/src/duckdb/extension/parquet/column_reader.cpp

Source
#include "column_reader.hpp"

#include "duckdb/common/vector/flat_vector.hpp"
#include "duckdb/common/vector/constant_vector.hpp"

#include <algorithm>
#include <memory>
#include <sstream>
#include <stdexcept>
#include <utility>

#include "parquet_statistics.hpp"

#include "reader/boolean_column_reader.hpp"
#include "brotli/decode.h"
#include "reader/callback_column_reader.hpp"
#include "reader/interval_column_reader.hpp"
#include "lz4.hpp"
#include "miniz_wrapper.hpp"
#include "reader/null_column_reader.hpp"
#include "parquet_reader.hpp"
#include "parquet_timestamp.hpp"
#include "parquet_float16.hpp"
#include "snappy.h"
#include "reader/string_column_reader.hpp"
#include "reader/templated_column_reader.hpp"
#include "reader/uuid_column_reader.hpp"
#include "zstd.h"
#include "duckdb/common/helper.hpp"
#include "duckdb/common/numeric_utils.hpp"
#include "duckdb/storage/table/column_segment.hpp"
#include "parquet_crypto.hpp"
#include "decode_utils.hpp"
#include "duckdb/common/enums/vector_type.hpp"
#include "duckdb/common/limits.hpp"
#include "duckdb/common/string.hpp"
#include "duckdb/common/types/date.hpp"
#include "duckdb/common/types/datetime.hpp"
#include "duckdb/common/types/timestamp.hpp"
#include "duckdb/common/types/vector.hpp"
#include "duckdb/common/vector/unified_vector_format.hpp"
#include "duckdb/common/vector_size.hpp"
#include "parquet_decimal_utils.hpp"
#include "parquet_file_metadata_cache.hpp"
#include "parquet_rle_bp_decoder.hpp"
#include "thrift/protocol/TProtocol.h"
#include "thrift_tools.hpp"

namespace duckdb_apache {
namespace thrift {
class TBase;
} // namespace thrift
} // namespace duckdb_apache

namespace duckdb {
class Allocator;
struct hugeint_t;

using duckdb_parquet::CompressionCodec;
using duckdb_parquet::ConvertedType;
using duckdb_parquet::Encoding;
using duckdb_parquet::PageType;
using duckdb_parquet::Type;

const uint64_t ParquetDecodeUtils::BITPACK_MASKS[] = {0,
                                                      1,
                                                      3,
                                                      7,
                                                      15,
                                                      31,
                                                      63,
                                                      127,
                                                      255,
                                                      511,
                                                      1023,
                                                      2047,
                                                      4095,
                                                      8191,
                                                      16383,
                                                      32767,
                                                      65535,
                                                      131071,
                                                      262143,
                                                      524287,
                                                      1048575,
                                                      2097151,
                                                      4194303,
                                                      8388607,
                                                      16777215,
                                                      33554431,
                                                      67108863,
                                                      134217727,
                                                      268435455,
                                                      536870911,
                                                      1073741823,
                                                      2147483647,
                                                      4294967295,
                                                      8589934591,
                                                      17179869183,
                                                      34359738367,
                                                      68719476735,
                                                      137438953471,
                                                      274877906943,
                                                      549755813887,
                                                      1099511627775,
                                                      2199023255551,
                                                      4398046511103,
                                                      8796093022207,
                                                      17592186044415,
                                                      35184372088831,
                                                      70368744177663,
                                                      140737488355327,
                                                      281474976710655,
                                                      562949953421311,
                                                      1125899906842623,
                                                      2251799813685247,
                                                      4503599627370495,
                                                      9007199254740991,
                                                      18014398509481983,
                                                      36028797018963967,
                                                      72057594037927935,
                                                      144115188075855871,
                                                      288230376151711743,
                                                      576460752303423487,
                                                      1152921504606846975,
                                                      2305843009213693951,
                                                      4611686018427387903,
                                                      9223372036854775807,
                                                      18446744073709551615ULL};

const uint64_t ParquetDecodeUtils::BITPACK_MASKS_SIZE = sizeof(ParquetDecodeUtils::BITPACK_MASKS) / sizeof(uint64_t);

const uint8_t ParquetDecodeUtils::BITPACK_DLEN = 8;

ColumnReader::ColumnReader(const ParquetReader &reader, const ParquetColumnSchema &schema_p)
    : column_schema(schema_p), reader(reader), page_rows_available(0), dictionary_decoder(*this),
      delta_binary_packed_decoder(*this), rle_decoder(*this), delta_length_byte_array_decoder(*this),
      delta_byte_array_decoder(*this), byte_stream_split_decoder(*this), aad_crypto_metadata(reader.allocator) {
}

ColumnReader::~ColumnReader() {
}

Allocator &ColumnReader::GetAllocator() {
  return reader.allocator;
}

const ParquetReader &ColumnReader::Reader() {
  return reader;
}

void ColumnReader::RegisterPrefetch(ThriftFileTransport &transport, bool allow_merge) {
  if (chunk) {
    uint64_t size = chunk->meta_data.total_compressed_size;
    transport.RegisterPrefetch(FileOffset(), size, allow_merge);
  }
}

unique_ptr<BaseStatistics> ColumnReader::Stats(idx_t row_group_idx_p, const vector<ColumnChunk> &columns) {
  return Schema().Stats(*reader.GetFileMetadata(), reader.parquet_options, row_group_idx_p, columns);
}

uint64_t ColumnReader::TotalCompressedSize() {
  if (IsSkipped()) {
    return 0;
  }

