Go to the previous, next section.

Protocols and Transports

When two modules of a program are in different address spaces, or use different data representations, ILU forms messages to send across the inter-module boundary; we call a particular way of forming and interpreting these messages an RPC protocol (sometimes simply protocol). These messages may be transported between address spaces in different ways; we call a particular way of moving messages a transport. This chapter describes the various kinds of available ILU protocols and transports. ILU is extensible: additional RPC protocols and transports can be added, either at compile-time or run-time; this chapter does not describe how to do so.

When an ILU kernel server exports objects, it does so via one or more contact stacks. Each stack has an RPC protocol at the top of the stack, forming and interpreting messages, and one or more layers of transport below the protocol layer, transforming or communicating the messages in various ways. A contact stack is specified by a protocol-info string and a sequence of transport-info strings; the syntax of these strings is defined in this chapter.

Protocols

The Abstract ILU Message Protocol

Before describing any particular protocol, we will describe the abstract ILU protocol, which is layered on top of each actual protocol. It is quite simple. Two types of messages are used, one to communicate parameters to a true method, and the other to communicate results and/or exceptions from the true method to surrogate caller. Parameters and values are encoded according to a simple abstract external data representation format. This abstract protocol identifies what information is passed between modules without specifying its exact mapping to bit patterns.

Messages

The first type of message is called a request. Each request consists of a code identifying the method being requested, an authentication block identifying the principal making the call, and a list of parameter inputs to the method being called. The method is identified by passing the one-based ordinal value (that is, the index of the method in the list of methods, beginning with one) of the method, in the list of methods as specified in the ISL description of the class which actually defines the method. No more than 65278 (1-0xFEFF) methods may be directly specified for any type (though more methods may be inherited by a type). Method codes 0xFF00 to 0xFFFF are reserved for ILU internal use. The principal is identified by a block of authentication credentials information which varies depending on the specific authentication protocol used. These credentials may be either in the request header, or may appear as a parameter of the request. (Note: There should also be an ILU protocol version number somewhere here, but there isn't (yet).)

The result message is used to convey return values and exception values from the true method back to the caller. It consists of a Boolean value, indicating whether the call was successful (for TRUE) or signalled an exception (for FALSE). If successful, the return value (if any), follows, followed by the values of any Out parameters, in the order they are specified as parameters. If an exception was signalled, a value between 1 and 2^16-1 follows, indicating the ordinal value specific exception in the list specified in the definition of the method, followed by a value of the exception type, if any was specified for the exception.

Parameter Types

Simple numerical values, of types integer, cardinal, real, or byte, are passed directly.

Character values are passed as integer values in the range [0,2^16-1]. Short character values are passed as integer values in the range [0,2^16-1]. Long character values are passed as integer values in the range [0,2^32-1].

Enumeration values are passed as integer values in the range [0,2^16-1], the value being the zero-based ordinal value of the corresponding enumeration value in the original list of enumeration values in the definition of the enumerated type.

Boolean values are passed as as integer values of either 0, for FALSE, or 1, for TRUE.

Optional values are passed by first passing a Boolean value, with TRUE indicating that a non-NIL value is being passed, and then only in the non-NIL case passing a value of the optional value's indicated type.

Sequence values are passed by first passing a count, as an integer in the range [0,2^32-1] for sequences without limits, or for sequences with limits greater than 2^16-1, or an integer in the range [0,2^16-1], for sequences with limits less than 2^16, indicating the number of elements in the sequence, and then that number of values of the sequence's base type.

Array values are passed by passing a number of elements of the array's base type corresponding to the size of the array.

Record values are passed by passing values of types corresponding to the fields of the record, following the order in which the fields are defined in the ISL definition of the record.

Union values are passed by passing a value of the discriminant type, which indicates which branch of the union constitutes the union's actual type, usually followed by a value of the union's actual type. If the discriminant value indicates a branch of the union which has no associate value, only the discriminant value is passed.

Object values are passed in several different forms, depending on whether or not the object value is in the discriminator position, whether or not the object's type is a singleton type, and whether or not the object reference is NIL.

