Network Working Group R. Winter Request for Comments: 515 Computer Corporation of America NIC 16446 6 June 1973
Specifications for Datalanguage, Version 0/9
Preface
Datalanguage is the language processed by the Datacomputer, a data utility system being developed for the Arpanet. The Datacomputer performs data storage and data management functions for the benefit of computers on the network.
Version 0/9 is currently running at CCA. This version is extremely primitive; however, it does offer an opportunity for experience with the Datacomputer and with fundamental Datalanguage concepts.
Subsequent versions will provide greater portions of the full Datalanguage capability, which has been described earlier (Datalanguage, Working Paper No. 3, Datacomputer Project, October, 1971, NIC 8028). For example, one of the primary restrictions in 0/9--elementary data items must be fixed-length ASCII strings--will be eliminated in Version 0/10, which is currently being implemented.
Based on the experience gained in the implementation of these early versions, and based on the feedback from their use, a revised specification of the full language will be issued.
This document presents a precise and complete specification of Datalanguage, Version 0/9. It is organized into 11 sections, of which this introduction is the first. Section 2 discusses the capabilities of Version 0/9 in general terms. Sections 3 and 4 are concerned with data description and the directory. Sections 5 through 8 cover the expression of data management operations. 9">Section 9 discusses the recognition of names. 10">Section 10 covers miscellaneous topics and 11">Section 11 specifies the syntax in BNF.
This specification is to be followed with a user manual, which will present the language in tutorial form and treat components of the Datacomputer-user interface other than the language.
Winter [Page 1]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
Version 0/9 of Datalanguage has capabilities for the storage of files; for addition of data to existing files, and for the deletion of files. Retrievals can output whole files as well as subsets of files. Data can be selected from files by content, using expressions formed from boolean and inequality operators.
At the option of the file creator, an inversion is constructed and maintained by the Datacomputer. The inversion increases the efficiency of selective retrieval, at the cost of storage space and file maintenance effort. Users other than the file creator need not be aware of the existence of the inversion, or of which fields are inverted file keys. The language is designed so that they state the desired result of a retrieval, and the Datacomputer uses the inversion as much as the request permits.
Elementary data items are fixed-length ASCII strings. Files are a restricted class of hierarchical structures.
Many of the restrictions mentioned in this memo will be short-lived. In particular, those statements followed with 3 asterisks (***) refer to restrictions that will be considerably weakened or eliminated entirely in the next version of the software.
A container is a variable whose value is a data object of general character and arbitrary size (In Version 0/9, size is restricted. See section 3.4). Examples of containers which are implemented in other systems are files, records, fields, groups, and entries.
The container is distinct from the data in the container. For example, space allocation is an operation on a container, while changing the unit price field from 25 to 50 is an operation on data in a container.
A container may enclose other containers. When a container is not enclosed by another container, it is said to be outermost. If container A encloses container B, and no other container in A also encloses B, then A immediately encloses B.
A Datalanguage description is a statement of the properties of a container.
All containers have the attributes ident and type. Ident is a character string by which users refer to the container. Type determines the form of the container's value; the value can be
Winter [Page 2]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
elementary, or it can consist of other containers. There are 3 types: LIST, STRUCT, and STRING(***). A LIST contains a group of containers having the same description. A STRUCT contains a group of containers, each of which has its own description. A STRING is a sequence of ASCII characters. While a STRING is not really an elementary item, it is handled as one in Version 0/9.
Certain containers can have other attributes. An outermost container has a function. The function attribute specifies whether the container is to be used for storage or for transmission.
Size is some meaningful dimension of the container, which is type- dependent. It is used for space allocation and data stream parsing.
An aggregate container (i.e., one that contains other containers) has as an attribute the description or descriptions of its components. Thus if S is a STRUCT containing A, B, and C, then the descriptions of A, B, and C are attributes of S.
A STRING defined in certain contexts can have an inversion attribute. This is an access property that is not really local to the STRING, but is associated with it for convenience.
