fm_index_cursor.hpp

Go to the documentation of this file.

// -----------------------------------------------------------------------------------------------------
// Copyright (c) 2006-2020, Knut Reinert & Freie Universität Berlin
// Copyright (c) 2016-2020, Knut Reinert & MPI für molekulare Genetik
// This file may be used, modified and/or redistributed under the terms of the 3-clause BSD-License
// shipped with this file and also available at: https://github.com/seqan/seqan3/blob/master/LICENSE.md
// -----------------------------------------------------------------------------------------------------
 
#pragma once
 
#include <array>
#include <type_traits>
 
#include <sdsl/suffix_trees.hpp>
 
#include <range/v3/view/slice.hpp>
 
#include <seqan3/alphabet/concept.hpp>
#include <seqan3/core/type_traits/range.hpp>
#include <seqan3/range/views/join.hpp>
#include <seqan3/range/views/slice.hpp>
#include <seqan3/search/fm_index/detail/csa_alphabet_strategy.hpp>
#include <seqan3/search/fm_index/detail/fm_index_cursor.hpp>
#include <seqan3/search/fm_index/fm_index.hpp>
#include <seqan3/std/ranges>
 
namespace seqan3
{
 
template <typename index_t>
class fm_index_cursor
{
public:
 
    using index_type = index_t;
    using size_type = typename index_type::size_type;
 
private:
 
    using node_type = detail::fm_index_cursor_node<index_t>;
    using sdsl_char_type = typename index_type::sdsl_char_type;
    using sdsl_sigma_type = typename index_type::sdsl_sigma_type;
    using index_alphabet_type = typename index_t::alphabet_type;
 
    index_type const * index;
    size_type parent_lb;
    size_type parent_rb;
    node_type node;
    sdsl_sigma_type sigma;
 
    template <typename _index_t>
    friend class bi_fm_index_cursor;
 
    size_type offset() const noexcept
    {
        assert(index->index.size() > query_length());
        return index->index.size() - query_length() - 1; // since the string is reversed during construction
    }
 
    template <detail::sdsl_index csa_t>
    bool backward_search(csa_t const & csa, sdsl_char_type const c, size_type & l, size_type & r) const noexcept
    {
        assert(l <= r && r < csa.size());
 
        size_type _l, _r;
 
        size_type cc = c;
        if constexpr(!std::same_as<typename csa_t::alphabet_type, sdsl::plain_byte_alphabet>)
        {
            cc = csa.char2comp[c];
            if (cc == 0 && c > 0) // [[unlikely]]
                return false;
        }
 
        size_type const c_begin = csa.C[cc];
        if (l == 0 && r + 1 == csa.size()) // [[unlikely]]
        {
            _l = c_begin;
            _r = csa.C[cc + 1] - 1;
            // if we use not the plain_byte_alphabet, we could return always return true here
        }
        else
        {
            _l = c_begin + csa.bwt.rank(l, c);         // count c in bwt[0..l-1]
            _r = c_begin + csa.bwt.rank(r + 1, c) - 1; // count c in bwt[0..r]
        }
 
        if (_r >= _l)
        {
            r = _r;
            l = _l;
            assert(r + 1 - l >= 0);
            return true;
        }
        return false;
    }
 
public:
 
    //        Default construction is necessary to make this class semi-regular and e.g., to allow construction of
    //        std::array of iterators.
    fm_index_cursor() noexcept = default;                                    
    fm_index_cursor(fm_index_cursor const &) noexcept = default;             
    fm_index_cursor & operator=(fm_index_cursor const &) noexcept = default; 
    fm_index_cursor(fm_index_cursor &&) noexcept = default;                  
    fm_index_cursor & operator=(fm_index_cursor &&) noexcept = default;      
    ~fm_index_cursor() = default;                                            
 
    fm_index_cursor(index_t const & _index) noexcept : index(&_index), node({0, _index.index.size() - 1, 0, 0}),
                                                       sigma(_index.index.sigma - index_t::text_layout_mode)
    {}
    //\}
 
    bool operator==(fm_index_cursor const & rhs) const noexcept
    {
        assert(index != nullptr);
        assert(node != rhs.node || (query_length() == 0 || (parent_lb == rhs.parent_lb && parent_rb == rhs.parent_rb)));
 
