Table of contents: ================== * Installation from sources * More details about ./configure --with-PACKAGE[=OPTION] * Platform compiler problems * Environment variables * How to test gnuplot * Installing WPS object for gnuplot on OS/2 * Compiling Gnuplot with the wxt (wxWidgets) terminal If you are installing a binary package, you may still want to check out some sections of this document, particularly the one discussing environment variables. Installation from sources ========================= For the impatient ----------------- Configuration options are in the Makefile and in src/term.h, which selects the set of terminal drivers to be compiled in. The recommended way to configure both of these is the GNU-style "./configure" script described below and described further in INSTALL.gnu. This script is provided as part of the distributed source package, but if you are using sources downloaded directly from a git snapshot you may need to regenerate it using the preliminary script "./prepare". The ./configure script inspects your system to see what support libraries are available, then modifies the gnuplot Makefile and header file config.h to use them. Assuming you have already installed the necessary support libraries, the process is: ./configure # prepare Makefile and config.h make # build the program and documentation make check # run unit tests to confirm successful build make install # install the program and documentation File location defaults ---------------------- PREFIX /usr/local gnuplot PREFIX/bin gnuplot_x11 PREFIX/libexec/gnuplot/$VERSION gnuplot_qt PREFIX/libexec/gnuplot/$VERSION gnuplot.1 PREFIX/man/man1 gnuplot.gih PREFIX/share/gnuplot/$VERSION terminal support files PREFIX/share/gnuplot/$VERSION/js PREFIX/share/gnuplot/$VERSION/PostScript PREFIX/share/gnuplot/$VERSION/lua PREFIX/share/gnuplot/$VERSION/qt The only files where the runtime location is defined at compile time are gnuplot.gih and show.c. This is important if gnuplot is not installed by running 'make install'. The default path for the help library, gnuplot.gih, can be controlled in several different ways: o use configure's --with-gihdir option to specify the full path to the directory where gnuplot.gih is to be installed, e.g. ./configure --with-gihdir=/opt/docs o use configure's --datadir= option to specify a version and package independent data dir prefix, e.g. ./configure --datadir='/opt/docs' gnuplot will look for the online help file, gnuplot.gih, in /opt/docs/gnuplot/$VERSION. o with configure's --prefix= option, eg. ./configure --prefix=/gnuplot gnuplot will look for the online help file, gnuplot.gih, in /gnuplot/share/gnuplot/$VERSION. Attention: This option affects the location of *all* installed files! o at make time, eg. make HELPFILE='/gnuplot/docs/gnuplot.gih' gnuplot will look for the online help file /gnuplot/docs/gnuplot.gih. o at execution time by specifying the helpfile with the environment variable GNUHELP (see "Environment Variables" below). The default location of the main executable can be controlled in different ways: o with configure's --prefix= option, eg. ./configure --prefix=/gnuplot gnuplot will be installed as /gnuplot/bin/gnuplot. Attention: This affects the location of *all* installed files! o with configure's --bindir= option, eg. ./configure --bindir='/gnuplot/bin' gnuplot will be installed as /gnuplot/bin/gnuplot. o at make time, eg. make bindir='/gnuplot/bin' gnuplot will be installed as /gnuplot/bin/gnuplot. The default location of the additional executable, like gnuplot_x11, can be controlled in different ways: o with configure's --prefix= option, eg. ./configure --prefix=/gnuplot gnuplot_x11 will be installed as /gnuplot/libexec/gnuplot/4.0/gnuplot_x11. Attention: This affects the location of *all* installed files! o with configure's --libexecdir= option, eg. ./configure --libexecdir='/gnuplot/lib' gnuplot_x11 will be installed as /gnuplot/lib/gnuplot/4.0/gnuplot_x11 o at make time, eg. make X11_DRIVER_DIR='/gnuplot/lib/4.0' gnuplot_x11 will be installed as /gnuplot/lib/4.0/gnuplot_x11 If you don't install the program at the place chosen by ./configure, e.g. by overriding the "prefix" at "make install" time, files written by the `save' command cannot be used as scripts on Un*x platforms, so this should be avoided. There is no way to specify the location of the executables during runtime. Other platforms are not affected by this. The gnuplot demo files are not installed by default, mainly because there is no universally agreed place where such files should go. If desired, they should be copied manually to a location of choice. Unix, configure --------------- On Unix, use $ ./configure $ make [ Optionally run demos--see "How to test gnuplot" below. ] $ make install