The function of a container is either FILE, PORT, or TEMPORARY PORT. When the function is FILE, then the container is used for storage of data at the Datacomputer. When the function is PORT, then the container is used for transmission of data into or out of the Datacomputer. When the function is TEMPORARY PORT (which may be abbreviated TEMP PORT), the container behaves like a PORT; however, its description is not retained in the Datacomputer beyond the session in which it is created.
Winter [Page 3]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
Type is one of: LIST, STRUCT, or STRING. These are defined on the preceding page.
In an occurrence of a STRUCT, the elements appear in the order in which their descriptions appear in the STRUCT description. All elements are present in each occurrence of the STRUCT.
An element of a STRUCT or LIST can be a container of any datatype. However, the outermost container must be a LIST(***).
The size of a STRING is the number of characters in it. The size of a STRUCT is not defined (***). The meaning of the size of a LIST depends upon other properties of the LIST (***).
Ordinarily, the size of a LIST is the number of LIST-members. An exception is the case of the outermost-LIST. In an outermost-LIST with a function of FILE, the size is the number of LIST-members for which space should be allocated. When no size is present in this case, the system computes a default. In an outermost-LIST with a function of PORT, the size is ignored (***).
Only outermost containers may be larger than a TENEX page (2560 ASCII characters)(***).
An inversion is an auxiliary data structure used to facilitate retrieval by content.
Its basic application is the fast retrieval of sets of outermost- LIST-members (this can be extended to other container sets, and will be after release 1). Consider a list of weather observations, stored as a file on the Datacomputer. If quick retrieval of observations by COUNTRY is desired, then this is indicated in the description of the COUNTRY container. According to common usage in information retrieval, this makes COUNTRY a key in the retrieval of observations.
Note that the inversion option only affects the efficiency of retrieval by COUNTRY, not the ability to retrieve by COUNTRY.
Winter [Page 4]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
There are restrictions on use of the inversion option. First, it can be applied only to STRINGs. Second a STRING having the inversion option must occur only once in each outermost-LIST-member. Third, it is ignored when applied to STRINGs in PORT descriptions.
Eventually there will be several types of inversion option; in Version 0/9 there is only the 'D' option (for distinct).
The description is simply an enumeration of properties; these properties are specified in the order:
<ident> <function> <type> <size> <other>
Properties which do not apply are omitted. An example:
F FILE LIST (25) A STR (10)
Here 'F' is the <ident>, 'FILE' is the <function>, 'LIST' is the <type>, '(25)' is the size, and 'A STR (10)' is the <other> of one description. Of course, 'A STR (10)' is itself another description: the description for members of the LIST named F.
An example of a complete description for a file of weather observations keyed on location:
WEATHER FILE LIST OBSERVATION STRUCT LOCATION STRUCT CITY STR (10), I=D COUNTRY STR (10), I=D END TIME STRUCT YEAR STR (2) DAY STR (3) HOUR STR (2) END DATE STRUCT TEMPERATURE STR (3) RAINFALL STR (3) HUMIDITY STR (2) END END The ENDs are needed to delimit the list of elements of a STRUCT. `, I=D' indicates that the string is to be an inversion key for the retrieval of outermost-LIST-members.
Winter [Page 5]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
The directory is a system file in which the names and descriptions of all user-defined containers are kept.
The directory is structured as a tree. Each node has an ident, which need not be unique. There is a single path from the root of the tree to any node. The idents of the nodes along this path are concatenated, separated by periods, to form a pathname, which unambiguously identifies the node (e.g., A.B.C could be a pathname for node with an ident of C).
In a later version of the software, the directory will be generalized to provide for links between nodes, so that it will not properly be a tree. For now, however, the tree model is convenient and adequate.
A node may represent a container, or it may simply hold a place in the space of pathnames. When it represents a container, it cannot (currently) have subordinate nodes.
Eventually, it is planned to model the directory as a structure of containers, with its description distributed throughout the structure. Most operations defined on the directory will be defined on user data, and vice versa. Access privileges and privacy locks will be part of the data description and will likewise be applicable both to directory nodes and data structures below the node level.
A CREATE-request either; (a) adds a node to the directory, optionally associating the description of either a PORT or a FILE with the node, or (b) creates a temporary container which is not entered in the directory, but has a description and can be referenced in requests. If the description defines a file, CREATE causes space to be allocated for the file.
