JPH02151971A - Method for searching node of tree structure data - Google Patents

Abstract

PURPOSE:To rapidly execute node search without saving addresses and to rapidly decide the existence of a loop by holding a succeeding node address necessary for node search in each node. CONSTITUTION:A node value expressing an applied tree structure data and a slave node address are set up on a prescribed position of a node information area. Then, the search order of the tree structure data is determined and the succeeding node address is set up in a prescribed position of the node information area. When a slave node address is not set up, the succeeding node address is used to successively refer respective node information of n-advanced three structure data. Consequently, address stacking at the time of searching nodes is omitted and the existence of a loop existing in the tree structure data is efficiently detected.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a method for representing and processing tree-structured data, and particularly relates to a processing method suitable for node searching of tree-structured data required for pattern matching processing. .

[Conventional technology]

Among the data structures handled in non-numerical processing fields.

Tree-structured data is a frequently used data structure. Here, the tree structure data refers to a set of node information. The node information includes information indicating the value of the node (node value) and the relationship between the node and another node (child node). Each tree-structured data has a node (
(referred to as the root node) is determined.

The basic process for this tree-structured data is a process of extracting the node value of each node in order of depth priority, and is called a node search process. For example, consider a case where a pattern-number comparison of two tree-structured data is performed. If all the node values are the same or at least one is [mon] (indication that any value is allowed; the value may be structured data), the two tree-structured data are considered to have a pattern match. In order to perform a turn matching comparison, node search processing must be performed on each of the two tree-structured data.

Figure 2 shows an example of tree-structured data. According to the above-mentioned definition of he'turn matching, the tree structure data in figure (a) becomes (
))), the pattern does not match with (Q), but the pattern matches with (d) in the same figure. This is because (a) and (b) have mismatched structures, and (a) and (c) have mismatched node values (the second node values c and m from the left on the fourth level from the top). .

Conventional data representation methods in memory for tree-structured data, particularly data representations suitable for pattern-number comparison of tree-structured data, are described in Danii, Research Report Nα, 39 (19
77), pages 36 to 44 (D, A,].

-Reasearch Report Nn, 39 (19
77) pp, :(6-4/I, I), 11. l).

Written by Warren: Impl, ementinF, PR
OLOG-compj Hngpredicate l
ologic programs Volume]). Figure 3 shows how the tree-structured data shown in Figure 2(a) is expressed using the conventional method.3゜A node with n child nodes is divided into n + 1 consecutive cells (memory elements). It is expressed as a sequence of Each cell is composed of a "tag field" and a [content field], and the tag fields IF+, IAI, II]1 are "Fanfuta cell", "A1 hem cell", "
``address cell'', i.e. the value of each content field is the node value of a node with r child nodes'',
Indicates that it is a "node value of a node that does not have a child node" and a "storage address of a cell representing a child node." This data representation method is hereinafter referred to as 1-type representation.

[Problem to be solved by the invention]

There are two problems with the tree-type theta representation.

The problem with this method is that during the node search process, it is necessary to sequentially store the addresses of the cells to be checked next. For example, in the case of FIG. 3, before checking node g, White 11 checks the cell address of node a, and then checks node h.
It is necessary to store the address of the cell pointed to by node °p. This address storage process is called address scooking.

The second problem with the tree-type theta representation is that internal loops existing in tree-structured data cannot be efficiently detected. For example, the tree structure data shown in FIG. 4 includes an internal loop. If the existence of this loop is not detected, the node search process will fall into an infinite loop. When representing tree-structured data using conventional data representation, in order to detect the existence of a loop, all addresses of nodes searched in the past (ancestor nodes) are memorized, and each time a new node is searched, , it is necessary to check whether the address of the node exists in the already stored address group of ancestor nodes.

Another conventional example of a search method for tree-structured data is JP-A-58.
-3035r tree structure information search control method".

This method is a method in which past search progress for tree-structured data is stored within the tree structure.

An object of the present invention is to solve the first problem of the conventional method described above, that is, to eliminate the need for address scooking when searching for a node.

Another object of the present invention is to efficiently detect the presence of loops in tree-structured data.

