Chapter 4. Foreign Interface

Chez Scheme provides two ways to interact with "foreign" code, i.e., code written in other languages. The first is via subprocess creation and communication, which is discussed in the Section 4.1. The second is via static or dynamic loading and invocation from Scheme of procedures written in C and invocation from C of procedures written in Scheme. These mechanisms are discussed in Sections 4.2 through 4.4.

The method for static loading of C object code is dependent upon which machine you are running; see the installation instructions distributed with Chez Scheme.

Section 4.1. Subprocess Communication

Two procedures, system and process, are used to create subprocesses. Both procedures accept a single string argument and create a subprocess to execute the shell command contained in the string. The system procedure waits for the process to exit before returning, however, while the process procedure returns immediately without waiting for the process to exit. The standard input and output files of a subprocess created by system may be used to communicate with the user's console. The standard input and output files of a subprocess created by process may be used to communicate with the Scheme process.

procedure: (system command)
returns: see below
libraries: (chezscheme)

command must be a string.

The system procedure creates a subprocess to perform the operation specified by command. The subprocess may communicate with the user through the same console input and console output files used by the Scheme process. After creating the subprocess, system waits for the process to exit before returning.

When the subprocess exits, system returns the exit code for the subprocess, unless (on Unix-based systems) a signal caused the subprocess to terminate, in which case system returns the negation of the signal that caused the termination, e.g., -1 for SIGHUP.

procedure: (open-process-ports command)
procedure: (open-process-ports command b-mode)
procedure: (open-process-ports command b-mode ?transcoder)
returns: see below
libraries: (chezscheme)

command must be a string. If ?transcoder is present and not #f, it must be a transcoder, and this procedure creates textual ports, each of whose transcoder is ?transcoder. Otherwise, this procedure returns binary ports. b-mode specifies the buffer mode used by each of the ports returned by this procedure and defaults to block. Buffer modes are described in Section 7.2 of The Scheme Programming Language, 4th Edition.

open-process-ports creates a subprocess to perform the operation specified by command. Unlike system, process returns immediately after creating the subprocess, i.e., without waiting for the subprocess to terminate. It returns four values:

  1. to-stdin is an output port to which Scheme can send output to the subprocess through the subprocess's standard input file.

  2. from-stdout is an input port from which Scheme can read input from the subprocess through the subprocess's standard output file.

  3. from-stderr is an input port from which Scheme can read input from the subprocess through the subprocess's standard error file.

  4. process-id is an integer identifying the created subprocess provided by the host operating system.

If the process exits or closes its standard output file descriptor, any procedure that reads input from from-stdout will return an end-of-file object. Similarly, if the process exits or closes its standard error file descriptor, any procedure that reads input from from-stderr will return an end-of-file object.

The predicate input-port-ready? may be used to detect whether input has been sent by the subprocess to Scheme.

It is sometimes necessary to force output to be sent immediately to the subprocess by invoking flush-output-port on to-stdin, since Chez Scheme buffers the output for efficiency.

On UNIX systems, the process-id is the process identifier for the shell created to execute command. If command is used to invoke an executable file rather than a shell command, it may be useful to prepend command with the string "exec ", which causes the shell to load and execute the named executable directly, without forking a new process---the shell equivalent of a tail call. This will reduce by one the number of subprocesses created and cause process-id to reflect the process identifier for the executable once the shell has transferred control.

procedure: (process command)
returns: see explanation
libraries: (chezscheme)

command must be a string.

process is similar to open-process-ports, but less general. It does not return a port from which the subproces's standard error output can be read, and it always creates textual ports. It returns a list of three values rather than the four separate values of open-process-ports. The returned list contains, in order: from-stdout, to-stdin, and process-id, which correspond to the second, first, and fourth return values of open-process-ports.

Section 4.2. Calling out of Scheme

Chez Scheme's foreign-procedure interface allows a Scheme program to invoke procedures written in C or in languages that obey the same calling conventions as C. Two steps are necessary before foreign procedures can be invoked from Scheme. First, the foreign procedure must be compiled and loaded, either statically or dynamically, as described in Section 4.6. Then, access to the foreign procedure must be established in Scheme, as described in this section. Once access to a foreign procedure has been established it may be called as an ordinary Scheme procedure.

Since foreign procedures operate independently of the Scheme memory management and exception handling system, great care must be taken when using them. Although the foreign-procedure interface provides type checking (at optimize levels less than 3) and type conversion, the programmer must ensure that the sharing of data between Scheme and foreign procedures is done safely by specifying proper argument and result types.

Scheme-callable wrappers for foreign procedures can also be created via ftype-ref and function ftypes (Section 4.5).

syntax: (foreign-procedure entry-exp (param-type ...) res-type)
syntax: (foreign-procedure conv entry-exp (param-type ...) res-type)
returns: a procedure
libraries: (chezscheme)

entry-exp must evaluate to a string representing a valid foreign procedure entry point or an integer representing the address of the foreign procedure. The param-types and res-type must be symbols or structured forms as described below. When a foreign-procedure expression is evaluated, a Scheme procedure is created that will invoke the foreign procedure specified by entry-exp. When the procedure is called each argument is checked and converted according to the specified param-type before it is passed to the foreign procedure. The result of the foreign procedure call is converted as specified by the res-type. Multiple procedures may be created for the same foreign entry.

If conv is present, it specifies the calling convention to be used. The default is #f, which specifies the default calling convention on the target machine. Three other conventions are currently supported, all only under Windows: __stdcall, __cdecl, and __com. Since __cdecl is the default, specifying __cdecl is equivalent to specifying #f or no convention.

Use __stdcall to access most Windows API procedures. Use __cdecl for Windows API varargs procedures, for C library procedures, and for most other procedures. Use __com to invoke COM interface methods; COM uses the __stdcall convention but additionally performs the indirections necessary to obtain the correct method from a COM instance. The address of the COM instance must be passed as the first argument, which should normally be declared as iptr. For the __com interface only, entry-exp must evaluate to the byte offset of the method in the COM vtable. For example,

(foreign-procedure __com 12 (iptr double-float) integer-32)

creates an interface to a COM method at offset 12 in the vtable encapsulated within the COM instance passed as the first argument, with the second argument being a double float and the return value being an integer.

Complete type checking and conversion is performed on the parameters. The types scheme-object, string, wstring, u8*, u16*, u32*, utf-8, utf-16le, utf-16be, utf-32le, and utf-32be, must be used with caution, however, since they allow allocated Scheme objects to be used in places the Scheme memory management system cannot control. No problems will arise as long as such objects are not retained in foreign variables or data structures while Scheme code is running, since garbage collection can occur only while Scheme code is running. All other parameter types are converted to equivalent foreign representations and consequently can be retained indefinitely in foreign variables and data structures. Following are the valid parameter types:

integer-8: Exact integers from -27 through 28 - 1 are valid. Integers in the range 27 through 28 - 1 are treated as two's complement representations of negative numbers, e.g., #xff is treated as -1. The argument is passed to C as an integer of the appropriate size (usually signed char).

unsigned-8: Exact integers from -27 to 28 - 1 are valid. Integers in the range -27 through -1 are treated as the positive equivalents of their two's complement representation, e.g., -1 is treated as #xff. The argument is passed to C as an unsigned integer of the appropriate size (usually unsigned char).

integer-16: Exact integers from -215 through 216 - 1 are valid. Integers in the range 215 through 216 - 1 are treated as two's complement representations of negative numbers, e.g., #xffff is treated as -1. The argument is passed to C as an integer of the appropriate size (usually short).

unsigned-16: Exact integers from -215 to 216 - 1 are valid. Integers in the range -215 through -1 are treated as the positive equivalents of their two's complement representation, e.g., -1 is treated as #xffff. The argument is passed to C as an unsigned integer of the appropriate size (usually unsigned short).

integer-32: Exact integers from -231 through 232 - 1 are valid. Integers in the range 231 through 232 - 1 are treated as two's complement representations of negative numbers, e.g., #xffffffff is treated as -1. The argument is passed to C as an integer of the appropriate size (usually int).

unsigned-32: Exact integers from -231 to 232 - 1 are valid. Integers in the range -231 through -1 are treated as the positive equivalents of their two's complement representation, e.g., -1 is treated as #xffffffff. The argument is passed to C as an unsigned integer of the appropriate size (usually unsigned int).

integer-64: Exact integers from -263 through 264 - 1 are valid. Integers in the range 263 through 264 - 1 are treated as two's complement representations of negative numbers. The argument is passed to C as an integer of the appropriate size (usually long long or, on many 64-bit platforms, long).

unsigned-64: Exact integers from -263 through 264 - 1 are valid. Integers in the range -263 through -1 are treated as the positive equivalents of their two's complement representation, The argument is passed to C as an integer of the appropriate size (usually unsigned long long or, on many 64-bit platforms, long).

double-float: Only Scheme flonums are valid---other Scheme numeric types are not automatically converted. The argument is passed to C as a double float.

single-float: Only Scheme flonums are valid---other Scheme numeric types are not automatically converted. The argument is passed to C as a single float. Since Chez Scheme represents flonums in double-float format, the parameter is first converted into single-float format.

short: This type is an alias for the appropriate fixed-size type above, depending on the size of a C short.

unsigned-short: This type is an alias for the appropriate fixed-size type above, depending on the size of a C unsigned short.

int: This type is an alias for the appropriate fixed-size type above, depending on the size of a C int.

unsigned: This type is an alias for the appropriate fixed-size type above, depending on the size of a C unsigned.

unsigned-int: This type is an alias unsigned. fixed-size type above, depending on the size of a C unsigned.

long: This type is an alias for the appropriate fixed-size type above, depending on the size of a C long.

unsigned-long: This type is an alias for the appropriate fixed-size type above, depending on the size of a C unsigned long.

long-long: This type is an alias for the appropriate fixed-size type above, depending on the size of the nonstandard C type long long.

unsigned-long-long: This type is an alias for the appropriate fixed-size type above, depending on the size of the nonstandard C type unsigned long long.

ptrdiff_t: This type is an alias for the appropriate fixed-size type above, depending on its definition in the host machine's stddef.h include file.

size_t: This type is an alias for the appropriate unsigned fixed-size type above, depending on its definition in the host machine's stddef.h include file.

ssize_t: This type is an alias for the appropriate signed fixed-size type above, depending on its definition in the host machine's stddef.h include file.

iptr: This type is an alias for the appropriate fixed-size type above, depending on the size of a C pointer.

uptr: This type is an alias for the appropriate (unsigned) fixed-size type above, depending on the size of a C pointer.

void*: This type is an alias for uptr.

fixnum: This type is equivalent to iptr, except only values in the fixnum range are valid. Transmission of fixnums is slightly faster than transmission of iptr values, but the fixnum range is smaller, so some iptr values do not have a fixnum representation.

boolean: Any Scheme object may be passed as a boolean. #f is converted to 0; all other objects are converted to 1. The argument is passed to C as an int.

char: Only Scheme characters with Unicode scalar values in the range 0 through 255 are valid char parameters. The character is converted to its Unicode scalar value, as with char->integer, and passed to C as an unsigned char.

wchar_t: Only Scheme characters are valid wchar_t parameters. Under Windows and any other system where wchar_t holds only 16-bit values rather than full Unicode scalar values, only characters with 16-bit Unicode scalar values are valid. On systems where wchar_t is a full 32-bit value, any Scheme character is valid. The character is converted to its Unicode scalar value, as with char->integer, and passed to C as a wchar_t.

wchar: This type is an alias for wchar_t.

double: This type is an alias for double-float.

float: This type is an alias for single-float.

scheme-object: The argument is passed directly to the foreign procedure; no conversion or type checking is performed. This form of parameter passing should be used with discretion. Scheme objects should not be preserved in foreign variables or data structures since the memory management system may relocate them between foreign procedure calls.

ptr: This type is an alias for scheme-object.

u8*: The argument must be a Scheme bytevector or #f. For #f, the null pointer (0) is passed to the foreign procedure. For a bytevector, a pointer to the first byte of the bytevector's data is passed. If the C routine to which the data is passed requires the input to be null-terminated, a null (0) byte must be included explicitly in the bytevector. The bytevector should not be retained in foreign variables or data structures, since the memory management system may relocate or discard them between foreign procedure calls, and use their storage for some other purpose.

u16*: Arguments of this type are treated just like arguments of type u8*. If the C routine to which the data is passed requires the input to be null-terminated, two null (0) bytes must be included explicitly in the bytevector, aligned on a 16-bit boundary.

u32*: Arguments of this type are treated just like arguments of type u8*. If the C routine to which the data is passed requires the input to be null-terminated, four null (0) bytes must be included explicitly in the bytevector, aligned on a 32-bit boundary.

utf-8: The argument must be a Scheme string or #f. For #f, the null pointer (0) is passed to the foreign procedure. A string is converted into a bytevector, as if via string->utf8, with an added null byte, and the address of the first byte of the bytevector is passed to C. The bytevector should not be retained in foreign variables or data structures, since the memory management system may relocate or discard them between foreign procedure calls, and use their storage for some other purpose.

utf-16le: Arguments of this type are treated like arguments of type utf-8, except they are converted as if via string->utf16 with endianness little, and they are extended by two null bytes rather than one.

utf-16be: Arguments of this type are treated like arguments of type utf-8, except they are converted as if via string->utf16 with endianness big, and they are extended by two null bytes rather than one.

utf-32le: Arguments of this type are treated like arguments of type utf-8, except they are converted as if via string->utf32 with endianness little, and they are extended by four null bytes rather than one.

utf-32be: Arguments of this type are treated like arguments of type utf-8, except they are converted as if via string->utf32 with endianness big, and they are extended by four null bytes rather than one.

string: This type is an alias for utf-8.

wstring: This type is an alias for utf-16le, utf-16be, utf-32le, or utf-32be as appropriate depending on the size of a C wchar_t and the endianness of the target machine. For example, wstring is equivalent to utf-16le under Windows running on Intel hardware.

