Go to the previous, next section.

Using configure

configure prepares source directories for building programs in them. "Configuring" is the process of preparing software to compile correctly on a given host, for a given target.

configure subsequently writes a configured `Makefile' from a pre-built template; configure uses variables that have been set in the configuring process to determine the values of some variables in the `Makefile'. Because of this we will refer to both configure variables and `Makefile' variables. This convention allows us to determine where the variable should be set initially, in either `configure.in' or `Makefile.in'.

What configure really does

Cygnus configure is a shell script that sets up an environment in which your programs will compile correctly for your machine and operating system, and will install in proper places. configure accomplishes this task by doing the following:

For the sake of safety (i.e., in order to prevent broken installations), the GNU coding standards call for software to be configured in such a way that an end user trying to build a given package will be able to do so by affecting a finite number of variables. All GNU software comes with an executable configure shell script which sets up an environment within a build directory which will correctly compile your new package for your host (or, alternatively, whatever host you specify to configure). For further background on this topic, see On Configuring Development Tools by K. Richard Pixley.

Use configure to set for the build process:

configure uses a few other files to complete its tasks. These are discussed in detail where noted.

configure.in
Input file for configure. Shell script fragments reside here. See section The configure.in input file.

Makefile.in
Template which configure uses to build a file called `Makefile' in the build directory. See section Makefile generation.

config.sub
Shell script used by configure to expand referents to the hosttype argument into a single specification of the form cpu-vendor-os. For instance, on the command line you can specify

eg$ ./configure sun4

to configure for a Sun SPARCstation running SunOS 4.x. configure consults config.sub to find that the three-part specification for this is

sparc-sun-sunos4.1.1

which notes the cpu as `sparc', the manufacturer as `sun' (Sun Microsystems), and the os (operating system) as `sunos4.1.1', the SunOS 4.1.1 release. See section Variables available to configure.in.

config.guess
If you do not put the hosttype argument on the command line, configure uses the config.guess shell script to make an analysis of your machine (it assumes that you wish to configure your software for the type of machine on which you are running). The output of config.guess is a three-part identifier as described above.

config.status
The final step in configuring a directory is to create a shell script, config.status. The main purpose of this file is to allow the `Makefile' for the current directory to rebuild itself, if necessary. See section config.status.

config/*
configure uses three types of `Makefile' fragments, which reside in the directory `srcdir/config/'. See section Adding information about local conventions.

Build variables

There are several variables in the build process which you can control through build programs such as make. These include machine definitions, local conventions, installation locations, locations for temporary files, etc. This data is accessible through certain variables which are configurable in the build process; we refer to them as build variables.

For lists of build variables which you can affect by using configure, see section Variables available to configure.in, and section Full descriptions of all installation subdirectories.

Generally, build variables, which are used by the `Makefile' to determine various aspects of the build and installation processes, are changeable with command-line options to configure. In most large suites of programs, like the Cygnus Support Developer's Kit, the individual programs reside in several subdirectories of a single source code "tree". All of these subdirectories need to be configured with information relative to the build directory, which is not known until configure is run. Unless specified otherwise, configure recursively configures every subdirectory in the source tree.

Build variables are passed from configure directly into the `Makefile', and use the same names (except that dashes are transformed into underbars; for example, when you specify the option `--exec-prefix' on the command line, the `Makefile' variable `exec_prefix' is set). In other words, if you specify

eg$ ./configure --prefix=/usr/gnu/local ... hosttype

on the command line, configure sets an variable called `prefix' to `/usr/gnu/local', and passes this into the `Makefile' in the same manner. After this command, each `Makefile' generated by configure will contain a line that reads:

prefix = /usr/gnu/local

For a list of the `Makefile' variables configure can change, and instructions on how to change them, see section Variables available to configure.in}, and section Invoking configure}.

Build directories

By default, configure builds a `Makefile' and symbolic links in the same directory as the source files. This default works for many cases, but it has limitations. For instance, using this approach, you can only build object code for one host at a time.

