src/array_list.c

Tue, 16 Dec 2025 12:07:01 +0100

author
Mike Becker <universe@uap-core.de>
date
Tue, 16 Dec 2025 12:07:01 +0100
changeset 1611
de21dd0d1426
parent 1608
46d8a8305948
permissions
-rw-r--r--

adds docstrings to the new array API - relates to #619

/*
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS HEADER.
 *
 * Copyright 2021 Mike Becker, Olaf Wintermann All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *
 *   1. Redistributions of source code must retain the above copyright
 *      notice, this list of conditions and the following disclaimer.
 *
 *   2. Redistributions in binary form must reproduce the above copyright
 *      notice, this list of conditions and the following disclaimer in the
 *      documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 */

#include "cx/array_list.h"
#include "cx/compare.h"
#include <assert.h>
#include <string.h>
#include <errno.h>

// LOW LEVEL ARRAY LIST FUNCTIONS

/**
 * Intelligently calculates a new capacity, reserving some more
 * elements than required to prevent too many allocations.
 *
 * @param current_capacity the current capacity of the array
 * @param needed_capacity the required capacity of the array
 * @return the new capacity
 */
static size_t cx_array_grow_capacity(
    size_t current_capacity,
    size_t needed_capacity
) {
    if (current_capacity >= needed_capacity) {
        return current_capacity;
    }
    size_t cap = needed_capacity;
    size_t alignment;
    if (cap < 128) alignment = 16;
    else if (cap < 1024) alignment = 64;
    else if (cap < 8192) alignment = 512;
    else alignment = 1024;
    return cap - (cap % alignment) + alignment;
}

int cx_array_init_(const CxAllocator *allocator, CxArray *array, size_t elem_size, size_t capacity) {
    memset(array, 0, sizeof(CxArray));
    return cx_array_reserve_(allocator, array, elem_size, capacity);
}

void cx_array_init_fixed_(CxArray *array, const void *data, size_t capacity, size_t size) {
    array->data = (void*) data;
    array->capacity = capacity;
    array->size = size;
}

int cx_array_reserve_(const CxAllocator *allocator, CxArray *array, size_t elem_size, size_t capacity) {
    if (cxReallocateArray(allocator, &array->data, capacity, elem_size)) {
        return -1; // LCOV_EXCL_LINE
    }
    array->capacity = capacity;
    if (array->size > capacity) {
        array->size = capacity;
    }
    return 0;
}

int cx_array_copy_to_new_(const CxAllocator *allocator, CxArray *array, size_t elem_size, size_t capacity) {
    CxArray heap_array;
    if (cx_array_init_(allocator, &heap_array, elem_size, capacity)) {
        return -1; // LCOV_EXCL_LINE
    }
    heap_array.size = array->size;
    memcpy(heap_array.data, array->data, elem_size * array->size);
    *array = heap_array;
    return 0;
}

int cx_array_insert_(const CxAllocator *allocator, CxArray *array,
        size_t elem_size, size_t index, const void *other, size_t n) {
    // out of bounds and special case check
    if (index > array->size) return -1;
    if (n == 0) return 0;

    // guarantee enough capacity
    if (array->capacity < array->size + n) {
        const size_t new_capacity = cx_array_grow_capacity(array->capacity,array->size + n);
        if (cxReallocateArray(allocator, &array->data, new_capacity, elem_size)) {
            return -1; // LCOV_EXCL_LINE
        }
        array->capacity = new_capacity;
    }

    // determine insert position
    char *dst = array->data;
    dst += index * elem_size;

    // do we need to move some elements?
    if (index < array->size) {
        size_t elems_to_move = array->size - index;
        char *target = dst + n * elem_size;
        memmove(target, dst, elems_to_move * elem_size);
    }

