path: root/doc/rfc/rfc4911.txt

Network Working Group                                            S. Legg
Request for Comments: 4911                                       eB2Bcom
Category: Experimental                                         July 2007


                     Encoding Instructions for the
                    Robust XML Encoding Rules (RXER)

Status of This Memo

   This memo defines an Experimental Protocol for the Internet
   community.  It does not specify an Internet standard of any kind.
   Discussion and suggestions for improvement are requested.
   Distribution of this memo is unlimited.

Copyright Notice

   Copyright (C) The IETF Trust (2007).

Abstract

   This document defines encoding instructions that may be used in an
   Abstract Syntax Notation One (ASN.1) specification to alter how ASN.1
   values are encoded by the Robust XML Encoding Rules (RXER) and
   Canonical Robust XML Encoding Rules (CRXER), for example, to encode a
   component of an ASN.1 value as an Extensible Markup Language (XML)
   attribute rather than as a child element.  Some of these encoding
   instructions also affect how an ASN.1 specification is translated
   into an Abstract Syntax Notation X (ASN.X) specification.  Encoding
   instructions that allow an ASN.1 specification to reference
   definitions in other XML schema languages are also defined.




















Legg                          Experimental                      [Page 1]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


Table of Contents

   1. Introduction ....................................................3
   2. Conventions .....................................................3
   3. Definitions .....................................................4
   4. Notation for RXER Encoding Instructions .........................4
   5. Component Encoding Instructions .................................6
   6. Reference Encoding Instructions .................................8
   7. Expanded Names of Components ...................................10
   8. The ATTRIBUTE Encoding Instruction .............................11
   9. The ATTRIBUTE-REF Encoding Instruction .........................12
   10. The COMPONENT-REF Encoding Instruction ........................13
   11. The ELEMENT-REF Encoding Instruction ..........................16
   12. The LIST Encoding Instruction .................................17
   13. The NAME Encoding Instruction .................................19
   14. The REF-AS-ELEMENT Encoding Instruction .......................19
   15. The REF-AS-TYPE Encoding Instruction ..........................20
   16. The SCHEMA-IDENTITY Encoding Instruction ......................22
   17. The SIMPLE-CONTENT Encoding Instruction .......................22
   18. The TARGET-NAMESPACE Encoding Instruction .....................23
   19. The TYPE-AS-VERSION Encoding Instruction ......................24
   20. The TYPE-REF Encoding Instruction .............................25
   21. The UNION Encoding Instruction ................................26
   22. The VALUES Encoding Instruction ...............................27
   23. Insertion Encoding Instructions ...............................29
   24. The VERSION-INDICATOR Encoding Instruction ....................32
   25. The GROUP Encoding Instruction ................................34
      25.1. Unambiguous Encodings ....................................36
           25.1.1. Grammar Construction ..............................37
           25.1.2. Unique Component Attribution ......................47
           25.1.3. Deterministic Grammars ............................52
           25.1.4. Attributes in Unknown Extensions ..................54
   26. Security Considerations .......................................56
   27. References ....................................................56
      27.1. Normative References .....................................56
      27.2. Informative References ...................................57
   Appendix A. GROUP Encoding Instruction Examples ...................58
   Appendix B. Insertion Encoding Instruction Examples ...............74
   Appendix C. Extension and Versioning Examples .....................87












Legg                          Experimental                      [Page 2]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


1.  Introduction

   This document defines encoding instructions [X.680-1] that may be
   used in an Abstract Syntax Notation One (ASN.1) [X.680] specification
   to alter how ASN.1 values are encoded by the Robust XML Encoding
   Rules (RXER) [RXER] and Canonical Robust XML Encoding Rules (CRXER)
   [RXER], for example, to encode a component of an ASN.1 value as an
   Extensible Markup Language (XML) [XML10] attribute rather than as a
   child element.  Some of these encoding instructions also affect how
   an ASN.1 specification is translated into an Abstract Syntax Notation
   X (ASN.X) specification [ASN.X].

   This document also defines encoding instructions that allow an ASN.1
   specification to incorporate the definitions of types, elements, and
   attributes in specifications written in other XML schema languages.
   References to XML Schema [XSD1] types, elements, and attributes,
   RELAX NG [RNG] named patterns and elements, and XML document type
   definition (DTD) [XML10] element types are supported.

   In most cases, the effect of an encoding instruction is only briefly
   mentioned in this document.  The precise effects of these encoding
   instructions are described fully in the specifications for RXER
   [RXER] and ASN.X [ASN.X], at the points where they apply.

2.  Conventions

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "SHOULD", "SHOULD NOT", "RECOMMENDED" and "MAY" in this document are
   to be interpreted as described in BCP 14, RFC 2119 [BCP14].  The key
   word "OPTIONAL" is exclusively used with its ASN.1 meaning.

   Throughout this document "type" shall be taken to mean an ASN.1 type,
   and "value" shall be taken to mean an ASN.1 abstract value, unless
   qualified otherwise.

   A reference to an ASN.1 production [X.680] (e.g., Type, NamedType) is
   a reference to text in an ASN.1 specification corresponding to that
   production.  Throughout this document, "component" is synonymous with
   NamedType.

   This document uses the namespace prefix "xsi:" to stand for the
   namespace name [XMLNS10] "http://www.w3.org/2001/XMLSchema-instance".

   Example ASN.1 definitions in this document are assumed to be defined
   in an ASN.1 module with a TagDefault of "AUTOMATIC TAGS" and an
   EncodingReferenceDefault [X.680-1] of "RXER INSTRUCTIONS".





Legg                          Experimental                      [Page 3]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


3.  Definitions

   The following definition of base type is used in specifying a number
   of encoding instructions.

   Definition (base type): If a type, T, is a constrained type, then the
   base type of T is the base type of the type that is constrained; else
   if T is a prefixed type, then the base type of T is the base type of
   the type that is prefixed; else if T is a type notation that
   references or denotes another type (i.e., DefinedType,
   ObjectClassFieldType, SelectionType, TypeFromObject, or
   ValueSetFromObjects), then the base type of T is the base type of the
   type that is referenced or denoted; otherwise, the base type of T is
   T itself.

      Aside: A tagged type is a special case of a prefixed type.

4.  Notation for RXER Encoding Instructions

   The grammar of ASN.1 permits the application of encoding instructions
   [X.680-1], through type prefixes and encoding control sections, that
   modify how abstract values are encoded by nominated encoding rules.

   The generic notation for type prefixes and encoding control sections
   is defined by the ASN.1 basic notation [X.680] [X.680-1], and
   includes an encoding reference to identify the specific encoding
   rules that are affected by the encoding instruction.

   The encoding reference that identifies the Robust XML Encoding rules
   is literally RXER.  An RXER encoding instruction applies equally to
   both RXER and CRXER encodings.

   The specific notation for an encoding instruction for a specific set
   of encoding rules is left to the specification of those encoding
   rules.  Consequently, this companion document to the RXER
   specification [RXER] defines the notation for RXER encoding
   instructions.  Specifically, it elaborates the EncodingInstruction
   and EncodingInstructionAssignmentList placeholder productions of the
   ASN.1 basic notation.

   In the context of the RXER encoding reference, the
   EncodingInstruction production is defined as follows, using the
   conventions of the ASN.1 basic notation:








Legg                          Experimental                      [Page 4]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      EncodingInstruction ::=
          AttributeInstruction |
          AttributeRefInstruction |
          ComponentRefInstruction |
          ElementRefInstruction |
          GroupInstruction |
          InsertionsInstruction |
          ListInstruction |
          NameInstruction |
          RefAsElementInstruction |
          RefAsTypeInstruction |
          SimpleContentInstruction |
          TypeAsVersionInstruction |
          TypeRefInstruction |
          UnionInstruction |
          ValuesInstruction |
          VersionIndicatorInstruction

   In the context of the RXER encoding reference, the
   EncodingInstructionAssignmentList production (which only appears in
   an encoding control section) is defined as follows:

      EncodingInstructionAssignmentList ::=
          SchemaIdentityInstruction ?
          TargetNamespaceInstruction ?
          TopLevelComponents ?

      TopLevelComponents ::= TopLevelComponent TopLevelComponents ?

      TopLevelComponent ::= "COMPONENT" NamedType

   Definition (top-level NamedType): A NamedType is a top-level
   NamedType (equivalently, a top-level component) if and only if it is
   the NamedType in a TopLevelComponent.  A NamedType nested within the
   Type of the NamedType of a TopLevelComponent is not itself a
   top-level NamedType.

      Aside: Specification writers should note that non-trivial types
      defined within a top-level NamedType will not be visible to ASN.1
      tools that do not understand RXER.

   Although a top-level NamedType only appears in an RXER encoding
   control section, the default encoding reference for the module
   [X.680-1] still applies when parsing a top-level NamedType.

   Each top-level NamedType within a module SHALL have a distinct
   identifier.




Legg                          Experimental                      [Page 5]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The NamedType production is defined by the ASN.1 basic notation.  The
   other productions are described in subsequent sections and make use
   of the following productions:

      NCNameValue ::= Value

      AnyURIValue ::= Value

      QNameValue ::= Value

      NameValue ::= Value

   The Value production is defined by the ASN.1 basic notation.

   The governing type for the Value in an NCNameValue is the NCName type
   from the AdditionalBasicDefinitions module [RXER].

   The governing type for the Value in an AnyURIValue is the AnyURI type
   from the AdditionalBasicDefinitions module.

   The governing type for the Value in a QNameValue is the QName type
   from the AdditionalBasicDefinitions module.

   The governing type for the Value in a NameValue is the Name type from
   the AdditionalBasicDefinitions module.

   The Value in an NCNameValue, AnyURIValue, QNameValue, or NameValue
   SHALL NOT be a DummyReference [X.683] and SHALL NOT textually contain
   a nested DummyReference.

      Aside: Thus, encoding instructions are not permitted to be
      parameterized in any way.  This restriction will become important
      if a future specification for ASN.X explicitly represents
      parameterized definitions and parameterized references instead of
      expanding out parameterized references as in the current
      specification.  A parameterized definition could not be directly
      translated into ASN.X if it contained encoding instructions that
      were not fully specified.

5.  Component Encoding Instructions

   Certain of the RXER encoding instructions are categorized as
   component encoding instructions.  The component encoding instructions
   are the ATTRIBUTE, ATTRIBUTE-REF, COMPONENT-REF, GROUP, ELEMENT-REF,
   NAME, REF-AS-ELEMENT, SIMPLE-CONTENT, TYPE-AS-VERSION, and
   VERSION-INDICATOR encoding instructions (whose notations are
   described respectively by AttributeInstruction,
   AttributeRefInstruction, ComponentRefInstruction, GroupInstruction,



Legg                          Experimental                      [Page 6]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   ElementRefInstruction, NameInstruction, RefAsElementInstruction,
   SimpleContentInstruction, TypeAsVersionInstruction, and
   VersionIndicatorInstruction).

   The Type in the EncodingPrefixedType for a component encoding
   instruction SHALL be either:

   (1) the Type in a NamedType, or

   (2) the Type in an EncodingPrefixedType in a PrefixedType in a
       BuiltinType in a Type that is one of (1) to (4), or

   (3) the Type in an TaggedType in a PrefixedType in a BuiltinType in a
       Type that is one of (1) to (4), or

   (4) the Type in a ConstrainedType (excluding a TypeWithConstraint) in
       a Type that is one of (1) to (4).

      Aside: The effect of this condition is to force the component
      encoding instructions to be textually within the NamedType to
      which they apply.  Only case (2) can be true on the first
      iteration as the Type belongs to an EncodingPrefixedType; however,
      any of (1) to (4) can be true on subsequent iterations.

   Case (4) is not permitted when the encoding instruction is the
   ATTRIBUTE-REF, COMPONENT-REF, ELEMENT-REF, or REF-AS-ELEMENT encoding
   instruction.

   The NamedType in case (1) is said to be "subject to" the component
   encoding instruction.

   A top-level NamedType SHALL NOT be subject to an ATTRIBUTE-REF,
   COMPONENT-REF, GROUP, ELEMENT-REF, REF-AS-ELEMENT, or SIMPLE-CONTENT
   encoding instruction.

      Aside: This condition does not preclude these encoding
      instructions being used on a nested NamedType.

   A NamedType SHALL NOT be subject to two or more component encoding
   instructions of the same kind, e.g., a NamedType is not permitted to
   be subject to two NAME encoding instructions.

   The ATTRIBUTE, ATTRIBUTE-REF, COMPONENT-REF, GROUP, ELEMENT-REF,
   REF-AS-ELEMENT, SIMPLE-CONTENT, and TYPE-AS-VERSION encoding
   instructions are mutually exclusive.  The NAME, ATTRIBUTE-REF,
   COMPONENT-REF, ELEMENT-REF, and REF-AS-ELEMENT encoding instructions
   are mutually exclusive.  A NamedType SHALL NOT be subject to two or
   more encoding instructions that are mutually exclusive.



Legg                          Experimental                      [Page 7]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   A SelectionType [X.680] SHALL NOT be used to select the Type from a
   NamedType that is subject to an ATTRIBUTE-REF, COMPONENT-REF,
   ELEMENT-REF or REF-AS-ELEMENT encoding instruction.  The other
   component encoding instructions are not inherited by the type denoted
   by a SelectionType.

   Definition (attribute component):  An attribute component is a
   NamedType that is subject to an ATTRIBUTE or ATTRIBUTE-REF encoding
   instruction, or subject to a COMPONENT-REF encoding instruction that
   references a top-level NamedType that is subject to an ATTRIBUTE
   encoding instruction.

   Definition (element component):  An element component is a NamedType
   that is not subject to an ATTRIBUTE, ATTRIBUTE-REF, GROUP, or
   SIMPLE-CONTENT encoding instruction, and not subject to a
   COMPONENT-REF encoding instruction that references a top-level
   NamedType that is subject to an ATTRIBUTE encoding instruction.

      Aside: A NamedType subject to a GROUP or SIMPLE-CONTENT encoding
      instruction is neither an attribute component nor an element
      component.

6.  Reference Encoding Instructions

   Certain of the RXER encoding instructions are categorized as
   reference encoding instructions.  The reference encoding instructions
   are the ATTRIBUTE-REF, COMPONENT-REF, ELEMENT-REF, REF-AS-ELEMENT,
   REF-AS-TYPE, and TYPE-REF encoding instructions (whose notations are
   described respectively by AttributeRefInstruction,
   ComponentRefInstruction, ElementRefInstruction,
   RefAsElementInstruction, RefAsTypeInstruction, and
   TypeRefInstruction).  These encoding instructions (except
   COMPONENT-REF) allow an ASN.1 specification to incorporate the
   definitions of types, elements, and attributes in specifications
   written in other XML schema languages, through implied constraints on
   the markup that may appear in values of the Markup ASN.1 type from
   the AdditionalBasicDefinitions module [RXER] (for ELEMENT-REF,
   REF-AS-ELEMENT, REF-AS-TYPE, and TYPE-REF) or the UTF8String type
   (for ATTRIBUTE-REF).  References to XML Schema [XSD1] types,
   elements, and attributes, RELAX NG [RNG] named patterns and elements,
   and XML document type definition (DTD) [XML10] element types are
   supported.  References to ASN.1 types and top-level components are
   also permitted.  The COMPONENT-REF encoding instruction provides a
   more direct method of referencing a top-level component.

   The Type in the EncodingPrefixedType for an ELEMENT-REF,
   REF-AS-ELEMENT, REF-AS-TYPE, or TYPE-REF encoding instruction SHALL
   be either:



Legg                          Experimental                      [Page 8]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   (1) a ReferencedType that is a DefinedType that is a typereference
       (not a DummyReference) or ExternalTypeReference that references
       the Markup ASN.1 type from the AdditionalBasicDefinitions module
       [RXER], or

   (2) a BuiltinType that is a PrefixedType that is a TaggedType where
       the Type in the TaggedType is one of (1) to (3), or

   (3) a BuiltinType that is a PrefixedType that is an
       EncodingPrefixedType where the Type in the EncodingPrefixedType
       is one of (1) to (3) and the EncodingPrefix in the
       EncodingPrefixedType does not contain a reference encoding
       instruction.

      Aside: Case (3) and similar cases for the ATTRIBUTE-REF and
      COMPONENT-REF encoding instructions have the effect of making the
      reference encoding instructions mutually exclusive as well as
      singly occurring.

   With respect to the REF-AS-TYPE and TYPE-REF encoding instructions,
   the DefinedType in case (1) is said to be "subject to" the encoding
   instruction.

   The restrictions on the Type in the EncodingPrefixedType for an
   ATTRIBUTE-REF encoding instruction are specified in Section 9.  The
   restrictions on the Type in the EncodingPrefixedType for a
   COMPONENT-REF encoding instruction are specified in Section 10.

   The reference encoding instructions make use of a common production
   defined as follows:

      RefParameters ::= ContextParameter ?

      ContextParameter ::= "CONTEXT" AnyURIValue

   A RefParameters instance provides extra information about a reference
   to a definition.  A ContextParameter is used when a reference is
   ambiguous, i.e., refers to definitions in more than one schema
   document or external DTD subset.  This situation would occur, for
   example, when importing types with the same name from independently
   developed XML Schemas defined without a target namespace [XSD1].
   When used in conjunction with a reference to an element type in an
   external DTD subset, the AnyURIValue in the ContextParameter is the
   system identifier (a Uniform Resource Identifier or URI [URI]) of the
   external DTD subset; otherwise, the AnyURIValue is a URI that
   indicates the intended schema document, either an XML Schema
   specification, a RELAX NG specification, or an ASN.1 or ASN.X
   specification.



Legg                          Experimental                      [Page 9]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


7.  Expanded Names of Components

   Each NamedType has an associated expanded name [XMLNS10], determined
   as follows:

   (1) if the NamedType is subject to a NAME encoding instruction, then
       the local name of the expanded name is the character string
       specified by the NCNameValue of the NAME encoding instruction,

   (2) else if the NamedType is subject to a COMPONENT-REF encoding
       instruction, then the expanded name is the same as the expanded
       name of the referenced top-level NamedType,

   (3) else if the NamedType is subject to an ATTRIBUTE-REF or
       ELEMENT-REF encoding instruction, then the namespace name of the
       expanded name is equal to the namespace-name component of the
       QNameValue of the encoding instruction, and the local name is
       equal to the local-name component of the QNameValue,

   (4) else if the NamedType is subject to a REF-AS-ELEMENT encoding
       instruction, then the local name of the expanded name is the
       LocalPart [XMLNS10] of the qualified name specified by the
       NameValue of the encoding instruction,

   (5) otherwise, the local name of the expanded name is the identifier
       of the NamedType.

   In cases (1) and (5), if the NamedType is a top-level NamedType and
   the module containing the NamedType has a TARGET-NAMESPACE encoding
   instruction, then the namespace name of the expanded name is the
   character string specified by the AnyURIValue of the TARGET-NAMESPACE
   encoding instruction; otherwise, the namespace name has no value.

      Aside: Thus, the TARGET-NAMESPACE encoding instruction applies to
      a top-level NamedType but not to any other NamedType.

   In case (4), if the encoding instruction contains a Namespace, then
   the namespace name of the expanded name is the character string
   specified by the AnyURIValue of the Namespace; otherwise, the
   namespace name has no value.

