I have a piece of C++ code in which I try to update a record for a
primary key. If the record does not exist, an exception is raised and
the data is treated as an insert by hitting the exception logic.
Someone pointed out that this is expensive because of the stack unwind
on the exception logic, but I argue that it is still faster because
finding out if an primary record exists takes time. Others argue that
exception logic should be used for exceptional events rather than as a
kludge which happens by circumstance.
Is this acceptable?

----------example ----------------
int foo::Upate(string stIn){


try{
m_Conn.sql = "update tab1 set foo=" + stIn.c_str() +" where
some=cond ";
m_Conn.exec();
catch(SQLException ex){
m_Conn.sql = "insert into tab1 values(" + stIn.c_str() +" , cond)
";
m_Conn.exec();
}

}

On 2011-02-23 21:34, Deodiaus wrote:
You don't mention what version of db2 you are using, but I recommend
that you have a look at MERGE:

http://publib.boulder.ibm.com/infoce...c/r0010873.htm

/Lennart

On 2011-02-23 23:27, Lennart Jonsson wrote:
I for whatever reason you can't or don't want to use MERGE, be more
specific in your catch clause. I.e. only insert when you catch a "no
rows found" exception.

If the key "rarely" exists it might be better to start with the insert,
catch "key violation", and do the update.

/Lennart