Difference between revisions of "Container Ideas"

From D Wiki
Jump to: navigation, search
(Methods)
(Methods)
Line 88: Line 88:
 
! align="left"style="border-left: 2px solid;" style="border-right: 2px solid;"|opSlice(KeyType, KeyType)
 
! align="left"style="border-left: 2px solid;" style="border-right: 2px solid;"|opSlice(KeyType, KeyType)
 
|-
 
|-
! align="left"style="border-left: 2px solid;" style="border-right: 2px solid;"| in || || ||
+
! align="left"style="border-left: 2px solid;" style="border-right: 2px solid;"| in || || || style="border-right: 2px solid;"|
|bgcolor="Iceberg" align="center" style="border-right: 2px solid;"|bool  
+
|bgcolor="Iceberg" align="center"|bool  
 
|bgcolor="Iceberg" align="center"|bool  
 
|bgcolor="Iceberg" align="center"|bool  
 
|bgcolor="Iceberg" align="center"|bool  
 
|bgcolor="Iceberg" align="center"|bool  

Revision as of 08:52, 15 February 2013

Variable Implementation

As the discussion on the mailing-list (Deque impl.) has shown the ideas about how the internals of the container should work differ. Not only for the general case but also for specific cases like small (<64) vectors. Another application for this are the mapping container. For some tasks a hashtable might be best suited for other a binary-tree. Sometimes even, the user might supply a even more used implementation for the given task.

Therefor I suggest a design that lets the user switch out the underlying implementation. To supply the implementation I see two possibilities. The first is to supply it as a template. The second is to supply it as an argument. The first possibility has the disadvantage that different underlying implementation introduce new types. On the plus side the compile can remove the delegation as it is static. The second possibility does not introduce a new type, this is at the cost of a vtable lookup for every operation. It also forces the range types to work for all implementations. This might lead to making ranges abstract class or interfaces. Therefor I prefer the template approach.

Benchmarking

Having different implementation for the same container raises the question when to use which. The answer to this is as complex as the usage of the container. Therefor I suggest an extensive benchmark suite that tests every method for with different payload sizes for different number of elements.

Methods

notconst const immutable const or immutable
exists means that name or type exists
Sequence Associative Special Access
Container Double Linked List(T) Single Linked List(T) Random Access(T) Map(T,S) Multimap(T,S) MapSet(T,S) Set(T) Stack(T) Queue(T) PriorityQueue(T)
Types RangeType exists exists exists exists exists exists exists exists exists
cRangeType exists exists exists exists exists exists exists exists exists
ElementType T T T S S S T T T T
KeyType size_t size_t size_t T T T T size_t size_t
Modify insertFront(T)
insertFront(T...)
insertBack(T)
insertBack(T...)
insertBefore(RangeType, T)
insertBefore(RangeType, T...)
insertAfter(RangeType, T) T
insertAfter(RangeType, T...)
removeFront(size_t = 1)
removeBack(size_t = 1)
remove(KeyType)
remove(Range)
remove(ElementType)
clear()
Access front()
back()
opIndex(KeyType)
find(KeyType)
opSlice(KeyType)
opSlice(KeyType, KeyType)
in bool bool bool bool
Properties length size_t size_t size_t size_t size_t size_t size_t size_t size_t size_t
empty bool bool bool bool bool bool bool bool bool bool
Capacity capacity
compact
reserve
max_size
growable
packed