If gcc is installed, it is used by default. A different compiler can be used by passing its name to configure: $ ./configure CC=c99 Any environment settings for CFLAGS are included into the Makefile, so please make sure that these are really needed. There are several options available for configure that you may want to change. A complete list of options is available through $ ./configure --help Some options relevant to gnuplot are listed below: --prefix=PREFIX Install architecture-independent files in PREFIX [/usr/local] --exec-prefix=EPREFIX Install architecture-specific file in EPREFIX [PREFIX] --bindir=DIR user executables in DIR [EPREFIX/bin] The gnuplot binary is installed this directory --libexecdir=DIR program executables in DIR [EPREFIX/libexec] gnuplot_x11 goes into $libexecdir/gnuplot/. --datadir=DIR Read-only architecture-independent data in DIR [PREFIX/share]. The gnuplot help file is installed in this directory. --libdir=DIR object code libraries in DIR [EPREFIX/lib] --includedir=DIR C header files in DIR [PREFIX/include] --mandir=DIR Man documentation in DIR [PREFIX/man] --disable-history-file do not use history file --disable-mouse disable mouse for interactive terminals --disable-x11-mbfonts disable multi-byte font support for x11 --disable-x11-external disable drawing to windows belonging to external apps --disable-raise-console spacebar in plot window does not raise console --disable-wxwidgets wxWidgets terminal (default enabled) --enable-backwards-compatibility enable deprecated syntax --with-bitmap-terminals enable support for HP deskjet era dot matrix printers and *.pbm files --with-readline={gnu|bsd|builtin} Override the automatic choice of terminal input modes. See below for more detail. Options marked with (*) are enabled by default, ie. these features or packages are used if configure detects them even if the corresponding option is not specified. Please note that the `--with-PACKAGE' options can have additional arguments: o `--with-PACKAGE' is equivalent to `--with-PACKAGE=yes' o `--with-PACKAGE=no' disables checking for PACKAGE. It has the same effect as `--without-PACKAGE'. o `--with-PACKAGE=DIR' checks for PACKAGE in DIR Linux ----- ./configure may fail to find lua support if the package configuration tool is installed as "lua5.1" or "lua5.3" rather than "lua". If necessary, you can fix this by adding a symlink prior to running ./configure as shown below: ln -s /usr/lib/pkgconfig/lua5.3.pc /usr/lib/pkgconfig/lua.pc Mac OSX ------- Readline issues: As I understand the situation, Apple ships OSX with a "fake" libreadline shared library. The file /usr/lib/libreadline.dylib is really a symlink to a compatibility layer over the BSD libedit library. But the compatibility isn't complete, and in particular it is missing some routines used for readline support by gnuplot. The ./configure script should be able to cope with this. But you will still be left without some of the functionality of the "real" libreadline. For one thing, libedit doesn't handle UTF-8 input. You have several options: 1) Delete the fake libreadline libraries from OSX and install the real gnu libreadline as a system library. 2) Install the real gnu libreadline into your personal account and tell gnuplot to use it: ./configure --with-readline=/my/private/readline/installdir This may require some playing around with additional -L and -I definitions in CFLAGS, and is probably harder than replacing the system copy of the library. 3) Use gnuplot's built-in readline routines. These now support UTF-8 input and tab-completion of file names. ./configure --with-readline=builtin wxt terminal: Cairo & pango need to be installed first on Mac, I recommend using MacPorts building universal (i386 libraries): sudo port install cairo +no_x11 +quartz +universal sudo port install pango +no_x11 +quartz +universal Next gnuplot may be compiled with shipped with Mac OS X 10.6 SL wxWidgets using: ./configure CPPFLAGS='-arch i386' LDFLAGS='-arch i386' NOTE: 10.6 SL by default generates x86_64 code, but wxWidgets stable is using Carbon API on Mac which is 32-bit only, that makes compiling gnuplot, cairo and pango to i386 necessary. linux: Some wxWidgets configurations require linking to the X11 library but fail to include "-lX11" in the string provided for auto-configuration. You may have to add this manually when configuring gnuplot: TERMLIBS="-lX11" ./configure Note that some people have reported runtime problems with wxgtk 3.0. If you have the option of linking to libraries for wxgtk 2.8 instead, this may be preferable. The -lX11 flag is not need in this case. VMS --- We used to test-build gnuplot pre-releases on VMS, but this has not been the case since about version 4.6. The old guidelines are included below