(* ftype): This type allows a pointer to a foreign type (ftype) to be passed. The argument must be an ftype pointer of with type ftype, and the actual argument is the address encapsulated in the ftype pointer. See Section 4.5 for a description of foreign types.

The result types are similar to the parameter types with the addition of a void type. In general, the type conversions are the inverse of the parameter type conversions. No error checking is performed on return, since the system cannot determine whether a foreign result is actually of the indicated type. Particular caution should be exercised with the result types scheme-object, double-float, double, single-float, float, and the types that result in the construction of bytevectors or strings, since invalid return values may lead to invalid memory references as well as incorrect computations. Following are the valid result types:

void: The result of the foreign procedure call is ignored and an unspecified Scheme object is returned. void should be used when foreign procedures are called for effect only.

integer-8: The result is interpreted as a signed 8-bit integer and is converted to a Scheme exact integer.

unsigned-8: The result is interpreted as an unsigned 8-bit integer and is converted to a Scheme nonnegative exact integer.

integer-16: The result is interpreted as a signed 16-bit integer and is converted to a Scheme exact integer.

unsigned-16: The result is interpreted as an unsigned 16-bit integer and is converted to a Scheme nonnegative exact integer.

integer-32: The result is interpreted as a signed 32-bit integer and is converted to a Scheme exact integer.

unsigned-32: The result is interpreted as an unsigned 32-bit integer and is converted to a Scheme nonnegative exact integer.

integer-64: The result is interpreted as a signed 64-bit integer and is converted to a Scheme exact integer.

unsigned-64: The result is interpreted as an unsigned 64-bit integer and is converted to a Scheme nonnegative exact integer.

double-float: The result is interpreted as a double float and is translated into a Chez Scheme flonum.

single-float: The result is interpreted as a single float and is translated into a Chez Scheme flonum. Since Chez Scheme represents flonums in double-float format, the result is first converted into double-float format.

short: This type is an alias for the appropriate fixed-size type above, depending on the size of a C short.

unsigned-short: This type is an alias for the appropriate fixed-size type above, depending on the size of a C unsigned short.

int: This type is an alias for the appropriate fixed-size type above, depending on the size of a C int.

unsigned: This type is an alias for the appropriate fixed-size type above, depending on the size of a C unsigned.

unsigned-int: This type is an alias unsigned. fixed-size type above, depending on the size of a C unsigned.

long: This type is an alias for the appropriate fixed-size type above, depending on the size of a C long.

unsigned-long: This type is an alias for the appropriate fixed-size type above, depending on the size of a C unsigned long.

long-long: This type is an alias for the appropriate fixed-size type above, depending on the size of the nonstandard C type long long.

unsigned-long-long: This type is an alias for the appropriate fixed-size type above, depending on the size of the nonstandard C type unsigned long long.

ptrdiff_t: This type is an alias for the appropriate fixed-size type above, depending on its definition in the host machine's stddef.h include file.

size_t: This type is an alias for the appropriate unsigned fixed-size type above, depending on its definition in the host machine's stddef.h include file.

ssize_t: This type is an alias for the appropriate signed fixed-size type above, depending on its definition in the host machine's stddef.h include file.

iptr: This type is an alias for the appropriate fixed-size type above, depending on the size of a C pointer.

uptr: This type is an alias for the appropriate (unsigned) fixed-size type above, depending on the size of a C pointer.

void*: This type is an alias for uptr.

boolean: This type converts a C int return value into a Scheme boolean. 0 is converted to #f; all other values are converted to #t.

char: This type converts a C unsigned char return value into a Scheme character, as if via integer->char.

wchar_t: This type converts a C wchar_t return value into a Scheme character, as if via integer->char. The wchar_t value must be a valid Unicode scalar value.

wchar: This type is an alias for wchar_t.

double: This type is an alias for double-float.

float: This type is an alias for single-float.

scheme-object: The result is assumed to be a valid Scheme object, and no conversion is performed. This type is inherently dangerous, since an invalid Scheme object can corrupt the memory management system with unpredictable (but always unpleasant) results. Since Scheme objects are actually typed pointers, even integers cannot safely be returned as type scheme-object unless they were created by the Scheme system.

ptr: This type is an alias for scheme-object.

u8*: The result is interpreted as a pointer to a null-terminated sequence of 8-bit unsigned integers (bytes). If the result is a null pointer, #f is returned. Otherwise, the sequence of bytes is stored in a freshly allocated bytevector of the appropriate length, and the bytevector is returned to Scheme.

u16*: The result is interpreted as a pointer to a null-terminated sequence of 16-bit unsigned integers. If the result is a null pointer, #f is returned. Otherwise, the sequence of 16-bit integers is stored in a freshly allocated bytevector of the appropriate length, and the bytevector is returned to Scheme. The null terminator must be a properly aligned 16-bit word, i.e., two bytes of zero aligned on a 16-bit boundary.

u32*: The result is interpreted as a pointer to a null-terminated sequence of 32-bit unsigned integers. If the result is a null pointer, #f is returned. Otherwise, the sequence of 16-bit integers is stored in a freshly allocated bytevector of the appropriate length, and the bytevector is returned to Scheme. The null terminator must be a properly aligned 32-bit word, i.e., four bytes of zero aligned on a 32-bit boundary.

utf-8: The result is interpreted as a pointer to a null-terminated sequence of 8-bit unsigned character values. If the result is a null pointer, #f is returned. Otherwise, the sequence of bytes is converted into a Scheme string, as if via utf8->string, and the string is returned to Scheme.

utf-16le: The result is interpreted as a pointer to a null-terminated sequence of 16-bit unsigned integers. If the result is a null pointer, #f is returned. Otherwise, the sequence of integers is converted into a Scheme string, as if via utf16->string with endianness little, and the string is returned to Scheme. A byte-order mark in the sequence of integers as treated as an ordinary character value and does not affect the byte ordering.

utf-16be: The result is interpreted as a pointer to a null-terminated sequence of 16-bit unsigned integers. If the result is a null pointer, #f is returned. Otherwise, the sequence of integers is converted into a Scheme string, as if via utf16->string with endianness big, and the string is returned to Scheme. A byte-order mark in the sequence of integers as treated as an ordinary character value and does not affect the byte ordering.

utf-32le: The result is interpreted as a pointer to a null-terminated sequence of 32-bit unsigned integers. If the result is a null pointer, #f is returned. Otherwise, the sequence of integers is converted into a Scheme string, as if via utf32->string with endianness little, and the string is returned to Scheme. A byte-order mark in the sequence of integers as treated as an ordinary character value and does not affect the byte ordering.

utf-32be: The result is interpreted as a pointer to a null-terminated sequence of 32-bit unsigned integers. If the result is a null pointer, #f is returned. Otherwise, the sequence of integers is converted into a Scheme string, as if via utf32->string with endianness big, and the string is returned to Scheme. A byte-order mark in the sequence of integers as treated as an ordinary character value and does not affect the byte ordering.

string: This type is an alias for utf-8.

wstring: This type is an alias for utf-16le, utf-16be, utf-32le, or utf-32be as appropriate depending on the size of a C wchar_t and the endianness of the target machine. For example, wstring is equivalent to utf-16le under Windows running on Intel hardware.

(* ftype): The result is interpreted as the address of a foreign object whose structure is described by ftype, and a freshly allocated ftype pointer encapsulating the address is returned. See Section 4.5 for a description of foreign types.

Consider a C identity procedure:

int id(x) int x; { return x; }

After a file containing this procedure has been compiled and loaded (see Section 4.6) it can be accessed as follows:

(foreign-procedure "id"
  (int) int) <graphic> #<procedure>
((foreign-procedure "id"
   (int) int)
 1) <graphic> 1
(define int-id
  (foreign-procedure "id"
    (int) int))
(int-id 1) <graphic> 1

The "id" entry can also be interpreted as accepting and returning a boolean:

(define bool-id
  (foreign-procedure "id"
    (boolean) boolean))
(bool-id #f) <graphic> #f
(bool-id #t) <graphic> #t
(bool-id 1) <graphic> #t

As the last example reveals, bool-id is actually a conversion procedure. When a Scheme object is passed as type boolean it is converted to 0 or 1, and when it is returned it is converted to #f or #t. As a result objects are converted to normalized boolean values. The "id" entry can be used to create other conversion procedures by varying the type specifications:

(define int->bool
  (foreign-procedure "id"
    (int) boolean))
(int->bool 0) <graphic> #f
(int->bool 5) <graphic> #t
(map (foreign-procedure "id"
       (boolean) int)
     '(#t #f)) <graphic> (1 0)
(define void
  (foreign-procedure "id"
    (int) void))
(void 10) <graphic> unspecified

There are, of course, simpler and more efficient ways of accomplishing these conversions directly in Scheme.

A foreign entry is resolved when a foreign-procedure expression is evaluated, rather than either when the code is loaded or each time the procedure is invoked. Thus, the following definition is always valid since the foreign-procedure expression is not immediately evaluated:

(define doit
  (lambda ()
    ((foreign-procedure "doit" () void))))

doit should not be invoked, however, before an entry for "doit" has been provided. Similarly, an entry for "doit" must exist before the following code is evaluated:

(define doit
  (foreign-procedure "doit" () void))

Although the second definition is more constraining on the load order of foreign files, it is more efficient since the entry resolution need be done only once.

It is often useful to define a template to be used in the creation of several foreign procedures with similar argument types and return values. For example, the following code creates two foreign procedures from a single foreign procedure expression, by abstracting out the foreign procedure name:

(define double->double
  (lambda (proc-name)
    (foreign-procedure proc-name
      (double)
      double)))

(define log10 (double->double "log10"))
(define gamma (double->double "gamma"))

Both "log10" and "gamma" must be available as foreign entries (see Section 4.6) before the corresponding definitions. The use of foreign procedure templates can simplify the coding process and reduce the amount of code generated when a large number of foreign procedures are involved, e.g., when an entire library of foreign procedures is imported into Scheme.

Section 4.3. Calling into Scheme

Section 4.2 describes the foreign-procedure form, which permits Scheme code to invoke C or C-compatible foreign procedures. This section describes the foreign-callable form, which permits C or C-compatible code to call Scheme procedures. A more primitive mechanism for calling Scheme procedures from C is described in Section 4.8.

As when calling foreign procedures from Scheme, great care must be taken when sharing data between Scheme and foreign code that calls Scheme to avoid corrupting Scheme's memory management system.

A foreign-callable wrapper for a Scheme procedure can also be created by passing the procedure to make-ftype-pointer with an appropriate function ftype (Section 4.5).

syntax: (foreign-callable proc-exp (param-type ...) res-type)
syntax: (foreign-callable conv proc-exp (param-type ...) res-type)
returns: a code object
libraries: (chezscheme)

proc-exp must evaluate to a procedure, the Scheme procedure that is to be invoked by foreign code. The parameter and result types are as described for foreign-procedure in Section 4.2, except that the requirements and conversions are effectively reversed, e.g., the conversions described for foreign-procedure arguments are performed for foreign-callable return values. Type checking is performed for result values but not argument values, since the parameter values are provided by the foreign code and must be assumed to be correct.

