guile/doc/ref/api-foreign.texi

@c -*-texinfo-*-
@c This is part of the GNU Guile Reference Manual.
@c Copyright (C)  1996, 1997, 2000-2004, 2007-2014, 2016-2017, 2021
@c   Free Software Foundation, Inc.
@c See the file guile.texi for copying conditions.

@node Foreign Function Interface
@section Foreign Function Interface
@cindex foreign function interface
@cindex ffi

Sometimes you need to use libraries written in C or Rust or some other
non-Scheme language.  More rarely, you might need to write some C to
extend Guile.  This section describes how to load these ``foreign
libraries'', look up data and functions inside them, and so on.

@menu
* Foreign Libraries::              Dynamically linking to libraries.
* Foreign Extensions::             Extending Guile in C with loadable modules.
* Foreign Pointers::               Pointers to C data or functions.
* Foreign Types::                  Expressing C types in Scheme.
* Foreign Functions::              Simple calls to C procedures.
* Void Pointers and Byte Access::  Pointers into the ether.
* Foreign Structs::                Packing and unpacking structs.
* More Foreign Functions::         Advanced examples.
@end menu


@node Foreign Libraries
@subsection Foreign Libraries

Just as Guile can load up Scheme libraries at run-time, Guile can also
load some system libraries written in C or other low-level languages.
We refer to these as dynamically-loadable modules as @dfn{foreign
libraries}, to distinguish them from native libraries written in Scheme
or other languages implemented by Guile.
@cindex foreign libraries
@cindex libraries, foreign

Foreign libraries usually come in two forms.  Some foreign libraries are
part of the operating system, such as the compression library
@code{libz}.  These shared libraries are built in such a way that many
programs can use their functionality without duplicating their code.
When a program written in C is built, it can declare that it uses a
specific set of shared libraries.
@cindex shared libraries
@cindex libraries, shared
When the program is run, the operating system takes care of locating and
loading the shared libraries.

The operating system components that can dynamically load and link
shared libraries when a program is run are also available
programmatically during a program's execution.  This is the interface
that's most useful for Guile, and this is what we mean in Guile when we
refer to @dfn{dynamic linking}.  Dynamic linking at run-time is
sometimes called @dfn{dlopening}, to distinguish it from the dynamic
linking that happens at program start-up.
@cindex dynamic linking
@cindex dlopening

The other kind of foreign library is sometimes known as a module,
plug-in, bundle, or an extension.  These foreign libraries aren't meant
to be linked to by C programs, but rather only to be dynamically loaded
at run-time -- they extend some main program with functionality, but
don't stand on their own.  Sometimes a Guile library will implement some
of its functionality in a loadable module.

In either case, the interface on the Guile side is the same.  You load
the interface using @code{load-foreign-library}.  The resulting foreign
library object implements a simple lookup interface whereby the user can
get addresses of data or code exported by the library.  There is no
facility to inspect foreign libraries; you have to know what's in there
already before you look.

Routines for loading foreign libraries and accessing their contents are
implemented in the @code{(system foreign-library)} module.

@example
(use-modules (system foreign-library))
@end example

@deffn {Scheme Procedure} load-foreign-library [library] @
       [#:extensions=system-library-extensions] @
       [#:search-ltdl-library-path?=#t] @
       [#:search-path=search-path] @
       [#:search-system-paths?=#t] [#:lazy?=#t] [#:global=#f]
Find the shared library denoted by @var{library} (a string or @code{#f})
and link it into the running Guile application.  When everything works
out, return a Scheme object suitable for representing the linked object
file.  Otherwise an error is thrown.

If @var{library} argument is omitted, it defaults to @code{#f}.  If
@code{library} is false, the resulting foreign library gives access to
all symbols available for dynamic linking in the main binary.

It is not necessary to include any extension such as @code{.so} in
@var{library}.  For each system, Guile has a default set of extensions
that it will try.  On GNU systems, the default extension set is just
@code{.so}; on Windows, just @code{.dll}; and on Darwin (Mac OS), it is
@code{.bundle}, @code{.so}, and @code{.dylib}.  Pass @code{#:extensions
@var{extensions}} to override the default extensions list.  If
@var{library} contains one of the extensions, no extensions are tried,
so it is possible to specify the extension if you know exactly what file
to load.

