buffer.hpp

Go to the documentation of this file.

#ifndef OSMIUM_MEMORY_BUFFER_HPP
#define OSMIUM_MEMORY_BUFFER_HPP
 
/*
 
This file is part of Osmium (https://osmcode.org/libosmium).
 
Copyright 2013-2021 Jochen Topf <jochen@topf.org> and others (see README).
 
Boost Software License - Version 1.0 - August 17th, 2003
 
Permission is hereby granted, free of charge, to any person or organization
obtaining a copy of the software and accompanying documentation covered by
this license (the "Software") to use, reproduce, display, distribute,
execute, and transmit the Software, and to prepare derivative works of the
Software, and to permit third-parties to whom the Software is furnished to
do so, all subject to the following:
 
The copyright notices in the Software and this entire statement, including
the above license grant, this restriction and the following disclaimer,
must be included in all copies of the Software, in whole or in part, and
all derivative works of the Software, unless such copies or derivative
works are solely in the form of machine-executable object code generated by
a source language processor.
 
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE, TITLE AND NON-INFRINGEMENT. IN NO EVENT
SHALL THE COPYRIGHT HOLDERS OR ANYONE DISTRIBUTING THE SOFTWARE BE LIABLE
FOR ANY DAMAGES OR OTHER LIABILITY, WHETHER IN CONTRACT, TORT OR OTHERWISE,
ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
DEALINGS IN THE SOFTWARE.
 
*/
 
#include <osmium/memory/item.hpp>
#include <osmium/memory/item_iterator.hpp>
#include <osmium/osm/entity.hpp>
#include <osmium/util/compatibility.hpp>
 
#include <algorithm>
#include <cassert>
#include <cstddef>
#include <cstring>
#include <functional>
#include <iterator>
#include <memory>
#include <stdexcept>
#include <utility>
 
namespace osmium {
 
    struct buffer_is_full : public std::runtime_error {
 
        buffer_is_full() :
            std::runtime_error{"Osmium buffer is full"} {
        }
 
    }; // struct buffer_is_full
 
    namespace memory {
 
        class Buffer {
 
        public:
 
            // This is needed so we can call std::back_inserter() on a Buffer.
            using value_type = Item;
 
            enum class auto_grow {
                no       = 0,
                yes      = 1,
                internal = 2
            }; // enum class auto_grow
 
        private:
 
            std::unique_ptr<Buffer> m_next_buffer;
            std::unique_ptr<unsigned char[]> m_memory{};
            unsigned char* m_data = nullptr;
            std::size_t m_capacity = 0;
            std::size_t m_written = 0;
            std::size_t m_committed = 0;
#ifndef NDEBUG
            uint8_t m_builder_count = 0;
#endif
            auto_grow m_auto_grow{auto_grow::no};
            std::function<void(Buffer&)> m_full;
 
            static std::size_t calculate_capacity(std::size_t capacity) noexcept {
                enum {
                    // The majority of all Nodes will fit into this size.
                    min_capacity = 64
                };
 
                if (capacity < min_capacity) {
                    return min_capacity;
                }
                return padded_length(capacity);
            }
 
            void grow_internal() {
                assert(m_data && "This must be a valid buffer");
                if (!m_memory) {
                    throw std::logic_error{"Can't grow Buffer if it doesn't use internal memory management."};
                }
 
                std::unique_ptr<Buffer> old{new Buffer{std::move(m_memory), m_capacity, m_committed}};
                m_memory = std::unique_ptr<unsigned char[]>{new unsigned char[m_capacity]};
                m_data = m_memory.get();
 
                m_written -= m_committed;
                std::copy_n(old->data() + m_committed, m_written, m_data);
                m_committed = 0;
 
                old->m_next_buffer = std::move(m_next_buffer);
                m_next_buffer = std::move(old);
            }
 
        public:
 
            Buffer() noexcept :
                m_next_buffer() {
            }
 
            explicit Buffer(unsigned char* data, std::size_t size) :
                m_next_buffer(),
                m_data(data),
                m_capacity(size),
                m_written(size),
                m_committed(size) {
                if (size % align_bytes != 0) {
                    throw std::invalid_argument{"buffer size needs to be multiple of alignment"};
                }
            }
 
            explicit Buffer(unsigned char* data, std::size_t capacity, std::size_t committed) :
                m_next_buffer(),
                m_data(data),
                m_capacity(capacity),
                m_written(committed),
                m_committed(committed) {
                if (capacity % align_bytes != 0) {
                    throw std::invalid_argument{"buffer capacity needs to be multiple of alignment"};
                }
                if (committed % align_bytes != 0) {
                    throw std::invalid_argument{"buffer parameter 'committed' needs to be multiple of alignment"};
                }
                if (committed > capacity) {
                    throw std::invalid_argument{"buffer parameter 'committed' can not be larger than capacity"};
                }
            }
 
