format_bam.hpp

Go to the documentation of this file.

// SPDX-FileCopyrightText: 2006-2025 Knut Reinert & Freie Universität Berlin
// SPDX-FileCopyrightText: 2016-2025 Knut Reinert & MPI für molekulare Genetik
// SPDX-License-Identifier: BSD-3-Clause
 
#pragma once
 
#include <bit>
#include <cstring>
#include <iterator>
#include <ranges>
#include <string>
#include <vector>
 
#include <seqan3/alphabet/nucleotide/dna16sam.hpp>
#include <seqan3/core/debug_stream/optional.hpp>
#include <seqan3/core/debug_stream/tuple.hpp>
#include <seqan3/io/sam_file/detail/cigar.hpp>
#include <seqan3/io/sam_file/detail/format_sam_base.hpp>
#include <seqan3/io/sam_file/header.hpp>
#include <seqan3/io/sam_file/input_options.hpp>
#include <seqan3/io/sam_file/sam_flag.hpp>
#include <seqan3/io/sam_file/sam_tag_dictionary.hpp>
#include <seqan3/io/stream/detail/fast_ostreambuf_iterator.hpp>
#include <seqan3/io/views/detail/istreambuf_view.hpp>
#include <seqan3/io/views/detail/take_exactly_view.hpp>
#include <seqan3/utility/views/slice.hpp>
 
namespace seqan3
{
 
class format_bam : private detail::format_sam_base
{
public:
    // string_buffer is of type std::string and has some problems with pre-C++11 ABI
    format_bam() = default;                               
    format_bam(format_bam const &) = default;             
    format_bam & operator=(format_bam const &) = default; 
    format_bam(format_bam &&) = default;                  
    format_bam & operator=(format_bam &&) = default;      
    ~format_bam() = default;                              
 
 
    static inline std::vector<std::string> file_extensions{{"bam"}};
 
protected:
    template <typename stream_type, // constraints checked by file
              typename seq_legal_alph_type,
              typename ref_seqs_type,
              typename ref_ids_type,
              typename stream_pos_type,
              typename seq_type,
              typename id_type,
              typename ref_seq_type,
              typename ref_id_type,
              typename ref_offset_type,
              typename cigar_type,
              typename flag_type,
              typename mapq_type,
              typename qual_type,
              typename mate_type,
              typename tag_dict_type,
              typename e_value_type,
              typename bit_score_type>
    void read_alignment_record(stream_type & stream,
                               sam_file_input_options<seq_legal_alph_type> const & SEQAN3_DOXYGEN_ONLY(options),
                               ref_seqs_type & ref_seqs,
                               sam_file_header<ref_ids_type> & header,
                               stream_pos_type & position_buffer,
                               seq_type & seq,
                               qual_type & qual,
                               id_type & id,
                               ref_seq_type & SEQAN3_DOXYGEN_ONLY(ref_seq),
                               ref_id_type & ref_id,
                               ref_offset_type & ref_offset,
                               cigar_type & cigar_vector,
                               flag_type & flag,
                               mapq_type & mapq,
                               mate_type & mate,
                               tag_dict_type & tag_dict,
                               e_value_type & SEQAN3_DOXYGEN_ONLY(e_value),
                               bit_score_type & SEQAN3_DOXYGEN_ONLY(bit_score));
 
    template <typename stream_type,
              typename header_type,
              typename seq_type,
              typename id_type,
              typename ref_seq_type,
              typename ref_id_type,
              typename cigar_type,
              typename qual_type,
              typename mate_type,
              typename tag_dict_type>
    void write_alignment_record([[maybe_unused]] stream_type & stream,
                                [[maybe_unused]] sam_file_output_options const & options,
                                [[maybe_unused]] header_type && header,
                                [[maybe_unused]] seq_type && seq,
                                [[maybe_unused]] qual_type && qual,
                                [[maybe_unused]] id_type && id,
                                [[maybe_unused]] ref_seq_type && SEQAN3_DOXYGEN_ONLY(ref_seq),
                                [[maybe_unused]] ref_id_type && ref_id,
                                [[maybe_unused]] std::optional<int32_t> ref_offset,
                                [[maybe_unused]] cigar_type && cigar_vector,
                                [[maybe_unused]] sam_flag const flag,
                                [[maybe_unused]] uint8_t const mapq,
                                [[maybe_unused]] mate_type && mate,
                                [[maybe_unused]] tag_dict_type && tag_dict,
                                [[maybe_unused]] double SEQAN3_DOXYGEN_ONLY(e_value),
                                [[maybe_unused]] double SEQAN3_DOXYGEN_ONLY(bit_score));
 
    template <typename stream_t, typename header_type>
    void write_header(stream_t & stream, sam_file_output_options const & options, header_type & header);
 
private:
    bool header_was_read{false};
 
    std::string string_buffer{};
 
    struct alignment_record_core
    {                             // naming corresponds to official SAM/BAM specifications
        int32_t block_size;       
        int32_t refID;            
        int32_t pos;              
        uint32_t l_read_name : 8; 
        uint32_t mapq : 8;        
        uint32_t bin : 16;        
        uint32_t n_cigar_op : 16; 
        sam_flag flag;            
        int32_t l_seq;            
        int32_t next_refID;       
        int32_t next_pos;         
        int32_t tlen;             
    };
 
    static_assert(sizeof(alignment_record_core) == 36);
 
    static constexpr std::array<uint8_t, 256> char_to_sam_rank{[]() constexpr
                                                               {
                                                                   std::array<uint8_t, 256> ret{};
 
                                                                   using index_t = std::make_unsigned_t<char>;
 
