Re: DBD::Mock

[jim cromie]

Tim Bunce wrote:

On Fri, Jan 30, 2004 at 12:27:02PM -0700, jim cromie wrote:
 

Tim Bunce wrote:

   

On Fri, Jan 30, 2004 at 03:00:37AM -0700, jim cromie wrote:

     

This sounds like it could plug into the trace mechanism, either 
increasing its
leverage, simplifying its implementation, or both.
       

I've long wanted the DBI to have code hooks for method call entry & exit
that could be used for building all sorts of funky tools.
     

And having just crept up to a line, let me cross it.  Can new DBI2 have
2 logging levels under transactions ?  Anything that commits (success)
logs at 1 or 2, things that rollback are logged at 5 or 6. This should
help unravel anomalous behavior by automatically putting your fovea
on it.
       

speculative logging:  The basic idea is to buffer up a log as a 
transaction proceeds,
and decide afterwards what details of the log you want to keep - ie emit 
to the logging device. 

Commit/rollback is the most obvious decision point, if youve got a
rollback, or an error, dump the log-trace since the start-transaction at 
a high level of detail, if all went smoothly, log only at a low trace-level.
   

Ah, okay. No, I wouldn't add that to the DBI, but if I add hooks
for method call entry & exit that could be used to implement what
you want - but do it for all drivers.
 

It seems that all tracing/logging activity could be implemented by these 
hooks.
assuming DBI dispatch is done by AUTOLOAD, (for sake of discussion)

sub AUTOLOAD {

   # call entry hooks
   $_->(@_) foreach ($callbacks{"${AUTOLOAD}_entry"});  

    # call method itself, whether its on $dbh, $drh, etc..
   @res = $AUTOLOAD->(@_);
   # call post hooks
   $_->([EMAIL PROTECTED], [EMAIL PROTECTED]) foreach $callbacks{"${AUTOLOAD}_exit"});
}
and trace() just controls those callback lists kept for each hook.

(That reminds me, anothet component is to allow tracing to a
 

filehandle so a tied handle can be used, for example to gether trace
output into a string.)
 

there seems (to me anyway) some advantage to treating each
trace/set_error/info item separately.  If autoflush on the handle can be 
turned off,
theres some possiblity of subclassing, and filtering the items when the 
queue is committed
to the file/socket/logging-object.

having named callbacks might be better than in an array, so that the 
trace-cb
is easy to replace wo disturbing info-cb, error-cb, etc..

But I confess, im not sure anymore whether were talking dbi-2 or 1 now.

Tim.

.