[Means to solve the problem]

In order to achieve the above object, a data representation method is introduced in which the node information corresponding to each node includes the address of the next node to be searched (next node address). This representation method is a non-conventional representation method, and will hereinafter be referred to as network-type data representation.

In this respect, it is completely different from the above-mentioned Japanese Patent Application Laid-Open No. 58-3035.

When using network type data representation, the above-mentioned next node address is set as data for each node at the time the tree structure data is stored in the memory. Since the next node address is always set to the address to be searched next, address scooking in the conventional method is no longer necessary. However, this method is useful when the same data structure that exists in two different parts of tree-structured data (tree-structured data, more specifically, subtree-structured data) cannot be expressed in a single data representation, such as in an inner loop. Applicable.

When the same data structure (tree-structured data) that exists in two different parts of tree-structured data is shared in one data representation to form a table (this includes cases where there is a loop inside), the tree-structured data Although it is not possible to set the next node address at the time of storing the structural data on the memory, the same effect can be obtained by setting it at the time of node search.

[Effect]

The information stored in each node information in network type data table type 1 is shown in Figure 5. The lo tree diagram represents the case where the tree structure data in Figure 2 is expressed.
) represents a node, and the value inside is the node value.

The dotted circle (○) is a node (terminal node) that virtually indicates the end of the newly set node search. The node value of the terminal node 1 is set to rendJ.

Arrows (→) in the figure indicate child node address information and next node address information stored in the node information of the node at the root of each arrow. Among the arrows, the thin arrows represent information stored in memory in the conventional tree-type theta representation. In addition to this information, the network type representation can store bold arrow information in the node information.

FIG. 1 shows a more specific example of network data representation. The node information of a node with n child nodes is
It is expressed as a sequence of 1,000 and 2 cells. By arranging the cells in the order of cells indicating the node value of the white node, cells indicating the first to nth child nodes, and cells indicating the next node, the child node search procedure and the next node search procedure can be realized by a single and simple processing procedure (hereinafter referred to as the new dereference processing procedure) that extracts the adjacent cell and, if that cell indicates an address, extracts the cell at the location of that address. .

Network-type data representation allows the next node address to be stored in correspondence with the node, so the same data structure (tree-structured data) that exists in two different parts of tree-structured data is not shared and expressed in one data representation. If -
The next node address once set can be used repeatedly. This is the first advantage derived from using network-based data representation. For example, when implementing a symbolic processing language 1) that handles tree-structured data (OL OG processing system,
If the tree-structured data of the formal argument in the block head is expressed as network data, there is no need to perform address stacking to search for nodes in the tree-structured data of the formal argument every time the predicate is called.

Assuming that the address of the cell currently being searched is p, it consists only of the following process: ``Find the address of the cell adjacent to p.Determine the type of cell pointed to by rpJ.'' The conventional search process The node search process can be realized with extremely simple processing compared to the case where it was necessary to determine the number of addresses and to recover the addresses saved in the stack.

The Fth advantage of using the network type data representation is that this representation can be interpreted as it is as a conventional 1-1 type data representation. The difference between the conventional tree-type data representation (Fig. 2) and the network-type data representation (Fig. 1) is that the former represents a node with T] child nodes as n-1-1-
While the latter is represented by a row of n+2 cells,
It is only a point that is represented by a sequence of cells. Furthermore, the contents are the same up to the (n+1)th cell among the (n+2) cells. Therefore, if the (n+2)th cell is ignored, it can be handled as a conventional tree-type data representation. This advantage is the premise for producing the fourth advantage described below, and also has the role of alleviating the disadvantages of network-type data representation described later.

In order to detect loops that exist within tree-structured data, it is necessary to check whether the parent node being searched has been revisited. In network-type data representation, the next node address representation area is stored in association with the node, but if the next node address has already been set for the node to be searched for, the parent node is revisited. Therefore, by determining whether or not the next node address is set in the search destination node, it is possible to determine whether or not there is an internal loop. This is the fourth advantage derived from using a network type data representation.

For example, consider the tree structure data shown in FIG.

