On Mon, 18 Sep 2006, Tim Bunce wrote:
Yes, and the same applies to AutoCommit. Whether it's a good thing or not is debatable but it's been that way forever. The 'workaround' is to explicitly state the attributes you want.
Hmmm, interesting. What would you think about making PrintError default to 0 when RaiseError is on? My reasoning is that if the error is getting croak()ed then there's no need to warn() it too. The warn() comes without a stack-trace and can be very difficult to track down when the croak() is being intentionally trapped (as was the case in DBIx::Timeout). When the croak() isn't being trapped it's just duplicate information on STDERR. -sam