If conv is present, it specifies the calling convention to be used. foreign-callable supports the same conventions as foreign-procedure with the exception of __com.

The value produced by foreign-callable is a Scheme code object, which contains some header information as well as code that performs the call to the encapsulated Scheme procedure. The code object may be converted into a foreign-callable address via foreign-callable-entry-point, which returns an integer representing the address of the entry point within the code object. (The C-callable library function Sforeign_callable_entry_point, described in Section 4.8, may be used to obtain the entry point as well.) This is an implicit pointer into a Scheme object, and in many cases, it is necessary to lock the code object (using lock-object) before converting it into an entry point to prevent Scheme's storage management system from relocating or destroying the code object, e.g., when the entry point is registered as a callback and retained in the "C" side indefinitely.

The following code creates a foreign-callable code object, locks the code object, and returns the entry point.

(let ([x (foreign-callable
           (lambda (x y) (pretty-print (cons x (* y 2))))
           (string integer-32)
           void)])
  (lock-object x)
  (foreign-callable-entry-point x))

Unless the entry point is intended to be permanent, a pointer to the code object returned by foreign-callable should be retained so that it can be unlocked when no longer needed.

Mixed use of foreign-callable and foreign-procedure may result in nesting of foreign and Scheme calls, and this results in some interesting considerations when continuations are involved, directly or indirectly (as via the default exception handler). See Section 4.4 for a discussion of the interaction between foreign calls and continuations.

The following example demonstrates how the "callback" functions required by many windowing systems might be defined in Scheme with the use of foreign-callable. Assume that the following C code has been compiled and loaded (see Section 4.6).

#include <stdio.h>

typedef void (*CB)(char);

CB callbacks[256];

void cb_init(void) {
   int i;

   for (i = 0; i < 256; i += 1)
       callbacks[i] = (CB)0;
}

void register_callback(char c, int cb) {
    callbacks[c] = (CB)cb;
}

void event_loop(void) {
    CB f; char c;

    for (;;) {
        c = getchar();
        if (c == EOF) break;
        f = callbacks[c];
        if (f != (CB)0) f(c);
    }
}

Interfaces to these functions may be defined in Scheme as follows.

(define cb-init
  (foreign-procedure "cb_init" () void))
(define register-callback
  (foreign-procedure "register_callback" (char int) void))
(define event-loop
  (foreign-procedure "event_loop" () void))

A callback for selected characters can then be defined.

(define callback
  (lambda (p)
    (let ([code (foreign-callable p (char) void)])
      (lock-object code)
      (foreign-callable-entry-point code))))
(define ouch
  (callback
    (lambda (c)
      (printf "Ouch! Hit by '~c'~%" c))))
(define rats
  (callback
    (lambda (c)
      (printf "Rats! Received '~c'~%" c))))

(cb-init)
(register-callback #\a ouch)
(register-callback #\c rats)
(register-callback #\e ouch)

This sets up the following interaction.

> (event-loop)
a
Ouch! Hit by 'a'
b
c
Rats! Received 'c'
d
e
Ouch! Hit by 'e'

A more well-behaved version of this example would save each code object returned by foreign-callable and unlock it when it is no longer registered as a callback.

procedure: (foreign-callable-entry-point code)
returns: the address of the foreign-callable entry point in code
libraries: (chezscheme)

code should be a code object produced by foreign-callable.

procedure: (foreign-callable-code-object address)
returns: the code object corresponding to the foreign-callable entry point address
libraries: (chezscheme)

address must be an exact integer and should be the address of the entry point of a code object produced by foreign-callable.

Section 4.4. Continuations and Foreign Calls

foreign-callable and foreign-procedure allow arbitrary nesting of foreign and Scheme calls. Because other languages do not support the fully general first-class continuations of Scheme, the interaction between continuations and nested calls among Scheme and foreign procedures is problematic. Chez Scheme handles this interaction in a general manner by trapping attempts to return to stale foreign contexts rather than by restricting the use of continuations directly. A foreign context is a foreign frame and return point corresponding to a particular call from a foreign language, e.g., C, into Scheme. A foreign context becomes stale after a normal return to the context or after a return to some other foreign context beneath it on the control stack.

As a result of this treatment, Scheme continuations may be used to throw control either upwards or downwards logically through any mix of Scheme and foreign frames. Furthermore, until some return to a foreign context is actually performed, all return points remain valid. In particular, this means that programs that use continuations exclusively for nonlocal exits never attempt to return to a stale foreign context. (Nonlocal exits themselves are no problem and are implemented by the C library function longjmp or the equivalent.) Programs that use continuations more generally also function properly as long as they never actually return to a stale foreign context, even if control logically moves past stale foreign contexts via invocation of continuations.

One implication of this mechanism is that the C stack pointer is not automatically restored to its base value when a continuation is used on the Scheme side to perform a nonlocal exit. If the program continues to run after the nonlocal exit, any further build-up of the C stack will add to the existing build up, which might result in a C stack overflow. To avoid this situation, a program can arrange to set up a single C call frame before obtaining the continuation and return to the C frame after the nonlocal exit. The procedure with-exit-proc below arranges to do this without involving any C code.

(define with-exit-proc
  (lambda (p)
    (define th (lambda () (call/cc p)))
    (define-ftype ->ptr (function () ptr))
    (let ([fptr (make-ftype-pointer ->ptr th)])
      (let ([v ((ftype-ref ->ptr () fptr))])
        (unlock-object
          (foreign-callable-code-object
            (ftype-pointer-address fptr)))
        v))))

with-exit-proc behaves like call/cc except it resets the C stack when the continuation is invoked. To do this, it creates an ftype-pointer representing a foreign-callable entry point for th and creates a Scheme-callable procedure for that entry point. This creates a wrapper for th that involves a C call. When a call to the wrapper returns, either by explicit invocation of the continuation passed to p or by a normal return from p, the C stack is reset to its original value.

Section 4.5. Foreign Data

The procedures described in this section directly create and manipulate foreign data, i.e., data that resides outside of the Scheme heap. With the exception of foreign-alloc and foreign-sizeof, these procedures are inherently unsafe in the sense that they do not (and cannot) check the validity of the addresses they are passed. Improper use of these procedures can result in invalid memory references, corrupted data, or system crashes.

This section also describes a higher-level syntactic mechanism for manipulating foreign data, including foreign structures, unions, arrays, and bit fields. The syntactic interface is safer than the procedural interface but must still assume that the addresses it's given are appropriate for the types of object being manipulated.

procedure: (foreign-alloc n)
returns: the address of a freshly allocated block of foreign data n bytes long
libraries: (chezscheme)

n must be a positive fixnum. The returned value is an exact integer and is guaranteed to be properly aligned for any type of value according to the requirements of the underlying hardware. An exception is raised with condition type &assertion if the block of foreign data cannot be allocated.

procedure: (foreign-free address)
returns: unspecified
libraries: (chezscheme)

This procedure frees the block of storage to which address points. address must be an exact integer in the range -2w-1 through 2w - 1, where w is the width in bits of a pointer, e.g., 64 for a 64-bit machine. It should be an address returned by an earlier call to foreign-alloc and not subsequently passed to foreign-free.

procedure: (foreign-ref type address offset)
returns: see below
libraries: (chezscheme)

foreign-ref extracts the value of type type offset bytes into the block of foreign data addressed by address.

type must be a symbol identifying the type of value to be extracted. The following types have machine-dependent sizes and correspond to the like-named C types:

The types long-long and unsigned-long-long correspond to the C types long long and unsigned long long. A value of type char is referenced as a single byte and converted (as if via integer->char) into a Scheme character. A value of type wchar_t is converted (as if via integer->char) into a Scheme character. The value must be a valid Unicode scalar value.

wchar is an alias for wchar_t.

Several additional machine-dependent types are recognized:

uptr is equivalent to void*; both are treated as unsigned integers the size of a pointer. iptr is treated as a signed integer the size of a pointer. fixnum is treated as an iptr, but with a range limited to the fixnum range. boolean is treated as an int, with zero converted to the Scheme value #f and all other values converted to #t.

Finally, several fixed-sized types are also supported:

address must be an exact integer in the range -2w-1 through 2w - 1, where w is the width in bits of a pointer, e.g., 64 for a 64-bit machine. offset must be an exact fixnum. The sum of address and offset should address a readable block of memory large enough to hold a value of type type, within a block of storage previously returned by foreign-alloc and not subsequently freed by foreign-free or within a block of storage obtained via some other mechanism, e.g., a foreign call. For multiple-byte values, the native endianness of the machine is assumed.

procedure: (foreign-set! type address offset value)
returns: see below
libraries: (chezscheme)

foreign-set! stores a representation of value as type type offset bytes into the block of foreign data addressed by address.

type must be a symbol identifying the type of value to be stored, one of those listed in the description of foreign-ref above. Scheme characters are converted to type char or wchar_t as if via char->integer. For type boolean, Scheme #f is converted to the int 0, and any other Scheme object is converted to 1.

address must be an exact integer in the range -2w-1 through 2w - 1, where w is the width in bits of a pointer, e.g., 64 for a 64-bit machine. offset must be an exact fixnum. The sum of address and offset should address a writable block of memory large enough to hold a value of type type, within a block of storage previously returned by foreign-alloc and not subsequently freed by foreign-free or within a block of storage obtained via some other mechanism, e.g., a foreign call. value must be an appropriate value for type, e.g., a floating-point number for the float types or an exact integer within the appropriate range for the integer types. For multiple-byte values, the native endianness of the machine is assumed.

procedure: (foreign-sizeof type)
returns: the size in bytes of type
libraries: (chezscheme)

type must be one of the symbols listed in the description of foreign-ref above.

syntax: (define-ftype ftype-name ftype)
syntax: (define-ftype (ftype-name ftype) ...)
returns: unspecified
libraries: (chezscheme)

A define-ftype form is a definition and can appear anywhere other definitions can appear. It establishes one or more foreign-type (ftype) bindings for the identifier ftype-name or identifiers ftype-name ... to the foreign type represented ftype or the foreign types represented by ftype .... Each ftype-name can be used to access foreign objects with the declared shape, and each can be used in the formation of other ftypes.

An ftype must take one of the following forms:

ftype-name
(struct (field-name ftype) ...)
(union (field-name ftype) ...)
(array length ftype)
(* ftype)
(bits (field-name signedness bits) ...)
(function (ftype ...) ftype)
(function conv (ftype ...) ftype)
(packed ftype)
(unpacked ftype)
(endian endianness ftype)

where length is an exact nonnegative integer, bits is an exact positive integer, field-name is an identifier, conv is #f or a string naming a valid convention as described on page 4.2, signedness is either signed or unsigned, and endianness is one of native, big, or little.

A restriction not reflected above is that function ftypes cannot be used as the types of field names or array elements. That is, function ftypes are valid only at the top level of an ftype, e.g,:

(define-ftype bvcopy_t (function (u8* u8* size_t) void))

or as the immediate sub-type of a pointer (*) ftype, as in the following definitions, which are equivalent assuming the definition of bvcopy_t above.

(define-ftype A
  (struct
    [x int]
    [f (* (function (u8* u8* size_t) void))]))

(define-ftype A
  (struct
    [x int]
    [f (* bvcopy_t)]))

That is, a function cannot be embedded within a struct, union, or array, but a pointer to a function can be so embedded.

The following definitions establish ftype bindings for F, A, and E.

(define-ftype F (function (wchar_t int) int))

(define-ftype A (array 10 wchar_t))

(define-ftype E
  (struct
    [a int]
    [b double]
    [c (array 25
         (struct
           [a short]
           [_ long]
           [b A]))]
    [d (endian big
         (union
           [v1 unsigned-32]
           [v2 (bits
                 [hi unsigned 12]
                 [lo unsigned 20])]))]
    [e (* A)]
    [f (* F)]))

The ftype F describes the type of a foreign function that takes two arguments, a wide character and an integer, and returns an integer. The ftype A is simply an array of 10 wchar_t values, and its size will be 10 times the size of a single wchar_t. The ftype E is a structure with five fields: an integer a, a double-float b, an array c, a union d, and a pointer e. The array c is an array of 25 structs, each of which contains a short integer, a long integer, and a A array. The size of the c array will be 25 times the size of a single A array, plus 25 times the space needed to store each of the short and long integers. The union d is either a 32-bit unsigned integer or a 32-bit unsigned integer split into high (12 bits) and low (20 bits) components. The fields of a union overlap so that writing to one effectively overlaps the other. Thus, one can use the d union type to split apart an unsigned integer by writing the integer into v1 and reading the pieces from hi and lo. The pointer e points to an A array; it is not itself an array, and its size is just the size of a single pointer. Similarly, f points to a function, and its size is also that of a single pointer.