When tracing from the root node (node f), the second child node of node 11 is looped. When searching for a node, set the next node address for that node, and delete it when the search ends. When searching for node f from node h, the existence of a loop is known because the next node address of node f is not erased.

The first drawback of network data representation is that it consumes more memory than tree-type data representation, but with the recent expansion of memory capacity and the development of data cache mechanisms, this drawback has been overcome. It no longer becomes -.

The second drawback of the network-type data representation is that the same data structure (tree-structured data) existing in two different parts of the main structural data cannot be shared and represented in one data representation. However, as mentioned in the third advantage, since the network type representation is an expression that completely includes the tree type data representation, by subtracting the processing part that handles the network type data representation as the tree type data representation, it is possible to This is not a problem.

For example, in the tree structure data shown in FIG. 2 (8), the subtrees below the node have the same number of 41 M'i. When this subtree is expressed using a single data representation, it becomes as shown in FIG. In the figure, cells with diagonal lines (next node cells) cannot be set when generating this tree structure data. However, this data can be handled as a tree-type data representation without any inconvenience.

In the PItOLOG processing system, the existence of loops is not allowed in tree-structured data included in a blockrum.

Therefore, network-type data representation is used to achieve high speed by taking advantage of the first advantage. On the other hand, since there is a possibility that shared subtrees may occur in data subjected to pattern matching processing, the screw 1-water data representation is processed as a conventional tree-type data representation.

Furthermore, when registering this data as a program, it is checked that there are no internal loops, taking advantage of the fourth advantage, and the next node address of each node is set to create a complete network-type data representation.

〔Example〕

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 shows a processing flow based on the present invention. In order to make the explanation more concrete, the following tree structure data will be processed by the present invention.

fk(h(b, c+ d) + P(e, 1.))
l a, g(, j+ P(e+ 1))) This tree structure data shows the tree structure shown in FIG. 2(a). That is, if there is a node X having node y and node Z as child nodes, their structure is expressed as x (y +2).

Six node information representing the above tree structure data as node information as shown in FIG. 1 130 is represented by an array of n+2 cells when each node has n child nodes. For example, node f has three nodes g+au g as child nodes, so the node information is represented by a sequence of five cells.

In step 100, node information is set as explained above. In the first cell of the node information, the tag field is set to F, and the content field is set to the value of the node and the number of child nodes that the node has. For example, regarding node f, it is set as 111-/3''. The value of the child node or the address of the node information area of the child node is set in the second to n+1 cells. When there is no child node, the child node value is set as au.In that case, the tag field is A.For example, in node f, its child node a does not have any child nodes, so the tag field is set to A, "a" is set in the content field.On the other hand, if the child node has further child nodes, the start address (child node address) of the node information area of the child node is set.For example, the start address (child node address) of the node information area of node f is set. In the second cell, P is set in its tag field, and the start address of the node information area of node g, which is a child node, is set in its content field.

In addition, in the case of a node information area that does not have any child node address, the (n+1)th cell.

Set the address of the unsearched cell in the node information area of the next node to be searched.

Next, in step 110, the search order of the tree structure data is determined, and the next node address based on the order is set at a predetermined position in the node information area. Figure 1 shows 130
n of each node information area when applying the left-first search order for the tree structure shown in
14 shows how it is set in +1st cell (last cell)
0. In addition, n from the first cell of each node information area
The reason why the +1st cell is shaded is to avoid complication of the diagram, and that part is the same as 130. In this step, the search order may be left-first or in any other order.In other words, the prior art may be sufficient.What characterizes the present invention in this step is that the next node address based on the search procedure is placed at the end of the node information area. It is to set it in the cell. Note that end is set in the last cell of the node that is the last node in the search.

In step 120, in addition to the child node address in the node information area, the next node address set in step 110'''C is used to search for information on each node. Since the address of the node information area of the node is set in the second cell, we move on to searching for the node.The address of the node information area of node h is set in the second cell of the No. 1 to information area of node g. Since it has been set, we move on to searching for node h.