Unless @var{library} denotes an absolute file name or otherwise contains
a directory separator (@code{/}, and also @code{\} on Windows), Guile
will search for the library in the directories listed in
@var{search-paths}.  The default search path has three components, which
can all be overriden by colon-delimited (semicolon on Windows)
environment variables:

@table @env
@item GUILE_EXTENSIONS_PATH
This is the main environment variable for users to add directories
containing Guile extensions.  The default value has no entries.  This
environment variable was added in Guile 3.0.6.
@item LTDL_LIBRARY_PATH
Before Guile 3.0.6, Guile loaded foreign libraries using @code{libltdl},
the dynamic library loader provided by libtool.  This loader used
@env{LTDL_LIBRARY_PATH}, and for backwards compatibility we still
support that path.

However, @code{libltdl} would not only open @code{.so} (or @code{.dll}
and so on) files, but also the @code{.la} files created by libtool.  In
installed libraries -- libraries that are in the target directories of
@code{make install} -- @code{.la} files are never needed, to the extent
that most GNU/Linux distributions remove them entirely.  It is
sufficient to just load the @code{.so} (or @code{.dll} and so on) files,
which are always located in the same directory as the @code{.la} files.

But for uninstalled dynamic libraries, like those in a build tree, the
situation is a bit of a mess.  If you have a project that uses libtool
to build libraries -- which is the case for Guile, and for most projects
using autotools -- and you build @file{foo.so} in directory @file{D},
libtool will put @file{foo.la} in @file{D}, but @file{foo.so} gets put
into @file{D/.libs}.

Users were mostly oblivious to this situation, as @code{libltdl} had
special logic to be able to read the @code{.la} file to know where to
find the @code{.so}, even from an uninstalled build tree, preventing the
existence of @file{.libs} from leaking out to the user.

We don't use libltdl now, essentially for flexibility and
error-reporting reasons.  But, to keep this old use-case working, if
@var{search-ltdl-library-path?} is true, we add each entry of
@code{LTDL_LIBRARY_PATH} to the default extensions load path,
additionally adding the @file{.libs} subdirextories for each entry, in
case there are @file{.so} files there instead of alongside the
@file{.la} files.
@item GUILE_SYSTEM_EXTENSIONS_PATH
The last path in Guile's search path belongs to Guile itself, and
defaults to the libdir and the extensiondir, in that order.  For
example, if you install to @file{/opt/guile}, these would probably be
@file{/opt/guile/lib} and
@code{/opt/guile/lib/guile/@value{EFFECTIVE-VERSION}/extensions},
respectively.  @xref{Parallel Installations}, for more details on
@code{extensionsdir}.
@end table

Finally, if no library is found in the search path, and if @var{library}
is not absolute and does not include directory separators, and if
@var{search-system-paths?} is true, the operating system may have its
own logic for where to locate @var{library}.  For example, on GNU, there
will be a default set of paths (often @file{/usr/lib} and @file{/lib},
though it depends on the system), and the @code{LD_LIBRARY_PATH}
environment variable can add additional paths.  Other operating systems
have other conventions.

Falling back to the operating system for search is usually not a great
thing; it is a recipe for making programs that work on one machine but
not on others.  Still, when wrapping system libraries, it can be the
only way to get things working at all.

If @var{lazy?} is true (the default), Guile will request the operating
system to resolve symbols used by the loaded library as they are first
used.  If @var{global?} is true, symbols defined by the loaded library
will be available when other modules need to resolve symbols; the
default is @code{#f}, which keeps symbols local.
@end deffn

The environment variables mentioned above are parsed when the
foreign-library module is first loaded and bound to parameters.  Null
path components, for example the three components of
@env{GUILE_SYSTEM_EXTENSIONS_PATH="::"}, are ignored.