We refer to each directory where configure builds a `Makefile' as a build directory.

The build directory for any given build is always the directory from which you call configure, or `.' relative to your prompt. The default source directory, the place configure looks to find source code, is also `.'. For instance, if we have a directory `/gnu-stuff/src/' that is the top branch of a tree of GNU source code we wish to configure, then the program we will use to configure this code is `/gnu-stuff/src/configure', as follows. (Assume for the sake of argument that our machine is a sun4.)

eg$ cd /gnu-stuff/src
eg$ ./configure sun4
Created "Makefile" in /gnu-stuff/src
eg$

We just configured the code in `/gnu-stuff/src' to run on a Sun SPARCstation using SunOS 4.x by creating a `Makefile' in `/gnu-stuff/src'. By default, we also specified that when this code is built, the object code should reside in the same directory, `/gnu-stuff/src'.

However, if we wanted to build this code for more than one host, we would be in trouble, because the new configuration would write over the old one, destroying it in the process. What we can do is to make a new build directory and configure from there. Running configure from the new directory will place a correct `Makefile' and a `config.status' in this new file. That is all configure does; we must run make to generate any object code.

The new `Makefile' in `/gnu-stuff/sun4-obj', created from the template file `/gnu-stuff/src/Makefile.in', contains all the information needed to build the program.

eg$ mkdir /gnu-stuff/sun4-obj
eg$ cd /gnu-stuff/sun4-obj
eg$ ../src/configure --srcdir=../src sun4
Created "Makefile" in /gnu-stuff/sun4-obj
eg$ ls
Makefile       config.status
eg$ make all info install install-info clean
compilation messages...
eg$ mkdir /gnu-stuff/solaris2
eg$ cd /gnu-stuff/solaris2
eg$ ../src/configure --srcdir=../src sol2
Created "Makefile" in /gnu-stuff/solaris2
eg$ ls
Makefile       config.status
eg$ make all info install install-info clean
compilation messages...

We can repeat this for other configurations of the same software simply by making a new build directory and reconfiguring from inside it. If you do not specify the hosttype argument, configure will attempt to figure out what kind of machine and operating system you happen to be using. See section Determining system information. Of course, this may not always be the configuration you wish to build.

Caution: If you build more than one configuration for a single program, remember that you must also specify a different `--prefix' for each configuration at configure-time. Otherwise, both configurations will be installed in the same default location (`/usr/local'); the configuration to be installed last would overwrite previously installed configurations.

Makefile generation

Cygnus configure creates a file called `Makefile' in the build directory which can be used with make to automatically build a given program or package. configure also builds a `Makefile' for each relevant subdirectory for a given program or package (irrelevant subdirectories would be those which contain no code which needs configuring, and which therefore have no configure input file `configure.in' and no `Makefile' template `Makefile.in'). @xref{Running, make Invocation, How to Run make, make, GNU Make}, for details on using make to compile your source code.

Each `Makefile' contains variables which have been configured for a specific build. These build variables are determined when configure is run. All build variables have defaults. By default, configure generates a `Makefile' which specifies:

Variables are changeable through command-line options to configure (see section Invoking configure).

If you are porting a new program and intend to use configure, see section Porting with configure, as well as section `Writing Makefiles' in GNU Make, and section `Makefile Conventions' in GNU Coding Standards.

Determining system information

The shell script config.guess is called when you do not specify a hosttype on the command line to configure. config.guess acquires available system information from your local machine through the shell command uname. It compares this information to a database and attempts to determine a usable three-part system identifier (known as a triple) to use as your hosttype. See section What configure really does really does}, to see how this information is used.

Note: If you do not specify a hosttype on the command line, configure will attempt to configure your software to run on the machine you happen to be using. This may not be the configuration you desire.

config.status

The final step in configuring a directory is to create an executable shell script, `config.status'. The main purpose of this file is to allow the `Makefile' for the current directory to rebuild itself, if necessary. It is usually run from within the `Makefile'. See section Extensions to the GNU coding standards coding standards}.