    // place the new elements, if any
    // otherwise, this function just reserved the memory (a.k.a emplace)
    if (other != NULL) {
        memcpy(dst, other, n * elem_size);
    }
    array->size += n;

    return 0;
}

int cx_array_insert_sorted_(
        const CxAllocator *allocator,
        CxArray *array,
        size_t elem_size,
        cx_compare_func cmp_func,
        const void *sorted_data,
        size_t n,
        bool allow_duplicates
) {
    // assert pointers
    assert(allocator != NULL);
    assert(array != NULL);
    assert(cmp_func != NULL);
    assert(sorted_data != NULL);

    // corner case
    if (n == 0) return 0;

    // overflow check
    // LCOV_EXCL_START
    if (n > SIZE_MAX - array->size) {
        errno = EOVERFLOW;
        return 1;
    }
    // LCOV_EXCL_STOP

    // store some counts
    const size_t old_size = array->size;
    const size_t old_capacity = array->capacity;
    // the necessary capacity is the worst case assumption, including duplicates
    const size_t needed_capacity = cx_array_grow_capacity(old_capacity, old_size + n);

    // if we need more than we have, try a reallocation
    if (needed_capacity > old_capacity) {
        if (cxReallocateArray(allocator, &array->data, needed_capacity, elem_size)) {
            return -1; // LCOV_EXCL_LINE
        }
        array->capacity = needed_capacity;
    }

    // now we have guaranteed that we can insert everything
    size_t new_size = old_size + n;
    array->size = new_size;

    // declare the source and destination indices/pointers
    size_t si = 0, di = 0;
    const char *src = sorted_data;
    char *dest = array->data;

    // find the first insertion point
    di = cx_array_binary_search_sup(dest, old_size, elem_size, src, cmp_func);
    dest += di * elem_size;

    // move the remaining elements in the array completely to the right
    // we will call it the "buffer" for parked elements
    size_t buf_size = old_size - di;
    size_t bi = new_size - buf_size;
    char *bptr = ((char *) array->data) + bi * elem_size;
    memmove(bptr, dest, buf_size * elem_size);

    // while there are both source and buffered elements left,
    // copy them interleaving
    while (si < n && bi < new_size) {
        // determine how many source elements can be inserted.
        // the first element that shall not be inserted is the smallest element
        // that is strictly larger than the first buffered element
        // (located at the index of the infimum plus one).
        // the infimum is guaranteed to exist:
        // - if all src elements are larger,
        //   there is no buffer, and this loop is skipped
        // - if any src element is smaller or equal, the infimum exists
        // - when all src elements that are smaller are copied, the second part
        //   of this loop body will copy the remaining buffer (emptying it)
        // Therefore, the buffer can never contain an element that is smaller
        // than any element in the source and the infimum exists.
        size_t copy_len, bytes_copied;
        copy_len = cx_array_binary_search_inf(
            src, n - si, elem_size, bptr, cmp_func
        );
        copy_len++;

        // copy the source elements
        if (copy_len > 0) {
            if (allow_duplicates) {
                // we can copy the entire chunk
                bytes_copied = copy_len * elem_size;
                memcpy(dest, src, bytes_copied);
                dest += bytes_copied;
                src += bytes_copied;
                si += copy_len;
                di += copy_len;
            } else {
                // first, check the end of the source chunk
                // for being a duplicate of the bptr
                const char *end_of_src = src + (copy_len - 1) * elem_size;
                size_t skip_len = 0;
                while (copy_len > 0 && cmp_func(bptr, end_of_src) == 0) {
                    end_of_src -= elem_size;
                    skip_len++;
                    copy_len--;
                }
                char *last = dest == array->data ? NULL : dest - elem_size;
                // then iterate through the source chunk
                // and skip all duplicates with the last element in the array
                size_t more_skipped = 0;
                for (unsigned j = 0; j < copy_len; j++) {
                    if (last != NULL && cmp_func(last, src) == 0) {
                        // duplicate - skip
                        src += elem_size;
                        si++;
                        more_skipped++;
                    } else {
                        memcpy(dest, src, elem_size);
                        src += elem_size;
                        last = dest;
                        dest += elem_size;
                        si++;
                        di++;
                    }
                }
                // skip the previously identified elements as well
                src += skip_len * elem_size;
                si += skip_len;
                skip_len += more_skipped;
                // reduce the actual size by the number of skipped elements
                array->size -= skip_len;
            }
        }

