Merge branch 'master' into faster_replicated_ddl

.clang-tidy

@@ -144,6 +144,7 @@ Checks: '-*,
     clang-analyzer-cplusplus.SelfAssignment,
     clang-analyzer-deadcode.DeadStores,
     clang-analyzer-cplusplus.Move,
+    clang-analyzer-optin.cplusplus.UninitializedObject,
     clang-analyzer-optin.cplusplus.VirtualCall,
     clang-analyzer-security.insecureAPI.UncheckedReturn,
     clang-analyzer-security.insecureAPI.bcmp,

.github/PULL_REQUEST_TEMPLATE.md

@@ -1,4 +1,4 @@
-Changelog category (leave one):
+### Changelog category (leave one):
 - New Feature
 - Improvement
 - Bug Fix (user-visible misbehaviour in official stable or prestable release)
@@ -9,7 +9,7 @@ Changelog category (leave one):
 - Not for changelog (changelog entry is not required)
 
 
-Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):
+### Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):
 ...
 
 

contrib/krb5-cmake/CMakeLists.txt

@@ -1,4 +1,4 @@
-set (ENABLE_KRB5_DEFAULT 1)
+set (ENABLE_KRB5_DEFAULT ${ENABLE_LIBRARIES})
 if (NOT CMAKE_SYSTEM_NAME MATCHES "Linux" AND NOT (CMAKE_SYSTEM_NAME MATCHES "Darwin" AND NOT CMAKE_CROSSCOMPILING))
     message (WARNING "krb5 disabled in non-Linux and non-native-Darwin environments")
     set (ENABLE_KRB5_DEFAULT 0)

contrib/poco

@@ -1 +1 @@
-Subproject commit 520a90e02e3e5cb90afeae1846d161dbc508a6f1
+Subproject commit 008b16469471d55b176db181756c94e3f14dd2dc

contrib/unixodbc

@@ -1 +1 @@
-Subproject commit b0ad30f7f6289c12b76f04bfb9d466374bb32168
+Subproject commit a2cd5395e8c7f7390025ec93af5bfebef3fb5fcd

docker/keeper/Dockerfile

@@ -20,6 +20,8 @@ ENV LANG=en_US.UTF-8 \
 COPY --from=glibc-donor /lib/linux-gnu/libc.so.6 /lib/linux-gnu/libdl.so.2 /lib/linux-gnu/libm.so.6 /lib/linux-gnu/libpthread.so.0 /lib/linux-gnu/librt.so.1 /lib/linux-gnu/libnss_dns.so.2 /lib/linux-gnu/libnss_files.so.2 /lib/linux-gnu/libresolv.so.2 /lib/linux-gnu/ld-2.31.so /lib/
 COPY --from=glibc-donor /etc/nsswitch.conf /etc/
 COPY entrypoint.sh /entrypoint.sh
+
+ARG TARGETARCH
 RUN arch=${TARGETARCH:-amd64} \
     && case $arch in \
         amd64) mkdir -p /lib64 && ln -sf /lib/ld-2.31.so /lib64/ld-linux-x86-64.so.2 ;; \

docker/server/Dockerfile.ubuntu

@@ -94,8 +94,9 @@ RUN arch=${TARGETARCH:-amd64} \
         && apt-get update \
         && apt-get --yes -o "Dpkg::Options::=--force-confdef" -o "Dpkg::Options::=--force-confold" upgrade \
         && for package in ${PACKAGES}; do \
-            apt-get install --allow-unauthenticated --yes --no-install-recommends "${package}=${VERSION}" || exit 1 \
+            packages="${packages} ${package}=${VERSION}" \
         ; done \
+        && apt-get install --allow-unauthenticated --yes --no-install-recommends ${packages} || exit 1 \
     ; fi \
     && clickhouse-local -q 'SELECT * FROM system.build_options' \
     && rm -rf \

docs/en/sql-reference/functions/index.md

@@ -77,7 +77,7 @@ A function configuration contains the following settings:
 -   `argument` - argument description with the `type`, and optional `name` of an argument. Each argument is described in a separate setting. Specifying name is necessary if argument names are part of serialization for user defined function format like [Native](../../interfaces/formats.md#native) or [JSONEachRow](../../interfaces/formats.md#jsoneachrow). Default argument name value is `c` + argument_number.
 -   `format` - a [format](../../interfaces/formats.md) in which arguments are passed to the command.
 -   `return_type` - the type of a returned value.
--   `return_name` - name of retuned value. Specifying return name is necessary if return name is part of serialization for user defined function format like [Native](../../interfaces/formats.md#native) or [JSONEachRow](../../interfaces/formats.md#jsoneachrow). Optional. Default value is `result`.
+-   `return_name` - name of returned value. Specifying return name is necessary if return name is part of serialization for user defined function format like [Native](../../interfaces/formats.md#native) or [JSONEachRow](../../interfaces/formats.md#jsoneachrow). Optional. Default value is `result`.
 -   `type` - an executable type. If `type` is set to `executable` then single command is started. If it is set to `executable_pool` then a pool of commands is created.
 -   `max_command_execution_time` - maximum execution time in seconds for processing block of data. This setting is valid for `executable_pool` commands only. Optional. Default value is `10`.
 -   `command_termination_timeout` - time in seconds during which a command should finish after its pipe is closed. After that time `SIGTERM` is sent to the process executing the command. Optional. Default value is `10`.

programs/client/Client.cpp

@@ -810,7 +810,7 @@ void Client::addOptions(OptionsDescription & options_description)
         ("quota_key", po::value<std::string>(), "A string to differentiate quotas when the user have keyed quotas configured on server")
 
         ("max_client_network_bandwidth", po::value<int>(), "the maximum speed of data exchange over the network for the client in bytes per second.")
-        ("compression", po::value<bool>(), "enable or disable compression")
+        ("compression", po::value<bool>(), "enable or disable compression (enabled by default for remote communication and disabled for localhost communication).")
 
         ("query-fuzzer-runs", po::value<int>()->default_value(0), "After executing every SELECT query, do random mutations in it and run again specified number of times. This is used for testing to discover unexpected corner cases.")
         ("interleave-queries-file", po::value<std::vector<std::string>>()->multitoken(),

src/CMakeLists.txt

@@ -49,6 +49,18 @@ if (COMPILER_GCC)
     add_definitions ("-fno-tree-loop-distribute-patterns")
 endif ()
 
+# ClickHouse developers may use platform-dependent code under some macro (e.g. `#ifdef ENABLE_MULTITARGET`).
+# If turned ON, this option defines such macro.
+# See `src/Common/TargetSpecific.h`
+option(ENABLE_MULTITARGET_CODE "Enable platform-dependent code" ON)
+
+if (ENABLE_MULTITARGET_CODE)
+    add_definitions(-DENABLE_MULTITARGET_CODE=1)
+else()
+    add_definitions(-DENABLE_MULTITARGET_CODE=0)
+endif()
+
+
 add_subdirectory (Access)
 add_subdirectory (Backups)
 add_subdirectory (Columns)

src/Common/ColumnsHashingImpl.h

@@ -125,7 +125,7 @@ class FindResultImpl : public FindResultImplBase, public FindResultImplOffsetBas
 
 public:
     FindResultImpl()
-        : FindResultImplBase(false), FindResultImplOffsetBase<need_offset>(0)
+        : FindResultImplBase(false), FindResultImplOffsetBase<need_offset>(0) // NOLINT(clang-analyzer-optin.cplusplus.UninitializedObject)  intentionally allow uninitialized value here
     {}
 
     FindResultImpl(Mapped * value_, bool found_, size_t off)

src/Common/CompactArray.h

@@ -214,6 +214,9 @@ private:
 
         /// offset in bits to the next to the rightmost bit at that byte; or zero if the rightmost bit is the rightmost bit in that byte.
         offset_r = (l + content_width) % 8;
+
+        content_l = nullptr;
+        content_r = nullptr;
     }
 
     UInt8 ALWAYS_INLINE read(UInt8 value_l) const

src/Common/JSONBuilder.h

@@ -61,7 +61,7 @@ private:
 class JSONBool : public IItem
 {
 public:
-    explicit JSONBool(bool value_) : value(std::move(value_)) {}
+    explicit JSONBool(bool value_) : value(value_) {}
     void format(const FormatSettings & settings, FormatContext & context) override;
 
 private:
@@ -74,7 +74,7 @@ public:
     void add(ItemPtr value) { values.push_back(std::move(value)); }
     void add(std::string value) { add(std::make_unique<JSONString>(std::move(value))); }
     void add(const char * value) { add(std::make_unique<JSONString>(value)); }
-    void add(bool value) { add(std::make_unique<JSONBool>(std::move(value))); }
+    void add(bool value) { add(std::make_unique<JSONBool>(value)); }
 
     template <typename T>
     requires std::is_arithmetic_v<T>
@@ -99,7 +99,7 @@ public:
     void add(std::string key, std::string value) { add(std::move(key), std::make_unique<JSONString>(std::move(value))); }
     void add(std::string key, const char * value) { add(std::move(key), std::make_unique<JSONString>(value)); }
     void add(std::string key, std::string_view value) { add(std::move(key), std::make_unique<JSONString>(value)); }
-    void add(std::string key, bool value) { add(std::move(key), std::make_unique<JSONBool>(std::move(value))); }
+    void add(std::string key, bool value) { add(std::move(key), std::make_unique<JSONBool>(value)); }
 
     template <typename T>
     requires std::is_arithmetic_v<T>

src/Common/NamePrompter.cpp

@@ -0,0 +1,15 @@
+#include <IO/WriteHelpers.h>
+#include <Common/NamePrompter.h>
+
+namespace DB::detail
+{
+void appendHintsMessageImpl(String & message, const std::vector<String> & hints)
+{
+    if (hints.empty())
+    {
+        return;
+    }
+
+    message += ". Maybe you meant: " + toString(hints);
+}
+}

src/Common/NamePrompter.h

@@ -90,6 +90,10 @@ private:
     }
 };
 
+namespace detail
+{
+void appendHintsMessageImpl(String & message, const std::vector<String> & hints);
+}
 
 template <size_t MaxNumHints, typename Self>
 class IHints
@@ -102,6 +106,12 @@ public:
         return prompter.getHints(name, getAllRegisteredNames());
     }
 
+    void appendHintsMessage(String & message, const String & name) const
+    {
+        auto hints = getHints(name);
+        detail::appendHintsMessageImpl(message, hints);
+    }
+
     IHints() = default;
 
     IHints(const IHints &) = default;
@@ -114,5 +124,4 @@ public:
 private:
     NamePrompter<MaxNumHints> prompter;
 };
-
 }

src/Common/TargetSpecific.cpp

@@ -1,4 +1,4 @@
-#include <Functions/TargetSpecific.h>
+#include <Common/TargetSpecific.h>
 
 #include <Common/CpuId.h>
 

src/Core/Block.cpp

@@ -46,7 +46,8 @@ static ReturnType checkColumnStructure(const ColumnWithTypeAndName & actual, con
         return onError<ReturnType>("Block structure mismatch in " + std::string(context_description) + " stream: different names of columns:\n"
             + actual.dumpStructure() + "\n" + expected.dumpStructure(), code);
 
-    if (!actual.type->equals(*expected.type))
+    if ((actual.type && !expected.type) || (!actual.type && expected.type)
+        || (actual.type && expected.type && !actual.type->equals(*expected.type)))
         return onError<ReturnType>("Block structure mismatch in " + std::string(context_description) + " stream: different types:\n"
             + actual.dumpStructure() + "\n" + expected.dumpStructure(), code);
 

src/Core/SortCursor.h

@@ -15,10 +15,6 @@
 
 namespace DB
 {
-namespace ErrorCodes
-{
-    extern const int LOGICAL_ERROR;
-}
 
 /** Cursor allows to compare rows in different blocks (and parts).
   * Cursor moves inside single block.
@@ -61,25 +57,21 @@ struct SortCursorImpl
         reset(block, perm);
     }
 
-    SortCursorImpl(const Columns & columns, const SortDescription & desc_, size_t order_ = 0, IColumn::Permutation * perm = nullptr)
+    SortCursorImpl(
+        const Block & header,
+        const Columns & columns,
+        const SortDescription & desc_,
+        size_t order_ = 0,
+        IColumn::Permutation * perm = nullptr)
         : desc(desc_), sort_columns_size(desc.size()), order(order_), need_collation(desc.size())
     {
-        for (auto & column_desc : desc)
+        reset(columns, header, perm);
-        {
-            if (!column_desc.column_name.empty())
-                throw Exception("SortDescription should contain column position if SortCursor was used without header.",
-                        ErrorCodes::LOGICAL_ERROR);
-        }
-        reset(columns, {}, perm);
     }
 
     bool empty() const { return rows == 0; }
 
     /// Set the cursor to the beginning of the new block.
-    void reset(const Block & block, IColumn::Permutation * perm = nullptr)
+    void reset(const Block & block, IColumn::Permutation * perm = nullptr) { reset(block.getColumns(), block, perm); }
-    {
-        reset(block.getColumns(), block, perm);
-    }
 
     /// Set the cursor to the beginning of the new block.
     void reset(const Columns & columns, const Block & block, IColumn::Permutation * perm = nullptr)
@@ -95,9 +87,7 @@ struct SortCursorImpl
         for (size_t j = 0, size = desc.size(); j < size; ++j)
         {
             auto & column_desc = desc[j];
-            size_t column_number = !column_desc.column_name.empty()
+            size_t column_number = block.getPositionByName(column_desc.column_name);
-                                   ? block.getPositionByName(column_desc.column_name)
-                                   : column_desc.column_number;
             sort_columns.push_back(columns[column_number].get());
 
             need_collation[j] = desc[j].collator != nullptr && sort_columns.back()->isCollationSupported();
@@ -367,12 +357,12 @@ private:
 };
 
 template <typename TLeftColumns, typename TRightColumns>
-bool less(const TLeftColumns & lhs, const TRightColumns & rhs, size_t i, size_t j, const SortDescription & descr)
+bool less(const TLeftColumns & lhs, const TRightColumns & rhs, size_t i, size_t j, const SortDescriptionWithPositions & descr)
 {
     for (const auto & elem : descr)
     {
         size_t ind = elem.column_number;
-        int res = elem.direction * lhs[ind]->compareAt(i, j, *rhs[ind], elem.nulls_direction);
+        int res = elem.base.direction * lhs[ind]->compareAt(i, j, *rhs[ind], elem.base.nulls_direction);
         if (res < 0)
             return true;
         else if (res > 0)

src/Core/SortDescription.cpp

@@ -1,12 +1,12 @@
-#include <Core/SortDescription.h>
 #include <Core/Block.h>
+#include <Core/SortDescription.h>
 #include <IO/Operators.h>
 #include <Common/JSONBuilder.h>
 
 namespace DB
 {
 
-void dumpSortDescription(const SortDescription & description, const Block & header, WriteBuffer & out)
+void dumpSortDescription(const SortDescription & description, WriteBuffer & out)
 {
     bool first = true;
 
@@ -16,17 +16,7 @@ void dumpSortDescription(const SortDescription & description, const Block & head
             out << ", ";
         first = false;
 
-        if (!desc.column_name.empty())
+        out << desc.column_name;
-            out << desc.column_name;
-        else
-        {
-            if (desc.column_number < header.columns())
-                out << header.getByPosition(desc.column_number).name;
-            else
-                out << "?";
-
-            out << " (pos " << desc.column_number << ")";
-        }
 
         if (desc.direction > 0)
             out << " ASC";
@@ -38,18 +28,9 @@ void dumpSortDescription(const SortDescription & description, const Block & head
     }
 }
 
-void SortColumnDescription::explain(JSONBuilder::JSONMap & map, const Block & header) const
+void SortColumnDescription::explain(JSONBuilder::JSONMap & map) const
 {
-    if (!column_name.empty())
+    map.add("Column", column_name);
-        map.add("Column", column_name);
-    else
-    {
-        if (column_number < header.columns())
-            map.add("Column", header.getByPosition(column_number).name);
-
-        map.add("Position", column_number);
-    }
-
     map.add("Ascending", direction > 0);
     map.add("With Fill", with_fill);
 }
@@ -57,17 +38,17 @@ void SortColumnDescription::explain(JSONBuilder::JSONMap & map, const Block & he
 std::string dumpSortDescription(const SortDescription & description)
 {
     WriteBufferFromOwnString wb;
-    dumpSortDescription(description, Block{}, wb);
+    dumpSortDescription(description, wb);
     return wb.str();
 }
 
-JSONBuilder::ItemPtr explainSortDescription(const SortDescription & description, const Block & header)
+JSONBuilder::ItemPtr explainSortDescription(const SortDescription & description)
 {
     auto json_array = std::make_unique<JSONBuilder::JSONArray>();
     for (const auto & descr : description)
     {
         auto json_map = std::make_unique<JSONBuilder::JSONMap>();
-        descr.explain(*json_map, header);
+        descr.explain(*json_map);
         json_array->add(std::move(json_map));
     }
 

src/Core/SortDescription.h

@@ -39,7 +39,6 @@ struct FillColumnDescription
 struct SortColumnDescription
 {
     std::string column_name; /// The name of the column.
-    size_t column_number;    /// Column number (used if no name is given).
     int direction;           /// 1 - ascending, -1 - descending.
     int nulls_direction;     /// 1 - NULLs and NaNs are greater, -1 - less.
                              /// To achieve NULLS LAST, set it equal to direction, to achieve NULLS FIRST, set it opposite.
@@ -48,23 +47,24 @@ struct SortColumnDescription
     FillColumnDescription fill_description;
 
     explicit SortColumnDescription(
-            size_t column_number_, int direction_ = 1, int nulls_direction_ = 1,
+        const std::string & column_name_,
-            const std::shared_ptr<Collator> & collator_ = nullptr,
+        int direction_ = 1,
-            bool with_fill_ = false, const FillColumnDescription & fill_description_ = {})
+        int nulls_direction_ = 1,
-            : column_number(column_number_), direction(direction_), nulls_direction(nulls_direction_), collator(collator_)
+        const std::shared_ptr<Collator> & collator_ = nullptr,
-            , with_fill(with_fill_), fill_description(fill_description_) {}
+        bool with_fill_ = false,
-
+        const FillColumnDescription & fill_description_ = {})
-    explicit SortColumnDescription(
+        : column_name(column_name_)
-            const std::string & column_name_, int direction_ = 1, int nulls_direction_ = 1,
+        , direction(direction_)
-            const std::shared_ptr<Collator> & collator_ = nullptr,
+        , nulls_direction(nulls_direction_)
-            bool with_fill_ = false, const FillColumnDescription & fill_description_ = {})
+        , collator(collator_)
-            : column_name(column_name_), column_number(0), direction(direction_), nulls_direction(nulls_direction_)
+        , with_fill(with_fill_)
-            , collator(collator_), with_fill(with_fill_), fill_description(fill_description_) {}
+        , fill_description(fill_description_)
+    {
+    }
 
     bool operator == (const SortColumnDescription & other) const
     {
-        return column_name == other.column_name && column_number == other.column_number
+        return column_name == other.column_name && direction == other.direction && nulls_direction == other.nulls_direction;
-            && direction == other.direction && nulls_direction == other.nulls_direction;
     }
 
     bool operator != (const SortColumnDescription & other) const
@@ -72,22 +72,30 @@ struct SortColumnDescription
         return !(*this == other);
     }
 
-    std::string dump() const
+    std::string dump() const { return fmt::format("{}:dir {}nulls {}", column_name, direction, nulls_direction); }
-    {
-        return fmt::format("{}:{}:dir {}nulls ", column_name, column_number, direction, nulls_direction);
-    }
 
-    void explain(JSONBuilder::JSONMap & map, const Block & header) const;
+    void explain(JSONBuilder::JSONMap & map) const;
+};
+
+struct SortColumnDescriptionWithColumnIndex
+{
+    SortColumnDescription base;
+    size_t column_number;
+
+    SortColumnDescriptionWithColumnIndex(SortColumnDescription description_, size_t column_number_)
+        : base(std::move(description_)), column_number(column_number_)
+    {
+    }
 };
 
 /// Description of the sorting rule for several columns.
 using SortDescription = std::vector<SortColumnDescription>;
+using SortDescriptionWithPositions = std::vector<SortColumnDescriptionWithColumnIndex>;
 
 /// Outputs user-readable description into `out`.
-void dumpSortDescription(const SortDescription & description, const Block & header, WriteBuffer & out);
+void dumpSortDescription(const SortDescription & description, WriteBuffer & out);
 
 std::string dumpSortDescription(const SortDescription & description);
 