                                                                   // ret['M'] = 0; set anyway by initialization
                                                                   ret[static_cast<index_t>('I')] = 1;
                                                                   ret[static_cast<index_t>('D')] = 2;
                                                                   ret[static_cast<index_t>('N')] = 3;
                                                                   ret[static_cast<index_t>('S')] = 4;
                                                                   ret[static_cast<index_t>('H')] = 5;
                                                                   ret[static_cast<index_t>('P')] = 6;
                                                                   ret[static_cast<index_t>('=')] = 7;
                                                                   ret[static_cast<index_t>('X')] = 8;
 
                                                                   return ret;
                                                               }()};
 
    static uint16_t reg2bin(int32_t beg, int32_t end) noexcept
    {
        --end;
        if (beg >> 14 == end >> 14)
            return ((1 << 15) - 1) / 7 + (beg >> 14);
        if (beg >> 17 == end >> 17)
            return ((1 << 12) - 1) / 7 + (beg >> 17);
        if (beg >> 20 == end >> 20)
            return ((1 << 9) - 1) / 7 + (beg >> 20);
        if (beg >> 23 == end >> 23)
            return ((1 << 6) - 1) / 7 + (beg >> 23);
        if (beg >> 26 == end >> 26)
            return ((1 << 3) - 1) / 7 + (beg >> 26);
        return 0;
    }
 
    template <typename stream_view_type, std::integral number_type>
    void read_integral_byte_field(stream_view_type && stream_view, number_type & target)
    {
        std::ranges::copy_n(std::ranges::begin(stream_view), sizeof(target), reinterpret_cast<char *>(&target));
    }
 
    template <std::integral number_type>
    void read_integral_byte_field(std::string_view const str, number_type & target)
    {
        std::memcpy(&target, str.data(), sizeof(target));
    }
 
    template <typename stream_view_type>
    void read_float_byte_field(stream_view_type && stream_view, float & target)
    {
        std::ranges::copy_n(std::ranges::begin(stream_view), sizeof(int32_t), reinterpret_cast<char *>(&target));
    }
 
    template <typename value_type>
    int32_t read_sam_dict_vector(seqan3::detail::sam_tag_variant & variant,
                                 std::string_view const str,
                                 value_type const & SEQAN3_DOXYGEN_ONLY(value));
 
    void read_sam_dict(std::string_view const tag_str, sam_tag_dictionary & target);
 
    std::vector<cigar> parse_binary_cigar(std::string_view const cigar_str) const;
 
    static std::string get_tag_dict_str(sam_tag_dictionary const & tag_dict);
};
class format_bam : private detail::format_sam_base {…};
 
template <typename stream_type, // constraints checked by file
          typename seq_legal_alph_type,
          typename ref_seqs_type,
          typename ref_ids_type,
          typename stream_pos_type,
          typename seq_type,
          typename id_type,
          typename ref_seq_type,
          typename ref_id_type,
          typename ref_offset_type,
          typename cigar_type,
          typename flag_type,
          typename mapq_type,
          typename qual_type,
          typename mate_type,
          typename tag_dict_type,
          typename e_value_type,
          typename bit_score_type>
inline void
format_bam::read_alignment_record(stream_type & stream,
                                  sam_file_input_options<seq_legal_alph_type> const & SEQAN3_DOXYGEN_ONLY(options),
                                  ref_seqs_type & ref_seqs,
                                  sam_file_header<ref_ids_type> & header,
                                  stream_pos_type & position_buffer,
                                  seq_type & seq,
                                  qual_type & qual,
                                  id_type & id,
                                  ref_seq_type & SEQAN3_DOXYGEN_ONLY(ref_seq),
                                  ref_id_type & ref_id,
                                  ref_offset_type & ref_offset,
                                  cigar_type & cigar_vector,
                                  flag_type & flag,
                                  mapq_type & mapq,
                                  mate_type & mate,
                                  tag_dict_type & tag_dict,
                                  e_value_type & SEQAN3_DOXYGEN_ONLY(e_value),
                                  bit_score_type & SEQAN3_DOXYGEN_ONLY(bit_score))
{
    static_assert(detail::decays_to_ignore_v<ref_offset_type>
                      || detail::is_type_specialisation_of_v<ref_offset_type, std::optional>,
                  "The ref_offset must be a specialisation of std::optional.");
 
    static_assert(detail::decays_to_ignore_v<mapq_type> || std::same_as<mapq_type, uint8_t>,
                  "The type of field::mapq must be uint8_t.");
 
    static_assert(detail::decays_to_ignore_v<flag_type> || std::same_as<flag_type, sam_flag>,
                  "The type of field::flag must be seqan3::sam_flag.");
 
    auto stream_view = seqan3::detail::istreambuf(stream);
 
    // Header
    // -------------------------------------------------------------------------------------------------------------
    if (!header_was_read)
    {
        // magic BAM string
        if (!std::ranges::equal(stream_view | detail::take_exactly_or_throw(4), std::string_view{"BAM\1"}))
            throw format_error{"File is not in BAM format."};
 
        int32_t l_text{}; // length of header text including \0 character
        int32_t n_ref{};  // number of reference sequences
        int32_t l_name{}; // 1 + length of reference name including \0 character
        int32_t l_ref{};  // length of reference sequence
 
        read_integral_byte_field(stream_view, l_text);
 
        if (l_text > 0) // header text is present
            read_header(stream_view | detail::take_exactly_or_throw(l_text), header, ref_seqs);
 
        read_integral_byte_field(stream_view, n_ref);
 
        for (int32_t ref_idx = 0; ref_idx < n_ref; ++ref_idx)
        {
            read_integral_byte_field(stream_view, l_name);
 
            string_buffer.resize(l_name - 1);
            std::ranges::copy_n(std::ranges::begin(stream_view),
                                l_name - 1,
                                string_buffer.data()); // copy without \0 character
            ++std::ranges::begin(stream_view);         // skip \0 character
 