@deffn {Scheme Parameter} guile-extensions-path
@deffnx {Scheme Parameter} ltdl-library-path
@deffnx {Scheme Parameter} guile-system-extensions-path
Parameters whose initial values are taken from
@env{GUILE_EXTENSIONS_PATH}, @env{LTDL_LIBRARY_PATH}, and
@env{GUILE_SYSTEM_EXTENSIONS_PATH}, respectively.  @xref{Parameters}.
The current values of these parameters are used when building the search
path when @code{load-foreign-library} is called, unless the caller
explicitly passes a @code{#:search-path} argument.
@end deffn

@deffn {Scheme Procedure} foreign-library? obj
Return @code{#t} if @var{obj} is a foreign library, or @code{#f}
otherwise.
@end deffn


@node Foreign Extensions
@subsection Foreign Extensions

One way to use shared libraries is to extend Guile.  Such loadable
modules generally define one distinguished initialization function that,
when called, will use the @code{libguile} API to define procedures in
the current module.

Concretely, you might extend Guile with an implementation of the Bessel
function, @code{j0}:

@smallexample
#include <math.h>
#include <libguile.h>

SCM
j0_wrapper (SCM x)
@{
  return scm_from_double (j0 (scm_to_double (x, "j0")));
@}

void
init_math_bessel (void)
@{
  scm_c_define_gsubr ("j0", 1, 0, 0, j0_wrapper);
@}
@end smallexample

The C source file would then need to be compiled into a shared library.
On GNU/Linux, the compiler invocation might look like this:

@smallexample
gcc -shared -o bessel.so -fPIC bessel.c
@end smallexample

A good default place to put shared libraries that extend Guile is into
the extensions dir.  From the command line or a build script, invoke
@code{pkg-config --variable=extensionsdir
guile-@value{EFFECTIVE-VERSION}} to print the extensions dir.
@xref{Parallel Installations}, for more details.

Guile can load up @code{bessel.so} via @code{load-extension}.

@deffn {Scheme Procedure} load-extension lib init
@deffnx {C Function} scm_load_extension (lib, init)
Load and initialize the extension designated by LIB and INIT.
@end deffn

The normal way for a extension to be used is to write a small Scheme
file that defines a module, and to load the extension into this
module.  When the module is auto-loaded, the extension is loaded as
well.  For example:

@lisp
(define-module (math bessel)
  #:export (j0))

(load-extension "bessel" "init_math_bessel")
@end lisp

This @code{load-extension} invocation loads the @code{bessel} library
via @code{(load-foreign-library "bessel")}, then looks up the
@code{init_math_bessel} symbol in the library, treating it as a function
of no arguments, and calls that function.

If you decide to put your extension outside the default search path for
@code{load-foreign-library}, probably you should adapt the Scheme module
to specify its absolute path.  For example, if you use @code{automake}
to build your extension and place it in @code{$(pkglibdir)}, you might
define a build-parameters module that gets created by the build system:

@example
(define-module (math config)
  #:export (extensiondir))
(define extensiondir "PKGLIBDIR")
@end example

This file would be @code{config.scm.in}.  You would define a @code{make}
rule to substitute in the absolute installed file name:

@example
config.scm: config.scm.in
        sed 's|PKGLIBDIR|$(pkglibdir)|' <$< >$@
@end example

Then your @code{(math bessel)} would import @code{(math config)}, then
@code{(load-extension (in-vicinity extensiondir "bessel")
"init_math_bessel")}.

An alternate approach would be to rebind the
@code{guile-extensions-path} parameter, or its corresponding environment
variable, but note that changing those parameters applies to other users
of @code{load-foreign-library} as well.

Note that the new primitives that the extension adds to Guile with
@code{scm_c_define_gsubr} (@pxref{Primitive Procedures}) or with any of
the other mechanisms are placed into the module that is current when the
@code{scm_c_define_gsubr} is executed, so to be clear about what goes
vwhere it's best to include the @code{load-extension} in a module, as
above.  Alternately, the C code can use @code{scm_c_define_module} to
specify which module is being created:

@smallexample
static void
do_init (void *unused)
@{
  scm_c_define_gsubr ("j0", 1, 0, 0, j0_wrapper);
  scm_c_export ("j0", NULL);
@}

void
init_math_bessel ()
@{
  scm_c_define_module ("math bessel", do_init, NULL);
@}
@end smallexample

And yet... if what we want is just the @code{j0} function, it seems like
a lot of ceremony to have to compile a Guile-specific wrapper library
complete with an initialization function and wraper module to allow
Guile users to call it.  There is another way, but to get there, we have
to talk about function pointers and function types first.  @xref{Foreign
Functions}, to skip to the good parts.


