US 6,292,804 B1Grant
Object relationship management system
Issue Date:2001-09-18
•23 Claims
•13 Drawing Sheets
Abstract
A method for maintaining relationships between entities in a computer system, each entity having a plurality nodes, includes the steps of: modifying one of the plurality of nodes; searching for a plurality of dependent nodes from the plurality of nodes coupled to the one node; ordering the plurality of dependent nodes into an order; and evaluating the plurality of dependent nodes in the order.
Metadata
Assignee
- Intergraph Corporation
Inventors
- Jean-Louis Ardoin
- Richard M. Eade
- Robert Patience
- Alain Falasse
- Dave L. Brann
- Gerard J. Attilio
- Alfredo Arce
Application Information
Application Number:US 09/459,812
Filing Date:1999-12-13
Priority Date:1995-05-09
Art Unit:7
Classifications
IPC:
G06F 1730
Field of Search:
707102
Patent Drawings (13 sheets)
Description
CROSS REFERENCE TO RELATED APPLICATIONS
[0002] This application is a continuation of U.S. Application Ser. No. 08/937,147, filed Sep. 24, 1997 (U.S. Pat. No. 6,052,691), which is a continuation of U.S. Ser. No. 08/437,942, filed May 9, 1995 (U.S. Pat. No. 5,692,184), which claims priority from 08/378,251, filed Jan. 23, 1995 (U.S. Pat. No. 5,682,468).
[0003] This patent application is related to and incorporates by reference the subject matter of commonly owned and assigned co-pending U.S. patent application Ser. No. 08/378,251 entitled “OLE For Design and Modeling”, filed Jan. 23, 1995, by Mark Fortenbery, Robert Patience, Dominique Payannet, and Cameron Stubbs, which is directed towards extensions of OLE enabling the transfer of two and three-dimensional data objects between computer-aided design and computer-aided manufacturing (CAD/CAM) software applications.
COPYRIGHT NOTICE
[0004] A portion of the disclosure of this patent contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever
BACKGROUND OF THE INVENTION
[0005] The present invention relates generally to the area of computer-aided design and computer-aided manufacturing (CAD/CAM) software, and specifically to a general object relationship management subsystem. More specifically, the present invention relates to a system that maintains data integrity when an object is modified, maintains referential integrity when an object is copied or deleted, defines the guidelines for programming behavior when an object is copied or deleted, encapsulates the details of how relationships are stored, and provides a mechanism for undoing or reversing actions taken on the object model through the relationship management system.
Object Linking and Embedding (OLE) Overview
[0006] Within the office environment, one method that has been developed to enable “cutting” and “pasting” of data between software applications is OLE. OLE defines standardized interfaces and functions enabling users to transfer objects supporting the Component Object Model (COM) between software applications. The following section is an abbreviated overview of some of the concepts used in COM and OLE version 2.0, from Microsoft Corporation of Belleview, Washington, and defines some of the terms that will be used in the disclosure. Further information and details about COM and OLE may be obtained from “Inside OLE 2” by Kraig Brockschmidt, 1994, Microsoft Press, hereby incorporated by reference.
[0007] An example of cutting and pasting data between software applications is illustrated in FIG. 1. FIG. 1 illustrates a two-dimensional object 1 created in a first software application being transferred into a second software application. The first and second software applications (not shown) are commonly specialized software applications such as spread-sheets, word processors, or graphics programs. Once two-dimensional object 1 has been transferred, the second software application can manipulate its own data, two-dimensional object 2, so that two-dimensional object 2 interacts with two-dimensional object 1. The resulting document is then output to the user.
[0008] OLE provides a set of “interfaces”, or groups of functions, which when combined provide the mechanics enabling the user to transfer data between programs. FIG. 2 illustrates the convention for representing an OLE interface 10, for an object 11 and a “consumer” 12 of the object. Object 11 is said to have an “interface implementation”, including interfaces 13 and 14, that are analogous to an object oriented programming “class.” Interfaces 13 and 14 include member functions 15 and 16, respectively, that are analogous to object oriented programming class “instances”.
[0009] Consumer 12 receives data from object 11 by calling functions of interface 13 and/or interface 14. In some cases the consumer may only be aware of one of several interfaces available in an object. In response to the function calls, object 11 may return specific data about itself to the consumer 12. Object 11, however maintains exclusive control of its own data 17. As further illustrated in FIG. 2, IUnknown is an interface available to all objects, that when queried for specific interfaces, returns pointers to the requested interface. For example, assuming consumer 12 knows which functions are available in interface 14, consumer 12 can ask for and receive a pointer to interface 14. Then, once consumer 12 receives a pointer to interface 14, consumer 12 can call member functions 16.
CAD/CAM Market Overview
[0010] Application software specifically designed for architectural and engineering and mechanical design and fabrication purposes are commonly labeled Computer Aided Design (CAD) and Computer Aided Manufacturing (CAM) software. Some of the standard features of CAD/CAM applications is the ability to create and manipulate three-dimensional objects and to create relationships between three-dimensional objects relative to other three-dimensional objects.
[0011] CAD/CAM applications typically fall into generic types, those built upon a standard graphics platform where data and code is shared among many applications and proprietary applications which have very little in common with other CAD/CAM applications. Within the domain of applications built on generic graphics platforms some have crude capabilities for defining relationships between CAD objects. Typically the user must place objects in a location that represents the results of a relationship acting on that object, that is, the user performs the computation external to the application to determine the physical location of the new object. In many proprietary applications, there exist software that manages geometric relationships between CAD objects. Typically, these capabilities (called constraint management capabilities) allow for the definition of geometric constraints on CAD objects and between CAD objects. For example, a typical constraint between two CAD objects is maintaining a given distance between the objects or maintaining one face parallel to another face. One problem with these type of limited constraint management systems is that they only represent spatial relationships and they cannot model or maintain a more general class of relationships.
CAD/CAM Applications and Database Structures
[0012] In the past, CAD/CAM software applications included specialized software which allowed for the modeling of a limited set of relationships between objects. In many cases CAD/CAM systems did not model general relationships at all and only allowed the user to define positional relationships which essentially placed objects at specific coordinates in 3D space. In some systems “constraint management” capabilities have been added, however these are restricted to only modeling and managing geometric relationships between objects. Other relationship types that needed to be in the model were handled in an ad hoc manner by the user as CAD/CAM applications did not have generalized relationship modeling and maintenance capabilities.
[0013] What is needed is a relationship management subsystem that allows generalized relationships (including geometric relationships commonly present in current CAD/CAM systems) to be defined and that maintains the integrity of the object model.
SUMMARY OF THE INVENTION
[0014] The present invention provides enhancements and extensions to OLE for the CAD/CAM environment that are used for a general object relationship management subsystem. The “Object Relationship Management System” discloses extensions to the OLE COM object model for modeling and maintaining relationships that can be modeled between COM objects. The relationship types and COM object types are not limited to CAD/CAM systems by the Object Relationship Management System. The relationship subsystem maintains data integrity when an object is modified, maintains referential integrity when an object is copied or deleted, defines the guidelines for programming behavior when an object is copied or deleted, encapsulates the details of how relationships are stored, and provides a mechanism for undoing or reversing any modification done by a modify, copy, or delete operation and is accomplished through the Object Relationship Management System.
BRIEF DESCRIPTION OF THE DRAWINGS
[0015] FIG. 1 illustrates a two-dimensional object 1 created in a first software application being transferred into a second software application;
[0016] FIG. 2 illustrates the convention for representing an OLE interface, for an object and a “consumer” of the object;
[0017] FIG. 3 is a block diagram of a system according to a preferred embodiment of the present invention;
[0018] FIG. 4 illustrates a flow diagram of one embodiment of the process of recomputing the graph when a modification is made;
[0019] FIG. 5 illustrates a flow diagram of one embodiment of the process of recomputing the graph after a node is edited;
[0020] FIG. 6 illustrates a simple graph with nodes and relations;
[0021] FIG. 7 illustrates new values of the graph in FIG. 6 in response to the modification of the value of node “a”;
[0022] FIG. 8 illustrates two COM objects: a dimension and a line;
[0023] FIG. 9 illustrates a graphic representation of the COM relations for a dimension and a line of FIG. 8;
[0024] FIG. 10 illustrates a graphic representation of the COM relations for a parallel constraint;
[0025] FIG. 11 illustrates two lines having a parallel constraint and a dimension having a graphic handle;
[0026] FIG. 12 illustrates a graphic representation of the COM relations of FIG. 11;
[0027] FIG. 13 illustrates an example of two pipes connected by a connector;
[0028] FIG. 14 illustrates a graphic representation of the COM objects of FIG. 13;
[0029] FIG. 15 illustrates a graphic representation of the COM relations of FIG. 13;
[0030] FIG. 16 illustrates a flow diagram of one embodiment of the process of searching the associative graph for all nodes that need to be recomputed and sorting them into the correct order for recompute;
[0031] FIG. 17 illustrates a flow diagram of one embodiment of the process of copying objects;
[0032] FIG. 18 illustrates a flow diagram of one embodiment of the process of the deleting process;
[0033] FIG. 19 illustrates three pipes connected to a tee connector in a constraint relationship;
[0034] FIG. 20 illustrates three pipes connected to a tee connector in a constraint relationship being modified;
[0035] FIG. 21 illustrates a graphic representation of the COM relations of FIG. 20;
[0036] FIG. 22 illustrates three pipes connected to a tee connector in a constraint relationship where two pipes are selected for being copied;
[0037] FIG. 23 illustrates a graphic representation of the COM relations of FIG. 22 before the copy process;
[0038] FIG. 24 illustrates three pipes connected to a tee connector in a constraint relationship where one pipe is selected for deletion;
[0039] FIG. 25 illustrates a graphic representation of the COM relations of FIG. 24 before the deletion process is started;
[0040] FIG. 26 illustrates a graphic representation of the COM relations of FIG. 24 after the pipe has been deleted;
[0041] FIG. 27 illustrates a flow diagram of one embodiment of the process of the Undo feature; and
[0042] FIG. 28. illustrates a flow diagram of one embodiment of the application of the Relations subsystem described above.
DESCRIPTION OF THE PREFERRED EMBODIMENT
[0043] The goals of the herein described Object Relationship Management Subsystem (Relations subsystem) are to maintain data integrity when an object is modified, maintain referential integrity when an object is copied or deleted, define the guidelines for programming behavior when an object is copied or deleted, encapsulate the details of how relationships are stored, and provides a mechanism for undoing or reversing any modification done by a modify, copy, or delete operation through the Relations subsystem
[0044] These goals are accomplished by unifying “Modify”, “Copy” and “Delete” “Semantics” in a single relationship model. Semantics defines the behavior for a known operation. For example, copy semantics of a relation are invoked by the Relations subsystem to define and operate on the values coupled to the relation. A single relationship model allows new types of relationships to be added to the system without breaking existing code. The alternative would be a special case solution in which adding a new type of relationship to the system would break existing copy, delete and/or modification commands.
[0045] COM objects are related to one another via their IUnknown pointers. The details of how the relationship is maintained persistently (if the objects it connects are persistent) are encapsulated behind two COM interfaces: IJRelationValue on the object side, and IJRelation on the relation side, as will be described below. The IUnknown pointer is the only object identity exposed in the API. The Data objects are given this ability by incorporation of the described OLE interfaces and functions below.
[0046] It is believed that one familiar with OLE, as described in “Inside OLE 2” by Kraig Brockschmidt, would be able to readily use the following COM interfaces based upon the following disclosure. Further details about the preferred method of enabling such transfer of data are detailed in the following sections.
[0047] The examples illustrated in this application are based on a typical CAD/CAM application data.
System Overview
[0048] FIG. 3 is a block diagram of a system 60 according to a preferred embodiment of the present invention. System 60 includes a display monitor 61, a computer 62, a keyboard 63, and a mouse 64. Computer 62 includes familiar computer components such as a processor 65, and memory storage devices such as a random access memory (RAM) 66, a disk drive 67, and a system bus 68 interconnecting the above components. Mouse 64 is but one example of a graphical input device, a digitizing tablet 65 is an example of another.
[0049] In a preferred embodiment, system 60 includes a IBM PC compatible personal computer, running Windows™ operating system Version 3.1 and OLE 2.0 by Microsoft Corporation, and Jupiter™ software, currently under development by Intergraph Corporation. The appendix includes preferred embodiments of COM interfaces described below, written in C++. The appendix also includes sample source code programs that incorporate the Jupiter™ OLE extensions.
[0050] FIG. 3 is representative of but one type of system for embodying the present invention. It will be readily apparent to one of ordinary skill in the art that many system types and configurations are suitable for use in conjunction with the present invention.
[0051] I. Overview of Relations
[0052] The Relations Subsystem includes new interfaces and functions given to COM objects, as well as functions executed by Relations machines. Four Relations machines currently exist: a Modify machine, a Copy machine, a Delete machine, and an undo machine.
[0053] 0. Definitions
[0054] 1. The Associative Graph
[0055] 2. The Associative Graph in COM
[0056] 0. Definitions
[0057] value: A value is an object, analogous to an entity in a Entity-Relation Model.
[0058] relation: A relation is the means by which one object refers to another object. A relation is defined by one or more connections and a set of semantics that define its behavior.
[0059] connection: A connection is a link between a value and a relation.
[0060] semantic: A semantic is a COM interface that defines some aspect of the relation such as its copy or delete behavior. Each value in a relationship can enumerate all the relations it participates in, and each relation semantic can enumerate the values that it defines behavior for.
[0061] Each relation must provide a “Copy” and a “Delete” semantic. A “Modify” semantic, known also as a “Compute” semantic, is optional and provides the relation with the ability to recompute an output value of the object based on changed input value(s). This will be discussed in greater detail later in this document.
[0062] 1. The Associative Graph
[0063] Associativity and constraints deal with relationship management. The purpose of the Relations subsystem is to establish, maintain and dissolve relationships between entities. The Relations subsystem notifies related entities whenever something in a relationship changes. There are two types of relationships: dependency and equivalence.
[0064] In a dependency relation (also called a directed associative relationship), an entity depends on one or more other entities. when something changes dependent entities are notified in order of dependency. A dependency relation is illustrated with a cell in a spreadsheet that calculates the sum of a row of numbers: whenever a number in the row is modified, the sum is updated. Because the sum depends on the numbers in the row, one cannot directly edit the sum for the sum only exists as the output of the dependency relation. To change the sum the user must change one or more of the values the cell depends upon or make the cell depend on a new set of numbers. Dependency is a one-way relation such that the output of a dependency is changed by changing its input. Further, dependency relations can be built upon other dependency relations in an ordered manner.
[0065] In an equivalence relation (also called a constraint relationship), the related entities are equally dependent and are notified simultaneously when something changes. This ordered or simultaneous notification distinguishes constraints from dependencies. An equivalence relation is illustrated with a CAD system that creates a parallel constraint between two lines: line1 and line2. When the user grabs line1 with a mouse and moves it, the system moves line2 so that it remains parallel to line1. Conversely, when the user moves line2, the system moves line1 to remain parallel. In Relations subsystems terms, the lines are in an equivalence relation. Notice that in the dependency example above the sum could not be directly modified, whereas in the equivalence relation example either line can be modified. Equivalence is not a one-way relation, thus there is no implied ordering. To update equivalence relations, a set of entities and their related constraints are examined as a whole, then a decision is made as to which entities can be modified to satisfy all of the constraints.
[0066] The Relations subsystem describes the associative and the constraint relationships in a conceptual graph called an associative graph. This associative graph is used to maintain the integrity of relationships when entities of the associative graph are modified, deleted, copied, etc. The Relations subsystem creates and manipulates these graphs and maintains the information used to create them.
[0067] The associative graph consists of nodes that are connected together by edges. These nodes represent the related entities and the edges represent the relationships (either dependency or equivalence) between the entities.
[0068] 1.1. Nodes
[0069] Each node in an associative graph is distinguished by a unique identifier. A node can represent a value, an associative evaluation function or a predicate as will be described below.
[0070] 1.1.1. Value Nodes
[0071] A “value” node represents a storage. It can be used as:
[0072] an input of an associative function or predicate;
[0073] an output of an associative function; or
[0074] a constrained argument of a predicate.
[0075] A value node can be thought of as a “passive node” of the associative graph. It is a repository to obtain or provide a value. This is in contrast to a node containing an associative function or predicate that is “active” in the sense that it recomputes.
[0076] 1.1.2. Associative Function Nodes
[0077] An “associative function” node is used to establish dependency relations between value nodes. The dependency is such that output values depend on input values. When an associative function is evaluated, it entirely replaces the output value nodes. For this to produce predictable results:
[0078] A value node cannot be the output of more that one associative function node. If this were the case, the output values would never be consistent. The value would only represent the result of which ever function was evaluated last.
[0079] The output value nodes can only be changed by re-evaluation of the associative function node. As seen in the spreadsheet example above, the user cannot directly edit an associative output. Any direct modification of an output would be over-written the next time the associative function node is evaluated.
[0080] As an example, suppose a function that sets the start and end points of a line: “LineBy2Pts(Line, P1,P2)” is an associative function node that has a line for its output. When called, this function updates the geometry of the line with the input values P1, P2. The node containing the function would have a dependency relation with the two points, P1 and P2. If either of the two points changed, the function node would be notified and the evaluation function would recompute a new value for the line. If the line is modified directly, without modifying the associative function, such modifications would be lost the next time the LineBy2Pts function is evaluated.
[0081] 1.1.3. Predicate Nodes
[0082] A “predicate” node is used to define an equation or a constraint between value nodes. A predicate node represents a boolean (TRUE/FALSE) value that indicates if the equation or constraint is satisfied, according to the inputs. Several predicates can be associated to the same input value node, thus unlike an associative function, a predicate only partially defines a value(s). An external mechanism, called a solver, is used to modify the input value nodes until all predicates relating to the input value nodes are TRUE. An exception to this is if an input value node that is the output of an associative function is used as an argument of a predicate, the input value node cannot be modified directly by a solver. In the preferred embodiment of the present invention, the solver treats these sort of input value nodes as “constant”.
[0083] As an example, a predicate node is TRUE or FALSE depending if the beginning point of a line L0 matches a given point P1: “IsBeginPoint(L0,P1)”. The geometry of the line is only partially defined by this predicate. If L0 or P1 changes, the predicate passes input values L0 and P1 to a solver. The solver modifies L0 or P1 until L0 has a beginning point equal with P1, i.e. until TRUE. If there are other input values that are related by other predicates to L0 and P1, these other input values along with the other predicates are also passed to the solver at the same time.
[0084] 1.2. Edges
[0085] An edge connects a value node to a function node. (Two value nodes or two function nodes cannot be connected by an edge.) Edges are:
[0086] oriented, forming a dependency relationship; or
[0087] non-oriented, forming an equivalence relationship.
[0088] A “non-oriented” edge always connects a value node (that is not the output of an associative function) to a node containing a predicate (constraint or equation). The value is a non-constant parameter of the predicate, meaning it can be modified by a solver. Such a relationship is an equivalence relationship because a set of values and their related predicates are treated equivalently by a solver during the evaluation of the associative graph and of the values may be modified.
[0089] There are two cases of oriented edges:
[0090] The edge is oriented from a value node to a function node, i.e. the function depends on the value node as input; and
[0091] The edge is oriented from a function node to a value node, i.e. the value node is the output of an associative function node.
[0092] When a value node is an input to a function node and the function node produces a new output, the value node is referred to as the “parent” and the function node as the “child”. Notice, children depend on the value of their parents, thus parents must be evaluated before their children. When an oriented edge connects an associative function node to its output value node(s), the function node is the “parent” and the value node is the “child”.
