mulle-container

🛄 Arrays, hashtables and a queue

A collection of C (C11) data structures to store integers, strings, pointers, structs.

All data structures can live (temporarily) on the stack, or permanently in the heap. None of them are thread-safe. mulle-allocator is used pervasively to simplify memory management. Some data structures utilize callbacks that are compatible to OS X's NSHashtable and friends.

Release Version	Release Notes	AI Documentation
	RELEASENOTES	DeepWiki for mulle-container

Data Structures

The names of the functions are consistent. Each function is a verb that is prefixed with the name of the data structure it handles. So for example the get function for mulle_array is mulle_array_get. The first parameter,

• except for creation functions - is always the container itself.

Check out SYNTAX.md for a somewhat formal explanation and VERBS.md for a list of common verbs being employed.

The various underscore prefixed functions may assert their parameters, but when compiled for "Release" there will be no runtime checks. The non underscore functions protect themselves against NULL pointers. For development it is wise, to use a debug version of the library.

Arrays

All arrays store their elements in a single block of memory.

Standard Arrays

mulle_array conveniently stores managed pointers

This is an array of void * that reallocs on power of two sizes. The elements equality is determined with a callback function. The elements can be copied/freed or reference counted using mulle_container_callbacks. It could be the basis for a NSMutableArray implementation (but isn't currently).

The array was allocated with an initial capacity of 4. The addition of a fifth value forced an expansion to 8 (yellow: initial cells in use, green: additional cells in use). The "notakey" is not used for arrays.

There is also an API Documentation.

mulle__array conveniently stores managed pointers with minimal overhead

This is mulle_array minus the allocator and callback.

mulle_pointerarray stores unmanaged pointers

This is a stripped down version of mulle_array to manage untyped void *. The pointers are simply compared with ==. You can also store integers casted as void *. The memory referenced by a pointer isn't managed (e.g. freed when removed from storage), like mulle-array does.

mulle__pointerarray stores pointers with minimal overhead

This is a stripped down version of mulle_pointerarray minus the allocator. It can be convenient for quickly building up lots of C arrays of void *.

flexarray, a replacement for alloca

The mulle_flexarray can be used as an replacement for alloca. The problem with alloca is always two-fold. 1.) It's non-standard and not available on all platforms. 2.) The amount of memory to alloca may exceed the available stack space. The mulle_flexarray solves this problem by using a small amount of stack space for low memory scenarios and moving to malloc, when it's needed.

Example:

 void  foo( int n, int *data)
{
   mulle_flexarray_do( copy, int, 32, n)
   {
      memcpy( copy, data, n * sizeof( int));    // using copy here for something
   }
}

Special Arrays

mulle__uniquepointerarray keeps unmanaged sorted pointers for reference

mulle__uniquepointerarray is based on mulle__pointerarray. It keeps its elements of void * in an array in sorted order for searching by pointer equality. The sorting is done lazily. It's use is as a set. It's advantage over a hashtable set are low memory requirement and superior performance for smaller quantities.

mulle__rangeset stores selected elements of an array

This is an array of mulle_range elements, that are always kept in sorted order. A mulle_range is defined as

struct mulle_range
{
   uintptr_t   location;
   uintptr_t   length;
};

which is identical to NSRange. Ranges are combined and split as necessary, so that mulle_rangeset can ensure that

• there are no overlapping ranges (they are coalesced)
• there are no adjacent ranges (they are coalesced)
• there are no ranges of size 0 (they are removed)

It could be the basis for an NSIndexSet.

mulle_structarray stores any kind of C type with proper alignment

This is a variant of mulle_pointerarray, but instead of using void * you can specify any C type as the element size. This can be useful for building arrays of float or any kind of struct for example.

Associative Arrays

mulle_pointerpairarray stores unmanaged key/value pairs

This is a modified mulle_pointerarray that holds key/value pairs of void *, void * instead of just a single void *. In comparison to using a map, order of addition is preserved and the space requirement is lower. But it is slow in searches.

mulle__pointerpairarray stores key/value pairs with minimal overhead

This is the mulle_pointerpairarray variant without the allocator.

Queues

mulle__pointerqueue stores unmanaged pointers without copying

This is a FIFO queue for void *. It does not realloc on addition. Instead it chains buckets of void * together. This makes large quantities of additions pretty much as fast as possible.

Filled cells indicate a cell containing a pointer value. All values are possible, so there are no holes. The cells are clustered in buckets and the buckets form a linked list. The _read_index maintains the number of values already read from the current _read bucket. Correspondingly the _write_index keeps a record of the next available index in the _write bucket. If a bucket is exhausted a new one is allocated, chained into the linked list and the _write pointer is set to it, resetting the _write_index.

It is the basis for the NSAutoreleasePool implementation.

mulle__structqueue stores structs instead of pointers

A key difference to mulle_structarray is, that maintaining pointers to mulle__structqueue elements are possible, whereas mulle_structarray elements may get reorganised.

Hashtables

A hashtable is about twice the size of the number of elements contained within, the rest being holes. A hash of the element is used to quickly locate the entry in the hashtable.

Hashtable realloc and redistribute their elements on power of two size boundaries.

A hashtables use "holes" to discriminate between valid entries and available space. Consequently a hashtable can not use all possible void * values as keys and needs a special notakey marker for such holes.

The "notakey" value is usually mulle_not_a_pointer, which is defined as (void *) INTPTR_MIN.

Maps

Hashtable data structures that provide associative storage (key/value) are called maps.

mulle_map conveniently stores managed keys and values

The elements equality is determined with a callback function and the elements can be copied/freed or reference counted using callbacks organized in a mulle_container_callback.