            read_integral_byte_field(stream_view, l_ref);
 
            if constexpr (detail::decays_to_ignore_v<ref_seqs_type>) // no reference information given
            {
                // If there was no header text, we parse reference sequences block as header information
                if (l_text == 0)
                {
                    auto & reference_ids = header.ref_ids();
                    // put the length of the reference sequence into ref_id_info
                    header.ref_id_info.emplace_back(l_ref, "");
                    // put the reference name into reference_ids
                    reference_ids.push_back(string_buffer);
                    // assign the reference name an ascending reference id (starts at index 0).
                    header.ref_dict.emplace(reference_ids.back(), reference_ids.size() - 1);
                    continue;
                }
            }
 
            auto id_it = header.ref_dict.find(string_buffer);
 
            // sanity checks of reference information to existing header object:
            if (id_it == header.ref_dict.end()) // [unlikely]
            {
                throw format_error{detail::to_string("Unknown reference name '" + string_buffer
                                                         + "' found in BAM file header (header.ref_ids():",
                                                     header.ref_ids(),
                                                     ").")};
            }
            else if (id_it->second != ref_idx) // [unlikely]
            {
                throw format_error{detail::to_string("Reference id '",
                                                     string_buffer,
                                                     "' at position ",
                                                     ref_idx,
                                                     " does not correspond to the position ",
                                                     id_it->second,
                                                     " in the header (header.ref_ids():",
                                                     header.ref_ids(),
                                                     ").")};
            }
            else if (std::get<0>(header.ref_id_info[id_it->second]) != l_ref) // [unlikely]
            {
                throw format_error{"Provided reference has unequal length as specified in the header."};
            }
        }
 
        header_was_read = true;
 
        if (std::ranges::begin(stream_view) == std::ranges::end(stream_view)) // no records follow
            return;
    }
 
    // read alignment record into buffer
    // -------------------------------------------------------------------------------------------------------------
    position_buffer = stream.tellg();
 
    auto stream_it = detail::fast_istreambuf_iterator{*stream.rdbuf()};
 
    alignment_record_core core;
    std::string_view const core_str = stream_it.cache_bytes(sizeof(core));
    std::ranges::copy(core_str, reinterpret_cast<char *>(&core));
 
    if (core.refID >= static_cast<int32_t>(header.ref_ids().size()) || core.refID < -1) // [[unlikely]]
    {
        throw format_error{detail::to_string("Reference id index '",
                                             core.refID,
                                             "' is not in range of ",
                                             "header.ref_ids(), which has size ",
                                             header.ref_ids().size(),
                                             ".")};
    }
    else if (core.refID > -1) // not unmapped
    {
        ref_id = core.refID; // field::ref_id
    }
 
    flag = core.flag;                       // field::flag
    mapq = static_cast<uint8_t>(core.mapq); // field::mapq
 
    if (core.pos > -1)         // [[likely]]
        ref_offset = core.pos; // field::ref_offset
 
    if constexpr (!detail::decays_to_ignore_v<mate_type>) // field::mate
    {
        if (core.next_refID > -1)
            get<0>(mate) = core.next_refID;
 
        if (core.next_pos > -1) // [[likely]]
            get<1>(mate) = core.next_pos;
 
        get<2>(mate) = core.tlen;
    }
 
    // read id
    // -------------------------------------------------------------------------------------------------------------
    std::string_view record_str = stream_it.cache_bytes(core.block_size - (sizeof(alignment_record_core) - 4));
    size_t considered_bytes{0};
 
    if constexpr (!detail::decays_to_ignore_v<id_type>)
        read_forward_range_field(record_str.substr(0, core.l_read_name - 1), id);
 
    considered_bytes += core.l_read_name;
 
    // read cigar string
    // -------------------------------------------------------------------------------------------------------------
    if constexpr (!detail::decays_to_ignore_v<cigar_type>)
        cigar_vector = parse_binary_cigar(record_str.substr(considered_bytes, core.n_cigar_op * 4));
 
    considered_bytes += core.n_cigar_op * 4;
 
    // read sequence
    // -------------------------------------------------------------------------------------------------------------
    if constexpr (!detail::decays_to_ignore_v<seq_type>)
    {
        size_t const number_of_bytes = (core.l_seq + 1) / 2;
        std::string_view const seq_str = record_str.substr(considered_bytes, number_of_bytes);
 
        seq.resize(
            core.l_seq
            + 1 /* reserve one more in case size is uneven. will be corrected */); // TODO: .resize() is not generic
 
        using alph_t = std::ranges::range_value_t<decltype(seq)>;
        constexpr auto from_dna16 = detail::convert_through_char_representation<dna16sam, alph_t>;
 
        // 1 byte encodes two sequence characters
        for (size_t i = 0, j = 0; i < number_of_bytes; ++i, j += 2)
        {
            seq[j] = from_dna16[to_rank(dna16sam{}.assign_rank(std::min(15, static_cast<uint8_t>(seq_str[i]) >> 4)))];
            seq[j + 1] =
                from_dna16[to_rank(dna16sam{}.assign_rank(std::min(15, static_cast<uint8_t>(seq_str[i]) & 0x0f)))];
        }
 
        seq.resize(core.l_seq); // remove extra letter
    }
 
    considered_bytes += (core.l_seq + 1) / 2;
 
    // read qual string
    // -------------------------------------------------------------------------------------------------------------
    if constexpr (!detail::decays_to_ignore_v<qual_type>)
    {
        std::string_view const qual_str = record_str.substr(considered_bytes, core.l_seq);
        qual.resize(core.l_seq); // TODO: this is not generic
 
        for (int32_t i = 0; i < core.l_seq; ++i)
            qual[i] = assign_char_to(static_cast<char>(qual_str[i] + 33), std::ranges::range_value_t<qual_type>{});
    }
 
    considered_bytes += core.l_seq;
 