   The expanded names for the attribute components of a CHOICE,
   SEQUENCE, or SET type MUST be distinct.  The expanded names for the
   components of a CHOICE, SEQUENCE, or SET type that are not attribute
   components MUST be distinct.  These tests are applied after the
   COMPONENTS OF transformation specified in X.680, Clause 24.4 [X.680].





Legg                          Experimental                     [Page 10]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      Aside: Two components of the same CHOICE, SEQUENCE, or SET type
      may have the same expanded name if one of them is an attribute
      component and the other is not.  Note that the "not" case includes
      components that are subject to a GROUP or SIMPLE-CONTENT encoding
      instruction.

   The expanded name of a top-level NamedType subject to an ATTRIBUTE
   encoding instruction MUST be distinct from the expanded name of every
   other top-level NamedType subject to an ATTRIBUTE encoding
   instruction in the same module.

   The expanded name of a top-level NamedType not subject to an
   ATTRIBUTE encoding instruction MUST be distinct from the expanded
   name of every other top-level NamedType not subject to an ATTRIBUTE
   encoding instruction in the same module.

      Aside: Two top-level components may have the same expanded name if
      one of them is an attribute component and the other is not.

8.  The ATTRIBUTE Encoding Instruction

   The ATTRIBUTE encoding instruction causes an RXER encoder to encode a
   value of the component to which it is applied as an XML attribute
   instead of as a child element.

   The notation for an ATTRIBUTE encoding instruction is defined as
   follows:

      AttributeInstruction ::= "ATTRIBUTE"

   The base type of the type of a NamedType that is subject to an
   ATTRIBUTE encoding instruction SHALL NOT be:

   (1) a CHOICE, SET, or SET OF type, or

   (2) a SEQUENCE type other than the one defining the QName type from
       the AdditionalBasicDefinitions module [RXER] (i.e., QName is
       allowed), or

   (3) a SEQUENCE OF type where the SequenceOfType is not subject to a
       LIST encoding instruction, or

   (4) an open type.








Legg                          Experimental                     [Page 11]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   Example

      PersonalDetails ::= SEQUENCE {
          firstName   [ATTRIBUTE] UTF8String,
          middleName  [ATTRIBUTE] UTF8String,
          surname     [ATTRIBUTE] UTF8String
      }

9.  The ATTRIBUTE-REF Encoding Instruction

   The ATTRIBUTE-REF encoding instruction causes an RXER encoder to
   encode a value of the component to which it is applied as an XML
   attribute instead of as a child element, where the attribute's name
   is a qualified name of the attribute declaration referenced by the
   encoding instruction.  In addition, the ATTRIBUTE-REF encoding
   instruction causes values of the UTF8String type to be restricted to
   conform to the type of the attribute declaration.

   The notation for an ATTRIBUTE-REF encoding instruction is defined as
   follows:

      AttributeRefInstruction ::=
          "ATTRIBUTE-REF" QNameValue RefParameters

   Taken together, the QNameValue and the ContextParameter in the
   RefParameters (if present) MUST reference an XML Schema attribute
   declaration or a top-level NamedType that is subject to an ATTRIBUTE
   encoding instruction.

   The type of a referenced XML Schema attribute declaration SHALL NOT
   be, either directly or by derivation, the XML Schema type QName,
   NOTATION, ENTITY, ENTITIES, or anySimpleType.

      Aside: Values of these types require information from the context
      of the attribute for interpretation.  Because an ATTRIBUTE-REF
      encoding instruction is restricted to prefixing the ASN.1
      UTF8String type, there is no mechanism to capture such context.

   The type of a referenced top-level NamedType SHALL NOT be, either
   directly or by subtyping, the QName type from the
   AdditionalBasicDefinitions module [RXER].

   The Type in the EncodingPrefixedType for an ATTRIBUTE-REF encoding
   instruction SHALL be either:

   (1) the UTF8String type, or





Legg                          Experimental                     [Page 12]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   (2) a BuiltinType that is a PrefixedType that is a TaggedType where
       the Type in the TaggedType is one of (1) to (3), or

   (3) a BuiltinType that is a PrefixedType that is an
       EncodingPrefixedType where the Type in the EncodingPrefixedType
       is one of (1) to (3) and the EncodingPrefix in the
       EncodingPrefixedType does not contain a reference encoding
       instruction.

   The identifier of a NamedType subject to an ATTRIBUTE-REF encoding
   instruction does not contribute to the name of attributes in an RXER
   encoding.  For the sake of consistency, the identifier SHOULD, where
   possible, be the same as the local name of the referenced attribute
   declaration.

10.  The COMPONENT-REF Encoding Instruction

   The ASN.1 basic notation does not have a concept of a top-level
   NamedType and therefore does not have a mechanism to reference a
   top-level NamedType.  The COMPONENT-REF encoding instruction provides
   a way to specify that a NamedType within a combining type definition
   is equivalent to a referenced top-level NamedType.

   The notation for a COMPONENT-REF encoding instruction is defined as
   follows:

      ComponentRefInstruction ::= "COMPONENT-REF" ComponentReference

      ComponentReference ::=
          InternalComponentReference |
          ExternalComponentReference

      InternalComponentReference ::= identifier FromModule ?

      FromModule ::= "FROM" GlobalModuleReference

      ExternalComponentReference ::= modulereference "." identifier

   The GlobalModuleReference production is defined by the ASN.1 basic
   notation [X.680].  If the GlobalModuleReference is absent from an
   InternalComponentReference, then the identifier MUST be the
   identifier of a top-level NamedType in the same module.  If the
   GlobalModuleReference is present in an InternalComponentReference,
   then the identifier MUST be the identifier of a top-level NamedType
   in the referenced module.






Legg                          Experimental                     [Page 13]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The modulereference in an ExternalComponentReference is used in the
   same way as a modulereference in an ExternalTypeReference.  The
   identifier in an ExternalComponentReference MUST be the identifier of
   a top-level NamedType in the referenced module.

   The Type in the EncodingPrefixedType for a COMPONENT-REF encoding
   instruction SHALL be either:

   (1) a ReferencedType that is a DefinedType that is a typereference
       (not a DummyReference) or an ExternalTypeReference, or

   (2) a BuiltinType or ReferencedType that is one of the productions in
       Table 1 in Section 5 of the specification for RXER [RXER], or

   (3) a BuiltinType that is a PrefixedType that is a TaggedType where
       the Type in the TaggedType is one of (1) to (4), or

   (4) a BuiltinType that is a PrefixedType that is an
       EncodingPrefixedType where the Type in the EncodingPrefixedType
       is one of (1) to (4) and the EncodingPrefix in the
       EncodingPrefixedType does not contain a reference encoding
       instruction.

   The restrictions on the use of RXER encoding instructions are such
   that no other RXER encoding instruction is permitted within a
   NamedType if the NamedType is subject to a COMPONENT-REF encoding
   instruction.

   The Type in the top-level NamedType referenced by the COMPONENT-REF
   encoding instruction MUST be either:

   (a) if the preceding case (1) is used, a ReferencedType that is a
       DefinedType that is a typereference or ExternalTypeReference that
       references the same type as the DefinedType in case (1), or

   (b) if the preceding case (2) is used, a BuiltinType or
       ReferencedType that is the same as the BuiltinType or
       ReferencedType in case (2), or

   (c) a BuiltinType that is a PrefixedType that is an
       EncodingPrefixedType where the Type in the EncodingPrefixedType
       is one of (a) to (c), and the EncodingPrefix in the
       EncodingPrefixedType contains an RXER encoding instruction.

   In principle, the COMPONENT-REF encoding instruction creates a
   notional NamedType where the expanded name is that of the referenced
   top-level NamedType and the Type in case (1) or (2) is substituted by
   the Type of the referenced top-level NamedType.



Legg                          Experimental                     [Page 14]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   In practice, it is sufficient for non-RXER encoders and decoders to
   use the original NamedType rather than the notional NamedType because
   the Type in case (1) or (2) can only differ from the Type of the
   referenced top-level NamedType by having fewer RXER encoding
   instructions, and RXER encoding instructions are ignored by non-RXER
   encoders and decoders.

   Although any prefixes for the Type in case (1) or (2) would be
   bypassed, it is sufficient for RXER encoders and decoders to use the
   referenced top-level NamedType instead of the notional NamedType
   because these prefixes cannot be RXER encoding instructions (except,
   of course, for the COMPONENT-REF encoding instruction) and can have
   no effect on an RXER encoding.

   Example

      Modules ::= SEQUENCE OF
          module [COMPONENT-REF module
                     FROM AbstractSyntaxNotation-X
                         { 1 3 6 1 4 1 21472 1 0 1 }]
                     ModuleDefinition

      Note that the "module" top-level NamedType in the
      AbstractSyntaxNotation-X module is defined like so:

         COMPONENT module ModuleDefinition

      The ASN.X translation of the SEQUENCE OF type definition provides
      a more natural representation:

         <namedType xmlns:asnx="urn:ietf:params:xml:ns:asnx"
                    name="Modules">
          <sequenceOf>
           <element ref="asnx:module"/>
          </sequenceOf>
         </namedType>

         Aside: The <namedType> element in ASN.X corresponds to a
         TypeAssignment, not a NamedType.

   The identifier of a NamedType subject to a COMPONENT-REF encoding
   instruction does not contribute to an RXER encoding.  For the sake of
   consistency with other encoding rules, the identifier SHOULD be the
   same as the identifier in the ComponentRefInstruction.







Legg                          Experimental                     [Page 15]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


11.  The ELEMENT-REF Encoding Instruction

   The ELEMENT-REF encoding instruction causes an RXER encoder to encode
   a value of the component to which it is applied as an element where
   the element's name is a qualified name of the element declaration
   referenced by the encoding instruction.  In addition, the ELEMENT-REF
   encoding instruction causes values of the Markup ASN.1 type to be
   restricted to conform to the type of the element declaration.

   The notation for an ELEMENT-REF encoding instruction is defined as
   follows:

      ElementRefInstruction ::= "ELEMENT-REF" QNameValue RefParameters

   Taken together, the QNameValue and the ContextParameter in the
   RefParameters (if present) MUST reference an XML Schema element
   declaration, a RELAX NG element definition, or a top-level NamedType
   that is not subject to an ATTRIBUTE encoding instruction.

   A referenced XML Schema element declaration MUST NOT have a type that
   requires the presence of values for the XML Schema ENTITY or ENTITIES
   types.

      Aside: Entity declarations are not supported by CRXER.

   Example

      AnySchema ::= CHOICE {
          module   [ELEMENT-REF {
                       namespace-name
                           "urn:ietf:params:xml:ns:asnx",
                       local-name "module" }]
                   Markup,
          schema   [ELEMENT-REF {
                       namespace-name
                           "http://www.w3.org/2001/XMLSchema",
                       local-name "schema" }]
                   Markup,
          grammar  [ELEMENT-REF {
                       namespace-name
                           "http://relaxng.org/ns/structure/1.0",
                       local-name "grammar" }]
                   Markup
      }

      The ASN.X translation of the choice type definition provides a
      more natural representation:




Legg                          Experimental                     [Page 16]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


         <namedType xmlns:asnx="urn:ietf:params:xml:ns:asnx"
                    xmlns:xs="http://www.w3.org/2001/XMLSchema"
                    xmlns:rng="http://relaxng.org/ns/structure/1.0"
                    name="AnySchema">
          <choice>
           <element ref="asnx:module" embedded="true"/>
           <element ref="xs:schema" embedded="true"/>
           <element ref="rng:grammar" embedded="true"/>
          </choice>
         </namedType>

   The identifier of a NamedType subject to an ELEMENT-REF encoding
   instruction does not contribute to the name of an element in an RXER
   encoding.  For the sake of consistency, the identifier SHOULD, where
   possible, be the same as the local name of the referenced element
   declaration.

12.  The LIST Encoding Instruction

   The LIST encoding instruction causes an RXER encoder to encode a
   value of a SEQUENCE OF type as a white-space-separated list of the
   component values.

   The notation for a LIST encoding instruction is defined as follows:

      ListInstruction ::= "LIST"

   The Type in an EncodingPrefixedType for a LIST encoding instruction
   SHALL be either:

   (1) a BuiltinType that is a SequenceOfType of the
       "SEQUENCE OF NamedType" form, or

   (2) a ConstrainedType that is a TypeWithConstraint of the
       "SEQUENCE Constraint OF NamedType" form or
       "SEQUENCE SizeConstraint OF NamedType" form, or

   (3) a ConstrainedType that is not a TypeWithConstraint where the Type
       in the ConstrainedType is one of (1) to (5), or

   (4) a BuiltinType that is a PrefixedType that is a TaggedType where
       the Type in the TaggedType is one of (1) to (5), or

   (5) a BuiltinType that is a PrefixedType that is an
       EncodingPrefixedType where the Type in the EncodingPrefixedType
       is one of (1) to (5).





Legg                          Experimental                     [Page 17]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The effect of this condition is to force the LIST encoding
   instruction to be textually co-located with the SequenceOfType or
   TypeWithConstraint to which it applies.

      Aside: This makes it clear to a reader that the encoding
      instruction applies to every use of the type no matter how it
      might be referenced.

   The SequenceOfType in case (1) and the TypeWithConstraint in case (2)
   are said to be "subject to" the LIST encoding instruction.

   A SequenceOfType or TypeWithConstraint SHALL NOT be subject to more
   than one LIST encoding instruction.

   The base type of the component type of a SequenceOfType or
   TypeWithConstraint that is subject to a LIST encoding instruction
   MUST be one of the following:

   (1) the BOOLEAN, INTEGER, ENUMERATED, REAL, OBJECT IDENTIFIER,
       RELATIVE-OID, GeneralizedTime, or UTCTime type, or

   (2) the NCName, AnyURI, Name, or QName type from the
       AdditionalBasicDefinitions module [RXER].

      Aside: While it would be feasible to allow the component type to
      also be any character string type that is constrained such that
      all its abstract values have a length greater than zero and none
      of its abstract values contain any white space characters, testing
      whether this condition is satisfied can be quite involved.  For
      the sake of simplicity, only certain immediately useful
      constrained UTF8String types, which are known to be suitable, are
      permitted (i.e., NCName, AnyURI, and Name).

   The NamedType in a SequenceOfType or TypeWithConstraint that is
   subject to a LIST encoding instruction MUST NOT be subject to an
   ATTRIBUTE, ATTRIBUTE-REF, COMPONENT-REF, GROUP, ELEMENT-REF,
   REF-AS-ELEMENT, SIMPLE-CONTENT, or TYPE-AS-VERSION encoding
   instruction.

   Example

      UpdateTimes ::= [LIST] SEQUENCE OF updateTime GeneralizedTime









Legg                          Experimental                     [Page 18]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


13.  The NAME Encoding Instruction

   The NAME encoding instruction causes an RXER encoder to use a
   nominated character string instead of a component's identifier
   wherever that identifier would otherwise appear in the encoding
   (e.g., as an element or attribute name).

   The notation for a NAME encoding instruction is defined as follows:

      NameInstruction ::= "NAME" "AS"? NCNameValue

   Example

      CHOICE {
          foo-att   [ATTRIBUTE] [NAME AS "Foo"] INTEGER,
          foo-elem  [NAME "Foo"] INTEGER
      }

14.  The REF-AS-ELEMENT Encoding Instruction

   The REF-AS-ELEMENT encoding instruction causes an RXER encoder to
   encode a value of the component to which it is applied as an element
   where the element's name is the name of the external DTD subset
   element type declaration referenced by the encoding instruction.  In
   addition, the REF-AS-ELEMENT encoding instruction causes values of
   the Markup ASN.1 type to be restricted to conform to the content and
   attributes permitted by that element type declaration and its
   associated attribute-list declarations.

   The notation for a REF-AS-ELEMENT encoding instruction is defined as
   follows:

      RefAsElementInstruction ::=
          "REF-AS-ELEMENT" NameValue Namespace ? RefParameters

      Namespace ::= "NAMESPACE" AnyURIValue

   Taken together, the NameValue and the ContextParameter in the
   RefParameters (if present) MUST reference an element type declaration
   in an external DTD subset that is conformant with Namespaces in XML
   1.0 [XMLNS10].

   The Namespace is present if and only if the Name of the referenced
   element type declaration conforms to a PrefixedName (a QName)
   [XMLNS10], in which case the Namespace specifies the namespace name
   to be associated with the Prefix of the PrefixedName.





Legg                          Experimental                     [Page 19]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The referenced element type declaration MUST NOT require the presence
   of attributes of type ENTITY or ENTITIES.

      Aside: Entity declarations are not supported by CRXER.

   Example

      Suppose that the following external DTD subset has been defined
      with a system identifier of "http://www.example.com/inventory":

         <?xml version='1.0'?>
         <!ELEMENT product EMPTY>
         <!ATTLIST product
             name       CDATA #IMPLIED
             partNumber CDATA #REQUIRED
             quantity   CDATA #REQUIRED >

      The product element type declaration can be referenced as an
      element in an ASN.1 type definition:

         CHOICE {
             product  [REF-AS-ELEMENT "product"
                          CONTEXT "http://www.example.com/inventory"]
                      Markup
         }

      Here is the ASN.X translation of this ASN.1 type definition:

         <type>
          <choice>
           <element elementType="product"
                    context="http://www.example.com/inventory"/>
          </choice>
         </type>

   The identifier of a NamedType subject to a REF-AS-ELEMENT encoding
   instruction does not contribute to the name of an element in an RXER
   encoding.  For the sake of consistency, the identifier SHOULD, where
   possible, be the same as the Name of the referenced element type
   declaration (or the LocalPart if the Name conforms to a
   PrefixedName).

15.  The REF-AS-TYPE Encoding Instruction

   The REF-AS-TYPE encoding instruction causes values of the Markup
   ASN.1 type to be restricted to conform to the content and attributes
   permitted by a nominated element type declaration and its associated
   attribute-list declarations in an external DTD subset.



Legg                          Experimental                     [Page 20]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The notation for a REF-AS-TYPE encoding instruction is defined as
   follows:

      RefAsTypeInstruction ::= "REF-AS-TYPE" NameValue RefParameters

   Taken together, the NameValue and the ContextParameter of the
   RefParameters (if present) MUST reference an element type declaration
   in an external DTD subset that is conformant with Namespaces in XML
   1.0 [XMLNS10].

   The referenced element type declaration MUST NOT require the presence
   of attributes of type ENTITY or ENTITIES.

      Aside: Entity declarations are not supported by CRXER.

   Example

      The product element type declaration can be referenced as a type
      in an ASN.1 definition:

         SEQUENCE OF
             inventoryItem
                 [REF-AS-TYPE "product"
                     CONTEXT "http://www.example.com/inventory"]
                 Markup

      Here is the ASN.X translation of this definition:

         <sequenceOf>
          <element name="inventoryItem">
           <type elementType="product"
                 context="http://www.example.com/inventory"/>
          </element>
         </sequenceOf>

      Note that when an element type declaration is referenced as a
      type, the Name of the element type declaration does not contribute
      to RXER encodings.  For example, child elements in the RXER
      encoding of values of the above SEQUENCE OF type would resemble
      the following:

         <inventoryItem name="hammer" partNumber="1543" quantity="29"/>









Legg                          Experimental                     [Page 21]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


16.  The SCHEMA-IDENTITY Encoding Instruction

   The SCHEMA-IDENTITY encoding instruction associates a unique
   identifier, a URI [URI], with the ASN.1 module containing the
   encoding instruction.  This encoding instruction has no effect on an
   RXER encoder but does have an effect on the translation of an ASN.1
   specification into an ASN.X representation.

