Merge branch 'master' into hive_style_partitioning

CMakeLists.txt

@@ -187,14 +187,6 @@ else ()
     set(NO_WHOLE_ARCHIVE --no-whole-archive)
 endif ()
 
-if (NOT CMAKE_BUILD_TYPE_UC STREQUAL "RELEASE")
-    # Can be lld or ld-lld or lld-13 or /path/to/lld.
-    if (LINKER_NAME MATCHES "lld")
-        set (CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -Wl,--gdb-index")
-        message (STATUS "Adding .gdb-index via --gdb-index linker option.")
-    endif ()
-endif()
-
 if (NOT (SANITIZE_COVERAGE OR WITH_COVERAGE)
     AND (CMAKE_BUILD_TYPE_UC STREQUAL "RELEASE"
         OR CMAKE_BUILD_TYPE_UC STREQUAL "RELWITHDEBINFO"
@@ -402,7 +394,7 @@ if ((NOT OS_LINUX AND NOT OS_ANDROID) OR (CMAKE_BUILD_TYPE_UC STREQUAL "DEBUG")
     set(ENABLE_GWP_ASAN OFF)
 endif ()
 
-option (ENABLE_FIU "Enable Fiu" ON)
+option (ENABLE_LIBFIU "Enable libfiu" ON)
 
 option(WERROR "Enable -Werror compiler option" ON)
 

contrib/CMakeLists.txt

@@ -179,7 +179,7 @@ else()
     message(STATUS "Not using QPL")
 endif ()
 
-if (OS_LINUX AND ARCH_AMD64)
+if (OS_LINUX AND ARCH_AMD64 AND NOT NO_SSE3_OR_HIGHER)
     option (ENABLE_QATLIB "Enable Intel® QuickAssist Technology Library (QATlib)" ${ENABLE_LIBRARIES})
 elseif(ENABLE_QATLIB)
     message (${RECONFIGURE_MESSAGE_LEVEL} "QATLib is only supported on x86_64")

contrib/QAT-ZSTD-Plugin-cmake/CMakeLists.txt

@@ -27,7 +27,7 @@ if (ENABLE_QAT_OUT_OF_TREE_BUILD)
                 ${QAT_AL_INCLUDE_DIR}
                 ${QAT_USDM_INCLUDE_DIR}
                 ${ZSTD_LIBRARY_DIR})
-    target_compile_definitions(_qatzstd_plugin PRIVATE -DDEBUGLEVEL=0 PUBLIC -DENABLE_ZSTD_QAT_CODEC)
+    target_compile_definitions(_qatzstd_plugin PRIVATE -DDEBUGLEVEL=0)
     add_library (ch_contrib::qatzstd_plugin ALIAS _qatzstd_plugin)
 else () # In-tree build
     message(STATUS "Intel QATZSTD in-tree build")
@@ -78,7 +78,7 @@ else () # In-tree build
         ${QAT_USDM_INCLUDE_DIR}
         ${ZSTD_LIBRARY_DIR}
         ${LIBQAT_HEADER_DIR})
-    target_compile_definitions(_qatzstd_plugin PRIVATE -DDEBUGLEVEL=0 PUBLIC -DENABLE_ZSTD_QAT_CODEC -DINTREE)
+    target_compile_definitions(_qatzstd_plugin PRIVATE -DDEBUGLEVEL=0 PUBLIC -DINTREE)
     target_include_directories(_qatzstd_plugin SYSTEM PUBLIC $<BUILD_INTERFACE:${QATZSTD_SRC_DIR}> $<INSTALL_INTERFACE:include>)
     add_library (ch_contrib::qatzstd_plugin ALIAS _qatzstd_plugin)
 endif ()

contrib/libfiu-cmake/CMakeLists.txt

@@ -1,20 +1,21 @@
-if (NOT ENABLE_FIU)
-  message (STATUS "Not using fiu")
+if (NOT ENABLE_LIBFIU)
+  message (STATUS "Not using libfiu")
   return ()
 endif ()
 
-set(FIU_DIR "${ClickHouse_SOURCE_DIR}/contrib/libfiu/")
+set(LIBFIU_DIR "${ClickHouse_SOURCE_DIR}/contrib/libfiu/")
 
-set(FIU_SOURCES
-  ${FIU_DIR}/libfiu/fiu.c
-  ${FIU_DIR}/libfiu/fiu-rc.c
-  ${FIU_DIR}/libfiu/backtrace.c
-  ${FIU_DIR}/libfiu/wtable.c
+set(LIBFIU_SOURCES
+    ${LIBFIU_DIR}/libfiu/fiu.c
+    ${LIBFIU_DIR}/libfiu/fiu-rc.c
+    ${LIBFIU_DIR}/libfiu/backtrace.c
+    ${LIBFIU_DIR}/libfiu/wtable.c
 )
 
-set(FIU_HEADERS "${FIU_DIR}/libfiu")
+set(LIBFIU_HEADERS "${LIBFIU_DIR}/libfiu")
 
-add_library(_fiu ${FIU_SOURCES})
-target_compile_definitions(_fiu PUBLIC DUMMY_BACKTRACE)
-target_include_directories(_fiu PUBLIC ${FIU_HEADERS})
-add_library(ch_contrib::fiu ALIAS _fiu)
+add_library(_libfiu ${LIBFIU_SOURCES})
+target_compile_definitions(_libfiu PUBLIC DUMMY_BACKTRACE)
+target_compile_definitions(_libfiu PUBLIC FIU_ENABLE)
+target_include_directories(_libfiu PUBLIC ${LIBFIU_HEADERS})
+add_library(ch_contrib::libfiu ALIAS _libfiu)

contrib/qpl-cmake/CMakeLists.txt

@@ -728,10 +728,6 @@ add_library(_qpl STATIC ${LIB_DEPS})
 target_include_directories(_qpl
         PUBLIC $<BUILD_INTERFACE:${QPL_PROJECT_DIR}/include/> $<INSTALL_INTERFACE:include>)
 
-
-target_compile_definitions(_qpl
-        PUBLIC -DENABLE_QPL_COMPRESSION)
-
 target_link_libraries(_qpl
         PRIVATE ch_contrib::accel-config)
 

docker/packager/cctools/Dockerfile

@@ -1,3 +1,5 @@
+# docker build -t clickhouse/cctools .
+
 # This is a hack to significantly reduce the build time of the clickhouse/binary-builder
 # It's based on the assumption that we don't care of the cctools version so much
 # It event does not depend on the clickhouse/fasttest in the `docker/images.json`
@@ -30,5 +32,29 @@ RUN git clone https://github.com/tpoechtrager/cctools-port.git \
     && cd ../.. \
     && rm -rf cctools-port
 
+#
+# GDB
+#
+# ld from binutils is 2.38, which has the following error:
+#
+#     DWARF error: invalid or unhandled FORM value: 0x23
+#
+ENV LD=ld.lld-${LLVM_VERSION}
+ARG GDB_VERSION=15.1
+RUN apt-get update \
+    && apt-get install --yes \
+        libgmp-dev \
+        libmpfr-dev \
+    && apt-get clean \
+    && rm -rf /var/lib/apt/lists/* /var/cache/debconf /tmp/*
+RUN wget https://sourceware.org/pub/gdb/releases/gdb-$GDB_VERSION.tar.gz \
+    && tar -xvf gdb-$GDB_VERSION.tar.gz \
+    && cd gdb-$GDB_VERSION \
+    && ./configure --prefix=/opt/gdb \
+    && make -j $(nproc) \
+    && make install \
+    && rm -fr gdb-$GDB_VERSION gdb-$GDB_VERSION.tar.gz
+
 FROM scratch
 COPY --from=builder /cctools /cctools
+COPY --from=builder /opt/gdb /opt/gdb

docker/test/fasttest/Dockerfile

@@ -83,7 +83,7 @@ RUN arch=${TARGETARCH:-amd64} \
 
 # Give suid to gdb to grant it attach permissions
 # chmod 777 to make the container user independent
-RUN chmod u+s /usr/bin/gdb \
+RUN chmod u+s /opt/gdb/bin/gdb \
   && mkdir -p /var/lib/clickhouse \
   && chmod 777 /var/lib/clickhouse
 

docker/test/integration/base/Dockerfile

@@ -11,7 +11,6 @@ RUN apt-get update \
         curl \
         default-jre \
         g++ \
-        gdb \
         iproute2 \
         krb5-user \
         libicu-dev \
@@ -73,3 +72,6 @@ maxClientCnxns=80' > /opt/zookeeper/conf/zoo.cfg && \
 
 ENV TZ=Etc/UTC
 RUN ln -snf /usr/share/zoneinfo/$TZ /etc/localtime && echo $TZ > /etc/timezone
+
+COPY --from=clickhouse/cctools:0d6b90a7a490 /opt/gdb /opt/gdb
+ENV PATH="/opt/gdb/bin:${PATH}"

docker/test/integration/runner/Dockerfile

@@ -30,7 +30,6 @@ RUN apt-get update \
     luajit \
     libssl-dev \
     libcurl4-openssl-dev \
-    gdb \
     default-jdk \
     software-properties-common \
     libkrb5-dev \
@@ -87,6 +86,8 @@ COPY modprobe.sh /usr/local/bin/modprobe
 COPY dockerd-entrypoint.sh /usr/local/bin/
 COPY misc/ /misc/
 
+COPY --from=clickhouse/cctools:0d6b90a7a490 /opt/gdb /opt/gdb
+ENV PATH="/opt/gdb/bin:${PATH}"
 
 # Same options as in test/base/Dockerfile
 # (in case you need to override them in tests)

docker/test/performance-comparison/Dockerfile

@@ -9,7 +9,6 @@ RUN apt-get update \
             curl \
             dmidecode \
             g++ \
-            gdb \
             git \
             gnuplot \
             imagemagick \
@@ -42,6 +41,9 @@ RUN pip3 --no-cache-dir install -r requirements.txt
 
 COPY run.sh /
 
+COPY --from=clickhouse/cctools:0d6b90a7a490 /opt/gdb /opt/gdb
+ENV PATH="/opt/gdb/bin:${PATH}"
+
 CMD ["bash", "/run.sh"]
 
 # docker run --network=host --volume <workspace>:/workspace --volume=<output>:/output -e PR_TO_TEST=<> -e SHA_TO_TEST=<> clickhouse/performance-comparison

docker/test/stateless/Dockerfile

@@ -69,8 +69,8 @@ ENV MAX_RUN_TIME=0
 
 # Unrelated to vars in setup_minio.sh, but should be the same there
 # to have the same binaries for local running scenario
-ARG MINIO_SERVER_VERSION=2022-01-03T18-22-58Z
-ARG MINIO_CLIENT_VERSION=2022-01-05T23-52-51Z
+ARG MINIO_SERVER_VERSION=2024-08-03T04-33-23Z
+ARG MINIO_CLIENT_VERSION=2024-07-31T15-58-33Z
 ARG TARGETARCH
 
 # Download Minio-related binaries

docker/test/stateless/setup_minio.sh

@@ -59,8 +59,8 @@ find_os() {
 download_minio() {
   local os
   local arch
-  local minio_server_version=${MINIO_SERVER_VERSION:-2022-09-07T22-25-02Z}
-  local minio_client_version=${MINIO_CLIENT_VERSION:-2022-08-28T20-08-11Z}
+  local minio_server_version=${MINIO_SERVER_VERSION:-2024-08-03T04-33-23Z}
+  local minio_client_version=${MINIO_CLIENT_VERSION:-2024-07-31T15-58-33Z}
 
   os=$(find_os)
   arch=$(find_arch)
@@ -82,10 +82,10 @@ setup_minio() {
   local test_type=$1
   ./mc alias set clickminio http://localhost:11111 clickhouse clickhouse
   ./mc admin user add clickminio test testtest
-  ./mc admin policy set clickminio readwrite user=test
+  ./mc admin policy attach clickminio readwrite --user=test
   ./mc mb --ignore-existing clickminio/test
   if [ "$test_type" = "stateless" ]; then
-    ./mc policy set public clickminio/test
+    ./mc anonymous set public clickminio/test
   fi
 }
 

docker/test/util/Dockerfile

@@ -44,7 +44,6 @@ RUN apt-get update \
         bash \
         bsdmainutils \
         build-essential \
-        gdb \
         git \
         gperf \
         moreutils \
@@ -58,3 +57,6 @@ RUN apt-get update \
     && rm -rf /var/lib/apt/lists/* /var/cache/debconf /tmp/*
 
 COPY process_functional_tests_result.py /
+
+COPY --from=clickhouse/cctools:0d6b90a7a490 /opt/gdb /opt/gdb
+ENV PATH="/opt/gdb/bin:${PATH}"

docs/en/interfaces/prometheus.md

@@ -75,7 +75,7 @@ Data are received by this protocol and written to a [TimeSeries](/en/engines/tab
         <my_rule_1>
             <url>/write</url>
             <handler>
-                <type>remote_write</type
+                <type>remote_write</type>
                 <database>db_name</database>
                 <table>time_series_table</table>
             </handler>
@@ -105,7 +105,7 @@ Data are read from a [TimeSeries](/en/engines/table-engines/special/time_series)
         <my_rule_1>
             <url>/read</url>
             <handler>
-                <type>remote_read</type
+                <type>remote_read</type>
                 <database>db_name</database>
                 <table>time_series_table</table>
             </handler>
@@ -144,14 +144,14 @@ Multiple protocols can be specified together in one place:
         <my_rule_2>
             <url>/write</url>
             <handler>
-                <type>remote_write</type
+                <type>remote_write</type>
                 <table>db_name.time_series_table</table>
             </handler>
         </my_rule_2>
         <my_rule_3>
             <url>/read</url>
             <handler>
-                <type>remote_read</type
+                <type>remote_read</type>
                 <table>db_name.time_series_table</table>
             </handler>
         </my_rule_3>

docs/en/sql-reference/data-types/geo.md

@@ -52,6 +52,48 @@ Result:
 └───────────────────────────────┴───────────────┘
 ```
 
+## LineString
+
+`LineString` is a line stored as an array of points: [Array](array.md)([Point](#point)).
+
+**Example**
+
+Query:
+
+```sql
+CREATE TABLE geo_linestring (l LineString) ENGINE = Memory();
+INSERT INTO geo_linestring VALUES([(0, 0), (10, 0), (10, 10), (0, 10)]);
+SELECT l, toTypeName(l) FROM geo_linestring;
+```
+Result:
+
+``` text
+┌─r─────────────────────────────┬─toTypeName(r)─┐
+│ [(0,0),(10,0),(10,10),(0,10)] │ LineString    │
+└───────────────────────────────┴───────────────┘
+```
+
+## MultiLineString
+
+`MultiLineString` is multiple lines stored as an array of `LineString`: [Array](array.md)([LineString](#linestring)).
+
+**Example**
+
+Query:
+
+```sql
+CREATE TABLE geo_multilinestring (l MultiLineString) ENGINE = Memory();
+INSERT INTO geo_multilinestring VALUES([[(0, 0), (10, 0), (10, 10), (0, 10)], [(1, 1), (2, 2), (3, 3)]]);
+SELECT l, toTypeName(l) FROM geo_multilinestring;
+```
+Result:
+
+``` text
+┌─l───────────────────────────────────────────────────┬─toTypeName(l)───┐
+│ [[(0,0),(10,0),(10,10),(0,10)],[(1,1),(2,2),(3,3)]] │ MultiLineString │
+└─────────────────────────────────────────────────────┴─────────────────┘
+```
+
 ## Polygon
 
 `Polygon` is a polygon with holes stored as an array of rings: [Array](array.md)([Ring](#ring)). First element of outer array is the outer shape of polygon and all the following elements are holes.

docs/en/sql-reference/functions/geo/polygon.md

@@ -11,6 +11,8 @@ Returns a WKT (Well Known Text) geometric object from various [Geo Data Types](.
 - POINT
 - POLYGON
 - MULTIPOLYGON
+- LINESTRING
+- MULTILINESTRING
 
 **Syntax**
 
@@ -26,12 +28,16 @@ WKT(geo_data)
 - [Ring](../../data-types/geo.md#ring)
 - [Polygon](../../data-types/geo.md#polygon)
 - [MultiPolygon](../../data-types/geo.md#multipolygon)
+- [LineString](../../data-types/geo.md#linestring)
+- [MultiLineString](../../data-types/geo.md#multilinestring)
 
 **Returned value**
 
 - WKT geometric object `POINT` is returned for a Point.
 - WKT geometric object `POLYGON` is returned for a Polygon
 - WKT geometric object `MULTIPOLYGON` is returned for a MultiPolygon.
+- WKT geometric object `LINESTRING` is returned for a LineString.
+- WKT geometric object `MULTILINESTRING` is returned for a MultiLineString.
 
 **Examples**
 
@@ -170,6 +176,34 @@ SELECT readWKTLineString('LINESTRING (1 1, 2 2, 3 3, 1 1)');
 [(1,1),(2,2),(3,3),(1,1)]
 ```
 
+## readWKTMultiLineString
+
+Parses a Well-Known Text (WKT) representation of a MultiLineString geometry and returns it in the internal ClickHouse format.
+
+### Syntax
+
+```sql
+readWKTMultiLineString(wkt_string)
+```
+
+### Arguments
+
+- `wkt_string`: The input WKT string representing a MultiLineString geometry.
+
+### Returned value
+
+The function returns a ClickHouse internal representation of the multilinestring geometry.
+
+### Example
+
+```sql
+SELECT readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3), (4 4, 5 5, 6 6))');
+```
+
+```response
+[[(1,1),(2,2),(3,3)],[(4,4),(5,5),(6,6)]]
+```
+
 ## readWKTRing
 
 Parses a Well-Known Text (WKT) representation of a Polygon geometry and returns a ring (closed linestring) in the internal ClickHouse format.

docs/en/sql-reference/statements/select/join.md

@@ -186,7 +186,7 @@ Otherwise, you'll get `INVALID_JOIN_ON_EXPRESSION`.
 
 :::
 
-Clickhouse currently supports `ALL INNER/LEFT/RIGHT/FULL JOIN` with inequality conditions in addition to equality conditions. The inequality conditions are supported only for `hash` and `grace_hash` join algorithms. The inequality conditions are not supported with `join_use_nulls`.
+Clickhouse currently supports `ALL/ANY/SEMI/ANTI INNER/LEFT/RIGHT/FULL JOIN` with inequality conditions in addition to equality conditions. The inequality conditions are supported only for `hash` and `grace_hash` join algorithms. The inequality conditions are not supported with `join_use_nulls`.
 
 **Example**
 

programs/local/LocalServer.cpp

@@ -143,7 +143,7 @@ void LocalServer::initialize(Poco::Util::Application & self)
 
     if (fs::exists(config_path))
     {
-        ConfigProcessor config_processor(config_path, false, true);
+        ConfigProcessor config_processor(config_path);
         ConfigProcessor::setConfigPath(fs::path(config_path).parent_path());
         auto loaded_config = config_processor.loadConfig();
         getClientConfiguration().add(loaded_config.configuration.duplicate(), PRIO_DEFAULT, false);

src/Backups/WithRetries.cpp

@@ -68,13 +68,19 @@ const WithRetries::KeeperSettings & WithRetries::getKeeperSettings() const
 
 WithRetries::FaultyKeeper WithRetries::getFaultyZooKeeper() const
 {
-    /// We need to create new instance of ZooKeeperWithFaultInjection each time a copy a pointer to ZooKeeper client there
+    zkutil::ZooKeeperPtr current_zookeeper;
+    {
+        std::lock_guard lock(zookeeper_mutex);
+        current_zookeeper = zookeeper;
+    }
+
+    /// We need to create new instance of ZooKeeperWithFaultInjection each time and copy a pointer to ZooKeeper client there
     /// The reason is that ZooKeeperWithFaultInjection may reset the underlying pointer and there could be a race condition
     /// when the same object is used from multiple threads.
     auto faulty_zookeeper = ZooKeeperWithFaultInjection::createInstance(
         settings.keeper_fault_injection_probability,
         settings.keeper_fault_injection_seed,
-        zookeeper,
+        current_zookeeper,
         log->name(),
         log);
 

src/CMakeLists.txt

@@ -353,8 +353,8 @@ target_link_libraries(clickhouse_common_io
             Poco::Foundation
 )
 
-if (TARGET ch_contrib::fiu)
-    target_link_libraries(clickhouse_common_io PUBLIC ch_contrib::fiu)
+if (TARGET ch_contrib::libfiu)
+    target_link_libraries(clickhouse_common_io PUBLIC ch_contrib::libfiu)
 endif()
 
 if (TARGET ch_contrib::cpuid)
@@ -556,14 +556,13 @@ target_link_libraries (clickhouse_common_io PRIVATE ch_contrib::lz4)
 
 if (TARGET ch_contrib::qpl)
     dbms_target_link_libraries(PUBLIC ch_contrib::qpl)
+    target_link_libraries (clickhouse_compression PUBLIC ch_contrib::qpl)
+    target_link_libraries (clickhouse_compression PUBLIC ch_contrib::accel-config)
 endif ()
 
-if (TARGET ch_contrib::accel-config)
-    dbms_target_link_libraries(PUBLIC ch_contrib::accel-config)
-endif ()
-
-if (TARGET ch_contrib::qatzstd_plugin)
+if (TARGET ch_contrib::accel-config AND TARGET ch_contrib::qatzstd_plugin)
     dbms_target_link_libraries(PUBLIC ch_contrib::qatzstd_plugin)
+    dbms_target_link_libraries(PUBLIC ch_contrib::accel-config)
     target_link_libraries(clickhouse_common_io PUBLIC ch_contrib::qatzstd_plugin)
 endif ()
 

src/Client/ClientApplicationBase.cpp

@@ -200,8 +200,6 @@ void ClientApplicationBase::init(int argc, char ** argv)
         ("pager", po::value<std::string>(), "Pipe all output into this command (less or similar)")
         ("max_memory_usage_in_client", po::value<std::string>(), "Set memory limit in client/local server")
 
-        ("fuzzer-args", po::value<std::string>(), "Command line arguments for the LLVM's libFuzzer driver. Only relevant if the application is compiled with libFuzzer.")
-
         ("client_logs_file", po::value<std::string>(), "Path to a file for writing client logs. Currently we only have fatal logs (when the client crashes)")
     ;
 

src/Common/FailPoint.cpp

@@ -7,6 +7,8 @@
 #include <condition_variable>
 #include <mutex>
 
+#include "config.h"
+
 namespace DB
 {
 
@@ -15,7 +17,7 @@ namespace ErrorCodes
 extern const int LOGICAL_ERROR;
 };
 
-#if FIU_ENABLE
+#if USE_LIBFIU
 static struct InitFiu
 {
     InitFiu()
@@ -135,7 +137,7 @@ void FailPointInjection::pauseFailPoint(const String & fail_point_name)
 
 void FailPointInjection::enableFailPoint(const String & fail_point_name)
 {
-#if FIU_ENABLE
+#if USE_LIBFIU
 #define SUB_M(NAME, flags, pause)                                                                               \
     if (fail_point_name == FailPoints::NAME)                                                                    \
     {                                                                                                           \

src/Common/FailPoint.h

@@ -1,17 +1,16 @@
 #pragma once
-#include "config.h"
 
 #include <Common/Exception.h>
 #include <Core/Types.h>
 #include <Poco/Util/AbstractConfiguration.h>
 
+#include "config.h"
+
 #pragma clang diagnostic push
 #pragma clang diagnostic ignored "-Wdocumentation"
 #pragma clang diagnostic ignored "-Wreserved-macro-identifier"
-
-#include <fiu.h>
-#include <fiu-control.h>
-
+#  include <fiu.h>
+#  include <fiu-control.h>
 #pragma clang diagnostic pop
 
 #include <unordered_map>

src/Common/ShellCommand.cpp

@@ -237,7 +237,14 @@ std::unique_ptr<ShellCommand> ShellCommand::executeImpl(
         res->write_fds.emplace(fd, fds.fds_rw[1]);
     }
 
-    LOG_TRACE(getLogger(), "Started shell command '{}' with pid {}", filename, pid);
+    LOG_TRACE(
+        getLogger(),
+        "Started shell command '{}' with pid {} and file descriptors: out {}, err {}",
+        filename,
+        pid,
+        res->out.getFD(),
+        res->err.getFD());
+
     return res;
 }
 

src/Common/config.h.in

@@ -32,6 +32,8 @@
 #cmakedefine01 USE_IDNA
 #cmakedefine01 USE_NLP
 #cmakedefine01 USE_VECTORSCAN
+#cmakedefine01 USE_QPL
+#cmakedefine01 USE_QATLIB
 #cmakedefine01 USE_LIBURING
 #cmakedefine01 USE_AVRO
 #cmakedefine01 USE_CAPNP
@@ -59,7 +61,7 @@
 #cmakedefine01 USE_SKIM
 #cmakedefine01 USE_PRQL
 #cmakedefine01 USE_ULID
-#cmakedefine01 FIU_ENABLE
+#cmakedefine01 USE_LIBFIU
 #cmakedefine01 USE_BCRYPT
 #cmakedefine01 USE_LIBARCHIVE
 #cmakedefine01 USE_POCKETFFT

src/Compression/CompressionCodecDeflateQpl.cpp

@@ -1,7 +1,3 @@
-#ifdef ENABLE_QPL_COMPRESSION
-
-#include <cstdio>
-#include <thread>
 #include <Compression/CompressionCodecDeflateQpl.h>
 #include <Compression/CompressionFactory.h>
 #include <Compression/CompressionInfo.h>
@@ -11,6 +7,10 @@
 #include <Common/randomSeed.h>
 #include <base/scope_guard.h>
 #include <base/getPageSize.h>
+#include <cstdio>
+#include <thread>
+
+#if USE_QPL
 
 #include "libaccel_config.h"
 

src/Compression/CompressionCodecDeflateQpl.h

@@ -4,6 +4,11 @@
 #include <map>
 #include <random>
 #include <pcg_random.hpp>
+
+#include "config.h"
+
+#if USE_QPL
+
 #include <qpl/qpl.h>
 
 namespace Poco
@@ -117,3 +122,4 @@ private:
 };
 
 }
+#endif

src/Compression/CompressionCodecZSTDQAT.cpp

@@ -1,4 +1,6 @@
-#ifdef ENABLE_ZSTD_QAT_CODEC
+#include "config.h"
+
+#if USE_QATLIB
 
 #include <Common/logger_useful.h>
 #include <Compression/CompressionCodecZSTD.h>
@@ -6,6 +8,7 @@
 #include <Parsers/ASTLiteral.h>
 #include <Parsers/IAST.h>
 #include <Poco/Logger.h>
+
 #include <qatseqprod.h>
 #include <zstd.h>
 

src/Compression/CompressionFactory.cpp

@@ -1,20 +1,20 @@
-#include "config.h"
-
 #include <Compression/CompressionFactory.h>
+#include <Compression/CompressionCodecMultiple.h>
+#include <Compression/CompressionCodecNone.h>
+#include <IO/ReadBuffer.h>
+#include <IO/WriteHelpers.h>
 #include <Parsers/ASTFunction.h>
 #include <Parsers/ASTIdentifier.h>
 #include <Parsers/ASTLiteral.h>
-#include <Poco/String.h>
-#include <IO/ReadBuffer.h>
-#include <Parsers/queryToString.h>
-#include <Parsers/parseQuery.h>
 #include <Parsers/ExpressionElementParsers.h>
-#include <Compression/CompressionCodecMultiple.h>
-#include <Compression/CompressionCodecNone.h>
-#include <IO/WriteHelpers.h>
+#include <Parsers/parseQuery.h>
+#include <Parsers/queryToString.h>
+#include <Poco/String.h>
 
 #include <boost/algorithm/string/join.hpp>
 
+#include "config.h"
+
 namespace DB
 {
 
@@ -175,11 +175,11 @@ void registerCodecNone(CompressionCodecFactory & factory);
 void registerCodecLZ4(CompressionCodecFactory & factory);
 void registerCodecLZ4HC(CompressionCodecFactory & factory);
 void registerCodecZSTD(CompressionCodecFactory & factory);
-#ifdef ENABLE_ZSTD_QAT_CODEC
+#if USE_QATLIB
 void registerCodecZSTDQAT(CompressionCodecFactory & factory);
 #endif
 void registerCodecMultiple(CompressionCodecFactory & factory);
-#ifdef ENABLE_QPL_COMPRESSION
+#if USE_QPL
 void registerCodecDeflateQpl(CompressionCodecFactory & factory);
 #endif
 