  1. The first form is used when the object is in the discriminator position (that is, is the instance upon which the method is being invoked), and is an instance of a singleton type. In this case, the object is already known to both sides, and the object is passed implicitly; that is, no actual bytes are transmitted.
  2. The second form is used when the object is in the discriminator position, but is not of a singleton type. In this case, the CRC-32 of the server ID of the object is passed as a cardinal value, followed by the instance handle of the object, as a sequence of short character value. Both the instance handle and server ID must be passed, as the true object previously at the "known address" for the object may have been replaced by a different object with the same instance handle, in a different kernel server.
  3. In the third case, the object is being passed as a normal parameter, that is, not in the discriminator position. In this case, the full string binding handle of the object is passed as a sequence of short character value.
  4. If the object being passed as a normal parameter is the CORBA Nil object reference, it is passed as the sequence of short character value of length zero.

The ONC RPC Protocol

This section describes the mapping of the abstract ILU protocol into the specific on-the-wire protocol used with ONC RPC(10) One of the major goals of this mapping is to preserve compatibility with existing Sun RPC services that can be described in ISL.

A protocol info string for ONC RPC has the form sunrpc_2_program-number_program-version where program-number and program-version may be specified either in decimal, or in hexidecimal with a leading string of 0x. The program number for non-native-ONC-RPC ILU object types is always the same (in ILU 2.0alpha1 to 2.0alpha7, 0x61A78; in ILU 2.0alpha8 and up, 0x61A79), and the program version varies depending on the specific object type.

Use of ONC RPC requires use of a boundaried transport below it.

Message Mappings

The request message used is that specified by the ONC RPC protocol. The ILU method index is encoded as a 32-bit number in the "proc" field in the ONC RPC request header. Principal identification is passed in the "cred" field of the ONC RPC request header. By default, ILU will pass the AUTH_UNIX authentication information, if no authentication method is specified for the method. (This default authentication can be disabled by defining the environment variable ILU_NO_SUNRPC_UNIX_AUTH to any value.) For non-singleton object types, the ONC RPC program number passed in the "prog" slot is always the same (for ILU 2.0alpha1 - 2.0alpha7, 0x00061a78; for ILU 2.0alpha8, 0x00061a79), and the version number passed in the "vers" slot is the CRC-32 hashed value of the MSTID for the object type on which the method being invoked is defined. For singleton classes, the program number and version specified in the singleton information is used. The "mtype" field is set to CALL. The indicated "rpcvers" is 2. A monotonically increasing 32-bit serial number is used in the "xid" field. For non-singleton, non-NIL objects, an extra argument identifying the discriminant of the message (the object on which the method is being invoked) is marshalled before any of the specified arguments. This discriminant is marshalled as an XDR Unsigned Integer, which is the CRC-32 of the server ID of the object, followed by an XDR string, which is the plain instance handle of the object.

The reply message used is that specified by the ONC RPC protocol. The "mtype" field is set to REPLY. The "stat" field is always set to MSG_ACCEPTED. In the accepted_reply, the authentication verifier is always NULL. The "stat" field may be non-zero, to signal one of a small number of "standard" exceptions, or may be zero. This header is then followed by one of three forms: If a "standard" exception was raised, nothing. If the method has no exceptions, the return values and out parameters (if any). If the method has any exceptions defined, a 32-bit value which specifies either successful completion (a value of 0), or an exception (a value greater than 0, which is the ordinal value of the particular exception being signalled in the list of exceptions specified for this method), followed by either the return value and out parameters (if any), in the case of successful completion, or the exception value (if any), in the case of an exception.

Mapping of Standard Types

The mapping of ILU types into ONC RPC types is accomplished primarily by using the appropriate XDR (11) representation for that type.

Short integer and integer types are represented with the XDR Integer type. Long integer types are represented as an XDR Hyper Integer.

Short cardinal, byte, and cardinal types are represented with the XDR Unsigned Integer type. Long cardinal types are represented as an XDR Unsigned Hyper Integer.

Short real numbers are encoded as XDR Floating-point. Real numbers are encoded as XDR Double-precision Floating-point. Long real numbers are encoded as XDR Fixed-length Opaque data of length 16.

Array values are encoded as XDR Fixed-length Array, except for two special cases. If the array is multi-dimensional, it is encoded as a flat rendering into a single-dimensional array in row-major order (the last specified index varying most rapidly). If the array is of element-type byte or short character, it is encoded as an array of one (in the one-dimensional case) or more (in the greater-than-one dimensional case) values of XDR Opaque Data.