  return chunk->meta_data.total_compressed_size;
}

// Note: It's not trivial to determine where all Column data is stored. Chunk->file_offset
// apparently is not the first page of the data. Therefore we determine the address of the first page by taking the
// minimum of all page offsets.
idx_t ColumnReader::FileOffset() const {
  if (IsSkipped()) {
    //! This column reader is skipped
    return 0;
  }
  auto min_offset = NumericLimits<idx_t>::Maximum();
  if (chunk->meta_data.__isset.dictionary_page_offset) {
    if (chunk->meta_data.dictionary_page_offset < 0) {
      throw InvalidInputException("Failed to read file \"%s\": metadata is corrupt. Column has invalid "
                                  "dictionary page offset (%lld)",
                                  reader.GetFileName(), chunk->meta_data.dictionary_page_offset);
    }
    min_offset = MinValue<idx_t>(min_offset, NumericCast<idx_t>(chunk->meta_data.dictionary_page_offset));
  }
  if (chunk->meta_data.__isset.index_page_offset) {
    if (chunk->meta_data.index_page_offset < 0) {
      throw InvalidInputException("Failed to read file \"%s\": metadata is corrupt. Column has invalid "
                                  "index page offset (%lld)",
                                  reader.GetFileName(), chunk->meta_data.index_page_offset);
    }
    min_offset = MinValue<idx_t>(min_offset, NumericCast<idx_t>(chunk->meta_data.index_page_offset));
  }
  if (chunk->meta_data.data_page_offset < 0) {
    throw InvalidInputException("Failed to read file \"%s\": metadata is corrupt. Column has invalid "
                                "data page offset (%lld)",
                                reader.GetFileName(), chunk->meta_data.data_page_offset);
  }
  min_offset = MinValue<idx_t>(min_offset, NumericCast<idx_t>(chunk->meta_data.data_page_offset));

  return min_offset;
}

idx_t ColumnReader::GroupRowsAvailable() {
  return group_rows_available;
}

bool ColumnReader::AllValuesAreNull() const {
  // for repeated columns the null_count/num_values statistics do not reliably indicate that every value is NULL
  // (num_values counts leaf slots, not rows), so we only trust this for non-repeated columns
  if (MaxRepeat() != 0 || !chunk || !chunk->__isset.meta_data) {
    return false;
  }
  auto &chunk_meta = chunk->meta_data;
  if (!chunk_meta.__isset.statistics || !chunk_meta.statistics.__isset.null_count) {
    return false;
  }
  return chunk_meta.statistics.null_count == chunk_meta.num_values;
}

void ColumnReader::PlainSkip(ByteBuffer &plain_data, uint8_t *defines, idx_t num_values) {
  throw NotImplementedException("PlainSkip not implemented");
}

void ColumnReader::Plain(ByteBuffer &plain_data, uint8_t *defines, idx_t num_values, // NOLINT
                         idx_t result_offset, Vector &result) {
  throw NotImplementedException("Plain not implemented");
}

void ColumnReader::Plain(shared_ptr<ResizeableBuffer> &plain_data, uint8_t *defines, idx_t num_values,
                         idx_t result_offset, Vector &result) {
  Plain(*plain_data, defines, num_values, result_offset, result);
}

void ColumnReader::PlainSelect(shared_ptr<ResizeableBuffer> &plain_data, uint8_t *defines, idx_t num_values,
                               Vector &result, const SelectionVector &sel, idx_t count) {
  throw NotImplementedException("PlainSelect not implemented");
}

void ColumnReader::InitializeRead(idx_t row_group_idx_p, const vector<ColumnChunk> &columns, TProtocol &protocol_p) {
  D_ASSERT(ColumnIndex() < columns.size());
  chunk = &columns[ColumnIndex()];
  protocol = &protocol_p;
  D_ASSERT(chunk);
  D_ASSERT(chunk->__isset.meta_data);

  if (chunk->__isset.file_path) {
    throw InvalidInputException("Failed to read file \"%s\": Only inlined data files are supported (no references)",
                                Reader().GetFileName());
  }

  if (chunk->meta_data.data_page_offset < 0) {
    throw InvalidInputException("Failed to read file \"%s\": metadata is corrupt. Column has invalid "
                                "data page offset (%lld)",
                                Reader().GetFileName(), chunk->meta_data.data_page_offset);
  }
  // ugh. sometimes there is an extra offset for the dict. sometimes it's wrong.
  chunk_read_offset = NumericCast<idx_t>(chunk->meta_data.data_page_offset);
  if (chunk->meta_data.__isset.dictionary_page_offset && chunk->meta_data.dictionary_page_offset >= 4) {
    // this assumes the data pages follow the dict pages directly.
    chunk_read_offset = NumericCast<idx_t>(chunk->meta_data.dictionary_page_offset);
  }
  group_rows_available = chunk->meta_data.num_values;
}

bool ColumnReader::PageIsFilteredOut(PageHeader &page_hdr, optional_ptr<const TableFilter> filter) {
  if (page_hdr.type != PageType::DATA_PAGE && page_hdr.type != PageType::DATA_PAGE_V2) {
    // we can only filter out data pages
    return false;
  }
  bool is_v1 = page_hdr.type == PageType::DATA_PAGE;
  auto &v1_header = page_hdr.data_page_header;
  auto &v2_header = page_hdr.data_page_header_v2;
  auto page_encoding = is_v1 ? v1_header.encoding : v2_header.encoding;

  if (page_encoding == Encoding::PLAIN_DICTIONARY || page_encoding == Encoding::RLE_DICTIONARY) {
    if (!dictionary_decoder.HasFilteredOutAllValues()) {
      return false;
    }
    encoding = ColumnEncoding::DICTIONARY;
    page_is_filtered_out = true;
  } else if (filter) {
    // try to use page statistics to skip this page if could.
    const duckdb_parquet::Statistics *page_stats = nullptr;
    if (is_v1 && v1_header.__isset.statistics) {
      page_stats = &v1_header.statistics;
    } else if (!is_v1 && v2_header.__isset.statistics) {
      page_stats = &v2_header.statistics;
    }

    if (!page_stats || !((page_stats->__isset.min_value || page_stats->__isset.min) &&
                         (page_stats->__isset.max_value || page_stats->__isset.max))) {
      return false;
    }
    auto stats =
        ParquetStatisticsUtils::TransformParquetStatistics(Type(), Schema(), *page_stats, /*can_have_nan=*/true);
    auto &expr_filter = filter->Cast<ExpressionFilter>();
    if (stats && expr_filter.CheckStatistics(*stats) == FilterPropagateResult::FILTER_ALWAYS_FALSE) {
      page_is_filtered_out = true;
    }
  }
  if (page_is_filtered_out) {
    // the page has been filtered out!
    // skip forward
    auto &trans = reinterpret_cast<ThriftFileTransport &>(*protocol->getTransport());
    trans.Skip(page_hdr.compressed_page_size);
    page_rows_available = is_v1 ? v1_header.num_values : v2_header.num_values;
  }

  return page_is_filtered_out;
}

void ColumnReader::ReadEncrypted(duckdb_apache::thrift::TBase &object) {
  aad_crypto_metadata.module = ParquetCrypto::GetModuleHeader(*chunk, aad_crypto_metadata.page_ordinal);
  aad_crypto_metadata.page_ordinal =
      ParquetCrypto::GetFinalPageOrdinal(*chunk, aad_crypto_metadata.module, aad_crypto_metadata.page_ordinal);
  reader.ReadEncrypted(object, *protocol, aad_crypto_metadata);
}