[0093] When a value node is connected to a predicate by oriented edges, the value cannot be modified by a solver. An example of this is when the value node is the output of an associative function, or when it comes from a read only document or for reasons dictated by application semantics.
[0094] A loop is a set of oriented edges that form a cycle in the graph. If a loop is detected in the graph, all nodes in the loop are treated as nodes connected by equivalence relations, thus all the edges in the loop are treated as non-oriented edges and a solver is used to evaluate the nodes.
[0095] 1.3. The Recompute Machine
[0096] Recompute is the process by which the value nodes in the graph are kept consistent. In the case of an associative evaluation function, the associative function re-evaluates (or computes) the value of its output node(s) whenever an input node changes. In the case of a predicate, when a value node changes, the set of values related via predicates is modified by a solver to make the predicates TRUE.
[0097] FIG. 4 illustrates a flow diagram of one embodiment of the process of recomputing the graph when a modification is made:
[0098] Accumulate modified nodes, step 100.
[0099] A node can be modified because:
[0100] a value node was modified by the user; or
[0101] a function node has its associative behavior modified by changing its evaluation function, changing its parameters, editing its parents or changing its equivalence relations.
[0102] Search for all nodes to evaluate, step 110. Nodes to evaluate are nodes which depend directly or transitively on the modified nodes. In other words, search for the nodes that are either directly or indirectly related to modified nodes as children of dependency relations or by equivalence relations.
[0103] Order the nodes to evaluate, step 120. This step insures that a function is re-evaluated after the values it depends upon are evaluated. Such ordering is analogous to ordering numeric variables related by operators “less than” (<) and “equal to” (==) into ascending order: (<) being the directed parent/child (dependency) relation, and (==) being the equivalence relation. Thus a parent is “less than” its child and is evaluated before the child, and all related predicates are “equal to” each other and are evaluated simultaneously by a solver.
[0104] Evaluate the ordered function nodes, step 130 (value nodes don't recompute)
[0105] A node with an associative evaluation function is evaluated by calling its evaluation function.
[0106] Predicate nodes and related value nodes are passed to a solver. It is up to the solver then to satisfy the predicates by modifying the value nodes that are related by the predicates.
[0107] 1.3.1. Editing The Graph During The Recompute of A Node
[0108] Re-ordering of an associative graph is necessary when a node changes the relations on itself or on other nodes during a recompute. When relations are changed, the order in which nodes are computed may have to be re-established. The changing of relationships is equivalent to editing the associative graph.
[0109] FIG. 5 illustrates a flow diagram of one embodiment of the process of recomputing the graph after a node is edited, step 150:
[0110] The graph is incrementally searched if the computed node adds some new dependency, step 160. Incremental means adding to the list of elements to be recomputed incrementally. Elements will not be removed from the recompute list because of this editing, but elements currently in the list to be recomputed could choose not to recompute if their parents did not change value either. This is an optimization mechanism known as “graph triggering”.
[0111] The graph is resorted if the computed node modifies dependency relations, step 170.
[0112] Editing the graph during recompute is always allowable if the edited element is the current node being computed or depends directly or transitively on the node being computed. This limitation ensures that the edited node has not yet been evaluated. If the edited node does not depend (before the edit) on the node being recomputed there are risks that:
[0113] The edited node could have been already computed thus the Relations subsystem will not re-evaluate the node to take into account its new relations. In the best case, the Relations subsystem will identify it as an error, in the worst case the data of the model will no longer be consistent.
[0114] The edited node could have dependent nodes already computed. This will also result in an error.
[0115] When the edited node is not the current node being computed or one of its dependents, the application must assume responsibility for the consequences of its graph edits.
[0116] 1.4. The Copy and Delete Machines
[0117] The relationships between nodes need to be modified whenever a node(s) is deleted, copied, or an action is “undone”. The copy and delete machines are responsible for ensuring the integrity of the graph is maintained. The actions taken are relationship specific, and the modification behavior may be different on a per node basis. For more detailed information on how the Relations subsystem maintains graph integrity, refer to section III, below.
[0118] On a node delete: A node related to a node being deleted is made aware of the deletion in order to rectify the information it stores and adjust its behavior. Some behaviors the node may exhibit are:
[0119] keep the relation to the node being deleted, i.e. to prevent the deletion;
[0120] modify its evaluation function so as not to refer to the node being deleted;
[0121] remove its evaluation function completely (transform to non-associative); and
[0122] delete itself.
[0123] On a node copy: a “prune graph” operation is performed on the initial graph to ensure that a meaningful set of elements is copied. The relationships between nodes in the copied graph should be the same as the relationships in the original graph. In other words, when a child is copied, the parents of the child should be copied as well. Alternatively the copy of the child should be hooked up to an existing parent.
[0124] 1.5. Simple Example
[0125] FIG. 6 illustrates a simple graph with nodes and relations. Nodes 200, 210, 220, 230, and 240 are value nodes, Nodes 250 and 260 are predicate nodes, and node 270 is an associative function node. Edges 280, 290, 300, and 310 represent equivalence relations, and edges 320, 330, and 340 represent dependency relations, in the direction shown.
[0126] FIG. 7 illustrates new values of the graph in FIG. 6 in response to the modification of the value of node “a” 200. Following the recompute process disclosed above, if the node “a” 200 is modified to have a value of “2”, the Relations subsystem performs the following steps:
[0127] accumulates the modified nodes. In this case, only one node is modified “a” 200.
[0128] searches for all related nodes to evaluate. This results in the nodes: “a” 200, “Z” 250, “c” 210, “Y” 270, “b” 220, “X” 260, “d” 230, and “e” 240.
[0129] orders the nodes to evaluate. Given the relationships above:
[0130] “a” 200, “Z” 250 and “c” 210 are equivalents;
[0131] “a” 200 must be evaluated before “Y” 270, “Y” 270 must be evaluated before “b” 220, and “b” 220 must be evaluated before “X” 260, thus are dependents; and
[0132] “d” 230, “X” 260 and “e” 240 are equivalents.
[0133] This ordering can be expressed as: a==Z==c<Y<b<X==d==f.
[0134] evaluates the ordered nodes:
[0135] 1. “a” 200 , “Z” 250 and “c” 210 are passed to a solver to satisfy the predicate “Z” 250. In this case, “c” 210 is changed to 10 to satisfy the predicate “Z” 250.
[0136] 2. The associative evaluation function “Y” 270 is then called on to calculate a new value for “b” 220. Since “a” 200 is 2, the new value of “b” 220 is 4.
[0137] 3. “d” 230, “X” 260 and “e” 240 are passed to a solver to satisfy the predicate “X” 260. Since “b” 220 is 4, “d” 230 can be set to 4 to satisfy the predicate. Alternatively “d” 230 could remain at 0, and “e” 240 can be set to 8.
[0138] Notice that even though “e” 240 was considered as part of the recompute “e” 240 may not need to change. Further, if there were other nodes dependent on “e” they would not necessarily change either. (A solver can have more than one way to change values to satisfy predicates.) It is an important optimization to recognize which values do not change, in a graphics system, for instance, this can avoid needless erases/displays.
[0139] 2. Associative Gragh in COM
[0140] The following section describes how the associative graph is realized using the component object model (COM). The following assumptions are initially made about the Relations subsystem in COM:
[0141] A single COM object can represent more than one node of the associative graph. Multiple nodes per COM object allows a “granularity” of recompute. For example, a recompute of a geometric dependency of an element would not be necessary if the symbology changes; and
[0142] All higher-level application relationships are broken down into the two basic relations: dependency and equivalence. These are the only relationships managed by the Relations subsystem.
[0143] 2.1. Nodes in COM
[0144] Each value node of the associative graph corresponds to a single COM interface. A COM object can support more than one interface and can represent more than one value node. Because of the interface granularity, a recompute semantic can enumerate as a connected value the same IUnknown pointer more than once with different IIDs. An interface represents a value node only if it is used as input or output argument of a function node. A function node is also known as a recompute semantic. A COM object can have more than one recompute semantic. For example, a recompute semantic between A and B could be represented by (A interface w) as input and for output (B interface x). another recompute semantic for the same relation could use (B interface y) as input and (A interface z) for output. Something that will not happen for the copy and delete semantics since a COM object can not be “half copied” or “half deleted”. These functions nodes are enumerated by the Relation COM object
[0145] The components of a node (identity, and value or function) are represented in COM as follows:
[0146] The identity of a value node is defined by the persistent identity of the COM object representing the node and the IID of the interface that is used to set or get the value.
[0147] The value of a node is not explicit. It is represented by the interface functions that are used to set or get the value. (This is in accordance with COM where all data is private.) Associative functions and predicate solvers use COM interfaces to request the values for input or output.
[0148] The identity of a function node is defined by the persistent identity of the COM object representing the node and a key returned from the COM object representing the node when the COM object it enumerates its function nodes.
[0149] The recompute semantic of a function node is defined by a Relations interface. The Relations subsystem uses a function on this interface to recompute the node.
[0150] Value nodes that represent the output of an associative function may come from the same COM object as the function node. Value nodes which are input arguments, or constrained argument of a predicate can come from different COM object than the function node.
[0151] 2.2. Edges in COM
[0152] A COM object representing value nodes can determine which other COM object(s) depend on it or have an equivalence relationship with it. Further, a COM object representing function nodes can determine the value nodes (COM identity, Interface id) that each function node that it contains depends upon or is in equivalence with.
[0153] Some important limitations of Edges in COM:
[0154] The information stored on both sides of a relation is not symmetrical:
[0155] A value node only knows the COM object that holds the relation not the identity of the function nodes that refer to it.
[0156] A function node knows the identity of the value nodes it refers to.
[0157] The Relations subsystem defines interfaces to access the relationship information. The implementation of these interfaces determines where the actual information is stored. The Relations subsystem provides a set of “helper” classes that implement the Relations interfaces (see Appendix):
[0158] The information stored by function nodes (COM identity, interface id) is stored in a COM “helper” object aggregated by the application object.
[0159] The information stored by value nodes (COM identity of function node holder) is maintained by a COM helper object aggregated by the application object.
[0160] II. Relationship Subsystem COM Interfaces
[0161] The following are a more detailed explanation of the OLE extensions to COM described above.
[0162] 1. IJRelation Interface
[0163] 2. IJRelationValue Interface
[0164] 3. IJCopy Interface
[0165] 4. IJDelete Interface
[0166] 5. IJCompute Interface
[0167] 6. IJRelationGraph Interface
[0168] 7. IJRelationNotify Interface
[0169] 8. IJRelationGlobal Interface
[0170] 9. IJRelationSolver Interface
[0171] 10. IJComputeCFunction Interface
[0172] 11. IJUndo Interface
[0173] 12. IEnumJCHKPNT Interface
[0174] 13. IJUndoManager Interface
[0175] 14. Other Interfaces
[0176] 1. IJRelation Interface
[0177] The IJRelation interface is supported by any object that wants to define a relationship with one or more other objects. A relationship is defined by its copy, delete, and recompute (modify) behavior. These three behaviors (modify, copy, and delete) are referred to collectively as the semantics of the relationship. The purpose of the IJRelation interface is to manage the relation semantics and to connect or disconnect values to and from the relation. Each type of semantic has a corresponding software module, called an machine, provided by the Relation subsystem. These machines use IJRelation::EnumSemantics to interrogate the semantics of the relationships. An object implementor is responsible for implementing IJRelation and the necessary semantic interfaces listed below.
[0178] The copy semantic of a relation defines what happens when objects connected to the relation or the relation itself is copied. The copy machine that the Relation subsystem provides is responsible for all copy operations. It enumerates relation semantics via IJRelation::EnumSemantics and queries each semantic for IJCopy. The copy machine uses the IJCopy interface to complete the copy operation.
[0179] The delete semantic of a relation defines what happens when objects connected to the relation or the relation itself is deleted. The delete machine that the Relation subsystem provides is responsible for all delete operations. It enumerates relation semantics via IJRelation::EnumSemantics and queries each semantic for IJDelete. The delete machine uses the IJDelete interface to complete the delete operation.
[0180] The recompute semantics of a relation define what happens when objects connected to the relation are modified. A relation can define multiple recompute semantics. The recompute machine that the core provides enumerates relation semantics via IJRelation::EnumSemantics and queries each semantic for IJCompute. The recompute machine uses the IJCompute interface to complete the recompute operation.
[0181] The copy/delete/recompute machines only access semantics through IJRelation::EnumSemantics. A relation can choose to store semantics as physical objects, or it can create the semantics dynamically when asked to enumerate them. An implementation of IJRelation can choose to return an error from AddSemantic and RemoveSemantic to disallow semantics to be added or removed.
[0182] The following is a description of the preferred embodiment of the formal IJRelation interface:
[0183] 1.1. IJRelation::AddSemantic
[0184] IJRelation::AddSemantic adds a semantic to a relation and connects values enumerated by the semantic to the relation. A single semantic object can support multiple semantic interfaces (e.g. IJCopy and IJDelete). In this case, the semantic should only be added to the relation once.
[0185] The implementation of this function performs the following:
[0186] Stores the specified semantic and handles its persistence;
[0187] Connects to all values enumerated by the specified semantic;
[0188] Performs QueryInterface for IJCopy, IJDelete, and IJCompute; and
[0189] Connects the values enumerated by these interfaces to the relation with IJRelationValue::ConnectRelation.
[0190] The following is a description of the preferred embodiment of the formal AddSemantic interface:
[0191] 1.2. IJRelation::EnumSemantics
[0192] IJRelation::EnumSemantics is used to enumerate the semantics of a relation. The core copy, delete, and recompute machines use this function to interrogate the behavior of a relationship.
[0193] The returned enumerator specifies a “key” identifying each semantic, as described above. This key can be used as input to GetSemantic or RemoveSemantic and is invalidated if the semantic is removed.
[0194] The following is a description of the preferred embodiment of the formal EnumSemantics interface:
[0195] 1.3. IJRelation::GetSemantic
[0196] IJRelation::GetSemantic returns a pointer to the IUnknown interface of the relation semantic identified by the key. The key is returned from AddSemantic or EnumSemantics. The returned pointer must be released before being discarded.
[0197] The following is a description of the preferred embodiment of the formal GetSemantic interface:
[0198] 1.4. IJRelation::GetType
[0199] IJRelation::GetType gets the “type” of the relation. Each relation has an associated type that can be used to infer information about the relation. The type is never interpreted by the MCD machines but it is defined by the relation implementor as a “hint” as to what the relation is used for.
[0200] The following is a description of the preferred embodiment of the formal GetType interface:
[0201] 1.5. IJRelation::RemoveSemantic
[0202] IJRelation::RemoveSemantic removes a semantic from a relation and disconnects any values enumerated by the semantic. This function is the inverse of AddSemantic. Any implementation of this function should QueryInterface for IJCopy, IJDelete, and IJCompute and disconnect the values they enumerate by calling IJRelationValue::DisconnectRelation.
[0203] The following is a description of the preferred embodiment of the formal RemoveSemantic interface:
[0204] 1.6. IJRelation::SetType
[0205] IJRelation::SetType sets the type of the relation. Each relation has an associated type that can be used to infer information about the relation. The type is never interpreted by the MCD machines, it is defined by the relation implementor as a hint to what the relation is used for. SetType can only be called once on a relation. Some relation definitions can choose to return an error from SetType.
[0206] SetType is most useful for the standard relation helper where a set of semantics and a type can be added from “outside” an object.
[0207] The following is a description of the preferred embodiment of the formal SetType interface:
[0208] 2. IJRelationValue Interface
[0209] The IJRelationValue interface should be supported by any object that wants to participate in a relationship with another object. It is recommended that all objects support this interface. The IJRelationValue interface manages the list of relations that are connected to a value.
[0210] A helper object (referred to above is obtained through ::CreateStdRelationValue) has been provided that implements this interface. This helper is intended to be a virtual aggregate, the aggregator does not store a pointer to the helper object. No overhead is required in the aggregator itself.
[0211] The following is a description of the preferred embodiment of the formal ReplaceValue interface:
[0212] 2.1. IJRelationValue::ConnectRelation
[0213] IJRelationValue::ConnectRelation establishes a connection between a value and a relation. This function is called by a relation as part of its implementation of IJRelation::AddSemantic from some other function that defines how objects are connected to a relationship.
[0214] The following is a description of the preferred embodiment of the formal ConnectRelation interface:
[0215] 2.2. IJRelationValue::DisconnectRelation
[0216] IJRelationValue::DisconnectRelation disconnects an object from a relation. This function is only called by a relation as part of its implementation of IJRelation::RemoveSemantic, by IJDelete::Disconnect or from some other interface function that defines how objects are disconnected from a relationship.
[0217] The following is a description of the preferred embodiment of the formal DisconnectRelation interface:
[0218] 2.3. IJRelationValue::EnumRelations
[0219] IJRelationValue::EnumRelations returns an enumerator that is used to iterate through all the relations that are connected to a value.
[0220] The following is a description of the preferred embodiment of the formal EnumRelations interface:
[0221] 2.4. IJRelationValue::GetRelation
[0222] IJRelationValue::GetRelation returns the relation of pType connected to this value. This can be used to get a particular relation connected to a value such as the owner or a symbology bundle. If more than one relation of the specified type is connected to this value, the first relation connected should be returned along with the return code, RELATION_S_DUPLICATE_TYPE. In this case, EnumRelations should be used initially to find all the relations of the specified type. A NULL pointer and S_FALSE are returned if no relations of the given type are found.
[0223] The following is a description of the preferred embodiment of the formal GetRelation interface:
[0224] 2.5. IJRelationValue::IsInterfaceIndependent
[0225] IJRelationValue::IsInterfaceIndependent identifies whether or not the interface riid is independent of all the other interfaces supported by this value. Interfaces are not independent if calling a function on the interface affects the results of any function on another interface. These interface dependencies must be known to the recompute machine so that the recompute semantics can be evaluated correctly. Interface dependencies create two ways for a recompute semantic to be evaluated: first, a recompute semantic is evaluated when one of its inputs changes (by the interface specified by the semantic) or second, when another interface of one of its inputs changes and this implicitly (through interface dependencies) causes an input to be changed.
[0226] The interfaces of an object can be divided into two categories: independent interfaces and non-independent interfaces. Independent interfaces are mutually exclusive of each other such that calling a function on an independent interface will NOT cause a change in any values returned from other independent interfaces. In contrast when a non-independent interface is used to modify an object, ALL interfaces on the object are considered to be modified. Note that when an independent interface is used to modify an object, all non-independent interfaces are also considered modified.
[0227] In sum, a recompute semantic will be evaluated in the following cases:
[0228] When one of its input values is modified by the interface enumerated from the recompute semantic. A recompute semantic always specifies its dependence on a particular interface of an input value.
[0229] When an input value is modified by any non-independent interface. Modification of a non-independent interface always triggers a recompute. This includes the recompute of semantics that input interfaces other than the one used to modify the object.
[0230] When an independent interface is used to modify a value, all semantics that depend on non-independent interfaces are recomputed.
[0231] Independent interfaces are used to prevent un-necessary recompute. Given an interface ID, IsInterfaceIndependent returns an HRESULT that describes the interfaces relationship with the other interfaces on the same object.
[0232] Note: The IJRelationValue helper object uses a callback function to implement this. The application should define a global C function with the following function signature: STDAPI IsInterfaceIndependent(REFIID riid).
[0233] A pointer to the application callback function is passed into CreateStdRelationValue. The recompute machine then calls this function, passing it the IID of the interface in question. In this way an application can re-use the provided helper implementation while still supplying the recompute machine with the interface dependencies of particular application classes.
