// Licensed to the Apache Software Foundation (ASF) under one

// or more contributor license agreements.  See the NOTICE file

// distributed with this work for additional information

// regarding copyright ownership.  The ASF licenses this file

// to you under the Apache License, Version 2.0 (the

// "License"); you may not use this file except in compliance

// with the License.  You may obtain a copy of the License at

//

//   http://www.apache.org/licenses/LICENSE-2.0

//

// Unless required by applicable law or agreed to in writing,

// software distributed under the License is distributed on an

// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY

// KIND, either express or implied.  See the License for the

// specific language governing permissions and limitations

// under the License.

#pragma once

#include <gen_cpp/Descriptors_types.h>

#include <gen_cpp/PaloInternalService_types.h>

#include <gen_cpp/PlanNodes_types.h>

#include <stddef.h>

#include <stdint.h>

#include <memory>

#include <set>

#include <string>

#include <unordered_set>

#include <utility>

#include <vector>

#include "agent/be_exec_version_manager.h"

#include "common/status.h"

#include "io/io_common.h"

#include "storage/delete/delete_handler.h"

#include "storage/iterators.h"

#include "storage/olap_common.h"

#include "storage/olap_tuple.h"

#include "storage/predicate/filter_olap_param.h"

#include "storage/row_cursor.h"

#include "storage/rowid_conversion.h"

#include "storage/rowset/rowset.h"

#include "storage/rowset/rowset_meta.h"

#include "storage/rowset/rowset_reader.h"

#include "storage/rowset/rowset_reader_context.h"

#include "storage/tablet/base_tablet.h"

#include "storage/tablet/tablet_fwd.h"

namespace doris {

class RuntimeState;

class BloomFilterFuncBase;

class ColumnPredicate;

class DeleteBitmap;

class HybridSetBase;

class RuntimeProfile;

class VCollectIterator;

class Block;

class VExpr;

class VExprContext;

// Used to compare row with input scan key. Scan key only contains key columns,

// row contains all key columns, which is superset of key columns.

// So we should compare the common prefix columns of lhs and rhs.

//

// NOTE: if you are not sure if you can use it, please don't use this function.

inline int compare_row_key(const RowCursor& lhs, const RowCursor& rhs) {

    auto cmp_cids = std::min(lhs.field_count(), rhs.field_count());

    for (uint32_t cid = 0; cid < cmp_cids; ++cid) {

        const auto& lf = lhs.field(cid);

        const auto& rf = rhs.field(cid);

        // Handle nulls: null < non-null

        if (lf.is_null() != rf.is_null()) {

            return lf.is_null() ? -1 : 1;

        }

        if (lf.is_null()) {

            continue; // both null

        }

        auto cmp = lf <=> rf;

        if (cmp < 0) return -1;

        if (cmp > 0) return 1;

    }

    return 0;

}

class TabletReader {

    struct KeysParam {

        std::vector<RowCursor> start_keys;

        std::vector<RowCursor> end_keys;

        bool start_key_include = false;

        bool end_key_include = false;

    };

public:

    // Params for Reader,

    // mainly include tablet, data version and fetch range.

    struct ReaderParams {

        bool has_single_version() const {

            return (rs_splits.size() == 1 &&

                    rs_splits[0].rs_reader->rowset()->start_version() == 0 &&

                    !rs_splits[0].rs_reader->rowset()->rowset_meta()->is_segments_overlapping()) ||

                   (rs_splits.size() == 2 &&

                    rs_splits[0].rs_reader->rowset()->rowset_meta()->num_rows() == 0 &&

                    rs_splits[1].rs_reader->rowset()->start_version() == 2 &&

                    !rs_splits[1].rs_reader->rowset()->rowset_meta()->is_segments_overlapping());

        }

        int get_be_exec_version() const {

            if (runtime_state) {

                return runtime_state->be_exec_version();

            }

            return BeExecVersionManager::get_newest_version();

        }

        void set_read_source(TabletReadSource read_source, bool skip_delete_bitmap = false) {

            rs_splits = std::move(read_source.rs_splits);

            delete_predicates = std::move(read_source.delete_predicates);

#ifndef BE_TEST

            if (!skip_delete_bitmap && tablet->need_read_delete_bitmap()) {

                delete_bitmap = std::move(read_source.delete_bitmap);