@node Foreign Pointers
@subsection Foreign Pointers

Foreign libraries are essentially key-value mappings, where the keys are
names of definitions and the values are the addresses of those
definitions.  To look up the address of a definition, use
@code{foreign-library-pointer} from the @code{(system foreign-library)}
module.

@deffn {Scheme Procedure} foreign-library-pointer lib name
Return a ``wrapped pointer'' for the symbol @var{name} in the shared
object referred to by @var{lib}.  The returned pointer points to a C
object.

As a convenience, if @var{lib} is not a foreign library, it will be
passed to @code{load-foreign-library}.
@end deffn

If we continue with the @code{bessel.so} example from before, we can get
the address of the @code{init_math_bessel} function via:

@example
(use-modules (system foreign-library))
(define init (foreign-library-pointer "bessel" "init_math_bessel"))
init
@result{} #<pointer 0x7fb35b1b4688>
@end example

A value returned by @code{foreign-library-pointer} is a Scheme wrapper
for a C pointer.  Pointers are a data type in Guile that is disjoint
from all other types.  The next section discusses ways to dereference
pointers, but before then we describe the usual type predicates and so
on.

Note that the rest of the interfaces in this section are part of the
@code{(system foreign)} library:

@example
(use-modules (system foreign))
@end example

@deffn {Scheme Procedure} pointer-address pointer
@deffnx {C Function} scm_pointer_address (pointer)
Return the numerical value of @var{pointer}.

@example
(pointer-address init)
@result{} 139984413364296 ; YMMV
@end example
@end deffn

@deffn {Scheme Procedure} make-pointer address [finalizer]
Return a foreign pointer object pointing to @var{address}.  If
@var{finalizer} is passed, it should be a pointer to a one-argument C
function that will be called when the pointer object becomes
unreachable.
@end deffn

@deffn {Scheme Procedure} pointer? obj
Return @code{#t} if @var{obj} is a pointer object, or @code{#f}
otherwise.
@end deffn

@defvr {Scheme Variable} %null-pointer
A foreign pointer whose value is 0.
@end defvr

@deffn {Scheme Procedure} null-pointer? pointer
Return @code{#t} if @var{pointer} is the null pointer, @code{#f} otherwise.
@end deffn

For the purpose of passing SCM values directly to foreign functions, and
allowing them to return SCM values, Guile also supports some unsafe
casting operators.

@deffn {Scheme Procedure} scm->pointer scm
Return a foreign pointer object with the @code{object-address}
of @var{scm}.
@end deffn

@deffn {Scheme Procedure} pointer->scm pointer
Unsafely cast @var{pointer} to a Scheme object.
Cross your fingers!
@end deffn

Sometimes you want to give C extensions access to the dynamic FFI.  At
that point, the names get confusing, because ``pointer'' can refer to a
@code{SCM} object that wraps a pointer, or to a @code{void*} value.  We
will try to use ``pointer object'' to refer to Scheme objects, and
``pointer value'' to refer to @code{void *} values.

@deftypefn {C Function} SCM scm_from_pointer (void *ptr, void (*finalizer) (void*))
Create a pointer object from a pointer value.

If @var{finalizer} is non-null, Guile arranges to call it on the pointer
value at some point after the pointer object becomes collectable.
@end deftypefn

@deftypefn {C Function} void* scm_to_pointer (SCM obj)
Unpack the pointer value from a pointer object.
@end deftypefn

@node Foreign Types
@subsection Foreign Types

From Scheme's perspective, foreign pointers are shards of chaos.  The
user can create a foreign pointer for any address, and do with it what
they will.  The only thing that lends a sense of order to the whole is a
shared hallucination that certain storage locations have certain types.
When making Scheme wrappers for foreign interfaces, we hide the madness
by explicitly representing the the data types of parameters and fields.

These ``foreign type values'' may be constructed using the constants and
procedures from the @code{(system foreign)} module, which may be loaded
like this:

@example
(use-modules (system foreign))
@end example

@code{(system foreign)} exports a number of values expressing the basic
C types.