@@ -198,7 +198,7 @@ CompressionCodecFactory::CompressionCodecFactory()
     registerCodecNone(*this);
     registerCodecLZ4(*this);
     registerCodecZSTD(*this);
-#ifdef ENABLE_ZSTD_QAT_CODEC
+#if USE_QATLIB
     registerCodecZSTDQAT(*this);
 #endif
     registerCodecLZ4HC(*this);
@@ -209,7 +209,7 @@ CompressionCodecFactory::CompressionCodecFactory()
     registerCodecGorilla(*this);
     registerCodecEncrypted(*this);
     registerCodecFPC(*this);
-#ifdef ENABLE_QPL_COMPRESSION
+#if USE_QPL
     registerCodecDeflateQpl(*this);
 #endif
     registerCodecGCD(*this);

src/Core/Settings.h

@@ -605,7 +605,7 @@ class IColumn;
     M(Bool, optimize_if_chain_to_multiif, false, "Replace if(cond1, then1, if(cond2, ...)) chains to multiIf. Currently it's not beneficial for numeric types.", 0) \
     M(Bool, optimize_multiif_to_if, true, "Replace 'multiIf' with only one condition to 'if'.", 0) \
     M(Bool, optimize_if_transform_strings_to_enum, false, "Replaces string-type arguments in If and Transform to enum. Disabled by default cause it could make inconsistent change in distributed query that would lead to its fail.", 0) \
-    M(Bool, optimize_functions_to_subcolumns, false, "Transform functions to subcolumns, if possible, to reduce amount of read data. E.g. 'length(arr)' -> 'arr.size0', 'col IS NULL' -> 'col.null' ", 0) \
+    M(Bool, optimize_functions_to_subcolumns, true, "Transform functions to subcolumns, if possible, to reduce amount of read data. E.g. 'length(arr)' -> 'arr.size0', 'col IS NULL' -> 'col.null' ", 0) \
     M(Bool, optimize_using_constraints, false, "Use constraints for query optimization", 0)                                                                                                                                           \
     M(Bool, optimize_substitute_columns, false, "Use constraints for column substitution", 0)                                                                                                                                         \
     M(Bool, optimize_append_index, false, "Use constraints in order to append index condition (indexHint)", 0) \

src/Core/SettingsChangesHistory.cpp

@@ -85,6 +85,7 @@ static std::initializer_list<std::pair<ClickHouseVersion, SettingsChangesHistory
             {"allow_archive_path_syntax", true, true, "Added new setting to allow disabling archive path syntax."},
             {"allow_experimental_time_series_table", false, false, "Added new setting to allow the TimeSeries table engine"},
             {"enable_analyzer", 1, 1, "Added an alias to a setting `allow_experimental_analyzer`."},
+            {"optimize_functions_to_subcolumns", false, true, "Enabled settings by default"},
         }
     },
     {"24.7",

src/DataTypes/DataTypeCustomGeo.cpp

@@ -24,6 +24,13 @@ void registerDataTypeDomainGeo(DataTypeFactory & factory)
             std::make_unique<DataTypeCustomDesc>(std::make_unique<DataTypeLineStringName>()));
     });
 
+    // Custom type for multiple lines stored as Array(LineString)
+    factory.registerSimpleDataTypeCustom("MultiLineString", []
+    {
+        return std::make_pair(DataTypeFactory::instance().get("Array(LineString)"),
+            std::make_unique<DataTypeCustomDesc>(std::make_unique<DataTypeMultiLineStringName>()));
+    });
+
     // Custom type for simple polygon without holes stored as Array(Point)
     factory.registerSimpleDataTypeCustom("Ring", []
     {

src/DataTypes/DataTypeCustomGeo.h

@@ -17,6 +17,12 @@ public:
     DataTypeLineStringName() : DataTypeCustomFixedName("LineString") {}
 };
 
+class DataTypeMultiLineStringName : public DataTypeCustomFixedName
+{
+public:
+    DataTypeMultiLineStringName() : DataTypeCustomFixedName("MultiLineString") {}
+};
+
 class DataTypeRingName : public DataTypeCustomFixedName
 {
 public:

src/Disks/ObjectStorages/S3/S3ObjectStorage.cpp

@@ -305,7 +305,8 @@ void S3ObjectStorage::listObjects(const std::string & path, RelativePathsWithMet
 
     S3::ListObjectsV2Request request;
     request.SetBucket(uri.bucket);
-    request.SetPrefix(path);
+    if (path != "/")
+        request.SetPrefix(path);
     if (max_keys)
         request.SetMaxKeys(static_cast<int>(max_keys));
     else

src/Functions/CastOverloadResolver.cpp

@@ -3,6 +3,7 @@
 #include <Functions/FunctionHelpers.h>
 #include <DataTypes/DataTypeFactory.h>
 #include <DataTypes/DataTypeNullable.h>
+#include <DataTypes/DataTypeString.h>
 #include <Columns/ColumnString.h>
 #include <Core/Settings.h>
 #include <Interpreters/parseColumnsListForTableFunction.h>
@@ -35,7 +36,7 @@ FunctionBasePtr createFunctionBaseCast(
 class CastOverloadResolverImpl : public IFunctionOverloadResolver
 {
 public:
-    const char * getNameImpl() const
+    static const char * getNameImpl(CastType cast_type, bool internal)
     {
         if (cast_type == CastType::accurate)
             return "accurateCast";
@@ -49,7 +50,7 @@ public:
 
     String getName() const override
     {
-        return getNameImpl();
+        return getNameImpl(cast_type, internal);
     }
 
     size_t getNumberOfArguments() const override { return 2; }
@@ -79,10 +80,22 @@ public:
         }
     }
 
+    static FunctionBasePtr createInternalCast(ColumnWithTypeAndName from, DataTypePtr to, CastType cast_type, std::optional<CastDiagnostic> diagnostic)
+    {
+        if (cast_type == CastType::accurateOrNull && !isVariant(to))
+            to = makeNullable(to);
+
+        ColumnsWithTypeAndName arguments;
+        arguments.emplace_back(std::move(from));
+        arguments.emplace_back().type = std::make_unique<DataTypeString>();
+
+        return createFunctionBaseCast(nullptr, getNameImpl(cast_type, true), arguments, to, diagnostic, cast_type);
+    }
+
 protected:
     FunctionBasePtr buildImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr & return_type) const override
     {
-        return createFunctionBaseCast(context, getNameImpl(), arguments, return_type, diagnostic, cast_type);
+        return createFunctionBaseCast(context, getNameImpl(cast_type, internal), arguments, return_type, diagnostic, cast_type);
     }
 
     DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override
@@ -130,9 +143,9 @@ private:
 };
 
 
-FunctionOverloadResolverPtr createInternalCastOverloadResolver(CastType type, std::optional<CastDiagnostic> diagnostic)
+FunctionBasePtr createInternalCast(ColumnWithTypeAndName from, DataTypePtr to, CastType cast_type, std::optional<CastDiagnostic> diagnostic)
 {
-    return CastOverloadResolverImpl::create(ContextPtr{}, type, true, diagnostic);
+    return CastOverloadResolverImpl::createInternalCast(std::move(from), std::move(to), cast_type, std::move(diagnostic));
 }
 
 REGISTER_FUNCTION(CastOverloadResolvers)

src/Functions/CastOverloadResolver.h

@@ -3,6 +3,7 @@
 #include <memory>
 #include <optional>
 #include <Interpreters/Context_fwd.h>
+#include <Core/ColumnWithTypeAndName.h>
 
 
 namespace DB
@@ -11,6 +12,9 @@ namespace DB
 class IFunctionOverloadResolver;
 using FunctionOverloadResolverPtr = std::shared_ptr<IFunctionOverloadResolver>;
 
+class IFunctionBase;
+using FunctionBasePtr = std::shared_ptr<const IFunctionBase>;
+
 enum class CastType : uint8_t
 {
     nonAccurate,
@@ -24,6 +28,6 @@ struct CastDiagnostic
     std::string column_to;
 };
 
-FunctionOverloadResolverPtr createInternalCastOverloadResolver(CastType type, std::optional<CastDiagnostic> diagnostic);
+FunctionBasePtr createInternalCast(ColumnWithTypeAndName from, DataTypePtr to, CastType cast_type, std::optional<CastDiagnostic> diagnostic);
 
 }

src/Functions/geometryConverters.h

@@ -31,6 +31,9 @@ namespace ErrorCodes
 template <typename Point>
 using LineString = boost::geometry::model::linestring<Point>;
 
+template <typename Point>
+using MultiLineString = boost::geometry::model::multi_linestring<LineString<Point>>;
+
 template <typename Point>
 using Ring = boost::geometry::model::ring<Point>;
 
@@ -42,12 +45,14 @@ using MultiPolygon = boost::geometry::model::multi_polygon<Polygon<Point>>;
 
 using CartesianPoint = boost::geometry::model::d2::point_xy<Float64>;
 using CartesianLineString = LineString<CartesianPoint>;
+using CartesianMultiLineString = MultiLineString<CartesianPoint>;
 using CartesianRing = Ring<CartesianPoint>;
 using CartesianPolygon = Polygon<CartesianPoint>;
 using CartesianMultiPolygon = MultiPolygon<CartesianPoint>;
 
 using SphericalPoint = boost::geometry::model::point<Float64, 2, boost::geometry::cs::spherical_equatorial<boost::geometry::degree>>;
 using SphericalLineString = LineString<SphericalPoint>;
+using SphericalMultiLineString = MultiLineString<SphericalPoint>;
 using SphericalRing = Ring<SphericalPoint>;
 using SphericalPolygon = Polygon<SphericalPoint>;
 using SphericalMultiPolygon = MultiPolygon<SphericalPoint>;
@@ -113,6 +118,28 @@ struct ColumnToLineStringsConverter
     }
 };
 
+/**
+ * Class which converts Column with type Array(Array(Tuple(Float64, Float64))) to a vector of boost multi_linestring type.
+*/
+template <typename Point>
+struct ColumnToMultiLineStringsConverter
+{
+    static std::vector<MultiLineString<Point>> convert(ColumnPtr col)
+    {
+        const IColumn::Offsets & offsets = typeid_cast<const ColumnArray &>(*col).getOffsets();
+        size_t prev_offset = 0;
+        std::vector<MultiLineString<Point>> answer(offsets.size());
+        auto all_linestrings = ColumnToLineStringsConverter<Point>::convert(typeid_cast<const ColumnArray &>(*col).getDataPtr());
+        for (size_t iter = 0; iter < offsets.size() && iter < all_linestrings.size(); ++iter)
+        {
+            for (size_t linestring_iter = prev_offset; linestring_iter < offsets[iter]; ++linestring_iter)
+                answer[iter].emplace_back(std::move(all_linestrings[linestring_iter]));
+            prev_offset = offsets[iter];
+        }
+        return answer;
+    }
+};
+
 /**
  * Class which converts Column with type Array(Tuple(Float64, Float64)) to a vector of boost ring type.
 */
@@ -268,6 +295,38 @@ private:
     ColumnUInt64::MutablePtr offsets;
 };
 
+/// Serialize Point, MultiLineString as MultiLineString
+template <typename Point>
+class MultiLineStringSerializer
+{
+public:
+    MultiLineStringSerializer()
+        : offsets(ColumnUInt64::create())
+    {}
+
+    explicit MultiLineStringSerializer(size_t n)
+        : offsets(ColumnUInt64::create(n))
+    {}
+
+    void add(const MultiLineString<Point> & multilinestring)
+    {
+        size += multilinestring.size();
+        offsets->insertValue(size);
+        for (const auto & linestring : multilinestring)
+            linestring_serializer.add(linestring);
+    }
+
+    ColumnPtr finalize()
+    {
+        return ColumnArray::create(linestring_serializer.finalize(), std::move(offsets));
+    }
+
+private:
+    size_t size = 0;
+    LineStringSerializer<Point> linestring_serializer;
+    ColumnUInt64::MutablePtr offsets;
+};
+
 /// Almost the same as LineStringSerializer
 /// Serialize Point, Ring as Ring
 template <typename Point>
@@ -411,6 +470,11 @@ static void callOnGeometryDataType(DataTypePtr type, F && f)
     else if (factory.get("LineString")->equals(*type) && type->getCustomName() && type->getCustomName()->getName() == "LineString")
         return f(ConverterType<ColumnToLineStringsConverter<Point>>());
 
+    /// We should take the name into consideration to avoid ambiguity.
+    /// Because for example both MultiLineString and Polygon are resolved to Array(Array(Point)).
+    else if (factory.get("MultiLineString")->equals(*type) && type->getCustomName() && type->getCustomName()->getName() == "MultiLineString")
+        return f(ConverterType<ColumnToMultiLineStringsConverter<Point>>());
+
     /// For backward compatibility if we call this function not on a custom type, we will consider Array(Tuple(Point)) as type Ring.
     else if (factory.get("Ring")->equals(*type))
         return f(ConverterType<ColumnToRingsConverter<Point>>());

src/Functions/polygonsIntersection.cpp

@@ -75,6 +75,8 @@ public:
                 throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Any argument of function {} must not be Point", getName());
             else if constexpr (std::is_same_v<ColumnToLineStringsConverter<Point>, LeftConverter> || std::is_same_v<ColumnToLineStringsConverter<Point>, RightConverter>)
                 throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Any argument of function {} must not be LineString", getName());
+            else if constexpr (std::is_same_v<ColumnToMultiLineStringsConverter<Point>, LeftConverter> || std::is_same_v<ColumnToMultiLineStringsConverter<Point>, RightConverter>)
+                throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Any argument of function {} must not be MultiLineString", getName());
             else
             {
                 auto first = LeftConverter::convert(arguments[0].column->convertToFullColumnIfConst());

src/Functions/polygonsSymDifference.cpp

@@ -73,6 +73,8 @@ public:
                 throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Any argument of function {} must not be Point", getName());
             else if constexpr (std::is_same_v<ColumnToLineStringsConverter<Point>, LeftConverter> || std::is_same_v<ColumnToLineStringsConverter<Point>, RightConverter>)
                 throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Any argument of function {} must not be LineString", getName());
+            else if constexpr (std::is_same_v<ColumnToMultiLineStringsConverter<Point>, LeftConverter> || std::is_same_v<ColumnToMultiLineStringsConverter<Point>, RightConverter>)
+                throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Any argument of function {} must not be MultiLineString", getName());
             else
             {
                 auto first = LeftConverter::convert(arguments[0].column->convertToFullColumnIfConst());

src/Functions/polygonsUnion.cpp

@@ -73,6 +73,8 @@ public:
                 throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Any argument of function {} must not be Point", getName());
             else if constexpr (std::is_same_v<ColumnToLineStringsConverter<Point>, LeftConverter> || std::is_same_v<ColumnToLineStringsConverter<Point>, RightConverter>)
                 throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Any argument of function {} must not be LineString", getName());
+            else if constexpr (std::is_same_v<ColumnToMultiLineStringsConverter<Point>, LeftConverter> || std::is_same_v<ColumnToMultiLineStringsConverter<Point>, RightConverter>)
+                throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Any argument of function {} must not be MultiLineString", getName());
             else
             {
                 auto first = LeftConverter::convert(arguments[0].column->convertToFullColumnIfConst());

src/Functions/polygonsWithin.cpp

@@ -77,6 +77,8 @@ public:
                 throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Any argument of function {} must not be Point", getName());
             else if constexpr (std::is_same_v<ColumnToLineStringsConverter<Point>, LeftConverter> || std::is_same_v<ColumnToLineStringsConverter<Point>, RightConverter>)
                 throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Any argument of function {} must not be LineString", getName());
+            else if constexpr (std::is_same_v<ColumnToMultiLineStringsConverter<Point>, LeftConverter> || std::is_same_v<ColumnToMultiLineStringsConverter<Point>, RightConverter>)
+                throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Any argument of function {} must not be MultiLineString", getName());
             else
             {
                 auto first = LeftConverter::convert(arguments[0].column->convertToFullColumnIfConst());

src/Functions/readWkt.cpp

@@ -87,6 +87,11 @@ struct ReadWKTLineStringNameHolder
     static constexpr const char * name = "readWKTLineString";
 };
 
+struct ReadWKTMultiLineStringNameHolder
+{
+    static constexpr const char * name = "readWKTMultiLineString";
+};
+
 struct ReadWKTRingNameHolder
 {
     static constexpr const char * name = "readWKTRing";
@@ -131,6 +136,31 @@ Parses a Well-Known Text (WKT) representation of a LineString geometry and retur
         },
         .categories{"Unique identifiers"}
     });
+    factory.registerFunction<FunctionReadWKT<DataTypeMultiLineStringName, CartesianMultiLineString, MultiLineStringSerializer<CartesianPoint>, ReadWKTMultiLineStringNameHolder>>(FunctionDocumentation
+    {
+        .description=R"(
+Parses a Well-Known Text (WKT) representation of a MultiLineString geometry and returns it in the internal ClickHouse format.
+)",
+        .syntax = "readWKTMultiLineString(wkt_string)",
+        .arguments{
+            {"wkt_string", "The input WKT string representing a MultiLineString geometry."}
+        },
+        .returned_value = "The function returns a ClickHouse internal representation of the multilinestring geometry.",
+        .examples{
+            {"first call", "SELECT readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3), (4 4, 5 5, 6 6))');", R"(
+┌─readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3), (4 4, 5 5, 6 6))')─┐
+│ [[(1,1),(2,2),(3,3)],[(4,4),(5,5),(6,6)]]                                    │
+└──────────────────────────────────────────────────────────────────────────────┘
+
+            )"},
+            {"second call", "SELECT toTypeName(readWKTLineString('MULTILINESTRING ((1 1, 2 2, 3 3, 1 1))'));", R"(
+┌─toTypeName(readWKTLineString('MULTILINESTRING ((1 1, 2 2, 3 3, 1 1))'))─┐
+│ MultiLineString                                                         │
+└─────────────────────────────────────────────────────────────────────────┘
+            )"},
+        },
+        .categories{"Unique identifiers"}
+    });
     factory.registerFunction<FunctionReadWKT<DataTypeRingName, CartesianRing, RingSerializer<CartesianPoint>, ReadWKTRingNameHolder>>();
     factory.registerFunction<FunctionReadWKT<DataTypePolygonName, CartesianPolygon, PolygonSerializer<CartesianPoint>, ReadWKTPolygonNameHolder>>();
     factory.registerFunction<FunctionReadWKT<DataTypeMultiPolygonName, CartesianMultiPolygon, MultiPolygonSerializer<CartesianPoint>, ReadWKTMultiPolygonNameHolder>>();

src/Functions/toBool.cpp

@@ -54,8 +54,7 @@ namespace
                 }
             };
 
-            FunctionOverloadResolverPtr func_builder_cast = createInternalCastOverloadResolver(CastType::nonAccurate, {});
-            auto func_cast = func_builder_cast->build(cast_args);
+            auto func_cast = createInternalCast(arguments[0], result_type, CastType::nonAccurate, {});
             return func_cast->execute(cast_args, result_type, arguments[0].column->size());
         }
     };

src/Interpreters/ActionsDAG.cpp

@@ -301,11 +301,11 @@ const ActionsDAG::Node & ActionsDAG::addCast(const Node & node_to_cast, const Da
     column.column = DataTypeString().createColumnConst(0, cast_type_constant_value);
     column.type = std::make_shared<DataTypeString>();
 
-    const auto * cast_type_constant_node = &addColumn(std::move(column));
+    const auto * cast_type_constant_node = &addColumn(column);
     ActionsDAG::NodeRawConstPtrs children = {&node_to_cast, cast_type_constant_node};
-    FunctionOverloadResolverPtr func_builder_cast = createInternalCastOverloadResolver(CastType::nonAccurate, {});
+    auto func_base_cast = createInternalCast(ColumnWithTypeAndName{node_to_cast.result_type, node_to_cast.result_name}, cast_type, CastType::nonAccurate, {});
 
-    return addFunction(func_builder_cast, std::move(children), result_name);
+    return addFunction(func_base_cast, std::move(children), result_name);
 }
 
 const ActionsDAG::Node & ActionsDAG::addFunctionImpl(
@@ -1547,11 +1547,11 @@ ActionsDAG ActionsDAG::makeConvertingActions(
             const auto * left_arg = dst_node;
 
             CastDiagnostic diagnostic = {dst_node->result_name, res_elem.name};
-            FunctionOverloadResolverPtr func_builder_cast
-                = createInternalCastOverloadResolver(CastType::nonAccurate, std::move(diagnostic));
+            ColumnWithTypeAndName left_column{nullptr, dst_node->result_type, {}};
+            auto func_base_cast = createInternalCast(std::move(left_column), res_elem.type, CastType::nonAccurate, std::move(diagnostic));
 
             NodeRawConstPtrs children = { left_arg, right_arg };
-            dst_node = &actions_dag.addFunction(func_builder_cast, std::move(children), {});
+            dst_node = &actions_dag.addFunction(func_base_cast, std::move(children), {});
         }
 
         if (dst_node->column && isColumnConst(*dst_node->column) && !(res_elem.column && isColumnConst(*res_elem.column)))

src/Interpreters/HashJoin/FullHashJoin.cpp

@@ -1,4 +1,4 @@
-#include <Interpreters/HashJoin/HashJoinMethods.h>
+#include <Interpreters/HashJoin/HashJoinMethodsImpl.h>
 
 namespace DB
 {

src/Interpreters/HashJoin/HashJoin.cpp

@@ -291,12 +291,13 @@ void HashJoin::dataMapInit(MapsVariant & map)
 {
     if (kind == JoinKind::Cross)
         return;
-    joinDispatchInit(kind, strictness, map);
-    joinDispatch(kind, strictness, map, [&](auto, auto, auto & map_) { map_.create(data->type); });
+    auto prefer_use_maps_all = table_join->getMixedJoinExpression() != nullptr;
+    joinDispatchInit(kind, strictness, map, prefer_use_maps_all);
+    joinDispatch(kind, strictness, map, prefer_use_maps_all, [&](auto, auto, auto & map_) { map_.create(data->type); });
 
     if (reserve_num)
     {
-        joinDispatch(kind, strictness, map, [&](auto, auto, auto & map_) { map_.reserve(data->type, reserve_num); });
+        joinDispatch(kind, strictness, map, prefer_use_maps_all, [&](auto, auto, auto & map_) { map_.reserve(data->type, reserve_num); });
     }
 
     if (!data)
@@ -327,9 +328,10 @@ size_t HashJoin::getTotalRowCount() const
     }
     else
     {
+        auto prefer_use_maps_all = table_join->getMixedJoinExpression() != nullptr;
         for (const auto & map : data->maps)
         {
-            joinDispatch(kind, strictness, map, [&](auto, auto, auto & map_) { res += map_.getTotalRowCount(data->type); });
+            joinDispatch(kind, strictness, map, prefer_use_maps_all, [&](auto, auto, auto & map_) { res += map_.getTotalRowCount(data->type); });
         }
     }
 
@@ -367,9 +369,10 @@ size_t HashJoin::getTotalByteCount() const
 
     if (data->type != Type::CROSS)
     {
+        auto prefer_use_maps_all = table_join->getMixedJoinExpression() != nullptr;
         for (const auto & map : data->maps)
         {
-            joinDispatch(kind, strictness, map, [&](auto, auto, auto & map_) { res += map_.getTotalByteCountImpl(data->type); });
+            joinDispatch(kind, strictness, map, prefer_use_maps_all, [&](auto, auto, auto & map_) { res += map_.getTotalByteCountImpl(data->type); });
         }
     }
     return res;
@@ -520,6 +523,8 @@ bool HashJoin::addBlockToJoin(const Block & source_block_, bool check_limits)
         return true;
     }
 
+    bool prefer_use_maps_all = table_join->getMixedJoinExpression() != nullptr;
+
     size_t total_rows = 0;
     size_t total_bytes = 0;
     {
@@ -592,7 +597,7 @@ bool HashJoin::addBlockToJoin(const Block & source_block_, bool check_limits)
             bool is_inserted = false;
             if (kind != JoinKind::Cross)
             {
-                joinDispatch(kind, strictness, data->maps[onexpr_idx], [&](auto kind_, auto strictness_, auto & map)
+                joinDispatch(kind, strictness, data->maps[onexpr_idx], prefer_use_maps_all, [&](auto kind_, auto strictness_, auto & map)
                 {
                     size_t size = HashJoinMethods<kind_, strictness_, std::decay_t<decltype(map)>>::insertFromBlockImpl(
                             *this,
@@ -608,10 +613,10 @@ bool HashJoin::addBlockToJoin(const Block & source_block_, bool check_limits)
                             is_inserted);
 
                     if (flag_per_row)
-                        used_flags->reinit<kind_, strictness_>(stored_block);
+                        used_flags->reinit<kind_, strictness_, std::is_same_v<std::decay_t<decltype(map)>, MapsAll>>(stored_block);
                     else if (is_inserted)
                         /// Number of buckets + 1 value from zero storage
-                        used_flags->reinit<kind_, strictness_>(size + 1);
+                        used_flags->reinit<kind_, strictness_, std::is_same_v<std::decay_t<decltype(map)>, MapsAll>>(size + 1);
                 });
             }
 
@@ -873,7 +878,7 @@ ColumnWithTypeAndName HashJoin::joinGet(const Block & block, const Block & block
         keys.insert(std::move(key));
     }
 
-    static_assert(!MapGetter<JoinKind::Left, JoinStrictness::Any>::flagged,
+    static_assert(!MapGetter<JoinKind::Left, JoinStrictness::Any, false>::flagged,
                   "joinGet are not protected from hash table changes between block processing");
 
     std::vector<const MapsOne *> maps_vector;
@@ -914,16 +919,34 @@ void HashJoin::joinBlock(Block & block, ExtraBlockPtr & not_processed)
         materializeBlockInplace(block);
     }
 
+    bool prefer_use_maps_all = table_join->getMixedJoinExpression() != nullptr;
     {
         std::vector<const std::decay_t<decltype(data->maps[0])> * > maps_vector;
         for (size_t i = 0; i < table_join->getClauses().size(); ++i)
             maps_vector.push_back(&data->maps[i]);
 
-        if (joinDispatch(kind, strictness, maps_vector, [&](auto kind_, auto strictness_, auto & maps_vector_)
+        if (joinDispatch(kind, strictness, maps_vector, prefer_use_maps_all, [&](auto kind_, auto strictness_, auto & maps_vector_)
         {
-            using MapType = typename MapGetter<kind_, strictness_>::Map;
-            Block remaining_block = HashJoinMethods<kind_, strictness_, MapType>::joinBlockImpl(
-                *this, block, sample_block_with_columns_to_add, maps_vector_);
+            Block remaining_block;
+            if constexpr (std::is_same_v<std::decay_t<decltype(maps_vector_)>, std::vector<const MapsAll *>>)
+            {
+                remaining_block = HashJoinMethods<kind_, strictness_, MapsAll>::joinBlockImpl(
+                    *this, block, sample_block_with_columns_to_add, maps_vector_);
+            }
+            else if constexpr (std::is_same_v<std::decay_t<decltype(maps_vector_)>, std::vector<const MapsOne *>>)
+            {
+                remaining_block = HashJoinMethods<kind_, strictness_, MapsOne>::joinBlockImpl(
+                    *this, block, sample_block_with_columns_to_add, maps_vector_);
+            }
+            else if constexpr (std::is_same_v<std::decay_t<decltype(maps_vector_)>, std::vector<const MapsAsof *>>)
+            {
+                remaining_block = HashJoinMethods<kind_, strictness_, MapsAsof>::joinBlockImpl(
+                    *this, block, sample_block_with_columns_to_add, maps_vector_);
+            }
+            else
+            {
+                throw Exception(ErrorCodes::LOGICAL_ERROR, "Unknown maps type");
+            }
             if (remaining_block.rows())
                 not_processed = std::make_shared<ExtraBlock>(ExtraBlock{std::move(remaining_block)});
             else
@@ -1023,7 +1046,8 @@ public:
                 rows_added = fillColumnsFromMap(map, columns_right);
             };
 
-            if (!joinDispatch(parent.kind, parent.strictness, parent.data->maps.front(), fill_callback))
+            bool prefer_use_maps_all = parent.table_join->getMixedJoinExpression() != nullptr;
+            if (!joinDispatch(parent.kind, parent.strictness, parent.data->maps.front(), prefer_use_maps_all, fill_callback))
                 throw Exception(ErrorCodes::LOGICAL_ERROR, "Unknown JOIN strictness '{}' (must be on of: ANY, ALL, ASOF)", parent.strictness);
         }
 
@@ -1220,11 +1244,12 @@ void HashJoin::reuseJoinedData(const HashJoin & join)
     if (flag_per_row)
         throw Exception(ErrorCodes::NOT_IMPLEMENTED, "StorageJoin with ORs is not supported");
 
+    bool prefer_use_maps_all = join.table_join->getMixedJoinExpression() != nullptr;
     for (auto & map : data->maps)
     {
-        joinDispatch(kind, strictness, map, [this](auto kind_, auto strictness_, auto & map_)
+        joinDispatch(kind, strictness, map, prefer_use_maps_all, [this](auto kind_, auto strictness_, auto & map_)
         {
-            used_flags->reinit<kind_, strictness_>(map_.getBufferSizeInCells(data->type) + 1);
+            used_flags->reinit<kind_, strictness_, std::is_same_v<std::decay_t<decltype(map_)>, MapsAll>>(map_.getBufferSizeInCells(data->type) + 1);
         });
     }
 }
@@ -1304,7 +1329,9 @@ void HashJoin::validateAdditionalFilterExpression(ExpressionActionsPtr additiona
             additional_filter_expression->dumpActions());
     }
 