[0234] The following is a description of the preferred embodiment of the formal IsInterfaceIndependent interface:
[0235] 3. IJCopy Interface
[0236] The IJCopy interface represents the copy behavior of a relationship. An object implements this interface to define the copy operation for a particular relation. The Relation subsystem copy machine uses this interface when it is copying objects. The Relation subsystem copy machine accesses the IJCopy interface through IJRelation::EnumSemantics. A relationship describes its modify, copy, and delete behavior by enumerating one or more semantics.
[0237] An IJCopy semantic defines a relationship between a set of values. A value is represented by the IUnknown pointer of a COM object and a status that describes whether or not the object is an input or output: (LPUNKNOWN, IID, flags). The structure that defines a value also contains an IID that is not used for the copy semantic. The IID is used only by the Compute semantics of the relation object.
[0238] The designation of objects as input or output affects how much control an application has over the copy operation. When a semantic enumerates a value as input it just guarantees that someone (another semantic or the Relation subsystem copy machine) will provide a copy of the object before the semantic is called. When a semantic enumerates a value as output, the semantic must provide a copy of the object. Outputs give an application a hook to make the choice of either cloning the value, locating an existing object as the copy, or replacing the output value with a new object. For example, a copy semantic on a document object would enumerate a symbology manager object as output so that it could decide where the copy of the symbology manager would come from. When copying to the clipboard, the semantic would clone a new symbology manager. When copying from the clipboard to an existing document, the semantic would locate the existing symbology manager instead of cloning a new one.
[0239] The following is a description of the preferred embodiment of the formal IJCopy interface:
[0240] 3.1. IJCopy::CreateClones
[0241] IJCopy::CreateClones provides a clone for each of the output values of the copy semantic. The clones can be located objects, objects created with IJPersist::Clone, a new object that replaces the output value, or NULL. The core copy machine calls IJCopy::CreateClones after input values enumerated by the copy semantic have been cloned. The copy semantic provides a clone for each of its outputs and adds them to a map (hash table). If the clone is created with IJPersist::Clone, the flag RELATION_CLONE must be set in the clone map.
[0242] The following is a description of the preferred embodiment of the formal CreateClones interface:
[0243] 3.2. IJCopy::EnumValues
[0244] IJCopy::EnumValues is called by the copy machine to build a graph of the objects related to a set of objects being copied. Returns an enumerator that is used to iterate through the values related by this copy semantic. Each value is qualified as input or output. (The IID is not used and is set to NULL.) Input values are objects that must be copied before this semantics CreateClones function is called. Outputs are objects that will be copied by the copy semantics CreateClones function. Typically, inputs are objects that are accessed during CreateClones to make a decision about an output. For example, a symbology manager's copy semantic would enumerate the symbology manager as an input. This would guarantee that the copy semantic could use the copy of the symbology manager during CreateClones to look for existing symbology bundles. Outputs are objects the application wants to control the copy of. Using the same example, the symbology bundles would be output of the symbology manager's copy semantic so that during copy, the semantic could look to see if an equivalent symbology bundle already exists before creating a new one. Note: the true IUnknown pointer must be returned for each value.
[0245] The following is a description of the preferred embodiment of the formal EnumValues interface:
[0246] Since any object enumerated as input by the copy semantic will be copied by the copy machine (or by another semantic), a semantic can choose to not enumerate all its input values. In this case, any values that enumerate this relation in IJRelationValue::EnumRelations (i.e. are connected to this relation via IJRelationValue::ConnectRelation) will be considered INPUT values and will consequently be added to the enumerator returned from IJCopy::EnumValues before IJCopy::CreateClones and IJCopy::ReconnectClones are called. This allows a semantic to “see” objects only when they are of interest (enumerated by another semantic or selected by the user).
[0247] 3.3. IJCopy::PreValueCopy
[0248] IJCopy::PreValueCopy is called by the copy machine when a value directly related to the relation containing this semantic is going to be copied. For efficiency in the copy machine, all copy semantics on a relation will receive PreValueCopy whether or not they enumerate the value being copied. It is the responsibility of the PreValueCopy implementation to test that the copy of the value will have some impact on the semantic. The implementation of PreValueCopy should call IJRelationNotify::Copy to add additional objects to the copy set.
[0249] The main purpose of PreValueCopy is to propagate the copy operation. For instance, when an owner is selected for copy, the copy operation is propagated to members via IJRelationNotify::Copy. An implementation of PreValueCopy can choose to do nothing. In this case, PreValueCopy should just return S_FALSE.
[0250] The following is a description of the preferred embodiment of the formal PreValueCopy interface:
[0251] 3.4. IJCopy::ReconnectClones
[0252] IJCopy::ReconnectClones re-establishes the relationships between the set of objects created by IJCopy::CreateClones. The copy machine calls this function after persistent identities are assigned to all copied values. Access to the clones is provided by iterating poriginals and looking up the clones in pCloneMap. During a copy operation, it is the responsibility of the original object to reconnect the relationships of its clone. This includes calling IJRelationValue::ConnectRelation to re-establish the relation and storing any necessary persist ids.
[0253] Note: since objects created with IJPersist::Clone are by definition in an incomplete state, care should be taken when accessing the cloned objects for reconnection. Some API's on the cloned object could fail. Objects in this state will have the flag RELATION CLONE set when they are looked up in pCloneMap.
[0254] The following is a description of the preferred embodiment of the formal ReconnectClones interface:
[0255] This enumerator provides the same list of values that were returned from IJCopy::EnumValues with any implicit values added
[0256] 4. IJDelete Interface
[0257] The IJDelete interface represents the delete behavior of a relationship. An object implements this interface to define the delete operation of a particular relation. This interface is only exposed via IJRelation::EnumSemantics. A relationship describes its modify, copy, and delete behavior by enumerating one or more semantics.
[0258] An IJDelete semantic defines a relationship between a set of values. A value is represented by the IUnknown pointer of a COM object and a status that describes whether or not the object is an input or output: (LPUNKNOWN, IID, flags). The structure that defines a value also contains an IID that is not used for the delete semantic.
[0259] The designation of objects as input or output affects how much control an application has over the delete operation.
[0260] The following is a description of the preferred embodiment of the formal IJDelete interface:
[0261] 4.1. IJDelete::Disconnect
[0262] The IJDelete::Disconnect interface disconnects objects from a relation before they are deleted.
[0263] The following is a description of the preferred embodiment of the formal Disconnect interface:
[0264] 4.2. IJDelete::EnumValues
[0265] The IJDelete::EnumValues returns an enumerator that is used to iterate through the values related by this delete semantic. Each value is qualified as input or output. (The IID is not used and is set to NULL.). Note the true IUnknown pointer must be returned for each value.
[0266] The following is a description of the preferred embodiment of the formal EnumValues interface:
[0267] 4.3. IJDelete::PreValueDelete
[0268] IJDelete::PreValueDelete is called by the delete machine when a value directly related to the relation containing this semantic is going to be deleted. For efficiency in the delete machine, all delete semantics on a relation will receive PreValueDelete whether or not they enumerate the value being deleted. It is the responsibility of the PreValueDelete implementation to test that the delete of the value will have some impact on the semantic. The implementation of PreValueDelete should call IJRelationNotify::Delete to add additional objects to the delete set. It should also call IJRelationNotify::Modify when the deletion of a connected element causes a modification of state such as when a member of an owner is deleted.
[0269] The main purpose of PreValueDelete is to propagate the delete operation: for instance when an owner is selected for delete, the delete operation is propagated to members via PreValueDelete. An implementation of PreValueDelete can choose to do nothing. In this case, PreValueDelete should just return S_FALSE. Note: the user cannot call the global batch function from within PreValueDelete.
[0270] The following is a description of the preferred embodiment of the formal PreValueDelete interface:
[0271] 5. IJCompute Interface
[0272] The IJCompute interface represents the recompute behavior of a relationship. An object implements this interface in order to be notified of changes to other objects in the system. A pointer to the IJCompute interface is obtained by calling IJRelation::EnumSemantics, as was previously described. A relationship describes its modify, copy, and delete behavior by enumerating one or more objects that implement the relation's semantics.
[0273] An IJCompute semantic defines a relationship between a set of values. Values are represented by the identity of the COM object representing the value, the identity of the interface used to access the value, and a status that describes whether or not the value is an input, an output, or constrained: (LPUNKNOWN, IID, and flags).
[0274] A semantic is evaluated when one of its input values is modified. When constrained values are modified, a set of related semantics and values are passed to a solver. Typically, the result of evaluating a semantic is the modification of one or more output values. The modification of output values can trigger the evaluation of other semantics, and so on.
[0275] The relationships between input, output, and constrained values can be modified during evaluation by editing the recompute semantics. If this happens, evaluation will stop; the semantics will be searched again; and evaluation will continue.
[0276] The following is a description of the preferred embodiment of the formal IJCompute interface:
[0277] 5.1. IJCompute::EnumValues
[0278] During modification (::BatchFlush) the Relation subsystem recompute machine calls this function to build a graph of all objects related to the objects being modified. The Enumvalues function returns an enumerator that can be used to iterate through the values related by this recompute semantic. Each value is qualified as input, output, or constrained. Input values are objects that the relation depends on, i.e. the recompute semantic is evaluated when one or more of its input values are modified. Output values are objects that can be changed as the result of evaluating the recompute semantic. Constrained values must be passed to a solver along with the related recompute semantics. Note: the true IUnknown pointer must be returned for each value.
[0279] The following is a description of the preferred embodiment of the formal EnumValues interface:
[0280] These three qualifiers are mutually exclusive. An additional qualifier may be added to any of these three:
[0281] 5.2. IJCompute::Evaluate
[0282] The IJCompute::Evaluate function is called by the Relation subsystem recompute machine after input values connected to the relation are modified. The recompute semantic has the option of modifying output values as the result of being evaluated. If a recompute semantic has constrained arguments, it will NOT get an Evaluate call, instead, the recompute semantic and its related values will be passed to a solver. The solver chooses which values to modify.
[0283] A recompute semantic can choose to change some relations during evaluation. In that case, Evaluate should return RELATION_S_SEARCH and specify the list of changed semantics in ppEditedSemantics.
[0284] The implementation of Evaluate performs the following functions:
[0285] Tests the status of input arguments (modified, not modified, or error);
[0286] Evaluates an output;
[0287] Notifies the recompute machine if the output value will change before the value is actually changed; and
[0288] Posts the new value of the output, after the recompute machine notification.
[0289] If Evaluate changes relationships (for example, edit the relationship graph), it should return RELATION_S_SEARCH to notify the recompute machine that the current graph is no longer valid. The enumerator ppEditedSemantics must be returned indicating the edited semantics. If the current semantic is included in the edited list, it is recomputed.
[0290] When calling this function, the recompute machine changes the status of output values as follows:
[0291] If Evaluate sends a modify for an output value to pNotify, the recompute machine changes the status of the output value to RELATION_MODIFIED; and
[0292] If the return code of Evaluate is an error, the state of the output value(s) is set to RELATION_ERROR.
[0293] The following is a description of the preferred embodiment of the formal Evaluate interface:
[0294] The flags field of the JIDREF structure indicates the status of the value with respect to recompute:
[0295] pvalues can be held by AddRefing or Cloning it. This allows this enumerator to be used to enumerate which values have been modified after returning from Evaluate. However, this enumerator will only be valid for the duration of the current graph compute. It will be automatically invalidated by the recompute machine at the end of the current recompute. After an enumerator has been invalidated it will return E_ACCESSDENIED.
[0296] The global batch functions, ::BatchInsert, ::BatchModify, and so on, should never be called from within Evaluate. Use pNotify instead.
[0297] 5.3. IJCompute::GetSolver
[0298] The IJCompute::GetSolver gets the solver associated with a recompute semantic. The solver is responsible for updating constrained values or updating the outputs of recompute semantics that are in a loop. This function is only called when the recompute machine detects constrained values or a set of semantics that are in a loop. (A loop results when a set of recompute semantics defines inputs and outputs such that evaluation would cause an infinite loop.)
[0299] The following is a description of the preferred embodiment of the formal GetSolver interface:
[0300] 5.4. IJCompute::PreValueModify
[0301] IJCompute::PreValueModify is called by the recompute machine when a value directly or transitively related to the relation containing this semantic is going to be modified. For efficiency in the recompute machine, all compute semantics on a relation will receive PreValueModify whether or not they enumerate the value being modified. It is the responsibility of the PreValueModify implementation to test that the modification of the value will have some impact on the semantic. (This could involve testing independent or non-independent interface information.) The implementation of PrevalueModify should call IJRelationNotify::Modify on any output values that will be modified as a result of the input value being modified.
[0302] If an interface modifies an object several times, the recompute machine notifies connected semantics only for the first modification. If different interfaces modify an object several times, the recompute machine notifies connected semantics each time a new interface is modified. The flag RELATION_MULTI_MODIF indicates that another interface has already modified the object.
[0303] By enumerating pStackedValues, the semantic can find out which object initiated the modification because of ::BatchModify and which objects forwarded it using IJRelationNotify::Modify. This allows access to the list of objects involved in the modification not just the object connected directly to this relation. PstackedValues enumerates values in the following manner:
[0304] 1. The first enumerated value is the value directly connected to this relation.
[0305] 2. All enumerated values but the last have been marked modified by IJRelationNotify::Modify.
[0306] 3. The value enumerated last has been marked modified by ::BatchModify.
[0307] A modification is always initiated with ::BatchModify. The relation(s) directly connected to the value to be modified will receive IJCompute::PreValueModify. Each of these semantics will call IJRelationNotify::Modify for each output that will be modified. These additional semantics will receive IJCompute::PreValueModify and be able to access the original object that was specified in ::BatchModify and the intermediate objects that were modified by IJRelationNotify::Modify. For example, when a listener that depends on the IJOwner interface of an owner is being notified that the owner is about to change, it can find which member of the owner is about to change.
[0308] Note that an implementation of PreValueModify can choose to do nothing. In this case, PreValueModify should just return S_FALSE.
[0309] The following is a description of the preferred embodiment of the formal PreValueModify interface:
[0310] 5.5. IJCompute::ReplaceValue
[0311] IJCompute::ReplaceValue replaces a value related enumerated by this semantic with another value. This function can be used to change or edit the relationship graph. Note that if the same value is depended on multiple times, this function replaces the first occurrence of that value with the new value.
[0312] The following is a description of the preferred embodiment of the formal ReplaceValue interface:
[0313] 6. IJRelationGraph Interface
[0314] The IJRelationGraph interface is supported by the Relation Manager to communicate with the relation semantics or the solvers that operate on relation semantics. It presents a “snap-shot” of a document's relationship graph.
[0315] The following is a description of the preferred embodiment of the formal IJRelationGraph interface:
[0316] 6.1. IJRelationGraph::AddElement
[0317] IJRelationGraph::AddElement indicates to the graph element to accumulate before a call to SearchAndSort. That is an alternate way to indicate the set of initial values to use for searching and sorting of a dependent graph.
[0318] The following is a description of the preferred embodiment of the formal AddElement interface:
[0319] 6.2. IJRelationGraph::Clear
[0320] IJRelationGraph::Clear clears the graph. Releases a pointer to the object that it contains and releases memory. Constraint system to solve.
[0321] The following is a description of the preferred embodiment of the formal Clear interface:
[0322] 6.3. IJRelationGraph::CommittedOff
[0323] IJRelationGraph::CommittedOff indicates that the graph operation performed will not affect the data base permanently. This mode gets elements which recompute when calling IJRelationNotify::IsCommitted
[0324] The following is a description of the preferred embodiment of the formal CommittedOff interface:
[0325] ULONG CommittedOff()
[0326] Return Number of CommittedOff requirement See Also: BatchCommittedOn, BatchCommittedOff, IsBatchCommitted, RelationNotify::IsBatchCommitted
[0327] 6.4. IJRelationGraph::CommittedOn
[0328] IJRelationGraph::CommittedOn indicates that the graph operation that will be performed will affect permanently the data base. This mode gets elements which recompute when calling IJRelationNotify::IsCommitted
[0329] The following is a description of the preferred embodiment of the formal CommittedOn interface:
[0330] ULONG CommittedOn()
[0331] Return Number of “CommittedOn” requirement left. The mode is back to “Committed” when this count is back to 0. See Also: BatchCommittedOn, BatchCommittedOff, IsBatchCommitted, RelationNotify::IsBatchCommitted
[0332] 6.5. IJRelationGraph::EnumValues
[0333] IJRelationGraph::EnumValues returns an enumerator that can be used to list the values that the graph contains. The flags field of the JIDREF structure has the following meaning:
[0334] The following is a description of the preferred embodiment of the formal EnumValues interface:
[0335] 6.6. IJRelationGraph::IncrementalSearchAndSort
[0336] IJRelationGraph::IncrementalSearchAndSoort is used in two cases:
[0337] After an initial SearchAndSort has been performed with the option RELATION_PARTIAL_CONSTRAINT, and a solver cannot find a solution without modifying values whose predicates are not in the system to solve (values with the state RELATION_FROZEN). The solver should then ask for an incremental search of the graph that takes all the predicates acting on these frozen values into consideration. The incremental search adds these predicates and adds any values that are arguments of these predicates. The solver can then attempt to solve again. This entire process is repeated until a solution is found.
[0338] During update if Evaluate returns RELATION_S_SEARCH. In this case, the graph needs to be searched and sorted again with respect to the list of edited semantics returned by Evaluate.
[0339] The following is a description of the preferred embodiment of the formal IncrementalSearchAndSort interface:
[0340] 6.7. IJRelationGraph::ModifyValueState
[0341] IJRelationGraph::ModifyValueState allows a callback to be used to modify the status of values in the current graph. This function is typically used to reset the status of values after an Update.
[0342] The following is a description of the preferred embodiment of the formal ModifyValueState interface:
[0343] 6.8. IJRelationGraph::SearchAndSort
[0344] IJRelationGraph::SearchAndSort forms a set of COM object identifiers and interface identifiers, and finds the elements transitively connected by dependency or equivalence relations and sorts them into the correct order for recompute. The following modes of searching are provided:
[0345] The set of initial elements can be passed as input argument pInitial, or accumulate by the graph itself by previous call to AddElement.
[0346] The following is a description of the preferred embodiment of the formal SearchAndSort interface:
[0347] Processing
[0348] The elements are searched with respect to the options defined above.
[0349] The following options are mutually exclusive:
[0350] RELATION_PARTIAL_CONSTRAINT/RELATION_FULL_CONSTRAINT. The elements are sorted in the dependency order. If the SearchAndSort is to be used for a graph update, the status of all pinitial values should be set to RELATION_MODIFIED.
[0351] 6.9. IJRelationGraph::Update
[0352] IJRelationGraph::Update maintains the data integrity of values by evaluating recompute semantics or by passing recompute semantics to a solver. Whenever elements with relations are modified, SearchAndSort should be called followed by an Update.
[0353] The following is a description of the preferred embodiment of the formal Update interface:
[0354] 7. IJRelationNotify Interface
[0355] The IJRelationNotify interface is used during update, copy, and delete to notify the relationship manager of additional objects to be processed. It is passed as a parameter to the compute, copy and delete semantics. The semantics are responsible for notifying the relationship manager.
[0356] The following is a description of the preferred embodiment of the formal IJRelationNotify interface:
[0357] 7.1. IJRelationNotify::Copy
[0358] IJRelationNotify::Copy is used by a relation semantic to notify the relationship manager that the COM object, pAnyInterface, needs to be copied as the result of the current operation. This is typically called from IJCopy::PreValueCopy to add additional objects to be copied. For instance, when an owner is being copied, its copy semantic uses this to recursively add its members to be copied.