        // when all source elements are in place, we are done
        if (si >= n) break;

        // determine how many buffered elements need to be restored
        copy_len = cx_array_binary_search_sup(
                bptr,
                new_size - bi,
                elem_size,
                src,
                cmp_func
        );

        // restore the buffered elements
        bytes_copied = copy_len * elem_size;
        memmove(dest, bptr, bytes_copied);
        dest += bytes_copied;
        bptr += bytes_copied;
        di += copy_len;
        bi += copy_len;
    }

    // still source elements left?
    if (si < n) {
        if (allow_duplicates) {
            // duplicates allowed or nothing inserted yet: simply copy everything
            memcpy(dest, src, elem_size * (n - si));
        } else {
            // we must check the remaining source elements one by one
            // to skip the duplicates.
            // Note that no source element can equal the last element in the
            // destination, because that would have created an insertion point
            // and a buffer, s.t. the above loop already handled the duplicates
            while (si < n) {
                // find a chain of elements that can be copied
                size_t copy_len = 1, skip_len = 0;
                {
                    const char *left_src = src;
                    while (si + copy_len + skip_len < n) {
                        const char *right_src = left_src + elem_size;
                        int d = cmp_func(left_src,  right_src);
                        if (d < 0) {
                            if (skip_len > 0) {
                                // new larger element found;
                                // handle it in the next cycle
                                break;
                            }
                            left_src += elem_size;
                            copy_len++;
                        } else if (d == 0) {
                            left_src += elem_size;
                            skip_len++;
                        } else {
                            break;
                        }
                    }
                }
                size_t bytes_copied = copy_len * elem_size;
                memcpy(dest, src, bytes_copied);
                dest += bytes_copied;
                src += bytes_copied + skip_len * elem_size;
                si += copy_len + skip_len;
                di += copy_len;
                array->size -= skip_len;
            }
        }
    }

    // buffered elements need to be moved when we skipped duplicates
    size_t total_skipped = new_size - array->size;
    if (bi < new_size && total_skipped > 0) {
        // move the remaining buffer to the end of the array
        memmove(dest, bptr, elem_size * (new_size - bi));
    }

    return 0;
}

CxIterator cx_array_iterator_(CxArray *array, size_t elem_size) {
    return cxIterator(array->data, elem_size, array->size);
}

CxIterator cx_array_iterator_ptr_(CxArray *array) {
    return cxIteratorPtr(array->data, array->size);
}

void cx_array_free_(const CxAllocator *allocator, CxArray *array) {
    cxFree(allocator, array->data);
    array->data = NULL;
    array->size = array->capacity = 0;
}


// implementation that finds ANY index
static size_t cx_array_binary_search_inf_impl(
        const void *arr,
        size_t size,
        size_t elem_size,
        const void *elem,
        cx_compare_func cmp_func
) {
    // special case: empty array
    if (size == 0) return 0;

    // declare a variable that will contain the compare results
    int result;

    // cast the array pointer to something we can use offsets with
    const char *array = arr;

    // check the first array element
    result = cmp_func(elem, array);
    if (result < 0) {
        return size;
    } else if (result == 0) {
        return 0;
    }

    // special case: there is only one element and that is smaller
    if (size == 1) return 0;

    // check the last array element
    result = cmp_func(elem, array + elem_size * (size - 1));
    if (result >= 0) {
        return size - 1;
    }

    // the element is now guaranteed to be somewhere in the list
    // so start the binary search
    size_t left_index = 1;
    size_t right_index = size - 1;
    size_t pivot_index;

    while (left_index <= right_index) {
        pivot_index = left_index + (right_index - left_index) / 2;
        const char *arr_elem = array + pivot_index * elem_size;
        result = cmp_func(elem, arr_elem);
        if (result == 0) {
            // found it!
            return pivot_index;
        } else if (result < 0) {
            // element is smaller than pivot, continue search left
            right_index = pivot_index - 1;
        } else {
            // element is larger than pivot, continue search right
            left_index = pivot_index + 1;
        }
    }

