Virtual Data Concept

The Virtual Data Pattern

TeeGrid’s TVirtualData class implements a virtual data pattern that decouples the grid’s rendering engine from the underlying data source. This design provides several key benefits:

Data Source Flexibility

Support any data source by implementing a single interface

Lazy Loading

Retrieve only visible cells on-demand for optimal performance

Large Datasets

Handle millions of rows without loading all data into memory

Framework Agnostic

Same data adapter works across VCL, FMX, and Lazarus

How Virtual Data Works

The virtual data pattern operates on a simple principle: the grid never holds the actual data.

Key Characteristics

On-Demand Retrieval

Values are fetched only when needed for display, not preloaded into memory.

No Data Duplication

The grid holds no copy of your data - it queries the virtual data adapter each time.

Efficient Updates

When data changes, notify the grid via events rather than transferring data.

Streaming Support

Data sources without a known row count work perfectly (return -1 from Count()).

TVirtualData Abstract Class

Defined in Tee.GridData.pas, this class establishes the contract for all data adapters:

Required Methods

Every TVirtualData implementation must provide:

Count

Integer

Returns the total number of rows. Return -1 if unknown (streaming data).

function Count: Integer; virtual; abstract;

AsString

String

Retrieves the display value for a specific cell. This is the most performance-critical method.

function AsString(const AColumn: TColumn; const ARow: Integer): String; virtual; abstract;

AddColumns

procedure

Populates the columns collection based on the data structure.

procedure AddColumns(const AColumns: TColumns); virtual; abstract;

Load

procedure

Associates manually-defined columns with data source fields.

procedure Load(const AColumns: TColumns); virtual; abstract;

SetValue

procedure

Updates a cell value when the user edits it.

procedure SetValue(const AColumn: TColumn; const ARow: Integer; const AText: String); virtual; abstract;

AutoWidth

Single

Calculates the optimal width for a column based on its content.

function AutoWidth(const APainter: TPainter; const AColumn: TColumn): Single; virtual; abstract;

Optional Override Methods

AsFloat

function AsFloat(const AColumn: TColumn; const ARow: Integer): TFloat; virtual;

Returns numeric values for calculations and sorting. Default implementation converts AsString result.

AutoHeight

function AutoHeight(const APainter: TPainter; const AColumn: TColumn; 
  const ARow: Integer; out AHeight: Single): Boolean; virtual;

Calculates custom row height. Return True if implemented, False to use default height.

CanExpand

function CanExpand(const Sender: TRender; const ARow: Integer): Boolean; virtual;

Indicates whether a row can be expanded to show detail data.

GetDetail

function GetDetail(const ARow: Integer; const AColumns: TColumns; 
  out AParent: TColumn): TVirtualData; virtual;

Returns a TVirtualData instance for detail rows in master-detail scenarios.

CanSortBy

function CanSortBy(const AColumn: TColumn): Boolean; virtual;

Returns True if the column supports sorting.

SortBy

procedure SortBy(const AColumn: TColumn); virtual;

Implements sorting logic for the specified column.

IsSorted

function IsSorted(const AColumn: TColumn; out Ascending: Boolean): Boolean; virtual;

Returns current sort state for a column.

ReadOnly

function ReadOnly(const AColumn: TColumn): Boolean; virtual;

Returns True if the column is read-only. Default returns False.

DataType

function DataType(const AColumn: TColumn): PTypeInfo; virtual;

Returns type information for better formatting and editing.

Built-in Implementations

TeeGrid provides three production-ready TVirtualData implementations:

TVirtualDBData (Tee.GridData.DB)

Connects to TDataSet and TDataSource components. Key Features:

Automatic field-to-column mapping
Full edit support with Post/Cancel
Native data type handling
Bookmark-based navigation
Master-detail relationships
Fetch mode configuration (All, Partial, Automatic)

Internal Architecture:

TVirtualDBData
  ├─ IDataSet: TDataSet          // Target dataset
  ├─ IDataSource: TDataSource    // Optional datasource
  └─ ILink: TVirtualDataLink     // TDataLink descendant
       ├─ Monitors dataset state
       ├─ Responds to data changes
       └─ Triggers grid refresh

Data Retrieval:

function TVirtualDBData.AsString(const AColumn: TColumn; const ARow: Integer): String;
var
  RecNo: Integer;
  Field: TField;
begin
  RecNo := BeginRow(ARow);  // Navigate to row
  try
    Field := FieldOf(AColumn);
    Result := Field.DisplayText;
  finally
    EndRow(RecNo);  // Restore position
  end;
end;

TVirtualData<T> (Tee.GridData.Rtti)

Generic adapter using RTTI for arrays, lists, and objects. Supported Types:

