ParserNode.cpp

Go to the documentation of this file.

/*
 * Copyright 2022 HEAVY.AI, Inc.
 *
 * Licensed 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.
 */

#include "ParserNode.h"

#include <boost/algorithm/string.hpp>
#include <boost/core/null_deleter.hpp>
#include <boost/filesystem.hpp>
#include <boost/function.hpp>

#include <rapidjson/document.h>
#include <rapidjson/stringbuffer.h>
#include <rapidjson/writer.h>

#include <cassert>
#include <cmath>
#include <limits>
#include <random>
#include <regex>
#include <stdexcept>
#include <string>
#include <type_traits>
#include <typeinfo>

#include "Analyzer/RangeTableEntry.h"
#include "Catalog/Catalog.h"
#include "Catalog/DataframeTableDescriptor.h"
#include "Catalog/SharedDictionaryValidator.h"
#include "Fragmenter/InsertOrderFragmenter.h"
#include "Fragmenter/SortedOrderFragmenter.h"
#include "Fragmenter/TargetValueConvertersFactories.h"
#include "Geospatial/Compression.h"
#include "Geospatial/Types.h"
#include "ImportExport/ForeignDataImporter.h"
#include "ImportExport/Importer.h"
#include "ImportExport/RenderGroupAnalyzer.h"
#include "LockMgr/LockMgr.h"
#include "QueryEngine/CalciteAdapter.h"
#include "QueryEngine/CalciteDeserializerUtils.h"
#include "QueryEngine/ErrorHandling.h"
#include "QueryEngine/Execute.h"
#include "QueryEngine/ExtensionFunctionsWhitelist.h"
#include "QueryEngine/JsonAccessors.h"
#include "QueryEngine/RelAlgExecutor.h"
#include "QueryEngine/TableOptimizer.h"
#include "ReservedKeywords.h"
#include "Shared/DbObjectKeys.h"
#include "Shared/StringTransform.h"
#include "Shared/SysDefinitions.h"
#include "Shared/enable_assign_render_groups.h"
#include "Shared/measure.h"
#include "Shared/shard_key.h"
#include "TableArchiver/TableArchiver.h"
#include "Utils/FsiUtils.h"

#include "gen-cpp/CalciteServer.h"

size_t g_leaf_count{0};
bool g_test_drop_column_rollback{false};
extern bool g_enable_string_functions;
extern bool g_enable_fsi;

bool g_enable_legacy_delimited_import{true};
#ifdef ENABLE_IMPORT_PARQUET
bool g_enable_legacy_parquet_import{false};
#endif
bool g_enable_fsi_regex_import{true};

bool g_enable_add_metadata_columns{true};

using Catalog_Namespace::SysCatalog;
using namespace std::string_literals;

using TableDefFuncPtr = boost::function<void(TableDescriptor&,
                                             const Parser::NameValueAssign*,
                                             const std::list<ColumnDescriptor>& columns)>;

using DataframeDefFuncPtr =
    boost::function<void(DataframeTableDescriptor&,
                         const Parser::NameValueAssign*,
                         const std::list<ColumnDescriptor>& columns)>;

namespace Parser {
bool check_session_interrupted(const QuerySessionId& query_session, Executor* executor) {
  // we call this function with unitary executor but is okay since
  // we know the exact session info from a global session map object
  // in the executor
  if (g_enable_non_kernel_time_query_interrupt) {
    heavyai::shared_lock<heavyai::shared_mutex> session_read_lock(
        executor->getSessionLock());
    return executor->checkIsQuerySessionInterrupted(query_session, session_read_lock);
  }
  return false;
}

std::vector<int> getTableChunkKey(const TableDescriptor* td,
                                  Catalog_Namespace::Catalog& catalog) {
  std::vector<int> table_chunk_key_prefix;
  if (td) {
    if (td->fragmenter) {
      table_chunk_key_prefix = td->fragmenter->getFragmentsForQuery().chunkKeyPrefix;
    } else {
      table_chunk_key_prefix.push_back(catalog.getCurrentDB().dbId);
      table_chunk_key_prefix.push_back(td->tableId);
    }
  }
  return table_chunk_key_prefix;
}

std::shared_ptr<Analyzer::Expr> NullLiteral::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  return makeExpr<Analyzer::Constant>(kNULLT, true);
}

std::shared_ptr<Analyzer::Expr> StringLiteral::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  return analyzeValue(*stringval_, false);
}

std::shared_ptr<Analyzer::Expr> StringLiteral::analyzeValue(const std::string& stringval,
                                                            const bool is_null) {
  if (!is_null) {
    const SQLTypeInfo ti(kVARCHAR, stringval.length(), 0, true);
    Datum d;
    d.stringval = new std::string(stringval);
    return makeExpr<Analyzer::Constant>(ti, false, d);
  }
  // Null value
  return makeExpr<Analyzer::Constant>(kVARCHAR, true);
}

std::shared_ptr<Analyzer::Expr> IntLiteral::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  return analyzeValue(intval_);
}

std::shared_ptr<Analyzer::Expr> IntLiteral::analyzeValue(const int64_t intval) {
  SQLTypes t;
  Datum d;
  if (intval >= INT16_MIN && intval <= INT16_MAX) {
    t = kSMALLINT;
    d.smallintval = (int16_t)intval;
  } else if (intval >= INT32_MIN && intval <= INT32_MAX) {
    t = kINT;
    d.intval = (int32_t)intval;
  } else {
    t = kBIGINT;
    d.bigintval = intval;
  }
  return makeExpr<Analyzer::Constant>(t, false, d);
}

std::shared_ptr<Analyzer::Expr> FixedPtLiteral::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  SQLTypeInfo ti(kNUMERIC, 0, 0, false);
  Datum d = StringToDatum(*fixedptval_, ti);
  return makeExpr<Analyzer::Constant>(ti, false, d);
}

std::shared_ptr<Analyzer::Expr> FixedPtLiteral::analyzeValue(const int64_t numericval,
                                                             const int scale,
                                                             const int precision) {
  SQLTypeInfo ti(kNUMERIC, 0, 0, false);
  ti.set_scale(scale);
  ti.set_precision(precision);
  Datum d;
  d.bigintval = numericval;
  return makeExpr<Analyzer::Constant>(ti, false, d);
}

std::shared_ptr<Analyzer::Expr> FloatLiteral::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  Datum d;
  d.floatval = floatval_;
  return makeExpr<Analyzer::Constant>(kFLOAT, false, d);
}

std::shared_ptr<Analyzer::Expr> DoubleLiteral::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  Datum d;
  d.doubleval = doubleval_;
  return makeExpr<Analyzer::Constant>(kDOUBLE, false, d);
}

std::shared_ptr<Analyzer::Expr> TimestampLiteral::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  return get(timestampval_);
}

std::shared_ptr<Analyzer::Expr> TimestampLiteral::get(const int64_t timestampval) {
  Datum d;
  d.bigintval = timestampval;
  return makeExpr<Analyzer::Constant>(kTIMESTAMP, false, d);
}

std::shared_ptr<Analyzer::Expr> UserLiteral::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  Datum d;
  return makeExpr<Analyzer::Constant>(kTEXT, false, d);
}

std::shared_ptr<Analyzer::Expr> UserLiteral::get(const std::string& user) {
  Datum d;
  d.stringval = new std::string(user);
  return makeExpr<Analyzer::Constant>(kTEXT, false, d);
}

std::shared_ptr<Analyzer::Expr> ArrayLiteral::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  SQLTypeInfo ti = SQLTypeInfo(kARRAY, true);
  bool set_subtype = true;
  std::list<std::shared_ptr<Analyzer::Expr>> value_exprs;
  for (auto& p : value_list_) {
    auto e = p->analyze(catalog, query, allow_tlist_ref);
    CHECK(e);
    auto c = std::dynamic_pointer_cast<Analyzer::Constant>(e);
    if (c != nullptr && c->get_is_null()) {
      value_exprs.push_back(c);
      continue;
    }
    auto subtype = e->get_type_info().get_type();
    if (subtype == kNULLT) {
      // NULL element
    } else if (set_subtype) {
      ti.set_subtype(subtype);
      set_subtype = false;
    }
    value_exprs.push_back(e);
  }
  std::shared_ptr<Analyzer::Expr> result =
      makeExpr<Analyzer::Constant>(ti, false, value_exprs);
  return result;
}

std::string ArrayLiteral::to_string() const {
  std::string str = "{";
  bool notfirst = false;
  for (auto& p : value_list_) {
    if (notfirst) {
      str += ", ";
    } else {
      notfirst = true;
    }
    str += p->to_string();
  }
  str += "}";
  return str;
}

std::shared_ptr<Analyzer::Expr> OperExpr::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  auto left_expr = left_->analyze(catalog, query, allow_tlist_ref);
  const auto& left_type = left_expr->get_type_info();
  if (right_ == nullptr) {
    return makeExpr<Analyzer::UOper>(
        left_type, left_expr->get_contains_agg(), optype_, left_expr->decompress());
  }
  if (optype_ == kARRAY_AT) {
    if (left_type.get_type() != kARRAY) {
      throw std::runtime_error(left_->to_string() + " is not of array type.");
    }
    auto right_expr = right_->analyze(catalog, query, allow_tlist_ref);
    const auto& right_type = right_expr->get_type_info();
    if (!right_type.is_integer()) {
      throw std::runtime_error(right_->to_string() + " is not of integer type.");
    }
    return makeExpr<Analyzer::BinOper>(
        left_type.get_elem_type(), false, kARRAY_AT, kONE, left_expr, right_expr);
  }
  auto right_expr = right_->analyze(catalog, query, allow_tlist_ref);
  return normalize(optype_, opqualifier_, left_expr, right_expr);
}

bool exprs_share_one_and_same_rte_idx(const std::shared_ptr<Analyzer::Expr>& lhs_expr,
                                      const std::shared_ptr<Analyzer::Expr>& rhs_expr) {
  std::set<int> lhs_rte_idx;
  lhs_expr->collect_rte_idx(lhs_rte_idx);
  CHECK(!lhs_rte_idx.empty());
  std::set<int> rhs_rte_idx;
  rhs_expr->collect_rte_idx(rhs_rte_idx);
  CHECK(!rhs_rte_idx.empty());
  return lhs_rte_idx.size() == 1UL && lhs_rte_idx == rhs_rte_idx;
}

SQLTypeInfo const& get_str_dict_cast_type(const SQLTypeInfo& lhs_type_info,
                                          const SQLTypeInfo& rhs_type_info,
                                          const Executor* executor) {
  CHECK(lhs_type_info.is_string());
  CHECK(lhs_type_info.get_compression() == kENCODING_DICT);
  CHECK(rhs_type_info.is_string());
  CHECK(rhs_type_info.get_compression() == kENCODING_DICT);
  const auto& lhs_dict_key = lhs_type_info.getStringDictKey();
  const auto& rhs_dict_key = rhs_type_info.getStringDictKey();
  CHECK_NE(lhs_dict_key, rhs_dict_key);
  if (lhs_dict_key.isTransientDict()) {
    return rhs_type_info;
  }
  if (rhs_dict_key.isTransientDict()) {
    return lhs_type_info;
  }
  // If here then neither lhs or rhs type was transient, we should see which
  // type has the largest dictionary and make that the destination type
  const auto lhs_sdp = executor->getStringDictionaryProxy(lhs_dict_key, true);
  const auto rhs_sdp = executor->getStringDictionaryProxy(rhs_dict_key, true);
  return lhs_sdp->entryCount() >= rhs_sdp->entryCount() ? lhs_type_info : rhs_type_info;
}

SQLTypeInfo common_string_type(const SQLTypeInfo& lhs_type_info,
                               const SQLTypeInfo& rhs_type_info,
                               const Executor* executor) {
  CHECK(lhs_type_info.is_string());
  CHECK(rhs_type_info.is_string());
  if (lhs_type_info.is_dict_encoded_string() && rhs_type_info.is_dict_encoded_string()) {
    const auto& lhs_dict_key = lhs_type_info.getStringDictKey();
    const auto& rhs_dict_key = rhs_type_info.getStringDictKey();
    if (lhs_dict_key == rhs_dict_key ||
        (lhs_dict_key.db_id == rhs_dict_key.db_id &&
         lhs_dict_key.dict_id == TRANSIENT_DICT(rhs_dict_key.dict_id))) {
      return lhs_dict_key.dict_id <= rhs_dict_key.dict_id ? lhs_type_info : rhs_type_info;
    }
    return get_str_dict_cast_type(lhs_type_info, rhs_type_info, executor);
  }
  CHECK(lhs_type_info.is_none_encoded_string() || rhs_type_info.is_none_encoded_string());
  SQLTypeInfo ret_ti =
      rhs_type_info.is_none_encoded_string() ? lhs_type_info : rhs_type_info;
  if (ret_ti.is_none_encoded_string()) {
    ret_ti.set_dimension(
        std::max(lhs_type_info.get_dimension(), rhs_type_info.get_dimension()));
  }
  return ret_ti;
}

std::shared_ptr<Analyzer::Expr> OperExpr::normalize(
    const SQLOps optype,
    const SQLQualifier qual,
    std::shared_ptr<Analyzer::Expr> left_expr,
    std::shared_ptr<Analyzer::Expr> right_expr,
    const Executor* executor) {
  if (left_expr->get_type_info().is_date_in_days() ||
      right_expr->get_type_info().is_date_in_days()) {
    // Do not propogate encoding
    left_expr = left_expr->decompress();
    right_expr = right_expr->decompress();
  }
  const auto& left_type = left_expr->get_type_info();
  auto right_type = right_expr->get_type_info();
  if (qual != kONE) {
    // subquery not supported yet.
    CHECK(!std::dynamic_pointer_cast<Analyzer::Subquery>(right_expr));
    if (right_type.get_type() != kARRAY) {
      throw std::runtime_error(
          "Existential or universal qualifiers can only be used in front of a subquery "
          "or an "
          "expression of array type.");
    }
    right_type = right_type.get_elem_type();
  }
  SQLTypeInfo new_left_type;
  SQLTypeInfo new_right_type;
  auto result_type = Analyzer::BinOper::analyze_type_info(
      optype, left_type, right_type, &new_left_type, &new_right_type);
  if (result_type.is_timeinterval()) {
    return makeExpr<Analyzer::BinOper>(
        result_type, false, optype, qual, left_expr, right_expr);
  }
  if (left_type != new_left_type) {
    left_expr = left_expr->add_cast(new_left_type);
  }
  if (right_type != new_right_type) {
    if (qual == kONE) {
      right_expr = right_expr->add_cast(new_right_type);
    } else {
      right_expr = right_expr->add_cast(new_right_type.get_array_type());
    }
  }

  if (IS_COMPARISON(optype)) {
    if (optype != kOVERLAPS && new_left_type.is_geometry() &&
        new_right_type.is_geometry()) {
      throw std::runtime_error(
          "Comparison operators are not yet supported for geospatial types.");
    }

    if (new_left_type.get_compression() == kENCODING_DICT &&
        new_right_type.get_compression() == kENCODING_DICT) {
      if (new_left_type.getStringDictKey() != new_right_type.getStringDictKey()) {
        if (optype == kEQ || optype == kNE) {
          // Join framework does its own string dictionary translation
          // (at least partly since the rhs table projection does not use
          // the normal runtime execution framework), so if we detect
          // that the rte idxs of the two tables are different, bail
          // on translating
          const bool should_translate_strings =
              exprs_share_one_and_same_rte_idx(left_expr, right_expr);
          if (should_translate_strings && (optype == kEQ || optype == kNE)) {
            CHECK(executor);
            // Make the type we're casting to the transient dictionary, if it exists,
            // otherwise the largest dictionary in terms of number of entries
            SQLTypeInfo ti(
                get_str_dict_cast_type(new_left_type, new_right_type, executor));
            auto& expr_to_cast = ti == new_left_type ? right_expr : left_expr;
            ti.set_fixed_size();
            ti.set_dict_intersection();
            expr_to_cast = expr_to_cast->add_cast(ti);
          } else {  // Ordered comparison operator
            // We do not currently support ordered (i.e. >, <=) comparisons between
            // dictionary-encoded columns, and need to decompress when translation
            // is turned off even for kEQ and KNE
            left_expr = left_expr->decompress();
            right_expr = right_expr->decompress();
          }
        } else {  // Ordered comparison operator
          // We do not currently support ordered (i.e. >, <=) comparisons between
          // dictionary-encoded columns, and need to decompress when translation
          // is turned off even for kEQ and KNE
          left_expr = left_expr->decompress();
          right_expr = right_expr->decompress();
        }
      } else {  // Strings shared comp param
        if (!(optype == kEQ || optype == kNE)) {
          // We do not currently support ordered (i.e. >, <=) comparisons between
          // encoded columns, so try to decode (will only succeed with watchdog off)
          left_expr = left_expr->decompress();
          right_expr = right_expr->decompress();
        } else {
          // do nothing, can directly support equals/non-equals comparisons between two
          // dictionary encoded columns sharing the same dictionary as these are
          // effectively integer comparisons in the same dictionary space
        }
      }
    } else if (new_left_type.get_compression() == kENCODING_DICT &&
               new_right_type.get_compression() == kENCODING_NONE) {
      SQLTypeInfo ti(new_right_type);
      ti.set_compression(new_left_type.get_compression());
      ti.set_comp_param(new_left_type.get_comp_param());
      ti.setStringDictKey(new_left_type.getStringDictKey());
      ti.set_fixed_size();
      right_expr = right_expr->add_cast(ti);
    } else if (new_right_type.get_compression() == kENCODING_DICT &&
               new_left_type.get_compression() == kENCODING_NONE) {
      SQLTypeInfo ti(new_left_type);
      ti.set_compression(new_right_type.get_compression());
      ti.set_comp_param(new_right_type.get_comp_param());
      ti.setStringDictKey(new_right_type.getStringDictKey());
      ti.set_fixed_size();
      left_expr = left_expr->add_cast(ti);
    } else {
      left_expr = left_expr->decompress();
      right_expr = right_expr->decompress();
    }
  } else {
    // Is this now a no-op just for pairs of none-encoded string columns
    left_expr = left_expr->decompress();
    right_expr = right_expr->decompress();
  }
  bool has_agg = (left_expr->get_contains_agg() || right_expr->get_contains_agg());
  return makeExpr<Analyzer::BinOper>(
      result_type, has_agg, optype, qual, left_expr, right_expr);
}

std::shared_ptr<Analyzer::Expr> SubqueryExpr::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  throw std::runtime_error("Subqueries are not supported yet.");
  return nullptr;
}

std::shared_ptr<Analyzer::Expr> IsNullExpr::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  auto arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
  auto result = makeExpr<Analyzer::UOper>(kBOOLEAN, kISNULL, arg_expr);
  if (is_not_) {
    result = makeExpr<Analyzer::UOper>(kBOOLEAN, kNOT, result);
  }
  return result;
}

std::shared_ptr<Analyzer::Expr> InSubquery::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  throw std::runtime_error("Subqueries are not supported yet.");
  return nullptr;
}

std::shared_ptr<Analyzer::Expr> InValues::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  auto arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
  SQLTypeInfo ti = arg_expr->get_type_info();
  bool dict_comp = ti.get_compression() == kENCODING_DICT;
  std::list<std::shared_ptr<Analyzer::Expr>> value_exprs;
  for (auto& p : value_list_) {
    auto e = p->analyze(catalog, query, allow_tlist_ref);
    if (ti != e->get_type_info()) {
      if (ti.is_string() && e->get_type_info().is_string()) {
        // Todo(todd): Can we have this leverage the cast framework as well
        ti = Analyzer::BinOper::common_string_type(ti, e->get_type_info());
      } else if (ti.is_number() && e->get_type_info().is_number()) {
        ti = Analyzer::BinOper::common_numeric_type(ti, e->get_type_info());
      } else {
        throw std::runtime_error("IN expressions must contain compatible types.");
      }
    }
    if (dict_comp) {
      value_exprs.push_back(e->add_cast(arg_expr->get_type_info()));
    } else {
      value_exprs.push_back(e);
    }
  }
  if (!dict_comp) {
    arg_expr = arg_expr->decompress();
    arg_expr = arg_expr->add_cast(ti);
    std::list<std::shared_ptr<Analyzer::Expr>> cast_vals;
    for (auto p : value_exprs) {
      cast_vals.push_back(p->add_cast(ti));
    }
    value_exprs.swap(cast_vals);
  }
  std::shared_ptr<Analyzer::Expr> result =
      makeExpr<Analyzer::InValues>(arg_expr, value_exprs);
  if (is_not_) {
    result = makeExpr<Analyzer::UOper>(kBOOLEAN, kNOT, result);
  }
  return result;
}

std::shared_ptr<Analyzer::Expr> BetweenExpr::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  auto arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
  auto lower_expr = lower_->analyze(catalog, query, allow_tlist_ref);
  auto upper_expr = upper_->analyze(catalog, query, allow_tlist_ref);
  SQLTypeInfo new_left_type, new_right_type;
  (void)Analyzer::BinOper::analyze_type_info(kGE,
                                             arg_expr->get_type_info(),
                                             lower_expr->get_type_info(),
                                             &new_left_type,
                                             &new_right_type);
  auto lower_pred =
      makeExpr<Analyzer::BinOper>(kBOOLEAN,
                                  kGE,
                                  kONE,
                                  arg_expr->add_cast(new_left_type)->decompress(),
                                  lower_expr->add_cast(new_right_type)->decompress());
  (void)Analyzer::BinOper::analyze_type_info(kLE,
                                             arg_expr->get_type_info(),
                                             lower_expr->get_type_info(),
                                             &new_left_type,
                                             &new_right_type);
  auto upper_pred = makeExpr<Analyzer::BinOper>(
      kBOOLEAN,
      kLE,
      kONE,
      arg_expr->deep_copy()->add_cast(new_left_type)->decompress(),
      upper_expr->add_cast(new_right_type)->decompress());
  std::shared_ptr<Analyzer::Expr> result =
      makeExpr<Analyzer::BinOper>(kBOOLEAN, kAND, kONE, lower_pred, upper_pred);
  if (is_not_) {
    result = makeExpr<Analyzer::UOper>(kBOOLEAN, kNOT, result);
  }
  return result;
}

std::shared_ptr<Analyzer::Expr> CharLengthExpr::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  auto arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
  if (!arg_expr->get_type_info().is_string()) {
    throw std::runtime_error(
        "expression in char_length clause must be of a string type.");
  }
  std::shared_ptr<Analyzer::Expr> result =
      makeExpr<Analyzer::CharLengthExpr>(arg_expr->decompress(), calc_encoded_length_);
  return result;
}

std::shared_ptr<Analyzer::Expr> CardinalityExpr::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  auto arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
  if (!arg_expr->get_type_info().is_array()) {
    throw std::runtime_error(
        "expression in cardinality clause must be of an array type.");
  }
  std::shared_ptr<Analyzer::Expr> result =
      makeExpr<Analyzer::CardinalityExpr>(arg_expr->decompress());
  return result;
}

void LikeExpr::check_like_expr(const std::string& like_str, char escape_char) {
  if (like_str.back() == escape_char) {
    throw std::runtime_error("LIKE pattern must not end with escape character.");
  }
}

bool LikeExpr::test_is_simple_expr(const std::string& like_str, char escape_char) {
  // if not bounded by '%' then not a simple string
  if (like_str.size() < 2 || like_str[0] != '%' || like_str[like_str.size() - 1] != '%') {
    return false;
  }
  // if the last '%' is escaped then not a simple string
  if (like_str[like_str.size() - 2] == escape_char &&
      like_str[like_str.size() - 3] != escape_char) {
    return false;
  }
  for (size_t i = 1; i < like_str.size() - 1; i++) {
    if (like_str[i] == '%' || like_str[i] == '_' || like_str[i] == '[' ||
        like_str[i] == ']') {
      if (like_str[i - 1] != escape_char) {
        return false;
      }
    }
  }
  return true;
}

void LikeExpr::erase_cntl_chars(std::string& like_str, char escape_char) {
  char prev_char = '\0';
  // easier to create new string of allowable chars
  // rather than erase chars from
  // existing string
  std::string new_str;
  for (char& cur_char : like_str) {
    if (cur_char == '%' || cur_char == escape_char) {
      if (prev_char != escape_char) {
        prev_char = cur_char;
        continue;
      }
    }
    new_str.push_back(cur_char);
    prev_char = cur_char;
  }
  like_str = new_str;
}

std::shared_ptr<Analyzer::Expr> LikeExpr::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  auto arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
  auto like_expr = like_string_->analyze(catalog, query, allow_tlist_ref);
  auto escape_expr = escape_string_ == nullptr
                         ? nullptr
                         : escape_string_->analyze(catalog, query, allow_tlist_ref);
  return LikeExpr::get(arg_expr, like_expr, escape_expr, is_ilike_, is_not_);
}

