To compile just the MySQL client libraries and client programs and not the server, use the -without-server option:

shell ./configure --without-server

If you don't have a C++ compiler, mysql will not compile (it is the one client program that requires C++). In this case, you can remove the code in configure that tests for the C++ compiler and then run ./configure with the -without-server option. The compile step will still try to build mysql, but you can ignore any warnings about mysql.cc. (If make stops, try make -k to tell it to continue with the rest of the build even if errors occur.)

If you want to get an embedded MySQL library (libmysqld.a) you should use the -with-embedded-server option.

If you don't want your log files and database directories located under /usr/local/var, use a configure command, something like one of these:

shell ./configure --prefix=/usr/local/mysql
shell ./configure --prefix=/usr/local \
           --localstatedir=/usr/local/mysql/data

The first command changes the installation prefix so that everything is installed under /usr/local/mysql rather than the default of /usr/local. The second command preserves the default installation prefix, but overrides the default location for database directories (normally /usr/local/var) and changes it to /usr/local/mysql/data. After you have compiled MySQL, you can change these options with option files. Section 4.1.2, “my.cnf Option Files ”.

If you are using Unix and you want the MySQL socket located somewhere other than the default location (normally in the directory /tmp or /var/run) use a configure command like this:

shell ./configure --with-unix-socket-path=/usr/local/mysql/tmp/mysql.sock

Note that the given file must be an absolute pathname. You can also later change the location mysql.sock by using the MySQL option files. Section A.4.5, “How to Protect or Change the MySQL Socket File /tmp/mysql.sock”.

If you want to compile statically linked programs (for example, to make a binary distribution, to get more speed, or to work around problems with some Red Hat Linux distributions), run configure like this:

shell ./configure --with-client-ldflags=-all-static \
           --with-mysqld-ldflags=-all-static

If you are using gcc and don't have libg++ or libstdc++ installed, you can tell configure to use gcc as your C++ compiler:

shell CC=gcc CXX=gcc ./configure

When you use gcc as your C++ compiler, it will not attempt to link in libg++ or libstdc++. This may be a good idea to do even if you have the above libraries installed, as some versions of these libraries have caused strange problems for MySQL users in the past.

Here are some common environment variables to set depending on the compiler you are using:

In most cases you can get a reasonably optimal MySQL binary by using the options from the preceding table and adding the following options to the configure line:

--prefix=/usr/local/mysql --enable-assembler \
--with-mysqld-ldflags=-all-static

The full configure line would, in other words, be something like the following for all recent gcc versions:

CFLAGS="-O3 -mpentiumpro" CXX=gcc CXXFLAGS="-O3 -mpentiumpro \
-felide-constructors -fno-exceptions -fno-rtti" ./configure \
--prefix=/usr/local/mysql --enable-assembler \
--with-mysqld-ldflags=-all-static

The binaries we provide on the MySQL web site at http://www.mysql.com/ are all compiled with full optimisation and should be perfect for most users. Section 2.2.10, “MySQL Binaries Compiled by MySQL AB ”. There are some things you can tweak to make an even faster binary, but this is only for advanced users. Section 5.5.3, “How Compiling and Linking Affects the Speed of MySQL ”.

If the build fails and produces errors about your compiler or linker not being able to create the shared library libmysqlclient.so.# (# is a version number), you can work around this problem by giving the -disable-shared option to configure. In this case, configure will not build a shared libmysqlclient.so.# library.

You can configure MySQL not to use DEFAULT column values for non-NULL columns (that is, columns that are not allowed to be NULL). Section 1.7.5.2, “Constraint NOT NULL and DEFAULT values ”.

shell CXXFLAGS=-DDONT_USE_DEFAULT_FIELDS ./configure

By default, MySQL uses the ISO-8859-1 (Latin1) character set. To change the default set, use the -with-charset option:

shell ./configure --with-charset=CHARSET

CHARSET may be one of big5, cp1251, cp1257, czech, danish, dec8, dos, euc_kr, gb2312, gbk, german1, hebrew, hp8, hungarian, koi8_ru, koi8_ukr, latin1, latin2, sjis, swe7, tis620, ujis, usa7, or win1251ukr. Section 4.6.1, “The Character Set Used for Data and Sorting ”.

If you want to convert characters between the server and the client, you should take a look at the SET CHARACTER SET command. Section 5.5.6, “SET Syntax ”.

Warning: If you change character sets after having created any tables, you will have to run myisamchk -r -q -set-character-set=charset on every table. Your indexes may be sorted incorrectly otherwise. (This can happen if you install MySQL, create some tables, then reconfigure MySQL to use a different character set and reinstall it.)

With the option -with-extra-charsets=LIST you can define which additional character sets should be compiled into the server.

Here LIST is either a list of character sets separated with spaces, complex to include all characters that can't be dynamically loaded, or all to include all character sets into the binaries.