void ColumnReader::ReadDataEncrypted(const data_ptr_t buffer, const uint32_t buffer_size, PageType::type page_type) {
  aad_crypto_metadata.module = ParquetCrypto::GetModule(*chunk, page_type, aad_crypto_metadata.page_ordinal);
  aad_crypto_metadata.page_ordinal =
      ParquetCrypto::GetFinalPageOrdinal(*chunk, aad_crypto_metadata.module, aad_crypto_metadata.page_ordinal);
  reader.ReadDataEncrypted(*protocol, buffer, buffer_size, aad_crypto_metadata);
}

void ColumnReader::Read(PageHeader &page_hdr) {
  if (reader.parquet_options.encryption_config) {
    ReadEncrypted(page_hdr);
  } else {
    reader.Read(page_hdr, *protocol);
  }
}

void ColumnReader::ReadData(const data_ptr_t buffer, const uint32_t buffer_size, PageType::type page_type) {
  if (reader.parquet_options.encryption_config) {
    ReadDataEncrypted(buffer, buffer_size, page_type);
  } else {
    reader.ReadData(*protocol, buffer, buffer_size);
  }
}

void ColumnReader::PrepareRead(optional_ptr<const TableFilter> filter, optional_ptr<TableFilterState> filter_state,
                               idx_t rows_to_skip) {
  encoding = ColumnEncoding::INVALID;
  defined_decoder.reset();
  page_is_filtered_out = false;
  block.reset();
  PageHeader page_hdr;
  auto &trans = reinterpret_cast<ThriftFileTransport &>(*protocol->getTransport());

  if (trans.HasPrefetch()) {
    // Already has some data prefetched, let's not mess with it
    Read(page_hdr);
  } else {
    // No prefetch yet, prefetch the full header in one go (so thrift won't read byte-by-byte from storage)
    // 256 bytes should cover almost all headers (unless it's a V2 header with really LONG string statistics)
    static constexpr idx_t ASSUMED_HEADER_SIZE = 256;
    const auto prefetch_size = MinValue(trans.GetSize() - trans.GetLocation(), ASSUMED_HEADER_SIZE);
    trans.Prefetch(trans.GetLocation(), prefetch_size);
    Read(page_hdr);
    trans.ClearPrefetch();
  }
  // some basic sanity check
  if (page_hdr.compressed_page_size < 0 || page_hdr.uncompressed_page_size < 0) {
    throw InvalidInputException("Failed to read file \"%s\": Page sizes must be >= 0", Reader().GetFileName());
  }

  if (PageIsFilteredOut(page_hdr, filter)) {
    return;
  }

  if (rows_to_skip > 0 && (page_hdr.type == PageType::DATA_PAGE || page_hdr.type == PageType::DATA_PAGE_V2)) {
    bool is_v1 = page_hdr.type == PageType::DATA_PAGE;
    idx_t page_num_values =
        NumericCast<idx_t>(is_v1 ? page_hdr.data_page_header.num_values : page_hdr.data_page_header_v2.num_values);
    if (rows_to_skip >= page_num_values) {
      trans.Skip(page_hdr.compressed_page_size);
      page_is_filtered_out = true;
      page_rows_available = page_num_values;
      return;
    }
  }

  switch (page_hdr.type) {
  case PageType::DATA_PAGE_V2:
    PreparePageV2(page_hdr);
    PrepareDataPage(page_hdr);
    break;
  case PageType::DATA_PAGE:
    PreparePage(page_hdr);
    PrepareDataPage(page_hdr);
    break;
  case PageType::DICTIONARY_PAGE: {
    PreparePage(page_hdr);
    auto dictionary_size = page_hdr.dictionary_page_header.num_values;
    if (dictionary_size < 0) {
      throw InvalidInputException("Failed to read file \"%s\": Invalid dictionary page header (num_values < 0)",
                                  Reader().GetFileName());
    }
    dictionary_decoder.InitializeDictionary(dictionary_size, filter, filter_state, HasDefines());
    break;
  }
  default:
    break; // ignore INDEX page type and any other custom extensions
  }
  ResetPage();
}

void ColumnReader::ResetPage() {
}

void ColumnReader::PreparePageV2(PageHeader &page_hdr) {
  D_ASSERT(page_hdr.type == PageType::DATA_PAGE_V2);

  AllocateBlock(page_hdr.uncompressed_page_size + 1);
  bool uncompressed = false;
  if (page_hdr.data_page_header_v2.__isset.is_compressed && !page_hdr.data_page_header_v2.is_compressed) {
    uncompressed = true;
  }
  if (chunk->meta_data.codec == CompressionCodec::UNCOMPRESSED) {
    if (page_hdr.compressed_page_size != page_hdr.uncompressed_page_size) {
      const auto &file_name = Reader().GetFileName();
      throw InvalidInputException(
          "Parquet file (%s) corrupted: uncompressed page size mismatch (expected %d, actual: %d)", file_name,
          page_hdr.uncompressed_page_size, page_hdr.compressed_page_size);
    }
    uncompressed = true;
  }
  if (uncompressed) {
    ReadData(block->ptr, page_hdr.compressed_page_size, page_hdr.type);
    return;
  }

  // copy repeats & defines as-is because FOR SOME REASON they are uncompressed.
  // the page sizes are already validated >= 0 by the caller, but the level lengths are not, so guard
  // them here. with all four i32 header fields non-negative, the sum below cannot overflow once widened
  // to uint64_t, and the uint64_t casts in the comparisons below are safe.
  if (page_hdr.data_page_header_v2.repetition_levels_byte_length < 0 ||
      page_hdr.data_page_header_v2.definition_levels_byte_length < 0) {
    throw InvalidInputException(
        "Failed to read file \"%s\": header inconsistency, repetition_levels_byte_length and "
        "definition_levels_byte_length must be >= 0",
        Reader().GetFileName());
  }
  uint64_t uncompressed_bytes = static_cast<uint64_t>(page_hdr.data_page_header_v2.repetition_levels_byte_length) +
                                page_hdr.data_page_header_v2.definition_levels_byte_length;
  if (uncompressed_bytes > static_cast<uint64_t>(page_hdr.uncompressed_page_size)) {
    throw InvalidInputException(
        "Failed to read file \"%s\": header inconsistency, uncompressed_page_size needs to be larger than "
        "repetition_levels_byte_length + definition_levels_byte_length",
        Reader().GetFileName());
  }
  if (static_cast<uint64_t>(page_hdr.compressed_page_size) < uncompressed_bytes) {
    throw InvalidInputException(
        "Failed to read file \"%s\": header inconsistency, compressed_page_size is smaller than "
        "repetition_levels_byte_length + definition_levels_byte_length",
        Reader().GetFileName());
  }