`config.status' also contains a record of the configure session which created it.

The configure.in input file

A `configure.in' file for Cygnus configure consists of a per-invocation section, followed by a per-host section, followed by a per-target section, optionally followed by a post-target section. Each section is a shell script fragment, which is executed by the configure shell script at an appropriate time. Values are passed among configure and the shell fragments through a set of shell variables. When each section is being interpreted by the shell, the shell's current directory is the build directory, and any files created by the section (or referred to by the section) will be relative to the build directory. To reference files in other places (such as the source directory), prepend a shell variable such as `$(srcdir)/' to the desired file name.

The beginning of the `configure.in' file begins the per-invocation section.

A line beginning with `# per-host:' begins the per-host section.

A line beginning with `# per-target:' begins the per-target section.

If it exists, the post-target section begins with `# post-target:'.

Variables available to configure.in

The following variables pass information between the standard parts of configure and the shell-script fragments in `configure.in':

srctrigger
Contains the name of a source file that is expected to live in the source directory. You must usually set this in the per-invocation section of `configure.in'. configure tests to see that this file exists. If the file does not exist, configure prints an error message. This is used as a sanity check that `configure.in' matches the source directory.

srcname
Contains the name of the source collection contained in the source directory. You must usually set this in the per-invocation section of `configure.in'. If the file named in `srctrigger' does not exist, configure uses the value of `srcname' when it prints the error message.

configdirs
Contains the names of any subdirectories in which configure should recurse. You must usually set this in the per-invocation section of `configure.in'. If `Makefile.in' contains a line starting with `SUBDIRS =', then it will be replaced with an assignment to `SUBDIRS' using the value of `configdirs' (if `subdirs' is empty). This can be used to determine which directories to configure and build depending on the host and target configurations. Use `configdirs' (instead of the `subdirs' variable described below) if you want to be able to partition the subdirectories, or use independent `Makefile' fragments. Each subdirectory can be independent, and independently reconfigured.

subdirs
Contains the names of any subdirectories where configure should create a `Makefile' (in addition to the current directory), without recursively running configure. Use `subdirs' (instead of the `configdirs' variable described above) if you want to configure all of the directories as a unit. Since there is a single invocation of configure that configures many directories, all the directories can use the same `Makefile' fragments, and the same configure.in.

host
Contains the full configuration name for the host (generated by the script `config.sub' from the name that you entered). This is a three-part name (commonly referred to as a triple) of the form cpu-vendor-os.

There are separate variables `host_cpu', `host_vendor', and `host_os' that you can use to test each of the three parts; this variable is useful, however, for error messages, and for testing combinations of the three components.

host_cpu
Contains the first element of the canonical triple representing the host as returned by `config.sub'. This is occasionally used to distinguish between minor variations of a particular vendor's operating system and sometimes to determine variations in binary format between the host and the target.

host_vendor
Contains the second element of the canonical triple representing the host as returned by `config.sub'. This is usually used to distinguish among the numerous variations of common operating systems.

host_os
Contains the the third element of the canonical triple representing the host as returned by `config.sub'.

target
Contains the full configuration name (generated by the script `config.sub' from the name that you entered) for the target. Like the host, this is a three-part name of the form cpu-vendor-os.

There are separate variables `target_cpu', `target_vendor', and `target_os' that you can use to test each of the three parts; this variable is useful, however, for error messages, and for testing combinations of the three components.

target_cpu
Contains the first element of the canonical triple representing the target as returned by `config.sub'. This variable is used heavily by programs which are involved in building other programs, like the compiler, assembler, linker, etc. Most programs will not need the `target' variables at all, but this one could conceivably be used to build a program, for instance, that operated on binary data files whose byte order or alignment differ from the system where the program is running.

target_vendor
Contains the second element of the canonical triple representing the target as returned by `config.sub'. This is usually used to distinguish among the numerous variations of common operating systems or object file formats. It is sometimes used to switch between different flavors of user interfaces.

