.TH "Marshal" 3 2024-05-31 OCamldoc "OCaml library"
.SH NAME
Marshal \- Marshaling of data structures.
.SH Module
Module   Marshal
.SH Documentation
.sp
Module
.BI "Marshal"
 : 
.B sig end

.sp
Marshaling of data structures\&.
.sp
This module provides functions to encode arbitrary data structures
as sequences of bytes, which can then be written on a file or
sent over a pipe or network connection\&.  The bytes can then
be read back later, possibly in another process, and decoded back
into a data structure\&. The format for the byte sequences
is compatible across all machines for a given version of OCaml\&.
.sp
Warning: marshaling is currently not type\-safe\&. The type
of marshaled data is not transmitted along the value of the data,
making it impossible to check that the data read back possesses the
type expected by the context\&. In particular, the result type of
the 
.ft B
Marshal\&.from_*
.ft R
functions is given as 
.ft B
\&'a
.ft R
, but this is
misleading: the returned OCaml value does not possess type 
.ft B
\&'a
.ft R
for all 
.ft B
\&'a
.ft R
; it has one, unique type which cannot be determined
at compile\-time\&.  The programmer should explicitly give the expected
type of the returned value, using the following syntax:
.sp
\-
.ft B
(Marshal\&.from_channel chan : type)
.ft R
\&.
Anything can happen at run\-time if the object in the file does not
belong to the given type\&.

Values of extensible variant types, for example exceptions (of
extensible type 
.ft B
exn
.ft R
), returned by the unmarshaller should not be
pattern\-matched over through 
.ft B
match \&.\&.\&. with
.ft R
or 
.ft B
try \&.\&.\&. with
.ft R
,
because unmarshalling does not preserve the information required for
matching their constructors\&. Structural equalities with other
extensible variant values does not work either\&.  Most other uses such
as Printexc\&.to_string, will still work as expected\&.
.sp
The representation of marshaled values is not human\-readable,
and uses bytes that are not printable characters\&. Therefore,
input and output channels used in conjunction with 
.ft B
Marshal\&.to_channel
.ft R
and 
.ft B
Marshal\&.from_channel
.ft R
must be opened in binary mode, using e\&.g\&.
.ft B
open_out_bin
.ft R
or 
.ft B
open_in_bin
.ft R
; channels opened in text mode will
cause unmarshaling errors on platforms where text channels behave
differently than binary channels, e\&.g\&. Windows\&.

.sp

.sp
.sp
.I type extern_flags 
=
 | No_sharing  (* Don\&'t preserve sharing
 *)
 | Closures  (* Send function closures
 *)
 | Compat_32  (* Ensure 32\-bit compatibility
 *)
 
.sp

.sp

.PP
The flags to the 
.ft B
Marshal\&.to_*
.ft R
functions below\&.
.PP

.I val to_channel 
: 
.B out_channel -> 'a -> extern_flags list -> unit
.sp

.ft B
Marshal\&.to_channel chan v flags
.ft R
writes the representation
of 
.ft B
v
.ft R
on channel 
.ft B
chan
.ft R
\&. The 
.ft B
flags
.ft R
argument is a
possibly empty list of flags that governs the marshaling
behavior with respect to sharing, functional values, and compatibility
between 32\- and 64\-bit platforms\&.
.sp
If 
.ft B
flags
.ft R
does not contain 
.ft B
Marshal\&.No_sharing
.ft R
, circularities
and sharing inside the value 
.ft B
v
.ft R
are detected and preserved
in the sequence of bytes produced\&. In particular, this
guarantees that marshaling always terminates\&. Sharing
between values marshaled by successive calls to
.ft B
Marshal\&.to_channel
.ft R
is neither detected nor preserved, though\&.
If 
.ft B
flags
.ft R
contains 
.ft B
Marshal\&.No_sharing
.ft R
, sharing is ignored\&.
This results in faster marshaling if 
.ft B
v
.ft R
contains no shared
substructures, but may cause slower marshaling and larger
byte representations if 
.ft B
v
.ft R
actually contains sharing,
or even non\-termination if 
.ft B
v
.ft R
contains cycles\&.
.sp
If 
.ft B
flags
.ft R
does not contain 
.ft B
Marshal\&.Closures
.ft R
, marshaling fails
when it encounters a functional value inside 
.ft B
v
.ft R
: only \&'pure\&' data
structures, containing neither functions nor objects, can safely be
transmitted between different programs\&. If 
.ft B
flags
.ft R
contains
.ft B
Marshal\&.Closures
.ft R
, functional values will be marshaled as a the
position in the code of the program together with the values
corresponding to the free variables captured in the closure\&.  In
this case, the output of marshaling can only be read back in
processes that run exactly the same program, with exactly the same
compiled code\&. (This is checked at un\-marshaling time, using an MD5
digest of the code transmitted along with the code position\&.)
.sp
The exact definition of which free variables are captured in a
closure is not specified and can vary between bytecode and native
code (and according to optimization flags)\&.  In particular, a
function value accessing a global reference may or may not include
the reference in its closure\&.  If it does, unmarshaling the
corresponding closure will create a new reference, different from
the global one\&.
.sp
If 
.ft B
flags
.ft R
contains 
.ft B
Marshal\&.Compat_32
.ft R
, marshaling fails when
it encounters an integer value outside the range 
.ft B
\-2
.ft R
^
.ft B
30
.ft R
, 
.ft B
2
.ft R
^
.ft B
30
.ft R
.ft B
\-1
.ft R
of integers that are representable on a 32\-bit platform\&.  This
ensures that marshaled data generated on a 64\-bit platform can be
safely read back on a 32\-bit platform\&.  If 
.ft B
flags
.ft R
does not
contain 
.ft B
Marshal\&.Compat_32
.ft R
, integer values outside the
range 
.ft B
\-2
.ft R
^
.ft B
30
.ft R
, 
.ft B
2
.ft R
^
.ft B
30
.ft R
.ft B
\-1
.ft R
are marshaled, and can be read back on
a 64\-bit platform, but will cause an error at un\-marshaling time
when read back on a 32\-bit platform\&.  The 
.ft B
Mashal\&.Compat_32
.ft R
flag
only matters when marshaling is performed on a 64\-bit platform;
it has no effect if marshaling is performed on a 32\-bit platform\&.

