Deferred RI trigger for non-key UPDATEs and subxacts

With some time to spare, I thought I'd submit a quick-fix patch to the
issue I reported here:

http://archives.postgresql.org/pgsql-hackers/2007-07/msg00339.php

This should preclude optimizing away a deferred RI trigger if the
UPDATEd row (in the FK table) was inserted by "current" transaction
(i.e. defined by TransactionIdIsCurrentTransactionId()) and not
necessarily "by our own transaction" as the code currently says.
________________________________________________________________________

backend/commands/trigger.c | 17 !!!!!!!!!!!!!!!!!
test/regress/expected/foreign_key.out | 15 +++++++++++++++
test/regress/sql/foreign_key.sql | 19 +++++++++++++++++++
3 files changed, 34 insertions(+), 17 modifications(!)
________________________________________________________________________

--
Affan Salman
EnterpriseDB Corporation http://www.enterprisedb.com