An underscore ( _ ) can be used as the field name for one or more fields of a struct, union, or bits ftype. Such fields are included in the layout but are considered unnamed and cannot be accessed via the ftype operators described below. Thus, in the example above, the long field within the c array is inaccessible.

Non-underscore field names are handled symbolically, i.e., they are treated as symbols rather than identifiers. Each symbol must be unique (as a symbol) with respect to the other field names within a single struct, union, or bits ftype but need not be unique with respect to field names in other struct, union, or bits ftypes within the same ftype.

Each ftype-name in an ftype must either (a) have been defined previously by define-ftype, (b) be defined by the current define-ftype, or (c) be a base-type name, i.e., one of the type names supported by foreign-ref and foreign-set!. In case (b), any reference within one ftype to the ftype-name of one of the earlier bindings is permissible, but a reference to the ftype-name of the current or a subsequent binding can appear only within a pointer field.

For example, in:

(define-ftype
  [Qlist (struct
           [head int]
           [tail (* Qlist)])])

the reference to Qlist is permissible since it appears within a pointer field. Similarly, in:

(define-ftype
  [Qfrob (struct
           [head int]
           [tail (* Qsnark)])]
  [Qsnark (struct
            [head int]
            [xtra Qfrob]
            [tail (* Qfrob)])])

the mutually recursive references to Qsnark and Qfrob are permissible. In the following, however:

(define-ftype
  [Qfrob (struct
           [head int]
           [xtra Qfrob]
           [tail (* Qsnark)])]
  [Qsnark (struct
            [head int]
            [tail (* Qfrob)])])

the reference to Qfrob within the ftype for Qfrob is invalid, and in:

(define-ftype
  [Qfrob (struct
           [head int]
           [xtra Qsnark]
           [tail (* Qsnark)])]
  [Qsnark (struct
            [head int]
            [tail (* Qfrob)])])

the reference to Qsnark is similarly invalid.

By default, padding is inserted where appropriate to maintain proper alignment of multiple-byte scalar values in an attempt to mirror the target machine's C struct layout conventions, where such layouts are adequately documented. For packed ftypes (ftypes wrapped in a packed form with no closer enclosing unpacked form), this padding is not inserted.

Multiple-byte scalar values are stored in memory using the target machine's native "endianness," e.g., little on X86 and X86_64-based platforms and big on Sparc-based platforms. Big-endian or little-endian representation can be forced via the endian ftype with a big or little endianness specifier. The native specifier can be used to force a return back to native representation. Each endian form affects only ftypes nested syntactically within it and not nested within a closer endian form.

The total size n of the fields within an ftype bits form must be 8, 16, 24, 32, 40, 48, 56, or 64. padding must be added manually if needed. In little-endian representation, the first field occupies the low-order bits of the containing 8, 16, 24, 32, 40, 48, 56, or 64-bit word, with each subsequent field just above the preceding field. In big-endian representation, the first field occupies the high-order bits, with each subsequent field just below the preceding field.

Two ftypes are considered equivalent only if defined by the same ftype binding. If two ftype definitions look identical but appear in two parts of the same program, the ftypes are not identical, and attempts to access one using the name of the other via the operators described below will fail with a run-time exception.

Array bounds must always be constant. If an array's length cannot be known until run time, the array can be placed at the end of the ftype (and any containing ftype) and declared to have size zero, as illustrated by the example below.

(define-ftype Vec
  (struct
    [len int]
    [data (array 0 double)]))
(define make-Vec
  (lambda (n)
    (let ([fptr (make-ftype-pointer Vec
                  (foreign-alloc
                    (+ (ftype-sizeof Vec)
                       (* (ftype-sizeof double) n))))])
      (ftype-set! Vec (len) fptr n)
      fptr)))
(define x (make-Vec 100))
(/ (- (ftype-pointer-address (ftype-&ref Vec (data 10) x))
      (ftype-pointer-address x)                            <graphic> 10
      (ftype-sizeof int))
   (ftype-sizeof double))
(foreign-free (ftype-pointer-address x))

No array bounds checks are performed for zero-length arrays. Only one variable-sized array can appear in a single foreign object, but one can work around this by treating the object as multiple individual objects.

To avoid specifying the constant length of an array in more than one place, a macro that binds both a variable to the size as well as an ftype name to the ftype can be used. For example,

(define-syntax define-array
  (syntax-rules ()
    [(_ array-name type size-name size)
     (begin
       (define size-name size)
       (define-ftype array-name
         (array size type)))]))
(define-array A int A-size 100)
A-size <graphic> 100
(ftype-pointer-ftype
  (make-ftype-pointer A
    (foreign-alloc (ftype-sizeof A)))) <graphic> (array 100 int)

This technique can be used to define arbitrary ftypes with arbitrary numbers of array fields.

A struct ftype is an implicit subtype of the type of the first field of the struct. Similarly, an array ftype is an implicit subtype of the type of its elements. Thus, the struct or array extends the type of first field or element with additional fields or elements. This allows an instance of the struct or array to be treated as an instance of the type of its first field or element, without the need to use ftype-&ref to allocate a new pointer to the field or element.

syntax: (ftype-sizeof ftype-name)
returns: the size in bytes of the ftype identified by ftype-name
libraries: (chezscheme)

The size includes the sizes of any ftypes directly embedded within the identified ftype but excludes those indirectly embedded via a pointer ftype. In the latter case, the size of the pointer is included.

ftype-name must not be defined as a function ftype, since the size of a function cannot generally be determined.

(define-ftype B
  (struct
    [b1 integer-32]
    [b2 (array 10 integer-32)]))
(ftype-sizeof B) <graphic> 44

(define-ftype C (* B))
(ftype-sizeof C) <graphic> 4  ; on 32-bit machines
(ftype-sizeof C) <graphic> 8  ; on 64-bit machines

(define-ftype BB
  (struct
    [bb1 B]
    [bb2 (* B)]))
(- (ftype-sizeof BB) (ftype-sizeof void*)) <graphic> 44

syntax: (make-ftype-pointer ftype-name expr)
returns: an ftype-pointer object
libraries: (chezscheme)

If ftype-name does not describe a function ftype, expr must evaluate to an address represented as an exact integer in the appropriate range for the target machine.

The ftype-pointer object returned by this procedure encapsulates the address and is tagged with a representation of the type identified by ftype-name to enable various forms of checking to be done by the access routines described below.

(make-ftype-pointer E #x80000000) <graphic> #<ftype-pointer #x80000000>

The address will not typically be a constant, as shown. Instead, it might instead come from a call to foreign-alloc, e.g.:

(make-ftype-pointer E (foreign-alloc (ftype-sizeof E)))

It might also come from source outside of Scheme such as from a C routine called from Scheme via the foreign-procedure interface.

If ftype-name describes a function ftype, expr must evaluate to an address, procedure, or string. If it evaluates to address, the call behaves like any other call to make-ftype-pointer with an address argument.

If it evaluates to a procedure, a foreign-callable code object is created for the procedure, as if via foreign-callable (Section 4.3). The address encapsulated in the resulting ftype-pointer object is the address of the procedure's entry point.

(define fact
  (lambda (n)
    (if (= n 0) 1 (fact (- n 1)))))
(define-ftype fact_t (function (int) int))
(define fact-fptr (make-ftype-pointer fact_t fact))

The resulting ftype pointer can be passed to a C routine, if the argument is declared to be a pointer to the same ftype, and the C routine can invoke the function pointer it receives as it would any other function pointer. Thus, make-ftype-pointer with a function ftype is an alternative to foreign-callable for creating C-callable wrappers for Scheme procedures.

Since all Scheme objects, including code objects, can be relocated or even reclaimed by the garbage collector the foreign-callable code object is automatically locked, as if via lock-object, before it is embedded in the ftype pointer. The code object should be unlocked after its last use from C, since locked objects take up space, cause fragmentation, and increase the cost of collection. Since the system cannot determine automatically when the last use from C occurs, the program must explicitly unlock the code object, which it can do by extracting the address from the ftype-pointer converting the address (back) into a code object, and passing it to unlock-object:

(unlock-object
  (foreign-callable-code-object
    (ftype-pointer-address fact-fptr)))

Once unlocked, the ftype pointer should not be used again, unless it is relocked, e.g., via:

(lock-object
  (foreign-callable-code-object
    (ftype-pointer-address fact-fptr)))

A program can determine whether an object is already locked via the locked-object? predicate.

A function ftype can be also used with make-ftype-pointer to create an ftype-pointer to a C function, either by providing the address of the C function or its name, represented as a string. For example, with the following definition of bvcopy_t,

(define-ftype bvcopy_t (function (u8* u8* size_t) void))

the two definitions of bvcopy-ftpr below are equivalent.

(define bvcopy-fptr (make-ftype-pointer bvcopy_t "memcpy"))
(define bvcopy-fptr (make-ftype-pointer bvcopy_t (foreign-entry "memcpy")))

A library that defines memcpy must be loaded first via load-shared-object, or memcpy must be registered via one of the methods described in Section  4.6.

syntax: (ftype-pointer? obj)
returns: #t if obj is an ftype pointer, otherwise #f
syntax: (ftype-pointer? ftype-name obj)
returns: #t if obj is an ftype-name, otherwise #f
libraries: (chezscheme)

(define-ftype Widget1 (struct [x int] [y int]))
(define-ftype Widget2 (struct [w Widget1] [b boolean]))

(define x1 (make-ftype-pointer Widget1 #x80000000))
(define x2 (make-ftype-pointer Widget2 #x80000000))

(ftype-pointer? x1) <graphic> #t
(ftype-pointer? x2) <graphic> #t

(ftype-pointer? Widget1 x1) <graphic> #t
(ftype-pointer? Widget1 x2) <graphic> #t

(ftype-pointer? Widget2 x1) <graphic> #f
(ftype-pointer? Widget2 x2) <graphic> #t

(ftype-pointer? #x80000000) <graphic> #f
(ftype-pointer? Widget1 #x80000000) <graphic> #f

procedure: (ftype-pointer-address fptr)
returns: the address encapsulated within fptr
libraries: (chezscheme)

fptr must be an ftype-pointer object.

(define x (make-ftype-pointer E #x80000000))
(ftype-pointer-address x) <graphic> #x80000000

syntax: (ftype-pointer=? fptr1 fptr2)
returns: #t if fptr1 and fptr2 have the same address, otherwise #f
libraries: (chezscheme)

fptr1 and fptr2 must be ftype-pointer objects.

ftype-pointer=? might be defined as follows:

(define ftype-pointer=?
  (lambda (fptr1 fptr2)
    (= (ftype-pointer-address fptr1) (ftype-pointer-address fptr2))))

It is, however, guaranteed not to allocate bignums for the addresses even if the addresses do not fit in fixnum range.

syntax: (ftype-pointer-null? fptr)
returns: #t if the address of fptr is 0, otherwise #f
libraries: (chezscheme)

fptr must be an ftype-pointer object.

ftype-pointer-null? might be defined as follows:

(define ftype-pointer-null?
  (lambda (fptr)
    (= (ftype-pointer-address fptr) 0)))

It is, however, guaranteed not to allocate a bignum for the address even if the address does not fit in fixnum range.

syntax: (ftype-&ref ftype-name (a ...) fptr-expr)
syntax: (ftype-&ref ftype-name (a ...) fptr-expr index)
returns: an ftype-pointer object
libraries: (chezscheme)

The ftype-pointer object returned by ftype-&ref encapsulates the address of some object embedded directly or indirectly within the foreign object pointed to by the value of fptr-expr, offset by index, if present. The value of fptr-expr must be an ftype pointer (fptr) of the ftype identified by ftype-name, and index must either be the identifier * or evaluate to a fixnum, possibly negative. The index is automatically scaled by the size of the ftype identified by ftype-name, which allows the fptr to be treated as an array of ftype-name objects and index as an index into that array. An index of * or 0 is the same as no index.