-    bool is_supported = (strictness == JoinStrictness::All) && (isInnerOrLeft(kind) || isRightOrFull(kind));
+    bool is_supported = ((strictness == JoinStrictness::All) && (isInnerOrLeft(kind) || isRightOrFull(kind)))
+        || ((strictness == JoinStrictness::Semi || strictness == JoinStrictness::Any || strictness == JoinStrictness::Anti)
+                && (isLeft(kind) || isRight(kind))) || (strictness == JoinStrictness::Any && (isInner(kind)));
     if (!is_supported)
     {
         throw Exception(ErrorCodes::INVALID_JOIN_ON_EXPRESSION,

src/Interpreters/HashJoin/HashJoinMethods.h

@@ -12,15 +12,8 @@
 #include <Poco/Logger.h>
 #include <Common/logger_useful.h>
 
-
 namespace DB
 {
-namespace ErrorCodes
-{
-    extern const int UNSUPPORTED_JOIN_KEYS;
-    extern const int LOGICAL_ERROR;
-}
-
 /// Inserting an element into a hash table of the form `key -> reference to a string`, which will then be used by JOIN.
 template <typename HashMap, typename KeyGetter>
 struct Inserter
@@ -64,7 +57,6 @@ struct Inserter
     }
 };
 
-
 /// MapsTemplate is one of MapsOne, MapsAll and MapsAsof
 template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
 class HashJoinMethods
@@ -81,27 +73,7 @@ public:
         ConstNullMapPtr null_map,
         UInt8ColumnDataPtr join_mask,
         Arena & pool,
-        bool & is_inserted)
-    {
-        switch (type)
-        {
-            case HashJoin::Type::EMPTY:
-                [[fallthrough]];
-            case HashJoin::Type::CROSS:
-                /// Do nothing. We will only save block, and it is enough
-                is_inserted = true;
-                return 0;
-
-        #define M(TYPE) \
-            case HashJoin::Type::TYPE: \
-                return insertFromBlockImplTypeCase<typename KeyGetterForType<HashJoin::Type::TYPE, std::remove_reference_t<decltype(*maps.TYPE)>>::Type>(\
-                    join, *maps.TYPE, rows, key_columns, key_sizes, stored_block, null_map, join_mask, pool, is_inserted); \
-                break;
-
-            APPLY_FOR_JOIN_VARIANTS(M)
-        #undef M
-        }
-    }
+        bool & is_inserted);
 
     using MapsTemplateVector = std::vector<const MapsTemplate *>;
 
@@ -110,280 +82,36 @@ public:
         Block & block,
         const Block & block_with_columns_to_add,
         const MapsTemplateVector & maps_,
-        bool is_join_get = false)
-    {
-        constexpr JoinFeatures<KIND, STRICTNESS> join_features;
-
-        std::vector<JoinOnKeyColumns> join_on_keys;
-        const auto & onexprs = join.table_join->getClauses();
-        for (size_t i = 0; i < onexprs.size(); ++i)
-        {
-            const auto & key_names = !is_join_get ? onexprs[i].key_names_left : onexprs[i].key_names_right;
-            join_on_keys.emplace_back(block, key_names, onexprs[i].condColumnNames().first, join.key_sizes[i]);
-        }
-        size_t existing_columns = block.columns();
-
-        /** If you use FULL or RIGHT JOIN, then the columns from the "left" table must be materialized.
-          * Because if they are constants, then in the "not joined" rows, they may have different values
-          *  - default values, which can differ from the values of these constants.
-          */
-        if constexpr (join_features.right || join_features.full)
-        {
-            materializeBlockInplace(block);
-        }
-
-        /** For LEFT/INNER JOIN, the saved blocks do not contain keys.
-          * For FULL/RIGHT JOIN, the saved blocks contain keys;
-          *  but they will not be used at this stage of joining (and will be in `AdderNonJoined`), and they need to be skipped.
-          * For ASOF, the last column is used as the ASOF column
-          */
-        AddedColumns<!join_features.is_any_join> added_columns(
-            block,
-            block_with_columns_to_add,
-            join.savedBlockSample(),
-            join,
-            std::move(join_on_keys),
-            join.table_join->getMixedJoinExpression(),
-            join_features.is_asof_join,
-            is_join_get);
-
-        bool has_required_right_keys = (join.required_right_keys.columns() != 0);
-        added_columns.need_filter = join_features.need_filter || has_required_right_keys;
-        added_columns.max_joined_block_rows = join.max_joined_block_rows;
-        if (!added_columns.max_joined_block_rows)
-            added_columns.max_joined_block_rows = std::numeric_limits<size_t>::max();
-        else
-            added_columns.reserve(join_features.need_replication);
-
-        size_t num_joined = switchJoinRightColumns(maps_, added_columns, join.data->type, *join.used_flags);
-        /// Do not hold memory for join_on_keys anymore
-        added_columns.join_on_keys.clear();
-        Block remaining_block = sliceBlock(block, num_joined);
-
-        added_columns.buildOutput();
-        for (size_t i = 0; i < added_columns.size(); ++i)
-            block.insert(added_columns.moveColumn(i));
-
-        std::vector<size_t> right_keys_to_replicate [[maybe_unused]];
-
-        if constexpr (join_features.need_filter)
-        {
-            /// If ANY INNER | RIGHT JOIN - filter all the columns except the new ones.
-            for (size_t i = 0; i < existing_columns; ++i)
-                block.safeGetByPosition(i).column = block.safeGetByPosition(i).column->filter(added_columns.filter, -1);
-
-            /// Add join key columns from right block if needed using value from left table because of equality
-            for (size_t i = 0; i < join.required_right_keys.columns(); ++i)
-            {
-                const auto & right_key = join.required_right_keys.getByPosition(i);
-                /// asof column is already in block.
-                if (join_features.is_asof_join && right_key.name == join.table_join->getOnlyClause().key_names_right.back())
-                    continue;
-
-                const auto & left_column = block.getByName(join.required_right_keys_sources[i]);
-                const auto & right_col_name = join.getTableJoin().renamedRightColumnName(right_key.name);
-                auto right_col = copyLeftKeyColumnToRight(right_key.type, right_col_name, left_column);
-                block.insert(std::move(right_col));
-            }
-        }
-        else if (has_required_right_keys)
-        {
-            /// Add join key columns from right block if needed.
-            for (size_t i = 0; i < join.required_right_keys.columns(); ++i)
-            {
-                const auto & right_key = join.required_right_keys.getByPosition(i);
-                auto right_col_name = join.getTableJoin().renamedRightColumnName(right_key.name);
-                /// asof column is already in block.
-                if (join_features.is_asof_join && right_key.name == join.table_join->getOnlyClause().key_names_right.back())
-                    continue;
-
-                const auto & left_column = block.getByName(join.required_right_keys_sources[i]);
-                auto right_col = copyLeftKeyColumnToRight(right_key.type, right_col_name, left_column, &added_columns.filter);
-                block.insert(std::move(right_col));
-
-                if constexpr (join_features.need_replication)
-                    right_keys_to_replicate.push_back(block.getPositionByName(right_col_name));
-            }
-        }
-
-        if constexpr (join_features.need_replication)
-        {
-            std::unique_ptr<IColumn::Offsets> & offsets_to_replicate = added_columns.offsets_to_replicate;
-
-            /// If ALL ... JOIN - we replicate all the columns except the new ones.
-            for (size_t i = 0; i < existing_columns; ++i)
-            {
-                block.safeGetByPosition(i).column = block.safeGetByPosition(i).column->replicate(*offsets_to_replicate);
-            }
-
-            /// Replicate additional right keys
-            for (size_t pos : right_keys_to_replicate)
-            {
-               block.safeGetByPosition(pos).column = block.safeGetByPosition(pos).column->replicate(*offsets_to_replicate);
-            }
-        }
-
-        return remaining_block;
-    }
-
+        bool is_join_get = false);
 private:
     template <typename KeyGetter, bool is_asof_join>
-    static KeyGetter createKeyGetter(const ColumnRawPtrs & key_columns, const Sizes & key_sizes)
-    {
-        if constexpr (is_asof_join)
-        {
-            auto key_column_copy = key_columns;
-            auto key_size_copy = key_sizes;
-            key_column_copy.pop_back();
-            key_size_copy.pop_back();
-            return KeyGetter(key_column_copy, key_size_copy, nullptr);
-        }
-        else
-            return KeyGetter(key_columns, key_sizes, nullptr);
-    }
+    static KeyGetter createKeyGetter(const ColumnRawPtrs & key_columns, const Sizes & key_sizes);
 
     template <typename KeyGetter, typename HashMap>
-    static size_t NO_INLINE insertFromBlockImplTypeCase(
+    static size_t insertFromBlockImplTypeCase(
         HashJoin & join, HashMap & map, size_t rows, const ColumnRawPtrs & key_columns,
-        const Sizes & key_sizes, Block * stored_block, ConstNullMapPtr null_map, UInt8ColumnDataPtr join_mask, Arena & pool, bool & is_inserted)
-    {
-        [[maybe_unused]] constexpr bool mapped_one = std::is_same_v<typename HashMap::mapped_type, RowRef>;
-        constexpr bool is_asof_join = STRICTNESS == JoinStrictness::Asof;
-
-        const IColumn * asof_column [[maybe_unused]] = nullptr;
-        if constexpr (is_asof_join)
-            asof_column = key_columns.back();
-
-        auto key_getter = createKeyGetter<KeyGetter, is_asof_join>(key_columns, key_sizes);
-
-        /// For ALL and ASOF join always insert values
-        is_inserted = !mapped_one || is_asof_join;
-
-        for (size_t i = 0; i < rows; ++i)
-        {
-            if (null_map && (*null_map)[i])
-            {
-                /// nulls are not inserted into hash table,
-                /// keep them for RIGHT and FULL joins
-                is_inserted = true;
-                continue;
-            }
-
-            /// Check condition for right table from ON section
-            if (join_mask && !(*join_mask)[i])
-                continue;
-
-            if constexpr (is_asof_join)
-                Inserter<HashMap, KeyGetter>::insertAsof(join, map, key_getter, stored_block, i, pool, *asof_column);
-            else if constexpr (mapped_one)
-                is_inserted |= Inserter<HashMap, KeyGetter>::insertOne(join, map, key_getter, stored_block, i, pool);
-            else
-                Inserter<HashMap, KeyGetter>::insertAll(join, map, key_getter, stored_block, i, pool);
-        }
-        return map.getBufferSizeInCells();
-    }
+        const Sizes & key_sizes, Block * stored_block, ConstNullMapPtr null_map, UInt8ColumnDataPtr join_mask, Arena & pool, bool & is_inserted);
 
     template <typename AddedColumns>
     static size_t switchJoinRightColumns(
         const std::vector<const MapsTemplate *> & mapv,
         AddedColumns & added_columns,
         HashJoin::Type type,
-        JoinStuff::JoinUsedFlags & used_flags)
-    {
-        constexpr bool is_asof_join = STRICTNESS == JoinStrictness::Asof;
-        switch (type)
-        {
-            case HashJoin::Type::EMPTY: {
-                if constexpr (!is_asof_join)
-                {
-                    using KeyGetter = KeyGetterEmpty<typename MapsTemplate::MappedType>;
-                    std::vector<KeyGetter> key_getter_vector;
-                    key_getter_vector.emplace_back();
-
-                    using MapTypeVal = typename KeyGetter::MappedType;
-                    std::vector<const MapTypeVal *> a_map_type_vector;
-                    a_map_type_vector.emplace_back();
-                    return joinRightColumnsSwitchNullability<KeyGetter>(
-                        std::move(key_getter_vector), a_map_type_vector, added_columns, used_flags);
-                }
-                throw Exception(ErrorCodes::UNSUPPORTED_JOIN_KEYS, "Unsupported JOIN keys. Type: {}", type);
-            }
-        #define M(TYPE) \
-            case HashJoin::Type::TYPE: \
-            {                                                           \
-                using MapTypeVal = const typename std::remove_reference_t<decltype(MapsTemplate::TYPE)>::element_type; \
-                using KeyGetter = typename KeyGetterForType<HashJoin::Type::TYPE, MapTypeVal>::Type; \
-                std::vector<const MapTypeVal *> a_map_type_vector(mapv.size()); \
-                std::vector<KeyGetter> key_getter_vector;                     \
-                for (size_t d = 0; d < added_columns.join_on_keys.size(); ++d)                   \
-                {       \
-                    const auto & join_on_key = added_columns.join_on_keys[d];     \
-                    a_map_type_vector[d] = mapv[d]->TYPE.get();              \
-                    key_getter_vector.push_back(std::move(createKeyGetter<KeyGetter, is_asof_join>(join_on_key.key_columns, join_on_key.key_sizes))); \
-                }                                                           \
-                return joinRightColumnsSwitchNullability<KeyGetter>( \
-                                std::move(key_getter_vector), a_map_type_vector, added_columns, used_flags); \
-            }
-            APPLY_FOR_JOIN_VARIANTS(M)
-        #undef M
-
-        default:
-            throw Exception(ErrorCodes::UNSUPPORTED_JOIN_KEYS, "Unsupported JOIN keys (type: {})", type);
-        }
-    }
+        JoinStuff::JoinUsedFlags & used_flags);
 
     template <typename KeyGetter, typename Map, typename AddedColumns>
     static size_t joinRightColumnsSwitchNullability(
         std::vector<KeyGetter> && key_getter_vector,
         const std::vector<const Map *> & mapv,
         AddedColumns & added_columns,
-        JoinStuff::JoinUsedFlags & used_flags)
-    {
-        if (added_columns.need_filter)
-        {
-            return joinRightColumnsSwitchMultipleDisjuncts<KeyGetter, Map, true>(
-            std::forward<std::vector<KeyGetter>>(key_getter_vector), mapv, added_columns, used_flags);
-        }
-        else
-        {
-            return joinRightColumnsSwitchMultipleDisjuncts<KeyGetter, Map, false>(
-            std::forward<std::vector<KeyGetter>>(key_getter_vector), mapv, added_columns, used_flags);
-        }
-    }
+        JoinStuff::JoinUsedFlags & used_flags);
 
     template <typename KeyGetter, typename Map, bool need_filter, typename AddedColumns>
     static size_t joinRightColumnsSwitchMultipleDisjuncts(
         std::vector<KeyGetter> && key_getter_vector,
         const std::vector<const Map *> & mapv,
         AddedColumns & added_columns,
-        JoinStuff::JoinUsedFlags & used_flags)
-    {
-        constexpr JoinFeatures<KIND, STRICTNESS> join_features;
-        if constexpr (join_features.is_all_join)
-        {
-        if (added_columns.additional_filter_expression)
-        {
-            bool mark_per_row_used = join_features.right || join_features.full || mapv.size() > 1;
-            return joinRightColumnsWithAddtitionalFilter<KeyGetter, Map, join_features.need_replication>(
-                std::forward<std::vector<KeyGetter>>(key_getter_vector),
-                mapv,
-                added_columns,
-                used_flags,
-                need_filter,
-                join_features.need_flags,
-                join_features.add_missing,
-                mark_per_row_used);
-        }
-        }
-
-        if (added_columns.additional_filter_expression)
-            throw Exception(ErrorCodes::LOGICAL_ERROR, "Additional filter expression is not supported for this JOIN");
-
-        return mapv.size() > 1 ? joinRightColumns<KeyGetter, Map, need_filter, true>(
-                   std::forward<std::vector<KeyGetter>>(key_getter_vector), mapv, added_columns, used_flags)
-                               : joinRightColumns<KeyGetter, Map, need_filter, false>(
-                                   std::forward<std::vector<KeyGetter>>(key_getter_vector), mapv, added_columns, used_flags);
-    }
+        JoinStuff::JoinUsedFlags & used_flags);
 
     /// Joins right table columns which indexes are present in right_indexes using specified map.
     /// Makes filter (1 if row presented in right table) and returns offsets to replicate (for ALL JOINS).
@@ -392,464 +120,30 @@ private:
         std::vector<KeyGetter> && key_getter_vector,
         const std::vector<const Map *> & mapv,
         AddedColumns & added_columns,
-        JoinStuff::JoinUsedFlags & used_flags)
-    {
-        constexpr JoinFeatures<KIND, STRICTNESS> join_features;
-
-        size_t rows = added_columns.rows_to_add;
-        if constexpr (need_filter)
-            added_columns.filter = IColumn::Filter(rows, 0);
-
-        Arena pool;
-
-        if constexpr (join_features.need_replication)
-            added_columns.offsets_to_replicate = std::make_unique<IColumn::Offsets>(rows);
-
-        IColumn::Offset current_offset = 0;
-        size_t max_joined_block_rows = added_columns.max_joined_block_rows;
-        size_t i = 0;
-        for (; i < rows; ++i)
-        {
-            if constexpr (join_features.need_replication)
-            {
-                if (unlikely(current_offset >= max_joined_block_rows))
-                {
-                    added_columns.offsets_to_replicate->resize_assume_reserved(i);
-                    added_columns.filter.resize_assume_reserved(i);
-                    break;
-                }
-            }
-
-            bool right_row_found = false;
-
-            KnownRowsHolder<flag_per_row> known_rows;
-            for (size_t onexpr_idx = 0; onexpr_idx < added_columns.join_on_keys.size(); ++onexpr_idx)
-            {
-                const auto & join_keys = added_columns.join_on_keys[onexpr_idx];
-                if (join_keys.null_map && (*join_keys.null_map)[i])
-                        continue;
-
-                bool row_acceptable = !join_keys.isRowFiltered(i);
-                using FindResult = typename KeyGetter::FindResult;
-                auto find_result = row_acceptable ? key_getter_vector[onexpr_idx].findKey(*(mapv[onexpr_idx]), i, pool) : FindResult();
-
-                if (find_result.isFound())
-                {
-                    right_row_found = true;
-                    auto & mapped = find_result.getMapped();
-                    if constexpr (join_features.is_asof_join)
-                    {
-                        const IColumn & left_asof_key = added_columns.leftAsofKey();
-
-                        auto row_ref = mapped->findAsof(left_asof_key, i);
-                        if (row_ref.block)
-                        {
-                            setUsed<need_filter>(added_columns.filter, i);
-                            if constexpr (flag_per_row)
-                                used_flags.template setUsed<join_features.need_flags, flag_per_row>(row_ref.block, row_ref.row_num, 0);
-                            else
-                                used_flags.template setUsed<join_features.need_flags, flag_per_row>(find_result);
-
-                            added_columns.appendFromBlock(*row_ref.block, row_ref.row_num, join_features.add_missing);
-                        }
-                        else
-                            addNotFoundRow<join_features.add_missing, join_features.need_replication>(added_columns, current_offset);
-                    }
-                    else if constexpr (join_features.is_all_join)
-                    {
-                        setUsed<need_filter>(added_columns.filter, i);
-                        used_flags.template setUsed<join_features.need_flags, flag_per_row>(find_result);
-                        auto used_flags_opt = join_features.need_flags ? &used_flags : nullptr;
-                        addFoundRowAll<Map, join_features.add_missing>(mapped, added_columns, current_offset, known_rows, used_flags_opt);
-                    }
-                    else if constexpr ((join_features.is_any_join || join_features.is_semi_join) && join_features.right)
-                    {
-                        /// Use first appeared left key + it needs left columns replication
-                        bool used_once = used_flags.template setUsedOnce<join_features.need_flags, flag_per_row>(find_result);
-                        if (used_once)
-                        {
-                            auto used_flags_opt = join_features.need_flags ? &used_flags : nullptr;
-                            setUsed<need_filter>(added_columns.filter, i);
-                            addFoundRowAll<Map, join_features.add_missing>(
-                                mapped, added_columns, current_offset, known_rows, used_flags_opt);
-                        }
-                    }
-                    else if constexpr (join_features.is_any_join && KIND == JoinKind::Inner)
-                    {
-                        bool used_once = used_flags.template setUsedOnce<join_features.need_flags, flag_per_row>(find_result);
-
-                        /// Use first appeared left key only
-                        if (used_once)
-                        {
-                            setUsed<need_filter>(added_columns.filter, i);
-                            added_columns.appendFromBlock(*mapped.block, mapped.row_num, join_features.add_missing);
-                        }
-
-                        break;
-                    }
-                    else if constexpr (join_features.is_any_join && join_features.full)
-                    {
-                        /// TODO
-                    }
-                    else if constexpr (join_features.is_anti_join)
-                    {
-                        if constexpr (join_features.right && join_features.need_flags)
-                            used_flags.template setUsed<join_features.need_flags, flag_per_row>(find_result);
-                    }
-                    else /// ANY LEFT, SEMI LEFT, old ANY (RightAny)
-                    {
-                        setUsed<need_filter>(added_columns.filter, i);
-                        used_flags.template setUsed<join_features.need_flags, flag_per_row>(find_result);
-                        added_columns.appendFromBlock(*mapped.block, mapped.row_num, join_features.add_missing);
-
-                        if (join_features.is_any_or_semi_join)
-                        {
-                            break;
-                        }
-                    }
-                }
-            }
-
-            if (!right_row_found)
-            {
-                if constexpr (join_features.is_anti_join && join_features.left)
-                    setUsed<need_filter>(added_columns.filter, i);
-                addNotFoundRow<join_features.add_missing, join_features.need_replication>(added_columns, current_offset);
-            }
-
-            if constexpr (join_features.need_replication)
-            {
-                (*added_columns.offsets_to_replicate)[i] = current_offset;
-            }
-        }
-
-        added_columns.applyLazyDefaults();
-        return i;
-    }
+        JoinStuff::JoinUsedFlags & used_flags);
 
     template <bool need_filter>
-    static void setUsed(IColumn::Filter & filter [[maybe_unused]], size_t pos [[maybe_unused]])
-    {
-        if constexpr (need_filter)
-            filter[pos] = 1;
-    }
+    static void setUsed(IColumn::Filter & filter [[maybe_unused]], size_t pos [[maybe_unused]]);
 
     template <typename AddedColumns>
     static ColumnPtr buildAdditionalFilter(
         size_t left_start_row,
         const std::vector<RowRef> & selected_rows,
         const std::vector<size_t> & row_replicate_offset,
-        AddedColumns & added_columns)
-    {
-        ColumnPtr result_column;
-        do
-        {
-            if (selected_rows.empty())
-            {
-                result_column = ColumnUInt8::create();
-                break;
-            }
-            const Block & sample_right_block = *selected_rows.begin()->block;
-            if (!sample_right_block || !added_columns.additional_filter_expression)
-            {
-                auto filter = ColumnUInt8::create();
-                filter->insertMany(1, selected_rows.size());
-                result_column = std::move(filter);
-                break;
-            }
-
-            auto required_cols = added_columns.additional_filter_expression->getRequiredColumnsWithTypes();
-            if (required_cols.empty())
-            {
-                Block block;
-                added_columns.additional_filter_expression->execute(block);
-                result_column = block.getByPosition(0).column->cloneResized(selected_rows.size());
-                break;
-            }
-            NameSet required_column_names;
-            for (auto & col : required_cols)
-                required_column_names.insert(col.name);
-
-            Block executed_block;
-            size_t right_col_pos = 0;
-            for (const auto & col : sample_right_block.getColumnsWithTypeAndName())
-            {
-                if (required_column_names.contains(col.name))
-                {
-                    auto new_col = col.column->cloneEmpty();
-                    for (const auto & selected_row : selected_rows)
-                    {
-                        const auto & src_col = selected_row.block->getByPosition(right_col_pos);
-                        new_col->insertFrom(*src_col.column, selected_row.row_num);
-                    }
-                    executed_block.insert({std::move(new_col), col.type, col.name});
-                }
-                right_col_pos += 1;
-            }
-            if (!executed_block)
-            {
-                result_column = ColumnUInt8::create();
-                break;
-            }
-
-            for (const auto & col_name : required_column_names)
-            {
-                const auto * src_col = added_columns.left_block.findByName(col_name);
-                if (!src_col)
-                    continue;
-                auto new_col = src_col->column->cloneEmpty();
-                size_t prev_left_offset = 0;
-                for (size_t i = 1; i < row_replicate_offset.size(); ++i)
-                {
-                    const size_t & left_offset = row_replicate_offset[i];
-                    size_t rows = left_offset - prev_left_offset;
-                    if (rows)
-                        new_col->insertManyFrom(*src_col->column, left_start_row + i - 1, rows);
-                    prev_left_offset = left_offset;
-                }
-                executed_block.insert({std::move(new_col), src_col->type, col_name});
-            }
-            if (!executed_block)
-            {
-                throw Exception(
-                    ErrorCodes::LOGICAL_ERROR,
-                    "required columns: [{}], but not found any in left/right table. right table: {}, left table: {}",
-                    required_cols.toString(),
-                    sample_right_block.dumpNames(),
-                    added_columns.left_block.dumpNames());
-            }
-
-            for (const auto & col : executed_block.getColumnsWithTypeAndName())
-                if (!col.column || !col.type)
-                    throw Exception(ErrorCodes::LOGICAL_ERROR, "Illegal nullptr column in input block: {}", executed_block.dumpStructure());
-
-            added_columns.additional_filter_expression->execute(executed_block);
-            result_column = executed_block.getByPosition(0).column->convertToFullColumnIfConst();
-            executed_block.clear();
-        } while (false);
-
-        result_column = result_column->convertToFullIfNeeded();
-        if (result_column->isNullable())
-        {
-            /// Convert Nullable(UInt8) to UInt8 ensuring that nulls are zeros
-            /// Trying to avoid copying data, since we are the only owner of the column.
-            ColumnPtr mask_column = assert_cast<const ColumnNullable &>(*result_column).getNullMapColumnPtr();
-
-            MutableColumnPtr mutable_column;
-            {
-                ColumnPtr nested_column = assert_cast<const ColumnNullable &>(*result_column).getNestedColumnPtr();
-                result_column.reset();
-                mutable_column = IColumn::mutate(std::move(nested_column));
-            }
-
-            auto & column_data = assert_cast<ColumnUInt8 &>(*mutable_column).getData();
-            const auto & mask_column_data = assert_cast<const ColumnUInt8 &>(*mask_column).getData();
-            for (size_t i = 0; i < column_data.size(); ++i)
-            {
-                if (mask_column_data[i])
-                    column_data[i] = 0;
-            }
-            return mutable_column;
-        }
-        return result_column;
-    }
+        AddedColumns & added_columns);
 