   The notation for a SCHEMA-IDENTITY encoding instruction is defined as
   follows:

      SchemaIdentityInstruction ::= "SCHEMA-IDENTITY" AnyURIValue

   The character string specified by the AnyURIValue of each
   SCHEMA-IDENTITY encoding instruction MUST be distinct.  In
   particular, successive versions of an ASN.1 module must each have a
   different schema identity URI value.

17.  The SIMPLE-CONTENT Encoding Instruction

   The SIMPLE-CONTENT encoding instruction causes an RXER encoder to
   encode a value of a component of a SEQUENCE or SET type without
   encapsulation in a child element.

   The notation for a SIMPLE-CONTENT encoding instruction is defined as
   follows:

      SimpleContentInstruction ::= "SIMPLE-CONTENT"

   A NamedType subject to a SIMPLE-CONTENT encoding instruction SHALL be
   in a ComponentType in a ComponentTypeList in a RootComponentTypeList.
   At most one such NamedType of a SEQUENCE or SET type is permitted to
   be subject to a SIMPLE-CONTENT encoding instruction.  If any
   component is subject to a SIMPLE-CONTENT encoding instruction, then
   all other components in the same SEQUENCE or SET type definition MUST
   be attribute components.  These tests are applied after the
   COMPONENTS OF transformation specified in X.680, Clause 24.4 [X.680].

      Aside: Child elements and simple content are mutually exclusive.
      Specification writers should note that use of the SIMPLE-CONTENT
      encoding instruction on a component of an extensible SEQUENCE or
      SET type means that all future extensions to the SEQUENCE or SET
      type are restricted to being attribute components with the limited
      set of types that are permitted for attribute components.  Using
      an ATTRIBUTE encoding instruction instead of a SIMPLE-CONTENT
      encoding instruction avoids this limitation.





Legg                          Experimental                     [Page 22]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The base type of the type of a NamedType that is subject to a
   SIMPLE-CONTENT encoding instruction SHALL NOT be:

   (1) a SET or SET OF type, or

   (2) a CHOICE type where the ChoiceType is not subject to a UNION
       encoding instruction, or

   (3) a SEQUENCE type other than the one defining the QName type from
       the AdditionalBasicDefinitions module [RXER] (i.e., QName is
       allowed), or

   (4) a SEQUENCE OF type where the SequenceOfType is not subject to a
       LIST encoding instruction, or

   (5) an open type.

   If the type of a NamedType subject to a SIMPLE-CONTENT encoding
   instruction has abstract values with an empty character data
   translation [RXER] (i.e., an empty encoding), then the NamedType
   SHALL NOT be marked OPTIONAL or DEFAULT.

   Example

      SEQUENCE {
          units   [ATTRIBUTE] UTF8String,
          amount  [SIMPLE-CONTENT] INTEGER
      }

18.  The TARGET-NAMESPACE Encoding Instruction

   The TARGET-NAMESPACE encoding instruction associates an XML namespace
   name [XMLNS10], a URI [URI], with the type, object class, value,
   object, and object set references defined in the ASN.1 module
   containing the encoding instruction.  In addition, it associates the
   namespace name with each top-level NamedType in the RXER encoding
   control section.

   The notation for a TARGET-NAMESPACE encoding instruction is defined
   as follows:

      TargetNamespaceInstruction ::=
          "TARGET-NAMESPACE" AnyURIValue Prefix ?

      Prefix ::= "PREFIX" NCNameValue

   The AnyURIValue SHALL NOT specify an empty string.




Legg                          Experimental                     [Page 23]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   Definition (target namespace):  If an ASN.1 module contains a
   TARGET-NAMESPACE encoding instruction, then the target namespace of
   the module is the character string specified by the AnyURIValue of
   the TARGET-NAMESPACE encoding instruction; otherwise, the target
   namespace of the module is said to be absent.

   Two or more ASN.1 modules MAY have the same non-absent target
   namespace if and only if the expanded names of the top-level
   attribute components are distinct across all those modules, the
   expanded names of the top-level element components are distinct
   across all those modules, and the defined type, object class, value,
   object, and object set references are distinct in their category
   across all those modules.

   The Prefix, if present, suggests an NCName to use as the namespace
   prefix in namespace declarations involving the target namespace.  An
   RXER encoder is not obligated to use the nominated namespace prefix.

   If there are no top-level components, then the RXER encodings
   produced using a module with a TARGET-NAMESPACE encoding instruction
   are backward compatible with the RXER encodings produced by the same
   module without the TARGET-NAMESPACE encoding instruction.

19.  The TYPE-AS-VERSION Encoding Instruction

   The TYPE-AS-VERSION encoding instruction causes an RXER encoder to
   include an xsi:type attribute in the encoding of a value of the
   component to which the encoding instruction is applied.  This
   attribute allows an XML Schema [XSD1] validator to select, if
   available, the appropriate XML Schema translation for the version of
   the ASN.1 specification used to create the encoding.

      Aside: Translations of an ASN.1 specification into a compatible
      XML Schema are expected to be slightly different across versions
      because of progressive extensions to the ASN.1 specification.  Any
      incompatibilities between these translations can be accommodated
      if each version uses a different target namespace.  The target
      namespace will be evident in the value of the xsi:type attribute
      and will cause an XML Schema validator to use the appropriate
      version.  This mechanism also accommodates an ASN.1 type that is
      renamed in a later version of the ASN.1 specification.

   The notation for a TYPE-AS-VERSION encoding instruction is defined as
   follows:

      TypeAsVersionInstruction ::= "TYPE-AS-VERSION"





Legg                          Experimental                     [Page 24]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The Type in a NamedType that is subject to a TYPE-AS-VERSION encoding
   instruction MUST be a namespace-qualified reference [RXER].

   The addition of a TYPE-AS-VERSION encoding instruction does not
   affect the backward compatibility of RXER encodings.

      Aside: In a translation of an ASN.1 specification into XML Schema,
      any Type in a NamedType that is subject to a TYPE-AS-VERSION
      encoding instruction is expected to be translated into the
      XML Schema anyType so that the xsi:type attribute acts as a switch
      to select the appropriate version.

20.  The TYPE-REF Encoding Instruction

   The TYPE-REF encoding instruction causes values of the Markup ASN.1
   type to be restricted to conform to a specific XML Schema named type,
   RELAX NG named pattern or an ASN.1 defined type.

      Aside: Referencing an ASN.1 type in a TYPE-REF encoding
      instruction does not have the effect of imposing a requirement to
      preserve the Infoset [INFOSET] representation of the RXER encoding
      of an abstract value of the type.  It is still sufficient to
      preserve just the abstract value.

   The notation for a TYPE-REF encoding instruction is defined as
   follows:

      TypeRefInstruction ::= "TYPE-REF" QNameValue RefParameters

   Taken together, the QNameValue and the ContextParameter of the
   RefParameters (if present) MUST reference an XML Schema named type, a
   RELAX NG named pattern, or an ASN.1 defined type.

   A referenced XML Schema type MUST NOT require the presence of values
   for the XML Schema ENTITY or ENTITIES types.

      Aside: Entity declarations are not supported by CRXER.

   The QNameValue SHALL NOT be a direct reference to the XML Schema
   NOTATION type [XSD2] (i.e., the namespace name
   "http://www.w3.org/2001/XMLSchema" and local name "NOTATION");
   however, a reference to an XML Schema type derived from the NOTATION
   type is permitted.

      Aside: This restriction is to ensure that the lexical space [XSD2]
      of the referenced type is actually populated with the names of
      notations [XSD1].




Legg                          Experimental                     [Page 25]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   Example

      MyDecimal ::=
          [TYPE-REF {
              namespace-name "http://www.w3.org/2001/XMLSchema",
              local-name     "decimal" }]
          Markup

      Note that the ASN.X translation of this ASN.1 type definition
      provides a more natural way to reference the XML Schema decimal
      type:

         <namedType xmlns:xs="http://www.w3.org/2001/XMLSchema"
                    name="MyDecimal">
          <type ref="xs:decimal" embedded="true"/>
         </namedType>

21.  The UNION Encoding Instruction

   The UNION encoding instruction causes an RXER encoder to encode the
   value of an alternative of a CHOICE type without encapsulation in a
   child element.  The chosen alternative is optionally indicated with a
   member attribute.  The optional PrecedenceList also allows a
   specification writer to alter the order in which an RXER decoder will
   consider the alternatives of the CHOICE as it determines which
   alternative has been used (if the actual alternative has not been
   specified through the member attribute).

   The notation for a UNION encoding instruction is defined as follows:

      UnionInstruction ::= "UNION" AlternativesPrecedence ?

      AlternativesPrecedence ::= "PRECEDENCE" PrecedenceList

      PrecedenceList ::= identifier PrecedenceList ?

   The Type in the EncodingPrefixedType for a UNION encoding instruction
   SHALL be either:

   (1) a BuiltinType that is a ChoiceType, or

   (2) a ConstrainedType that is not a TypeWithConstraint where the Type
       in the ConstrainedType is one of (1) to (4), or

   (3) a BuiltinType that is a PrefixedType that is a TaggedType where
       the Type in the TaggedType is one of (1) to (4), or





Legg                          Experimental                     [Page 26]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   (4) a BuiltinType that is a PrefixedType that is an
       EncodingPrefixedType where the Type in the EncodingPrefixedType
       is one of (1) to (4).

   The ChoiceType in case (1) is said to be "subject to" the UNION
   encoding instruction.

   The base type of the type of each alternative of a ChoiceType that is
   subject to a UNION encoding instruction SHALL NOT be:

   (1) a CHOICE, SET, or SET OF type, or

   (2) a SEQUENCE type other than the one defining the QName type from
       the AdditionalBasicDefinitions module [RXER] (i.e., QName is
       allowed), or

   (3) a SEQUENCE OF type where the SequenceOfType is not subject to a
       LIST encoding instruction, or

   (4) an open type.

   Each identifier in the PrecedenceList MUST be the identifier of a
   NamedType in the ChoiceType.

   A particular identifier SHALL NOT appear more than once in the same
   PrecedenceList.

   Every NamedType in a ChoiceType that is subject to a UNION encoding
   instruction MUST NOT be subject to an ATTRIBUTE, ATTRIBUTE-REF,
   COMPONENT-REF, GROUP, ELEMENT-REF, REF-AS-ELEMENT, SIMPLE-CONTENT, or
   TYPE-AS-VERSION encoding instruction.

   Example

      [UNION PRECEDENCE basicName] CHOICE {
          extendedName  UTF8String,
          basicName     PrintableString
      }

22.  The VALUES Encoding Instruction

   The VALUES encoding instruction causes an RXER encoder to use
   nominated names instead of the identifiers that would otherwise
   appear in the encoding of a value of a BIT STRING, ENUMERATED, or
   INTEGER type.






Legg                          Experimental                     [Page 27]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The notation for a VALUES encoding instruction is defined as follows:

      ValuesInstruction ::=
          "VALUES" AllValuesMapped ? ValueMappingList ?

      AllValuesMapped ::= AllCapitalized | AllUppercased

      AllCapitalized ::= "ALL" "CAPITALIZED"

      AllUppercased ::= "ALL" "UPPERCASED"

      ValueMappingList ::= ValueMapping ValueMappingList ?

      ValueMapping ::= "," identifier "AS" NCNameValue

   The Type in the EncodingPrefixedType for a VALUES encoding
   instruction SHALL be either:

   (1) a BuiltinType that is a BitStringType with a NamedBitList, or

   (2) a BuiltinType that is an EnumeratedType, or

   (3) a BuiltinType that is an IntegerType with a NamedNumberList, or

   (4) a ConstrainedType that is not a TypeWithConstraint where the Type
       in the ConstrainedType is one of (1) to (6), or

   (5) a BuiltinType that is a PrefixedType that is a TaggedType where
       the Type in the TaggedType is one of (1) to (6), or

   (6) a BuiltinType that is a PrefixedType that is an
       EncodingPrefixedType where the Type in the EncodingPrefixedType
       is one of (1) to (6).

   The effect of this condition is to force the VALUES encoding
   instruction to be textually co-located with the type definition to
   which it applies.

   The BitStringType, EnumeratedType, or IntegerType in case (1), (2),
   or (3), respectively, is said to be "subject to" the VALUES encoding
   instruction.

   A BitStringType, EnumeratedType, or IntegerType SHALL NOT be subject
   to more than one VALUES encoding instruction.

   Each identifier in a ValueMapping MUST be an identifier appearing in
   the NamedBitList, Enumerations, or NamedNumberList, as the case may
   be.



Legg                          Experimental                     [Page 28]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The identifier in a ValueMapping SHALL NOT be the same as the
   identifier in any other ValueMapping for the same ValueMappingList.

   Definition (replacement name):  Each identifier in a BitStringType,
   EnumeratedType, or IntegerType subject to a VALUES encoding
   instruction has a replacement name.  If there is a ValueMapping for
   the identifier, then the replacement name is the character string
   specified by the NCNameValue in the ValueMapping; else if
   AllCapitalized is used, then the replacement name is the identifier
   with the first character uppercased; else if AllUppercased is used,
   then the replacement name is the identifier with all its characters
   uppercased; otherwise, the replacement name is the identifier.

   The replacement names for the identifiers in a BitStringType subject
   to a VALUES encoding instruction MUST be distinct.

   The replacement names for the identifiers in an EnumeratedType
   subject to a VALUES encoding instruction MUST be distinct.

   The replacement names for the identifiers in an IntegerType subject
   to a VALUES encoding instruction MUST be distinct.

   Example

      Traffic-Light ::= [VALUES ALL CAPITALIZED, red AS "RED"]
          ENUMERATED {
              red,    -- Replacement name is RED.
              amber,  -- Replacement name is Amber.
              green   -- Replacement name is Green.
          }

23.  Insertion Encoding Instructions

   Certain of the RXER encoding instructions are categorized as
   insertion encoding instructions.  The insertion encoding instructions
   are the NO-INSERTIONS, HOLLOW-INSERTIONS, SINGULAR-INSERTIONS,
   UNIFORM-INSERTIONS, and MULTIFORM-INSERTIONS encoding instructions
   (whose notations are described respectively by
   NoInsertionsInstruction, HollowInsertionsInstruction,
   SingularInsertionsInstruction, UniformInsertionsInstruction, and
   MultiformInsertionsInstruction).

   The notation for the insertion encoding instructions is defined as
   follows:







Legg                          Experimental                     [Page 29]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      InsertionsInstruction ::=
          NoInsertionsInstruction |
          HollowInsertionsInstruction |
          SingularInsertionsInstruction |
          UniformInsertionsInstruction |
          MultiformInsertionsInstruction

      NoInsertionsInstruction ::= "NO-INSERTIONS"

      HollowInsertionsInstruction ::= "HOLLOW-INSERTIONS"

      SingularInsertionsInstruction ::= "SINGULAR-INSERTIONS"

      UniformInsertionsInstruction ::= "UNIFORM-INSERTIONS"

      MultiformInsertionsInstruction ::= "MULTIFORM-INSERTIONS"

   Using the GROUP encoding instruction on components with extensible
   types can lead to situations where an unknown extension could be
   associated with more than one extension insertion point.  The
   insertion encoding instructions remove this ambiguity by limiting the
   form that extensions can take.  That is, the insertion encoding
   instructions indicate what extensions can be made to an ASN.1
   specification without breaking forward compatibility for RXER
   encodings.

      Aside: Forward compatibility means the ability for a decoder to
      successfully decode an encoding containing extensions introduced
      into a version of the specification that is more recent than the
      one used by the decoder.

   In the most general case, an extension to a CHOICE, SET, or SEQUENCE
   type will generate zero or more attributes and zero or more elements,
   due to the potential use of the GROUP and ATTRIBUTE encoding
   instructions by the extension.

   The MULTIFORM-INSERTIONS encoding instruction indicates that the RXER
   encodings produced by forward-compatible extensions to a type will
   always consist of one or more elements and zero or more attributes.
   No restriction is placed on the names of the elements.

      Aside: Of necessity, the names of the attributes will all be
      different in any given encoding.

   The UNIFORM-INSERTIONS encoding instruction indicates that the RXER
   encodings produced by forward-compatible extensions to a type will
   always consist of one or more elements having the same expanded name,
   and zero or more attributes.  The expanded name shared by the



Legg                          Experimental                     [Page 30]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   elements in one particular encoding is not required to be the same as
   the expanded name shared by the elements in any other encoding of the
   extension.  For example, in one encoding of the extension the
   elements might all be called "foo", while in another encoding of the
   extension they might all be called "bar".

   The SINGULAR-INSERTIONS encoding instruction indicates that the RXER
   encodings produced by forward-compatible extensions to a type will
   always consist of a single element and zero or more attributes.  The
   name of the single element is not required to be the same in every
   possible encoding of the extension.

   The HOLLOW-INSERTIONS encoding instruction indicates that the RXER
   encodings produced by forward-compatible extensions to a type will
   always consist of zero elements and zero or more attributes.

   The NO-INSERTIONS encoding instruction indicates that no forward-
   compatible extensions can be made to a type.

   Examples of forward-compatible extensions are provided in Appendix C.

   The Type in the EncodingPrefixedType for an insertion encoding
   instruction SHALL be either:

   (1) a BuiltinType that is a ChoiceType where the ChoiceType is not
       subject to a UNION encoding instruction, or

   (2) a BuiltinType that is a SequenceType or SetType, or

   (3) a ConstrainedType that is not a TypeWithConstraint where the Type
       in the ConstrainedType is one of (1) to (5), or

   (4) a BuiltinType that is a PrefixedType that is a TaggedType where
       the Type in the TaggedType is one of (1) to (5), or

   (5) a BuiltinType that is a PrefixedType that is an
       EncodingPrefixedType where the Type in the EncodingPrefixedType
       is one of (1) to (5).

   Case (2) is not permitted when the insertion encoding instruction is
   the SINGULAR-INSERTIONS, UNIFORM-INSERTIONS, or MULTIFORM-INSERTIONS
   encoding instruction.









Legg                          Experimental                     [Page 31]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      Aside: Because extensions to a SET or SEQUENCE type are serial and
      effectively optional, the SINGULAR-INSERTIONS, UNIFORM-INSERTIONS,
      and MULTIFORM-INSERTIONS encoding instructions offer no advantage
      over unrestricted extensions (for a SET or SEQUENCE).  For
      example, an optional series of singular insertions generates zero
      or more elements and zero or more attributes, just like an
      unrestricted extension.

   The Type in case (1) or case (2) is said to be "subject to" the
   insertion encoding instruction.

   The Type in case (1) or case (2) MUST be extensible, either
   explicitly or by default.

   A Type SHALL NOT be subject to more than one insertion encoding
   instruction.

   The insertion encoding instructions indicate what kinds of extensions
   can be made to a type without breaking forward compatibility, but
   they do not prohibit extensions that do break forward compatibility.
   That is, it is not an error for a type's base type to contain
   extensions that do not satisfy an insertion encoding instruction
   affecting the type.  However, if any such extensions are made, then a
   new value SHOULD be introduced into the extensible set of permitted
   values for a version indicator attribute, or attributes (see
   Section 24), whose scope encompasses the extensions.  An example is
   provided in Appendix C.