    // report the largest upper bound
    return result < 0 ? (pivot_index - 1) : pivot_index;
}

size_t cx_array_binary_search_inf(
        const void *arr,
        size_t size,
        size_t elem_size,
        const void *elem,
        cx_compare_func cmp_func
) {
    size_t index = cx_array_binary_search_inf_impl(
        arr, size, elem_size, elem, cmp_func);
    // in case of equality, report the largest index
    const char *e = ((const char *) arr) + (index + 1) * elem_size;
    while (index + 1 < size && cmp_func(e, elem) == 0) {
        e += elem_size;
        index++;
    }
    return index;
}

size_t cx_array_binary_search(
        const void *arr,
        size_t size,
        size_t elem_size,
        const void *elem,
        cx_compare_func cmp_func
) {
    size_t index = cx_array_binary_search_inf(
            arr, size, elem_size, elem, cmp_func
    );
    if (index < size &&
            cmp_func(((const char *) arr) + index * elem_size, elem) == 0) {
        return index;
    } else {
        return size;
    }
}

size_t cx_array_binary_search_sup(
        const void *arr,
        size_t size,
        size_t elem_size,
        const void *elem,
        cx_compare_func cmp_func
) {
    size_t index = cx_array_binary_search_inf_impl(
            arr, size, elem_size, elem, cmp_func
    );
    const char *e = ((const char *) arr) + index * elem_size;
    if (index == size) {
        // no infimum means the first element is supremum
        return 0;
    } else if (cmp_func(e, elem) == 0) {
        // found an equal element, search the smallest index
        e -= elem_size; // e now contains the element at index-1
        while (index > 0 && cmp_func(e, elem) == 0) {
            e -= elem_size;
            index--;
        }
        return index;
    } else {
        // we already have the largest index of the infimum (by design)
        // the next element is the supremum (or there is no supremum)
        return index + 1;
    }
}

#ifndef CX_ARRAY_SWAP_SBO_SIZE
#define CX_ARRAY_SWAP_SBO_SIZE 128
#endif
const unsigned cx_array_swap_sbo_size = CX_ARRAY_SWAP_SBO_SIZE;

void cx_array_swap(
        void *arr,
        size_t elem_size,
        size_t idx1,
        size_t idx2
) {
    assert(arr != NULL);

    // short circuit
    if (idx1 == idx2) return;

    char sbo_mem[CX_ARRAY_SWAP_SBO_SIZE];
    void *tmp;

    // decide if we can use the local buffer
    if (elem_size > CX_ARRAY_SWAP_SBO_SIZE) {
        tmp = cxMallocDefault(elem_size);
        // we don't want to enforce error handling
        if (tmp == NULL) abort();
    } else {
        tmp = sbo_mem;
    }

    // calculate memory locations
    char *left = arr, *right = arr;
    left += idx1 * elem_size;
    right += idx2 * elem_size;

    // three-way swap
    memcpy(tmp, left, elem_size);
    memcpy(left, right, elem_size);
    memcpy(right, tmp, elem_size);

    // free dynamic memory, if it was needed
    if (tmp != sbo_mem) {
        cxFreeDefault(tmp);
    }
}

