Skip to main content

A pure-python module to read and write binary data

Project description


A pure-python and lightweight module to read and write binary data


Rawutil is a module aimed at reading and writing binary data in python in the same way as the built-in struct module, but with more features. The rawutil's interface is thus compatible with struct, with a few small exceptions, and many things added. It does not have any non-builtin dependency.

What is already in struct

  • Unpack and pack fixed structures from/to bytes (pack, pack_into, unpack, unpack_from, iter_unpack, calcsize)
  • Struct objects that allow to parse one and for all a structure that may be used several times

What is different compared to struct

  • Some rarely-used format characters are not in rawutil (N, P and p are not available, n is used for a different purpose)
  • There is no consideration for native size and alignment, thus the @ characters simply applies system byte order with standard sizes and no alignment, just like =
  • There are several differences in error handling that are described below

What has been added to struct

  • Reading and writing files and file-like objects
  • New format characters, to handle padding, alignment, strings, ...
  • Internal references in structures
  • Loops in structures


Rawutil exports more or less the same interface as struct. In all those functions, structure may be a simple format string or a Struct object.


unpack(structure, data, names=None, refdata=())

Unpacks the given data according to the structure, and returns the unpacked values as a list.

  • structure is the structure of the data to unpack, as a format string or a Struct object
  • data may be a bytes-like or a file-like object. If it is a file-like object, the data will be unpacked starting from the current position in the file, and will leave the cursor at the end of the data that has been read (effectively reading the data to unpack from the file).
  • names may be a list of field names, or a namedtuple. If it is defined, the values will then be returned in a namedtuple with the given names or definition.
  • refdata may be used to easily input external data into the structure, as #n references. This will be described in the References part below

Unlike struct, this function does not raises any error if the data is larger than the structure expected size.

Examples :

>>> unpack("4B 3s 3s", b"\x01\x02\x03\x04foobar")
(1, 2, 3, 4, b"foo", b"bar")
>>> unpack("<4s #0I", b"ABCD\x10\x00\x00\x00\x20\x00\x00\x00", names=("string", "num1", "num2"), refdata=(2, ))
RawutilNameSpace(string=b'ABCD', num1=16, num2=32)


unpack_from(structure, data, position=None, names=None, refdata=(), getptr=False)

Unpacks the given data according to the structure starting from the given position, and returns the unpacked values as a list

This function works exactly like unpack, with two more optional arguments :

  • position can be used to specify a starting position to read. In a file-like object, the cursor is moved to the given position, then the data to unpack is read and the cursor is left at the end of the data that has been read. If this parameter is not set, it works like unpack and reads from the current position
  • getptr can be set to True to return the final position in the data, after the unpacked data. The function will then return (values, end_position). If left to False, it works like unpack and only returns the values.

Examples :

>>> unpack_from("<4s #0I", b"ABCD\x10\x00\x00\x00\x20\x00\x00\x00", names=("string", "num1", "num2"), refdata=(2, ))
RawutilNameSpace(string=b'ABCD', num1=16, num2=32)
>>> values, endpos = unpack_from("<2I", b"ABCD\x10\x00\x00\x00\x20\x00\x00\x00EFGH", 4, getptr=True)
>>> values
[16, 32]
>>> endpos


iter_unpack(structure, data, names=None, refdata=())

Returns an iterator that will unpack according to the structure and return the values as a list at each iteration. The data must be of a multiple of the structure’s length. If names is defined, each iteration will return a namedtuple, most like unpack and unpack_from. refdata also works the same.

This function is present mostly to ensure compatibility with struct. It is rather recommended to use iterators in structures, that are faster and offer much more control.

Examples :

>>> for a, b, c in iter_unpack("3c", b"abcdefghijkl"):
...     print(a.decode("ascii"), b.decode("ascii"), c.decode("ascii"))
a b c
d e f
g h i
j k l


pack(self, *data, refdata=())

Packs the given data in the binary format defined by structure, and returns the packed data as a bytes object. refdata is still there to insert external data in the structure using the #n references, and is a named argument only.

Note that if the last element of data is a writable file-like object, the data will be written into it instead of being returned. This behaviour is deprecated and kept only for backwards-compatibility, to pack into a file you should rather use pack_file.

Examples :

>>> pack("<2In", 10, 100, b"String")
>>> pack(">#0B #1I", 10, 100, 1000, 10000, 100000, refdata=(2, 3))
>>> unpack(">2B3I", _)
[10, 100, 1000, 10000, 100000]


pack_into(structure, buffer, offset, *data, refdata=())

Packs the given data into the given buffer at the given offset according to the given structure. Refdata still has the same usage as everywhere else.

  • buffer must be a mutable bytes-like object (typically a bytearray). The data will be written directly into it at the given position
  • offset specifies the position to write the data to. It is a required argument.

Examples :

>>> b = bytearray(b"AB----GH")
>>> pack_into("4s", b, 2, b"CDEF")
>>> b


pack_file(structure, file, *data, position=None, refdata=())

Packs the given data into the given file according to the given structure. refdata is still there for the external references data.

  • file can be any binary writable file-like object.
  • position can be set to pack the data at a specific position in the file. If it is left to None, the data will be packed at the current position in the file. In either case, the cursor will end up at the end of the packed data.

Examples :

>>> file = io.BytesIO(b"\x00\x00\x00\x00\x00\x00\x00\x00")
>>> rawutil.pack_file("2B", file, 60, 61)  # Writes at the current position (0)
>>> rawutil.pack_file("c", file, b"A")     # Writes at the current position (now 2)
>>> rawutil.pack_file("2c", file, b"y", b"z", position=6)  # Writes at the given position (6)


calcsize(structure, refdata=())

Returns the size of the data represented by the given structure.