     /// First to collect all matched rows refs by join keys, then filter out rows which are not true in additional filter expression.
-    template <typename KeyGetter, typename Map, bool need_replication, typename AddedColumns>
+    template <typename KeyGetter, typename Map, typename AddedColumns>
     static size_t joinRightColumnsWithAddtitionalFilter(
         std::vector<KeyGetter> && key_getter_vector,
         const std::vector<const Map *> & mapv,
         AddedColumns & added_columns,
         JoinStuff::JoinUsedFlags & used_flags [[maybe_unused]],
         bool need_filter [[maybe_unused]],
-        bool need_flags [[maybe_unused]],
-        bool add_missing [[maybe_unused]],
-        bool flag_per_row [[maybe_unused]])
-    {
-        size_t left_block_rows = added_columns.rows_to_add;
-        if (need_filter)
-            added_columns.filter = IColumn::Filter(left_block_rows, 0);
-
-        std::unique_ptr<Arena> pool;
-
-        if constexpr (need_replication)
-            added_columns.offsets_to_replicate = std::make_unique<IColumn::Offsets>(left_block_rows);
-
-        std::vector<size_t> row_replicate_offset;
-        row_replicate_offset.reserve(left_block_rows);
-
-        using FindResult = typename KeyGetter::FindResult;
-        size_t max_joined_block_rows = added_columns.max_joined_block_rows;
-        size_t left_row_iter = 0;
-        PreSelectedRows selected_rows;
-        selected_rows.reserve(left_block_rows);
-        std::vector<FindResult> find_results;
-        find_results.reserve(left_block_rows);
-        bool exceeded_max_block_rows = false;
-        IColumn::Offset total_added_rows = 0;
-        IColumn::Offset current_added_rows = 0;
-
-        auto collect_keys_matched_rows_refs = [&]()
-        {
-            pool = std::make_unique<Arena>();
-            find_results.clear();
-            row_replicate_offset.clear();
-            row_replicate_offset.push_back(0);
-            current_added_rows = 0;
-            selected_rows.clear();
-            for (; left_row_iter < left_block_rows; ++left_row_iter)
-            {
-                if constexpr (need_replication)
-                {
-                    if (unlikely(total_added_rows + current_added_rows >= max_joined_block_rows))
-                    {
-                        break;
-                    }
-                }
-                KnownRowsHolder<true> all_flag_known_rows;
-                KnownRowsHolder<false> single_flag_know_rows;
-                for (size_t join_clause_idx = 0; join_clause_idx < added_columns.join_on_keys.size(); ++join_clause_idx)
-                {
-                    const auto & join_keys = added_columns.join_on_keys[join_clause_idx];
-                    if (join_keys.null_map && (*join_keys.null_map)[left_row_iter])
-                        continue;
-
-                    bool row_acceptable = !join_keys.isRowFiltered(left_row_iter);
-                    auto find_result = row_acceptable
-                        ? key_getter_vector[join_clause_idx].findKey(*(mapv[join_clause_idx]), left_row_iter, *pool)
-                        : FindResult();
-
-                    if (find_result.isFound())
-                    {
-                        auto & mapped = find_result.getMapped();
-                        find_results.push_back(find_result);
-                        if (flag_per_row)
-                            addFoundRowAll<Map, false, true>(mapped, selected_rows, current_added_rows, all_flag_known_rows, nullptr);
-                        else
-                            addFoundRowAll<Map, false, false>(mapped, selected_rows, current_added_rows, single_flag_know_rows, nullptr);
-                    }
-                }
-                row_replicate_offset.push_back(current_added_rows);
-            }
-        };
-
-        auto copy_final_matched_rows = [&](size_t left_start_row, ColumnPtr filter_col)
-        {
-            const PaddedPODArray<UInt8> & filter_flags = assert_cast<const ColumnUInt8 &>(*filter_col).getData();
-
-            size_t prev_replicated_row = 0;
-            auto selected_right_row_it = selected_rows.begin();
-            size_t find_result_index = 0;
-            for (size_t i = 1, n = row_replicate_offset.size(); i < n; ++i)
-            {
-                bool any_matched = false;
-                /// For all right join, flag_per_row is true, we need mark used flags for each row.
-                if (flag_per_row)
-                {
-                    for (size_t replicated_row = prev_replicated_row; replicated_row < row_replicate_offset[i]; ++replicated_row)
-                    {
-                        if (filter_flags[replicated_row])
-                        {
-                            any_matched = true;
-                            added_columns.appendFromBlock(*selected_right_row_it->block, selected_right_row_it->row_num, add_missing);
-                            total_added_rows += 1;
-                            if (need_flags)
-                                used_flags.template setUsed<true, true>(selected_right_row_it->block, selected_right_row_it->row_num, 0);
-                        }
-                        ++selected_right_row_it;
-                    }
-                }
-                else
-                {
-                    for (size_t replicated_row = prev_replicated_row; replicated_row < row_replicate_offset[i]; ++replicated_row)
-                    {
-                        if (filter_flags[replicated_row])
-                        {
-                            any_matched = true;
-                            added_columns.appendFromBlock(*selected_right_row_it->block, selected_right_row_it->row_num, add_missing);
-                            total_added_rows += 1;
-                        }
-                        ++selected_right_row_it;
-                    }
-                }
-                if (!any_matched)
-                {
-                    if (add_missing)
-                        addNotFoundRow<true, need_replication>(added_columns, total_added_rows);
-                    else
-                        addNotFoundRow<false, need_replication>(added_columns, total_added_rows);
-                }
-                else
-                {
-                    if (!flag_per_row && need_flags)
-                        used_flags.template setUsed<true, false>(find_results[find_result_index]);
-                    if (need_filter)
-                        setUsed<true>(added_columns.filter, left_start_row + i - 1);
-                    if (add_missing)
-                        added_columns.applyLazyDefaults();
-                }
-                find_result_index += (prev_replicated_row != row_replicate_offset[i]);
-
-                if constexpr (need_replication)
-                {
-                    (*added_columns.offsets_to_replicate)[left_start_row + i - 1] = total_added_rows;
-                }
-                prev_replicated_row = row_replicate_offset[i];
-            }
-        };
-
-        while (left_row_iter < left_block_rows && !exceeded_max_block_rows)
-        {
-            auto left_start_row = left_row_iter;
-            collect_keys_matched_rows_refs();
-            if (selected_rows.size() != current_added_rows || row_replicate_offset.size() != left_row_iter - left_start_row + 1)
-            {
-                throw Exception(
-                    ErrorCodes::LOGICAL_ERROR,
-                    "Sizes are mismatched. selected_rows.size:{}, current_added_rows:{}, row_replicate_offset.size:{}, left_row_iter: {}, "
-                    "left_start_row: {}",
-                    selected_rows.size(),
-                    current_added_rows,
-                    row_replicate_offset.size(),
-                    left_row_iter,
-                    left_start_row);
-            }
-            auto filter_col = buildAdditionalFilter(left_start_row, selected_rows, row_replicate_offset, added_columns);
-            copy_final_matched_rows(left_start_row, filter_col);
-
-            if constexpr (need_replication)
-            {
-                // Add a check for current_added_rows to avoid run the filter expression on too small size batch.
-                if (total_added_rows >= max_joined_block_rows || current_added_rows < 1024)
-                    exceeded_max_block_rows = true;
-            }
-        }
-
-        if constexpr (need_replication)
-        {
-            added_columns.offsets_to_replicate->resize_assume_reserved(left_row_iter);
-            added_columns.filter.resize_assume_reserved(left_row_iter);
-        }
-        added_columns.applyLazyDefaults();
-        return left_row_iter;
-    }
+        bool flag_per_row [[maybe_unused]]);
 
     /// Cut first num_rows rows from block in place and returns block with remaining rows
-    static Block sliceBlock(Block & block, size_t num_rows)
-    {
-        size_t total_rows = block.rows();
-        if (num_rows >= total_rows)
-            return {};
-        size_t remaining_rows = total_rows - num_rows;
-        Block remaining_block = block.cloneEmpty();
-        for (size_t i = 0; i < block.columns(); ++i)
-        {
-            auto & col = block.getByPosition(i);
-            remaining_block.getByPosition(i).column = col.column->cut(num_rows, remaining_rows);
-            col.column = col.column->cut(0, num_rows);
-        }
-        return remaining_block;
-    }
+    static Block sliceBlock(Block & block, size_t num_rows);
 
     /** Since we do not store right key columns,
       * this function is used to copy left key columns to right key columns.
@@ -864,70 +158,22 @@ private:
         const DataTypePtr & right_key_type,
         const String & renamed_right_column,
         const ColumnWithTypeAndName & left_column,
-        const IColumn::Filter * null_map_filter = nullptr)
-    {
-        ColumnWithTypeAndName right_column = left_column;
-        right_column.name = renamed_right_column;
+        const IColumn::Filter * null_map_filter = nullptr);
 
-        if (null_map_filter)
-            right_column.column = JoinCommon::filterWithBlanks(right_column.column, *null_map_filter);
+    static void correctNullabilityInplace(ColumnWithTypeAndName & column, bool nullable);
 
-        bool should_be_nullable = isNullableOrLowCardinalityNullable(right_key_type);
-        if (null_map_filter)
-            correctNullabilityInplace(right_column, should_be_nullable, *null_map_filter);
-        else
-            correctNullabilityInplace(right_column, should_be_nullable);
-
-        if (!right_column.type->equals(*right_key_type))
-        {
-            right_column.column = castColumnAccurate(right_column, right_key_type);
-            right_column.type = right_key_type;
-        }
-
-        right_column.column = right_column.column->convertToFullColumnIfConst();
-        return right_column;
-    }
-
-    static void correctNullabilityInplace(ColumnWithTypeAndName & column, bool nullable)
-    {
-        if (nullable)
-        {
-            JoinCommon::convertColumnToNullable(column);
-        }
-        else
-        {
-            /// We have to replace values masked by NULLs with defaults.
-            if (column.column)
-                if (const auto * nullable_column = checkAndGetColumn<ColumnNullable>(&*column.column))
-                    column.column = JoinCommon::filterWithBlanks(column.column, nullable_column->getNullMapColumn().getData(), true);
-
-            JoinCommon::removeColumnNullability(column);
-        }
-    }
-
-    static void correctNullabilityInplace(ColumnWithTypeAndName & column, bool nullable, const IColumn::Filter & negative_null_map)
-    {
-        if (nullable)
-        {
-            JoinCommon::convertColumnToNullable(column);
-            if (column.type->isNullable() && !negative_null_map.empty())
-            {
-                MutableColumnPtr mutable_column = IColumn::mutate(std::move(column.column));
-                assert_cast<ColumnNullable &>(*mutable_column).applyNegatedNullMap(negative_null_map);
-                column.column = std::move(mutable_column);
-            }
-        }
-        else
-            JoinCommon::removeColumnNullability(column);
-    }
+    static void correctNullabilityInplace(ColumnWithTypeAndName & column, bool nullable, const IColumn::Filter & negative_null_map);
 };
 
 /// Instantiate template class ahead in different .cpp files to avoid `too large translation unit`.
 extern template class HashJoinMethods<JoinKind::Left, JoinStrictness::RightAny, HashJoin::MapsOne>;
 extern template class HashJoinMethods<JoinKind::Left, JoinStrictness::Any, HashJoin::MapsOne>;
+extern template class HashJoinMethods<JoinKind::Left, JoinStrictness::Any, HashJoin::MapsAll>;
 extern template class HashJoinMethods<JoinKind::Left, JoinStrictness::All, HashJoin::MapsAll>;
 extern template class HashJoinMethods<JoinKind::Left, JoinStrictness::Semi, HashJoin::MapsOne>;
+extern template class HashJoinMethods<JoinKind::Left, JoinStrictness::Semi, HashJoin::MapsAll>;
 extern template class HashJoinMethods<JoinKind::Left, JoinStrictness::Anti, HashJoin::MapsOne>;
+extern template class HashJoinMethods<JoinKind::Left, JoinStrictness::Anti, HashJoin::MapsAll>;
 extern template class HashJoinMethods<JoinKind::Left, JoinStrictness::Asof, HashJoin::MapsAsof>;
 
 extern template class HashJoinMethods<JoinKind::Right, JoinStrictness::RightAny, HashJoin::MapsOne>;
@@ -939,6 +185,7 @@ extern template class HashJoinMethods<JoinKind::Right, JoinStrictness::Asof, Has
 
 extern template class HashJoinMethods<JoinKind::Inner, JoinStrictness::RightAny, HashJoin::MapsOne>;
 extern template class HashJoinMethods<JoinKind::Inner, JoinStrictness::Any, HashJoin::MapsOne>;
+extern template class HashJoinMethods<JoinKind::Inner, JoinStrictness::Any, HashJoin::MapsAll>;
 extern template class HashJoinMethods<JoinKind::Inner, JoinStrictness::All, HashJoin::MapsAll>;
 extern template class HashJoinMethods<JoinKind::Inner, JoinStrictness::Semi, HashJoin::MapsOne>;
 extern template class HashJoinMethods<JoinKind::Inner, JoinStrictness::Anti, HashJoin::MapsOne>;