To create a node with a description: CREATE <pathname> <description> ; To create a node with no description: CREATE <pathname> ; Note that the description determines whether or not the container is temporary (see section 3.2 for details).
Winter [Page 6]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
A CREATE-request adds a single node to the directory. Thus to add CCA.RAW.F to an empty directory, three requests are needed: CREATE CCA ; CREATE CCA.RAW ; CREATE CCA.RAW.F ; Notice that the last ident of the pathname doubles as the first ident of the description: CREATE CCA.RAW.G FILE LIST A STR (5) ;
That is, G is both the ident of a node and the ident of an outermost container of type LIST.
A DELETE-request deletes a tree of nodes and any associated descriptions or data. The syntax is: DELETE <pathname> ; The named node and any subordinates are deleted.
Note that to delete data while retaining the directory entry and description, DELETE should not be used (see section 6.3 for the proper method).
The LIST-request is used to display system data of interest to a user. It causes the data specified to be transmitted through the Datalanguage output port.
Several arguments of LIST apply to the directory. LIST %ALL transmits all pathnames in the directory. LIST %ALL.%SOURCE transmits all descriptions in the directory. Instead of %ALL, a pathname can be used: LIST <pn>.%ALL Lists pathnames subordinate to <pn>. LIST <pn>.%SOURCE lists descriptions subordinate to the node represented by <pn>.
For details about the LIST-request, see section 10.1.
Containers must be open before they can be operated on.
A container is open when it is first created. It remains open until closed explicitly by a CLOSE-request or implicitly by a DELETE- request or by session end.
Winter [Page 7]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
A closed container is opened by an OPEN-request. A temporary container is always open; a CLOSE-request deletes it.
An open container has a mode, which is one of: READ, WRITE, or APPEND. The mode determines the meaning and/or legitimacy of certain operations on the container.
The mode is established by the operation which opens the container. It can be changed at any time by a MODE-request. A CREATE leaves the container in WRITE mode. An OPEN either specifies the mode explicitly or implicitly sets the mode to READ.
LIST %OPEN transmits name, mode and connection status of each open outermost container through the Datalanguage output port. (The Datalanguage output port is the destination to which all Datacomputer diagnostics and replies are sent. It is established when the user initially connects to the Datacomputer.) For details of the LIST-request, see section 10.1.
Assignment transfers data from one container to another.
The equal sign ('=') is the symbol for assignment. The value of the operand on the right of the equal sign is transferred to the operand on the left. (Eventually, both operands will be weakly-restricted
Winter [Page 8]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
Datalanguage expressions, which may evaluate to sets as well as to single containers. Now, the left must be a container name, the right may be a container name or a constant.)
Assignment is defined for all types of containers. When the containers are aggregates, their elements are paired and data is transferred between paired elements. Elements of the target container that do not pair with some source element are handled with a default operation (currently they are filled with blanks).
The operands of an assignment must have descriptions that match. The idea of matching is that the descriptions must be similar enough so that it is obvious how to map one into the other.
Assignment must reference objects, not sets. An object is: (a) an outermost container, or (b) a constant, or (c) in the body of a FOR-loop, either (c1) a member of a set defined by a FOR-OPERAND, or (c2) a container which occurs once in such a member
In the case of a reference of type (c1), the object referenced is taken to be the current member. In the case of (c2), the object referenced is that which occurs in the current member. This is explained further in section 7.
The left operand of an assignment is subject to further restriction. If it is an outermost container, it must be open in either WRITE- or APPEND-mode. If it is not an outermost container, then the reference is of type (c), which means that some FOR-operand has established a context in which the assign- operand is an object. The FOR-operand which establishes this context must be the output-operand of the FOR.
When the assign-operand is an outermost container, it must be open. Such an operand must be referenced by its simple container ident(***), not its directory pathname.
In the body of a loop nested in one or more other loops, assignments are further restricted, due to a 0/9 implementation problem. See section 7.2 for details.
