The general approach is to make a snapshot object that represents the state of the application. After the user performs some action that you want to be undoable, the application creates a snapshot representing its new state, and saves the snapshot in a collection.
To undo an action, the application pulls the previous snapshot out of the collection and applies it, resetting the application's state to whatever it was when the snapshot was taken. To undo the next action, the application applies the snapshot before that one, and so on until it has moved back to the beginning of the snapshot collection.
To redo an action, the program moves forward to the next snapshot (the one most recently undone) and applies it. It can continue moving forward through the collection until it reaches the most recent snapshot.
If the user undoes some actions and then performs a new action, the program removes all of the snapshots after the one that was most recently applied. It then adds a snapshot representing the new action to the collection.
The program adds and removes snapshots from the same end of the collection, so the items added first are the last ones you can remove. This kind of data structure is called a first-in-last-out (FILO) list, and is analogous to a stack of plates on a table: You can add a plate to the top and remove a plate from the top, but you cannot add or remove a plate from the middle or bottom. Because of this similarity, a FILO list is often called a stack.
Sometimes, you may want to bend the definition of the stack slightly and allow the program to remove items from the bottom of the stack. For a complex application, a snapshot may take up a lot of memory. In that case, you might not want to keep every snapshot in the undo stack forever. For example, you might decide to keep only the 10 most recent snapshots. Whenever the program added an 11th snapshot, it would delete the oldest one from the collection. A more flexible approach might add up the sizes of the snapshots and only start deleting the oldest ones when the total memory used was more than a few megabytes.
A Kodak Moment
The general approach of building an undo stack is relatively simple. You make a collection, add snapshots to it, and move back and forth through the collection to undo and redo actions.
The tricky part is taking the snapshot. The snapshot must contain all the information you need to restore the application to a particular state. Depending on the application, a snapshot might need to include a huge variety of information. For example:
In a simple text file editor, a snapshot can simply be a string containing the current text.
In a word processor, a snapshot can include a whole swarm of objects representing paragraphs, words, margins, figures, text styles, footnotes, and other pieces of the document.
In a drawing program, a snapshot includes information about the size, position, orientation, color, line style, fill style, and other drawing attributes of all of the currently displayed graphical objects.
Given the diversity of information you might need to store for a particular application, you might think this article can give you little help in building snapshots. Actually the XML serialization tools provided by VB.NET make building snapshots easier than you might think.
A serialization is a streamed, often textual representation of an object. When you serialize an object, you convert it into a serialization. When you deserialize a serialization, you convert it back into an object. Some of the more whimsically minded programmers call this dehydration and rehydration.
For example, suppose you have a Person object that has variables FirstName, LastName, and Age. If you have a particular object with FirstName = Andrew, LastName = Anderson, and Age = 25, then one possible serialization is the following:
If you know that the fields in this serialization are LastName, FirstName, and Age, you can pull this string apart to build a new Person object containing the original values.
An XML-based serialization uses a sequence of XML (extensible markup language) tags to represent an object. In this example, the serialization might be the following:
<Person> <LastName>Andrew</LastName> <LastName>Anderson</LastName> <Age>25</Age> </Person>
XML is a standard data representation language, so there is some reason to believe this serialization is more readable than the previous one. The real clincher for using this version, however, is that VB.NET's XmlSerializer object can automatically serialize and deserialize objects for you.
The XmlSerializer class automatically serializes and deserializes objects without you needing to tell it all about the objects. The serializer uses VB.NET's reflection tools to learn about the object all by itself. That means you can use an XmlSerializer to create a textual representation of any kind of object quickly and easily.
The following code serializes a Person object named customer and displays the result in a message box.
Dim customer As Person Dim xml_serializer As XmlSerializer Dim string_writer As New StringWriter() ' Initialize the customer object somehow. ... ' Create the XmlSerializer. xml_serializer As New XmlSerializer(GetType(Person)) ' Make the serialization, writing it into the StringWriter. xml_serializer.Serialize(string_writer, customer) ' Display the results. MsgBox(string_writer.ToString)
The following code deserializes a serialization string to re-create a Person object.
Dim customer As Person Dim serialization As String Dim string_reader As StringReader Dim xml_serializer As New XmlSerializer(GetType(Bouncer)) ' Load the serialization string somehow. ... ' Make a StringReader holding the serialization. string_reader = New StringReader(serialization) ' Create the new Person object from the serialization. customer = xml_serializer.Deserialize(string_reader)
Notice that this code contains no information about the structure of the Person class. The XmlSerializer figures out what information it needs to save in the serialization and how to recover that information without any help from you.
In fact, the XmlSerializer will even climb down into any sub-objects the main object might contain. For instance, suppose the Person object contains an array of Assignment objects representing projects to which the person is assigned. The XmlSerializer will dig through the array and pull out the information it needs to serialize all of the Assignment objects that are present. XmlSerializer can climb as deeply into an object hierarchy as it needs to serialize all of the objects attached to the Person object, with a few constraints.
First, XmlSerializer cannot serialize disconnected objects. If you have two Person objects that are not connected in any way, XmlSerializer does not include both objects in the serialization of either one. For an example that's more relevant to an undo/redo discussion, if your drawing application uses a bunch of drawing objects (line, circle, polygon), the XmlSerializer cannot serialize them all if they are not connected.
An easy way to solve this problem is to create a new object that simply contains references to all the objects you want to serialize. This object can be from a new class you build, or it can be a simple array.
The second restriction on XmlSerializer is that it cannot serialize an object graph that contains a loop. If a Person object contains a reference to another object, and that object contains a reference to the original Person object, the XmlSerializer gets confused.
VB.NET provides other serialization classes that can solve this problem. The SoapFormatter and BinaryFormatter classes can both serialize and deserialize complex networks that include loops. These classes also sometimes provide better error messages than XmlSerializer, and they are a bit more robust in other ways, so the example program described shortly uses the SoapFormatter. For more information on using all of these classes, see my book Visual Basic.NET and XML (Wiley, 2002, http://www.vb-helper.com/xml.htm.
Now that you know what XML serialization can do, you've probably figured out the rest of the undo/redo puzzle. If you store all of the application's information in a single object, possibly referring to other objects, you can serialize that object fairly easily. You can then use the serialization for your snapshot.