YapDatabaseViewSorting

@interface YapDatabaseViewSorting : NSObject

The sorting block handles sorting of objects within their group.

After the view invokes the grouping block to determine what group a database row belongs to (if any), the view then needs to determine what index within that group the row should be. In order to do this, it needs to compare the new/updated row with existing rows in the same view group. This is what the sorting block is used for. So the sorting block will be invoked automatically during this process until the view has come to a conclusion.

You should choose a block type that takes the minimum number of required parameters. The view can make various optimizations based on required parameters of the block.

For example, if sorting is based on the object, and the metadata of a row is updated, then the view can deduce that the index hasn’t changed (if the group hans’t), and can skip this step.

Performance Note:

The view uses various optimizations (based on common patterns) to reduce the number of times it needs to invoke the sorting block.

• Pattern : row is updated, but its index in the view doesn’t change. Optimization : if an updated row doesn’t change groups, the view will first compare it with objects to the left and right.

• Pattern : rows are added to the beginning or end or a view Optimization : if the last change put an object at the beginning of the view, then it will test this quickly. if the last change put an object at the end of the view, then it will test this quickly.

These optimizations offer huge performance benefits to many common cases. For example, adding objects to a view that are sorted by timestamp of when they arrived.

The optimizations are not always performed. For example, if the last change didn’t place an item at the beginning or end of the view.

If optimizations fail, or are skipped, then the view uses a binary search algorithm.

Although this may be considered internal information, I feel it is important to explain for the following reason:

Another common pattern is to fetch a number of objects in a batch, and then insert them into the database. Now imagine a situation in which the view is sorting posts based on timestamp, and you just fetched the most recent 10 posts. You can enumerate these 10 posts either forwards or backwards while adding them to the database. One direction will hit the optimization every time. The other will cause the view to perform a binary search every time. These little one-liner optimzations are easy (given this internal information is known).