Remove hard-coded paths in Interpreters/Compiler.cpp

Interpreters/Compiler.cpp contained hard-coded paths for system's
includes needed for query compiler. These paths were not portable
between different Linux distros and gcc/clang versions. For example,
Debian/Ubuntu use /usr/lib/gcc/x86_64-linux-gnu/*/include,
RHEL/Fedora use /usr/lib/gcc/x86_64-redhat-linux/*/include,
others use /usr/lib/gcc/*/include (without x86_64-XXX triplet).

Patch 68850012b "Embedded compiler fixes" attempted to fix this problem
by adding CMAKE_LIBRARY_ARCHITECTURE after /usr/lib. Unfortunally,
CMAKE_LIBRARY_ARCHITECTURE is not defined on RHEL/Fedora because someone
decided to omit "-gnu" from x86_64-redhat-linux (see RHBZ#1531678).

Patch 70e35d0bc "Build fixes (#1718)" added a workaround for
undefined CMAKE_LIBRARY_ARCHITECTURE on RHEL/Fedora, but hasn't fixed
problem with missing /usr/lib/gcc/x86_64-redhat-linux/*/include/
in the list of hardcoded paths.

Remove hard-coded paths and get the list of `-isystem` includes directly
from bundled clickhouse-clang.

Other changes:

- Enable RPATH for the build directory to get working binaries
  without installing them by `make install`.

.gitignore

@@ -187,6 +187,7 @@ debian/files
 dbms/src/Server/data/*
 dbms/src/Server/metadata/*
 dbms/src/Server/status
+dbms/src/Interpreters/CompilerIncludes.h
 config-9001.xml
 
 *-preprocessed.xml

CMakeLists.txt

@@ -79,6 +79,13 @@ if (USE_STATIC_LIBRARIES)
     list(REVERSE CMAKE_FIND_LIBRARY_SUFFIXES)
 endif ()
 
+# Don't skip the full RPATH for the build tree to get working binaries
+# without running `make install`.
+set (CMAKE_SKIP_BUILD_RPATH OFF)
+
+# The RPATH to be used when installing
+set (CMAKE_INSTALL_RPATH "")
+
 if (CMAKE_LIBRARY_ARCHITECTURE MATCHES "amd64.*|x86_64.*|AMD64.*")
     option (USE_INTERNAL_MEMCPY "Use internal implementation of 'memcpy' function instead of provided by libc. Only for x86_64." ON)
 

dbms/CMakeLists.txt

@@ -88,6 +88,9 @@ list (APPEND dbms_headers
 list (APPEND dbms_sources src/TableFunctions/TableFunctionFactory.cpp)
 list (APPEND dbms_headers src/TableFunctions/ITableFunction.h src/TableFunctions/TableFunctionFactory.h)
 
+if (USE_EMBEDDED_COMPILER)
+    list (APPEND dbms_headers ${CMAKE_CURRENT_SOURCE_DIR}/src/Interpreters/CompilerIncludes.h)
+endif ()
 
 list(REMOVE_ITEM clickhouse_common_io_sources
     src/Common/StringUtils.cpp)

dbms/scripts/gen-compiler-includes.sh

@@ -0,0 +1,16 @@
+#!/bin/sh
+#
+# Generates the list of -isystem includes from gcc or clang compiler
+# Usage: ./gen-includes.sh [COMPILER]
+#
+
+# Set local to C to capture GCC's "search starts here" text in English
+export LANG=C
+export LC_ALL=C
+export LC_MESSAGES=C
+CC=$1 || cc
+if test -z "$CC"; then CC=cc; fi
+
+echo "" | $CC -x c++ -E -Wp,-v - 2>&1|
+    sed -n '/#include <...> search starts here:/,/End of search list./p' |
+    sed -e '1d;$d' -e 's/^[ \t ]*/" -isystem /;s/$/"/'

dbms/src/Interpreters/CMakeLists.txt

@@ -5,12 +5,21 @@ else ()
     set (PATH_SHARE "/usr/share" CACHE STRING "")
 endif ()
 
