Marshall,

This is an offshoot of the topic where you are exploring the comparison of
graph databases with systems of relations.
It's so tangential that I chose to start a new thread, rather than hijack
your thread.

When DEC built Rdb, in the early 1980s,  they included a datatype called
DBKEY.  The contents of a DBKEY, in some address space, was really the
address of another record.  Yeah, "record".

Their nomenclature at the time was to talk about "relations", "records",
"fields", and "global fields".  A few years later, they were talking about
"tables", "rows", "columns" and "domains", after they adopted SQL.   But the
SW engineering was really the same idea.

In any event,  the availability of DBKEY as a datatype meant that a DB
designer could, if he (meaning he or she) wanted to, extend the relational
model by implementing his own arbitrary graph  scheme.  The documentation
warned people against using this facility,  but there it was... enough rope
to hang yourself.

The consequences of "pinning" rows of a table in the DB address space were,
of course, devastating.  But the people who went down that road generally
had to learn it for themselves.

BTW,  the SW engineering that went into Rdb, and earlier into VAX DBMS (a
CODASYL style database)  was truly superb.  Some of the cleanest
implementation ever.  DEC Rdb still survives as "Oracle Rdb".