src/Interpreters/HashJoin/HashJoinMethodsImpl.h

@@ -0,0 +1,936 @@
+#pragma once
+#include <Interpreters/HashJoin/HashJoinMethods.h>
+
+namespace DB
+{
+namespace ErrorCodes
+{
+extern const int UNSUPPORTED_JOIN_KEYS;
+extern const int LOGICAL_ERROR;
+}
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+size_t HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::insertFromBlockImpl(
+    HashJoin & join,
+    HashJoin::Type type,
+    MapsTemplate & maps,
+    size_t rows,
+    const ColumnRawPtrs & key_columns,
+    const Sizes & key_sizes,
+    Block * stored_block,
+    ConstNullMapPtr null_map,
+    UInt8ColumnDataPtr join_mask,
+    Arena & pool,
+    bool & is_inserted)
+{
+    switch (type)
+    {
+        case HashJoin::Type::EMPTY:
+            [[fallthrough]];
+        case HashJoin::Type::CROSS:
+            /// Do nothing. We will only save block, and it is enough
+            is_inserted = true;
+            return 0;
+
+#define M(TYPE) \
+    case HashJoin::Type::TYPE: \
+        return insertFromBlockImplTypeCase< \
+            typename KeyGetterForType<HashJoin::Type::TYPE, std::remove_reference_t<decltype(*maps.TYPE)>>::Type>( \
+            join, *maps.TYPE, rows, key_columns, key_sizes, stored_block, null_map, join_mask, pool, is_inserted); \
+        break;
+
+            APPLY_FOR_JOIN_VARIANTS(M)
+#undef M
+    }
+}
+
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+Block HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::joinBlockImpl(
+    const HashJoin & join, Block & block, const Block & block_with_columns_to_add, const MapsTemplateVector & maps_, bool is_join_get)
+{
+    constexpr JoinFeatures<KIND, STRICTNESS, MapsTemplate> join_features;
+
+    std::vector<JoinOnKeyColumns> join_on_keys;
+    const auto & onexprs = join.table_join->getClauses();
+    for (size_t i = 0; i < onexprs.size(); ++i)
+    {
+        const auto & key_names = !is_join_get ? onexprs[i].key_names_left : onexprs[i].key_names_right;
+        join_on_keys.emplace_back(block, key_names, onexprs[i].condColumnNames().first, join.key_sizes[i]);
+    }
+    size_t existing_columns = block.columns();
+
+    /** If you use FULL or RIGHT JOIN, then the columns from the "left" table must be materialized.
+      * Because if they are constants, then in the "not joined" rows, they may have different values
+      *  - default values, which can differ from the values of these constants.
+      */
+    if constexpr (join_features.right || join_features.full)
+    {
+        materializeBlockInplace(block);
+    }
+
+    /** For LEFT/INNER JOIN, the saved blocks do not contain keys.
+      * For FULL/RIGHT JOIN, the saved blocks contain keys;
+      *  but they will not be used at this stage of joining (and will be in `AdderNonJoined`), and they need to be skipped.
+      * For ASOF, the last column is used as the ASOF column
+      */
+    AddedColumns<!join_features.is_any_join> added_columns(
+        block,
+        block_with_columns_to_add,
+        join.savedBlockSample(),
+        join,
+        std::move(join_on_keys),
+        join.table_join->getMixedJoinExpression(),
+        join_features.is_asof_join,
+        is_join_get);
+
+    bool has_required_right_keys = (join.required_right_keys.columns() != 0);
+    added_columns.need_filter = join_features.need_filter || has_required_right_keys;
+    added_columns.max_joined_block_rows = join.max_joined_block_rows;
+    if (!added_columns.max_joined_block_rows)
+        added_columns.max_joined_block_rows = std::numeric_limits<size_t>::max();
+    else
+        added_columns.reserve(join_features.need_replication);
+
+    size_t num_joined = switchJoinRightColumns(maps_, added_columns, join.data->type, *join.used_flags);
+    /// Do not hold memory for join_on_keys anymore
+    added_columns.join_on_keys.clear();
+    Block remaining_block = sliceBlock(block, num_joined);
+
+    added_columns.buildOutput();
+    for (size_t i = 0; i < added_columns.size(); ++i)
+        block.insert(added_columns.moveColumn(i));
+
+    std::vector<size_t> right_keys_to_replicate [[maybe_unused]];
+
+    if constexpr (join_features.need_filter)
+    {
+        /// If ANY INNER | RIGHT JOIN - filter all the columns except the new ones.
+        for (size_t i = 0; i < existing_columns; ++i)
+            block.safeGetByPosition(i).column = block.safeGetByPosition(i).column->filter(added_columns.filter, -1);
+
+        /// Add join key columns from right block if needed using value from left table because of equality
+        for (size_t i = 0; i < join.required_right_keys.columns(); ++i)
+        {
+            const auto & right_key = join.required_right_keys.getByPosition(i);
+            /// asof column is already in block.
+            if (join_features.is_asof_join && right_key.name == join.table_join->getOnlyClause().key_names_right.back())
+                continue;
+
+            const auto & left_column = block.getByName(join.required_right_keys_sources[i]);
+            const auto & right_col_name = join.getTableJoin().renamedRightColumnName(right_key.name);
+            auto right_col = copyLeftKeyColumnToRight(right_key.type, right_col_name, left_column);
+            block.insert(std::move(right_col));
+        }
+    }
+    else if (has_required_right_keys)
+    {
+        /// Add join key columns from right block if needed.
+        for (size_t i = 0; i < join.required_right_keys.columns(); ++i)
+        {
+            const auto & right_key = join.required_right_keys.getByPosition(i);
+            auto right_col_name = join.getTableJoin().renamedRightColumnName(right_key.name);
+            /// asof column is already in block.
+            if (join_features.is_asof_join && right_key.name == join.table_join->getOnlyClause().key_names_right.back())
+                continue;
+
+            const auto & left_column = block.getByName(join.required_right_keys_sources[i]);
+            auto right_col = copyLeftKeyColumnToRight(right_key.type, right_col_name, left_column, &added_columns.filter);
+            block.insert(std::move(right_col));
+
+            if constexpr (join_features.need_replication)
+                right_keys_to_replicate.push_back(block.getPositionByName(right_col_name));
+        }
+    }
+
+    if constexpr (join_features.need_replication)
+    {
+        std::unique_ptr<IColumn::Offsets> & offsets_to_replicate = added_columns.offsets_to_replicate;
+
+        /// If ALL ... JOIN - we replicate all the columns except the new ones.
+        for (size_t i = 0; i < existing_columns; ++i)
+        {
+            block.safeGetByPosition(i).column = block.safeGetByPosition(i).column->replicate(*offsets_to_replicate);
+        }
+
+        /// Replicate additional right keys
+        for (size_t pos : right_keys_to_replicate)
+        {
+            block.safeGetByPosition(pos).column = block.safeGetByPosition(pos).column->replicate(*offsets_to_replicate);
+        }
+    }
+    return remaining_block;
+}
+
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+template <typename KeyGetter, bool is_asof_join>
+KeyGetter HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::createKeyGetter(const ColumnRawPtrs & key_columns, const Sizes & key_sizes)
+{
+    if constexpr (is_asof_join)
+    {
+        auto key_column_copy = key_columns;
+        auto key_size_copy = key_sizes;
+        key_column_copy.pop_back();
+        key_size_copy.pop_back();
+        return KeyGetter(key_column_copy, key_size_copy, nullptr);
+    }
+    else
+        return KeyGetter(key_columns, key_sizes, nullptr);
+}
+
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+template <typename KeyGetter, typename HashMap>
+size_t HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::insertFromBlockImplTypeCase(
+    HashJoin & join,
+    HashMap & map,
+    size_t rows,
+    const ColumnRawPtrs & key_columns,
+    const Sizes & key_sizes,
+    Block * stored_block,
+    ConstNullMapPtr null_map,
+    UInt8ColumnDataPtr join_mask,
+    Arena & pool,
+    bool & is_inserted)
+{
+    [[maybe_unused]] constexpr bool mapped_one = std::is_same_v<typename HashMap::mapped_type, RowRef>;
+    constexpr bool is_asof_join = STRICTNESS == JoinStrictness::Asof;
+
+    const IColumn * asof_column [[maybe_unused]] = nullptr;
+    if constexpr (is_asof_join)
+        asof_column = key_columns.back();
+
+    auto key_getter = createKeyGetter<KeyGetter, is_asof_join>(key_columns, key_sizes);
+
+    /// For ALL and ASOF join always insert values
+    is_inserted = !mapped_one || is_asof_join;
+
+    for (size_t i = 0; i < rows; ++i)
+    {
+        if (null_map && (*null_map)[i])
+        {
+            /// nulls are not inserted into hash table,
+            /// keep them for RIGHT and FULL joins
+            is_inserted = true;
+            continue;
+        }
+
+        /// Check condition for right table from ON section
+        if (join_mask && !(*join_mask)[i])
+            continue;
+
+        if constexpr (is_asof_join)
+            Inserter<HashMap, KeyGetter>::insertAsof(join, map, key_getter, stored_block, i, pool, *asof_column);
+        else if constexpr (mapped_one)
+            is_inserted |= Inserter<HashMap, KeyGetter>::insertOne(join, map, key_getter, stored_block, i, pool);
+        else
+            Inserter<HashMap, KeyGetter>::insertAll(join, map, key_getter, stored_block, i, pool);
+    }
+    return map.getBufferSizeInCells();
+}
+
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+template <typename AddedColumns>
+size_t HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::switchJoinRightColumns(
+    const std::vector<const MapsTemplate *> & mapv,
+    AddedColumns & added_columns,
+    HashJoin::Type type,
+    JoinStuff::JoinUsedFlags & used_flags)
+{
+    constexpr bool is_asof_join = STRICTNESS == JoinStrictness::Asof;
+    switch (type)
+    {
+        case HashJoin::Type::EMPTY: {
+            if constexpr (!is_asof_join)
+            {
+                using KeyGetter = KeyGetterEmpty<typename MapsTemplate::MappedType>;
+                std::vector<KeyGetter> key_getter_vector;
+                key_getter_vector.emplace_back();
+
+                using MapTypeVal = typename KeyGetter::MappedType;
+                std::vector<const MapTypeVal *> a_map_type_vector;
+                a_map_type_vector.emplace_back();
+                return joinRightColumnsSwitchNullability<KeyGetter>(
+                    std::move(key_getter_vector), a_map_type_vector, added_columns, used_flags);
+            }
+            throw Exception(ErrorCodes::UNSUPPORTED_JOIN_KEYS, "Unsupported JOIN keys. Type: {}", type);
+        }
+#define M(TYPE) \
+    case HashJoin::Type::TYPE: { \
+        using MapTypeVal = const typename std::remove_reference_t<decltype(MapsTemplate::TYPE)>::element_type; \
+        using KeyGetter = typename KeyGetterForType<HashJoin::Type::TYPE, MapTypeVal>::Type; \
+        std::vector<const MapTypeVal *> a_map_type_vector(mapv.size()); \
+        std::vector<KeyGetter> key_getter_vector; \
+        for (size_t d = 0; d < added_columns.join_on_keys.size(); ++d) \
+        { \
+            const auto & join_on_key = added_columns.join_on_keys[d]; \
+            a_map_type_vector[d] = mapv[d]->TYPE.get(); \
+            key_getter_vector.push_back( \
+                std::move(createKeyGetter<KeyGetter, is_asof_join>(join_on_key.key_columns, join_on_key.key_sizes))); \
+        } \
+        return joinRightColumnsSwitchNullability<KeyGetter>(std::move(key_getter_vector), a_map_type_vector, added_columns, used_flags); \
+    }
+            APPLY_FOR_JOIN_VARIANTS(M)
+#undef M
+
+        default:
+            throw Exception(ErrorCodes::UNSUPPORTED_JOIN_KEYS, "Unsupported JOIN keys (type: {})", type);
+    }
+}
+
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+template <typename KeyGetter, typename Map, typename AddedColumns>
+size_t HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::joinRightColumnsSwitchNullability(
+    std::vector<KeyGetter> && key_getter_vector,
+    const std::vector<const Map *> & mapv,
+    AddedColumns & added_columns,
+    JoinStuff::JoinUsedFlags & used_flags)
+{
+    if (added_columns.need_filter)
+    {
+        return joinRightColumnsSwitchMultipleDisjuncts<KeyGetter, Map, true>(
+            std::forward<std::vector<KeyGetter>>(key_getter_vector), mapv, added_columns, used_flags);
+    }
+    else
+    {
+        return joinRightColumnsSwitchMultipleDisjuncts<KeyGetter, Map, false>(
+            std::forward<std::vector<KeyGetter>>(key_getter_vector), mapv, added_columns, used_flags);
+    }
+}
+
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+template <typename KeyGetter, typename Map, bool need_filter, typename AddedColumns>
+size_t HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::joinRightColumnsSwitchMultipleDisjuncts(
+    std::vector<KeyGetter> && key_getter_vector,
+    const std::vector<const Map *> & mapv,
+    AddedColumns & added_columns,
+    JoinStuff::JoinUsedFlags & used_flags)
+{
+    constexpr JoinFeatures<KIND, STRICTNESS, MapsTemplate> join_features;
+    if constexpr (join_features.is_maps_all)
+    {
+        if (added_columns.additional_filter_expression)
+        {
+            bool mark_per_row_used = join_features.right || join_features.full || mapv.size() > 1;
+            return joinRightColumnsWithAddtitionalFilter<KeyGetter, Map>(
+                std::forward<std::vector<KeyGetter>>(key_getter_vector), mapv, added_columns, used_flags, need_filter, mark_per_row_used);
+        }
+    }
+
+    if (added_columns.additional_filter_expression)
+        throw Exception(ErrorCodes::LOGICAL_ERROR, "Additional filter expression is not supported for this JOIN");
+
+    return mapv.size() > 1 ? joinRightColumns<KeyGetter, Map, need_filter, true>(
+               std::forward<std::vector<KeyGetter>>(key_getter_vector), mapv, added_columns, used_flags)
+                           : joinRightColumns<KeyGetter, Map, need_filter, false>(
+                               std::forward<std::vector<KeyGetter>>(key_getter_vector), mapv, added_columns, used_flags);
+}
+
+
+/// Joins right table columns which indexes are present in right_indexes using specified map.
+/// Makes filter (1 if row presented in right table) and returns offsets to replicate (for ALL JOINS).
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+template <typename KeyGetter, typename Map, bool need_filter, bool flag_per_row, typename AddedColumns>
+size_t HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::joinRightColumns(
+    std::vector<KeyGetter> && key_getter_vector,
+    const std::vector<const Map *> & mapv,
+    AddedColumns & added_columns,
+    JoinStuff::JoinUsedFlags & used_flags)
+{
+    constexpr JoinFeatures<KIND, STRICTNESS, MapsTemplate> join_features;
+
+    size_t rows = added_columns.rows_to_add;
+    if constexpr (need_filter)
+        added_columns.filter = IColumn::Filter(rows, 0);
+
+    Arena pool;
+
+    if constexpr (join_features.need_replication)
+        added_columns.offsets_to_replicate = std::make_unique<IColumn::Offsets>(rows);
+
+    IColumn::Offset current_offset = 0;
+    size_t max_joined_block_rows = added_columns.max_joined_block_rows;
+    size_t i = 0;
+    for (; i < rows; ++i)
+    {
+        if constexpr (join_features.need_replication)
+        {
+            if (unlikely(current_offset >= max_joined_block_rows))
+            {
+                added_columns.offsets_to_replicate->resize_assume_reserved(i);
+                added_columns.filter.resize_assume_reserved(i);
+                break;
+            }
+        }
+
+        bool right_row_found = false;
+        KnownRowsHolder<flag_per_row> known_rows;
+        for (size_t onexpr_idx = 0; onexpr_idx < added_columns.join_on_keys.size(); ++onexpr_idx)
+        {
+            const auto & join_keys = added_columns.join_on_keys[onexpr_idx];
+            if (join_keys.null_map && (*join_keys.null_map)[i])
+                continue;
+
+            bool row_acceptable = !join_keys.isRowFiltered(i);
+            using FindResult = typename KeyGetter::FindResult;
+            auto find_result = row_acceptable ? key_getter_vector[onexpr_idx].findKey(*(mapv[onexpr_idx]), i, pool) : FindResult();
+
+            if (find_result.isFound())
+            {
+                right_row_found = true;
+                auto & mapped = find_result.getMapped();
+                if constexpr (join_features.is_asof_join)
+                {
+                    const IColumn & left_asof_key = added_columns.leftAsofKey();
+
+                    auto row_ref = mapped->findAsof(left_asof_key, i);
+                    if (row_ref.block)
+                    {
+                        setUsed<need_filter>(added_columns.filter, i);
+                        if constexpr (flag_per_row)
+                            used_flags.template setUsed<join_features.need_flags, flag_per_row>(row_ref.block, row_ref.row_num, 0);
+                        else
+                            used_flags.template setUsed<join_features.need_flags, flag_per_row>(find_result);
+
+                        added_columns.appendFromBlock(*row_ref.block, row_ref.row_num, join_features.add_missing);
+                    }
+                    else
+                        addNotFoundRow<join_features.add_missing, join_features.need_replication>(added_columns, current_offset);
+                }
+                else if constexpr (join_features.is_all_join)
+                {
+                    setUsed<need_filter>(added_columns.filter, i);
+                    used_flags.template setUsed<join_features.need_flags, flag_per_row>(find_result);
+                    auto used_flags_opt = join_features.need_flags ? &used_flags : nullptr;
+                    addFoundRowAll<Map, join_features.add_missing>(mapped, added_columns, current_offset, known_rows, used_flags_opt);
+                }
+                else if constexpr ((join_features.is_any_join || join_features.is_semi_join) && join_features.right)
+                {
+                    /// Use first appeared left key + it needs left columns replication
+                    bool used_once = used_flags.template setUsedOnce<join_features.need_flags, flag_per_row>(find_result);
+                    if (used_once)
+                    {
+                        auto used_flags_opt = join_features.need_flags ? &used_flags : nullptr;
+                        setUsed<need_filter>(added_columns.filter, i);
+                        addFoundRowAll<Map, join_features.add_missing>(mapped, added_columns, current_offset, known_rows, used_flags_opt);
+                    }
+                }
+                else if constexpr (join_features.is_any_join && join_features.inner)
+                {
+                    bool used_once = used_flags.template setUsedOnce<join_features.need_flags, flag_per_row>(find_result);
+
+                    /// Use first appeared left key only
+                    if (used_once)
+                    {
+                        setUsed<need_filter>(added_columns.filter, i);
+                        added_columns.appendFromBlock(*mapped.block, mapped.row_num, join_features.add_missing);
+                    }
+
+                    break;
+                }
+                else if constexpr (join_features.is_any_join && join_features.full)
+                {
+                    /// TODO
+                }
+                else if constexpr (join_features.is_anti_join)
+                {
+                    if constexpr (join_features.right && join_features.need_flags)
+                        used_flags.template setUsed<join_features.need_flags, flag_per_row>(find_result);
+                }
+                else /// ANY LEFT, SEMI LEFT, old ANY (RightAny)
+                {
+                    setUsed<need_filter>(added_columns.filter, i);
+                    used_flags.template setUsed<join_features.need_flags, flag_per_row>(find_result);
+                    added_columns.appendFromBlock(*mapped.block, mapped.row_num, join_features.add_missing);
+
+                    if (join_features.is_any_or_semi_join)
+                    {
+                        break;
+                    }
+                }
+            }
+        }
+
+        if (!right_row_found)
+        {
+            if constexpr (join_features.is_anti_join && join_features.left)
+                setUsed<need_filter>(added_columns.filter, i);
+            addNotFoundRow<join_features.add_missing, join_features.need_replication>(added_columns, current_offset);
+        }
+
+        if constexpr (join_features.need_replication)
+        {
+            (*added_columns.offsets_to_replicate)[i] = current_offset;
+        }
+    }
+
+    added_columns.applyLazyDefaults();
+    return i;
+}
+
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+template <bool need_filter>
+void HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::setUsed(IColumn::Filter & filter [[maybe_unused]], size_t pos [[maybe_unused]])
+{
+    if constexpr (need_filter)
+        filter[pos] = 1;
+}
+
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+template <typename AddedColumns>
+ColumnPtr HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::buildAdditionalFilter(
+    size_t left_start_row,
+    const std::vector<RowRef> & selected_rows,
+    const std::vector<size_t> & row_replicate_offset,
+    AddedColumns & added_columns)
+{
+    ColumnPtr result_column;
+    do
+    {
+        if (selected_rows.empty())
+        {
+            result_column = ColumnUInt8::create();
+            break;
+        }
+        const Block & sample_right_block = *selected_rows.begin()->block;
+        if (!sample_right_block || !added_columns.additional_filter_expression)
+        {
+            auto filter = ColumnUInt8::create();
+            filter->insertMany(1, selected_rows.size());
+            result_column = std::move(filter);
+            break;
+        }
+
+        auto required_cols = added_columns.additional_filter_expression->getRequiredColumnsWithTypes();
+        if (required_cols.empty())
+        {
+            Block block;
+            added_columns.additional_filter_expression->execute(block);
+            result_column = block.getByPosition(0).column->cloneResized(selected_rows.size());
+            break;
+        }
+        NameSet required_column_names;
+        for (auto & col : required_cols)
+            required_column_names.insert(col.name);
+
+        Block executed_block;
+        size_t right_col_pos = 0;
+        for (const auto & col : sample_right_block.getColumnsWithTypeAndName())
+        {
+            if (required_column_names.contains(col.name))
+            {
+                auto new_col = col.column->cloneEmpty();
+                for (const auto & selected_row : selected_rows)
+                {
+                    const auto & src_col = selected_row.block->getByPosition(right_col_pos);
+                    new_col->insertFrom(*src_col.column, selected_row.row_num);
+                }
+                executed_block.insert({std::move(new_col), col.type, col.name});
+            }
+            right_col_pos += 1;
+        }
+        if (!executed_block)
+        {
+            result_column = ColumnUInt8::create();
+            break;
+        }
+
+        for (const auto & col_name : required_column_names)
+        {
+            const auto * src_col = added_columns.left_block.findByName(col_name);
+            if (!src_col)
+                continue;
+            auto new_col = src_col->column->cloneEmpty();
+            size_t prev_left_offset = 0;
+            for (size_t i = 1; i < row_replicate_offset.size(); ++i)
+            {
+                const size_t & left_offset = row_replicate_offset[i];
+                size_t rows = left_offset - prev_left_offset;
+                if (rows)
+                    new_col->insertManyFrom(*src_col->column, left_start_row + i - 1, rows);
+                prev_left_offset = left_offset;
+            }
+            executed_block.insert({std::move(new_col), src_col->type, col_name});
+        }
+        if (!executed_block)
+        {
+            throw Exception(
+                ErrorCodes::LOGICAL_ERROR,
+                "required columns: [{}], but not found any in left/right table. right table: {}, left table: {}",
+                required_cols.toString(),
+                sample_right_block.dumpNames(),
+                added_columns.left_block.dumpNames());
+        }
+
+        for (const auto & col : executed_block.getColumnsWithTypeAndName())
+            if (!col.column || !col.type)
+                throw Exception(ErrorCodes::LOGICAL_ERROR, "Illegal nullptr column in input block: {}", executed_block.dumpStructure());
+
+        added_columns.additional_filter_expression->execute(executed_block);
+        result_column = executed_block.getByPosition(0).column->convertToFullColumnIfConst();
+        executed_block.clear();
+    } while (false);
+
+    result_column = result_column->convertToFullIfNeeded();
+    if (result_column->isNullable())
+    {
+        /// Convert Nullable(UInt8) to UInt8 ensuring that nulls are zeros
+        /// Trying to avoid copying data, since we are the only owner of the column.
+        ColumnPtr mask_column = assert_cast<const ColumnNullable &>(*result_column).getNullMapColumnPtr();
+
+        MutableColumnPtr mutable_column;
+        {
+            ColumnPtr nested_column = assert_cast<const ColumnNullable &>(*result_column).getNestedColumnPtr();
+            result_column.reset();
+            mutable_column = IColumn::mutate(std::move(nested_column));
+        }
+
+        auto & column_data = assert_cast<ColumnUInt8 &>(*mutable_column).getData();
+        const auto & mask_column_data = assert_cast<const ColumnUInt8 &>(*mask_column).getData();
+        for (size_t i = 0; i < column_data.size(); ++i)
+        {
+            if (mask_column_data[i])
+                column_data[i] = 0;
+        }
+        return mutable_column;
+    }
+    return result_column;
+}
+
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+template <typename KeyGetter, typename Map, typename AddedColumns>
+size_t HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::joinRightColumnsWithAddtitionalFilter(
+    std::vector<KeyGetter> && key_getter_vector,
+    const std::vector<const Map *> & mapv,
+    AddedColumns & added_columns,
+    JoinStuff::JoinUsedFlags & used_flags [[maybe_unused]],
+    bool need_filter [[maybe_unused]],
+    bool flag_per_row [[maybe_unused]])
+{
+    constexpr JoinFeatures<KIND, STRICTNESS, MapsTemplate> join_features;
+    size_t left_block_rows = added_columns.rows_to_add;
+    if (need_filter)
+        added_columns.filter = IColumn::Filter(left_block_rows, 0);
+
+    std::unique_ptr<Arena> pool;
+
+    if constexpr (join_features.need_replication)
+        added_columns.offsets_to_replicate = std::make_unique<IColumn::Offsets>(left_block_rows);
+
+    std::vector<size_t> row_replicate_offset;
+    row_replicate_offset.reserve(left_block_rows);
+
+    using FindResult = typename KeyGetter::FindResult;
+    size_t max_joined_block_rows = added_columns.max_joined_block_rows;
+    size_t left_row_iter = 0;
+    PreSelectedRows selected_rows;
+    selected_rows.reserve(left_block_rows);
+    std::vector<FindResult> find_results;
+    find_results.reserve(left_block_rows);
+    bool exceeded_max_block_rows = false;
+    IColumn::Offset total_added_rows = 0;
+    IColumn::Offset current_added_rows = 0;
+
+    auto collect_keys_matched_rows_refs = [&]()
+    {
+        pool = std::make_unique<Arena>();
+        find_results.clear();
+        row_replicate_offset.clear();
+        row_replicate_offset.push_back(0);
+        current_added_rows = 0;
+        selected_rows.clear();
+        for (; left_row_iter < left_block_rows; ++left_row_iter)
+        {
+            if constexpr (join_features.need_replication)
+            {
+                if (unlikely(total_added_rows + current_added_rows >= max_joined_block_rows))
+                {
+                    break;
+                }
+            }
+            KnownRowsHolder<true> all_flag_known_rows;
+            KnownRowsHolder<false> single_flag_know_rows;
+            for (size_t join_clause_idx = 0; join_clause_idx < added_columns.join_on_keys.size(); ++join_clause_idx)
+            {
+                const auto & join_keys = added_columns.join_on_keys[join_clause_idx];
+                if (join_keys.null_map && (*join_keys.null_map)[left_row_iter])
+                    continue;
+
+                bool row_acceptable = !join_keys.isRowFiltered(left_row_iter);
+                auto find_result = row_acceptable
+                    ? key_getter_vector[join_clause_idx].findKey(*(mapv[join_clause_idx]), left_row_iter, *pool)
+                    : FindResult();
+
+                if (find_result.isFound())
+                {
+                    auto & mapped = find_result.getMapped();
+                    find_results.push_back(find_result);
+                    /// We don't add missing in addFoundRowAll here. we will add it after filter is applied.
+                    /// it's different from `joinRightColumns`.
+                    if (flag_per_row)
+                        addFoundRowAll<Map, false, true>(mapped, selected_rows, current_added_rows, all_flag_known_rows, nullptr);
+                    else
+                        addFoundRowAll<Map, false, false>(mapped, selected_rows, current_added_rows, single_flag_know_rows, nullptr);
+                }
+            }
+            row_replicate_offset.push_back(current_added_rows);
+        }
+    };
+
+    auto copy_final_matched_rows = [&](size_t left_start_row, ColumnPtr filter_col)
+    {
+        const PaddedPODArray<UInt8> & filter_flags = assert_cast<const ColumnUInt8 &>(*filter_col).getData();
+
+        size_t prev_replicated_row = 0;
+        auto selected_right_row_it = selected_rows.begin();
+        size_t find_result_index = 0;
+        for (size_t i = 1, n = row_replicate_offset.size(); i < n; ++i)
+        {
+            bool any_matched = false;
+            /// right/full join or multiple disjuncts, we need to mark used flags for each row.
+            if (flag_per_row)
+            {
+                for (size_t replicated_row = prev_replicated_row; replicated_row < row_replicate_offset[i]; ++replicated_row)
+                {
+                    if (filter_flags[replicated_row])
+                    {
+                        if constexpr (join_features.is_semi_join || join_features.is_any_join)
+                        {
+                            /// For LEFT/INNER SEMI/ANY JOIN, we need to add only first appeared row from left,
+                            if constexpr (join_features.left || join_features.inner)
+                            {
+                                if (!any_matched)
+                                {
+                                    // For inner join, we need mark each right row'flag, because we only use each right row once.
+                                    auto used_once = used_flags.template setUsedOnce<join_features.need_flags, true>(
+                                        selected_right_row_it->block, selected_right_row_it->row_num, 0);
+                                    if (used_once)
+                                    {
+                                        any_matched = true;
+                                        total_added_rows += 1;
+                                        added_columns.appendFromBlock(
+                                            *selected_right_row_it->block, selected_right_row_it->row_num, join_features.add_missing);
+                                    }
+                                }
+                            }
+                            else
+                            {
+                                auto used_once = used_flags.template setUsedOnce<join_features.need_flags, true>(
+                                    selected_right_row_it->block, selected_right_row_it->row_num, 0);
+                                if (used_once)
+                                {
+                                    any_matched = true;
+                                    total_added_rows += 1;
+                                    added_columns.appendFromBlock(
+                                        *selected_right_row_it->block, selected_right_row_it->row_num, join_features.add_missing);
+                                }
+                            }
+                        }
+                        else if constexpr (join_features.is_anti_join)
+                        {
+                            any_matched = true;
+                            if constexpr (join_features.right && join_features.need_flags)
+                                used_flags.template setUsed<true, true>(selected_right_row_it->block, selected_right_row_it->row_num, 0);
+                        }
+                        else
+                        {
+                            any_matched = true;
+                            total_added_rows += 1;
+                            added_columns.appendFromBlock(
+                                *selected_right_row_it->block, selected_right_row_it->row_num, join_features.add_missing);
+                            used_flags.template setUsed<join_features.need_flags, true>(
+                                selected_right_row_it->block, selected_right_row_it->row_num, 0);
+                        }
+                    }
+
+                    ++selected_right_row_it;
+                }
+            }
+            else
+            {
+                for (size_t replicated_row = prev_replicated_row; replicated_row < row_replicate_offset[i]; ++replicated_row)
+                {
+                    if constexpr (join_features.is_anti_join)
+                    {
+                        any_matched |= filter_flags[replicated_row];
+                    }
+                    else if constexpr (join_features.need_replication)
+                    {
+                        if (filter_flags[replicated_row])
+                        {
+                            any_matched = true;
+                            added_columns.appendFromBlock(
+                                *selected_right_row_it->block, selected_right_row_it->row_num, join_features.add_missing);
+                            total_added_rows += 1;
+                        }
+                        ++selected_right_row_it;
+                    }
+                    else
+                    {
+                        if (filter_flags[replicated_row])
+                        {
+                            any_matched = true;
+                            added_columns.appendFromBlock(
+                                *selected_right_row_it->block, selected_right_row_it->row_num, join_features.add_missing);
+                            total_added_rows += 1;
+                            selected_right_row_it = selected_right_row_it + row_replicate_offset[i] - replicated_row;
+                            break;
+                        }
+                        else
+                            ++selected_right_row_it;
+                    }
+                }
+            }
+
+
+            if constexpr (join_features.is_anti_join)
+            {
+                if (!any_matched)
+                {
+                    if constexpr (join_features.left)
+                        if (need_filter)
+                            setUsed<true>(added_columns.filter, left_start_row + i - 1);
+                    addNotFoundRow<join_features.add_missing, join_features.need_replication>(added_columns, total_added_rows);
+                }
+            }
+            else
+            {
+                if (!any_matched)
+                {
+                    addNotFoundRow<join_features.add_missing, join_features.need_replication>(added_columns, total_added_rows);
+                }
+                else
+                {
+                    if (!flag_per_row)
+                        used_flags.template setUsed<join_features.need_flags, false>(find_results[find_result_index]);
+                    if (need_filter)
+                        setUsed<true>(added_columns.filter, left_start_row + i - 1);
+                    if constexpr (join_features.add_missing)
+                        added_columns.applyLazyDefaults();
+                }
+            }
+            find_result_index += (prev_replicated_row != row_replicate_offset[i]);
+
+            if constexpr (join_features.need_replication)
+            {
+                (*added_columns.offsets_to_replicate)[left_start_row + i - 1] = total_added_rows;
+            }
+            prev_replicated_row = row_replicate_offset[i];
+        }
+    };
+
+    while (left_row_iter < left_block_rows && !exceeded_max_block_rows)
+    {
+        auto left_start_row = left_row_iter;
+        collect_keys_matched_rows_refs();
+        if (selected_rows.size() != current_added_rows || row_replicate_offset.size() != left_row_iter - left_start_row + 1)
+        {
+            throw Exception(
+                ErrorCodes::LOGICAL_ERROR,
+                "Sizes are mismatched. selected_rows.size:{}, current_added_rows:{}, row_replicate_offset.size:{}, left_row_iter: {}, "
+                "left_start_row: {}",
+                selected_rows.size(),
+                current_added_rows,
+                row_replicate_offset.size(),
+                left_row_iter,
+                left_start_row);
+        }
+        auto filter_col = buildAdditionalFilter(left_start_row, selected_rows, row_replicate_offset, added_columns);
+        copy_final_matched_rows(left_start_row, filter_col);
+
+        if constexpr (join_features.need_replication)
+        {
+            // Add a check for current_added_rows to avoid run the filter expression on too small size batch.
+            if (total_added_rows >= max_joined_block_rows || current_added_rows < 1024)
+                exceeded_max_block_rows = true;
+        }
+    }
+
+    if constexpr (join_features.need_replication)
+    {
+        added_columns.offsets_to_replicate->resize_assume_reserved(left_row_iter);
+        added_columns.filter.resize_assume_reserved(left_row_iter);
+    }
+    added_columns.applyLazyDefaults();
+    return left_row_iter;
+}
+
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+Block HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::sliceBlock(Block & block, size_t num_rows)
+{
+    size_t total_rows = block.rows();
+    if (num_rows >= total_rows)
+        return {};
+    size_t remaining_rows = total_rows - num_rows;
+    Block remaining_block = block.cloneEmpty();
+    for (size_t i = 0; i < block.columns(); ++i)
+    {
+        auto & col = block.getByPosition(i);
+        remaining_block.getByPosition(i).column = col.column->cut(num_rows, remaining_rows);
+        col.column = col.column->cut(0, num_rows);
+    }
+    return remaining_block;
+}
+
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+ColumnWithTypeAndName HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::copyLeftKeyColumnToRight(
+    const DataTypePtr & right_key_type,
+    const String & renamed_right_column,
+    const ColumnWithTypeAndName & left_column,
+    const IColumn::Filter * null_map_filter)
+{
+    ColumnWithTypeAndName right_column = left_column;
+    right_column.name = renamed_right_column;
+
+    if (null_map_filter)
+        right_column.column = JoinCommon::filterWithBlanks(right_column.column, *null_map_filter);
+
+    bool should_be_nullable = isNullableOrLowCardinalityNullable(right_key_type);
+    if (null_map_filter)
+        correctNullabilityInplace(right_column, should_be_nullable, *null_map_filter);
+    else
+        correctNullabilityInplace(right_column, should_be_nullable);
+
+    if (!right_column.type->equals(*right_key_type))
+    {
+        right_column.column = castColumnAccurate(right_column, right_key_type);
+        right_column.type = right_key_type;
+    }
+
+    right_column.column = right_column.column->convertToFullColumnIfConst();
+    return right_column;
+}
+
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+void HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::correctNullabilityInplace(ColumnWithTypeAndName & column, bool nullable)
+{
+    if (nullable)
+    {
+        JoinCommon::convertColumnToNullable(column);
+    }
+    else
+    {
+        /// We have to replace values masked by NULLs with defaults.
+        if (column.column)
+            if (const auto * nullable_column = checkAndGetColumn<ColumnNullable>(&*column.column))
+                column.column = JoinCommon::filterWithBlanks(column.column, nullable_column->getNullMapColumn().getData(), true);
+
+        JoinCommon::removeColumnNullability(column);
+    }
+}
+
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename MapsTemplate>
+void HashJoinMethods<KIND, STRICTNESS, MapsTemplate>::correctNullabilityInplace(
+    ColumnWithTypeAndName & column, bool nullable, const IColumn::Filter & negative_null_map)
+{
+    if (nullable)
+    {
+        JoinCommon::convertColumnToNullable(column);
+        if (column.type->isNullable() && !negative_null_map.empty())
+        {
+            MutableColumnPtr mutable_column = IColumn::mutate(std::move(column.column));
+            assert_cast<ColumnNullable &>(*mutable_column).applyNegatedNullMap(negative_null_map);
+            column.column = std::move(mutable_column);
+        }
+    }
+    else
+        JoinCommon::removeColumnNullability(column);
+}
+}

src/Interpreters/HashJoin/InnerHashJoin.cpp

@@ -1,10 +1,11 @@
 
-#include <Interpreters/HashJoin/HashJoinMethods.h>
+#include <Interpreters/HashJoin/HashJoinMethodsImpl.h>
 
 namespace DB
 {
 template class HashJoinMethods<JoinKind::Inner, JoinStrictness::RightAny, HashJoin::MapsOne>;
 template class HashJoinMethods<JoinKind::Inner, JoinStrictness::Any, HashJoin::MapsOne>;
+template class HashJoinMethods<JoinKind::Inner, JoinStrictness::Any, HashJoin::MapsAll>;
 template class HashJoinMethods<JoinKind::Inner, JoinStrictness::All, HashJoin::MapsAll>;
 template class HashJoinMethods<JoinKind::Inner, JoinStrictness::Semi, HashJoin::MapsOne>;
 template class HashJoinMethods<JoinKind::Inner, JoinStrictness::Anti, HashJoin::MapsOne>;

src/Interpreters/HashJoin/JoinFeatures.h

@@ -3,15 +3,15 @@
 #include <Interpreters/joinDispatch.h>
 namespace DB
 {
-template <JoinKind KIND, JoinStrictness STRICTNESS>
+template <JoinKind KIND, JoinStrictness STRICTNESS, typename Map>
 struct JoinFeatures
 {
     static constexpr bool is_any_join = STRICTNESS == JoinStrictness::Any;
-    static constexpr bool is_any_or_semi_join = STRICTNESS == JoinStrictness::Any || STRICTNESS == JoinStrictness::RightAny || (STRICTNESS == JoinStrictness::Semi && KIND == JoinKind::Left);
     static constexpr bool is_all_join = STRICTNESS == JoinStrictness::All;
     static constexpr bool is_asof_join = STRICTNESS == JoinStrictness::Asof;
     static constexpr bool is_semi_join = STRICTNESS == JoinStrictness::Semi;
     static constexpr bool is_anti_join = STRICTNESS == JoinStrictness::Anti;
+    static constexpr bool is_any_or_semi_join = is_any_join || STRICTNESS == JoinStrictness::RightAny || (is_semi_join && KIND == JoinKind::Left);
 
     static constexpr bool left = KIND == JoinKind::Left;
     static constexpr bool right = KIND == JoinKind::Right;
@@ -22,7 +22,8 @@ struct JoinFeatures
     static constexpr bool need_filter = !need_replication && (inner || right || (is_semi_join && left) || (is_anti_join && left));
     static constexpr bool add_missing = (left || full) && !is_semi_join;
 
-    static constexpr bool need_flags = MapGetter<KIND, STRICTNESS>::flagged;
+    static constexpr bool need_flags = MapGetter<KIND, STRICTNESS, std::is_same_v<std::decay_t<Map>, HashJoin::MapsAll>>::flagged;
+    static constexpr bool is_maps_all = std::is_same_v<std::decay_t<Map>, HashJoin::MapsAll>;
 };
 
 }

src/Interpreters/HashJoin/JoinUsedFlags.h

@@ -26,10 +26,10 @@ public:
     /// Update size for vector with flags.
     /// Calling this method invalidates existing flags.
     /// It can be called several times, but all of them should happen before using this structure.
-    template <JoinKind KIND, JoinStrictness STRICTNESS>
+    template <JoinKind KIND, JoinStrictness STRICTNESS, bool prefer_use_maps_all>
     void reinit(size_t size)
     {
-        if constexpr (MapGetter<KIND, STRICTNESS>::flagged)
+        if constexpr (MapGetter<KIND, STRICTNESS, prefer_use_maps_all>::flagged)
         {
             assert(flags[nullptr].size() <= size);
             need_flags = true;
@@ -43,10 +43,10 @@ public:
         }
     }
 
-    template <JoinKind KIND, JoinStrictness STRICTNESS>
+    template <JoinKind KIND, JoinStrictness STRICTNESS, bool prefer_use_maps_all>
     void reinit(const Block * block_ptr)
     {
-        if constexpr (MapGetter<KIND, STRICTNESS>::flagged)
+        if constexpr (MapGetter<KIND, STRICTNESS, prefer_use_maps_all>::flagged)
         {
             assert(flags[block_ptr].size() <= block_ptr->rows());
             need_flags = true;
@@ -148,6 +148,31 @@ public:
         }
 
     }
+    template <bool use_flags, bool flag_per_row>
+    bool setUsedOnce(const Block * block, size_t row_num, size_t offset)
+    {
+        if constexpr (!use_flags)
+            return true;
+
+        if constexpr (flag_per_row)
+        {
+            /// fast check to prevent heavy CAS with seq_cst order
+            if (flags[block][row_num].load(std::memory_order_relaxed))
+                return false;
+
+            bool expected = false;
+            return flags[block][row_num].compare_exchange_strong(expected, true);
+        }
+        else
+        {
+            /// fast check to prevent heavy CAS with seq_cst order
+            if (flags[nullptr][offset].load(std::memory_order_relaxed))
+                return false;
+
+            bool expected = false;
+            return flags[nullptr][offset].compare_exchange_strong(expected, true);
+        }
+    }
 };
 
 }

src/Interpreters/HashJoin/LeftHashJoin.cpp

@@ -1,11 +1,14 @@
-#include <Interpreters/HashJoin/HashJoinMethods.h>
+#include <Interpreters/HashJoin/HashJoinMethodsImpl.h>
 
 namespace DB
 {
 template class HashJoinMethods<JoinKind::Left, JoinStrictness::RightAny, HashJoin::MapsOne>;
 template class HashJoinMethods<JoinKind::Left, JoinStrictness::Any, HashJoin::MapsOne>;
+template class HashJoinMethods<JoinKind::Left, JoinStrictness::Any, HashJoin::MapsAll>;
 template class HashJoinMethods<JoinKind::Left, JoinStrictness::All, HashJoin::MapsAll>;
 template class HashJoinMethods<JoinKind::Left, JoinStrictness::Semi, HashJoin::MapsOne>;
+template class HashJoinMethods<JoinKind::Left, JoinStrictness::Semi, HashJoin::MapsAll>;
 template class HashJoinMethods<JoinKind::Left, JoinStrictness::Anti, HashJoin::MapsOne>;
+template class HashJoinMethods<JoinKind::Left, JoinStrictness::Anti, HashJoin::MapsAll>;
 template class HashJoinMethods<JoinKind::Left, JoinStrictness::Asof, HashJoin::MapsAsof>;
 }

src/Interpreters/HashJoin/RightHashJoin.cpp

@@ -1,4 +1,4 @@
-#include <Interpreters/HashJoin/HashJoinMethods.h>
+#include <Interpreters/HashJoin/HashJoinMethodsImpl.h>
 
 namespace DB
 {

src/Interpreters/castColumn.cpp

@@ -26,11 +26,9 @@ static ColumnPtr castColumn(CastType cast_type, const ColumnWithTypeAndName & ar
             ""
         }
     };
-    auto get_cast_func = [cast_type, &arguments]
+    auto get_cast_func = [from = arg, to = type, cast_type]
     {
-
-        FunctionOverloadResolverPtr func_builder_cast = createInternalCastOverloadResolver(cast_type, {});
-        return func_builder_cast->build(arguments);
+        return createInternalCast(from, to, cast_type, {});
     };
 
     FunctionBasePtr func_cast = cache ? cache->getOrSet(cast_type, from_name, to_name, std::move(get_cast_func)) : get_cast_func();

src/Interpreters/joinDispatch.h

@@ -12,38 +12,53 @@
 namespace DB
 {
 
-template <JoinKind kind, JoinStrictness join_strictness>
+/// HashJoin::MapsOne is more efficient, it only store one row for each key in the map. It is recommended to use it whenever possible.
+/// When only need to match only one row from right table, use HashJoin::MapsOne. For example, LEFT ANY/SEMI/ANTI.
+///
+/// HashJoin::MapsAll will store all rows for each key in the map. It is used when need to match multiple rows from right table.
+/// For example, LEFT ALL, INNER ALL, RIGHT ALL/ANY.
+///
+/// prefer_use_maps_all is true when there is mixed inequal condition in the join condition. For example, `t1.a = t2.a AND t1.b > t2.b`.
+/// In this case, we need to use HashJoin::MapsAll to store all rows for each key in the map. We will select all matched rows from the map
+/// and filter them by `t1.b > t2.b`.
+///
+/// flagged indicates whether we need to store flags for each row whether it has been used in the join. See JoinUsedFlags.h.
+template <JoinKind kind, JoinStrictness join_strictness, bool prefer_use_maps_all>
 struct MapGetter;
 
-template <> struct MapGetter<JoinKind::Left, JoinStrictness::RightAny>  { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
-template <> struct MapGetter<JoinKind::Inner, JoinStrictness::RightAny> { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
-template <> struct MapGetter<JoinKind::Right, JoinStrictness::RightAny> { using Map = HashJoin::MapsOne; static constexpr bool flagged = true; };
-template <> struct MapGetter<JoinKind::Full, JoinStrictness::RightAny>  { using Map = HashJoin::MapsOne; static constexpr bool flagged = true; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Left, JoinStrictness::RightAny, prefer_use_maps_all>  { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Inner, JoinStrictness::RightAny, prefer_use_maps_all> { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Right, JoinStrictness::RightAny, prefer_use_maps_all> { using Map = HashJoin::MapsOne; static constexpr bool flagged = true; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Full, JoinStrictness::RightAny, prefer_use_maps_all>  { using Map = HashJoin::MapsOne; static constexpr bool flagged = true; };
 