// HIGH LEVEL ARRAY LIST FUNCTIONS

typedef struct {
    struct cx_list_s base;
    void *data;
    size_t capacity;
} cx_array_list;

static void cx_arl_destructor(struct cx_list_s *list) {
    cx_array_list *arl = (cx_array_list *) list;

    char *ptr = arl->data;

    if (list->collection.simple_destructor) {
        for (size_t i = 0; i < list->collection.size; i++) {
            cx_invoke_simple_destructor(list, ptr);
            ptr += list->collection.elem_size;
        }
    }
    if (list->collection.advanced_destructor) {
        for (size_t i = 0; i < list->collection.size; i++) {
            cx_invoke_advanced_destructor(list, ptr);
            ptr += list->collection.elem_size;
        }
    }

    cxFree(list->collection.allocator, arl->data);
    cxFree(list->collection.allocator, list);
}

static size_t cx_arl_insert_array(
        struct cx_list_s *list,
        size_t index,
        const void *array,
        size_t n
) {
    cx_array_list *arl = (cx_array_list *) list;
    CxArray wrap = {
        arl->data, list->collection.size, arl->capacity
    };
    if (cx_array_insert_(list->collection.allocator, &wrap,
            list->collection.elem_size, index, array, n)) {
        return 0;
    }
    arl->data = wrap.data;
    arl->capacity = wrap.capacity;
    list->collection.size = wrap.size;
    return n;
}

static size_t cx_arl_insert_sorted_impl(
        struct cx_list_s *list,
        const void *sorted_data,
        size_t n,
        bool allow_duplicates
) {
    cx_array_list *arl = (cx_array_list *) list;
    CxArray wrap = {
        arl->data, list->collection.size, arl->capacity
    };

    if (cx_array_insert_sorted_(
            list->collection.allocator,
            &wrap,
            list->collection.elem_size,
            list->collection.cmpfunc,
            sorted_data,
            n,
            allow_duplicates
    )) {
        // array list implementation is "all or nothing"
        return 0;  // LCOV_EXCL_LINE
    }
    arl->data = wrap.data;
    arl->capacity = wrap.capacity;
    list->collection.size = wrap.size;
    return n;
}

static size_t cx_arl_insert_sorted(
        struct cx_list_s *list,
        const void *sorted_data,
        size_t n
) {
    return cx_arl_insert_sorted_impl(list, sorted_data, n, true);
}

static size_t cx_arl_insert_unique(
        struct cx_list_s *list,
        const void *sorted_data,
        size_t n
) {
    return cx_arl_insert_sorted_impl(list, sorted_data, n, false);
}

static void *cx_arl_insert_element(
        struct cx_list_s *list,
        size_t index,
        const void *element
) {
    if (cx_arl_insert_array(list, index, element, 1) == 1) {
        return ((char*)((cx_array_list *) list)->data) + index * list->collection.elem_size;
    } else {
        return NULL;
    }
}

static int cx_arl_insert_iter(
        struct cx_iterator_s *iter,
        const void *elem,
        int prepend
) {
    struct cx_list_s *list = iter->src_handle;
    if (iter->index < list->collection.size) {
        if (cx_arl_insert_element(list,
                iter->index + 1 - prepend, elem) == NULL) {
            return 1; // LCOV_EXCL_LINE
        }
        iter->elem_count++;
        if (prepend != 0) {
            iter->index++;
            iter->elem_handle = ((char *) iter->elem_handle) + list->collection.elem_size;
        }
        return 0;
    } else {
        if (cx_arl_insert_element(list, list->collection.size, elem) == NULL) {
            return 1;  // LCOV_EXCL_LINE
        }
        iter->elem_count++;
        iter->index = list->collection.size;
        return 0;
    }
}

static size_t cx_arl_remove(
        struct cx_list_s *list,
        size_t index,
        size_t num,
        void *targetbuf
) {
    cx_array_list *arl = (cx_array_list *) list;

    // out-of-bounds check
    size_t remove;
    if (index >= list->collection.size) {
        remove = 0;
    } else if (index + num > list->collection.size) {
        remove = list->collection.size - index;
    } else {
        remove = num;
    }

    // easy exit
    if (remove == 0) return 0;

    // destroy or copy contents
    if (targetbuf == NULL) {
        for (size_t idx = index; idx < index + remove; idx++) {
            cx_invoke_destructor(
                    list,
                    ((char *) arl->data) + idx * list->collection.elem_size
            );
        }
    } else {
        memcpy(
                targetbuf,
                ((char *) arl->data) + index * list->collection.elem_size,
                remove * list->collection.elem_size
        );
    }