24.  The VERSION-INDICATOR Encoding Instruction

   The VERSION-INDICATOR encoding instruction provides a mechanism for
   RXER decoders to be alerted that an encoding contains extensions that
   break forward compatibility (see the preceding section).

   The notation for a VERSION-INDICATOR encoding instruction is defined
   as follows:

      VersionIndicatorInstruction ::= "VERSION-INDICATOR"

   A NamedType that is subject to a VERSION-INDICATOR encoding
   instruction MUST also be subject to an ATTRIBUTE encoding
   instruction.

   The type of the NamedType that is subject to the VERSION-INDICATOR
   encoding instruction MUST be directly or indirectly a constrained
   type where the set of permitted values is defined to be extensible.
   Each value represents a different version of the ASN.1 specification.
   Ordinarily, an application will set the value of a version indicator



Legg                          Experimental                     [Page 32]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   attribute to be the last of these permitted values.  An application
   MAY set the value of the version indicator attribute to the value
   corresponding to an earlier version of the specification if it has
   not used any of the extensions added in a subsequent version.

   If an RXER decoder encounters a value of the type that is not one of
   the root values or extension additions (but that is still allowed
   since the set of permitted values is extensible), then this indicates
   that the decoder is using a version of the ASN.1 specification that
   is not compatible with the version used to produce the encoding.  In
   such cases, the decoder SHOULD treat the element containing the
   attribute as having an unknown ASN.1 type.

      Aside: A version indicator attribute only indicates an
      incompatibility with respect to RXER encodings.  Other encodings
      are not affected because the GROUP encoding instruction does not
      apply to them.

   Examples

      In this first example, the decoder is using an incompatible older
      version if the value of the version attribute in a received RXER
      encoding is not 1, 2, or 3.

         SEQUENCE {
             version  [ATTRIBUTE] [VERSION-INDICATOR]
                          INTEGER (1, ..., 2..3),
             message  MessageType
         }

      In this second example, the decoder is using an incompatible older
      version if the value of the format attribute in a received RXER
      encoding is not "1.0", "1.1", or "2.0".

         SEQUENCE {
             format   [ATTRIBUTE] [VERSION-INDICATOR]
                          UTF8String ("1.0", ..., "1.1" | "2.0"),
             message  MessageType
         }

      An extensive example is provided in Appendix C.

   It is not necessary for every extensible type to have its own version
   indicator attribute.  It would be typical for only the types of
   top-level element components to include a version indicator
   attribute, which would serve as the version indicator for all of the
   nested components.




Legg                          Experimental                     [Page 33]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


25.  The GROUP Encoding Instruction

   The GROUP encoding instruction causes an RXER encoder to encode a
   value of the component to which it is applied without encapsulation
   as an element.  It allows the construction of non-trivial content
   models for element content.

   The notation for a GROUP encoding instruction is defined as follows:

      GroupInstruction ::= "GROUP"

   The base type of the type of a NamedType that is subject to a GROUP
   encoding instruction SHALL be either:

   (1) a SEQUENCE, SET, or SET OF type, or

   (2) a CHOICE type where the ChoiceType is not subject to a UNION
       encoding instruction, or

   (3) a SEQUENCE OF type where the SequenceOfType is not subject to a
       LIST encoding instruction.

   The SEQUENCE type in case (1) SHALL NOT be the associated type for a
   built-in type, SHALL NOT be a type from the
   AdditionalBasicDefinitions module [RXER], and SHALL NOT contain a
   component that is subject to a SIMPLE-CONTENT encoding instruction.

      Aside: Thus, the CHARACTER STRING, EMBEDDED PDV, EXTERNAL, REAL,
      and QName types are excluded.

   The CHOICE type in case (2) SHALL NOT be a type from the
   AdditionalBasicDefinitions module.

      Aside: Thus, the Markup type is excluded.

   Definition (visible component): Ignoring all type constraints, the
   visible components for a type that is directly or indirectly a
   combining ASN.1 type (i.e., SEQUENCE, SET, CHOICE, SEQUENCE OF, or
   SET OF) is the set of components of the combining type definition
   plus, for each NamedType (of the combining type definition) that is
   subject to a GROUP encoding instruction, the visible components for
   the type of the NamedType.  The visible components are determined
   after the COMPONENTS OF transformation specified in X.680, Clause
   24.4 [X.680].







Legg                          Experimental                     [Page 34]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      Aside: The set of visible attribute and element components for a
      type is the set of all the components of the type, and any nested
      types, that describe attributes and child elements appearing in
      the RXER encodings of values of the outer type.

   A GROUP encoding instruction MUST NOT be used where it would cause a
   NamedType to be a visible component of the type of that same
   NamedType (which is only possible if the type definition is
   recursive).

      Aside: Components subject to a GROUP encoding instruction might be
      translated into a compatible XML Schema [XSD1] as group
      definitions.  A NamedType that is visible to its own type is
      analogous to a circular group, which XML Schema disallows.

   Section 25.1 imposes additional conditions on the use of the GROUP
   encoding instruction.

   In any use of the GROUP encoding instruction, there is a type, the
   including type, that contains the component subject to the GROUP
   encoding instruction, and a type, the included type, that is the base
   type of that component.  Either type can have an extensible content
   model, either by directly using ASN.1 extensibility or by including
   through another GROUP encoding instruction some other type that is
   extensible.

   The including and included types may be defined in different ASN.1
   modules, in which case the owner of the including type, i.e., the
   person or organization having the authority to add extensions to the
   including type's definition, may be different from the owner of the
   included type.

   If the owner of the including type is not using the most recent
   version of the included type's definition, then the owner of the
   including type might add an extension to the including type that is
   valid with respect to the older version of the included type, but is
   later found to be invalid when the latest versions of the including
   and included type definitions are brought together (perhaps by a
   third party).  Although the owner of the including type must
   necessarily be aware of the existence of the included type, the
   reverse is not necessarily true.  The owner of the included type
   could add an extension to the included type without realizing that it
   invalidates someone else's including type.








Legg                          Experimental                     [Page 35]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   To avoid these problems, a GROUP encoding instruction MUST NOT be
   used if:

   (1) the included type is defined in a different module from the
       including type, and

   (2) the included type has an extensible content model, and

   (3) changes to the included type are not coordinated with the owner
       of the including type.

   Changes in the included type are coordinated with the owner of the
   including type if:

   (1) the owner of the included type is also the owner of the including
       type, or

   (2) the owner of the including type is collaborating with the owner
       of the included type, or

   (3) all changes will be vetted by a common third party before being
       approved and published.

25.1.  Unambiguous Encodings

   Unregulated use of the GROUP encoding instruction can easily lead to
   specifications in which distinct abstract values have
   indistinguishable RXER encodings, i.e., ambiguous encodings.  This
   section imposes restrictions on the use of the GROUP encoding
   instruction to ensure that distinct abstract values have distinct
   RXER encodings.  In addition, these restrictions ensure that an
   abstract value can be easily decoded in a single pass without
   back-tracking.

   An RXER decoder for an ASN.1 type can be abstracted as a recognizer
   for a notional language, consisting of element and attribute expanded
   names, where the type definition describes the grammar for that
   language (in fact it is a context-free grammar).  The restrictions on
   a type definition to ensure easy, unambiguous decoding are more
   conveniently, completely, and simply expressed as conditions on this
   associated grammar.  Implementations are not expected to verify type
   definitions exactly in the manner to be described; however, the
   procedure used MUST produce the same result.

   Section 25.1.1 describes the procedure for recasting as a grammar a
   type definition containing components subject to the GROUP encoding
   instruction.  Sections 25.1.2 and 25.1.3 specify conditions that the




Legg                          Experimental                     [Page 36]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   grammar must satisfy for the type definition to be valid.  Section
   25.1.4 describes how unrecognized attributes are accepted by the
   grammar for an extensible type.

   Appendices A and B have extensive examples.

25.1.1.  Grammar Construction

   A grammar consists of a collection of productions.  A production has
   a left-hand side and a right-hand side (in this document, separated
   by the "::=" symbol).  The left-hand side (in a context-free grammar)
   is a single non-terminal symbol.  The right-hand side is a sequence
   of non-terminal and terminal symbols.  The terminal symbols are the
   lexical items of the language that the grammar describes.  One of the
   non-terminals is nominated to be the start symbol.  A valid sequence
   of terminals for the language can be generated from the grammar by
   beginning with the start symbol and repeatedly replacing any
   non-terminal with the right-hand side of one of the productions where
   that non-terminal is on the production's left-hand side.  The final
   sequence of terminals is achieved when there are no remaining
   non-terminals to replace.

      Aside: X.680 describes the ASN.1 basic notation using a
      context-free grammar.

   Each NamedType has an associated primary and secondary non-terminal.

      Aside: The secondary non-terminal for a NamedType is used when the
      base type of the type in the NamedType is a SEQUENCE OF type or
      SET OF type.

   Each ExtensionAddition and ExtensionAdditionAlternative has an
   associated non-terminal.  There is a non-terminal associated with the
   extension insertion point of each extensible type.  There is also a
   primary start non-terminal (this is the start symbol) and a secondary
   start non-terminal.  The exact nature of the non-terminals is not
   important, however all the non-terminals MUST be mutually distinct.

   It is adequate for most of the examples in this document (though not
   in the most general case) for the primary non-terminal for a
   NamedType to be the identifier of the NamedType, for the primary
   start non-terminal to be S, for the non-terminals for the instances
   of ExtensionAddition and ExtensionAdditionAlternative to be E1, E2,
   E3, and so on, and for the non-terminals for the extension insertion
   points to be I1, I2, I3, and so on.  The secondary non-terminals are
   labelled by appending a "'" character to the primary non-terminal
   label, e.g., the primary and secondary start non-terminals are S and
   S', respectively.



Legg                          Experimental                     [Page 37]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   Each NamedType and extension insertion point has an associated
   terminal.  There exists a terminal called the general extension
   terminal that is not associated with any specific notation.  The
   general extension terminal and the terminals for the extension
   insertion points are used to represent elements in unknown
   extensions.  The exact nature of the terminals is not important;
   however, the aforementioned terminals MUST be mutually distinct.  The
   terminals are further categorized as either element terminals or
   attribute terminals.  A terminal for a NamedType is an attribute
   terminal if its associated NamedType is an attribute component;
   otherwise, it is an element terminal.  The general extension terminal
   and the terminals for the extension insertion points are categorized
   as element terminals.

   Terminals for attributes in unknown extensions are not explicitly
   provided in the grammar.  Certain productions in the grammar are
   categorized as insertion point productions, and their role in
   accepting unknown attributes is described in Section 25.1.4.

   In the examples in this document, the terminal for a component other
   than an attribute component will be represented as the local name of
   the expanded name of the component enclosed in double quotes, and the
   terminal for an attribute component will be represented as the local
   name of the expanded name of the component prefixed by the '@'
   character and enclosed in double quotes.  The general extension
   terminal will be represented as "*" and the terminals for the
   extension insertion points will be represented as "*1", "*2", "*3",
   and so on.

   The productions generated from a NamedType depend on the base type of
   the type of the NamedType.  The productions for the start
   non-terminals depend on the combining type definition being tested.
   In either case, the procedure for generating productions takes a
   primary non-terminal, a secondary non-terminal (sometimes), and a
   type definition.

   The grammar is constructed beginning with the start non-terminals and
   the combining type definition being tested.

   A grammar is constructed after the COMPONENTS OF transformation
   specified in X.680, Clause 24.4 [X.680].

   Given a primary non-terminal, N, and a type where the base type is a
   SEQUENCE or SET type, a production is added to the grammar with N as
   the left-hand side.  The right-hand side is constructed from an
   initial empty state according to the following cases considered in
   order:




Legg                          Experimental                     [Page 38]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   (1) If an initial RootComponentTypeList is present in the base type,
       then the sequence of primary non-terminals for the components
       nested in that RootComponentTypeList are appended to the right-
       hand side in the order of their definition.

   (2) If an ExtensionAdditions instance is present in the base type and
       not empty, then the non-terminal for the first ExtensionAddition
       nested in the ExtensionAdditions instance is appended to the
       right-hand side.

   (3) If an ExtensionAdditions instance is empty or not present in the
       base type, and the base type is extensible (explicitly or by
       default), and the base type is not subject to a NO-INSERTIONS or
       HOLLOW-INSERTIONS encoding instruction, then the non-terminal for
       the extension insertion point of the base type is appended to the
       right-hand side.

   (4) If a final RootComponentTypeList is present in the base type,
       then the primary non-terminals for the components nested in that
       RootComponentTypeList are appended to the right-hand side in the
       order of their definition.

   The production is an insertion point production if an
   ExtensionAdditions instance is empty or not present in the base type,
   and the base type is extensible (explicitly or by default), and the
   base type is not subject to a NO-INSERTIONS encoding instruction.

   If a component in a ComponentTypeList (in either a
   RootComponentTypeList or an ExtensionAdditionGroup) is marked
   OPTIONAL or DEFAULT, then a production with the primary non-terminal
   of the component as the left-hand side and an empty right-hand side
   is added to the grammar.

   If a component (regardless of the ASN.1 combining type containing it)
   is subject to a GROUP encoding instruction, then one or more
   productions constructed according to the component's type are added
   to the grammar.  Each of these productions has the primary
   non-terminal of the component as the left-hand side.

   If a component (regardless of the ASN.1 combining type containing it)
   is not subject to a GROUP encoding instruction, then a production is
   added to the grammar with the primary non-terminal of the component
   as the left-hand side and the terminal of the component as the
   right-hand side.







Legg                          Experimental                     [Page 39]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   Example

      Consider the following ASN.1 type definition:

         SEQUENCE {
             -- Start of initial RootComponentTypeList.
             one    [ATTRIBUTE] UTF8String,
             two    BOOLEAN OPTIONAL,
             three  INTEGER
             -- End of initial RootComponentTypeList.
         }

      Here is the grammar derived from this type:

         S ::= one two three
         one ::= "@one"
         two ::= "two"
         two ::=
         three ::= "three"

   For each ExtensionAddition (of a SEQUENCE or SET base type), a
   production is added to the grammar where the left-hand side is the
   non-terminal for the ExtensionAddition and the right-hand side is
   initially empty.  If the ExtensionAddition is a ComponentType, then
   the primary non-terminal for the NamedType in the ComponentType is
   appended to the right-hand side; otherwise (an
   ExtensionAdditionGroup), the sequence of primary non-terminals for
   the components nested in the ComponentTypeList in the
   ExtensionAdditionGroup are appended to the right-hand side in the
   order of their definition.  If the ExtensionAddition is followed by
   another ExtensionAddition, then the non-terminal for the next
   ExtensionAddition is appended to the right-hand side; otherwise, if
   the base type is not subject to a NO-INSERTIONS or HOLLOW-INSERTIONS
   encoding instruction, then the non-terminal for the extension
   insertion point of the base type is appended to the right-hand side.
   If the ExtensionAddition is not followed by another ExtensionAddition
   and the base type is not subject to a NO-INSERTIONS encoding
   instruction, then the production is an insertion point production.
   If the empty sequence of terminals cannot be generated from the
   production (it may be necessary to wait until the grammar is
   otherwise complete before making this determination), then another
   production is added to the grammar where the left-hand side is the
   non-terminal for the ExtensionAddition and the right-hand side is
   empty.

      Aside: An extension is always effectively optional since a sender
      may be using an earlier version of the ASN.1 specification where
      none, or only some, of the extensions have been defined.



Legg                          Experimental                     [Page 40]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      Aside: The grammar generated for ExtensionAdditions is structured
      to take account of the condition that an extension can only be
      used if all the earlier extensions are also used [X.680].

   If a SEQUENCE or SET base type is extensible (explicitly or by
   default) and is not subject to a NO-INSERTIONS or HOLLOW-INSERTIONS
   encoding instruction, then:

   (1) a production is added to the grammar where the left-hand side is
       the non-terminal for the extension insertion point of the base
       type and the right-hand side is the general extension terminal
       followed by the non-terminal for the extension insertion point,
       and

   (2) a production is added to the grammar where the left-hand side is
       the non-terminal for the extension insertion point and the
       right-hand side is empty.

   Example

      Consider the following ASN.1 type definition:

         SEQUENCE {
             -- Start of initial RootComponentTypeList.
             one    BOOLEAN,
             two    INTEGER OPTIONAL,
             -- End of initial RootComponentTypeList.
             ...,
             -- Start of ExtensionAdditions.
             four  INTEGER,  -- First ExtensionAddition (E1).
             five  BOOLEAN OPTIONAL,  -- Second ExtensionAddition (E2).
             [[ -- An ExtensionAdditionGroup.
                 six    UTF8String,
                 seven  INTEGER OPTIONAL
             ]], -- Third ExtensionAddition (E3).
             -- End of ExtensionAdditions.
             -- The extension insertion point is here (I1).
             ...,
             -- Start of final RootComponentTypeList.
             three  INTEGER
         }

      Here is the grammar derived from this type:

         S ::= one two E1 three

         E1 ::= four E2
         E1 ::=



Legg                          Experimental                     [Page 41]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


         E2 ::= five E3
         E3 ::= six seven I1
         E3 ::=

         I1 ::= "*" I1
         I1 ::=

         one ::= "one"
         two ::= "two"
         two ::=
         three ::= "three"
         four ::= "four"
         five ::= "five"
         five ::=
         six ::= "six"
         seven ::= "seven"
         seven ::=

      If the SEQUENCE type were subject to a NO-INSERTIONS or
      HOLLOW-INSERTIONS encoding instruction, then the productions for
      I1 would not appear, and the first production for E3 would be:

         E3 ::= six seven

   Given a primary non-terminal, N, and a type where the base type is a
   CHOICE type:

   (1) A production is added to the grammar for each NamedType nested in
       the RootAlternativeTypeList of the base type, where the left-hand
       side is N and the right-hand side is the primary non-terminal for
       the NamedType.

   (2) A production is added to the grammar for each
       ExtensionAdditionAlternative of the base type, where the left-
       hand side is N and the right-hand side is the non-terminal for
       the ExtensionAdditionAlternative.

   (3) If the base type is extensible (explicitly or by default) and the
       base type is not subject to an insertion encoding instruction,
       then:

       (a) A production is added to the grammar where the left-hand side
           is N and the right-hand side is the non-terminal for the
           extension insertion point of the base type.  This production
           is an insertion point production.






Legg                          Experimental                     [Page 42]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


       (b) A production is added to the grammar where the left-hand side
           is the non-terminal for the extension insertion point of the
           base type and the right-hand side is the general extension
           terminal followed by the non-terminal for the extension
           insertion point.

       (c) A production is added to the grammar where the left-hand side
           is the non-terminal for the extension insertion point of the
           base type and the right-hand side is empty.

   (4) If the base type is subject to a HOLLOW-INSERTIONS encoding
       instruction, then a production is added to the grammar where the
       left-hand side is N and the right-hand side is empty.  This
       production is an insertion point production.

   (5) If the base type is subject to a SINGULAR-INSERTIONS encoding
       instruction, then a production is added to the grammar where the
       left-hand side is N and the right-hand side is the general
       extension terminal.  This production is an insertion point
       production.

