Unpivot

Turns each event into one event per value field, each holding the key, the field’s name and its value.

Unpivot is the reverse of Pivot. It takes four arguments:

• a constructor for the output type, such as () => new Triple();
• a key selector, such as p => p.Key, which copies the input’s key field into the output field of the same name;
• an attribute selector, t => t.Attribute, naming the output field that receives the name of the input field;
• a value selector, t => t.Value, naming the output field that receives the input field’s value.

With a value field of a non-nullable value type, such as the int Value here, each other input field of the matching nullable type (int?) becomes one output event with the input event’s lifetime, unless it is null. A field of another type, such as string, is ignored. Other kinds of value field follow other rules, and some combinations fail: an int input field with an object value field throws when the query is built.

In the diagram, a Select first builds a Pair with the nullable fields Low and High. Unpivot then turns each Pair into two events: a:1 becomes High 10 and Low 1, both over [1,5).

Known issues in this fork. With an int value field, an input field of type int is always skipped, although an int? field is unpivoted. Nullable properties are handled the wrong way round: an int? property with a value is skipped, and a null one throws an InvalidOperationException. The causes are in the engine, and the fixes belong there. The diagram uses nullable fields, which work.

See also: Pivot, SelectMany