fm_index.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 <sdsl/suffix_trees.hpp>
 
#include <seqan3/core/type_traits/range.hpp>
#include <seqan3/std/filesystem>
#include <seqan3/range/shortcuts.hpp>
#include <seqan3/range/views/join.hpp>
#include <seqan3/range/views/to_rank.hpp>
#include <seqan3/range/views/to.hpp>
#include <seqan3/search/fm_index/concept.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_cursor.hpp>
#include <seqan3/std/algorithm>
#include <seqan3/std/ranges>
 
namespace seqan3
{
 
// forward declarations
template <typename index_t>
class fm_index_cursor;
 
template <typename index_t>
class bi_fm_index_cursor;
 
using sdsl_wt_index_type =
    sdsl::csa_wt<sdsl::wt_blcd<sdsl::bit_vector,
                               sdsl::rank_support_v<>,
                               sdsl::select_support_scan<>,
                               sdsl::select_support_scan<0>>,
                 16,
                 10000000,
                 sdsl::sa_order_sa_sampling<>,
                 sdsl::isa_sampling<>,
                 sdsl::plain_byte_alphabet>;
 
using default_sdsl_index_type = sdsl_wt_index_type;
 
template <semialphabet alphabet_t,
          text_layout text_layout_mode_,
          detail::sdsl_index sdsl_index_type_ = default_sdsl_index_type>
class fm_index
{
private:
    using sdsl_index_type = sdsl_index_type_;
    using sdsl_char_type = typename sdsl_index_type::alphabet_type::char_type;
    using sdsl_sigma_type = typename sdsl_index_type::alphabet_type::sigma_type;
 
    sdsl_index_type index;
 
    sdsl::sd_vector<> text_begin;
    sdsl::select_support_sd<1> text_begin_ss;
    sdsl::rank_support_sd<1> text_begin_rs;
 
    template <std::ranges::range text_t>
        requires text_layout_mode_ == text_layout::single
    void construct(text_t && text)
    {
        static_assert(std::ranges::bidirectional_range<text_t>, "The text must model bidirectional_range.");
        static_assert(alphabet_size<innermost_value_type_t<text_t>> <= 256, "The alphabet is too big.");
        static_assert(std::convertible_to<innermost_value_type_t<text_t>, alphabet_t>,
                     "The alphabet of the text collection must be convertible to the alphabet of the index.");
        static_assert(dimension_v<text_t> == 1, "The input cannot be a text collection.");
 
        // text must not be empty
        if (std::ranges::begin(text) == std::ranges::end(text))
            throw std::invalid_argument("The text that is indexed cannot be empty.");
 
        constexpr auto sigma = alphabet_size<alphabet_t>;
 
        // TODO:
        // * check what happens in sdsl when constructed twice!
        // * choose between in-memory/external and construction algorithms
        // * sdsl construction currently only works for int_vector, std::string and char *, not ranges in general
        // uint8_t largest_char = 0;
        sdsl::int_vector<8> tmp_text(std::ranges::distance(text));
 
        std::ranges::move(text
                          | views::to_rank
                          | std::views::transform([] (uint8_t const r)
                          {
                              if constexpr (sigma == 256)
                              {
                                  if (r == 255)
                                      throw std::out_of_range("The input text cannot be indexed, because for full"
                                                              "character alphabets the last one/two values are reserved"
                                                              "(single sequence/collection).");
                              }
                              return r + 1;
                          })
                          | std::views::reverse,
                          std::ranges::begin(tmp_text)); // reverse and increase rank by one
 
        sdsl::construct_im(index, tmp_text, 0);
 
        // TODO: would be nice but doesn't work since it's private and the public member references are const
        // index.m_C.resize(largest_char);
        // index.m_C.shrink_to_fit();
        // index.m_sigma = largest_char;
    }
 
    template <std::ranges::range text_t>
        requires text_layout_mode_ == text_layout::collection
    void construct(text_t && text)
    {
        static_assert(std::ranges::bidirectional_range<text_t>, "The text collection must model bidirectional_range.");
        static_assert(std::ranges::bidirectional_range<reference_t<text_t>>,
                      "The elements of the text collection must model bidirectional_range.");
        static_assert(alphabet_size<innermost_value_type_t<text_t>> <= 256, "The alphabet is too big.");
        static_assert(std::convertible_to<innermost_value_type_t<text_t>, alphabet_t>,
                     "The alphabet of the text collection must be convertible to the alphabet of the index.");
        static_assert(dimension_v<text_t> == 2, "The input must be a text collection.");
 
        // text collection must not be empty
        if (std::ranges::begin(text) == std::ranges::end(text))
            throw std::invalid_argument("The text that is indexed cannot be empty.");
 
        size_t text_size{0}; // text size including delimiters
 
        // there must be at least one non-empty text in the collection
        bool all_empty = true;
 
        for (auto && t : text)
        {
            if (std::ranges::begin(t) != std::ranges::end(t))
            {
                all_empty = false;
            }
            text_size += 1 + std::ranges::distance(t); // text size and delimiter (sum will be 1 for empty texts)
        }
 
        if (all_empty)
            throw std::invalid_argument("A text collection that only contains empty texts cannot be indexed.");
 
        constexpr auto sigma = alphabet_size<alphabet_t>;
 