   (6) If the base type is subject to a UNIFORM-INSERTIONS encoding
       instruction, then:

       (a) A production is added to the grammar where the left-hand side
           is N and the right-hand side is the general extension
           terminal.

              Aside: This production is used to verify the correctness
              of an ASN.1 type definition, but would not be used in the
              implementation of an RXER decoder.  The next production
              takes precedence over it for accepting an unknown element.

       (b) A production is added to the grammar where the left-hand side
           is N and the right-hand side is the terminal for the
           extension insertion point of the base type followed by the
           non-terminal for the extension insertion point.  This
           production is an insertion point production.

       (c) A production is added to the grammar where the left-hand side
           is the non-terminal for the extension insertion point of the
           base type and the right-hand side is the terminal for the
           extension insertion point followed by the non-terminal for
           the extension insertion point.

       (d) A production is added to the grammar where the left-hand side
           is the non-terminal for the extension insertion point of the
           base type and the right-hand side is empty.



Legg                          Experimental                     [Page 43]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   (7) If the base type is subject to a MULTIFORM-INSERTIONS encoding
       instruction, then:

       (a) A production is added to the grammar where the left-hand side
           is N and the right-hand side is the general extension
           terminal followed by the non-terminal for the extension
           insertion point of the base type.  This production is an
           insertion point production.

       (b) A production is added to the grammar where the left-hand side
           is the non-terminal for the extension insertion point of the
           base type and the right-hand side is the general extension
           terminal followed by the non-terminal for the extension
           insertion point.

       (c) A production is added to the grammar where the left-hand side
           is the non-terminal for the extension insertion point of the
           base type and the right-hand side is empty.

   If an ExtensionAdditionAlternative is a NamedType, then a production
   is added to the grammar where the left-hand side is the non-terminal
   for the ExtensionAdditionAlternative and the right-hand side is the
   primary non-terminal for the NamedType.

   If an ExtensionAdditionAlternative is an
   ExtensionAdditionAlternativesGroup, then a production is added to the
   grammar for each NamedType nested in the
   ExtensionAdditionAlternativesGroup, where the left-hand side is the
   non-terminal for the ExtensionAdditionAlternative and the right-hand
   side is the primary non-terminal for the NamedType.





















Legg                          Experimental                     [Page 44]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   Example

      Consider the following ASN.1 type definition:

         CHOICE {
             -- Start of RootAlternativeTypeList.
             one    BOOLEAN,
             two    INTEGER,
             -- End of RootAlternativeTypeList.
             ...,
             -- Start of ExtensionAdditionAlternatives.
             three  INTEGER, -- First ExtensionAdditionAlternative (E1).
             [[ -- An ExtensionAdditionAlternativesGroup.
                 four  UTF8String,
                 five  INTEGER
             ]] -- Second ExtensionAdditionAlternative (E2).
             -- The extension insertion point is here (I1).
         }

      Here is the grammar derived from this type:

         S ::= one
         S ::= two
         S ::= E1
         S ::= E2
         S ::= I1

         I1 ::= "*" I1
         I1 ::=

         E1 ::= three
         E2 ::= four
         E2 ::= five

         one ::= "one"
         two ::= "two"
         three ::= "three"
         four ::= "four"
         five ::= "five"

      If the CHOICE type were subject to a NO-INSERTIONS encoding
      instruction, then the fifth, sixth, and seventh productions would
      be removed.

      If the CHOICE type were subject to a HOLLOW-INSERTIONS encoding
      instruction, then the fifth, sixth, and seventh productions would
      be replaced by:




Legg                          Experimental                     [Page 45]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


         S ::=

      If the CHOICE type were subject to a SINGULAR-INSERTIONS encoding
      instruction, then the fifth, sixth, and seventh productions would
      be replaced by:

         S ::= "*"

      If the CHOICE type were subject to a UNIFORM-INSERTIONS encoding
      instruction, then the fifth and sixth productions would be
      replaced by:

         S ::= "*"
         S ::= "*1" I1

         I1 ::= "*1" I1

      If the CHOICE type were subject to a MULTIFORM-INSERTIONS encoding
      instruction, then the fifth production would be replaced by:

         S ::= "*" I1

   Constraints on a SEQUENCE, SET, or CHOICE type are ignored.  They do
   not affect the grammar being generated.

      Aside: This avoids an awkward situation where values of a subtype
      have to be decoded differently from values of the parent type.  It
      also simplifies the verification procedure.

   Given a primary non-terminal, N, and a type that has a SEQUENCE OF or
   SET OF base type and that permits a value of size zero (i.e., an
   empty sequence or set):

   (1) a production is added to the grammar where the left-hand side of
       the production is N and the right-hand side is the primary
       non-terminal for the NamedType of the component of the
       SEQUENCE OF or SET OF base type, followed by N, and

   (2) a production is added to the grammar where the left-hand side of
       the production is N and the right-hand side is empty.

   Given a primary non-terminal, N, a secondary non-terminal, N', and a
   type that has a SEQUENCE OF or SET OF base type and that does not
   permit a value of size zero:







Legg                          Experimental                     [Page 46]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   (1) a production is added to the grammar where the left-hand side of
       the production is N and the right-hand side is the primary
       non-terminal for the NamedType of the component of the
       SEQUENCE OF or SET OF base type, followed by N', and

   (2) a production is added to the grammar where the left-hand side of
       the production is N' and the right-hand side is the primary
       non-terminal for the NamedType of the component of the
       SEQUENCE OF or SET OF base type, followed by N', and

   (3) a production is added to the grammar where the left-hand side of
       the production is N' and the right-hand side is empty.

   Example

      Consider the following ASN.1 type definition:

         SEQUENCE SIZE(1..MAX) OF number INTEGER

      Here is the grammar derived from this type:

         S ::= number S'
         S' ::= number S'
         S' ::=

         number ::= "number"

   All inner subtyping (InnerTypeContraints) is ignored for the purposes
   of deciding whether a value of size zero is permitted by a
   SEQUENCE OF or SET OF type.

   This completes the description of the transformation of ASN.1
   combining type definitions into a grammar.

25.1.2.  Unique Component Attribution

   This section describes conditions that the grammar must satisfy so
   that each element and attribute in a received RXER encoding can be
   uniquely associated with an ASN.1 component definition.

   Definition (used by the grammar):  A non-terminal, N, is used by the
   grammar if:

   (1) N is the start symbol or

   (2) N appears on the right-hand side of a production where the
       non-terminal on the left-hand side is used by the grammar.




Legg                          Experimental                     [Page 47]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   Definition (multiple derivation paths):  A non-terminal, N, has
   multiple derivation paths if:

   (1) N appears on the right-hand side of a production where the
       non-terminal on the left-hand side has multiple derivation paths,
       or

   (2) N appears on the right-hand side of more than one production
       where the non-terminal on the left-hand side is used by the
       grammar, or

   (3) N is the start symbol and it appears on the right-hand side of a
       production where the non-terminal on the left-hand side is used
       by the grammar.

   For every ASN.1 type with a base type containing components that are
   subject to a GROUP encoding instruction, the grammar derived by the
   method described in this document MUST NOT have:

   (1) two or more primary non-terminals that are used by the grammar
       and are associated with element components having the same
       expanded name, or

   (2) two or more primary non-terminals that are used by the grammar
       and are associated with attribute components having the same
       expanded name, or

   (3) a primary non-terminal that has multiple derivation paths and is
       associated with an attribute component.

      Aside: Case (1) is in response to component referencing notations
      that are evaluated with respect to the XML encoding of an abstract
      value.  Case (1) guarantees, without having to do extensive
      testing (which would necessarily have to take account of encoding
      instructions for all other encoding rules), that all sibling
      elements with the same expanded name will be associated with
      equivalent type definitions.  Such equivalence allows a component
      referenced by element name to be re-encoded using a different set
      of ASN.1 encoding rules without ambiguity as to which type
      definition and encoding instructions apply.

      Cases (2) and (3) ensure that an attribute name is always uniquely
      associated with one component that can occur at most once and is
      always nested in the same part of an abstract value.







Legg                          Experimental                     [Page 48]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   Example

      The following example types illustrate various uses and misuses of
      the GROUP encoding instruction with respect to unique component
      attribution:

         TA ::= SEQUENCE {
             a  [GROUP] TB,
             b  [GROUP] CHOICE {
                 a  [GROUP] TB,
                 b  [NAME AS "c"] [ATTRIBUTE] INTEGER,
                 c  INTEGER,
                 d  TB,
                 e  [GROUP] TD,
                 f  [ATTRIBUTE] UTF8String
             },
             c  [ATTRIBUTE] INTEGER,
             d  [GROUP] SEQUENCE OF
                 a [GROUP] SEQUENCE {
                     a  [ATTRIBUTE] OBJECT IDENTIFIER,
                     b  INTEGER
                 },
             e  [NAME AS "c"] INTEGER,
             COMPONENTS OF TD
         }

         TB ::= SEQUENCE {
             a  INTEGER,
             b  [ATTRIBUTE] BOOLEAN,
             COMPONENTS OF TC
         }

         TC ::= SEQUENCE {
             f  OBJECT IDENTIFIER
         }

         TD ::= SEQUENCE {
             g  OBJECT IDENTIFIER
         }

      The grammar for TA is constructed after performing the
      COMPONENTS OF transformation.  The result of this transformation
      is shown next.  This example will depart from the usual convention
      of using just the identifier of a NamedType to represent the
      primary non-terminal for that NamedType.  A label relative to the
      outermost type will be used instead to better illustrate unique
      component attribution.  The labels used for the non-terminals are
      shown down the right-hand side.



Legg                          Experimental                     [Page 49]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


         TA ::= SEQUENCE {
             a  [GROUP] TB,                             -- TA.a
             b  [GROUP] CHOICE {                        -- TA.b
                 a  [GROUP] TB,                         -- TA.b.a
                 b  [NAME AS "c"] [ATTRIBUTE] INTEGER,  -- TA.b.b
                 c  INTEGER,                            -- TA.b.c
                 d  TB,                                 -- TA.b.d
                 e  [GROUP] TD,                         -- TA.b.e
                 f  [ATTRIBUTE] UTF8String              -- TA.b.f
             },
             c  [ATTRIBUTE] INTEGER,                    -- TA.c
             d  [GROUP] SEQUENCE OF                     -- TA.d
                 a [GROUP] SEQUENCE {                   -- TA.d.a
                     a  [ATTRIBUTE] OBJECT IDENTIFIER,  -- TA.d.a.a
                     b  INTEGER                         -- TA.d.a.b
                 },
             e  [NAME AS "c"] INTEGER,                  -- TA.e
             g  OBJECT IDENTIFIER                       -- TA.g
         }

         TB ::= SEQUENCE {
             a  INTEGER,                                -- TB.a
             b  [ATTRIBUTE] BOOLEAN,                    -- TB.b
             f  OBJECT IDENTIFIER                       -- TB.f
         }

         -- Type TC is no longer of interest. --

         TD ::= SEQUENCE {
             g  OBJECT IDENTIFIER                       -- TD.g
         }

      The associated grammar is:

         S ::= TA.a TA.b TA.c TA.d TA.e TA.g

         TA.a ::= TB.a TB.b TB.f

         TB.a ::= "a"
         TB.b ::= "@b"
         TB.f ::= "f"

         TA.b ::= TA.b.a
         TA.b ::= TA.b.b
         TA.b ::= TA.b.c
         TA.b ::= TA.b.d
         TA.b ::= TA.b.e
         TA.b ::= TA.b.f



Legg                          Experimental                     [Page 50]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


         TA.b.a ::= TB.a TB.b TB.f
         TA.b.b ::= "@c"
         TA.b.c ::= "c"
         TA.b.d ::= "d"
         TA.b.e ::= TD.g
         TA.b.f ::= "@f"

         TD.g ::= "g"

         TA.c ::= "@c"

         TA.d ::= TA.d.a TA.d
         TA.d ::=

         TA.d.a ::= TA.d.a.a TA.d.a.b

         TA.d.a.a := "@a"
         TA.d.a.b ::= "b"

         TA.e ::= "c"

         TA.g ::= "g"

      All the non-terminals are used by the grammar.

      The type definition for TA is invalid because there are two
      instances where two or more primary non-terminals are associated
      with element components having the same expanded name:

      (1) TA.b.c and TA.e (both generate the terminal "c"), and

      (2) TD.g and TA.g (both generate the terminal "g").

      In case (2), TD.g and TA.g are derived from the same instance of
      NamedType notation, but become distinct components following the
      COMPONENTS OF transformation.  AUTOMATIC tagging is applied after
      the COMPONENTS OF transformation, which means that the types of
      the components corresponding to TD.g and TA.g will end up with
      different tags, and therefore the types will not be equivalent.

      The type definition for TA is also invalid because there is one
      instance where two or more primary non-terminals are associated
      with attribute components having the same expanded name:  TA.b.b
      and TA.c (both generate the terminal "@c").







Legg                          Experimental                     [Page 51]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      The non-terminals with multiple derivation paths are:  TA.d,
      TA.d.a, TA.d.a.a, TA.d.a.b, TB.a, TB.b, and TB.f.  The type
      definition for TA is also invalid because TA.d.a.a and TB.b are
      primary non-terminals that are associated with an attribute
      component.

25.1.3.  Deterministic Grammars

   Let the First Set of a production P, denoted First(P), be the set of
   all element terminals T where T is the first element terminal in a
   sequence of terminals that can be generated from the right-hand side
   of P.  There can be any number of leading attribute terminals before
   T.

   Let the Follow Set of a non-terminal N, denoted Follow(N), be the set
   of all element terminals T where T is the first element terminal
   following N in a sequence of non-terminals and terminals that can be
   generated from the grammar.  There can be any number of attribute
   terminals between N and T.  If a sequence of non-terminals and
   terminals can be generated from the grammar where N is not followed
   by any element terminals, then Follow(N) also contains a special end
   terminal, denoted by "$".

      Aside: If N does not appear on the right-hand side of any
      production, then Follow(N) will be empty.

   For a production P, let the predicate Empty(P) be true if and only if
   the empty sequence of terminals can be generated from P.  Otherwise,
   Empty(P) is false.

   Definition (base grammar):  The base grammar is a rewriting of the
   grammar in which the non-terminals for every ExtensionAddition and
   ExtensionAdditionAlternative are removed from the right-hand side of
   all productions.

   For a production P, let the predicate Preselected(P) be true if and
   only if every sequence of terminals that can be generated from the
   right-hand side of P using only the base grammar contains at least
   one attribute terminal.  Otherwise, Preselected(P) is false.

   The Select Set of a production P, denoted Select(P), is empty if
   Preselected(P) is true; otherwise, it contains First(P).  Let N be
   the non-terminal on the left-hand side of P.  If Empty(P) is true,
   then Select(P) also contains Follow(N).







Legg                          Experimental                     [Page 52]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      Aside: It may appear somewhat dubious to include the attribute
      components in the grammar because, in reality, attributes appear
      unordered within the start tag of an element, and not interspersed
      with the child elements as the grammar would suggest.  This is why
      attribute terminals are ignored in composing the First Sets and
      Follow Sets.  However, the attribute terminals are important in
      composing the Select Sets because they can preselect a production
      and can prevent a production from being able to generate an empty
      sequence of terminals.  In real terms, this corresponds to an RXER
      decoder using the attributes to determine the presence or absence
      of optional components and to select between the alternatives of a
      CHOICE, even before considering the child elements.

      An attribute appearing in an extension isn't used to preselect a
      production since, in general, a decoder using an earlier version
      of the specification would not be able to associate the attribute
      with any particular extension insertion point.

   Let the Reach Set of a non-terminal N, denoted Reach(N), be the set
   of all element terminals T where T appears in a sequence of terminals
   that can be generated from N.

      Aside: It can be readily shown that all the optional attribute
      components and all but one of the mandatory attribute components
      of a SEQUENCE or SET type can be ignored in constructing the
      grammar because their omission does not alter the First, Follow,
      Select, or Reach Sets, or the evaluation of the Preselected and
      Empty predicates.

   A grammar is deterministic (for the purposes of an RXER decoder) if
   and only if:

   (1) there do not exist two productions P and Q, with the same
       non-terminal on the left-hand side, where the intersection of
       Select(P) and Select(Q) is not empty, and

   (2) there does not exist a non-terminal E for an ExtensionAddition or
       ExtensionAdditionAlternative where the intersection of Reach(E)
       and Follow(E) is not empty.

      Aside: In case (1), if the intersection is not empty, then a
      decoder would have two or more possible ways to attempt to decode
      the input into an abstract value.  In case (2), if the
      intersection is not empty, then a decoder using an earlier version
      of the ASN.1 specification would confuse an element in an unknown
      (to that decoder) extension with a known component following the
      extension.




Legg                          Experimental                     [Page 53]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      Aside: In the absence of any attribute components, case (1) is the
      test for an LL(1) grammar.

   For every ASN.1 type with a base type containing components that are
   subject to a GROUP encoding instruction, the grammar derived by the
   method described in this document MUST be deterministic.

25.1.4.  Attributes in Unknown Extensions

   An insertion point production is able to accept unknown attributes if
   the non-terminal on the left-hand side of the production does not
   have multiple derivation paths.

      Aside: If the non-terminal has multiple derivation paths, then any
      future extension cannot possibly contain an attribute component
      because that would violate the requirements of Section 25.1.2.

   For a deterministic grammar, there is only one possible way to
   construct a sequence of element terminals matching the element
   content of an element in a correctly formed RXER encoding.  Any
   unknown attributes of the element are accepted if at least one
   insertion point production that is able to accept unknown attributes
   is used in that construction.

   Example

      Consider this type definition:

         CHOICE {
             one  UTF8String,
             two  [GROUP] SEQUENCE {
                  three  INTEGER,
                  ...
             }
         }

      The associated grammar is:

         S ::= one
         S ::= two

         two ::= three I1

         I1 ::= "*" I1
         I1 ::=

         one ::= "one"
         three ::= "three"



Legg                          Experimental                     [Page 54]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      The third production is an insertion point production, and it is
      able to accept unknown attributes.

      When decoding a value of this type, if the element content
      contains a <one> child element, then any unrecognized attribute
      would be illegal as the insertion point production would not be
      used to recognize the input (the "one" alternative does not admit
      an extension insertion point).  If the element content contains a
      <three> element, then an unrecognized attribute would be accepted
      because the insertion point production would be used to recognize
      the input (the "two" alternative that generates the <three>
      element has an extensible type).

      If the SEQUENCE type were prefixed by a NO-INSERTIONS encoding
      instruction, then the third, fourth, and fifth productions would
      be replaced by:

         two ::= three

      With this change, any unrecognized attribute would be illegal for
      the "two" alternative also, since the replacement production is
      not an insertion point production.

   If more than one insertion point production that is able to accept
   unknown attributes is used in constructing a matching sequence of
   element terminals, then a decoder is free to associate an
   unrecognized attribute with any one of the extension insertion points
   corresponding to those insertion point productions.  The
   justification for doing so comes from the following two observations:

   (1) If the encoding of an abstract value contains an extension where
       the type of the extension is unknown to the receiver, then it is
       generally impossible to re-encode the value using a different set
       of encoding rules, including the canonical variant of the
       received encoding.  This is true no matter which encoding rules
       are being used.  It is desirable for a decoder to be able to
       accept and store the raw encoding of an extension without raising
       an error, and to re-insert the raw encoding of the extension when
       re-encoding the abstract value using the same non-canonical
       encoding rules.  However, there is little more that an
       application can do with an unknown extension.