@defvr {Scheme Variable} int8
@defvrx {Scheme Variable} uint8
@defvrx {Scheme Variable} uint16
@defvrx {Scheme Variable} int16
@defvrx {Scheme Variable} uint32
@defvrx {Scheme Variable} int32
@defvrx {Scheme Variable} uint64
@defvrx {Scheme Variable} int64
@defvrx {Scheme Variable} float
@defvrx {Scheme Variable} double
These values represent the C numeric types of the specified sizes and
signednesses.
@end defvr

In addition there are some convenience bindings for indicating types of
platform-dependent size.

@defvr {Scheme Variable} int
@defvrx {Scheme Variable} unsigned-int
@defvrx {Scheme Variable} long
@defvrx {Scheme Variable} unsigned-long
@defvrx {Scheme Variable} short
@defvrx {Scheme Variable} unsigned-short
@defvrx {Scheme Variable} size_t
@defvrx {Scheme Variable} ssize_t
@defvrx {Scheme Variable} ptrdiff_t
@defvrx {Scheme Variable} intptr_t
@defvrx {Scheme Variable} uintptr_t
Values exported by the @code{(system foreign)} module, representing C
numeric types. For example, @code{long} may be @code{equal?} to
@code{int64} on a 64-bit platform.
@end defvr

@defvr {Scheme Variable} void
The @code{void} type.  It can be used as the first argument to
@code{pointer->procedure} to wrap a C function that returns nothing.
@end defvr

In addition, the symbol @code{*} is used by convention to denote pointer
types.  Procedures detailed in the following sections, such as
@code{pointer->procedure}, accept it as a type descriptor.

@node Foreign Functions
@subsection Foreign Functions

The most natural thing to do with a dynamic library is to grovel around
in it for a function pointer: a @dfn{foreign function}.  Load the
@code{(system foreign)} module to use these Scheme interfaces.

@example
(use-modules (system foreign))
@end example

@deffn {Scheme Procedure} pointer->procedure return_type func_ptr arg_types @
                                             [#:return-errno?=#f]
@deffnx {C Function} scm_pointer_to_procedure (return_type, func_ptr, arg_types)
@deffnx {C Function} scm_pointer_to_procedure_with_errno (return_type, func_ptr, arg_types)

Make a foreign function.

Given the foreign void pointer @var{func_ptr}, its argument and
return types @var{arg_types} and @var{return_type}, return a
procedure that will pass arguments to the foreign function
and return appropriate values.

@var{arg_types} should be a list of foreign types.
@code{return_type} should be a foreign type. @xref{Foreign Types}, for
more information on foreign types.

If @var{return-errno?} is true, or when calling
@code{scm_pointer_to_procedure_with_errno}, the returned procedure will
return two values, with @code{errno} as the second value.
@end deffn

Finally, in @code{(system foreign-library)} there is a convenient
wrapper function, joining together @code{foreign-libary-pointer} and
@code{procedure->pointer}:

@deffn {Scheme Procedure} foreign-library-function lib name @
       [#:return-type=void] [#:arg-types='()] [#:return-errno?=#f]
Load the address of @var{name} from @var{lib}, and treat it as a
function taking arguments @var{arg-types} and returning
@var{return-type}, optionally also with errno.

An invocation of @code{foreign-library-function} is entirely equivalent
to:
@example
(pointer->procedure @var{return-type}
                    (foreign-library-pointer @var{lib} @var{name})
                    @var{arg-types}
                    #:return-errno? @var{return-errno?}).
@end example
@end deffn

Pulling all this together, here is a better definition of @code{(math
bessel)}:

@example
(define-module (math bessel)
  #:use-module (system foreign)
  #:use-module (system foreign-library)
  #:export (j0))

(define j0
  (foreign-library-function "libm" "j0"
                            #:return-type double
                            #:arg-types (list double)))
@end example

That's it! No C at all.

Before going on to more detailed examples, the next two sections discuss
how to deal with data that is more complex than, say, @code{int8}.
@xref{More Foreign Functions}, to continue with foreign function examples.