  ReadData(block->ptr, uncompressed_bytes, page_hdr.type);

  auto compressed_bytes = page_hdr.compressed_page_size - uncompressed_bytes;

  if (compressed_bytes == 0 && static_cast<uint64_t>(page_hdr.uncompressed_page_size) > uncompressed_bytes) {
    throw InvalidInputException(
        "Failed to read file \"%s\": header inconsistency, compressed_page_size is too small for the "
        "declared value region",
        Reader().GetFileName());
  }

  if (compressed_bytes > 0) {
    ResizeableBuffer compressed_buffer;
    compressed_buffer.resize(GetAllocator(), compressed_bytes);

    ReadData(compressed_buffer.ptr, compressed_bytes, page_hdr.type);

    DecompressInternal(chunk->meta_data.codec, compressed_buffer.ptr, compressed_bytes,
                       block->ptr + uncompressed_bytes, page_hdr.uncompressed_page_size - uncompressed_bytes);
  }
}

void ColumnReader::AllocateBlock(idx_t size) {
  if (!block) {
    block = make_shared_ptr<ResizeableBuffer>(GetAllocator(), size);
  } else {
    block->resize(GetAllocator(), size);
  }
}

void ColumnReader::PreparePage(PageHeader &page_hdr) {
  AllocateBlock(page_hdr.uncompressed_page_size + 1);
  uint32_t compressed_page_size = page_hdr.compressed_page_size;

  if (chunk->__isset.crypto_metadata) {
    auto const file_aad = reader.GetUniqueFileIdentifier(reader.metadata->crypto_metadata->encryption_algorithm);
    if (!file_aad.empty()) {
      // If there is a file aad (identifier), this means that the Encrypted file is written by Arrow
      // Arrow adds the bytes for encryption (len + nonce + tag)
      // to the compressed page size
      compressed_page_size -=
          (ParquetCrypto::LENGTH_BYTES + ParquetCrypto::NONCE_BYTES + ParquetCrypto::TAG_BYTES);
    }
  }

  if (chunk->meta_data.codec == CompressionCodec::UNCOMPRESSED) {
    if (compressed_page_size != NumericCast<uint32_t>(page_hdr.uncompressed_page_size)) {
      const auto &file_name = Reader().GetFileName();
      throw InvalidInputException(
          "Parquet file (%s) corrupted: uncompressed page size mismatch (expected %d, actual: %d)", file_name,
          page_hdr.uncompressed_page_size, compressed_page_size);
    }
    ReadData(block->ptr, compressed_page_size, page_hdr.type);
    return;
  }

  ResizeableBuffer compressed_buffer;
  compressed_buffer.resize(GetAllocator(), compressed_page_size + 1);
  ReadData(compressed_buffer.ptr, compressed_page_size, page_hdr.type);

  DecompressInternal(chunk->meta_data.codec, compressed_buffer.ptr, compressed_page_size, block->ptr,
                     page_hdr.uncompressed_page_size);
}

void ColumnReader::DecompressInternal(CompressionCodec::type codec, const_data_ptr_t src, idx_t src_size,
                                      data_ptr_t dst, idx_t dst_size) {
  switch (codec) {
  case CompressionCodec::UNCOMPRESSED:
    throw InternalException("Parquet data unexpectedly uncompressed");
  case CompressionCodec::GZIP: {
    MiniZStream s;
    s.Decompress(const_char_ptr_cast(src), src_size, char_ptr_cast(dst), dst_size);
    break;
  }
  case CompressionCodec::LZ4_RAW: {
    auto res =
        duckdb_lz4::LZ4_decompress_safe(const_char_ptr_cast(src), char_ptr_cast(dst),
                                        UnsafeNumericCast<int32_t>(src_size), UnsafeNumericCast<int32_t>(dst_size));
    if (res != NumericCast<int>(dst_size)) {
      throw InvalidInputException("Failed to read file \"%s\": LZ4 decompression failure",
                                  Reader().GetFileName());
    }
    break;
  }
  case CompressionCodec::SNAPPY: {
    {
      size_t uncompressed_size = 0;
      auto res = duckdb_snappy::GetUncompressedLength(const_char_ptr_cast(src), src_size, &uncompressed_size);
      if (!res) {
        throw InvalidInputException("Failed to read file \"%s\": Snappy decompression failure",
                                    Reader().GetFileName());
      }
      if (uncompressed_size != dst_size) {
        throw InvalidInputException(
            "Failed to read file \"%s\": Snappy decompression failure: Uncompressed data size mismatch",
            Reader().GetFileName());
      }
    }
    auto res = duckdb_snappy::RawUncompress(const_char_ptr_cast(src), src_size, char_ptr_cast(dst));
    if (!res) {
      throw InvalidInputException("Failed to read file \"%s\": Snappy decompression failure",
                                  Reader().GetFileName());
    }
    break;
  }
  case CompressionCodec::ZSTD: {
    auto res = duckdb_zstd::ZSTD_decompress(dst, dst_size, src, src_size);
    if (duckdb_zstd::ZSTD_isError(res) || res != dst_size) {
      throw InvalidInputException("Failed to read file \"%s\": ZSTD Decompression failure",
                                  Reader().GetFileName());
    }
    break;
  }
  case CompressionCodec::BROTLI: {
    auto state = duckdb_brotli::BrotliDecoderCreateInstance(nullptr, nullptr, nullptr);
    size_t total_out = 0;
    auto src_size_size_t = NumericCast<size_t>(src_size);
    auto dst_size_size_t = NumericCast<size_t>(dst_size);

    auto res = duckdb_brotli::BrotliDecoderDecompressStream(state, &src_size_size_t, &src, &dst_size_size_t, &dst,
                                                            &total_out);
    if (res != duckdb_brotli::BROTLI_DECODER_RESULT_SUCCESS) {
      throw InvalidInputException("Failed to read file \"%s\": Brotli Decompression failure",
                                  Reader().GetFileName());
    }
    duckdb_brotli::BrotliDecoderDestroyInstance(state);
    break;
  }

