Scaling Up With the Java Foundation Classes JFCSwing API Large Data Sets and Complex Beans

1
Scaling Up With the Java Foundation Classes
(JFC/Swing) APILarge Data Sets and Complex Beans

• David Zeleznik
• Manager Gantt RD
• ILOG, Inc.

Yunpeng Zhao, PhD Software Engineer ILOG, Inc.
2
Origins of This Talk

• Project to create a complex Gantt chart bean
• Large hierarchical data sets
• Time constraints forced to use JTable and JTree
  for the JFC/Swing 1.03 release
• Plan to migrate to later releases of JFC/Swing
  and/or custom table and tree components

3
Roadmap

• JFC/Swing Architecture David
• Data Caching Guidelines David
• JTable David
• JTree David
• TreeTable David
• Complex Beans Yunpeng

4
Roadmap

• JFC/Swing Architecture
• MVC Overview
• Scrolling With JFC/Swing
• Data Caching Guidelines
• JTable
• JTree
• TreeTable
• Complex Beans

5
MVC Overview(the 60 second recap)

• Model-View-Controller architecture decouples 3
  aspects of a user-interface
• Model the data being displayed
• View how the data is displayed
• Controller how the user interacts with the
  displayed data

• JFC/Swing variant of MVC separable model
  architecture.
• Provides benefits of complete MV separation.
• Easier to use because it bundles view and
  controller together.

6
Scrolling With JFC/Swing

• Why?
• Standard means to display subsets of large
  dataset
• Intuitive navigation through dataset
• Scrolling architecture impacts scalability

• Attributes
• General purpose
• No dependency on underlying data
• Scalable
• Architecture
• Child component to be scrolled
• JViewport
• JScrollPane

7
JViewport

• Provides a clipped and translated view of another
  component
• Can be used independently of a JScrollPane
• Scrolling modes in order of increasing
  performance
• SIMPLE_SCROLL_MODE
• BACKINGSTORE_SCROLL_MODE
• This was default for JTable in earlier JDK's
• BLIT_SCROLL_MODE
• Very buggy version was available in JFC/Swing
  1.1.1
• There are still some bugs with this mode,
  especially using javaw inside IDEs

8
JScrollPane

• Responsible for up to 9 children
• 3 JViewports
• main viewport
• row header
• column header
• 2 scrollbars
• 4 corner components
• ScrollPaneLayout controls the arrangement of the
  9 components

9
Scrolling a JTable
10
Roadmap

• JFC/Swing Architecture
• Data Caching
• JTable
• JTree
• TreeTable
• Complex Beans

11
Data Caching

• Large Data Sets
• GIS geographical data
• Project or factory scheduling and planning
• High Retrieval Latencies
• Slow connections to back-end datastore
• Database queries
• Optimizes speed in exchange for memory usage
• Can combine with eager or lazy background
  evaluation

12
UI Data Caching Guidelines

• User interface should not cache model data
• duplicate views duplicate caching
• difficulty replacing caching strategy to suit
  different deployment needs
• caching may already be implemented in ORB,
  database, server, etc.
• User interface should only access data model for
  items being painted
• UI should pull data from model as needed during
  redraw
• Redrawing a data item should not require access
  to data not being redrawn
• Good example JTable paints a cell using only the
  data for that cell
• Poor example JTree calculates horizontal indent
  position to paint a node by querying for nodes
  ancestors

13
Where to Cache

• Strategy pattern a single class to encapsulate
  policy
• ability to create deterministic policy
• easier tuning and analysis
• single interface pluggability
• Model interface provides existing interface
• The Proxy or Decorator pattern
• The Cache Management Pattern
• Patterns in JavaTM by Mark Grand, John Wiley
  and Sons

14
JTable Caching Example
JTable
TableModel Interface
15
DataModel Caching - Static
16
DataModel Caching - Collaboration
17
Roadmap

• JFC/Swing Architecture
• Data Caching
• JTable
• Scalability
• Demos
• JTree
• TreeTable
• Complex Beans