TArray<T> - Static arrays
TList<T> - Generic lists
TObjectList<T> - Object lists
Single records/objects
2D arrays (TArray<TArray<T>>)

RTTI-Based Column Creation:

procedure TVirtualDataRtti.AddColumns(const AColumns: TColumns);
var
  RttiType: TRttiType;
  Member: TRttiMember;
  Column: TColumn;
begin
  RttiType := Context.GetType(FTypeInfo);
  
  // Iterate fields and properties
  for Member in RttiType.GetFields + RttiType.GetProperties do
    if IsVisible(Member) then  // Check visibility level
    begin
      Column := AColumns.Add;
      Column.Header.Text := Member.Name;
      LinkTo(Column, Member, TypeOf(Member));
    end;
end;

Value Access:

function TVirtualData<T>.AsString(const AColumn: TColumn; const ARow: Integer): String;
var
  Value: TValue;
  Member: TRttiMember;
  P: Pointer;
begin
  P := GetPointerOf(AColumn, ARow);  // Get data pointer
  Member := MemberOf(AColumn);        // Get RTTI member
  Value := ValueOf(Member, P);        // Read value via RTTI
  Result := Value.ToString;           // Convert to string
end;

Helper Classes:

// Convenient aliases
TVirtualArrayData<T> = class(TVirtualData<TArray<T>>);
TVirtualListData<T> = class(TVirtualData<TList<T>>);
TVirtualObjectListData<T:class> = class(TVirtualData<TObjectList<T>>);

TVirtualStringData (Tee.GridData.Strings)

String grid emulation with Cells[Col, Row] indexing. Usage:

var
  Data: TVirtualStringData;
begin
  Data := TVirtualStringData.Create;
  Data.RowCount := 100;
  Data.Columns.Add('Name');
  Data.Columns.Add('Value');
  
  Data.Cells[0, 0] := 'First Cell';
  Data.Cells[1, 0] := 'Second Cell';
  
  TeeGrid1.Data := Data;
end;

Creating Custom Virtual Data

Implementing a custom adapter requires careful attention to performance and correctness.

Basic Implementation Template

unit MyVirtualData;

interface

uses
  Tee.GridData, Tee.Grid.Columns, Tee.Painter;

type
  TMyVirtualData = class(TVirtualData)
  private
    FDataSource: TMyDataSource;
  protected
    function Empty: Boolean; override;
    function KnownCount: Boolean; override;
  public
    constructor Create(ADataSource: TMyDataSource);
    
    procedure AddColumns(const AColumns: TColumns); override;
    function AsString(const AColumn: TColumn; const ARow: Integer): String; override;
    function AutoWidth(const APainter: TPainter; const AColumn: TColumn): Single; override;
    function Count: Integer; override;
    procedure Load(const AColumns: TColumns); override;
    procedure SetValue(const AColumn: TColumn; const ARow: Integer; const AText: String); override;
  end;

implementation

constructor TMyVirtualData.Create(ADataSource: TMyDataSource);
begin
  inherited Create;
  FDataSource := ADataSource;
end;

function TMyVirtualData.Empty: Boolean;
begin
  Result := (FDataSource = nil) or (FDataSource.RecordCount = 0);
end;

function TMyVirtualData.KnownCount: Boolean;
begin
  // Return True if Count is known, False for streaming data
  Result := True;
end;

function TMyVirtualData.Count: Integer;
begin
  if FDataSource = nil then
    Result := 0
  else
    Result := FDataSource.RecordCount;
end;

function TMyVirtualData.AsString(const AColumn: TColumn; const ARow: Integer): String;
var
  FieldName: string;
begin
  // Extract field identifier from TagObject
  FieldName := TMyField(AColumn.TagObject).Name;
  
  // Retrieve value from data source
  Result := FDataSource.GetValue(FieldName, ARow);
end;

procedure TMyVirtualData.AddColumns(const AColumns: TColumns);
var
  I: Integer;
  Col: TColumn;
  Field: TMyField;
begin
  AColumns.Clear;
  
  for I := 0 to FDataSource.FieldCount - 1 do
  begin
    Field := FDataSource.Fields[I];
    Col := AColumns.Add;
    Col.Header.Text := Field.DisplayName;
    Col.TagObject := Field;  // Store field reference
    