Winter [Page 9]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
Finally, the descriptions of the operands must match. If one is a constant, then the other must be a STRING(***). If both are containers, then in the expression: A = B; the descriptions of containers A and B match if: 1. A and B have the same type 2. If A and B are LISTs, then they have equal numbers of LIST-members, or else A is an outermost-LIST. 3. If A and B are aggregates, then at least one container immediately enclosed in A matches, and has the same ident as, one container immediately enclosed in B.
If the operands are STRINGs, then the value of B, left-justified, replaces the value of A. If B is longer than A, the value is truncated. If B is shorter than A, then A is filled on the right with blanks as necessary.
If the operands are STRUCTs, then assignment is defined in terms of the STRUCT members. If a member of A, mA, matches and has the same name as a member of B, mB, then mB is assigned to mA. If no such mB exists, then mA is filled with blanks.
If the operands are LISTs, the result depends on several factors. First, notice that the descriptions of the LIST-members must match; otherwise the assignment would not be legitimate by the matching rules of 6.1.
If A is an outermost-LIST, then it can be in either of two modes: WRITE or APPEND. If A is in WRITE-mode, its previous contents are first discarded; it is then handled as though it were in APPEND- mode.
If A is not an outermost-LIST, then it is always effectively in WRITE-mode.
After taking the mode of A into account, as described above, the procedure is: for each member of LIST B (a) add a new member to the end of A (b) assign the current number of B to the new member of A
Winter [Page 10]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
If A is an outermost container in WRITE-mode, and B is a container with description that matches A, and if B contains no data, then A=B has the effect of deleting all data from A. Note that if A is in APPEND-mode in these circumstances, then A=B is a no-operation (i.e., has no effect).
FOR <output set spec>, <input set spec> <body> END ;
The output set is optional: that is, FOR need not produce output. When the output set is omitted, the syntax is:
FOR <input set spec> <body> END ;
The operations specified in the body are performed once for each member of the input set. References in the body to the input set member are treated as references to the current input set member. When an output set is present, a new member is created and added to the output set for each iteration (i.e., for each input set member). References to the output set member, similarly, are treated as references to the current output set member.
The output set spec must be the name of a LIST member. Each iteration of the FOR will create one such member, and add it to its LIST (hereafter called the output LIST). The body terminates the value that the new member receives. Any STRING in the new member which is not given a value by the body receives he default value of all blanks.
The input set spec must be an expression evaluating to a set of LIST-members (see section 7.1 for details of input set specification). Each iteration for the FOR will input one such member; the FOR will terminate when each member of the set has been processed. The LIST from which the input set members are drawn is called the input LIST.
FOR is effectively a means of accomplishing variants of assignment between a pair of LISTS. FOR is less concise than assignment, but offers more flexibility. Its advantages are: (a) not all the input LIST-members need be transferred to the output LIST. A subset can be selected by content. (b) the user has explicit control over the assignment of values to output LIST-members.
Winter [Page 11]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
This is most easily understood by an example: P PORT LIST F FILE LIST R STRUCT R STRUCT B STR A STRUCT C STR A1 STR END A2 STR B STR C STR END
(1) P = F ; (2) FOR P.R, F.R P.R = F.R ; END ; (3) FOR P.R, F.R WITH A1 EQ 'XY' OR A2 GE 'AB' B = C ; C = A2 ; END
Here, (1) and (2) are entirely equivalent requests. However, (3) is quite different and is not expressible as assignment. It selects a subset of F.Rs. The values it gives to the P.Rs could not result from application of the matching rules to F and P.
Because FOR is effectively assignment between a pair of LISTs, the LISTs referenced by a legitimate FOR-operation are largely subject to the same restrictions as LISTs referenced in an assignment. One exception is that the descriptions of the LIST-members need not match.
These restrictions are: (a) both LISTs must be objects in the context in which the FOR appears. (b) both LISTs must be open or contained in open outermost containers. (c) if the output LIST is an outermost container, it must be in WRITE- or APPEND-mode. (d) If the output LIST is not outermost, the LIST which most immediately encloses it must be the output LIST of an enclosing FOR.
The mode of the output LIST of the FOR affects the result much as it would in an assignment: that is, a FOR outputting to a LIST in WRITE-mode overwrites previous contents, while a FOR outputting to a LIST in APPEND-mode adds to previous contents.