-JSONBuilder::ItemPtr explainSortDescription(const SortDescription & description, const Block & header);
+JSONBuilder::ItemPtr explainSortDescription(const SortDescription & description);
-
 }

src/DataTypes/ObjectUtils.cpp

@@ -128,22 +128,21 @@ static auto extractVector(const std::vector<Tuple> & vec)
     return res;
 }
 
-void convertObjectsToTuples(NamesAndTypesList & columns_list, Block & block, const NamesAndTypesList & extended_storage_columns)
+void convertObjectsToTuples(Block & block, const NamesAndTypesList & extended_storage_columns)
 {
     std::unordered_map<String, DataTypePtr> storage_columns_map;
     for (const auto & [name, type] : extended_storage_columns)
         storage_columns_map[name] = type;
 
-    for (auto & name_type : columns_list)
+    for (auto & column : block)
     {
-        if (!isObject(name_type.type))
+        if (!isObject(column.type))
             continue;
 
-        auto & column = block.getByName(name_type.name);
         if (!isObject(column.type))
             throw Exception(ErrorCodes::TYPE_MISMATCH,
                 "Type for column '{}' mismatch in columns list and in block. In list: {}, in block: {}",
-                name_type.name, name_type.type->getName(), column.type->getName());
+                column.name, column.type->getName(), column.type->getName());
 
         const auto & column_object = assert_cast<const ColumnObject &>(*column.column);
         const auto & subcolumns = column_object.getSubcolumns();
@@ -151,7 +150,7 @@ void convertObjectsToTuples(NamesAndTypesList & columns_list, Block & block, con
         if (!column_object.isFinalized())
             throw Exception(ErrorCodes::LOGICAL_ERROR,
                 "Cannot convert to tuple column '{}' from type {}. Column should be finalized first",
-                name_type.name, name_type.type->getName());
+                column.name, column.type->getName());
 
         PathsInData tuple_paths;
         DataTypes tuple_types;
@@ -164,12 +163,11 @@ void convertObjectsToTuples(NamesAndTypesList & columns_list, Block & block, con
             tuple_columns.emplace_back(entry->data.getFinalizedColumnPtr());
         }
 
-        auto it = storage_columns_map.find(name_type.name);
+        auto it = storage_columns_map.find(column.name);
         if (it == storage_columns_map.end())
-            throw Exception(ErrorCodes::LOGICAL_ERROR, "Column '{}' not found in storage", name_type.name);
+            throw Exception(ErrorCodes::LOGICAL_ERROR, "Column '{}' not found in storage", column.name);
 
         std::tie(column.column, column.type) = unflattenTuple(tuple_paths, tuple_types, tuple_columns);
-        name_type.type = column.type;
 
         /// Check that constructed Tuple type and type in storage are compatible.
         getLeastCommonTypeForObject({column.type, it->second}, true);

src/DataTypes/ObjectUtils.h

@@ -38,7 +38,7 @@ DataTypePtr getDataTypeByColumn(const IColumn & column);
 
 /// Converts Object types and columns to Tuples in @columns_list and @block
 /// and checks that types are consistent with types in @extended_storage_columns.
-void convertObjectsToTuples(NamesAndTypesList & columns_list, Block & block, const NamesAndTypesList & extended_storage_columns);
+void convertObjectsToTuples(Block & block, const NamesAndTypesList & extended_storage_columns);
 
 /// Checks that each path is not the prefix of any other path.
 void checkObjectHasNoAmbiguosPaths(const PathsInData & paths);

src/Disks/IO/CachedReadBufferFromRemoteFS.cpp

@@ -334,15 +334,17 @@ SeekableReadBufferPtr CachedReadBufferFromRemoteFS::getImplementationBuffer(File
                 read_buffer_for_file_segment->seek(file_offset_of_buffer_end, SEEK_SET);
             }
 
-            auto impl_range = read_buffer_for_file_segment->getRemainingReadRange();
             auto download_offset = file_segment->getDownloadOffset();
             if (download_offset != static_cast<size_t>(read_buffer_for_file_segment->getPosition()))
+            {
+                auto impl_range = read_buffer_for_file_segment->getRemainingReadRange();
                 throw Exception(
                     ErrorCodes::LOGICAL_ERROR,
                     "Buffer's offsets mismatch; cached buffer offset: {}, download_offset: {}, position: {}, implementation buffer offset: {}, "
                     "implementation buffer reading until: {}, file segment info: {}",
                     file_offset_of_buffer_end, download_offset, read_buffer_for_file_segment->getPosition(),
                     impl_range.left, *impl_range.right, file_segment->getInfoForLog());
+            }
 
             break;
         }
@@ -802,12 +804,14 @@ std::optional<size_t> CachedReadBufferFromRemoteFS::getLastNonDownloadedOffset()
 
 String CachedReadBufferFromRemoteFS::getInfoForLog()
 {
-    auto implementation_buffer_read_range_str =
+    String implementation_buffer_read_range_str;
-        implementation_buffer ?
+    if (implementation_buffer)
-        std::to_string(implementation_buffer->getRemainingReadRange().left)
+    {
-        + '-'
+        auto read_range = implementation_buffer->getRemainingReadRange();
-        + (implementation_buffer->getRemainingReadRange().right ? std::to_string(*implementation_buffer->getRemainingReadRange().right) : "None")
+        implementation_buffer_read_range_str = std::to_string(read_range.left) + '-' + (read_range.right ? std::to_string(*read_range.right) : "None");
-        : "None";
+    }
+    else
+        implementation_buffer_read_range_str = "None";
 
     auto current_file_segment_info = current_file_segment_it == file_segments_holder->file_segments.end() ? "None" : (*current_file_segment_it)->getInfoForLog();
 

src/Functions/CMakeLists.txt

@@ -96,17 +96,6 @@ if (TARGET ch_contrib::rapidjson)
     target_link_libraries(clickhouse_functions PRIVATE ch_contrib::rapidjson)
 endif()
 
-# ClickHouse developers may use platform-dependent code under some macro (e.g. `#ifdef ENABLE_MULTITARGET`).
-# If turned ON, this option defines such macro.
-# See `src/Functions/TargetSpecific.h`
-option(ENABLE_MULTITARGET_CODE "Enable platform-dependent code" ON)
-
-if (ENABLE_MULTITARGET_CODE)
-    add_definitions(-DENABLE_MULTITARGET_CODE=1)
-else()
-    add_definitions(-DENABLE_MULTITARGET_CODE=0)
-endif()
-
 add_subdirectory(GatherUtils)
 target_link_libraries(clickhouse_functions PRIVATE clickhouse_functions_gatherutils)
 

src/Functions/FunctionStartsEndsWith.h

@@ -1,12 +1,12 @@
 #pragma once
 #include <base/map.h>
 
+#include <Common/TargetSpecific.h>
 #include <Functions/FunctionHelpers.h>
 #include <Functions/GatherUtils/GatherUtils.h>
 #include <Functions/GatherUtils/Sources.h>
 #include <Functions/IFunction.h>
 #include <Functions/PerformanceAdaptors.h>
-#include <Functions/TargetSpecific.h>
 #include <DataTypes/DataTypeString.h>
 #include <DataTypes/DataTypesNumber.h>
 #include <DataTypes/getLeastSupertype.h>

src/Functions/FunctionsHashing.h

@@ -38,8 +38,8 @@
 #include <Columns/ColumnTuple.h>
 #include <Functions/IFunction.h>
 #include <Functions/FunctionHelpers.h>
-#include <Functions/TargetSpecific.h>
 #include <Functions/PerformanceAdaptors.h>
+#include <Common/TargetSpecific.h>
 #include <base/range.h>
 #include <base/bit_cast.h>
 

src/Functions/FunctionsRandom.h

@@ -1,9 +1,9 @@
 #pragma once
 
+#include <Common/TargetSpecific.h>
 #include <DataTypes/DataTypesNumber.h>
 #include <Columns/ColumnVector.h>
 #include <Functions/IFunction.h>
-#include <Functions/TargetSpecific.h>
 #include <Functions/PerformanceAdaptors.h>
 #include <IO/WriteHelpers.h>
 

src/Functions/GatherUtils/Algorithms.h

@@ -7,6 +7,8 @@
 #include <Core/AccurateComparison.h>
 #include <base/range.h>
 #include "GatherUtils.h"
+#include "sliceEqualElements.h"
+#include "sliceHasImplAnyAll.h"
 
 
 namespace DB::ErrorCodes
@@ -461,39 +463,19 @@ void NO_INLINE conditional(SourceA && src_a, SourceB && src_b, Sink && sink, con
 }
 
 
-/// Methods to check if first array has elements from second array, overloaded for various combinations of types.
+template <typename T>
-template <
+bool insliceEqualElements(const NumericArraySlice<T> & first [[maybe_unused]],
-    ArraySearchType search_type,
+                          size_t first_ind [[maybe_unused]],
-    typename FirstSliceType,
+                          size_t second_ind [[maybe_unused]])
-    typename SecondSliceType,
-          bool (*isEqual)(const FirstSliceType &, const SecondSliceType &, size_t, size_t)>
-bool sliceHasImplAnyAll(const FirstSliceType & first, const SecondSliceType & second, const UInt8 * first_null_map, const UInt8 * second_null_map)
 {
-    const bool has_first_null_map = first_null_map != nullptr;
+    if constexpr (is_decimal<T>)
-    const bool has_second_null_map = second_null_map != nullptr;
+        return accurate::equalsOp(first.data[first_ind].value, first.data[second_ind].value);
-
+    else
-    for (size_t i = 0; i < second.size; ++i)
+        return accurate::equalsOp(first.data[first_ind], first.data[second_ind]);
-    {
+}
-        bool has = false;
+inline ALWAYS_INLINE bool insliceEqualElements(const GenericArraySlice & first, size_t first_ind, size_t second_ind)
-        for (size_t j = 0; j < first.size && !has; ++j)
+{
-        {
+    return first.elements->compareAt(first_ind + first.begin, second_ind + first.begin, *first.elements, -1) == 0;
-            const bool is_first_null = has_first_null_map && first_null_map[j];
-            const bool is_second_null = has_second_null_map && second_null_map[i];
-
-            if (is_first_null && is_second_null)
-                has = true;
-
-            if (!is_first_null && !is_second_null && isEqual(first, second, j, i))
-                has = true;
-        }
-
-        if (has && search_type == ArraySearchType::Any)
-            return true;
-
-        if (!has && search_type == ArraySearchType::All)
-            return false;
-    }
-    return search_type == ArraySearchType::All;
 }
 
 template <
@@ -620,55 +602,6 @@ bool sliceHasImpl(const FirstSliceType & first, const SecondSliceType & second,
         return sliceHasImplAnyAll<search_type, FirstSliceType, SecondSliceType, isEqual>(first, second, first_null_map, second_null_map);
 }
 
-
-template <typename T, typename U>
-bool sliceEqualElements(const NumericArraySlice<T> & first [[maybe_unused]],
-                        const NumericArraySlice<U> & second [[maybe_unused]],
-                        size_t first_ind [[maybe_unused]],
-                        size_t second_ind [[maybe_unused]])
-{
-    /// TODO: Decimal scale
-    if constexpr (is_decimal<T> && is_decimal<U>)
-        return accurate::equalsOp(first.data[first_ind].value, second.data[second_ind].value);
-    else if constexpr (is_decimal<T> || is_decimal<U>)
-        return false;
-    else
-        return accurate::equalsOp(first.data[first_ind], second.data[second_ind]);
-}
-
-template <typename T>
-bool sliceEqualElements(const NumericArraySlice<T> &, const GenericArraySlice &, size_t, size_t)
-{
-    return false;
-}
-
-template <typename U>
-bool sliceEqualElements(const GenericArraySlice &, const NumericArraySlice<U> &, size_t, size_t)
-{
-    return false;
-}
-
-inline ALWAYS_INLINE bool sliceEqualElements(const GenericArraySlice & first, const GenericArraySlice & second, size_t first_ind, size_t second_ind)
-{
-    return first.elements->compareAt(first_ind + first.begin, second_ind + second.begin, *second.elements, -1) == 0;
-}
-
-template <typename T>
-bool insliceEqualElements(const NumericArraySlice<T> & first [[maybe_unused]],
-                          size_t first_ind [[maybe_unused]],
-                          size_t second_ind [[maybe_unused]])
-{
-    if constexpr (is_decimal<T>)
-        return accurate::equalsOp(first.data[first_ind].value, first.data[second_ind].value);
-    else
-        return accurate::equalsOp(first.data[first_ind], first.data[second_ind]);
-}
-
-inline ALWAYS_INLINE bool insliceEqualElements(const GenericArraySlice & first, size_t first_ind, size_t second_ind)
-{
-    return first.elements->compareAt(first_ind + first.begin, second_ind + first.begin, *first.elements, -1) == 0;
-}
-
 template <ArraySearchType search_type, typename T, typename U>
 bool sliceHas(const NumericArraySlice<T> & first, const NumericArraySlice<U> & second)
 {
@@ -854,4 +787,3 @@ void resizeConstantSize(ArraySource && array_source, ValueSource && value_source
 }
 
 }
-

src/Functions/GatherUtils/CMakeLists.txt

@@ -1,4 +1,5 @@
 include("${ClickHouse_SOURCE_DIR}/cmake/dbms_glob_sources.cmake")
+
 add_headers_and_sources(clickhouse_functions_gatherutils .)
 add_library(clickhouse_functions_gatherutils ${clickhouse_functions_gatherutils_sources} ${clickhouse_functions_gatherutils_headers})
 target_link_libraries(clickhouse_functions_gatherutils PRIVATE dbms)
@@ -14,3 +15,5 @@ endif()
 if (STRIP_DEBUG_SYMBOLS_FUNCTIONS)
     target_compile_options(clickhouse_functions_gatherutils PRIVATE "-g0")
 endif()
+
+set_target_properties(clickhouse_functions_gatherutils PROPERTIES COMPILE_FLAGS "${X86_INTRINSICS_FLAGS}")

src/Functions/GatherUtils/sliceEqualElements.h

@@ -0,0 +1,41 @@
+#pragma once
+
+#include <Core/AccurateComparison.h>
+#include "Slices.h"
+
+namespace DB::GatherUtils
+{
+
+template <typename T, typename U>
+bool sliceEqualElements(const NumericArraySlice<T> & first [[maybe_unused]],
+                        const NumericArraySlice<U> & second [[maybe_unused]],
+                        size_t first_ind [[maybe_unused]],
+                        size_t second_ind [[maybe_unused]])
+{
+    /// TODO: Decimal scale
+    if constexpr (is_decimal<T> && is_decimal<U>)
+        return accurate::equalsOp(first.data[first_ind].value, second.data[second_ind].value);
+    else if constexpr (is_decimal<T> || is_decimal<U>)
+        return false;
+    else
+        return accurate::equalsOp(first.data[first_ind], second.data[second_ind]);
+}
+
+template <typename T>
+bool sliceEqualElements(const NumericArraySlice<T> &, const GenericArraySlice &, size_t, size_t)
+{
+    return false;
+}
+
+template <typename U>
+bool sliceEqualElements(const GenericArraySlice &, const NumericArraySlice<U> &, size_t, size_t)
+{
+    return false;
+}
+
+inline ALWAYS_INLINE bool sliceEqualElements(const GenericArraySlice & first, const GenericArraySlice & second, size_t first_ind, size_t second_ind)
+{
+    return first.elements->compareAt(first_ind + first.begin, second_ind + second.begin, *second.elements, -1) == 0;
+}
+
+}