       An application using RXER can successfully accept, store, and
       re-encode an unrecognized attribute regardless of which extension
       insertion point it might be ascribed to.






Legg                          Experimental                     [Page 55]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   (2) Even if there is a single extension insertion point, an unknown
       extension could still be the encoding of a value of any one of an
       infinite number of valid type definitions.  For example, an
       attribute or element component could be nested to any arbitrary
       depth within CHOICEs whose components are subject to GROUP
       encoding instructions.

          Aside: A similar series of nested CHOICEs could describe an
          unknown extension in a Basic Encoding Rules (BER) encoding
          [X.690].

26.  Security Considerations

   ASN.1 compiler implementors should take special care to be thorough
   in checking that the GROUP encoding instruction has been correctly
   used; otherwise, ASN.1 specifications with ambiguous RXER encodings
   could be deployed.

   Ambiguous encodings mean that the abstract value recovered by a
   decoder may differ from the original abstract value that was encoded.
   If that is the case, then a digital signature generated with respect
   to the original abstract value (using a canonical encoding other than
   CRXER) will not be successfully verified by a receiver using the
   decoded abstract value.  Also, an abstract value may have
   security-sensitive fields, and in particular, fields used to grant or
   deny access.  If the decoded abstract value differs from the encoded
   abstract value, then a receiver using the decoded abstract value will
   be applying different security policy than that embodied in the
   original abstract value.

27.  References

27.1.  Normative References

   [BCP14]    Bradner, S., "Key words for use in RFCs to Indicate
              Requirement Levels", BCP 14, RFC 2119, March 1997.

   [URI]      Berners-Lee, T., Fielding, R. and L. Masinter, "Uniform
              Resource Identifiers (URI): Generic Syntax", STD 66, RFC
              3986, January 2005.

   [RXER]     Legg, S. and D. Prager, "Robust XML Encoding Rules (RXER)
              for Abstract Syntax Notation One (ASN.1)", RFC 4910, July
              2007.

   [ASN.X]    Legg, S., "Abstract Syntax Notation X (ASN.X)", RFC 4912,
              July 2007.




Legg                          Experimental                     [Page 56]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   [X.680]    ITU-T Recommendation X.680 (07/02) | ISO/IEC 8824-1,
              Information technology - Abstract Syntax Notation One
              (ASN.1):  Specification of basic notation.

   [X.680-1]  ITU-T Recommendation X.680 (2002) Amendment 1 (10/03) |
              ISO/IEC 8824-1:2002/Amd 1:2004, Support for EXTENDED-XER.

   [X.683]    ITU-T Recommendation X.683 (07/02) | ISO/IEC 8824-4,
              Information technology - Abstract Syntax Notation One
              (ASN.1):  Parameterization of ASN.1 specifications.

   [XML10]    Bray, T., Paoli, J., Sperberg-McQueen, C., Maler, E. and
              F. Yergeau, "Extensible Markup Language (XML) 1.0 (Fourth
              Edition)", W3C Recommendation,
              http://www.w3.org/TR/2006/REC-xml-20060816, August 2006.

   [XMLNS10]  Bray, T., Hollander, D., Layman, A., and R. Tobin,
              "Namespaces in XML 1.0 (Second Edition)", W3C
              Recommendation,
              http://www.w3.org/TR/2006/REC-xml-names-20060816, August
              2006.

   [XSD1]     Thompson, H., Beech, D., Maloney, M. and N. Mendelsohn,
              "XML Schema Part 1: Structures Second Edition", W3C
              Recommendation,
              http://www.w3.org/TR/2004/REC-xmlschema-1-20041028/,
              October 2004.

   [XSD2]     Biron, P. and A. Malhotra, "XML Schema Part 2: Datatypes
              Second Edition", W3C Recommendation,
              http://www.w3.org/TR/2004/REC-xmlschema-2-20041028/,
              October 2004.

   [RNG]      Clark, J. and M. Makoto, "RELAX NG Tutorial", OASIS
              Committee Specification, http://www.oasis-open.org/
              committees/relax-ng/tutorial-20011203.html, December 2001.

27.2.  Informative References

   [INFOSET]  Cowan, J. and R. Tobin, "XML Information Set (Second
              Edition)", W3C Recommendation, http://www.w3.org/
              TR/2004/REC-xml-infoset-20040204, February 2004.

   [X.690]    ITU-T Recommendation X.690 (07/02) | ISO/IEC 8825-1,
              Information technology - ASN.1 encoding rules:
              Specification of Basic Encoding Rules (BER), Canonical
              Encoding Rules (CER) and Distinguished Encoding Rules
              (DER).



Legg                          Experimental                     [Page 57]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


Appendix A.  GROUP Encoding Instruction Examples

   This appendix is non-normative.

   This appendix contains examples of both correct and incorrect use of
   the GROUP encoding instruction, determined with respect to the
   grammars derived from the example type definitions.  The productions
   of the grammars are labeled for convenience.  Sets and predicates for
   non-terminals with only one production will be omitted from the
   examples since they never indicate non-determinism.

   The requirements of Section 25.1.2 ("Unique Component Attribution")
   are satisfied by all the examples in this appendix and the appendices
   that follow it.

A.1.  Example 1

   Consider this type definition:

      SEQUENCE {
          one    [GROUP] SEQUENCE {
              two    UTF8String OPTIONAL
          } OPTIONAL,
          three  INTEGER
      }

   The associated grammar is:

      P1:  S ::= one three
      P2:  one ::= two
      P3:  one ::=
      P4:  two ::= "two"
      P5:  two ::=
      P6:  three ::= "three"

   Select Sets have to be evaluated to test the validity of the type
   definition.  The grammar leads to the following sets and predicates:

      First(P2) = { "two" }
      First(P3) = { }
      Preselected(P2) = Preselected(P3) = false
      Empty(P2) = Empty(P3) = true
      Follow(one) = { "three" }
      Select(P2) = First(P2) + Follow(one) = { "two", "three" }
      Select(P3) = First(P3) + Follow(one) = { "three" }






Legg                          Experimental                     [Page 58]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      First(P4) = { "two" }
      First(P5) = { }
      Preselected(P4) = Preselected(P5) = Empty(P4) = false
      Empty(P5) = true
      Follow(two) = { "three" }
      Select(P4) = First(P4) = { "two" }
      Select(P5) = First(P5) + Follow(two) = { "three" }

   The intersection of Select(P2) and Select(P3) is not empty; hence,
   the grammar is not deterministic, and the type definition is not
   valid.  If the RXER encoding of a value of the type does not have a
   child element <two>, then it is not possible to determine whether the
   "one" component is present or absent in the value.

   Now consider this type definition with attributes in the "one"
   component:

      SEQUENCE {
          one    [GROUP] SEQUENCE {
              two    UTF8String OPTIONAL,
              four   [ATTRIBUTE] BOOLEAN,
              five   [ATTRIBUTE] BOOLEAN OPTIONAL
          } OPTIONAL,
          three  INTEGER
      }

   The associated grammar is:

      P1:  S ::= one three
      P2:  one ::= two four five
      P3:  one ::=
      P4:  two ::= "two"
      P5:  two ::=
      P6:  four ::= "@four"
      P7:  five ::= "@five"
      P8:  five ::=
      P9:  three ::= "three"

   This grammar leads to the following sets and predicates:

      First(P2) = { "two" }
      First(P3) = { }
      Preselected(P3) = Empty(P2) = false
      Preselected(P2) = Empty(P3) = true
      Follow(one) = { "three" }
      Select(P2) = { }
      Select(P3) = First(P3) + Follow(one) = { "three" }




Legg                          Experimental                     [Page 59]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      First(P4) = { "two" }
      First(P5) = { }
      Preselected(P4) = Preselected(P5) = Empty(P4) = false
      Empty(P5) = true
      Follow(two) = { "three" }
      Select(P4) = First(P4) = { "two" }
      Select(P5) = First(P5) + Follow(two) = { "three" }

      First(P7) = { }
      First(P8) = { }
      Preselected(P8) = Empty(P7) = false
      Preselected(P7) = Empty(P8) = true
      Follow(five) = { "three" }
      Select(P7) = { }
      Select(P8) = First(P8) + Follow(five) = { "three" }

   The intersection of Select(P2) and Select(P3) is empty, as is the
   intersection of Select(P4) and Select(P5) and the intersection of
   Select(P7) and Select(P8); hence, the grammar is deterministic, and
   the type definition is valid.  In a correct RXER encoding, the "one"
   component will be present if and only if the "four" attribute is
   present.

A.2.  Example 2

   Consider this type definition:

      CHOICE {
          one    [GROUP] SEQUENCE {
              two    [ATTRIBUTE] BOOLEAN OPTIONAL
          },
          three  INTEGER,
          four   [GROUP] SEQUENCE {
              five   BOOLEAN OPTIONAL
          }
      }

   The associated grammar is:

      P1:  S ::= one
      P2:  S ::= three
      P3:  S ::= four
      P4:  one ::= two
      P5:  two ::= "@two"
      P6:  two ::=
      P7:  three ::= "three"
      P8:  four ::= five
      P9:  five ::= "five"



Legg                          Experimental                     [Page 60]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      P10: five ::=

   This grammar leads to the following sets and predicates:

      First(P1) = { }
      First(P2) = { "three" }
      First(P3) = { "five" }
      Preselected(P1) = Preselected(P2) = Preselected(P3) = false
      Empty(P2) = false
      Empty(P1) = Empty(P3) = true
      Follow(S) = { "$" }
      Select(P1) = First(P1) + Follow(S) = { "$" }
      Select(P2) = First(P2) = { "three" }
      Select(P3) = First(P3) + Follow(S) = { "five", "$" }

      First(P5) = { }
      First(P6) = { }
      Preselected(P6) = Empty(P5) = false
      Preselected(P5) = Empty(P6) = true
      Follow(two) = { "$" }
      Select(P5) = { }
      Select(P6) = First(P6) + Follow(two) = { "$" }

      First(P9) = { "five" }
      First(P10) = { }
      Preselected(P9) = Preselected(P10) = Empty(P9) = false
      Empty(P10) = true
      Follow(five) = { "$" }
      Select(P9) = First(P9) = { "five" }
      Select(P10) = First(P10) + Follow(five) = { "$" }

   The intersection of Select(P1) and Select(P3) is not empty; hence,
   the grammar is not deterministic, and the type definition is not
   valid.  If the RXER encoding of a value of the type is empty, then it
   is not possible to determine whether the "one" alternative or the
   "four" alternative has been chosen.

   Now consider this slightly different type definition:

      CHOICE {
          one    [GROUP] SEQUENCE {
              two    [ATTRIBUTE] BOOLEAN
          },
          three  INTEGER,
          four   [GROUP] SEQUENCE {
              five   BOOLEAN OPTIONAL
          }
      }



Legg                          Experimental                     [Page 61]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The associated grammar is:

      P1:  S ::= one
      P2:  S ::= three
      P3:  S ::= four
      P4:  one ::= two
      P5:  two ::= "@two"
      P6:  three ::= "three"
      P7:  four ::= five
      P8:  five ::= "five"
      P9:  five ::=

   This grammar leads to the following sets and predicates:

      First(P1) = { }
      First(P2) = { "three" }
      First(P3) = { "five" }
      Preselected(P2) = Preselected(P3) = false
      Empty(P1) = Empty(P2) = false
      Preselected(P1) = Empty(P3) = true
      Follow(S) = { "$" }
      Select(P1) = { }
      Select(P2) = First(P2) = { "three" }
      Select(P3) = First(P3) + Follow(S) = { "five", "$" }

      First(P8) = { "five" }
      First(P9) = { }
      Preselected(P8) = Preselected(P9) = Empty(P8) = false
      Empty(P9) = true
      Follow(five) = { "$" }
      Select(P8) = First(P8) = { "five" }
      Select(P9) = First(P9) + Follow(five) = { "$" }

   The intersection of Select(P1) and Select(P2) is empty, the
   intersection of Select(P1) and Select(P3) is empty, the intersection
   of Select(P2) and Select(P3) is empty, and the intersection of
   Select(P8) and Select(P9) is empty; hence, the grammar is
   deterministic, and the type definition is valid.  The "one" and
   "four" alternatives can be distinguished because the "one"
   alternative has a mandatory attribute.











Legg                          Experimental                     [Page 62]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


A.3.  Example 3

   Consider this type definition:

      SEQUENCE {
          one  [GROUP] CHOICE {
              two    [ATTRIBUTE] BOOLEAN,
              three  [GROUP] SEQUENCE OF number INTEGER
          } OPTIONAL
      }

   The associated grammar is:

      P1:  S ::= one
      P2:  one ::= two
      P3:  one ::= three
      P4:  one ::=
      P5:  two ::= "@two"
      P6:  three ::= number three
      P7:  three ::=
      P8:  number ::= "number"

   This grammar leads to the following sets and predicates:

      First(P2) = { }
      First(P3) = { "number" }
      First(P4) = { }
      Preselected(P3) = Preselected(P4) = Empty(P2) = false
      Preselected(P2) = Empty(P3) = Empty(P4) = true
      Follow(one) = { "$" }
      Select(P2) = { }
      Select(P3) = First(P3) + Follow(one) = { "number", "$" }
      Select(P4) = First(P4) + Follow(one) = { "$" }

      First(P6) = { "number" }
      First(P7) = { }
      Preselected(P6) = Preselected(P7) = Empty(P6) = false
      Empty(P7) = true
      Follow(three) = { "$" }
      Select(P6) = First(P6) = { "number" }
      Select(P7) = First(P7) + Follow(three) = { "$" }

   The intersection of Select(P3) and Select(P4) is not empty; hence,
   the grammar is not deterministic, and the type definition is not
   valid.  If the RXER encoding of a value of the type is empty, then it
   is not possible to determine whether the "one" component is absent or
   the empty "three" alternative has been chosen.




Legg                          Experimental                     [Page 63]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


A.4.  Example 4

   Consider this type definition:

      SEQUENCE {
          one  [GROUP] CHOICE {
              two    [ATTRIBUTE] BOOLEAN,
              three  [ATTRIBUTE] BOOLEAN
          } OPTIONAL
      }

   The associated grammar is:

      P1:  S ::= one
      P2:  one ::= two
      P3:  one ::= three
      P4:  one ::=
      P5:  two ::= "@two"
      P6:  three ::= "@three"

   This grammar leads to the following sets and predicates:

      First(P2) = { }
      First(P3) = { }
      First(P4) = { }
      Preselected(P4) = Empty(P2) = Empty(P3) = false
      Preselected(P2) = Preselected(P3) = Empty(P4) = true
      Follow(one) = { "$" }
      Select(P2) = { }
      Select(P3) = { }
      Select(P4) = First(P4) + Follow(one) = { "$" }

   The intersection of Select(P2) and Select(P3) is empty, the
   intersection of Select(P2) and Select(P4) is empty, and the
   intersection of Select(P3) and Select(P4) is empty; hence, the
   grammar is deterministic, and the type definition is valid.

A.5.  Example 5

   Consider this type definition:

      SEQUENCE {
          one  [GROUP] SEQUENCE OF number INTEGER OPTIONAL
      }







Legg                          Experimental                     [Page 64]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The associated grammar is:

      P1:  S ::= one
      P2:  one ::= number one
      P3:  one ::=
      P4:  one ::=
      P5:  number ::= "number"

   P3 is generated during the processing of the SEQUENCE OF type.  P4 is
   generated because the "one" component is optional.

   This grammar leads to the following sets and predicates:

      First(P2) = { "number" }
      First(P3) = { }
      First(P4) = { }
      Preselected(P2) = Preselected(P3) = Preselected(P4) = false
      Empty(P2) = false
      Empty(P3) = Empty(P4) = true
      Follow(one) = { "$" }
      Select(P2) = First(P2) = { "number" }
      Select(P3) = First(P3) + Follow(one) = { "$" }
      Select(P4) = First(P4) + Follow(one) = { "$" }

   The intersection of Select(P3) and Select(P4) is not empty; hence,
   the grammar is not deterministic, and the type definition is not
   valid.  If the RXER encoding of a value of the type does not have any
   <number> child elements, then it is not possible to determine whether
   the "one" component is present or absent in the value.

   Consider this similar type definition with a SIZE constraint:

      SEQUENCE {
          one  [GROUP] SEQUENCE SIZE(1..MAX) OF number INTEGER OPTIONAL
      }

   The associated grammar is:

      P1:  S ::= one
      P2:  one ::= number one'
      P3:  one' ::= number one'
      P4:  one' ::=
      P5:  one ::=
      P6:  number ::= "number"







Legg                          Experimental                     [Page 65]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   This grammar leads to the following sets and predicates:

      First(P2) = { "number" }
      First(P5) = { }
      Preselected(P2) = Preselected(P5) = Empty(P2) = false
      Empty(P5) = true
      Follow(one) = { "$" }
      Select(P2) = First(P2) = { "number" }
      Select(P5) = First(P5) + Follow(one) = { "$" }

      First(P3) = { "number" }
      First(P4) = { }
      Preselected(P3) = Preselected(P4) = Empty(P3) = false
      Empty(P4) = true
      Follow(one') = { "$" }
      Select(P3) = First(P3) = { "number" }
      Select(P4) = First(P4) + Follow(one') = { "$" }

   The intersection of Select(P2) and Select(P5) is empty, as is the
   intersection of Select(P3) and Select(P4); hence, the grammar is
   deterministic, and the type definition is valid.  If there are no
   <number> child elements, then the "one" component is necessarily
   absent and there is no ambiguity.

A.6.  Example 6

   Consider this type definition:

      SEQUENCE {
          beginning  [GROUP] List,
          middle     UTF8String OPTIONAL,
          end        [GROUP] List
      }

      List ::= SEQUENCE OF string UTF8String

   The associated grammar is:

      P1:  S ::= beginning middle end
      P2:  beginning ::= string beginning
      P3:  beginning ::=
      P4:  middle ::= "middle"
      P5:  middle ::=
      P6:  end ::= string end
      P7:  end ::=
      P8:  string ::= "string"





Legg                          Experimental                     [Page 66]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   This grammar leads to the following sets and predicates:

      First(P2) = { "string" }
      First(P3) = { }
      Preselected(P2) = Preselected(P3) = Empty(P2) = false
      Empty(P3) = true
      Follow(beginning) = { "middle", "string", "$" }
      Select(P2) = First(P2) = { "string" }
      Select(P3) = First(P3) + Follow(beginning)
                 = { "middle", "string", "$" }

      First(P4) = { "middle" }
      First(P5) = { }
      Preselected(P4) = Preselected(P5) = Empty(P4) = false
      Empty(P5) = true
      Follow(middle) = { "string", "$" }
      Select(P4) = First(P4) = { "middle" }
      Select(P5) = First(P5) + Follow(middle) = { "string", "$" }

      First(P6) = { "string" }
      First(P7) = { }
      Preselected(P6) = Preselected(P7) = Empty(P6) = false
      Empty(P7) = true
      Follow(end) = { "$" }
      Select(P6) = First(P6) = { "string" }
      Select(P7) = First(P7) + Follow(end) = { "$" }

   The intersection of Select(P2) and Select(P3) is not empty; hence,
   the grammar is not deterministic, and the type definition is not
   valid.