Record values are encoded as XDR Structures.

Union values are encoded as XDR Discriminated Unions, with a discriminant of type "unsigned int" containing the ILU short cardinal discriminant.

Enumeration values are encoded as XDR Unsigned Integer (note that this is different from XDR Enumerations, which are encoded as XDR Integer).

Boolean values are encoded as XDR Unsigned Integer, using the value 0 for FALSE and the value 1 for TRUE.

Sequence values are encoded as XDR Variable-length Arrays, except for several special cases. Sequences of short character are encoded as XDR String, sequences of byte are encoded as XDR Variable-length Opaque Data, and sequences of character are encoded as XDR String, where the string is the UTF-2 encoding of the Unicode characters in the sequence.

Optional values are encoded as an XDR Boolean value, followed by another encoded value, if the Boolean value is TRUE.

Instances of an object type are encoded as either zero (in the case of a method discriminant of a singleton type), or one, values of type XDR String.

The Xerox Courier Protocol

This section describes the mapping of the abstract ILU protocol into the specific on-the-wire protocol used with Xerox Courier(12). One of the major goals of this mapping is to preserve compatibility with existing Xerox Courier services that can be described in ISL. Unfortunately, many if not most important Courier services use bulk data transfer, something that is still only planned for ILU.

A protocol info string for Xerox Courier has the form courier_program-number_program-version where program-number and program-version may be specified either in decimal, or in hexidecimal with a leading string of 0x. The program number for non-singleton ILU object types is always (in ILU 2.0) 0x001yxxxx, where y is currently 1; the specific program number and the program version varies depending on the specific object type.

Use of Xerox Courier requires use of a boundaried transport below it.

Message Mappings -- Courier Layer 3

The request message used is the CallMessageBody specified in section 4.3.1 of the Courier protocol. A monotonically increasing 16-bit serial number is passed in the transactionID field; a 32-bit program number is passed in the programNumber field, a 16-bit number is passed in the versionNumber field; the ILU method index is passed as a 16-bit value in the procedureValue field. The program number is calculated by computing the CRC-32 hash value of the MSTID of the object type on which the method is defined, then forming a program number by using the value 0x0011 for the high-order 16 bits, and the high-order 16 bits of the CRC for the low-order 16 bits of the program number. The version number is the low-order 16 bits of the CRC.

Successful replies are sent using the Courier ReturnMessageBody specified in section 4.3.3 of the Courier specification. The procedureResults field contains the return value, if any, followed by the INOUT and OUT parameter values, if any.

User exceptions are signalled using the AbortMessageBody specified in section 4.3.4 of the Courier specification. The errorValue field contains a value greater than 0, which is the ordinal value of the particular exception being signalled in the list of exceptions specified for this method. The errorArguments field contain the exception value, if any.

System exceptions (of exception type ilu.ProtocolError) are signalled using the RejectMessageBody message of section 4.3.2. The rejectionDetail field of the message contains the ProtocolError detail.

Mapping of Standard Types -- Courier Layer 2

The mapping of ILU types into Courier types is accomplished primarily by using the appropriate Courier Layer 2 representation for that type.

Short integer and integer types are represented with the Courier Integer and Long Integer types. Long integer types are represented as an integer followed by a cardinal.

Short cardinal, byte, and cardinal types are represented with the Courier cardinal, cardinal, and long cardinal types, respectively. Long cardinal types are represented as a big-endian (most significant 16 bits first) Courier array of 4 cardinals.

As the Courier protocol does not have any mapping for floating point values, short real numbers are passed as a Courier long cardinal, real numbers are encoded as a big-endian array of two Courier long cardinal values, and long real numbers are encoded as big-endian array of four Courier long cardinal values.

Array values are encoded as Courier one-dimensional arrays. If the array is multi-dimensional, it is encoded as a flat rendering into a single-dimensional array in row-major order (the last specified index varying most rapidly). If the array is of type byte or short character, the contents of the ILU value are packed into a Courier array of unspecified two values per array element, so that the Courier array is half the length of the actual ILU array.

Record values are encoded as Courier record values.

Union values of union types whose discriminant type can be mapped to a 16-bit value type in the range [0,2^16-1] are passed as Courier choice values. Other unions are passed as a Courier long cardinal, followed by the value of the union's indicated type (if any).

Enumeration values are encoded as Courier enumeration values.