    // All remaining optional fields if any: SAM tags dictionary
    // -------------------------------------------------------------------------------------------------------------
    if constexpr (!detail::decays_to_ignore_v<tag_dict_type>)
        read_sam_dict(record_str.substr(considered_bytes), tag_dict);
 
    // DONE READING - wrap up
    // -------------------------------------------------------------------------------------------------------------
    if constexpr (!detail::decays_to_ignore_v<cigar_type>)
    {
        int32_t const sc_front = soft_clipping_at_front(cigar_vector);
 
        // Check cigar, if it matches ‘kSmN’, where ‘k’ equals lseq, ‘m’ is the reference sequence length in the
        // alignment, and ‘S’ and ‘N’ are the soft-clipping and reference-clip, then the cigar string was larger
        // than 65535 operations and is stored in the sam_tag_dictionary (tag GC).
        if (core.l_seq != 0 && sc_front == core.l_seq)
        {
            if constexpr (detail::decays_to_ignore_v<tag_dict_type> | detail::decays_to_ignore_v<seq_type>)
            { // maybe only throw in debug mode and otherwise return an empty alignment?
                throw format_error{
                    detail::to_string("The cigar string '",
                                      detail::get_cigar_string(cigar_vector),
                                      "' suggests that the cigar string exceeded 65535 elements and was therefore ",
                                      "stored in the optional field CG. You need to read in the field::tags and "
                                      "field::seq in order to access this information.")};
            }
            else
            {
                auto it = tag_dict.find("CG"_tag);
 
                if (it == tag_dict.end())
                    throw format_error{
                        detail::to_string("The cigar string '",
                                          detail::get_cigar_string(cigar_vector),
                                          "' suggests that the cigar string exceeded 65535 elements and was therefore ",
                                          "stored in the optional field CG but this tag is not present in the given ",
                                          "record.")};
 
                cigar_vector = detail::parse_cigar(std::get<std::string>(it->second));
                tag_dict.erase(it); // remove redundant information
            }
        }
    }
}
format_bam::read_alignment_record(stream_type & stream, {…}
 
template <typename stream_type,
          typename header_type,
          typename seq_type,
          typename id_type,
          typename ref_seq_type,
          typename ref_id_type,
          typename cigar_type,
          typename qual_type,
          typename mate_type,
          typename tag_dict_type>
inline void format_bam::write_alignment_record([[maybe_unused]] stream_type & stream,
                                               [[maybe_unused]] sam_file_output_options const & options,
                                               [[maybe_unused]] header_type && header,
                                               [[maybe_unused]] seq_type && seq,
                                               [[maybe_unused]] qual_type && qual,
                                               [[maybe_unused]] id_type && id,
                                               [[maybe_unused]] ref_seq_type && SEQAN3_DOXYGEN_ONLY(ref_seq),
                                               [[maybe_unused]] ref_id_type && ref_id,
                                               [[maybe_unused]] std::optional<int32_t> ref_offset,
                                               [[maybe_unused]] cigar_type && cigar_vector,
                                               [[maybe_unused]] sam_flag const flag,
                                               [[maybe_unused]] uint8_t const mapq,
                                               [[maybe_unused]] mate_type && mate,
                                               [[maybe_unused]] tag_dict_type && tag_dict,
                                               [[maybe_unused]] double SEQAN3_DOXYGEN_ONLY(e_value),
                                               [[maybe_unused]] double SEQAN3_DOXYGEN_ONLY(bit_score))
{
    // ---------------------------------------------------------------------
    // Type Requirements (as static asserts for user friendliness)
    // ---------------------------------------------------------------------
    static_assert((std::ranges::forward_range<seq_type> && alphabet<std::ranges::range_reference_t<seq_type>>),
                  "The seq object must be a std::ranges::forward_range over "
                  "letters that model seqan3::alphabet.");
 
    static_assert((std::ranges::forward_range<id_type> && alphabet<std::ranges::range_reference_t<id_type>>),
                  "The id object must be a std::ranges::forward_range over "
                  "letters that model seqan3::alphabet.");
 
    static_assert((std::ranges::forward_range<ref_seq_type> && alphabet<std::ranges::range_reference_t<ref_seq_type>>),
                  "The ref_seq object must be a std::ranges::forward_range "
                  "over letters that model seqan3::alphabet.");
 
    if constexpr (!detail::decays_to_ignore_v<ref_id_type>)
    {
        static_assert((std::ranges::forward_range<ref_id_type> || std::integral<std::remove_reference_t<ref_id_type>>
                       || detail::is_type_specialisation_of_v<std::remove_cvref_t<ref_id_type>, std::optional>),
                      "The ref_id object must be a std::ranges::forward_range "
                      "over letters that model seqan3::alphabet or an integral or a std::optional<integral>.");
    }
 
    static_assert((std::ranges::forward_range<qual_type> && alphabet<std::ranges::range_reference_t<qual_type>>),
                  "The qual object must be a std::ranges::forward_range "
                  "over letters that model seqan3::alphabet.");
 
    static_assert(tuple_like<std::remove_cvref_t<mate_type>>,
                  "The mate object must be a std::tuple of size 3 with "
                  "1) a std::ranges::forward_range with a value_type modelling seqan3::alphabet, "
                  "2) a std::integral or std::optional<std::integral>, and "
                  "3) a std::integral.");
 