src/Functions/GatherUtils/sliceHasImplAnyAll.h

@@ -0,0 +1,943 @@
+#pragma once
+
+#include "GatherUtils.h"
+#include "Slices.h"
+#include "sliceEqualElements.h"
+
+#if defined(__SSE4_2__)
+    #include <emmintrin.h>
+    #include <smmintrin.h>
+    #include <nmmintrin.h>
+#endif
+
+#if defined(__AVX2__)
+    #include <immintrin.h>
+#endif
+
+#include <Common/TargetSpecific.h>
+
+
+namespace DB::GatherUtils
+{
+
+inline ALWAYS_INLINE bool hasNull(const UInt8 * null_map, size_t null_map_size)
+{
+    if (null_map == nullptr)
+        return false;
+
+    for (size_t i = 0; i < null_map_size; ++i)
+    {
+        if (null_map[i])
+            return true;
+    }
+
+    return false;
+}
+
+template<class T>
+inline ALWAYS_INLINE bool hasAllIntegralLoopRemainder(
+    size_t j, const NumericArraySlice<T> & first, const NumericArraySlice<T> & second, const UInt8 * first_null_map, const UInt8 * second_null_map)
+{
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map != nullptr;
+
+    for (; j < second.size; ++j)
+    {
+        // skip null elements since both have at least one - assuming it was checked earlier that at least one element in 'first' is null
+        if (has_second_null_map && second_null_map[j])
+            continue;
+
+        bool found = false;
+
+        for (size_t i = 0; i < first.size; ++i)
+        {
+            if (has_first_null_map && first_null_map[i])
+                continue;
+
+            if (first.data[i] == second.data[j])
+            {
+                found = true;
+                break;
+            }
+        }
+
+        if (!found)
+            return false;
+    }
+    return true;
+}
+
+
+#if defined(__AVX2__)
+
+DECLARE_AVX2_SPECIFIC_CODE (
+
+// AVX2 Int64, UInt64 specialization
+template<typename IntType>
+requires (std::is_same_v<IntType, Int64> || std::is_same_v<IntType, UInt64>)
+inline ALWAYS_INLINE bool sliceHasImplAnyAllImplInt64(
+    const NumericArraySlice<IntType> & first,
+    const NumericArraySlice<IntType> & second,
+    const UInt8 * first_null_map,
+    const UInt8 * second_null_map)
+{
+    if (second.size == 0)
+        return true;
+
+    if (!hasNull(first_null_map, first.size) && hasNull(second_null_map, second.size))
+        return false;
+
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map  != nullptr;
+
+    size_t j = 0;
+    int has_mask = 1;
+    static constexpr Int64 full = -1, none = 0;
+    const __m256i ones = _mm256_set1_epi64x(full);
+    const __m256i zeros = _mm256_setzero_si256();
+
+    if (second.size > 3 && first.size > 3)
+    {
+        for (; j < second.size - 3 && has_mask; j += 4)
+        {
+            has_mask = 0;
+            const __m256i second_data = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(second.data + j));
+            // bits of the bitmask are set to one if considered as null in the corresponding null map, 0 otherwise;
+            __m256i bitmask = has_second_null_map ?
+                _mm256_set_epi64x(
+                    (second_null_map[j + 3])? full : none,
+                    (second_null_map[j + 2])? full : none,
+                    (second_null_map[j + 1])? full : none,
+                    (second_null_map[j]) ? full : none)
+                : zeros;
+
+            size_t i = 0;
+            for (; i < first.size - 3 && !has_mask; has_mask = _mm256_testc_si256(bitmask, ones), i += 4)
+            {
+                const __m256i first_data = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(first.data + i));
+                const __m256i first_nm_mask = has_first_null_map?
+                    _mm256_set_m128i(
+                        _mm_cvtepi8_epi64(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i + 2))),
+                        _mm_cvtepi8_epi64(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i))))
+                    : zeros;
+                bitmask =
+                    _mm256_or_si256(
+                        _mm256_or_si256(
+                            _mm256_or_si256(
+                                    _mm256_andnot_si256(
+                                        first_nm_mask,
+                                        _mm256_cmpeq_epi64(second_data, first_data)),
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(5,4,3,2,1,0,7,6)),
+                                        _mm256_cmpeq_epi64(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(5,4,3,2,1,0,7,6))))),
+
+                            _mm256_or_si256(
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(3,2,1,0,7,6,5,4)),
+                                        _mm256_cmpeq_epi64(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(3,2,1,0,7,6,5,4)))),
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(1,0,7,6,5,4,3,2)),
+                                        _mm256_cmpeq_epi64(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(1,0,7,6,5,4,3,2)))))),
+                        bitmask);
+            }
+
+            if (i < first.size)
+            {
+                for (; i < first.size && !has_mask; ++i)
+                {
+                    if (has_first_null_map && first_null_map[i])
+                        continue;
+
+                    __m256i v_i = _mm256_set1_epi64x(first.data[i]);
+                    bitmask = _mm256_or_si256(bitmask, _mm256_cmpeq_epi64(second_data, v_i));
+                    has_mask = _mm256_testc_si256(bitmask, ones);
+                }
+            }
+        }
+    }
+
+    if (!has_mask && second.size > 3)
+        return false;
+
+    return hasAllIntegralLoopRemainder(j, first, second, first_null_map, second_null_map);
+}
+
+// AVX2 Int32, UInt32 specialization
+template<typename IntType>
+requires (std::is_same_v<IntType, Int32> || std::is_same_v<IntType, UInt32>)
+inline ALWAYS_INLINE bool sliceHasImplAnyAllImplInt32(
+    const NumericArraySlice<IntType> & first,
+    const NumericArraySlice<IntType> & second,
+    const UInt8 * first_null_map,
+    const UInt8 * second_null_map)
+{
+    if (second.size == 0)
+        return true;
+
+    if (!hasNull(first_null_map, first.size) && hasNull(second_null_map, second.size))
+        return false;
+
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map != nullptr;
+
+    size_t j = 0;
+    int has_mask = 1;
+    static constexpr int full = -1, none = 0;
+
+    const __m256i ones = _mm256_set1_epi32(full);
+    const __m256i zeros = _mm256_setzero_si256();
+
+    if (second.size > 7 && first.size > 7)
+    {
+        for (; j < second.size - 7 && has_mask; j += 8)
+        {
+            has_mask = 0;
+            const __m256i second_data = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(second.data + j));
+            // bits of the bitmask are set to one if considered as null in the corresponding null map, 0 otherwise;
+            __m256i bitmask = has_second_null_map ?
+                _mm256_set_epi32(
+                    (second_null_map[j + 7]) ? full : none,
+                    (second_null_map[j + 6]) ? full : none,
+                    (second_null_map[j + 5]) ? full : none,
+                    (second_null_map[j + 4]) ? full : none,
+                    (second_null_map[j + 3]) ? full : none,
+                    (second_null_map[j + 2]) ? full : none,
+                    (second_null_map[j + 1]) ? full : none,
+                    (second_null_map[j]) ? full : none)
+                : zeros;
+
+            size_t i = 0;
+            for (; i < first.size - 7 && !has_mask; has_mask = _mm256_testc_si256(bitmask, ones), i += 8)
+            {
+                const __m256i first_data = _mm256_loadu_si256(reinterpret_cast<const __m256i *>(first.data + i));
+                // Create a mask to avoid to compare null elements
+                // set_m128i takes two arguments: (high segment, low segment) that are two __m128i convert from 8bits to 32bits to match with next operations
+                const __m256i first_nm_mask = has_first_null_map?
+                    _mm256_set_m128i(
+                        _mm_cvtepi8_epi32(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i + 4))),
+                        _mm_cvtepi8_epi32(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i))))
+                    : zeros;
+                bitmask =
+                    _mm256_or_si256(
+                        _mm256_or_si256(
+                            _mm256_or_si256(
+                                _mm256_or_si256(
+                                    _mm256_andnot_si256(
+                                        first_nm_mask,
+                                        _mm256_cmpeq_epi32(second_data, first_data)),
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(6,5,4,3,2,1,0,7)),
+                                        _mm256_cmpeq_epi32(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(6,5,4,3,2,1,0,7))))),
+                                _mm256_or_si256(
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(5,4,3,2,1,0,7,6)),
+                                        _mm256_cmpeq_epi32(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(5,4,3,2,1,0,7,6)))),
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(4,3,2,1,0,7,6,5)),
+                                        _mm256_cmpeq_epi32(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(4,3,2,1,0,7,6,5)))))
+                            ),
+                            _mm256_or_si256(
+                                _mm256_or_si256(
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(3,2,1,0,7,6,5,4)),
+                                        _mm256_cmpeq_epi32(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(3,2,1,0,7,6,5,4)))),
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(2,1,0,7,6,5,4,3)),
+                                        _mm256_cmpeq_epi32(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(2,1,0,7,6,5,4,3))))),
+                                _mm256_or_si256(
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(1,0,7,6,5,4,3,2)),
+                                        _mm256_cmpeq_epi32(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(1,0,7,6,5,4,3,2)))),
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(0,7,6,5,4,3,2,1)),
+                                        _mm256_cmpeq_epi32(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(0,7,6,5,4,3,2,1))))))),
+                        bitmask);
+            }
+
+            if (i < first.size)
+            {
+                for (; i < first.size && !has_mask; ++i)
+                {
+                    if (has_first_null_map && first_null_map[i])
+                        continue;
+
+                    __m256i v_i = _mm256_set1_epi32(first.data[i]);
+                    bitmask = _mm256_or_si256(bitmask, _mm256_cmpeq_epi32(second_data, v_i));
+                    has_mask = _mm256_testc_si256(bitmask, ones);
+                }
+            }
+        }
+    }
+
+    if (!has_mask && second.size > 7)
+        return false;
+
+    return hasAllIntegralLoopRemainder(j, first, second, first_null_map, second_null_map);
+}
+
+// AVX2 Int16, UInt16 specialization
+template<typename IntType>
+requires (std::is_same_v<IntType, Int16> || std::is_same_v<IntType, UInt16>)
+inline ALWAYS_INLINE bool sliceHasImplAnyAllImplInt16(
+    const NumericArraySlice<IntType> & first,
+    const NumericArraySlice<IntType> & second,
+    const UInt8 * first_null_map,
+    const UInt8 * second_null_map)
+{
+    if (second.size == 0)
+        return true;
+
+    if (!hasNull(first_null_map, first.size) && hasNull(second_null_map, second.size))
+        return false;
+
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map  != nullptr;
+
+    size_t j = 0;
+    int has_mask = 1;
+    static constexpr int16_t full = -1, none = 0;
+    const __m256i ones = _mm256_set1_epi16(full);
+    const __m256i zeros = _mm256_setzero_si256();
+    if (second.size > 15 && first.size > 15)
+    {
+        for (; j < second.size - 15 && has_mask; j += 16)
+        {
+            has_mask = 0;
+            const __m256i second_data = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(second.data + j));
+            __m256i bitmask = has_second_null_map ?
+                _mm256_set_epi16(
+                    (second_null_map[j + 15]) ? full : none, (second_null_map[j + 14]) ? full : none,
+                    (second_null_map[j + 13]) ? full : none, (second_null_map[j + 12]) ? full : none,
+                    (second_null_map[j + 11]) ? full : none, (second_null_map[j + 10]) ? full : none,
+                    (second_null_map[j + 9]) ? full : none, (second_null_map[j + 8])? full : none,
+                    (second_null_map[j + 7]) ? full : none, (second_null_map[j + 6])? full : none,
+                    (second_null_map[j + 5]) ? full : none, (second_null_map[j + 4])? full : none,
+                    (second_null_map[j + 3]) ? full : none, (second_null_map[j + 2])? full : none,
+                    (second_null_map[j + 1]) ? full : none, (second_null_map[j]) ? full : none)
+                : zeros;
+
+            size_t i = 0;
+            for (; i < first.size - 15 && !has_mask; has_mask = _mm256_testc_si256(bitmask, ones), i += 16)
+            {
+                const __m256i first_data = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(first.data + i));
+                const __m256i first_nm_mask = has_first_null_map?
+                    _mm256_set_m128i(
+                        _mm_cvtepi8_epi16(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i + 8))),
+                        _mm_cvtepi8_epi16(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i))))
+                    : zeros;
+
+                bitmask =
+                    _mm256_or_si256(
+                        _mm256_or_si256(
+                            _mm256_or_si256(
+                                _mm256_or_si256(
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            first_nm_mask,
+                                            _mm256_cmpeq_epi16(second_data, first_data)),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(first_nm_mask, _mm256_set_epi8(29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(first_data, _mm256_set_epi8(29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30))))),
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(first_nm_mask, _mm256_set_epi8(27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(first_data, _mm256_set_epi8(27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28)))),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(first_nm_mask, _mm256_set_epi8(25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(first_data, _mm256_set_epi8(25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26)))))
+                                ),
+                                _mm256_or_si256(
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(first_nm_mask, _mm256_set_epi8(23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(first_data, _mm256_set_epi8(23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24)))),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(first_nm_mask, _mm256_set_epi8(21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(first_data, _mm256_set_epi8(21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22))))),
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(first_nm_mask, _mm256_set_epi8(19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(first_data, _mm256_set_epi8(19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20)))),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(first_nm_mask, _mm256_set_epi8(17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(first_data, _mm256_set_epi8(17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18))))))
+                            ),
+                            _mm256_or_si256(
+                                _mm256_or_si256(
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            _mm256_permute2x128_si256(first_nm_mask, first_nm_mask,1),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_permute2x128_si256(first_data, first_data, 1))),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_nm_mask, first_nm_mask, 1), _mm256_set_epi8(13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_data, first_data, 1), _mm256_set_epi8(13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14))))),
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_nm_mask, first_nm_mask, 1), _mm256_set_epi8(11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_data, first_data, 1), _mm256_set_epi8(11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12)))),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_nm_mask, first_nm_mask, 1), _mm256_set_epi8(9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_data, first_data, 1), _mm256_set_epi8(9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10)))))
+                                ),
+                                _mm256_or_si256(
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_nm_mask, first_nm_mask, 1), _mm256_set_epi8(7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_data ,first_data, 1), _mm256_set_epi8(7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8)))),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_nm_mask, first_nm_mask, 1), _mm256_set_epi8(5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_data, first_data, 1), _mm256_set_epi8(5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6))))),
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_nm_mask, first_nm_mask, 1), _mm256_set_epi8(3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_data ,first_data ,1), _mm256_set_epi8(3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4)))),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_nm_mask, first_nm_mask, 1), _mm256_set_epi8(1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_data, first_data, 1), _mm256_set_epi8(1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2))))))
+                        )
+                    ),
+                    bitmask);
+            }
+
+            if (i < first.size)
+            {
+                for (; i < first.size && !has_mask; ++i)
+                {
+                    if (has_first_null_map && first_null_map[i])
+                        continue;
+
+                    __m256i v_i = _mm256_set1_epi16(first.data[i]);
+                    bitmask = _mm256_or_si256(bitmask, _mm256_cmpeq_epi16(second_data, v_i));
+                    has_mask = _mm256_testc_si256(bitmask, ones);
+                }
+            }
+        }
+    }
+
+    if (!has_mask && second.size > 15)
+        return false;
+
+    return hasAllIntegralLoopRemainder(j, first, second, first_null_map, second_null_map);
+}
+
+)
+
+#endif
+
+#if defined(__SSE4_2__)
+
+DECLARE_SSE42_SPECIFIC_CODE (
+
+// SSE4.2 Int64, UInt64 specialization
+template<typename IntType>
+requires (std::is_same_v<IntType, Int64> || std::is_same_v<IntType, UInt64>)
+inline ALWAYS_INLINE bool sliceHasImplAnyAllImplInt64(
+    const NumericArraySlice<IntType> & first,
+    const NumericArraySlice<IntType> & second,
+    const UInt8 * first_null_map,
+    const UInt8 * second_null_map)
+{
+    if (second.size == 0)
+        return true;
+
+    if (!hasNull(first_null_map, first.size) && hasNull(second_null_map, second.size))
+        return false;
+
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map  != nullptr;
+
+    size_t j = 0;
+    int has_mask = 1;
+    static constexpr Int64 full = -1, none = 0;
+    const __m128i zeros = _mm_setzero_si128();
+    if (second.size > 1 && first.size > 1)
+    {
+        for (; j < second.size - 1 && has_mask; j += 2)
+        {
+            has_mask = 0;
+            const __m128i second_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(second.data + j));
+            __m128i bitmask = has_second_null_map ?
+                _mm_set_epi64x(
+                    (second_null_map[j + 1]) ? full : none,
+                    (second_null_map[j]) ? full : none)
+                : zeros;
+
+            size_t i = 0;
+
+            for (; i < first.size - 1 && !has_mask; has_mask = _mm_test_all_ones(bitmask), i += 2)
+            {
+                const __m128i first_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(first.data + i));
+                const __m128i first_nm_mask = has_first_null_map ?
+                    _mm_cvtepi8_epi64(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i)))
+                    : zeros;
+
+                bitmask =
+                    _mm_or_si128(
+                            _mm_or_si128(
+                                _mm_andnot_si128(
+                                    first_nm_mask,
+                                    _mm_cmpeq_epi64(second_data, first_data)),
+                                _mm_andnot_si128(
+                                    _mm_shuffle_epi32(first_nm_mask, _MM_SHUFFLE(1,0,3,2)),
+                                    _mm_cmpeq_epi64(second_data, _mm_shuffle_epi32(first_data, _MM_SHUFFLE(1,0,3,2))))),
+                        bitmask);
+            }
+
+            if (i < first.size)
+            {
+                for (; i < first.size && !has_mask; ++i)
+                {
+                    if (has_first_null_map && first_null_map[i])
+                        continue;
+
+                    __m128i v_i = _mm_set1_epi64x(first.data[i]);
+                    bitmask = _mm_or_si128(bitmask, _mm_cmpeq_epi64(second_data, v_i));
+                    has_mask = _mm_test_all_ones(bitmask);
+                }
+            }
+        }
+    }
+
+    if (!has_mask && second.size > 1)
+        return false;
+
+    return hasAllIntegralLoopRemainder(j, first, second, first_null_map, second_null_map);
+}
+
+// SSE4.2 Int32, UInt32 specialization
+template<typename IntType>
+requires (std::is_same_v<IntType, Int32> || std::is_same_v<IntType, UInt32>)
+inline ALWAYS_INLINE bool sliceHasImplAnyAllImplInt32(
+    const NumericArraySlice<IntType> & first,
+    const NumericArraySlice<IntType> & second,
+    const UInt8 * first_null_map,
+    const UInt8 * second_null_map)
+{
+    if (second.size == 0)
+        return true;
+
+    if (!hasNull(first_null_map, first.size) && hasNull(second_null_map, second.size))
+        return false;
+
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map  != nullptr;
+
+    size_t j = 0;
+    int has_mask = 1;
+    static constexpr int full = -1, none = 0;
+    const __m128i zeros = _mm_setzero_si128();
+    if (second.size > 3 && first.size > 3)
+    {
+        for (; j < second.size - 3 && has_mask; j += 4)
+        {
+            has_mask = 0;
+            const __m128i second_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(second.data + j));
+            __m128i bitmask = has_second_null_map ?
+                _mm_set_epi32(
+                    (second_null_map[j + 3]) ? full : none,
+                    (second_null_map[j + 2]) ? full : none,
+                    (second_null_map[j + 1]) ? full : none,
+                    (second_null_map[j]) ? full : none)
+                : zeros;
+
+            size_t i = 0;
+            for (; i < first.size - 3 && !has_mask; has_mask = _mm_test_all_ones(bitmask), i += 4)
+            {
+                const __m128i first_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(first.data + i));
+                const __m128i first_nm_mask = has_first_null_map ?
+                    _mm_cvtepi8_epi32(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i)))
+                    : zeros;
+
+                bitmask =
+                    _mm_or_si128(
+                        _mm_or_si128(
+                            _mm_or_si128(
+                                _mm_andnot_si128(
+                                        first_nm_mask,
+                                        _mm_cmpeq_epi32(second_data, first_data)),
+                                _mm_andnot_si128(
+                                    _mm_shuffle_epi32(first_nm_mask, _MM_SHUFFLE(2,1,0,3)),
+                                    _mm_cmpeq_epi32(second_data, _mm_shuffle_epi32(first_data, _MM_SHUFFLE(2,1,0,3))))),
+                            _mm_or_si128(
+                                _mm_andnot_si128(
+                                    _mm_shuffle_epi32(first_nm_mask, _MM_SHUFFLE(1,0,3,2)),
+                                    _mm_cmpeq_epi32(second_data, _mm_shuffle_epi32(first_data, _MM_SHUFFLE(1,0,3,2)))),
+                                _mm_andnot_si128(
+                                    _mm_shuffle_epi32(first_nm_mask, _MM_SHUFFLE(0,3,2,1)),
+                                    _mm_cmpeq_epi32(second_data, _mm_shuffle_epi32(first_data, _MM_SHUFFLE(0,3,2,1)))))
+                        ),
+                        bitmask);
+            }
+
+            if (i < first.size)
+            {
+                for (; i < first.size && !has_mask; ++i)
+                {
+                    if (has_first_null_map && first_null_map[i])
+                        continue;
+                    __m128i r_i = _mm_set1_epi32(first.data[i]);
+                    bitmask = _mm_or_si128(bitmask, _mm_cmpeq_epi32(second_data, r_i));
+                    has_mask = _mm_test_all_ones(bitmask);
+                }
+            }
+        }
+    }
+
+    if (!has_mask && second.size > 3)
+        return false;
+
+    return hasAllIntegralLoopRemainder(j, first, second, first_null_map, second_null_map);
+}
+
+// SSE4.2 Int16, UInt16 specialization
+template<typename IntType>
+requires (std::is_same_v<IntType, Int16> || std::is_same_v<IntType, UInt16>)
+inline ALWAYS_INLINE bool sliceHasImplAnyAllImplInt16(
+    const NumericArraySlice<IntType> & first,
+    const NumericArraySlice<IntType> & second,
+    const UInt8 * first_null_map,
+    const UInt8 * second_null_map)
+{
+    if (second.size == 0)
+        return true;
+
+    if (!hasNull(first_null_map, first.size) && hasNull(second_null_map, second.size))
+        return false;
+
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map  != nullptr;
+
+    size_t j = 0;
+    int has_mask = 1;
+    static constexpr int16_t full = -1, none = 0;
+    const __m128i zeros = _mm_setzero_si128();
+    if (second.size > 6 && first.size > 6)
+    {
+        for (; j < second.size - 7 && has_mask; j += 8)
+        {
+            has_mask = 0;
+            const __m128i second_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(second.data + j));
+            __m128i bitmask = has_second_null_map ?
+                _mm_set_epi16(
+                    (second_null_map[j + 7]) ? full : none, (second_null_map[j + 6]) ? full : none,
+                    (second_null_map[j + 5]) ? full : none, (second_null_map[j + 4]) ? full : none,
+                    (second_null_map[j + 3]) ? full : none, (second_null_map[j + 2]) ? full : none,
+                    (second_null_map[j + 1]) ? full : none, (second_null_map[j]) ? full: none)
+                : zeros;
+
+            size_t i = 0;
+            for (; i < first.size-7 && !has_mask; has_mask = _mm_test_all_ones(bitmask), i += 8)
+            {
+                const __m128i first_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(first.data + i));
+                const __m128i first_nm_mask = has_first_null_map ?
+                    _mm_cvtepi8_epi16(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i)))
+                    : zeros;
+                bitmask =
+                    _mm_or_si128(
+                            _mm_or_si128(
+                                _mm_or_si128(
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            first_nm_mask,
+                                            _mm_cmpeq_epi16(second_data, first_data)),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(13,12,11,10,9,8,7,6,5,4,3,2,1,0,15,14)),
+                                            _mm_cmpeq_epi16(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(13,12,11,10,9,8,7,6,5,4,3,2,1,0,15,14))))),
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(11,10,9,8,7,6,5,4,3,2,1,0,15,14,13,12)),
+                                            _mm_cmpeq_epi16(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(11,10,9,8,7,6,5,4,3,2,1,0,15,14,13,12)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(9,8,7,6,5,4,3,2,1,0,15,14,13,12,11,10)),
+                                            _mm_cmpeq_epi16(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(9,8,7,6,5,4,3,2,1,0,15,14,13,12,11,10)))))
+                                ),
+                                _mm_or_si128(
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8)),
+                                            _mm_cmpeq_epi16(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(5,4,3,2,1,0,15,14,13,12,11,10,9,8,7,6)),
+                                        _mm_cmpeq_epi16(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(5,4,3,2,1,0,15,14,13,12,11,10,9,8,7,6))))),
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(3,2,1,0,15,14,13,12,11,10,9,8,7,6,5,4)),
+                                            _mm_cmpeq_epi16(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(3,2,1,0,15,14,13,12,11,10,9,8,7,6,5,4)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(1,0,15,14,13,12,11,10,9,8,7,6,5,4,3,2)),
+                                            _mm_cmpeq_epi16(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(1,0,15,14,13,12,11,10,9,8,7,6,5,4,3,2))))))
+                        ),
+                        bitmask);
+            }
+
+            if (i < first.size)
+            {
+                for (; i < first.size && !has_mask; ++i)
+                {
+                    if (has_first_null_map && first_null_map[i])
+                        continue;
+                    __m128i v_i = _mm_set1_epi16(first.data[i]);
+                    bitmask = _mm_or_si128(bitmask, _mm_cmpeq_epi16(second_data, v_i));
+                    has_mask = _mm_test_all_ones(bitmask);
+                }
+            }
+        }
+    }
+
+    if (!has_mask && second.size > 6)
+        return false;
+
+    return hasAllIntegralLoopRemainder(j, first, second, first_null_map, second_null_map);
+}
+
+// Int8/UInt8 version is faster with SSE than with AVX2
+// SSE2 Int8, UInt8 specialization
+template<typename IntType>
+requires (std::is_same_v<IntType, Int8> || std::is_same_v<IntType, UInt8>)
+inline ALWAYS_INLINE bool sliceHasImplAnyAllImplInt8(
+    const NumericArraySlice<IntType> & first,
+    const NumericArraySlice<IntType> & second,
+    const UInt8 * first_null_map,
+    const UInt8 * second_null_map)
+{
+    if (second.size == 0)
+        return true;
+
+    if (!hasNull(first_null_map, first.size) && hasNull(second_null_map, second.size))
+        return false;
+
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map != nullptr;
+
+    size_t j = 0;
+    int has_mask = 1;
+    static constexpr int8_t full = -1, none = 0;
+    const __m128i zeros = _mm_setzero_si128();
+
+    if (second.size > 15 && first.size > 15)
+    {
+        for (; j < second.size - 15 && has_mask; j += 16)
+        {
+            has_mask = 0;
+            const __m128i second_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(second.data + j));
+            __m128i bitmask = has_second_null_map ?
+                _mm_set_epi8(
+                    (second_null_map[j + 15]) ? full : none, (second_null_map[j + 14]) ? full : none,
+                    (second_null_map[j + 13]) ? full : none, (second_null_map[j + 12]) ? full : none,
+                    (second_null_map[j + 11]) ? full : none, (second_null_map[j + 10]) ? full : none,
+                    (second_null_map[j + 9]) ? full : none, (second_null_map[j + 8]) ? full : none,
+                    (second_null_map[j + 7]) ? full : none, (second_null_map[j + 6]) ? full : none,
+                    (second_null_map[j + 5]) ? full : none, (second_null_map[j + 4]) ? full : none,
+                    (second_null_map[j + 3]) ? full : none, (second_null_map[j + 2]) ? full : none,
+                    (second_null_map[j + 1]) ? full : none, (second_null_map[j]) ? full : none)
+                : zeros;
+
+            size_t i = 0;
+            for (; i < first.size - 15 && !has_mask; has_mask = _mm_test_all_ones(bitmask), i += 16)
+            {
+                const __m128i first_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(first.data + i));
+                const __m128i first_nm_mask = has_first_null_map ?
+                    _mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i))
+                    : zeros;
+                bitmask =
+                    _mm_or_si128(
+                        _mm_or_si128(
+                            _mm_or_si128(
+                                _mm_or_si128(
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            first_nm_mask,
+                                            _mm_cmpeq_epi8(second_data, first_data)),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,15)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,15))))),
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(13,12,11,10,9,8,7,6,5,4,3,2,1,0,15,14)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(13,12,11,10,9,8,7,6,5,4,3,2,1,0,15,14)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(12,11,10,9,8,7,6,5,4,3,2,1,0,15,14,13)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(12,11,10,9,8,7,6,5,4,3,2,1,0,15,14,13)))))
+                                ),
+                                _mm_or_si128(
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(11,10,9,8,7,6,5,4,3,2,1,0,15,14,13,12)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(11,10,9,8,7,6,5,4,3,2,1,0,15,14,13,12)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(10,9,8,7,6,5,4,3,2,1,0,15,14,13,12,11)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(10,9,8,7,6,5,4,3,2,1,0,15,14,13,12,11))))),
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(9,8,7,6,5,4,3,2,1,0,15,14,13,12,11,10)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(9,8,7,6,5,4,3,2,1,0,15,14,13,12,11,10)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(8,7,6,5,4,3,2,1,0,15,14,13,12,11,10,9)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(8,7,6,5,4,3,2,1,0,15,14,13,12,11,10,9))))))),
+                            _mm_or_si128(
+                                _mm_or_si128(
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(6,5,4,3,2,1,0,15,14,13,12,11,10,9,8,7)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(6,5,4,3,2,1,0,15,14,13,12,11,10,9,8,7))))),
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(5,4,3,2,1,0,15,14,13,12,11,10,9,8,7,6)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(5,4,3,2,1,0,15,14,13,12,11,10,9,8,7,6)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(4,3,2,1,0,15,14,13,12,11,10,9,8,7,6,5)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(4,3,2,1,0,15,14,13,12,11,10,9,8,7,6,5)))))),
+                                _mm_or_si128(
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(3,2,1,0,15,14,13,12,11,10,9,8,7,6,5,4)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(3,2,1,0,15,14,13,12,11,10,9,8,7,6,5,4)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(2,1,0,15,14,13,12,11,10,9,8,7,6,5,4,3)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(2,1,0,15,14,13,12,11,10,9,8,7,6,5,4,3))))),
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(1,0,15,14,13,12,11,10,9,8,7,6,5,4,3,2)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(1,0,15,14,13,12,11,10,9,8,7,6,5,4,3,2)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(0,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(0,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1)))))))),
+                        bitmask);
+            }
+
+            if (i < first.size)
+            {
+                for (; i < first.size && !has_mask; ++i)
+                {
+                    if (has_first_null_map && first_null_map[i])
+                        continue;
+
+                    __m128i v_i = _mm_set1_epi8(first.data[i]);
+                    bitmask = _mm_or_si128(bitmask, _mm_cmpeq_epi8(second_data, v_i));
+                    has_mask = _mm_test_all_ones(bitmask);
+                }
+            }
+        }
+    }
+
+    if (!has_mask && second.size > 15)
+        return false;
+
+    return hasAllIntegralLoopRemainder(j, first, second, first_null_map, second_null_map);
+}
+
+)
+
+#endif
+
+template <
+    ArraySearchType search_type,
+    typename FirstSliceType,
+    typename SecondSliceType,
+    bool (*isEqual)(const FirstSliceType &, const SecondSliceType &, size_t, size_t)>
+bool sliceHasImplAnyAllGenericImpl(const FirstSliceType & first, const SecondSliceType & second, const UInt8 * first_null_map, const UInt8 * second_null_map)
+{
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map != nullptr;
+
+    const bool has_second_null = hasNull(second_null_map, second.size);
+    if (has_second_null)
+    {
+        const bool has_first_null = hasNull(first_null_map, first.size);
+
+        if (has_first_null && search_type == ArraySearchType::Any)
+            return true;
+
+        if (!has_first_null && search_type == ArraySearchType::All)
+            return false;
+    }
+
+    for (size_t i = 0; i < second.size; ++i)
+    {
+        if (has_second_null_map && second_null_map[i])
+            continue;
+
+        bool has = false;
+
+        for (size_t j = 0; j < first.size && !has; ++j)
+        {
+            if (has_first_null_map && first_null_map[j])
+                continue;
+
+            if (isEqual(first, second, j, i))
+            {
+                has = true;
+                break;
+            }
+        }
+
+        if (has && search_type == ArraySearchType::Any)
+            return true;
+
+        if (!has && search_type == ArraySearchType::All)
+            return false;
+    }
+
+    return search_type == ArraySearchType::All;
+}
+
+/// Methods to check if first array has elements from second array, overloaded for various combinations of types.
+template <
+    ArraySearchType search_type,
+    typename FirstSliceType,
+    typename SecondSliceType,
+    bool (*isEqual)(const FirstSliceType &, const SecondSliceType &, size_t, size_t)>
+inline ALWAYS_INLINE bool sliceHasImplAnyAll(const FirstSliceType & first, const SecondSliceType & second, const UInt8 * first_null_map, const UInt8 * second_null_map)
+{
+#if USE_MULTITARGET_CODE
+    if constexpr (search_type == ArraySearchType::All && std::is_same_v<FirstSliceType, SecondSliceType>)
+    {
+
+#if defined(__AVX2__)
+        if (isArchSupported(TargetArch::AVX2))
+        {
+            if constexpr (std::is_same_v<FirstSliceType, NumericArraySlice<Int16>> || std::is_same_v<FirstSliceType, NumericArraySlice<UInt16>>)
+            {
+                return GatherUtils::TargetSpecific::AVX2::sliceHasImplAnyAllImplInt16(first, second, first_null_map, second_null_map);
+            }
+            else if constexpr (std::is_same_v<FirstSliceType, NumericArraySlice<Int32>> || std::is_same_v<FirstSliceType, NumericArraySlice<UInt32>>)
+            {
+                return GatherUtils::TargetSpecific::AVX2::sliceHasImplAnyAllImplInt32(first, second, first_null_map, second_null_map);
+            }
+            else if constexpr (std::is_same_v<FirstSliceType, NumericArraySlice<Int64>> || std::is_same_v<FirstSliceType, NumericArraySlice<UInt64>>)
+            {
+                return GatherUtils::TargetSpecific::AVX2::sliceHasImplAnyAllImplInt64(first, second, first_null_map, second_null_map);
+            }
+        }
+#endif
+
+        if (isArchSupported(TargetArch::SSE42))
+        {
+            if constexpr (std::is_same_v<FirstSliceType, NumericArraySlice<Int8>> || std::is_same_v<FirstSliceType, NumericArraySlice<UInt8>>)
+            {
+                return TargetSpecific::SSE42::sliceHasImplAnyAllImplInt8(first, second, first_null_map, second_null_map);
+            }
+            else if constexpr (std::is_same_v<FirstSliceType, NumericArraySlice<Int16>> || std::is_same_v<FirstSliceType, NumericArraySlice<UInt16>>)
+            {
+                return GatherUtils::TargetSpecific::SSE42::sliceHasImplAnyAllImplInt16(first, second, first_null_map, second_null_map);
+            }
+            else if constexpr (std::is_same_v<FirstSliceType, NumericArraySlice<Int32>> || std::is_same_v<FirstSliceType, NumericArraySlice<UInt32>>)
+            {
+                return GatherUtils::TargetSpecific::SSE42::sliceHasImplAnyAllImplInt32(first, second, first_null_map, second_null_map);
+            }
+            else if constexpr (std::is_same_v<FirstSliceType, NumericArraySlice<Int64>> || std::is_same_v<FirstSliceType, NumericArraySlice<UInt64>>)
+            {
+                return GatherUtils::TargetSpecific::SSE42::sliceHasImplAnyAllImplInt64(first, second, first_null_map, second_null_map);
+            }
+        }
+    }
+#endif
+
+    return sliceHasImplAnyAllGenericImpl<search_type, FirstSliceType, SecondSliceType, isEqual>(first, second, first_null_map, second_null_map);
+}
+
+
+}

