What Is a Hashtable?
A hashtable is a data construct that stores a set of items. Each item has a key that identifies the item. Items are found, added, and removed from the hashtable by using the key. Hashtables may seem like arrays, but there are important differences:
|Keys:||integer: arrays are always keyed on integers, and must be contiguous.||any type: almost any datatype can be used as key, including strings, integers, XPCOM interface pointers, IIDs, and almost anything else. Keys can be disjunct (i.e. you can store entries with keys 1, 5, and 3000).|
|Lookup Time:||O(1): lookup time is a simple constant||O(1): lookup time is mostly-constant, but the constant time can be larger than an array lookup|
|Sorting:||sorted: stored sorted; enumerated in a sorted fashion.||unsorted: stored unsorted; cannot be enumerated in a sorted manner.|
|Inserting/Removing:||O(n): adding and removing items from a large array can be time-consuming||O(1): adding and removing items from hashtables is a quick operation|
|Wasted space:||none: Arrays are packed structures, so there is no wasted space.||some: hashtables are not packed structures; depending on on the implementation, there may be significant wasted memory.|
In their implementation, hashtables take the key and apply a mathematical hash function to “randomize�? the key and then use the hash to find the location in the hashtable. Good hashtable implementations will automatically resize the hashtable in memory if extra space is needed, or if too much space has been allocated.
When Should I Use a Hashtable?
Hashtables are useful for
- sets of data that need swift random access;
- with non-integral keys or non-contiguous integral keys;
- or where items will be frequently added or removed.
Hashtables should not be used for
- Sets that need to be sorted;
- Very small datasets (less than 12-16 items);
- Data that does not need random access.
In these situations, an array, a linked-list, or various tree data structures are more efficient.
Mozilla's Hashtable Implementations
Mozilla has several hashtable implementations, which have been tested and, tuned, and hide the inner complexities of hashtable implementations:
PLDHash- low-level C API; stores keys and data in one large memory structure; uses the heap efficiently; client must declare an "entry class" and may not hold onto entry pointers.
PLHashTable- low-level C API; entry class pointers are constant; more efficient for large entry structures; often wastes memory making many small heap allocations.
nsTHashtable- low-level C++ wrapper around
PLDHash; generates callback functions and handles most casting automagically. Client writes their own entry class which can include complex key and data types.
nsDataHashtable/nsInterfaceHashtable/nsClassHashtable- high-level C++ wrappers around
PLDHash; simplifies the common usage pattern mapping a simple keytype to a simple datatype; client does not need to declare or manage an entry class;
nsDataHashtabledatatype is a scalar such as
nsInterfaceHashtabledatatype is an interface;
nsClassHashtabledatatype is a class pointer owned by the hashtable.
Which Hashtable Should I Use?
|Data Type:||None (Hash Set)||
JSDHash are the same thing; one is linked from the XPCOM libraries, the other from the JS libraries.
JSDHash is used extensively in the SpiderMonkey JS engine.
PLDHash implementation is a fairly low-level implementation, written in C. It is extremely flexible, but requires some time to understand and use. A basic guide is included here, but you should read most of
xpcom/ds/pldhash.h if you intend to use
PLDHash. The C++ wrappers for
PLDHash (see below) are often much easier and safer to use in C++ code, as many potential casting errors are easily avoided.
You must declare an entry struct type, deriving from
PLDHashEntryHdr. This entry struct should contain whatever data you wish to store in the hashtable (any pointer or fixed-length data type). Note: because of the double-hashing implementation, entries may move in memory when the hashtable is altered. If you need entry pointers to remain constant, you may want to consider using
You must also initialize a
PLDHashTableOps structure. This serves similarly to a vtable in C++, with pointers to appropriate user-defined functions that initialize, compare, and match entries. Because
PLDHash does not know what datatype your key is, all functions that work with keys are declared using
const void*, and your client code must cast these pointers to the appropriate type.
PLDHashTables can be allocated on the stack or the heap:
- When allocated on the stack, or as a C++ class member, the table must be initialized using
PL_DHashTableInit, and finalized using
- When allocated on the heap, use
PL_DHashTableDestroyto allocate and delete the table.