Boolean values are encoded as Courier boolean values.

Sequence values are encoded as Courier sequences, except for several special cases. Sequences of N short characters or bytes are encoded as either a Courier cardinal, for sequences with limits less than 2^16, or long cardinal, for sequences with no limits or limits greater than 2^16-1, value of N, followed by (N+1)/2 values of Courier unspecified, each such value containing two short character or byte values, packed in big-endian order.

Optional values are encoded as an Courier boolean value, followed by another encoded value, if the Boolean value is TRUE.

Instances of an object type are encoded as either zero (in the case of a method discriminant of a singleton type), or one values of ISL short sequence of short character. CORBA Nil object references are represented as a zero-length short sequence of short character.

The OMG Internet Inter-Orb Protocol (IIOP)

This section describes the mapping of the abstract ILU protocol into the specific on-the-wire protocol prescribed by the OMG's CORBA Internet Inter-ORB Protocol (IIOP), version 1.0.

A protocol info string for the IIOP version 1.0, with the ILU-to-IIOP mapping version 1, has the form iiop_1_0_1.

The IIOP may be used on top of either a reliable, boundaried or non-boundaried, transport stack.

Message Mappings -- GIOP

ILU request and reply messages are mapped to GIOP Request and Reply messages fairly directly. The byte order used is that native to the machine on which the message is being formed. A zero-length service context is always sent.

In a Request message, the operation name is the ISL operation name for the method, with all hyphen characters in the operation name changed to underscore characters. The Principal field is always sent as a zero-length field.

The GIOP CancelRequest, LocateRequest, MessageError, and CloseConnection messages are never sent by ILU, though one or more of them may be used in the future. ILU will send GIOP LocateReply messages in response to LocateRequest messages.

Mapping of Standard Types -- GIOP

The mapping of ILU types into IIOP types is accomplished primarily by using the mapping for the corresponding CORBA type.

Short integer and integer types are marshaled as CORBA short and long types. Long integer types are represented as an integer followed by a cardinal.

Short cardinal, byte, and cardinal types are marshaled as the CORBA unsigned short, octet, and unsigned long types, respectively. Long cardinal types are marshalled as two CORBA unsigned long values, and the byte order of the message determines which is marshalled first.

short real numbers are passed as CORBA float values. real numbers are passed as CORBA double values, and long real numbers are encoded as big-endian array of 16 bytes.

Array values are encoded as CORBA array values.

Record values are encoded as CORBA struct values.

Union values are encoded as CORBA union values.

Enumeration values are encoded as CORBA enum values.

Boolean values are encoded as CORBA boolean values.

Sequence values are encoded as CORBA sequence values.

Optional values are encoded as a CORBA sequence of the base type, with an upper limit of one value.

Object values are passed as an IIOP Interoperable Object Reference (IOR), containing at least an Internet Profile. The IOR may also contain an ILU Profile. In the case of the Internet Profile, the object key contains four strings, separated by NUL (zero octet) characters. The first string is always "ilu". The second string is the most specific type ID of the object (in case some intervening ORB decides to re-write the IOR's type_id field). The third string is the server ID of the object's server. The fourth string is the instance handle of the object.

The Hyper Text Transfer Protocol (HTTP)

HTTP in ILU allows an ILU application to interact with an existing Web resource. That is, Web Browser to ILU, ILU to Web Server, and general ILU to ILU over HTTP is possible.

For HTTP interaction with existing web services, an ILU application must be able to not only get an object (a surrogate actually) representing the resource. It must also have some means by which to specify the HTTP headers and entity body that should be sent with the request. Similarly, an ILU server functioning as a HTTP accessible Web resource must be able to set status, header and entity body content.

Arbitrary programmers interpretations of these HTTP components cannot be generally mapped into HTTP. A specific signature is needed for the GET HEAD and POST methods so that the ILU implementation of the HTTP protocol can know how to map arguments into actual HTTP format. In addition, a way is needed to distinguish these methods intended for use with existing Web services from other methods that may happen to have the same name but different signatures.