Winter [Page 12]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
CAUTION TO THE READER: For convenience, these specifications use phrases such as 'LISTs referenced by a FOR'. Recall that such a phrase is not literally correct, in the sense that the operands of a FOR are always LIST members, not LISTs.
The input set is specified by a Datalanguage expression that evaluates to a set of LIST-members. Such an expression can be simply the set of all members of a LIST, or it can be a subset of the members of a LIST, specified by content. For example, with the description:
F FILE LIST R STRUCT A STR (1) B STR (2) END
the expression: F.R references all R's on the LIST F. However: F.R WITH A EQ '5' references only those R's containing an A having the value '5'. The expressions permitted as input set specifications are of the form: <list-member-name> WITH <boolexp> The <boolexp> is constructed of comparison expressions joined by the Boolean operators AND and OR. Any expression can be negated with NOT.
Comparison operators have the highest precedence. Next highest is AND, then OR, then NOT.
The comparison expressions are restricted to the form: <container name> <comop> <constant> where: (a) <constant> is a string constant enclosed in single quotes (see section 10.2 for a discussion of constants) (b) <comop> is one of six operators: EQ equal NE not equal LT less than GT greater than LE less than or equal to GE greater than or equal to (c) <container name> is the name of a STRING that appears once in each LIST-member.
Winter [Page 13]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
The constant is truncated or padded with blanks on the right to make it equal in size to the container to which it is being compared. Notice that padding on the right is not always desirable (users will have control over the padding in a future release). In particular, care must be exercised when using numbers in Version 0/9. (A number represented as a STRING should actually be described as a number; eventually it will be possible to do this).
Two operations are legitimate in a FOR-body: FOR and assignment.
These are subject to the restrictions discussed in Section 6.1 and in the introduction to Section 7. The restrictions are related to three requirements: (1) that the names be recognizable (see Section 9 for details), (2) that a request be consistent regarding direction of data transfer between containers, both within itself and with the MODE of outermost containers, and (3) that transfers occur between objects, not sets of objects. The first two requirements are permanent, but will become weaker in later versions of the language. The last requirement is temporary and will be present only in early versions.
Due to an implementation problem associated with Version 0/9, there is a somewhat bizarre restriction applied to references made in the body of a loop nested in another loop. This restriction is not expected to pose any practical problems for users, and is not part of the language design, but is discussed here for completeness.
The restriction is most easily understood by example:
given the description F LIST R STRUCT A STR (3) BL LIST (3) B STR (3) C STR (3) END and the request fragment: FOR ...,R FOR ...,B ... = A ; ... = C ; END END
Winter [Page 14]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
observe: (a) The outer loop processes the set of R's in F. (b) For each R in F, the inner loop processes the set of B's in the BL contained in that R. (c) In the body of the inner loop, there are references to A and C, which do not occur in B, but do occur in R. That is, the objects referenced in the inner loop body are defined by the outer loop, not the inner loop. In general, this is fine; in the case of C, however, we have a problem. (d) C occurs beyond the end of BL.
The 0/9 compiler is capable of neither (1) looking ahead enough to locate C before it compiles code for the loop, nor (2) while generating code to loop on the B's in BL, generating a separate body of code that skips to the end of BL to locate C. Thus it can handle A, which has been located before it begins loop generation, but it cannot handle C, which requires a little foresight.
The request fragment shown would not cause problems if the description were changed to:
F LIST R STRUCT A STR (3) C STR (3) BL LIST (3) B STR (3) END
Then both A and C would have been found before code for the inner loop was generated.
Data is transferred from container to container by execution of assignment and FOR operations. The outermost containers involved in transfers can be files or ports. If both are files, then the transfer is internal to the Datacomputer. If either is a port, then an address in the external world is needed to accomplish the data transmission.
Winter [Page 15]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
Such an address is supplied through a CONNECT-request, which associates a container (having a function of PORT) with an external address: CONNECT <ident> TO <address> ; Here <address> is either a specifications of host and socket number, or a TENEX file designator (for CCA's TENEX) enclosed in single quotes. The host and socket form is: <socket> AT <host> where <socket> is a decimal number, and <host> is either a decimal number or a standard host name (since standard host names don't exist right now, it has to be the TENEX 'standard' name for the host. Contact the author for the latest list.) If <host> is omitted, it is taken to be the host from which the Datalanguage is being transmitted.