There are two situations where
PLHashTable may be preferable to
- You need entry-pointers to remain constant.
- The entries stored in the table are very large (larger than 12 words).
PLDHashdoes not handle large entry structures efficiently.
nsTHashtable is a C++ template that wraps
PLDHash. It hides many of the complexities of
PLDHash (callback functions, the ops structure, etc). You should read
nsTHashtable, you must declare an entry-class in a pre-defined format. This entry class contains the key and the data that you are hashing (just like
PLDHash, above). It also declares functions that manipulate the key. In most cases, the functions of this entry class can be entirely inline. For examples of entry classes, see the declarations at
The template parameter is the entry class. You must use the
Init() function to initalize the table properly. At this point, use the functions
PutEntry/GetEntry/RemoveEntry to alter the hashtable.
EnumerateEntries will do enumeration, but beware that the enumeration will occur in a seemingly-random order (no sorting).
nsTHashtables can be allocated on the stack, as class members, or on the heap.
- Entry pointers can and do change when items are added to the hashtable, or removed. Do not keep long-lasting pointers to entries.
- because of this,
nsTHashtableis not inherently thread-safe. If you use a hashtable in a multi-thread environment, you must provide locking as appropriate.
nsTHashtable, see if
nsBaseHashtable and relatives will work for you. They are much easier to use, because you do not have to declare an entry class. If you are hashing a simple key type to a simple data type, they are generally a better choice.
nsBaseHashtable and friends: nsDataHashtable, nsInterfaceHashtable, and nsClassHashtable
These C++ templates provide a high-level interface for using hashtables that hides most of the complexities of
PLDHash. They provide the following features:
- hashtable operations can be completed without using an entry class, making code easier to read;
- optional thread-safety: the hashtable can manage a read-write lock around the table;
- predefined key classes provide automatic cleanup of strings/interfaces
nsClassHashtableautomatically release/delete data pointers to avoid leaks.
nsBaseHashtable is not used directly; choose one of the three derivative classes based on the data type you want to store. The
KeyClass is taken from
nsHashKeys.h and is the same for all three classes:
DataTypeis a simple type such as
Interfaceis an XPCOM interface such as
Tis any C++ class. The hashtable stores a pointer to the class, and deletes it when the entry is removed.
The important files to read are
xpcom/ds/nsHashKeys.h. These classes can be used on the stack, as a class member, or on the heap. Initialize using the
Init() function; you can specify whether you need thread-safety at this time. Use the
Remove() methods to alter the table.
There are two enumeration functions:
EnumerateRead()performs a read-only enumeration, where entries cannot be changed or removed;
Enumerate()performs a read-write enumeration, where entries may be changed or removed as necessary.
A hash set only tracks the existence of keys: it does not associate data with the keys.
nsHashSets has predefined hash sets for common keys, which are trivially easy to use. See
xpcom/ds/nsHashSets.h. The appropriate functions are
Although some predefined hash sets are already available in
nsHashSets.h, a generic solution using templates (similar to
nsDataHashtable) would be easy to implement. When I have time...
nsClassHashtable keeps owning pointers to C++ objects. However, for many uses such as keeping
nsCOMArray as data members, the class can be allocated within the hashtable, rather than as a pointer. This can reduce heap allocations, though it prevents thread-safety.
nsHashtable has a wrapper that exposes an
nsISimpleEnumerator on its items. I will add this support to the various
nsBaseHashtable classes as well, as needed.
All of the above hashtables need a Hash Function. This function converts the key into a semi-unique integer. The mozilla codebase already contains hash functions for most key types, including narrow and wide strings, pointers, and most binary data:
Writing a good hash function is well beyond the scope of this document, and has been discussed extensively in computer-science circles for many years. There are many different types of hash functions. Mozilla has tuned a good general-purpose hash algorithm for strings and
Mozilla's Old/Obsolete/Deprecated/Decrepit Hashtables
nsHashtable was a C++ wrapper around
PLHashTable, and now wraps
PLDHash. The design of the key classes is not optimal, however, and
nsHashtable has been deprecated in favor of
nsDataHashtable and friends.
Original Document Information
- Author(s): Benjamin Smedberg <email@example.com>