.sp
.B "Raises Failure"
if 
.ft B
chan
.ft R
is not in binary mode\&.

.sp

.I val to_bytes 
: 
.B 'a -> extern_flags list -> bytes
.sp

.ft B
Marshal\&.to_bytes v flags
.ft R
returns a byte sequence containing
the representation of 
.ft B
v
.ft R
\&.
The 
.ft B
flags
.ft R
argument has the same meaning as for
.ft B
Marshal\&.to_channel
.ft R
\&.

.sp
.B "Since"
4.02

.sp

.I val to_string 
: 
.B 'a -> extern_flags list -> string
.sp
Same as 
.ft B
to_bytes
.ft R
but return the result as a string instead of
a byte sequence\&.

.sp

.I val to_buffer 
: 
.B bytes -> int -> int -> 'a -> extern_flags list -> int
.sp

.ft B
Marshal\&.to_buffer buff ofs len v flags
.ft R
marshals the value 
.ft B
v
.ft R
,
storing its byte representation in the sequence 
.ft B
buff
.ft R
,
starting at index 
.ft B
ofs
.ft R
, and writing at most
.ft B
len
.ft R
bytes\&.  It returns the number of bytes
actually written to the sequence\&. If the byte representation
of 
.ft B
v
.ft R
does not fit in 
.ft B
len
.ft R
characters, the exception 
.ft B
Failure
.ft R
is raised\&.

.sp

.I val from_channel 
: 
.B in_channel -> 'a
.sp

.ft B
Marshal\&.from_channel chan
.ft R
reads from channel 
.ft B
chan
.ft R
the
byte representation of a structured value, as produced by
one of the 
.ft B
Marshal\&.to_*
.ft R
functions, and reconstructs and
returns the corresponding value\&.

.sp
.B "Raises End_of_file"
if 
.ft B
chan
.ft R
is already at the end of the file\&.

.sp
.B "Raises Failure"
if the end of the file is reached during
unmarshalling itself or if 
.ft B
chan
.ft R
is not in binary mode\&.

.sp

.I val from_bytes 
: 
.B bytes -> int -> 'a
.sp

.ft B
Marshal\&.from_bytes buff ofs
.ft R
unmarshals a structured value
like 
.ft B
Marshal\&.from_channel
.ft R
does, except that the byte
representation is not read from a channel, but taken from
the byte sequence 
.ft B
buff
.ft R
, starting at position 
.ft B
ofs
.ft R
\&.
The byte sequence is not mutated\&.

.sp
.B "Since"
4.02

.sp

.I val from_string 
: 
.B string -> int -> 'a
.sp
Same as 
.ft B
from_bytes
.ft R
but take a string as argument instead of a
byte sequence\&.

.sp

.I val header_size 
: 
.B int
.sp
The bytes representing a marshaled value are composed of
a fixed\-size header and a variable\-sized data part,
whose size can be determined from the header\&.
.ft B
Marshal\&.header_size
.ft R
is the size, in bytes, of the header\&.
.ft B
Marshal\&.data_size
.ft R
.ft B
buff ofs
.ft R
is the size, in bytes,
of the data part, assuming a valid header is stored in
.ft B
buff
.ft R
starting at position 
.ft B
ofs
.ft R
\&.
Finally, 
.ft B
Marshal\&.total_size
.ft R
.ft B
buff ofs
.ft R
is the total size,
in bytes, of the marshaled value\&.
Both 
.ft B
Marshal\&.data_size
.ft R
and 
.ft B
Marshal\&.total_size
.ft R
raise 
.ft B
Failure
.ft R
if 
.ft B
buff
.ft R
, 
.ft B
ofs
.ft R
does not contain a valid header\&.
.sp
To read the byte representation of a marshaled value into
a byte sequence, the program needs to read first
.ft B
Marshal\&.header_size
.ft R
bytes into the sequence,
then determine the length of the remainder of the
representation using 
.ft B
Marshal\&.data_size
.ft R
,
make sure the sequence is large enough to hold the remaining
data, then read it, and finally call 
.ft B
Marshal\&.from_bytes
.ft R
to unmarshal the value\&.

.sp

.I val data_size 
: 
.B bytes -> int -> int
.sp
See 
.ft B
Marshal\&.header_size
.ft R
\&.

.sp

.I val total_size 
: 
.B bytes -> int -> int
.sp
See 
.ft B
Marshal\&.header_size
.ft R
\&.

.sp

.PP
.SS Marshal and domain safety
.sp
Care must be taken when marshaling a mutable value that may be modified by
a different domain\&. Mutating a value that is being marshaled (i\&.e\&., turned
into a sequence of bytes) is a programming error and might result in
suprising values (when unmarshaling) due to tearing, since marshaling
involves byte\-per\-byte copy\&.
.PP