+if (USE_EMBEDDED_COMPILER)
+    # Generate CompilerIncludes.h which contains all
+    # `-isystem` includes for bundled clickhouse-clang
+    add_custom_command(OUTPUT CompilerIncludes.h
+                       COMMAND ${PROJECT_SOURCE_DIR}/dbms/scripts/gen-compiler-includes.sh
+                               ${PROJECT_BINARY_DIR}/dbms/src/Server/clickhouse-clang
+                               > ${CMAKE_CURRENT_BINARY_DIR}/CompilerIncludes.h
+                       DEPENDS clickhouse-clang)
+endif ()
+
 set (INTERNAL_COMPILER_EXECUTABLE "clickhouse-clang" CACHE STRING "")
 set (INTERNAL_LINKER_EXECUTABLE "clickhouse-lld" CACHE STRING "")
 set (INTERNAL_COMPILER_NO_WARNING OFF CACHE INTERNAL "")
 set (INTERNAL_COMPILER_HEADERS "${PATH_SHARE}/clickhouse/headers" CACHE STRING "")
 set (INTERNAL_COMPILER_HEADERS_ROOT "${INTERNAL_COMPILER_HEADERS}" CACHE STRING "")
-set (INTERNAL_COMPILER_CUSTOM_ROOT ON CACHE INTERNAL "")
 
 set (INTERNAL_COMPILER_FLAGS "${CMAKE_CXX_FLAGS} ${CMAKE_CXX_FLAGS_RELEASE} ${CXX_FLAGS_INTERNAL_COMPILER} -x c++ -march=native -fPIC -fvisibility=hidden -fno-implement-inlines -nostdinc -nostdinc++ -Wno-unused-command-line-argument -Bprefix=${PATH_SHARE}/clickhouse -isysroot=${INTERNAL_COMPILER_HEADERS_ROOT}" CACHE STRING "")
 string(REPLACE "-no-pie" "" INTERNAL_COMPILER_FLAGS ${INTERNAL_COMPILER_FLAGS})

dbms/src/Interpreters/Compiler.cpp

@@ -222,23 +222,10 @@ void Compiler::compile(
             INTERNAL_COMPILER_EXECUTABLE
             " " INTERNAL_COMPILER_FLAGS
 
-
+            #if USE_EMBEDDED_COMPILER
-    #if INTERNAL_COMPILER_CUSTOM_ROOT
+            /// Contains `echo "" | clickhouse-clang -x c++ -E -Wp,-v -`
-            /// To get correct order merge this results carefully:
+            #include "CompilerIncludes.h"
-            /// echo | clang -x c++ -E -Wp,-v -
+            #endif /* USE_EMBEDDED_COMPILER */
-            /// echo | g++ -x c++ -E -Wp,-v -
-
-            " -isystem " INTERNAL_COMPILER_HEADERS_ROOT "/usr/include/c++/*"
-            " -isystem " INTERNAL_COMPILER_HEADERS_ROOT "/usr/include/" CMAKE_LIBRARY_ARCHITECTURE "/c++/*"
-            " -isystem " INTERNAL_COMPILER_HEADERS_ROOT "/usr/include/c++/*/backward"
-            " -isystem " INTERNAL_COMPILER_HEADERS_ROOT "/usr/include/clang/*/include"                  /// if compiler is clang (from package)
-            " -isystem " INTERNAL_COMPILER_HEADERS_ROOT "/usr/local/lib/clang/*/include"                /// if clang installed manually
-            " -isystem " INTERNAL_COMPILER_HEADERS_ROOT "/usr/lib/gcc/" CMAKE_LIBRARY_ARCHITECTURE "/*/include-fixed"
-            " -isystem " INTERNAL_COMPILER_HEADERS_ROOT "/usr/lib/gcc/" CMAKE_LIBRARY_ARCHITECTURE "/*/include"
-            " -isystem " INTERNAL_COMPILER_HEADERS_ROOT "/usr/local/include"                            /// if something installed manually
-            " -isystem " INTERNAL_COMPILER_HEADERS_ROOT "/usr/include/" CMAKE_LIBRARY_ARCHITECTURE
-            " -isystem " INTERNAL_COMPILER_HEADERS_ROOT "/usr/include"
-    #endif
             " -I " INTERNAL_COMPILER_HEADERS "/dbms/src/"
             " -I " INTERNAL_COMPILER_HEADERS "/contrib/libcityhash/include/"
             " -I " INTERNAL_COMPILER_HEADERS "/contrib/libpcg-random/include/"