            explicit Buffer(std::unique_ptr<unsigned char[]> data, std::size_t capacity, std::size_t committed) :
                m_next_buffer(),
                m_memory(std::move(data)),
                m_data(m_memory.get()),
                m_capacity(capacity),
                m_written(committed),
                m_committed(committed) {
                if (capacity % align_bytes != 0) {
                    throw std::invalid_argument{"buffer capacity needs to be multiple of alignment"};
                }
                if (committed % align_bytes != 0) {
                    throw std::invalid_argument{"buffer parameter 'committed' needs to be multiple of alignment"};
                }
                if (committed > capacity) {
                    throw std::invalid_argument{"buffer parameter 'committed' can not be larger than capacity"};
                }
            }
 
            explicit Buffer(std::size_t capacity, auto_grow auto_grow = auto_grow::yes) :
                m_next_buffer(),
                m_memory(new unsigned char[calculate_capacity(capacity)]),
                m_data(m_memory.get()),
                m_capacity(calculate_capacity(capacity)),
                m_auto_grow(auto_grow) {
            }
 
            // buffers can not be copied
            Buffer(const Buffer&) = delete;
            Buffer& operator=(const Buffer&) = delete;
 
            // buffers can be moved
            Buffer(Buffer&& other) noexcept :
                m_next_buffer(std::move(other.m_next_buffer)),
                m_memory(std::move(other.m_memory)),
                m_data(other.m_data),
                m_capacity(other.m_capacity),
                m_written(other.m_written),
                m_committed(other.m_committed),
#ifndef NDEBUG
                m_builder_count(other.m_builder_count),
#endif
                m_auto_grow(other.m_auto_grow),
                m_full(std::move(other.m_full)) {
                other.m_data = nullptr;
                other.m_capacity = 0;
                other.m_written = 0;
                other.m_committed = 0;
#ifndef NDEBUG
                other.m_builder_count = 0;
#endif
            }
 
            Buffer& operator=(Buffer&& other) noexcept {
                m_next_buffer = std::move(other.m_next_buffer);
                m_memory = std::move(other.m_memory);
                m_data = other.m_data;
                m_capacity = other.m_capacity;
                m_written = other.m_written;
                m_committed = other.m_committed;
#ifndef NDEBUG
                m_builder_count = other.m_builder_count;
#endif
                m_auto_grow = other.m_auto_grow;
                m_full = std::move(other.m_full);
                other.m_data = nullptr;
                other.m_capacity = 0;
                other.m_written = 0;
                other.m_committed = 0;
#ifndef NDEBUG
                other.m_builder_count = 0;
#endif
                return *this;
            }
 
            ~Buffer() noexcept = default;
 
#ifndef NDEBUG
            void increment_builder_count() noexcept {
                ++m_builder_count;
            }
 
            void decrement_builder_count() noexcept {
                assert(m_builder_count > 0);
                --m_builder_count;
            }
 
            uint8_t builder_count() const noexcept {
                return m_builder_count;
            }
#endif
 
            unsigned char* data() const noexcept {
                assert(m_data && "This must be a valid buffer");
                return m_data;
            }
 
            std::size_t capacity() const noexcept {
                return m_capacity;
            }
 
            std::size_t committed() const noexcept {
                return m_committed;
            }
 
            std::size_t written() const noexcept {
                return m_written;
            }
 
            bool is_aligned() const noexcept {
                assert(m_data && "This must be a valid buffer");
                return (m_written % align_bytes == 0) && (m_committed % align_bytes == 0);
            }
 
            OSMIUM_DEPRECATED void set_full_callback(const std::function<void(Buffer&)>& full) {
                assert(m_data && "This must be a valid buffer");
                m_full = full;
            }
 
            void grow(std::size_t size) {
                assert(m_data && "This must be a valid buffer");
                if (!m_memory) {
                    throw std::logic_error{"Can't grow Buffer if it doesn't use internal memory management."};
                }
                size = calculate_capacity(size);
                if (m_capacity < size) {
                    std::unique_ptr<unsigned char[]> memory{new unsigned char[size]};
                    std::copy_n(m_memory.get(), m_capacity, memory.get());
                    using std::swap;
                    swap(m_memory, memory);
                    m_data = m_memory.get();
                    m_capacity = size;
                }
            }
 
            bool has_nested_buffers() const noexcept {
                return m_next_buffer != nullptr;
            }
 