The sequence of accessors a ... must specify a valid path through the identified ftype. For struct, union, and bits ftypes, an accessor must be a valid field name for the ftype, while for pointer and array ftypes, an accessor must be the identifier * or evaluate to a fixnum index. For array ftypes, an index must be nonnegative, and for array ftypes with nonzero length, an index must also be less than the length.

The examples below assume the definitions of B and BB shown above in the description of ftype-sizeof. Fixed addresses are shown for illustrative purposes and are assumed to be valid, although addresses are generally determined at run time via foreign-alloc or some other mechanism.

(define x (make-ftype-pointer B #x80000000))
(ftype-&ref B () x) <graphic> #<ftype-pointer #x80000000>
(let ([idx 1])             <graphic> #<ftype-pointer #x8000002C>
  (ftype-&ref B () x idx))
(let ([idx -1])            <graphic> #<ftype-pointer #x7FFFFFD4>
  (ftype-&ref B () x idx))
(ftype-&ref B (b1) x) <graphic> #<ftype-pointer #x80000000>
(ftype-&ref B (b2) x) <graphic> #<ftype-pointer #x80000004>
(ftype-&ref B (b2 5) x) <graphic> #<ftype-pointer #x80000018>
(let ([n 5]) (ftype-&ref B (b2 n) x)) <graphic> #<ftype-pointer #x80000018>

(ftype-&ref B (b1 b2) x) <graphic> syntax error
(ftype-&ref B (b2 15) x) <graphic> run-time exception

(define y (make-ftype-pointer BB #x90000000))
(ftype-set! BB (bb2) y x)
(ftype-&ref BB (bb1 b2) y) <graphic> #<ftype-pointer #x90000004>
(ftype-&ref BB (bb2 * b2) y) <graphic> #<ftype-pointer #x80000004>
(let ([idx 1])                    <graphic> #<ftype-pointer #x80000030>
  (ftype-&ref BB (bb2 idx b2) y))

With no accessors and no index, as in the first use of ftype-&ref above, the returned ftype-pointer might be eq? to the input. Otherwise, the ftype-pointer is freshly allocated.

syntax: (ftype-set! ftype-name (a ...) fptr-expr val-expr)
syntax: (ftype-set! ftype-name (a ...) fptr-expr index val-expr)
returns: unspecified
syntax: (ftype-ref ftype-name (a ...) fptr-expr)
syntax: (ftype-ref ftype-name (a ...) fptr-expr index)
returns: an ftype-pointer object
libraries: (chezscheme)

These forms are used to store values into or retrieve values from the object pointed to by the value of fptr-expr, offset by index, if present. The value of fptr-expr must be an ftype pointer (fptr) of the ftype identified by ftype-name, and index must either be the identifier * or evaluate to a fixnum, possibly negative. The index is automatically scaled by the size of the ftype identified by ftype-name, which allows the fptr to be treated as an array of ftype-name objects and index as an index into that array. An index of * or 0 is the same as no index.

The sequence of accessors a ... must specify a valid path through the identified ftype. For struct, union, and bits ftypes, an accessor must be a valid field name for the ftype, while for pointer and array ftypes, an accessor must be the identifier * or evaluate to a fixnum index. For array ftypes, an index must be nonnegative, and for array ftypes with nonzero length, an index must also be less than the length. The field or element specified by the sequence of accessors must be a scalar field, e.g., a pointer field or a field containing a base type such as an int, char, or double.

For ftype-set!, val-expr must evaluate to a value of the appropriate type for the specified field, e.g., an ftype pointer of the appropriate type or an appropriate base-type value.

For both signed and unsigned integer fields, values in the range -2w-1 through 2w - 1 are accepted, where w is the width in bits of the integer field. For signed integer fields, values in the range 2w-1 through 2w - 1 are treated as two's complement representations of the corresponding negative numbers. For unsigned integer fields, values in the range -2w-1 through -1 are similarly treated as two's complement representations of the corresponding positive numbers.

char and wchar_t (wchar) field values are converted from (ftype-set!) or to (ftype-ref) Scheme characters, as if with char->integer and integer->char. Characters stored by ftype-set! into a char field must have Unicode scalar values in the range 0 through 255. Under Windows and any other system where wchar_t (wchar) is a 16-bit value, characters stored by ftype-set! into a whar_t (wchar) field must have Unicode scalar values in the range 0 through 216 - 1. On systems where wchar_t is a 32-bit value, any character can be stored in a wchar_t (wchar) field.

The examples below assume that B and C have been defined as shown in the description of ftype-sizeof above.

(define b
  (make-ftype-pointer B
    (foreign-alloc
      (* (ftype-sizeof B) 3))))
(define c
  (make-ftype-pointer C
    (foreign-alloc (ftype-sizeof C))))

(ftype-set! B (b1) b 5)
(ftype-set! B (b1) b 1 6)
(ftype-set! B (b1) c 5) <graphic> exception: ftype mismatch
(ftype-set! B (b2) b 0) <graphic> exception: not a scalar
(ftype-set! B (b2 -1) b 0) <graphic> exception: invalid index
(ftype-set! B (b2 0) b 50)
(ftype-set! B (b2 4) b 55)
(ftype-set! B (b2 10) b 55) <graphic> exception: invalid index

(ftype-set! C () c (ftype-&ref B () b 1))

(= (ftype-pointer-address (ftype-ref C () c))      <graphic> #t
   (+ (ftype-pointer-address b) (ftype-sizeof B)))
(= (ftype-pointer-address (ftype-&ref C (*) c)) <graphic> #t
   (+ (ftype-pointer-address b) (ftype-sizeof B)))
(= (ftype-pointer-address (ftype-&ref C (-1) c)) <graphic> #t
   (ftype-pointer-address b))

(ftype-ref C (-1 b1) c) <graphic> 5
(ftype-ref C (* b1) c) <graphic> 6
(ftype-ref C (-1 b2 0) c) <graphic> 50
(let ([i 4]) (ftype-ref C (-1 b2 i) c)) <graphic> 55

(ftype-set! C (-1 b2 0) c 75)
(ftype-ref B (b2 0) b) <graphic> 75
(foreign-free (ftype-pointer-address c))
(foreign-free (ftype-pointer-address b))

A function ftype pointer can be converted into a Scheme-callable procedure via ftype-ref. Assuming that a library defining memcpy has been loaded via load-shared-object or memcpy has been registered via one of the methods described in Section  4.6, A Scheme-callable memcpy can be defined as follows.

(define-ftype bvcopy_t (function (u8* u8* size_t) void))
(define bvcopy-fptr (make-ftype-pointer bvcopy_t "memcpy"))
(define bvcopy (ftype-ref bvcopy_t () bvcopy-fptr))

(define bv1 (make-bytevector 8 0))
(define bv2 (make-bytevector 8 57))
bv1 <graphic> #vu8(0 0 0 0 0 0 0 0)
bv2 <graphic> #vu8(57 57 57 57 57 57 57 57)
(bvcopy bv1 bv2 5)
bv1 <graphic> #vu8(57 57 57 57 57 0 0 0)

An ftype pointer can also be obtained as a return value from a C function declared to return a pointer to a function ftype.

Thus, ftype-ref with a function ftype is an alternative to foreign-procedure (Section 4.2) for creating Scheme-callable wrappers for C functions.

procedure: (ftype-pointer-ftype fptr)
returns: fptr's ftype, represented as an s-expression
libraries: (chezscheme)

fptr must be an ftype-pointer object.

(define-ftype Q0
  (struct
    [x int]
    [y int]))
(define-ftype Q1
  (struct
    [x double]
    [y char]
    [z (endian big
         (bits
           [_ unsigned 3]
           [a unsigned 9]
           [b unsigned 4]))]
    [w (* Q0)]))
(define q1 (make-ftype-pointer Q1 0))
(ftype-pointer-ftype q1) <graphic> (struct 
                            [x double]
                            [y char]
                            [z (endian big
                                 (bits
                                   [_ unsigned 3]
                                   [a unsigned 9]
                                   [b unsigned 4]))]
                            [w (* Q0)])

procedure: (ftype-pointer->sexpr fptr)
returns: an s-expression representation of the object to which fptr points
libraries: (chezscheme)

fptr must be an ftype-pointer object.

For each unnamed field, i.e., each whose field name is an underscore, the corresponding field value in the resulting s-expression is also an underscore. Similarly, if a field is inaccessible, i.e., if its address is invalid, the value is the symbol invalid.

(define-ftype Frob
  (struct
    [p boolean]
    [q char]))
(define-ftype Snurk
  (struct
    [a Frob]
    [b (* Frob)]
    [c (* Frob)]
    [d (bits
         [_ unsigned 15]
         [dx signed 17])]
    [e (array 5 double)]))
(define x
  (make-ftype-pointer Snurk
    (foreign-alloc (ftype-sizeof Snurk))))
(ftype-set! Snurk (b) x
  (make-ftype-pointer Frob
    (foreign-alloc (ftype-sizeof Frob))))
(ftype-set! Snurk (c) x
  (make-ftype-pointer Frob 0))
(ftype-set! Snurk (a p) x #t)
(ftype-set! Snurk (a q) x #\A)
(ftype-set! Snurk (b * p) x #f)
(ftype-set! Snurk (b * q) x #\B)
(ftype-set! Snurk (d dx) x -2500)
(do ([i 0 (fx+ i 1)])
    ((fx= i 5))
  (ftype-set! Snurk (e i) x (+ (* i 5.0) 3.0)))
(ftype-pointer->sexpr x) <graphic> (struct
                            [a (struct [p #t] [q #\A])]
                            [b (* (struct [p #f] [q #\B]))]
                            [c (* (struct [p invalid] [q invalid]))]
                            [d (bits [_ _] [dx -2500])]
                            [e (array 5 3.0 8.0 13.0 18.0 23.0)])

Section 4.6. Providing Access to Foreign Procedures

Access to foreign procedures can be provided in several ways:

procedure: (foreign-entry? entry-name)
returns: #t if entry-name is an existing foreign procedure entry point, #f otherwise
libraries: (chezscheme)

entry-name must be a string. foreign-entry? may be used to determine if an entry exists for a foreign procedure.

The following examples assume that a library that defines strlen has been loaded via load-shared-object or that strlen has been registered via one of the other methods described in this section.

(foreign-entry? "strlen") <graphic> #t
((foreign-procedure "strlen"
    (string) size_t)
 "hey!") <graphic> 4

procedure: (foreign-entry entry-name)
returns: the address of entry-name as an exact integer
libraries: (chezscheme)

entry-name must be a string naming an existing foreign entry point.

The following examples assume that a library that defines strlen has been loaded via load-shared-object or that strlen has been registered via one of the other methods described in this section.

(let ([addr (foreign-entry "strlen")])
  (and (integer? addr) (exact? addr))) <graphic> #t

(define-ftype strlen-type (function (string) size_t))
(define strlen
  (ftype-ref strlen-type ()
    (make-ftype-pointer strlen-type "strlen")))
(strlen "hey!") <graphic> 4

procedure: (foreign-address-name address)
returns: the entry name corresponding to address, if known, otherwise #f
libraries: (chezscheme)

The following examples assume that a library that defines strlen has been loaded via load-shared-object or that strlen has been registered via one of the other methods described in this section.

(foreign-address-name (foreign-entry "strlen")) <graphic> "strlen"

procedure: (load-shared-object path)
returns: unspecified
libraries: (chezscheme)

path must be a string. load-shared-object loads the shared object named by path. Shared objects may be system libraries or files created from ordinary C programs. All external symbols in the shared object, along with external symbols available in other shared objects linked with the shared object, are made available as foreign entries.

This procedure is supported for most platforms upon which Chez Scheme runs.

If path does not begin with a "." or "/", the shared object is searched for in a default set of directories determined by the system.

On most Unix systems, load-shared-object is based on the system routine dlopen. Under Windows, load-shared-object is based on LoadLibrary. Refer to the documentation for these routines and for the C compiler and loader for precise rules for locating and building shared objects.

load-shared-object can be used to access built-in C library functions, such as getenv. The name of the shared object varies from one system to another. On Linux systems:

(load-shared-object "libc.so.6")

On Solaris, OpenSolaris, FreeBSD, NetBSD, and OpenBSD systems:

(load-shared-object "libc.so")

On MacOS X systems:

(load-shared-object "libc.dylib")

On Windows:

(load-shared-object "crtdll.dll")

Once the C library has been loaded, getenv should be available as a foreign entry.