            }

#endif

        }

        BaseTabletSPtr tablet;

        TabletSchemaSPtr tablet_schema;

        ReaderType reader_type = ReaderType::READER_QUERY;

        bool read_row_binlog = false;

        bool direct_mode = false;

        bool aggregation = false;

        // for compaction, schema_change, check_sum: we don't use page cache

        // for query, when the BE config disable_storage_page_cache is false, we use page cache

        bool use_page_cache = false;

        Version version = Version(-1, 0);

        std::vector<OlapTuple> start_key;

        std::vector<OlapTuple> end_key;

        bool start_key_include = false;

        bool end_key_include = false;

        std::vector<std::shared_ptr<ColumnPredicate>> predicates;

        std::vector<RowsetMetaSharedPtr> delete_predicates;

        // slots that cast may be eliminated in storage layer

        std::map<std::string, DataTypePtr> target_cast_type_for_variants;

        std::map<int32_t, TColumnAccessPaths> all_access_paths;

        std::map<int32_t, TColumnAccessPaths> predicate_access_paths;

        std::vector<RowSetSplits> rs_splits;

        // For unique key table with merge-on-write

        DeleteBitmapPtr delete_bitmap = nullptr;

        // The read schema of every Block at the TabletReader boundary.

        // Query scanners construct the FE-slot prefix in scan-tuple order;

        // TabletReader may append storage-only delete-predicate columns.

        ReadSchemaSPtr read_schema;

        // output_columns only contain columns in OrderByExprs and outputExprs

        // (column unique ids, not ordinals)

        std::set<int32_t> output_columns;

        // Ordinals (positions in read_schema) of extra storage key columns

        // that are present only for scan-schema alignment.

        // Example: for AGG keys (k1, k2), a query that returns k2 can scan

        // (k1, k2) and project away k1. Direct readers may avoid reading such

        // columns only if the lower iterator proves their real values are not

        // required by predicates, delete conditions, or expressions.

        std::set<ColumnId> extra_columns;

        RuntimeProfile* profile = nullptr;

        RuntimeState* runtime_state = nullptr;

        TPushAggOp::type push_down_agg_type_opt = TPushAggOp::NONE;

        VExprContextSPtrs common_expr_ctxs_push_down;

        // used for compaction to record row ids

        bool record_rowids = false;

        RowIdConversion* rowid_conversion = nullptr;

        std::vector<int> topn_filter_source_node_ids;

        // used for special optimization for query : ORDER BY key LIMIT n

        bool read_orderby_key = false;

        // used for special optimization for query : ORDER BY key DESC LIMIT n

        bool read_orderby_key_reverse = false;

        // For rows with the same key, use ascending order (small-to-large) for tie-breakers.

        // For example, use lower rowset version / segment id first.

        bool use_insert_order_when_same = false;

        // Force a key-ordered merge across all segments even when their key ranges do not

        // overlap. By default a rowset reader can skip the merge heap if its segments are

        // mono-ascending and disjoint, but row-binlog scans require strict global key order

        // (e.g. so MIN_DELTA can group consecutive same-key changes), so this flag is set.

        // See BetaRowsetReader::is_merge_iterator() in beta_rowset_reader.h:62.

        bool force_key_ordered_read = false;

        // num of columns for orderby key

        size_t read_orderby_key_num_prefix_columns = 0;

        // limit of rows for read_orderby_key

        size_t read_orderby_key_limit = 0;

        // for vertical compaction

        bool is_key_column_group = false;

        std::vector<uint32_t> key_group_cluster_key_idxes;

        // For sparse column compaction optimization

        // When true, use optimized path for sparse wide tables

        bool enable_sparse_optimization = false;

        bool is_segcompaction = false;

        // Enable value predicate pushdown for MOR tables

        bool enable_mor_value_predicate_pushdown = false;

        std::vector<RowwiseIteratorUPtr>* segment_iters_ptr = nullptr;

        void check_validation() const;

        int64_t batch_size = -1;

        // virtual column ordinal (position in read_schema) -> expression

        std::map<ColumnId, VExprContextSPtr> virtual_column_exprs;

        std::shared_ptr<ScoreRuntime> score_runtime;