This need is addressed by defining a specific type of object that has declarations for how an application should structure the arguments / return values for the GET HEAD and POST operations. Any GET HEAD or POST operation invoked on an object that is an instance of this base type (or an instance of a type derived directly or indirectly from that base type) has a particular signature that the ILU protocol implementation knows how to map to HTTP. This type, called Resource in the http interface, is defined in the http.isl file, and any application wishing to supply Web compatible objects should make the objects a direct or derived instance of it. A server for objects acessible via HTTP should be created with the protocol info string http_1_0, and should use the tcp transport. See the httest example for a sample use of HTTP in ILU.

A method named GET HEAD or POST, invoked on an object that is a direct or indirect instance of the Resource type, automatically has its Request and Response mapped to/from HTTP in a manner compatible with existing Web services. The fairly straightforward mapping from the ILU http Interface to HTTP Protocol is outlined described below:


ILU Method Name                       Method name in Request's Request line

(if using a Proxy server, scheme + 
location of object +) ILU Object ID +
any params/queries present in the 
Request.URI field                     Request-URI in Request's Request line

Request.headers                       Headers in Request

Request.body                          Entity-Body in Request

Response.status                       Status-Code and Reason-Phrase in Response's Status-Line

Response.headers                      Headers in Response

Response.body                         Entity-Body in Response

The HTTP implementation will automatically insert a Content-Length header when necessary and possible, and takes care of the colon separators between header names and values. It will also deal with older servers that sometimes omits the CR from the required CRLF line termination.

Note that existing Web tools (e.g. browsers) will always send the 'path' of the resource. On the ILU HTTP end, this means that the object identifier will always begin with a forward slash. For example, asking a browser to retrieve http://www.foo.bar.com/hello.html where www.foo.bar.com is serviced by an ILU HTTP server, will result in that server trying to invoke a GET operation on the object whose object identifier is /hello.html. Omitting any path info, i.e. asking the browser to retrieve http://www.foo.bar.com would result in a GET on an object whose object identifier is simply /.

Regarding the Request-URI field on the client side, it is really only necessary to put in any 'param's and/or 'queries'. Any path information in this field is just ignored, since the path info needed is to form the request is based on the object's instance handle. So for example, a client my simply put the string ";param1;param2?query" into the Request-URI field instead of "http://www.foo.bar.com/hello.html;param1;param2?query".

If operations need to occur through a proxy server, the environment variable ILU_HTTP_PROXY_INFO should be set to the proxy server name, colon, and port number e.g. ourproxyserver.foo.bar.com:8000.

For other situations, i.e. general ILU to ILU communication that just happens to be occurring over HTTP, the mapping is still consistent with HTTP protocol, but a more general format is used. ILU specific information such as the ilu_Server ID is placed in a header, and the marshaling of arguments is done entirely within the entity body. In keeping with some idea of human readability, marshaled arguments, with the exception of potentially huge byte-vectors, are encoded as readable ASCII strings - e.g. 3.1416 encodes as "3.1416". Readers concerned about utmost efficiency should note that for general ILU-ILU communication, another protocol such as ONC RPC is a much better choice than the current HTTP implementation. The HTTP protocol implementation could however be easily changed to use a more efficient encoding, similar to what's used in ONC RPC for example.

Transports

A transport stack consists of a sequence of transport layers. The last, or "bottom", layer does some kind of low-level I/O; the other layers are "filters" or "modifiers" on the transport services provided by the lower part of the stack. That is, every tail of a transport stack implements an abstraction called simply "a transport" (in English; in C, it is ilu_Transport); each transport layer (except the bottom) implements a transport in terms of another transport (the one implemented by the rest of the stack).

A transport stack is specified by a sequence of strings, each one of which specifies a transport layer. This section is a catalog of the built-in kinds of transport layers and their specification strings.

Some transports convey delimited messages (where each message is a byte sequence), others simply convey a byte sequence (that must be parsed into messages by something else). The former are called boundaried, the latter are not. Some RPC protocols require a boundaried transport, others require non-boundaried transports.

Some transports are reliable, and some aren't. Unreliable transports are deprecated in ILU, but included for interoperability with existing software that uses only unreliable transports. With these transport, messages may be delivered more than once. The ILU implementation of UDP on the server side filters out multiple receipts of the same request. Asynchronous methods may not be called over this transport mechanism, as reliable delivery of the request packet cannot be recognized by the client side. Non-asynchronous methods use the reply message as an acknowledgement that the request was received. (13)

TCP

A TCP transport layer is reliable, not boundaried, and goes on the bottom.