-template <> struct MapGetter<JoinKind::Left, JoinStrictness::Any>  { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
-template <> struct MapGetter<JoinKind::Inner, JoinStrictness::Any> { using Map = HashJoin::MapsOne; static constexpr bool flagged = true; };
-template <> struct MapGetter<JoinKind::Right, JoinStrictness::Any> { using Map = HashJoin::MapsAll; static constexpr bool flagged = true; };
-template <> struct MapGetter<JoinKind::Full, JoinStrictness::Any>  { using Map = HashJoin::MapsAll; static constexpr bool flagged = true; };
+template <> struct MapGetter<JoinKind::Left, JoinStrictness::Any, false>  { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
+template <> struct MapGetter<JoinKind::Left, JoinStrictness::Any, true>  { using Map = HashJoin::MapsAll; static constexpr bool flagged = false; };
+template <> struct MapGetter<JoinKind::Inner, JoinStrictness::Any, true> { using Map = HashJoin::MapsAll; static constexpr bool flagged = true; };
+template <> struct MapGetter<JoinKind::Inner, JoinStrictness::Any, false> { using Map = HashJoin::MapsOne; static constexpr bool flagged = true; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Right, JoinStrictness::Any, prefer_use_maps_all> { using Map = HashJoin::MapsAll; static constexpr bool flagged = true; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Full, JoinStrictness::Any, prefer_use_maps_all>  { using Map = HashJoin::MapsAll; static constexpr bool flagged = true; };
 
-template <> struct MapGetter<JoinKind::Left, JoinStrictness::All>  { using Map = HashJoin::MapsAll; static constexpr bool flagged = false; };
-template <> struct MapGetter<JoinKind::Inner, JoinStrictness::All> { using Map = HashJoin::MapsAll; static constexpr bool flagged = false; };
-template <> struct MapGetter<JoinKind::Right, JoinStrictness::All> { using Map = HashJoin::MapsAll; static constexpr bool flagged = true; };
-template <> struct MapGetter<JoinKind::Full, JoinStrictness::All>  { using Map = HashJoin::MapsAll; static constexpr bool flagged = true; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Left, JoinStrictness::All, prefer_use_maps_all>  { using Map = HashJoin::MapsAll; static constexpr bool flagged = false; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Inner, JoinStrictness::All, prefer_use_maps_all> { using Map = HashJoin::MapsAll; static constexpr bool flagged = false; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Right, JoinStrictness::All, prefer_use_maps_all> { using Map = HashJoin::MapsAll; static constexpr bool flagged = true; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Full, JoinStrictness::All, prefer_use_maps_all>  { using Map = HashJoin::MapsAll; static constexpr bool flagged = true; };
 
 /// Only SEMI LEFT and SEMI RIGHT are valid. INNER and FULL are here for templates instantiation.
-template <> struct MapGetter<JoinKind::Left, JoinStrictness::Semi>  { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
-template <> struct MapGetter<JoinKind::Inner, JoinStrictness::Semi> { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
-template <> struct MapGetter<JoinKind::Right, JoinStrictness::Semi> { using Map = HashJoin::MapsAll; static constexpr bool flagged = true; };
-template <> struct MapGetter<JoinKind::Full, JoinStrictness::Semi>  { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
+template <> struct MapGetter<JoinKind::Left, JoinStrictness::Semi, false>  { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
+template <> struct MapGetter<JoinKind::Left, JoinStrictness::Semi, true>  { using Map = HashJoin::MapsAll; static constexpr bool flagged = false; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Inner, JoinStrictness::Semi, prefer_use_maps_all> { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Right, JoinStrictness::Semi, prefer_use_maps_all> { using Map = HashJoin::MapsAll; static constexpr bool flagged = true; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Full, JoinStrictness::Semi, prefer_use_maps_all>  { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
 
-/// Only SEMI LEFT and SEMI RIGHT are valid. INNER and FULL are here for templates instantiation.
-template <> struct MapGetter<JoinKind::Left, JoinStrictness::Anti>  { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
-template <> struct MapGetter<JoinKind::Inner, JoinStrictness::Anti> { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
-template <> struct MapGetter<JoinKind::Right, JoinStrictness::Anti> { using Map = HashJoin::MapsAll; static constexpr bool flagged = true; };
-template <> struct MapGetter<JoinKind::Full, JoinStrictness::Anti>  { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
+/// Only ANTI LEFT and ANTI RIGHT are valid. INNER and FULL are here for templates instantiation.
+template <> struct MapGetter<JoinKind::Left, JoinStrictness::Anti, false>  { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
+template <> struct MapGetter<JoinKind::Left, JoinStrictness::Anti, true>  { using Map = HashJoin::MapsAll; static constexpr bool flagged = true; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Inner, JoinStrictness::Anti, prefer_use_maps_all> { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Right, JoinStrictness::Anti, prefer_use_maps_all> { using Map = HashJoin::MapsAll; static constexpr bool flagged = true; };
+template <bool prefer_use_maps_all> struct MapGetter<JoinKind::Full, JoinStrictness::Anti, prefer_use_maps_all>  { using Map = HashJoin::MapsOne; static constexpr bool flagged = false; };
 
-template <JoinKind kind>
-struct MapGetter<kind, JoinStrictness::Asof> { using Map = HashJoin::MapsAsof; static constexpr bool flagged = false; };
+template <JoinKind kind, bool prefer_use_maps_all>
+struct MapGetter<kind, JoinStrictness::Asof, prefer_use_maps_all> { using Map = HashJoin::MapsAsof; static constexpr bool flagged = false; };
 
 static constexpr std::array<JoinStrictness, 6> STRICTNESSES = {
     JoinStrictness::RightAny,
@@ -62,7 +77,7 @@ static constexpr std::array<JoinKind, 4> KINDS = {
 };
 
 /// Init specified join map
-inline bool joinDispatchInit(JoinKind kind, JoinStrictness strictness, HashJoin::MapsVariant & maps)
+inline bool joinDispatchInit(JoinKind kind, JoinStrictness strictness, HashJoin::MapsVariant & maps, bool prefer_use_maps_all = false)
 {
     return static_for<0, KINDS.size() * STRICTNESSES.size()>([&](auto ij)
     {
@@ -70,7 +85,10 @@ inline bool joinDispatchInit(JoinKind kind, JoinStrictness strictness, HashJoin:
         constexpr auto j = ij % STRICTNESSES.size();
         if (kind == KINDS[i] && strictness == STRICTNESSES[j])
         {
-            maps = typename MapGetter<KINDS[i], STRICTNESSES[j]>::Map();
+            if (prefer_use_maps_all)
+                maps = typename MapGetter<KINDS[i], STRICTNESSES[j], true>::Map();
+            else
+                maps = typename MapGetter<KINDS[i], STRICTNESSES[j], false>::Map();
             return true;
         }
         return false;
@@ -79,7 +97,7 @@ inline bool joinDispatchInit(JoinKind kind, JoinStrictness strictness, HashJoin:
 
 /// Call function on specified join map
 template <typename MapsVariant, typename Func>
-inline bool joinDispatch(JoinKind kind, JoinStrictness strictness, MapsVariant & maps, Func && func)
+inline bool joinDispatch(JoinKind kind, JoinStrictness strictness, MapsVariant & maps, bool prefer_use_maps_all, Func && func)
 {
     return static_for<0, KINDS.size() * STRICTNESSES.size()>([&](auto ij)
     {
@@ -89,10 +107,16 @@ inline bool joinDispatch(JoinKind kind, JoinStrictness strictness, MapsVariant &
         constexpr auto j = ij % STRICTNESSES.size();
         if (kind == KINDS[i] && strictness == STRICTNESSES[j])
         {
-            func(
-                std::integral_constant<JoinKind, KINDS[i]>(),
-                std::integral_constant<JoinStrictness, STRICTNESSES[j]>(),
-                std::get<typename MapGetter<KINDS[i], STRICTNESSES[j]>::Map>(maps));
+            if (prefer_use_maps_all)
+                func(
+                    std::integral_constant<JoinKind, KINDS[i]>(),
+                    std::integral_constant<JoinStrictness, STRICTNESSES[j]>(),
+                    std::get<typename MapGetter<KINDS[i], STRICTNESSES[j], true>::Map>(maps));
+            else
+                func(
+                    std::integral_constant<JoinKind, KINDS[i]>(),
+                    std::integral_constant<JoinStrictness, STRICTNESSES[j]>(),
+                    std::get<typename MapGetter<KINDS[i], STRICTNESSES[j], false>::Map>(maps));
             return true;
         }
         return false;
@@ -101,7 +125,7 @@ inline bool joinDispatch(JoinKind kind, JoinStrictness strictness, MapsVariant &
 
 /// Call function on specified join map
 template <typename MapsVariant, typename Func>
-inline bool joinDispatch(JoinKind kind, JoinStrictness strictness, std::vector<const MapsVariant *> & mapsv, Func && func)
+inline bool joinDispatch(JoinKind kind, JoinStrictness strictness, std::vector<const MapsVariant *> & mapsv, bool prefer_use_maps_all, Func && func)
 {
     return static_for<0, KINDS.size() * STRICTNESSES.size()>([&](auto ij)
     {
@@ -111,17 +135,31 @@ inline bool joinDispatch(JoinKind kind, JoinStrictness strictness, std::vector<c
         constexpr auto j = ij % STRICTNESSES.size();
         if (kind == KINDS[i] && strictness == STRICTNESSES[j])
         {
-            using MapType = typename MapGetter<KINDS[i], STRICTNESSES[j]>::Map;
-            std::vector<const MapType *> v;
-            v.reserve(mapsv.size());
-            for (const auto & el : mapsv)
-                v.push_back(&std::get<MapType>(*el));
+            if (prefer_use_maps_all)
+            {
+                using MapType = typename MapGetter<KINDS[i], STRICTNESSES[j], true>::Map;
+                std::vector<const MapType *> v;
+                v.reserve(mapsv.size());
+                for (const auto & el : mapsv)
+                    v.push_back(&std::get<MapType>(*el));
 
-            func(
-                std::integral_constant<JoinKind, KINDS[i]>(),
-                std::integral_constant<JoinStrictness, STRICTNESSES[j]>(),
-                v
-                /*std::get<typename MapGetter<KINDS[i], STRICTNESSES[j]>::Map>(maps)*/);
+                func(
+                    std::integral_constant<JoinKind, KINDS[i]>(), std::integral_constant<JoinStrictness, STRICTNESSES[j]>(), v
+                    /*std::get<typename MapGetter<KINDS[i], STRICTNESSES[j]>::Map>(maps)*/);
+            }
+            else
+            {
+                using MapType = typename MapGetter<KINDS[i], STRICTNESSES[j], false>::Map;
+                std::vector<const MapType *> v;
+                v.reserve(mapsv.size());
+                for (const auto & el : mapsv)
+                    v.push_back(&std::get<MapType>(*el));
+
+                func(
+                    std::integral_constant<JoinKind, KINDS[i]>(), std::integral_constant<JoinStrictness, STRICTNESSES[j]>(), v
+                    /*std::get<typename MapGetter<KINDS[i], STRICTNESSES[j]>::Map>(maps)*/);
+
+            }
             return true;
         }
         return false;

src/Processors/QueryPlan/Optimizations/optimizeUseAggregateProjection.cpp

@@ -255,20 +255,13 @@ static void appendAggregateFunctions(
 
         const auto * node = input;
 
-        if (node->result_name != aggregate.column_name)
-        {
-            if (DataTypeAggregateFunction::strictEquals(type, node->result_type))
-            {
-                node = &proj_dag.addAlias(*node, aggregate.column_name);
-            }
-            else
-            {
-                /// Cast to aggregate types specified in query if it's not
-                /// strictly the same as the one specified in projection. This
-                /// is required to generate correct results during finalization.
-                node = &proj_dag.addCast(*node, type, aggregate.column_name);
-            }
-        }
+        if (!DataTypeAggregateFunction::strictEquals(type, node->result_type))
+            /// Cast to aggregate types specified in query if it's not
+            /// strictly the same as the one specified in projection. This
+            /// is required to generate correct results during finalization.
+            node = &proj_dag.addCast(*node, type, aggregate.column_name);
+        else if (node->result_name != aggregate.column_name)
+            node = &proj_dag.addAlias(*node, aggregate.column_name);
 
         proj_dag_outputs.push_back(node);
     }

src/Processors/QueryPlan/Optimizations/removeRedundantSorting.cpp

@@ -8,6 +8,7 @@
 #include <Processors/QueryPlan/JoinStep.h>
 #include <Processors/QueryPlan/LimitByStep.h>
 #include <Processors/QueryPlan/LimitStep.h>
+#include <Processors/QueryPlan/OffsetStep.h>
 #include <Processors/QueryPlan/Optimizations/Optimizations.h>
 #include <Processors/QueryPlan/QueryPlanVisitor.h>
 #include <Processors/QueryPlan/ReadFromMergeTree.h>
@@ -59,9 +60,10 @@ public:
 
         if (typeid_cast<LimitStep *>(current_step)
             || typeid_cast<LimitByStep *>(current_step) /// (1) if there are LIMITs on top of ORDER BY, the ORDER BY is non-removable
-            || typeid_cast<FillingStep *>(current_step) /// (2) if ORDER BY is with FILL WITH, it is non-removable
-            || typeid_cast<SortingStep *>(current_step) /// (3) ORDER BY will change order of previous sorting
-            || typeid_cast<AggregatingStep *>(current_step)) /// (4) aggregation change order
+            || typeid_cast<OffsetStep *>(current_step) /// (2) OFFSET on top of ORDER BY, the ORDER BY is non-removable
+            || typeid_cast<FillingStep *>(current_step) /// (3) if ORDER BY is with FILL WITH, it is non-removable
+            || typeid_cast<SortingStep *>(current_step) /// (4) ORDER BY will change order of previous sorting
+            || typeid_cast<AggregatingStep *>(current_step)) /// (5) aggregation change order
         {
             logStep("nodes_affect_order/push", current_node);
             nodes_affect_order.push_back(current_node);

src/Processors/Sources/ShellCommandSource.cpp

@@ -8,13 +8,15 @@
 #include <IO/WriteHelpers.h>
 #include <IO/ReadHelpers.h>
 
-#include <QueryPipeline/Pipe.h>
-#include <Processors/ISimpleTransform.h>
-#include <Processors/Formats/IOutputFormat.h>
-#include <Processors/Executors/CompletedPipelineExecutor.h>
 #include <Interpreters/Context.h>
+#include <Processors/Executors/CompletedPipelineExecutor.h>
+#include <Processors/Formats/IOutputFormat.h>
+#include <Processors/ISimpleTransform.h>
+#include <QueryPipeline/Pipe.h>
+
 #include <boost/circular_buffer.hpp>
 
+#include <ranges>
 
 namespace DB
 {
@@ -68,11 +70,17 @@ static void makeFdBlocking(int fd)
 
 static int pollWithTimeout(pollfd * pfds, size_t num, size_t timeout_milliseconds)
 {
+    auto logger = getLogger("TimeoutReadBufferFromFileDescriptor");
+    auto describe_fd = [](const auto & pollfd) { return fmt::format("(fd={}, flags={})", pollfd.fd, fcntl(pollfd.fd, F_GETFL)); };
+
     int res;
 
     while (true)
     {
         Stopwatch watch;
+
+        LOG_TEST(logger, "Polling descriptors: {}", fmt::join(std::span(pfds, pfds + num) | std::views::transform(describe_fd), ", "));
+
         res = poll(pfds, static_cast<nfds_t>(num), static_cast<int>(timeout_milliseconds));
 
         if (res < 0)
@@ -82,7 +90,10 @@ static int pollWithTimeout(pollfd * pfds, size_t num, size_t timeout_millisecond
 
             const auto elapsed = watch.elapsedMilliseconds();
             if (timeout_milliseconds <= elapsed)
+            {
+                LOG_TEST(logger, "Timeout exceeded: elapsed={}, timeout={}", elapsed, timeout_milliseconds);
                 break;
+            }
             timeout_milliseconds -= elapsed;
         }
         else
@@ -91,6 +102,12 @@ static int pollWithTimeout(pollfd * pfds, size_t num, size_t timeout_millisecond
         }
     }
 
+    LOG_TEST(
+        logger,
+        "Poll for descriptors: {} returned {}",
+        fmt::join(std::span(pfds, pfds + num) | std::views::transform(describe_fd), ", "),
+        res);
+
     return res;
 }
 
@@ -200,12 +217,6 @@ public:
         return true;
     }
 
-    void reset() const
-    {
-        makeFdBlocking(stdout_fd);
-        makeFdBlocking(stderr_fd);
-    }
-
     ~TimeoutReadBufferFromFileDescriptor() override
     {
         tryMakeFdBlocking(stdout_fd);

src/Processors/Transforms/WindowTransform.cpp

@@ -2337,22 +2337,9 @@ struct WindowFunctionLagLeadInFrame final : public WindowFunction
                 argument_types[2]->getName());
         }
 
-        const auto from_name = argument_types[2]->getName();
-        const auto to_name = argument_types[0]->getName();
-        ColumnsWithTypeAndName arguments
+        auto get_cast_func = [from = argument_types[2], to = argument_types[0]]
         {
-            { argument_types[2], "" },
-            {
-                DataTypeString().createColumnConst(0, to_name),
-                std::make_shared<DataTypeString>(),
-                ""
-            }
-        };
-
-        auto get_cast_func = [&arguments]
-        {
-            FunctionOverloadResolverPtr func_builder_cast = createInternalCastOverloadResolver(CastType::accurate, {});
-            return func_builder_cast->build(arguments);
+            return createInternalCast({from, {}}, to, CastType::accurate, {});
         };
 
         func_cast = get_cast_func();

src/Storages/MergeTree/IMergeTreeDataPart.cpp

@@ -749,8 +749,16 @@ void IMergeTreeDataPart::loadColumnsChecksumsIndexes(bool require_columns_checks
                 /// Probably there is something wrong with files of this part.
                 /// So it can be helpful to add to the error message some information about those files.
                 String files_in_part;
+
                 for (auto it = getDataPartStorage().iterate(); it->isValid(); it->next())
-                    files_in_part += fmt::format("{}{} ({} bytes)", (files_in_part.empty() ? "" : ", "), it->name(), getDataPartStorage().getFileSize(it->name()));
+                {
+                    std::string file_info;
+                    if (!getDataPartStorage().isDirectory(it->name()))
+                        file_info = fmt::format(" ({} bytes)", getDataPartStorage().getFileSize(it->name()));
+
+                    files_in_part += fmt::format("{}{}{}", (files_in_part.empty() ? "" : ", "), it->name(), file_info);
+
+                }
                 if (!files_in_part.empty())
                     e->addMessage("Part contains files: {}", files_in_part);
                 if (isEmpty())
@@ -2141,7 +2149,27 @@ void IMergeTreeDataPart::checkConsistencyBase() const
             }
         }
 
-        checksums.checkSizes(getDataPartStorage());
+        const auto & data_part_storage = getDataPartStorage();
+        for (const auto & [filename, checksum] : checksums.files)
+        {
+            try
+            {
+                checksum.checkSize(data_part_storage, filename);
+            }
+            catch (const Exception & ex)
+            {
+                /// For projection parts check will mark them broken in loadProjections
+                if (!parent_part && filename.ends_with(".proj"))
+                {
+                    std::string projection_name = fs::path(filename).stem();
+                    LOG_INFO(storage.log, "Projection {} doesn't exist on start for part {}, marking it as broken", projection_name, name);
+                    if (hasProjection(projection_name))
+                        markProjectionPartAsBroken(projection_name, ex.message(), ex.code());
+                }
+                else
+                    throw;
+            }
+        }
     }
     else
     {

src/Storages/MergeTree/KeyCondition.cpp

@@ -1956,11 +1956,8 @@ bool KeyCondition::extractAtomFromTree(const RPNBuilderTreeNode & node, RPNEleme
                         auto common_type_maybe_nullable = (key_expr_type_is_nullable && !common_type->isNullable())
                             ? DataTypePtr(std::make_shared<DataTypeNullable>(common_type))
                             : common_type;
-                        ColumnsWithTypeAndName arguments{
-                            {nullptr, key_expr_type, ""},
-                            {DataTypeString().createColumnConst(1, common_type_maybe_nullable->getName()), common_type_maybe_nullable, ""}};
-                        FunctionOverloadResolverPtr func_builder_cast = createInternalCastOverloadResolver(CastType::nonAccurate, {});
-                        auto func_cast = func_builder_cast->build(arguments);
+
+                        auto func_cast = createInternalCast({key_expr_type, {}}, common_type_maybe_nullable, CastType::nonAccurate, {});
 
                         /// If we know the given range only contains one value, then we treat all functions as positive monotonic.
                         if (!single_point && !func_cast->hasInformationAboutMonotonicity())

src/Storages/MergeTree/MergeTreeData.cpp

@@ -1146,7 +1146,7 @@ std::optional<UInt64> MergeTreeData::totalRowsByPartitionPredicateImpl(
     auto metadata_snapshot = getInMemoryMetadataPtr();
     auto virtual_columns_block = getBlockWithVirtualsForFilter(metadata_snapshot, {parts[0]});
 
-    auto filter_dag = VirtualColumnUtils::splitFilterDagForAllowedInputs(filter_actions_dag.getOutputs().at(0), nullptr);
+    auto filter_dag = VirtualColumnUtils::splitFilterDagForAllowedInputs(filter_actions_dag.getOutputs().at(0), nullptr, /*allow_partial_result=*/ false);
     if (!filter_dag)
         return {};
 
@@ -6932,7 +6932,8 @@ Block MergeTreeData::getMinMaxCountProjectionBlock(
         const auto * predicate = filter_dag->getOutputs().at(0);
 
         // Generate valid expressions for filtering
-        VirtualColumnUtils::filterBlockWithPredicate(predicate, virtual_columns_block, query_context);
+        VirtualColumnUtils::filterBlockWithPredicate(
+            predicate, virtual_columns_block, query_context, /*allow_filtering_with_partial_predicate =*/true);
 
         rows = virtual_columns_block.rows();
         part_name_column = virtual_columns_block.getByName("_part").column;

src/Storages/MergeTree/MergeTreeDataPartChecksum.cpp

@@ -100,12 +100,6 @@ void MergeTreeDataPartChecksums::checkEqual(const MergeTreeDataPartChecksums & r
     }
 }
 
-void MergeTreeDataPartChecksums::checkSizes(const IDataPartStorage & storage) const
-{
-    for (const auto & [name, checksum] : files)
-        checksum.checkSize(storage, name);
-}
-
 UInt64 MergeTreeDataPartChecksums::getTotalSizeOnDisk() const
 {
     UInt64 res = 0;

src/Storages/MergeTree/MergeTreeDataPartChecksum.h

@@ -65,9 +65,6 @@ struct MergeTreeDataPartChecksums
 
     static bool isBadChecksumsErrorCode(int code);
 
-    /// Checks that the directory contains all the needed files of the correct size. Does not check the checksum.
-    void checkSizes(const IDataPartStorage & storage) const;
-
     /// Returns false if the checksum is too old.
     bool read(ReadBuffer & in);
     /// Assume that header with version (the first line) is read

src/Storages/MergeTree/MergeTreeSplitPrewhereIntoReadSteps.cpp

@@ -152,23 +152,15 @@ const ActionsDAG::Node & addFunction(
 const ActionsDAG::Node & addCast(
         const ActionsDAGPtr & dag,
         const ActionsDAG::Node & node_to_cast,
-        const String & type_name,
+        const DataTypePtr & to_type,
         OriginalToNewNodeMap & node_remap)
 {
-    if (node_to_cast.result_type->getName() == type_name)
+    if (!node_to_cast.result_type->equals(*to_type))
         return node_to_cast;
 
-    Field cast_type_constant_value(type_name);
-
-    ColumnWithTypeAndName column;
-    column.column = DataTypeString().createColumnConst(0, cast_type_constant_value);
-    column.type = std::make_shared<DataTypeString>();
-
-    const auto * cast_type_constant_node = &dag->addColumn(std::move(column));
-    ActionsDAG::NodeRawConstPtrs children = {&node_to_cast, cast_type_constant_node};
-    FunctionOverloadResolverPtr func_builder_cast = createInternalCastOverloadResolver(CastType::nonAccurate, {});
-
-    return addFunction(dag, func_builder_cast, std::move(children), node_remap);
+    const auto & new_node = dag->addCast(node_to_cast, to_type, {});
+    node_remap[new_node.result_name] = {dag.get(), &new_node};
+    return new_node;
 }
 
 /// Normalizes the filter node by adding AND with a constant true.
@@ -332,7 +324,7 @@ bool tryBuildPrewhereSteps(PrewhereInfoPtr prewhere_info, const ExpressionAction
             /// Build AND(last_step_result_node, true)
             const auto & and_node = addAndTrue(last_step_dag, *last_step_result_node_info.node, node_remap);
             /// Build CAST(and_node, type of PREWHERE column)
-            const auto & cast_node = addCast(last_step_dag, and_node, output->result_type->getName(), node_remap);
+            const auto & cast_node = addCast(last_step_dag, and_node, output->result_type, node_remap);
             /// Add alias for the result with the name of the PREWHERE column
             const auto & prewhere_result_node = last_step_dag->addAlias(cast_node, output->result_name);
             last_step_dag->addOrReplaceInOutputs(prewhere_result_node);

src/Storages/StorageJoin.cpp

@@ -546,7 +546,11 @@ protected:
             return {};
 