@node Void Pointers and Byte Access
@subsection Void Pointers and Byte Access

Wrapped pointers are untyped, so they are essentially equivalent to C
@code{void} pointers.  As in C, the memory region pointed to by a
pointer can be accessed at the byte level.  This is achieved using
@emph{bytevectors} (@pxref{Bytevectors}).  The @code{(rnrs bytevectors)}
module contains procedures that can be used to convert byte sequences to
Scheme objects such as strings, floating point numbers, or integers.

Load the @code{(system foreign)} module to use these Scheme interfaces.

@example
(use-modules (system foreign))
@end example

@deffn {Scheme Procedure} pointer->bytevector pointer len [offset [uvec_type]]
@deffnx {C Function} scm_pointer_to_bytevector (pointer, len, offset, uvec_type)
Return a bytevector aliasing the @var{len} bytes pointed to by
@var{pointer}.

The user may specify an alternate default interpretation for the memory
by passing the @var{uvec_type} argument, to indicate that the memory is
an array of elements of that type.  @var{uvec_type} should be something
that @code{array-type} would return, like @code{f32} or @code{s16}.

When @var{offset} is passed, it specifies the offset in bytes relative
to @var{pointer} of the memory region aliased by the returned
bytevector.

Mutating the returned bytevector mutates the memory pointed to by
@var{pointer}, so buckle your seatbelts.
@end deffn

@deffn {Scheme Procedure} bytevector->pointer bv [offset]
@deffnx {C Function} scm_bytevector_to_pointer (bv, offset)
Return a pointer pointer aliasing the memory pointed to by @var{bv} or
@var{offset} bytes after @var{bv} when @var{offset} is passed.
@end deffn

In addition to these primitives, convenience procedures are available:

@deffn {Scheme Procedure} dereference-pointer pointer
Assuming @var{pointer} points to a memory region that holds a pointer,
return this pointer.
@end deffn

@deffn {Scheme Procedure} string->pointer string [encoding]
Return a foreign pointer to a nul-terminated copy of @var{string} in the
given @var{encoding}, defaulting to the current locale encoding.  The C
string is freed when the returned foreign pointer becomes unreachable.

This is the Scheme equivalent of @code{scm_to_stringn}.
@end deffn

@deffn {Scheme Procedure} pointer->string pointer [length] [encoding]
Return the string representing the C string pointed to by @var{pointer}.
If @var{length} is omitted or @code{-1}, the string is assumed to be
nul-terminated.  Otherwise @var{length} is the number of bytes in memory
pointed to by @var{pointer}.  The C string is assumed to be in the given
@var{encoding}, defaulting to the current locale encoding.

This is the Scheme equivalent of @code{scm_from_stringn}.
@end deffn

@cindex wrapped pointer types
Most object-oriented C libraries use pointers to specific data
structures to identify objects.  It is useful in such cases to reify the
different pointer types as disjoint Scheme types.  The
@code{define-wrapped-pointer-type} macro simplifies this.

@deffn {Scheme Syntax} define-wrapped-pointer-type type-name pred wrap unwrap print
Define helper procedures to wrap pointer objects into Scheme objects
with a disjoint type.  Specifically, this macro defines:

@itemize
@item @var{pred}, a predicate for the new Scheme type;
@item @var{wrap}, a procedure that takes a pointer object and returns an
object that satisfies @var{pred};
@item @var{unwrap}, which does the reverse.
@end itemize

@var{wrap} preserves pointer identity, for two pointer objects @var{p1}
and @var{p2} that are @code{equal?}, @code{(eq? (@var{wrap} @var{p1})
(@var{wrap} @var{p2})) @result{} #t}.

Finally, @var{print} should name a user-defined procedure to print such
objects.  The procedure is passed the wrapped object and a port to write
to.

For example, assume we are wrapping a C library that defines a type,
@code{bottle_t}, and functions that can be passed @code{bottle_t *}
pointers to manipulate them.  We could write:

@example
(define-wrapped-pointer-type bottle
  bottle?
  wrap-bottle unwrap-bottle
  (lambda (b p)
    (format p "#<bottle of ~a ~x>"
            (bottle-contents b)
            (pointer-address (unwrap-bottle b)))))

(define grab-bottle
  ;; Wrapper for `bottle_t *grab (void)'.
  (let ((grab (foreign-library-function libbottle "grab_bottle"
                                        #:return-type '*)))
    (lambda ()
      "Return a new bottle."
      (wrap-bottle (grab)))))