   Now consider the following type definition:

      SEQUENCE {
          beginning     [GROUP] List,
          middleAndEnd  [GROUP] SEQUENCE {
              middle        UTF8String,
              end           [GROUP] List
          } OPTIONAL
      }

   The associated grammar is:

      P1:  S ::= beginning middleAndEnd
      P2:  beginning ::= string beginning
      P3:  beginning ::=
      P4:  middleAndEnd ::= middle end
      P5:  middleAndEnd ::=



Legg                          Experimental                     [Page 67]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      P6:  middle ::= "middle"
      P7:  end ::= string end
      P8:  end ::=
      P9:  string ::= "string"

   This grammar leads to the following sets and predicates:

      First(P2) = { "string" }
      First(P3) = { }
      Preselected(P2) = Preselected(P3) = Empty(P2) = false
      Empty(P3) = true
      Follow(beginning) = { "middle", "$" }
      Select(P2) = First(P2) = { "string" }
      Select(P3) = First(P3) + Follow(beginning) = { "middle", "$" }

      First(P4) = { "middle" }
      First(P5) = { }
      Preselected(P4) = Preselected(P5) = Empty(P4) = false
      Empty(P5) = true
      Follow(middleAndEnd) = { "$" }
      Select(P4) = First(P4) = { "middle" }
      Select(P5) = First(P5) + Follow(middleAndEnd) = { "$" }

      First(P7) = { "string" }
      First(P8) = { }
      Preselected(P7) = Preselected(P8) = Empty(P7) = false
      Empty(P8) = true
      Follow(end) = { "$" }
      Select(P7) = First(P7) = { "string" }
      Select(P8) = First(P8) + Follow(end) = { "$" }

   The intersection of Select(P2) and Select(P3) is empty, as is the
   intersection of Select(P4) and Select(P5) and the intersection of
   Select(P7) and Select(P8); hence, the grammar is deterministic, and
   the type definition is valid.

A.7.  Example 7

   Consider the following type definition:

      SEQUENCE SIZE(1..MAX) OF
          one  [GROUP] SEQUENCE {
              two    INTEGER OPTIONAL
          }







Legg                          Experimental                     [Page 68]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The associated grammar is:

      P1:  S ::= one S'
      P2:  S' ::= one S'
      P3:  S' ::=
      P4:  one ::= two
      P5:  two ::= "two"
      P6:  two ::=

   This grammar leads to the following sets and predicates:

      First(P2) = { "two" }
      First(P3) = { }
      Preselected(P2) = Preselected(P3) = false
      Empty(P2) = Empty(P3) = true
      Follow(S') = { "$" }
      Select(P2) = First(P2) + Follow(S') = { "two", "$" }
      Select(P3) = First(P3) + Follow(S') = { "$" }

      First(P5) = { "two" }
      First(P6) = { }
      Preselected(P5) = Preselected(P6) = Empty(P5) = false
      Empty(P6) = true
      Follow(two) = { "two", "$" }
      Select(P5) = First(P5) = { "two" }
      Select(P6) = First(P6) + Follow(two) = { "two", "$" }

   The intersection of Select(P2) and Select(P3) is not empty and the
   intersection of Select(P5) and Select(P6) is not empty; hence, the
   grammar is not deterministic, and the type definition is not valid.
   The encoding of a value of the type contains an indeterminate number
   of empty instances of the component type.

A.8.  Example 8

   Consider the following type definition:

      SEQUENCE OF
          list [GROUP] SEQUENCE SIZE(1..MAX) OF number INTEGER

   The associated grammar is:

      P1:  S ::= list S
      P2:  S ::=
      P3:  list ::= number list'
      P4:  list' ::= number list'
      P5:  list' ::=
      P6:  number ::= "number"



Legg                          Experimental                     [Page 69]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   This grammar leads to the following sets and predicates:

      First(P1) = { "number" }
      First(P2) = { }
      Preselected(P1) = Preselected(P2) = Empty(P1) = false
      Empty(P2) = true
      Follow(S) = { "$" }
      Select(P1) = First(P1) = { "number" }
      Select(P2) = First(P2) + Follow(S) = { "$" }

      First(P4) = { "number" }
      First(P5) = { }
      Preselected(P4) = Preselected(P5) = Empty(P4) = false
      Empty(P5) = true
      Follow(list') = { "number", "$" }
      Select(P4) = First(P4) = { "number" }
      Select(P5) = First(P5) + Follow(list') = { "number", "$" }

   The intersection of Select(P4) and Select(P5) is not empty; hence,
   the grammar is not deterministic, and the type definition is not
   valid.  The type describes a list of lists, but it is not possible
   for a decoder to determine where the outer lists begin and end.

A.9.  Example 9

   Consider the following type definition:

      SEQUENCE OF item [GROUP] SEQUENCE {
          before  [GROUP] OneAndTwo,
          core    UTF8String,
          after   [GROUP] OneAndTwo OPTIONAL
      }

      OneAndTwo ::= SEQUENCE {
          non-core  UTF8String
      }

   The associated grammar is:

      P1:  S ::= item S
      P2:  S ::=
      P3:  item ::= before core after
      P4:  before ::= non-core
      P5:  non-core ::= "non-core"
      P6:  core ::= "core"
      P7:  after ::= non-core
      P8:  after ::=




Legg                          Experimental                     [Page 70]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   This grammar leads to the following sets and predicates:

      First(P1) = { "non-core" }
      First(P2) = { }
      Preselected(P1) = Preselected(P2) = Empty(P1) = false
      Empty(P2) = true
      Follow(S) = { "$" }
      Select(P1) = First(P1) = { "non-core" }
      Select(P2) = First(P2) + Follow(S) = { "$" }

      First(P7) = { "non-core" }
      First(P8) = { }
      Preselected(P7) = Preselected(P8) = Empty(P7) = false
      Empty(P8) = true
      Follow(after) = { "non-core", "$" }
      Select(P7) = First(P7) = { "non-core" }
      Select(P8) = First(P8) + Follow(after) = { "non-core", "$" }

   The intersection of Select(P7) and Select(P8) is not empty; hence,
   the grammar is not deterministic, and the type definition is not
   valid.  There is ambiguity between the end of one item and the start
   of the next.  Without looking ahead in an encoding, it is not
   possible to determine whether a <non-core> element belongs with the
   preceding or following <core> element.

A.10.  Example 10

   Consider the following type definition:

      CHOICE {
          one   [GROUP] List,
          two   [GROUP] SEQUENCE {
              three  [ATTRIBUTE] UTF8String,
              four   [GROUP] List
          }
      }

      List ::= SEQUENCE OF string UTF8String

   The associated grammar is:

      P1:  S ::= one
      P2:  S ::= two
      P3:  one ::= string one
      P4:  one ::=
      P5:  two ::= three four
      P6:  three ::= "@three"
      P7:  four ::= string four



Legg                          Experimental                     [Page 71]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      P8:  four ::=
      P9:  string ::= "string"

   This grammar leads to the following sets and predicates:

      First(P1) = { "string" }
      First(P2) = { "string" }
      Preselected(P1) = Empty(P2) = false
      Preselected(P2) = Empty(P1) = true
      Follow(S) = { "$" }
      Select(P1) = First(P1) + Follow(S) = { "string", "$" }
      Select(P2) = { }

      First(P3) = { "string" }
      First(P4) = { }
      Preselected(P3) = Preselected(P4) = Empty(P3) = false
      Empty(P4) = true
      Follow(one) = { "$" }
      Select(P3) = First(P3) = { "string" }
      Select(P4) = First(P4) + Follow(one) = { "$" }

      First(P7) = { "string" }
      First(P8) = { }
      Preselected(P7) = Preselected(P8) = Empty(P7) = false
      Empty(P8) = true
      Follow(four) = { "$" }
      Select(P7) = First(P7) = { "string" }
      Select(P8) = First(P8) + Follow(four) = { "$" }

   The intersection of Select(P1) and Select(P2) is empty, as is the
   intersection of Select(P3) and Select(P4) and the intersection of
   Select(P7) and Select(P8); hence, the grammar is deterministic, and
   the type definition is valid.  Although both alternatives of the
   CHOICE can begin with a <string> element, an RXER decoder would use
   the presence of a "three" attribute to decide whether to select or
   disregard the "two" alternative.

   However, an attribute in an extension cannot be used to select
   between alternatives.  Consider the following type definition:












Legg                          Experimental                     [Page 72]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      [SINGULAR-INSERTIONS] CHOICE {
          one   [GROUP] List,
          ...,
          two   [GROUP] SEQUENCE {
              three  [ATTRIBUTE] UTF8String,
              four   [GROUP] List
          } -- ExtensionAdditionAlternative (E1).
          -- The extension insertion point is here (I1).
      }

      List ::= SEQUENCE OF string UTF8String

   The associated grammar is:

      P1:  S ::= one
      P10: S ::= E1
      P11: S ::= "*"
      P12: E1 ::= two
      P3:  one ::= string one
      P4:  one ::=
      P5:  two ::= three four
      P6:  three ::= "@three"
      P7:  four ::= string four
      P8:  four ::=
      P9:  string ::= "string"

   This grammar leads to the following sets and predicates for P1, P10
   and P11:

      First(P1) = { "string" }
      First(P10) = { "string" }
      First(P11) = { "*" }
      Preselected(P1) = Preselected(P10) = Preselected(P11) = false
      Empty(P10) = Empty(P11) = false
      Empty(P1) = true
      Follow(S) = { "$" }
      Select(P1) = First(P1) + Follow(S) = { "string", "$" }
      Select(P10) = First(P10) = { "string" }
      Select(P11) = First(P11) = { "*" }

   Preselected(P10) evaluates to false because Preselected(P10) is
   evaluated on the base grammar, wherein P10 is rewritten as:

      P10: S ::=







Legg                          Experimental                     [Page 73]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The intersection of Select(P1) and Select(P10) is not empty; hence,
   the grammar is not deterministic, and the type definition is not
   valid.  An RXER decoder using the original, unextended version of the
   definition would not know that the "three" attribute selects between
   the "one" alternative and the extension.

Appendix B.  Insertion Encoding Instruction Examples

   This appendix is non-normative.

   This appendix contains examples showing the use of insertion encoding
   instructions to remove extension ambiguity arising from use of the
   GROUP encoding instruction.

B.1.  Example 1

   Consider the following type definition:

      SEQUENCE {
          one    [GROUP] SEQUENCE {
              two    UTF8String,
              ... -- Extension insertion point (I1).
          },
          three  INTEGER OPTIONAL,
          ... -- Extension insertion point (I2).
      }

   The associated grammar is:

      P1:  S ::= one three I2
      P2:  one ::= two I1
      P3:  two ::= "two"
      P4:  I1 ::= "*" I1
      P5:  I1 ::=
      P6:  three ::= "three"
      P7:  three ::=
      P8:  I2 ::= "*" I2
      P9:  I2 ::=

   This grammar leads to the following sets and predicates:

      First(P4) = { "*" }
      First(P5) = { }
      Preselected(P4) = Preselected(P5) = Empty(P4) = false
      Empty(P5) = true
      Follow(I1) = { "three", "*", "$" }
      Select(P4) = First(P4) = { "*" }
      Select(P5) = First(P5) + Follow(I1) = { "three", "*", "$" }



Legg                          Experimental                     [Page 74]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      First(P6) = { "three" }
      First(P7) = { }
      Preselected(P6) = Preselected(P7) = Empty(P6) = false
      Empty(P7) = true
      Follow(three) = { "*", "$" }
      Select(P6) = First(P6) = { "three" }
      Select(P7) = First(P7) + Follow(three) = { "*", "$" }

      First(P8) = { "*" }
      First(P9) = { }
      Preselected(P8) = Preselected(P9) = Empty(P8) = false
      Empty(P9) = true
      Follow(I2) = { "$" }
      Select(P8) = First(P8) = { "*" }
      Select(P9) = First(P9) + Follow(I2) = { "$" }

   The intersection of Select(P4) and Select(P5) is not empty; hence,
   the grammar is not deterministic, and the type definition is not
   valid.  If an RXER decoder encounters an unrecognized element
   immediately after a <two> element, then it will not know whether to
   associate it with extension insertion point I1 or I2.

   The non-determinism can be resolved with either a NO-INSERTIONS or
   HOLLOW-INSERTIONS encoding instruction.  Consider this revised type
   definition:

      SEQUENCE {
          one    [GROUP] [HOLLOW-INSERTIONS] SEQUENCE {
              two    UTF8String,
              ... -- Extension insertion point (I1).
          },
          three  INTEGER OPTIONAL,
          ... -- Extension insertion point (I2).
      }

   The associated grammar is:

      P1:  S ::= one three I2
      P10: one ::= two
      P3:  two ::= "two"
      P6:  three ::= "three"
      P7:  three ::=
      P8:  I2 ::= "*" I2
      P9:  I2 ::=

   With the addition of the HOLLOW-INSERTIONS encoding instruction, the
   P4 and P5 productions are no longer generated, and the conflict
   between Select(P4) and Select(P5) no longer exists.  The Select Sets



Legg                          Experimental                     [Page 75]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   for P6, P7, P8, and P9 are unchanged.  A decoder will now assume that
   an unrecognized element is to be associated with extension insertion
   point I2.  It is still free to associate an unrecognized attribute
   with either extension insertion point.  If a NO-INSERTIONS encoding
   instruction had been used, then an unrecognized attribute could only
   be associated with extension insertion point I2.

   The non-determinism could also be resolved by adding a NO-INSERTIONS
   or HOLLOW-INSERTIONS encoding instruction to the outer SEQUENCE:

      [HOLLOW-INSERTIONS] SEQUENCE {
          one    [GROUP] SEQUENCE {
              two    UTF8String,
              ... -- Extension insertion point (I1).
          },
          three  INTEGER OPTIONAL,
          ... -- Extension insertion point (I2).
      }

   The associated grammar is:

      P11: S ::= one three
      P2:  one ::= two I1
      P3:  two ::= "two"
      P4:  I1 ::= "*" I1
      P5:  I1 ::=
      P6:  three ::= "three"
      P7:  three ::=

   This grammar leads to the following sets and predicates:

      First(P4) = { "*" }
      First(P5) = { }
      Preselected(P4) = Preselected(P5) = Empty(P4) = false
      Empty(P5) = true
      Follow(I1) = { "three", "$" }
      Select(P4) = First(P4) = { "*" }
      Select(P5) = First(P5) + Follow(I1) = { "three", "$" }

      First(P6) = { "three" }
      First(P7) = { }
      Preselected(P6) = Preselected(P7) = Empty(P6) = false
      Empty(P7) = true
      Follow(three) = { "$" }
      Select(P6) = First(P6) = { "three" }
      Select(P7) = First(P7) + Follow(three) = { "$" }





Legg                          Experimental                     [Page 76]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The intersection of Select(P4) and Select(P5) is empty, as is the
   intersection of Select(P6) and Select(P7); hence, the grammar is
   deterministic, and the type definition is valid.  A decoder will now
   assume that an unrecognized element is to be associated with
   extension insertion point I1.  It is still free to associate an
   unrecognized attribute with either extension insertion point.  If a
   NO-INSERTIONS encoding instruction had been used, then an
   unrecognized attribute could only be associated with extension
   insertion point I1.

B.2.  Example 2

   Consider the following type definition:

      SEQUENCE {
          one  [GROUP] CHOICE {
              two  UTF8String,
              ... -- Extension insertion point (I1).
          } OPTIONAL
      }

   The associated grammar is:

      P1:  S ::= one
      P2:  one ::= two
      P3:  one ::= I1
      P4:  one ::=
      P5:  two ::= "two"
      P6:  I1 ::= "*" I1
      P7:  I1 ::=

   This grammar leads to the following sets and predicates:

      First(P2) = { "two" }
      First(P3) = { "*" }
      First(P4) = { }
      Preselected(P2) = Preselected(P3) = Preselected(P4) = false
      Empty(P2) = false
      Empty(P3) = Empty(P4) = true
      Follow(one) = { "$" }
      Select(P2) = First(P2) = { "two" }
      Select(P3) = First(P3) + Follow(one) = { "*", "$" }
      Select(P4) = First(P4) + Follow(one) = { "$" }

      First(P6) = { "*" }
      First(P7) = { }
      Preselected(P6) = Preselected(P7) = Empty(P6) = false
      Empty(P7) = true



Legg                          Experimental                     [Page 77]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      Follow(I1) = { "$" }
      Select(P6) = First(P6) = { "*" }
      Select(P7) = First(P7) + Follow(I1) = { "$" }

   The intersection of Select(P3) and Select(P4) is not empty; hence,
   the grammar is not deterministic, and the type definition is not
   valid.  If the <two> element is not present, then a decoder cannot
   determine whether the "one" alternative is absent, or present with an
   unknown extension that generates no elements.

   The non-determinism can be resolved with either a
   SINGULAR-INSERTIONS, UNIFORM-INSERTIONS, or MULTIFORM-INSERTIONS
   encoding instruction.  The MULTIFORM-INSERTIONS encoding instruction
   is the least restrictive.  Consider this revised type definition:

      SEQUENCE {
          one  [GROUP] [MULTIFORM-INSERTIONS] CHOICE {
              two  UTF8String,
              ... -- Extension insertion point (I1).
          } OPTIONAL
      }

   The associated grammar is:

      P1:  S ::= one
      P2:  one ::= two
      P8:  one ::= "*" I1
      P4:  one ::=
      P5:  two ::= "two"
      P6:  I1 ::= "*" I1
      P7:  I1 ::=

   This grammar leads to the following sets and predicates:

      First(P2) = { "two" }
      First(P8) = { "*" }
      First(P4) = { }
      Preselected(P2) = Preselected(P8) = Preselected(P4) = false
      Empty(P2) = Empty(P8) = false
      Empty(P4) = true
      Follow(one) = { "$" }
      Select(P2) = First(P2) = { "two" }
      Select(P8) = First(P8) = { "*" }
      Select(P4) = First(P4) + Follow(one) = { "$" }

      First(P6) = { "*" }
      First(P7) = { }
      Preselected(P6) = Preselected(P7) = Empty(P6) = false



Legg                          Experimental                     [Page 78]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      Empty(P7) = true
      Follow(I1) = { "$" }
      Select(P6) = First(P6) = { "*" }
      Select(P7) = First(P7) + Follow(I1) = { "$" }

   The intersection of Select(P2) and Select(P8) is empty, as is the
   intersection of Select(P2) and Select(P4), the intersection of
   Select(P8) and Select(P4), and the intersection of Select(P6) and
   Select(P7); hence, the grammar is deterministic, and the type
   definition is valid.  A decoder will now assume the "one" alternative
   is present if it sees at least one unrecognized element, and absent
   otherwise.