         Chunk chunk;
-        if (!joinDispatch(join->kind, join->strictness, join->data->maps.front(),
+        if (!joinDispatch(
+                join->kind,
+                join->strictness,
+                join->data->maps.front(),
+                join->table_join->getMixedJoinExpression() != nullptr,
                 [&](auto kind, auto strictness, auto & map) { chunk = createChunk<kind, strictness>(map); }))
             throw Exception(ErrorCodes::LOGICAL_ERROR, "Unknown JOIN strictness");
         return chunk;

src/Storages/System/StorageSystemBuildOptions.cpp.in

@@ -21,7 +21,7 @@ const char * auto_config_build[]
     "BUILD_COMPILE_DEFINITIONS", "@BUILD_COMPILE_DEFINITIONS@",
     "USE_EMBEDDED_COMPILER", "@USE_EMBEDDED_COMPILER@",
     "USE_GLIBC_COMPATIBILITY", "@GLIBC_COMPATIBILITY@",
-    "USE_JEMALLOC", "@ENABLE_JEMALLOC@",
+    "USE_JEMALLOC", "@USE_JEMALLOC@",
     "USE_ICU", "@USE_ICU@",
     "USE_H3", "@USE_H3@",
     "USE_MYSQL", "@USE_MYSQL@",
@@ -36,7 +36,7 @@ const char * auto_config_build[]
     "USE_SSL", "@USE_SSL@",
     "OPENSSL_VERSION", "@OPENSSL_VERSION@",
     "OPENSSL_IS_BORING_SSL", "@OPENSSL_IS_BORING_SSL@",
-    "USE_VECTORSCAN", "@ENABLE_VECTORSCAN@",
+    "USE_VECTORSCAN", "@USE_VECTORSCAN@",
     "USE_SIMDJSON", "@USE_SIMDJSON@",
     "USE_ODBC", "@USE_ODBC@",
     "USE_GRPC", "@USE_GRPC@",
@@ -62,8 +62,8 @@ const char * auto_config_build[]
     "USE_ARROW", "@USE_ARROW@",
     "USE_ORC", "@USE_ORC@",
     "USE_MSGPACK", "@USE_MSGPACK@",
-    "USE_QPL", "@ENABLE_QPL@",
-    "USE_QAT", "@ENABLE_QATLIB@",
+    "USE_QPL", "@USE_QPL@",
+    "USE_QATLIB", "@USE_QATLIB@",
     "GIT_HASH", "@GIT_HASH@",
     "GIT_BRANCH", R"IRjaNsZIL9Yh7FQ4(@GIT_BRANCH@)IRjaNsZIL9Yh7FQ4",
     "GIT_DATE", "@GIT_DATE@",

src/Storages/VirtualColumnUtils.cpp

@@ -1,39 +1,43 @@
 #include <memory>
 #include <stack>
+#include <Columns/ColumnConst.h>
+#include <Columns/ColumnSet.h>
+#include <Columns/ColumnsCommon.h>
+#include <Columns/ColumnsNumber.h>
+#include <Columns/FilterDescription.h>
 #include <Core/NamesAndTypes.h>
 #include <Core/TypeId.h>
-
+#include <DataTypes/DataTypeDateTime.h>
+#include <DataTypes/DataTypeLowCardinality.h>
+#include <DataTypes/DataTypeString.h>
+#include <DataTypes/DataTypesNumber.h>
+#include <Functions/FunctionHelpers.h>
+#include <Functions/FunctionsLogical.h>
+#include <Functions/IFunction.h>
+#include <Functions/IFunctionAdaptors.h>
+#include <Functions/indexHint.h>
+#include <IO/WriteHelpers.h>
+#include <Interpreters/ActionsDAG.h>
+#include <Interpreters/ActionsVisitor.h>
 #include <Interpreters/Context.h>
-#include <Interpreters/TreeRewriter.h>
-#include <Interpreters/ExpressionAnalyzer.h>
 #include <Interpreters/ExpressionActions.h>
+#include <Interpreters/ExpressionAnalyzer.h>
 #include <Interpreters/IdentifierSemantic.h>
+#include <Interpreters/TreeRewriter.h>
 #include <Interpreters/misc.h>
-
-#include <Parsers/ASTIdentifier.h>
 #include <Parsers/ASTExpressionList.h>
-#include <Parsers/ASTLiteral.h>
 #include <Parsers/ASTFunction.h>
+#include <Parsers/ASTIdentifier.h>
+#include <Parsers/ASTLiteral.h>
 #include <Parsers/ASTSelectQuery.h>
 #include <Parsers/ASTSubquery.h>
-
-#include <Columns/ColumnConst.h>
-#include <Columns/ColumnsNumber.h>
-#include <Columns/ColumnsCommon.h>
-#include <Columns/FilterDescription.h>
-
-#include <DataTypes/DataTypesNumber.h>
-#include <DataTypes/DataTypeString.h>
-#include <DataTypes/DataTypeLowCardinality.h>
-#include <DataTypes/DataTypeDateTime.h>
-
-#include <Processors/QueryPlan/QueryPlan.h>
+#include <Parsers/makeASTForLogicalFunction.h>
+#include <Processors/Executors/CompletedPipelineExecutor.h>
 #include <Processors/QueryPlan/BuildQueryPipelineSettings.h>
 #include <Processors/QueryPlan/Optimizations/QueryPlanOptimizationSettings.h>
+#include <Processors/QueryPlan/QueryPlan.h>
 #include <Processors/Sinks/EmptySink.h>
-#include <Processors/Executors/CompletedPipelineExecutor.h>
 #include <QueryPipeline/QueryPipelineBuilder.h>
-
 #include <Storages/VirtualColumnUtils.h>
 #include <IO/WriteHelpers.h>
 #include <Common/re2.h>
@@ -337,9 +341,7 @@ bool isDeterministicInScopeOfQuery(const ActionsDAG::Node * node)
 }
 
 static const ActionsDAG::Node * splitFilterNodeForAllowedInputs(
-    const ActionsDAG::Node * node,
-    const Block * allowed_inputs,
-    ActionsDAG::Nodes & additional_nodes)
+    const ActionsDAG::Node * node, const Block * allowed_inputs, ActionsDAG::Nodes & additional_nodes, bool allow_partial_result)
 {
     if (node->type == ActionsDAG::ActionType::FUNCTION)
     {
@@ -348,8 +350,15 @@ static const ActionsDAG::Node * splitFilterNodeForAllowedInputs(
             auto & node_copy = additional_nodes.emplace_back(*node);
             node_copy.children.clear();
             for (const auto * child : node->children)
-                if (const auto * child_copy = splitFilterNodeForAllowedInputs(child, allowed_inputs, additional_nodes))
+                if (const auto * child_copy
+                    = splitFilterNodeForAllowedInputs(child, allowed_inputs, additional_nodes, allow_partial_result))
                     node_copy.children.push_back(child_copy);
+                /// Expression like (now_allowed AND allowed) is not allowed if allow_partial_result = true. This is important for
+                /// trivial count optimization, otherwise we can get incorrect results. For example, if the query is
+                /// SELECT count() FROM table WHERE _partition_id = '0' AND rowNumberInBlock() = 1, we cannot apply
+                /// trivial count.
+                else if (!allow_partial_result)
+                    return nullptr;
 
             if (node_copy.children.empty())
                 return nullptr;
@@ -357,7 +366,7 @@ static const ActionsDAG::Node * splitFilterNodeForAllowedInputs(
             if (node_copy.children.size() == 1)
             {
                 const ActionsDAG::Node * res = node_copy.children.front();
-                /// Expression like (not_allowed AND 256) can't be resuced to (and(256)) because AND requires
+                /// Expression like (not_allowed AND 256) can't be reduced to (and(256)) because AND requires
                 /// at least two arguments; also it can't be reduced to (256) because result type is different.
                 if (!res->result_type->equals(*node->result_type))
                 {
@@ -375,7 +384,7 @@ static const ActionsDAG::Node * splitFilterNodeForAllowedInputs(
         {
             auto & node_copy = additional_nodes.emplace_back(*node);
             for (auto & child : node_copy.children)
-                if (child = splitFilterNodeForAllowedInputs(child, allowed_inputs, additional_nodes); !child)
+                if (child = splitFilterNodeForAllowedInputs(child, allowed_inputs, additional_nodes, allow_partial_result); !child)
                     return nullptr;
 
             return &node_copy;
@@ -389,7 +398,8 @@ static const ActionsDAG::Node * splitFilterNodeForAllowedInputs(
                     auto index_hint_dag = index_hint->getActions().clone();
                     ActionsDAG::NodeRawConstPtrs atoms;
                     for (const auto & output : index_hint_dag.getOutputs())
-                        if (const auto * child_copy = splitFilterNodeForAllowedInputs(output, allowed_inputs, additional_nodes))
+                        if (const auto * child_copy
+                            = splitFilterNodeForAllowedInputs(output, allowed_inputs, additional_nodes, allow_partial_result))
                             atoms.push_back(child_copy);
 
                     if (!atoms.empty())
@@ -423,22 +433,24 @@ static const ActionsDAG::Node * splitFilterNodeForAllowedInputs(
     return node;
 }
 
-std::optional<ActionsDAG> splitFilterDagForAllowedInputs(const ActionsDAG::Node * predicate, const Block * allowed_inputs)
+std::optional<ActionsDAG>
+splitFilterDagForAllowedInputs(const ActionsDAG::Node * predicate, const Block * allowed_inputs, bool allow_partial_result)
 {
     if (!predicate)
         return {};
 
     ActionsDAG::Nodes additional_nodes;
-    const auto * res = splitFilterNodeForAllowedInputs(predicate, allowed_inputs, additional_nodes);
+    const auto * res = splitFilterNodeForAllowedInputs(predicate, allowed_inputs, additional_nodes, allow_partial_result);
     if (!res)
         return {};
 
     return ActionsDAG::cloneSubDAG({res}, true);
 }
 
-void filterBlockWithPredicate(const ActionsDAG::Node * predicate, Block & block, ContextPtr context)
+void filterBlockWithPredicate(
+    const ActionsDAG::Node * predicate, Block & block, ContextPtr context, bool allow_filtering_with_partial_predicate)
 {
-    auto dag = splitFilterDagForAllowedInputs(predicate, &block);
+    auto dag = splitFilterDagForAllowedInputs(predicate, &block, /*allow_partial_result=*/allow_filtering_with_partial_predicate);
     if (dag)
         filterBlockWithExpression(buildFilterExpression(std::move(*dag), context), block);
 }

src/Storages/VirtualColumnUtils.h

@@ -27,9 +27,13 @@ namespace VirtualColumnUtils
 ///
 /// Otherwise calling filter*() outside applyFilters() will throw "Not-ready Set is passed"
 /// if there are subqueries.
+///
+/// Similar to filterBlockWithExpression(buildFilterExpression(splitFilterDagForAllowedInputs(...)))./// Similar to filterBlockWithQuery, but uses ActionsDAG as a predicate.
+/// Basically it is filterBlockWithDAG(splitFilterDagForAllowedInputs).
+/// If allow_filtering_with_partial_predicate is true, then the filtering will be done even if some part of the predicate
+/// cannot be evaluated using the columns from the block.
+void filterBlockWithPredicate(const ActionsDAG::Node * predicate, Block & block, ContextPtr context, bool allow_filtering_with_partial_predicate = true);
 
-/// Similar to filterBlockWithExpression(buildFilterExpression(splitFilterDagForAllowedInputs(...))).
-void filterBlockWithPredicate(const ActionsDAG::Node * predicate, Block & block, ContextPtr context);
 
 /// Just filters block. Block should contain all the required columns.
 ExpressionActionsPtr buildFilterExpression(ActionsDAG dag, ContextPtr context);
@@ -42,7 +46,15 @@ void buildSetsForDAG(const ActionsDAG & dag, const ContextPtr & context);
 bool isDeterministicInScopeOfQuery(const ActionsDAG::Node * node);
 
 /// Extract a part of predicate that can be evaluated using only columns from input_names.
-std::optional<ActionsDAG> splitFilterDagForAllowedInputs(const ActionsDAG::Node * predicate, const Block * allowed_inputs);
+/// When allow_partial_result is false, then the result will be empty if any part of if cannot be evaluated deterministically
+/// on the given inputs.
+/// allow_partial_result must be false when we are going to use the result to filter parts in
+/// MergeTreeData::totalRowsByPartitionPredicateImp. For example, if the query is
+/// `SELECT count() FROM table  WHERE _partition_id = '0' AND rowNumberInBlock() = 1`
+/// The predicate will be `_partition_id = '0' AND rowNumberInBlock() = 1`, and `rowNumberInBlock()` is
+/// non-deterministic. If we still extract the part `_partition_id = '0'` for filtering parts, then trivial
+/// count optimization will be mistakenly applied to the query.
+std::optional<ActionsDAG> splitFilterDagForAllowedInputs(const ActionsDAG::Node * predicate, const Block * allowed_inputs, bool allow_partial_result = true);
 
 /// Extract from the input stream a set of `name` column values
 template <typename T>

src/configure_config.cmake

@@ -135,6 +135,12 @@ endif()
 if (TARGET ch_contrib::vectorscan)
     set(USE_VECTORSCAN 1)
 endif()
+if (TARGET ch_contrib::qpl)
+    set(USE_QPL 1)
+endif()
+if (TARGET ch_contrib::qatlib)
+    set(USE_QATLIB 1)
+endif()
 if (TARGET ch_contrib::avrocpp)
     set(USE_AVRO 1)
 endif()
@@ -161,8 +167,8 @@ endif()
 if (TARGET ch_contrib::ssh)
     set(USE_SSH 1)
 endif()
-if (TARGET ch_contrib::fiu)
-    set(FIU_ENABLE 1)
+if (TARGET ch_contrib::libfiu)
+    set(USE_LIBFIU 1)
 endif()
 if (TARGET ch_contrib::libarchive)
     set(USE_LIBARCHIVE 1)

tests/ci/ci.py

@@ -985,6 +985,7 @@ def _run_test(job_name: str, run_command: str) -> int:
     else:
         print("Use run command from the workflow")
     env["CHECK_NAME"] = job_name
+    env["MAX_RUN_TIME"] = str(timeout or 0)
     print(f"Going to start run command [{run_command}]")
     stopwatch = Stopwatch()
     job_log = Path(TEMP_PATH) / "job_log.txt"

tests/ci/functional_test_check.py

@@ -114,6 +114,9 @@ def get_run_command(
     if flaky_check:
         envs.append("-e NUM_TRIES=50")
         envs.append("-e MAX_RUN_TIME=2800")
+    else:
+        max_run_time = os.getenv("MAX_RUN_TIME", "0")
+        envs.append(f"-e MAX_RUN_TIME={max_run_time}")
 
     envs += [f"-e {e}" for e in additional_envs]
 

tests/ci/report.py

@@ -738,7 +738,7 @@ def create_test_html_report(
     if test_results:
         rows_part = []
         num_fails = 0
-        has_test_time = False
+        has_test_time = any(tr.time is not None for tr in test_results)
         has_log_urls = False
 
         # Display entires with logs at the top (they correspond to failed tests)
@@ -770,9 +770,11 @@ def create_test_html_report(
             row.append(f'<td {fail_id}style="{style}">{test_result.status}</td>')
             colspan += 1
 
-            if test_result.time is not None:
-                has_test_time = True
-                row.append(f"<td>{test_result.time}</td>")
+            if has_test_time:
+                if test_result.time is not None:
+                    row.append(f"<td>{test_result.time}</td>")
+                else:
+                    row.append("<td></td>")
                 colspan += 1
 
             if test_result.log_urls is not None:

tests/integration/test_broken_projections/config.d/dont_start_broken.xml

@@ -0,0 +1,6 @@
+<?xml version="1.0" encoding="utf-8"?>
+<clickhouse>
+  <merge_tree>
+    <max_suspicious_broken_parts_bytes>0</max_suspicious_broken_parts_bytes>
+  </merge_tree>
+</clickhouse>

tests/integration/test_broken_projections/test.py

@@ -4,6 +4,7 @@ import logging
 import string
 import random
 from helpers.cluster import ClickHouseCluster
+from multiprocessing.dummy import Pool
 
 cluster = ClickHouseCluster(__file__)
 
@@ -18,6 +19,12 @@ def cluster():
             stay_alive=True,
             with_zookeeper=True,
         )
+        cluster.add_instance(
+            "node_restart",
+            main_configs=["config.d/dont_start_broken.xml"],
+            stay_alive=True,
+            with_zookeeper=True,
+        )
 
         logging.info("Starting cluster...")
         cluster.start()
@@ -632,6 +639,49 @@ def test_broken_on_start(cluster):
     check(node, table_name, 0)
 
 
+def test_disappeared_projection_on_start(cluster):
+    node = cluster.instances["node_restart"]
+
+    table_name = "test_disapperead_projection"
+    create_table(node, table_name, 1)
+
+    node.query(f"SYSTEM STOP MERGES {table_name}")
+
+    insert(node, table_name, 0, 5)
+    insert(node, table_name, 5, 5)
+    insert(node, table_name, 10, 5)
+    insert(node, table_name, 15, 5)
+
+    assert ["all_0_0_0", "all_1_1_0", "all_2_2_0", "all_3_3_0"] == get_parts(
+        node, table_name
+    )
+
+    def drop_projection():
+        node.query(
+            f"ALTER TABLE {table_name} DROP PROJECTION proj2",
+            settings={"mutations_sync": "0"},
+        )
+
+    p = Pool(2)
+    p.apply_async(drop_projection)
+
+    for i in range(30):
+        create_query = node.query(f"SHOW CREATE TABLE {table_name}")
+        if "proj2" not in create_query:
+            break
+        time.sleep(0.5)
+
+    assert "proj2" not in create_query
+
+    # Remove 'proj2' for part all_2_2_0
+    break_projection(node, table_name, "proj2", "all_2_2_0", "part")
+
+    node.restart_clickhouse()
+
+    # proj2 is not broken, it doesn't exist, but ok
+    check(node, table_name, 0, expect_broken_part="proj2", do_check_command=0)
+
+
 def test_mutation_with_broken_projection(cluster):
     node = cluster.instances["node"]
 

tests/integration/test_drop_is_lock_free/test.py

@@ -176,7 +176,7 @@ def test_query_is_permanent(transaction, permanent, exclusive_table):
 
     select_handler = node.get_query_request(
         f"""
-            SELECT sleepEachRow(3) FROM {exclusive_table} SETTINGS function_sleep_max_microseconds_per_block = 0;
+            SELECT sleepEachRow(3) FROM {exclusive_table} SETTINGS function_sleep_max_microseconds_per_block = 0, max_threads=1;
         """,
         query_id=query_id,
     )

tests/integration/test_storage_s3_queue/test.py

@@ -771,7 +771,11 @@ def test_multiple_tables_streaming_sync_distributed(started_cluster, mode):
             table_name,
             mode,
             files_path,
-            additional_settings={"keeper_path": keeper_path, "s3queue_buckets": 2},
+            additional_settings={
+                "keeper_path": keeper_path,
+                "s3queue_buckets": 2,
+                **({"s3queue_processing_threads_num": 1} if mode == "ordered" else {}),
+            },
         )
 
     for instance in [node, node_2]:
@@ -806,6 +810,10 @@ def test_multiple_tables_streaming_sync_distributed(started_cluster, mode):
         list(map(int, l.split())) for l in run_query(node_2, get_query).splitlines()
     ]
 
+    logging.debug(
+        f"res1 size: {len(res1)}, res2 size: {len(res2)}, total_rows: {total_rows}"
+    )
+
     assert len(res1) + len(res2) == total_rows
 
     # Checking that all engines have made progress

tests/performance/optimize_functions_to_subcolumns.xml

@@ -0,0 +1,26 @@
+<test>
+    <settings>
+        <max_insert_threads>4</max_insert_threads>
+    </settings>
+
+    <create_query>
+        CREATE TABLE t_subcolumns (a Array(UInt64), s Nullable(String), m Map(String, UInt64)) ENGINE = MergeTree ORDER BY tuple()
+    </create_query>
+
+    <fill_query>
+        INSERT INTO t_subcolumns SELECT range(number % 20), toString(number), mapFromArrays(range(number % 20), range(number % 20)) FROM numbers_mt(50000000)
+    </fill_query>
+
+    <fill_query>
+        OPTIMIZE TABLE t_subcolumns FINAL
+    </fill_query>
+
+    <query>SELECT count() FROM t_subcolumns WHERE NOT ignore(length(a))</query>
+    <query>SELECT count() FROM t_subcolumns WHERE notEmpty(a)</query>
+    <query>SELECT count() FROM t_subcolumns WHERE NOT ignore(length(m))</query>
+    <query>SELECT count() FROM t_subcolumns WHERE notEmpty(m)</query>
+    <query>SELECT count() FROM t_subcolumns WHERE isNotNull(s)</query>
+    <query>SELECT count(s) FROM t_subcolumns</query>
+
+    <drop_query>DROP TABLE t_subcolumns</drop_query>
+</test>

tests/queries/0_stateless/00900_long_parquet_load.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-# Tags: long, no-fasttest, no-debug
+# Tags: long, no-fasttest, no-debug, no-asan, no-msan, no-tsan
 
 #
 # Load all possible .parquet files found in submodules.

tests/queries/0_stateless/01246_buffer_flush.sh

@@ -27,7 +27,7 @@ function wait_until()
 function get_buffer_delay()
 {
     local buffer_insert_id=$1 && shift
-    query "SYSTEM FLUSH LOGS"
+    $CLICKHOUSE_CLIENT -q "SYSTEM FLUSH LOGS"
     query "
         WITH
             (SELECT event_time_microseconds FROM system.query_log WHERE current_database = '$CLICKHOUSE_DATABASE' AND type = 'QueryStart' AND query_id = '$buffer_insert_id') AS begin_,

tests/queries/0_stateless/01600_parts_states_metrics_long.sh

@@ -1,10 +1,16 @@
 #!/usr/bin/env bash
-# Tags: long
+# Tags: long, no-parallel
 
 CURDIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
 # shellcheck source=../shell_config.sh
 . "$CURDIR"/../shell_config.sh
 
+function query()
+{
+    # NOTE: database_atomic_wait_for_drop_and_detach_synchronously needed only for local env, CI has it ON
+    ${CLICKHOUSE_CURL} -sS "${CLICKHOUSE_URL}&database_atomic_wait_for_drop_and_detach_synchronously=1" -d "$*"
+}
+
 # NOTE: database = $CLICKHOUSE_DATABASE is unwanted
 verify_sql="SELECT
     (SELECT sumIf(value, metric = 'PartsActive'), sumIf(value, metric = 'PartsOutdated') FROM system.metrics)
@@ -18,13 +24,13 @@ verify()
 {
     for i in {1..5000}
     do
-        result=$( $CLICKHOUSE_CLIENT --query="$verify_sql" )
+        result=$( query "$verify_sql" )
         [ "$result" = "1" ] && echo "$result" && break
         sleep 0.1
 
         if [[ $i -eq 5000 ]]
         then
-            $CLICKHOUSE_CLIENT "
+            query "
               SELECT sumIf(value, metric = 'PartsActive'), sumIf(value, metric = 'PartsOutdated') FROM system.metrics;
               SELECT sum(active), sum(NOT active) FROM system.parts;
               SELECT sum(active), sum(NOT active) FROM system.projection_parts;
@@ -34,17 +40,17 @@ verify()
     done
 }
 
-$CLICKHOUSE_CLIENT --database_atomic_wait_for_drop_and_detach_synchronously=1 --query="DROP TABLE IF EXISTS test_table"
-$CLICKHOUSE_CLIENT --query="CREATE TABLE test_table (data Date) ENGINE = MergeTree PARTITION BY toYear(data) ORDER BY data;"
+query "DROP TABLE IF EXISTS test_table"
+query "CREATE TABLE test_table (data Date) ENGINE = MergeTree PARTITION BY toYear(data) ORDER BY data;"
 
-$CLICKHOUSE_CLIENT --query="INSERT INTO test_table VALUES ('1992-01-01')"
+query "INSERT INTO test_table VALUES ('1992-01-01')"
 verify
 
-$CLICKHOUSE_CLIENT --query="INSERT INTO test_table VALUES ('1992-01-02')"
+query "INSERT INTO test_table VALUES ('1992-01-02')"
 verify
 
-$CLICKHOUSE_CLIENT --query="OPTIMIZE TABLE test_table FINAL"
+query "OPTIMIZE TABLE test_table FINAL"
 verify
 
-$CLICKHOUSE_CLIENT --database_atomic_wait_for_drop_and_detach_synchronously=1 --query="DROP TABLE test_table"
+query "DROP TABLE test_table"
 verify

tests/queries/0_stateless/01656_test_query_log_factories_info.reference

@@ -17,7 +17,7 @@ used_functions
 ['repeat']
 
 arraySort(used_data_type_families)
-['Array','Int32','Nullable','String']
+['Int32','Nullable','String']
 
 used_database_engines
 ['Atomic']

tests/queries/0_stateless/01710_aggregate_projection_with_normalized_states.reference

@@ -1,2 +1,3 @@
 3
 950	990	500	2000
+[950]	[999]

tests/queries/0_stateless/01710_aggregate_projection_with_normalized_states.sql

@@ -29,4 +29,6 @@ FROM cluster('test_cluster_two_shards', currentDatabase(), r)
 WHERE a = 'x'
 settings prefer_localhost_replica=0;
 
+SELECT quantilesTimingMerge(0.95)(q), quantilesTimingMerge(toInt64(1))(q) FROM remote('127.0.0.{1,2}', currentDatabase(), r);
+
 DROP TABLE r;

tests/queries/0_stateless/01710_projection_vertical_merges.sql

@@ -1,4 +1,5 @@
--- Tags: long
+-- Tags: long, no-parallel
+-- set no-parallel tag is to prevent timeout of this test
 
 drop table if exists t;
 

tests/queries/0_stateless/02122_join_group_by_timeout.reference

@@ -1,4 +1,6 @@
-Code: 159
-0
+Code: 159
+query_duration	1
+0
+query_duration	1
 Code: 159
 0

tests/queries/0_stateless/02122_join_group_by_timeout.sh

@@ -1,27 +1,23 @@
 #!/usr/bin/env bash
-# Tags: no-debug
-
-# no-debug: Query is canceled by timeout after max_execution_time,
-#           but sending an exception to the client may hang
-#           for more than MAX_PROCESS_WAIT seconds in a slow debug build,
-#           and test will fail.
 
 CURDIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
 # shellcheck source=../shell_config.sh
 . "$CURDIR"/../shell_config.sh
 
-MAX_PROCESS_WAIT=5
-
-IS_SANITIZER=$($CLICKHOUSE_CLIENT -q "SELECT count() FROM system.warnings WHERE message like '%built with sanitizer%'")
-if [ "$IS_SANITIZER" -gt 0 ]; then
-    # Query may hang for more than 5 seconds, especially in tsan build
-    MAX_PROCESS_WAIT=15
+TIMEOUT=5
+IS_SANITIZER_OR_DEBUG=$($CLICKHOUSE_CLIENT -q "SELECT count() FROM system.warnings WHERE message like '%built with sanitizer%' or message like '%built in debug mode%'")
+if [ "$IS_SANITIZER_OR_DEBUG" -gt 0 ]; then
+    # Increase the timeout due to in debug/sanitizers build:
+    # - client is slow
+    # - stacktrace resolving is slow
+    TIMEOUT=15
 fi
 
 # TCP CLIENT: As of today (02/12/21) uses PullingAsyncPipelineExecutor
 ### Should be cancelled after 1 second and return a 159 exception (timeout)
-timeout -s KILL $MAX_PROCESS_WAIT $CLICKHOUSE_CLIENT --max_execution_time 1 -q \
-    "SELECT * FROM
+query_id=$(random_str 12)
+$CLICKHOUSE_CLIENT --query_id "$query_id" --max_execution_time 1 -q "
+    SELECT * FROM
     (
         SELECT a.name as n
         FROM
@@ -34,28 +30,35 @@ timeout -s KILL $MAX_PROCESS_WAIT $CLICKHOUSE_CLIENT --max_execution_time 1 -q \
         GROUP BY n
     )
     LIMIT 20
-    FORMAT Null" 2>&1 | grep -o "Code: 159" | sort | uniq
+    FORMAT Null
+" 2>&1 | grep -m1 -o "Code: 159"
+$CLICKHOUSE_CLIENT -q "system flush logs"
+${CLICKHOUSE_CURL} -q -sS "$CLICKHOUSE_URL" -d "select 'query_duration', round(query_duration_ms/1000) from system.query_log where current_database = '$CLICKHOUSE_DATABASE' and query_id = '$query_id' and type != 'QueryStart'"
+
 
 ### Should stop pulling data and return what has been generated already (return code 0)
-timeout -s KILL $MAX_PROCESS_WAIT $CLICKHOUSE_CLIENT -q \
-    "SELECT a.name as n
-     FROM
-     (
-         SELECT 'Name' as name, number FROM system.numbers LIMIT 2000000
-     ) AS a,
-     (
-         SELECT 'Name' as name2, number FROM system.numbers LIMIT 2000000
-     ) as b
-     FORMAT Null
-     SETTINGS max_execution_time = 1, timeout_overflow_mode = 'break'
-    "
+query_id=$(random_str 12)
+$CLICKHOUSE_CLIENT --query_id "$query_id" -q "
+    SELECT a.name as n
+    FROM
+    (
+        SELECT 'Name' as name, number FROM system.numbers LIMIT 2000000
+    ) AS a,
+    (
+        SELECT 'Name' as name2, number FROM system.numbers LIMIT 2000000
+    ) as b
+    FORMAT Null
+    SETTINGS max_execution_time = 1, timeout_overflow_mode = 'break'
+"
 echo $?
+$CLICKHOUSE_CLIENT -q "system flush logs"
+${CLICKHOUSE_CURL} -q -sS "$CLICKHOUSE_URL" -d "select 'query_duration', round(query_duration_ms/1000) from system.query_log where current_database = '$CLICKHOUSE_DATABASE' and query_id = '$query_id' and type != 'QueryStart'"
 
 
 # HTTP CLIENT: As of today (02/12/21) uses PullingPipelineExecutor
 ### Should be cancelled after 1 second and return a 159 exception (timeout)
-${CLICKHOUSE_CURL} -q --max-time $MAX_PROCESS_WAIT -sS "$CLICKHOUSE_URL&max_execution_time=1" -d \
-    "SELECT * FROM
+${CLICKHOUSE_CURL} -q --max-time $TIMEOUT -sS "$CLICKHOUSE_URL&max_execution_time=1" -d "
+    SELECT * FROM
     (
         SELECT a.name as n
         FROM
@@ -68,12 +71,13 @@ ${CLICKHOUSE_CURL} -q --max-time $MAX_PROCESS_WAIT -sS "$CLICKHOUSE_URL&max_exec
         GROUP BY n
     )
     LIMIT 20
-    FORMAT Null" 2>&1 | grep -o "Code: 159" | sort | uniq
+    FORMAT Null
+" 2>&1 | grep -o "Code: 159" | sort | uniq
 
 
 ### Should stop pulling data and return what has been generated already (return code 0)
-${CLICKHOUSE_CURL} -q --max-time $MAX_PROCESS_WAIT -sS "$CLICKHOUSE_URL" -d \
-    "SELECT a.name as n
+${CLICKHOUSE_CURL} -q --max-time $TIMEOUT -sS "$CLICKHOUSE_URL" -d "
+    SELECT a.name as n
           FROM
           (
               SELECT 'Name' as name, number FROM system.numbers LIMIT 2000000
@@ -83,5 +87,5 @@ ${CLICKHOUSE_CURL} -q --max-time $MAX_PROCESS_WAIT -sS "$CLICKHOUSE_URL" -d \
           ) as b
           FORMAT Null
           SETTINGS max_execution_time = 1, timeout_overflow_mode = 'break'
-    "
+"
 echo $?

tests/queries/0_stateless/02496_remove_redundant_sorting.reference

@@ -465,6 +465,37 @@ Expression ((Projection + Before ORDER BY))
             ReadFromStorage (SystemOne)
 -- execute
 Float64	9007199254740994
+-- presence of an inner OFFSET retains the ORDER BY
+-- query
+WITH
+  t1 AS (
+    SELECT a, b
+    FROM
+      VALUES (
+        'b UInt32, a Int32',
+        (1, 1),
+        (2, 0)
+      )
+  )
+SELECT
+  SUM(a)
+FROM (
+  SELECT a, b
+  FROM t1
+  ORDER BY 1 DESC, 2
+  OFFSET 1
+) t2
+-- explain
+Expression ((Projection + Before ORDER BY))
+  Aggregating
+    Expression (Before GROUP BY)
+      Offset
+        Expression (Projection)
+          Sorting (Sorting for ORDER BY)
+            Expression ((Before ORDER BY + (Projection + Before ORDER BY)))
+              ReadFromStorage (Values)
+-- execute
+0
 -- disable common optimization to avoid functions to be lifted up (liftUpFunctions optimization), needed for testing with stateful function
 -- neighbor() as stateful function prevents removing inner ORDER BY since its result depends on order
 -- query

tests/queries/0_stateless/02496_remove_redundant_sorting.sh

@@ -302,6 +302,27 @@ FROM
 )"
 run_query "$query"
 