dbms/src/Interpreters/config_compile.h.in

@@ -1,16 +1,12 @@
 #pragma once
 
-#cmakedefine CMAKE_LIBRARY_ARCHITECTURE "@CMAKE_LIBRARY_ARCHITECTURE@"
-#if !defined(CMAKE_LIBRARY_ARCHITECTURE)
-#define CMAKE_LIBRARY_ARCHITECTURE ""
-#endif
 #cmakedefine PATH_SHARE "@PATH_SHARE@"
 #cmakedefine INTERNAL_COMPILER_FLAGS "@INTERNAL_COMPILER_FLAGS@"
 #cmakedefine INTERNAL_COMPILER_EXECUTABLE "@INTERNAL_COMPILER_EXECUTABLE@"
 #cmakedefine INTERNAL_LINKER_EXECUTABLE "@INTERNAL_LINKER_EXECUTABLE@"
 #cmakedefine INTERNAL_COMPILER_HEADERS "@INTERNAL_COMPILER_HEADERS@"
 #cmakedefine INTERNAL_COMPILER_HEADERS_ROOT "@INTERNAL_COMPILER_HEADERS_ROOT@"
-#cmakedefine01 INTERNAL_COMPILER_CUSTOM_ROOT
+#cmakedefine01 USE_EMBEDDED_COMPILER
 #cmakedefine INTERNAL_DOUBLE_CONVERSION_INCLUDE_DIR "@INTERNAL_DOUBLE_CONVERSION_INCLUDE_DIR@"
 #cmakedefine INTERNAL_Poco_Foundation_INCLUDE_DIR "@INTERNAL_Poco_Foundation_INCLUDE_DIR@"
 #cmakedefine INTERNAL_Poco_Util_INCLUDE_DIR "@INTERNAL_Poco_Util_INCLUDE_DIR@"

dbms/src/Server/CMakeLists.txt

@@ -46,6 +46,10 @@ target_link_libraries (clickhouse-format-lib clickhouse_common_io ${Boost_PROGRA
 if (USE_EMBEDDED_COMPILER)
     link_directories (${LLVM_LIBRARY_DIRS})
     add_subdirectory ("Compiler-${LLVM_VERSION}")
+    # Link clickhouse-clang binary even if CLICKHOUSE_SPLIT_BINARY is OFF
+    # in order to get system include paths for Interpreters/Compiler.cpp
+    add_executable (clickhouse-clang clickhouse-clang.cpp)
+    target_link_libraries (clickhouse-clang clickhouse-compiler-lib)
 endif ()
 
 if (CLICKHOUSE_SPLIT_BINARY)
@@ -70,8 +74,7 @@ if (CLICKHOUSE_SPLIT_BINARY)
     set (CLICKHOUSE_ALL_TARGETS clickhouse-server clickhouse-client clickhouse-local clickhouse-benchmark clickhouse-performance-test clickhouse-extract-from-config clickhouse-format)
 
     if (USE_EMBEDDED_COMPILER)
-        add_executable (clickhouse-clang clickhouse-clang.cpp)
+        # clickhouse-clang is linked above
-        target_link_libraries (clickhouse-clang clickhouse-compiler-lib)
         add_executable (clickhouse-lld clickhouse-lld.cpp)
         target_link_libraries (clickhouse-lld clickhouse-compiler-lib)
         install (TARGETS clickhouse-clang clickhouse-lld RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR} COMPONENT clickhouse)