target_os
Contains the the third element of the canonical triple representing the target as returned by `config.sub'. This variable is used by development tools to distinguish between subtle variations in object file formats that some vendors use across operating system releases. It might also be use to decide which libraries to build or what user interface the tool should provide.

floating_point
Set to `no' if you invoked configure with the `--nfp' command-line option, otherwise it is empty. This is a request to target machines with no floating point unit, even if the targets ordinarily have floating point units available.

gas
Set to `true' if you invoked configure with the `--with-gnu-as' command line option, otherwise it is empty. This is a request to assume that the specified hosttype machine has GNU as available even if it ordinarily does not.

srcdir
Set to the name of the directory containing the source for this program. This will be different from `.' if you have specified the `--srcdir=dir' option. `srcdir' can indicate either an absolute path or a path relative to the build directory.

package_makefile_frag
If set in `configure.in', this variable should be the name a file relative to `srcdir' to be included in the resulting `Makefile'. If the named file does not exist, configure will print a warning message. This variable is not set by configure.

host_makefile_frag
If set in `configure.in', this variable should be the name a file relative to `srcdir' to be included in the resulting `Makefile'. If the named file does not exist, configure will print a warning message. This variable is not set by configure.

target_makefile_frag
If set in `configure.in', this variable should be the name of a file, relative to `srcdir', to be included in the resulting `Makefile'. If the named file does not exist, configure will print a warning message. This variable is not set by configure.

site_makefile_frag
Set to a file name representing to the default `Makefile' fragment for this host. It may be set in `configure.in' to override this default. Normally `site_makefile_frag' is empty, but will have a value if you specify `--site=site' on the command line.

Makefile
Set to the name of the generated `Makefile'. Normally this value is precisely `Makefile', but some programs may want something else.

removing
Normally empty but will be set to some non-null value if you specified `--rm' on the command line. That is, if `removing' is not empty, then configure is removing a configuration rather than creating one.

files
If this variable is not empty following the per-target section, then each word in its value will be the target of a symbolic link named in the corresponding word from the `links' variable.

links
If the `files' variable is not empty following the per-target section, then configure creates symbolic links with the first word of `links' pointing to the first word of `files', the second word of `links' pointing to the second word of `files', and so on.

A minimal configure.in

A minimal `configure.in' consists of four lines.

srctrigger=foo.c
srcname="source for the foo program"
# per-host:
# per-target:

The `# per-host:' and `# per-target:' lines divide the file into the three required sections. The `srctrigger' line names a file. configure checks to see that this file exists in the source directory before configuring. If the `srctrigger' file does not exist, configure uses the value of `srcname' to print an error message about not finding the source.

This particular example uses no links, and only the default host, target, and site-specific `Makefile' fragments if they exist.

For each invocation

configure invokes the entire shell script fragment from the start of `configure.in' up to a line beginning with `# per-host:' immediately after parsing command line arguments. The variables `srctrigger' and `srcname' must be set here.

You might also want to set the variables `configdirs' and `package_makefile_frag' here.

Host-specific instructions

The per-host section of `configure.in' starts with the line that begins with `# per-host:' and ends before a line beginning with `# per-target:'. configure invokes the commands in the per-host section when determining host-specific information.

This section usually contains a big case statement using the variable `host' to determine appropriate values for `host_makefile_frag' and `files', although `files' is not usually set here. Usually, it is set at the end of the per-target section after determining the names of the target specific configuration files.

Target-specific instructions

The per-target section of `configure.in' starts with the line that begins with `# per-target:' and ends before the line that begins with `# post-target:', if there is such a line. Otherwise the per-target section extends to the end of the file. configure invokes the commands in the per-target section when determining target-specific information, and before building any files, directories, or links.

This section usually contains a big case statement using the variable `target' to determine appropriate values for `target_makefile_frag' and `files'. The last lines in the per-target section normally set the variables `files' and `links'.

Instructions to be executed after target info

The post-target section is optional. If it exists, the `post-target' section starts with a line beginning with `# Post-target:' and extends to the end of the file. If it exists, configure invokes this section once for each target after building all files, directories, or links.