            std::unique_ptr<Buffer> get_last_nested() {
                assert(has_nested_buffers());
                Buffer* buffer = this;
                while (buffer->m_next_buffer->has_nested_buffers()) {
                    buffer = buffer->m_next_buffer.get();
                }
                return std::move(buffer->m_next_buffer);
            }
 
            std::size_t commit() {
                assert(m_data && "This must be a valid buffer");
                assert(m_builder_count == 0 && "Make sure there are no Builder objects still in scope");
                assert(is_aligned());
 
                const std::size_t offset = m_committed;
                m_committed = m_written;
                return offset;
            }
 
            void rollback() {
                assert(m_data && "This must be a valid buffer");
                assert(m_builder_count == 0 && "Make sure there are no Builder objects still in scope");
                m_written = m_committed;
            }
 
            std::size_t clear() {
                assert(m_builder_count == 0 && "Make sure there are no Builder objects still in scope");
                const std::size_t num_used_bytes = m_committed;
                m_written = 0;
                m_committed = 0;
                return num_used_bytes;
            }
 
            template <typename T>
            T& get(const std::size_t offset) const {
                assert(m_data && "This must be a valid buffer");
                assert(offset % alignof(T) == 0 && "Wrong alignment");
                return *reinterpret_cast<T*>(&m_data[offset]);
            }
 
            unsigned char* reserve_space(const std::size_t size) {
                assert(m_data && "This must be a valid buffer");
                // try to flush the buffer empty first.
                if (m_written + size > m_capacity && m_full) {
                    m_full(*this);
                }
                // if there's still not enough space, then try growing the buffer.
                if (m_written + size > m_capacity) {
                    if (!m_memory || m_auto_grow == auto_grow::no) {
                        throw osmium::buffer_is_full{};
                    }
                    if (m_auto_grow == auto_grow::internal && m_committed != 0) {
                        grow_internal();
                    }
                    if (m_written + size > m_capacity) {
                        // double buffer size until there is enough space
                        std::size_t new_capacity = m_capacity * 2;
                        while (m_written + size > new_capacity) {
                            new_capacity *= 2;
                        }
                        grow(new_capacity);
                    }
                }
                unsigned char* reserved_space = &m_data[m_written];
                m_written += size;
                return reserved_space;
            }
 
            template <typename T>
            T& add_item(const T& item) {
                assert(m_data && "This must be a valid buffer");
                unsigned char* target = reserve_space(item.padded_size());
                std::copy_n(reinterpret_cast<const unsigned char*>(&item), item.padded_size(), target);
                return *reinterpret_cast<T*>(target);
            }
 
            void add_buffer(const Buffer& buffer) {
                assert(m_data && "This must be a valid buffer");
                assert(buffer && "Buffer parameter must be a valid buffer");
                assert(m_builder_count == 0 && "Make sure there are no Builder objects still in scope");
                unsigned char* target = reserve_space(buffer.committed());
                std::copy_n(buffer.data(), buffer.committed(), target);
            }
 
            void push_back(const osmium::memory::Item& item) {
                assert(m_data && "This must be a valid buffer");
                assert(m_builder_count == 0 && "Make sure there are no Builder objects still in scope");
                add_item(item);
                commit();
            }
 
            template <typename T>
            using t_iterator = osmium::memory::ItemIterator<T>;
 
            template <typename T>
            using t_const_iterator = osmium::memory::ItemIterator<const T>;
 
            using iterator = t_iterator<osmium::OSMEntity>;
 
            using const_iterator = t_const_iterator<osmium::OSMEntity>;
 
            template <typename T>
            ItemIteratorRange<T> select() {
                return ItemIteratorRange<T>{m_data, m_data + m_committed};
            }
 
            template <typename T>
            ItemIteratorRange<const T> select() const {
                return ItemIteratorRange<const T>{m_data, m_data + m_committed};
            }
 
            template <typename T>
            t_iterator<T> begin() {
                assert(m_data && "This must be a valid buffer");
                return t_iterator<T>(m_data, m_data + m_committed);
            }
 
            iterator begin() {
                assert(m_data && "This must be a valid buffer");
                return {m_data, m_data + m_committed};
            }
 
            template <typename T>
            t_iterator<T> get_iterator(std::size_t offset) {
                assert(m_data && "This must be a valid buffer");
                assert(offset % alignof(T) == 0 && "Wrong alignment");
                return {m_data + offset, m_data + m_committed};
            }
 
            iterator get_iterator(std::size_t offset) {
                assert(m_data && "This must be a valid buffer");
                assert(offset % alignof(OSMEntity) == 0 && "Wrong alignment");
                return {m_data + offset, m_data + m_committed};
            }
 
            template <typename T>
            t_iterator<T> end() {
                assert(m_data && "This must be a valid buffer");
                return {m_data + m_committed, m_data + m_committed};
            }
 