src/Functions/PerformanceAdaptors.h

@@ -1,8 +1,8 @@
 #pragma once
 
-#include <Functions/TargetSpecific.h>
 #include <Functions/IFunction.h>
 
+#include <Common/TargetSpecific.h>
 #include <Common/Stopwatch.h>
 #include <Interpreters/Context.h>
 

src/Functions/greatCircleDistance.cpp

@@ -6,8 +6,8 @@
 #include <Functions/IFunction.h>
 #include <Functions/FunctionHelpers.h>
 #include <Functions/FunctionFactory.h>
-#include <Functions/TargetSpecific.h>
 #include <Functions/PerformanceAdaptors.h>
+#include <Common/TargetSpecific.h>
 #include <base/range.h>
 #include <cmath>
 

src/Functions/tests/gtest_has_all.cpp

@@ -0,0 +1,139 @@
+#include <random>
+#include <gtest/gtest.h>
+#include <Functions/GatherUtils/Algorithms.h>
+
+using namespace DB::GatherUtils;
+
+
+auto uni_int_dist(int min, int max)
+{
+    std::random_device rd;
+    std::mt19937 mt(rd());
+    std::uniform_int_distribution<> dist(min, max);
+    return std::make_pair(dist, mt);
+}
+
+template<class T>
+void arrayInit(T* elements_to_have, size_t nb_elements_to_have, T* array_elements, size_t array_size, bool all_elements_present)
+{
+    for (size_t i = 0; i < array_size; ++i)
+    {
+        array_elements[i] = i;
+    }
+    auto [dist, gen] = uni_int_dist(0, array_size - 1);
+    for (size_t i = 0; i < nb_elements_to_have; ++i)
+    {
+        elements_to_have[i] = array_elements[dist(gen)];
+    }
+    if (!all_elements_present)
+    {
+        /// make one element to be searched for missing from the target array
+        elements_to_have[nb_elements_to_have - 1] = array_size + 1;
+    }
+}
+
+void nullMapInit(UInt8 * null_map, size_t null_map_size, size_t nb_null_elements)
+{
+    /// -2 to keep the last element of the array non-null
+    auto [dist, gen] = uni_int_dist(0, null_map_size - 2);
+    for (size_t i = 0; i < null_map_size; ++i)
+    {
+        null_map[i] = 0;
+    }
+    for (size_t i = 0; i < null_map_size - 1 && i < nb_null_elements; ++i)
+    {
+        null_map[dist(gen)] = 1;
+    }
+}
+
+template<class T>
+bool testHasAll(size_t nb_elements_to_have, size_t array_size, bool with_null_maps, bool all_elements_present)
+{
+    auto array_elements = std::make_unique<T[]>(array_size);
+    auto elements_to_have = std::make_unique<T[]>(nb_elements_to_have);
+
+    std::unique_ptr<UInt8[]> first_nm = nullptr, second_nm = nullptr;
+    if (with_null_maps)
+    {
+        first_nm = std::make_unique<UInt8[]>(array_size);
+        second_nm = std::make_unique<UInt8[]>(nb_elements_to_have);
+        /// add a null to elements to have, but not to the target array, making the answer negative
+        nullMapInit(first_nm.get(), array_size, 0);
+        nullMapInit(second_nm.get(), nb_elements_to_have, 1);
+    }
+
+    arrayInit(elements_to_have.get(), nb_elements_to_have, array_elements.get(), array_size, all_elements_present);
+
+    NumericArraySlice<T> first = {array_elements.get(), array_size};
+    NumericArraySlice<T> second = {elements_to_have.get(), nb_elements_to_have};
+
+    /// check whether all elements of the second array are also elements of the first array, overloaded for various combinations of types.
+    return sliceHasImplAnyAll<ArraySearchType::All, NumericArraySlice<T>, NumericArraySlice<T>, sliceEqualElements<T,T> >(
+        first, second, first_nm.get(), second_nm.get());
+}
+
+TEST(HasAll, integer)
+{
+    bool test1 = testHasAll<int>(4, 100, false, true);
+    bool test2 = testHasAll<int>(4, 100, false, false);
+    bool test3 = testHasAll<int>(100, 4096, false, true);
+    bool test4 = testHasAll<int>(100, 4096, false, false);
+
+    ASSERT_EQ(test1, true);
+    ASSERT_EQ(test2, false);
+    ASSERT_EQ(test3, true);
+    ASSERT_EQ(test4, false);
+}
+
+
+TEST(HasAll, int64)
+{
+    bool test1 = testHasAll<int64_t>(2, 100, false, true);
+    bool test2 = testHasAll<int64_t>(2, 100, false, false);
+    bool test3 = testHasAll<int64_t>(100, 4096, false, true);
+    bool test4 = testHasAll<int64_t>(100, 4096, false, false);
+
+    ASSERT_EQ(test1, true);
+    ASSERT_EQ(test2, false);
+    ASSERT_EQ(test3, true);
+    ASSERT_EQ(test4, false);
+}
+
+TEST(HasAll, int16)
+{
+    bool test1 = testHasAll<int16_t>(2, 100, false, true);
+    bool test2 = testHasAll<int16_t>(2, 100, false, false);
+    bool test3 = testHasAll<int16_t>(100, 4096, false, true);
+    bool test4 = testHasAll<int16_t>(100, 4096, false, false);
+
+    ASSERT_EQ(test1, true);
+    ASSERT_EQ(test2, false);
+    ASSERT_EQ(test3, true);
+    ASSERT_EQ(test4, false);
+}
+
+TEST(HasAll, int8)
+{
+    bool test1 = testHasAll<int8_t>(2, 100, false, true);
+    bool test2 = testHasAll<int8_t>(2, 100, false, false);
+    bool test3 = testHasAll<int8_t>(50, 125, false, true);
+    bool test4 = testHasAll<int8_t>(50, 125, false, false);
+
+    ASSERT_EQ(test1, true);
+    ASSERT_EQ(test2, false);
+    ASSERT_EQ(test3, true);
+    ASSERT_EQ(test4, false);
+}
+
+TEST(HasAllSingleNullElement, all)
+{
+    bool test1 = testHasAll<int>(4, 100, true, true);
+    bool test2 = testHasAll<int64_t>(4, 100, true, true);
+    bool test3 = testHasAll<int16_t>(4, 100, true, true);
+    bool test4 = testHasAll<int8_t>(4, 100, true, true);
+
+    ASSERT_EQ(test1, false);
+    ASSERT_EQ(test2, false);
+    ASSERT_EQ(test3, false);
+    ASSERT_EQ(test4, false);
+}

src/IO/ReadBufferFromFileDescriptor.h

@@ -39,10 +39,6 @@ public:
     {
     }
 
-    virtual ~ReadBufferFromFileDescriptor() override
-    {
-    }
-
     int getFD() const
     {
         return fd;
@@ -84,9 +80,6 @@ public:
     {
         use_pread = true;
     }
-    virtual ~ReadBufferFromFileDescriptorPRead() override
-    {
-    }
 };
 
 }

src/IO/ReadBufferFromS3.cpp

@@ -240,7 +240,7 @@ void ReadBufferFromS3::setReadUntilPosition(size_t position)
 
 SeekableReadBuffer::Range ReadBufferFromS3::getRemainingReadRange() const
 {
-    return Range{.left = static_cast<size_t>(offset), .right = read_until_position ? std::optional{read_until_position - 1} : std::nullopt};
+    return Range{ .left = static_cast<size_t>(offset), .right = read_until_position ? std::optional{read_until_position - 1} : std::nullopt };
 }
 
 std::unique_ptr<ReadBuffer> ReadBufferFromS3::initialize()

src/IO/ReadBufferFromS3.h

@@ -33,8 +33,11 @@ private:
     String key;
     UInt64 max_single_read_retries;
 
-    off_t offset = 0;
+    /// These variables are atomic because they can be used for `logging only`
-    off_t read_until_position = 0;
+    /// (where it is not important to get consistent result)
+    /// from separate thread other than the one which uses the buffer for s3 reading.
+    std::atomic<off_t> offset = 0;
+    std::atomic<off_t> read_until_position = 0;
 
     Aws::S3::Model::GetObjectResult read_result;
     std::unique_ptr<ReadBuffer> impl;

src/Interpreters/ActionsDAG.cpp

@@ -1527,6 +1527,21 @@ ActionsDAG::SplitResult ActionsDAG::splitActionsBeforeArrayJoin(const NameSet &
     return res;
 }
 
+ActionsDAG::SplitResult ActionsDAG::splitActionsBySortingDescription(const NameSet & sort_columns) const
+{
+    std::unordered_set<const Node *> split_nodes;
+    for (const auto & sort_column : sort_columns)
+        if (const auto * node = tryFindInIndex(sort_column))
+            split_nodes.insert(node);
+        else
+            throw Exception(
+                ErrorCodes::LOGICAL_ERROR, "Sorting column {} wasn't found in the ActionsDAG's index. DAG:\n{}", sort_column, dumpDAG());
+
+    auto res = split(split_nodes);
+    res.second->project_input = project_input;
+    return res;
+}
+
 ActionsDAG::SplitResult ActionsDAG::splitActionsForFilter(const std::string & column_name) const
 {
     const auto * node = tryFindInIndex(column_name);

src/Interpreters/ActionsDAG.h

@@ -274,6 +274,10 @@ public:
     /// Index of initial actions must contain column_name.
     SplitResult splitActionsForFilter(const std::string & column_name) const;
 
+    /// Splits actions into two parts. The first part contains all the calculations required to calculate sort_columns.
+    /// The second contains the rest.
+    SplitResult splitActionsBySortingDescription(const NameSet & sort_columns) const;
+
     /// Create actions which may calculate part of filter using only available_inputs.
     /// If nothing may be calculated, returns nullptr.
     /// Otherwise, return actions which inputs are from available_inputs.