The address associated with a port can be changed by issuing another CONNECT-request.
A DISCONNECT-request simply breaks an existing port/address association without establishing a new one. (A CLOSE-request that references an open port executes a DISCONNECT.) The syntax of DISCONNECT is: DISCONNECT <ident> ;
A port is disconnected when: (a) no successful CONNECT-request has ever been issued for it, or (b) a DISCONNECT for the port has been executed since the last successful CONNECT.
When a disconnected port is referenced in an assignment, it is connected by default either to: (a) the connection used for the transmission of Datalanguage to the Datacomputer, or (b) the connection used for the transmission of Datacomputer diagnostics to the user The choice between (a) and (b), of course, depends on whether the reference is for input or output. These connections are established by the network user's ICP to the Datacomputer at the beginning of the session.
Note that CONNECT and DISCONNECT do not open files or network connections. They simply make address associations within the Datacomputer. The files and connections are opened before each request and closed after each request.
Winter [Page 16]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
A name is recognized when it has been associated with a particular data container or set of containers.
Datalanguage has mechanisms for the recognition of names in contexts. That is, the meaning of the name is related to where it appears.
This makes it possible to attach natural meanings to partially qualified names.
For example:
WEATHER FILE LIST STATION STRUCT CITY STR (15) STATE STR (15) DATA LIST (24) OBSERVATION STRUCT HOUR STR (2) TEMPERATURE STR (3) HUMIDITY STR (2) PRESSURE STR (4) END END
RESULTS PORT LIST RESULT STRUCT CITY STR (15) HOUR STR (2) TEMPERATURE STR (3) END
FOR STATION WITH STATE EQ 'CALIFORNIA' FOR RESULT, OBSERVATION WITH HOUR GT '12' AND HUMIDITY LT '75' CITY = CITY ; HOUR = HOUR ; TEMPERATURE = TEMPERATURE ; END ; END ;
in the assignment 'CITY = CITY', the first CITY is understood to be RESULT.CITY and the second is understood to be STATION.CITY.
Winter [Page 17]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
'Ident' is used in the sense of section 3. For example, in the description: F FILE LIST R STRUCT A STR (1) B STR (1) END F, R, A and B are idents.
A context is a tree whose nodes are idents. In such a tree, the terminal nodes are idents of STRINGs. The ident of a LIST is superior to the ident of the LIST-member. The ident of a STRUCT is superior to the idents of the STRUCT elements. The context whose top node is F is said to be the context of F.
+-----+ ! F ! +-----+ ! ! +-----+ ! R ! +-----+ ! ! +----------+----------+ ! ! ! ! +-----+ +-----+ ! A ! ! B ! +-----+ +-----+
Figure 9.1-1 The context of F
A pathname is a sequence of idents, naming nodes along a path from one node to another. A full pathname in the context starts at the topmost node. Thus F.R.B is a full pathname in the context of F. A partial pathname starts at a node other than the topmost node (e.g. R.B, B).
In Datalanguage, pathnames omitting intermediate nodes, such as F.B (which omits 'R'), are not permitted. Thus partial pathnames are partial only in that additional names are implied on the left.
Winter [Page 18]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
Three attempts at recognition of a pathname, PN, in a context, CX, are made:
(a) recognition of PN as a full pathname in CX (b) recognition of PN as a partial pathname in which only the topmost node of CX is omitted (c) recognition of PN as an arbitrary partial pathname occurring only once in CX.
The attempts are made in the above order, and the recognition process halts with the first successful attempt.
As an example, consider the description:
F FILE LIST R STRUCT A STR B STR S STRUCT R STR
which defines the context in Figure 9.1-2.