but may be of only historical interest. If you have successfully built version 5 on VMS please drop us a note, particularly if there were any changes you needed to make it work. John Hasstedt has written configure.vms, a command file that creates the necessary make and option files to build gnuplot. See the file for instructions on using it. If you have problems with it, the old build files are still included; instructions for using them follow. On VMS, you can use MMS, MMK, or another make utility, or you can use BUILDVMS.COM. The supplied files work with Alpha/VMS V6.2 and DECC V5.7; you may get warnings or more serious errors depending on the versions of the C compiler, the C run-time libraries, and VMS on your system. To compile using MMK: MMK/DESCRIPTION=MAKEFILE.VMS To compile using MMS on VAX: MMS/DESCRIPTION=MAKEFILE.VMS To compile using MMS on Alpha: MMS/DESCRIPTION=MAKEFILE.VMS/MACRO=__ALPHA__=1 Alternatively, you can use MAKE_VMS.COM. This command file will run MMK or MMS to build gnuplot. If you don't have either one installed, it will use the command file. The first parameter on the command line is the compiler you want to use (default DECC). It may sometimes be useful to add `/IGNORE=WARNING' to the `MMS' call. To compile using GNUC or VAXC: add /MACRO=GNUC or /MACRO=VAXC to the above command To compile with another make utility: check the documentation of your utility to specify the description file and any necessary macros (__ALPHA__, GNUC, or VAXC) Or if you don't have a suitable make: @BUILDVMS To tell gnuplot where to find the help library: $ define gnuplot$help disk:[directory]gnuplot.hlb Alternatively, put the help in the main system help library. MS-Windows ---------- We provide binary packages with an installer or as "portable" versions packed in zip / 7z format on SourceForge. We currently support and test building with the following compilers and Makefiles: Mingw-w64/MSYS2: config/mingw/Makefile Microsoft Visual C++: config/msvc/Makefile OpenWatcom: config/watcom/Makefile Cygwin: config/cygwin/Makefile These Makefiles contain a configuration section which you have to adopt to your actual setup. Further building instructions are given in Makefiles. For MSYS2/Mingw-w64 you can find detailed instructions at https://sourceforge.net/p/gnuplot/support-requests/199/ Some of the Makefiles include an "install" target (e.g. config/mingw/Makefile). In order to build the help file you require the "Microsoft HTML Help 1.4 SDK", which is freely downloadable here: http://go.microsoft.com/fwlink/?LinkId=154968 MSDOS ----- Using DJGPP (32-bit DOS port of GNU CC) Basically the same as Unix+configure. Andris Pavenis has contributed a shell script which makes the whole process even easier: just run djconfig.sh, and then make. Only if you don't have a complete Unix toolset installed with your DJGPP, it is recommended to use makefile.dj2 instead: copy makefile.dj2 makefile make [NOTE: the 16-bit DOS versions of gnuplot no longer work. gnuplot has outgrown this platform.] OS/2 ---- To compile under OS/2 (2.x and above) you need the development suite EMX 0.9 (including gcc). You should also have GNU Make and IBM's IPFC (Information Presentation Facility Compiler, available from the Developer's Toolkit; nowadays it's accessible through an IBM website for free!). At the beginning of makefile.os2 you will find a configuration section where you have to adjust all settings which control the build process. Most important is probably to select the terminal devices which should be supported. You can create a version offering PM graphics as well as X11 support (to use with XFree86). Support for these two terminals is provided by additional executables. Be sure to enable only those devices for which you have the necessary software already installed. Also ensure that the libraries and the sources are all compiled (not) using the '-Zmt' flags. Executing make -f makefile.os2 should create a default build while make -f makefile.os2 help will show you all pre-defined targets. See other sections of the manuals for more information about installing/using gnuplot on OS/2. More details about ./configure --with-PACKAGE[=OPTION] ====================================================== Every `--with-PACKAGE' option sets a `with_package' variable in configure. Depending on how `--with-PACKAGE' was invoked, there are only three different possible values for the `with_package' variable: Option $with_package ---------------------------------------- (not specified) yes or no; default set in configure --with-package yes --with-package=yes yes --with-package=no no --with-package=DIR DIR --without-package no In gnuplot, the following --with-PACKAGE options are available. The --with-PACKAGE=DIR form is always