A TCP transport is specified by a transport-info of the form tcp_host_port.

The host needs to convey an IP address. The host can either be a dotted decimal notation of an IP address (e.g., 13.2.116.14), or be a hostname that can be mapped into an IP address.

The host in a TCP transport-info used to create a mooring (i.e., one used to tell a server how to export itself) can use special notations to mean "any IP address of this host". Those notations are 0, 0.0.0.0, and localhost. These special notations are replaced with either a name or address of the host when an SBH is produced; an SBH cannot contain such special notations. 0 and 0.0.0.0 are replaced by an address; localhost is replaced by a name, if possible, and perhaps an address otherwise. The first replacement name considered is "the hostname" of the machine. Exactly what this is (and how it is set) is system specific, but you should beware that it may or may not be a Fully Qualified Domain Name. If it can be converted to an IP address by the usual means, and a socket bound to that address, that name is used. Otherwise, the replacement is "127.0.0.1" (the loopback address -- an address that means "this machine" everywhere) or "localhost" (the canonical name for the loopback address) and the mooring's socket is bound to 0.0.0.0 (the "any" address). We bind to 0.0.0.0 rather than 127.0.0.1 because some systems (e.g., Linux) won't let us bind to 127.0.0.1.

If you use a host name instead of an address, think about how widely it can be interpreted. If it's a Fully Qualified Domain Name, the client has to be able to use the DNS to resolve it -- but not all systems include DNS support. If it's not a FQDN, the resulting SBH can only be distributed within the organization that manages the name's mapping.

The port can either be a decimal string identifying a specific port (which, of course, must not be used for anything else), or, for a mooring, be 0, which constitutes a request for a new, unused port. In the latter case, a decimal string for that port will be substituted in the transport-info.

Examples are:

tcp_augustus_0
tcp_13.2.116.14_12321
tcp_localhost_12321
tcp_0_0

UDP

A UDP transport is boundaried, not reliable, and on the bottom.

A UDP transport supports only messages that are no longer than the maximum size of a UDP packet; an attempt to send a longer message will cause an I/O error to be raised in the sender.

One could imagine creating a reliable boundaried transport layer, either using RDP (an IP protocol similar to UDP; see RFC 1151) or building on UDP. Of course, such a transport layer could not be used to communicate with ONC RPC/UDP peers.

A UDP transport is specified by a transport-info of the form udp_host_port. The host and port convey IP host address and port just as for TCP.

Examples are:

udp_augustus_0
udp_13.2.116.14_12321
udp_localhost_12321
udp_0_0

SunRPC Record Marking

A SunRPC Record Marking transport is reliable and boundaried, and is built on top of some other transport that is reliable and not boundaried. A SunRPC Record Marking transport layer is specified by a transport-info of the form sunrpcrm.

The canonical form is:

sunrpcrm

In-Memory

In-memory transport layers are not normally used or seen by users or applications; they are automatically created and used by ILU for cross-representation calls within an address space.

An In-Memory transport layer is reliable, boundaried, and on the bottom. An In-Memory transport layer is specified by the string inmem.

The canonical form is:

inmem

Security

Security transport layers may be added to a transport stack to provide some form of authenticated connection. It uses the IETF Common Authentication Technology Working Group's Generic Security Service (GSS) API to add various flavors of security to the messages that flow back and forth over the transport. Generally speaking, each outgoing message will be "wrapped" by the standard GSS routine gss_wrap, and each incoming message will be "unwrapped" by the standard GSS routine gss_unwrap. This transport also includes a mechanism for identifying callers that is integrated with the specific security scheme being used.

Use of this transport requires linking against a GSS library, implemented according to the ANSI C mapping for the GSS spec, and against an implementation of the specific GSS scheme being used.

The security transport layer is reliable and unboundaried, and requires a reliable, boundaried, transport stack below it. It is specified, on the server side, by a string of the form security_1_scheme-name, where scheme-name identifies some specific GSS security scheme. Scheme names are typically dotted-decimal strings, representing OIDs for specific schemes. Two special names are also understood, "Xerox.ILU.GSS.NIL" and "Xerox.ILU.GSS.SSL". Examples are:

security_1_Xerox.ILU.GSS.NIL -- use security with the ILU GSS NIL scheme
security_1_1.2.840.113550.9.1.3 -- another way of saying the same thing

Go to the previous, next section.