  default: {
    duckdb::stringstream codec_name;
    codec_name << codec;
    throw InvalidInputException("Failed to read file \"%s\": Unsupported compression codec \"%s\". Supported "
                                "options are uncompressed, brotli, gzip, lz4_raw, snappy or zstd",
                                Reader().GetFileName(), codec_name.str());
  }
  }
}

void ColumnReader::PrepareDataPage(PageHeader &page_hdr) {
  if (page_hdr.type == PageType::DATA_PAGE && !page_hdr.__isset.data_page_header) {
    throw InvalidInputException("Failed to read file \"%s\": Missing data page header from data page",
                                Reader().GetFileName());
  }
  if (page_hdr.type == PageType::DATA_PAGE_V2 && !page_hdr.__isset.data_page_header_v2) {
    throw InvalidInputException("Failed to read file \"%s\": Missing data page header from data page v2",
                                Reader().GetFileName());
  }

  bool is_v1 = page_hdr.type == PageType::DATA_PAGE;
  bool is_v2 = page_hdr.type == PageType::DATA_PAGE_V2;
  auto &v1_header = page_hdr.data_page_header;
  auto &v2_header = page_hdr.data_page_header_v2;

  page_rows_available = is_v1 ? v1_header.num_values : v2_header.num_values;
  auto page_encoding = is_v1 ? v1_header.encoding : v2_header.encoding;

  if (HasRepeats()) {
    uint32_t rep_length = is_v1 ? block->read<uint32_t>() : v2_header.repetition_levels_byte_length;
    block->available(rep_length);
    repeated_decoder =
        make_uniq<RleBpDecoder>(block->ptr, rep_length, RleBpDecoder::ComputeBitWidthFromMaxValue(MaxRepeat()));
    block->inc(rep_length);
  } else if (is_v2 && v2_header.repetition_levels_byte_length > 0) {
    block->inc(v2_header.repetition_levels_byte_length);
  }

  if (HasDefines()) {
    uint32_t def_length = is_v1 ? block->read<uint32_t>() : v2_header.definition_levels_byte_length;
    block->available(def_length);
    defined_decoder =
        make_uniq<RleBpDecoder>(block->ptr, def_length, RleBpDecoder::ComputeBitWidthFromMaxValue(MaxDefine()));
    block->inc(def_length);
  } else if (is_v2 && v2_header.definition_levels_byte_length > 0) {
    block->inc(v2_header.definition_levels_byte_length);
  }

  switch (page_encoding) {
  case Encoding::RLE_DICTIONARY:
  case Encoding::PLAIN_DICTIONARY: {
    encoding = ColumnEncoding::DICTIONARY;
    dictionary_decoder.InitializePage();
    break;
  }
  case Encoding::RLE: {
    encoding = ColumnEncoding::RLE;
    rle_decoder.InitializePage();
    break;
  }
  case Encoding::DELTA_BINARY_PACKED: {
    encoding = ColumnEncoding::DELTA_BINARY_PACKED;
    delta_binary_packed_decoder.InitializePage();
    break;
  }
  case Encoding::DELTA_LENGTH_BYTE_ARRAY: {
    encoding = ColumnEncoding::DELTA_LENGTH_BYTE_ARRAY;
    delta_length_byte_array_decoder.InitializePage();
    break;
  }
  case Encoding::DELTA_BYTE_ARRAY: {
    encoding = ColumnEncoding::DELTA_BYTE_ARRAY;
    delta_byte_array_decoder.InitializePage();
    break;
  }
  case Encoding::BYTE_STREAM_SPLIT: {
    encoding = ColumnEncoding::BYTE_STREAM_SPLIT;
    byte_stream_split_decoder.InitializePage();
    break;
  }
  case Encoding::PLAIN:
    // nothing to do here, will be read directly below
    encoding = ColumnEncoding::PLAIN;
    break;

  default:
    throw InvalidInputException("Failed to read file \"%s\": Unsupported page encoding", Reader().GetFileName());
  }
}

void ColumnReader::BeginRead(data_ptr_t define_out, data_ptr_t repeat_out) {
  // we need to reset the location because multiple column readers share the same protocol
  auto &trans = reinterpret_cast<ThriftFileTransport &>(*protocol->getTransport());
  trans.SetLocation(chunk_read_offset);

  // Perform any skips that were not applied yet.
  if (define_out && repeat_out) {
    ApplyPendingSkips(define_out, repeat_out);
  }
}

idx_t ColumnReader::ReadPageHeaders(idx_t max_read, optional_ptr<const TableFilter> filter,
                                    optional_ptr<TableFilterState> filter_state, idx_t rows_to_skip) {
  int8_t page_ordinal = 0;
  while (page_rows_available == 0) {
    aad_crypto_metadata.page_ordinal = page_ordinal;
    PrepareRead(filter, filter_state, rows_to_skip);
    page_ordinal++;
  }
  return MinValue<idx_t>(MinValue<idx_t>(max_read, page_rows_available), STANDARD_VECTOR_SIZE);
}

bool ColumnReader::PrepareRead(idx_t read_now, data_ptr_t define_out, data_ptr_t repeat_out, idx_t result_offset) {
  D_ASSERT(block);

  D_ASSERT(read_now + result_offset <= STANDARD_VECTOR_SIZE);
  D_ASSERT(!page_is_filtered_out);

  if (HasRepeats()) {
    D_ASSERT(repeated_decoder);
    repeated_decoder->GetBatch<uint8_t>(repeat_out + result_offset, read_now);
  }

  if (HasDefines()) {
    D_ASSERT(defined_decoder);
    const auto max_define = NumericCast<uint8_t>(MaxDefine());
    if (!HasRepeats() && defined_decoder->HasRepeatedBatch<uint8_t>(read_now, max_define)) {
      // Fast path: no repeats and all valid
      defined_decoder->GetRepeatedBatch<uint8_t>(read_now, max_define);
      return true;
    }
    defined_decoder->GetBatch<uint8_t>(define_out + result_offset, read_now);
    return false;
  }