(foreign-entry? "getenv") <graphic> #t

An equivalent Scheme procedure may be defined and invoked as follows.

(define getenv
  (foreign-procedure "getenv"
    (string)
    string))
(getenv "HOME") <graphic> "/home/elmer/fudd"
(getenv "home") <graphic> #f

load-shared-object can be used to access user-created libraries as well. Suppose the C file "even.c" contains

int even(n) int n; { return n == 0 || odd(n - 1); }

and the C file "odd.c" contains

int odd(n) int n; { return n != 0 && even(n - 1); }

The files must be compiled and linked into a shared object before they can be loaded. How this is done depends upon the host system. On Linux, FreeBSD, OpenBSD, and OpenSolaris systems:

(system "cc -fPIC -shared -o evenodd.so even.c odd.c")

Depending on the host configuration, the -m32 or -m64 option might be needed to specify 32-bit or 64-bit compilation as appropriate.

On MacOS X (Intel or PowerPC) systems:

(system "cc -dynamiclib -o evenodd.so even.c odd.c")

Depending on the host configuration, the -m32 or -m64 option might be needed to specify 32-bit or 64-bit compilation as appropriate.

On 32-bit Sparc Solaris:

(system "cc -KPIC -G -o evenodd.so even.c odd.c")

On 64-bit Sparc Solaris:

(system "cc -xarch=v9 -KPIC -G -o evenodd.so even.c odd.c")

On Windows, we build a DLL (dynamic link library) file. In order to make the compiler generate the appropriate entry points, we alter even.c to read

#ifdef WIN32
#define EXPORT extern __declspec (dllexport)
#else
#define EXPORT extern
#endif

EXPORT int even(n) int n; { return n == 0 || odd(n - 1); }

and odd.c to read

#ifdef WIN32
#define EXPORT extern __declspec (dllexport)
#else
#define EXPORT extern
#endif

EXPORT int odd(n) int n; { return n != 0 && even(n - 1); }

We can then build the DLL as follows, giving it the extension ".so" rather than ".dll" for consistency with the other systems.

(system "cl -c -DWIN32 even.c")
(system "cl -c -DWIN32 odd.c")
(system "link -dll -out:evenodd.so even.obj odd.obj")

The resulting ".so" file can be loaded into Scheme and even and odd made available as foreign procedures:

(load-shared-object "./evenodd.so")
(let ([odd (foreign-procedure "odd"
             (integer-32) boolean)]
      [even (foreign-procedure "even"
              (integer-32) boolean)])
  (list (even 100) (odd 100))) <graphic> (#t #f)

The filename is given as "./evenodd.so" rather than simply "evenodd.so", because some systems look for shared libraries in a standard set of system directories that does not include the current directory.

procedure: (remove-foreign-entry entry-name)
returns: unspecified
libraries: (chezscheme)

remove-foreign-entry blocks further access to the entry specified by the string entry-name. An exception is raised with condition type &assertion if the entry does not exist. Since access previously established by foreign-procedure is not affected, remove-foreign-entry may be used to clean up after the desired interface to a group of foreign procedures has been established.

remove-foreign-entry can be used to remove entries registered using Sforeign_symbol and Sregister_symbol but not entries created as a result of a call to load-shared-object.

Section 4.7. Using Other Foreign Languages

Although the Chez Scheme foreign procedure interface is oriented primarily toward procedures defined in C or available in C libraries, it is possible to invoke procedures defined in other languages that follow C calling conventions. One source of difficulty may be the interpretation of names. Since Unix-based C compilers often prepend an underscore to external names, the foreign interface attempts to interpret entry names in a manner consistent with the host C compiler. Occasionally, such as for assembly coded files, this entry name interpretation may not be desired. It can be prevented by prefixing the entry name with an "=" character. For example, after loading an assembly file containing a procedure "foo" one might have:

(foreign-entry? "foo") <graphic> #f
(foreign-entry? "=foo") <graphic> #t

Section 4.8. C Library Routines

Additional foreign interface support is provided via a set of C preprocessor macros and C-callable library functions. Some of these routines allow C programs to examine, allocate, and alter Scheme objects. Others permit C functions to call Scheme procedures via a more primitive interface than that defined in Section 4.3. Still others permit the development of custom executable images and use of the Scheme system as a subordinate program within another program, e.g., for use as an extension language.

C code that uses these routines must include the "scheme.h" header file distributed with Chez Scheme and must be linked (statically or dynamically) with the Chez Scheme kernel. The header file contains definitions for the preprocessor macros and extern declarations for the library functions. The file is customized to the release of Chez Scheme and machine type with which it is distributed; it should be left unmodified to facilitate switching among Chez Scheme releases, and the proper version of the header file should always be used with C code compiled for use with a particular version of Chez Scheme. The version and machine type are defined in "scheme.h" under the names VERSION and MACHINE_TYPE.

The name of each routine begins with a capital S, e.g., Sfixnump. Many of the names are simple translations of the names of closely related Scheme procedures, e.g., Sstring_to_symbol is the C interface equivalent of string->symbol. Most externally visible entries in the Chez Scheme executable that are not documented here begin with capital S followed by an underscore (S_); their use should be avoided.

In addition to the various macros and external declarations given in scheme.h, the header file also defines (typedefs) several types used in the header file:

These types may vary depending upon the platform, although ptr is typically void *, iptr is typically long int, and uptr is typically unsigned long int.

Under Windows, defining SCHEME_IMPORT before including scheme.h causes scheme.h to declare its entry points using extern declspec (dllimport) rather than extern declspec (dllexport) (the default). Not defining SCHEME_IMPORT and instead defining SCHEME_STATIC causes scheme.h to declare exports using just extern. The static libraries distributed with Chez Scheme are built using SCHEME_STATIC.

The remainder of this section describes each of the C interface routines in turn. A declaration for each routine is given in ANSI C function prototype notation to precisely specify the argument and result types. Scheme objects have the C type ptr, which is defined in "scheme.h". Where appropriate, C values are accepted as arguments or returned as values in place of Scheme objects.

The preprocessor macros may evaluate their arguments more than once (or not at all), so care should be taken to ensure that this does not cause problems.

Customization.  The functions described here are used to initialize the Scheme system, build the Scheme heap, and run the Scheme system from a separate program.

[func] char * Skernel_version(void)
[func] void Sscheme_init(void (*abnormal_exit)(void))
[func] void Sset_verbose(int v)
[func] void Sregister_boot_file(const char *name)
[func] void Sbuild_heap(const char *exec, void (*custom_init)(void))
[func] void Senable_expeditor(const char *history_file)
[func] void Sretain_static_relocation(void)
[func] int Sscheme_start(int argc, char *argv[])
[func] int Sscheme_script(char *scriptfile, int argc, char *argv[])
[func] int Sscheme_program(char *programfile, int argc, char *argv[])
[func] void Scompact_heap(void)
[func] void Sscheme_deinit(void)

Skernel_version returns a string representing the Scheme version. It should be compared against the value of the VERSION preprocessor macro before any of the initialization functions listed above are used to verify that the correct "scheme.h" header file has been used.

Sscheme_init causes the Scheme system to initialize its static memory in preparation for boot file registration. The abnormal_exit parameter should be a (possibly null) pointer to a C function of no arguments that takes appropriate action if the initialization or subsequent heap-building process fails. If null, the default action is to call exit(1).

Sset_verbose sets verbose mode on for nonzero values of v and off when v is zero. In verbose mode, the system displays a trace of the search process for subsequently registered boot files.

Sregister_boot_file searches for the named boot file and register it for loading. The file is opened but not loaded until the heap is built via Sbuild_heap. For the first boot file registered only, the system also searches for the boot files upon which the named file depends, either directly or indirectly.

Sbuild_heap creates the Scheme heap from the registered boot files. exec is assumed to be the name of or path to the executable image and is used when no boot files have been registered as the base name for the boot-file search process. exec may be null only if one or more boot files have been registered. custom_init must be a (possibly null) pointer to a C function of no arguments; if non-null, it is called before any boot files are loaded.

Sscheme_start invokes the interactive startup procedure, i.e., the value of the parameter scheme-start, with one Scheme string argument for the first argc elements of argv, not including argv[0]. Sscheme_script similarly invokes the script startup procedure, i.e., the value of the parameter scheme-script, with one Scheme string argument for scriptfile and the first argc elements of argv, not including argv[0]. Sscheme_program similarly invokes the program startup procedure, i.e., the value of the parameter scheme-program, with one Scheme string argument for programfile and the first argc elements of argv, not including argv[0].

Senable_expeditor enables the expression editor (Section 2.2, Chapter 14), which is disabled by default, and determines the history file from which it restores and to which it saves the history. This procedure must be called after the heap is built, or an error will result. It must also be called before Sscheme_start in order to be effective. If the history_file argument is the null pointer, the history is not restored or saved. The preprocessor variable FEATURE_EXPEDITOR is defined in scheme.h if support for the expression editor has been compiled into the system.

Sretain_static_relocation causes relocation information to be retained for static generation code objects created by heap compaction for the benefit of compute-size and related procedures.

Scompact_heap compacts the Scheme heap and places all objects currently in the heap into a static generation. Objects in the static generation are never collected. That is, they are never moved during collection and the storage used for them is never reclaimed even if they become inaccessible. Scompact_heap is called implicitly after any boot files have been loaded.

Sscheme_deinit closes any open files, tears down the Scheme heap, and puts the Scheme system in an uninitialized state.

Predicates.  The predicates described here correspond to the similarly named Scheme predicates. A trailing letter p, for "predicate," is used in place of the question mark that customarily appears at the end of a Scheme predicate name. Each predicate accepts a single Scheme object and returns a boolean (C integer) value.

[macro] int Sfixnump(ptr obj)
[macro] int Scharp(ptr obj)
[macro] int Snullp(ptr obj)
[macro] int Seof_objectp(ptr obj)
[macro] int Sbwp_objectp(ptr obj)
[macro] int Sbooleanp(ptr obj)
[macro] int Spairp(ptr obj)
[macro] int Ssymbolp(ptr obj)
[macro] int Sprocedurep(ptr obj)
[macro] int Sflonump(ptr obj)
[macro] int Svectorp(ptr obj)
[macro] int Sbytevectorp(ptr obj)
[macro] int Sfxvectorp(ptr obj)
[macro] int Sstringp(ptr obj)
[macro] int Sbignump(ptr obj)
[macro] int Sboxp(ptr obj)
[macro] int Sinexactnump(ptr obj)
[macro] int Sexactnump(ptr obj)
[macro] int Sratnump(ptr obj)
[macro] int Sinputportp(ptr obj)
[macro] int Soutputportp(ptr obj)
[macro] int Srecordp(ptr obj)

Accessors.  Some of the accessors described here correspond to similarly named Scheme procedures, while others are unique to this interface. Sfixnum_value, Schar_value, Sboolean_value, and Sflonum_value return the C equivalents of the given Scheme value.

[macro] iptr Sfixnum_value(ptr fixnum)
[macro] uptr Schar_value(ptr character)
[macro] int Sboolean_value(ptr obj)
[macro] double Sflonum_value(ptr flonum)

Sinteger_value and Sunsigned_value are similar to Sfixnum_value, except they accept not only fixnum arguments but bignum arguments in the range of C integer or unsigned values. Sinteger_value and Sunsigned_value accept the same range of Scheme integer values. They differ only in the result type, and so allow differing interpretations of negative and large unsigned values.

[func] iptr Sinteger_value(ptr integer)
[macro] uptr Sunsigned_value(ptr integer)

Sinteger32_value, Sunsigned32_value, Sinteger64_value, and Sunsigned64_value accept signed or unsigned Scheme integers in the 32- or 64-bit range and return integers of the appropriate type for the machine type.

[func] <32-bit int type> Sinteger32_value(ptr integer)
[macro] <32-bit unsigned type> Sunsigned32_value(ptr integer)
[func] <64-bit int type> Sinteger64_value(ptr integer)
[macro] <64-bit unsigned type> Sunsigned64_value(ptr integer)

Scar, Scdr, Ssymbol_to_string (corresponding to symbol->string), and Sunbox are identical to their Scheme counterparts.