required if a package is installed in a non-default location that is not searched by the preprocessor or linker. --with-readline Use whatever readline support is detected automatically, either an external library or gnuplot's built-in version. This is the default. --with-readline=builtin Use gnuplot's own builtin readline routines even if a GNU or BSD readline library is detected. --without-readline (same as --with-readline=no) Do not use any readline processing. --with-readline=gnu Use the GNU readline library instead of gnuplot's builtin readline. --with-readline[=DIR] Use the GNU readline library instead of gnuplot's builtin readline. This form is required if GNU readline is not installed in a default location. --with-gd[=DIR] Thomas Boutell's gd library is needed to support jpeg and gif output. It is also one of several options for producing png output. This option is on by default, i.e. configure will check if the library exists, and use it if found. Platform compiler problems ========================== This section addresses trouble shooting and testing issues. Userland questions are answered in the FAQ. Platform and compiler notes --------------------------- Generally, if you think that configure has made a mistake in detecting platform features, you should consider that a bug either in GNU autoconf or in gnuplot's use of it, which you should report. As a workaround you can override such faulty tests in a site-wide collection of preset configure test results, or manually override them after running configure. Example: configure was for some reason unable to detect the memset() function, but you are sure it is ok to use on your platform. To fix that, you'll have to edit the generated config.h and change the line /* #undef HAVE_MEMSET */ to #define HAVE_MEMSET 1 Note that changing such defines from outside config.h, eg. via 'make DEFS=-DHAVE_MEMSET' is wrong, because config.h will usually override such changes. Also note that this change will be lost whenever you re-run configure. - gcc -ansi If you wish to compile with gcc -ansi (CC='gcc -ansi -pedantic' ./configure), additional platform specific defines may be necessary to include non-ANSI function prototypes. E.g. on Solaris, -D__EXTENSIONS__ is needed. Otherwise, non-declared functions not returning int will be assumed to return int, with possibly dangerous consequences. Especially scpecfun.c is vulnerable here. - SunOS / Solaris There are known issues with regard to compiler support for Booleans. You may have to tweak gnuplot's definition of TBOOLEAN. The plugin demo Makefile may also require editing as noted in comments. Environment variables ===================== See 'help environment'. If the environment variable GNUTERM is found, it is used as the terminal type. Otherwise, in some cases the variable TERM will be used, or the hardware may be detected automatically. If all else fails, a compile-time default will be used. The PC version looks for the environment variable GNUPLOT to contain the name of the directory from which to load the initialization file GNUPLOT.INI and also to be used as the home directory in ~ substitution. See the help on 'start_up' for more information. If defined, the environment variable GNUHELP is used for the name of the .gih help file, otherwise HELPFILE (defined in makefile or command.c) is used. The VMS version looks for the logical name GNUPLOT$HELP to locate the help library. The usual system-wide environment variable is used to locate a command line shell to be used for executing commands and starting sub-shells. On all versions of Windows and OS/2 the variable looked at is COMSPEC. Other platforms will consult SHELL. The x11 and qt terminals each require an auxilliary program (gnuplot_x11 gnuplot_qt) to manage the displayed plots. These are installed by default in /usr/local/libexec/gnuplot// but this default can be overridden by the environmental variable GNUPLOT_DRIVER_DIR. GDFONTPATH is the directory where the libgd terminals (jpeg png gif) search for fonts. You should set it to directory where files like arial.ttf and Symbol.pfa exist. GNUPLOT_DEFAULT_GDFONT is the default font to be used by the png and jpeg terminal types. This must either be a full path to the font file or the face name of a font found in the GDFONTPATH search path. GNUPLOT_FONTPATH is used by the postscript driver as search path for postscript (Adobe Type 1 and Type 2) fonts. GNUPLOT_PS_DIR is used by the postscript driver to use external prologue files. Depending on the build process, gnuplot contains either a builtin copy of those files or simply a default hardcoded path. Use this variable to test the postscript terminal with custom prologue files. GNUPLOT_LIB may be used to define additional search directories