On Sat, Mar 26, 2011 at 4:36 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> ** Selecting a field from a record-returning function's output.
> Currently, we'll use the field's declared collation; except that
> if the field has default collation, we'll replace that with the common
> collation of the function's inputs, if any. Is either part of that
> sane? Do we need to make this work for functions invoked with other
> syntax than a plain function call, eg operator or cast syntax?
Either of those sounds reasonable but the combination seems weird.
Some example functions might help:
list_words('foo bar bar') -> ('foo'), ('bar'), ('baz')
fetch_users_by_lastname('Smith') -> (1,'John','Smith','Great
Britain','GB'), (2,'Jürgen','Smith','DE')
fetch_users_by_countrycode('DE') -> (2,'Jürgen','Smith','DE')
The first looks like it should definitely inherit. The second the
result set is heterogeneous and inheriting might be the best
compromise but it would produce very strange results for columns like
the country-code which should just use their defined collation of C.
The third case inheriting the country code's collation of C would be
very strange and definitely wrong.
It occurs to me that if we have any inherited cases people might come
to depend on that behaviour and there would be no out for us. Whereas
if we say record return values always use the record type's field's
collations then we could always later add a collation of type
_inherited or _anycollation or some such that indicated that that
column should inherit the arguments' collation and it wouldn't affect
any existing code.
--
greg