        // Instead of creating a bitvector of size `text_size`, setting the bits to 1 and then compressing it, we can
        // use the `sd_vector_builder(text_size, number_of_ones)` because we know the parameters and the 1s we want to
        // set are in a strictly increasing order. This inplace construction of the compressed vector saves memory.
        sdsl::sd_vector_builder builder(text_size, std::ranges::distance(text));
        size_t prefix_sum{0};
 
        for (auto && t : text)
        {
            builder.set(prefix_sum);
            prefix_sum += std::ranges::distance(t) + 1;
        }
 
        text_begin    = sdsl::sd_vector<>(builder);
        text_begin_ss = sdsl::select_support_sd<1>(&text_begin);
        text_begin_rs = sdsl::rank_support_sd<1>(&text_begin);
 
        // last text in collection needs no delimiter if we have more than one text in the collection
        sdsl::int_vector<8> tmp_text(text_size - (std::ranges::distance(text) > 1));
 
        constexpr uint8_t delimiter = sigma >= 255 ? 255 : sigma + 1;
 
 
        std::ranges::move(text
                          | views::deep{views::to_rank}
                          | views::deep
                          {
                              std::views::transform([] (uint8_t const r)
                              {
                                  if constexpr (sigma >= 255)
                                  {
                                      if (r >= 254)
                                          throw std::out_of_range("The input text cannot be indexed, because"
                                                                  " for full character alphabets the last one/"
                                                                  "two values are reserved (single sequence/"
                                                                  "collection).");
                                  }
                                  return r + 1;
                              })
                          }
                          | views::join(delimiter),
                          std::ranges::begin(tmp_text));
 
 
        // we need at least one delimiter
        if (std::ranges::distance(text) == 1)
            tmp_text.back() = delimiter;
 
        std::ranges::reverse(tmp_text);
 
        sdsl::construct_im(index, tmp_text, 0);
    }
 
public:
    static constexpr text_layout text_layout_mode = text_layout_mode_;
 
    using alphabet_type = alphabet_t;
    using size_type = typename sdsl_index_type::size_type;
    using cursor_type = fm_index_cursor<fm_index>;
 
    template <typename bi_fm_index_t>
    friend class bi_fm_index_cursor;
 
    template <typename fm_index_t>
    friend class fm_index_cursor;
 
    template <typename fm_index_t>
    friend class detail::fm_index_cursor_node;
 
    fm_index() = default;              
 
    fm_index(fm_index const & rhs) :
        index{rhs.index}, text_begin{rhs.text_begin}, text_begin_ss{rhs.text_begin_ss}, text_begin_rs{rhs.text_begin_rs}
    {
        text_begin_ss.set_vector(&text_begin);
        text_begin_rs.set_vector(&text_begin);
    }
 
    fm_index(fm_index && rhs) :
        index{std::move(rhs.index)}, text_begin{std::move(rhs.text_begin)},text_begin_ss{std::move(rhs.text_begin_ss)},
        text_begin_rs{std::move(rhs.text_begin_rs)}
    {
        text_begin_ss.set_vector(&text_begin);
        text_begin_rs.set_vector(&text_begin);
    }
 
    fm_index & operator=(fm_index rhs)
    {
        index = std::move(rhs.index);
        text_begin = std::move(rhs.text_begin);
        text_begin_ss = std::move(rhs.text_begin_ss);
        text_begin_rs = std::move(rhs.text_begin_rs);
 
        text_begin_ss.set_vector(&text_begin);
        text_begin_rs.set_vector(&text_begin);
 
        return *this;
    }
 
    ~fm_index() = default;             
 
    template <std::ranges::range text_t>
    fm_index(text_t && text)
    {
        construct(std::forward<text_t>(text));
    }
 
    size_type size() const noexcept
    {
        return index.size();
    }
 
    bool empty() const noexcept
    {
        return size() == 0;
    }
 
    bool operator==(fm_index const & rhs) const noexcept
    {
        // (void) rhs;
        return (index == rhs.index) && (text_begin == rhs.text_begin);
    }
 
    bool operator!=(fm_index const & rhs) const noexcept
    {
        return !(*this == rhs);
    }
 
    cursor_type begin() const noexcept
    {
        return {*this};
    }
 
    template <cereal_archive archive_t>
    void CEREAL_SERIALIZE_FUNCTION_NAME(archive_t & archive)
    {
        archive(index);
        archive(text_begin);
        archive(text_begin_ss);
        text_begin_ss.set_vector(&text_begin);
        archive(text_begin_rs);
        text_begin_rs.set_vector(&text_begin);
 
        auto sigma = alphabet_size<alphabet_t>;
        archive(sigma);
        if (sigma != alphabet_size<alphabet_t>)
        {
            throw std::logic_error{"The fm_index was built over an alphabet of size " + std::to_string(sigma) +
                                   " but it is being read into an fm_index with an alphabet of size " +
                                   std::to_string(alphabet_size<alphabet_t>) + "."};
        }
 
        bool tmp = text_layout_mode;
        archive(tmp);
        if (tmp != text_layout_mode)
        {
            throw std::logic_error{std::string{"The fm_index was built over a "} +
                                   (tmp ? "text collection" : "single text") +
                                   " but it is being read into an fm_index expecting a " +
                                   (text_layout_mode ? "text collection." : "single text.")};
        }
    }
 
};
 
template <std::ranges::range text_t>
fm_index(text_t &&) -> fm_index<innermost_value_type_t<text_t>, text_layout{dimension_v<text_t> != 1}>;
 
} // namespace seqan3