B.3.  Example 3

   Consider the following type definition:

      SEQUENCE {
          one    [GROUP] CHOICE {
              two    UTF8String,
              ... -- Extension insertion point (I1).
          },
          three  [GROUP] CHOICE {
              four   UTF8String,
              ... -- Extension insertion point (I2).
          }
      }

   The associated grammar is:

      P1:  S ::= one three
      P2:  one ::= two
      P3:  one ::= I1
      P4:  two ::= "two"
      P5:  I1 ::= "*" I1
      P6:  I1 ::=
      P7:  three ::= four
      P8:  three ::= I2
      P9:  four ::= "four"
      P10: I2 ::= "*" I2
      P11: I2 ::=

   This grammar leads to the following sets and predicates:

      First(P2) = { "two" }
      First(P3) = { "*" }
      Preselected(P2) = Preselected(P3) = Empty(P2) = false
      Empty(P3) = true



Legg                          Experimental                     [Page 79]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      Follow(one) = { "four", "*", "$" }
      Select(P2) = First(P2) = { "two" }
      Select(P3) = First(P3) + Follow(one) = { "*", "four", "$" }

      First(P5) = { "*" }
      First(P6) = { }
      Preselected(P5) = Preselected(P6) = Empty(P5) = false
      Empty(P6) = true
      Follow(I1) = { "four", "*", "$" }
      Select(P5) = First(P5) = { "*" }
      Select(P6) = First(P6) + Follow(I1) = { "four", "*", "$" }

      First(P7) = { "four" }
      First(P8) = { "*" }
      Preselected(P7) = Preselected(P8) = Empty(P7) = false
      Empty(P8) = true
      Follow(three) = { "$" }
      Select(P7) = First(P7) = { "four" }
      Select(P8) = First(P8) + Follow(three) = { "*", "$" }

      First(P10) = { "*" }
      First(P11) = { }
      Preselected(P10) = Preselected(P11) = Empty(P10) = false
      Empty(P11) = true
      Follow(I2) = { "$" }
      Select(P10) = First(P10) = { "*" }
      Select(P11) = First(P11) + Follow(I2) = { "$" }

   The intersection of Select(P5) and Select(P6) is not empty; hence,
   the grammar is not deterministic, and the type definition is not
   valid.  If the first child element is an unrecognized element, then a
   decoder cannot determine whether to associate it with extension
   insertion point I1, or to associate it with extension insertion point
   I2 by assuming that the "one" component has an unknown extension that
   generates no elements.

   The non-determinism can be resolved with either a SINGULAR-INSERTIONS
   or UNIFORM-INSERTIONS encoding instruction.  Consider this revised
   type definition using the SINGULAR-INSERTIONS encoding instruction:












Legg                          Experimental                     [Page 80]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      SEQUENCE {
          one    [GROUP] [SINGULAR-INSERTIONS] CHOICE {
              two    UTF8String,
              ... -- Extension insertion point (I1).
          },
          three  [GROUP] CHOICE {
              four   UTF8String,
              ... -- Extension insertion point (I2).
          }
      }

   The associated grammar is:

      P1:  S ::= one three
      P2:  one ::= two
      P12: one ::= "*"
      P4:  two ::= "two"
      P7:  three ::= four
      P8:  three ::= I2
      P9:  four ::= "four"
      P10: I2 ::= "*" I2
      P11: I2 ::=

   With the addition of the SINGULAR-INSERTIONS encoding instruction,
   the P5 and P6 productions are no longer generated.  The grammar leads
   to the following sets and predicates for the P2 and P12 productions:

      First(P2) = { "two" }
      First(P12) = { "*" }
      Preselected(P2) = Preselected(P12) = false
      Empty(P2) = Empty(P12) = false
      Follow(one) = { "four", "*", "$" }
      Select(P2) = First(P2) = { "two" }
      Select(P12) = First(P12) = { "*" }

   The sets for P5 and P6 are no longer generated, and the remaining
   sets are unchanged.

   The intersection of Select(P2) and Select(P12) is empty, as is the
   intersection of Select(P7) and Select(P8) and the intersection of
   Select(P10) and Select(P11); hence, the grammar is deterministic, and
   the type definition is valid.  If the first child element is an
   unrecognized element, then a decoder will now assume that it is
   associated with extension insertion point I1.  Whatever follows,
   possibly including another unrecognized element, will belong to the
   "three" component.





Legg                          Experimental                     [Page 81]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   Now consider the type definition using the UNIFORM-INSERTIONS
   encoding instruction instead:

      SEQUENCE {
          one    [GROUP] [UNIFORM-INSERTIONS] CHOICE {
              two    UTF8String,
              ... -- Extension insertion point (I1).
          },
          three  [GROUP] CHOICE {
              four   UTF8String,
              ... -- Extension insertion point (I2).
          }
      }

   The associated grammar is:

      P1:  S ::= one three
      P2:  one ::= two
      P13: one ::= "*"
      P14: one ::= "*1" I1
      P4:  two ::= "two"
      P15: I1 ::= "*1" I1
      P6:  I1 ::=
      P7:  three ::= four
      P8:  three ::= I2
      P9:  four ::= "four"
      P10: I2 ::= "*" I2
      P11: I2 ::=

   This grammar leads to the following sets and predicates for the P2,
   P13, P14, P15, and P6 productions:

      First(P2) = { "two" }
      First(P13) = { "*" }
      First(P14) = { "*1" }
      Preselected(P2) = Preselected(P13) = Preselected(P14) = false
      Empty(P2) = Empty(P13) = Empty(P14) = false
      Follow(one) = { "four", "*", "$" }
      Select(P2) = First(P2) = { "two" }
      Select(P13) = First(P13) = { "*" }
      Select(P14) = First(P14) = { "*1" }










Legg                          Experimental                     [Page 82]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      First(P15) = { "*1" }
      First(P6) = { }
      Preselected(P15) = Preselected(P6) = Empty(P15) = false
      Empty(P6) = true
      Follow(I1) = { "four", "*", "$" }
      Select(P15) = First(P15) = { "*1" }
      Select(P6) = First(P6) + Follow(I1) = { "four", "*", "$" }

   The remaining sets are unchanged.

   The intersection of Select(P2) and Select(P13) is empty, as is the
   intersection of Select(P2) and Select(P14), the intersection of
   Select(P13) and Select(P14) and the intersection of Select(P15) and
   Select(P6); hence, the grammar is deterministic, and the type
   definition is valid.  If the first child element is an unrecognized
   element, then a decoder will now assume that it and every subsequent
   unrecognized element with the same name are associated with I1.
   Whatever follows, possibly including another unrecognized element
   with a different name, will belong to the "three" component.

   A consequence of using the UNIFORM-INSERTIONS encoding instruction is
   that any future extension to the "three" component will be required
   to generate elements with names that are different from the names of
   the elements generated by the "one" component.  With the
   SINGULAR-INSERTIONS encoding instruction, extensions to the "three"
   component are permitted to generate elements with names that are the
   same as the names of the elements generated by the "one" component.

B.4.  Example 4

   Consider the following type definition:

      SEQUENCE OF one [GROUP] CHOICE {
          two    UTF8String,
          ... -- Extension insertion point (I1).
      }

   The associated grammar is:

      P1:  S ::= one S
      P2:  S ::=
      P3:  one ::= two
      P4:  one ::= I1
      P5:  two ::= "two"
      P6:  I1 ::= "*" I1
      P7:  I1 ::=





Legg                          Experimental                     [Page 83]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   This grammar leads to the following sets and predicates:

      First(P1) = { "two", "*" }
      First(P2) = { }
      Preselected(P1) = Preselected(P2) = false
      Empty(P1) = Empty(P2) = true
      Follow(S) = { "$" }
      Select(P1) = First(P1) + Follow(S) = { "two", "*", "$" }
      Select(P2) = First(P2) + Follow(S) = { "$" }

      First(P3) = { "two" }
      First(P4) = { "*" }
      Preselected(P3) = Preselected(P4) = Empty(P3) = false
      Empty(P4) = true
      Follow(one) = { "two", "*", "$" }
      Select(P3) = First(P3) = { "two" }
      Select(P4) = First(P4) + Follow(one) = { "*", "two", "$" }

      First(P6) = { "*" }
      First(P7) = { }
      Preselected(P6) = Preselected(P7) = Empty(P6) = false
      Empty(P7) = true
      Follow(I1) = { "two", "*", "$" }
      Select(P6) = First(P6) = { "*" }
      Select(P7) = First(P7) + Follow(I1) = { "two", "*", "$" }

   The intersection of Select(P1) and Select(P2) is not empty, as is the
   intersection of Select(P3) and Select(P4) and the intersection of
   Select(P6) and Select(P7); hence, the grammar is not deterministic,
   and the type definition is not valid.  If a decoder encounters two or
   more unrecognized elements in a row, then it cannot determine whether
   this represents one instance or more than one instance of the "one"
   component.  Even without unrecognized elements, there is still a
   problem that an encoding could contain an indeterminate number of
   "one" components using an extension that generates no elements.

   The non-determinism cannot be resolved with a UNIFORM-INSERTIONS
   encoding instruction.  Consider this revised type definition using
   the UNIFORM-INSERTIONS encoding instruction:

      SEQUENCE OF one [GROUP] [UNIFORM-INSERTIONS] CHOICE {
          two    UTF8String,
          ... -- Extension insertion point (I1).
      }







Legg                          Experimental                     [Page 84]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The associated grammar is:

      P1:  S ::= one S
      P2:  S ::=
      P3:  one ::= two
      P8:  one ::= "*"
      P9:  one ::= "*1" I1
      P5:  two ::= "two"
      P10: I1 ::= "*1" I1
      P7:  I1 ::=

   This grammar leads to the following sets and predicates:

      First(P1) = { "two", "*", "*1" }
      First(P2) = { }
      Preselected(P1) = Preselected(P2) = Empty(P1) = false
      Empty(P2) = true
      Follow(S) = { "$" }
      Select(P1) = First(P1) = { "two", "*", "*1" }
      Select(P2) = First(P2) + Follow(S) = { "$" }

      First(P3) = { "two" }
      First(P8) = { "*" }
      First(P9) = { "*1" }
      Preselected(P3) = Preselected(P8) = Preselected(P9) = false
      Empty(P3) = Empty(P8) = Empty(P9) = false
      Follow(one) = { "two", "*", "*1", "$" }
      Select(P3) = First(P3) = { "two" }
      Select(P8) = First(P8) = { "*" }
      Select(P9) = First(P9) = { "*1" }

      First(P10) = { "*1" }
      First(P7) = { }
      Preselected(P10) = Preselected(P7) = Empty(P10) = false
      Empty(P7) = true
      Follow(I1) = { "two", "*", "*1", "$" }
      Select(P10) = First(P10) = { "*1" }
      Select(P7) = First(P7) + Follow(I1) = { "two", "*", "*1", "$" }

   The intersection of Select(P1) and Select(P2) is now empty, but the
   intersection of Select(P10) and Select(P7) is not; hence, the grammar
   is not deterministic, and the type definition is not valid.  The
   problem of an indeterminate number of "one" components from an
   extension that generates no elements has been solved.  However, if a
   decoder encounters a series of elements with the same name, it cannot
   determine whether this represents one instance or more than one
   instance of the "one" component.




Legg                          Experimental                     [Page 85]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The non-determinism can be fully resolved with a SINGULAR-INSERTIONS
   encoding instruction.  Consider this revised type definition:

      SEQUENCE OF one [GROUP] [SINGULAR-INSERTIONS] CHOICE {
          two    UTF8String,
          ... -- Extension insertion point (I1).
      }

   The associated grammar is:

      P1:  S ::= one S
      P2:  S ::=
      P3:  one ::= two
      P8:  one ::= "*"
      P5:  two ::= "two"

   This grammar leads to the following sets and predicates:

      First(P1) = { "two", "*" }
      First(P2) = { }
      Preselected(P1) = Preselected(P2) = Empty(P1) = false
      Empty(P2) = true
      Follow(S) = { "$" }
      Select(P1) = First(P1) = { "two", "*" }
      Select(P2) = First(P2) + Follow(S) = { "$" }

      First(P3) = { "two" }
      First(P8) = { "*" }
      Preselected(P3) = Preselected(P8) = false
      Empty(P3) = Empty(P8) = false
      Follow(one) = { "two", "*", "$" }
      Select(P3) = First(P3) = { "two" }
      Select(P8) = First(P8) = { "*" }

   The intersection of Select(P1) and Select(P2) is empty, as is the
   intersection of Select(P3) and Select(P8); hence, the grammar is
   deterministic, and the type definition is valid.  A decoder now knows
   that every extension to the "one" component will generate a single
   element, so the correct number of "one" components will be decoded.












Legg                          Experimental                     [Page 86]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


Appendix C.  Extension and Versioning Examples

   This appendix is non-normative.

C.1.  Valid Extensions for Insertion Encoding Instructions

   The first example shows extensions that satisfy the HOLLOW-INSERTIONS
   encoding instruction.

      [HOLLOW-INSERTIONS] CHOICE {
          one    BOOLEAN,
          ...,
          two    [ATTRIBUTE] INTEGER,
          three  [GROUP] SEQUENCE {
              four  [ATTRIBUTE] UTF8String,
              five  [ATTRIBUTE] INTEGER OPTIONAL,
              ...
          },
          six    [GROUP] CHOICE {
              seven  [ATTRIBUTE] BOOLEAN,
              eight  [ATTRIBUTE] INTEGER
          }
      }

   The "two" and "six" components generate only attributes.

   The "three" component in its current form does not generate elements.
   Any extension to the "three" component will need to do likewise to
   avoid breaking forward compatibility.

   The second example shows extensions that satisfy the
   SINGULAR-INSERTIONS encoding instruction.

      [SINGULAR-INSERTIONS] CHOICE {
          one    BOOLEAN,
          ...,
          two    INTEGER,
          three  [GROUP] SEQUENCE {
              four   [ATTRIBUTE] UTF8String,
              five   INTEGER
          },
          six    [GROUP] CHOICE {
              seven  BOOLEAN,
              eight  INTEGER
          }
      }





Legg                          Experimental                     [Page 87]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


   The "two" component will always generate a single <two> element.

   The "three" component will always generate a single <five> element.
   It will also generate a "four" attribute, but any number of
   attributes is allowed by the SINGULAR-INSERTIONS encoding
   instruction.

   The "six" component will either generate a single <seven> element or
   a single <eight> element.  Either case will satisfy the requirement
   that there will be a single element in any given encoding of the
   extension.

   The third example shows extensions that satisfy the
   UNIFORM-INSERTIONS encoding instruction.

      [UNIFORM-INSERTIONS] CHOICE {
          one    BOOLEAN,
          ...,
          two    INTEGER,
          three  [GROUP] SEQUENCE SIZE(1..MAX) OF four INTEGER,
          five   [GROUP] SEQUENCE {
              six    [ATTRIBUTE] UTF8String OPTIONAL,
              seven  INTEGER
          },
          eight  [GROUP] CHOICE {
              nine   BOOLEAN,
              ten    [GROUP] SEQUENCE SIZE(1..MAX) OF eleven INTEGER
          }
      }

   The "two" component will always generate a single <two> element.

   The "three" component will always generate one or more <four>
   elements.

   The "five" component will always generate a single <seven> element.
   It may also generate a "six" attribute, but any number of attributes
   is allowed by the UNIFORM-INSERTIONS encoding instruction.

   The "eight" component will either generate a single <nine> element or
   one or more <eleven> elements.  Either case will satisfy the
   requirement that there must be one or more elements with the same
   name in any given encoding of the extension.








Legg                          Experimental                     [Page 88]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


C.2.  Versioning Example

   Making extensions that are not forward compatible is permitted
   provided that the incompatibility is signalled with a version
   indicator attribute.

   Suppose that version 1.0 of a specification contains the following
   type definition:

      MyMessageType ::= SEQUENCE {
         version  [ATTRIBUTE] [VERSION-INDICATOR]
                      UTF8String ("1.0", ...) DEFAULT "1.0",
         one      [GROUP] [SINGULAR-INSERTIONS] CHOICE {
             two  BOOLEAN,
             ...
         },
         ...
      }

   An attribute is to be added to the CHOICE for version 1.1.  This
   change is not forward compatible since it does not satisfy the
   SINGULAR-INSERTIONS encoding instruction.  Therefore, the version
   indicator attribute must be updated at the same time (or added if it
   wasn't already present).  This results in the following new type
   definition for version 1.1:

      MyMessageType ::= SEQUENCE {
         version  [ATTRIBUTE] [VERSION-INDICATOR]
                      UTF8String ("1.0", ..., "1.1") DEFAULT "1.0",
         one      [GROUP] [SINGULAR-INSERTIONS] CHOICE {
             two    BOOLEAN,
             ...,
             three  [ATTRIBUTE] INTEGER -- Added in Version 1.1
         },
         ...
      }

   If a version 1.1 conformant application hasn't used the version 1.1
   extension in a value of MyMessageType, then it is allowed to set the
   value of the version attribute to "1.0".

   A pair of elements is added to the CHOICE for version 1.2.  Again the
   change does not satisfy the SINGULAR-INSERTIONS encoding instruction.
   The type definition for version 1.2 is:







Legg                          Experimental                     [Page 89]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


      MyMessageType ::= SEQUENCE {
         version  [ATTRIBUTE] [VERSION-INDICATOR]
                      UTF8String ("1.0", ..., "1.1" | "1.2")
                          DEFAULT "1.0",
         one      [GROUP] [SINGULAR-INSERTIONS] CHOICE {
             two    BOOLEAN,
             ...,
             three  [ATTRIBUTE] INTEGER, -- Added in Version 1.1
             four   [GROUP] SEQUENCE {
                 five  UTF8String,
                 six   GeneralizedTime
             } -- Added in version 1.2
         },
         ...
      }

   If a version 1.2 conformant application hasn't used the version 1.2
   extension in a value of MyMessageType, then it is allowed to set the
   value of the version attribute to "1.1".  If it hasn't used either of
   the extensions, then it is allowed to set the value of the version
   attribute to "1.0".

Author's Address

   Dr. Steven Legg
   eB2Bcom
   Suite 3, Woodhouse Corporate Centre
   935 Station Street
   Box Hill North, Victoria 3129
   AUSTRALIA

   Phone: +61 3 9896 7830
   Fax:   +61 3 9896 7801
   EMail: steven.legg@eb2bcom.com

















Legg                          Experimental                     [Page 90]
^L
RFC 4911             Encoding Instructions for RXER            July 2007


Full Copyright Statement

   Copyright (C) The IETF Trust (2007).

   This document is subject to the rights, licenses and restrictions
   contained in BCP 78, and except as set forth therein, the authors
   retain all their rights.

   This document and the information contained herein are provided on an
   "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS
   OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE IETF TRUST AND
   THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS
   OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF
   THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED
   WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Intellectual Property

   The IETF takes no position regarding the validity or scope of any
   Intellectual Property Rights or other rights that might be claimed to
   pertain to the implementation or use of the technology described in
   this document or the extent to which any license under such rights
   might or might not be available; nor does it represent that it has
   made any independent effort to identify any such rights.  Information
   on the procedures with respect to rights in RFC documents can be
   found in BCP 78 and BCP 79.

   Copies of IPR disclosures made to the IETF Secretariat and any
   assurances of licenses to be made available, or the result of an
   attempt made to obtain a general license or permission for the use of
   such proprietary rights by implementers or users of this
   specification can be obtained from the IETF on-line IPR repository at
   http://www.ietf.org/ipr.

   The IETF invites any interested party to bring to its attention any
   copyrights, patents or patent applications, or other proprietary
   rights that may cover technology that may be required to implement
   this standard.  Please address the information to the IETF at
   ietf-ipr@ietf.org.

Acknowledgement

   Funding for the RFC Editor function is currently provided by the
   Internet Society.







Legg                          Experimental                     [Page 91]
^L