Creates a new empty skip list.
Creates a new empty skip list. Type parameter S specifies the STM system to use. Type parameter A
specifies the type of the keys stored in the list.
the minimum gap-size used for the skip list. This value must be between 1 and 126 inclusive.
an object which observes key promotions and demotions. Use NoKeyObserver (default) if
key motions do not need to be monitored. The monitoring allows the use of the skip list
for synchronized decimations of related data structures, such as the deterministic
skip quadtree.
the transaction in which to initialize the structure
the ordering of the keys. This is an instance of txn.Ordering to allow
for specialized versions and transactional restrictions.
the serializer for the elements, in case a persistent STM is used.
Creates a new empty skip list with default minimum gap parameter of 2 and no key observer.
Creates a new empty skip list with default minimum gap parameter of 2 and no key observer.
Type parameter S specifies the STM system to use. Type parameter A
specifies the type of the keys stored in the list.
the transaction in which to initialize the structure
the ordering of the keys. This is an instance of txn.Ordering to allow
for specialized versions and transactional restrictions.
the serializer for the elements, in case a persistent STM is used.
A transactional version of the deterministic k-(2k+1) top-down operated skip list as described in T. Papadakis, Skip Lists and Probabilistic Analysis of Algorithms. Ch. 4 (Deterministic Skip Lists), pp. 55--78. Waterloo (CA) 1993
It uses the horizontal array technique with a parameter for k (minimum gap size). It uses a modified top-down removal algorithm that avoids the need for a second pass as in the original algorithm, and is careful about object creations, so that it will be able to persist the data structure without any unnecessary reads or writes to the store.
Three implementation notes: (1) We treat the nodes as immutable at the moment, storing them directly in the S#Val child pointers of their parents. While this currently seems to have a performance advantage (?), we could try to avoid this by using S#Refs for the child pointers, making the nodes becomes mutables. We could avoid copying the arrays for each insertion or deletion, at the cost of more space, but maybe better performance.
(2) The special treatment of
isRightkind of sucks. Since now that information is also persisted, we might just have two types of branches and leaves, and avoid passing around this flag.(3) Since there is a bug with the top-down one-pass removal, we might end up removing the creation of instances of virtual branches altogether again when replacing the current algorithm by a two-pass one.
TODO: nodes or at least leaves should be horizontally connected for a faster iterator and fast pair (interval) search