+-----+ ! F ! +-----+ ! ! +-----+ ! R ! +-----+ ! ! +----------------+----------------+ ! ! ! ! ! ! +-----+ +-----+ +-----+ ! A ! ! B ! ! S ! +-----+ +-----+ +-----+ ! ! +-----+ ! R ! +-----+
Figure 9.1-2 Example Context
Winter [Page 19]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
In this context, F.R.A is a full pathname. Thus, F.R.A is recognized in attempt (a). R is a partial pathname in which only the topmost node is omitted. Thus R is recognized in attempt (b). Note carefully that R is recognized as a reference to F.R, not to F.R.S.R. Finally, B is an arbitrary partial pathname occurring only once in the context. Thus B is recognized in attempt (c).
Two stacks of contexts are maintained: one for names used in an input sense, and one for names used in an output sense. When a name is to be recognized, it is first decided whether the reference is an input reference or an output reference. An input reference is (a) the right hand operand of an assign, or (b) a name in the input set spec of a FOR. An output reference is (a) the left operand of an assign, or (b) the output operand of a FOR. The first context on the appropriate context stack is then searched, according to the procedure outlined on the previous page. If the name is neither recognized nor ambiguous in that context, search continues in the next context on the stack. If the name can be recognized in none of the contexts on the appropriate stack, it is unrecognizable.
When a stack is empty, the recognition procedure is different. The search is carried on in a special context: The context of %OPEN. Its top node, %OPEN, is a built in system ident. Subordinate to %OPEN is a context for each open directory node. Each such context represents all the idents defined in the directory nodes having data descriptions: F FILE LIST R STRUCT A STR (1) B STR (1) and: P PORT LIST R STRUCT A STR (1) B STR (1) then the context of %OPEN would be as in Figure 9.1-3.
Winter [Page 20]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
+-------+ ! %OPEN ! +-------+ ! ! +-----------+-----------+ ! ! ! ! +-----+ +-----+ ! A ! ! S ! +-----+ +-----+ ! ! ! ! +-----+ +-----+ ! R ! ! R ! +-----+ +-----+ ! ! ! ! +-----+-----+ +-----+-----+ ! ! ! ! ! ! ! ! +-----+ +-----+ +-----+ +-----+ ! A ! ! B ! ! A ! ! B ! +-----+ +-----+ +-----+ +-----+
Figure 9.1-3 The Context of %OPEN
When a directory node is closed, the corresponding context is removed from the context of %OPEN. When a node is opened, the associated context is added as the rightmost context subordinate to %OPEN.
The context stacks are always empty between requests. The FOR- operator adds entries to the stacks. FOR A adds the context of A to the input context stack. FOR A, B ... adds the context of A to the output stack and the context of B to the input stack.
When adding to an empty stack, FOR adds two contexts instead of one. The second addition to the stack is the context of the looparg; the first addition is the context of the outermost container which encloses the looparg.
Winter [Page 21]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
For example, given a context of %OPEN as in Figure 9.1-3, and empty context stacks, the fragment 'FOR F.R ...' adds two contexts: (1) the context of F, and (2) the context of F.R.
Contexts are removed from the stacks by the END matching the FOR which added them.
The LIST-request provides a means for the user to inspect system data of interest to him. The user can determine the contents of the directory, the source or parsed form of any data description in the system, and the mode and connection status of any open file or port.
The LIST operator has a single operand, which can have any of several forms. The action of the operator is to output a list of values on the Datalanguage output port.
To output a list of pathnames, the operand used is '%ALL'. When '%ALL' appears alone, all pathnames in the directory are listed. When '%ALL' appears after the last ident in a directory pathname, the full pathnames of all nodes subordinate to the named node are listed.
To output one or more source descriptions, the operand '.%SOURCE' is used. '.%SOURCE' is preceded with one of (a) '%ALL', (b) '%OPEN', or (c) the ident of an open outermost container. The output is either (a) all descriptions, (b) all open descriptions, or (c) a particular open description.
To output a parsed description, the operand '.%DESC' is used ('%DESCRIPTION' is also accepted). This operand is preceded either with (a) '%OPEN', or (b) the ident of an open outermost container.
Winter [Page 22]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
Examples:
Let P be the ident of an open PORT. Let A.B.C be a directory pathname. LIST %ALL LIST A.B.C.%ALL LIST %OPEN LIST %ALL.%SOURCE LIST %OPEN.%SOURCE LIST P.%SOURCE LIST %OPEN.%DESC LIST P.%DESC
Note that 'LIST A.B.C.%SOURCE' is not legal - '.%SOURCE' must be preceded with the ident of an open container, not a pathname. A similar restriction applies to '.%DESC'.