std::shared_ptr<Analyzer::Expr> LikeExpr::get(std::shared_ptr<Analyzer::Expr> arg_expr,
                                              std::shared_ptr<Analyzer::Expr> like_expr,
                                              std::shared_ptr<Analyzer::Expr> escape_expr,
                                              const bool is_ilike,
                                              const bool is_not) {
  if (!arg_expr->get_type_info().is_string()) {
    throw std::runtime_error("expression before LIKE must be of a string type.");
  }
  if (!like_expr->get_type_info().is_string()) {
    throw std::runtime_error("expression after LIKE must be of a string type.");
  }
  char escape_char = '\\';
  if (escape_expr != nullptr) {
    if (!escape_expr->get_type_info().is_string()) {
      throw std::runtime_error("expression after ESCAPE must be of a string type.");
    }
    if (!escape_expr->get_type_info().is_string()) {
      throw std::runtime_error("expression after ESCAPE must be of a string type.");
    }
    auto c = std::dynamic_pointer_cast<Analyzer::Constant>(escape_expr);
    if (c != nullptr && c->get_constval().stringval->length() > 1) {
      throw std::runtime_error("String after ESCAPE must have a single character.");
    }
    escape_char = (*c->get_constval().stringval)[0];
  }
  auto c = std::dynamic_pointer_cast<Analyzer::Constant>(like_expr);
  bool is_simple = false;
  if (c != nullptr) {
    std::string& pattern = *c->get_constval().stringval;
    if (is_ilike) {
      std::transform(pattern.begin(), pattern.end(), pattern.begin(), ::tolower);
    }
    check_like_expr(pattern, escape_char);
    is_simple = test_is_simple_expr(pattern, escape_char);
    if (is_simple) {
      erase_cntl_chars(pattern, escape_char);
    }
  }
  std::shared_ptr<Analyzer::Expr> result = makeExpr<Analyzer::LikeExpr>(
      arg_expr->decompress(), like_expr, escape_expr, is_ilike, is_simple);
  if (is_not) {
    result = makeExpr<Analyzer::UOper>(kBOOLEAN, kNOT, result);
  }
  return result;
}

void RegexpExpr::check_pattern_expr(const std::string& pattern_str, char escape_char) {
  if (pattern_str.back() == escape_char) {
    throw std::runtime_error("REGEXP pattern must not end with escape character.");
  }
}

bool RegexpExpr::translate_to_like_pattern(std::string& pattern_str, char escape_char) {
  char prev_char = '\0';
  char prev_prev_char = '\0';
  std::string like_str;
  for (char& cur_char : pattern_str) {
    if (prev_char == escape_char || isalnum(cur_char) || cur_char == ' ' ||
        cur_char == '.') {
      like_str.push_back((cur_char == '.') ? '_' : cur_char);
      prev_prev_char = prev_char;
      prev_char = cur_char;
      continue;
    }
    if (prev_char == '.' && prev_prev_char != escape_char) {
      if (cur_char == '*' || cur_char == '+') {
        if (cur_char == '*') {
          like_str.pop_back();
        }
        // .* --> %
        // .+ --> _%
        like_str.push_back('%');
        prev_prev_char = prev_char;
        prev_char = cur_char;
        continue;
      }
    }
    return false;
  }
  pattern_str = like_str;
  return true;
}

std::shared_ptr<Analyzer::Expr> RegexpExpr::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  auto arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
  auto pattern_expr = pattern_string_->analyze(catalog, query, allow_tlist_ref);
  auto escape_expr = escape_string_ == nullptr
                         ? nullptr
                         : escape_string_->analyze(catalog, query, allow_tlist_ref);
  return RegexpExpr::get(arg_expr, pattern_expr, escape_expr, is_not_);
}

std::shared_ptr<Analyzer::Expr> RegexpExpr::get(
    std::shared_ptr<Analyzer::Expr> arg_expr,
    std::shared_ptr<Analyzer::Expr> pattern_expr,
    std::shared_ptr<Analyzer::Expr> escape_expr,
    const bool is_not) {
  if (!arg_expr->get_type_info().is_string()) {
    throw std::runtime_error("expression before REGEXP must be of a string type.");
  }
  if (!pattern_expr->get_type_info().is_string()) {
    throw std::runtime_error("expression after REGEXP must be of a string type.");
  }
  char escape_char = '\\';
  if (escape_expr != nullptr) {
    if (!escape_expr->get_type_info().is_string()) {
      throw std::runtime_error("expression after ESCAPE must be of a string type.");
    }
    if (!escape_expr->get_type_info().is_string()) {
      throw std::runtime_error("expression after ESCAPE must be of a string type.");
    }
    auto c = std::dynamic_pointer_cast<Analyzer::Constant>(escape_expr);
    if (c != nullptr && c->get_constval().stringval->length() > 1) {
      throw std::runtime_error("String after ESCAPE must have a single character.");
    }
    escape_char = (*c->get_constval().stringval)[0];
    if (escape_char != '\\') {
      throw std::runtime_error("Only supporting '\\' escape character.");
    }
  }
  auto c = std::dynamic_pointer_cast<Analyzer::Constant>(pattern_expr);
  if (c != nullptr) {
    std::string& pattern = *c->get_constval().stringval;
    if (translate_to_like_pattern(pattern, escape_char)) {
      return LikeExpr::get(arg_expr, pattern_expr, escape_expr, false, is_not);
    }
  }
  std::shared_ptr<Analyzer::Expr> result =
      makeExpr<Analyzer::RegexpExpr>(arg_expr->decompress(), pattern_expr, escape_expr);
  if (is_not) {
    result = makeExpr<Analyzer::UOper>(kBOOLEAN, kNOT, result);
  }
  return result;
}

std::shared_ptr<Analyzer::Expr> LikelihoodExpr::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  auto arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
  return LikelihoodExpr::get(arg_expr, likelihood_, is_not_);
}

std::shared_ptr<Analyzer::Expr> LikelihoodExpr::get(
    std::shared_ptr<Analyzer::Expr> arg_expr,
    float likelihood,
    const bool is_not) {
  if (!arg_expr->get_type_info().is_boolean()) {
    throw std::runtime_error("likelihood expression expects boolean type.");
  }
  std::shared_ptr<Analyzer::Expr> result = makeExpr<Analyzer::LikelihoodExpr>(
      arg_expr->decompress(), is_not ? 1 - likelihood : likelihood);
  return result;
}

std::shared_ptr<Analyzer::Expr> WidthBucketExpr::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  auto target_value = target_value_->analyze(catalog, query, allow_tlist_ref);
  auto lower_bound = lower_bound_->analyze(catalog, query, allow_tlist_ref);
  auto upper_bound = upper_bound_->analyze(catalog, query, allow_tlist_ref);
  auto partition_count = partition_count_->analyze(catalog, query, allow_tlist_ref);
  return WidthBucketExpr::get(target_value, lower_bound, upper_bound, partition_count);
}

std::shared_ptr<Analyzer::Expr> WidthBucketExpr::get(
    std::shared_ptr<Analyzer::Expr> target_value,
    std::shared_ptr<Analyzer::Expr> lower_bound,
    std::shared_ptr<Analyzer::Expr> upper_bound,
    std::shared_ptr<Analyzer::Expr> partition_count) {
  std::shared_ptr<Analyzer::Expr> result = makeExpr<Analyzer::WidthBucketExpr>(
      target_value, lower_bound, upper_bound, partition_count);
  return result;
}

std::shared_ptr<Analyzer::Expr> ExistsExpr::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  throw std::runtime_error("Subqueries are not supported yet.");
  return nullptr;
}

std::shared_ptr<Analyzer::Expr> ColumnRef::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  int table_id{0};
  int rte_idx{0};
  const ColumnDescriptor* cd{nullptr};
  if (column_ == nullptr) {
    throw std::runtime_error("invalid column name *.");
  }
  if (table_ != nullptr) {
    rte_idx = query.get_rte_idx(*table_);
    if (rte_idx < 0) {
      throw std::runtime_error("range variable or table name " + *table_ +
                               " does not exist.");
    }
    Analyzer::RangeTableEntry* rte = query.get_rte(rte_idx);
    cd = rte->get_column_desc(catalog, *column_);
    if (cd == nullptr) {
      throw std::runtime_error("Column name " + *column_ + " does not exist.");
    }
    table_id = rte->get_table_id();
  } else {
    bool found = false;
    int i = 0;
    for (auto rte : query.get_rangetable()) {
      cd = rte->get_column_desc(catalog, *column_);
      if (cd != nullptr && !found) {
        found = true;
        rte_idx = i;
        table_id = rte->get_table_id();
      } else if (cd != nullptr && found) {
        throw std::runtime_error("Column name " + *column_ + " is ambiguous.");
      }
      i++;
    }
    if (cd == nullptr && allow_tlist_ref != TlistRefType::TLIST_NONE) {
      // check if this is a reference to a targetlist entry
      bool found = false;
      int varno = -1;
      int i = 1;
      std::shared_ptr<Analyzer::TargetEntry> tle;
      for (auto p : query.get_targetlist()) {
        if (*column_ == p->get_resname() && !found) {
          found = true;
          varno = i;
          tle = p;
        } else if (*column_ == p->get_resname() && found) {
          throw std::runtime_error("Output alias " + *column_ + " is ambiguous.");
        }
        i++;
      }
      if (found) {
        if (dynamic_cast<Analyzer::Var*>(tle->get_expr())) {
          Analyzer::Var* v = static_cast<Analyzer::Var*>(tle->get_expr());
          if (v->get_which_row() == Analyzer::Var::kGROUPBY) {
            return v->deep_copy();
          }
        }
        if (allow_tlist_ref == TlistRefType::TLIST_COPY) {
          return tle->get_expr()->deep_copy();
        } else {
          return makeExpr<Analyzer::Var>(
              tle->get_expr()->get_type_info(), Analyzer::Var::kOUTPUT, varno);
        }
      }
    }
    if (cd == nullptr) {
      throw std::runtime_error("Column name " + *column_ + " does not exist.");
    }
  }
  return makeExpr<Analyzer::ColumnVar>(
      cd->columnType,
      shared::ColumnKey{catalog.getDatabaseId(), table_id, cd->columnId},
      rte_idx);
}

std::shared_ptr<Analyzer::Expr> FunctionRef::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  SQLTypeInfo result_type;
  SQLAgg agg_type;
  std::shared_ptr<Analyzer::Expr> arg_expr;
  bool is_distinct = false;
  if (boost::iequals(*name_, "count")) {
    result_type = SQLTypeInfo(kBIGINT, false);
    agg_type = kCOUNT;
    if (arg_) {
      arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
      const SQLTypeInfo& ti = arg_expr->get_type_info();
      if (ti.is_string() && (ti.get_compression() != kENCODING_DICT || !distinct_)) {
        throw std::runtime_error(
            "Strings must be dictionary-encoded in COUNT(DISTINCT).");
      }
      if (ti.get_type() == kARRAY && !distinct_) {
        throw std::runtime_error("Only COUNT(DISTINCT) is supported on arrays.");
      }
    }
    is_distinct = distinct_;
  } else {
    if (!arg_) {
      throw std::runtime_error("Cannot compute " + *name_ + " with argument '*'.");
    }
    if (boost::iequals(*name_, "min")) {
      agg_type = kMIN;
      arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
      arg_expr = arg_expr->decompress();
      result_type = arg_expr->get_type_info();
    } else if (boost::iequals(*name_, "max")) {
      agg_type = kMAX;
      arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
      arg_expr = arg_expr->decompress();
      result_type = arg_expr->get_type_info();
    } else if (boost::iequals(*name_, "avg")) {
      agg_type = kAVG;
      arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
      if (!arg_expr->get_type_info().is_number()) {
        throw std::runtime_error("Cannot compute AVG on non-number-type arguments.");
      }
      arg_expr = arg_expr->decompress();
      result_type = SQLTypeInfo(kDOUBLE, false);
    } else if (boost::iequals(*name_, "sum")) {
      agg_type = kSUM;
      arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
      if (!arg_expr->get_type_info().is_number()) {
        throw std::runtime_error("Cannot compute SUM on non-number-type arguments.");
      }
      arg_expr = arg_expr->decompress();
      result_type = arg_expr->get_type_info().is_integer() ? SQLTypeInfo(kBIGINT, false)
                                                           : arg_expr->get_type_info();
    } else if (boost::iequals(*name_, "unnest")) {
      arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
      const SQLTypeInfo& arg_ti = arg_expr->get_type_info();
      if (arg_ti.get_type() != kARRAY) {
        throw std::runtime_error(arg_->to_string() + " is not of array type.");
      }
      return makeExpr<Analyzer::UOper>(arg_ti.get_elem_type(), false, kUNNEST, arg_expr);
    } else {
      throw std::runtime_error("invalid function name: " + *name_);
    }
    if (arg_expr->get_type_info().is_string() ||
        arg_expr->get_type_info().get_type() == kARRAY) {
      throw std::runtime_error(
          "Only COUNT(DISTINCT ) aggregate is supported on strings and arrays.");
    }
  }
  int naggs = query.get_num_aggs();
  query.set_num_aggs(naggs + 1);
  return makeExpr<Analyzer::AggExpr>(
      result_type, agg_type, arg_expr, is_distinct, nullptr);
}

std::shared_ptr<Analyzer::Expr> CastExpr::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  target_type_->check_type();
  auto arg_expr = arg_->analyze(catalog, query, allow_tlist_ref);
  SQLTypeInfo ti(target_type_->get_type(),
                 target_type_->get_param1(),
                 target_type_->get_param2(),
                 arg_expr->get_type_info().get_notnull());
  if (arg_expr->get_type_info().get_type() != target_type_->get_type() &&
      arg_expr->get_type_info().get_compression() != kENCODING_NONE) {
    arg_expr->decompress();
  }
  return arg_expr->add_cast(ti);
}

std::shared_ptr<Analyzer::Expr> CaseExpr::analyze(
    const Catalog_Namespace::Catalog& catalog,
    Analyzer::Query& query,
    TlistRefType allow_tlist_ref) const {
  SQLTypeInfo ti;
  std::list<std::pair<std::shared_ptr<Analyzer::Expr>, std::shared_ptr<Analyzer::Expr>>>
      expr_pair_list;
  for (auto& p : when_then_list_) {
    auto e1 = p->get_expr1()->analyze(catalog, query, allow_tlist_ref);
    if (e1->get_type_info().get_type() != kBOOLEAN) {
      throw std::runtime_error("Only boolean expressions can be used after WHEN.");
    }
    auto e2 = p->get_expr2()->analyze(catalog, query, allow_tlist_ref);
    expr_pair_list.emplace_back(e1, e2);
  }
  auto else_e =
      else_expr_ ? else_expr_->analyze(catalog, query, allow_tlist_ref) : nullptr;
  return normalize(expr_pair_list, else_e);
}

namespace {

bool bool_from_string_literal(const Parser::StringLiteral* str_literal) {
  const std::string* s = str_literal->get_stringval();
  if (*s == "t" || *s == "true" || *s == "T" || *s == "True") {
    return true;
  } else if (*s == "f" || *s == "false" || *s == "F" || *s == "False") {
    return false;
  } else {
    throw std::runtime_error("Invalid string for boolean " + *s);
  }
}

void parse_copy_params(const std::list<std::unique_ptr<NameValueAssign>>& options_,
                       import_export::CopyParams& copy_params,
                       std::vector<std::string>& warnings,
                       std::string& deferred_copy_from_partitions_) {
  if (!options_.empty()) {
    for (auto& p : options_) {
      if (boost::iequals(*p->get_name(), "max_reject")) {
        const IntLiteral* int_literal = dynamic_cast<const IntLiteral*>(p->get_value());
        if (int_literal == nullptr) {
          throw std::runtime_error("max_reject option must be an integer.");
        }
        copy_params.max_reject = int_literal->get_intval();
      } else if (boost::iequals(*p->get_name(), "max_import_batch_row_count")) {
        const IntLiteral* int_literal = dynamic_cast<const IntLiteral*>(p->get_value());
        if (int_literal == nullptr) {
          throw std::runtime_error(
              "max_import_batch_row_count option must be an integer.");
        }
        if (int_literal->get_intval() <= 0) {
          throw std::runtime_error(
              "max_import_batch_row_count option must be a positive integer (greater "
              "than 0).");
        }
        copy_params.max_import_batch_row_count = int_literal->get_intval();
      } else if (boost::iequals(*p->get_name(), "buffer_size")) {
        const IntLiteral* int_literal = dynamic_cast<const IntLiteral*>(p->get_value());
        if (int_literal == nullptr) {
          throw std::runtime_error("buffer_size option must be an integer.");
        }
        copy_params.buffer_size = int_literal->get_intval();
      } else if (boost::iequals(*p->get_name(), "threads")) {
        const IntLiteral* int_literal = dynamic_cast<const IntLiteral*>(p->get_value());
        if (int_literal == nullptr) {
          throw std::runtime_error("Threads option must be an integer.");
        }
        copy_params.threads = int_literal->get_intval();
      } else if (boost::iequals(*p->get_name(), "delimiter")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Delimiter option must be a string.");
        } else if (str_literal->get_stringval()->length() != 1) {
          throw std::runtime_error("Delimiter must be a single character string.");
        }
        copy_params.delimiter = (*str_literal->get_stringval())[0];
      } else if (boost::iequals(*p->get_name(), "nulls")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Nulls option must be a string.");
        }
        copy_params.null_str = *str_literal->get_stringval();
      } else if (boost::iequals(*p->get_name(), "header")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Header option must be a boolean.");
        }
        copy_params.has_header = bool_from_string_literal(str_literal)
                                     ? import_export::ImportHeaderRow::kHasHeader
                                     : import_export::ImportHeaderRow::kNoHeader;
#ifdef ENABLE_IMPORT_PARQUET
      } else if (boost::iequals(*p->get_name(), "parquet")) {
        warnings.push_back(
            "Deprecation Warning: COPY FROM WITH (parquet='true') is deprecated. Use "
            "WITH (source_type='parquet_file') instead.");
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("'parquet' option must be a boolean.");
        }
        if (bool_from_string_literal(str_literal)) {
          // not sure a parquet "table" type is proper, but to make code
          // look consistent in some places, let's set "table" type too
          copy_params.source_type = import_export::SourceType::kParquetFile;
        }
#endif  // ENABLE_IMPORT_PARQUET
      } else if (boost::iequals(*p->get_name(), "s3_access_key")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option s3_access_key must be a string.");
        }
        copy_params.s3_access_key = *str_literal->get_stringval();
      } else if (boost::iequals(*p->get_name(), "s3_secret_key")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option s3_secret_key must be a string.");
        }
        copy_params.s3_secret_key = *str_literal->get_stringval();
      } else if (boost::iequals(*p->get_name(), "s3_session_token")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option s3_session_token must be a string.");
        }
        copy_params.s3_session_token = *str_literal->get_stringval();
      } else if (boost::iequals(*p->get_name(), "s3_region")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option s3_region must be a string.");
        }
        copy_params.s3_region = *str_literal->get_stringval();
      } else if (boost::iequals(*p->get_name(), "s3_endpoint")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option s3_endpoint must be a string.");
        }
        copy_params.s3_endpoint = *str_literal->get_stringval();
      } else if (boost::iequals(*p->get_name(), "s3_max_concurrent_downloads")) {
        const IntLiteral* int_literal = dynamic_cast<const IntLiteral*>(p->get_value());
        if (int_literal == nullptr) {
          throw std::runtime_error(
              "'s3_max_concurrent_downloads' option must be an integer");
        }
        const int s3_max_concurrent_downloads = int_literal->get_intval();
        if (s3_max_concurrent_downloads > 0) {
          copy_params.s3_max_concurrent_downloads = s3_max_concurrent_downloads;
        } else {
          throw std::runtime_error(
              "Invalid value for 's3_max_concurrent_downloads' option (must be > 0): " +
              std::to_string(s3_max_concurrent_downloads));
        }
      } else if (boost::iequals(*p->get_name(), "quote")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Quote option must be a string.");
        } else if (str_literal->get_stringval()->length() != 1) {
          throw std::runtime_error("Quote must be a single character string.");
        }
        copy_params.quote = (*str_literal->get_stringval())[0];
      } else if (boost::iequals(*p->get_name(), "escape")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Escape option must be a string.");
        } else if (str_literal->get_stringval()->length() != 1) {
          throw std::runtime_error("Escape must be a single character string.");
        }
        copy_params.escape = (*str_literal->get_stringval())[0];
      } else if (boost::iequals(*p->get_name(), "line_delimiter")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Line_delimiter option must be a string.");
        } else if (str_literal->get_stringval()->length() != 1) {
          throw std::runtime_error("Line_delimiter must be a single character string.");
        }
        copy_params.line_delim = (*str_literal->get_stringval())[0];
      } else if (boost::iequals(*p->get_name(), "quoted")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Quoted option must be a boolean.");
        }
        copy_params.quoted = bool_from_string_literal(str_literal);
      } else if (boost::iequals(*p->get_name(), "plain_text")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("plain_text option must be a boolean.");
        }
        copy_params.plain_text = bool_from_string_literal(str_literal);
      } else if (boost::iequals(*p->get_name(), "trim_spaces")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("trim_spaces option must be a boolean.");
        }
        copy_params.trim_spaces = bool_from_string_literal(str_literal);
      } else if (boost::iequals(*p->get_name(), "array_marker")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Array Marker option must be a string.");
        } else if (str_literal->get_stringval()->length() != 2) {
          throw std::runtime_error(
              "Array Marker option must be exactly two characters.  Default is {}.");
        }
        copy_params.array_begin = (*str_literal->get_stringval())[0];
        copy_params.array_end = (*str_literal->get_stringval())[1];
      } else if (boost::iequals(*p->get_name(), "array_delimiter")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Array Delimiter option must be a string.");
        } else if (str_literal->get_stringval()->length() != 1) {
          throw std::runtime_error("Array Delimiter must be a single character string.");
        }
        copy_params.array_delim = (*str_literal->get_stringval())[0];
      } else if (boost::iequals(*p->get_name(), "lonlat")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Lonlat option must be a boolean.");
        }
        copy_params.lonlat = bool_from_string_literal(str_literal);
      } else if (boost::iequals(*p->get_name(), "geo")) {
        warnings.push_back(
            "Deprecation Warning: COPY FROM WITH (geo='true') is deprecated. Use WITH "
            "(source_type='geo_file') instead.");
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("'geo' option must be a boolean.");
        }
        if (bool_from_string_literal(str_literal)) {
          copy_params.source_type = import_export::SourceType::kGeoFile;
        }
      } else if (boost::iequals(*p->get_name(), "source_type")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("'source_type' option must be a string.");
        }
        const std::string* s = str_literal->get_stringval();
        if (boost::iequals(*s, "delimited_file")) {
          copy_params.source_type = import_export::SourceType::kDelimitedFile;
        } else if (boost::iequals(*s, "geo_file")) {
          copy_params.source_type = import_export::SourceType::kGeoFile;
#if ENABLE_IMPORT_PARQUET
        } else if (boost::iequals(*s, "parquet_file")) {
          copy_params.source_type = import_export::SourceType::kParquetFile;
#endif
        } else if (boost::iequals(*s, "raster_file")) {
          copy_params.source_type = import_export::SourceType::kRasterFile;
        } else if (boost::iequals(*s, "regex_parsed_file")) {
          copy_params.source_type = import_export::SourceType::kRegexParsedFile;
        } else {
          throw std::runtime_error(
              "Invalid string for 'source_type' option (must be 'GEO_FILE', 'RASTER_FILE'"
#if ENABLE_IMPORT_PARQUET
              ", 'PARQUET_FILE'"
#endif
              ", 'REGEX_PARSED_FILE'"
              " or 'DELIMITED_FILE'): " +
              *s);
        }
      } else if (boost::iequals(*p->get_name(), "geo_coords_type")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("'geo_coords_type' option must be a string");
        }
        const std::string* s = str_literal->get_stringval();
        if (boost::iequals(*s, "geography")) {
          throw std::runtime_error(
              "GEOGRAPHY coords type not yet supported. Please use GEOMETRY.");
          // copy_params.geo_coords_type = kGEOGRAPHY;
        } else if (boost::iequals(*s, "geometry")) {
          copy_params.geo_coords_type = kGEOMETRY;
        } else {
          throw std::runtime_error(
              "Invalid string for 'geo_coords_type' option (must be 'GEOGRAPHY' or "
              "'GEOMETRY'): " +
              *s);
        }
      } else if (boost::iequals(*p->get_name(), "raster_point_type")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("'raster_point_type' option must be a string");
        }
        const std::string* s = str_literal->get_stringval();
        if (boost::iequals(*s, "none")) {
          copy_params.raster_point_type = import_export::RasterPointType::kNone;
        } else if (boost::iequals(*s, "auto")) {
          copy_params.raster_point_type = import_export::RasterPointType::kAuto;
        } else if (boost::iequals(*s, "smallint")) {
          copy_params.raster_point_type = import_export::RasterPointType::kSmallInt;
        } else if (boost::iequals(*s, "int")) {
          copy_params.raster_point_type = import_export::RasterPointType::kInt;
        } else if (boost::iequals(*s, "float")) {
          copy_params.raster_point_type = import_export::RasterPointType::kFloat;
        } else if (boost::iequals(*s, "double")) {
          copy_params.raster_point_type = import_export::RasterPointType::kDouble;
        } else if (boost::iequals(*s, "point")) {
          copy_params.raster_point_type = import_export::RasterPointType::kPoint;
        } else {
          throw std::runtime_error(
              "Invalid string for 'raster_point_type' option (must be 'NONE', 'AUTO', "
              "'SMALLINT', 'INT', 'FLOAT', 'DOUBLE' or 'POINT'): " +
              *s);
        }
      } else if (boost::iequals(*p->get_name(), "raster_point_transform")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("'raster_point_transform' option must be a string");
        }
        const std::string* s = str_literal->get_stringval();
        if (boost::iequals(*s, "none")) {
          copy_params.raster_point_transform = import_export::RasterPointTransform::kNone;
        } else if (boost::iequals(*s, "auto")) {
          copy_params.raster_point_transform = import_export::RasterPointTransform::kAuto;
        } else if (boost::iequals(*s, "file")) {
          copy_params.raster_point_transform = import_export::RasterPointTransform::kFile;
        } else if (boost::iequals(*s, "world")) {
          copy_params.raster_point_transform =
              import_export::RasterPointTransform::kWorld;
        } else {
          throw std::runtime_error(
              "Invalid string for 'raster_point_transform' option (must be 'NONE', "
              "'AUTO', 'FILE' or 'WORLD'): " +
              *s);
        }
      } else if (boost::iequals(*p->get_name(), "raster_import_bands")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("'raster_import_bands' option must be a string");
        }
        const std::string* raster_import_bands = str_literal->get_stringval();
        if (raster_import_bands) {
          copy_params.raster_import_bands = *raster_import_bands;
        } else {
          throw std::runtime_error("Invalid value for 'raster_import_bands' option");
        }
      } else if (boost::iequals(*p->get_name(), "raster_import_dimensions")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("'raster_import_dimensions' option must be a string");
        }
        const std::string* raster_import_dimensions = str_literal->get_stringval();
        if (raster_import_dimensions) {
          copy_params.raster_import_dimensions = *raster_import_dimensions;
        } else {
          throw std::runtime_error("Invalid value for 'raster_import_dimensions' option");
        }
      } else if (boost::iequals(*p->get_name(), "geo_coords_encoding")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("'geo_coords_encoding' option must be a string");
        }
        const std::string* s = str_literal->get_stringval();
        if (boost::iequals(*s, "none")) {
          copy_params.geo_coords_encoding = kENCODING_NONE;
          copy_params.geo_coords_comp_param = 0;
        } else if (boost::iequals(*s, "compressed(32)")) {
          copy_params.geo_coords_encoding = kENCODING_GEOINT;
          copy_params.geo_coords_comp_param = 32;
        } else {
          throw std::runtime_error(
              "Invalid string for 'geo_coords_encoding' option (must be 'NONE' or "
              "'COMPRESSED(32)'): " +
              *s);
        }
      } else if (boost::iequals(*p->get_name(), "raster_scanlines_per_thread")) {
        const IntLiteral* int_literal = dynamic_cast<const IntLiteral*>(p->get_value());
        if (int_literal == nullptr) {
          throw std::runtime_error(
              "'raster_scanlines_per_thread' option must be an integer");
        }
        const int raster_scanlines_per_thread = int_literal->get_intval();
        if (raster_scanlines_per_thread < 0) {
          throw std::runtime_error(
              "'raster_scanlines_per_thread' option must be >= 0, with 0 denoting auto "
              "sizing");
        }
        copy_params.raster_scanlines_per_thread = raster_scanlines_per_thread;
      } else if (boost::iequals(*p->get_name(), "geo_coords_srid")) {
        const IntLiteral* int_literal = dynamic_cast<const IntLiteral*>(p->get_value());
        if (int_literal == nullptr) {
          throw std::runtime_error("'geo_coords_srid' option must be an integer");
        }
        const int srid = int_literal->get_intval();
        if (srid == 4326 || srid == 3857 || srid == 900913) {
          copy_params.geo_coords_srid = srid;
        } else {
          throw std::runtime_error(
              "Invalid value for 'geo_coords_srid' option (must be 4326, 3857, or "
              "900913): " +
              std::to_string(srid));
        }
      } else if (boost::iequals(*p->get_name(), "geo_layer_name")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("'geo_layer_name' option must be a string");
        }
        const std::string* layer_name = str_literal->get_stringval();
        if (layer_name) {
          copy_params.geo_layer_name = *layer_name;
        } else {
          throw std::runtime_error("Invalid value for 'geo_layer_name' option");
        }
      } else if (boost::iequals(*p->get_name(), "partitions")) {
        const auto partitions =
            static_cast<const StringLiteral*>(p->get_value())->get_stringval();
        CHECK(partitions);
        const auto partitions_uc = boost::to_upper_copy<std::string>(*partitions);
        if (partitions_uc != "REPLICATED") {
          throw std::runtime_error(
              "Invalid value for 'partitions' option. Must be 'REPLICATED'.");
        }
        deferred_copy_from_partitions_ = partitions_uc;
      } else if (boost::iequals(*p->get_name(), "geo_assign_render_groups")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("geo_assign_render_groups option must be a boolean.");
        }
        copy_params.geo_assign_render_groups = bool_from_string_literal(str_literal);
      } else if (boost::iequals(*p->get_name(), "geo_explode_collections")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("geo_explode_collections option must be a boolean.");
        }
        copy_params.geo_explode_collections = bool_from_string_literal(str_literal);
      } else if (boost::iequals(*p->get_name(), "source_srid")) {
        const IntLiteral* int_literal = dynamic_cast<const IntLiteral*>(p->get_value());
        if (int_literal == nullptr) {
          throw std::runtime_error("'source_srid' option must be an integer");
        }
        const int srid = int_literal->get_intval();
        if (copy_params.source_type == import_export::SourceType::kDelimitedFile) {
          copy_params.source_srid = srid;
        } else {
          throw std::runtime_error(
              "'source_srid' option can only be used on csv/tsv files");
        }
      } else if (boost::iequals(*p->get_name(), "regex_path_filter")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option regex_path_filter must be a string.");
        }
        const auto string_val = *str_literal->get_stringval();
        copy_params.regex_path_filter =
            string_val.empty() ? std::nullopt : std::optional<std::string>{string_val};
      } else if (boost::iequals(*p->get_name(), "file_sort_order_by")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option file_sort_order_by must be a string.");
        }
        const auto string_val = *str_literal->get_stringval();
        copy_params.file_sort_order_by =
            string_val.empty() ? std::nullopt : std::optional<std::string>{string_val};
      } else if (boost::iequals(*p->get_name(), "file_sort_regex")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option file_sort_regex must be a string.");
        }
        const auto string_val = *str_literal->get_stringval();
        copy_params.file_sort_regex =
            string_val.empty() ? std::nullopt : std::optional<std::string>{string_val};
      } else if (boost::iequals(*p->get_name(), "raster_point_compute_angle")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error(
              "'raster_point_compute_angle' option must be a boolean.");
        }
        if (bool_from_string_literal(str_literal)) {
          copy_params.raster_point_compute_angle = true;
        }
      } else if (boost::iequals(*p->get_name(), "sql_order_by")) {
        if (auto str_literal = dynamic_cast<const StringLiteral*>(p->get_value())) {
          copy_params.sql_order_by = *str_literal->get_stringval();
        } else {
          throw std::runtime_error("Option sql_order_by must be a string.");
        }
      } else if (boost::iequals(*p->get_name(), "username")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option username must be a string.");
        }
        const auto string_val = *str_literal->get_stringval();
        copy_params.username = string_val;
      } else if (boost::iequals(*p->get_name(), "password")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option password must be a string.");
        }
        const auto string_val = *str_literal->get_stringval();
        copy_params.password = string_val;
      } else if (boost::iequals(*p->get_name(), "credential_string")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option credential_string must be a string.");
        }
        const auto string_val = *str_literal->get_stringval();
        copy_params.credential_string = string_val;
      } else if (boost::iequals(*p->get_name(), "data_source_name")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option data_source_name must be a string.");
        }
        const auto string_val = *str_literal->get_stringval();
        copy_params.dsn = string_val;
      } else if (boost::iequals(*p->get_name(), "connection_string")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option connection_string must be a string.");
        }
        const auto string_val = *str_literal->get_stringval();
        copy_params.connection_string = string_val;
      } else if (boost::iequals(*p->get_name(), "line_start_regex")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option line_start_regex must be a string.");
        }
        const auto string_val = *str_literal->get_stringval();
        copy_params.line_start_regex = string_val;
      } else if (boost::iequals(*p->get_name(), "line_regex")) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("Option line_regex must be a string.");
        }
        const auto string_val = *str_literal->get_stringval();
        copy_params.line_regex = string_val;
      } else if (boost::iequals(*p->get_name(), "add_metadata_columns") &&
                 g_enable_add_metadata_columns) {
        const StringLiteral* str_literal =
            dynamic_cast<const StringLiteral*>(p->get_value());
        if (str_literal == nullptr) {
          throw std::runtime_error("'add_metadata_columns' option must be a string.");
        }
        copy_params.add_metadata_columns = *str_literal->get_stringval();
      } else {
        throw std::runtime_error("Invalid option for COPY: " + *p->get_name());
      }
    }
  }
}

