From: Warren Young
Date: August 15 2007 6:27am
Subject: Eyeballs needed on new reference counted pointer template
List-Archive: http://lists.mysql.com/plusplus/6930
Message-Id: <46C29CC6.9060005@etr-usa.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Smart pointers are tricky, and automatic memory management is even 
trickier, and I just created a template that does both.  I swapped it in 
for existing code in MySQL++ that used value semantics and it passes the 
test suite, but I'd still appreciate someone else taking a look at it.

http://svn.gna.org/viewcvs/mysqlpp/trunk/lib/refcounted.h?view=markup

This is part of a general effort toward improving the way MySQL++ 
handles memory and object copies.

The particular need that sparked the creation of this template is that 
in released MySQL++ versions, you cannot use a Row object after the 
Result/ResUse object that created it is destroyed or reused itself. 
This doesn't seem to catch too many newbies, but we do see it from time 
to time.

I built a similar, but more primitive, mechanism last week for ColData, 
which should improve performance of field access greatly: v2.3 copied 
each field two or three times on every access, because it had to do 
everything with value semantics to avoid similar problems to the one 
described above.  I hope to move ColData's buffer management to this new 
template, but I'm not yet certain how it's going to work yet.

And yes, I am aware that this wheel has already been invented.  It's in 
Boost, for instance, but the last time I brought up the possibility of 
making Boost a dependency of MySQL++, it was shot down so violently that 
the detritus from the explosion settled in a thin even layer of powdery 
ash over six counties.