(define bottle-contents
  ;; Wrapper for `const char *bottle_contents (bottle_t *)'.
  (let ((contents (foreign-library-function libbottle "bottle_contents"
                                            #:return-type '*
                                            #:arg-types  '(*))))
    (lambda (b)
      "Return the contents of B."
      (pointer->string (contents (unwrap-bottle b))))))

(write (grab-bottle))
@result{} #<bottle of Ch@^ateau Haut-Brion 803d36>
@end example

In this example, @code{grab-bottle} is guaranteed to return a genuine
@code{bottle} object satisfying @code{bottle?}.  Likewise,
@code{bottle-contents} errors out when its argument is not a genuine
@code{bottle} object.
@end deffn

As another example, currently Guile has a variable, @code{scm_numptob},
as part of its API. It is declared as a C @code{long}. So, to read its
value, we can do:

@example
(use-modules (system foreign))
(use-modules (rnrs bytevectors))
(define numptob
  (foreign-library-pointer #f "scm_numptob"))
numptob
(bytevector-uint-ref (pointer->bytevector numptob (sizeof long))
                     0 (native-endianness)
                     (sizeof long))
@result{} 8
@end example

If we wanted to corrupt Guile's internal state, we could set
@code{scm_numptob} to another value; but we shouldn't, because that
variable is not meant to be set.  Indeed this point applies more widely:
the C API is a dangerous place to be.  Not only might setting a value
crash your program, simply accessing the data pointed to by a dangling
pointer or similar can prove equally disastrous.

@node Foreign Structs
@subsection Foreign Structs

Finally, one last note on foreign values before moving on to actually
calling foreign functions. Sometimes you need to deal with C structs,
which requires interpreting each element of the struct according to the
its type, offset, and alignment. The @code{(system foreign)} module has
some primitives to support this.

@example
(use-modules (system foreign))
@end example

@deffn {Scheme Procedure} sizeof type
@deffnx {C Function} scm_sizeof (type)
Return the size of @var{type}, in bytes.

@var{type} should be a valid C type, like @code{int}.
Alternately @var{type} may be the symbol @code{*}, in which
case the size of a pointer is returned. @var{type} may
also be a list of types, in which case the size of a
@code{struct} with ABI-conventional packing is returned.
@end deffn

@deffn {Scheme Procedure} alignof type
@deffnx {C Function} scm_alignof (type)
Return the alignment of @var{type}, in bytes.

@var{type} should be a valid C type, like @code{int}.
Alternately @var{type} may be the symbol @code{*}, in which
case the alignment of a pointer is returned. @var{type} may
also be a list of types, in which case the alignment of a
@code{struct} with ABI-conventional packing is returned.
@end deffn

Guile also provides some convenience methods to pack and unpack foreign
pointers wrapping C structs.

@deffn {Scheme Procedure} make-c-struct types vals
Create a foreign pointer to a C struct containing @var{vals} with types
@code{types}.

@var{vals} and @code{types} should be lists of the same length.
@end deffn

@deffn {Scheme Procedure} parse-c-struct foreign types
Parse a foreign pointer to a C struct, returning a list of values.

@code{types} should be a list of C types.
@end deffn

For example, to create and parse the equivalent of a @code{struct @{
int64_t a; uint8_t b; @}}:

@example
(parse-c-struct (make-c-struct (list int64 uint8)
                               (list 300 43))
                (list int64 uint8))
@result{} (300 43)
@end example

As yet, Guile only has convenience routines to support
conventionally-packed structs. But given the @code{bytevector->pointer}
and @code{pointer->bytevector} routines, one can create and parse
tightly packed structs and unions by hand. See the code for
@code{(system foreign)} for details.