src/Interpreters/ExternalDictionariesLoader.cpp

@@ -136,7 +136,7 @@ std::string ExternalDictionariesLoader::resolveDictionaryNameFromDatabaseCatalog
 
     if (qualified_name->database.empty())
     {
-        /// Ether database name is not specified and we should use current one
+        /// Either database name is not specified and we should use current one
         /// or it's an XML dictionary.
         bool is_xml_dictionary = has(name);
         if (is_xml_dictionary)

src/Interpreters/InterpreterSelectQuery.cpp

@@ -2249,10 +2249,6 @@ static bool windowDescriptionComparator(const WindowDescription * _left, const W
             return true;
         else if (left[i].column_name > right[i].column_name)
             return false;
-        else if (left[i].column_number < right[i].column_number)
-            return true;
-        else if (left[i].column_number > right[i].column_number)
-            return false;
         else if (left[i].direction < right[i].direction)
             return true;
         else if (left[i].direction > right[i].direction)

src/Interpreters/MutationsInterpreter.cpp

@@ -1025,7 +1025,7 @@ std::optional<SortDescription> MutationsInterpreter::getStorageSortDescriptionIf
     for (size_t i = 0; i < sort_columns_size; ++i)
     {
         if (header.has(sort_columns[i]))
-            sort_description.emplace_back(header.getPositionByName(sort_columns[i]), 1, 1);
+            sort_description.emplace_back(sort_columns[i], 1, 1);
         else
             return {};
     }

src/Interpreters/Set.cpp

@@ -430,8 +430,8 @@ MergeTreeSetIndex::MergeTreeSetIndex(const Columns & set_elements, std::vector<K
     SortDescription sort_description;
     for (size_t i = 0; i < tuple_size; ++i)
     {
-        block_to_sort.insert({ ordered_set[i], nullptr, "_" + toString(i) });
+        block_to_sort.insert({ordered_set[i], nullptr, ordered_set[i]->getName()});
-        sort_description.emplace_back(i, 1, 1);
+        sort_description.emplace_back(ordered_set[i]->getName(), 1, 1);
     }
 
     sortBlock(block_to_sort, sort_description);

src/Interpreters/SystemLog.cpp

@@ -379,7 +379,7 @@ void SystemLog<LogElement>::flushImpl(const std::vector<LogElement> & to_flush,
         for (const auto & name_and_type : log_element_names_and_types)
             log_element_columns.emplace_back(name_and_type.type, name_and_type.name);
 
-        Block block(log_element_columns);
+        Block block(std::move(log_element_columns));
 
         MutableColumns columns = block.mutateColumns();
         for (const auto & elem : to_flush)

src/Interpreters/TableJoin.cpp

@@ -512,14 +512,6 @@ TableJoin::createConvertingActions(const ColumnsWithTypeAndName & left_sample_co
 template <typename LeftNamesAndTypes, typename RightNamesAndTypes>
 void TableJoin::inferJoinKeyCommonType(const LeftNamesAndTypes & left, const RightNamesAndTypes & right, bool allow_right)
 {
-    if (strictness() == ASTTableJoin::Strictness::Asof)
-    {
-        if (clauses.size() != 1)
-            throw DB::Exception("ASOF join over multiple keys is not supported", ErrorCodes::NOT_IMPLEMENTED);
-        if (right.back().type->isNullable())
-            throw DB::Exception("ASOF join over right table Nullable column is not implemented", ErrorCodes::NOT_IMPLEMENTED);
-    }
-
     if (!left_type_map.empty() || !right_type_map.empty())
         return;
 
@@ -531,6 +523,15 @@ void TableJoin::inferJoinKeyCommonType(const LeftNamesAndTypes & left, const Rig
     for (const auto & col : right)
         right_types[renamedRightColumnName(col.name)] = col.type;
 
+    if (strictness() == ASTTableJoin::Strictness::Asof)
+    {
+        if (clauses.size() != 1)
+            throw DB::Exception("ASOF join over multiple keys is not supported", ErrorCodes::NOT_IMPLEMENTED);
+
+        auto asof_key_type = right_types.find(clauses.back().key_names_right.back());
+        if (asof_key_type != right_types.end() && asof_key_type->second->isNullable())
+            throw DB::Exception("ASOF join over right table Nullable column is not implemented", ErrorCodes::NOT_IMPLEMENTED);
+    }
 
     forAllKeys(clauses, [&](const auto & left_key_name, const auto & right_key_name)
     {

src/Interpreters/sortBlock.cpp

@@ -98,9 +98,7 @@ ColumnsWithSortDescriptions getColumnsWithSortDescription(const Block & block, c
     {
         const auto & sort_column_description = description[i];
 
-        const IColumn * column = !sort_column_description.column_name.empty()
+        const IColumn * column = block.getByName(sort_column_description.column_name).column.get();
-            ? block.getByName(sort_column_description.column_name).column.get()
-            : block.safeGetByPosition(sort_column_description.column_number).column.get();
 
         if (isCollationRequired(sort_column_description))
         {

src/Parsers/ASTInsertQuery.cpp

@@ -81,9 +81,17 @@ void ASTInsertQuery::formatImpl(const FormatSettings & settings, FormatState & s
 
     if (infile)
     {
-        settings.ostr << (settings.hilite ? hilite_keyword : "") << " FROM INFILE " << (settings.hilite ? hilite_none : "") << infile->as<ASTLiteral &>().value.safeGet<std::string>();
+        settings.ostr
+            << (settings.hilite ? hilite_keyword : "")
+            << " FROM INFILE "
+            << (settings.hilite ? hilite_none : "")
+            << quoteString(infile->as<ASTLiteral &>().value.safeGet<std::string>());
         if (compression)
-            settings.ostr << (settings.hilite ? hilite_keyword : "") << " COMPRESSION " << (settings.hilite ? hilite_none : "") << compression->as<ASTLiteral &>().value.safeGet<std::string>();
+            settings.ostr
+                << (settings.hilite ? hilite_keyword : "")
+                << " COMPRESSION "
+                << (settings.hilite ? hilite_none : "")
+                << quoteString(compression->as<ASTLiteral &>().value.safeGet<std::string>());
     }
 
     if (select)

src/Parsers/ASTPartition.h

@@ -15,6 +15,7 @@ public:
     size_t fields_count = 0;
 
     String id;
+    bool all = false;
 
     String getID(char) const override;
     ASTPtr clone() const override;

src/Parsers/ParserPartition.cpp

@@ -5,6 +5,7 @@
 #include <Parsers/ASTPartition.h>
 #include <Parsers/ASTLiteral.h>
 #include <Parsers/ASTFunction.h>
+#include <Parsers/ASTIdentifier.h>
 #include <Common/typeid_cast.h>
 
 namespace DB
@@ -13,6 +14,7 @@ namespace DB
 bool ParserPartition::parseImpl(Pos & pos, ASTPtr & node, Expected & expected)
 {
     ParserKeyword s_id("ID");
+    ParserKeyword s_all("ALL");
     ParserStringLiteral parser_string_literal;
     ParserExpression parser_expr;
 
@@ -28,6 +30,14 @@ bool ParserPartition::parseImpl(Pos & pos, ASTPtr & node, Expected & expected)
 
         partition->id = partition_id->as<ASTLiteral &>().value.get<String>();
     }
+    else if (s_all.ignore(pos, expected))
+    {
+        ASTPtr value = makeASTFunction("tuple");
+        partition->value = value;
+        partition->children.push_back(value);
+        partition->fields_count = 0;
+        partition->all = true;
+    }
     else
     {
         ASTPtr value;

src/Processors/LimitTransform.cpp

@@ -38,12 +38,7 @@ LimitTransform::LimitTransform(
     }
 
     for (const auto & desc : description)
-    {
+        sort_column_positions.push_back(header_.getPositionByName(desc.column_name));
-        if (!desc.column_name.empty())
-            sort_column_positions.push_back(header_.getPositionByName(desc.column_name));
-        else
-            sort_column_positions.push_back(desc.column_number);
-    }
 }
 
 Chunk LimitTransform::makeChunkWithPreviousRow(const Chunk & chunk, UInt64 row) const

src/Processors/Merges/Algorithms/AggregatingSortedAlgorithm.cpp

@@ -104,7 +104,7 @@ static AggregatingSortedAlgorithm::ColumnsDefinition defineColumns(
         /// Included into PK?
         auto it = description.begin();
         for (; it != description.end(); ++it)
-            if (it->column_name == column.name || (it->column_name.empty() && it->column_number == i))
+            if (it->column_name == column.name)
                 break;
 
         if (it != description.end())
@@ -290,11 +290,10 @@ void AggregatingSortedAlgorithm::AggregatingMergedData::initAggregateDescription
 
 
 AggregatingSortedAlgorithm::AggregatingSortedAlgorithm(
-    const Block & header, size_t num_inputs,
+    const Block & header_, size_t num_inputs, SortDescription description_, size_t max_block_size)
-    SortDescription description_, size_t max_block_size)
+    : IMergingAlgorithmWithDelayedChunk(header_, num_inputs, description_)
-    : IMergingAlgorithmWithDelayedChunk(num_inputs, description_)
+    , columns_definition(defineColumns(header_, description_))
-    , columns_definition(defineColumns(header, description_))
+    , merged_data(getMergedColumns(header_, columns_definition), max_block_size, columns_definition)
-    , merged_data(getMergedColumns(header, columns_definition), max_block_size, columns_definition)
 {
 }
 

src/Processors/Merges/Algorithms/CollapsingSortedAlgorithm.cpp

@@ -21,7 +21,7 @@ namespace ErrorCodes
 }
 
 CollapsingSortedAlgorithm::CollapsingSortedAlgorithm(
-    const Block & header,
+    const Block & header_,
     size_t num_inputs,
     SortDescription description_,
     const String & sign_column,
@@ -30,9 +30,9 @@ CollapsingSortedAlgorithm::CollapsingSortedAlgorithm(
     Poco::Logger * log_,
     WriteBuffer * out_row_sources_buf_,
     bool use_average_block_sizes)
-    : IMergingAlgorithmWithSharedChunks(num_inputs, std::move(description_), out_row_sources_buf_, max_row_refs)
+    : IMergingAlgorithmWithSharedChunks(header_, num_inputs, std::move(description_), out_row_sources_buf_, max_row_refs)
-    , merged_data(header.cloneEmptyColumns(), use_average_block_sizes, max_block_size)
+    , merged_data(header_.cloneEmptyColumns(), use_average_block_sizes, max_block_size)
-    , sign_column_number(header.getPositionByName(sign_column))
+    , sign_column_number(header_.getPositionByName(sign_column))
     , only_positive_sign(only_positive_sign_)
     , log(log_)
 {

src/Processors/Merges/Algorithms/FinishAggregatingInOrderAlgorithm.cpp

@@ -14,11 +14,8 @@ namespace ErrorCodes
     extern const int LOGICAL_ERROR;
 }
 
-FinishAggregatingInOrderAlgorithm::State::State(
+FinishAggregatingInOrderAlgorithm::State::State(const Chunk & chunk, const SortDescriptionWithPositions & desc, Int64 total_bytes_)
-    const Chunk & chunk, const SortDescription & desc, Int64 total_bytes_)
+    : all_columns(chunk.getColumns()), num_rows(chunk.getNumRows()), total_bytes(total_bytes_)
-    : all_columns(chunk.getColumns())
-    , num_rows(chunk.getNumRows())
-    , total_bytes(total_bytes_)
 {
     if (!chunk)
         return;
@@ -32,25 +29,13 @@ FinishAggregatingInOrderAlgorithm::FinishAggregatingInOrderAlgorithm(
     const Block & header_,
     size_t num_inputs_,
     AggregatingTransformParamsPtr params_,
-    SortDescription description_,
+    const SortDescription & description_,
     size_t max_block_size_,
     size_t max_block_bytes_)
-    : header(header_)
+    : header(header_), num_inputs(num_inputs_), params(params_), max_block_size(max_block_size_), max_block_bytes(max_block_bytes_)
-    , num_inputs(num_inputs_)
-    , params(params_)
-    , description(std::move(description_))
-    , max_block_size(max_block_size_)
-    , max_block_bytes(max_block_bytes_)
 {
-    /// Replace column names in description to positions.
+    for (const auto & column_description : description_)
-    for (auto & column_description : description)
+        description.emplace_back(column_description, header_.getPositionByName(column_description.column_name));
-    {
-        if (!column_description.column_name.empty())
-        {
-            column_description.column_number = header_.getPositionByName(column_description.column_name);
-            column_description.column_name.clear();
-        }
-    }
 }
 
 void FinishAggregatingInOrderAlgorithm::initialize(Inputs inputs)

src/Processors/Merges/Algorithms/FinishAggregatingInOrderAlgorithm.h

@@ -41,7 +41,7 @@ public:
         const Block & header_,
         size_t num_inputs_,
         AggregatingTransformParamsPtr params_,
-        SortDescription description_,
+        const SortDescription & description_,
         size_t max_block_size_,
         size_t max_block_bytes_);
 
@@ -69,7 +69,7 @@ private:
         /// Number of bytes in all columns + number of bytes in arena, related to current chunk.
         size_t total_bytes = 0;
 
-        State(const Chunk & chunk, const SortDescription & description, Int64 total_bytes_);
+        State(const Chunk & chunk, const SortDescriptionWithPositions & description, Int64 total_bytes_);
         State() = default;
 
         bool isValid() const { return current_row < num_rows; }
@@ -78,7 +78,7 @@ private:
     Block header;
     size_t num_inputs;
     AggregatingTransformParamsPtr params;
-    SortDescription description;
+    SortDescriptionWithPositions description;
     size_t max_block_size;
     size_t max_block_bytes;
 

src/Processors/Merges/Algorithms/GraphiteRollupSortedAlgorithm.cpp

@@ -30,12 +30,16 @@ static GraphiteRollupSortedAlgorithm::ColumnsDefinition defineColumns(
 }
 
 GraphiteRollupSortedAlgorithm::GraphiteRollupSortedAlgorithm(
-    const Block & header, size_t num_inputs,
+    const Block & header_,
-    SortDescription description_, size_t max_block_size,
+    size_t num_inputs,
-    Graphite::Params params_, time_t time_of_merge_)
+    SortDescription description_,
-    : IMergingAlgorithmWithSharedChunks(num_inputs, std::move(description_), nullptr, max_row_refs)
+    size_t max_block_size,
-    , merged_data(header.cloneEmptyColumns(), false, max_block_size)
+    Graphite::Params params_,
-    , params(std::move(params_)), time_of_merge(time_of_merge_)
+    time_t time_of_merge_)
+    : IMergingAlgorithmWithSharedChunks(header_, num_inputs, std::move(description_), nullptr, max_row_refs)
+    , merged_data(header_.cloneEmptyColumns(), false, max_block_size)
+    , params(std::move(params_))
+    , time_of_merge(time_of_merge_)
 {
     size_t max_size_of_aggregate_state = 0;
     size_t max_alignment_of_aggregate_state = 1;
@@ -50,7 +54,7 @@ GraphiteRollupSortedAlgorithm::GraphiteRollupSortedAlgorithm(
     }
 
     merged_data.allocMemForAggregates(max_size_of_aggregate_state, max_alignment_of_aggregate_state);
-    columns_definition = defineColumns(header, params);
+    columns_definition = defineColumns(header_, params);
 }
 
 UInt32 GraphiteRollupSortedAlgorithm::selectPrecision(const Graphite::Retentions & retentions, time_t time) const

src/Processors/Merges/Algorithms/IMergingAlgorithmWithDelayedChunk.cpp

@@ -4,12 +4,8 @@
 namespace DB
 {
 
-IMergingAlgorithmWithDelayedChunk::IMergingAlgorithmWithDelayedChunk(
+IMergingAlgorithmWithDelayedChunk::IMergingAlgorithmWithDelayedChunk(Block header_, size_t num_inputs, SortDescription description_)
-    size_t num_inputs,
+    : description(std::move(description_)), header(std::move(header_)), current_inputs(num_inputs), cursors(num_inputs)
-    SortDescription description_)
-    : description(std::move(description_))
-    , current_inputs(num_inputs)
-    , cursors(num_inputs)
 {
 }
 
@@ -22,7 +18,8 @@ void IMergingAlgorithmWithDelayedChunk::initializeQueue(Inputs inputs)
         if (!current_inputs[source_num].chunk)
             continue;
 
-        cursors[source_num] = SortCursorImpl(current_inputs[source_num].chunk.getColumns(), description, source_num, current_inputs[source_num].permutation);
+        cursors[source_num] = SortCursorImpl(
+            header, current_inputs[source_num].chunk.getColumns(), description, source_num, current_inputs[source_num].permutation);
     }
 
     queue = SortingHeap<SortCursor>(cursors);
@@ -37,7 +34,7 @@ void IMergingAlgorithmWithDelayedChunk::updateCursor(Input & input, size_t sourc
     last_chunk_sort_columns = std::move(cursors[source_num].sort_columns);
 
     current_input.swap(input);
-    cursors[source_num].reset(current_input.chunk.getColumns(), {}, current_input.permutation);
+    cursors[source_num].reset(current_input.chunk.getColumns(), header, current_input.permutation);
 
     queue.push(cursors[source_num]);
 }

src/Processors/Merges/Algorithms/IMergingAlgorithmWithDelayedChunk.h

@@ -10,9 +10,7 @@ namespace DB
 class IMergingAlgorithmWithDelayedChunk : public IMergingAlgorithm
 {
 public:
-    IMergingAlgorithmWithDelayedChunk(
+    IMergingAlgorithmWithDelayedChunk(Block header_, size_t num_inputs, SortDescription description_);
-        size_t num_inputs,
-        SortDescription description_);
 
 protected:
     SortingHeap<SortCursor> queue;
@@ -28,6 +26,8 @@ protected:
     bool skipLastRowFor(size_t input_number) const { return current_inputs[input_number].skip_last_row; }
 
 private:
+    Block header;
+
     /// Inputs currently being merged.
     Inputs current_inputs;
     SortCursorImpls cursors;

src/Processors/Merges/Algorithms/IMergingAlgorithmWithSharedChunks.cpp

@@ -4,11 +4,9 @@ namespace DB
 {
 
 IMergingAlgorithmWithSharedChunks::IMergingAlgorithmWithSharedChunks(
-    size_t num_inputs,
+    Block header_, size_t num_inputs, SortDescription description_, WriteBuffer * out_row_sources_buf_, size_t max_row_refs)
-    SortDescription description_,
+    : header(std::move(header_))
-    WriteBuffer * out_row_sources_buf_,
+    , description(std::move(description_))
-    size_t max_row_refs)
-    : description(std::move(description_))
     , chunk_allocator(num_inputs + max_row_refs)
     , cursors(num_inputs)
     , sources(num_inputs)
@@ -39,7 +37,7 @@ void IMergingAlgorithmWithSharedChunks::initialize(Inputs inputs)
 
         source.skip_last_row = inputs[source_num].skip_last_row;
         source.chunk = chunk_allocator.alloc(inputs[source_num].chunk);
-        cursors[source_num] = SortCursorImpl(source.chunk->getColumns(), description, source_num, inputs[source_num].permutation);
+        cursors[source_num] = SortCursorImpl(header, source.chunk->getColumns(), description, source_num, inputs[source_num].permutation);
 
         source.chunk->all_columns = cursors[source_num].all_columns;
         source.chunk->sort_columns = cursors[source_num].sort_columns;
@@ -55,7 +53,7 @@ void IMergingAlgorithmWithSharedChunks::consume(Input & input, size_t source_num
     auto & source = sources[source_num];
     source.skip_last_row = input.skip_last_row;
     source.chunk = chunk_allocator.alloc(input.chunk);
-    cursors[source_num].reset(source.chunk->getColumns(), {}, input.permutation);
+    cursors[source_num].reset(source.chunk->getColumns(), header, input.permutation);
 
     source.chunk->all_columns = cursors[source_num].all_columns;
     source.chunk->sort_columns = cursors[source_num].sort_columns;

src/Processors/Merges/Algorithms/IMergingAlgorithmWithSharedChunks.h

@@ -10,15 +10,13 @@ class IMergingAlgorithmWithSharedChunks : public IMergingAlgorithm
 {
 public:
     IMergingAlgorithmWithSharedChunks(
-        size_t num_inputs,
+        Block header_, size_t num_inputs, SortDescription description_, WriteBuffer * out_row_sources_buf_, size_t max_row_refs);
-        SortDescription description_,
-        WriteBuffer * out_row_sources_buf_,
-        size_t max_row_refs);
 
     void initialize(Inputs inputs) override;
     void consume(Input & input, size_t source_num) override;
 
 private:
+    Block header;
     SortDescription description;
 
     /// Allocator must be destroyed after source_chunks.