[macro] ptr Scar(ptr pair)
[macro] ptr Scdr(ptr pair)
[macro] ptr Ssymbol_to_string(ptr sym)
[macro] ptr Sunbox(ptr box)

Sstring_length, Svector_length, Sbytevector_length, and Sfxvector_length each return a C integer representing the length (in elements) of the object.

[macro] iptr Sstring_length(ptr str)
[macro] iptr Svector_length(ptr vec)
[macro] iptr Sbytevector_length(ptr bytevec)
[macro] iptr Sfxvector_length(ptr fxvec)

Sstring_ref, Svector_ref, Sbytevector_u8_ref, and Sfxvector_ref correspond to their Scheme counterparts, except that the index arguments are C integers, the return value for Sstring_ref is a C character, and the return value for Sbytevector_u8_ref is an octet (unsigned char).

[macro] char Sstring_ref(ptr str, iptr i)
[macro] ptr Svector_ref(ptr vec, iptr i)
[macro] octet Sbytevector_u8_ref(ptr fxvec, iptr i)
[macro] ptr Sfxvector_ref(ptr fxvec, iptr i)

A Scheme bytevector is represented as a length field followed by a sequence of octets (unsignec chars). Sbytevector_data returns a pointer to the start of the sequence of octets. Extreme care should be taken to stop dereferencing the pointer returned by Sbytevector_data or to lock the bytevector into memory (see Slock_object below) before any Scheme code is executed, whether by calling into Scheme or returning to a Scheme caller. The storage manager may otherwise relocate or discard the object into which the pointer points and may copy other data over the object.

[macro] octet * Sbytevector_data(ptr bytevec)

Mutators.  Changes to mutable objects that contain pointers, such as pairs and vectors, must be tracked on behalf of the storage manager, as described in one of the references [13]. The operations described here perform this tracking automatically where necessary.

[func] void Sset_box(ptr box, ptr obj)
[func] void Sset_car(ptr pair, ptr obj)
[func] void Sset_cdr(ptr pair, ptr obj)
[macro] void Sstring_set(ptr str, iptr i, char c)
[func] void Svector_set(ptr vec, iptr i, ptr obj)
[macro] void Sbytevector_u8_set(ptr bytevec, iptr i, octet n)
[macro] void Sfxvector_set(ptr fxvec, iptr i, ptr fixnum)

Some Scheme objects, such as procedures and numbers, are not mutable, so no operators are provided for altering the contents of those objects.

Constructors.  The constructors described here create Scheme objects. Some objects, such as fixnums and the empty list, are represented as immediate values that do not require any heap allocation; others, such as pairs and vectors, are represented as pointers to heap allocated objects.

Snil, Strue, Sfalse, Sbwp_object, Seof_object, and Svoid construct constant immediate values representing the empty list ( () ), the boolean values (#t and #f), the broken-weak-pointer object (#!bwp), the eof object (#!eof), and the void object.

[macro] ptr Snil
[macro] ptr Strue
[macro] ptr Sfalse
[macro] ptr Sbwp_object
[macro] ptr Seof_object
[macro] ptr Svoid

Fixnums, characters, booleans, flonums, and strings may be created from their C equivalents.

[macro] ptr Sfixnum(iptr n)
[macro] ptr Schar(char c)
[macro] ptr Sboolean(int b)
[func] ptr Sflonum(double x)
[func] ptr Sstring(const char *s)
[func] ptr Sstring_of_length(const char *s, iptr n)

Sstring creates a Scheme copy of the C string s, while Sstring_of_length creates a Scheme string of length n and copies the first n bytes from s into the new Scheme string.

It is possible to determine whether a C integer is within fixnum range by comparing the fixnum value of a fixnum created from a C integer with the C integer:

#define fixnum_rangep(x) (Sfixnum_value(Sfixnum(x)) == x)

Sinteger and Sunsigned may be used to create Scheme integers whether they are in fixnum range or not.

[func] ptr Sinteger(iptr n)
[func] ptr Sunsigned(uptr n)

Sinteger and Sunsigned differ in their treatment of negative C integer values as well as C unsigned integer values that would appear negative if cast to integers. Sinteger converts such values into negative Scheme values, whereas Sunsigned converts such values into the appropriate positive Scheme values. For example, assuming a 32-bit, two's complement representation for iptrs, Sinteger(-1) and Sunsigned((iptr)0xffffffff) both evaluate to the Scheme integer -1, whereas Sunsigned(0xffffffff) and Sunsigned((uptr)-1) both evaluate to the Scheme integer #xffffffff (4294967295).

Whichever routine is used, Sinteger_value and Sunsigned_value always reproduce the corresponding C input value, thus the following are all equivalent to x if x is an iptr.

Sinteger_value(Sinteger(x))
(iptr)Sunsigned_value(Sinteger(x))
Sinteger_value(Sunsigned((uptr)x))
(iptr)Sunsigned_value(Sunsigned((uptr)x))

Similarly, the following are all equivalent to x if x is a uptr.

(uptr)Sinteger_value(Sinteger((iptr)x))
Sunsigned_value(Sinteger((iptr)x))
(uptr)Sinteger_value(Sunsigned(x))
Sunsigned_value(Sunsigned(x))

Sinteger32, Sunsigned32, Sinteger64, and Sunsigned64 are like the generic equivalents but restrict their arguments to the 32- or 64-bit range.

[func] ptr Sinteger32(<32-bit int type> n)
[func] ptr Sunsigned32(<32-bit unsigned type> n)
[func] ptr Sinteger64(<64-bit int type> n)
[func] ptr Sunsigned64(<64-bit unsigned type> n)

Scons and Sbox are identical to their Scheme counterparts.

[func] ptr Scons(ptr obj1, ptr obj2)
[func] ptr Sbox(ptr obj)

Sstring_to_symbol is similar to its Scheme counterpart, string->symbol, except that it takes a C string (character pointer) as input.

[func] ptr Sstring_to_symbol(const char *s)

Smake_string, Smake_vector, Smake_bytevector, and Smake_fxvector are similar to their Scheme counterparts.

[func] ptr Smake_string(iptr n, int c)
[func] ptr Smake_vector(iptr n, ptr obj)
[func] ptr Smake_bytevector(iptr n, int fill)
[func] ptr Smake_fxvector(iptr n, ptr fixnum)

Smake_uninitialized_string is similar to the one-argument make-string.

[func] ptr Smake_uninitialized_string(iptr n)

Accessing top-level values.  Top-level variable bindings may be accessed or assigned via Stop_level_value and Sset_top_level_value.

[func] ptr Stop_level_value(ptr sym)
[func] void Sset_top_level_value(ptr sym, ptr obj)

These procedures give fast access to the bindings in the original interaction environment and do not reflect changes to the interaction-environment parameter or top-level module imports. To access the current interaction-environment binding for a symbol, it is necessary to call the Scheme top-level-value and set-top-level-value! procedures instead.

Locking Scheme objects.  The storage manager periodically relocates objects in order to reclaim storage and compact the heap. This relocation is completely transparent to Scheme programs, since all pointers to a relocated object are updated to refer to the new location of the object. The storage manager cannot, however, update Scheme pointers that reside outside of the Scheme heap.

As a general rule, all pointers from C variables or data structures to Scheme objects should be discarded before entry (or reentry) into Scheme. That is, if a C procedure receives an object from Scheme or obtains it via the mechanisms described in this section, all pointers to the object should be considered invalid once the C procedure calls into Scheme or returns back to Scheme. Dereferencing an invalid pointer or passing it back to Scheme can have disastrous effects, including unrecoverable memory faults. The foregoing does not apply to immediate objects, e.g., fixnums, characters, booleans, or the empty list. It does apply to all heap-allocated objects, including pairs, vectors, strings, all numbers other than fixnums, ports, procedures, and records.

In practice, the best way to ensure that C code does not retain pointers to Scheme objects is to immediately convert the Scheme objects into C equivalents, if possible. In certain cases, it is not possible to do so, yet retention of the Scheme object is essential to the design of the C portions of the program. In these cases, the object may be locked via the library routine Slock_object (or from Scheme, the equivalent procedure lock-object).

[func] void Slock_object(ptr obj)

Locking an object prevents the storage manager from reclaiming or relocating the object. Locking should be used sparingly, as it introduces memory fragmentation and increases storage management overhead. Locking can also lead to accidental retention of storage if objects are not unlocked. Locking objects that have been made static via heap compaction (see Scompact_heap above) is unnecessary but harmless.

Objects may be unlocked via Sunlock_object (unlock-object).

[func] void Sunlock_object(ptr obj)

An object may be locked more than once by successive calls to Slock_object or lock-object, in which case it must be unlocked by an equal number of calls to Sunlock_object or unlock-object before it is truly unlocked.

The function Sunlocked_objectp can be used to determine if an object is locked.

[func] int Sunlocked_objectp(ptr obj)

When a foreign procedure call is made into Scheme, a return address pointing into the Scheme code object associated with the foreign procedure is passed implicitly to the C routine. The system therefore locks the code object before calls are made from C back into Scheme and unlocks it upon return from Scheme. This locking is performed automatically; user code should never need to lock such code objects.

An object contained within a locked object, such as an object in the car of a locked pair, need not also be locked unless a separate C pointer to the object exists.

Registering foreign entry points.  Foreign entry points may be made visible to Scheme via Sforeign_symbol or Sregister_symbol.

[func] void Sforeign_symbol(const char *name, void *addr)
[func] void Sregister_symbol(const char *name, void *addr)

External entry points in object files or shared objects loaded as a result of a call to load-shared-object are automatically made visible by the system. Once a foreign entry point is made visible, it may be named in a foreign-procedure expression to create a Scheme-callable version of the entry point. Sforeign_symbol and Sregister_symbol allow programs to register nonexternal entry points, entry points in code linked statically with Chez Scheme, and entry points into code loaded directly from C, i.e., without load-shared-object. Sforeign_symbol and Sregister_symbol differ only in that Sforeign_symbol raises an exception when an attempt is made to register an existing name, whereas Sregister_symbol permits existing names to be redefined.

Obtaining Scheme entry points.  Sforeign_callable_entry_point extracts the entry point from a code object produced by foreign-callable, performing the same operation as its Scheme counterpart, i.e., the Scheme procedure foreign-callable-entry-point.

[func] (void (*) (void)) Sforeign_callable_entry_point(ptr code)

This can be used to avoid converting the code object into an address until just when it is needed, which may eliminate the need to lock the code object in some circumstances, assuming that the code object is not saved across any calls back into Scheme.

The inverse translation can be made via Sforeign_callable_code_object.

[func] ptr Sforeign_callable_code_object((void (*addr)(void)))

Low-level support for calls into Scheme.  Support for calling Scheme procedures from C is provided by the set of routines documented below. Calling a Scheme procedure that expects a small number of arguments (0-3) involves the use of one of the following routines.

[func] ptr Scall0(ptr procedure)
[func] ptr Scall1(ptr procedure, ptr obj1)
[func] ptr Scall2(ptr procedure, ptr obj1, ptr obj2)
[func] ptr Scall3(ptr procedure, ptr obj1, ptr obj2, ptr obj3)

In each case, the first argument, procedure, should be a Scheme procedure. The remaining arguments, which should be Scheme objects, are passed to the procedure. The tools described earlier in this section may be used to convert C datatypes into their Scheme equivalents. A program that automatically generates conversion code from declarations that are similar to foreign-procedure expressions is distributed with Chez Scheme. It can be found in the Scheme library directory on most systems in the file "foreign.ss".

A Scheme procedure may be obtained in a number of ways. For example, it may be received as an argument in a call from Scheme into C, obtained via another call to Scheme, extracted from a Scheme data structure, or obtained from the top-level environment via Stop_level_value.

A more general interface involving the following routines is available for longer argument lists.

[func] void Sinitframe(iptr n)
[func] void Sput_arg(iptr i, ptr obj)
[func] ptr Scall(ptr procedure, iptr n)

A C procedure first calls Sinitframe with one argument, the number of arguments to be passed to Scheme. It then calls Sput_arg once for each argument (in any order), passing Sput_arg the argument number (starting with 1) and the argument. Finally, it calls Scall to perform the call, passing it the Scheme procedure and the number of arguments (the same number as in the call to Sinitframe). Programmers should ensure a Scheme call initiated via Sinitframe is completed via Scall before any other calls to Scheme are made and before a return to Scheme is attempted. If for any reason the call is not completed after Sinitframe has been called, it may not be possible to return to Scheme.

