Re: [boost] Boost.SQL?

Subject: Re: [boost] Boost.SQL?
From: Roland Bock (rbock_at_[hidden])
Date: 2010-09-14 10:38:21

On 09/14/2010 04:12 PM, Dominique Devienne wrote:
> On Tue, Sep 14, 2010 at 8:26 AM, Roland Bock<rbock_at_[hidden]>  wrote:
>    
>> currently I am working on a set of header files which are supposed to make
>> the life of C++ developers easier when in comes to interacting with
>> databases (based on SQL). Goals are:
>>
>> * Stay close to SQL syntax
>> * Do as much type checking as possible
>>      
> Static type checking is always a plus. I tried going that route too,
> but found it a bit cumbersome (and my C++ lacking too ;) One of the
> problem with trying to generate queries like this is also the
> complexity of the SQL language itself, with nested queries, and all
> the various extensions and idiosyncrasies of the different SQL
> vendors. I've lurked on the Oracle forums for a few months, and it's
> open my eyes to the complexity of SQL (the Oracle dialect at least).
>    
Yes, the code looks "interesting" so far. And I have learned quite a lot 
about boost.mpl in the past few weeks :-)

I think I have solved (conceptually) the problem of nested queries, but 
the differences between the different vendors are certainly a huge problem.

>> The basic idea is, that given a table definition, it should be possible to
>> * offer a lot of checks at compile time for building a query
>> * make it hard to misinterpret the results
>>      
> The few C++ libraries I've seen in this area take a less ambitious
> approach, and typically provide instead an easy way to pass typed C++
> values as SQL bind values, and get back the records as typed tuples
> for examples. The SQL text must still be correct, and will be parsed
> and type checked only at runtime by the DB, but in practice, with unit
> testing, I find that it works fairly well.
>    
Sure, but you have to do quite some testing to prevent all the possible 
oversights from happening. For instance, if you have a tuple of strings, 
it is fairly simple to mix them up. If the compiler could give you names 
for it, it would be harder to misuse.

> A good and established library already mentioned on this list in this
> domain is http://soci.sourceforge.net/, which uses modern C++ and
> Boost.
>    
I've seen it. It is certainly one of the best I've seen so far.

> If you go that route, you'll need a way to generate the static C++
> definition of the tables from existing Schemas, otherwise maintaining
> these to be in sync might be problematic. Then again the same applies
> to all the SQL text in more traditional approaches.
>    
Right, as of now, we are using a small DDL->Header conversion program.

> My own feeling is that you should explore this subject into your own
> public project for while, and try to attract a community of
> volunteers. Then again, once you've reached some kind of milestone you
> may want to revisit a possible submission to Boost. But I'm certainly
> no expert in the matter. Good luck to you, --DD
>    

I would have assumed one year minimum before it comes even close. And I 
am pretty sure I could not cover all aspects all on my own (e.g. OS, 
DBS, compiler).

Regards,

Roland