Re^2: DBI bind_param_inout trick

[mje]

in reply to Re: DBI bind_param_inout trick
in thread DBI bind_param_inout trick

Looks like a bug in DBD::ODBC. Parameters are bound initially in ODBC at bind time and if they change at execute time they have to be rebound ODBC-wise. The following code is what tests whether a parameter needs to be rebound:

if (SvTYPE(phs->sv) != phs->sv_type /* has the type changed? */
   || (SvOK(phs->sv) && !SvPOK(phs->sv)) /* is there still a string? *
+/
   || SvPVX(phs->sv) != phs->sv_buf /* has the string buffer moved? */
                ) {
                if (!rebind_param(sth, imp_sth, imp_dbh, phs))
                    croak("Can't rebind placeholder %s", phs->name);
               }

Trouble is I didn't write that code which attempts to rebind the parameter ODBC-wise and although I understand the first and last test in the if condition the comment next to "(SvOK(phs->sv) && !SvPOK(phs->sv))" does not seem to agree with the code. I wonder what the original author meant. It also does not cater for a parameter being undef then a string.

[mje]

This is now fixed in 1.40_1. There is a small discussion on the dbi-dev mailing list if you are interested.

[runrig]

Awesome! I expected this to not be supported in some DBDs, but if it fleshes out bugs where it is supposed to be supported, then it's done more than originally intended!

[mje]

Thanks for the example runrig. The problem was fairly specific in that it only went wrong it you called bind_param_inout with undef values initially then changed them to strings shorter than 28 characters. However, it was certainly a bug and I was a little surprised no one has hit it before.