for data and command files. The variable may contain a single directory name, or a list of directories separated by a platform-specific path separator, eg. ':' on Unix, or ';' on DOS/Windows/OS2 platforms. The contents of GNUPLOT_LIB are appended to the `loadpath` variable but are not saved with the `save` and `save set` commands. GNUFITLOG holds the name of a file that saves fit results. The default is fit.log. If the name ends with a "/" or "\", it is treated as a directory name, and "fit.log" will be created as a file in that directory. The CGI drivers need the CGIPATH environment variable to set the path to the CGI agents, and the CGIDISP and/or CGIPRNT environment variables to the CGI agents, and the CGIDISP and/or CGIPRNT environment variables to set the output devices. If creating dynamically linked executables, it may be necessary to add flags to the LIBS variable in Makefile to make sure gnuplot finds all required libraries at runtime. Systems like SunOS and Solaris use -R to specify the runtime library search path, whereas OSF/Dec Unix, Irix and Linux use -rpath. If this is not possible, the LD_LIBRARY_PATH environment variable should be set. Generally it's recommended to install shared libraries only in places that are already searched by the linker without such options. LD_LIBRARY_PATH is the second-best choice. How to test gnuplot =================== The demo files distributed with the program source act also as unit tests for various gnuplot syntax, plotting, and terminal features. Output can be compared to the on-line demo output collection on the gnuplot web site. All command line examples below assume Unix Bourne shell syntax. Individual demo files can be run interactively by eg. $ cd gnuplot/demo $ gnuplot simple.dem or $ GNUPLOT_DRIVER_DIR=../src ../src/gnuplot simple.dem and gnuplot prompts the user to "Hit return to continue" to cycle through all the plots (the GNUPLOT_DRIVER_DIR= part is only required if you are using the X11 or BeOS plotting device and the gnuplot_x11/gnuplot_be binary has not been installed yet, or if you want to test the external binaries you just compiled at the same time). To run the demos in a specified file without interaction, one can use $ gnuplot simple.dem set term dumb gnuplot> load 'all.dem' To test the capabilities of the terminal you are using, there is the 'test' command: gnuplot> test `test` creates a display of line and point styles and other useful things appropriate for the terminal you are using. Installing WPS object for gnuplot on OS/2 ========================================= GNUPLOT can be run from the command line of an OS/2 session, or can be set up as a program object in the WPS. GNUPLOT starts up as a command line session which accepts the usual GNUPLOT input. The Presentation Manager (PM) window for graphics display is only opened when the first 'plot' or 'splot' command is entered. If GNUPLOT is set up as a program object, it can be configured for 'drag and drop' in the usual way, by specifying '%*' as the program parameter. The PM window has facilities for printing the plot, for changing the fonts used in the plot, and so on. These of course are handled through the usual PM interface. In addition, you can 'drop' a font onto the window to change the font. The usual GNUPLOT commands can also be used to produce output on supported devices. Settings and the positions and sizes of the windows can be saved in the file gnushell.ini. This file is saved in the program's working directory; several different directories can be set up (as program objects) with different options, if you wish. PM terminal is fully mouse-capable. Type 'm' to switch the mouse on/off (see 'help mouse' for more details), or 'h' for the full list of current hotkeys (see also 'help bind'). Compiling Gnuplot with the wxWidgets terminal ============================================== The `wxt` terminal provides an interactive windowed output. The code is cross-platform and can be compiled as soon as the following requirements are met: First, this terminal is based on the wxWidgets library. This library can be found on http://www.wxwidgets.org. It comes in different flavours corresponding to the underlying API : Microsoft Windows ("wxMSW"), GTK+ ("wxGTK"), X11 ("wxX11"), MacOS ("wxMAC"), etc. Then, the terminal needs the Cairo library (no specific backend is required) which can be found on http://www.cairographics.org, and the Pango library, which can be found on http://www.pango.org and need to be compiled with its Cairo backend. On Unix, use the usual `./configure; make; make install' build process to compile this terminal. On Windows, please use the MinGW/MSYS makefile, namely config/makefile.mgw. MinGW and MSYS can be found on http://www.mingw.org.