The contigous memory is split into two equal sized parts "KEYS" and "VALUES". Gray cells indicate a "hole", they contain notakey. The corresponding value is undefined (white). The map was allocated with an initial capacity of 4. The addition of a third value forced an expansion to 8, as the hashtable needs enough holes to operate. (yellow: initial cells in use, green: additional cells in use). On growth all the keys and values are redistributed. The cells are placed according to the hash of the key. On collision the next available hole is used.

mulle__map stores managed keys and values with minimal overhead

This is mulle_map minus the allocator and the callback. It is the basis for the NSMutableDictionary implementation.

mulle__pointermap stores unmanaged keys and values

This is a simplified version of mulle__map. Equality is determined by comparing pointers with == and holes are always filled with mulle_not_a_pointer (which is INTPTR_MIN). Therefore no callbacks are used by mulle__pointermap.

Sets

A set is the most simple hashtable. It's like a map minus the value part. The storage of pointers and objects in a set is not very memory efficient, compared to an array, but searching is fast for sets with larger item quantities. (ca. > 100)

mulle_set conveniently stores managed pointers for quick lookup

mulle__set stores managed pointers with minimal overhead

The mulle__set is a mulle_set without the allocator and the callback.

mulle__pointerset stores unmanaged pointers with minimal overhead

This is a simplified version of mulle__set. Equality is determined by comparing pointers with == and holes are always filled with mulle_not_a_pointer (which is INTPTR_MIN). Therefore no callbacks are used by mulle__pointerset.

Predefined Callbacks

Key Callbacks

As used by mulle_array, mulle_map, mulle_set. Each callback name is to be prefixed with mulle_container_keycallback_, which has been elided for brevity:

Name	Type	Notakey	Equality	Copies	Frees
_int	int	INTPTR_MIN	==	NO	NO
_intptr	intptr_t	INTPTR_MIN	==	NO	NO
_nonowned_pointer_or_null	void *	INTPTR_MIN	==	NO	NO
_nonowned_pointer	void *	NULL	==	NO	NO
_owned_pointer	void *	NULL	==	NO	YES
_copied_cstring	char *	NULL	strcmp	YES	YES
_nonowned_cstring	char *	NULL	strcmp	NO	NO
_owned_cstring	char *	NULL	strcmp	NO	YES

Value Callbacks

As used by mulle_map. Value callbacks don't need to hash and check for equality. Each callback name is to be prefixed with mulle_container_valuecallback, which has been elided for brevity.

Name	Type	Copies	Frees
_int	int	NO	NO
_intptr	intptr_t	NO	NO
_nonowned_pointer	void *	NO	NO
_owned_pointer	void *	NO	YES
_copied_cstring	char *	YES	YES
_nonowned_cstring	char *	NO	NO
_owned_cstring	char *	NO	YES

Efficiency

As time critical sections are inlined, the performance of the library code is as good as your compiler (within the runtime limitations of the algorithm).

Here is an example of mulle-container code, compiled with -Os:

int   _mulle__pointerset_count_zeroes_generic( struct mulle__pointerset *src,
                                               struct mulle_container_keycallback *callback,
                                               struct mulle_allocator *allocator)
{
   struct mulle__genericpointersetenumerator  rover;
   void                                       *item;
   int                                        rval;

   rval  = 0;
   rover = _mulle__pointerset_enumerate_generic( src, callback);
   while( _mulle__genericpointersetenumerator_next( &rover, &item))
   {
       if( item == 0)
           ++rval;
   }
   mulle__genericpointersetenumerator_done( &rover);
   return( rval);
}

produces:

_mulle__pointerset_count_zeroes_generic(mulle__pointerset*, mulle_container_keycallback*, mulle_allocator*): # @_mulle__pointerset_count_zeroes_generic(mulle__pointerset*, mulle_container_keycallback*, mulle_allocator*)
        mov     rcx, qword ptr [rdi + 8]
        test    rcx, rcx
        je      .LBB0_1
        mov     rdx, qword ptr [rsi + 40]
        mov     rsi, qword ptr [rdi]
        xor     eax, eax
.LBB0_3:                                # =>This Inner Loop Header: Depth=1
        mov     rdi, qword ptr [rsi]
        add     rsi, 8
        cmp     rdi, rdx
        je      .LBB0_3
        cmp     rdi, 1
        adc     eax, 0
        dec     rcx
        jne     .LBB0_3
        ret
.LBB0_1:
        xor     eax, eax
        ret

You are here

Add

This project is a component of the mulle-core library. As such you usually will not add or install it individually, unless you specifically do not want to link against mulle-core.

Add as an individual component

Use mulle-sde to add mulle-container to your project:

mulle-sde add github:mulle-c/mulle-container

To only add the sources of mulle-container with dependency sources use clib:

clib install --out src/mulle-c mulle-c/mulle-container

Add -isystem src/mulle-c to your CFLAGS and compile all the sources that were downloaded with your project.

Install

Use mulle-sde to build and install mulle-container and all dependencies:

mulle-sde install --prefix /usr/local \
   https://github.com/mulle-c/mulle-container/archive/latest.tar.gz

Legacy Installation

Install the requirements:

Requirements	Description
mulle-allocator	🔄 Flexible C memory allocation scheme
mulle-data	#️⃣ A collection of hash functions

Download the latest tar or zip archive and unpack it.

Install mulle-container into /usr/local with cmake:

PREFIX_DIR="/usr/local"
cmake -B build                               \
      -DMULLE_SDK_PATH="${PREFIX_DIR}"       \
      -DCMAKE_INSTALL_PREFIX="${PREFIX_DIR}" \
      -DCMAKE_PREFIX_PATH="${PREFIX_DIR}"    \
      -DCMAKE_BUILD_TYPE=Release &&
cmake --build build --config Release &&
cmake --install build --config Release

Author

Nat! for Mulle kybernetiK