        // position in the implicit suffix tree is defined by the SA interval and depth.
        // No need to compare parent intervals
        return node == rhs.node;
    }
 
    bool operator!=(fm_index_cursor const & rhs) const noexcept
    {
        assert(index != nullptr);
 
        return !(*this == rhs);
    }
 
    bool extend_right() noexcept
    {
        // TODO: specialize extend_right() and cycle_back() for EPR-dictionaries
        // store all cursors at once in a private std::array of cursors
        assert(index != nullptr);
 
        sdsl_char_type c = 1; // NOTE: start with 0 or 1 depending on implicit_sentintel
        size_type _lb = node.lb, _rb = node.rb;
        while (c < sigma && !backward_search(index->index, index->index.comp2char[c], _lb, _rb))
        {
            ++c;
        }
 
        if (c != sigma)
        {
            parent_lb = node.lb;
            parent_rb = node.rb;
            node = {_lb, _rb, node.depth + 1, c};
            return true;
        }
        return false;
    }
 
    template <typename char_t>
    bool extend_right(char_t const c) noexcept
    {
        static_assert(std::convertible_to<char_t, index_alphabet_type>,
                     "The character must be convertible to the alphabet of the index.");
 
        assert(index != nullptr);
 
        size_type _lb = node.lb, _rb = node.rb;
 
        sdsl_char_type c_char = to_rank(static_cast<index_alphabet_type>(c)) + 1;
 
        if (backward_search(index->index, c_char, _lb, _rb))
        {
            parent_lb = node.lb;
            parent_rb = node.rb;
            node = {_lb, _rb, node.depth + 1, c_char};
            return true;
        }
        return false;
    }
 
    template <std::ranges::range seq_t>
    bool extend_right(seq_t && seq) noexcept
    {
        static_assert(std::ranges::forward_range<seq_t>, "The query must model forward_range.");
        static_assert(std::convertible_to<innermost_value_type_t<seq_t>, index_alphabet_type>,
                     "The alphabet of the sequence must be convertible to the alphabet of the index.");
 
        assert(index != nullptr); // range must not be empty!
 
        size_type _lb = node.lb, _rb = node.rb;
        size_type new_parent_lb = parent_lb, new_parent_rb = parent_rb;
 
        sdsl_char_type c{};
        size_t len{0};
 
        for (auto it = std::ranges::begin(seq); it != std::ranges::end(seq); ++len, ++it)
        {
            c = to_rank(static_cast<index_alphabet_type>(*it)) + 1;
 
            new_parent_lb = _lb;
            new_parent_rb = _rb;
            if (!backward_search(index->index, c, _lb, _rb))
                return false;
        }
 
        parent_lb = new_parent_lb;
        parent_rb = new_parent_rb;
        node = {_lb, _rb, len + node.depth, c};
        return true;
    }
 
    bool cycle_back() noexcept
    {
        assert(index != nullptr && query_length() > 0);
        // parent_lb > parent_rb --> invalid interval
        assert(parent_lb <= parent_rb);
 
        sdsl_char_type c = node.last_char + 1;
        size_type _lb = parent_lb, _rb = parent_rb;
 
        while (c < sigma && !backward_search(index->index, index->index.comp2char[c], _lb, _rb))
        {
            ++c;
        }
 
        if (c != sigma) // Collection has additional sentinel as delimiter
        {
            node = {_lb, _rb, node.depth, c};
            return true;
        }
        return false;
    }
 
