On some platforms GHC supports building Haskell code into shared libraries. Shared libraries are also sometimes known as dynamic libraries, in particular on Windows they are referred to as dynamic link libraries (DLLs).
Shared libraries allow a single instance of some pre-compiled code to be shared between several programs. In contrast, with static linking the code is copied into each program. Using shared libraries can thus save disk space. They also allow a single copy of code to be shared in memory between several programs that use it. Shared libraires are often used as a way of structuring large projects, especially where different parts are written in different programming languages. Shared libraries are also commonly used as a plugin mechanism by various applications. This is particularly common on Windows using COM.
In GHC version 6.12 building shared libraries is supported for Linux on x86 and x86-64 architectures and there is partial support on Windows (see Section 11.6, “Building and using Win32 DLLs ”). The crucial difference in support on Windows is that it is not currently possible to build each Haskell package as a separate DLL, it is only possible to link an entire Haskell program as one massive DLL.
Building and using shared libraries is slightly more complicated than
building and using static libraries. When using Cabal much of the detail
is hidden, just use --enable-shared when configuring a
package to build it into a shared library, or to link it against other
packages built as shared libraries. The additional complexity when
building code is to distinguish whether the code will be used in a shared
library or will use shared library versions of other packages it depends
on. There is additional complexity when installing and distributing
shared libraries or programs that use shared libraries, to ensure that
all shared libraries that are required at runtime are present in suitable
locations.
To build a simple program and have it use shared libraries for the
runtime system and the base libraries use the
-dynamic flag:
ghc --make -dynamic Main.hs
This has two effects. The first is to compile the code in such a way
that it can be linked against shared library versions of Haskell
packages (such as base). The second is when linking, to link against
the shared versions of the packages' libraries rather than the static
versions. Obviously this requires that the packages were build with
shared libraries. On supported platforms GHC comes with shared
libraries for all the core packages, but if you install extra packages
(e.g. with Cabal) then they would also have to be built with shared
libraries (--enable-shared for Cabal).
You can build Haskell code into a shared library and make a package to be
used by other Haskell programs. The easiest way is using Cabal, simply
configure the Cabal package with the --enable-shared
flag.
If you want to do the steps manually or are writing your own build system then there are certain conventions that must be followed. Building a shared library that exports Haskell code, to be used by other Haskell code is a bit more complicated than it is for one that exports a C API and will be used by C code. If you get it wrong you will usually end up with linker errors.
In particular Haskell shared libraries must be made into packages. You cannot freely assign which modules go in which shared libraries. The Haskell shared libraries must match the package boundaries. Most of the conventions GHC expects when using packages are described in Section 4.8.7, “Building a package from Haskell source”.
GHC handles references to symbols within the same
shared library (or main executable binary) differently from references
to symbols between different shared libraries. GHC
needs to know for each imported module if that module lives locally in
the same shared lib or in a separate shared lib. The way it does this
is by using packages. When using -dynamic, a module
from a separate package is assumed to come from a separate shared lib,
while modules from the same package (or the default "main" package) are
assumed to be within the same shared lib (or main executable binary).
Most of the conventions GHC expects when using packages are described
in Section 4.8.7, “Building a package from Haskell source”. In addition note that GHC
expects the .hi files to use the extension
.dyn_hi. The other requirements are the same as for
C libraries and are described below, in particular the use of the flags
-dynamic, -fPIC and
-shared.
Building Haskell code into a shared library is a good way to include Haskell code in a larger mixed-language project. While with static linking it is recommended to use GHC to perform the final link step, with shared libaries a Haskell library can be treated just like any other shared libary. The linking can be done using the normal system C compiler or linker.
It is possible to load shared libraries generated by GHC in other programs not written in Haskell, so they are suitable for using as plugins. Of course to construct a plugin you will have to use the FFI to export C functions and follow the rules about initialising the RTS. See Section 8.2.1.2, “Making a Haskell library that can be called from foreign code”. In particular you will probably want to export a C function from your shared library to initialise the plugin before any Haskell functions are called.
To build Haskell modules that export a C API into a shared library use
the -dynamic, -fPIC and
-shared flags:
ghc --make -dynamic -shared -fPIC Foo.hs -o libfoo.so
As before, the -dynamic flag specifies that this
library links against the shared library versions of the rts and base
package. The -fPIC flag is required for all code
that will end up in a shared library. The -shared
flag specifies to make a shared library rather than a program. To make
this clearer we can break this down into separate compliation and link
steps:
ghc -dynamic -fPIC -c Foo.hs ghc -dynamic -shared Foo.o -o libfoo.so
In principle you can use -shared without
-dynamic in the link step. That means to
statically link the rts all the base libraries into your new shared
library. This would make a very big, but standalone shared library.
Indeed this is exactly what we must currently do on Windows where
-dynamic is not yet supported (see Section 11.6, “Building and using Win32 DLLs
”).
On most platforms however that would require all the static libraries
to have been built with -fPIC so that the code is
suitable to include into a shared library and we do not do that at the
moment.
Warning: if your shared library exports a Haskell API then you cannot directly link it into another Haskell program and use that Haskell API. You will get linker errors. You must instead make it into a package as described in the section above.
The primary difficulty with managing shared libraries is arranging things such that programs can find the libraries they need at runtime. The details of how this works varies between platforms, in particular the three major systems: Unix ELF platforms, Windows and Mac OS X.
On Unix there are two mechanisms. Shared libraries can be installed
into standard locations that the dynamic linker knows about. For
example /usr/lib or
/usr/local/lib on most systems. The other mechanism
is to use a "runtime path" or "rpath" embedded into programs and
libraries themselves. These paths can either be absolute paths or on at
least Linux and Solaris they can be paths relative to the program or
libary itself. In principle this makes it possible to construct fully
relocatable sets of programs and libraries.
GHC has a -dynload linking flag to select the method
that is used to find shared libraries at runtime. There are currently
three modes:
A system-dependent mode. This is also the default mode. On Unix ELF systems this embeds rpaths into the shared library or executable. In particular it uses absolute paths to where the shared libraries for the rts and each package can be found. This means the program can immediately be run and it will be able to find the libraries it needs. However it may not be suitable for deployment if the libraries are installed in a different location on another machine.
This does not embed any runtime paths. It relies on the shared
libraries being available in a standard location or in a
directory given by the LD_LIBRARY_PATH
environment variable.
This mode generates a wrapper program which in turn calls the real program (in the same directory but with a .dyn extension) in such a way that it can find the shared libraries that it needs. At the current time this mode is somewhat experimental.
To use relative paths for dependent libraries on Linux and Solaris you
can use the deploy mode and pass suitable a -rpath
flag to the linker:
ghc -dynamic Main.hs -o main -lfoo -L. -optl-Wl,-rpath,'$ORIGIN'
This assumes that the library libfoo.so is in the
current directory and will be able to be found in the same directory as
the executable main once the program is deployed.
Similarly it would be possible to use a subdirectory relative to the
executable e.g. -optl-Wl,-rpath,'$ORIGIN/lib'.