G.1 Binary format of a FASL file

A FASL object is a header followed by one or more object fields followed by an end–of–FASL marker. After the end–of–FASL marker the file must be at its EOF.

The header is the string #@IK0, followed by 1 if fixnum width is 30 bits, else followed by 2; the end of fasl marker is @.

If the same object is referenced multiple times in the FASL file: it can be included only once and “marked” with a fixnum; such mark is later dereferenced to insert a reference to the object in the loaded code.

A data word is an exact signed integer of 32-bit or 64-bit depending on the word size of the underlying platform.

An object field is a character in ASCII encoding, optionally followed by data representing the serialisation of a Scheme value:

"N"

Denotes the empty list.

"T"

Denotes #t.

"F"

Denotes #f.

"E"

Denotes the end of file object.

"U"

Denotes the unspecified value (void).

"I" + word

A fixnum stored as a big endian word.

"s" + word(N) + octet ...

An ASCII string of N characters followed by N octets representing the characters in ASCII encoding. The data word N must represent an exact integer in the range of fixnums.

"S" + word(N) + int32 ...

A Unicode string of N characters followed by N 32-bit integers in native order representing the characters as Unicode code points. The data word N must represent an exact integer in the range of fixnums.

"M" + symbol-name

A symbol. symbol-name is a string field.

"G" + pretty-name + unique-name

A gensym. Both pretty-name and unique-name are strings.

"c" + octet

A character in the ASCII range (<= 255).

"C" + int32

A character stored as 32-bit integer representing a Unicode code point.

"P" + object1 + object2

A pair.

"V" + word(N) + object ...

A vector of length N followed by N object fields. The data word N must represent an exact integer in the range of fixnums.

"v" + word(N) + octet ...

A bytevector of length N followed by N octets. The data word N must represent an exact integer in the range of fixnums.

"R" + rtd-name + rtd-symbol + field-count + name ...

A struct type descriptor. rtd-name must be a symbol. rtd-symbol must be a symbol. field-count must be a word in the fixnum range. The name object fields must be symbols representing the field name and there must be field-count of them.

"{" + field-count + rtd + field ...

A structure instance. field-count must be a word in the fixnum range. rtd must be a struct type descriptor. The field object fields must be the fields of the structure instance and there must be field-count of them.

"r" + denominator + numerator

Ratnum.

"f" + 8-byte

IEEE double precision flonum. The bytes are stored: most significant first, least significant last.

"b" + word(N) + octet ...

Denotes a bignum. N is a signed integer word whose sign is the sign of the bignum and whose modulo is the number of octets representing the bignum.

"i" + real-part + imag-part

Complex numbers, both cflonum and compnum.

"l" + octet(N) + object ...

A short chain of pairs followed by its elements, including the cdr of the last pair; the number N <= 255 is 2 less than the number of elements. As example, the list:

(#\A . (#\B . (#\C . #\D)))

has N = 2, so it is serialised as:

"l" octet(2) #\A #\B #\C #\D

As other example, the standalone pair ‘(#\A . #\B)’ has N = 0, so it is serialised as:

"l" octet(0) #\A #\B

"L" + word(N) + object ...

A long chain of pairs followed by its elements, including the cdr of the last pair; the number N > 255 is 2 less than the number of elements. See the format of ‘l’ for details.

"h" + vector + vector

eq? hashtable, first vector keys, second vector values.

"H" + vector + vector

eqv? hashtable, first vector keys, second vector values.

"W" + name + parent + uid + sealed? + opaque? + count + (bool + field) ...

An R6RS record type descriptor.

name

Must be the return value of record-type-name.

parent

Must be the return value of record-type-parent.

uid

Must be the return value of record-type-uid.

sealed?

Must be the return value of record-type-sealed?.

opaque?

Must be the return value of record-type-opaque?.

count

Must be a fixnum representing the number of fields.

Each bool + field sequence must be a boolean representing the mutability of the record field, followed by symbol representing the name of the record field.

"x" + int + int + annotation + bytes + relocation vector

Denotes code. A serialised code object is represented as follows, after the ‘x’ header:

1. An exact integer representing the number of bytes actually used in the data area of the code object; this is the number of bytes used by the executable machine code.
   • On 32-bit platforms: a big endian 32-bit raw integer.
   • On 64-bit platforms: a sequence of two big endian 32-bit raw integers; the first integer represents the least significant 32 bits, the second integer represents the most significant 32 bits.
2. An exact integer representing the number of free variables in the code.
   • On 32-bit platforms: a big endian 32-bit integer representing the fixnum.
   • On 64-bit platforms: a sequence of two big endian 32-bit integers representing the fixnum; the first integer represents the least significant 32 bits, the second integer represents the most significant 32 bits.
3. A Scheme object representing the code annotation.
4. An array of bytes being the binary code.
5. The code relocation vector as an ordinary Scheme vector.

"Q" + "x" + code object

Procedure. A procedure is represented by the header Q followed by the serialisation of a code object, header x included.

">" + int32(I)

Mark the next object with index I.

"<" + int32(I)

Dereference the object marked with index I.

"O" + libid

Foreign library identifier. libid must be a string representing the foreign shared library identifier: on Unix–like systems it is prefixed with lib and suffixed with .so to compose a library file name.