Cells 2 to 4 [1 in the node information area of node h have the node values b, c, and d set respectively, so
Search these values sequentially. Next, in the fifth (last) cell, the address of the node information area of no FP is set as the next node address, so the search for node P is started. After searching the second and third cells of node P, the last cell has the third cell of node information area of node f.
Since the address of the th cell is set, the node f
Let's move on to searching for child node a. After searching for node a, the search moves on to node K (there are two nodes g in 130 or 140 in FIG. 1, and the node g on the right side). When searching as above, the output is No. 1 as shown at the bottom of Figure 1.
You can get a sequence of values.

The following cases are omitted from the above explanation. for example,
The point is that the address set in the last cell of node g on the left in FIG. 1 is not used. This point is utilized when the search based on the present invention is used for pattern matching of a tree structure when the node value, which will be explained later using FIG. 12, is a variable (indeterminate). That is, it is used when node g makes it unnecessary to search for node h and node P in the tree structure that is the partner (target) of pattern matching.

Furthermore, although it has been explained that the process of step 120 is started after the process of step 110 is completed for all nodes,
Steps 110 and 120 may be repeated for each node. That is, this is a method of setting a next node address for a certain node and moving on to searching for the set next node.

Hereinafter, the present invention will be described as a structural pattern search system and a pro
An example in which the present invention is applied to a log language processing system will be described.

FIG. 8 is a block diagram of the tree structure pattern investigation system I constructed on a computer system.

The computer 11 searches for pattern data that matches the input data (tree structure data not including loops) input from the keyboard I 3 and outputs it to the display device 12 .

The computer 11 is provided with an input/output program 14, a conversion programmer 15, a control program 16, a pattern matching program 17, an input category area 22, and a pattern data area 19.

In the pattern data area 19, there are a plurality of pattern data 21 and a pattern data table 20 that stores their start addresses. The dashed arrows in the figure indicate the flow of control, and the solid arrows indicate the flow of data. keyboard 1
When data is input from 3, input/output coupler A i
4 passes the input data to the conversion program. The conversion program 15 converts the input data into a network type representation and stores it in the input data area 18. At this time, the next node address is statically set. Static setting is the same as the setting in step 100 of the first time. The input/output program 14 then activates the control programmer 1116. The control program 16 sequentially extracts the start addresses of pattern data one by one from the pattern data table 20 and passes them to the pattern matching program. Pattern data 2
J- is previously converted into a network type representation and stored in the pattern data area 19. (The next node address has also been set.) The pattern matching program 17 combines the human data 22 and the control program 16.
A matching process is performed with one of the pattern data 21 ' designated by .

FIG. 9 shows the processing flow of the pattern matching program.

In the initial setting 24, the starting cell address of the input data and the starting cell address of the pattern data are respectively set in the processing pointer T'+'l.

If the cell contents pointed to by the pointer P (cell TJ) and the cell pointed to by the pointer q (cell q) match in the match determination process 26 and are circles, then node search processes 28 and 30 are performed for each of the input data and pattern data. . Since both input data and pattern data are network type expressions with the next node address statically set, node search advances the pointer to the next cell, and if the cell type is a pointer cell, the point destination is moved to the next cell. It may be a cell.

If the cell after the search is not the terminal cell in end determination 32, the process returns to process 26.

FIG. 10 is a block diagram of an embodiment of the prolog language processing system.

The computer 34 reads a program 38 stored on a disk device 36 and outputs an answer 44 to a CRT display device 46 in response to a question 42 input from a keyboard 40.

The program 38 is composed of an arbitrary number of factual sentences and rule sentences. Here, in order to simplify the explanation, it is assumed that only factual sentences are constructed. Figure 10 shows two factual sentences 48.
50 is shown.

In the question input, tree structure data and operation instructions for the data can be described. The question human power 52 is f(g(z,
The question is whether there is a factual sentence that pattern matches z) and x). Here, 2°X represents a variable. The system sequentially performs pattern matching with the factual sentence 48.degree. 50, and outputs as a solution 54 that pattern matching is possible with the factual sentence 50 and that "2=a,X-indeterminate" at that time.