A constant of type STRING can be included in a Datalanguage request by enclosing it in single quotes: 'ABC' A single quote is included in a constant by preceding it with a double quote: 'FATHER"'S' Likewise, a double quote is included by preceding it with a double quote: 'JOHN SAID ""HELLO""'
Such constants can be used on the right of comparison operators and of assignment operators which reference strings.
Eventually, Datalanguage will contain facilities for the inclusion of constants of all datatypes; such constants are simply a special case of the Datalanguage expression and will be permitted wherever such expressions are permitted.
Internally, Version 0/9 of the Datacomputer software operates in 7-bit ASCII characters. Its output to the ARPANET is converted to 8-bit ASCII. On input from the ARPANET, it expects 8-bit characters, which it converts to 7-bit characters.
Winter [Page 23]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
To convert from 7- to 8-bit characters, a '0' bit is prefixed. To convert from 8- to 7-bit characters, the high-order bit is checked. If the high-order bit is a '0', the bit is discarded and the character is accepted as a 7-bit character. If the high-order bit is a '1', then the character is discarded.
(In the following discussion, as elsewhere in this memo, all references to numerical character codes are in decimal).
The remainder of this section discusses treatment of codes 0 through 127, when they appear in Datalanguage requests.
In general, printing characters are acceptable in requests, while control characters are not. There are some exceptions, which are detailed below. The printing characters are codes 32-126. The control characters are codes 0-31 and 127.
Certain control characters are accepted:
Tab(9) is accepted freely in requests. It functions as a separator (explained below).
EOL(31), meaning end-of-line, is accepted in requests, functioning both as a separator and an activator (a). EOL has a special meaning in data, and should not be introduced into STRING constants(***).
Control-L(12) is an activator and a high-level request delimiter. It terminates the test of any request being processed when it is encounter in the input stream. It is useful in Datacomputer-user program synchronization.
Control-Z(26) means end-of-session when encountered in Datalanguage. It has the properties of control-L, and in addition, causes the Datacomputer to execute an end-of-session procedure, which results in closing the Datalanguage connections, closing any open files or ports, etc. The effect is equivalent to a [LOGOUT(which] does not exist yet).
The two-character sequence <carriage return(13), line feed(10)> is equivalent to EOL (and is translated to EOL on input from the network). The reverse sequence, as well as either character alone, is treated simply as other control characters (ignored).
All other control characters are ignored.
Winter [Page 24]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973
The printing characters are further divided into four groups: special characters, uppercase letters, lower case letters, and digits (the membership of these groups is defined in section 11).
Corresponding upper and lower case letters are equivalent in requests, except with quoted strings.
Certain special characters have a lexical function, which is either break or separator. A break character terminates the current lexical item and returned itself as the next item. A separator character terminates the current item but does not begin a new item (i.e., its only function is to separate items). Multiple separators are equivalent to a single separator. A separator can always be inserted before or after a break character, without altering the meaning of the request.
The separators are tab(9), space(32), and end-of-line(31).
The break characters are left parenthesis(40), right parenthesis(41), equals(61), semicolon(59), period(46), comma(44), quote(39), and slash(47).
(a) An activator character causes the Datacomputer to process whatever has been received since the previous activator or the beginning of the request. The meaning of a request is independent of the presence/absence of activators. However, a request will not be executed until an activator beyond the end of the request is received.
While Version 0/9 defines (carriage return, linefeed) in terms of EOL, network users should not think in terms of sending EOL's to the Datacomputer over the network. EOL is not part of the network ASCII character set, and has no definite permanent place in Datacomputer implementation plans.
Comments can be included in Datalanguage requests. A comment is begun with the two-character sequence '/*', and ended with the two-character sequence '*/'. Since slash is a break character, a comment does cause a lexical break; its overall effect is that of a separator.
Winter [Page 25]
RFC 515 Specifications for Datalanguage, Version 0/9 6 June 1973