Oliver Jowett wrote:
> I guess that we should tighten the checks in the query executor so that
> it will only consider types "compatible enough" if the new value is both
> of unspecified type *and* null?
With that change the test just fails differently:
ERROR: could not determine data type of parameter $1
since the test query ("SELECT ?") is actually one of the cases where
string-as-unspecified-type does not work.
I changed it to use a different query that still appears to exercise the
type-changing behaviour correctly ("SELECT CAST (? AS TEXT)") but
doesn't fail with a parameter of unspecified type.
-O