@node More Foreign Functions
@subsection More Foreign Functions

It is possible to pass pointers to foreign functions, and to return them
as well.  In that case the type of the argument or return value should
be the symbol @code{*}, indicating a pointer. For example, the following
code makes @code{memcpy} available to Scheme:

@example
(use-modules (system foreign))
(define memcpy
  (foreign-library-function #f "memcpy"
                            #:return-type '*
                            #:arg-types (list '* '* size_t)))
@end example

To invoke @code{memcpy}, one must pass it foreign pointers:

@example
(use-modules (rnrs bytevectors))

(define src-bits
  (u8-list->bytevector '(0 1 2 3 4 5 6 7)))
(define src
  (bytevector->pointer src-bits))
(define dest
  (bytevector->pointer (make-bytevector 16 0)))

(memcpy dest src (bytevector-length src-bits))

(bytevector->u8-list (pointer->bytevector dest 16))
@result{} (0 1 2 3 4 5 6 7 0 0 0 0 0 0 0 0)
@end example

One may also pass structs as values, passing structs as foreign
pointers. @xref{Foreign Structs}, for more information on how to express
struct types and struct values.

``Out'' arguments are passed as foreign pointers. The memory pointed to
by the foreign pointer is mutated in place.

@example
;; struct timeval @{
;;      time_t      tv_sec;     /* seconds */
;;      suseconds_t tv_usec;    /* microseconds */
;; @};
;; assuming fields are of type "long"

(define gettimeofday
  (let ((f (foreign-library-function #f "gettimeofday"
                                     #:return-type int
                                     #:arg-types (list '* '*)))
        (tv-type (list long long)))
    (lambda ()
      (let* ((timeval (make-c-struct tv-type (list 0 0)))
             (ret (f timeval %null-pointer)))
        (if (zero? ret)
            (apply values (parse-c-struct timeval tv-type))
            (error "gettimeofday returned an error" ret))))))

(gettimeofday)    
@result{} 1270587589
@result{} 499553
@end example

As you can see, this interface to foreign functions is at a very low,
somewhat dangerous level@footnote{A contribution to Guile in the form of
a high-level FFI would be most welcome.}.

@cindex callbacks
The FFI can also work in the opposite direction: making Scheme
procedures callable from C.  This makes it possible to use Scheme
procedures as ``callbacks'' expected by C function.

@deffn {Scheme Procedure} procedure->pointer return-type proc arg-types
@deffnx {C Function} scm_procedure_to_pointer (return_type, proc, arg_types)
Return a pointer to a C function of type @var{return-type}
taking arguments of types @var{arg-types} (a list) and
behaving as a proxy to procedure @var{proc}.  Thus
@var{proc}'s arity, supported argument types, and return
type should match @var{return-type} and @var{arg-types}.
@end deffn

As an example, here's how the C library's @code{qsort} array sorting
function can be made accessible to Scheme (@pxref{Array Sort Function,
@code{qsort},, libc, The GNU C Library Reference Manual}):

@example
(define qsort!
  (let ((qsort (foreign-library-function
                #f "qsort" #:arg-types (list '* size_t size_t '*))))
    (lambda (bv compare)
      ;; Sort bytevector BV in-place according to comparison
      ;; procedure COMPARE.
      (let ((ptr (procedure->pointer int
                                     (lambda (x y)
                                       ;; X and Y are pointers so,
                                       ;; for convenience, dereference
                                       ;; them before calling COMPARE.
                                       (compare (dereference-uint8* x)
                                                (dereference-uint8* y)))
                                     (list '* '*))))
        (qsort (bytevector->pointer bv)
               (bytevector-length bv) 1 ;; we're sorting bytes
               ptr)))))

(define (dereference-uint8* ptr)
  ;; Helper function: dereference the byte pointed to by PTR.
  (let ((b (pointer->bytevector ptr 1)))
    (bytevector-u8-ref b 0)))

(define bv
  ;; An unsorted array of bytes.
  (u8-list->bytevector '(7 1 127 3 5 4 77 2 9 0)))

;; Sort BV.
(qsort! bv (lambda (x y) (- x y)))

;; Let's see what the sorted array looks like:
(bytevector->u8-list bv)
@result{} (0 1 2 3 4 5 7 9 77 127)
@end example

And voil@`a!

Note that @code{procedure->pointer} is not supported (and not defined)
on a few exotic architectures.  Thus, user code may need to check
@code{(defined? 'procedure->pointer)}.  Nevertheless, it is available on
many architectures, including (as of libffi 3.0.9) x86, ia64, SPARC,
PowerPC, ARM, and MIPS, to name a few.

@c Local Variables:
@c TeX-master: "guile.texi"
@c End: