Sorry, meant to send the previous message to pgsql-patches.
Here's a version cut using cvs diff so it's usable with -p0
I added one more fixup. There was a silly test in toast_fetch_datum_slice()
which handled compressed datums. Returning a slice of a compressed datum is
nonsensical with toast since the resulting datum would be useless. I also
added an assertion in this function that the datum is external before we treat
it as a toast_pointer.
(Incidentally, I did eventually manage to construct a case to reach all the pfrees.)
--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com