  return true; // No defines, so everything is valid
}

void ColumnReader::ReadData(idx_t read_now, data_ptr_t define_out, data_ptr_t repeat_out, Vector &result,
                            idx_t result_offset) {
  // flatten the result vector if required
  if (result_offset != 0 && result.GetVectorType() != VectorType::FLAT_VECTOR) {
    result.Flatten();
    result.Reserve(STANDARD_VECTOR_SIZE);
  }
  if (page_is_filtered_out) {
    // page is filtered out - emit NULL for any rows
    auto &validity = FlatVector::ValidityMutable(result);
    for (idx_t i = 0; i < read_now; i++) {
      validity.SetInvalid(result_offset + i);
    }
    page_rows_available -= read_now;
    return;
  }
  // read the defines/repeats
  const auto all_valid = PrepareRead(read_now, define_out, repeat_out, result_offset);
  if (!IsRoot() && AllValuesAreNull()) {
    // every value is NULL: the parent still needs the define/repeat levels we just read, but there are no
    // values to decode - set the result to NULL and skip the encoding read
    if (result_offset == 0) {
      // we own the entire vector - emit a constant NULL
      ConstantVector::SetNull(result, count_t(read_now));
    } else {
      for (idx_t i = 0; i < read_now; i++) {
        FlatVector::SetNull(result, result_offset + i, true);
      }
    }
    page_rows_available -= read_now;
    return;
  }
  // read the data according to the encoder
  const auto define_ptr = all_valid ? nullptr : static_cast<uint8_t *>(define_out);
  switch (encoding) {
  case ColumnEncoding::DICTIONARY:
    dictionary_decoder.Read(define_ptr, read_now, result, result_offset);
    break;
  case ColumnEncoding::DELTA_BINARY_PACKED:
    delta_binary_packed_decoder.Read(define_ptr, read_now, result, result_offset);
    break;
  case ColumnEncoding::RLE:
    rle_decoder.Read(define_ptr, read_now, result, result_offset);
    break;
  case ColumnEncoding::DELTA_LENGTH_BYTE_ARRAY:
    delta_length_byte_array_decoder.Read(block, define_ptr, read_now, result, result_offset);
    break;
  case ColumnEncoding::DELTA_BYTE_ARRAY:
    delta_byte_array_decoder.Read(define_ptr, read_now, result, result_offset);
    break;
  case ColumnEncoding::BYTE_STREAM_SPLIT:
    byte_stream_split_decoder.Read(define_ptr, read_now, result, result_offset);
    break;
  default:
    Plain(block, define_ptr, read_now, result_offset, result);
    break;
  }
  page_rows_available -= read_now;
}

void ColumnReader::FinishRead(idx_t read_count) {
  auto &trans = reinterpret_cast<ThriftFileTransport &>(*protocol->getTransport());
  chunk_read_offset = trans.GetLocation();

  group_rows_available -= read_count;
}

idx_t ColumnReader::ReadInternal(ColumnReaderInput &input, Vector &result) {
  idx_t result_offset = 0;

  auto &num_values = input.num_values;
  auto &define_out = input.define_out;
  auto &repeat_out = input.repeat_out;

  auto to_read = num_values;
  D_ASSERT(to_read <= STANDARD_VECTOR_SIZE);

  while (to_read > 0) {
    auto read_now = ReadPageHeaders(to_read);

    ReadData(read_now, define_out, repeat_out, result, result_offset);

    result_offset += read_now;
    to_read -= read_now;
  }
  FinishRead(num_values);

  return num_values;
}

idx_t ColumnReader::Read(ColumnReaderInput &input, Vector &result) {
  if (IsRoot() && AllValuesAreNull()) {
    // a top-level column that is entirely NULL - emit a constant NULL vector without reading anything.
    // (nested columns are handled in ReadData: they still need to emit their define/repeat levels)
    ConstantVector::SetNull(result, count_t(input.num_values));
    return input.num_values;
  }
  BeginRead(input.define_out, input.repeat_out);
  return ReadInternal(input, result);
}

void ColumnReader::Select(ColumnReaderInput &input, Vector &result, const SelectionVector &sel,
                          idx_t approved_tuple_count) {
  auto &num_values = input.num_values;
  if (SupportsDirectSelect() && approved_tuple_count < num_values && !(IsRoot() && AllValuesAreNull())) {
    DirectSelect(input, result, sel, approved_tuple_count);
    return;
  }
  Read(input, result);
}

void ColumnReader::DirectSelect(ColumnReaderInput &input, Vector &result, const SelectionVector &sel,
                                idx_t approved_tuple_count) {
  auto &num_values = input.num_values;
  auto &define_out = input.define_out;
  auto &repeat_out = input.repeat_out;

  auto to_read = num_values;

  // prepare the first read if we haven't yet
  BeginRead(define_out, repeat_out);
  auto read_now = ReadPageHeaders(to_read);

  // we can only push the filter into the decoder if we are reading the ENTIRE vector in one go
  if (read_now == to_read && encoding == ColumnEncoding::PLAIN) {
    const auto all_valid = PrepareRead(read_now, define_out, repeat_out, 0);
    const auto define_ptr = all_valid ? nullptr : static_cast<uint8_t *>(define_out);
    PlainSelect(block, define_ptr, read_now, result, sel, approved_tuple_count);

    page_rows_available -= read_now;
    FinishRead(to_read);
    return;
  }
  // fallback to regular read + filter
  ReadInternal(input, result);
}

void ColumnReader::Filter(ColumnReaderInput &input, Vector &result, const TableFilter &filter,
                          TableFilterState &filter_state, SelectionVector &sel, idx_t &approved_tuple_count,
                          bool is_first_filter) {
  auto &num_values = input.num_values;
  if (SupportsDirectFilter() && is_first_filter && !(IsRoot() && AllValuesAreNull())) {
    DirectFilter(input, result, filter, filter_state, sel, approved_tuple_count);
    return;
  }
  Select(input, result, sel, approved_tuple_count);
  ApplyFilter(result, filter, filter_state, num_values, sel, approved_tuple_count);
}

void ColumnReader::DirectFilter(ColumnReaderInput &input, Vector &result, const TableFilter &filter,
                                TableFilterState &filter_state, SelectionVector &sel, idx_t &approved_tuple_count) {
  auto &num_values = input.num_values;
  auto &define_out = input.define_out;
  auto &repeat_out = input.repeat_out;

  auto to_read = num_values;

  // prepare the first read if we haven't yet
  BeginRead(define_out, repeat_out);
  auto read_now = ReadPageHeaders(to_read, &filter, &filter_state);

