Re: [boost] Interest in an ODBC library?

From: Maciej Sobczak (prog_at_[hidden])
Date: 2005-07-27 07:31:33

Hajo Kirchhoff wrote:

> I'd like to gauge interest in an odbc wrapper library.

This subject is recurring, with some peaks divided by long periods of 
complete silence. :)

> I am currently
> writing one and the most important difference to other libraries I've 
> seen is the ease with which my library allows binding variables to odbc 
> columns and parameters.
> 
> Given a table in the database named 'my_table' and a struct my_table,
> this would be the shortest code to retrieve all rows of that table and
> do something with them:
> 
> odbc::table t;
> my_table data;
> t.bind(data);	//!!! This does all the binding!
> t.open();
> while (t.fetch().ok()) {
> }

What I don't like in your approach is that you require that there is a 
type (my_table above) for each different rowset traversal "profile" - 
this is a table in your case, but why not query just a subset of rows? I 
have seen an application with a table containing 150 columns (no joke, 
and add to this joins with other tables) and of course, all queries were 
interested in only some of the columns - from 2 to 5, typically. Do you 
require that the application always retrieves all columns?

Just imagine a single table with 4 columns - there are 15 different 
subsets of columns there (not counting the queries with no columns). 
Now, imagine a table with 10 columns. What about 150? What about joins?
You does your approach scale in this aspect?

In my humble opinion, a DB library needs to be usable also with the 
assumption that each query in a program can potentially retrieve a 
*different* set of attributes (columns).

> Note that this is the full, actual working code, not pseudo code.

Take a look at this:

http://soci.sourceforge.net/

In the context of Boost, you might be interested in this informal page:

http://www.crystalclearsoftware.com/cgi-bin/boost_wiki/wiki.pl?Relational_Database_Access

-- 
Maciej Sobczak
http://www.msobczak.com