[0359] The following is a description of the preferred embodiment of the formal Copy interface:
[0360] 7.2. IJRelationNotify::Delete
[0361] IJRelationNotify::Delete used by a relation semantic to notify the relationship manager that the COM object, pAnyInterface, needs to be deleted as the result of the current operation. This is typically called from IJDelete::PreValueDelete to add additional objects to be deleted. For instance, when an owner is being delete, its delete semantic uses this to recursively add its members to be deleted.
[0362] The following is a description of the preferred embodiment of the formal Delete interface:
[0363] 7.3. IJRelationNotify::Insert
[0364] IJRelationNotify::Insert is used by a relation semantic to notify the relationship manager that the COM object, pAnyInterface, was created as the result of the current operation. This can only be called from an implementation of IJDeleteIJDeletePreValueDelete. A delete semantic sometimes has the need to create objects. For example, to delete an elbow and create an end cap when a pipe is being deleted.
[0365] The following is a description of the preferred embodiment of the formal Insert interface:
[0366] 7.4. IJRelationNotify::IsCommitted
[0367] IJRelationNotify::IsCommitted is used by functions of relation semantic which are given a LPRELATIONNOTIFY as argument to inquire if the current modification if committed or not, i.e. will it affect the database permanently.
[0368] The following is a description of the preferred embodiment of the formal IsCommitted interface:
[0369] 7.5. IJRelationNotify::Modify
[0370] IJRelationNotify::Modify is used by a relation semantic to notify the relationship manager that the COM object, pAnyInterface, will be modified via interface, refIID, as the result of the current operation. This is typically called from IJCompute::Evaluate to notify the relationship manager that the outputs of a recompute semantic are going to be modified. This function should be called before the actual modification of pAnyInterface.
[0371] The following is a description of the preferred embodiment of the formal Modify interface:
[0372] 8. IJRelationGlobal Interface
[0373] The IJRelationGlobal interface is supported by the Relation Manager to set and get how the Relation Manager behaves during recompute.
[0374] The following is a description of the preferred embodiment of the formal IJRelationGlobal interface:
[0375] 8.1. IJRelationGlobal::GetUpdateMode
[0376] IJRelationGlobal::GetUpdateMode inquires how Relation Manager recomputes dependency after a modification: RELATION_PARTIAL_CONSTRAINT solves the constraint systems incrementally. RELATION_FULL_CONSTRAINT searches the entire constraints system before trying to solve them.
[0377] The following is a description of the preferred embodiment of the formal GetUpdateMode interface:
[0378] 8.2. IJRelationGlobal::SetUpdateMode
[0379] IJRelationGlobal::SetUpdateMode defines how Relation Manager recomputes dependency after a modification: RELATION_PARTIAL_CONSTRAINT solves the constraint systems incrementally. RELATION_FULL_CONSTRAINT searches the entire constraints system before trying to solve them.
[0380] The following is a description of the preferred embodiment of the formal SetUpdateMode interface:
[0381] 9. IJRelationSolver Interface
[0382] The IJRelationSolver interface is implemented by the Relation Manager to communicate with an application solver. The solver should have knowledge of a particular set of constraints or knowledge of how to solve a list of semantics in a loop.
[0383] The following is a description of the preferred embodiment of the formal IJRelationSolver interface:
[0384] 9.1. IJRelationSolver:Solve
[0385] IJRelationSolver::Solve solves a constraint system. A set of values and a set of semantics defines the system to solve. The solver should update values to ensure that all the constraints are TRUE.
[0386] The following is a description of the preferred embodiment of the formal Solve interface:
[0387] To reduce the size of the system to solve, the solver can test for the status RELATION_MODIFIED to see which values have been modified by the user or by some re-evaluation function. The graph to recompute can contain values that have not changed.
[0388] The solver must not modify values with the status RELATION_INPUT. Such a value is a constant input of a predicate, for example, if the value is the output of a recompute semantic.
[0389] If the solver needs to modify RELATION_FROZEN values to find a solution, it should return the values in ppAddedValues and return the status RELATION_E_NO_SOLUTION to ask the recompute machine for a more complete graph. All the semantics acting on these values and all the values that are values related to these semantics are then added to the graph. After the graph is re-ordered, the evaluation functions with the added values for output are called before the constraint solver is called again with the extended system.
[0390] An alternate solution is to ask for a complete search of the constraint system by returning RELATION_S_FULL_CONSTRAINT.
[0391] If a solution is found, the solver should pre-notify the recompute machine, update adequate geometry, and return S_OK.
[0392] 10. IJComputeCFunction Interface
[0393] The IJComputeCFunction interface represents a recompute semantic that is implemented as a C function. It also provides local storage of intrinsic values like int, double, and so on to be used as function arguments. This interface should be implemented by an object representing the recompute semantic of a relation. This interface is accessed through IJRelation::EnumSemantics. The standard C function helper supports IJComputeCFunction.
[0394] The idea is for an application to write a set of C functions that the Relationship Manager can call during recompute. These functions take values as inputs and write to output values. This allows an application to model a wide variety of recompute behavior without having to implement the associative machine itself. This is analogous to a spreadsheet where end-users can write their own functions to be placed in a cell.
[0395] The following is a description of the preferred embodiment of the formal Solve interface:
[0396] 10.1. IJComputeCFunction::GetArgumentCount
[0397] IJComputeCFunction::GetArgumentCount gets the count of the arguments for the specified function. The count is determined from the argument format passed to IJComputeCFunction::SetDefinition.
[0398] The following is a description of the preferred embodiment of the formal GetArgumentCount interface:
[0399] 10.2. IJComputeCFunction::GetArguments
[0400] IJComputeCFunction::GetArguments gets the actual arguments for the specified function. The buffer passed in should be large enough to hold all the arguments. The arguments returned in the buffer could contain interface pointers that need to be released and memory pointers that need to be freed via IMalloc→Free. The Relations subsystem provides a helper function, ::ReleaseJCFUNCARGDESCon pArgs to release any interfaces and free any allocated memory it might contain.
[0401] The following is a description of the preferred embodiment of the formal GetArguments interface:
[0402] 10.3. IJComputeCFunction::GetDefinitionV
[0403] IJComputeCFunction::GetDefinitionV gets the parameters used to define the C function.
[0404] The following is a description of the preferred embodiment of the formal GetDefinitionV interface:
[0405] 10.4. IJComputeCFunction::ReplaceArgument
[0406] IJComputeCFunction::ReplaceArgument replaces an actual argument for the specified function.
[0407] The following is a description of the preferred embodiment of the formal ReplaceArgument interface:
[0408] 10.5. IJComputeCFunction::SetDefinition
[0409] IJComputeCFunction::SetDefinition establishes a C function as an associative evaluation function. The C function is called from IJCompute::Evaluate during a graph update. The C function to call and its arguments are specified.
[0410] A function can have the following intrinsic types as input arguments: char, unsigned char, char[], unsigned char[], short, unsigned short, short[], unsigned short[], long, unsigned long, long[], unsigned long[], double, double[], and string (NULL terminated char[]).
[0411] A function can have COM objects as input and output arguments. These arguments can be passed to the evaluation function in three ways: as an interface pointer, as a pointer to a JIDREF structure (LPJIDREF) or as an array of JIDREF structures. In the first case, IJCompute::Evaluate is responsible for querying and releasing the interface. In the second and third cases, the C evaluation function receives the IUnknown pointer in the JIDREF struct. Since the flags field of the JIDREF contains a status: modified, error, and so on, using an JIDREF as input allows a function to determine which of its parents has changed.
[0412] A function can also have a special “run-time” argument: a pointer to the relation manager notification interface. This argument is set during evaluate and can be used by the function to notify the recompute machine of modify events for outputs. This allows a function to optimize its recompute behavior by sending modify events for outputs only when necessary.
[0413] Argument types are specified by a “printf” style format string. The argument format string is composed of a contiguous list of one or more argument formats. Each argument format consists of three parts:
[0414] A ‘%’ character to indicate the start of a new argument.
[0415] For intrinsic types: zero, one or two modifiers that indicate whether or not the argument is an array and/or unsigned. For COM object types: one or two status characters that indicate whether or not the argument is an array and input or output.
[0416] A single character that indicates the argument type.
[0417] The syntax for an argument format is:
[0418] Arrays are passed as two arguments: a ULONG indicating the number of elements in the array and a pointer to the first element. Zero size arrays are passed with a NULL pointer. On creation of the C function, array arguments are copied into a buffer within the C function. When called, the C function is passed a pointer into this buffer. Changing the contents of arrays and strings can only be done with IJComputeCFunction::ReplaceArgument.
[0419] The actual arguments, as passed to SetDefinition, must exactly match the specified argument format (except for the relation notification pointer, which should be NULL. The current relation notification pointer is substituted when the C function is called.). Arrays require two actual arguments: a ULONG count indicating the number of elements in the array and a pointer to the first element. COM object arguments also require two actual arguments: a pointer to any interface on the value (LPUNKNOWN) and a pointer to the iid of the interface used on the value (LPIID). These arguments are passed to the evaluation function as a pointer to the specified interface or as a pointer to a JIDREF structure. In the later case, the JIDREF structure contains the IUnknown pointer, interface ID, and a status flag. An array of JIDREF structures is required for the “%#>r” and “%#<r” argument types.
[0420] COM object arguments are optional. In this case, the actual arguments are passed as NULL (both the LPUNKNOWN and LPIID). When called, the evaluation function is passed a NULL interface pointer or NULL LPJIDREF for these arguments. Optional arguments can be set with IJComputeCFunction::ReplaceArgument.
[0421] Normally, the Relation Manager assumes that Evaluate modifies all output values and sets their status. If the C function wants to set the status of the outputs itself, it should declare the relation notification interface, “%n”, as a run-time argument and declare its output arguments as LPJIDREF. (The notification for modify requires the COM identity of an object and the IID of the interface used to modify the object.) This allows a C function to optimize its recompute behavior, for example by deciding whether or not an output value has changed. Note: the user cannot call the global batch functions during evaluate of the C function.
[0422] The following is a description of the preferred embodiment of the formal SetDefinition interface:
[0423] 10.6. IJComputeCFunction::SetDefinitionV
[0424] IJComputeCFunction::SetDefinitionV returns a va_list version of IJComputeCFunction::SetDefinition.
[0425] The following is a description of the preferred embodiment of the formal SetDefinitionV interface:
[0426] HRESULT SetDefinitionV (REFCLSID dllServerClsid, LPSTR pFunctionName, LPSTR pArgFormat, va_list args);
[0427] 11.IJUndo Interface
[0428] The Undo Interface writes object modifications to a log (stream) and reads from the log to restore/revert a value. This is an optional Interface.
[0429] The following is a description of the preferred embodiment of the formal IJUndo interface:
[0430] 11.1. IJUndo::Log
[0431] IJUndo::Log Method writes to the undo stream the data associated with a given interface. IJUndo::Log(IID,pStream) is the logging method. Given the IID of the interface slated for modification, the object saves (writes) the data associated with the given IID, including any dependent interfaces, to the provided Steam pointer, pStream→write().
[0432] The following is a description of the preferred embodiment of the formal Log interface:
[0433] 11.2. IJUndo::Revert
[0434] IJUndo::Revert reads from the undo stream the data and reverts object data associated with a given interface STDMETHOD(Revert) (THIS_IJUndo::Revert(IID,pStream) is the reloading method. Given the IID of the interface slated for Undo, the object reloads the data associated with the given IID, including any dependent interfaces, from the provided Steam pointer, pStream→read().
[0435] The following is a description of the preferred embodiment of the formal Revert interface:
[0436] 12. IEnumJCHKPNT Interface
[0437] IEnumJCHKPNT Interface is an enumerator to be returned by the undo manager for selecting a checkpoint. It is supported by IJUndoManager.
[0438] The following is a description of the preferred embodiment of the formal IEnumJCHKPNT interface:
[0439] 12.1. IEnumJCHKPNT::Clone
[0440] IEnumJCHKPNT::Clone creates a copy of the enumerator.
[0441] The following is a description of the preferred embodiment of the formal Clone interface:
[0442] 12.2. IEnumJCHKPNT::Next
[0443] IEnumJCHKPNT::Next returns the next checkpoint structure(s).
[0444] The following is a description of the preferred embodiment of the formal Next interface:
[0445] 12.3. IEnumJCHKPNT::Reset
[0446] IEnumJCHKPNT::Reset Reverts the enumerator to the first entry in the list.
[0447] The following is a description of the preferred embodiment of the formal Reset interface:
[0448] 12.4. IEnumJCHKPNT::Skip
[0449] IEnumJCHKPNT::Skip skips the next count of checkpoint structures.
[0450] The following is a description of the preferred embodiment of the formal skip interface:
[0451] 13. IJUndoManager Interface
[0452] The Undo Manager maintains an undo log for the document. In addition it keeps the transaction checkpoints and initiates the undo and redo processing. IJUndoManager is supported by IJUndoManager
[0453] The following is a description of the preferred embodiment of the formal IJUndoManager interface:
[0454] 13.1. IJUndoManager::Checkpoint
[0455] IJUndoManager::Checkpoint sets a checkpoint in the undo log for the next set of document modifications. The document's undo manager is informed of the beginning of a command by calling its checkpointing method, IJundoManager::Checkpoint(LPJUSERTEXT, DWORD). The command manager (One per document makes this call at the beginning of a transaction with the name to associate with the log of the next set of changes (using Level=0). The implementation of a command may add additional checkpoints that may be selected by the user during that command. When a command has completed, only the top level checkpoints must be accessible. Any checkpoints issued to the UndoMgr without subsequent modifications to the document will not be accessible.
[0456] The following is a description of the preferred embodiment of the formal Checkpoint interface:
[0457] 13.2. IJUndoManager::CheckpointData
[0458] IJUndoManager::GetCheckPointData returns a stream from which a command may reuse data in an undo log. GetCheckpointData(JChkPnt, LPStream) is to be used by commands that have explicit knowledge of the undo log written by a data entity. The purpose for providing this method is to allow a command access to data that it modified, for reuse in its processing and not maintain multiple copies of that data. As mentioned earlier, the undo log is application entity specific, therefore, the log may contain references to other storage media (such as a memory mapped files . . . ) which the entity (and command if so programmed) will know how to access.
[0459] The following is a description of the preferred embodiment of the formal CheckpointData interface:
[0460] 13.3. IJUndoManager::GetUndoCheckpoints
[0461] IJUndoManager::GetUndoCheckPoints returns an enumerator of checkpoints from the Undo Log. A command (like the Undo command itself), gets a list of checkpoints with associated keys from the UndoMgr by calling the method IJUndoManager::GetUndoCheckPoints (LPJEnumChkPnt). The command could then use the information to either provide in a dialog to the user for the selection to undo or to access the stream at the point where that data was logged (more on IJUndoManager::GetCheckPointData() later).
[0462] The following is a description of the preferred embodiment of the formal GetUndoCheckpoints interface:
[0463] 13.4. IJUndoManager::GetRedoCheckpoints
[0464] IJUndoManager::GetRedoCheckPoints returns an enumerator of checkpoints from the Redo Log. Same as GetUndoCheckPoints for Undoing an Undo operation (Redo).
[0465] The following is a description of the preferred embodiment of the formal GetRedoCheckpoints interface:
[0466] 13.5. IJUndoManager::Log
[0467] IJUndoManager::Log logs a transaction to the Undo Stream. IJUndoMgr::log(LOGTYPE, LPIUNKNOWN, IID) is the logging method. Given the logtype and IUnknown of the modified object, the undo manager performs the log write. In processing it searches for Modify, Create, and Delete operations sorting them to eliminate any unnecessary steps (e.g. create and subsequent delete need only perform a DELETE_MARKONLY operation).
[0468] The following is a description of the preferred embodiment of the formal Log interface:
[0469] 13.6. IJUndoManager::Undo
[0470] IJUndoManager::Undo Method reverts the document to the state at which the specified checkpoint was logged. IJUndoMgr::Undo(JChkPnt, LPENUMJIDREF) is the undo method. Given the checkpoint key, the undo manager performs the undo operation. In processing it searches for Modify, Create, and Delete operations sorting them to eliminate any unnecessary steps (e.g. create and subsequent delete), and performs the modify to the document in an order necessary to restore data and referential integrity. As changes are made, the UndoMgr will set the document undo stream to point to its redo pointer.
[0471] The following is a description of the preferred embodiment of the formal Undo interface:
[0472] 13.7. IJUndoManager::Redo
[0473] IJUndoManager::Redo Method reverses an undo to the specified document state. IJUndoMgr::Redo(JChkPnt, LPENUMJIDREF) is the redo method. It operates just as Undo except that changes are logged to the document's redo stream.
[0474] The following is a description of the preferred embodiment of the formal Redo interface:
[0475] 13.8. JCHKPNT Structure
[0476] This structure contains the checkpoint name associated with a set of document changes and a key used to identify the checkpoint in the undo log.
[0477] 14. Other Interfaces
[0478] The IJMapUnknownToUnknown interface defines a hash table that maps a key to a value. In this case, the key is an IUnknown pointer, and the value is an IUnknown pointer and a ULONG flags word.
[0479] The following is a description of the preferred embodiment of the formal IJMapUnknownToUnknown interface:
[0480] Each of the functions in the IJMapUnknownToUnknown interface is included in the Appendix for sake of completeness.
[0481] The IEnumJRelationSemanticDesc interface defines an enumerator object for a list of relation semantic descriptor structures. This structure contains a pointer to the IUnknown interface of a relation semantic and a key that uniquely identifies the semantic. The interface pointer must be released before discarding the contents of the structure. For more information on enumerators, please refer to IEnumX in the OLE documentation.
[0482] The following is a description of the preferred embodiment of the formal IEnumJRelationSemanticDesc interface:
[0483] The IEnumJIDREF interface defines the Enumerator object for the JIDREF structure. The JIDREF structure associates an object's IUnknown pointer to an interface ID and a set of bit flags. The bit flags determine the current status of the object. The IID field is used to indicate which interface (if any) was used to modify the object.
[0484] The following is a description of the preferred embodiment of the formal IEnumJRelationSemanticDesc interface:
[0485] For more information on enumerators, please refer to IEnumX in the OLE documentation previously cited.
[0486] The Relations subsystem provides several global functions. These functions serve three purposes: to provide access to global Relations objects, to create “helper” objects that implement Relations interfaces, and to provide common functions to interrogate the relationships between objects.
[0487] The following is a description of the preferred embodiment of the formal IEnumJRelationSemanticDesc interface;
[0488] Each of the global functions is included in the Appendix for sake of completeness.
[0489] III. The Relations Subsystem
[0490] 1. Examples of Associative Graphs in COM
[0491] 2. Detailed Description of Relations
[0492] 3. Detailed Description of Modify Copy Delete Machines
[0493] 4. Example of the Relations Subsystem
[0494] 5. Detailed Description of the Undo Feature
[0495] 1. Examples of Associative Graphs in COM
[0496] In the following example, names of certain interfaces are contrived where necessary.
[0497] 1.1. Associative Dimension on a Line
[0498] FIG. 8 illustrates two COM objects: a dimension and a line. The dimension 480 depends on the line 490 such that whenever the length of 490 line changes dimension 480 is re-calculated.
[0499] FIG. 9 illustrates a graphic representation of the COM relations for dimension 480 and line 490 of FIG. 8. IJRelationValue 500 is a COM interface that represents value nodes. IJRelation 510 is a COM interface that represents function nodes. Finally, IJCompute 520 is a COM interface that represents a single function node.
[0500] The three nodes in the associative graph are a value node 530 representing the line's geometry (identified by: line 540, IJLine 550), a function node 560 representing the evaluation function, and a value node 570 representing the output of the function: dimension value (identified by: dimension 580, IJDimension 590).