To configure MySQL with debugging code, use the -with-debug option:

shell ./configure --with-debug

This causes a safe memory allocator to be included that can find some errors and that provides output about what is happening. Section E.1, “Debugging a MySQL server ”.

If your client programs are using threads, you need to also compile a thread-safe version of the MySQL client library with the -enable-thread-safe-client configure options. This will create a libmysqlclient_r library with which you should link your threaded applications. Section 9.1.14, “How to Make a Threaded Client ”.

Options that pertain to particular systems can be found in the system-specific section of this manual. Section 2.6, “Operating System Specific Notes ”.

If configure is run after it already has been run, it may use information that was gathered during its previous invocation. This information is stored in config.cache. When configure starts up, it looks for that file and reads its contents if it exists, on the assumption that the information is still correct. That assumption is invalid when you reconfigure.

Each time you run configure, you must run make again to recompile. However, you may want to remove old object files from previous builds first because they were compiled using different configuration options.

If you get errors when compiling sql_yacc.cc, such as the ones shown here, you have probably run out of memory or swap space:

Internal compiler error: program cc1plus got fatal signal 11
  or
Out of virtual memory
  or
Virtual memory exhausted

The problem is that gcc requires huge amounts of memory to compile sql_yacc.cc with inline functions. Try running configure with the -with-low-memory option:

shell ./configure --with-low-memory

This option causes -fno-inline to be added to the compile line if you are using gcc and -O0 if you are using something else. You should try the -with-low-memory option even if you have so much memory and swap space that you think you can't possibly have run out. This problem has been observed to occur even on systems with generous hardware configurations, and the -with-low-memory option usually fixes it.

By default, configure picks c++ as the compiler name and GNU c++ links with -lg++. If you are using gcc, that behaviour can cause problems during configuration such as this:

configure: error: installation or configuration problem:
C++ compiler cannot create executables.

You might also observe problems during compilation related to g++, libg++, or libstdc++.

One cause of these problems is that you may not have g++, or you may have g++ but not libg++, or libstdc++. Take a look at the config.log file. It should contain the exact reason why your C++ compiler didn't work. To work around these problems, you can use gcc as your C++ compiler. Try setting the environment variable CXX to "gcc -O3". For example:

shell CXX="gcc -O3" ./configure

This works because gcc compiles C++ sources as well as g++ does, but does not link in libg++ or libstdc++ by default.

Another way to fix these problems, of course, is to install g++, libg++, and libstdc++. We would however like to recommend you to not use libg++ or libstdc++ with MySQL as this will only increase the binary size of mysqld without giving you any benefits. Some versions of these libraries have also caused strange problems for MySQL users in the past.

Using gcc as the C++ compiler is also required, if you want to compile MySQL with RAID functionality (see Section 6.5.3, “CREATE TABLE Syntax ” for more info on RAID table type) and you are using GNU gcc version 3 and above. If you get errors like the ones below during the linking stage when you configure MySQL to compile with the option -with-raid, try to use gcc as your C++ compiler by defining the above mentioned environment variable CXX:

gcc -O3 -DDBUG_OFF -rdynamic -o isamchk isamchk.o sort.o  libnisam.a
../mysys/libmysys.a ../dbug/libdbug.a ../strings/libmystrings.a -lpthread
-lz -lcrypt -lnsl -lm -lpthread
../mysys/libmysys.a(raid.o)(.text+0x79): In function `my_raid_create':
: undefined reference to `operator new(unsigned)'
../mysys/libmysys.a(raid.o)(.text+0xdd): In function `my_raid_create':
: undefined reference to `operator delete(void*)'
../mysys/libmysys.a(raid.o)(.text+0x129): In function `my_raid_open':
: undefined reference to `operator new(unsigned)'
../mysys/libmysys.a(raid.o)(.text+0x189): In function `my_raid_open':
: undefined reference to `operator delete(void*)'
../mysys/libmysys.a(raid.o)(.text+0x64b): In function `my_raid_close':
: undefined reference to `operator delete(void*)'
collect2: ld returned 1 exit status

If your compile fails with errors, such as any of the following, you must upgrade your version of make to GNU make:

making all in mit-pthreads
make: Fatal error in reader: Makefile, line 18:
Badly formed macro assignment
  or
make: file `Makefile' line 18: Must be a separator (:
  or
pthread.h: No such file or directory

Solaris and FreeBSD are known to have troublesome make programs.

GNU make Version 3.75 is known to work.

If you want to define flags to be used by your C or C++ compilers, do so by adding the flags to the CFLAGS and CXXFLAGS environment variables. You can also specify the compiler names this way using CC and CXX. For example:

shell CC=gcc
shell CFLAGS=-O3
shell CXX=gcc
shell CXXFLAGS=-O3
shell export CC CFLAGS CXX CXXFLAGS

See Section 2.2.10, “MySQL Binaries Compiled by MySQL AB ”, for a list of flag definitions that have been found to be useful on various systems.