18
JTable Is Very Scalable

• Does not cache any data objects from TableModel
• Painting a cell uses only data object for that
  cell
• Fixed row height in JDKTM 1.2 release, variable
  row heights in JDKTM 1.3 release
• SelectionModel references cell addresses
• TableModel events reference cell addresses

19
Verifying That JTable Does Not Cache
20
Demos

• Code available on ILOG website
• www.ilog.com/products/jviews/javaone
• Static data models
• No insert/delete events

21
Table Demo 1

• 100 x 100
• No scrolling
• table is fixed size
• null layout
• Data objects from TableModel are not cached
• A new data object instance is created and
  returned for every call to TableModel.getValueAt()
  
• Data objects do not have to implement logical
  equality because JTable does no caching or
  comparison testing

22
Table Demo 1 Notes

• JFC/Swing double-buffering is enabled
• All cells are repainted and data model is
  accessed during frame resize
• Double-buffering eliminates flashing
• Does not deal with backing store and
  intelligent management of invalid regions like
  JScrollPane
• Data model access during mouse movement is due to
  tooltip mechanism
• In JTable and JTree, tooltip text is obtained
  from the cell renderer
• Every mouse move JTable.prepareRenderer()
• Override JTable.getToolTipText(MouseEvent) or
  unregister table from ToolTipManager

23
Table Demo 2

• JTable is in a JScrollPane
• TableModel is same as demo 1
• New JViewport behavior in JDK 1.3 release
• BitBlt mode is default in JDK 1.3 release
• Backing store mode is default in JDK 1.2 release

24
Table Demo 3

• Cached data
• TableModel.getValueAt() is only invoked once per
  table cell for lifetime of data model
• Optimized for scrolling over previously viewed
  areas
• at expense of memory usage
• Does not help when scrolling over new areas of
  table when time delay obtaining data is large

25
Table Demo 3 Notes

• Each cell is cached individually
• keyed by row,col cell address
• Alternative cache by rows
• Each column is a database field
• RecordSets
• java.util.Hashtable is inefficient when using
  primitive ints
• Keys must be Objects
• Expensive to create Integer or other object keys
  from int rows and columns
• Custom Hashtable that uses primitive ints or
  longs as lookup keys

26
Table Demo 3L

• 10,000 x 10,000 table
• JFC/Swing 1.1.1
• large memory usage and time spent initializing
  TableColumn objects
• Individual headerRenderers, Borders, etc.
• DefaultTableColumnModel continually recalcs total
  column width
• JDK 1.3 release
• Default TableColumns share a common
  headerRenderer in the JTableHeader component
• DefaultTableColumnModel lazy calculates total
  column width

27
JTable Performance Issues

• Startup
• AbstractTableModel.getColumnName()uses String
  operator
• Instantiation of TableColumns
• Scrolling
• JTable.getCellRect iterates through all columns
  to accumulate column x position
• Moving first column is faster than moving last
  column
• New SizeSequence class is used to store variable
  row heights in an efficient manner. Use same
  approach to store column positions
• DefaultTableColumnModel uses a synchronized
  Vector to store the TableColumns.

28
Table Demo 4

• Cached data model which defers fetching data in a
  separate thread
• Future Pattern supply a surrogate value
  immediately. Defer actual data asynchronously
• JTable must be notified on the JFC/Swing event
  thread
• SwingWorker

29
Roadmap

• JFC/Swing Architecture
• Data Caching
• JTable
• JTree
• TreeTable
• Complex Beans

30
JTree

• Is JTree as scalable as JTable?