[0501] The associative evaluation function of function node 560 is called every time the IJLine interface 550 is changed. The evaluation function of function node 560 gets the line geometry via the IJLine interface 550 and sets the dimension value using the IJDimension interface 590. There can be many more interfaces on the line that can be modified without affecting the dimension. For example, the symbology interface, IJSymbology, could be changed without affecting the dimension.
[0502] 1.2. Parallel Constraint Between Two Lines
[0503] FIG. 10 illustrates a graphic representation of the COM relations for a parallel constraint. Whenever the geometry of line-1 changes, line-2's geometry is changed to maintain the parallel constraint. Furthermore, if the geometry of line-2 changes, then line-1 geometry is also changed to maintain the parallel constraint. In this example there are no graphic handles to represent the constraint 605. A predicate evaluation function 610 has an equivalence relation 620 and 630 with the IJLine interfaces 640 and 650 of line-1 and line-2, respectively.
[0504] The associative graph contains three nodes: a value node 660 for line-1's geometry (identified by: line-1 680, IJLine 640), a value node 670 for line-2's geometry (identified by: line-2 690, IJLine 650), and a function node IJCompute 700 for the constraint.
[0505] 1.3. Parallel Constraint with Graphic Handles
[0506] FIG. 11 illustrates two lines having a parallel constraint and a dimension having a graphic handle. As the size of dimension 720 is edited, the distance between the two parallel lines 730 and 740 is determined.
[0507] FIG. 12 illustrates a graphic representation of the COM relations of FIG. 11. A predicate evaluation function 750 has an equivalence relation 760 and 770 with the IJLine interfaces 780 and 790 of line-1 and line-2, respectively. Further, the constraint 800 encapsulates the dimension 810 that acts as a graphic handle for the end-user.
[0508] There are five nodes in the associative graph: a value node for line-1's geometry (identified by: line-1 830, IJLine 780), a value node for line-2's geometry (identified by: line-2 850, IJLine 790), a function node 750 for the parallel constraint, a function node 820 for the graphic handle, and a value node 820 for the dimension value (identified by: constraint 800, IJDimension 810).
[0509] 1.4. Connector Between Two Pipes
[0510] FIG. 13 illustrates an example of two pipes connected by a connector. Pipes 870 and 880 are represented by “multi-lines” that are each created by constructing two parallel lines to respective center-lines 900 and 910. Pipes 870 and 880 are connected by a connector 890. In this example, when the connector 890 or pipes 870 and 880 are modified: the center-lines 900 and 910 need to be kept connected and the multi-lines that represent the pipes need to be evaluated (trimmed to the proper length).
[0511] FIG. 14 illustrates a graphic representation of the COM objects of FIG. 13.
[0512] FIG. 15 illustrates a graphic representation of the COM relations of FIG. 13.
[0513] The connector 930 represents a constraint 940 to ensure that the end-point of pipe-1's center line and the begin-point of pipe-2's center line are coincident. The connector 930 represents another constraint 950 that acts as a cleanup function to trim the pipes (multi-lines) appropriately for this type of connector. The connector 930 also represents an associative function 960 that generates the connector geometry based on the center lines of the pipes.
[0514] Pipe-1 and Pipe-2 each contain a constraint 970 and 980, respectively to create their multi-lines by copying parallel the center lines. The multi-line representing each pipe is defined by two constraints: a constraint in the pipe that maintains two lines parallel to the center-line 970 and 980 (to determine the pipe thickness) and a constraint 950 in the connector 930 to do trimming.
[0515] Thus there are ten nodes in the associative graph:
[0516] 1. A value node 990 for pipe-1's center-line geometry (identified by: pipel 1000, IJLine 1010).
[0517] 2. A value node 990 for pipe-1's multi-line geometry (identified by: pipel 1000, IJMultiLine 1020).
[0518] 3. A function node 990 for pipe-1's copy parallel constraint 970.
[0519] 4. A value node 1030 for pipe-2's center-line geometry (identified by: pipe2 1040, IJLine 1050).
[0520] 5. A value node 1030 for pipe-2's multi-line geometry (identified by: pipe2 1040, IJMultiLine 1060).
[0521] 6. A function node 1030 for pipe-2's copy parallel constraint 980.
[0522] 7. A function node 930 for the connector's end point constraint 940.
[0523] 8. A function node 930 for the connector's trimming constraint 950.
[0524] 9. A function node 930 for the connector's evaluation function 960.
[0525] 10. A value node 930 for the connector's geometry (identified by: connector 1070, IJConnector 1080)
[0526] There are two constraint systems in the graph:
[0527] 1. One containing the center-lines of the pipes and the connector's end-point constraint (nodes 990, 940, 1030).
[0528] 2. One for the generation of the multi-lines and their trimming (nodes 970, 980, 930).
[0529] Each of these “systems” would be passed to a solver independently of each other. In fact, they probably would use a different solver.
[0530] Given the above relations, the Relations subsystem manages each relation in response to a modify/copy/ or delete semantic.
[0531] 2. Detailed Description of Associative Graphs in COM
[0532] 2.1. The Relationship on the Value Side
[0533] The only requirement for an object to participate in a relationship is that it must be able to store and enumerate the relations that it participates in. This protocol is defined by the interface IJRelationValue. IJRelationValue defines how to connect, disconnect, and enumerate relations. The actual storage of the relation is implementation specific; the IJRelationValue API deals strictly with IUnknown pointers. Values should allow definition of relations contained within a persistent or non-persistent object. In the persistent case the relation is stored using the persistid of the relation container, in the non-persistent case the relation is stored using its unknown pointer and is not written to file.
[0534] 2.2. The Relationship on the Relation Side
[0535] Objects that define/store a relation have to store and enumerate the semantics that apply to the relation. This is achieved by implementing he interface IJRelation. Only the few objects that store relations and efine their behavior need to implement/contain/aggregate this interface. An object can be both a relation and be used as a value by another relation. In other words, an object can support both IJRelation and IJRelationValue.
[0536] 2.3. Connections
[0537] A relation contains one or more connections. A connection is represented to the external world by the IUnknown pointer of the value that the relation connects. Only this IUnknown pointer is used by API to communicate object identity.
[0538] Connections are typically stored internally using the persistid of the connected element, if this element is persistent. A connection elevates the persistid reference count of the object that it connects. When an object with relationships is deleted, the relations may or may not disconnect from the object. If any of the relations choose to keep a connection, the object will not be deleted. Instead, it will become “support only”, that is, it will not be seen by the user and will be deleted when the last relation disconnects.
[0539] There are cases where more than just the persistid might be necessary to establish a connection. For example, a relation that goes across a document boundary probably needs to store a moniker. Also for relations to temporary objects which are not given a persistid, the connection will be stored by keeping the unknown pointer of the connected value. It is up to the relation implementation to allow or not such connections.
[0540] The interface for adding a connection on the relation side is implementation specific; the user cannot directly access connected objects from the relation, and must go through a relation semantic. To avoid redundancy, a unique storage of the connection information could be shared by all the relation semantics.
[0541] 2.4. Semantics
[0542] A relation contains several semantics that describe how the connected objects interact together. These semantics store and manage connections (i.e. the identity of the connected values):
[0543] “MCD semantics” for modify/copy/delete operations. These semantics are understood by the generic relationship management subsystem and are used by the corresponding Relations machine to ensure system wide integrity by a uniform handling of these operations.
[0544] “Application semantics”. The IJRelation interface allows any semantic to be enumerated. Applications define their own semantics to control relation properties like cardinality, ordering, unicity, etc., or to apply special meaning to element connectivity. These added semantics should never conflict with the Relations semantics. Application specific commands can use these semantics for special behavior. Only the MCD semantics are discussed herein.
[0545] A MCD semantic is composed of three parts:
[0546] First, PreNotification Behavior: The pre-notification of a semantic is used for the relation to close on behavior that would not be accessed during transitive closure.
[0547] When a value is selected for a MCD operation (by a batch operation or pRelationNotify→operation), it is interrogated (IJRelationValue::EnumRelations()) for all connections/relationships. Each relation returned in the value's relation enumerator is notified of the pending operation by sending its MCD semantic(s) a pre-notification message with the IUnknown pointer of the selected value as a parameter. The relations/semantics then have an opportunity to mark additional values as RELATION_MODIFIED, RELATION_COPIED, or RELATION_DELETED, depending on the semantic's behavioral requirements.
[0548] Second, Oriented Graph behavior: The “oriented graph behavior” of a semantic is used to search and sort the domain of interest of an MCD operation by building the “domain of interest”.
[0549] The transitive closure (or catching) of elements and relations on which the operation has some impact and should be performed. When an element is selected for modification copy or delete, the relations that it participates in are asked to introduce themselves and to introduce other elements (if they wish) in the list of selected elements. Building this “domain of interest” is semantic specific: It will not be the same for a modify or a copy operation. The different MCD machines will detail later their domain of interest.
[0550] The MDC semantic defines in which order the relation's semantic should be activated to accomplish the MCD operation. A relation semantic sees its connections as oriented and presents them as input, constrained and output values. Oriented connections ensures a re-evaluation function gets triggered after the functions that evaluate its inputs. Oriented connections are a new concept for a copy or delete operation. The idea is to use this ordering to reduce conflict when handling the copy or delete, or in other words let a relation semantic define how an element that it connects to should be copied/deleted and be sure that it will not conflict with any other copy/delete semantic defined by other relations on this object.
[0551] A relation semantic can only act on a connected value that it declares as output and a value can be output of one relation only. If several relations want to influence the same value, they have to declare it “constrained, and they have to present a common referee (the constraint solver), that will be in charge of performing the operation. The constraint solver is passed the entire set of “constrained values” and “constraints” on them. The different semantics of a relation can qualify the same connection differently: For the relation R, the object A could be an input of the delete semantic, and an output of the copy semantic. For a modify operation (IJCompute semantic), independent interfaces (as verified by IJRelationValue::IsInterfaceIndependent()) are viewed as if separate values, hence are allowed to be enumerated as OUTPUT by multiple Compute semantics as long as the interfaces specified in the JIDREF structures are different and independent of each other.
[0552] When the MCD operation is performed, it should be possible for some semantics to stop the operation in progress and cause the corresponding MCD machine to search and sort the domain of interest again before proceeding with the operation. (This is currently only implemented for the modify semantic).
[0553] Third, Operational behavior: This describes what should happen to the relation, or to the objects on the output side of the relation, when an MCD operation happens on any object connected to the relation. Each specific MCD machine will detail its operational behavior.
[0554] 2.5. Typing Relations
[0555] Relations can be disambiguated by assigning each “kind” of relation a unique type. For example, the ownership relation would have a unique type as would the symbology relation. Once the relations connected to a value are disambiguated, an application can unambiguously retrieve a relation by type. The connected relations can also be hashed on type to provide efficient access.
[0556] A relations type is optional and is never interpreted by the Relations engine itself. Relations participate in the modify/copy/delete operations by the semantics that they enumerate. A relation's type is there to give the relation user a fast and reliable way to access connected objects.
[0557] 2.5.1. Assigning Tpye to A Relation
[0558] Relation types are defined by GUIDs. Note that this type is independent on the interfaces supported by the relation container: The symbology bundle, obtained by IJRelationValue::GetRelation(&pRel, SYMTYPE) could support IJOwner.
[0559] The two following member functions of IJRelation deal with typing:
[0560] HRESULT IJRelation::GetType(LPGUID pType);
[0561] HRESULT IJRelation::SetType(LPGUID pType);
[0562] The person(s) responsible for creating a relation would:
[0563] define a GUID for that relation type. For example: “extern GUID OwnerRelationType;”.
[0564] return that GUID from IJRelation::GetType and/or set the type with IJRelation::SetType.
[0565] Note that SetType is mainly for use by the standard relation helper object. A relation can only support one type. An implementor of IJRelation can choose to “hard code” the relation type and return an error from SetType. SetType can only be called once per relation. The Relations subsystem will define a generic relation type: StdRelationType. It should be returned from GetType if no one has called SetType.
[0566] 2.5.2. Finding A Relation by Type
[0567] The following function of IJRelationValue is used to retrieve a relation by type:
[0568] HRESULT IJRelationValue::GetRelation(LPUNKNOWN* ppRelation, LPGUID pType);
[0569] EnumRelations also takes an optional GUID. If the GUID is not NULL, then only relations of the specified type should be enumerated.
[0570] HRESULT IJRelationValue::EnumRelations(LPUNKNOWN* ppRelation, LPGUID pType);
[0571] If more than one relation connected to the value has the same type, GetRelation should return the first relation and the return code RELATION_S_DUPLICATETYPE. All duplicates can be found with IJRelationValue::EnumRelations.
[0572] 2.5.3. Making A Value Non-Dependent on the Implementation
[0573] Since the physical location of a relation is arbitrary with respect to its enumerated values (i.e. it can be contained in a separate object), obtaining a relation from a value does not always guarantee access to the desired object. For example, there could be an implementation of the ownership model where the owner relation was not contained in the owner. In this case more work has to be done to get the owner.
[0574] In such a case information based on relation type is implied. For example, if it is assumed that a relation of type, OwnerRelationType, will always be contained in the owner itself, the owner relation can be retrieved via JRelationValue::GetRelation(&pOwner, &OwnerRelationType). In other words the relation type is a contract, if the owner relation were not contained in the owner, the relation would have a different type
[0575] 2.6. How the Relation Management Subsystem is Made Aware of Element Deletion/Insertion/Copy for Recompute
[0576] The different associative machines are not totally independent the one from the other. When inserting an element in an owner, or when deleting one of its members, the owner itself is modified, hence its dependents in he recompute sense should be re-evaluated. That is, for example, how a “display manager” hooked to the owner interface of a group that is displays, is aware of elements to display/erase. This is achieved as follow:
[0577] 2.6.1. Recompute When an Element Gets Deleted
[0578] When an element is deleted by the user, or marked deleted by the delete machine BatchDelete() (by the command) or pRelationNotify→Delete() (by a semantic) is called. It causes a call to PreValueDelete (deleted_obj,pRelationNotify) for connected relations' delete semantics.
[0579] The implementation of this function tells the MCD machine that one its interfaces will be modified because of this deletion. This is done by callin pRelationNotify→Modify(This,output intf). By doing this, recompute semantics that are directly connected, as well as those that are transitively connected, can be prenotified.
[0580] In this way the notification going down the dependency graph goes up the dependency tree. When getting the PreValueModify message, the dependent could determine by walking the stack (enumerator) of JIDREFs passed as argument, if the element that initiates the notification is modified inserted or deleted.
[0581] The delete machine then complete (close on) the delete operation.
[0582] It is then up to the compute machine to complete the recompute operation. When by the compute machine, the display manager for example could redisplay invalidated regions. Since deleted elements got disconnected from their owner, and the regions' redisplays go down the owner tree, deleted elements will not get redisplayed.
[0583] 2.6.2. Recompute When an Element Gets Copied
[0584] Copying an element is similar to the insertion of a new element. Hence after the copy is performed this is the responsibility of the command to notify the Relations Machine of the copy by using BatchInsert(Clone). This will retrigger an evaluation of the recompute semantic of the relations connected to that clone. As above in the delete case, it will walk up the ownership graph and owner listener will get an Evaluate call as part of this recompute.
[0585] 3. The MCD Machines
[0586] 3.1. The Recompute Machine (Modify)
[0587] 3.1.1. Transitive Closure
[0588] Transitive closure is the term used to describe the process of searching the associative graph for all nodes that need to be recomputed and sorting them (if necessary) into the correct order for recompute. Searching the associative graph is also called “catching” the graph. This search is typically instigated by the Relations subsystem when an element with an associative relationship is modified or deleted. After a modification or delete, related nodes must be re-evaluated to reflect the current status of their relationship. It is transitive because nodes that are not directly related (i.e. they are transitively related) are involved in the search.
[0589] The catch of a graph always starts with a set of modified value nodes (represented by COM identity and interface id (IID)). These value nodes are gathered by the batch or by some application command. The Relations subsystem defines a set of interfaces that are used to find and sort the remaining nodes of the graph. The interfaces of primary interest, that will be explained further below, are:
[0590] IJRelationValue: an interface supported by COM objects that want to represent one or more value nodes.
[0591] IJRelation: an interface supported by COM objects that want to represent one or more function nodes.
[0592] IJCompute: an interface supported by each function node.
[0593] FIG. 16 illustrates a flow diagram of one embodiment of the process of searching the associative graph for all nodes that need to be recomputed and sorting them into the correct order for recompute. The steps include:
[0594] Enumerate dependency. Using IJRelationValue, ask each COM object representing one of the original modified value nodes for the list of COM objects connected to it by a dependency relation or an equivalence relation, step 360. Repeat this process until all transitive relations are found.
[0595] Enumerate nodes. Using IJRelation, expand each of the COM objects above into a set of function nodes, step 370.
[0596] Enumerate values. Using IJCompute, ask each function node for the value nodes it outputs, the value nodes it depends upon or the value nodes it is in an equivalence relation with. The result is a list containing the COM identity, interface id and status of all value nodes related to each function node, step 380. The status describes whether or not a value is an output, an input, or is constrained.
[0597] Sort. The function nodes and related value nodes are sorted with respect to the found relationship: Parents are “less than” children and predicates are “equal to” each other, step 390. (See the simple Example section earlier.)
[0598] After Closure, the nodes are then ready for evaluation.
[0599] 3.1.2. Recompute of Associative Function Nodes
[0600] Associative function nodes are evaluated by calling the function Evaluate on the IJCompute interface. This function is responsible for reading any input value nodes and updating any output value nodes.
[0601] 3.1.2.1. Related Value Nodes
[0602] A sociative function nodes must implement a function on IJCompute to enumerate all values related to the node. Each value is given a status as an input or output.
[0603] Input values represent parents of dependency relations, and they are read from during evaluation using COM interfaces. Output values represent children of dependency relations, and they are written to during evaluation using COM interfaces.
[0604] 3.1.2.2. Return Codes
[0605] The function on IJCompute that is used to evaluate associative functions should return a special return code in the following cases: an evaluation failure; an evaluation does not change the value of the output nodes; and an evaluation function changed the associative graph. If any of these errors occur, typically the graph needs to be searched and sorted again.
[0606] 3.1.2.3. Batch Notification
[0607] The batch is passed into the evaluation function on IJCompute. The batch should be notified during evaluation before an output value is changed.
[0608] 3.1.3. Recompute of Predicate Nodes
[0609] Predicate nodes and their related value nodes are passed to a solver for recomputation. The solver is responsible for updating the value nodes related by the predicates until all predicates are satisfied. The solver communicates with predicate nodes via an application defined interface. This interface represents the semantics specific to a particular solver. For instance, an interface to the D-Cubed solver.
[0610] 3.1.3.1. Related Value Nodes
[0611] Predicate nodes must implement a function on IJCompute to enumerate all values related to the node. Each value is given a status of an input or constrained.
[0612] Input values represent parents of dependency relations. The solver can only read from these values using COM interfaces. Constrained values represent equivalence relations. The solver reads from and/or writes to these values using COM interfaces. A set of constrained values is chosen for writing such that a set of predicates is satisfied. This solver can choose different techniques for satisfying a set of predicates:
[0613] If all predicates on a value node can be seen as accessing independent information, the value node can be independently modified. For example, a begin point and an end point are two independent interfaces on a line; the begin and end points of a line are independent.
[0614] Solve of all the predicates like a system of simultaneous equations. For example EndPoint, BeginPoint, Length are three inter-dependent interfaces of a line that should be solved simultaneously (the length of a line affects the begin and/or end point).