    static_assert(
        ((std::ranges::forward_range<decltype(std::get<0>(mate))>
          || std::integral<std::remove_cvref_t<decltype(std::get<0>(mate))>>
          || detail::is_type_specialisation_of_v<std::remove_cvref_t<decltype(std::get<0>(mate))>, std::optional>)
         && (std::integral<std::remove_cvref_t<decltype(std::get<1>(mate))>>
             || detail::is_type_specialisation_of_v<std::remove_cvref_t<decltype(std::get<1>(mate))>, std::optional>)
         && std::integral<std::remove_cvref_t<decltype(std::get<2>(mate))>>),
        "The mate object must be a std::tuple of size 3 with "
        "1) a std::ranges::forward_range with a value_type modelling seqan3::alphabet, "
        "2) a std::integral or std::optional<std::integral>, and "
        "3) a std::integral.");
 
    static_assert(std::same_as<std::remove_cvref_t<tag_dict_type>, sam_tag_dictionary>,
                  "The tag_dict object must be of type seqan3::sam_tag_dictionary.");
 
    if constexpr (detail::decays_to_ignore_v<header_type>)
    {
        throw format_error{"BAM can only be written with a header but you did not provide enough information! "
                           "You can either construct the output file with reference names and reference length "
                           "information and the header will be created for you, or you can access the `header` member "
                           "directly."};
    }
    else
    {
        // ---------------------------------------------------------------------
        // logical Requirements
        // ---------------------------------------------------------------------
 
        if (ref_offset.has_value() && (ref_offset.value() + 1) < 0)
            throw format_error{detail::to_string("The ref_offset object must be >= -1 but is: ", ref_offset)};
 
        detail::fast_ostreambuf_iterator stream_it{*stream.rdbuf()};
 
        // ---------------------------------------------------------------------
        // Writing the BAM Header on first call
        // ---------------------------------------------------------------------
        if (!header_was_written)
        {
            write_header(stream, options, header);
            header_was_written = true;
        }
 
        // ---------------------------------------------------------------------
        // Writing the Record
        // ---------------------------------------------------------------------
        int32_t ref_length{};
 
        // Compute the ref_length from given cigar_vector which is needed to fill field `bin`.
        if (!std::ranges::empty(cigar_vector))
        {
            int32_t dummy_seq_length{};
            for (auto & [count, operation] : cigar_vector)
                detail::update_alignment_lengths(ref_length, dummy_seq_length, operation.to_char(), count);
        }
 
        if (cigar_vector.size() >= (1 << 16)) // must be written into the sam tag CG
        {
            tag_dict["CG"_tag] = detail::get_cigar_string(cigar_vector);
            cigar_vector.resize(2);
            cigar_vector[0] = cigar{static_cast<uint32_t>(std::ranges::distance(seq)), 'S'_cigar_operation};
            cigar_vector[1] = cigar{static_cast<uint32_t>(ref_length), 'N'_cigar_operation};
        }
 
        std::string tag_dict_binary_str = get_tag_dict_str(tag_dict);
 
        // Compute the value for the l_read_name field for the bam record.
        // This value is stored including a trailing `0`, so at most 254 characters of the id can be stored, since
        // the data type to store the value is uint8_t and 255 is the maximal size.
        // If the id is empty a '*' is written instead, i.e. the written id is never an empty string and stores at least
        // 2 bytes.
        uint8_t read_name_size = std::min<uint8_t>(std::ranges::distance(id), 254) + 1;
        read_name_size += static_cast<uint8_t>(read_name_size == 1); // need size two since empty id is stored as '*'.
 
        alignment_record_core core{/* block_size  */ 0,  // will be initialised right after
                                   /* refID       */ -1, // will be initialised right after
                                   /* pos         */ ref_offset.value_or(-1),
                                   /* l_read_name */ read_name_size,
                                   /* mapq        */ mapq,
                                   /* bin         */ reg2bin(ref_offset.value_or(-1), ref_length),
                                   /* n_cigar_op  */ static_cast<uint16_t>(cigar_vector.size()),
                                   /* flag        */ flag,
                                   /* l_seq       */ static_cast<int32_t>(std::ranges::distance(seq)),
                                   /* next_refId  */ -1, // will be initialised right after
                                   /* next_pos    */ get<1>(mate).value_or(-1),
                                   /* tlen        */ get<2>(mate)};
 
        auto check_and_assign_id_to = [&header]([[maybe_unused]] auto & id_source, [[maybe_unused]] auto & id_target)
        {
            using id_t = std::remove_reference_t<decltype(id_source)>;
 
            if constexpr (!detail::decays_to_ignore_v<id_t>)
            {
                if constexpr (std::integral<id_t>)
                {
                    id_target = id_source;
                }
                else if constexpr (detail::is_type_specialisation_of_v<id_t, std::optional>)
                {
                    id_target = id_source.value_or(-1);
                }
                else
                {
                    if (!std::ranges::empty(id_source)) // otherwise default will remain (-1)
                    {
                        auto id_it = header.ref_dict.end();
 
                        if constexpr (std::ranges::contiguous_range<decltype(id_source)>
                                      && std::ranges::sized_range<decltype(id_source)>
                                      && std::ranges::borrowed_range<decltype(id_source)>)
                        {
                            id_it = header.ref_dict.find(
                                std::span{std::ranges::data(id_source), std::ranges::size(id_source)});
                        }
                        else
                        {
                            using header_ref_id_type = std::remove_reference_t<decltype(header.ref_ids()[0])>;
 
                            static_assert(
                                implicitly_convertible_to<decltype(id_source), header_ref_id_type>,
                                "The ref_id type is not convertible to the reference id information stored in the "
                                "reference dictionary of the header object.");
 