  // we can only push the filter into the decoder if we are reading the ENTIRE vector in one go
  if (encoding == ColumnEncoding::DICTIONARY && read_now == to_read && dictionary_decoder.HasFilter()) {
    if (page_is_filtered_out) {
      // the page has been filtered out entirely - skip
      approved_tuple_count = 0;
    } else {
      // Push filter into dictionary directly
      // read the defines/repeats
      const auto all_valid = PrepareRead(read_now, define_out, repeat_out, 0);
      const auto define_ptr = all_valid ? nullptr : static_cast<uint8_t *>(define_out);
      dictionary_decoder.Filter(define_ptr, read_now, result, sel, approved_tuple_count);
    }
    page_rows_available -= read_now;
    FinishRead(to_read);
    return;
  }
  // fallback to regular read + filter
  ReadInternal(input, result);
  ApplyFilter(result, filter, filter_state, num_values, sel, approved_tuple_count);
}

void ColumnReader::ApplyFilter(Vector &v, const TableFilter &filter, TableFilterState &filter_state, idx_t scan_count,
                               SelectionVector &sel, idx_t &approved_tuple_count) {
  FlatVector::SetSize(v, count_t(scan_count));
  ColumnSegment::FilterSelection(sel, v, filter_state, scan_count, approved_tuple_count);
}

void ColumnReader::Skip(idx_t num_values) {
  pending_skips += num_values;
}

void ColumnReader::ApplyPendingSkips(data_ptr_t define_out, data_ptr_t repeat_out) {
  if (pending_skips == 0) {
    return;
  }
  idx_t num_values = pending_skips;
  pending_skips = 0;

  auto to_skip = num_values;
  data_t skip_defines[STANDARD_VECTOR_SIZE] = {};
  data_t skip_repeats[STANDARD_VECTOR_SIZE];
  data_ptr_t skip_define_out = HasDefines() ? skip_defines : define_out;
  data_ptr_t skip_repeat_out = HasRepeats() ? skip_repeats : repeat_out;
  // start reading but do not apply skips (we are skipping now)
  BeginRead(nullptr, nullptr);

  while (to_skip > 0) {
    auto skip_now = ReadPageHeaders(to_skip, nullptr, nullptr, to_skip);
    if (page_is_filtered_out) {
      // the page has been filtered out entirely - skip
      page_rows_available -= skip_now;
      to_skip -= skip_now;
      continue;
    }
    const auto all_valid = PrepareRead(skip_now, skip_define_out, skip_repeat_out, 0);

    const auto define_ptr = all_valid ? nullptr : static_cast<uint8_t *>(skip_define_out);
    switch (encoding) {
    case ColumnEncoding::DICTIONARY:
      dictionary_decoder.Skip(define_ptr, skip_now);
      break;
    case ColumnEncoding::DELTA_BINARY_PACKED:
      delta_binary_packed_decoder.Skip(define_ptr, skip_now);
      break;
    case ColumnEncoding::RLE:
      rle_decoder.Skip(define_ptr, skip_now);
      break;
    case ColumnEncoding::DELTA_LENGTH_BYTE_ARRAY:
      delta_length_byte_array_decoder.Skip(define_ptr, skip_now);
      break;
    case ColumnEncoding::DELTA_BYTE_ARRAY:
      delta_byte_array_decoder.Skip(define_ptr, skip_now);
      break;
    case ColumnEncoding::BYTE_STREAM_SPLIT:
      byte_stream_split_decoder.Skip(define_ptr, skip_now);
      break;
    default:
      PlainSkip(*block, define_ptr, skip_now);
      break;
    }
    page_rows_available -= skip_now;
    to_skip -= skip_now;
  }
  FinishRead(num_values);
}

//===--------------------------------------------------------------------===//
// Create Column Reader
//===--------------------------------------------------------------------===//
template <class T>
static unique_ptr<ColumnReader> CreateDecimalReader(const ParquetReader &reader, const ParquetColumnSchema &schema) {
  switch (schema.type.InternalType()) {
  case PhysicalType::INT16:
    return make_uniq<TemplatedColumnReader<int16_t, TemplatedParquetValueConversion<T>>>(reader, schema);
  case PhysicalType::INT32:
    return make_uniq<TemplatedColumnReader<int32_t, TemplatedParquetValueConversion<T>>>(reader, schema);
  case PhysicalType::INT64:
    return make_uniq<TemplatedColumnReader<int64_t, TemplatedParquetValueConversion<T>>>(reader, schema);
  case PhysicalType::INT128:
    return make_uniq<TemplatedColumnReader<hugeint_t, TemplatedParquetValueConversion<T>>>(reader, schema);
  default:
    throw NotImplementedException("Unimplemented internal type for CreateDecimalReader");
  }
}