bool expr_is_null(const Analyzer::Expr* expr) {
  if (expr->get_type_info().get_type() == kNULLT) {
    return true;
  }
  const auto const_expr = dynamic_cast<const Analyzer::Constant*>(expr);
  return const_expr && const_expr->get_is_null();
}

}  // namespace

std::shared_ptr<Analyzer::Expr> CaseExpr::normalize(
    const std::list<std::pair<std::shared_ptr<Analyzer::Expr>,
                              std::shared_ptr<Analyzer::Expr>>>& expr_pair_list,
    const std::shared_ptr<Analyzer::Expr> else_e_in,
    const Executor* executor) {
  SQLTypeInfo ti;
  bool has_agg = false;
  // We need to keep track of whether there was at
  // least one none-encoded string literal expression
  // type among any of the case sub-expressions separately
  // from rest of type determination logic, as it will
  // be casted to the output dictionary type if all output
  // types are either dictionary encoded or none-encoded
  // literals, or a transient encoded dictionary if all
  // types are none-encoded (column or literal)
  SQLTypeInfo none_encoded_literal_ti;

  for (auto& p : expr_pair_list) {
    auto e1 = p.first;
    CHECK(e1->get_type_info().is_boolean());
    auto e2 = p.second;
    if (e2->get_contains_agg()) {
      has_agg = true;
    }
    const auto& e2_ti = e2->get_type_info();
    const auto col_var = std::dynamic_pointer_cast<const Analyzer::ColumnVar>(e2);
    if (e2_ti.is_string() && !e2_ti.is_dict_encoded_string() && !col_var) {
      CHECK(e2_ti.is_none_encoded_string());
      none_encoded_literal_ti =
          none_encoded_literal_ti.get_type() == kNULLT
              ? e2_ti
              : common_string_type(none_encoded_literal_ti, e2_ti, executor);
      continue;
    }
    if (ti.get_type() == kNULLT) {
      ti = e2_ti;
    } else if (e2_ti.get_type() == kNULLT) {
      ti.set_notnull(false);
      e2->set_type_info(ti);
    } else if (ti != e2_ti) {
      if (ti.is_string() && e2_ti.is_string()) {
        // Executor is needed to determine which dictionary is the largest
        // in case of two dictionary types with different encodings
        ti = common_string_type(ti, e2_ti, executor);
      } else if (ti.is_number() && e2_ti.is_number()) {
        ti = Analyzer::BinOper::common_numeric_type(ti, e2_ti);
      } else if (ti.is_boolean() && e2_ti.is_boolean()) {
        ti = Analyzer::BinOper::common_numeric_type(ti, e2_ti);
      } else {
        throw std::runtime_error(
            "Expressions in THEN clause must be of the same or compatible types.");
      }
    }
  }
  auto else_e = else_e_in;
  const auto& else_ti = else_e->get_type_info();
  if (else_e) {
    const auto col_var = std::dynamic_pointer_cast<const Analyzer::ColumnVar>(else_e);
    if (else_e->get_contains_agg()) {
      has_agg = true;
    }
    if (else_ti.is_string() && !else_ti.is_dict_encoded_string() && !col_var) {
      CHECK(else_ti.is_none_encoded_string());
      none_encoded_literal_ti =
          none_encoded_literal_ti.get_type() == kNULLT
              ? else_ti
              : common_string_type(none_encoded_literal_ti, else_ti, executor);
    } else {
      if (ti.get_type() == kNULLT) {
        ti = else_ti;
      } else if (expr_is_null(else_e.get())) {
        ti.set_notnull(false);
        else_e->set_type_info(ti);
      } else if (ti != else_ti) {
        ti.set_notnull(false);
        if (ti.is_string() && else_ti.is_string()) {
          // Executor is needed to determine which dictionary is the largest
          // in case of two dictionary types with different encodings
          ti = common_string_type(ti, else_ti, executor);
        } else if (ti.is_number() && else_ti.is_number()) {
          ti = Analyzer::BinOper::common_numeric_type(ti, else_ti);
        } else if (ti.is_boolean() && else_ti.is_boolean()) {
          ti = Analyzer::BinOper::common_numeric_type(ti, else_ti);
        } else if (get_logical_type_info(ti) != get_logical_type_info(else_ti)) {
          throw std::runtime_error(
              // types differing by encoding will be resolved at decode
              "Expressions in ELSE clause must be of the same or compatible types as "
              "those in the THEN clauses.");
        }
      }
    }
  }

  if (ti.get_type() == kNULLT && none_encoded_literal_ti.get_type() != kNULLT) {
    // If we haven't set a type so far it's because
    // every case sub-expression has a none-encoded
    // literal output. Output a transient-encoded dictionary
    // so we can use the output downstream
    ti = SQLTypeInfo(kTEXT, kENCODING_DICT, TRANSIENT_DICT_ID, kNULLT);
  }

  std::list<std::pair<std::shared_ptr<Analyzer::Expr>, std::shared_ptr<Analyzer::Expr>>>
      cast_expr_pair_list;
  for (auto p : expr_pair_list) {
    ti.set_notnull(false);
    cast_expr_pair_list.emplace_back(p.first, p.second->add_cast(ti));
  }
  if (else_e != nullptr) {
    else_e = else_e->add_cast(ti);
  } else {
    Datum d;
    // always create an else expr so that executor doesn't need to worry about it
    ti.set_notnull(false);
    else_e = makeExpr<Analyzer::Constant>(ti, true, d);
  }
  if (ti.get_type() == kNULLT) {
    throw std::runtime_error(
        "Cannot deduce the type for case expressions, all branches null");
  }

  auto case_expr = makeExpr<Analyzer::CaseExpr>(ti, has_agg, cast_expr_pair_list, else_e);
  return case_expr;
}

std::string CaseExpr::to_string() const {
  std::string str("CASE ");
  for (auto& p : when_then_list_) {
    str += "WHEN " + p->get_expr1()->to_string() + " THEN " +
           p->get_expr2()->to_string() + " ";
  }
  if (else_expr_ != nullptr) {
    str += "ELSE " + else_expr_->to_string();
  }
  str += " END";
  return str;
}

void UnionQuery::analyze(const Catalog_Namespace::Catalog& catalog,
                         Analyzer::Query& query) const {
  left_->analyze(catalog, query);
  Analyzer::Query* right_query = new Analyzer::Query();
  right_->analyze(catalog, *right_query);
  query.set_next_query(right_query);
  query.set_is_unionall(is_unionall_);
}

void QuerySpec::analyze_having_clause(const Catalog_Namespace::Catalog& catalog,
                                      Analyzer::Query& query) const {
  std::shared_ptr<Analyzer::Expr> p;
  if (having_clause_ != nullptr) {
    p = having_clause_->analyze(catalog, query, Expr::TlistRefType::TLIST_COPY);
    if (p->get_type_info().get_type() != kBOOLEAN) {
      throw std::runtime_error("Only boolean expressions can be in HAVING clause.");
    }
    p->check_group_by(query.get_group_by());
  }
  query.set_having_predicate(p);
}

void QuerySpec::analyze_group_by(const Catalog_Namespace::Catalog& catalog,
                                 Analyzer::Query& query) const {
  std::list<std::shared_ptr<Analyzer::Expr>> groupby;
  if (!groupby_clause_.empty()) {
    int gexpr_no = 1;
    std::shared_ptr<Analyzer::Expr> gexpr;
    const std::vector<std::shared_ptr<Analyzer::TargetEntry>>& tlist =
        query.get_targetlist();
    for (auto& c : groupby_clause_) {
      // special-case ordinal numbers in GROUP BY
      if (dynamic_cast<Literal*>(c.get())) {
        IntLiteral* i = dynamic_cast<IntLiteral*>(c.get());
        if (!i) {
          throw std::runtime_error("Invalid literal in GROUP BY clause.");
        }
        int varno = (int)i->get_intval();
        if (varno <= 0 || varno > static_cast<int>(tlist.size())) {
          throw std::runtime_error("Invalid ordinal number in GROUP BY clause.");
        }
        if (tlist[varno - 1]->get_expr()->get_contains_agg()) {
          throw std::runtime_error(
              "Ordinal number in GROUP BY cannot reference an expression containing "
              "aggregate "
              "functions.");
        }
        gexpr = makeExpr<Analyzer::Var>(
            tlist[varno - 1]->get_expr()->get_type_info(), Analyzer::Var::kOUTPUT, varno);
      } else {
        gexpr = c->analyze(catalog, query, Expr::TlistRefType::TLIST_REF);
      }
      const SQLTypeInfo gti = gexpr->get_type_info();
      bool set_new_type = false;
      SQLTypeInfo ti(gti);
      if (gti.is_string() && gti.get_compression() == kENCODING_NONE) {
        set_new_type = true;
        ti.set_compression(kENCODING_DICT);
        ti.set_comp_param(TRANSIENT_DICT_ID);
        ti.set_fixed_size();
      }
      std::shared_ptr<Analyzer::Var> v;
      if (std::dynamic_pointer_cast<Analyzer::Var>(gexpr)) {
        v = std::static_pointer_cast<Analyzer::Var>(gexpr);
        int n = v->get_varno();
        gexpr = tlist[n - 1]->get_own_expr();
        auto cv = std::dynamic_pointer_cast<Analyzer::ColumnVar>(gexpr);
        if (cv != nullptr) {
          // inherit all ColumnVar info for lineage.
          *std::static_pointer_cast<Analyzer::ColumnVar>(v) = *cv;
        }
        v->set_which_row(Analyzer::Var::kGROUPBY);
        v->set_varno(gexpr_no);
        tlist[n - 1]->set_expr(v);
      }
      if (set_new_type) {
        auto new_e = gexpr->add_cast(ti);
        groupby.push_back(new_e);
        if (v != nullptr) {
          v->set_type_info(new_e->get_type_info());
        }
      } else {
        groupby.push_back(gexpr);
      }
      gexpr_no++;
    }
  }
  if (query.get_num_aggs() > 0 || !groupby.empty()) {
    for (auto t : query.get_targetlist()) {
      auto e = t->get_expr();
      e->check_group_by(groupby);
    }
  }
  query.set_group_by(groupby);
}

void QuerySpec::analyze_where_clause(const Catalog_Namespace::Catalog& catalog,
                                     Analyzer::Query& query) const {
  if (where_clause_ == nullptr) {
    query.set_where_predicate(nullptr);
    return;
  }
  auto p = where_clause_->analyze(catalog, query, Expr::TlistRefType::TLIST_COPY);
  if (p->get_type_info().get_type() != kBOOLEAN) {
    throw std::runtime_error("Only boolean expressions can be in WHERE clause.");
  }
  query.set_where_predicate(p);
}

void QuerySpec::analyze_select_clause(const Catalog_Namespace::Catalog& catalog,
                                      Analyzer::Query& query) const {
  std::vector<std::shared_ptr<Analyzer::TargetEntry>>& tlist =
      query.get_targetlist_nonconst();
  if (select_clause_.empty()) {
    // this means SELECT *
    int rte_idx = 0;
    for (auto rte : query.get_rangetable()) {
      rte->expand_star_in_targetlist(catalog, tlist, rte_idx++);
    }
  } else {
    for (auto& p : select_clause_) {
      const Parser::Expr* select_expr = p->get_select_expr();
      // look for the case of range_var.*
      if (typeid(*select_expr) == typeid(ColumnRef) &&
          dynamic_cast<const ColumnRef*>(select_expr)->get_column() == nullptr) {
        const std::string* range_var_name =
            dynamic_cast<const ColumnRef*>(select_expr)->get_table();
        int rte_idx = query.get_rte_idx(*range_var_name);
        if (rte_idx < 0) {
          throw std::runtime_error("invalid range variable name: " + *range_var_name);
        }
        Analyzer::RangeTableEntry* rte = query.get_rte(rte_idx);
        rte->expand_star_in_targetlist(catalog, tlist, rte_idx);
      } else {
        auto e = select_expr->analyze(catalog, query);
        std::string resname;

        if (p->get_alias() != nullptr) {
          resname = *p->get_alias();
        } else if (std::dynamic_pointer_cast<Analyzer::ColumnVar>(e) &&
                   !std::dynamic_pointer_cast<Analyzer::Var>(e)) {
          auto colvar = std::static_pointer_cast<Analyzer::ColumnVar>(e);
          const auto& column_key = colvar->getColumnKey();
          const ColumnDescriptor* col_desc =
              catalog.getMetadataForColumn(column_key.table_id, column_key.column_id);
          resname = col_desc->columnName;
        }
        if (e->get_type_info().get_type() == kNULLT) {
          throw std::runtime_error(
              "Untyped NULL in SELECT clause.  Use CAST to specify a type.");
        }
        auto o = std::static_pointer_cast<Analyzer::UOper>(e);
        bool unnest = (o != nullptr && o->get_optype() == kUNNEST);
        auto tle = std::make_shared<Analyzer::TargetEntry>(resname, e, unnest);
        tlist.push_back(tle);
      }
    }
  }
}

void QuerySpec::analyze_from_clause(const Catalog_Namespace::Catalog& catalog,
                                    Analyzer::Query& query) const {
  Analyzer::RangeTableEntry* rte;
  for (auto& p : from_clause_) {
    const TableDescriptor* table_desc;
    table_desc = catalog.getMetadataForTable(*p->get_table_name());
    if (table_desc == nullptr) {
      throw std::runtime_error("Table " + *p->get_table_name() + " does not exist.");
    }
    std::string range_var;
    if (p->get_range_var() == nullptr) {
      range_var = *p->get_table_name();
    } else {
      range_var = *p->get_range_var();
    }
    rte = new Analyzer::RangeTableEntry(range_var, table_desc, nullptr);
    query.add_rte(rte);
  }
}

void QuerySpec::analyze(const Catalog_Namespace::Catalog& catalog,
                        Analyzer::Query& query) const {
  query.set_is_distinct(is_distinct_);
  analyze_from_clause(catalog, query);
  analyze_select_clause(catalog, query);
  analyze_where_clause(catalog, query);
  analyze_group_by(catalog, query);
  analyze_having_clause(catalog, query);
}

namespace {

// clean known escape'd chars without having to do a full json parse
std::string unescape(std::string s) {
  boost::replace_all(s, "\\\\t", "\t");
  boost::replace_all(s, "\\t", "\t");
  boost::replace_all(s, "\\\\n", "\n");
  boost::replace_all(s, "\\n", "\n");

  // handle numerics
  std::smatch m;

  // "\x00"
  std::regex e1("(\\\\x[0-9A-Fa-f][0-9A-Fa-f])");
  while (std::regex_search(s, m, e1)) {
    std::string original(m[0].first, m[0].second);
    std::string replacement;
    long val = strtol(original.substr(2, 2).c_str(), NULL, 16);
    replacement.push_back(val);
    boost::replace_all(s, original, replacement);
  }