This section is seldom needed, but you can use it to edit the `Makefile' generated by configure.

An example configure.in

Here is a small example of a `configure.in' file.

# This file is a collection of shell script fragments
# used to tailor a template configure script as
# appropriate for this directory.  For more information,
# see configure.texi.

configdirs=
srctrigger=warshall.c
srcname="bison"

# per-host:
case "${host}" in
m88k-motorola-*)
        host_makefile_frag=config/mh-delta88
        ;;
esac

# per-target:
files="bison_in.hairy"
links="bison.hairy"

# post-target:

Install locations

Using the default configuration, `make install' creates a single tree of files, some of which are programs. The location of this tree is determined by the value of the variable `prefix'. The default value of `prefix' is `/usr/local'. This is often correct for native tools installed on only one host.

Changing the default install directory

In the default configuration, all files are installed in subdirectories of `/usr/local'. The location is determined by the value of the configure variable `prefix'; in turn, this determines the value of the `Makefile' variable of the same name (`prefix').

You can also set the value of the `Makefile' variable `prefix' explicitly each time you invoke make if you are so inclined. However, because many programs have this location compiled in, you must specify the `prefix' value consistently on each invocation of make, or you will end up with a broken installation.

To make this easier, the value of the configure variable `prefix' can be set on the command line to configure using the option `--prefix='.

Installing for multiple hosts

By default, host dependent files are installed in subdirectories of `$(exec_prefix)'. The location is determined by the value of the configure variable `exec_prefix', which determines the value of the `Makefile' variable `exec_prefix'. This makes it easier to install for a single host, and simplifies changing the default location for the install tree. The default doesn't allow for multiple hosts to effectively share host independent files, however.

To configure so that multiple hosts can share common files, use something like:

configure host1 -prefix=/usr/gnu -exec_prefix=/usr/gnu/H-host1
make all info install install-info clean

configure host2 -prefix=/usr/gnu -exec_prefix=/usr/gnu/H-host2
make all info install install-info

The first line configures the source for host1 to place host-specific programs in subdirectories of `/usr/gnu/H-host1'.

The second line builds and installs all programs for host1, including both host-independent and host-specific files, as well as removing the host-specific object files from of the build directory.

The third line reconfigures the source for host2 to place host specific programs in subdirectories of `/usr/gnu/H-host2'.

The fourth line builds and installs all programs for host2. Host specific files are installed in new directories, but the host independent files are installed on top of the host independent files installed for host1. This results in a single copy of the host independent files, suitable for use by both hosts.

See section Extensions to the GNU coding standards, for more information.

Full descriptions of all installation subdirectories

During any install, a number of standard directories are created. Their names are determined by `Makefile' variables. Some of the defaults for `Makefile' variables can be changed at configuration time using command line options to configure. For more information on the standard directories or the `Makefile' variables, please refer to section `Makefile Conventions' in GNU Coding Standards. See also section Extensions to the GNU coding standards coding standards}.

Note that configure does not create the directory indicated by the variable `srcdir' at any time. $(srcdir) is not an installation directory.

You can override all `Makefile' variables on the command line to make. (See section `Overriding Variables' in GNU Make.) If you do so, you will need to specify the value precisely the same way for each invocation of make, or you risk ending up with a broken installation. This is because many programs have the locations of other programs or files compiled into them. If you find yourself overriding any of the variables frequently, you should consider site dependent `Makefile' fragments. See also section Adding site info.