unique_ptr<ColumnReader> ColumnReader::CreateReader(const ParquetReader &reader, const ParquetColumnSchema &schema) {
  switch (schema.type.id()) {
  case LogicalTypeId::BOOLEAN:
    return make_uniq<BooleanColumnReader>(reader, schema);
  case LogicalTypeId::UTINYINT:
    return make_uniq<TemplatedColumnReader<uint8_t, TemplatedParquetValueConversion<uint32_t>>>(reader, schema);
  case LogicalTypeId::USMALLINT:
    return make_uniq<TemplatedColumnReader<uint16_t, TemplatedParquetValueConversion<uint32_t>>>(reader, schema);
  case LogicalTypeId::UINTEGER:
    return make_uniq<TemplatedColumnReader<uint32_t, TemplatedParquetValueConversion<uint32_t>>>(reader, schema);
  case LogicalTypeId::UBIGINT:
    return make_uniq<TemplatedColumnReader<uint64_t, TemplatedParquetValueConversion<uint64_t>>>(reader, schema);
  case LogicalTypeId::TINYINT:
    return make_uniq<TemplatedColumnReader<int8_t, TemplatedParquetValueConversion<int32_t>>>(reader, schema);
  case LogicalTypeId::SMALLINT:
    return make_uniq<TemplatedColumnReader<int16_t, TemplatedParquetValueConversion<int32_t>>>(reader, schema);
  case LogicalTypeId::INTEGER:
    return make_uniq<TemplatedColumnReader<int32_t, TemplatedParquetValueConversion<int32_t>>>(reader, schema);
  case LogicalTypeId::BIGINT:
    return make_uniq<TemplatedColumnReader<int64_t, TemplatedParquetValueConversion<int64_t>>>(reader, schema);
  case LogicalTypeId::FLOAT:
    if (schema.type_info == ParquetExtraTypeInfo::FLOAT16) {
      return make_uniq<CallbackColumnReader<uint16_t, float, Float16ToFloat32>>(reader, schema);
    }
    return make_uniq<TemplatedColumnReader<float, TemplatedParquetValueConversion<float>>>(reader, schema);
  case LogicalTypeId::DOUBLE:
    if (schema.type_info == ParquetExtraTypeInfo::DECIMAL_BYTE_ARRAY) {
      return ParquetDecimalUtils::CreateReader(reader, schema);
    }
    return make_uniq<TemplatedColumnReader<double, TemplatedParquetValueConversion<double>>>(reader, schema);
  case LogicalTypeId::TIMESTAMP:
  case LogicalTypeId::TIMESTAMP_TZ:
    switch (schema.type_info) {
    case ParquetExtraTypeInfo::IMPALA_TIMESTAMP:
      return make_uniq<CallbackColumnReader<Int96, timestamp_t, ImpalaTimestampToTimestamp>>(reader, schema);
    case ParquetExtraTypeInfo::UNIT_MS:
      return make_uniq<CallbackColumnReader<int64_t, timestamp_t, ParquetTimestampMsToTimestamp>>(reader, schema);
    case ParquetExtraTypeInfo::UNIT_MICROS:
      return make_uniq<CallbackColumnReader<int64_t, timestamp_t, ParquetTimestampMicrosToTimestamp>>(reader,
                                                                                                      schema);
    case ParquetExtraTypeInfo::UNIT_NS:
      return make_uniq<CallbackColumnReader<int64_t, timestamp_t, ParquetTimestampNsToTimestamp>>(reader, schema);
    default:
      throw InternalException("TIMESTAMP requires type info");
    }
  case LogicalTypeId::TIMESTAMP_NS:
  case LogicalTypeId::TIMESTAMP_TZ_NS:
    switch (schema.type_info) {
    case ParquetExtraTypeInfo::IMPALA_TIMESTAMP:
      return make_uniq<CallbackColumnReader<Int96, timestamp_ns_t, ImpalaTimestampToTimestampNS>>(reader, schema);
    case ParquetExtraTypeInfo::UNIT_MS:
      return make_uniq<CallbackColumnReader<int64_t, timestamp_ns_t, ParquetTimestampMsToTimestampNs>>(reader,
                                                                                                       schema);
    case ParquetExtraTypeInfo::UNIT_MICROS:
      return make_uniq<CallbackColumnReader<int64_t, timestamp_ns_t, ParquetTimestampUsToTimestampNs>>(reader,
                                                                                                       schema);
    case ParquetExtraTypeInfo::UNIT_NS:
      return make_uniq<CallbackColumnReader<int64_t, timestamp_ns_t, ParquetTimestampNsToTimestampNs>>(reader,
                                                                                                       schema);
    default:
      throw InternalException("TIMESTAMP_NS requires type info");
    }
  case LogicalTypeId::DATE:
    return make_uniq<CallbackColumnReader<int32_t, date_t, ParquetIntToDate>>(reader, schema);
  case LogicalTypeId::TIME:
    switch (schema.type_info) {
    case ParquetExtraTypeInfo::UNIT_MS:
      return make_uniq<CallbackColumnReader<int32_t, dtime_t, ParquetMsIntToTime>>(reader, schema);
    case ParquetExtraTypeInfo::UNIT_MICROS:
      return make_uniq<CallbackColumnReader<int64_t, dtime_t, ParquetIntToTime>>(reader, schema);
    case ParquetExtraTypeInfo::UNIT_NS:
      return make_uniq<CallbackColumnReader<int64_t, dtime_t, ParquetNsIntToTime>>(reader, schema);
    default:
      throw InternalException("TIME requires type info");
    }
  case LogicalTypeId::TIME_NS:
    switch (schema.type_info) {
    case ParquetExtraTypeInfo::UNIT_MS:
      return make_uniq<CallbackColumnReader<int32_t, dtime_ns_t, ParquetMsIntToTimeNs>>(reader, schema);
    case ParquetExtraTypeInfo::UNIT_MICROS:
      return make_uniq<CallbackColumnReader<int64_t, dtime_ns_t, ParquetUsIntToTimeNs>>(reader, schema);
    case ParquetExtraTypeInfo::UNIT_NS:
      return make_uniq<CallbackColumnReader<int64_t, dtime_ns_t, ParquetIntToTimeNs>>(reader, schema);
    default:
      throw InternalException("TIME requires type info");
    }
  case LogicalTypeId::TIME_TZ:
    switch (schema.type_info) {
    case ParquetExtraTypeInfo::UNIT_MS:
      return make_uniq<CallbackColumnReader<int32_t, dtime_tz_t, ParquetIntToTimeMsTZ>>(reader, schema);
    case ParquetExtraTypeInfo::UNIT_MICROS:
      return make_uniq<CallbackColumnReader<int64_t, dtime_tz_t, ParquetIntToTimeTZ>>(reader, schema);
    case ParquetExtraTypeInfo::UNIT_NS:
      return make_uniq<CallbackColumnReader<int64_t, dtime_tz_t, ParquetIntToTimeNsTZ>>(reader, schema);
    default:
      throw InternalException("TIME_TZ requires type info");
    }
  case LogicalTypeId::BLOB:
  case LogicalTypeId::VARCHAR:
    return make_uniq<StringColumnReader>(reader, schema);
  case LogicalTypeId::DECIMAL:
    // we have to figure out what kind of int we need
    switch (schema.type_info) {
    case ParquetExtraTypeInfo::DECIMAL_INT32:
      return CreateDecimalReader<int32_t>(reader, schema);
    case ParquetExtraTypeInfo::DECIMAL_INT64:
      return CreateDecimalReader<int64_t>(reader, schema);
    case ParquetExtraTypeInfo::DECIMAL_BYTE_ARRAY:
      return ParquetDecimalUtils::CreateReader(reader, schema);
    default:
      throw NotImplementedException("Unrecognized Parquet type for Decimal");
    }
  case LogicalTypeId::UUID:
    return make_uniq<UUIDColumnReader>(reader, schema);
  case LogicalTypeId::INTERVAL:
    return make_uniq<IntervalColumnReader>(reader, schema);
  case LogicalTypeId::SQLNULL:
    return make_uniq<NullColumnReader>(reader, schema);
  default:
    break;
  }
  throw NotImplementedException(schema.type.ToString());
}

} // namespace duckdb

Coverage Report

Created: 2026-06-30 06:38