        CollectionStatisticsPtr collection_statistics;

        std::shared_ptr<segment_v2::AnnTopNRuntime> ann_topn_runtime;

        uint64_t condition_cache_digest = 0;

        // General LIMIT budget forwarded to SegmentIterator. -1 means no limit.

        int64_t general_read_limit = -1;

        TBinlogScanType::type binlog_scan_type = TBinlogScanType::NONE;

    };

    TabletReader() = default;

    virtual ~TabletReader() = default;

    TabletReader(const TabletReader&) = delete;

    void operator=(const TabletReader&) = delete;

    // Initialize TabletReader with tablet, data version and fetch range.

    virtual Status init(const ReaderParams& read_params);

    // Read next block with aggregation.

    // Return OK and set `*eof` to false when next block is read

    // Return OK and set `*eof` to true when no more rows can be read.

    // Return others when unexpected error happens.

    virtual Status next_block_with_aggregation(Block* block, bool* eof) {

        return Status::Error<ErrorCode::READER_INITIALIZE_ERROR>(

                "TabletReader not support next_block_with_aggregation");

    }

    virtual uint64_t merged_rows() const { return _merged_rows; }

    uint64_t filtered_rows() const {

        return _stats.rows_del_filtered + _stats.rows_del_by_bitmap +

               _stats.rows_conditions_filtered + _stats.rows_vec_del_cond_filtered +

               _stats.rows_vec_cond_filtered + _stats.rows_short_circuit_cond_filtered;

    }

    void set_batch_size(int batch_size) { _reader_context.batch_size = batch_size; }

    size_t batch_max_rows() const { return _reader_context.batch_size; }

    void set_preferred_block_size_bytes(size_t bytes) {

        _reader_context.preferred_block_size_bytes = bytes;

    }

    // Returns the preferred output block byte budget. Subclasses that support adaptive batch size

    // should override this; the base returns 0 (disabled) so VCollectIterator degrades safely

    // when called through a TabletReader* that has not been configured.

    virtual size_t preferred_block_size_bytes() const { return 0; }

    const OlapReaderStatistics& stats() const { return _stats; }

    OlapReaderStatistics* mutable_stats() { return &_stats; }

    virtual void update_profile(RuntimeProfile* profile) {}

    // Remove the delete-condition columns from `all_access_paths` so they fall back to a full

    // read (a meta-only read would make the storage delete predicate match nothing and leak

    // deleted rows).

    static void remove_delete_columns_from_access_paths(

            const DeleteHandler& delete_handler, const ReadSchema& read_schema,

            std::map<int32_t, TColumnAccessPaths>& all_access_paths);

protected:

    friend class VCollectIterator;

    Status _init_params(const ReaderParams& read_params);

    Status _capture_rs_readers(const ReaderParams& read_params);

    Status _init_keys_param(const ReaderParams& read_params);

    Status _init_orderby_keys_param(const ReaderParams& read_params);

    Status _init_column_predicates(const ReaderParams& read_params);

    Status _init_delete_condition(const ReaderParams& read_params);

    const BaseTabletSPtr& tablet() { return _tablet; }

    // The read schema shared by the caller, TabletReader, VCollect and each rowset reader.

    ReadSchemaSPtr _read_schema;

    // used for special optimization for query : ORDER BY key [ASC|DESC] LIMIT n

    // columns for orderby keys

    std::vector<uint32_t> _orderby_key_columns;

    BaseTabletSPtr _tablet;

    RowsetReaderContext _reader_context;

    TabletSchemaSPtr _tablet_schema;

    KeysParam _keys_param;

    std::vector<bool> _is_lower_keys_included;

    std::vector<bool> _is_upper_keys_included;

    std::vector<std::shared_ptr<ColumnPredicate>> _col_predicates;

    std::vector<std::shared_ptr<ColumnPredicate>> _value_col_predicates;

    DeleteHandler _delete_handler;

    // Indicates whether the tablets has do a aggregation in storage engine.

    bool _aggregation = false;

    // for agg query, we don't need to finalize when scan agg object data

    ReaderType _reader_type = ReaderType::READER_QUERY;

    bool _delete_sign_available = false;

    bool _filter_delete = false;

    bool _direct_mode = false;

    uint64_t _merged_rows = 0;

    OlapReaderStatistics _stats;

};

} // namespace doris