mirror of
https://github.com/MariaDB/server.git
synced 2025-01-16 03:52:35 +01:00
e904d0e750
The patch implements the idea suggested by Olaf van der Spek in thread "Client: many small reads?" (internals@lists.mysql.com). Now small reads performed by the client library are buffered. The buffering gives up to 2 times speedup when retrieving one-column tables. BUILD/SETUP.sh: Remove --with-vio option which no longer exist. BUILD/compile-pentium64-valgrind-max: Remove --with-vio option which no longer exist. config/ac-macros/misc.m4: Removed --with-vio configure switch: we always use VIO. The switch, in fact, only saved us one pointer dereferencing per call in case we had only one transport type in VIO enabled. config/ac-macros/openssl.m4: Removed HAVE_VIO. include/config-win.h: Removed HAVE_VIO (not needed anymore) Added HAVE_VIO_READ_BUFF (define buffered client reads for Windows clients). include/violite.h: Removed HAVE_VIO, as currently VIO is always in use. Added declaration for vio_read_buff and related members in struct VIO. sql-common/client.c: Use flags to set up vio read buffering in mysql_real_connect. sql/mysqld.cc: Use flags to disable vio read buffering when creating a server connection. vio/vio.c: Optionally set up vio read buffer when creating a new VIO structure. vio/viosocket.c: Implementation of client-side buffered reads in VIO: the idea is to buffer small reads in a client buffer to save amount of syscalls per retrieved result set. The implementation relies on the fact that read/recv will return as soon as there is some data in the kernel buffer, no matter how big the given user's buffer is. To be able to disable it in case recv/read don't have such semantics, the new calls are guarded with #define HAVE_VIO_READ_BUFF. Currently buffered reading is switched on only for BSD sockets and named pipes, both on Windows and UNIXes.
29 lines
1.2 KiB
Bash
29 lines
1.2 KiB
Bash
#! /bin/sh
|
|
|
|
path=`dirname $0`
|
|
. "$path/SETUP.sh"
|
|
|
|
extra_flags="$pentium64_cflags $debug_cflags -USAFEMALLOC -UFORCE_INIT_OF_VARS -DHAVE_purify -DMYSQL_SERVER_SUFFIX=-valgrind-max"
|
|
c_warnings="$c_warnings $debug_extra_warnings"
|
|
cxx_warnings="$cxx_warnings $debug_extra_warnings"
|
|
extra_configs="$pentium_configs $debug_configs"
|
|
|
|
# We want to test isam when building with valgrind
|
|
extra_configs="$extra_configs --with-berkeley-db --with-innodb --with-isam --with-embedded-server --with-openssl --with-raid --with-ndbcluster"
|
|
|
|
. "$path/FINISH.sh"
|
|
|
|
if test -z "$just_print"
|
|
then
|
|
set +v +x
|
|
echo "\
|
|
******************************************************************************
|
|
Note that by default BUILD/compile-pentium-valgrind-max calls 'configure' with
|
|
--enable-assembler. When Valgrind detects an error involving an assembly
|
|
function (for example an uninitialized value used as an argument of an
|
|
assembly function), Valgrind will not print the stacktrace and 'valgrind
|
|
--gdb-attach=yes' will not work either. If you need a stacktrace in those
|
|
cases, you have to run BUILD/compile-pentium-valgrind-max with the
|
|
--disable-assembler argument.
|
|
******************************************************************************"
|
|
fi
|