The following examples serve to illustrate both the simpler and more general interfaces.

/* a particularly silly way to multiply two floating-point numbers */
double mul(double x, double y) {
    ptr times = Stop_level_value(Sstring_to_symbol("*"));

    return Sflonum_value(Scall2(times, Sflonum(x), Sflonum(y)));
}

/* an equally silly way to call printf with five arguments */

/* it is best to define interfaces such as the one below to handle
 * calls into Scheme to prevent accidental attempts to nest frame
 * creation and to help ensure that initiated calls are completed
 * as discussed above.  Specialized versions tailored to particular
 * C argument types may be defined as well, with embedded conversions
 * to Scheme objects. */
ptr Scall5(ptr p, ptr x1, ptr x2, ptr x3, ptr x4, ptr x5) {
    Sinitframe(5);
    Sput_arg(1, x1);
    Sput_arg(2, x2);
    Sput_arg(3, x3);
    Sput_arg(4, x4);
    Sput_arg(5, x5);
    Scall(p, 5);
}

static void dumpem(char *s, int a, double b, ptr c, char *d) {
    printf(s, a, b, c, d);
}

static void foo(int x, double y, ptr z, char *s) {
    ptr ois, sip, read, expr, eval, c_dumpem;
    char *sexpr = "(foreign-procedure \"dumpem\" (string integer-32\
 double-float scheme-object string) void)";

  /* this series of statements is carefully crafted to avoid referencing
     variables holding Scheme objects after calls into Scheme */
    ois = Stop_level_value(Sstring_to_symbol("open-input-string"));
    sip = Scall1(ois, Sstring(sexpr));
    read = Stop_level_value(Sstring_to_symbol("read"));
    expr = Scall1(read, sip);
    eval = Stop_level_value(Sstring_to_symbol("eval"));
    Sforeign_symbol("dumpem", (void *)dumpem);
    c_dumpem = Scall1(eval, expr);
    Scall5(c_dumpem,
           Sstring("x = %d, y = %g, z = %x, s = %s\n"),
           Sinteger(x),
           Sflonum(y),
           z,
           Sstring(s));
}

Calls from C to Scheme should not be made from C interrupt handlers. When Scheme calls into C, the system saves the contents of certain dedicated machine registers in a register save area. When C then calls into Scheme, the registers are restored from the register save area. Because an interrupt can occur at any point in a computation, the contents of the register save locations would typically contain invalid information that would cause the Scheme system to fail to operate properly.

Activating, deactivating, and destroying threads.  Three functions are provided by the threaded versions of Scheme to allow C code to notify Scheme when a thread should be activated, deactivated, or destroyed.

[func] int Sactivate_thread(void)
[func] void Sdeactivate_thread(void)
[func] int Sdestroy_thread(void)

A thread created via the Scheme procedure fork-thread starts in the active state and need not be activated. Any thread that has been deactivated, and any thread created by some mechanism other than fork-thread must, however, be activated before before it can access Scheme data or execute Scheme code. Sactivate_thread is used for this purpose. It returns 1 the first time the thread is activated and 0 on each subsequent call.

Since active threads operating in C code prevent the storage management system from garbage collecting, a thread should be deactivated via Sdeactivate_thread whenever it may spend a significant amount of time in C code. This is especially important whenever the thread calls a C library function, like read, that may block indefinitely. Once deactivated, the thread must not touch any Scheme data or execute any Scheme code until it is reactivated, with one exception. The exception is that the thread may access or even modify a locked Scheme object, such as a locked string, that contains no pointers to other, unlocked Scheme objects. (Objects that are not locked may be relocated by the garbage collector while the thread is inactive.)

Sdestroy_thread is used to notify the Scheme system that the thread is shut down and any thread-specific data can be released.

Low-level synchronization primitives.  The header file defines several preprocessor macros that can be used to lock memory locations in a manner identical to the corresponding ftype lock operations (sections  15.4 and 15.5).

[macro] void INITLOCK(void *addr)
[macro] void SPINLOCK(void *addr)
[macro] void UNLOCK(void *addr)
[macro] void LOCKED_INCR(void *addr, int *ret)
[macro] void LOCKED_DECR(void *addr, int *ret)

LOCKED_INCR and LOCKED_DECR set ret to a nonzero (true) value if the incremented or decremented value is 0. Otherwise they set ret to 0.

Section 4.9. Example: Socket Operations

This section presents a simple socket interface that employs a combination of Scheme and C code. The C code defines a set of convenient low-level operating-system interfaces that can be used in the higher-level Scheme code to open, close, read from, and write to sockets.

The C code (csocket.c) is given below, followed by the Scheme code (socket.ss). The code should require little or no modification to run on most Unix systems and can be modified to work under Windows (using the Windows WinSock interface).

A sample session demonstrating the socket interface follows the code. See Section 9.17 for an example that demonstrates how to use the same socket interface to build a process port that allows transparent input from and output to a subprocess via a Scheme port.

C code.  

/* csocket.c */

#include <sys/types.h>
#include <sys/socket.h>
#include <sys/un.h>
#include <string.h>
#include <errno.h>
#include <signal.h>
#include <sys/ioctl.h>
#include <stdio.h>

/* c_write attempts to write the entire buffer, pushing through
   interrupts, socket delays, and partial-buffer writes */
int c_write(int fd, char *buf, unsigned n) {
    unsigned i, m;

    m = n;
    while (m > 0) {
        if ((i = write(fd, buf, m)) < 0) {
            if (errno != EAGAIN && errno != EINTR)
                return i;
        } else {
            m -= i;
            buf += i;
        }
    }
    return n;
}

/* c_read pushes through interrupts and socket delays */
int c_read(int fd, char *buf, unsigned n) {
    int i;

    for (;;) {
        i = read(fd, buf, n);
        if (i >= 0) return i;
        if (errno != EAGAIN && errno != EINTR) return -1;
    }
}

/* bytes_ready(fd) returns true if there are bytes available
   to be read from the socket identified by fd */
int bytes_ready(int fd) {
    int n;

    (void) ioctl(fd, FIONREAD, &n);
    return n;
}

/* socket support */

/* do_socket() creates a new AF_UNIX socket */
int do_socket(void) {

    return socket(AF_UNIX, SOCK_STREAM, 0);
}

/* do_bind(s, name) binds name to the socket s */
int do_bind(int s, char *name) {
    struct sockaddr_un sun;
    int length;

    sun.sun_family = AF_UNIX;
    (void) strcpy(sun.sun_path, name);
    length = sizeof(sun.sun_family) + sizeof(sun.sun_path);

    return bind(s, &sun, length);
}

/* do_accept accepts a connection on socket s */
int do_accept(int s) {
    struct sockaddr_un sun;
    int length;

    length = sizeof(sun.sun_family) + sizeof(sun.sun_path);

    return accept(s, &sun, &length);
}

/* do_connect initiates a socket connection */
int do_connect(int s, char *name) {
    struct sockaddr_un sun;
    int length;

    sun.sun_family = AF_UNIX;
    (void) strcpy(sun.sun_path, name);
    length = sizeof(sun.sun_family) + sizeof(sun.sun_path);

    return connect(s, &sun, length);
}

/* get_error returns the operating system's error status */
char* get_error(void) {
    extern int errno;
    return strerror(errno);
}

Scheme code.  

;;; socket.ss

;;; Requires csocket.so, built from csocket.c.
(case (machine-type)
  [(i3le ti3le) (load-shared-object "libc.so.6")]
  [(i3osx ti3osx) (load-shared-object "libc.dylib")]
  [else (load-shared-object "libc.so")])

;;; Requires from C library:
;;;   close, dup, execl, fork, kill, listen, tmpnam, unlink
(load-shared-object "libc.so")

;;; basic C-library stuff

(define close
  (foreign-procedure "close" (integer-32)
    integer-32))

(define dup
  (foreign-procedure "dup" (integer-32)
    integer-32))

(define execl4
  (let ([execl-help
         (foreign-procedure "execl"
           (string string string string integer-32)
           integer-32)])
    (lambda (s1 s2 s3 s4)
      (execl-help s1 s2 s3 s4 0))))

(define fork
  (foreign-procedure "fork" ()
    integer-32))

(define kill
  (foreign-procedure "kill" (integer-32 integer-32)
    integer-32))

(define listen
  (foreign-procedure "listen" (integer-32 integer-32)
    integer-32))

(define tmpnam
  (foreign-procedure "tmpnam" (integer-32)
    string))

(define unlink
  (foreign-procedure "unlink" (string)
    integer-32))

;;; routines defined in csocket.c

(define accept
  (foreign-procedure "do_accept" (integer-32)
    integer-32))

(define bytes-ready?
  (foreign-procedure "bytes_ready" (integer-32)
    boolean))

(define bind
  (foreign-procedure "do_bind" (integer-32 string)
    integer-32))

(define c-error
  (foreign-procedure "get_error" ()
    string))

(define c-read
  (foreign-procedure "c_read" (integer-32 string integer-32)
    integer-32))

(define c-write
  (foreign-procedure "c_write" (integer-32 string integer-32)
    integer-32))

(define connect
  (foreign-procedure "do_connect" (integer-32 string)
    integer-32))

(define socket
  (foreign-procedure "do_socket" ()
    integer-32))

;;; higher-level routines

(define dodup
 ; (dodup old new) closes old and dups new, then checks to
 ; make sure that resulting fd is the same as old
  (lambda (old new)
    (check 'close (close old))
    (unless (= (dup new) old)
      (errorf 'dodup
        "couldn't set up child process io for fd ~s" old))))

(define dofork
 ; (dofork child parent) forks a child process and invokes child
 ; without arguments and parent with the child's pid
  (lambda (child parent)
    (let ([pid (fork)])
      (cond
        [(= pid 0) (child)]
        [(> pid 0) (parent pid)]
        [else (errorf 'fork (c-error))]))))

(define setup-server-socket
 ; create a socket, bind it to name, and listen for connections
  (lambda (name)
    (let ([sock (check 'socket (socket))])
      (unlink name)
      (check 'bind (bind sock name))
      (check 'listen (listen sock 1))
      sock)))

(define setup-client-socket
 ; create a socket and attempt to connect to server
  (lambda (name)
    (let ([sock (check 'socket (socket))])
      (check 'connect (connect sock name))
      sock)))

(define accept-socket
 ; accept a connection
  (lambda (sock)
    (check 'accept (accept sock))))

(define check
 ; raise an exception if status x is negative, using c-error to
 ; obtain the operating-system's error message
  (lambda (who x)
    (if (< x 0)
        (errorf who (c-error))
        x)))

(define terminate-process
 ; kill the process identified by pid
  (lambda (pid)
    (define sigterm 15)
    (kill pid sigterm)
    (void)))

Sample session.  

> (define client-pid)
> (define client-socket)
> (let* ([server-socket-name (tmpnam 0)]
         [server-socket (setup-server-socket server-socket-name)])
   ; fork a child, use it to exec a client Scheme process, and set
   ; up server-side client-pid and client-socket variables.
    (dofork   ; child
      (lambda () 
       ; the child establishes the socket input/output fds as
       ; stdin and stdout, then starts a new Scheme session
        (check 'close (close server-socket))
        (let ([sock (setup-client-socket server-socket-name)])
          (dodup 0 sock)
          (dodup 1 sock))
        (check 'execl (execl4 "/bin/sh" "/bin/sh" "-c" "exec scheme -q"))
        (errorf 'client "returned!"))
      (lambda (pid) ; parent
       ; the parent waits for a connection from the client
        (set! client-pid pid)
        (set! client-socket (accept-socket server-socket))
        (check 'close (close server-socket)))))
> (define put ; procedure to send data to client
    (lambda (x)
      (let ([s (format "~s~%" x)])
        (c-write client-socket s (string-length s)))
      (void)))
> (define get ; procedure to read data from client
    (let ([buff (make-string 1024)])
      (lambda ()
        (let ([n (c-read client-socket buff (string-length buff))])
          (printf "client:~%~a~%server:~%" (substring buff 0 n))))))
> (get)
server:
> (put '(let ([x 3]) x))
> (get)
client:
3
server:
> (terminate-process client-pid)
> (exit)

Chez Scheme Version 9 User's Guide
Copyright © 2017 Cisco Systems, Inc.
Licensed under the Apache License Version 2.0 (full copyright notice.).
Revised March 2017 for Chez Scheme Version 9.4.1
about this book