                            id_it = header.ref_dict.find(id_source);
                        }
 
                        if (id_it == header.ref_dict.end())
                        {
                            throw format_error{detail::to_string("Unknown reference name '",
                                                                 id_source,
                                                                 "' could "
                                                                 "not be found in BAM header ref_dict: ",
                                                                 header.ref_dict,
                                                                 ".")};
                        }
 
                        id_target = id_it->second;
                    }
                }
            }
        };
 
        // initialise core.refID
        check_and_assign_id_to(ref_id, core.refID);
 
        // initialise core.next_refID
        check_and_assign_id_to(get<0>(mate), core.next_refID);
 
        // initialise core.block_size
        core.block_size = sizeof(core) - 4 /*block_size excluded*/ + core.l_read_name + core.n_cigar_op * 4
                        +                        // each int32_t has 4 bytes
                          (core.l_seq + 1) / 2 + // bitcompressed seq
                          core.l_seq +           // quality string
                          tag_dict_binary_str.size();
 
        std::ranges::copy_n(reinterpret_cast<char *>(&core), sizeof(core), stream_it); // write core
 
        if (std::ranges::empty(id)) // empty id is represented as * for backward compatibility
            stream_it = '*';
        else
            std::ranges::copy_n(std::ranges::begin(id), core.l_read_name - 1, stream_it); // write read id
        stream_it = '\0';
 
        // write cigar
        for (auto [cigar_count, op] : cigar_vector)
        {
            cigar_count = cigar_count << 4;
            cigar_count |= static_cast<int32_t>(char_to_sam_rank[op.to_char()]);
            std::ranges::copy_n(reinterpret_cast<char *>(&cigar_count), 4, stream_it);
        }
 
        // write seq (bit-compressed: dna16sam characters go into one byte)
        using alph_t = std::ranges::range_value_t<seq_type>;
        constexpr auto to_dna16 = detail::convert_through_char_representation<alph_t, dna16sam>;
 
        auto sit = std::ranges::begin(seq);
        for (int32_t sidx = 0; sidx < ((core.l_seq & 1) ? core.l_seq - 1 : core.l_seq); ++sidx, ++sit)
        {
            uint8_t compressed_chr = to_rank(to_dna16[to_rank(*sit)]) << 4;
            ++sidx, ++sit;
            compressed_chr |= to_rank(to_dna16[to_rank(*sit)]);
            stream_it = static_cast<char>(compressed_chr);
        }
 
        if (core.l_seq & 1) // write one more
            stream_it = static_cast<char>(to_rank(to_dna16[to_rank(*sit)]) << 4);
 
        // write qual
        if (std::ranges::empty(qual))
        {
            auto v = views::repeat_n(static_cast<char>(255), core.l_seq);
            std::ranges::copy_n(v.begin(), core.l_seq, stream_it);
        }
        else
        {
            if (std::ranges::distance(qual) != core.l_seq)
                throw format_error{detail::to_string("Expected quality of same length as sequence with size ",
                                                     core.l_seq,
                                                     ". Got quality with size ",
                                                     std::ranges::distance(qual),
                                                     " instead.")};
 
            auto v = qual
                   | std::views::transform(
                         [](auto chr)
                         {
                             return static_cast<char>(to_rank(chr));
                         });
            std::ranges::copy_n(v.begin(), core.l_seq, stream_it);
        }
 
        // write optional fields
        stream << tag_dict_binary_str;
    } // if constexpr (!detail::decays_to_ignore_v<header_type>)
}
inline void format_bam::write_alignment_record([[maybe_unused]] stream_type & stream, {…}
 
template <typename stream_t, typename header_type>
inline void format_bam::write_header(stream_t & stream, sam_file_output_options const & options, header_type & header)
{
    if constexpr (detail::decays_to_ignore_v<header_type>)
    {
        throw format_error{"BAM can only be written with a header but you did not provide enough information! "
                           "You can either construct the output file with reference names and reference length "
                           "information and the header will be created for you, or you can access the `header` member "
                           "directly."};
    }
    else
    {
        detail::fast_ostreambuf_iterator stream_it{*stream.rdbuf()};
 
        std::ranges::copy_n("BAM\1", 4, stream_it); // Do not copy the null terminator
 
        // write SAM header to temporary stream first to query its size.
        std::ostringstream os;
        detail::format_sam_base::write_header(os, options, header);
#if SEQAN3_WORKAROUND_GCC_NO_CXX11_ABI
        int32_t const l_text{static_cast<int32_t>(os.str().size())};
#else
        int32_t const l_text{static_cast<int32_t>(os.view().size())};
#endif
        std::ranges::copy_n(reinterpret_cast<char const *>(&l_text), 4, stream_it); // write text length
 
#if SEQAN3_WORKAROUND_GCC_NO_CXX11_ABI
        auto header_view = os.str();
#else
        auto header_view = os.view();
#endif
        std::ranges::copy(header_view, stream_it);
 
        assert(header.ref_ids().size() < (1ull << 32));
        int32_t const n_ref{static_cast<int32_t>(header.ref_ids().size())};
        std::ranges::copy_n(reinterpret_cast<char const *>(&n_ref), 4, stream_it); // write number of references
 
        for (int32_t ridx = 0; ridx < n_ref; ++ridx)
        {
            assert(header.ref_ids()[ridx].size() + 1 < (1ull << 32));
            int32_t const l_name{static_cast<int32_t>(header.ref_ids()[ridx].size()) + 1}; // plus null character
            std::ranges::copy_n(reinterpret_cast<char const *>(&l_name), 4, stream_it);    // write l_name
            // write reference name:
            std::ranges::copy(header.ref_ids()[ridx], stream_it);
            stream_it = '\0'; // ++ is not necessary for ostream_iterator
            // write reference sequence length:
            std::ranges::copy_n(reinterpret_cast<char *>(&get<0>(header.ref_id_info[ridx])), 4, stream_it);
        }
    }
}
 
template <typename value_type>
inline int32_t format_bam::read_sam_dict_vector(seqan3::detail::sam_tag_variant & variant,
                                                std::string_view const str,
                                                value_type const & SEQAN3_DOXYGEN_ONLY(value))
{
    auto it = str.begin();
 