    // calculate how many elements would need to be moved
    size_t remaining = list->collection.size - index - remove;

    // short-circuit removal of last elements
    if (remaining == 0) {
        list->collection.size -= remove;
        return remove;
    }

    // just move the elements to the left
    char *first_remaining = arl->data;
    first_remaining += (index + remove) * list->collection.elem_size;
    char *dst_move = arl->data;
    dst_move += index * list->collection.elem_size;
    memmove(dst_move, first_remaining, remaining * list->collection.elem_size);

    // decrease the size
    list->collection.size -= remove;

    return remove;
}

static void cx_arl_clear(struct cx_list_s *list) {
    if (list->collection.size == 0) return;

    cx_array_list *arl = (cx_array_list *) list;
    char *ptr = arl->data;

    if (list->collection.simple_destructor) {
        for (size_t i = 0; i < list->collection.size; i++) {
            cx_invoke_simple_destructor(list, ptr);
            ptr += list->collection.elem_size;
        }
    }
    if (list->collection.advanced_destructor) {
        for (size_t i = 0; i < list->collection.size; i++) {
            cx_invoke_advanced_destructor(list, ptr);
            ptr += list->collection.elem_size;
        }
    }

    memset(arl->data, 0, list->collection.size * list->collection.elem_size);
    list->collection.size = 0;
}

static int cx_arl_swap(
        struct cx_list_s *list,
        size_t i,
        size_t j
) {
    if (i >= list->collection.size || j >= list->collection.size) return 1;
    cx_array_list *arl = (cx_array_list *) list;
    cx_array_swap(arl->data, list->collection.elem_size, i, j);
    return 0;
}

static void *cx_arl_at(
        const struct cx_list_s *list,
        size_t index
) {
    if (index < list->collection.size) {
        const cx_array_list *arl = (const cx_array_list *) list;
        char *space = arl->data;
        return space + index * list->collection.elem_size;
    } else {
        return NULL;
    }
}

static size_t cx_arl_find_remove(
        struct cx_list_s *list,
        const void *elem,
        bool remove
) {
    assert(list != NULL);
    assert(list->collection.cmpfunc != NULL);
    if (list->collection.size == 0) return 0;
    char *cur = ((const cx_array_list *) list)->data;

    // optimize with binary search, when sorted
    if (list->collection.sorted) {
        size_t i = cx_array_binary_search(
            cur,
            list->collection.size,
            list->collection.elem_size,
            elem,
            list->collection.cmpfunc
        );
        if (remove && i < list->collection.size) {
            cx_arl_remove(list, i, 1, NULL);
        }
        return i;
    }

    // fallback: linear search
    for (size_t i = 0; i < list->collection.size; i++) {
        if (0 == list->collection.cmpfunc(elem, cur)) {
            if (remove) {
                cx_arl_remove(list, i, 1, NULL);
            }
            return i;
        }
        cur += list->collection.elem_size;
    }
    return list->collection.size;
}

static void cx_arl_sort(struct cx_list_s *list) {
    assert(list->collection.cmpfunc != NULL);
    qsort(((cx_array_list *) list)->data,
          list->collection.size,
          list->collection.elem_size,
          list->collection.cmpfunc
    );
}

static int cx_arl_compare(
        const struct cx_list_s *list,
        const struct cx_list_s *other
) {
    assert(list->collection.cmpfunc != NULL);
    if (list->collection.size == other->collection.size) {
        const char *left = ((const cx_array_list *) list)->data;
        const char *right = ((const cx_array_list *) other)->data;
        for (size_t i = 0; i < list->collection.size; i++) {
            int d = list->collection.cmpfunc(left, right);
            if (d != 0) {
                return d;
            }
            left += list->collection.elem_size;
            right += other->collection.elem_size;
        }
        return 0;
    } else {
        return list->collection.size < other->collection.size ? -1 : 1;
    }
}