[0615] 3.1.3.2. Return Codes
[0616] The evaluation function on IJCompute is not called for predicate nodes. For Predicate nodes, the solve method that is defined in the IJRelation solver interface is called. If the predicate node cannot be satisfied, the solver will return E_Fail; if the predicate node can be satisfied, the solver will return S_OK. Other return codes are possible and are discussed in further detail in section II.
[0617] 3.1.3.3. Batch Notification
[0618] The batch is passed into the solver. The solver should notify the batch before it updates any value nodes.
[0619] 3.1.4. Information Held During Recompute
[0620] This section describes the types of data stored and manipulated by the Relations subsystem during recompute. This information is stored as status in the graph during recompute.
[0621] 3.1.4.1. Status Information
[0622] For graph triggering and incremental graph
[0623] The nodes indicate whether they change value or not when the evaluation function is recomputed. The graph can be pruned during recompute if a node does not change value.
[0624] When a dependent node recomputes, it should be able to access the recompute state of its parent.
[0625] 3.1.4.2. Error Management
[0626] The nodes indicate if their re-evaluation fails. There are several possible means of failure:
[0627] The evaluation function stored with the node fails to recompute. This is detected by the returned HRESULT.
[0628] A solver failed to satisfy a predicate.
[0629] The node is in error because one of its parents is in error. This means that a node of the associative graph should be able to retrieve the state of its parents when it recomputes. Even though there may be an error, In the preferred embodiment recompute is still sent to the dependent node even if one of their parent's is in error. In such a situation, the dependent decides its own error state rather the recompute engine setting it on behalf of the dependent.
[0630] In addition to the temporary error status stored in the graph, error information must be persistent. Applications need to be able to control if an element is in error, and to find all the elements that are in error.
[0631] In the preferred embodiment, the entire COM object turns to error when one of its node fails to recompute, although it means that some granularity in ormation will be lost when saving this error state. That is, after saving the error at the COM level, all of the nodes that the COM object generate will also appear in error for next recompute.
[0632] To address this granularity issue, the application could perform some interactive and automatic handling of nodes in error, by asking the end user to fix the problems that occurred during last recompute. Fo example this could be done by:
[0633] having a list for the elements in error that will be built up when the element turns to error.
[0634] using an “error manager listener” that is called to handle the error list in a way similar to batch listeners that are notified when elements are modified.
[0635] 3.2. The Copy Machine
[0636] The copy machine is used by commands when they want to copy elements. The copy machine is not part of the batch mechanism and is never triggered automatically.
[0637] The result of a copy operation must be self-sufficient: It must have enough context information to stand alone. For example, when copying elements to the clipboard, one cannot predict where the results will be pasted. It could be to some external container like Word. In this case, the copied elements should include a document object. Our worst case for copy is the EMS™ “file fence command” where the only way to be sure enough context information was copied meant copying the entire document then deleting the elements that were not supposed to be copied.
[0638] FIG. 17 illustrates a flow diagram of one embodiment of the process of copying objects. The steps include:
[0639] 1. Starting from an initial list of objects to copy, step 410, (for example, the select set defined by the user), scan all relationships for their copy semantics to build a set of objects to participate in the copy operation, step 420. This is called getting the domain of interest, step. (Also known as the “catching” the graph).
[0640] 2. Mark the set of objects to be copied, step 430.
[0641] 3. Clone any objects that are input only using IJPersist::Clone, step 440.
[0642] 4. Ojects that are marked as output of a copy semantic are cloned by the copy semantic that declares the output using IJCopy::Createclones, step 450.
[0643] 5. All of the objects cloned are assigned persist IDs (if the original object had a persist ID).
[0644] 6. The relationships between the clones are re-established using IJCopy::ReconnectClones, step 460.
[0645] 3.2.1. Extend the Set of Elements Selected for Copy.
[0646] When a command copies elements, the first step is to use the CopyJObjects() global function to indicate which elements are operated on. Other elements will then be selected simultaneously for copy because of existing relationships. For the MCD machine, this is very similar to the other elements being selected directly by the command.
[0647] When an element is selected for copy, and if the element was not previously selected (marked RELATION_COPIED), the copy semantics (contained by elations that the element is connected to) receive a message. Thes semantics can then decide to select other values for copy, by calling pRelationNotify→Copy(OtherElem). This marks OtherElem RELATION_COPIED.
[0648] This copy process is transitive on elements newly marked RELATION_COPIED by calling recursively the PreValueCopy function of copy semantics of relations that they are connected to. Since the RELATION_COPIED bit is not set on the objects returned from IJCopy::EnumValues, a copy semantic can distinguish between these elements and the ones selected by the user or added via IJCopy::PreValueCopy.
[0649] 3.2.2. Search and Sort the Domain of Interest
[0650] The aim of this step is to build a self sufficient set of elements, even if the elements have not been selected for copy, and to order the copy semantics of their relations. It is not desirable to capture the entire document. For example if a symbology bundle enumerates all the objects using that symbology, this will end up getting the entire document in the “domain of interest” and it defeats the goal.
[0651] An OUTPUT means that the modify /copy /delete action of the semantic will redefine this element. Thus an element can only be output of a single semantic for each type of MCD action.
[0652] An INPUT means that the MCD action of the semantic needs to access this element in order to implement the action and give a value to the OUTPUT.
[0653] Building the domain of interest with transitive outputs applies correctly for the Compute machine. It is however difficult to apply in the copy and delete machines. For example, a symbology bundle is declared input argument of the relation: bundle→graphic object. In order to implement copy correctly, the domain of interest must contain:
[0654] the relation between the symbology bundle and the symbology manager;
[0655] the relation between the symbology manager and the document;
[0656] the symbology manager;
[0657] and the document.
[0658] In addition to the qualification RELATION_INPUT/RELATION_OUTPUT, a semantic can qualify the connected argument as NO_TRANSITIVE_EXPANSION.
[0659] Th copy and delete engine will visit transitively all enumerated values that are not qualified NO_TRANSITIVE_EXPANSION. By visiting transitively, it means adding in the domain of interest, the relations that this value contains (IJRelation interface) plus the relation this value participates to (IJRelationValue interface).
[0660] As an optimization, a semantic may not enumerate its input arguments. It must then return from the EnumValues function, the return code RELATION_S_IMPLICIT_VALUES (The symbology bundle does not enumerate the graphic objects using it). This is very similar to enumerate arguments and qualify them as RELATION_INPUT|NOT_TRANSITIVE_EXPANSION. Any value that is for other reasons in the domain of interest and enumerates the relation will then be considered as one of its input arguments and passed accordingly to the CreateClone/ReconnectClone function.
[0661] Searching and sorting of the domain is done by starting with an initial set of values (objects) to be copied (copyJObjects arguments+the expansion done by PreValueCopy). The initial set is represented by the value array {Vinit}, containing IUnknown pointers. Each of its elements is marked RELATION_COPIED.
[0662] Next, find any additional objects part of the domain of interest when copying the initial set, and find the relations between all these element. (These elements are not marked RELATION_COPIED.) This is achieved by:
[0663] Each object enumerating the relations it is connected to by IJRelationValue::EnumRelations. These relations are accumulated in the relation array {R}.
[0664] If the object itself contains a relation, this relation is also added to {R}.
[0665] Each relation copy semantic enumerates the objects (IJCopy::EnumValues) that participate in the copy operation and qualifies them as INPUT, OUTPUT, plus the optional qualifier: NO_TRANSITIVE_EXPANSION and optionally returns RELATION_S_IMPLICIT_VALUES.
[0666] This process is then repeated until no more objects/relations are added. The result is a set of values/relations to be copied: {Vinit}==>(Copy Semantics)==>{Vclosure, Rclosure}
[0667] Next, the values and relations are ordered in such a way that a relation that declares a value as output is listed before a relation that takes it as input.
[0668] 3.2.3. Clone the Elements
[0669] The objective in this step to establish a “clone graph” from the “original graph” gathered in the step above. Each object in the original graph will have a representative in the clone graph. This representative is context dependent (i.e. depends on the intent of the copy command (copying from a symbol library into a document is different than copying to the clipboard). The representative can be:
[0670] the original object itself,
[0671] a clone of the original,
[0672] an element selected by the user, etc.
[0673] As a result of the ordering done above, when asked to copy, each copy sematic has in hand the information it needs about its target context, because its input values will already be represented in the clone graph.
[0674] The distinction between elements initially selected, or just in the domain of interest can be made by looking at the RELATION_COPIED state. Only elements of the first kind have this status. By testing it, it allows the IJCopy::CreateClones function to behave differently as exemplified by the copy semantic of owner relation, and copy semantic of the symbology manager.
[0675] Any objects not declared as output or constrained gets its representative in the clone graph from either: the copy command itself, or the copy machine. The copy command can clone the element using IJPersist::Clone or choose another representative such as an existing object or user selected object. The copy machine will clone the object via IJPersist::Clone if the object is not marked RELATION_OUTPUT (i.e. no semantic is responsible for producing the clone), or if the object has not been cloned already by the command.
[0676] Each copy semantic, is then asked to produce a representative for its output values. The semantic could: Clone the output by IJPersist::Clone; Use the original element as the representative in the clone graph; Produce/retrieve an output using the clones of its inputs; or Produce NO output. (The corresponding element will be represented as NULL in the clone graph.
[0677] This step will not define the persistid of elements in the clone graph.
[0678] 3.2.4. Assign Persistent Identity to the Clones
[0679] Only after all the elements have been copied, the elements in the clone graph get a persistid as follows. First, for at least one object, the document clone gets a persistid during clone, then for all the other objects, if the original does not have a persistid, or if the clone already has a persistid, do nothing. If the original has a persistid, the following steps are performed: find the persist manager of the original; find the clone of the persist manager; use the persist manager clone to give the element clone a persistid.
[0680] 3.2.5. Establish Connections in the Clone Graph
[0681] The intent of this step is to complete the copy action by connecting all “new” clone objects to their relations in the target environment.
[0682] This is done by asking each original relation in the clone graph to establish connections for its clone in the target environment. (All clone objects now have a persistid if the original object did.) Each copy semantic is asked to reestablish connection by IJCopy::Reconnect. The relation reconnects by using the clone map to identify its clone and the clones of its connected values.
[0683] At this point, the behavior is relation specific. The copy semantic may use public or private interfaces to accomplish its task. At some point in the process, it must tell the values to reconnect by sending them IJRelationValue::ConnectRelation(R punk) (except for the values in the same object if any and object that are already connected).
[0684] Since every relation and value in the clone graph at this point has a clone and the clone could be an already existing object or a newly created one depending on the context of the MCD operation, the semantic must check if its connected values' clones are connected, if not, make the connection.
[0685] 3.3. The Delete Machine
[0686] The Delete machine is activated by a command when it calls BatchDelete(elem), hence the batch activates the delete machine. The extension of the set of elements for delete and the searching and sorting of the domain of interest is similar to what was described for the copy engine. The goal of the delete engine is to close on a delete operation such that the objects that are selected for delete are removed from the document and that referential integrity is maintained. Objects that rely an object marked for delete, must either be deleted or the marked object must remain as a support object (non-viewable, non-locatable) until the application can resolve the reference (maybe provide an alternate connection).
[0687] FIG. 18 illustrates a flow diagram of one embodiment of the process of the deleting process.
[0688] 3.3.1. Extend the Set of Elements Selected for Delete
[0689] When a semantic gets PreValueDelete message because one of its connected values has been selected for delete by a call to BatchDelete(elem), the function decides to extend the deletion to other arguments of the semantic by calling RelationNotify→Delete(otherObj), step 1100. This occurs recursively if the other objects participates with other relations. All the elements are marked RELATION_DELETED.
[0690] 3.3.2. Search and Sort the Domain of Interest
[0691] Next, the domain of interest is searched and sorted, step 1110, in a way similar to the search and sort of the copy engine. First, all values enumerated by a semantic are visited transitively if not specified NO_TRANSITIVE_EXPANSION. Next, if a semantic's EnumValues function return RELATION_S_IMPLICIT_VALUES, all the values in the domain of interest that enumerate this relation as a connected relation will be considered as INPUT argument of the semantic and passed accordingly to its disconnect function.
[0692] Elem ents introduced by EnumValues are not marked RELATION_DELETED. Hence delete semantics can distinguish between elements that are part of the initial selection and PreValueDelete from elements introduced as part of EnumValues.
[0693] The search and sort is implemented as follows: Starting with an initial set of objects to be deleted, defined by BatchDelete() and its expansion performed as 3.2.1. The initial set is represented by an array of IUnknown pointers, {Vinit}. These objects are already marked RELATION_DELETED.
[0694] Next, find the relations on these elements: each object enumerates the relations it is connected to by IJRelationValue::EnumRelations. These relations are accumulated in the relation array {R}, and if the object itself contains a relation, this relation is added to the same array; then each relation will then be asked to enumerate its related objects (IJDelete::EnumValues). This process is repeated until no more objects/relations are added. The result is a set of values/relations to be deleted/disconnected.: {Vinit}==>{Delete Semantics}==>{Vclosure,Rclosure}
[0695] Next, order the values and relations in such a way that a relation that declares a value as output is listed before a relation that takes it as input, step 1120.
[0696] 3.3.3. Disconnect
[0697] This stage involves telling the affected relations to take whatever action is appropriate for any related object marked as deleted, step 1130. As previously mentioned the object may not actually be disconnected from all its relationships, but becomes support only, this decision being made by the delete semantics. After testing the RELATION_DELETE state of its inputs/output, each delete semantic can: disconnect from its input/output by IJRelationValue::DisconnectRelation or keep the connection, the input becoming support only; and mark its output deleted, or transform it. (This could require a recompute of the dependency that will be accumulated by the batch for future handling.)
[0698] It should be noted that an element disappears for the user (becomes support only) when it is selected for delete because the semantic of its owner disconnects unconditionally from elements marked RELATION_DELETED
[0699] 3.3.4. Release Last OID Reference Count
[0700] This only applies to persistent objects, step i.e. objects for which the following calls succeed:
[0701] GetPersistManager(pObject, &pPersistMgr); and
[0702] pPersistMgr→GetPersistID(pObject&Oid, NULL).
[0703] When an element has been initially made persistent by calling MakeobjectPersistent(), its oid reference count has been set to 1. Each time another objects connected to it, its oid reference count has been elevated when the relation was established and stored as a persistid.
[0704] If after calling all the delete semantics of the relations that an object participates, its oid ref count is back to one, it means that nobody cares any longer about this object. It could be removed from the control of the Persist Manager by releasing this last oid ref count. The object will not be permanent any longer and will be physically destructed when its COM ref count goes to 0.
[0705] If the object still has an elevated OID, some relations did not disconnect from it, therefore, it is still known to the PM and becomes a “support only element” (i.e. not existent from the user's point of view) if disconnected from its owner.
[0706] 4. Example of the Relations Subsystem
[0707] The Relationship Subsystem is responsible for ensuring that the integrity of the data model is maintained during the manipulation of objects that are related. This section provides a description of the steps that are taken to create relationships between objects, and the steps taken when these relations and objects are modified, copied or deleted via examples
[0708] 4.1. Creating Relationships
[0709] FIG. 19 illustrates three pipes 1210, 1220, and 1230 are connected to a tee connector 1024 in a constraint relationship. This means that when pipe 1230 moves, the tee 1240 must also move, and if tee 1240 moves, pipe 1230 must also move. One additional relation depicted in FIG. 19 is the dependent relation between tee 1240 and the callout text object 1250. In this relation, if tee 1240 is moved, then the call out 1250 will stay connected to tee 1240; however, if call out text 1250 is moved tee 1240 will not be moved.
[0710] 4.1.1. Establishing the Pipes to the Tee Constraint Relation
[0711] The first step is establishing the pipes to the tee constraint relation. This is done by the following steps:
[0712] 1. Obtain a IUnknown pointer to the pipe objects 1210, 1220, and 1230 and tee 1240.
[0713] 2. The command will create a Tee Connector semantic object. This semantic object will implement the IJCompute, IJCopy, and IJDelete interfaces. The implementation of these interfaces will be described below.
[0714] 3. Give the IUnknown pointers obtained in step 1 and the associated interface IID's to the compute semantic object. The interface IID for the pipes is the IID_IJPipeEndPoint and the IID for the Tee is IID_IJTeeConnector.
[0715] 4. Create the relation object and obtain IUnknown pointer to the relation.
[0716] 5. Obtain the IJRelation interface for the Relation object.
[0717] 6. Add the semantic to the relation object by calling IJRelation::AddSemantics and passing in the IUnknown pointer to the semantic object.
[0718] Th relation then:
[0719] keeps the persistent id of the semantic;
[0720] obtains the IJRelationValue interface of the three pipes 1210, 1220, and 1230 and tee object 1240
[0721] gives the relation values a pointer to itself by calling their ConnectRelation methods; and
[0722] release each IJRelationValue interface.
[0723] 7. Call::BatchInsert() for any new values the command created and subsequently call ::BatchFlush(). For this example, the relation object connecting the tee to the pipes is added to the batch by calling ::BatchInsert and passing in the IUnknown pointer to the relation object.
[0724] 8. The relation manager then builds and sorts a graph of the newly inserted values guaranteeing that values are stable and the data model intact. The building of the graph is done by performing the following step for each value IUnknown pointer:
[0725] obtain a pointer to the IJRelationValue interface.
[0726] enumerate relations that the value is connected to by calling IJRelationValue::EnumRelations.
[0727] for each IJRelation interface in the enumerator, enumerate semantics by calling IJRelation::EnumSemantics.
[0728] for each compute semantic call the IJCompute::EnumValues method, and determine if the value is an input to the semantic.
[0729] if the value is an input to the semantic, add all values to the graph and repeat these steps on all subsequently added values.
[0730] The graph is then sorted by determining by connection type an order as follows:
[0731] Inputs (or Parents) <(before) Outputs (or Children)
[0732] Constrained (or equivalence) are equal in order
[0733] Each semantic that has an output in the graph is then evaluated.
[0734] 9. The command then releases the following pointers:
[0735] IUnknown pointers to the pipe objects 1210, 1220, and 1230 and the tee 1240.
[0736] IJRelation interface.
[0737] The IUnknown pointer to the relation object.
[0738] The IUnknown pointer to the semantic object.
[0739] 4.1.2. Establish the Dependency Relation Between the Tee and the Call Out Text Object
[0740] The next step is establishing the dependency relation between Tee 124 and the call out text object 1250. The text is dependent upon the tee. his is done by the following steps:
[0741] 1. Obtain a IUnknown pointer to the tee 1240 and the text object 1250.
[0742] 2 The command will create a Text with Leader Line compute semantic object. This semantic object will implement the IJCompute, IJCopy, and IJDelete interfaces. The implementation of these interfaces will be described below.
[0743] 3. Give the IUnknown pointers obtained in step 1 and the IID_IJTeeConne tor interface IID for the Tee 1240 and the IID_IJText interface for the text object 1250.
[0744] 4. Create the relation object and obtain IUnknown pointer to the relation.
[0745] 5. Obtain the IJRelation interface for the Relation object.
[0746] 6. Add the semantic to the relation object by calling IJRelation::AddSemantics and passing in the IUnknown pointer to the semantic object. The relation then
[0747] keeps the persistent id of the semantic
[0748] obtains the IJRelationValue interface of the three pipes 1210, 1220, and 1230 and tee objects 1240.
[0749] gives the relation values a pointer to itself by calling their ConnectRelation methods.
[0750] releases each IJRelationValue interface.
[0751] 7. Call ::BatchInsert() for any new values the command created such as the text object and subsequently call ::BatchFlush().
[0752] 8. The relation manager will then build and sort a graph of the newly inserted values guaranteeing that values are stable and the data model intact. The building of the graph is done by performing the following steps for each value IUnknown pointer:
[0753] obtain a pointer to the IJRelationValue interface.