• Architecture
• TreePaths
• Internal Caches

31
JTree Architecture
32
TreePaths (1)
lastPathComponent
parentPath
33
TreePaths (2)

• A child TreePath prevents all of its ancestors
  from being gcd
• Used throughout JTree as a unique address for a
  tree node
• Allows equal() node objects to be used in the
  same tree
• Some duplicate functionality with the data model
• trace node parentage
• indent level
• higher speed, more memory

34
JTree expandedState Cache

• Treepath used as key, value is Boolean
• Cache entry is not removed when node is collapsed
• Child entry is not removed if parent is collapsed
• JTree "remembers" expansion state of child nodes

Boolean.TRUE
Boolean.TRUE
Boolean.TRUE
Boolean.TRUE
35
JTree Layout Cache

• BasicTreeUI treePathMapping table
• Position and bounding box info for visible nodes
• VariableHeightLayoutCache
• Is JTree default
• Caches all visible nodes
• FixedHeightLayoutCache
• Caches only expanded visible nodes
• Only enabled if JTree largeModel true and rows
  are explicitly set to a fixed height

36
JTree Caching Summary
expanded state
variable height
fixed height
previously expanded
37
JTable and JTree Summary

• JDK 1.3 release
• JViewport BLIT_SCROLL_MODE
• Fixed row heights
• JTable is inherently scalable
• JTree can be made scalable
• Keep node objects small (db keys)
• Set largeModel true
• Consider
• SoftReferences
• Deleting invisible nodes

38
Roadmap

• JFC/Swing Architecture
• Data Caching
• JTable
• JTree
• TreeTable
• Complex Beans

39
JavaSoft TreeTable Demo
40
TreeTable Design

• Pros
• Subclass of JTable
• JTree is used as a TableCellRenderer
• Can place tree in any column
• Limitations
• Only 1 tree column per table
• Input event handling is mixture of JTree and
  JTable
• TreeTableModel is complex mixture of TreeModel
  and TableModel
• TreeTableModel has no row, column events
• Unneeded complexity of synchronizing selections
  in JTree and JTable
• Tree column is not editable

41
TreeTableModel

• TreeModel
• Object getRoot()
• Object getChild(Object parent, int index)
• int getChildCount(Object parent)
• boolean isLeaf(Object node)
• void valueForPathChanged(TreePath path, Object
  newValue)
• int getIndexOfChild(Object parent, Object child)
• void addTreeModelListener(TreeModelListener l)
• void removeTreeModelListener(TreeModelListener l)
• Table methods
• int getColumnCount()
• String getColumnName(int column)
• Class getColumnClass(int column)
• Object getValueAt(Object node, int column)
• boolean isCellEditable(Object node, int column)
• void setValueAt(Object aValue, Object node, int
  column)

42
TreeTable Solutions

• Allow users to install multiple JTree
  cellRenderers
• Keep JTree passive, allow JTable to handle all
  input events
• Navigation is consistent with remainder of table
• Editing is handled by table
• JTree does not need to track tables selection
• JTable simulates mouse handling of
  expand/collapse
• Simplify the TreeTableModel

43
Roadmap

• JFC/Swing Architecture
• Data Caching
• JTable
• JTree
• TreeTable
• Complex Beans

44
Complex Beans Overview

• What are complex beans?
• Internal bean behaviors
• Behavior synchronization
• Demonstration
• Conclusions

45
Are JTree and JTable complex beans?

• JTree and JTable are complex
• Are JTree and JTable complex beans?

46
JTree and JTable Model/View Architecture (1)

• JTree
• TreeModel representing hierarchical data
• A tree view on the data
• JTable
• TableModel Data can be arranged by rows and by
  columns
• A table view on the data

47
JTree and JTable Model/View Architecture (2)

• JTree and JTable both have
• One data model having a specific structure
• One data view designed to show the data model

Dedicated Data View
Specific Data Model

• JTree and JTable are simple beans, they are not
  complex beans

48
A Scheduling Bean (1)User Interface
2000 01 02 03 04 05 06 07 08 09 10 11 12
Start End Jan 2000 Dec 2000 Jan 2000 Apr 2000 Mar
2000 Jun 2000 May 2000 Aug 2000 Jul 2000 Sep
2000 Apr 2000 Jul 2000 Aug 2000 Dec 2000 Aug
2000 Nov 2000
Name Project Summary Gather Requirements Talk
to customers Compile customer list Contact
customers Write up requirements Marketing
Specifications First specification
49
A Scheduling Bean (2)Complex Data Model