If you get an error message like this, you need to upgrade your gcc compiler:

client/libmysql.c:273: parse error before `__attribute__'

gcc 2.8.1 is known to work, but we recommend using gcc 2.95.2 or egcs 1.0.3a instead.

If you get errors such as those shown here when compiling mysqld, configure didn't correctly detect the type of the last argument to accept(), getsockname(), or getpeername():

cxx: Error: mysqld.cc, line 645: In this statement, the referenced
     type of the pointer value ''length'' is ''unsigned long'', which
     is not compatible with ''int''.
new_sock = accept(sock, (struct sockaddr *)cAddr, length);

To fix this, edit the config.h file (which is generated by configure). Look for these lines:

/* Define as the base type of the last arg to accept */
#define SOCKET_SIZE_TYPE XXX

Change XXX to size_t or int, depending on your operating system. (Note that you will have to do this each time you run configure because configure regenerates config.h.)

The sql_yacc.cc file is generated from sql_yacc.yy. Normally the build process doesn't need to create sql_yacc.cc, because MySQL comes with an already generated copy. However, if you do need to re-create it, you might encounter this error:

"sql_yacc.yy", line xxx fatal: default action causes potential...

This is a sign that your version of yacc is deficient. You probably need to install bison (the GNU version of yacc) and use that instead.

If you need to debug mysqld or a MySQL client, run configure with the -with-debug option, then recompile and link your clients with the new client library. Section E.2, “Debugging a MySQL client ”.

If you get a compilation error on Linux (e.g. SuSE Linux 8.1 or Red Hat Linux 7.3) similar to the following one:

libmysql.c:1329: warning: passing arg 5 of `gethostbyname_r' from incompatible pointer type
libmysql.c:1329: too few arguments to function `gethostbyname_r'
libmysql.c:1329: warning: assignment makes pointer from integer without a cast
make[2]: *** [libmysql.lo] Error 1

By default, the configure script attempts to determine the correct number of arguments by using g++ the GNU C++ compiler. This test yields wrong results, if g++ is not installed. There are two ways to work around this problem:

Please note that you need to run configure again afterwards.

On most systems, you can force MIT-pthreads to be used by running configure with the -with-mit-threads option:

shell ./configure --with-mit-threads

Building in a non-source directory is not supported when using MIT-pthreads because we want to minimise our changes to this code.

The checks that determine whether to use MIT-pthreads occur only during the part of the configuration process that deals with the server code. If you have configured the distribution using -without-server to build only the client code, clients will not know whether MIT-pthreads is being used and will use Unix socket connections by default. Because Unix sockets do not work under MIT-pthreads on some platforms, this means you will need to use -h or -host when you run client programs.

When MySQL is compiled using MIT-pthreads, system locking is disabled by default for performance reasons. You can tell the server to use system locking with the -external-locking option. This is only needed if you want to be able to run two MySQL servers against the same datafiles (not recommended).

Sometimes the pthread bind() command fails to bind to a socket without any error message (at least on Solaris). The result is that all connections to the server fail. For example:

shell mysqladmin version
mysqladmin: connect to server at '' failed;
error: 'Can't connect to mysql server on localhost (146)'

The solution to this is to kill the mysqld server and restart it. This has only happened to us when we have forced the server down and done a restart immediately.

With MIT-pthreads, the sleep() system call isn't interruptible with SIGINT (break). This is only noticeable when you run mysqladmin -sleep. You must wait for the sleep() call to terminate before the interrupt is served and the process stops.

When linking, you may receive warning messages like these (at least on Solaris); they can be ignored:

ld: warning: symbol `_iob' has differing sizes:
    (file /my/local/pthreads/lib/libpthread.a(findfp.o) value=0x4;
file /usr/lib/libc.so value=0x140);
    /my/local/pthreads/lib/libpthread.a(findfp.o) definition taken
ld: warning: symbol `__iob' has differing sizes:
    (file /my/local/pthreads/lib/libpthread.a(findfp.o) value=0x4;
file /usr/lib/libc.so value=0x140);
    /my/local/pthreads/lib/libpthread.a(findfp.o) definition taken

Some other warnings also can be ignored:

implicit declaration of function `int strtoll(...)'
implicit declaration of function `int strtoul(...)'

We haven't gotten readline to work with MIT-pthreads. (This isn't needed, but may be interesting for someone.)

VC++ 6.0 compiler (updated with 4 or 5 SP and Pre-processor package) The Pre-processor package is necessary for the macro assembler. More details at: http://msdn.microsoft.com/vstudio/downloads/updates/sp/vs6/sp5/faq.aspx.

The MySQL source distribution for Windows, which can be downloaded from http://www.mysql.com/downloads/.