    // Set alignment based on field type
    if Field.IsNumeric then
    begin
      Col.TextAlignment := TColumnTextAlign.Custom;
      Col.TextAlign.Horizontal := THorizontalAlign.Right;
    end;
  end;
end;

procedure TMyVirtualData.Load(const AColumns: TColumns);
var
  I: Integer;
  Field: TMyField;
begin
  // Associate existing columns with fields
  for I := 0 to AColumns.Count - 1 do
  begin
    Field := FDataSource.FindField(AColumns[I].Header.Text);
    if Field <> nil then
      AColumns[I].TagObject := Field;
  end;
end;

function TMyVirtualData.AutoWidth(const APainter: TPainter; const AColumn: TColumn): Single;
var
  I: Integer;
  S: String;
  W: Single;
begin
  Result := 0;
  
  // Sample first 100 rows for width calculation
  for I := 0 to Min(99, Count - 1) do
  begin
    S := AsString(AColumn, I);
    W := APainter.TextWidth(S);
    if W > Result then
      Result := W;
  end;
  
  // Add padding
  Result := Result + 20;
end;

procedure TMyVirtualData.SetValue(const AColumn: TColumn; const ARow: Integer; const AText: String);
var
  Field: TMyField;
begin
  Field := TMyField(AColumn.TagObject);
  FDataSource.SetValue(Field.Name, ARow, AText);
  
  // Notify grid of change
  Repaint;
end;

end.

Performance Optimization

AsString Optimization
Count Caching
AutoWidth Sampling

AsString is called for every visible cell on every paint. Optimize it ruthlessly:

// BAD - slow
function AsString(const AColumn: TColumn; const ARow: Integer): String;
begin
  Result := Format('%.2f', [CalculateComplexValue(ARow)]);
end;

// GOOD - cache results
function AsString(const AColumn: TColumn; const ARow: Integer): String;
begin
  if not FCache.TryGetValue(ARow, Result) then
  begin
    Result := Format('%.2f', [CalculateComplexValue(ARow)]);
    FCache.Add(ARow, Result);
  end;
end;

Cache the count if it’s expensive to calculate:

private
  FCount: Integer;
  FCounted: Boolean;

function Count: Integer;
begin
  if not FCounted then
  begin
    FCount := ExpensiveCountCalculation;
    FCounted := True;
  end;
  Result := FCount;
end;

procedure Refresh;
begin
  FCounted := False;  // Invalidate cache
  inherited;
end;

Don’t scan all rows for width calculation:

function AutoWidth(const APainter: TPainter; const AColumn: TColumn): Single;
const
  SampleSize = 100;
var
  I, Step: Integer;
  S: String;
  W: Single;
begin
  Result := 0;
  
  if Count <= SampleSize then
    Step := 1
  else
    Step := Count div SampleSize;
  
  I := 0;
  while I < Count do
  begin
    S := AsString(AColumn, I);
    W := APainter.TextWidth(S);
    if W > Result then
      Result := W;
    Inc(I, Step);
  end;
end;

Data Change Notifications

Use events to notify the grid of changes:

procedure TMyVirtualData.DataSourceChanged;
begin
  // Structure changed (columns added/removed)
  Refresh;  // Calls FOnRefresh
end;

procedure TMyVirtualData.DataSourceModified;
begin
  // Values changed but structure same
  Repaint;  // Calls FOnRepaint
end;

procedure TMyVirtualData.CurrentRowChanged(ARow: Integer);
begin
  // Current row changed
  ChangeSelectedRow(ARow);  // Calls FOnChangeRow
end;

Advanced Features

Master-Detail Support

function TMyVirtualData.HasDetail(const ARow: Integer): Boolean;
begin
  Result := FDataSource.HasDetailRecords(ARow);
end;

function TMyVirtualData.GetDetail(const ARow: Integer; const AColumns: TColumns; 
  out AParent: TColumn): TVirtualData;
var
  DetailData: TMyDataSource;
begin
  DetailData := FDataSource.GetDetailDataSource(ARow);
  Result := TMyVirtualData.Create(DetailData);
  AParent := nil;  // Or specify which column triggers expansion
end;

Custom Sorting

function TMyVirtualData.CanSortBy(const AColumn: TColumn): Boolean;
begin
  Result := True;  // All columns sortable
end;

procedure TMyVirtualData.SortBy(const AColumn: TColumn);
var
  Field: TMyField;
  Ascending: Boolean;
begin
  Field := TMyField(AColumn.TagObject);
  
  // Toggle sort direction
  if IsSorted(AColumn, Ascending) then
    Ascending := not Ascending
  else
    Ascending := True;
  
  FDataSource.Sort(Field.Name, Ascending);
  Refresh;
end;

function TMyVirtualData.IsSorted(const AColumn: TColumn; out Ascending: Boolean): Boolean;
var
  Field: TMyField;
begin
  Field := TMyField(AColumn.TagObject);
  Result := FDataSource.IsSorted(Field.Name, Ascending);
end;

