Re: Class::DBI::Factory and Apache::DBI don't play nice

[prev] [thread] [next] [Date index for 2004/11/21]

On 21 Nov 2004, at 03:12, Perrin Harkins wrote:

> William Ross wrote:
>> The version of CDF that's on CPAN isn't up to date with Perrin's 
>> object cache. You need to add a:
>> $Class::DBI::Weaken_Is_Available = 0;     # disables unique-object 
>> stash
>
> Why is that?  Because the object index is class data but CDF is 
> changing the database underneath?  Maybe we could incorporate a unique 
> identifier for the database connection into the object index key.

Exactly that, yes. And an arbitrary connection-identifier would suit 
CDF's way of doing things very well, but perhaps it would be simpler 
just to route everything through an object_index accessor that can be 
overridden?

>> It also, separately, doesn't get on with Apache::DBI. I believe 
>> that's a general issue with Ima::DBI and Apache::DBI disagreeing 
>> about the caching of handles and not something that I've introduced, 
>> but if anyone would like to disagree I'd be delighted to fix it.
>
> It's probably not your fault, but you'd have to describe the problem 
> to be sure.  The big issue with Apache::DBI and Ima::DBI is that 
> Ima::DBI caches the handles in closures and doesn't go back to 
> Apache::DBI when you ask for them again, so the cleanup handler with 
> the automatic rollback for transactional databases doesn't get pushed 
> onto the stack by Apache::DBI.

Yes. CDF makes its own call to Ima::DBI->_mk_db_closure, in order to 
provide the kind of enclosured DBIx::ContextualFetch object that cdbi 
expects, so it will definitely inherit those problems but hopefully add 
no more of its own.

thanks

will