+echo "-- presence of an inner OFFSET retains the ORDER BY"
+query="WITH
+  t1 AS (
+    SELECT a, b
+    FROM
+      VALUES (
+        'b UInt32, a Int32',
+        (1, 1),
+        (2, 0)
+      )
+  )
+SELECT
+  SUM(a)
+FROM (
+  SELECT a, b
+  FROM t1
+  ORDER BY 1 DESC, 2
+  OFFSET 1
+) t2"
+run_query "$query"
+
 echo "-- disable common optimization to avoid functions to be lifted up (liftUpFunctions optimization), needed for testing with stateful function"
 ENABLE_OPTIMIZATION="SET query_plan_enable_optimizations=0;$ENABLE_OPTIMIZATION"
 echo "-- neighbor() as stateful function prevents removing inner ORDER BY since its result depends on order"

tests/queries/0_stateless/02496_remove_redundant_sorting_analyzer.reference

@@ -464,6 +464,36 @@ Expression ((Project names + Projection))
             ReadFromStorage (SystemOne)
 -- execute
 Float64	9007199254740994
+-- presence of an inner OFFSET retains the ORDER BY
+-- query
+WITH
+  t1 AS (
+    SELECT a, b
+    FROM
+      VALUES (
+        'b UInt32, a Int32',
+        (1, 1),
+        (2, 0)
+      )
+  )
+SELECT
+  SUM(a)
+FROM (
+  SELECT a, b
+  FROM t1
+  ORDER BY 1 DESC, 2
+  OFFSET 1
+) t2
+-- explain
+Expression ((Project names + Projection))
+  Aggregating
+    Expression ((Before GROUP BY + (Change column names to column identifiers + Project names)))
+      Offset
+        Sorting (Sorting for ORDER BY)
+          Expression ((Before ORDER BY + (Projection + (Change column names to column identifiers + (Project names + (Projection + Change column names to column identifiers))))))
+            ReadFromStorage (Values)
+-- execute
+0
 -- disable common optimization to avoid functions to be lifted up (liftUpFunctions optimization), needed for testing with stateful function
 -- neighbor() as stateful function prevents removing inner ORDER BY since its result depends on order
 -- query

tests/queries/0_stateless/03006_join_on_inequal_expression_fast.reference

@@ -382,6 +382,253 @@ key1	e	5	5	5	key1	C	3	4	5
 key2	a2	1	1	1			0	0	\N
 key4	f	2	3	4	key4	F	1	1	1
 SET join_algorithm='hash';
+SELECT t1.*, t2.* FROM t1 LEFT ANY JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	c	3	2	1	key1	C	3	4	5
+key1	d	4	7	2	key1	D	4	1	6
+key1	e	5	5	5			0	0	\N
+key2	a2	1	1	1			0	0	\N
+key4	f	2	3	4	key4	F	1	1	1
+SELECT t1.*, t2.* from t1 LEFT ANY  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	c	3	2	1	key1	B	2	1	2
+key1	d	4	7	2	key1	D	4	1	6
+key1	e	5	5	5			0	0	\N
+key2	a2	1	1	1			0	0	\N
+key4	f	2	3	4			0	0	\N
+SELECT t1.*, t2.* from t1 LEFT ANY  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	B	2	1	2
+key1	b	2	3	2	key1	C	3	4	5
+key1	c	3	2	1	key1	D	4	1	6
+key1	d	4	7	2			0	0	\N
+key1	e	5	5	5			0	0	\N
+key2	a2	1	1	1			0	0	\N
+key4	f	2	3	4			0	0	\N
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 LEFT ANY JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+1	1	1	1	1	1
+SELECT t1.*, t2.* FROM t1 LEFT SEMI JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	c	3	2	1	key1	C	3	4	5
+key1	d	4	7	2	key1	D	4	1	6
+key4	f	2	3	4	key4	F	1	1	1
+SELECT t1.*, t2.* from t1 LEFT SEMI  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	c	3	2	1	key1	B	2	1	2
+key1	d	4	7	2	key1	D	4	1	6
+SELECT t1.*, t2.* from t1 LEFT SEMI  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	B	2	1	2
+key1	b	2	3	2	key1	C	3	4	5
+key1	c	3	2	1	key1	D	4	1	6
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 LEFT SEMI JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+1	1	1	1	1	1
+SELECT t1.*, t2.* FROM t1 LEFT ANTI JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	e	5	5	5	key1		0	0	\N
+key2	a2	1	1	1	key2		0	0	\N
+SELECT t1.*, t2.* from t1 LEFT ANTI  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	e	5	5	5	key1		0	0	\N
+key2	a2	1	1	1	key2		0	0	\N
+key4	f	2	3	4	key4		0	0	\N
+SELECT t1.*, t2.* from t1 LEFT ANTI  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	d	4	7	2	key1		0	0	\N
+key1	e	5	5	5	key1		0	0	\N
+key2	a2	1	1	1	key2		0	0	\N
+key4	f	2	3	4	key4		0	0	\N
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 LEFT ANTI JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+SELECT t1.*, t2.* FROM t1 RIGHT ANY JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+		0	0	\N	key3	a3	1	1	1
+key1	a	1	1	2	key1	A	1	2	1
+key1	a	1	1	2	key1	B	2	1	2
+key1	a	1	1	2	key1	C	3	4	5
+key1	a	1	1	2	key1	D	4	1	6
+key4	f	2	3	4	key4	F	1	1	1
+SELECT t1.*, t2.* from t1 RIGHT ANY  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+		0	0	\N	key3	a3	1	1	1
+		0	0	\N	key4	F	1	1	1
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	b	2	3	2	key1	C	3	4	5
+key1	d	4	7	2	key1	D	4	1	6
+SELECT t1.*, t2.* from t1 RIGHT ANY  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+		0	0	\N	key1	A	1	2	1
+		0	0	\N	key3	a3	1	1	1
+		0	0	\N	key4	F	1	1	1
+key1	a	1	1	2	key1	B	2	1	2
+key1	a	1	1	2	key1	C	3	4	5
+key1	a	1	1	2	key1	D	4	1	6
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 RIGHT ANY JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+1	1	1	1	1	1
+SELECT t1.*, t2.* FROM t1 RIGHT SEMI JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	a	1	1	2	key1	B	2	1	2
+key1	a	1	1	2	key1	C	3	4	5
+key1	a	1	1	2	key1	D	4	1	6
+key4	f	2	3	4	key4	F	1	1	1
+SELECT t1.*, t2.* from t1 RIGHT SEMI  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	b	2	3	2	key1	C	3	4	5
+key1	d	4	7	2	key1	D	4	1	6
+SELECT t1.*, t2.* from t1 RIGHT SEMI  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	B	2	1	2
+key1	a	1	1	2	key1	C	3	4	5
+key1	a	1	1	2	key1	D	4	1	6
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 RIGHT SEMI JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+1	1	1	1	1	1
+SELECT t1.*, t2.* FROM t1 RIGHT ANTI JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+		0	0	\N	key3	a3	1	1	1
+SELECT t1.*, t2.* from t1 RIGHT ANTI  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+		0	0	\N	key3	a3	1	1	1
+		0	0	\N	key4	F	1	1	1
+SELECT t1.*, t2.* from t1 RIGHT ANTI  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+		0	0	\N	key1	A	1	2	1
+		0	0	\N	key3	a3	1	1	1
+		0	0	\N	key4	F	1	1	1
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 RIGHT ANTI JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+SET join_algorithm='grace_hash';
+SELECT t1.*, t2.* FROM t1 LEFT ANY JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	c	3	2	1	key1	C	3	4	5
+key1	d	4	7	2	key1	D	4	1	6
+key1	e	5	5	5			0	0	\N
+key2	a2	1	1	1			0	0	\N
+key4	f	2	3	4	key4	F	1	1	1
+SELECT t1.*, t2.* from t1 LEFT ANY  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	c	3	2	1	key1	B	2	1	2
+key1	d	4	7	2	key1	D	4	1	6
+key1	e	5	5	5			0	0	\N
+key2	a2	1	1	1			0	0	\N
+key4	f	2	3	4			0	0	\N
+SELECT t1.*, t2.* from t1 LEFT ANY  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	B	2	1	2
+key1	b	2	3	2	key1	C	3	4	5
+key1	c	3	2	1	key1	D	4	1	6
+key1	d	4	7	2			0	0	\N
+key1	e	5	5	5			0	0	\N
+key2	a2	1	1	1			0	0	\N
+key4	f	2	3	4			0	0	\N
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 LEFT ANY JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+1	1	1	1	1	1
+SELECT t1.*, t2.* FROM t1 LEFT SEMI JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	c	3	2	1	key1	C	3	4	5
+key1	d	4	7	2	key1	D	4	1	6
+key4	f	2	3	4	key4	F	1	1	1
+SELECT t1.*, t2.* from t1 LEFT SEMI  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	c	3	2	1	key1	B	2	1	2
+key1	d	4	7	2	key1	D	4	1	6
+SELECT t1.*, t2.* from t1 LEFT SEMI  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	B	2	1	2
+key1	b	2	3	2	key1	C	3	4	5
+key1	c	3	2	1	key1	D	4	1	6
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 LEFT SEMI JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+1	1	1	1	1	1
+SELECT t1.*, t2.* FROM t1 LEFT ANTI JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	e	5	5	5	key1		0	0	\N
+key2	a2	1	1	1	key2		0	0	\N
+SELECT t1.*, t2.* from t1 LEFT ANTI  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	e	5	5	5	key1		0	0	\N
+key2	a2	1	1	1	key2		0	0	\N
+key4	f	2	3	4	key4		0	0	\N
+SELECT t1.*, t2.* from t1 LEFT ANTI  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	d	4	7	2	key1		0	0	\N
+key1	e	5	5	5	key1		0	0	\N
+key2	a2	1	1	1	key2		0	0	\N
+key4	f	2	3	4	key4		0	0	\N
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 LEFT ANTI JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+SELECT t1.*, t2.* FROM t1 RIGHT ANY JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+		0	0	\N	key3	a3	1	1	1
+key1	a	1	1	2	key1	A	1	2	1
+key1	a	1	1	2	key1	B	2	1	2
+key1	a	1	1	2	key1	C	3	4	5
+key1	a	1	1	2	key1	D	4	1	6
+key4	f	2	3	4	key4	F	1	1	1
+SELECT t1.*, t2.* from t1 RIGHT ANY  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+		0	0	\N	key3	a3	1	1	1
+		0	0	\N	key4	F	1	1	1
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	b	2	3	2	key1	C	3	4	5
+key1	d	4	7	2	key1	D	4	1	6
+SELECT t1.*, t2.* from t1 RIGHT ANY  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+		0	0	\N	key1	A	1	2	1
+		0	0	\N	key3	a3	1	1	1
+		0	0	\N	key4	F	1	1	1
+key1	a	1	1	2	key1	B	2	1	2
+key1	a	1	1	2	key1	C	3	4	5
+key1	a	1	1	2	key1	D	4	1	6
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 RIGHT ANY JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+1	1	1	1	1	1
+SELECT t1.*, t2.* FROM t1 RIGHT SEMI JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	a	1	1	2	key1	B	2	1	2
+key1	a	1	1	2	key1	C	3	4	5
+key1	a	1	1	2	key1	D	4	1	6
+key4	f	2	3	4	key4	F	1	1	1
+SELECT t1.*, t2.* from t1 RIGHT SEMI  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	b	2	3	2	key1	C	3	4	5
+key1	d	4	7	2	key1	D	4	1	6
+SELECT t1.*, t2.* from t1 RIGHT SEMI  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	B	2	1	2
+key1	a	1	1	2	key1	C	3	4	5
+key1	a	1	1	2	key1	D	4	1	6
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 RIGHT SEMI JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+1	1	1	1	1	1
+SELECT t1.*, t2.* FROM t1 RIGHT ANTI JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+		0	0	\N	key3	a3	1	1	1
+SELECT t1.*, t2.* from t1 RIGHT ANTI  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+		0	0	\N	key3	a3	1	1	1
+		0	0	\N	key4	F	1	1	1
+SELECT t1.*, t2.* from t1 RIGHT ANTI  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+		0	0	\N	key1	A	1	2	1
+		0	0	\N	key3	a3	1	1	1
+		0	0	\N	key4	F	1	1	1
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 RIGHT ANTI JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+SET join_algorithm='hash';
+SELECT t1.* FROM t1 LEFT ANY JOIN t2 ON t1.key = t2.key AND t1.a < t2.a OR t1.a = t2.a ORDER BY ALL;
+key1	a	1	1	2
+key1	b	2	3	2
+key1	c	3	2	1
+key1	d	4	7	2
+key1	e	5	5	5
+key2	a2	1	1	1
+key4	f	2	3	4
+SELECT t1.* FROM t1 LEFT SEMI JOIN t2 ON t1.key = t2.key AND t1.a < t2.a OR t1.a = t2.a ORDER BY ALL;
+key1	a	1	1	2
+key1	b	2	3	2
+key1	c	3	2	1
+key1	d	4	7	2
+key2	a2	1	1	1
+key4	f	2	3	4
+SELECT t1.* FROM t1 LEFT ANTI JOIN t2 ON t1.key = t2.key AND t1.a < t2.a OR t1.a = t2.a ORDER BY ALL;
+key1	e	5	5	5
+SELECT t1.* FROM t1 RIGHT ANY JOIN t2 ON t1.key = t2.key AND t1.a < t2.a OR t1.a = t2.a ORDER BY ALL;
+key1	a	1	1	2
+key1	a	1	1	2
+key1	a	1	1	2
+key1	a	1	1	2
+key1	a	1	1	2
+key1	a	1	1	2
+SELECT t1.* FROM t1 RIGHT SEMI JOIN t2 ON t1.key = t2.key AND t1.a < t2.a OR t1.a = t2.a ORDER BY ALL;
+key1	a	1	1	2
+key1	a	1	1	2
+key1	a	1	1	2
+key1	a	1	1	2
+key1	a	1	1	2
+key1	a	1	1	2
+SELECT t1.* FROM t1 RIGHT ANTI JOIN t2 ON t1.key = t2.key AND t1.a < t2.a OR t1.a = t2.a ORDER BY ALL;
+SET join_algorithm='hash';
 SELECT t1.*, t2.* FROM t1 LEFT JOIN t2 ON t1.key = t2.key AND t1.a < t2.a OR t1.a = t2.a ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
 key1	a	1	1	2	key1	A	1	2	1
 key1	a	1	1	2	key1	B	2	1	2
@@ -452,3 +699,46 @@ key2	a2	1	1	1	key1	A	1	2	1
 key2	a2	1	1	1	key3	a3	1	1	1
 key2	a2	1	1	1	key4	F	1	1	1
 key4	f	2	3	4	key1	B	2	1	2
+SET join_algorithm='hash';
+SELECT t1.*, t2.* FROM t1 INNER ANY JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	c	3	2	1	key1	C	3	4	5
+key1	d	4	7	2	key1	D	4	1	6
+key4	f	2	3	4	key4	F	1	1	1
+SELECT t1.*, t2.* from t1 INNER ANY  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	c	3	2	1	key1	B	2	1	2
+key1	d	4	7	2	key1	D	4	1	6
+SELECT t1.*, t2.* from t1 INNER ANY  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	B	2	1	2
+key1	b	2	3	2	key1	C	3	4	5
+key1	c	3	2	1	key1	D	4	1	6
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 INNER ANY JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+1	1	1	1	1	1
+SET join_algorithm='grace_hash';
+SELECT t1.*, t2.* FROM t1 INNER ANY JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	c	3	2	1	key1	C	3	4	5
+key1	d	4	7	2	key1	D	4	1	6
+key4	f	2	3	4	key4	F	1	1	1
+SELECT t1.*, t2.* from t1 INNER ANY  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	A	1	2	1
+key1	b	2	3	2	key1	B	2	1	2
+key1	c	3	2	1	key1	B	2	1	2
+key1	d	4	7	2	key1	D	4	1	6
+SELECT t1.*, t2.* from t1 INNER ANY  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+key1	a	1	1	2	key1	B	2	1	2
+key1	b	2	3	2	key1	C	3	4	5
+key1	c	3	2	1	key1	D	4	1	6
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 INNER ANY JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+1	1	1	1	1	1
+SET join_algorithm='hash';
+SELECT t1.* FROM t1 INNER ANY JOIN t2 ON t1.key = t2.key AND t1.a < t2.a OR t1.a = t2.a ORDER BY ALL;
+key1	a	1	1	2
+key1	b	2	3	2
+key1	c	3	2	1
+key1	d	4	7	2
+key2	a2	1	1	1

tests/queries/0_stateless/03006_join_on_inequal_expression_fast.sql.j2

@@ -22,6 +22,26 @@ SELECT t1.*, t2.* FROM t1 {{ join_type }} JOIN t2 ON t1.key = t2.key AND (t1.a=2
 {% endfor -%}
 {% endfor -%}
 
+{% for algorithm in ['hash', 'grace_hash'] -%}
+SET join_algorithm='{{ algorithm }}';
+{% for join_type in ['LEFT', 'RIGHT'] -%}
+{% for join_strictness in ['ANY', 'SEMI', 'ANTI'] -%}
+SELECT t1.*, t2.* FROM t1 {{ join_type }} {{ join_strictness }} JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+SELECT t1.*, t2.* from t1 {{ join_type }} {{ join_strictness }}  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+SELECT t1.*, t2.* from t1 {{ join_type }} {{ join_strictness }}  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 {{ join_type }} {{ join_strictness }} JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+{% endfor -%}
+{% endfor -%}
+{% endfor -%}
+
+{% for algorithm in ['hash'] -%}
+SET join_algorithm='{{ algorithm }}';
+{% for join_type in ['LEFT', 'RIGHT'] -%}
+{% for join_strictness in ['ANY', 'SEMI', 'ANTI'] -%}
+SELECT t1.* FROM t1 {{ join_type }} {{ join_strictness }} JOIN t2 ON t1.key = t2.key AND t1.a < t2.a OR t1.a = t2.a ORDER BY ALL;
+{% endfor -%}
+{% endfor -%}
+{% endfor -%}
 
 {% for algorithm in ['hash'] -%}
 SET join_algorithm='{{ algorithm }}';
@@ -29,6 +49,28 @@ SET join_algorithm='{{ algorithm }}';
 SELECT t1.*, t2.* FROM t1 {{ join_type }} JOIN t2 ON t1.key = t2.key AND t1.a < t2.a OR t1.a = t2.a ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
 {% endfor -%}
 {% endfor -%}
+
+{% for algorithm in ['hash', 'grace_hash'] -%}
+SET join_algorithm='{{ algorithm }}';
+{% for join_type in ['INNER'] -%}
+{% for join_strictness in ['ANY'] -%}
+SELECT t1.*, t2.* FROM t1 {{ join_type }} {{ join_strictness }} JOIN t2 ON (t1.a < t2.a OR lower(t1.attr) == lower(t2.attr)) AND t1.key = t2.key ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+SELECT t1.*, t2.* from t1 {{ join_type }} {{ join_strictness }}  JOIN t2 ON t1.key = t2.key and (t1.b + t2.b == t1.c + t2.c) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+SELECT t1.*, t2.* from t1 {{ join_type }} {{ join_strictness }}  JOIN t2 ON t1.key = t2.key and (t1.a < t2.a) ORDER BY (t1.key, t1.attr, t2.key, t2.attr);
+SELECT * FROM (SELECT 1 AS a, 1 AS b, 1 AS c) AS t1 {{ join_type }} {{ join_strictness }} JOIN (SELECT 1 AS a, 1 AS b, 1 AS c) AS t2 ON t1.a = t2.a AND (t1.b > 0 OR t2.b > 0);
+{% endfor -%}
+{% endfor -%}
+{% endfor -%}
+
+{% for algorithm in ['hash'] -%}
+SET join_algorithm='{{ algorithm }}';
+{% for join_type in ['INNER'] -%}
+{% for join_strictness in ['ANY'] -%}
+SELECT t1.* FROM t1 {{ join_type }} {{ join_strictness }} JOIN t2 ON t1.key = t2.key AND t1.a < t2.a OR t1.a = t2.a ORDER BY ALL;
+{% endfor -%}
+{% endfor -%}
+{% endfor -%}
+
 -- { echoOff }
 
 -- test error messages

tests/queries/0_stateless/03203_count_with_non_deterministic_function.reference

@@ -0,0 +1,2 @@
+1
+1

tests/queries/0_stateless/03203_count_with_non_deterministic_function.sql

@@ -0,0 +1,4 @@
+CREATE TABLE t (p UInt8, x UInt64) Engine = MergeTree PARTITION BY p ORDER BY x;
+INSERT INTO t SELECT 0, number FROM numbers(10) SETTINGS max_block_size = 100;
+SELECT count() FROM t WHERE p = 0 AND rowNumberInAllBlocks() = 1 SETTINGS allow_experimental_analyzer = 0;
+SELECT count() FROM t WHERE p = 0 AND rowNumberInAllBlocks() = 1 SETTINGS allow_experimental_analyzer = 1;

tests/queries/0_stateless/03215_multilinestring_geometry.reference

@@ -0,0 +1,30 @@
+-- { echoOn }
+SELECT readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3, 1 1))');
+[[(1,1),(2,2),(3,3),(1,1)]]
+SELECT toTypeName(readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3, 1 1))'));
+MultiLineString
+SELECT wkt(readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3, 1 1))'));
+MULTILINESTRING((1 1,2 2,3 3,1 1))
+SELECT readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3, 1 1), (1 0, 2 0, 3 0))');
+[[(1,1),(2,2),(3,3),(1,1)],[(1,0),(2,0),(3,0)]]
+SELECT toTypeName(readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3, 1 1), (1 0, 2 0, 3 0))'));
+MultiLineString
+SELECT wkt(readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3, 1 1), (1 0, 2 0, 3 0))'));
+MULTILINESTRING((1 1,2 2,3 3,1 1),(1 0,2 0,3 0))
+-- Native Array(Array(Tuple(Float64, Float64))) is treated as Polygon, not as MultiLineString.
+WITH wkt(CAST([[(1, 1), (2, 2), (3, 3), (1, 1)]], 'Array(Array(Tuple(Float64, Float64)))')) as x
+SELECT x, toTypeName(x), readWKTPolygon(x) as y, toTypeName(y);
+POLYGON((1 1,2 2,3 3,1 1))	String	[[(1,1),(2,2),(3,3),(1,1)]]	Polygon
+-- Non constant tests
+
+DROP TABLE IF EXISTS t;
+CREATE TABLE IF NOT EXISTS t (shape Array(Array(Tuple(Float64, Float64))), wkt_string String, ord Float64) Engine = Memory;
+INSERT INTO t (ord, shape, wkt_string) VALUES (1, [[(1, 1), (2, 2), (3, 3), (1, 1)]], 'MULTILINESTRING ((1 1, 2 2, 3 3, 1 1))');
+INSERT INTO t (ord, shape, wkt_string) VALUES (2, [[(1, 1), (2, 2), (3, 3), (1, 1)], [(1, 0), (2, 0), (3, 0)]], 'MULTILINESTRING ((1 1, 2 2, 3 3, 1 1), (1 0, 2 0, 3 0))');
+INSERT INTO t (ord, shape, wkt_string) VALUES (3, [[(1, 0), (2, 1), (3, 0), (4, 1), (5, 0), (6, 1), (7, 0), (8, 1), (9, 0), (10, 1)]], 'MULTILINESTRING ((1 0, 2 1, 3 0, 4 1, 5 0, 6 1, 7 0, 8 1, 9 0, 10 1))');
+-- Native Array(Array(Tuple(Float64, Float64))) is treated as Polygon, not as MultiLineString.
+-- but reading MultiLineString should still return an Array(Array(Tuple(Float64, Float64)))
+select wkt(shape), readWKTMultiLineString(wkt_string), readWKTMultiLineString(wkt_string) = shape from t order by ord;
+POLYGON((1 1,2 2,3 3,1 1))	[[(1,1),(2,2),(3,3),(1,1)]]	1
+POLYGON((1 1,2 2,3 3,1 1),(1 0,2 0,3 0,1 0))	[[(1,1),(2,2),(3,3),(1,1)],[(1,0),(2,0),(3,0)]]	1
+POLYGON((1 0,2 1,3 0,4 1,5 0,6 1,7 0,8 1,9 0,10 1,1 0))	[[(1,0),(2,1),(3,0),(4,1),(5,0),(6,1),(7,0),(8,1),(9,0),(10,1)]]	1

tests/queries/0_stateless/03215_multilinestring_geometry.sql

@@ -0,0 +1,26 @@
+-- { echoOn }
+SELECT readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3, 1 1))');
+SELECT toTypeName(readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3, 1 1))'));
+SELECT wkt(readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3, 1 1))'));
+
+SELECT readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3, 1 1), (1 0, 2 0, 3 0))');
+SELECT toTypeName(readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3, 1 1), (1 0, 2 0, 3 0))'));
+SELECT wkt(readWKTMultiLineString('MULTILINESTRING ((1 1, 2 2, 3 3, 1 1), (1 0, 2 0, 3 0))'));
+
+-- Native Array(Array(Tuple(Float64, Float64))) is treated as Polygon, not as MultiLineString.
+WITH wkt(CAST([[(1, 1), (2, 2), (3, 3), (1, 1)]], 'Array(Array(Tuple(Float64, Float64)))')) as x
+SELECT x, toTypeName(x), readWKTPolygon(x) as y, toTypeName(y);
+
+-- Non constant tests
+
+DROP TABLE IF EXISTS t;
+CREATE TABLE IF NOT EXISTS t (shape Array(Array(Tuple(Float64, Float64))), wkt_string String, ord Float64) Engine = Memory;
+INSERT INTO t (ord, shape, wkt_string) VALUES (1, [[(1, 1), (2, 2), (3, 3), (1, 1)]], 'MULTILINESTRING ((1 1, 2 2, 3 3, 1 1))');
+INSERT INTO t (ord, shape, wkt_string) VALUES (2, [[(1, 1), (2, 2), (3, 3), (1, 1)], [(1, 0), (2, 0), (3, 0)]], 'MULTILINESTRING ((1 1, 2 2, 3 3, 1 1), (1 0, 2 0, 3 0))');
+INSERT INTO t (ord, shape, wkt_string) VALUES (3, [[(1, 0), (2, 1), (3, 0), (4, 1), (5, 0), (6, 1), (7, 0), (8, 1), (9, 0), (10, 1)]], 'MULTILINESTRING ((1 0, 2 1, 3 0, 4 1, 5 0, 6 1, 7 0, 8 1, 9 0, 10 1))');
+
+-- Native Array(Array(Tuple(Float64, Float64))) is treated as Polygon, not as MultiLineString.
+-- but reading MultiLineString should still return an Array(Array(Tuple(Float64, Float64)))
+select wkt(shape), readWKTMultiLineString(wkt_string), readWKTMultiLineString(wkt_string) = shape from t order by ord;
+
+

tests/queries/0_stateless/03217_filtering_in_storage_merge.reference

@@ -0,0 +1,6 @@
+Expression ((Project names + Projection))
+  Aggregating
+    Expression (Before GROUP BY)
+      ReadFromMerge
+        Filter (( + ( + )))
+          ReadFromMergeTree (default.test_03217_merge_replica_1)

tests/queries/0_stateless/03217_filtering_in_storage_merge.sql

@@ -0,0 +1,16 @@
+CREATE TABLE test_03217_merge_replica_1(x UInt32)
+    ENGINE ReplicatedMergeTree('/clickhouse/tables/{database}/test_03217_merge_replica', 'r1')
+    ORDER BY x;
+CREATE TABLE test_03217_merge_replica_2(x UInt32)
+    ENGINE ReplicatedMergeTree('/clickhouse/tables/{database}/test_03217_merge_replica', 'r2')
+    ORDER BY x;
+
+
+CREATE TABLE test_03217_all_replicas (x UInt32)
+    ENGINE = Merge(currentDatabase(), 'test_03217_merge_replica_*');
+
+INSERT INTO test_03217_merge_replica_1 SELECT number AS x FROM numbers(10);
+SYSTEM SYNC REPLICA test_03217_merge_replica_2;
+
+-- If the filter on _table is not applied, then the plan will show both replicas
+EXPLAIN SELECT _table, count() FROM test_03217_all_replicas WHERE  _table = 'test_03217_merge_replica_1' AND x >= 0 GROUP BY _table SETTINGS allow_experimental_analyzer=1;

tests/queries/0_stateless/03217_filtering_in_system_tables.reference

@@ -0,0 +1,6 @@
+information_schema	tables
+both	default	test_03217_system_tables_replica_1	r1
+both	default	test_03217_system_tables_replica_2	r2
+default	test_03217_system_tables_replica_1	r1
+1
+1