Column Calculations

// Built-in calculation support
var
  Total: Double;
begin
  Total := TeeGrid1.Data.Calculate(TeeGrid1.Columns[2], TColumnCalculation.Sum);
  Average := TeeGrid1.Data.Calculate(TeeGrid1.Columns[2], TColumnCalculation.Average);
end;

// Implement in your TVirtualData:
function Calculate(const AColumn: TColumn; const ACalculation: TColumnCalculation): TFloat;
var
  I: Integer;
  Value: TFloat;
begin
  case ACalculation of
    TColumnCalculation.Sum:
      begin
        Result := 0;
        for I := 0 to Count - 1 do
          Result := Result + AsFloat(AColumn, I);
      end;
    // ... other calculations
  end;
end;

Testing Virtual Data

procedure TestMyVirtualData;
var
  Data: TMyVirtualData;
  Columns: TColumns;
  S: String;
begin
  Data := TMyVirtualData.Create(MyDataSource);
  try
    // Test column creation
    Columns := TColumns.Create(nil, TColumn);
    try
      Data.AddColumns(Columns);
      Assert(Columns.Count > 0, 'No columns created');
      
      // Test data retrieval
      if Data.Count > 0 then
      begin
        S := Data.AsString(Columns[0], 0);
        Assert(S <> '', 'Empty value returned');
      end;
      
      // Test edit
      Data.SetValue(Columns[0], 0, 'Test');
      S := Data.AsString(Columns[0], 0);
      Assert(S = 'Test', 'Value not updated');
    finally
      Columns.Free;
    end;
  finally
    Data.Free;
  end;
end;

Best Practices

Memory Management

Store only essential state in TVirtualData
Let the original data source own the actual data
Clear caches in Refresh method
Don’t store TColumn references (they can be freed/recreated)

Error Handling

Protect against invalid row/column indices
Handle data source disconnection gracefully
Return empty string from AsString on errors (don’t raise)
Raise exceptions only in SetValue for validation errors

Thread Safety

TVirtualData methods are called from UI thread only
If data source uses background threads, synchronize access
Use critical sections for shared state

Type Information

Implement DataType() for proper formatting and editors
Use TagObject to store field/property references
Return appropriate THorizontalAlign for column types

Next Steps

Data Binding

Understanding the data binding architecture

Custom Data Source

Complete guide to custom implementations

Database Grids

Working with TVirtualDBData

API Reference

Full TVirtualData API documentation

Get Started

Core Concepts

Data Sources

Features

UI Customization

Framework Integration

Virtual Data Concept

The Virtual Data Pattern

Data Source Flexibility

Lazy Loading

Large Datasets

Framework Agnostic

How Virtual Data Works

Key Characteristics

TVirtualData Abstract Class

Required Methods

Optional Override Methods

Built-in Implementations

TVirtualDBData (Tee.GridData.DB)

TVirtualData<T> (Tee.GridData.Rtti)

TVirtualStringData (Tee.GridData.Strings)

Creating Custom Virtual Data

Basic Implementation Template

Performance Optimization

Data Change Notifications

Advanced Features

Master-Detail Support

Custom Sorting

Column Calculations

Testing Virtual Data

Best Practices

Next Steps

Data Binding

Custom Data Source

Database Grids

API Reference

Build docs developers (and LLMs) love

Get Started

Core Concepts

Data Sources

Features

UI Customization

Framework Integration

​The Virtual Data Pattern

Data Source Flexibility

Lazy Loading

Large Datasets

Framework Agnostic

​How Virtual Data Works

​Key Characteristics

​TVirtualData Abstract Class

​Required Methods

​Optional Override Methods

​Built-in Implementations

​TVirtualDBData (Tee.GridData.DB)

​TVirtualData<T> (Tee.GridData.Rtti)

​TVirtualStringData (Tee.GridData.Strings)

​Creating Custom Virtual Data

​Basic Implementation Template

​Performance Optimization

​Data Change Notifications

​Advanced Features

​Master-Detail Support

​Custom Sorting

​Column Calculations

​Testing Virtual Data

​Best Practices

​Next Steps

Data Binding

Custom Data Source

Database Grids

API Reference

Build docs developers (and LLMs) love

The Virtual Data Pattern

How Virtual Data Works

Key Characteristics

TVirtualData Abstract Class

Required Methods

Optional Override Methods

Built-in Implementations

TVirtualDBData (Tee.GridData.DB)

TVirtualData<T> (Tee.GridData.Rtti)

TVirtualStringData (Tee.GridData.Strings)

Creating Custom Virtual Data

Basic Implementation Template

Performance Optimization

Data Change Notifications

Advanced Features

Master-Detail Support

Custom Sorting

Column Calculations

Testing Virtual Data

Best Practices

Next Steps