   </para>

   <para>
    If the last SQL statement was successful, then
    <literal>sqlca.sqlerrd[1]</literal> contains the OID of the
    processed row, if applicable, and
    <literal>sqlca.sqlerrd[2]</literal> contains the number of
    processed or returned rows, if applicable to the command.
   </para>

   <para>
    In case of an error or warning,
    <literal>sqlca.sqlerrm.sqlerrmc</literal> will contain a string
    that describes the error.  The field
    <literal>sqlca.sqlerrm.sqlerrml</literal> contains the length of
    the error message that is stored in
    <literal>sqlca.sqlerrm.sqlerrmc</literal> (the result of
    <function>strlen()</function>, not really interesting for a C
    programmer).  Note that some messages are too long to fit in the
    fixed-size <literal>sqlerrmc</literal> array; they will be truncated.
   </para>

   <para>
    In case of a warning, <literal>sqlca.sqlwarn[2]</literal> is set
    to <literal>W</literal>.  (In all other cases, it is set to
    something different from <literal>W</literal>.)  If
    <literal>sqlca.sqlwarn[1]</literal> is set to
    <literal>W</literal>, then a value was truncated when it was
    stored in a host variable.  <literal>sqlca.sqlwarn[0]</literal> is
    set to <literal>W</literal> if any of the other elements are set
    to indicate a warning.
   </para>

   <para>
    The fields <structfield>sqlcaid</structfield>,
    <structfield>sqlabc</structfield>,
    <structfield>sqlerrp</structfield>,