[0754] enumerate relations that the value is connected to by calling IJRelationValue::EnumRelations.
[0755] for each IJRelation interface in the enumerator, enumerate semantics by calling IJRelation::EnumSemantics.
[0756] for each compute semantic call the IJCompute::EnumValues method, and determine if the value is an input to the semantic.
[0757] if the value is an input to the semantic, add all values to the graph and repeat these steps on all subsequently added values.
[0758] The graph is then sorted by determining by connection type an order a follows:
[0759] Inputs (or Parents)<(before) Outputs (or Children)
[0760] Constrained (or equivalence) are equal in order
[0761] Each semantic that has an output in the graph is then evaluated.
[0762] 9. Release the following pointers:
[0763] IUnknown pointers to the tee 1240 and the text object 1250.
[0764] IJRelation interface
[0765] The IUnknown pointer to the relation object.
[0766] The IUnknown pointer to the semantic object.
[0767] 4.2. Modify Operation
[0768] The Relationship subsystem is responsible for maintaining referential integrity of the entity-relationship data model during manipulations of that data model. The initial set of objects to be modified is provided by a command. The Relationship Manager will then determine which related objects will also need to be modified.
[0769] FIG. 20 illustrates three pipes connected to a tee connector in a constraint relationship being modified. When tee 1240 is moved upwards to become tee 1245, pipes 1210, 1220, and 1230 follow. Furthermore, the callout text object 1250 also follows tee 1240. The steps that are taken during this move operation are described below.
[0770] FIG. 21 illustrates a graphic representation of the COM relations of FIG. 20. The Modify Operation is performed by the following steps.
[0771] 1. The Relation Manager is provided with a set of objects that are to be modified. Note, before Tee 1240 is modified, the command calls the global function BatchModify passing in the IUnknown pointer to provided by the Relation manager. The pointer to the Tee object will be added to a list of modified values maintained by the Relation manager.
[0772] 2. The Relation manager obtains the IJRelationValue interface 129 for object passed into ::BatchModify() and gets an enumerator for all of the relations that the object is connected.
[0773] 3. From the IJRelation interface 1290 pointers obtained from the enumerator, the Relation manager obtains the IJCompute interfaces connected to the value. The IJCompute::PreValueModify method for each compute semantic is called passing in the IUnknown pointer to the related values, the interface IID to be modified and a flag that indicates the type of connection. In addition, a pointer to the batch which allows the semantic add additional values to the graph. Note, for this example, the tee-pipe compute semantics will simply return S_OK when its IJCompute::PreValueModify method is called.
[0774] 4. The objects are then modified by the command. In this case the geometry definition of the tee would be transformed to the new location.
[0775] 5. The Relation Manager is informed that the command done with all pending modify operations. This is done by calling ::BatchFlush().
[0776] 6. The Relation manager will build a graph starting with the modified values as follows:
[0777] For each semantic obtained in step 3, call IJCompute::Enumvalues() to obtain an enumerator for the values connected to that semantic. Note, the tee-pipe compute semantics will return an enumerator that contains pointers to the IUnknown pointer and interface to the values that it is connected.
[0778] For each value in the enumerator, determine if the value is already in the graph, if not add it to the graph and execute steps 3 and 6 recursively for any newly added values.
[0779] 7. The Relation manager will sort the graph. This means that it will determine by input, output and constrained values an ordering where parents (inputs) are before children (outputs), and constrained values are equal. Equal values are passed to a constraint solver.
[0780] 8. The Relation manager will then process the graph. For constraint relations, the IJRelationSolver::Solve() method will be called for the solver associated with the constraint relations. For parent/child relationships the Relation Manager will call the IJCompute::Evaluate.
[0781] 9. The solve method is implemented per a private contract between the relation-values and the constraint solver.
[0782] Note, the IJRelationSolver::Solve() method for the constraint relation will be called by the Relation Manager first. The input to this function includes an enumerator of the values that participate in the system to be solved. In this example, the values will be the three pipes with the interface being IID_IJPipeEndPoint 1300, 1310, and 1320, and the tee object with the interface being IID_IJTeeConnector 1330. Furthermore, the two modified pipes 1210 and 1220 will be tagged as being modified. An enumerator of semantics that participates in the system is also an input argument to the Solve method. In this example, the compute semantics for the constrain relation will be the only semantic in the enumerator,
[0783] Note additionally, input to the Solve method is the mode which can be set to either RELATION_FULL_CONSTRAINT or RELATION_PARTIAL_CONSTRAINT. The mode for full constraint indicates that a full set of values and constraints acting between them has been searched. The mode for partial constraints indicates that some of the values passed to the solver can be frozen because the Relation Manager was set up for incremental solving. In this case it is the responsibility of the solver to determine if the required values are present and writable to solve the system. If a frozen value is required to be modified in order to solve the system, the solver will return a status code of RELATION_S_INCREMENTAL_SEARCH and return the values that need to be modified in th ppAddedValues enumerator. The solve method will need to ensure that any of the values that need to be modified are not frozen. This means that none of the pipes 1210, 1220, and 1230 or tee 1240 can be frozen, if any one of them has been modified. Any values that are frozen will need to be added to the ppAddedValues enumerator.
[0784] If the pipes and tee objects are not frozen, the solver for the tee relation compute semantics will compute the new location using the location of the modified pipes as a basis for the new values.
[0785] 10. After the constraint system has been solved, the dependency relation between the tee object 1240 and the callout text 1250 will be evaluated. Using the relative location of the callout text with respect to the tee that was determined when the relation was established, the new text location will be computed when the IJCompute::Evaluate method is called. Once the new location is computed, the callout text's geometry will be updated, after calling pNotify→Modify passing in the text's LPUNKNOWN pointer and the IJGeometry as arguments.
[0786] 4.3. Copy Operation
[0787] The copy machine is used by commands to copy objects. The initial set of objects to be copied is provided by the command. The copy machine will then determine which related objects will also need to be copied.
[0788] FIG. 22 illustrates three pipes connected to a tee connector in a constraint relationship where two pipes are selected for being copied. Pipes 1210, 1220, and 1230, tee connector 1240, and call out text 1250 are similar to FIG. 19, and pipes 1210 and 1230 are selected for being copied.
[0789] The Relation machine provides a global function, ::CopyJObjects(), that copies the selected objects. The following steps are taken to ensure that all related objects that should be copied are included in the copy operation.
[0790] FIG. 23 illustrates a graphic representation of the COM relations of FIG. 22 before the copy process. Note that only the copy semantics are shown for the relation objects. The Copy Operation is performed by the following steps.
[0791] 1. The Copy Machine is provided with a set of objects that are to be copied (Pipes 1210 and 1230). Note, the command calls ::CopyJObjects passing in the objects to be copied in an array and the copy map.
[0792] 2. The Copy Machine builds an initial graph using the objects selected for copy in pObjectsArray by for each value in the list:
[0793] obtain a pointer to the IJRelationValue interfaces 1310 and 1320.
[0794] enumerate relations that the value is connected to by calling IJRelationValue::EnumRelations.
[0795] for each IJRelation interface in the enumerator, enumerate semantics by calling IJRelation::EnumSemantics; obtaining copy semantics (IJCopy).
[0796] 3. Recursively ask each object in the graph for additional objects that must be copied. This is done by sending a IJCopy::PreValuecopy() message to each of the selected objects. The IJCopy::PreValuecopy will add additional objects (that it depends on) to the copy operation. Note, in this case the connector relation will not add any other values to the copy set so nothing is done in this method.
[0797] 4. The copy machine then obtains transitive closure by sending each of the relation objects in the graph an IJCopy::EnumValues() message to obtain transitive closure. Note, the tee relation object would typically enumurate the pipes as input values and the tee as an output value for the copy operation, but in this case however, only the output Tee is enumerated (outputs are mandatory) and the RELATION_S_IMPLICIT_INPUTS code is returned. The inputs are not enumerated by the relation here because the Copy Engine will guarantee clones of all inputs in the graph and the copy semantic implementation wants the flexibility to maintain relations between objects marked RELATION_COPIED, without having to copy extraneous objects.
[0798] 5. The Copy Machine will sort the graph
[0799] 6. The Copy machine will copy any objects that are input-only using the objects IJPersist::Clone method.
[0800] 7. The Copy machine will send all other objects in the graph an IJCopy::CreateClones message. The copy semantic that declares the output (outputs such as members or dependent values) creates the clone. Note, it is at this stage that the tee relation will determine that only two of the pipes were copied, and that the tee must be converted to a elbow. An elbow object 1280 will be created and added to the clone map in place of the tee object.
[0801] 8. The relation manager then assigns persist IDs to all objects in th clone map (if the original object had a persist ID).
[0802] 9. Send each original object in the clone map an IJCopy::ReconnectClones message. This method will establish a connection between the copied values and the copied relation. This process will be similar to the steps taken when the relation was created.
[0803] obtain the IJRelationValue interface of the two pipes 1260 and 1270 and elbow object 1280
[0804] give the relation values 1260, 1270, and 1280 a pointer to the relation object by calling their ConnectRelation methods.
[0805] release each IJRelationValue interface.
[0806] 4.4. Delete Operation
[0807] The delete machine is used by commands to delete objects from a document. The initial set of objects to be deleted is provided by the command. The delete machine will then determine which related objects will also need to be deleted.
[0808] FIG. 24 illustrates three pipes connected to a tee connector in a constrain relationship where one pipe is selected for deletion. If Pipe 1210 is deleted, the behavior that is desired is that Pipe 1220 and Pipe 1230 remain and Tee 1240 is changed to Elbow 10.
[0809] FIG. 25 illustrates a graphic representation of the COM relations of FIG. 24 before the deletion process is started.
[0810] In order to provide this behavior, the delete machine implements the following procedure which allows the relation objects to provide their desired delete semantics. The Delete Operation is performed by the following steps.
[0811] 1. The Relation Manager is provided with a set of objects that are to be deleted when the command calls ::BatchDelete and pass in the IUnknown pointer for all objects that it selected to be deleted.
[0812] 2. The Delete Machine will send a PreValueDelete message to each relation object directly connected to the values that are passed into ::BatchDelete. For our example, nothing is done by the Tee relation on this method.
[0813] 3. The Relation Manager is informed that all pending delete requests by the command are completed when the command calls ::BatchPlush()
[0814] 4. The Relation Manager gets transitive closure on the delete graph by getting from each relation's delete semantics its enumerator, making sure all outputs have been checked for possible dependencies. It obtains the enumerator from each relation by calling the IJDelete::EnumValues() method. Note, the IJDelete::Enumvalues() method for the Tee would enumerate the three pipes 1210, 1220, and 1230 as inputs and the Tee 1240 as its output.
[0815] 5. Upon closure, each relation in the select set (including those who where IJRelationNotify::Deleted) is sent an IJDelete::Disconnect message, and expected to disconnect from the values it is dependent on (so that they may be deleted and have no outstanding reference count).
[0816] In the example, since the tee relation is given only one of its dependencies pipe 1210 for disconnect, it determines that it must convert the output tee 1240 into an elbow 1300.
[0817] FIG. 26 illustrates a graphic representation of the COM relations of FIG. 24 after the pipe has been deleted. The delete semantic performs the conversion as follows:
[0818] Creates an elbow object 1300.
[0819] Connect the input values not marked RELATION_DELETED pipes 1220 and 1230 to the elbow,
[0820] Notify the batch that the elbow was created (pNotify→Insert)
[0821] Notify the batch that the tee 1240 must be deleted (pNotify→Delete) and that it is replaced by elbow 1300
[0822] Restart the batch—As in IJCompute::Evaluate, IJDelete::Disconnect will return the RELATION_S_SEARCH code to tell the relation manager to accumulate additional objects for delete.
[0823] 6. At this point the Text Dependent object will receive a IJDelete::PreValueDelete() message. It will not do anything at this point.
[0824] 7. At the IJDelete::EnumValues() method, the text dependent delete semantic will enumerate the Tee as input and the Text object as output
[0825] 8. The second time that the Tee's delete semantic receives the IJDelete::DisConnect() message, the delete semantic will disconnect from the Tee.
[0826] 9. The Tee will have one remaining connection, the Text Dependent 1250, which will receive an IJDelete::Disconnect message with the Tee in the enumerator passed in.
[0827] 10. The Text dependent will look at the structure identifying the Tee for disconnect and test the flag field of the JIDREF structure for (RELATION_DELETED|RELATION_REPLACED).
[0828] If true, the relation will QueryInterface the enumerator object for an interface of type EnumJIDCHANGED, which will indicated any replacement object IUnknown pointers for objects that were deleted. The Text relation will then disconnect from the Tee and connect to the object(s) identified in the EnumJIDCHANGED object.
[0829] If not true, the text relation may pNotify→deleted the text object and return the RELATION_S_SEARCH code OR return and error code indicating that the relation was not disconnected (an implementation detail).
[0830] 5. Detailed Description of the Undo Feature
[0831] The following section provides a detailed walkthrough of the mechanics of the Undo facility. It lays out the steps that must be taken by all object to achieve the desired undo effect. The Undo API argument specifics are found in section II. FIG. 27 illustrates a flow diagram of one embodiment of the process of the Undo feature.
[0832] In this document it is assumed that each document object that will support an undo feature will aggregate an undo manager (supporting the IJUndoManager interface) for the logging and processing of recovery and undo information.
[0833] The Undo Manager maintains a log (the undo log) of document modifications by logging the identity of modified objects (or entities) with the data associated with object state before modification.
[0834] 5.1. Undo Checkpoints and Commands
[0835] For Undo to be usable, commands or the command manager/dispatcher must be involved in identifying and naming distinct points in a document's change history. This is done via the Undo Manager's IJUndoManager::Checkpoint() method. When a command is selected by the user a wide character string identifying the command is passes to the undo manager to indicated the beginning of a transaction. All subsequent modification logs will be associated with the checkpoint until another call to the IJUndoManager::Checkpoint() method is made. This is illustrated in FIG. 27, step 1300.
[0836] Th checkpoint that is maintained is not a complete copy of the document, but a point in the undo log for which to revert all changes to the document (the collection of all persistent objects associated with the document object.)
[0837] 5.2. The Batch/Relation Manager's Role
[0838] Undo is tightly integrated into the Relations subsystem Batch. (See the Relation Subsystem API document in the appendix for a detailed description of the Batch and other relation interfaces). Every time that a command intends to modify, create, or delete an object or relation, it will go through that Batch and call BatchModify(), BatchInsert, or BatchDelete respectively. This is illustrated in FIG. 27, step 1310. These APIs will consequently inform the Relation Manager of the pending modification. At this level, and the RelationNotify interface, that undo logging is initiated by the Relation Manager.
[0839] BatchInsert() and RelationNotify::Insert() are called after an object (entity) is created and relationships have been established with it. It is al called when a non-aggregated relation object is created.
[0840] BatchModify() and RelationNotify::Modify() are called before an object (entity) to be modified is modified. (RelationNotify::ModifyRelation() is called before a relation (or semantic) has a static value to be modified or the identify of a connected value is being changed.)
[0841] BatchDelete() and RelationNotify::Delete() are called to asked the relation subsystem to remove (or delete) the selected objects).
[0842] During the processing of these functions, the relation manager will call LogInsert(), LogModify(), or LogDelete(), which will:
[0843] Determine what document the object is associated with by looking for a Persist Manager associated with the object. If no Persist Manager found then no logging will be performed for the object (return(NOERROR).
[0844] Determine if the document associated with the object has aggregated an Undo Manager by calling QueryInerface(IID_IJUndoManager, &pUndoMgr). If NO Undo Manager then no logging will be performed for the object (return(NOERROR).
[0845] Tell the Undo Manager to log the creation of the object by calling the interface member function IJRelationManager::Log().
[0846] The IJRelationManager::Log() method writes information to the undo log indicating the type of action (Insert or Create) and the persistent identity of the object.
[0847] The IJRelationManager::Log() method writes information to the undo log indicating the type of action (Modify) and the persistent identity of the object. It then asks the object for a pointer to its IJUndo interface.
[0848] If the IJUndo Interface is supported, it calls IJUndo::Log(iid, pStream) asking the object to save the data associated with the give interface to the specified Istream interface pointer (the Undo Log).
[0849] If the IJUndo Interface is not supported, the IJPersist interface is requested from the Object (all persistent objects will support IJPersist to save their contents to a file). Subsequently, the IJPersist::Save method will be called telling the object to save its data to the undo log.
[0850] The IJRelationManager::Log() method writes information to the undo log indicating the type of action (Delete) and the persistent identity of the object. It then asks the object for a pointer to its IJPersist interface.
[0851] The object is then told to save itself to the undo log by calling IJPersist::Save();
[0852] 5.3 The Undo Command and Undo Processing
[0853] Using the Undo Manager, an Undo Command can be written as follows:
[0854] Obtain a list of checkpoints from the Undo Manager—This is done by calling IJUndoManager::GetUndoCheckpoints()
[0855] Present a list of checkpoint names (from the returned enumerator) to the user
[0856] Call IJUndoManager::Undo with the checkpoint structure associated with the name selected by the user. There may be more than one checkpoint with the same name, therefore the corresponding structure will provide the user with the desired result.
[0857] For each of the checkpoints that were logged between the selected checkpoint and the current document state, a redo checkpoint will be created and a discrete set of objects will be modified. Therefore, for each of the checkpoints the following will be performed:
[0858] If a redo log stream does not exist, create one
[0859] Hold the Undo Log pointer in local storage and set the redo log as the undo log checkpoint the transaction by calling IJUndoManager::Checkpoint() with the name of the undo checkpoint.
[0860] read the Undo Stream and record the identity, modification type and stream location of modified data
[0861] process the recorded information as follows:
[0862] “RELATION_DELETED”
[0863] Create the object that was deleted restoring its original persistent identifier.
[0864] Call the object's IJPersist::Load method with a pointer to the Undo Stream (Stream set to the location recorded in step 4)
[0865] Call BatchInsert(pObjectUnknown)—this writes a log to the Redo Stream.
[0866] “RELATION_MODIFIED”
[0867] Determine if the object was saved via IJUndo of IJPersist:
[0868] if IJUndo—Call BatchModify(pObjectUnknown, iid) then Call IJUndo::Revert(pStream) (Stream set to the location recorded in step 4).
[0869] else—Call BatchModify(pObjectUnknow, TBDiid) indicating all interfaces will change and then call IJPersist::Load(pStream). (Stream set to the location recorded in step 4)
[0870] “RELATION_INSERTED”
[0871] Call BatchDelete(pObjectUnknown)
[0872] In each case, all batch operations and subsequent RelationNotify:: . . . calls are logged to the Redo Stream.
[0873] Call BatchFlush() for each of the checkpoints. This will guarantee a closed set of elements changed for each checkpoint and the ability to use the redo log for undoing an undo operation (redo).
[0874] Upon completion of the last checkpoint (the one selected by the user), set the Undo Log pointer back to the address stored in step 2 and set the stream to point to the location of the user selected checkpoint.
[0875] 5.4. The Redo Command and Redo Processing
[0876] Using the Undo Manager, a Redo Command can be written as follows:
[0877] Obtain a list of checkpoints from the Undo Manager—This is done by calling IJUndoManager::GetRedoCheckpoints()
[0878] If the last command issued by the user for the given document is not an Undo Command, no checkpoints will be returned and the HRESULT code returned by the call will be E_FAILED.
[0879] The Redo Log will always be deleted after the first non-Undo co mmand is issued.
[0880] Present a list of checkpoint names (from the returned enumerator) to the user
[0881] Call IJUndoManager::Redo with the checkpoint structure associated with the name selected by the user. There may be more than one checkpoint with the same name, therefore the corresponding structure will provide the user with the desired result.
