Synthetic Infosets

I believe it is a very poor idea to countenance infosets that do not maintain
consistency. Invariants are the key to interoperability, and defining
properties that have certain relationships, but permitting those relationships
to be violated is just causing problems. Either make all the properties
completely independent and self-contained, or assert the invariants, and declare
any infoset that fails to uphold them ill-formed.

The notion that a specification that permits the creation of infosets that
violate the invariants merely needs to say how the inconsistencies should be
resolved is not a help, in practice. All it does is make it so that one has to
know where an infoset *came* from in order to know how to process it properly,
because different specifications are free to come up with different rules.

I would strongly urge you to forbid such inconsistencies in synthetic infosets.

	-- Mary
	   Holstege@mathling.com

Received on Tuesday, 29 May 2001 12:46:35 UTC