  // "\u0000"
  std::regex e2("(\\\\u[0-9A-Fa-f][0-9A-Fa-f][0-9A-Fa-f][0-9A-Fa-f])");
  while (std::regex_search(s, m, e2)) {
    std::string original(m[0].first, m[0].second);
    std::string replacement;
    long val = strtol(original.substr(2, 4).c_str(), NULL, 16);
    replacement.push_back(val);
    boost::replace_all(s, original, replacement);
  }

  return s;
}

void parse_options(const rapidjson::Value& payload,
                   std::list<std::unique_ptr<NameValueAssign>>& nameValueList,
                   bool stringToNull = false,
                   bool stringToInteger = false) {
  if (payload.HasMember("options") && payload["options"].IsObject()) {
    const auto& options = payload["options"];
    for (auto itr = options.MemberBegin(); itr != options.MemberEnd(); ++itr) {
      auto option_name = std::make_unique<std::string>(itr->name.GetString());
      std::unique_ptr<Literal> literal_value;
      if (itr->value.IsString()) {
        std::string str = itr->value.GetString();
        if (stringToNull && str == "") {
          literal_value = std::make_unique<NullLiteral>();
        } else if (stringToInteger && std::all_of(str.begin(), str.end(), ::isdigit)) {
          int iVal = std::stoi(str);
          literal_value = std::make_unique<IntLiteral>(iVal);
        } else {
          // Rapidjson will deliberately provide escape'd strings when accessed
          //   ... but the literal should have a copy of the raw unescaped string
          auto unique_literal_string = std::make_unique<std::string>(unescape(str));
          literal_value =
              std::make_unique<StringLiteral>(unique_literal_string.release());
        }
      } else if (itr->value.IsInt() || itr->value.IsInt64()) {
        literal_value = std::make_unique<IntLiteral>(json_i64(itr->value));
      } else if (itr->value.IsNull()) {
        literal_value = std::make_unique<NullLiteral>();
      } else {
        throw std::runtime_error("Unable to handle literal for " + *option_name);
      }
      CHECK(literal_value);

      nameValueList.emplace_back(std::make_unique<NameValueAssign>(
          option_name.release(), literal_value.release()));
    }
  }
}
}  // namespace

void SelectStmt::analyze(const Catalog_Namespace::Catalog& catalog,
                         Analyzer::Query& query) const {
  query.set_stmt_type(kSELECT);
  query.set_limit(limit_);
  if (offset_ < 0) {
    throw std::runtime_error("OFFSET cannot be negative.");
  }
  query.set_offset(offset_);
  query_expr_->analyze(catalog, query);
  if (orderby_clause_.empty() && !query.get_is_distinct()) {
    query.set_order_by(nullptr);
    return;
  }
  const std::vector<std::shared_ptr<Analyzer::TargetEntry>>& tlist =
      query.get_targetlist();
  std::list<Analyzer::OrderEntry>* order_by = new std::list<Analyzer::OrderEntry>();
  if (!orderby_clause_.empty()) {
    for (auto& p : orderby_clause_) {
      int tle_no = p->get_colno();
      if (tle_no == 0) {
        // use column name
        // search through targetlist for matching name
        const std::string* name = p->get_column()->get_column();
        tle_no = 1;
        bool found = false;
        for (auto tle : tlist) {
          if (tle->get_resname() == *name) {
            found = true;
            break;
          }
          tle_no++;
        }
        if (!found) {
          throw std::runtime_error("invalid name in order by: " + *name);
        }
      }
      order_by->push_back(
          Analyzer::OrderEntry(tle_no, p->get_is_desc(), p->get_nulls_first()));
    }
  }
  if (query.get_is_distinct()) {
    // extend order_by to include all targetlist entries.
    for (int i = 1; i <= static_cast<int>(tlist.size()); i++) {
      bool in_orderby = false;
      std::for_each(order_by->begin(),
                    order_by->end(),
                    [&in_orderby, i](const Analyzer::OrderEntry& oe) {
                      in_orderby = in_orderby || (i == oe.tle_no);
                    });
      if (!in_orderby) {
        order_by->push_back(Analyzer::OrderEntry(i, false, false));
      }
    }
  }
  query.set_order_by(order_by);
}

std::string SelectEntry::to_string() const {
  std::string str = select_expr_->to_string();
  if (alias_ != nullptr) {
    str += " AS " + *alias_;
  }
  return str;
}

std::string TableRef::to_string() const {
  std::string str = *table_name_;
  if (range_var_ != nullptr) {
    str += " " + *range_var_;
  }
  return str;
}

std::string ColumnRef::to_string() const {
  std::string str;
  if (table_ == nullptr) {
    str = *column_;
  } else if (column_ == nullptr) {
    str = *table_ + ".*";
  } else {
    str = *table_ + "." + *column_;
  }
  return str;
}

std::string OperExpr::to_string() const {
  std::string op_str[] = {
      "=", "===", "<>", "<", ">", "<=", ">=", " AND ", " OR ", "NOT", "-", "+", "*", "/"};
  std::string str;
  if (optype_ == kUMINUS) {
    str = "-(" + left_->to_string() + ")";
  } else if (optype_ == kNOT) {
    str = "NOT (" + left_->to_string() + ")";
  } else if (optype_ == kARRAY_AT) {
    str = left_->to_string() + "[" + right_->to_string() + "]";
  } else if (optype_ == kUNNEST) {
    str = "UNNEST(" + left_->to_string() + ")";
  } else if (optype_ == kIN) {
    str = "(" + left_->to_string() + " IN " + right_->to_string() + ")";
  } else {
    str = "(" + left_->to_string() + op_str[optype_] + right_->to_string() + ")";
  }
  return str;
}

std::string InExpr::to_string() const {
  std::string str = arg_->to_string();
  if (is_not_) {
    str += " NOT IN ";
  } else {
    str += " IN ";
  }
  return str;
}

std::string ExistsExpr::to_string() const {
  return "EXISTS (" + query_->to_string() + ")";
}

std::string SubqueryExpr::to_string() const {
  std::string str;
  str = "(";
  str += query_->to_string();
  str += ")";
  return str;
}

std::string IsNullExpr::to_string() const {
  std::string str = arg_->to_string();
  if (is_not_) {
    str += " IS NOT NULL";
  } else {
    str += " IS NULL";
  }
  return str;
}

std::string InSubquery::to_string() const {
  std::string str = InExpr::to_string();
  str += subquery_->to_string();
  return str;
}

std::string InValues::to_string() const {
  std::string str = InExpr::to_string() + "(";
  bool notfirst = false;
  for (auto& p : value_list_) {
    if (notfirst) {
      str += ", ";
    } else {
      notfirst = true;
    }
    str += p->to_string();
  }
  str += ")";
  return str;
}

std::string BetweenExpr::to_string() const {
  std::string str = arg_->to_string();
  if (is_not_) {
    str += " NOT BETWEEN ";
  } else {
    str += " BETWEEN ";
  }
  str += lower_->to_string() + " AND " + upper_->to_string();
  return str;
}

std::string CharLengthExpr::to_string() const {
  std::string str;
  if (calc_encoded_length_) {
    str = "CHAR_LENGTH (" + arg_->to_string() + ")";
  } else {
    str = "LENGTH (" + arg_->to_string() + ")";
  }
  return str;
}

std::string CardinalityExpr::to_string() const {
  std::string str = "CARDINALITY(" + arg_->to_string() + ")";
  return str;
}

std::string LikeExpr::to_string() const {
  std::string str = arg_->to_string();
  if (is_not_) {
    str += " NOT LIKE ";
  } else {
    str += " LIKE ";
  }
  str += like_string_->to_string();
  if (escape_string_ != nullptr) {
    str += " ESCAPE " + escape_string_->to_string();
  }
  return str;
}

std::string RegexpExpr::to_string() const {
  std::string str = arg_->to_string();
  if (is_not_) {
    str += " NOT REGEXP ";
  } else {
    str += " REGEXP ";
  }
  str += pattern_string_->to_string();
  if (escape_string_ != nullptr) {
    str += " ESCAPE " + escape_string_->to_string();
  }
  return str;
}

std::string WidthBucketExpr::to_string() const {
  std::string str = " WIDTH_BUCKET ";
  str += target_value_->to_string();
  str += " ";
  str += lower_bound_->to_string();
  str += " ";
  str += upper_bound_->to_string();
  str += " ";
  str += partition_count_->to_string();
  str += " ";
  return str;
}

std::string LikelihoodExpr::to_string() const {
  std::string str = " LIKELIHOOD ";
  str += arg_->to_string();
  str += " ";
  str += boost::lexical_cast<std::string>(is_not_ ? 1.0 - likelihood_ : likelihood_);
  return str;
}

std::string FunctionRef::to_string() const {
  std::string str = *name_ + "(";
  if (distinct_) {
    str += "DISTINCT ";
  }
  if (arg_ == nullptr) {
    str += "*)";
  } else {
    str += arg_->to_string() + ")";
  }
  return str;
}

std::string QuerySpec::to_string() const {
  std::string query_str = "SELECT ";
  if (is_distinct_) {
    query_str += "DISTINCT ";
  }
  if (select_clause_.empty()) {
    query_str += "* ";
  } else {
    bool notfirst = false;
    for (auto& p : select_clause_) {
      if (notfirst) {
        query_str += ", ";
      } else {
        notfirst = true;
      }
      query_str += p->to_string();
    }
  }
  query_str += " FROM ";
  bool notfirst = false;
  for (auto& p : from_clause_) {
    if (notfirst) {
      query_str += ", ";
    } else {
      notfirst = true;
    }
    query_str += p->to_string();
  }
  if (where_clause_) {
    query_str += " WHERE " + where_clause_->to_string();
  }
  if (!groupby_clause_.empty()) {
    query_str += " GROUP BY ";
    bool notfirst = false;
    for (auto& p : groupby_clause_) {
      if (notfirst) {
        query_str += ", ";
      } else {
        notfirst = true;
      }
      query_str += p->to_string();
    }
  }
  if (having_clause_) {
    query_str += " HAVING " + having_clause_->to_string();
  }
  query_str += ";";
  return query_str;
}

void InsertStmt::analyze(const Catalog_Namespace::Catalog& catalog,
                         Analyzer::Query& query) const {
  query.set_stmt_type(kINSERT);
  const TableDescriptor* td = catalog.getMetadataForTable(*table_);
  if (td == nullptr) {
    throw std::runtime_error("Table " + *table_ + " does not exist.");
  }
  if (td->isView) {
    throw std::runtime_error("Insert to views is not supported yet.");
  }
  foreign_storage::validate_non_foreign_table_write(td);
  query.set_result_table_id(td->tableId);
  std::list<int> result_col_list;
  if (column_list_.empty()) {
    const std::list<const ColumnDescriptor*> all_cols =
        catalog.getAllColumnMetadataForTable(td->tableId, false, false, true);
    for (auto cd : all_cols) {
      result_col_list.push_back(cd->columnId);
    }
  } else {
    for (auto& c : column_list_) {
      const ColumnDescriptor* cd = catalog.getMetadataForColumn(td->tableId, *c);
      if (cd == nullptr) {
        throw std::runtime_error("Column " + *c + " does not exist.");
      }
      result_col_list.push_back(cd->columnId);
      const auto& col_ti = cd->columnType;
      if (col_ti.get_physical_cols() > 0) {
        CHECK(cd->columnType.is_geometry());
        for (auto i = 1; i <= col_ti.get_physical_cols(); i++) {
          const ColumnDescriptor* pcd =
              catalog.getMetadataForColumn(td->tableId, cd->columnId + i);
          if (pcd == nullptr) {
            throw std::runtime_error("Column " + *c + "'s metadata is incomplete.");
          }
          result_col_list.push_back(pcd->columnId);
        }
      }
    }
  }
  query.set_result_col_list(result_col_list);
}

namespace {
Literal* parse_insert_literal(const rapidjson::Value& literal) {
  CHECK(literal.IsObject());
  CHECK(literal.HasMember("literal"));
  CHECK(literal.HasMember("type"));
  auto type = json_str(literal["type"]);
  if (type == "NULL") {
    return new NullLiteral();
  } else if (type == "CHAR" || type == "BOOLEAN") {
    auto* val = new std::string(json_str(literal["literal"]));
    return new StringLiteral(val);
  } else if (type == "DECIMAL") {
    CHECK(literal.HasMember("scale"));
    CHECK(literal.HasMember("precision"));
    auto scale = json_i64(literal["scale"]);
    auto precision = json_i64(literal["precision"]);
    if (scale == 0) {
      auto int_val = std::stol(json_str(literal["literal"]));
      return new IntLiteral(int_val);
    } else if (precision > sql_constants::kMaxNumericPrecision) {
      auto dbl_val = std::stod(json_str(literal["literal"]));
      return new DoubleLiteral(dbl_val);
    } else {
      auto* val = new std::string(json_str(literal["literal"]));
      return new FixedPtLiteral(val);
    }
  } else if (type == "DOUBLE") {
    auto dbl_val = std::stod(json_str(literal["literal"]));
    return new DoubleLiteral(dbl_val);
  } else {
    CHECK(false) << "Unexpected calcite data type: " << type;
  }
  return nullptr;
}

ArrayLiteral* parse_insert_array_literal(const rapidjson::Value& array) {
  CHECK(array.IsArray());
  auto json_elements = array.GetArray();
  auto* elements = new std::list<Expr*>();
  for (const auto& e : json_elements) {
    elements->push_back(parse_insert_literal(e));
  }
  return new ArrayLiteral(elements);
}
}  // namespace

InsertValuesStmt::InsertValuesStmt(const rapidjson::Value& payload)
    : InsertStmt(nullptr, nullptr) {
  CHECK(payload.HasMember("name"));
  table_ = std::make_unique<std::string>(json_str(payload["name"]));

  if (payload.HasMember("columns")) {
    CHECK(payload["columns"].IsArray());
    for (auto& column : payload["columns"].GetArray()) {
      std::string s = json_str(column);
      column_list_.emplace_back(std::make_unique<std::string>(s));
    }
  }

  CHECK(payload.HasMember("values") && payload["values"].IsArray());
  auto tuples = payload["values"].GetArray();
  if (tuples.Empty()) {
    throw std::runtime_error("Values statement cannot be empty");
  }
  values_lists_.reserve(tuples.Size());
  for (const auto& json_tuple : tuples) {
    auto values_list = std::make_unique<ValuesList>();
    CHECK(json_tuple.IsArray());
    auto tuple = json_tuple.GetArray();
    for (const auto& value : tuple) {
      CHECK(value.IsObject());
      if (value.HasMember("array")) {
        values_list->push_back(parse_insert_array_literal(value["array"]));
      } else {
        values_list->push_back(parse_insert_literal(value));
      }
    }
    values_lists_.push_back(std::move(values_list));
  }
}

void InsertValuesStmt::analyze(const Catalog_Namespace::Catalog& catalog,
                               Analyzer::Query& query) const {
  InsertStmt::analyze(catalog, query);
  size_t list_size = values_lists_[0]->get_value_list().size();
  if (!column_list_.empty()) {
    if (list_size != column_list_.size()) {
      throw std::runtime_error(
          "Numbers of columns and values don't match for the "
          "insert.");
    }
  } else {
    const auto tableId = query.get_result_table_id();
    const std::list<const ColumnDescriptor*> non_phys_cols =
        catalog.getAllColumnMetadataForTable(tableId, false, false, false);
    if (non_phys_cols.size() != list_size) {
      throw std::runtime_error(
          "Number of columns in table does not match the list of values given in the "
          "insert.");
    }
  }
  std::vector<const ColumnDescriptor*> cds;
  cds.reserve(query.get_result_col_list().size());
  for (auto id : query.get_result_col_list()) {
    const auto* cd = catalog.getMetadataForColumn(query.get_result_table_id(), id);
    CHECK(cd);
    cds.push_back(cd);
  }
  auto& query_values_lists = query.get_values_lists();
  query_values_lists.resize(values_lists_.size());
  for (size_t i = 0; i < values_lists_.size(); ++i) {
    const auto& values_list = values_lists_[i]->get_value_list();
    if (values_list.size() != list_size) {
      throw std::runtime_error(
          "Insert values lists should be of the same size. Expected: " +
          std::to_string(list_size) + ", Got: " + std::to_string(values_list.size()));
    }
    auto& query_values_list = query_values_lists[i];
    size_t cds_id = 0;
    for (auto& v : values_list) {
      auto e = v->analyze(catalog, query);
      const auto* cd = cds[cds_id];
      const auto& col_ti = cd->columnType;
      if (col_ti.get_notnull()) {
        auto c = std::dynamic_pointer_cast<Analyzer::Constant>(e);
        if (c != nullptr && c->get_is_null()) {
          throw std::runtime_error("Cannot insert NULL into column " + cd->columnName);
        }
      }
      e = e->add_cast(col_ti);
      query_values_list.emplace_back(new Analyzer::TargetEntry("", e, false));
      ++cds_id;

      if (col_ti.get_physical_cols() > 0) {
        CHECK(cd->columnType.is_geometry());
        auto c = dynamic_cast<const Analyzer::Constant*>(e.get());
        if (!c) {
          auto uoper = std::dynamic_pointer_cast<Analyzer::UOper>(e);
          if (uoper && uoper->get_optype() == kCAST) {
            c = dynamic_cast<const Analyzer::Constant*>(uoper->get_operand());
          }
        }
        bool is_null = false;
        std::string* geo_string{nullptr};
        if (c) {
          is_null = c->get_is_null();
          if (!is_null) {
            geo_string = c->get_constval().stringval;
          }
        }
        if (!is_null && !geo_string) {
          throw std::runtime_error("Expecting a WKT or WKB hex string for column " +
                                   cd->columnName);
        }
        std::vector<double> coords;
        std::vector<double> bounds;
        std::vector<int> ring_sizes;
        std::vector<int> poly_rings;
        int render_group =
            0;  // @TODO simon.eves where to get render_group from in this context?!
        SQLTypeInfo import_ti{cd->columnType};
        if (!is_null) {
          if (!Geospatial::GeoTypesFactory::getGeoColumns(
                  *geo_string, import_ti, coords, bounds, ring_sizes, poly_rings)) {
            throw std::runtime_error("Cannot read geometry to insert into column " +
                                     cd->columnName);
          }
          if (coords.empty()) {
            // Importing from geo_string WKT resulted in empty coords: dealing with a NULL
            is_null = true;
          }
          if (cd->columnType.get_type() != import_ti.get_type()) {
            // allow POLYGON to be inserted into MULTIPOLYGON column
            if (!(import_ti.get_type() == SQLTypes::kPOLYGON &&
                  cd->columnType.get_type() == SQLTypes::kMULTIPOLYGON)) {
              throw std::runtime_error(
                  "Imported geometry doesn't match the type of column " + cd->columnName);
            }
          }
        } else {
          // Special case for NULL POINT, push NULL representation to coords
          if (cd->columnType.get_type() == kPOINT) {
            if (!coords.empty()) {
              throw std::runtime_error(
                  "NULL POINT with unexpected coordinates in column " + cd->columnName);
            }
            coords.push_back(NULL_ARRAY_DOUBLE);
            coords.push_back(NULL_DOUBLE);
          }
        }

        // TODO: check if import SRID matches columns SRID, may need to transform before
        // inserting

        const auto* cd_coords = cds[cds_id];
        CHECK_EQ(cd_coords->columnType.get_type(), kARRAY);
        CHECK_EQ(cd_coords->columnType.get_subtype(), kTINYINT);
        std::list<std::shared_ptr<Analyzer::Expr>> value_exprs;
        if (!is_null || cd->columnType.get_type() == kPOINT) {
          auto compressed_coords = Geospatial::compress_coords(coords, col_ti);
          for (auto cc : compressed_coords) {
            Datum d;
            d.tinyintval = cc;
            auto e = makeExpr<Analyzer::Constant>(kTINYINT, false, d);
            value_exprs.push_back(e);
          }
        }
        query_values_list.emplace_back(new Analyzer::TargetEntry(
            "",
            makeExpr<Analyzer::Constant>(cd_coords->columnType, is_null, value_exprs),
            false));
        ++cds_id;

        if (cd->columnType.get_type() == kMULTILINESTRING ||
            cd->columnType.get_type() == kPOLYGON ||
            cd->columnType.get_type() == kMULTIPOLYGON) {
          // Put [linest]ring sizes array into separate physical column
          const auto* cd_ring_sizes = cds[cds_id];
          CHECK(cd_ring_sizes);
          CHECK_EQ(cd_ring_sizes->columnType.get_type(), kARRAY);
          CHECK_EQ(cd_ring_sizes->columnType.get_subtype(), kINT);
          std::list<std::shared_ptr<Analyzer::Expr>> value_exprs;
          if (!is_null) {
            for (auto c : ring_sizes) {
              Datum d;
              d.intval = c;
              auto e = makeExpr<Analyzer::Constant>(kINT, false, d);
              value_exprs.push_back(e);
            }
          }
          query_values_list.emplace_back(new Analyzer::TargetEntry(
              "",
              makeExpr<Analyzer::Constant>(
                  cd_ring_sizes->columnType, is_null, value_exprs),
              false));
          ++cds_id;

          if (cd->columnType.get_type() == kMULTIPOLYGON) {
            // Put poly_rings array into separate physical column
            const auto* cd_poly_rings = cds[cds_id];
            CHECK(cd_poly_rings);
            CHECK_EQ(cd_poly_rings->columnType.get_type(), kARRAY);
            CHECK_EQ(cd_poly_rings->columnType.get_subtype(), kINT);
            std::list<std::shared_ptr<Analyzer::Expr>> value_exprs;
            if (!is_null) {
              for (auto c : poly_rings) {
                Datum d;
                d.intval = c;
                auto e = makeExpr<Analyzer::Constant>(kINT, false, d);
                value_exprs.push_back(e);
              }
            }
            query_values_list.emplace_back(new Analyzer::TargetEntry(
                "",
                makeExpr<Analyzer::Constant>(
                    cd_poly_rings->columnType, is_null, value_exprs),
                false));
            ++cds_id;
          }
        }

        if (cd->columnType.get_type() == kMULTIPOINT ||
            cd->columnType.get_type() == kLINESTRING ||
            cd->columnType.get_type() == kMULTILINESTRING ||
            cd->columnType.get_type() == kPOLYGON ||
            cd->columnType.get_type() == kMULTIPOLYGON) {
          const auto* cd_bounds = cds[cds_id];
          CHECK(cd_bounds);
          CHECK_EQ(cd_bounds->columnType.get_type(), kARRAY);
          CHECK_EQ(cd_bounds->columnType.get_subtype(), kDOUBLE);
          std::list<std::shared_ptr<Analyzer::Expr>> value_exprs;
          if (!is_null) {
            for (auto b : bounds) {
              Datum d;
              d.doubleval = b;
              auto e = makeExpr<Analyzer::Constant>(kDOUBLE, false, d);
              value_exprs.push_back(e);
            }
          }
          query_values_list.emplace_back(new Analyzer::TargetEntry(
              "",
              makeExpr<Analyzer::Constant>(cd_bounds->columnType, is_null, value_exprs),
              false));
          ++cds_id;
        }

        if (cd->columnType.get_type() == kPOLYGON ||
            cd->columnType.get_type() == kMULTIPOLYGON) {
          // Put render group into separate physical column
          const auto* cd_render_group = cds[cds_id];
          CHECK(cd_render_group);
          CHECK_EQ(cd_render_group->columnType.get_type(), kINT);
          Datum d;
          d.intval = render_group;
          query_values_list.emplace_back(new Analyzer::TargetEntry(
              "",
              makeExpr<Analyzer::Constant>(cd_render_group->columnType, is_null, d),
              false));
          ++cds_id;
        }
      }
    }
  }
}

void InsertValuesStmt::execute(const Catalog_Namespace::SessionInfo& session,
                               bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("INSERT values invalid in read only mode.");
  }
  auto execute_read_lock =
      heavyai::shared_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
          *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
              legacylockmgr::ExecutorOuterLock, true));
  auto& catalog = session.getCatalog();
  const auto td_with_lock =
      lockmgr::TableSchemaLockContainer<lockmgr::ReadLock>::acquireTableDescriptor(
          catalog, *table_);
  if (!session.checkDBAccessPrivileges(DBObjectType::TableDBObjectType,
                                       AccessPrivileges::INSERT_INTO_TABLE,
                                       *table_)) {
    throw std::runtime_error("User has no insert privileges on " + *table_ + ".");
  }
  Analyzer::Query query;
  analyze(catalog, query);

  // Take an insert data write lock, which prevents concurrent inserts.
  const auto insert_data_lock =
      lockmgr::InsertDataLockMgr::getWriteLockForTable(catalog, *table_);