    // Read vector size from string_view and advance `it`.
    int32_t const vector_size = [&]()
    {
        int32_t size{};
        read_integral_byte_field(std::string_view{it, str.end()}, size);
        it += sizeof(size);
        return size;
    }();
 
    int32_t bytes_left{vector_size};
 
    std::vector<value_type> tmp_vector;
    tmp_vector.reserve(vector_size);
 
    value_type tmp{};
 
    while (bytes_left > 0)
    {
        if constexpr (std::integral<value_type>)
            read_integral_byte_field(std::string_view{it, str.end()}, tmp);
        else if constexpr (std::same_as<value_type, float>)
            read_float_byte_field(std::string_view{it, str.end()}, tmp);
        else
            static_assert(std::is_same_v<value_type, void>, "format_bam::read_sam_dict_vector: unsupported value_type");
 
        it += sizeof(tmp);
        tmp_vector.push_back(std::move(tmp));
        --bytes_left;
    }
 
    variant = std::move(tmp_vector);
 
    return vector_size;
}
 
inline void format_bam::read_sam_dict(std::string_view const tag_str, sam_tag_dictionary & target)
{
    /* Every BAM tag has the format "[TAG][TYPE_ID][VALUE]", where TAG is a two letter
       name tag which is converted to a unique integer identifier and TYPE_ID is one character in [A,i,Z,H,B,f]
       describing the type for the upcoming VALUES. If TYPE_ID=='B' it signals an array of
       VALUE's and the inner value type is identified by the next character, one of [cCsSiIf], followed
       by the length (int32_t) of the array, followed by the values.
    */
    auto it = tag_str.begin();
 
    // Deduces int_t from passed argument.
    auto parse_integer_into_target = [&]<std::integral int_t>(uint16_t const tag, int_t)
    {
        int_t tmp{};
        read_integral_byte_field(std::string_view{it, tag_str.end()}, tmp);
        target[tag] = static_cast<int32_t>(tmp); // readable sam format only allows int32_t
        it += sizeof(tmp);
    };
 
    // Deduces array_value_t from passed argument.
    auto parse_array_into_target = [&]<arithmetic array_value_t>(uint16_t const tag, array_value_t)
    {
        int32_t const count = read_sam_dict_vector(target[tag], std::string_view{it, tag_str.end()}, array_value_t{});
        it += sizeof(int32_t) /*length is stored within the vector*/ + sizeof(array_value_t) * count;
    };
 
    // Read uint16_t from string_view and advance `it`.
    auto parse_tag = [&]()
    {
        uint16_t tag = static_cast<uint16_t>(*it) << 8;
        ++it; // skip char read before
        tag |= static_cast<uint16_t>(*it);
        ++it; // skip char read before
        return tag;
    };
 
    while (it != tag_str.end())
    {
        uint16_t const tag = parse_tag();
 
        char const type_id{*it};
        ++it; // skip char read before
 
        switch (type_id)
        {
        case 'A': // char
        {
            target[tag] = *it;
            ++it; // skip char that has been read
            break;
        }
        // all integer sizes are possible
        case 'c': // int8_t
        {
            parse_integer_into_target(tag, int8_t{});
            break;
        }
        case 'C': // uint8_t
        {
            parse_integer_into_target(tag, uint8_t{});
            break;
        }
        case 's': // int16_t
        {
            parse_integer_into_target(tag, int16_t{});
            break;
        }
        case 'S': // uint16_t
        {
            parse_integer_into_target(tag, uint16_t{});
            break;
        }
        case 'i': // int32_t
        {
            parse_integer_into_target(tag, int32_t{});
            break;
        }
        case 'I': // uint32_t
        {
            parse_integer_into_target(tag, uint32_t{});
            break;
        }
        case 'f': // float
        {
            float tmp{};
            read_float_byte_field(std::string_view{it, tag_str.end()}, tmp);
            target[tag] = tmp;
            it += sizeof(float);
            break;
        }
        case 'Z': // string
        {
            std::string const v{static_cast<char const *>(it)}; // parses until '\0'
            it += v.size() + 1;
            target[tag] = std::move(v);
            break;
        }
        case 'H': // byte array, represented as null-terminated string; specification requires even number of bytes
        {
            std::string_view const str{static_cast<char const *>(it)}; // parses until '\0'
 
            std::vector<std::byte> tmp_vector{};
            // std::from_chars cannot directly parse into a std::byte
            uint8_t dummy_byte{};
 
            if (str.size() % 2 != 0)
                throw format_error{"[CORRUPTED BAM FILE]  Hexadecimal tag must have even number of digits."};
 