            iterator end() {
                assert(m_data && "This must be a valid buffer");
                return {m_data + m_committed, m_data + m_committed};
            }
 
            template <typename T>
            t_const_iterator<T> cbegin() const {
                assert(m_data && "This must be a valid buffer");
                return {m_data, m_data + m_committed};
            }
 
            const_iterator cbegin() const {
                assert(m_data && "This must be a valid buffer");
                return {m_data, m_data + m_committed};
            }
 
            template <typename T>
            t_const_iterator<T> get_iterator(std::size_t offset) const {
                assert(m_data && "This must be a valid buffer");
                assert(offset % alignof(T) == 0 && "Wrong alignment");
                return {m_data + offset, m_data + m_committed};
            }
 
            const_iterator get_iterator(std::size_t offset) const {
                assert(m_data && "This must be a valid buffer");
                assert(offset % alignof(OSMEntity) == 0 && "Wrong alignment");
                return {m_data + offset, m_data + m_committed};
            }
 
            template <typename T>
            t_const_iterator<T> cend() const {
                assert(m_data && "This must be a valid buffer");
                return {m_data + m_committed, m_data + m_committed};
            }
 
            const_iterator cend() const {
                assert(m_data && "This must be a valid buffer");
                return {m_data + m_committed, m_data + m_committed};
            }
 
            template <typename T>
            t_const_iterator<T> begin() const {
                return cbegin<T>();
            }
 
            const_iterator begin() const {
                return cbegin();
            }
 
            template <typename T>
            t_const_iterator<T> end() const {
                return cend<T>();
            }
 
            const_iterator end() const {
                return cend();
            }
 
            explicit operator bool() const noexcept {
                return m_data != nullptr;
            }
 
            void swap(Buffer& other) {
                using std::swap;
 
                swap(m_next_buffer, other.m_next_buffer);
                swap(m_memory, other.m_memory);
                swap(m_data, other.m_data);
                swap(m_capacity, other.m_capacity);
                swap(m_written, other.m_written);
                swap(m_committed, other.m_committed);
                swap(m_auto_grow, other.m_auto_grow);
                swap(m_full, other.m_full);
            }
 
            template <typename TCallbackClass>
            void purge_removed(TCallbackClass* callback) {
                assert(m_data && "This must be a valid buffer");
                assert(callback);
 
                if (begin() == end()) {
                    return;
                }
 
                iterator it_write = begin();
 
                iterator next;
                for (iterator it_read = begin(); it_read != end(); it_read = next) {
                    next = std::next(it_read);
                    if (!it_read->removed()) {
                        if (it_read != it_write) {
                            assert(it_read.data() >= data());
                            assert(it_write.data() >= data());
                            const auto old_offset = static_cast<std::size_t>(it_read.data() - data());
                            const auto new_offset = static_cast<std::size_t>(it_write.data() - data());
                            callback->moving_in_buffer(old_offset, new_offset);
                            std::memmove(it_write.data(), it_read.data(), it_read->padded_size());
                        }
                        it_write.advance_once();
                    }
                }
 
                assert(it_write.data() >= data());
                m_written = static_cast<std::size_t>(it_write.data() - data());
                m_committed = m_written;
            }
 
            void purge_removed() {
                assert(m_data && "This must be a valid buffer");
                if (begin() == end()) {
                    return;
                }
 
                iterator it_write = begin();
 
                iterator next;
                for (iterator it_read = begin(); it_read != end(); it_read = next) {
                    next = std::next(it_read);
                    if (!it_read->removed()) {
                        if (it_read != it_write) {
                            assert(it_read.data() >= data());
                            assert(it_write.data() >= data());
                            std::memmove(it_write.data(), it_read.data(), it_read->padded_size());
                        }
                        it_write.advance_once();
                    }
                }
 
                assert(it_write.data() >= data());
                m_written = static_cast<std::size_t>(it_write.data() - data());
                m_committed = m_written;
            }
 
        }; // class Buffer
 
        inline void swap(Buffer& lhs, Buffer& rhs) {
            lhs.swap(rhs);
        }
 
        inline bool operator==(const Buffer& lhs, const Buffer& rhs) noexcept {
            if (!lhs || !rhs) {
                return !lhs && !rhs;
            }
            return lhs.data() == rhs.data() && lhs.capacity() == rhs.capacity() && lhs.committed() == rhs.committed();
        }
 
        inline bool operator!=(const Buffer& lhs, const Buffer& rhs) noexcept {
            return !(lhs == rhs);
        }
 
    } // namespace memory
 
} // namespace osmium
 
#endif // OSMIUM_MEMORY_BUFFER_HPP