You've put your finger on a weakness in the REC -- the prevasive
impact of the necessity of allowing for late binding was imperfectly
realised in the prose.  The bit you quote from 4.2.1 [1]:

  "As discussed in Missing Sub-components (=A75.3) [2], =B7QName=B7s in X=
ML
  representations may fail to =B7resolve=B7, rendering components
  incomplete and unusable because of missing subcomponents. During
  schema construction, implementations must retain =B7QName=B7 values for
  such references, in case an appropriately-named component becomes
  available to discharge the reference by the time it is actually
  needed."

is meant to have a universal impact.  So, somewhat informally, I'd say
that in your example the inclusion of B into A results in a schema,
some of whose references are unresolved.  Once C is integrated, we get
an enlarged schema, and resolution becomes possible.

The XML Schema WG is at work to make this cleaner, clearer and more
systematic in Schema 1.1

ht

[1] http://www.w3.org/TR/2004/PER-xmlschema-1-20040318/#compound-schema
[2] http://www.w3.org/TR/2004/PER-xmlschema-1-20040318/#conformance-missi=
ng
-- 
 Henry S. Thompson, HCRC Language Technology Group, University of Edinbur=
gh
                     Half-time member of W3C Team
    2 Buccleuch Place, Edinburgh EH8 9LW, SCOTLAND -- (44) 131 650-4440
            Fax: (44) 131 650-4587, e-mail: ht@i...
                   URL: http://www.ltg.ed.ac.uk/~ht/
[mail really from me _always_ has this .sig -- mail without it is forged =
spam]

From mhk@m... Mon Sep 27 09:04:13 2004
Received: from lisa.w3.org ([1