The next question input 56 inputs tree structure data f(g(z
, zL x) and y=f(g(z, z).

x) + variable y to find a factual statement that pattern matches ca Q Q (y )), assign variable y to variable Add a value to the program as a fact statement (asse
nt(y)). The system is ass.
At the time of ert(y), it is detected that the value of the variable y is tree-structured data including a loop, and an unfeasible (n&) is output as the solution 58.

FIG. 11 shows the software configuration executed by the computer 34 in the embodiment shown in FIG. Software controls ProGuerra! 160, a reading processing program 62, an inference program 64, a program area 66, and a data area 68. The dashed arrows in the figure indicate the flow of control, and the solid arrows indicate the flow of data.

An overview of the operation of the embodiment system will be explained using FIG. 11. The control program 60 starts a reading processing program 62 or an inference program 64 in response to a question input from the keyboard 40. Loading processing progera 11
62 is activated when "?--Read program" is input as a question input. The read processing program 62 reads the program 38 in the disk device 36 and stores it in the program area 66 in network type representation. Since there is no loop in the tree structure %11 of the progera 11, the next node address part in this network type representation is set statically.

If a question other than the above is input, the control program 6
0 stores the input contents in the data area and starts the inference program 64. Reasoning program 64 has 4th grade
- conversion processing 641, operation processing 643, and rainbow update processing 645, each of which performs pattern matching of programs and data, data changes, and block diagram update processing.

The data stored in the data area is also a network type representation, but the next node address therein is not set in advance.

FIG. 12 shows the detailed causes of the program area 66 and data area 768. The figure shows the state at the time when the program 38 in FIG. 10 is read and the question input 56 is read.

The program area includes programs 661 and 662 expressed in network format, their storage locations [and a program table 667 to manage them], and a program table 667 and a block [1 program 11 chain 6
65 is stored.

The data area 68 includes a stack area 681 used in C11 conversion processing described later and a heap area 68 for storing data.
Consists of 3. In the heap area 683, data existing in the question text or data generated during processing is stored in a network type representation. However, the next node address of each expression content is not statically set in advance.

(Shaded area in the figure) This is because data may require an expression that has a shared part or an expression that includes a loop.

FIG. 13 shows neutralization processing 641 in the inference progera 1164.
The flow of processing is shown below. The main difference from the pattern matching process (No. 9, Part 1) in the previous embodiment is the node search process 73 for cell q. Here, cell p refers to data in program area 66, and cell q refers to data in data area 68. Both the data pointed to by cell p and the data pointed to by cell q are network type expressions, but the next node address of the former is set in advance, whereas the next node address of the latter is not set. Therefore, here, the latter node search for data is assumed to be expressed in a tree format, and is performed using the conventional address saving (puch) and repopulation (pop) processing to the restack. As another method, it is also possible to search for nodes by dynamically setting the next node address of each node, similar to the update process described later, without performing push or pop.

FIG. 14 shows the flow of update processing 645. The role of the update process is to copy the data in the data area to the program area. However, the data in the data area may contain loops. On the other hand, it is not allowed to place data with loops in the program area. The update logic must check for the existence of a loop at the same time as copy processing.

The processing in FIG. 14 uses T] as a processing pointer.

d and Wk are used. p is the cell address of the data area, d is the cell address of the program (copy destination), w
k is a work pointer for temporary saving. In addition, in this example, the cell types are factor cells (cells that indicate the node values of nodes that have child nodes), atom cells (cells that indicate the node values of nodes that do not have child nodes), and pointer cells (cells that indicate the node values of nodes that have no child nodes).
A new cell type (link cell) indicating that the cell stores the next node address has been added to simplify processing.

In the initial setting 75, an area for two cells is secured in the block area 66, the first cell address is set in d, and
The node value r indicating that the th cell is the terminal node
Set endJ. ■) also has data area 68.
Set the address of the data to be copied within.