src/Processors/Merges/Algorithms/MergingSortedAlgorithm.cpp

@@ -11,30 +11,22 @@ namespace ErrorCodes
 }
 
 MergingSortedAlgorithm::MergingSortedAlgorithm(
-    const Block & header,
+    Block header_,
     size_t num_inputs,
     SortDescription description_,
     size_t max_block_size,
     UInt64 limit_,
     WriteBuffer * out_row_sources_buf_,
     bool use_average_block_sizes)
-    : merged_data(header.cloneEmptyColumns(), use_average_block_sizes, max_block_size)
+    : header(std::move(header_))
+    , merged_data(header.cloneEmptyColumns(), use_average_block_sizes, max_block_size)
     , description(std::move(description_))
     , limit(limit_)
+    , has_collation(std::any_of(description.begin(), description.end(), [](const auto & descr) { return descr.collator != nullptr; }))
     , out_row_sources_buf(out_row_sources_buf_)
     , current_inputs(num_inputs)
     , cursors(num_inputs)
 {
-    /// Replace column names in description to positions.
-    for (auto & column_description : description)
-    {
-        has_collation |= column_description.collator != nullptr;
-        if (!column_description.column_name.empty())
-        {
-            column_description.column_number = header.getPositionByName(column_description.column_name);
-            column_description.column_name.clear();
-        }
-    }
 }
 
 void MergingSortedAlgorithm::addInput()
@@ -65,7 +57,7 @@ void MergingSortedAlgorithm::initialize(Inputs inputs)
             continue;
 
         prepareChunk(chunk);
-        cursors[source_num] = SortCursorImpl(chunk.getColumns(), description, source_num);
+        cursors[source_num] = SortCursorImpl(header, chunk.getColumns(), description, source_num);
     }
 
     if (has_collation)
@@ -78,7 +70,7 @@ void MergingSortedAlgorithm::consume(Input & input, size_t source_num)
 {
     prepareChunk(input.chunk);
     current_inputs[source_num].swap(input);
-    cursors[source_num].reset(current_inputs[source_num].chunk.getColumns(), {});
+    cursors[source_num].reset(current_inputs[source_num].chunk.getColumns(), header);
 
     if (has_collation)
         queue_with_collation.push(cursors[source_num]);

src/Processors/Merges/Algorithms/MergingSortedAlgorithm.h

@@ -14,7 +14,7 @@ class MergingSortedAlgorithm final : public IMergingAlgorithm
 {
 public:
     MergingSortedAlgorithm(
-        const Block & header,
+        Block header_,
         size_t num_inputs,
         SortDescription description_,
         size_t max_block_size,
@@ -31,6 +31,8 @@ public:
     const MergedData & getMergedData() const { return merged_data; }
 
 private:
+    Block header;
+
     MergedData merged_data;
 
     /// Settings

src/Processors/Merges/Algorithms/ReplacingSortedAlgorithm.cpp

@@ -5,16 +5,18 @@ namespace DB
 {
 
 ReplacingSortedAlgorithm::ReplacingSortedAlgorithm(
-        const Block & header, size_t num_inputs,
+    const Block & header_,
-        SortDescription description_, const String & version_column,
+    size_t num_inputs,
-        size_t max_block_size,
+    SortDescription description_,
-        WriteBuffer * out_row_sources_buf_,
+    const String & version_column,
-        bool use_average_block_sizes)
+    size_t max_block_size,
-        : IMergingAlgorithmWithSharedChunks(num_inputs, std::move(description_), out_row_sources_buf_, max_row_refs)
+    WriteBuffer * out_row_sources_buf_,
-        , merged_data(header.cloneEmptyColumns(), use_average_block_sizes, max_block_size)
+    bool use_average_block_sizes)
+    : IMergingAlgorithmWithSharedChunks(header_, num_inputs, std::move(description_), out_row_sources_buf_, max_row_refs)
+    , merged_data(header_.cloneEmptyColumns(), use_average_block_sizes, max_block_size)
 {
     if (!version_column.empty())
-        version_column_number = header.getPositionByName(version_column);
+        version_column_number = header_.getPositionByName(version_column);
 }
 
 void ReplacingSortedAlgorithm::insertRow()

src/Processors/Merges/Algorithms/SummingSortedAlgorithm.cpp

@@ -101,10 +101,10 @@ struct SummingSortedAlgorithm::AggregateDescription
 };
 
 
-static bool isInPrimaryKey(const SortDescription & description, const std::string & name, const size_t number)
+static bool isInPrimaryKey(const SortDescription & description, const std::string & name)
 {
     for (const auto & desc : description)
-        if (desc.column_name == name || (desc.column_name.empty() && desc.column_number == number))
+        if (desc.column_name == name)
             return true;
 
     return false;
@@ -251,7 +251,7 @@ static SummingSortedAlgorithm::ColumnsDefinition defineColumns(
             }
 
             /// Are they inside the primary key or partition key?
-            if (isInPrimaryKey(description, column.name, i) ||  isInPartitionKey(column.name, partition_key_columns))
+            if (isInPrimaryKey(description, column.name) || isInPartitionKey(column.name, partition_key_columns))
             {
                 def.column_numbers_not_to_aggregate.push_back(i);
                 continue;
@@ -307,7 +307,7 @@ static SummingSortedAlgorithm::ColumnsDefinition defineColumns(
         /// no elements of map could be in primary key
         auto column_num_it = map.second.begin();
         for (; column_num_it != map.second.end(); ++column_num_it)
-            if (isInPrimaryKey(description, header.safeGetByPosition(*column_num_it).name, *column_num_it))
+            if (isInPrimaryKey(description, header.safeGetByPosition(*column_num_it).name))
                 break;
         if (column_num_it != map.second.end())
         {
@@ -687,14 +687,15 @@ Chunk SummingSortedAlgorithm::SummingMergedData::pull()
 
 
 SummingSortedAlgorithm::SummingSortedAlgorithm(
-    const Block & header, size_t num_inputs,
+    const Block & header_,
+    size_t num_inputs,
     SortDescription description_,
     const Names & column_names_to_sum,
     const Names & partition_key_columns,
     size_t max_block_size)
-    : IMergingAlgorithmWithDelayedChunk(num_inputs, std::move(description_))
+    : IMergingAlgorithmWithDelayedChunk(header_, num_inputs, std::move(description_))
-    , columns_definition(defineColumns(header, description, column_names_to_sum, partition_key_columns))
+    , columns_definition(defineColumns(header_, description, column_names_to_sum, partition_key_columns))
-    , merged_data(getMergedDataColumns(header, columns_definition), max_block_size, columns_definition)
+    , merged_data(getMergedDataColumns(header_, columns_definition), max_block_size, columns_definition)
 {
 }
 

src/Processors/Merges/Algorithms/VersionedCollapsingAlgorithm.cpp

@@ -8,19 +8,20 @@ namespace DB
 static const size_t MAX_ROWS_IN_MULTIVERSION_QUEUE = 8192;
 
 VersionedCollapsingAlgorithm::VersionedCollapsingAlgorithm(
-    const Block & header, size_t num_inputs,
+    const Block & header_,
-    SortDescription description_, const String & sign_column_,
+    size_t num_inputs,
+    SortDescription description_,
+    const String & sign_column_,
     size_t max_block_size,
     WriteBuffer * out_row_sources_buf_,
     bool use_average_block_sizes)
-    : IMergingAlgorithmWithSharedChunks(
+    : IMergingAlgorithmWithSharedChunks(header_, num_inputs, std::move(description_), out_row_sources_buf_, MAX_ROWS_IN_MULTIVERSION_QUEUE)
-            num_inputs, std::move(description_), out_row_sources_buf_, MAX_ROWS_IN_MULTIVERSION_QUEUE)
+    , merged_data(header_.cloneEmptyColumns(), use_average_block_sizes, max_block_size)
-    , merged_data(header.cloneEmptyColumns(), use_average_block_sizes, max_block_size)
     /// -1 for +1 in FixedSizeDequeWithGaps's internal buffer. 3 is a reasonable minimum size to collapse anything.
     , max_rows_in_queue(std::min(std::max<size_t>(3, max_block_size), MAX_ROWS_IN_MULTIVERSION_QUEUE) - 1)
     , current_keys(max_rows_in_queue)
 {
-    sign_column_number = header.getPositionByName(sign_column_);
+    sign_column_number = header_.getPositionByName(sign_column_);
 }
 
 inline ALWAYS_INLINE static void writeRowSourcePart(WriteBuffer & buffer, RowSourcePart row_source)

src/Processors/QueryPlan/FillingStep.cpp

@@ -48,13 +48,13 @@ void FillingStep::transformPipeline(QueryPipelineBuilder & pipeline, const Build
 void FillingStep::describeActions(FormatSettings & settings) const
 {
     settings.out << String(settings.offset, ' ');
-    dumpSortDescription(sort_description, input_streams.front().header, settings.out);
+    dumpSortDescription(sort_description, settings.out);
     settings.out << '\n';
 }
 
 void FillingStep::describeActions(JSONBuilder::JSONMap & map) const
 {
-    map.add("Sort Description", explainSortDescription(sort_description, input_streams.front().header));
+    map.add("Sort Description", explainSortDescription(sort_description));
 }
 
 }

src/Processors/QueryPlan/ITransformingStep.cpp

@@ -70,4 +70,9 @@ void ITransformingStep::describePipeline(FormatSettings & settings) const
     IQueryPlanStep::describePipeline(processors, settings);
 }
 
+void ITransformingStep::appendExtraProcessors(const Processors & extra_processors)
+{
+    processors.insert(processors.end(), extra_processors.begin(), extra_processors.end());
+}
+
 }

src/Processors/QueryPlan/ITransformingStep.h

@@ -57,6 +57,9 @@ public:
 
     void describePipeline(FormatSettings & settings) const override;
 
+    /// Append extra processors for this step.
+    void appendExtraProcessors(const Processors & extra_processors);
+
 protected:
     /// Clear distinct_columns if res_header doesn't contain all of them.
     static void updateDistinctColumns(const Block & res_header, NameSet & distinct_columns);

src/Processors/QueryPlan/Optimizations/Optimizations.h

@@ -44,16 +44,20 @@ size_t tryMergeExpressions(QueryPlan::Node * parent_node, QueryPlan::Nodes &);
 /// May split FilterStep and push down only part of it.
 size_t tryPushDownFilter(QueryPlan::Node * parent_node, QueryPlan::Nodes & nodes);
 
+/// Move ExpressionStep after SortingStep if possible.
+/// May split ExpressionStep and lift up only a part of it.
+size_t tryExecuteFunctionsAfterSorting(QueryPlan::Node * parent_node, QueryPlan::Nodes & nodes);
+
 inline const auto & getOptimizations()
 {
-    static const std::array<Optimization, 5> optimizations =
+    static const std::array<Optimization, 6> optimizations = {{
-    {{
         {tryLiftUpArrayJoin, "liftUpArrayJoin", &QueryPlanOptimizationSettings::optimize_plan},
         {tryPushDownLimit, "pushDownLimit", &QueryPlanOptimizationSettings::optimize_plan},
         {trySplitFilter, "splitFilter", &QueryPlanOptimizationSettings::optimize_plan},
         {tryMergeExpressions, "mergeExpressions", &QueryPlanOptimizationSettings::optimize_plan},
         {tryPushDownFilter, "pushDownFilter", &QueryPlanOptimizationSettings::filter_push_down},
-     }};
+        {tryExecuteFunctionsAfterSorting, "liftUpFunctions", &QueryPlanOptimizationSettings::optimize_plan},
+    }};
 
     return optimizations;
 }

src/Processors/QueryPlan/Optimizations/liftUpFunctions.cpp

@@ -0,0 +1,77 @@
+#include <Interpreters/ActionsDAG.h>
+#include <Processors/QueryPlan/ExpressionStep.h>
+#include <Processors/QueryPlan/Optimizations/Optimizations.h>
+#include <Processors/QueryPlan/SortingStep.h>
+#include <Common/Exception.h>
+
+namespace DB
+{
+namespace ErrorCodes
+{
+    extern const int LOGICAL_ERROR;
+}
+}
+
+namespace
+{
+
+const DB::DataStream & getChildOutputStream(DB::QueryPlan::Node & node)
+{
+    if (node.children.size() != 1)
+        throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Node \"{}\" is expected to have only one child.", node.step->getName());
+    return node.children.front()->step->getOutputStream();
+}
+
+}
+
+namespace DB::QueryPlanOptimizations
+{
+
+size_t tryExecuteFunctionsAfterSorting(QueryPlan::Node * parent_node, QueryPlan::Nodes & nodes)
+{
+    if (parent_node->children.size() != 1)
+        return 0;
+
+    QueryPlan::Node * child_node = parent_node->children.front();
+
+    auto & parent_step = parent_node->step;
+    auto & child_step = child_node->step;
+    auto * sorting_step = typeid_cast<SortingStep *>(parent_step.get());
+    auto * expression_step = typeid_cast<ExpressionStep *>(child_step.get());
+
+    if (!sorting_step || !expression_step)
+        return 0;
+
+    NameSet sort_columns;
+    for (const auto & col : sorting_step->getSortDescription())
+        sort_columns.insert(col.column_name);
+    auto [needed_for_sorting, unneeded_for_sorting] = expression_step->getExpression()->splitActionsBySortingDescription(sort_columns);
+
+    // No calculations can be postponed.
+    if (unneeded_for_sorting->trivial())
+        return 0;
+
+    // Sorting (parent_node) -> Expression (child_node)
+    auto & node_with_needed = nodes.emplace_back();
+    std::swap(node_with_needed.children, child_node->children);
+    child_node->children = {&node_with_needed};
+
+    node_with_needed.step = std::make_unique<ExpressionStep>(getChildOutputStream(node_with_needed), std::move(needed_for_sorting));
+    node_with_needed.step->setStepDescription(child_step->getStepDescription());
+    // Sorting (parent_node) -> so far the origin Expression (child_node) -> NeededCalculations (node_with_needed)
+
+    std::swap(parent_step, child_step);
+    // so far the origin Expression (parent_node) -> Sorting (child_node) -> NeededCalculations (node_with_needed)
+
+    sorting_step->updateInputStream(getChildOutputStream(*child_node));
+    auto input_header = sorting_step->getInputStreams().at(0).header;
+    sorting_step->updateOutputStream(std::move(input_header));
+
+    auto description = parent_step->getStepDescription();
+    parent_step = std::make_unique<DB::ExpressionStep>(child_step->getOutputStream(), std::move(unneeded_for_sorting));
+    parent_step->setStepDescription(description + " [lifted up part]");
+    // UneededCalculations (parent_node) -> Sorting (child_node) -> NeededCalculations (node_with_needed)
+
+    return 3;
+}
+}

src/Processors/QueryPlan/ReadFromMergeTree.cpp

@@ -612,14 +612,8 @@ static void addMergingFinal(
 
     ColumnNumbers key_columns;
     key_columns.reserve(sort_description.size());
-
     for (const auto & desc : sort_description)
-    {
+        key_columns.push_back(header.getPositionByName(desc.column_name));
-        if (!desc.column_name.empty())
-            key_columns.push_back(header.getPositionByName(desc.column_name));
-        else
-            key_columns.emplace_back(desc.column_number);
-    }
 
     pipe.addSimpleTransform([&](const Block & stream_header)
     {
@@ -774,9 +768,8 @@ Pipe ReadFromMergeTree::spreadMarkRangesAmongStreamsFinal(
 
         Names partition_key_columns = metadata_for_reading->getPartitionKey().column_names;
 
-        const auto & header = pipe.getHeader();
         for (size_t i = 0; i < sort_columns_size; ++i)
-            sort_description.emplace_back(header.getPositionByName(sort_columns[i]), 1, 1);
+            sort_description.emplace_back(sort_columns[i], 1, 1);
 
         addMergingFinal(
             pipe,

src/Processors/QueryPlan/SortingStep.cpp

@@ -1,11 +1,12 @@
+#include <stdexcept>
+#include <IO/Operators.h>
+#include <Processors/Merges/MergingSortedTransform.h>
 #include <Processors/QueryPlan/SortingStep.h>
-#include <QueryPipeline/QueryPipelineBuilder.h>
+#include <Processors/Transforms/FinishSortingTransform.h>
+#include <Processors/Transforms/LimitsCheckingTransform.h>
 #include <Processors/Transforms/MergeSortingTransform.h>
 #include <Processors/Transforms/PartialSortingTransform.h>
-#include <Processors/Transforms/FinishSortingTransform.h>
+#include <QueryPipeline/QueryPipelineBuilder.h>
-#include <Processors/Merges/MergingSortedTransform.h>
-#include <Processors/Transforms/LimitsCheckingTransform.h>
-#include <IO/Operators.h>
 #include <Common/JSONBuilder.h>
 
 namespace DB
@@ -88,6 +89,18 @@ SortingStep::SortingStep(
     output_stream->sort_mode = DataStream::SortMode::Stream;
 }
 
+void SortingStep::updateInputStream(DataStream input_stream)
+{
+    input_streams.clear();
+    input_streams.emplace_back(std::move(input_stream));
+}
+
+void SortingStep::updateOutputStream(Block result_header)
+{
+    output_stream = createOutputStream(input_streams.at(0), std::move(result_header), getDataStreamTraits());
+    updateDistinctColumns(output_stream->header, output_stream->distinct_columns);
+}
+
 void SortingStep::updateLimit(size_t limit_)
 {
     if (limit_ && (limit == 0 || limit_ < limit))
@@ -206,17 +219,17 @@ void SortingStep::describeActions(FormatSettings & settings) const
     if (!prefix_description.empty())
     {
         settings.out << prefix << "Prefix sort description: ";
-        dumpSortDescription(prefix_description, input_streams.front().header, settings.out);
+        dumpSortDescription(prefix_description, settings.out);
         settings.out << '\n';
 
         settings.out << prefix << "Result sort description: ";
-        dumpSortDescription(result_description, input_streams.front().header, settings.out);
+        dumpSortDescription(result_description, settings.out);
         settings.out << '\n';
     }
     else
     {
         settings.out << prefix << "Sort description: ";
-        dumpSortDescription(result_description, input_streams.front().header, settings.out);
+        dumpSortDescription(result_description, settings.out);
         settings.out << '\n';
     }
 
@@ -228,11 +241,11 @@ void SortingStep::describeActions(JSONBuilder::JSONMap & map) const
 {
     if (!prefix_description.empty())
     {
-        map.add("Prefix Sort Description", explainSortDescription(prefix_description, input_streams.front().header));
+        map.add("Prefix Sort Description", explainSortDescription(prefix_description));
-        map.add("Result Sort Description", explainSortDescription(result_description, input_streams.front().header));
+        map.add("Result Sort Description", explainSortDescription(result_description));
     }
     else
-        map.add("Sort Description", explainSortDescription(result_description, input_streams.front().header));
+        map.add("Sort Description", explainSortDescription(result_description));
 
     if (limit)
         map.add("Limit", limit);

src/Processors/QueryPlan/SortingStep.h

@@ -49,6 +49,11 @@ public:
     /// Add limit or change it to lower value.
     void updateLimit(size_t limit_);
 
+    void updateInputStream(DataStream input_stream);
+    void updateOutputStream(Block result_header);
+
+    SortDescription getSortDescription() const { return result_description; }
+
 private:
 
     enum class Type

src/Processors/QueryPlan/WindowStep.cpp

@@ -129,7 +129,7 @@ void WindowStep::describeActions(JSONBuilder::JSONMap & map) const
     }
 
     if (!window_description.order_by.empty())
-        map.add("Sort Description", explainSortDescription(window_description.order_by, {}));
+        map.add("Sort Description", explainSortDescription(window_description.order_by));
 
     auto functions_array = std::make_unique<JSONBuilder::JSONArray>();
     for (const auto & func : window_functions)

src/Processors/Transforms/AggregatingInOrderTransform.cpp

@@ -26,7 +26,6 @@ AggregatingInOrderTransform::AggregatingInOrderTransform(
     , max_block_size(max_block_size_)
     , max_block_bytes(max_block_bytes_)
     , params(std::move(params_))
-    , group_by_description(group_by_description_)
     , aggregate_columns(params->params.aggregates_size)
     , many_data(std::move(many_data_))
     , variants(*many_data->variants[current_variant])
@@ -34,15 +33,8 @@ AggregatingInOrderTransform::AggregatingInOrderTransform(
     /// We won't finalize states in order to merge same states (generated due to multi-thread execution) in AggregatingSortedTransform
     res_header = params->getCustomHeader(false);
 
-    /// Replace column names to column position in description_sorted.
+    for (const auto & column_description : group_by_description_)
-    for (auto & column_description : group_by_description)
+        group_by_description.emplace_back(column_description, res_header.getPositionByName(column_description.column_name));
-    {
-        if (!column_description.column_name.empty())
-        {
-            column_description.column_number = res_header.getPositionByName(column_description.column_name);
-            column_description.column_name.clear();
-        }
-    }
 }
 
 AggregatingInOrderTransform::~AggregatingInOrderTransform() = default;

src/Processors/Transforms/AggregatingInOrderTransform.h

@@ -51,7 +51,7 @@ private:
     MutableColumns res_aggregate_columns;
 
     AggregatingTransformParamsPtr params;
-    SortDescription group_by_description;
+    SortDescriptionWithPositions group_by_description;
 