static void cx_arl_reverse(struct cx_list_s *list) {
    if (list->collection.size < 2) return;
    void *data = ((const cx_array_list *) list)->data;
    size_t half = list->collection.size / 2;
    for (size_t i = 0; i < half; i++) {
        cx_array_swap(data, list->collection.elem_size, i, list->collection.size - 1 - i);
    }
}

static bool cx_arl_iter_valid(const void *it) {
    const struct cx_iterator_s *iter = it;
    const struct cx_list_s *list = iter->src_handle;
    return iter->index < list->collection.size;
}

static void *cx_arl_iter_current(const void *it) {
    const struct cx_iterator_s *iter = it;
    return iter->elem_handle;
}

static void cx_arl_iter_next(void *it) {
    struct cx_iterator_s *iter = it;
    if (iter->base.remove) {
        iter->base.remove = false;
        cx_arl_remove(iter->src_handle, iter->index, 1, NULL);
        iter->elem_count--;
    } else {
        iter->index++;
        iter->elem_handle =
                ((char *) iter->elem_handle)
                + ((const struct cx_list_s *) iter->src_handle)->collection.elem_size;
    }
}

static void cx_arl_iter_prev(void *it) {
    struct cx_iterator_s *iter = it;
    if (iter->base.remove) {
        iter->base.remove = false;
        cx_arl_remove(iter->src_handle, iter->index, 1, NULL);
        iter->elem_count--;
    }
    iter->index--;
    cx_array_list *list = iter->src_handle;
    if (iter->index < list->base.collection.size) {
        iter->elem_handle = ((char *) list->data)
                            + iter->index * list->base.collection.elem_size;
    }
}

static int cx_arl_change_capacity(
        struct cx_list_s *list,
        size_t new_capacity
) {
    cx_array_list *arl = (cx_array_list *)list;
    return cxReallocateArray(list->collection.allocator,
        &arl->data, new_capacity, list->collection.elem_size);
}

static struct cx_iterator_s cx_arl_iterator(
        const struct cx_list_s *list,
        size_t index,
        bool backwards
) {
    struct cx_iterator_s iter;

    iter.index = index;
    iter.src_handle = (void*)list;
    iter.elem_handle = cx_arl_at(list, index);
    iter.elem_size = list->collection.elem_size;
    iter.elem_count = list->collection.size;
    iter.base.valid = cx_arl_iter_valid;
    iter.base.current = cx_arl_iter_current;
    iter.base.next = backwards ? cx_arl_iter_prev : cx_arl_iter_next;
    iter.base.remove = false;
    iter.base.allow_remove = true;

    return iter;
}

static cx_list_class cx_array_list_class = {
        cx_arl_destructor,
        cx_arl_insert_element,
        cx_arl_insert_array,
        cx_arl_insert_sorted,
        cx_arl_insert_unique,
        cx_arl_insert_iter,
        cx_arl_remove,
        cx_arl_clear,
        cx_arl_swap,
        cx_arl_at,
        cx_arl_find_remove,
        cx_arl_sort,
        cx_arl_compare,
        cx_arl_reverse,
        cx_arl_change_capacity,
        cx_arl_iterator,
};

CxList *cxArrayListCreate(
        const CxAllocator *allocator,
        size_t elem_size,
        size_t initial_capacity
) {
    if (allocator == NULL) {
        allocator = cxDefaultAllocator;
    }

    cx_array_list *list = cxCalloc(allocator, 1, sizeof(cx_array_list));
    if (list == NULL) return NULL;
    cx_list_init((CxList*)list, &cx_array_list_class,
        allocator, elem_size);
    list->capacity = initial_capacity;

    // allocate the array after the real elem_size is known
    list->data = cxCalloc(allocator, initial_capacity,
        list->base.collection.elem_size);
    if (list->data == NULL) { // LCOV_EXCL_START
        cxFree(allocator, list);
        return NULL;
    } // LCOV_EXCL_STOP

    return (CxList *) list;
}

mercurial