i am using pg7.3.4. i have written a table function in c that tokenizes a text, returning a setof text. i can use this function either in constructs like:
select * from tokenize( 'some text that has to be tokenized' );
or in costructs like:
select tokenize( text_field ) from some_table;
the latter gives me the whole result of all words contained in every "text_field" of "some_table". i wonder if this is a forseen behaviour, since doing the same with a plpgsql function does not work (the plpgsql interpreter raises an error when it reaches the "return next" statement telling i'm using a "set-valued function in a context that can not accept a set"). but the c function works fine. now i'm trying the 8beta, where my "tokenize" function does not work anymore in the second construct. the server does not alert me, it simply returns an empy record set. this behaviour for sure not the right one since it should raise an error, but mainly i would really like it to work at least as it did on 7.3.4 since, as i believe my little example shows, this can be very useful. does anybody know what is the forseen behavior for using a set returning function in the field list of a select and, if the answer is "it should raise an error", if it is possible to work around it?
thanks in advance,
max.
ps: i did not subscribe the mailing list, so please use "replay all"..
---------------------------(end of broadcast)---------------------------
TIP 9: the planner will ignore your desire to choose an index scan if your
joining column's datatypes do not match