     Aggregator::AggregateColumns aggregate_columns;
 

src/Processors/Transforms/CheckSortedTransform.cpp

@@ -12,33 +12,13 @@ namespace ErrorCodes
 extern const int LOGICAL_ERROR;
 }
 
-CheckSortedTransform::CheckSortedTransform(
+CheckSortedTransform::CheckSortedTransform(const Block & header, const SortDescription & sort_description)
-    const Block & header_,
+    : ISimpleTransform(header, header, false)
-    const SortDescription & sort_description_)
-    : ISimpleTransform(header_, header_, false)
-    , sort_description_map(addPositionsToSortDescriptions(sort_description_))
 {
+    for (const auto & column_description : sort_description)
+        sort_description_map.emplace_back(column_description, header.getPositionByName(column_description.column_name));
 }
 
-SortDescriptionsWithPositions
-CheckSortedTransform::addPositionsToSortDescriptions(const SortDescription & sort_description)
-{
-    SortDescriptionsWithPositions result;
-    result.reserve(sort_description.size());
-    const auto & header = getInputPort().getHeader();
-
-    for (SortColumnDescription description_copy : sort_description)
-    {
-        if (!description_copy.column_name.empty())
-            description_copy.column_number = header.getPositionByName(description_copy.column_name);
-
-        result.push_back(description_copy);
-    }
-
-    return result;
-}
-
-
 void CheckSortedTransform::transform(Chunk & chunk)
 {
     size_t num_rows = chunk.getNumRows();
@@ -54,7 +34,7 @@ void CheckSortedTransform::transform(Chunk & chunk)
             const IColumn * left_col = left[column_number].get();
             const IColumn * right_col = right[column_number].get();
 
-            int res = elem.direction * left_col->compareAt(left_index, right_index, *right_col, elem.nulls_direction);
+            int res = elem.base.direction * left_col->compareAt(left_index, right_index, *right_col, elem.base.nulls_direction);
             if (res < 0)
             {
                 return;

src/Processors/Transforms/CheckSortedTransform.h

@@ -5,16 +5,12 @@
 
 namespace DB
 {
-using SortDescriptionsWithPositions = std::vector<SortColumnDescription>;
-
 /// Streams checks that flow of blocks is sorted in the sort_description order
 /// Othrewise throws exception in readImpl function.
 class CheckSortedTransform : public ISimpleTransform
 {
 public:
-    CheckSortedTransform(
+    CheckSortedTransform(const Block & header, const SortDescription & sort_description);
-        const Block & header_,
-        const SortDescription & sort_description_);
 
     String getName() const override { return "CheckSortedTransform"; }
 
@@ -23,10 +19,7 @@ protected:
     void transform(Chunk & chunk) override;
 
 private:
-    SortDescriptionsWithPositions sort_description_map;
+    SortDescriptionWithPositions sort_description_map;
     Columns last_row;
-
-    /// Just checks, that all sort_descriptions has column_number
-    SortDescriptionsWithPositions addPositionsToSortDescriptions(const SortDescription & sort_description);
 };
 }

src/Processors/Transforms/DistinctSortedTransform.cpp

@@ -9,8 +9,9 @@ namespace ErrorCodes
 }
 
 DistinctSortedTransform::DistinctSortedTransform(
-    const Block & header, SortDescription sort_description, const SizeLimits & set_size_limits_, UInt64 limit_hint_, const Names & columns)
+    Block header_, SortDescription sort_description, const SizeLimits & set_size_limits_, UInt64 limit_hint_, const Names & columns)
-    : ISimpleTransform(header, header, true)
+    : ISimpleTransform(header_, header_, true)
+    , header(std::move(header_))
     , description(std::move(sort_description))
     , columns_names(columns)
     , limit_hint(limit_hint_)
@@ -24,7 +25,7 @@ void DistinctSortedTransform::transform(Chunk & chunk)
         if (column_ptrs.empty())
             return;
 
-        ColumnRawPtrs clearing_hint_columns(getClearingColumns(chunk, column_ptrs));
+        ColumnRawPtrs clearing_hint_columns(getClearingColumns(column_ptrs));
 
         if (data.type == ClearableSetVariants::Type::EMPTY)
             data.init(ClearableSetVariants::chooseMethod(column_ptrs, key_sizes));
@@ -139,13 +140,13 @@ ColumnRawPtrs DistinctSortedTransform::getKeyColumns(const Chunk & chunk) const
     return column_ptrs;
 }
 
-ColumnRawPtrs DistinctSortedTransform::getClearingColumns(const Chunk & chunk, const ColumnRawPtrs & key_columns) const
+ColumnRawPtrs DistinctSortedTransform::getClearingColumns(const ColumnRawPtrs & key_columns) const
 {
     ColumnRawPtrs clearing_hint_columns;
     clearing_hint_columns.reserve(description.size());
     for (const auto & sort_column_description : description)
     {
-        const auto * sort_column_ptr = chunk.getColumns().at(sort_column_description.column_number).get();
+        const auto * sort_column_ptr = header.getByName(sort_column_description.column_name).column.get();
         const auto it = std::find(key_columns.cbegin(), key_columns.cend(), sort_column_ptr);
         if (it != key_columns.cend()) /// if found in key_columns
             clearing_hint_columns.emplace_back(sort_column_ptr);

src/Processors/Transforms/DistinctSortedTransform.h

@@ -22,7 +22,8 @@ class DistinctSortedTransform : public ISimpleTransform
 {
 public:
     /// Empty columns_ means all columns.
-    DistinctSortedTransform(const Block & header, SortDescription sort_description, const SizeLimits & set_size_limits_, UInt64 limit_hint_, const Names & columns);
+    DistinctSortedTransform(
+        Block header_, SortDescription sort_description, const SizeLimits & set_size_limits_, UInt64 limit_hint_, const Names & columns);
 
     String getName() const override { return "DistinctSortedTransform"; }
 
@@ -33,7 +34,7 @@ private:
     ColumnRawPtrs getKeyColumns(const Chunk & chunk) const;
     /// When clearing_columns changed, we can clean HashSet to memory optimization
     /// clearing_columns is a left-prefix of SortDescription exists in key_columns
-    ColumnRawPtrs getClearingColumns(const Chunk & chunk, const ColumnRawPtrs & key_columns) const;
+    ColumnRawPtrs getClearingColumns(const ColumnRawPtrs & key_columns) const;
     static bool rowsEqual(const ColumnRawPtrs & lhs, size_t n, const ColumnRawPtrs & rhs, size_t m);
 
     /// return true if has new data
@@ -46,6 +47,7 @@ private:
         size_t rows,
         ClearableSetVariants & variants) const;
 
+    Block header;
     SortDescription description;
 
     struct PreviousChunk

src/Processors/Transforms/FinishSortingTransform.cpp

@@ -21,9 +21,11 @@ static bool isPrefix(const SortDescription & pref_descr, const SortDescription &
 }
 
 FinishSortingTransform::FinishSortingTransform(
-    const Block & header, const SortDescription & description_sorted_,
+    const Block & header,
+    const SortDescription & description_sorted_,
     const SortDescription & description_to_sort_,
-    size_t max_merged_block_size_, UInt64 limit_)
+    size_t max_merged_block_size_,
+    UInt64 limit_)
     : SortingTransform(header, description_to_sort_, max_merged_block_size_, limit_)
 {
     /// Check for sanity non-modified descriptions
@@ -34,7 +36,8 @@ FinishSortingTransform::FinishSortingTransform(
     /// The target description is modified in SortingTransform constructor.
     /// To avoid doing the same actions with description_sorted just copy it from prefix of target description.
     size_t prefix_size = description_sorted_.size();
-    description_sorted.assign(description.begin(), description.begin() + prefix_size);
+    for (size_t i = 0; i < prefix_size; ++i)
+        description_with_positions.emplace_back(description[i], header_without_constants.getPositionByName(description[i].column_name));
 }
 
 void FinishSortingTransform::consume(Chunk chunk)
@@ -62,7 +65,7 @@ void FinishSortingTransform::consume(Chunk chunk)
         while (high - low > 1)
         {
             ssize_t mid = (low + high) / 2;
-            if (!less(last_chunk.getColumns(), chunk.getColumns(), last_chunk.getNumRows() - 1, mid, description_sorted))
+            if (!less(last_chunk.getColumns(), chunk.getColumns(), last_chunk.getNumRows() - 1, mid, description_with_positions))
                 low = mid;
             else
                 high = mid;
@@ -100,7 +103,8 @@ void FinishSortingTransform::generate()
 {
     if (!merge_sorter)
     {
-        merge_sorter = std::make_unique<MergeSorter>(std::move(chunks), description, max_merged_block_size, limit);
+        merge_sorter
+            = std::make_unique<MergeSorter>(header_without_constants, std::move(chunks), description, max_merged_block_size, limit);
         generated_prefix = true;
     }
 

src/Processors/Transforms/FinishSortingTransform.h

@@ -11,9 +11,12 @@ class FinishSortingTransform : public SortingTransform
 {
 public:
     /// limit - if not 0, allowed to return just first 'limit' rows in sorted order.
-    FinishSortingTransform(const Block & header, const SortDescription & description_sorted_,
+    FinishSortingTransform(
+        const Block & header,
+        const SortDescription & description_sorted_,
         const SortDescription & description_to_sort_,
-        size_t max_merged_block_size_, UInt64 limit_);
+        size_t max_merged_block_size_,
+        UInt64 limit_);
 
     String getName() const override { return "FinishSortingTransform"; }
 
@@ -22,7 +25,7 @@ protected:
     void generate() override;
 
 private:
-    SortDescription description_sorted;
+    SortDescriptionWithPositions description_with_positions;
 
     Chunk tail_chunk;
 };

src/Processors/Transforms/MergeSortingTransform.cpp

@@ -90,16 +90,21 @@ private:
 MergeSortingTransform::MergeSortingTransform(
     const Block & header,
     const SortDescription & description_,
-    size_t max_merged_block_size_, UInt64 limit_,
+    size_t max_merged_block_size_,
+    UInt64 limit_,
     size_t max_bytes_before_remerge_,
     double remerge_lowered_memory_bytes_ratio_,
-    size_t max_bytes_before_external_sort_, VolumePtr tmp_volume_,
+    size_t max_bytes_before_external_sort_,
+    VolumePtr tmp_volume_,
     size_t min_free_disk_space_)
     : SortingTransform(header, description_, max_merged_block_size_, limit_)
     , max_bytes_before_remerge(max_bytes_before_remerge_)
     , remerge_lowered_memory_bytes_ratio(remerge_lowered_memory_bytes_ratio_)
-    , max_bytes_before_external_sort(max_bytes_before_external_sort_), tmp_volume(tmp_volume_)
+    , max_bytes_before_external_sort(max_bytes_before_external_sort_)
-    , min_free_disk_space(min_free_disk_space_) {}
+    , tmp_volume(tmp_volume_)
+    , min_free_disk_space(min_free_disk_space_)
+{
+}
 
 Processors MergeSortingTransform::expandPipeline()
 {
@@ -180,7 +185,8 @@ void MergeSortingTransform::consume(Chunk chunk)
         temporary_files.emplace_back(createTemporaryFile(tmp_path));
 
         const std::string & path = temporary_files.back()->path();
-        merge_sorter = std::make_unique<MergeSorter>(std::move(chunks), description, max_merged_block_size, limit);
+        merge_sorter
+            = std::make_unique<MergeSorter>(header_without_constants, std::move(chunks), description, max_merged_block_size, limit);
         auto current_processor = std::make_shared<BufferingToFileTransform>(header_without_constants, log, path);
 
         processors.emplace_back(current_processor);
@@ -223,7 +229,8 @@ void MergeSortingTransform::generate()
     if (!generated_prefix)
     {
         if (temporary_files.empty())
-            merge_sorter = std::make_unique<MergeSorter>(std::move(chunks), description, max_merged_block_size, limit);
+            merge_sorter
+                = std::make_unique<MergeSorter>(header_without_constants, std::move(chunks), description, max_merged_block_size, limit);
         else
         {
             ProfileEvents::increment(ProfileEvents::ExternalSortMerge);
@@ -251,7 +258,7 @@ void MergeSortingTransform::remerge()
     LOG_DEBUG(log, "Re-merging intermediate ORDER BY data ({} blocks with {} rows) to save memory consumption", chunks.size(), sum_rows_in_blocks);
 
     /// NOTE Maybe concat all blocks and partial sort will be faster than merge?
-    MergeSorter remerge_sorter(std::move(chunks), description, max_merged_block_size, limit);
+    MergeSorter remerge_sorter(header_without_constants, std::move(chunks), description, max_merged_block_size, limit);
 
     Chunks new_chunks;
     size_t new_sum_rows_in_blocks = 0;

src/Processors/Transforms/MergeSortingTransform.h

@@ -18,13 +18,16 @@ class MergeSortingTransform : public SortingTransform
 {
 public:
     /// limit - if not 0, allowed to return just first 'limit' rows in sorted order.
-    MergeSortingTransform(const Block & header,
+    MergeSortingTransform(
-                          const SortDescription & description_,
+        const Block & header,
-                          size_t max_merged_block_size_, UInt64 limit_,
+        const SortDescription & description_,
-                          size_t max_bytes_before_remerge_,
+        size_t max_merged_block_size_,
-                          double remerge_lowered_memory_bytes_ratio_,
+        UInt64 limit_,
-                          size_t max_bytes_before_external_sort_, VolumePtr tmp_volume_,
+        size_t max_bytes_before_remerge_,
-                          size_t min_free_disk_space_);
+        double remerge_lowered_memory_bytes_ratio_,
+        size_t max_bytes_before_external_sort_,
+        VolumePtr tmp_volume_,
+        size_t min_free_disk_space_);
 
     String getName() const override { return "MergeSortingTransform"; }
 

src/Processors/Transforms/PartialSortingTransform.cpp

@@ -22,9 +22,7 @@ static ColumnRawPtrs extractColumns(const Block & block, const SortDescription &
 
     for (size_t i = 0; i < size; ++i)
     {
-        const IColumn * column = !description[i].column_name.empty()
+        const IColumn * column = block.getByName(description[i].column_name).column.get();
-            ? block.getByName(description[i].column_name).column.get()
-            : block.safeGetByPosition(description[i].column_number).column.get();
         res.emplace_back(column);
     }
 

src/Processors/Transforms/SortingTransform.cpp

@@ -22,7 +22,7 @@ namespace ErrorCodes
     extern const int LOGICAL_ERROR;
 }
 
-MergeSorter::MergeSorter(Chunks chunks_, SortDescription & description_, size_t max_merged_block_size_, UInt64 limit_)
+MergeSorter::MergeSorter(const Block & header, Chunks chunks_, SortDescription & description_, size_t max_merged_block_size_, UInt64 limit_)
     : chunks(std::move(chunks_)), description(description_), max_merged_block_size(max_merged_block_size_), limit(limit_)
 {
     Chunks nonempty_chunks;
@@ -36,7 +36,7 @@ MergeSorter::MergeSorter(Chunks chunks_, SortDescription & description_, size_t
         /// which can be inefficient.
         convertToFullIfSparse(chunk);
 
-        cursors.emplace_back(chunk.getColumns(), description);
+        cursors.emplace_back(header, chunk.getColumns(), description);
         has_collation |= cursors.back().has_collation;
 
         nonempty_chunks.emplace_back(std::move(chunk));
@@ -139,16 +139,6 @@ SortingTransform::SortingTransform(
 {
     const auto & sample = inputs.front().getHeader();
 
-    /// Replace column names to column position in sort_description.
-    for (auto & column_description : description)
-    {
-        if (!column_description.column_name.empty())
-        {
-            column_description.column_number = sample.getPositionByName(column_description.column_name);
-            column_description.column_name.clear();
-        }
-    }
-
     /// Remove constants from header and map old indexes to new.
     size_t num_columns = sample.columns();
     ColumnNumbers map(num_columns, num_columns);
@@ -169,13 +159,10 @@ SortingTransform::SortingTransform(
     description_without_constants.reserve(description.size());
     for (const auto & column_description : description)
     {
-        auto old_pos = column_description.column_number;
+        auto old_pos = header.getPositionByName(column_description.column_name);
         auto new_pos = map[old_pos];
         if (new_pos < num_columns)
-        {
             description_without_constants.push_back(column_description);
-            description_without_constants.back().column_number = new_pos;
-        }
     }
 
     description.swap(description_without_constants);

src/Processors/Transforms/SortingTransform.h

@@ -15,7 +15,7 @@ namespace DB
 class MergeSorter
 {
 public:
-    MergeSorter(Chunks chunks_, SortDescription & description_, size_t max_merged_block_size_, UInt64 limit_);
+    MergeSorter(const Block & header, Chunks chunks_, SortDescription & description_, size_t max_merged_block_size_, UInt64 limit_);
 
     Chunk read();
 
@@ -45,8 +45,10 @@ private:
 class MergeSorterSource : public ISource
 {
 public:
-    MergeSorterSource(Block header, Chunks chunks, SortDescription & description, size_t max_merged_block_size, UInt64 limit)
+    MergeSorterSource(const Block & header, Chunks chunks, SortDescription & description, size_t max_merged_block_size, UInt64 limit)
-        : ISource(std::move(header)), merge_sorter(std::move(chunks), description, max_merged_block_size, limit) {}
+        : ISource(header), merge_sorter(header, std::move(chunks), description, max_merged_block_size, limit)
+    {
+    }
 
     String getName() const override { return "MergeSorterSource"; }
 

src/QueryPipeline/QueryPipelineBuilder.cpp

@@ -1,5 +1,6 @@
 #include <QueryPipeline/QueryPipelineBuilder.h>
 
+#include <Processors/QueryPlan/ExpressionStep.h>
 #include <Processors/ResizeProcessor.h>
 #include <Processors/LimitTransform.h>
 #include <Processors/Transforms/TotalsHavingTransform.h>
@@ -307,7 +308,15 @@ std::unique_ptr<QueryPipelineBuilder> QueryPipelineBuilder::joinPipelines(
     right->pipe.dropExtremes();
 
     left->pipe.collected_processors = collected_processors;
-    right->pipe.collected_processors = collected_processors;
+
+    /// Collect the NEW processors for the right pipeline.
+    QueryPipelineProcessorsCollector collector(*right);
+    /// Remember the last step of the right pipeline.
+    ExpressionStep* step = typeid_cast<ExpressionStep*>(right->pipe.processors.back()->getQueryPlanStep());
+    if (!step)
+    {
+        throw Exception(ErrorCodes::LOGICAL_ERROR, "The top step of the right pipeline should be ExpressionStep");
+    }
 
     /// In case joined subquery has totals, and we don't, add default chunk to totals.
     bool default_totals = false;
@@ -377,6 +386,10 @@ std::unique_ptr<QueryPipelineBuilder> QueryPipelineBuilder::joinPipelines(
         left->pipe.processors.emplace_back(std::move(joining));
     }
 
+    /// Move the collected processors to the last step in the right pipeline.
+    Processors processors = collector.detachProcessors();
+    step->appendExtraProcessors(processors);
+
     left->pipe.processors.insert(left->pipe.processors.end(), right->pipe.processors.begin(), right->pipe.processors.end());
     left->pipe.holder = std::move(right->pipe.holder);
     left->pipe.header = left->pipe.output_ports.front()->getHeader();

src/Storages/AlterCommands.cpp

@@ -696,22 +696,24 @@ namespace
 /// The function works for Arrays and Nullables of the same structure.
 bool isMetadataOnlyConversion(const IDataType * from, const IDataType * to)
 {
-    if (from->equals(*to))
+    auto is_compatible_enum_types_conversion = [](const IDataType * from_type, const IDataType * to_type)
-        return true;
-
-    if (const auto * from_enum8 = typeid_cast<const DataTypeEnum8 *>(from))
     {
-        if (const auto * to_enum8 = typeid_cast<const DataTypeEnum8 *>(to))
+        if (const auto * from_enum8 = typeid_cast<const DataTypeEnum8 *>(from_type))
-            return to_enum8->contains(*from_enum8);
+        {
-    }
+            if (const auto * to_enum8 = typeid_cast<const DataTypeEnum8 *>(to_type))
+                return to_enum8->contains(*from_enum8);
+        }
 
-    if (const auto * from_enum16 = typeid_cast<const DataTypeEnum16 *>(from))
+        if (const auto * from_enum16 = typeid_cast<const DataTypeEnum16 *>(from_type))
-    {
+        {
-        if (const auto * to_enum16 = typeid_cast<const DataTypeEnum16 *>(to))
+            if (const auto * to_enum16 = typeid_cast<const DataTypeEnum16 *>(to_type))
-            return to_enum16->contains(*from_enum16);
+                return to_enum16->contains(*from_enum16);
-    }
+        }
 