  // NOTE(max): we do the same checks as below just a few calls earlier in analyze().
  // Do we keep those intentionally to make sure nothing changed in between w/o
  // catalog locks or is it just a duplicate work?
  auto td = td_with_lock();
  CHECK(td);
  if (td->isView) {
    throw std::runtime_error("Singleton inserts on views is not supported.");
  }
  foreign_storage::validate_non_foreign_table_write(td);

  auto executor = Executor::getExecutor(Executor::UNITARY_EXECUTOR_ID);
  RelAlgExecutor ra_executor(executor.get());

  if (!leafs_connector_) {
    leafs_connector_ = std::make_unique<Fragmenter_Namespace::LocalInsertConnector>();
  }
  Fragmenter_Namespace::InsertDataLoader insert_data_loader(*leafs_connector_);
  try {
    ra_executor.executeSimpleInsert(query, insert_data_loader, session);
  } catch (...) {
    try {
      leafs_connector_->rollback(session, td->tableId);
    } catch (std::exception& e) {
      LOG(ERROR) << "An error occurred during insert rollback attempt. Table id: "
                 << td->tableId << ", Error: " << e.what();
    }
    throw;
  }
  if (!td->isTemporaryTable()) {
    leafs_connector_->checkpoint(session, td->tableId);
  }
}

void UpdateStmt::analyze(const Catalog_Namespace::Catalog& catalog,
                         Analyzer::Query& query) const {
  throw std::runtime_error("UPDATE statement not supported yet.");
}

void DeleteStmt::analyze(const Catalog_Namespace::Catalog& catalog,
                         Analyzer::Query& query) const {
  throw std::runtime_error("DELETE statement not supported yet.");
}

namespace {

void validate_shard_column_type(const ColumnDescriptor& cd) {
  const auto& col_ti = cd.columnType;
  if (!col_ti.is_integer() && !col_ti.is_time() &&
      !(col_ti.is_string() && col_ti.get_compression() == kENCODING_DICT)) {
    throw std::runtime_error("Cannot shard on type " + col_ti.get_type_name() +
                             ", encoding " + col_ti.get_compression_name());
  }
}

size_t shard_column_index(const std::string& name,
                          const std::list<ColumnDescriptor>& columns) {
  size_t index = 1;
  for (const auto& cd : columns) {
    if (cd.columnName == name) {
      validate_shard_column_type(cd);
      return index;
    }
    ++index;
    if (cd.columnType.is_geometry()) {
      index += cd.columnType.get_physical_cols();
    }
  }
  // Not found, return 0
  return 0;
}

size_t sort_column_index(const std::string& name,
                         const std::list<ColumnDescriptor>& columns) {
  size_t index = 1;
  for (const auto& cd : columns) {
    if (boost::to_upper_copy<std::string>(cd.columnName) == name) {
      return index;
    }
    ++index;
    if (cd.columnType.is_geometry()) {
      index += cd.columnType.get_physical_cols();
    }
  }
  // Not found, return 0
  return 0;
}

void set_string_field(rapidjson::Value& obj,
                      const std::string& field_name,
                      const std::string& field_value,
                      rapidjson::Document& document) {
  rapidjson::Value field_name_json_str;
  field_name_json_str.SetString(
      field_name.c_str(), field_name.size(), document.GetAllocator());
  rapidjson::Value field_value_json_str;
  field_value_json_str.SetString(
      field_value.c_str(), field_value.size(), document.GetAllocator());
  obj.AddMember(field_name_json_str, field_value_json_str, document.GetAllocator());
}

std::string serialize_key_metainfo(
    const ShardKeyDef* shard_key_def,
    const std::vector<SharedDictionaryDef>& shared_dict_defs) {
  rapidjson::Document document;
  auto& allocator = document.GetAllocator();
  rapidjson::Value arr(rapidjson::kArrayType);
  if (shard_key_def) {
    rapidjson::Value shard_key_obj(rapidjson::kObjectType);
    set_string_field(shard_key_obj, "type", "SHARD KEY", document);
    set_string_field(shard_key_obj, "name", shard_key_def->get_column(), document);
    arr.PushBack(shard_key_obj, allocator);
  }
  for (const auto& shared_dict_def : shared_dict_defs) {
    rapidjson::Value shared_dict_obj(rapidjson::kObjectType);
    set_string_field(shared_dict_obj, "type", "SHARED DICTIONARY", document);
    set_string_field(shared_dict_obj, "name", shared_dict_def.get_column(), document);
    set_string_field(
        shared_dict_obj, "foreign_table", shared_dict_def.get_foreign_table(), document);
    set_string_field(shared_dict_obj,
                     "foreign_column",
                     shared_dict_def.get_foreign_column(),
                     document);
    arr.PushBack(shared_dict_obj, allocator);
  }
  rapidjson::StringBuffer buffer;
  rapidjson::Writer<rapidjson::StringBuffer> writer(buffer);
  arr.Accept(writer);
  return buffer.GetString();
}

template <typename LITERAL_TYPE,
          typename ASSIGNMENT,
          typename VALIDATE = DefaultValidate<LITERAL_TYPE>>
decltype(auto) get_property_value(const NameValueAssign* p,
                                  ASSIGNMENT op,
                                  VALIDATE validate = VALIDATE()) {
  const auto val = validate(p);
  return op(val);
}

decltype(auto) get_storage_type(TableDescriptor& td,
                                const NameValueAssign* p,
                                const std::list<ColumnDescriptor>& columns) {
  auto assignment = [&td](const auto val) { td.storageType = val; };
  return get_property_value<StringLiteral, decltype(assignment), CaseSensitiveValidate>(
      p, assignment);
}

decltype(auto) get_frag_size_def(TableDescriptor& td,
                                 const NameValueAssign* p,
                                 const std::list<ColumnDescriptor>& columns) {
  return get_property_value<IntLiteral>(p,
                                        [&td](const auto val) { td.maxFragRows = val; });
}

decltype(auto) get_frag_size_dataframe_def(DataframeTableDescriptor& df_td,
                                           const NameValueAssign* p,
                                           const std::list<ColumnDescriptor>& columns) {
  return get_property_value<IntLiteral>(
      p, [&df_td](const auto val) { df_td.maxFragRows = val; });
}

decltype(auto) get_max_chunk_size_def(TableDescriptor& td,
                                      const NameValueAssign* p,
                                      const std::list<ColumnDescriptor>& columns) {
  return get_property_value<IntLiteral>(p,
                                        [&td](const auto val) { td.maxChunkSize = val; });
}

decltype(auto) get_max_chunk_size_dataframe_def(
    DataframeTableDescriptor& df_td,
    const NameValueAssign* p,
    const std::list<ColumnDescriptor>& columns) {
  return get_property_value<IntLiteral>(
      p, [&df_td](const auto val) { df_td.maxChunkSize = val; });
}

decltype(auto) get_delimiter_def(DataframeTableDescriptor& df_td,
                                 const NameValueAssign* p,
                                 const std::list<ColumnDescriptor>& columns) {
  return get_property_value<StringLiteral>(p, [&df_td](const auto val) {
    if (val.size() != 1) {
      throw std::runtime_error("Length of DELIMITER must be equal to 1.");
    }
    df_td.delimiter = val;
  });
}

decltype(auto) get_header_def(DataframeTableDescriptor& df_td,
                              const NameValueAssign* p,
                              const std::list<ColumnDescriptor>& columns) {
  return get_property_value<StringLiteral>(p, [&df_td](const auto val) {
    if (val == "FALSE") {
      df_td.hasHeader = false;
    } else if (val == "TRUE") {
      df_td.hasHeader = true;
    } else {
      throw std::runtime_error("Option HEADER support only 'true' or 'false' values.");
    }
  });
}

decltype(auto) get_page_size_def(TableDescriptor& td,
                                 const NameValueAssign* p,
                                 const std::list<ColumnDescriptor>& columns) {
  return get_property_value<IntLiteral>(p,
                                        [&td](const auto val) { td.fragPageSize = val; });
}
decltype(auto) get_max_rows_def(TableDescriptor& td,
                                const NameValueAssign* p,
                                const std::list<ColumnDescriptor>& columns) {
  return get_property_value<IntLiteral>(p, [&td](const auto val) { td.maxRows = val; });
}

decltype(auto) get_skip_rows_def(DataframeTableDescriptor& df_td,
                                 const NameValueAssign* p,
                                 const std::list<ColumnDescriptor>& columns) {
  return get_property_value<IntLiteral>(
      p, [&df_td](const auto val) { df_td.skipRows = val; });
}

decltype(auto) get_partions_def(TableDescriptor& td,
                                const NameValueAssign* p,
                                const std::list<ColumnDescriptor>& columns) {
  return get_property_value<StringLiteral>(p, [&td](const auto partitions_uc) {
    if (partitions_uc != "SHARDED" && partitions_uc != "REPLICATED") {
      throw std::runtime_error("PARTITIONS must be SHARDED or REPLICATED");
    }
    if (td.shardedColumnId != 0 && partitions_uc == "REPLICATED") {
      throw std::runtime_error(
          "A table cannot be sharded and replicated at the same time");
    };
    td.partitions = partitions_uc;
  });
}
decltype(auto) get_shard_count_def(TableDescriptor& td,
                                   const NameValueAssign* p,
                                   const std::list<ColumnDescriptor>& columns) {
  if (!td.shardedColumnId) {
    throw std::runtime_error("SHARD KEY must be defined.");
  }
  return get_property_value<IntLiteral>(p, [&td](const auto shard_count) {
    if (g_leaf_count && shard_count % g_leaf_count) {
      throw std::runtime_error(
          "SHARD_COUNT must be a multiple of the number of leaves in the cluster.");
    }
    td.nShards = g_leaf_count ? shard_count / g_leaf_count : shard_count;
    if (!td.shardedColumnId && !td.nShards) {
      throw std::runtime_error(
          "Must specify the number of shards through the SHARD_COUNT option");
    };
  });
}

decltype(auto) get_vacuum_def(TableDescriptor& td,
                              const NameValueAssign* p,
                              const std::list<ColumnDescriptor>& columns) {
  return get_property_value<StringLiteral>(p, [&td](const auto vacuum_uc) {
    if (vacuum_uc != "IMMEDIATE" && vacuum_uc != "DELAYED") {
      throw std::runtime_error("VACUUM must be IMMEDIATE or DELAYED");
    }
    td.hasDeletedCol = boost::iequals(vacuum_uc, "IMMEDIATE") ? false : true;
  });
}

decltype(auto) get_sort_column_def(TableDescriptor& td,
                                   const NameValueAssign* p,
                                   const std::list<ColumnDescriptor>& columns) {
  return get_property_value<StringLiteral>(p, [&td, &columns](const auto sort_upper) {
    td.sortedColumnId = sort_column_index(sort_upper, columns);
    if (!td.sortedColumnId) {
      throw std::runtime_error("Specified sort column " + sort_upper + " doesn't exist");
    }
  });
}

decltype(auto) get_max_rollback_epochs_def(TableDescriptor& td,
                                           const NameValueAssign* p,
                                           const std::list<ColumnDescriptor>& columns) {
  auto assignment = [&td](const auto val) {
    td.maxRollbackEpochs =
        val < 0 ? -1 : val;  // Anything < 0 means unlimited rollbacks. Note that 0
                             // still means keeping a shadow copy of data/metdata
                             // between epochs so bad writes can be rolled back
  };
  return get_property_value<IntLiteral, decltype(assignment), PositiveOrZeroValidate>(
      p, assignment);
}

static const std::map<const std::string, const TableDefFuncPtr> tableDefFuncMap = {
    {"fragment_size"s, get_frag_size_def},
    {"max_chunk_size"s, get_max_chunk_size_def},
    {"page_size"s, get_page_size_def},
    {"max_rows"s, get_max_rows_def},
    {"partitions"s, get_partions_def},
    {"shard_count"s, get_shard_count_def},
    {"vacuum"s, get_vacuum_def},
    {"sort_column"s, get_sort_column_def},
    {"storage_type"s, get_storage_type},
    {"max_rollback_epochs", get_max_rollback_epochs_def}};

void get_table_definitions(TableDescriptor& td,
                           const std::unique_ptr<NameValueAssign>& p,
                           const std::list<ColumnDescriptor>& columns) {
  const auto it = tableDefFuncMap.find(boost::to_lower_copy<std::string>(*p->get_name()));
  if (it == tableDefFuncMap.end()) {
    throw std::runtime_error(
        "Invalid CREATE TABLE option " + *p->get_name() +
        ". Should be FRAGMENT_SIZE, MAX_CHUNK_SIZE, PAGE_SIZE, MAX_ROLLBACK_EPOCHS, "
        "MAX_ROWS, "
        "PARTITIONS, SHARD_COUNT, VACUUM, SORT_COLUMN, STORAGE_TYPE.");
  }
  return it->second(td, p.get(), columns);
}

void get_table_definitions_for_ctas(TableDescriptor& td,
                                    const std::unique_ptr<NameValueAssign>& p,
                                    const std::list<ColumnDescriptor>& columns) {
  const auto it = tableDefFuncMap.find(boost::to_lower_copy<std::string>(*p->get_name()));
  if (it == tableDefFuncMap.end()) {
    throw std::runtime_error(
        "Invalid CREATE TABLE AS option " + *p->get_name() +
        ". Should be FRAGMENT_SIZE, MAX_CHUNK_SIZE, PAGE_SIZE, MAX_ROLLBACK_EPOCHS, "
        "MAX_ROWS, "
        "PARTITIONS, SHARD_COUNT, VACUUM, SORT_COLUMN, STORAGE_TYPE,  "
        "USE_SHARED_DICTIONARIES or FORCE_GEO_COMPRESSION.");
  }
  return it->second(td, p.get(), columns);
}

static const std::map<const std::string, const DataframeDefFuncPtr> dataframeDefFuncMap =
    {{"fragment_size"s, get_frag_size_dataframe_def},
     {"max_chunk_size"s, get_max_chunk_size_dataframe_def},
     {"skip_rows"s, get_skip_rows_def},
     {"delimiter"s, get_delimiter_def},
     {"header"s, get_header_def}};

void get_dataframe_definitions(DataframeTableDescriptor& df_td,
                               const std::unique_ptr<NameValueAssign>& p,
                               const std::list<ColumnDescriptor>& columns) {
  const auto it =
      dataframeDefFuncMap.find(boost::to_lower_copy<std::string>(*p->get_name()));
  if (it == dataframeDefFuncMap.end()) {
    throw std::runtime_error(
        "Invalid CREATE DATAFRAME option " + *p->get_name() +
        ". Should be FRAGMENT_SIZE, MAX_CHUNK_SIZE, SKIP_ROWS, DELIMITER or HEADER.");
  }
  return it->second(df_td, p.get(), columns);
}

std::unique_ptr<ColumnDef> column_from_json(const rapidjson::Value& element) {
  CHECK(element.HasMember("name"));
  auto col_name = std::make_unique<std::string>(json_str(element["name"]));
  CHECK(element.HasMember("sqltype"));
  const auto sql_types = to_sql_type(json_str(element["sqltype"]));

  // decimal / numeric precision / scale
  int precision = -1;
  int scale = -1;
  if (element.HasMember("precision")) {
    precision = json_i64(element["precision"]);
  }
  if (element.HasMember("scale")) {
    scale = json_i64(element["scale"]);
  }

  std::optional<int64_t> array_size;
  if (element.HasMember("arraySize")) {
    // We do not yet support geo arrays
    array_size = json_i64(element["arraySize"]);
  }
  std::unique_ptr<SQLType> sql_type;
  if (element.HasMember("subtype")) {
    CHECK(element.HasMember("coordinateSystem"));
    const auto subtype_sql_types = to_sql_type(json_str(element["subtype"]));
    sql_type =
        std::make_unique<SQLType>(subtype_sql_types,
                                  static_cast<int>(sql_types),
                                  static_cast<int>(json_i64(element["coordinateSystem"])),
                                  false);
  } else if (precision > 0 && scale > 0) {
    sql_type = std::make_unique<SQLType>(sql_types,
                                         precision,
                                         scale,
                                         /*is_array=*/array_size.has_value(),
                                         array_size ? *array_size : -1);
  } else if (precision > 0) {
    sql_type = std::make_unique<SQLType>(sql_types,
                                         precision,
                                         0,
                                         /*is_array=*/array_size.has_value(),
                                         array_size ? *array_size : -1);
  } else {
    sql_type = std::make_unique<SQLType>(sql_types,
                                         /*is_array=*/array_size.has_value(),
                                         array_size ? *array_size : -1);
  }
  CHECK(sql_type);

  CHECK(element.HasMember("nullable"));
  const auto nullable = json_bool(element["nullable"]);
  std::unique_ptr<ColumnConstraintDef> constraint_def;
  StringLiteral* str_literal = nullptr;
  if (element.HasMember("default") && !element["default"].IsNull()) {
    std::string* defaultval = new std::string(json_str(element["default"]));
    boost::algorithm::trim_if(*defaultval, boost::is_any_of(" \"'`"));
    str_literal = new StringLiteral(defaultval);
  }

  constraint_def = std::make_unique<ColumnConstraintDef>(/*notnull=*/!nullable,
                                                         /*unique=*/false,
                                                         /*primarykey=*/false,
                                                         /*defaultval=*/str_literal);
  std::unique_ptr<CompressDef> compress_def;
  if (element.HasMember("encodingType") && !element["encodingType"].IsNull()) {
    std::string encoding_type = json_str(element["encodingType"]);
    CHECK(element.HasMember("encodingSize"));
    auto encoding_name = std::make_unique<std::string>(json_str(element["encodingType"]));
    compress_def = std::make_unique<CompressDef>(encoding_name.release(),
                                                 json_i64(element["encodingSize"]));
  }
  return std::make_unique<ColumnDef>(col_name.release(),
                                     sql_type.release(),
                                     compress_def ? compress_def.release() : nullptr,
                                     constraint_def ? constraint_def.release() : nullptr);
}

void parse_elements(const rapidjson::Value& payload,
                    std::string element_name,
                    std::string& table_name,
                    std::list<std::unique_ptr<TableElement>>& table_element_list) {
  const auto elements = payload[element_name].GetArray();
  for (const auto& element : elements) {
    CHECK(element.IsObject());
    CHECK(element.HasMember("type"));
    if (json_str(element["type"]) == "SQL_COLUMN_DECLARATION") {
      auto col_def = column_from_json(element);
      table_element_list.emplace_back(std::move(col_def));
    } else if (json_str(element["type"]) == "SQL_COLUMN_CONSTRAINT") {
      CHECK(element.HasMember("name"));
      if (json_str(element["name"]) == "SHARD_KEY") {
        CHECK(element.HasMember("columns"));
        CHECK(element["columns"].IsArray());
        const auto& columns = element["columns"].GetArray();
        if (columns.Size() != size_t(1)) {
          throw std::runtime_error("Only one shard column is currently supported.");
        }
        auto shard_key_def = std::make_unique<ShardKeyDef>(json_str(columns[0]));
        table_element_list.emplace_back(std::move(shard_key_def));
      } else if (json_str(element["name"]) == "SHARED_DICT") {
        CHECK(element.HasMember("columns"));
        CHECK(element["columns"].IsArray());
        const auto& columns = element["columns"].GetArray();
        if (columns.Size() != size_t(1)) {
          throw std::runtime_error(
              R"(Only one column per shared dictionary entry is currently supported. Use multiple SHARED DICT statements to share dictionaries from multiple columns.)");
        }
        CHECK(element.HasMember("references") && element["references"].IsObject());
        const auto& references = element["references"].GetObject();
        std::string references_table_name;
        if (references.HasMember("table")) {
          references_table_name = json_str(references["table"]);
        } else {
          references_table_name = table_name;
        }
        CHECK(references.HasMember("column"));

        auto shared_dict_def = std::make_unique<SharedDictionaryDef>(
            json_str(columns[0]), references_table_name, json_str(references["column"]));
        table_element_list.emplace_back(std::move(shared_dict_def));

      } else {
        LOG(FATAL) << "Unsupported type for SQL_COLUMN_CONSTRAINT: "
                   << json_str(element["name"]);
      }
    } else {
      LOG(FATAL) << "Unsupported element type for CREATE TABLE: "
                 << element["type"].GetString();
    }
  }
}
}  // namespace

CreateTableStmt::CreateTableStmt(const rapidjson::Value& payload) {
  CHECK(payload.HasMember("name"));
  table_ = std::make_unique<std::string>(json_str(payload["name"]));
  CHECK(payload.HasMember("elements"));
  CHECK(payload["elements"].IsArray());

  is_temporary_ = false;
  if (payload.HasMember("temporary")) {
    is_temporary_ = json_bool(payload["temporary"]);
  }

  if_not_exists_ = false;
  if (payload.HasMember("ifNotExists")) {
    if_not_exists_ = json_bool(payload["ifNotExists"]);
  }

  parse_elements(payload, "elements", *table_, table_element_list_);

  parse_options(payload, storage_options_);
}

void CreateTableStmt::executeDryRun(const Catalog_Namespace::SessionInfo& session,
                                    TableDescriptor& td,
                                    std::list<ColumnDescriptor>& columns,
                                    std::vector<SharedDictionaryDef>& shared_dict_defs) {
  std::unordered_set<std::string> uc_col_names;
  const auto& catalog = session.getCatalog();
  const ShardKeyDef* shard_key_def{nullptr};
  for (auto& e : table_element_list_) {
    if (dynamic_cast<SharedDictionaryDef*>(e.get())) {
      auto shared_dict_def = static_cast<SharedDictionaryDef*>(e.get());
      validate_shared_dictionary(
          this, shared_dict_def, columns, shared_dict_defs, catalog);
      shared_dict_defs.push_back(*shared_dict_def);
      continue;
    }
    if (dynamic_cast<ShardKeyDef*>(e.get())) {
      if (shard_key_def) {
        throw std::runtime_error("Specified more than one shard key");
      }
      shard_key_def = static_cast<const ShardKeyDef*>(e.get());
      continue;
    }
    if (!dynamic_cast<ColumnDef*>(e.get())) {
      throw std::runtime_error("Table constraints are not supported yet.");
    }
    ColumnDef* coldef = static_cast<ColumnDef*>(e.get());
    ColumnDescriptor cd;
    cd.columnName = *coldef->get_column_name();
    ddl_utils::validate_non_duplicate_column(cd.columnName, uc_col_names);
    setColumnDescriptor(cd, coldef);
    columns.push_back(cd);
  }

  ddl_utils::set_default_table_attributes(*table_, td, columns.size());

  if (shard_key_def) {
    td.shardedColumnId = shard_column_index(shard_key_def->get_column(), columns);
    if (!td.shardedColumnId) {
      throw std::runtime_error("Specified shard column " + shard_key_def->get_column() +
                               " doesn't exist");
    }
  }
  if (is_temporary_) {
    td.persistenceLevel = Data_Namespace::MemoryLevel::CPU_LEVEL;
  } else {
    td.persistenceLevel = Data_Namespace::MemoryLevel::DISK_LEVEL;
  }
  if (!storage_options_.empty()) {
    for (auto& p : storage_options_) {
      get_table_definitions(td, p, columns);
    }
  }
  if (td.shardedColumnId && !td.nShards) {
    throw std::runtime_error("SHARD_COUNT needs to be specified with SHARD_KEY.");
  }
  td.keyMetainfo = serialize_key_metainfo(shard_key_def, shared_dict_defs);
}

void CreateTableStmt::execute(const Catalog_Namespace::SessionInfo& session,
                              bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("CREATE TABLE invalid in read only mode.");
  }
  auto& catalog = session.getCatalog();

  // Until we create the table we don't have a table descriptor to lock and guarantee
  // exclusive use of.  Because of that we need a global write lock to make sure we have
  // exclusive access to the system for now.
  const auto execute_write_lock =
      heavyai::unique_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
          *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
              legacylockmgr::ExecutorOuterLock, true));

  // check access privileges
  if (!session.checkDBAccessPrivileges(DBObjectType::TableDBObjectType,
                                       AccessPrivileges::CREATE_TABLE)) {
    throw std::runtime_error("Table " + *table_ +
                             " will not be created. User has no create privileges.");
  }

  if (!catalog.validateNonExistentTableOrView(*table_, if_not_exists_)) {
    return;
  }

  TableDescriptor td;
  std::list<ColumnDescriptor> columns;
  std::vector<SharedDictionaryDef> shared_dict_defs;

  executeDryRun(session, td, columns, shared_dict_defs);
  td.userId = session.get_currentUser().userId;

  catalog.createShardedTable(td, columns, shared_dict_defs);
  // TODO (max): It's transactionally unsafe, should be fixed: we may create object w/o
  // privileges
  SysCatalog::instance().createDBObject(
      session.get_currentUser(), td.tableName, TableDBObjectType, catalog);
}

CreateDataframeStmt::CreateDataframeStmt(const rapidjson::Value& payload) {
  CHECK(payload.HasMember("name"));
  table_ = std::make_unique<std::string>(json_str(payload["name"]));

  CHECK(payload.HasMember("elementList"));
  parse_elements(payload, "elementList", *table_, table_element_list_);

  CHECK(payload.HasMember("filePath"));
  std::string fs = json_str(payload["filePath"]);
  // strip leading/trailing spaces/quotes/single quotes
  boost::algorithm::trim_if(fs, boost::is_any_of(" \"'`"));
  filename_ = std::make_unique<std::string>(fs);

  parse_options(payload, storage_options_);
}

void CreateDataframeStmt::execute(const Catalog_Namespace::SessionInfo& session,
                                  bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("CREATE DATAFRAME invalid in read only mode.");
  }
  auto& catalog = session.getCatalog();

  const auto execute_write_lock =
      heavyai::unique_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
          *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
              legacylockmgr::ExecutorOuterLock, true));