During `make install', a number of standard directories are created and populated. The following `Makefile' variables define them. Those whose defaults are set by corresponding configure variables are marked "Makefile and configure".

prefix (Makefile and configure)
The root of the installation tree. You can set its `Makefile' default with the `--prefix=' command line option to configure (see section Invoking configure). The default value for `prefix' is `/usr/local'.

bindir
A directory for binary programs that users can run. The default value for `bindir' depends on `prefix'; `bindir' is normally changed only indirectly through `prefix'. The default value for `bindir' is `$(prefix)/bin'.

exec_prefix (Makefile and configure)
A directory for host dependent files. You can specify the `Makefile' default value by using the `--exec_prefix=' option to configure. (See section Invoking configure.) The default value for `exec_prefix' is `$(prefix)'.

libdir
A directory for libraries and support programs. The default value for `libdir' depends on `prefix'; `libdir' is normally changed only indirectly through `prefix'. The default value for `libdir' is `$(prefix)/lib'.

mandir
A directory for man format documentation ("man pages"). The default value for `mandir' depends on `prefix'; `mandir' is normally changed only indirectly through `prefix'. The default value for `mandir' is `$(prefix)/man'.

manNdir
These are eight variables named `man1dir', `man2dir', etc. They name the specific directories for each man page section. For example, `man1dir' by default holds the filename `$(mandir)/man1'; this directory contains `emacs.1' (the man page for GNU Emacs). Similarly, `man5dir' contains the value `$(mandir)/man5', indicating the directory which holds `rcsfile.5' (the man page describing the rcs data file format). The default value for any of the `manNdir' variables depends indirectly on `prefix', and is normally changed only through `prefix'. The default value for `manNdir' is `$(mandir)/manN'.

manNext
Not supported by Cygnus configure. The GNU Coding Standards do not call for `man1ext', `man2ext', so the intended use for manext is apparently not parallel to `mandir'. Its use is not clear. (See also section Extensions to the GNU coding standards.)

infodir
A directory for info format documentation. The default value for `infodir' depends indirectly on `prefix'; `infodir' is normally changed only through `prefix'. The default value for `infodir' is `$(prefix)/info'.

docdir
A directory for any documentation that is in a format other than those used by info or man. The default value for `docdir' depends indirectly on `prefix'; `docdir' is normally changed only through `prefix'. The default value for `docdir' is `$(datadir)/doc'. This variable is an extension to the GNU coding standards. (See also section Extensions to the GNU coding standards.)

includedir
A directory for the header files accompanying the libraries installed in `libdir'. The default value for `includedir' depends on `prefix'; `includedir' is normally changed only indirectly through `prefix'. The default value for `includedir' is `$(prefix)/include'.

Host

The arguments to configure are hosttypes. By hosttype we mean the environment in which the source will be compiled. This need not necessarily be the same as the physical machine involved, although it usually is.

For example, if some obscure machine had the GNU POSIX emulation libraries available, it would be possible to configure most GNU source for a POSIX system and build it on the obscure host.

For more on this topic, see section `Host Environments' in On Configuring Development Tools.

Target

For building native development tools, or most of the other GNU tools, you need not worry about the target. The target of a configuration defaults to the same as the host.

For building cross development tools, please see section `Building Development Environments' in On Configuring Development Tools.

Adding information about local conventions

If you find that a tool does not get configured to your liking, or if configure's conventions differ from your local conventions, you should probably consider site-specific `Makefile' fragments. See also section Adding site info.

These are probably not the right choice for options that can be set from the configure command line or for differences that are host or target dependent.

Cygnus configure uses three types of `Makefile' fragments. In a generated `Makefile' they appear in the order: target fragment, host fragment, and site fragment. This allows host fragments to override target fragments, and site fragments to override both.

Host-specific `Makefile' fragments conventionally reside in the `./config/' subdirectory with names of the form `mh-hosttype'. They are used for hosts that require odd options to the standard compiler and for compile time options based on the host configuration.

Target-specific `Makefile' fragments conventionally reside in the `./config/' subdirectory with names of the form `mt-target'. They are used for target dependent compile time options.

Site specific `Makefile' fragments conventionally reside in the `./config/' subdirectory with names of the form `ms-site'. They are used to override host- and target-independent compile time options. Note that you can also override these options on the make invocation line.

Extensions to the GNU coding standards

The following additions to the GNU coding standards are required for Cygnus configure to work properly.

Go to the previous, next section.