• The scheduling data is complex
• It is arranged by rows and columns
• It has a hierarchy
• The scheduling data can not be completely
  presented by a simple bean such as a JTree or a
  JTable
• The solution is to combine a JTree and a JTable
  into a complex bean

50
A Complex Bean to Show Scheduling Data
Complex Bean
Scheduling Data Model
51
Complex Beans vs. Multiple-Document Interface
(MDI)

• Internal beans are managed by a layout
• Internal beans usually have different
  architectures
• Internal beans must work harmoniously
• Complex beans expose common behaviors of the
  internal beans (Façade pattern)

52
A Complex Bean Is

• Complex beans are made by combining simple
  internal beans
• Internal beans show different views of a complex
  data model
• Internal beans must have synchronized behaviors
• Internal beans should be easily added or replaced
  to satisfy new UI requirements
• Complex beans must have a high level API

53
Internal Bean Behaviors

• What are behaviors? Ex
• Selecting
• Scrolling (HV)
• Zooming
• Visibility
• Behaviors are more general than properties
• Behaviors consist of
• APIs to query the state
• APIs to change the state
• APIs to add state listeners
• The state can be listened to
• Fire event when the state has changed

54
JTree Selection Behavior

• Get if a node is selected
• isPathSelected(TreePath path)
• Select a node
• addSelectionPath(TreePath path)
• Add selection listener
• addTreeSelectionListener(
• TreeSelectionListener tsl)

55
JTable Selection Behavior

• Get if a row is selected
• isRowSelected(int row)
• Select a row or rows
• addRowSelectionInterval(int from,
• int to)
• Add selection listener
• getSelectionModel().
• addListSelectionListener(
• ListSelectionListener lsl)

56
Selection Behaviors of JTree and JTable

• JTree and JTable selections work separately
• How to synchronize the selection behaviors of
  JTree and JTable?

57
Selection SynchronizationMediator Pattern
JTable
Mediator
JTree
List Selection Model
ListSelectionListener
TreeSelectionListener
58
Selection SynchronizationInfinite Loop(1)
JTable
Mediator
JTree
List Selection Model
ListSelectionListener
TreeSelectionListener
59
Selection SynchronizationMediator Pros and Cons

• Easy to implement
• Difficult to expose the selection behavior
  through the façade component
• Suitable for two components
• Not scalable

60
Selection SynchronizationScaling Up
JTable
Mediator
JTree
List Selection Model
61
Selection SynchronizationObserver/Adapter
Patterns
JTable
JTree
RowSelectable Adapter
RowSelectable Adapter
RowSelectionController
62
Selection SynchronizationRow-Selectable
Interface
public interface RowSelectable public void
isRowSelected(int row) public void
selectRow(int row, boolean select) public void
addRowSelectionListener (RowSelectionListener
rsl)
63
Selection SynchronizationRow Selection Listener
public interface RowSelectionListener extends
java.util.EventListener public void
rowSelected(RowSelectionEvent event)
64
Selection SynchronizationRow Selection Event
public RowSelectionEvent extends
java.util.EventObject private int
_row private boolean _selected public
RowSelectionEvent(Object source, int row,
boolean selected) super(source) _row
row _selected selected public int
getSelectedRow() return _row) public
boolean isSelected() return _selected
65
Selection SynchronizationInfinite Loops
JTable
JTree
RowSelectable Adapter
RowSelectable Adapter
2
3
5
1
4
RowSelectionController
66
Exposing Row Selection Behavior (Facade Pattern)
JTable
JTree
Other Internal Components ...
RowSelectable Adapter
RowSelectable Adapter
RowSelectable Adapter
RowSelectable Adapter
RowSelectionController
RowSelectable Interface
67
Behavior Synchronization Demonstrations

• Selecting
• Scrolling
• Zooming
• Visibility

68
Conclusions

• Concentrate your efforts on modeling your data
• Combine appropriate JFC API-based beans to
  present your data model
• Use Observer/Adapter patterns to make internal
  beans work harmoniously
• JFC technology-based beans can be scaled up!

69

• David Zeleznik
• zeleznik_at_ilog.com

Yunpeng Zhao zhao_at_ilog.fr