wiki:Building/Installing

Version 9 (modified by simonmar, 5 years ago) (diff)

--

NOTE: this page relates to the new GHC build system, due to be added to GHC in April 2009. For documentation of the "old" build system, see Building/Old/Using? and Building/BuildSystem?.

Installing GHC from a build tree

After GHC has been built, it can be installed on your system with

$ make install

This will build anything that isn't up-to-date, copy the files into the right places (see below) and make sure all the packages are registered properly.

Layout of the installed files

This section describes how the files of a GHC installation are laid out. The root of the GHC installation is specified via the --prefix flag to configure (see Running the configure script), and we refer to that location as $(prefix). If you are wondering what $(prefix) is, and hence where the make system is going to install GHC, just say (see debugging the build system)

$ make show VALUE=prefix

A GHC installation typically has three parts:

bindir
(default: $(prefix)/bin)
This is where the programs that you can run are installed, such as ghc, ghci, ghc-pkg, and haddock.
libdir
(default: Unix: $(prefix)/lib/ghc-<version>, Windows: $(prefix)/lib)
Where all of GHC's support files are kept, including package.conf, the header files, and the libraries. The location of libdir can be found by asking GHC: ghc --print-libdir. Normally you shouldn't have to look in here, and you shouldn't install extra files here. The only reason you might need to know the location of libdir at all is for passing to the GHC API, and the best way to do that is to use the ghc-paths package.
docdir
(default: Unix: $(prefix)/share/doc/ghc, Windows: $(prefix)/doc)
Where the documentation is installed.

On Unix systems you can change libdir and bindir using the --libdir and --bindir options respectively, and the location of the documentation can be changed using --datadir. On Windows all you can do is change $(prefix), because GHC finds the rest of its files by knowing their location relative to the ghc.exe binary, so the layout of the install tree is fixed (see How GHC finds its files, below).

To see how the install directories are derived from $(prefix), look in mk/config.mk.in.

The installed copy of MinGW on Windows

On Windows, GHC also comes with a copy of (most of) MinGW, in $(prefix)/mingw. So for instance, you can invoke the gcc that comes with GHC as $(prefix)/mingw/bin/gcc (replacing $(prefix) appropriately).

How GHC finds its files

GHC, when it starts, needs to find libdir, so that it can read package.conf, and find things like the unlit program. It does this in one of two ways:

  • On Windows, libdir is in a fixed location relative to the ghc.exe binary, namely ../lib. The advantage of the Windows way is that the installed GHC on Windows is independent of its location; it can be moved anywhere on the system, as long as the layout of the tree remains intact (however, links from the start menu and other shortcuts will break if you do this).
  • On Unix, it is standard to be able to change libdir relative to bindir, and also it is typically hard (or at least non-portable) to find the pathname to a running binary. So on Unix systems we use a different method: we make a wrapper script that passes the location of libdir to the GHC binary proper using the -B flag. Therefore, $(bindir)/ghc is a script that invokes $(libdir)/ghc passing it -B<libdir>, and the rest of the command-line arguments. All the other files that GHC needs are found by reading the package.conf file.

Relocating a GHC installation

On Windows, because GHC finds its files relative to itself, the whole installed tree can be relocated elsewhere in the filesystem, and everything will continue to work (except that if there are shortcuts or links from elsewhere, such as start menu items, these will need to be updated to point to the new location).

On Unix, GHC installations contain hardcoded paths in the package.conf file and also in the wrapper scripts for ghc and ghc-pkg in $bindir. So relocating the bits of a GHC installation on a Unix system is much harder; these paths would have to be updated manually.