  // check access privileges
  if (!session.checkDBAccessPrivileges(DBObjectType::TableDBObjectType,
                                       AccessPrivileges::CREATE_TABLE)) {
    throw std::runtime_error("Table " + *table_ +
                             " will not be created. User has no create privileges.");
  }

  if (catalog.getMetadataForTable(*table_) != nullptr) {
    throw std::runtime_error("Table " + *table_ + " already exists.");
  }
  DataframeTableDescriptor df_td;
  std::list<ColumnDescriptor> columns;
  std::vector<SharedDictionaryDef> shared_dict_defs;

  std::unordered_set<std::string> uc_col_names;
  for (auto& e : table_element_list_) {
    if (dynamic_cast<SharedDictionaryDef*>(e.get())) {
      auto shared_dict_def = static_cast<SharedDictionaryDef*>(e.get());
      validate_shared_dictionary(
          this, shared_dict_def, columns, shared_dict_defs, catalog);
      shared_dict_defs.push_back(*shared_dict_def);
      continue;
    }
    if (!dynamic_cast<ColumnDef*>(e.get())) {
      throw std::runtime_error("Table constraints are not supported yet.");
    }
    ColumnDef* coldef = static_cast<ColumnDef*>(e.get());
    ColumnDescriptor cd;
    cd.columnName = *coldef->get_column_name();
    const auto uc_col_name = boost::to_upper_copy<std::string>(cd.columnName);
    const auto it_ok = uc_col_names.insert(uc_col_name);
    if (!it_ok.second) {
      throw std::runtime_error("Column '" + cd.columnName + "' defined more than once");
    }
    setColumnDescriptor(cd, coldef);
    columns.push_back(cd);
  }

  df_td.tableName = *table_;
  df_td.nColumns = columns.size();
  df_td.isView = false;
  df_td.fragmenter = nullptr;
  df_td.fragType = Fragmenter_Namespace::FragmenterType::INSERT_ORDER;
  df_td.maxFragRows = DEFAULT_FRAGMENT_ROWS;
  df_td.maxChunkSize = DEFAULT_MAX_CHUNK_SIZE;
  df_td.fragPageSize = DEFAULT_PAGE_SIZE;
  df_td.maxRows = DEFAULT_MAX_ROWS;
  df_td.persistenceLevel = Data_Namespace::MemoryLevel::CPU_LEVEL;
  if (!storage_options_.empty()) {
    for (auto& p : storage_options_) {
      get_dataframe_definitions(df_td, p, columns);
    }
  }
  df_td.keyMetainfo = serialize_key_metainfo(nullptr, shared_dict_defs);
  df_td.userId = session.get_currentUser().userId;
  df_td.storageType = *filename_;

  catalog.createShardedTable(df_td, columns, shared_dict_defs);
  // TODO (max): It's transactionally unsafe, should be fixed: we may create object w/o
  // privileges
  SysCatalog::instance().createDBObject(
      session.get_currentUser(), df_td.tableName, TableDBObjectType, catalog);
}

std::shared_ptr<ResultSet> getResultSet(QueryStateProxy query_state_proxy,
                                        const std::string select_stmt,
                                        std::vector<TargetMetaInfo>& targets,
                                        bool validate_only = false,
                                        std::vector<size_t> outer_fragment_indices = {},
                                        bool allow_interrupt = false) {
  auto const session = query_state_proxy->getConstSessionInfo();
  auto& catalog = session->getCatalog();

  auto executor = Executor::getExecutor(Executor::UNITARY_EXECUTOR_ID);
#ifdef HAVE_CUDA
  const auto device_type = session->get_executor_device_type();
#else
  const auto device_type = ExecutorDeviceType::CPU;
#endif  // HAVE_CUDA
  auto calcite_mgr = catalog.getCalciteMgr();

  // TODO MAT this should actually get the global or the session parameter for
  // view optimization
  const auto calciteQueryParsingOption =
      calcite_mgr->getCalciteQueryParsingOption(true, false, true);
  const auto calciteOptimizationOption = calcite_mgr->getCalciteOptimizationOption(
      false,
      g_enable_watchdog,
      {},
      Catalog_Namespace::SysCatalog::instance().isAggregator());
  const auto query_ra = calcite_mgr
                            ->process(query_state_proxy,
                                      pg_shim(select_stmt),
                                      calciteQueryParsingOption,
                                      calciteOptimizationOption)
                            .plan_result;
  RelAlgExecutor ra_executor(
      executor.get(), query_ra, query_state_proxy->shared_from_this());
  CompilationOptions co = CompilationOptions::defaults(device_type);
  // TODO(adb): Need a better method of dropping constants into this ExecutionOptions
  // struct
  ExecutionOptions eo = {false,
                         false,
                         true,
                         false,
                         true,
                         false,
                         false,
                         validate_only,
                         false,
                         10000,
                         false,
                         false,
                         1000,
                         allow_interrupt,
                         g_running_query_interrupt_freq,
                         g_pending_query_interrupt_freq,
                         g_optimize_cuda_block_and_grid_sizes,
                         std::numeric_limits<size_t>::max(),
                         ExecutorType::Native,
                         outer_fragment_indices};

  ExecutionResult result{std::make_shared<ResultSet>(std::vector<TargetInfo>{},
                                                     ExecutorDeviceType::CPU,
                                                     QueryMemoryDescriptor(),
                                                     nullptr,
                                                     0,
                                                     0),
                         {}};
  result = ra_executor.executeRelAlgQuery(co, eo, false, nullptr);
  targets = result.getTargetsMeta();

  return result.getRows();
}

size_t LocalQueryConnector::getOuterFragmentCount(QueryStateProxy query_state_proxy,
                                                  std::string& sql_query_string) {
  auto const session = query_state_proxy->getConstSessionInfo();
  auto& catalog = session->getCatalog();

  auto executor = Executor::getExecutor(Executor::UNITARY_EXECUTOR_ID);
#ifdef HAVE_CUDA
  const auto device_type = session->get_executor_device_type();
#else
  const auto device_type = ExecutorDeviceType::CPU;
#endif  // HAVE_CUDA
  auto calcite_mgr = catalog.getCalciteMgr();

  // TODO MAT this should actually get the global or the session parameter for
  // view optimization
  const auto calciteQueryParsingOption =
      calcite_mgr->getCalciteQueryParsingOption(true, false, true);
  const auto calciteOptimizationOption = calcite_mgr->getCalciteOptimizationOption(
      false,
      g_enable_watchdog,
      {},
      Catalog_Namespace::SysCatalog::instance().isAggregator());
  const auto query_ra = calcite_mgr
                            ->process(query_state_proxy,
                                      pg_shim(sql_query_string),
                                      calciteQueryParsingOption,
                                      calciteOptimizationOption)
                            .plan_result;
  RelAlgExecutor ra_executor(executor.get(), query_ra);
  CompilationOptions co = {device_type, true, ExecutorOptLevel::Default, false};
  // TODO(adb): Need a better method of dropping constants into this ExecutionOptions
  // struct
  ExecutionOptions eo = {false,
                         false,
                         true,
                         false,
                         true,
                         false,
                         false,
                         false,
                         false,
                         10000,
                         false,
                         false,
                         0.9,
                         false,
                         false};
  return ra_executor.getOuterFragmentCount(co, eo);
}

AggregatedResult LocalQueryConnector::query(QueryStateProxy query_state_proxy,
                                            std::string& sql_query_string,
                                            std::vector<size_t> outer_frag_indices,
                                            bool validate_only,
                                            bool allow_interrupt) {
  // TODO(PS): Should we be using the shimmed query in getResultSet?
  std::string pg_shimmed_select_query = pg_shim(sql_query_string);

  std::vector<TargetMetaInfo> target_metainfos;
  auto executor = Executor::getExecutor(Executor::UNITARY_EXECUTOR_ID);
  auto const session = query_state_proxy->getConstSessionInfo();
  auto query_session = session ? session->get_session_id() : "";
  auto query_submitted_time = query_state_proxy->getQuerySubmittedTime();
  if (allow_interrupt && !validate_only && !query_session.empty()) {
    executor->enrollQuerySession(query_session,
                                 sql_query_string,
                                 query_submitted_time,
                                 Executor::UNITARY_EXECUTOR_ID,
                                 QuerySessionStatus::QueryStatus::PENDING_EXECUTOR);
  }
  auto result_rows = getResultSet(query_state_proxy,
                                  sql_query_string,
                                  target_metainfos,
                                  validate_only,
                                  outer_frag_indices,
                                  allow_interrupt);
  AggregatedResult res = {result_rows, target_metainfos};
  return res;
}

std::vector<AggregatedResult> LocalQueryConnector::query(
    QueryStateProxy query_state_proxy,
    std::string& sql_query_string,
    std::vector<size_t> outer_frag_indices,
    bool allow_interrupt) {
  auto res = query(
      query_state_proxy, sql_query_string, outer_frag_indices, false, allow_interrupt);
  return {res};
}

std::list<ColumnDescriptor> LocalQueryConnector::getColumnDescriptors(
    AggregatedResult& result,
    bool for_create) {
  std::list<ColumnDescriptor> column_descriptors;
  std::list<ColumnDescriptor> column_descriptors_for_create;

  int rowid_suffix = 0;
  for (const auto& target_metainfo : result.targets_meta) {
    ColumnDescriptor cd;
    cd.columnName = target_metainfo.get_resname();
    if (cd.columnName == "rowid") {
      cd.columnName += std::to_string(rowid_suffix++);
    }
    cd.columnType = target_metainfo.get_physical_type_info();

    ColumnDescriptor cd_for_create = cd;

    if (cd.columnType.get_compression() == kENCODING_DICT) {
      // we need to reset the comp param (as this points to the actual dictionary)
      if (cd.columnType.is_array()) {
        // for dict encoded arrays, it is always 4 bytes
        cd_for_create.columnType.set_comp_param(32);
      } else {
        cd_for_create.columnType.set_comp_param(cd.columnType.get_size() * 8);
      }
    }

    if (cd.columnType.is_date() && !cd.columnType.is_date_in_days()) {
      // default to kENCODING_DATE_IN_DAYS encoding
      cd_for_create.columnType.set_compression(kENCODING_DATE_IN_DAYS);
      cd_for_create.columnType.set_comp_param(0);
    }

    column_descriptors_for_create.push_back(cd_for_create);
    column_descriptors.push_back(cd);
  }

  if (for_create) {
    return column_descriptors_for_create;
  }

  return column_descriptors;
}

InsertIntoTableAsSelectStmt::InsertIntoTableAsSelectStmt(
    const rapidjson::Value& payload) {
  CHECK(payload.HasMember("name"));
  table_name_ = json_str(payload["name"]);

  CHECK(payload.HasMember("query"));
  select_query_ = json_str(payload["query"]);

  boost::replace_all(select_query_, "\n", " ");
  select_query_ = "(" + select_query_ + ")";

  if (payload.HasMember("columns")) {
    CHECK(payload["columns"].IsArray());
    for (auto& column : payload["columns"].GetArray()) {
      std::string s = json_str(column);
      column_list_.emplace_back(std::unique_ptr<std::string>(new std::string(s)));
    }
  }
}

void InsertIntoTableAsSelectStmt::populateData(QueryStateProxy query_state_proxy,
                                               const TableDescriptor* td,
                                               bool validate_table,
                                               bool for_CTAS) {
  auto const session = query_state_proxy->getConstSessionInfo();
  auto& catalog = session->getCatalog();
  foreign_storage::validate_non_foreign_table_write(td);
  bool populate_table = false;

  if (leafs_connector_) {
    populate_table = true;
  } else {
    leafs_connector_ = std::make_unique<LocalQueryConnector>();
    if (!g_cluster) {
      populate_table = true;
    }
  }

  auto get_target_column_descriptors = [this, &catalog](const TableDescriptor* td) {
    std::vector<const ColumnDescriptor*> target_column_descriptors;
    if (column_list_.empty()) {
      auto list = catalog.getAllColumnMetadataForTable(td->tableId, false, false, false);
      target_column_descriptors = {std::begin(list), std::end(list)};
    } else {
      for (auto& c : column_list_) {
        const ColumnDescriptor* cd = catalog.getMetadataForColumn(td->tableId, *c);
        if (cd == nullptr) {
          throw std::runtime_error("Column " + *c + " does not exist.");
        }
        target_column_descriptors.push_back(cd);
      }
    }

    return target_column_descriptors;
  };

  bool is_temporary = table_is_temporary(td);

  if (validate_table) {
    // check access privileges
    if (!td) {
      throw std::runtime_error("Table " + table_name_ + " does not exist.");
    }
    if (td->isView) {
      throw std::runtime_error("Insert to views is not supported yet.");
    }

    if (!session->checkDBAccessPrivileges(DBObjectType::TableDBObjectType,
                                          AccessPrivileges::INSERT_INTO_TABLE,
                                          table_name_)) {
      throw std::runtime_error("User has no insert privileges on " + table_name_ + ".");
    }

    // only validate the select query so we get the target types
    // correctly, but do not populate the result set
    LocalQueryConnector local_connector;
    auto result = local_connector.query(query_state_proxy, select_query_, {}, true, true);
    auto source_column_descriptors = local_connector.getColumnDescriptors(result, false);

    std::vector<const ColumnDescriptor*> target_column_descriptors =
        get_target_column_descriptors(td);

    if (source_column_descriptors.size() != target_column_descriptors.size()) {
      throw std::runtime_error("The number of source and target columns does not match.");
    }

    for (int i = 0; i < source_column_descriptors.size(); i++) {
      const ColumnDescriptor* source_cd =
          &(*std::next(source_column_descriptors.begin(), i));
      const ColumnDescriptor* target_cd = target_column_descriptors.at(i);

      if (source_cd->columnType.get_type() != target_cd->columnType.get_type()) {
        auto type_cannot_be_cast = [](const auto& col_type) {
          return (col_type.is_time() || col_type.is_geometry() || col_type.is_array() ||
                  col_type.is_boolean());
        };

        if (type_cannot_be_cast(source_cd->columnType) ||
            type_cannot_be_cast(target_cd->columnType)) {
          throw std::runtime_error("Source '" + source_cd->columnName + " " +
                                   source_cd->columnType.get_type_name() +
                                   "' and target '" + target_cd->columnName + " " +
                                   target_cd->columnType.get_type_name() +
                                   "' column types do not match.");
        }
      }
      if (source_cd->columnType.is_array()) {
        if (source_cd->columnType.get_subtype() != target_cd->columnType.get_subtype()) {
          throw std::runtime_error("Source '" + source_cd->columnName + " " +
                                   source_cd->columnType.get_type_name() +
                                   "' and target '" + target_cd->columnName + " " +
                                   target_cd->columnType.get_type_name() +
                                   "' array column element types do not match.");
        }
      }

      if (target_cd->columnType.is_string() && !source_cd->columnType.is_string()) {
        throw std::runtime_error("Source '" + source_cd->columnName + " " +
                                 source_cd->columnType.get_type_name() +
                                 "' and target '" + target_cd->columnName + " " +
                                 target_cd->columnType.get_type_name() +
                                 "' column types do not match.");
      }

      if (source_cd->columnType.is_decimal() ||
          source_cd->columnType.get_elem_type().is_decimal()) {
        SQLTypeInfo sourceType = source_cd->columnType;
        SQLTypeInfo targetType = target_cd->columnType;

        if (source_cd->columnType.is_array()) {
          sourceType = source_cd->columnType.get_elem_type();
          targetType = target_cd->columnType.get_elem_type();
        }

        if (sourceType.get_scale() != targetType.get_scale()) {
          throw std::runtime_error("Source '" + source_cd->columnName + " " +
                                   source_cd->columnType.get_type_name() +
                                   "' and target '" + target_cd->columnName + " " +
                                   target_cd->columnType.get_type_name() +
                                   "' decimal columns scales do not match.");
        }
      }

      if (source_cd->columnType.is_string()) {
        if (!target_cd->columnType.is_string()) {
          throw std::runtime_error("Source '" + source_cd->columnName + " " +
                                   source_cd->columnType.get_type_name() +
                                   "' and target '" + target_cd->columnName + " " +
                                   target_cd->columnType.get_type_name() +
                                   "' column types do not match.");
        }
        if (source_cd->columnType.get_compression() !=
            target_cd->columnType.get_compression()) {
          throw std::runtime_error("Source '" + source_cd->columnName + " " +
                                   source_cd->columnType.get_type_name() +
                                   "' and target '" + target_cd->columnName + " " +
                                   target_cd->columnType.get_type_name() +
                                   "' columns string encodings do not match.");
        }
      }

      if (source_cd->columnType.is_timestamp() && target_cd->columnType.is_timestamp()) {
        if (source_cd->columnType.get_dimension() !=
            target_cd->columnType.get_dimension()) {
          throw std::runtime_error("Source '" + source_cd->columnName + " " +
                                   source_cd->columnType.get_type_name() +
                                   "' and target '" + target_cd->columnName + " " +
                                   target_cd->columnType.get_type_name() +
                                   "' timestamp column precisions do not match.");
        }
      }

      if (!source_cd->columnType.is_string() && !source_cd->columnType.is_geometry() &&
          !source_cd->columnType.is_integer() && !source_cd->columnType.is_decimal() &&
          !source_cd->columnType.is_date() && !source_cd->columnType.is_time() &&
          !source_cd->columnType.is_timestamp() &&
          source_cd->columnType.get_size() > target_cd->columnType.get_size()) {
        throw std::runtime_error("Source '" + source_cd->columnName + " " +
                                 source_cd->columnType.get_type_name() +
                                 "' and target '" + target_cd->columnName + " " +
                                 target_cd->columnType.get_type_name() +
                                 "' column encoding sizes do not match.");
      }
    }
  }

  if (!populate_table) {
    return;
  }

  int64_t total_row_count = 0;
  int64_t total_source_query_time_ms = 0;
  int64_t total_target_value_translate_time_ms = 0;
  int64_t total_data_load_time_ms = 0;

  Fragmenter_Namespace::InsertDataLoader insertDataLoader(*leafs_connector_);
  auto target_column_descriptors = get_target_column_descriptors(td);
  auto outer_frag_count =
      leafs_connector_->getOuterFragmentCount(query_state_proxy, select_query_);

  size_t outer_frag_end = outer_frag_count == 0 ? 1 : outer_frag_count;
  auto query_session = session ? session->get_session_id() : "";
  auto executor = Executor::getExecutor(Executor::UNITARY_EXECUTOR_ID).get();
  std::string work_type_str = for_CTAS ? "CTAS" : "ITAS";
  try {
    for (size_t outer_frag_idx = 0; outer_frag_idx < outer_frag_end; outer_frag_idx++) {
      std::vector<size_t> allowed_outer_fragment_indices;

      if (outer_frag_count) {
        allowed_outer_fragment_indices.push_back(outer_frag_idx);
      }

      const auto query_clock_begin = timer_start();
      std::vector<AggregatedResult> query_results =
          leafs_connector_->query(query_state_proxy,
                                  select_query_,
                                  allowed_outer_fragment_indices,
                                  g_enable_non_kernel_time_query_interrupt);
      total_source_query_time_ms += timer_stop(query_clock_begin);

      auto start_time = query_state_proxy->getQuerySubmittedTime();
      auto query_str = "INSERT_DATA for " + work_type_str;
      if (g_enable_non_kernel_time_query_interrupt) {
        // In the clean-up phase of the query execution for collecting aggregated result
        // of SELECT query, we remove its query session info, so we need to enroll the
        // session info again
        executor->enrollQuerySession(query_session,
                                     query_str,
                                     start_time,
                                     Executor::UNITARY_EXECUTOR_ID,
                                     QuerySessionStatus::QueryStatus::RUNNING_IMPORTER);
      }

      ScopeGuard clearInterruptStatus = [executor, &query_session, &start_time] {
        // this data population is non-kernel operation, so we manually cleanup
        // the query session info in the cleanup phase
        if (g_enable_non_kernel_time_query_interrupt) {
          executor->clearQuerySessionStatus(query_session, start_time);
        }
      };

      for (auto& res : query_results) {
        if (UNLIKELY(check_session_interrupted(query_session, executor))) {
          throw std::runtime_error(
              "Query execution has been interrupted while performing " + work_type_str);
        }
        auto& result_rows = res.rs;
        result_rows->setGeoReturnType(ResultSet::GeoReturnType::GeoTargetValue);
        const auto num_rows = result_rows->rowCount();

        if (0 == num_rows) {
          continue;
        }

        total_row_count += num_rows;

        size_t leaf_count = leafs_connector_->leafCount();

        // ensure that at least 1 row is processed per block up to a maximum of 65536 rows
        const size_t rows_per_block =
            std::max(std::min(num_rows / leaf_count, size_t(64 * 1024)), size_t(1));

        std::vector<std::unique_ptr<TargetValueConverter>> value_converters;

        TargetValueConverterFactory factory;

        const int num_worker_threads = std::thread::hardware_concurrency();

        std::vector<size_t> thread_start_idx(num_worker_threads),
            thread_end_idx(num_worker_threads);
        bool can_go_parallel = !result_rows->isTruncated() && rows_per_block > 20000;

        std::atomic<size_t> crt_row_idx{0};

        auto do_work = [&result_rows, &value_converters, &crt_row_idx](
                           const size_t idx,
                           const size_t block_end,
                           const size_t num_cols,
                           const size_t thread_id,
                           bool& stop_convert) {
          const auto result_row = result_rows->getRowAtNoTranslations(idx);
          if (!result_row.empty()) {
            size_t target_row = crt_row_idx.fetch_add(1);
            if (target_row >= block_end) {
              stop_convert = true;
              return;
            }
            for (unsigned int col = 0; col < num_cols; col++) {
              const auto& mapd_variant = result_row[col];
              value_converters[col]->convertToColumnarFormat(target_row, &mapd_variant);
            }
          }
        };

        auto convert_function = [&thread_start_idx,
                                 &thread_end_idx,
                                 &value_converters,
                                 &executor,
                                 &query_session,
                                 &work_type_str,
                                 &do_work](const int thread_id, const size_t block_end) {
          const int num_cols = value_converters.size();
          const size_t start = thread_start_idx[thread_id];
          const size_t end = thread_end_idx[thread_id];
          size_t idx = 0;
          bool stop_convert = false;
          if (g_enable_non_kernel_time_query_interrupt) {
            size_t local_idx = 0;
            for (idx = start; idx < end; ++idx, ++local_idx) {
              if (UNLIKELY((local_idx & 0xFFFF) == 0 &&
                           check_session_interrupted(query_session, executor))) {
                throw std::runtime_error(
                    "Query execution has been interrupted while performing " +
                    work_type_str);
              }
              do_work(idx, block_end, num_cols, thread_id, stop_convert);
              if (stop_convert) {
                break;
              }
            }
          } else {
            for (idx = start; idx < end; ++idx) {
              do_work(idx, block_end, num_cols, thread_id, stop_convert);
              if (stop_convert) {
                break;
              }
            }
          }
          thread_start_idx[thread_id] = idx;
        };

        auto single_threaded_value_converter =
            [&crt_row_idx, &value_converters, &result_rows](const size_t idx,
                                                            const size_t block_end,
                                                            const size_t num_cols,
                                                            bool& stop_convert) {
              size_t target_row = crt_row_idx.fetch_add(1);
              if (target_row >= block_end) {
                stop_convert = true;
                return;
              }
              const auto result_row = result_rows->getNextRow(false, false);
              CHECK(!result_row.empty());
              for (unsigned int col = 0; col < num_cols; col++) {
                const auto& mapd_variant = result_row[col];
                value_converters[col]->convertToColumnarFormat(target_row, &mapd_variant);
              }
            };

        auto single_threaded_convert_function = [&value_converters,
                                                 &thread_start_idx,
                                                 &thread_end_idx,
                                                 &executor,
                                                 &query_session,
                                                 &work_type_str,
                                                 &single_threaded_value_converter](
                                                    const int thread_id,
                                                    const size_t block_end) {
          const int num_cols = value_converters.size();
          const size_t start = thread_start_idx[thread_id];
          const size_t end = thread_end_idx[thread_id];
          size_t idx = 0;
          bool stop_convert = false;
          if (g_enable_non_kernel_time_query_interrupt) {
            size_t local_idx = 0;
            for (idx = start; idx < end; ++idx, ++local_idx) {
              if (UNLIKELY((local_idx & 0xFFFF) == 0 &&
                           check_session_interrupted(query_session, executor))) {
                throw std::runtime_error(
                    "Query execution has been interrupted while performing " +
                    work_type_str);
              }
              single_threaded_value_converter(idx, block_end, num_cols, stop_convert);
              if (stop_convert) {
                break;
              }
            }
          } else {
            for (idx = start; idx < end; ++idx) {
              single_threaded_value_converter(idx, end, num_cols, stop_convert);
              if (stop_convert) {
                break;
              }
            }
          }
          thread_start_idx[thread_id] = idx;
        };

        if (can_go_parallel) {
          const size_t entry_count = result_rows->entryCount();
          for (size_t
                   i = 0,
                   start_entry = 0,
                   stride = (entry_count + num_worker_threads - 1) / num_worker_threads;
               i < num_worker_threads && start_entry < entry_count;
               ++i, start_entry += stride) {
            const auto end_entry = std::min(start_entry + stride, entry_count);
            thread_start_idx[i] = start_entry;
            thread_end_idx[i] = end_entry;
          }
        } else {
          thread_start_idx[0] = 0;
          thread_end_idx[0] = result_rows->entryCount();
        }

        RenderGroupAnalyzerMap render_group_analyzer_map;

        for (size_t block_start = 0; block_start < num_rows;
             block_start += rows_per_block) {
          const auto num_rows_this_itr = block_start + rows_per_block < num_rows
                                             ? rows_per_block
                                             : num_rows - block_start;
          crt_row_idx = 0;  // reset block tracker
          value_converters.clear();
          int colNum = 0;
          for (const auto targetDescriptor : target_column_descriptors) {
            auto sourceDataMetaInfo = res.targets_meta[colNum++];
            ConverterCreateParameter param{
                num_rows_this_itr,
                sourceDataMetaInfo,
                targetDescriptor,
                catalog,
                targetDescriptor->columnType,
                !targetDescriptor->columnType.get_notnull(),
                result_rows->getRowSetMemOwner()->getLiteralStringDictProxy(),
                g_enable_string_functions &&
                        sourceDataMetaInfo.get_type_info().is_dict_encoded_string()
                    ? executor->getStringDictionaryProxy(
                          sourceDataMetaInfo.get_type_info().getStringDictKey(),
                          result_rows->getRowSetMemOwner(),
                          true)
                    : nullptr,
                IS_GEO_POLY(targetDescriptor->columnType.get_type()) &&
                        g_enable_assign_render_groups
                    ? &render_group_analyzer_map
                    : nullptr};
            auto converter = factory.create(param);
            value_converters.push_back(std::move(converter));
          }

          const auto translate_clock_begin = timer_start();
          if (can_go_parallel) {
            std::vector<std::future<void>> worker_threads;
            for (int i = 0; i < num_worker_threads; ++i) {
              worker_threads.push_back(
                  std::async(std::launch::async, convert_function, i, num_rows_this_itr));
            }

            for (auto& child : worker_threads) {
              child.wait();
            }
            for (auto& child : worker_threads) {
              child.get();
            }

