Considering first just the addition of the table-name or correlation-name,
without a schema-name qualifier:
UPDATE table-name [ AS correlation-name ]
SET [ { table-name | correlation-name } '.' ] column-name opt_indirection
'=' value-expression ';'
we take the first sql92identifier after the SET as the presumed column-name,
and we create a list for the indirection elements, if any.
So for the ambiguous case (t.c versus c.f), the first token must match the
table-name (or correlation-name, if supplied), and the second token must be
the name of a column of table t (i.e. that t.t exists), and also t.c must
have a composite type. If all of these are true, we can parse this as c.f,
just as we do now, so existing PostgreSQL applications would have the same
semantics.
if an Oracle application has an UPDATE statement written as:
myTable.myTable = scalarValue
when myTable is a column of myTable, and myTable.myTable has a composite
type, but scalarValue is not coercible into the full composite value of that
type, then you'd get an error. So, don't do that, because we will interpret
the meaning of this corner case like PostgreSQL has always interpreted it.
Adding the optional schema-name doesn't expand the scope of that corner case
very much, although there are the usual issues with search_path.
-----
Jim Finnerty, AWS, Amazon Aurora PostgreSQL
--
Sent from: http://www.postgresql-archive.org/PostgreSQL-hackers-f1928748.html