Next, the type of cell p is determined (76), and if it is a pointer cell, the next address is saved to wk, and the light pointed to by cell p is changed to 71 (response) of p (77). Link cell If so, the search for the node has ended, so the address pointed to by that cell is set as the next cell address, and the value of the cell is set to be empty (for example, the value is zero).In the case of a factor cell, its Determine the next node cell at the end of the cell (79). If it is not empty, a loop has occurred and the process ends. If it is empty, a new area is secured in the program area 66. , a pointer to the new area is set in the cell pointed to by d, and the contents of cell p (fan cover cell) are copied to the beginning of the new area (80). The address saved in Wk is set in the next node cell, which is the last cell (81).If cell p is A1~Mucell, its contents are copied to the cell pointed to by d.Atom cell,
After completing the processing of the fan lid cell, advance the p and d pointers by 1 (83), and determine whether or not the process has ended (84).
After that, the process is repeated again from process 76.

(:+0) [Effects of the Invention] According to the present invention, the next node address required when searching a node is held for each node, so in the case of tree-structured data that has no possibility of containing loops, node search can be performed. This method can be implemented at high speed without saving addresses, and in the case of tree-structured data that may include loops, it has the effect of being able to quickly determine the presence or absence of loops.

[Brief explanation of the drawing]

Figures 1, 6, and 7 are examples of data representation using the invention method.
Figures 2 and 4 are examples of wooden bamboo construction data, Figure 3 is an example of conventional data representation, Figure 5 is a conceptual diagram of the inventive method, and Figures 8 and 9 are system inputs of one embodiment. Output related diagrams and processing flowcharts, Figures 1Q, 11, 12, 13, and 14 are input/output related diagrams, software, component factors, and processing flowcharts of other embodiments. .

Claims (1)

[Claims] 1. A node value of a node of tree-structured data, a child node address indicating a storage location in memory of node information of a child node belonging to the node, and a node to be searched next after the node. A node information area consisting of a next node address indicating a storage location of information in memory is provided for each node of the tree structure data, and a node value representing the given tree structure data and a child node address are stored in the node information area. a first step of setting the search order of the tree structure data to a predetermined position in the node information area; a second step of determining the search order of the tree structure data and setting the next node address to a predetermined position of the node information area; a third step of sequentially referring to each node information of the n-ary tree structure data using the next node address when the next node address is used. 2. Tree-structured data according to claim 1, characterized in that node information is provided with a node information area consisting of a sequence of m+2 cells corresponding to a node having zero or more m child nodes. node search method. 3. As types of cells representing node information, funta cells representing the node values of nodes with child nodes and cell atom cells representing the node values of nodes without child nodes are provided, and node information of nodes without child nodes is one cell. In the node information that is expressed by an atom cell and has a node that does not have a child node as a child node, an atom cell that indicates the value of the child node is used instead of an address cell that indicates the child node address. A node search method for tree-structured data. 4. A method for searching for nodes in tree-structured data according to claim 1, characterized in that the second step and the third step are executed alternately for each node. 5. In the node search method for tree-structured data described in claim 4, in the third step, when the next node is reached according to the next node address, the referenced next node address is deleted and the next node address is deleted. In step 2, it is determined whether or not a next node address is set, and when the next node address is set, a recursive loop within the tree structure is determined to exist. Search method. 6. In a processing method for a symbolic processing language that handles tree-structured data, a node value of a node of given tree-structured data and a child node address indicating a storage location in memory of node information of a child node belonging to the node; A node information area is provided for each node of the given tree structure data to store node information consisting of a next node address indicating a storage location in memory of the node information of the node to be searched next to the given node; a first step of setting a node value and child node address representing each node of the tree-structured data in a predetermined position in the node information area; and setting a next node address of the node in a predetermined position in the node information area of the node. a second step of determining whether or not a next node address is set at the position; and, if the result of the determination is that the next node address is not set, a predetermined search procedure for the given tree structure data; a third step of setting the next node address of the node in a predetermined position of the node information area of the node based on the node information area of the node; Therefore, a fourth step of searching for the next node and erasing the next node address; repeating the second to fourth steps for each node, and determining the next node address as a result of the determination in the second step. is set, a fifth step of assuming that an inner loop exists in the given tree structure data and displaying a node corresponding to the inner loop in a manner different from others. A node search method for tree-structured data.