          } else {
            single_threaded_convert_function(0, num_rows_this_itr);
          }

          // finalize the insert data
          auto finalizer_func =
              [](std::unique_ptr<TargetValueConverter>::pointer targetValueConverter) {
                targetValueConverter->finalizeDataBlocksForInsertData();
              };

          std::vector<std::future<void>> worker_threads;
          for (auto& converterPtr : value_converters) {
            worker_threads.push_back(
                std::async(std::launch::async, finalizer_func, converterPtr.get()));
          }

          for (auto& child : worker_threads) {
            child.wait();
          }
          for (auto& child : worker_threads) {
            child.get();
          }

          Fragmenter_Namespace::InsertData insert_data;
          insert_data.databaseId = catalog.getCurrentDB().dbId;
          CHECK(td);
          insert_data.tableId = td->tableId;
          insert_data.numRows = num_rows_this_itr;

          for (int col_idx = 0; col_idx < target_column_descriptors.size(); col_idx++) {
            if (UNLIKELY(g_enable_non_kernel_time_query_interrupt &&
                         check_session_interrupted(query_session, executor))) {
              throw std::runtime_error(
                  "Query execution has been interrupted while performing " +
                  work_type_str);
            }
            value_converters[col_idx]->addDataBlocksToInsertData(insert_data);
          }
          total_target_value_translate_time_ms += timer_stop(translate_clock_begin);

          const auto data_load_clock_begin = timer_start();
          auto data_memory_holder =
              import_export::fill_missing_columns(&catalog, insert_data);
          insertDataLoader.insertData(*session, insert_data);
          total_data_load_time_ms += timer_stop(data_load_clock_begin);
        }
      }
    }
  } catch (...) {
    try {
      leafs_connector_->rollback(*session, td->tableId);
    } catch (std::exception& e) {
      LOG(ERROR) << "An error occurred during ITAS rollback attempt. Table id: "
                 << td->tableId << ", Error: " << e.what();
    }
    throw;
  }

  int64_t total_time_ms = total_source_query_time_ms +
                          total_target_value_translate_time_ms + total_data_load_time_ms;

  VLOG(1) << "CTAS/ITAS " << total_row_count << " rows loaded in " << total_time_ms
          << "ms (outer_frag_count=" << outer_frag_count
          << ", query_time=" << total_source_query_time_ms
          << "ms, translation_time=" << total_target_value_translate_time_ms
          << "ms, data_load_time=" << total_data_load_time_ms
          << "ms)\nquery: " << select_query_;

  if (!is_temporary) {
    leafs_connector_->checkpoint(*session, td->tableId);
  }
}

namespace {
shared::TableKey get_table_key(const std::vector<std::string>& table) {
  const auto catalog = SysCatalog::instance().getCatalog(table[1]);
  CHECK(catalog);
  const auto table_id = catalog->getTableId(table[0]);
  if (!table_id.has_value()) {
    throw std::runtime_error{"Table \"" + table[0] +
                             "\" does not exist in catalog: " + table[1] + "."};
  }
  return {catalog->getDatabaseId(), table_id.value()};
}

lockmgr::LockedTableDescriptors acquire_query_table_locks(
    const std::string& insert_table_db_name,
    const std::string& query_str,
    const QueryStateProxy& query_state_proxy,
    const std::optional<std::string>& insert_table_name = {}) {
  auto& sys_catalog = SysCatalog::instance();
  auto& calcite_mgr = sys_catalog.getCalciteMgr();
  const auto calciteQueryParsingOption =
      calcite_mgr.getCalciteQueryParsingOption(true, false, true);
  const auto calciteOptimizationOption = calcite_mgr.getCalciteOptimizationOption(
      false, g_enable_watchdog, {}, sys_catalog.isAggregator());
  const auto result = calcite_mgr.process(query_state_proxy,
                                          pg_shim(query_str),
                                          calciteQueryParsingOption,
                                          calciteOptimizationOption);
  // force sort into tableid order in case of name change to guarantee fixed order of
  // mutex access
  auto comparator = [](const std::vector<std::string>& table_1,
                       const std::vector<std::string>& table_2) {
    return get_table_key(table_1) < get_table_key(table_2);
  };
  std::set<std::vector<std::string>, decltype(comparator)> tables(comparator);
  for (auto& tab : result.resolved_accessed_objects.tables_selected_from) {
    tables.emplace(tab);
  }
  if (insert_table_name.has_value()) {
    tables.emplace(
        std::vector<std::string>{insert_table_name.value(), insert_table_db_name});
  }
  lockmgr::LockedTableDescriptors locks;
  for (const auto& table : tables) {
    const auto catalog = sys_catalog.getCatalog(table[1]);
    CHECK(catalog);
    locks.emplace_back(
        std::make_unique<lockmgr::TableSchemaLockContainer<lockmgr::ReadLock>>(
            lockmgr::TableSchemaLockContainer<lockmgr::ReadLock>::acquireTableDescriptor(
                *catalog, table[0])));
    if (insert_table_name.has_value() && table[0] == insert_table_name.value() &&
        table[1] == insert_table_db_name) {
      locks.emplace_back(
          std::make_unique<lockmgr::TableInsertLockContainer<lockmgr::WriteLock>>(
              lockmgr::TableInsertLockContainer<lockmgr::WriteLock>::acquire(
                  catalog->getDatabaseId(), (*locks.back())())));
    } else {
      locks.emplace_back(
          std::make_unique<lockmgr::TableDataLockContainer<lockmgr::ReadLock>>(
              lockmgr::TableDataLockContainer<lockmgr::ReadLock>::acquire(
                  catalog->getDatabaseId(), (*locks.back())())));
    }
  }
  return locks;
}
}  // namespace

void InsertIntoTableAsSelectStmt::execute(const Catalog_Namespace::SessionInfo& session,
                                          bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("INSERT INTO TABLE invalid in read only mode.");
  }
  auto session_copy = session;
  auto session_ptr = std::shared_ptr<Catalog_Namespace::SessionInfo>(
      &session_copy, boost::null_deleter());
  auto query_state = query_state::QueryState::create(session_ptr, select_query_);
  auto stdlog = STDLOG(query_state);
  auto& catalog = session_ptr->getCatalog();

  const auto execute_read_lock =
      heavyai::shared_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
          *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
              legacylockmgr::ExecutorOuterLock, true));

  if (catalog.getMetadataForTable(table_name_) == nullptr) {
    throw std::runtime_error("ITAS failed: table " + table_name_ + " does not exist.");
  }

  auto locks = acquire_query_table_locks(
      catalog.name(), select_query_, query_state->createQueryStateProxy(), table_name_);
  const TableDescriptor* td = catalog.getMetadataForTable(table_name_);

  Executor::clearExternalCaches(true, td, catalog.getCurrentDB().dbId);

  try {
    populateData(query_state->createQueryStateProxy(), td, true, false);
  } catch (...) {
    throw;
  }
}

CreateTableAsSelectStmt::CreateTableAsSelectStmt(const rapidjson::Value& payload)
    : InsertIntoTableAsSelectStmt(payload) {
  if (payload.HasMember("temporary")) {
    is_temporary_ = json_bool(payload["temporary"]);
  } else {
    is_temporary_ = false;
  }

  if (payload.HasMember("ifNotExists")) {
    if_not_exists_ = json_bool(payload["ifNotExists"]);
  } else {
    if_not_exists_ = false;
  }

  parse_options(payload, storage_options_);
}

void CreateTableAsSelectStmt::execute(const Catalog_Namespace::SessionInfo& session,
                                      bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("CREATE TABLE invalid in read only mode.");
  }
  auto session_copy = session;
  auto session_ptr = std::shared_ptr<Catalog_Namespace::SessionInfo>(
      &session_copy, boost::null_deleter());
  auto query_state = query_state::QueryState::create(session_ptr, select_query_);
  auto stdlog = STDLOG(query_state);
  LocalQueryConnector local_connector;
  auto& catalog = session.getCatalog();
  bool create_table = nullptr == leafs_connector_;

  std::set<std::string> select_tables;
  if (create_table) {
    const auto execute_write_lock =
        heavyai::unique_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
            *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
                legacylockmgr::ExecutorOuterLock, true));

    // check access privileges
    if (!session.checkDBAccessPrivileges(DBObjectType::TableDBObjectType,
                                         AccessPrivileges::CREATE_TABLE)) {
      throw std::runtime_error("CTAS failed. Table " + table_name_ +
                               " will not be created. User has no create privileges.");
    }

    if (catalog.getMetadataForTable(table_name_) != nullptr) {
      if (if_not_exists_) {
        return;
      }
      throw std::runtime_error("Table " + table_name_ +
                               " already exists and no data was loaded.");
    }

    // only validate the select query so we get the target types
    // correctly, but do not populate the result set
    // we currently have exclusive access to the system so this is safe
    auto validate_result = local_connector.query(
        query_state->createQueryStateProxy(), select_query_, {}, true, false);

    auto column_descriptors_for_create =
        local_connector.getColumnDescriptors(validate_result, true);

    // some validation as the QE might return some out of range column types
    for (auto& cd : column_descriptors_for_create) {
      if (cd.columnType.is_decimal() &&
          cd.columnType.get_precision() > sql_constants::kMaxNumericPrecision) {
        throw std::runtime_error(cd.columnName + ": Precision too high, max " +
                                 std::to_string(sql_constants::kMaxNumericPrecision) +
                                 ".");
      }
    }

    TableDescriptor td;
    td.tableName = table_name_;
    td.userId = session.get_currentUser().userId;
    td.nColumns = column_descriptors_for_create.size();
    td.isView = false;
    td.fragmenter = nullptr;
    td.fragType = Fragmenter_Namespace::FragmenterType::INSERT_ORDER;
    td.maxFragRows = DEFAULT_FRAGMENT_ROWS;
    td.maxChunkSize = DEFAULT_MAX_CHUNK_SIZE;
    td.fragPageSize = DEFAULT_PAGE_SIZE;
    td.maxRows = DEFAULT_MAX_ROWS;
    td.maxRollbackEpochs = DEFAULT_MAX_ROLLBACK_EPOCHS;
    if (is_temporary_) {
      td.persistenceLevel = Data_Namespace::MemoryLevel::CPU_LEVEL;
    } else {
      td.persistenceLevel = Data_Namespace::MemoryLevel::DISK_LEVEL;
    }

    bool use_shared_dictionaries = true;
    bool force_geo_compression = true;

    if (!storage_options_.empty()) {
      for (auto& p : storage_options_) {
        if (boost::to_lower_copy<std::string>(*p->get_name()) ==
            "use_shared_dictionaries") {
          const StringLiteral* literal =
              dynamic_cast<const StringLiteral*>(p->get_value());
          if (nullptr == literal) {
            throw std::runtime_error(
                "USE_SHARED_DICTIONARIES must be a string parameter");
          }
          std::string val = boost::to_lower_copy<std::string>(*literal->get_stringval());
          use_shared_dictionaries = val == "true" || val == "1" || val == "t";
        } else if (boost::to_lower_copy<std::string>(*p->get_name()) ==
                   "force_geo_compression") {
          const StringLiteral* literal =
              dynamic_cast<const StringLiteral*>(p->get_value());
          if (nullptr == literal) {
            throw std::runtime_error("FORCE_GEO_COMPRESSION must be a string parameter");
          }
          std::string val = boost::to_lower_copy<std::string>(*literal->get_stringval());
          force_geo_compression = val == "true" || val == "1" || val == "t";
        } else {
          get_table_definitions_for_ctas(td, p, column_descriptors_for_create);
        }
      }
    }

    std::vector<SharedDictionaryDef> sharedDictionaryRefs;

    if (use_shared_dictionaries) {
      const auto source_column_descriptors =
          local_connector.getColumnDescriptors(validate_result, false);
      const auto mapping = catalog.getDictionaryToColumnMapping();

      for (auto& source_cd : source_column_descriptors) {
        const auto& ti = source_cd.columnType;
        if (ti.is_string()) {
          if (ti.get_compression() == kENCODING_DICT) {
            int dict_id = ti.get_comp_param();
            auto it = mapping.find(dict_id);
            if (mapping.end() != it) {
              const auto targetColumn = it->second;
              auto targetTable =
                  catalog.getMetadataForTable(targetColumn->tableId, false);
              CHECK(targetTable);
              LOG(INFO) << "CTAS: sharing text dictionary on column "
                        << source_cd.columnName << " with " << targetTable->tableName
                        << "." << targetColumn->columnName;
              sharedDictionaryRefs.emplace_back(
                  source_cd.columnName, targetTable->tableName, targetColumn->columnName);
            }
          }
        }
      }
    }

    if (force_geo_compression) {
      for (auto& cd_for_create : column_descriptors_for_create) {
        auto& ti = cd_for_create.columnType;
        if (ti.is_geometry() && ti.get_output_srid() == 4326) {
          // turn on GEOINT32 compression
          ti.set_compression(kENCODING_GEOINT);
          ti.set_comp_param(32);
        }
      }
    }

    // currently no means of defining sharding in CTAS
    td.keyMetainfo = serialize_key_metainfo(nullptr, sharedDictionaryRefs);

    catalog.createTable(td, column_descriptors_for_create, sharedDictionaryRefs, true);
    // TODO (max): It's transactionally unsafe, should be fixed: we may create object
    // w/o privileges
    SysCatalog::instance().createDBObject(
        session.get_currentUser(), td.tableName, TableDBObjectType, catalog);
  }

  // note there is a time where we do not have any executor outer lock here. someone could
  // come along and mess with the data or other tables.
  const auto execute_read_lock =
      heavyai::shared_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
          *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
              legacylockmgr::ExecutorOuterLock, true));

  auto locks = acquire_query_table_locks(
      catalog.name(), select_query_, query_state->createQueryStateProxy(), table_name_);
  const TableDescriptor* td = catalog.getMetadataForTable(table_name_);
  try {
    populateData(query_state->createQueryStateProxy(), td, false, true);
  } catch (...) {
    if (!g_cluster) {
      const TableDescriptor* created_td = catalog.getMetadataForTable(table_name_);
      if (created_td) {
        catalog.dropTable(created_td);
      }
    }
    throw;
  }
}

DropTableStmt::DropTableStmt(const rapidjson::Value& payload) {
  CHECK(payload.HasMember("tableName"));
  table_ = std::make_unique<std::string>(json_str(payload["tableName"]));

  if_exists_ = false;
  if (payload.HasMember("ifExists")) {
    if_exists_ = json_bool(payload["ifExists"]);
  }
}

void DropTableStmt::execute(const Catalog_Namespace::SessionInfo& session,
                            bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("DROP TABLE invalid in read only mode.");
  }
  // Because we are able to acquire a unique_lock on the table descriptor to be dropped we
  // can get away with only using a shared_lock on the executor, as anything that will
  // touch the table being dropped with block, but other transactions are ok.
  const auto execute_read_lock =
      heavyai::shared_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
          *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
              legacylockmgr::ExecutorOuterLock, true));
  auto& catalog = session.getCatalog();
  const TableDescriptor* td{nullptr};
  std::unique_ptr<lockmgr::TableSchemaLockContainer<lockmgr::WriteLock>> td_with_lock;
  try {
    td_with_lock =
        std::make_unique<lockmgr::TableSchemaLockContainer<lockmgr::WriteLock>>(
            lockmgr::TableSchemaLockContainer<lockmgr::WriteLock>::acquireTableDescriptor(
                catalog, *table_, false));
    td = (*td_with_lock)();
  } catch (const std::runtime_error& e) {
    if (if_exists_) {
      return;
    } else {
      throw e;
    }
  }

  CHECK(td);
  CHECK(td_with_lock);

  // check access privileges
  if (!session.checkDBAccessPrivileges(
          DBObjectType::TableDBObjectType, AccessPrivileges::DROP_TABLE, *table_)) {
    throw std::runtime_error("Table " + *table_ +
                             " will not be dropped. User has no proper privileges.");
  }

  ddl_utils::validate_table_type(td, ddl_utils::TableType::TABLE, "DROP");

  {
    auto table_data_read_lock =
        lockmgr::TableDataLockMgr::getReadLockForTable(catalog, *table_);
    Executor::clearExternalCaches(false, td, catalog.getCurrentDB().dbId);
  }

  auto table_data_write_lock =
      lockmgr::TableDataLockMgr::getWriteLockForTable(catalog, *table_);
  catalog.dropTable(td);
}

void AlterTableStmt::execute(const Catalog_Namespace::SessionInfo& session,
                             bool read_only_mode) {}

std::unique_ptr<DDLStmt> AlterTableStmt::delegate(const rapidjson::Value& payload) {
  CHECK(payload.HasMember("tableName"));
  auto tableName = json_str(payload["tableName"]);

  CHECK(payload.HasMember("alterType"));
  auto type = json_str(payload["alterType"]);

  if (type == "RENAME_TABLE") {
    CHECK(payload.HasMember("newTableName"));
    auto newTableName = json_str(payload["newTableName"]);
    return std::unique_ptr<DDLStmt>(new Parser::RenameTableStmt(
        new std::string(tableName), new std::string(newTableName)));

  } else if (type == "RENAME_COLUMN") {
    CHECK(payload.HasMember("columnName"));
    auto columnName = json_str(payload["columnName"]);
    CHECK(payload.HasMember("newColumnName"));
    auto newColumnName = json_str(payload["newColumnName"]);
    return std::unique_ptr<DDLStmt>(
        new Parser::RenameColumnStmt(new std::string(tableName),
                                     new std::string(columnName),
                                     new std::string(newColumnName)));

  } else if (type == "ALTER_COLUMN") {
    CHECK(payload.HasMember("alterData"));
    CHECK(payload["alterData"].IsArray());
    throw std::runtime_error("ALTER TABLE ALTER COLUMN is not implemented.");
  } else if (type == "ADD_COLUMN") {
    CHECK(payload.HasMember("columnData"));
    CHECK(payload["columnData"].IsArray());

    // New Columns go into this list
    std::list<ColumnDef*>* table_element_list_ = new std::list<ColumnDef*>;

    const auto elements = payload["columnData"].GetArray();
    for (const auto& element : elements) {
      CHECK(element.IsObject());
      CHECK(element.HasMember("type"));
      if (json_str(element["type"]) == "SQL_COLUMN_DECLARATION") {
        auto col_def = column_from_json(element);
        table_element_list_->emplace_back(col_def.release());
      } else {
        LOG(FATAL) << "Unsupported element type for ALTER TABLE: "
                   << element["type"].GetString();
      }
    }

    return std::unique_ptr<DDLStmt>(
        new Parser::AddColumnStmt(new std::string(tableName), table_element_list_));

  } else if (type == "DROP_COLUMN") {
    CHECK(payload.HasMember("columnData"));
    auto columnData = json_str(payload["columnData"]);
    // Convert columnData to std::list<std::string*>*
    //    allocate std::list<> as DropColumnStmt will delete it;
    std::list<std::string*>* cols = new std::list<std::string*>;
    std::vector<std::string> cols1;
    boost::split(cols1, columnData, boost::is_any_of(","));
    for (auto s : cols1) {
      // strip leading/trailing spaces/quotes/single quotes
      boost::algorithm::trim_if(s, boost::is_any_of(" \"'`"));
      std::string* str = new std::string(s);
      cols->emplace_back(str);
    }

    return std::unique_ptr<DDLStmt>(
        new Parser::DropColumnStmt(new std::string(tableName), cols));

  } else if (type == "ALTER_OPTIONS") {
    CHECK(payload.HasMember("options"));
    const auto& options = payload["options"];
    if (options.IsObject()) {
      for (auto itr = options.MemberBegin(); itr != options.MemberEnd(); ++itr) {
        std::string* option_name = new std::string(json_str(itr->name));
        Literal* literal_value;
        if (itr->value.IsString()) {
          std::string literal_string = json_str(itr->value);

          // iff this string can be converted to INT
          //   ... do so because it is necessary for AlterTableParamStmt
          std::size_t sz;
          int iVal = std::stoi(literal_string, &sz);
          if (sz == literal_string.size()) {
            literal_value = new IntLiteral(iVal);
          } else {
            literal_value = new StringLiteral(&literal_string);
          }
        } else if (itr->value.IsInt() || itr->value.IsInt64()) {
          literal_value = new IntLiteral(json_i64(itr->value));
        } else if (itr->value.IsNull()) {
          literal_value = new NullLiteral();
        } else {
          throw std::runtime_error("Unable to handle literal for " + *option_name);
        }
        CHECK(literal_value);

        NameValueAssign* nv = new NameValueAssign(option_name, literal_value);
        return std::unique_ptr<DDLStmt>(
            new Parser::AlterTableParamStmt(new std::string(tableName), nv));
      }
    } else {
      CHECK(options.IsNull());
    }
  }
  return nullptr;
}

TruncateTableStmt::TruncateTableStmt(const rapidjson::Value& payload) {
  CHECK(payload.HasMember("tableName"));
  table_ = std::make_unique<std::string>(json_str(payload["tableName"]));
}

void TruncateTableStmt::execute(const Catalog_Namespace::SessionInfo& session,
                                bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("TRUNCATE TABLE invalid in read only mode.");
  }
  const auto execute_read_lock =
      heavyai::shared_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
          *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
              legacylockmgr::ExecutorOuterLock, true));
  auto& catalog = session.getCatalog();
  const auto td_with_lock =
      lockmgr::TableSchemaLockContainer<lockmgr::WriteLock>::acquireTableDescriptor(
          catalog, *table_, true);
  const auto td = td_with_lock();
  if (!td) {
    throw std::runtime_error("Table " + *table_ + " does not exist.");
  }

  // check access privileges
  std::vector<DBObject> privObjects;
  DBObject dbObject(*table_, TableDBObjectType);
  dbObject.loadKey(catalog);
  dbObject.setPrivileges(AccessPrivileges::TRUNCATE_TABLE);
  privObjects.push_back(dbObject);
  if (!SysCatalog::instance().checkPrivileges(session.get_currentUser(), privObjects)) {
    throw std::runtime_error("Table " + *table_ + " will not be truncated. User " +
                             session.get_currentUser().userLoggable() +
                             " has no proper privileges.");
  }

  if (td->isView) {
    throw std::runtime_error(*table_ + " is a view.  Cannot Truncate.");
  }
  foreign_storage::validate_non_foreign_table_write(td);

  // invalidate cached item
  {
    auto table_data_read_lock =
        lockmgr::TableDataLockMgr::getReadLockForTable(catalog, *table_);
    Executor::clearExternalCaches(false, td, catalog.getCurrentDB().dbId);
  }

  auto table_data_write_lock =
      lockmgr::TableDataLockMgr::getWriteLockForTable(catalog, *table_);
  catalog.truncateTable(td);
}

OptimizeTableStmt::OptimizeTableStmt(const rapidjson::Value& payload) {
  CHECK(payload.HasMember("tableName"));
  table_ = std::make_unique<std::string>(json_str(payload["tableName"]));
  parse_options(payload, options_);
}

namespace {
bool user_can_access_table(const Catalog_Namespace::SessionInfo& session_info,
                           const TableDescriptor* td,
                           const AccessPrivileges access_priv) {
  CHECK(td);
  auto& cat = session_info.getCatalog();
  std::vector<DBObject> privObjects;
  DBObject dbObject(td->tableName, TableDBObjectType);
  dbObject.loadKey(cat);
  dbObject.setPrivileges(access_priv);
  privObjects.push_back(dbObject);
  return SysCatalog::instance().checkPrivileges(session_info.get_currentUser(),
                                                privObjects);
};
}  // namespace

void OptimizeTableStmt::execute(const Catalog_Namespace::SessionInfo& session,
                                bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("OPTIMIZE TABLE invalid in read only mode.");
  }
  auto& catalog = session.getCatalog();

  const auto execute_read_lock =
      heavyai::shared_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
          *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
              legacylockmgr::ExecutorOuterLock, true));

  const auto td_with_lock =
      lockmgr::TableSchemaLockContainer<lockmgr::WriteLock>::acquireTableDescriptor(
          catalog, *table_);
  const auto td = td_with_lock();

  if (!td || !user_can_access_table(session, td, AccessPrivileges::DELETE_FROM_TABLE)) {
    throw std::runtime_error("Table " + *table_ + " does not exist.");
  }

  if (td->isView) {
    throw std::runtime_error("OPTIMIZE TABLE command is not supported on views.");
  }

  // invalidate cached item
  std::vector<int> table_key{catalog.getCurrentDB().dbId, td->tableId};
  ResultSetCacheInvalidator::invalidateCachesByTable(boost::hash_value(table_key));

  auto executor = Executor::getExecutor(Executor::UNITARY_EXECUTOR_ID).get();
  const TableOptimizer optimizer(td, executor, catalog);
  if (shouldVacuumDeletedRows()) {
    optimizer.vacuumDeletedRows();
  }
  optimizer.recomputeMetadata();
}

bool repair_type(std::list<std::unique_ptr<NameValueAssign>>& options) {
  for (const auto& opt : options) {
    if (boost::iequals(*opt->get_name(), "REPAIR_TYPE")) {
      const auto repair_type =
          static_cast<const StringLiteral*>(opt->get_value())->get_stringval();
      CHECK(repair_type);
      if (boost::iequals(*repair_type, "REMOVE")) {
        return true;
      } else {
        throw std::runtime_error("REPAIR_TYPE must be REMOVE.");
      }
    } else {
      throw std::runtime_error("The only VALIDATE WITH options is REPAIR_TYPE.");
    }
  }
  return false;
}