            // H encodes bytes in a hexadecimal format. Two hex values are stored for each byte as characters.
            // E.g., '1' and 'A' need one byte each and are read as `\x1A`, which is 27 in decimal.
            for (auto hex_begin = str.begin(), hex_end = str.begin() + 2; hex_begin != str.end();
                 hex_begin += 2, hex_end += 2)
            {
                auto res = std::from_chars(hex_begin, hex_end, dummy_byte, 16);
 
                if (res.ec == std::errc::invalid_argument)
                    throw format_error{std::string("[CORRUPTED BAM FILE] The string '")
                                       + std::string(hex_begin, hex_end) + "' could not be cast into type uint8_t."};
 
                if (res.ec == std::errc::result_out_of_range)
                    throw format_error{std::string("[CORRUPTED BAM FILE] Casting '") + std::string(str)
                                       + "' into type uint8_t would cause an overflow."};
 
                tmp_vector.push_back(std::byte{dummy_byte});
            }
 
            target[tag] = std::move(tmp_vector);
 
            it += str.size() + 1;
 
            break;
        }
        case 'B': // Array. Value type depends on second char [cCsSiIf]
        {
            char array_value_type_id = *it;
            ++it; // skip char read before
 
            switch (array_value_type_id)
            {
            case 'c': // int8_t
                parse_array_into_target(tag, int8_t{});
                break;
            case 'C': // uint8_t
                parse_array_into_target(tag, uint8_t{});
                break;
            case 's': // int16_t
                parse_array_into_target(tag, int16_t{});
                break;
            case 'S': // uint16_t
                parse_array_into_target(tag, uint16_t{});
                break;
            case 'i': // int32_t
                parse_array_into_target(tag, int32_t{});
                break;
            case 'I': // uint32_t
                parse_array_into_target(tag, uint32_t{});
                break;
            case 'f': // float
                parse_array_into_target(tag, float{});
                break;
            default:
                throw format_error{detail::to_string("The first character in the numerical id of a SAM tag ",
                                                     "must be one of [cCsSiIf] but '",
                                                     array_value_type_id,
                                                     "' was given.")};
            }
            break;
        }
        default:
            throw format_error{detail::to_string("The second character in the numerical id of a "
                                                 "SAM tag must be one of [A,i,Z,H,B,f] but '",
                                                 type_id,
                                                 "' was given.")};
        }
    }
}
 
inline std::vector<cigar> format_bam::parse_binary_cigar(std::string_view const cigar_str) const
{
    // The cigar operation is encoded in 4 bits.
    constexpr std::array<char, 16>
        cigar_operation_mapping{'M', 'I', 'D', 'N', 'S', 'H', 'P', '=', 'X', '*', '*', '*', '*', '*', '*', '*'};
    // The rightmost 4 bits encode the operation, the other bits encode the count.
    constexpr uint32_t cigar_operation_mask = 0x0f; // rightmost 4 bits are set to one
 
    std::vector<cigar> cigar_vector{};
    char operation{'\0'};
    uint32_t count{};
    uint32_t operation_and_count{}; // In BAM, operation and count values are stored within one 32 bit integer.
 
    assert(cigar_str.size() % 4 == 0); // One cigar letter is stored in 4 bytes (uint32_t).
 
    for (auto it = cigar_str.begin(); it != cigar_str.end(); it += sizeof(operation_and_count))
    {
        std::memcpy(&operation_and_count, it, sizeof(operation_and_count));
        operation = cigar_operation_mapping[operation_and_count & cigar_operation_mask];
        count = operation_and_count >> 4;
 
        cigar_vector.emplace_back(count, seqan3::assign_char_strictly_to(operation, cigar::operation{}));
    }
 
    return cigar_vector;
}
 
inline std::string format_bam::get_tag_dict_str(sam_tag_dictionary const & tag_dict)
{
    std::string result{};
 
    auto stream_variant_fn = [&result](auto && arg) // helper to print a std::variant
    {
        // T is either char, int32_t, float, std::string, or a std::vector<some int>
        using T = std::remove_cvref_t<decltype(arg)>;
 
        if constexpr (std::same_as<T, int32_t>)
        {
            // always choose the smallest possible representation [cCsSiI]
            size_t const absolute_arg = std::abs(arg);
            auto n = std::countr_zero(std::bit_ceil(absolute_arg + 1u) >> 1u) / 8u;
            bool const negative = arg < 0;
            n = n * n + 2 * negative; // for switch case order
 
            switch (n)
            {
            case 0:
            {
                result[result.size() - 1] = 'C';
                result.append(reinterpret_cast<char const *>(&arg), 1);
                break;
            }
            case 1:
            {
                result[result.size() - 1] = 'S';
                result.append(reinterpret_cast<char const *>(&arg), 2);
                break;
            }
            case 2:
            {
                result[result.size() - 1] = 'c';
                int8_t tmp = static_cast<int8_t>(arg);
                result.append(reinterpret_cast<char const *>(&tmp), 1);
                break;
            }
            case 3:
            {
                result[result.size() - 1] = 's';
                int16_t tmp = static_cast<int16_t>(arg);
                result.append(reinterpret_cast<char const *>(&tmp), 2);
                break;
            }
            default:
            {
                result.append(reinterpret_cast<char const *>(&arg), 4); // always i
                break;
            }
            }
        }
        else if constexpr (std::same_as<T, std::string>)
        {
            result.append(reinterpret_cast<char const *>(arg.data()), arg.size() + 1 /*+ null character*/);
        }
        else if constexpr (!std::ranges::range<T>) // char, float
        {
            result.append(reinterpret_cast<char const *>(&arg), sizeof(arg));
        }
        else // std::vector of some arithmetic_type type
        {
            int32_t sz{static_cast<int32_t>(arg.size())};
            result.append(reinterpret_cast<char *>(&sz), 4);
            result.append(reinterpret_cast<char const *>(arg.data()),
                          arg.size() * sizeof(std::ranges::range_value_t<T>));
        }
    };
 
    for (auto & [tag, variant] : tag_dict)
    {
        result.push_back(static_cast<char>(tag / 256));
        result.push_back(static_cast<char>(tag % 256));
 
        result.push_back(detail::sam_tag_type_char[variant.index()]);
 
        if (!is_char<'\0'>(detail::sam_tag_type_char_extra[variant.index()]))
            result.push_back(detail::sam_tag_type_char_extra[variant.index()]);
 
        std::visit(stream_variant_fn, variant);
    }
 
    return result;
}
 
} // namespace seqan3