[0882] For each of the checkpoints that were logged between the selected check point and the current document state, an undo checkpoint will be created and a discrete set of objects will be modified. Therefore, for each of the checkpoints the following will be performed.
[0883] checkpoint the transaction by calling IJUndoManager::Checkpoint() with the name of the redo checkpoint.
[0884] read the Redo Stream and record the identity, modification type and stream location of modified data
[0885] process the recorded information as follows:
[0886] “RELATION_DELETED”
[0887] Create the object that was deleted restoring its original persistent identifier. This is illustrated in FIG. 27, step 1320
[0888] Call the object's IJPersist::Load method with a pointer to the Undo Stream (Stream set to the location recorded in step 4)
[0889] Call BatchInsert(pObjectUnknown)—this writes a log to the Redo Stream.
[0890] “RELATION_MODIFIED”
[0891] Determine if the object was saved via IJUndo of IJPersist. This is illustrated in FIG. 27, step 1330
[0892] if IJUndo—Call BatchModify(Pobjectunknown, iid) then Call IJUndo::Revert(Pstream) (Stream set to the location recorded in step 4)
[0893] else—Call BatchModify(Pobjectunknow, TBDiid) indicating all interfaces will change and then call IJPersist::Load(Pstream). (Stream set to the location recorded in step 4)
[0894] “RELATION_INSERTED”
[0895] Call BatchDelete(Pobjectunknown). This is illustrated in FIG. 27, step 1350.
[0896] In each case, all batch operations and subsequent RelationNotify:: . . . calls are logged to the Undo Stream.
[0897] Call BatchFlush() for each of the checkpoints. This will guarantee a closed set of elements changed for each checkpoint and the ability to use the redo log for undoing an undo operation (redo).
[0898] 5.5. Object Implementation Considerations
[0899] Obects that do not support IJUndo must take special care in the implementation of IJPersist::Load() method in the area of aggregated COM objects. When an aggregator is created, its persist manager simply creates the IUnknown that will serve a the Punkouter to the aggregates. punkouter is responsible for creating its aggregates and this is typically done during the IJPersist::Load() call made by the persist manager. If IJPersist::Load() is called by the Undo Manager, the aggregator (pUnkOuter) will have been newly created in most cases and the aggregate creation is correct behavior. But if the IJPersist::Load() is called because of a modify undo (or the delete operation left a support-only object), reloading of the aggregates will be necessary without creating them. And further more, creating them in the latter case will cause multiple redundant aggregates that will consume memory and be lost in the system.
[0900] An TBA API for informing the Relation Manager that a relation or semantic has been modified is needed to maintain referential integrity and allow the recovery and undo of changes directed at relationships rather than values.
[0901] Persistent objects that do not inform the relation manager that their state will be changed, such as a Rigid Owner's Proxy, should incorporate a special TBD API that would allow for their capture in an undo log without incurring the overhead of relation management. When a proxy's reference count reaches a threshold value, the proxy deletes itself because it knows its outstanding reference count is due to its owner. If that member is requested by a command or other object and the proxy does not exist, the owner will create another proxy. The fact that the Proxy deletes itself without using BatchDelete() or RelationNotify::Delete() means that when an object that was previously dependent upon an interface on the proxy is restored by an Undo-Delete, the relation will not find the dependency and Undo will fail. Restoring of all previous data is essential for a robust undo capability.
CONCLUSION
[0902] In the foregoing specification, the invention has been described with reference to a specific exemplary embodiments thereof. In sum, FIG. 28. llustrates a flow diagram of one embodiment of the application of the Relations subsystem described above. As illustrated in FIG. 28, a first and a second entity are provided in the Relations subsystem, steps 1360 and 1370. Each entity contains a plurality of nodes including value nodes, function nodes, predicate nodes, delete semantic nodes, and copy semantic nodes, as previously described. The granularity of the Relations subsystem is controlled by allowing each entity to have a plurality of nodes and to establish relationships among the plurality of nodes. Step 1380, illustrates that with the plurality of nodes, in an entity, relationships can be established between nodes in the first entity, as previously described. Step 1390, illustrates that relationships can also be established across entity boundaries, and between entities, as previously described. By providing each entity with delete semantic nodes, each entity knows how to modify the relationships of its own nodes upon a delete call, as previously described, step 1400. Further, by providing each entity with copy semantic nodes, each entity knows how to copy the relationships of its own nodes upon a copy call, as previously described, step 1410.
[0903] Many changes, modifications, and additional extensions to COM facilitating the Relations information between objects are readily envisioned and are included within other embodiments of the present invention.
[0904] The specification and drawings are, accordingly, to be regarded in an illustrative rather than in a restrictive sense. It will, however, be evident that various modifications and changes may be made thereunto without departing from the broader spirit and scope of the invention as set forth in the claims.
Claims
What is claimed:
1. A method of managing relationships between objects in a computer system the objects having a first plurality of nodes, the method comprising:
determining a first object which is modified;
determining a first node in the first object which is modified;
determiming a second plurality of nodes related directly or transitively to the first node;
ordering the second plurality of nodes into an order; and
processing the second plurality of nodes in the order.
2. The method of claim 1,
wherein the second plurality of nodes comprises function nodes and value nodes, each function node having an evaluation function; and
wherein processing the second plurality of nodes comprises:
determining value nodes associated with the function nodes; and
evaluating the evaluation functions of the function nodes.
3. The method of claim 2 wherein evaluating the evaluation functions comprises modifying the value nodes associated with the function nodes by applying the evaluation functions.
4. The method of claim 2,
wherein determining the value nodes associated with the functions nodes comprises:
determining input value nodes from the second plurality of nodes, the input value nodes configured to store data which is input to the evaluation functions of the function nodes; and
determining output value nodes from the second plurality of nodes, the output value nodes configured to store data obtained from evaluating the evaluation functions; and
wherein evaluating the evaluation functions comprises modifying the output value nodes by applying the evaluation functions to the data stored in the input value nodes.
5. The method of claim 2,
wherein the function nodes and the value nodes associated with the function nodes are in a parent and child dependency relation; and
wherein determining the value nodes associated with the function nodes comprises identifying value nodes directly or indirectly coupled to the first node as children of dependency relations.
6. The method of claim 5 wherein ordering the second plurality of nodes comprises ordering parent value nodes before children value nodes.
7. The method of claim 1,
wherein the second plurality of nodes comprises predicate nodes and value nodes, each predicate nodes having value nodes associated therewith; and
wherein processing the second plurality of nodes comprises:
identifying value nodes associated with the predicate nodes; and
solving the predicate nodes.
8. The method of claim 7 wherein solving the predicate nodes comprises:
passing the predicate nodes and the associated value nodes to a solver; and
for each predicate node, modifying a value node associated with the predicate node to satisfy a predicate indicated by the predicate node.
9. The method of claim 7,
wherein the predicate nodes and the value nodes associated with the predicate nodes are in an equivalence relation; and
identifying the value nodes associated with the predicate nodes comprises identifying value nodes directly or indirectly coupled to the first node in equivalence relations.
10. The method of claim 1 wherein the first plurality of objects include COM objects.
11. The method of claim 10 wherein the nodes include interfaces provided by the COM objects.
12. A method of copying objects in a computer system, and for copying relationships between the objects, the method comprising:
receiving a first object to be copied;
determining a first plurality of objects related directly or transitively to the first object;
creating a first plurality of duplicate objects based on the first plurality of objects; and
establishing relationships between objects in the first plurality of duplicate objects based on relationships between objects in the first plurality of objects.
13. The method of claim 12,
wherein the objects comprise a plurality of nodes, and a plurality of copy semantic nodes, each of the plurality of nodes having value nodes in a parent and child dependency relation with a copy semantic node; and
wherein determining the first plurality of objects related to the first object comprises:
identifying objects having nodes directly coupled to the first object; and
identifying objects having nodes indirectly coupled to the first object as children of a copy semantic node.
14. The method of claim 12 wherein the objects include COM objects.
15. A method for deleting objects from a first plurality of objects in a computer system, the method comprising:
receiving a first object to be deleted from the first plurality of objects;
determining a second plurality of objects related directly or transitively to the first object;
ordering the objects in the second plurality of objects based on the relationships of the objects; and
disconnecting objects in the second plurality of objects from objects in the first plurality of objects which are not included in the second plurality of objects.
16. The method of claim 15, wherein the objects in the first plurality of objects comprise a plurality of value nodes and delete semantic nodes, each of the plurality of delete semantic nodes having value nodes in a parent and child dependency relation; and
wherein determining the second plurality of objects comprises:
identifying objects from the first plurality of objects having nodes directly coupled the first object; and
identifying objects from the first plurality of objects having nodes indirectly coupled to the first object as children of dependency relations.
17. The method of claim 15 wherein the objects include COM objects.
18. A computer program product for managing relationships between objects in a computer system, the objects having a first plurality of nodes, the computer program product comprising:
code for determining a first object which is modified;
code for determining a first node in the first object which is modified;
code for determining a second plurality of nodes related directly or transitively to the first node;
code for ordering the second plurality of nodes into an order;
code for processing the second plurality of nodes in the order; and
a computer readable medium for storing the codes.
19. The computer program product of claim 18,
wherein the second plurality of nodes comprises function nodes and value nodes, each function node having an evaluation function;
wherein code for determining the second plurality of nodes comprises:
code for determining input value nodes from the value nodes in the second plurality of nodes, the input value nodes configured to store data which is input to the evaluation functions of the function nodes;
code for determining output value nodes from the value nodes in the second plurality of nodes, the output value nodes configured to store data obtained from evaluating the evaluation functions; and
wherein code for processing the second plurality of nodes comprises code for modifying the output value nodes by applying the evaluation functions to the data stored in the input value nodes.
20. The computer program product of claim 18 wherein the objects include COM objects.
21. A computer system for managing relationships between objects in a computer system, the objects having a first plurality of nodes, the system comprising:
a memory coupled to a bus; and
a processor coupled to the memory via the bus, the processor configured to determine a first object which is modified, to determine a first node in the first object which is modified, to determine a second plurality of nodes related directly or transitively to the first node, to order the second plurality of nodes into an order, and to process the second plurality of nodes in the order.
22. The computer system of claim 21,
wherein the second plurality of nodes comprises function nodes and value nodes, each function node having an evaluation function;
wherein the processor is configured to determine the second plurality of nodes by determining input value nodes from the value nodes in the second plurality of nodes, the input value nodes configured to store data which is input to the evaluation functions of the function nodes, and by determining output value nodes from the value nodes in the second plurality of nodes, the output value nodes configured to store data obtained from evaluating the evaluation functions; and
wherein the processor is configured to process the second plurality of nodes by modifying the output value nodes by applying the evaluation functions to the data stored in the input value nodes.
23. The computer system of claim 21 wherein the objects include COM objects.
Patent Citations (25)
| Patent | Date | Inventor | Cited By |
|---|---|---|---|
| US5019961 | 1991-05-01 | Addesso et al. | Applicant |
| US5187660 | 1993-02-01 | Civanlar et al. | Applicant |
| US5278954 | 1994-01-01 | Hohlfeld et al. | Applicant |
| US5299297 | 1994-03-01 | Reynolds et al. | Applicant |
| US5307452 | 1994-04-01 | Hahn et al. | Applicant |
| US5414801 | 1995-05-01 | Smith et al. | Applicant |
| US5414806 | 1995-05-01 | Richards | Applicant |
| US5428744 | 1995-06-01 | Webb et al. | Applicant |
| US5437008 | 1995-07-01 | Gay et al. | Applicant |
| US5455599 | 1995-10-01 | Cabral et al. | Applicant |
| US5471541 | 1995-11-01 | Burtnyk et al. | Applicant |
| US5511157 | 1996-04-01 | Wang | Applicant |
| US5513310 | 1996-04-01 | Megard et al. | Applicant |
| US5530864 | 1996-06-01 | Matheny et al. | Applicant |
| US5542036 | 1996-07-01 | Schroeder et al. | Applicant |
| US5544301 | 1996-08-01 | Orton et al. | Applicant |
| US5564119 | 1996-10-01 | Krawchuk et al. | Examiner |
| US5619692 | 1997-04-01 | Malkemus et al. | Examiner |
| US5682468 | 1997-10-01 | Fortenbery et al. | Examiner |
| US5692184 | 1997-11-01 | Ardoin et al. | Examiner |
| US5742738 | 1998-04-01 | Koza et al. | Examiner |
| US5761511 | 1998-06-01 | Gibbons et al. | Examiner |
| US6052691 | 2000-04-01 | Ardoin et al. | Examiner |
| EP360387A3 | 1990-03-01 | Applicant | |
| EP422877A2 | 1991-04-01 | Applicant |
Non-Patent Literature (38)
- “Geometric Models for CAD/CAM”, Machine Design, pp. 99-105, Jul. 24, 1990.Applicant
- P. Summers et al., “AML: A Programming Language for Automation”, IEEE Computer Society's Fifth International Computer Software and Applications Conference, 1981.Applicant
- Russell H. Taylor et al., “An Integrated Robot System Architecture”, Proceedings of the IEEE, vol. 71, No. 7, pp. 842-856, Jul. 1983.Applicant
- David Grossman, “AML as a Plant Floor Language”, Robotic & Computer-Integrated Manufacturing, vol. 2, No. 3/4, pp. 215-217, 1985.Applicant
- Shinji Kawabe et al., “A Framework for 3D Modeling: Constraint-Based Description and Non-Manifold Geometric Modeling”, A Collection of Contributions based on lectures presented at the 2d Toyota Conference, Japan, Oct. 2-5, 1988, pp. 325-357.Applicant
- Hiromasa Suzuki et al., “Genometric Modeling for Modeling Products”, Proceedings ICEGDG, vol. 2, pp. 237-243, Vienna 1988.Applicant
- J.H. Garret, Jr. et al., “An Object Oriented Environment for Representing Building Design and Construction Data”, Advanced Construction Technology Center, Document No. 89-37-04, pp. 1-34, Jun. 1989.Applicant
- Walter C. Dietrich, Jr., et al., “TGMS: An Object-Oriented System for Programming Geometry”, Software—Practice and Experience, vol. 19(10), pp. 979-1013, Oct. 1989.Applicant
- M. Olumolade, et al., “Object-Oriented Integration and Simulation of Manufacturing”, Proceedings of the 1990 Summer Computer Simulation Conference, pp. 249-252, Jul. 16-18, 1990.Applicant
- Randy H. Katz, “Toward a Unified Framework for Version Modeling in Engineering Databases”, ACM Computing Surveys, vol. 22, No. 4, pp. 375-408, Dec. 1990.Applicant
- Marco Baldassari, et al., “PROTOB: An Object Oriented Methodology for Developing Discrete Event Dynamic Systems”, Computer Languages, vol. 16, No. 1, pp. 39-63, 1991.Applicant
- Andrew R. LeBlanc, “Design Data Sotrage and Extraction Using Objects”, ASME Winter Annual Meeting Dec. 1-6, 1991, Atlanta, Georgia, pp. 1-7.Applicant
- N. Sreenath, “A Hybrid Computation Environment for Multibody Simulation”, Mathematics and Computers in Simulation 34, pp. 121-140, 1992.Applicant
- Kunihiko Kaneko et al., “Design of 3D CG Data Model of “Move” Animation Database System”, Advanced Database Research and Development Series, vol. 3, Proceedings of the Second Far-East Workshop On Future Database Systems, pp. 364-372, 1992.Applicant
- Anoop Singhal et al., “DDB: An Object Oriented Design Data Manager for VLSI CAD”, pp. 467-470, Association for Computing Machinery, 1993.Applicant
- Andrew R. LeBlanc et al., “Design Data Storage and Extraction Using Objects”, Concurrent Engineering: Research and Applications, 1, pp. 31-38, 1993.Applicant
- Kunihiko Kaneko et al., “Towards Dynamics Animation on Object-Oriented Animation Database System “Move””, Advanced Database Research and Development Series, vol. 4, Database Systems for Advanced Applications '93, pp. 3-10, 1993.Applicant
- Jose L.T. Santos et al., “Computer Aided Mechanical Engineering Design Environment for Concurrent Design Process”, Proceedings of the 1993 ITEC Workshop on Concurrent Engineering May 4-6, 1993, Simulation in Concurrent Engineering, pp. 71-83, May 1993.Applicant
- Uday Sreekanth et al., “A Specification Environment for Configuring a Discrete-Part Manufacturing System Simulation Infrastructure”, 1993, International Conference on Systems, Man and Cybernetics, Systems Engineering in the Service of Humans, Conference Proceedings, vol. 1, pp. 349-354, Oct. 17-20, 1993.Applicant
- Shun Watanabe, “Knowledge Integration for Architectural Design”, Knowledge-Based Computer-Aided Architectural Design, pp. 123-146, 1994.Applicant
- Chryssostomos Chryssostomidis, et al., “Geometric Modeling Issues in Computer Aided Design of Marine Structures”, MTS Journal, vol. 22, No. 2, pp. 15-33.Applicant
- Foley et al., “Computer Graphics: Principles and Practices”, Second Edition, Addison-Wesley Publishing Company, 1990, Ch. 5 and 6, pp. 201-283.Applicant
- Prosise, Jeff, “2-D drafting: Why pay more?”, PC Magazine, vol. 12, No. 4, pp. 255-289, Feb. 23, 1993.Applicant
- Gross, Christian, “What about Windows NT?”, Computer-Aided Engineering, vol. 12, No. 4, pp. 58-61, Apr. 1993.Applicant
- Rouse, Nancy E., “CAD pioneers are still trailblazing”, Machine Design, vol. 59, No. 25, pp. 117-122, Oct. 22, 1987.Applicant
- “Getting packaging specifications on track”, Food & Drug Packing, May 1994, p. 28.Applicant
- “Mechanical design software” (Buyers Guide), Computer-Aided Engineering, vol. 12, No. 12, pp. 32-36, Dec. 1993.Applicant
- “Turning Pro. (Visual Information Development Inc.'s Presenter Professional 1.5 promotes the Macintosh as a serious CAD tool)”, Cadcam, vol. 12, No. 6, pp. 33-36, Jun. 1993.Applicant
- Nadile, Lisa, “Computervision debuts development platform”, PC Week, vol. 12, No. 2, p. 10, Jan. 16, 1993.Applicant
- “Computervision Launches Design Automation Development Platform for Windows”, PR Newswire, Financial News, Jan. 10, 1995.Applicant
- “Next stop Jupiter for CADCAM users”, Engineer, No. 7252, p. 24, May 25, 1995.Applicant
- “Value in 3-D”, Industry Week, p. 45, Jan. 8, 1995.Applicant
- Kempfer, Lisa, “CAD jumps on Windows 3.1 bandwagon”, Computer-Aided Engineering, vol. 12, No. 11, pp. 24-25, Nov. 1993.Applicant
- “Link 3D, 2D to 3D conversion announced”, Businesswire, Mar. 23, 1993.Applicant
- Dorsch, Jeff, “EDA Licensing”, Electronic New, p. 44, Jan. 16, 1995.Applicant
- “CAD/CAM/CAE consortium agrees on OLE specifications”, Silicon Graphics World, vol. 5, No. 5, p. 5, May 1995.Applicant
- Brockschmidt, Inside OLE 2, 1994, Microsoft Press.Applicant
- Strickland, T.M., “Intersection of Relational and Object”, Proceedings AM/FM International Annual Conference XVII, Proceedings of AM/FM International Conference, Denver, Colorado, USA, Mar. 14-17, 1994, 1994, Aurora, Colorado, USA, AM/FM/INT, USA, pp. 69-75.Applicant