ValidateStmt::ValidateStmt(std::string* type, std::list<NameValueAssign*>* with_opts)
    : type_(type) {
  if (!type) {
    throw std::runtime_error("Validation Type is required for VALIDATE command.");
  }
  std::list<std::unique_ptr<NameValueAssign>> options;
  if (with_opts) {
    for (const auto e : *with_opts) {
      options.emplace_back(e);
    }
    delete with_opts;

    isRepairTypeRemove_ = repair_type(options);
  }
}

ValidateStmt::ValidateStmt(const rapidjson::Value& payload) {
  CHECK(payload.HasMember("type"));
  type_ = std::make_unique<std::string>(json_str(payload["type"]));

  std::list<std::unique_ptr<NameValueAssign>> options;
  parse_options(payload, options);

  isRepairTypeRemove_ = repair_type(options);
}

void check_alter_table_privilege(const Catalog_Namespace::SessionInfo& session,
                                 const TableDescriptor* td) {
  if (session.get_currentUser().isSuper ||
      session.get_currentUser().userId == td->userId) {
    return;
  }
  std::vector<DBObject> privObjects;
  DBObject dbObject(td->tableName, TableDBObjectType);
  dbObject.loadKey(session.getCatalog());
  dbObject.setPrivileges(AccessPrivileges::ALTER_TABLE);
  privObjects.push_back(dbObject);
  if (!SysCatalog::instance().checkPrivileges(session.get_currentUser(), privObjects)) {
    throw std::runtime_error("Current user does not have the privilege to alter table: " +
                             td->tableName);
  }
}

RenameUserStmt::RenameUserStmt(const rapidjson::Value& payload) {
  CHECK(payload.HasMember("name"));
  username_ = std::make_unique<std::string>(json_str(payload["name"]));
  CHECK(payload.HasMember("newName"));
  new_username_ = std::make_unique<std::string>(json_str(payload["newName"]));
}

void RenameUserStmt::execute(const Catalog_Namespace::SessionInfo& session,
                             bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("RENAME TABLE invalid in read only mode.");
  }
  if (!session.get_currentUser().isSuper) {
    throw std::runtime_error("Only a super user can rename users.");
  }

  Catalog_Namespace::UserMetadata user;
  if (!SysCatalog::instance().getMetadataForUser(*username_, user)) {
    throw std::runtime_error("User " + *username_ + " does not exist.");
  }

  SysCatalog::instance().renameUser(*username_, *new_username_);
}

RenameDBStmt::RenameDBStmt(const rapidjson::Value& payload) {
  CHECK(payload.HasMember("name"));
  database_name_ = std::make_unique<std::string>(json_str(payload["name"]));
  CHECK(payload.HasMember("newName"));
  new_database_name_ = std::make_unique<std::string>(json_str(payload["newName"]));
}

void RenameDBStmt::execute(const Catalog_Namespace::SessionInfo& session,
                           bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("RENAME DATABASE invalid in read only mode.");
  }
  Catalog_Namespace::DBMetadata db;

  // TODO: use database lock instead
  const auto execute_write_lock =
      heavyai::unique_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
          *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
              legacylockmgr::ExecutorOuterLock, true));

  if (!SysCatalog::instance().getMetadataForDB(*database_name_, db)) {
    throw std::runtime_error("Database " + *database_name_ + " does not exist.");
  }

  if (!session.get_currentUser().isSuper &&
      session.get_currentUser().userId != db.dbOwner) {
    throw std::runtime_error("Only a super user or the owner can rename the database.");
  }

  SysCatalog::instance().renameDatabase(*database_name_, *new_database_name_);
}

RenameTableStmt::RenameTableStmt(const rapidjson::Value& payload) {
  CHECK(payload.HasMember("tableNames"));
  CHECK(payload["tableNames"].IsArray());
  const auto elements = payload["tableNames"].GetArray();
  for (const auto& element : elements) {
    CHECK(element.HasMember("name"));
    CHECK(element.HasMember("newName"));
    tablesToRename_.emplace_back(new std::string(json_str(element["name"])),
                                 new std::string(json_str(element["newName"])));
  }
}

RenameTableStmt::RenameTableStmt(std::string* tab_name, std::string* new_tab_name) {
  tablesToRename_.emplace_back(tab_name, new_tab_name);
}

RenameTableStmt::RenameTableStmt(
    std::list<std::pair<std::string, std::string>> tableNames) {
  for (auto item : tableNames) {
    tablesToRename_.emplace_back(new std::string(item.first),
                                 new std::string(item.second));
  }
}

using SubstituteMap = std::map<std::string, std::string>;

// Namespace fns used to track a left-to-right execution of RENAME TABLE
//   and verify that the command should be (entirely/mostly) valid
//
namespace {

static constexpr char const* EMPTY_NAME{""};

std::string generateUniqueTableName(std::string name) {
  // TODO - is there a "better" way to create a tmp name for the table
  std::time_t result = std::time(nullptr);
  return name + "_tmp" + std::to_string(result);
}

void recordRename(SubstituteMap& sMap, std::string oldName, std::string newName) {
  sMap[oldName] = newName;
}

std::string loadTable(Catalog_Namespace::Catalog& catalog,
                      SubstituteMap& sMap,
                      std::string tableName) {
  if (sMap.find(tableName) != sMap.end()) {
    if (sMap[tableName] == EMPTY_NAME) {
      return tableName;
    }
    return sMap[tableName];
  } else {
    // lookup table in src catalog
    const TableDescriptor* td = catalog.getMetadataForTable(tableName);
    if (td) {
      sMap[tableName] = tableName;
    } else {
      sMap[tableName] = EMPTY_NAME;
    }
  }
  return tableName;
}

bool hasData(SubstituteMap& sMap, std::string tableName) {
  // assumes loadTable has been previously called
  return (sMap[tableName] != EMPTY_NAME);
}

void checkNameSubstition(SubstituteMap& sMap) {
  // Substition map should be clean at end of rename:
  //    all items in map must (map to self) or (map to EMPTY_STRING) by end

  for (auto it : sMap) {
    if ((it.second) != EMPTY_NAME && (it.first) != (it.second)) {
      throw std::runtime_error(
          "Error: Attempted to overwrite and lose data in table: \'" + (it.first) + "\'");
    }
  }
}
}  // namespace

namespace {
void disable_foreign_tables(const TableDescriptor* td) {
  if (td->storageType == StorageType::FOREIGN_TABLE) {
    throw std::runtime_error(td->tableName + " is a foreign table. " +
                             "Use ALTER FOREIGN TABLE.");
  }
}
}  // namespace

void RenameTableStmt::execute(const Catalog_Namespace::SessionInfo& session,
                              bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("RENAME TABLE invalid in read only mode.");
  }
  auto& catalog = session.getCatalog();

  // TODO(adb): the catalog should be handling this locking (see AddColumStmt)
  const auto execute_write_lock =
      heavyai::unique_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
          *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
              legacylockmgr::ExecutorOuterLock, true));

  // accumulated vector of table names: oldName->newName
  std::vector<std::pair<std::string, std::string>> names;

  SubstituteMap tableSubtituteMap;

  for (auto& item : tablesToRename_) {
    std::string curTableName = *(item.first);
    std::string newTableName = *(item.second);

    // Note: if rename (a->b, b->a)
    //    requires a tmp name change (a->tmp, b->a, tmp->a),
    //    inject that here because
    //         catalog.renameTable() assumes cleanliness else will fail

    std::string altCurTableName = loadTable(catalog, tableSubtituteMap, curTableName);
    std::string altNewTableName = loadTable(catalog, tableSubtituteMap, newTableName);

    if (altCurTableName != curTableName && altCurTableName != EMPTY_NAME) {
      // rename is a one-shot deal, reset the mapping once used
      recordRename(tableSubtituteMap, curTableName, curTableName);
    }

    // Check to see if the command (as-entered) will likely execute cleanly (logic-wise)
    //     src tables exist before coping from
    //     destination table collisions
    //         handled (a->b, b->a)
    //         or flagged (pre-existing a,b ... "RENAME TABLE a->c, b->c" )
    //     handle mulitple chained renames, tmp names (a_>tmp, b->a, tmp->a)
    //     etc.
    //
    if (hasData(tableSubtituteMap, altCurTableName)) {
      const TableDescriptor* td = catalog.getMetadataForTable(altCurTableName);
      if (td) {
        // Tables *and* views may be renamed here, foreign tables not
        //    -> just block foreign tables
        disable_foreign_tables(td);
        check_alter_table_privilege(session, td);
      }

      if (hasData(tableSubtituteMap, altNewTableName)) {
        std::string tmpNewTableName = generateUniqueTableName(altNewTableName);
        // rename: newTableName to tmpNewTableName to get it out of the way
        //    because it was full
        recordRename(tableSubtituteMap, altCurTableName, EMPTY_NAME);
        recordRename(tableSubtituteMap, altNewTableName, tmpNewTableName);
        recordRename(tableSubtituteMap, tmpNewTableName, tmpNewTableName);
        names.emplace_back(altNewTableName, tmpNewTableName);
        names.emplace_back(altCurTableName, altNewTableName);
      } else {
        // rename: curNewTableName to newTableName
        recordRename(tableSubtituteMap, altCurTableName, EMPTY_NAME);
        recordRename(tableSubtituteMap, altNewTableName, altNewTableName);
        names.emplace_back(altCurTableName, altNewTableName);
      }
    } else {
      throw std::runtime_error("Source table \'" + curTableName + "\' does not exist.");
    }
  }
  checkNameSubstition(tableSubtituteMap);

  catalog.renameTable(names);

  // just to be explicit, clean out the list, the unique_ptr will delete
  while (!tablesToRename_.empty()) {
    tablesToRename_.pop_front();
  }
}  // namespace Parser

void DDLStmt::setColumnDescriptor(ColumnDescriptor& cd, const ColumnDef* coldef) {
  bool not_null;
  const ColumnConstraintDef* cc = coldef->get_column_constraint();
  if (cc == nullptr) {
    not_null = false;
  } else {
    not_null = cc->get_notnull();
  }
  std::string default_value;
  const std::string* default_value_ptr = nullptr;
  if (cc) {
    if (auto def_val_literal = cc->get_defaultval()) {
      auto defaultsp = dynamic_cast<const StringLiteral*>(def_val_literal);
      default_value =
          defaultsp ? *defaultsp->get_stringval() : def_val_literal->to_string();
      // The preprocessing below is needed because:
      // a) TypedImportBuffer expects arrays in the {...} format
      // b) TypedImportBuffer expects string literals inside arrays w/o any quotes
      if (coldef->get_column_type()->get_is_array()) {
        std::regex array_re(R"(^ARRAY\s*\[(.*)\]$)", std::regex_constants::icase);
        default_value = std::regex_replace(default_value, array_re, "{$1}");
        boost::erase_all(default_value, "\'");
      }
      default_value_ptr = &default_value;
    }
  }
  ddl_utils::set_column_descriptor(*coldef->get_column_name(),
                                   cd,
                                   coldef->get_column_type(),
                                   not_null,
                                   coldef->get_compression(),
                                   default_value_ptr);
}

void AddColumnStmt::check_executable(const Catalog_Namespace::SessionInfo& session,
                                     const TableDescriptor* td) {
  auto& catalog = session.getCatalog();
  if (!td) {
    throw std::runtime_error("Table " + *table_ + " does not exist.");
  } else {
    if (td->isView) {
      throw std::runtime_error("Adding columns to a view is not supported.");
    }
    validate_table_type(td, ddl_utils::TableType::TABLE, "ALTER");
    if (table_is_temporary(td)) {
      throw std::runtime_error(
          "Adding columns to temporary tables is not yet supported.");
    }
  }

  check_alter_table_privilege(session, td);

  if (0 == coldefs_.size()) {
    coldefs_.push_back(std::move(coldef_));
  }

  for (const auto& coldef : coldefs_) {
    auto& new_column_name = *coldef->get_column_name();
    if (catalog.getMetadataForColumn(td->tableId, new_column_name) != nullptr) {
      throw std::runtime_error("Column " + new_column_name + " already exists.");
    }
  }
}

void AddColumnStmt::execute(const Catalog_Namespace::SessionInfo& session,
                            bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("ADD COLUMN invalid in read only mode.");
  }
  // TODO: Review add and drop column implementation
  const auto execute_write_lock =
      heavyai::unique_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
          *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
              legacylockmgr::ExecutorOuterLock, true));
  auto& catalog = session.getCatalog();
  const auto td_with_lock =
      lockmgr::TableSchemaLockContainer<lockmgr::WriteLock>::acquireTableDescriptor(
          catalog, *table_, true);
  const auto td = td_with_lock();

  check_executable(session, td);

  CHECK(td->fragmenter);
  if (std::dynamic_pointer_cast<Fragmenter_Namespace::SortedOrderFragmenter>(
          td->fragmenter)) {
    throw std::runtime_error(
        "Adding columns to a table is not supported when using the \"sort_column\" "
        "option.");
  }

  // invalidate cached item
  std::vector<int> table_key{catalog.getCurrentDB().dbId, td->tableId};
  ResultSetCacheInvalidator::invalidateCachesByTable(boost::hash_value(table_key));

  // Do not take a data write lock, as the fragmenter may call `deleteFragments`
  // during a cap operation. Note that the schema write lock will prevent concurrent
  // inserts along with all other queries.

  catalog.getSqliteConnector().query("BEGIN TRANSACTION");
  try {
    std::map<const std::string, const ColumnDescriptor> cds;
    std::map<const int, const ColumnDef*> cid_coldefs;
    for (const auto& coldef : coldefs_) {
      ColumnDescriptor cd;
      setColumnDescriptor(cd, coldef.get());
      catalog.addColumn(*td, cd);
      cds.emplace(*coldef->get_column_name(), cd);
      cid_coldefs.emplace(cd.columnId, coldef.get());

      // expand geo column to phy columns
      if (cd.columnType.is_geometry()) {
        std::list<ColumnDescriptor> phy_geo_columns;
        catalog.expandGeoColumn(cd, phy_geo_columns);
        for (auto& cd : phy_geo_columns) {
          catalog.addColumn(*td, cd);
          cds.emplace(cd.columnName, cd);
          cid_coldefs.emplace(cd.columnId, nullptr);
        }
      }
    }

    std::unique_ptr<import_export::Loader> loader(new import_export::Loader(catalog, td));
    std::vector<std::unique_ptr<import_export::TypedImportBuffer>> import_buffers;
    for (const auto& cd : cds) {
      import_buffers.emplace_back(std::make_unique<import_export::TypedImportBuffer>(
          &cd.second, loader->getStringDict(&cd.second)));
    }
    loader->setAddingColumns(true);

    // set_geo_physical_import_buffer below needs a sorted import_buffers
    std::sort(import_buffers.begin(),
              import_buffers.end(),
              [](decltype(import_buffers[0])& a, decltype(import_buffers[0])& b) {
                return a->getColumnDesc()->columnId < b->getColumnDesc()->columnId;
              });

    size_t nrows = td->fragmenter->getNumRows();
    // if sharded, get total nrows from all sharded tables
    if (td->nShards > 0) {
      const auto physical_tds = catalog.getPhysicalTablesDescriptors(td);
      nrows = 0;
      std::for_each(physical_tds.begin(), physical_tds.end(), [&nrows](const auto& td) {
        nrows += td->fragmenter->getNumRows();
      });
    }
    if (nrows > 0) {
      int skip_physical_cols = 0;
      for (const auto cit : cid_coldefs) {
        const auto cd = catalog.getMetadataForColumn(td->tableId, cit.first);
        const auto coldef = cit.second;
        const bool is_null = !cd->default_value.has_value();

        if (cd->columnType.get_notnull() && is_null) {
          throw std::runtime_error("Default value required for column " + cd->columnName +
                                   " because of NOT NULL constraint");
        }

        for (auto it = import_buffers.begin(); it < import_buffers.end(); ++it) {
          auto& import_buffer = *it;
          if (cd->columnId == import_buffer->getColumnDesc()->columnId) {
            if (coldef != nullptr ||
                skip_physical_cols-- <= 0) {  // skip non-null phy col
              import_buffer->add_value(cd,
                                       cd->default_value.value_or("NULL"),
                                       is_null,
                                       import_export::CopyParams());
              if (cd->columnType.is_geometry()) {
                std::vector<double> coords, bounds;
                std::vector<int> ring_sizes, poly_rings;
                int render_group = 0;
                SQLTypeInfo tinfo{cd->columnType};
                if (!Geospatial::GeoTypesFactory::getGeoColumns(
                        cd->default_value.value_or("NULL"),
                        tinfo,
                        coords,
                        bounds,
                        ring_sizes,
                        poly_rings,
                        false)) {
                  throw std::runtime_error("Bad geometry data: '" +
                                           cd->default_value.value_or("NULL") + "'");
                }
                size_t col_idx = 1 + std::distance(import_buffers.begin(), it);
                import_export::Importer::set_geo_physical_import_buffer(catalog,
                                                                        cd,
                                                                        import_buffers,
                                                                        col_idx,
                                                                        coords,
                                                                        bounds,
                                                                        ring_sizes,
                                                                        poly_rings,
                                                                        render_group);
                // skip following phy cols
                skip_physical_cols = cd->columnType.get_physical_cols();
              }
            }
            break;
          }
        }
      }
    }

    if (!loader->loadNoCheckpoint(import_buffers, nrows, &session)) {
      throw std::runtime_error("loadNoCheckpoint failed!");
    }
    catalog.roll(true);
    catalog.resetTableEpochFloor(td->tableId);
    loader->checkpoint();
    catalog.getSqliteConnector().query("END TRANSACTION");
  } catch (...) {
    catalog.roll(false);
    catalog.getSqliteConnector().query("ROLLBACK TRANSACTION");
    throw;
  }
}

void DropColumnStmt::execute(const Catalog_Namespace::SessionInfo& session,
                             bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("DROP COLUMN invalid in read only mode.");
  }
  // TODO: Review add and drop column implementation
  const auto execute_write_lock =
      heavyai::unique_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
          *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
              legacylockmgr::ExecutorOuterLock, true));
  auto& catalog = session.getCatalog();
  const auto td_with_lock =
      lockmgr::TableSchemaLockContainer<lockmgr::WriteLock>::acquireTableDescriptor(
          catalog, *table_, true);
  const auto td = td_with_lock();
  if (!td) {
    throw std::runtime_error("Table " + *table_ + " does not exist.");
  }
  validate_table_type(td, ddl_utils::TableType::TABLE, "ALTER");
  if (td->isView) {
    throw std::runtime_error("Dropping a column from a view is not supported.");
  }
  if (table_is_temporary(td)) {
    throw std::runtime_error(
        "Dropping a column from a temporary table is not yet supported.");
  }

  check_alter_table_privilege(session, td);

  for (const auto& column : columns_) {
    if (nullptr == catalog.getMetadataForColumn(td->tableId, *column)) {
      throw std::runtime_error("Column " + *column + " does not exist.");
    }
  }

  if (td->nColumns <= (td->hasDeletedCol ? 3 : 2)) {
    throw std::runtime_error("Table " + *table_ + " has only one column.");
  }

  // invalidate cached item
  Executor::clearExternalCaches(false, td, catalog.getCurrentDB().dbId);

  catalog.getSqliteConnector().query("BEGIN TRANSACTION");
  try {
    std::vector<int> columnIds;
    for (const auto& column : columns_) {
      ColumnDescriptor cd = *catalog.getMetadataForColumn(td->tableId, *column);
      if (td->nShards > 0 && td->shardedColumnId == cd.columnId) {
        throw std::runtime_error("Dropping sharding column " + cd.columnName +
                                 " is not supported.");
      }
      catalog.dropColumn(*td, cd);
      columnIds.push_back(cd.columnId);
      for (int i = 0; i < cd.columnType.get_physical_cols(); i++) {
        const auto pcd = catalog.getMetadataForColumn(td->tableId, cd.columnId + i + 1);
        CHECK(pcd);
        catalog.dropColumn(*td, *pcd);
        columnIds.push_back(cd.columnId + i + 1);
      }
    }

    for (auto shard : catalog.getPhysicalTablesDescriptors(td)) {
      shard->fragmenter->dropColumns(columnIds);
    }
    // if test forces to rollback
    if (g_test_drop_column_rollback) {
      throw std::runtime_error("lol!");
    }
    catalog.roll(true);
    if (td->persistenceLevel == Data_Namespace::MemoryLevel::DISK_LEVEL) {
      catalog.resetTableEpochFloor(td->tableId);
      catalog.checkpoint(td->tableId);
    }
    catalog.getSqliteConnector().query("END TRANSACTION");
  } catch (...) {
    catalog.setForReload(td->tableId);
    catalog.roll(false);
    catalog.getSqliteConnector().query("ROLLBACK TRANSACTION");
    throw;
  }
}

void RenameColumnStmt::execute(const Catalog_Namespace::SessionInfo& session,
                               bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("RENAME COLUMN invalid in read only mode.");
  }
  auto& catalog = session.getCatalog();

  const auto execute_read_lock =
      heavyai::shared_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
          *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
              legacylockmgr::ExecutorOuterLock, true));

  const auto td_with_lock =
      lockmgr::TableSchemaLockContainer<lockmgr::WriteLock>::acquireTableDescriptor(
          catalog, *table_, false);
  const auto td = td_with_lock();
  CHECK(td);
  validate_table_type(td, ddl_utils::TableType::TABLE, "ALTER");

  check_alter_table_privilege(session, td);
  const ColumnDescriptor* cd = catalog.getMetadataForColumn(td->tableId, *column_);
  if (cd == nullptr) {
    throw std::runtime_error("Column " + *column_ + " does not exist.");
  }
  if (catalog.getMetadataForColumn(td->tableId, *new_column_name_) != nullptr) {
    throw std::runtime_error("Column " + *new_column_name_ + " already exists.");
  }
  catalog.renameColumn(td, cd, *new_column_name_);
}

void AlterTableParamStmt::execute(const Catalog_Namespace::SessionInfo& session,
                                  bool read_only_mode) {
  if (read_only_mode) {
    throw std::runtime_error("ALTER TABLE invalid in read only mode.");
  }
  enum TableParamType { MaxRollbackEpochs, Epoch, MaxRows };
  static const std::unordered_map<std::string, TableParamType> param_map = {
      {"max_rollback_epochs", TableParamType::MaxRollbackEpochs},
      {"epoch", TableParamType::Epoch},
      {"max_rows", TableParamType::MaxRows}};
  const auto execute_read_lock =
      heavyai::shared_lock<legacylockmgr::WrapperType<heavyai::shared_mutex>>(
          *legacylockmgr::LockMgr<heavyai::shared_mutex, bool>::getMutex(
              legacylockmgr::ExecutorOuterLock, true));
  auto& catalog = session.getCatalog();
  const auto td_with_lock =
      lockmgr::TableSchemaLockContainer<lockmgr::WriteLock>::acquireTableDescriptor(
          catalog, *table_, false);
  const auto td = td_with_lock();
  if (!td) {
    throw std::runtime_error("Table " + *table_ + " does not exist.");
  }
  if (td->isView) {
    throw std::runtime_error("Setting parameters for a view is not supported.");
  }
  if (table_is_temporary(td)) {
    throw std::runtime_error(
        "Setting parameters for a temporary table is not yet supported.");
  }
  check_alter_table_privilege(session, td);

  // invalidate cached item
  std::vector<int> table_key{catalog.getCurrentDB().dbId, td->tableId};
  ResultSetCacheInvalidator::invalidateCachesByTable(boost::hash_value(table_key));

  std::string param_name(*param_->get_name());
  boost::algorithm::to_lower(param_name);
  const IntLiteral* val_int_literal =
      dynamic_cast<const IntLiteral*>(param_->get_value());
  if (val_int_literal == nullptr) {
    throw std::runtime_error("Table parameters should be integers.");
  }
  const int64_t param_val = val_int_literal->get_intval();

  const auto param_it = param_map.find(param_name);
  if (param_it == param_map.end()) {
    throw std::runtime_error(param_name + " is not a settable table parameter.");
  }
  switch (param_it->second) {
    case MaxRollbackEpochs: {
      catalog.setMaxRollbackEpochs(td->tableId, param_val);
      break;
    }
    case Epoch: {
      catalog.setTableEpoch(catalog.getDatabaseId(), td->tableId, param_val);
      break;
    }
    case MaxRows: {
      catalog.setMaxRows(td->tableId, param_val);
      break;
    }
    default: {
      UNREACHABLE() << "Unexpected TableParamType value: " << param_it->second
                    << ", key: " << param_it->first;
    }
  }
}

CopyTableStmt::CopyTableStmt(std::string* t,
                             std::string* f,
                             std::list<NameValueAssign*>* o)
    : table_(t), copy_from_source_pattern_(f), success_(true) {
  if (o) {
    for (const auto e : *o) {
      options_.emplace_back(e);
    }
    delete o;