Actually, I think it is a fault in the PHP driver. My hypothesis is
that after a transaction error all cursors on the attachment are
invalidated that the driver then attempts to drop it once more.

There are IBDEBUG statements in the source around the allocation and
release of statements, so I assume it was something that Art was
working on.

What I need is a complete but simple script that shows the bug. Your
example is nearly it, it just needs the initial connection and some
table defs / data rows.

Paul