-    static const std::unordered_multimap<std::type_index, const std::type_info &> ALLOWED_CONVERSIONS =
+        return false;
+    };
+
+    static const std::unordered_multimap<std::type_index, const std::type_info &> allowed_conversions =
         {
             { typeid(DataTypeEnum8),    typeid(DataTypeInt8)     },
             { typeid(DataTypeEnum16),   typeid(DataTypeInt16)    },
@@ -721,12 +723,19 @@ bool isMetadataOnlyConversion(const IDataType * from, const IDataType * to)
             { typeid(DataTypeUInt16),   typeid(DataTypeDate)     },
         };
 
+    /// Unwrap some nested and check for valid conevrsions
     while (true)
     {
+        /// types are equal, obviously pure metadata alter
         if (from->equals(*to))
             return true;
 
-        auto it_range = ALLOWED_CONVERSIONS.equal_range(typeid(*from));
+        /// We just adding something to enum, nothing changed on disk
+        if (is_compatible_enum_types_conversion(from, to))
+            return true;
+
+        /// Types changed, but representation on disk didn't
+        auto it_range = allowed_conversions.equal_range(typeid(*from));
         for (auto it = it_range.first; it != it_range.second; ++it)
         {
             if (it->second == typeid(*to))
@@ -1046,8 +1055,12 @@ void AlterCommands::validate(const StorageInMemoryMetadata & metadata, ContextPt
             if (!all_columns.has(column_name))
             {
                 if (!command.if_exists)
-                    throw Exception{"Wrong column name. Cannot find column " + backQuote(column_name) + " to modify",
+                {
-                                    ErrorCodes::NOT_FOUND_COLUMN_IN_BLOCK};
+                    String exception_message = fmt::format("Wrong column. Cannot find column {} to modify", backQuote(column_name));
+                    all_columns.appendHintsMessage(exception_message, column_name);
+                    throw Exception{exception_message,
+                        ErrorCodes::NOT_FOUND_COLUMN_IN_BLOCK};
+                }
                 else
                     continue;
             }
@@ -1152,17 +1165,22 @@ void AlterCommands::validate(const StorageInMemoryMetadata & metadata, ContextPt
                 all_columns.remove(command.column_name);
             }
             else if (!command.if_exists)
-                throw Exception(
+            {
-                    "Wrong column name. Cannot find column " + backQuote(command.column_name) + " to drop",
+                String exception_message = fmt::format("Wrong column name. Cannot find column {} to drop", backQuote(command.column_name));
-                    ErrorCodes::NOT_FOUND_COLUMN_IN_BLOCK);
+                all_columns.appendHintsMessage(exception_message, command.column_name);
+                throw Exception(exception_message, ErrorCodes::NOT_FOUND_COLUMN_IN_BLOCK);
+            }
         }
         else if (command.type == AlterCommand::COMMENT_COLUMN)
         {
             if (!all_columns.has(command.column_name))
             {
                 if (!command.if_exists)
-                    throw Exception{"Wrong column name. Cannot find column " + backQuote(command.column_name) + " to comment",
+                {
-                                    ErrorCodes::NOT_FOUND_COLUMN_IN_BLOCK};
+                    String exception_message = fmt::format("Wrong column name. Cannot find column {} to comment", backQuote(command.column_name));
+                    all_columns.appendHintsMessage(exception_message, command.column_name);
+                    throw Exception(exception_message, ErrorCodes::NOT_FOUND_COLUMN_IN_BLOCK);
+                }
             }
         }
         else if (command.type == AlterCommand::MODIFY_SETTING || command.type == AlterCommand::RESET_SETTING)
@@ -1196,8 +1214,11 @@ void AlterCommands::validate(const StorageInMemoryMetadata & metadata, ContextPt
             if (!all_columns.has(command.column_name))
             {
                 if (!command.if_exists)
-                    throw Exception{"Wrong column name. Cannot find column " + backQuote(command.column_name) + " to rename",
+                {
-                                    ErrorCodes::NOT_FOUND_COLUMN_IN_BLOCK};
+                    String exception_message = fmt::format("Wrong column name. Cannot find column {} to rename", backQuote(command.column_name));
+                    all_columns.appendHintsMessage(exception_message, command.column_name);
+                    throw Exception(exception_message, ErrorCodes::NOT_FOUND_COLUMN_IN_BLOCK);
+                }
                 else
                     continue;
             }

src/Storages/ColumnsDescription.cpp

@@ -122,7 +122,7 @@ void ColumnDescription::readText(ReadBuffer & buf)
             if (col_ast->default_expression)
             {
                 default_desc.kind = columnDefaultKindFromString(col_ast->default_specifier);
-                default_desc.expression = col_ast->default_expression;
+                default_desc.expression = std::move(col_ast->default_expression);
             }
 
             if (col_ast->comment)
@@ -230,8 +230,11 @@ void ColumnsDescription::remove(const String & column_name)
 {
     auto range = getNameRange(columns, column_name);
     if (range.first == range.second)
-        throw Exception("There is no column " + column_name + " in table.",
+    {
-            ErrorCodes::NO_SUCH_COLUMN_IN_TABLE);
+        String exception_message = fmt::format("There is no column {} in table", column_name);
+        appendHintsMessage(exception_message, column_name);
+        throw Exception(exception_message, ErrorCodes::NO_SUCH_COLUMN_IN_TABLE);
+    }
 
     for (auto list_it = range.first; list_it != range.second;)
     {
@@ -244,7 +247,11 @@ void ColumnsDescription::rename(const String & column_from, const String & colum
 {
     auto it = columns.get<1>().find(column_from);
     if (it == columns.get<1>().end())
-        throw Exception("Cannot find column " + column_from + " in ColumnsDescription", ErrorCodes::LOGICAL_ERROR);
+    {
+        String exception_message = fmt::format("Cannot find column {} in ColumnsDescription", column_from);
+        appendHintsMessage(exception_message, column_from);
+        throw Exception(exception_message, ErrorCodes::LOGICAL_ERROR);
+    }
 
     columns.get<1>().modify_key(it, [&column_to] (String & old_name)
     {
@@ -745,6 +752,18 @@ void ColumnsDescription::removeSubcolumns(const String & name_in_storage)
         subcolumns.get<1>().erase(range.first, range.second);
 }
 
+std::vector<String> ColumnsDescription::getAllRegisteredNames() const
+{
+    std::vector<String> names;
+    names.reserve(columns.size());
+    for (const auto & column : columns)
+    {
+        if (column.name.find('.') == std::string::npos)
+            names.push_back(column.name);
+    }
+    return names;
+}
+
 Block validateColumnsDefaultsAndGetSampleBlock(ASTPtr default_expr_list, const NamesAndTypesList & all_columns, ContextPtr context)
 {
     for (const auto & child : default_expr_list->children)

src/Storages/ColumnsDescription.h

@@ -91,7 +91,7 @@ struct ColumnDescription
 
 
 /// Description of multiple table columns (in CREATE TABLE for example).
-class ColumnsDescription
+class ColumnsDescription : public IHints<1, ColumnsDescription>
 {
 public:
     ColumnsDescription() = default;
@@ -149,7 +149,11 @@ public:
     {
         auto it = columns.get<1>().find(column_name);
         if (it == columns.get<1>().end())
-            throw Exception("Cannot find column " + column_name + " in ColumnsDescription", ErrorCodes::LOGICAL_ERROR);
+        {
+            String exception_message = fmt::format("Cannot find column {} in ColumnsDescription", column_name);
+            appendHintsMessage(exception_message, column_name);
+            throw Exception(exception_message, ErrorCodes::LOGICAL_ERROR);
+        }
 
         removeSubcolumns(it->name);
         if (!columns.get<1>().modify(it, std::forward<F>(f)))
@@ -196,6 +200,8 @@ public:
         return columns.empty();
     }
 
+    std::vector<String> getAllRegisteredNames() const override;
+
     /// Keep the sequence of columns and allow to lookup by name.
     using ColumnsContainer = boost::multi_index_container<
         ColumnDescription,

src/Storages/IndicesDescription.h

@@ -74,7 +74,6 @@ struct IndicesDescription : public std::vector<IndexDescription>, IHints<1, Indi
     /// Return common expression for all stored indices
     ExpressionActionsPtr getSingleExpressionForIndices(const ColumnsDescription & columns, ContextPtr context) const;
 
-public:
     Names getAllRegisteredNames() const override;
 };
 

src/Storages/Kafka/StorageKafka.cpp

@@ -779,11 +779,16 @@ void registerStorageKafka(StorageFactory & factory)
         #undef CHECK_KAFKA_STORAGE_ARGUMENT
 
         auto num_consumers = kafka_settings->kafka_num_consumers.value;
-        auto physical_cpu_cores = getNumberOfPhysicalCPUCores();
+        auto max_consumers = std::max<uint32_t>(getNumberOfPhysicalCPUCores(), 16);
 
-        if (num_consumers > physical_cpu_cores)
+        if (num_consumers > max_consumers)
         {
-            throw Exception(ErrorCodes::BAD_ARGUMENTS, "Number of consumers can not be bigger than {}", physical_cpu_cores);
+            throw Exception(ErrorCodes::BAD_ARGUMENTS, "The number of consumers can not be bigger than {}. "
+                            "A single consumer can read any number of partitions. Extra consumers are relatively expensive, "
+                            "and using a lot of them can lead to high memory and CPU usage. To achieve better performance "
+                            "of getting data from Kafka, consider using a setting kafka_thread_per_consumer=1, "
+                            "and ensure you have enough threads in MessageBrokerSchedulePool (background_message_broker_schedule_pool_size). "
+                            "See also https://clickhouse.com/docs/integrations/kafka/kafka-table-engine#tuning-performance", max_consumers);
         }
         else if (num_consumers < 1)
         {

src/Storages/MergeTree/MergeTask.cpp

@@ -782,7 +782,7 @@ void MergeTask::ExecuteAndFinalizeHorizontalPart::createMergedStream()
 
     Block header = pipes.at(0).getHeader();
     for (size_t i = 0; i < sort_columns_size; ++i)
-        sort_description.emplace_back(header.getPositionByName(sort_columns[i]), 1, 1);
+        sort_description.emplace_back(sort_columns[i], 1, 1);
 
     /// The order of the streams is important: when the key is matched, the elements go in the order of the source stream number.
     /// In the merged part, the lines with the same key must be in the ascending order of the identifier of original part,

src/Storages/MergeTree/MergeTreeData.cpp

@@ -1909,6 +1909,7 @@ void MergeTreeData::checkAlterIsPossible(const AlterCommands & commands, Context
     StorageInMemoryMetadata old_metadata = getInMemoryMetadata();
 
     const auto & settings = local_context->getSettingsRef();
+    const auto & settings_from_storage = getSettings();
 
     if (!settings.allow_non_metadata_alters)
     {
@@ -2099,6 +2100,14 @@ void MergeTreeData::checkAlterIsPossible(const AlterCommands & commands, Context
 
             dropped_columns.emplace(command.column_name);
         }
+        else if (command.type == AlterCommand::RESET_SETTING)
+        {
+            for (const auto & reset_setting : command.settings_resets)
+            {
+                if (!settings_from_storage->has(reset_setting))
+                    throw Exception(ErrorCodes::BAD_ARGUMENTS, "Cannot reset setting '{}' because it doesn't exist for MergeTree engines family", reset_setting);
+            }
+        }
         else if (command.isRequireMutationStage(getInMemoryMetadata()))
         {
             /// This alter will override data on disk. Let's check that it doesn't
@@ -2953,7 +2962,8 @@ void MergeTreeData::tryRemovePartImmediately(DataPartPtr && part)
     {
         auto lock = lockParts();
 
-        LOG_TRACE(log, "Trying to immediately remove part {}", part->getNameWithState());
+        auto part_name_with_state = part->getNameWithState();
+        LOG_TRACE(log, "Trying to immediately remove part {}", part_name_with_state);
 
         if (part->getState() != DataPartState::Temporary)
         {
@@ -2964,7 +2974,16 @@ void MergeTreeData::tryRemovePartImmediately(DataPartPtr && part)
             part.reset();
 
             if (!((*it)->getState() == DataPartState::Outdated && it->unique()))
+            {
+                if ((*it)->getState() != DataPartState::Outdated)
+                    LOG_WARNING(log, "Cannot immediately remove part {} because it's not in Outdated state "
+                             "usage counter {}", part_name_with_state, it->use_count());
+
+                if (!it->unique())
+                    LOG_WARNING(log, "Cannot immediately remove part {} because someone using it right now "
+                             "usage counter {}", part_name_with_state, it->use_count());
                 return;
+            }
 
             modifyPartState(it, DataPartState::Deleting);
 
@@ -3375,7 +3394,12 @@ void MergeTreeData::checkAlterPartitionIsPossible(
 void MergeTreeData::checkPartitionCanBeDropped(const ASTPtr & partition)
 {
     const String partition_id = getPartitionIDFromQuery(partition, getContext());
-    auto parts_to_remove = getDataPartsVectorInPartition(MergeTreeDataPartState::Active, partition_id);
+    DataPartsVector parts_to_remove;
+    const auto * partition_ast = partition->as<ASTPartition>();
+    if (partition_ast && partition_ast->all)
+        parts_to_remove = getDataPartsVector();
+    else
+        parts_to_remove = getDataPartsVectorInPartition(MergeTreeDataPartState::Active, partition_id);
 
     UInt64 partition_size = 0;
 
@@ -3826,6 +3850,8 @@ String MergeTreeData::getPartitionIDFromQuery(const ASTPtr & ast, ContextPtr loc
 
     auto metadata_snapshot = getInMemoryMetadataPtr();
     const Block & key_sample_block = metadata_snapshot->getPartitionKey().sample_block;
+    if (partition_ast.all)
+        return "ALL";
     size_t fields_count = key_sample_block.columns();
     if (partition_ast.fields_count != fields_count)
         throw Exception(ErrorCodes::INVALID_PARTITION_VALUE,

src/Storages/MergeTree/MergeTreeDataPartTTLInfo.h

@@ -81,7 +81,7 @@ struct MergeTreeDataPartTTLInfos
     bool empty() const
     {
         /// part_min_ttl in minimum of rows, rows_where and group_by TTLs
-        return !part_min_ttl && moves_ttl.empty() && recompression_ttl.empty();
+        return !part_min_ttl && moves_ttl.empty() && recompression_ttl.empty() && columns_ttl.empty();
     }
 };
 

src/Storages/MergeTree/MergeTreeDataWriter.cpp

@@ -145,7 +145,7 @@ void MergeTreeDataWriter::TemporaryPart::finalize()
 }
 
 BlocksWithPartition MergeTreeDataWriter::splitBlockIntoParts(
-        const Block & block, size_t max_parts, const StorageMetadataPtr & metadata_snapshot, ContextPtr context)
+    const Block & block, size_t max_parts, const StorageMetadataPtr & metadata_snapshot, ContextPtr context)
 {
     BlocksWithPartition result;
     if (!block || !block.rows())
@@ -282,16 +282,12 @@ MergeTreeDataWriter::TemporaryPart MergeTreeDataWriter::writeTempPart(
 {
     TemporaryPart temp_part;
     Block & block = block_with_partition.block;
+
     auto columns = metadata_snapshot->getColumns().getAllPhysical().filter(block.getNames());
-    auto storage_snapshot = data.getStorageSnapshot(metadata_snapshot);
 
-    if (!storage_snapshot->object_columns.empty())
+    for (auto & column : columns)
-    {
+        if (isObject(column.type))
-        auto extended_storage_columns = storage_snapshot->getColumns(
+            column.type = block.getByName(column.name).type;
-            GetColumnsOptions(GetColumnsOptions::AllPhysical).withExtendedObjects());
-
-        convertObjectsToTuples(columns, block, extended_storage_columns);
-    }
 
     static const String TMP_PREFIX = "tmp_insert_";
 
@@ -333,7 +329,7 @@ MergeTreeDataWriter::TemporaryPart MergeTreeDataWriter::writeTempPart(
     sort_description.reserve(sort_columns_size);
 
     for (size_t i = 0; i < sort_columns_size; ++i)
-        sort_description.emplace_back(block.getPositionByName(sort_columns[i]), 1, 1);
+        sort_description.emplace_back(sort_columns[i], 1, 1);
 
     ProfileEvents::increment(ProfileEvents::MergeTreeDataWriterBlocks);
 
@@ -466,6 +462,16 @@ MergeTreeDataWriter::TemporaryPart MergeTreeDataWriter::writeTempPart(
     return temp_part;
 }
 
+void MergeTreeDataWriter::deduceTypesOfObjectColumns(const StorageSnapshotPtr & storage_snapshot, Block & block)
+{
+    if (!storage_snapshot->object_columns.empty())
+    {
+        auto options = GetColumnsOptions(GetColumnsOptions::AllPhysical).withExtendedObjects();
+        auto storage_columns = storage_snapshot->getColumns(options);
+        convertObjectsToTuples(block, storage_columns);
+    }
+}
+
 MergeTreeDataWriter::TemporaryPart MergeTreeDataWriter::writeProjectionPartImpl(
     const String & part_name,
     MergeTreeDataPartType part_type,
@@ -521,7 +527,7 @@ MergeTreeDataWriter::TemporaryPart MergeTreeDataWriter::writeProjectionPartImpl(
     sort_description.reserve(sort_columns_size);
 
     for (size_t i = 0; i < sort_columns_size; ++i)
-        sort_description.emplace_back(block.getPositionByName(sort_columns[i]), 1, 1);
+        sort_description.emplace_back(sort_columns[i], 1, 1);
 
     ProfileEvents::increment(ProfileEvents::MergeTreeDataProjectionWriterBlocks);
 

src/Storages/MergeTree/MergeTreeDataWriter.h

@@ -42,14 +42,12 @@ public:
       */
     static BlocksWithPartition splitBlockIntoParts(const Block & block, size_t max_parts, const StorageMetadataPtr & metadata_snapshot, ContextPtr context);
 
-    /** All rows must correspond to same partition.
+    static void deduceTypesOfObjectColumns(const StorageSnapshotPtr & storage_snapshot, Block & block);
-      * Returns part with unique name starting with 'tmp_', yet not added to MergeTreeData.
-      */
-    MergeTreeData::MutableDataPartPtr writeTempPart(BlockWithPartition & block, const StorageMetadataPtr & metadata_snapshot, bool optimize_on_insert);
 
     /// This structure contains not completely written temporary part.
     /// Some writes may happen asynchronously, e.g. for blob storages.
     /// You should call finalize() to wait until all data is written.
+
     struct TemporaryPart
     {
         MergeTreeData::MutableDataPartPtr part;
@@ -65,6 +63,9 @@ public:
         void finalize();
     };
 
+    /** All rows must correspond to same partition.
+      * Returns part with unique name starting with 'tmp_', yet not added to MergeTreeData.
+      */
     TemporaryPart writeTempPart(BlockWithPartition & block, const StorageMetadataPtr & metadata_snapshot, ContextPtr context);
 
     /// For insertion.

src/Storages/MergeTree/MergeTreeSink.cpp

@@ -50,7 +50,9 @@ struct MergeTreeSink::DelayedChunk
 void MergeTreeSink::consume(Chunk chunk)
 {
     auto block = getHeader().cloneWithColumns(chunk.detachColumns());
+    auto storage_snapshot = storage.getStorageSnapshot(metadata_snapshot);
 
+    storage.writer.deduceTypesOfObjectColumns(storage_snapshot, block);
     auto part_blocks = storage.writer.splitBlockIntoParts(block, max_parts_per_block, metadata_snapshot, context);
 
     using DelayedPartitions = std::vector<MergeTreeSink::DelayedChunk::Partition>;

src/Storages/MergeTree/ReplicatedMergeTreeSink.cpp

@@ -150,7 +150,8 @@ void ReplicatedMergeTreeSink::consume(Chunk chunk)
     if (quorum)
         checkQuorumPrecondition(zookeeper);
 
-    const Settings & settings = context->getSettingsRef();
+    auto storage_snapshot = storage.getStorageSnapshot(metadata_snapshot);
+    storage.writer.deduceTypesOfObjectColumns(storage_snapshot, block);
     auto part_blocks = storage.writer.splitBlockIntoParts(block, max_parts_per_block, metadata_snapshot, context);
 
     using DelayedPartitions = std::vector<ReplicatedMergeTreeSink::DelayedChunk::Partition>;
@@ -158,6 +159,7 @@ void ReplicatedMergeTreeSink::consume(Chunk chunk)
 
     size_t streams = 0;
     bool support_parallel_write = false;
+    const Settings & settings = context->getSettingsRef();
 
     for (auto & current_block : part_blocks)
     {