    size_type last_rank() const noexcept
    {
        // parent_lb > parent_rb --> invalid interval
        assert(index != nullptr && query_length() > 0 && parent_lb <= parent_rb);
 
        return index->index.comp2char[node.last_char] - 1; // text is not allowed to contain ranks of 0
    }
 
    size_type query_length() const noexcept
    {
        assert(index != nullptr);
        assert(node.depth != 0 || (node.lb == 0 && node.rb == index->size() - 1)); // depth == 0 -> root node
 
        return node.depth;
    }
 
    template <std::ranges::range text_t>
    auto path_label(text_t && text) const noexcept
        requires index_t::text_layout_mode == text_layout::single
    {
        static_assert(std::ranges::input_range<text_t>, "The text must model input_range.");
        static_assert(dimension_v<text_t> == 1, "The input cannot be a text collection.");
        static_assert(std::same_as<innermost_value_type_t<text_t>, index_alphabet_type>,
                      "The alphabet types of the given text and index differ.");
        assert(index != nullptr);
 
        size_type const query_begin = offset() - index->index[node.lb];
        return text | views::slice(query_begin, query_begin + query_length());
    }
 
    template <std::ranges::range text_t>
    auto path_label(text_t && text) const noexcept
        requires index_t::text_layout_mode == text_layout::collection
    {
        static_assert(std::ranges::input_range<text_t>, "The text collection must model input_range.");
        static_assert(dimension_v<text_t> == 2, "The input must be a text collection.");
        static_assert(std::same_as<innermost_value_type_t<text_t>, index_alphabet_type>,
                      "The alphabet types of the given text and index differ.");
        assert(index != nullptr);
 
        size_type const loc = offset() - index->index[node.lb];
        size_type const query_begin = loc - index->text_begin_rs.rank(loc + 1) + 1; // Substract delimiters
        return text | views::join | views::slice(query_begin, query_begin + query_length());
    }
 
    size_type count() const noexcept
    {
        assert(index != nullptr);
 
        return 1 + node.rb - node.lb;
    }
 
    std::vector<size_type> locate() const
        requires index_t::text_layout_mode == text_layout::single
    {
        assert(index != nullptr);
 
        std::vector<size_type> occ(count());
        for (size_type i = 0; i < occ.size(); ++i)
        {
            occ[i] = offset() - index->index[node.lb + i];
        }
        return occ;
    }
 
    std::vector<std::pair<size_type, size_type>> locate() const
        requires index_t::text_layout_mode == text_layout::collection
    {
        assert(index != nullptr);
 
        std::vector<std::pair<size_type, size_type>> occ;
        occ.reserve(count());
        for (size_type i = 0; i < count(); ++i)
        {
            size_type loc = offset() - index->index[node.lb + i];
            size_type sequence_rank = index->text_begin_rs.rank(loc + 1);
            size_type sequence_position = loc - index->text_begin_ss.select(sequence_rank);
            occ.emplace_back(sequence_rank - 1, sequence_position);
        }
        return occ;
    }
 
    auto lazy_locate() const
        requires index_t::text_layout_mode == text_layout::single
    {
        assert(index != nullptr);
 
        return std::views::iota(node.lb, node.lb + count())
               | std::views::transform([*this, _offset = offset()] (auto sa_pos) { return _offset - index->index[sa_pos]; });
    }
 
    auto lazy_locate() const
        requires index_t::text_layout_mode == text_layout::collection
    {
        assert(index != nullptr);
 
        return std::views::iota(node.lb, node.lb + count())
               | std::views::transform([*this, _offset = offset()] (auto sa_pos) { return _offset - index->index[sa_pos]; })
               | std::views::transform([*this] (auto loc)
               {
                   size_type sequence_rank = index->text_begin_rs.rank(loc + 1);
                   size_type sequence_position = loc - index->text_begin_ss.select(sequence_rank);
                   return std::make_pair(sequence_rank-1, sequence_position);
               });
    }
 
};
 
 
} // namespace seqan3