This function is kept to ensure compatibility with struct. However, rawutil structure are not always of a fixed length, as they use internal references and variable length formats. Hence calcsize only works on fixed-length structures, thus structures that only use :

  • Fixed-length format characters (basic types with set repeat count)
  • External references (#0 type references, if you provide their value in refdata)
  • Iterators with fixed number of repeats (2(…) or 5[…] will work)
  • Alignments (structures with a and |). As long as everything else is fixed, alignments are too.

Trying to compute the size of a structure that includes any of those elements will raise an FormatError (basically, anything that depends on the data to read / write) :

  • Variable-length format characters (namely n and $)
  • {…} iterators, as they depend on the amount of data remaining.
  • Internal references (any /1 or /p1 types references)


Struct(format, names=None)

Struct objects allow to pre-parse format strings once and for all. Indeed, using only format strings will force to parse them every time you use them. If a structure is used more than once, it will thus save time to wrap it in a Struct object. You can also set the element names once, they will then be used by default every time you unpack data with that structure. Any function that accepts a format string also accepts Struct objects. For convenience, Struct also defines the module-level functions, for the structure it represents (without the structure argument as it is for the represented structure) :

unpack(self, data, names=None, refdata=())
unpack_from(self, data, offset=None, names=None, refdata=(), getptr=False)
iter_unpack(self, data, names=None, refdata=())
pack(self, *data, refdata=())
pack_into(self, buffer, offset, *data, refdata=())
pack_file(self, file, *data, position=None, refdata=())
calcsize(self, refdata=None, tokens=None)

You can retrieve the byte order with the byteorder attribute (can be "little" or "big"), and the format string with the format attribute.

It is also possible to add structures (it can add Struct and format strings transparently), and multiply a Struct object :

>>> part1 = Struct("<4s")
>>> part2 = Struct("I /0(#0B #0b)")
>>> part3 = "I /0s #0a"
>>> part1 + part2 + part3
Struct("<4s I /1(#0B #0b) I /3s #1a")
>>> part2 * 3
Struct("<I /0(#0B #0b) I /2(#1B #1b) I /4(#2B #2b)")

As you can see, the references are automatically fixed : all absolute references in the resulting structure point on the element they pointed to previously. External references are fixed too, and supposed to be in sequence in refdata.

Note that if the added structures have different byte order marks, the resulting structure will always retain the byte order of the left operand.


Rawutil defines two types of exception :

  • FormatError : Raised when the format string parsing fails, or if the structure is invalid
  • OperationError : Raised when data fails to be read or written

Format strings

In the same way as the struct module, binary data structures are defined with format strings.

Byte order marks

The first character of the format string may be used to specify the byte order to read the data in. Those are the same as in struct, except @ that is equivalent to = instead of setting native sizes and alignments.

Chr. Description
= System byte order (as defined by sys.byteorder)
@ Equivalent to =, system byte order
> Big endian (most significant byte last)
< Little endian (least significant byte first)
! Network byte order (big endian as defined by RFC 1700

If no byte order is defined in a structure, it is set to system byte order by default.


There are several format characters, that define various data types. Simple data types are described in the following table :

Chr. Type Size Description
? bool 1 Boolean value, 0 for False and any other value for True (packed as 0 and 1)
b int8 1 8 bits signed integer (7 bits + 1 sign bit)
B uint8 1 8 bits unsigned integer
h int16 2 16 bits signed integer
H uint16 2 16 bits unsigned integer
u int24 3 24 bits signed integer
U uint24 3 24 bits unsigned integer
i int32 4 32 bits signed integer
I uint32 4 32 bits unsigned integer
l int32 4 32 bits signed integer (same as i)
L uint32 4 32 bits unsigned integer (same as I)
q int64 8 64 bits signed integer
Q uint64 8 64 bits unsigned integer
e half 2 IEEE 754 half-precision floating-point number
f float 4 IEEE 754 single-precision floating-point number
d double 8 IEEE 754 double-precision floating-point number
F quad 16 IEEE 754 quadruple-precision floating-point number
c char 1 Character (returned as a 1-byte bytes object)
x void 1 Padding byte. Takes no data to pack (it simply inserts a null byte) nor returns anything.

A number before a simple format character may be added to indicate a repetition : "4I" means four 32-bits unsigned integers, and is equivalent to "IIII".

There also exist "special" format characters that define more complex types and behaviours :

Chr. Type Description
s char[] Fixed-length string. Represents a string of a given length, for example "16s" represents a 16-byte string. Returned as a single bytes object (as a contrary to c that only returns individual characters)
n string Null-terminated string. To unpack, reads until a null byte is found and returns the result as a bytes object, without the null byte. Packs the given bytes, and adds a null byte at the end.
X hex Works like s, but returns the result as an hexadecimal string.
a Inserts null bytes / reads until the data length reaches the next multiple of the given number (for example, "4a" goes to the next multiple of 4). Does not return anything and does not take input data to pack.
$ char[] When unpacking, returns all remaining data as a bytes object. When packing, simply packs the given bytes object. Must be the last element of the structure.

You can also set the base position for alignment with the | character. An alignment will then be performed according to the latest |. For example, "QBBB 4a" represents 1 uint64, 3 bytes and one alignment byte to get to the next multiple of 4 (12), whereas "QB| BB 4a" will align according to the | and give 1 uint64, 3 bytes and 2 alignment bytes, to get to 4 bytes since the last |.

Note that $ must be at the end of the structure. Any other element after a $ element will cause a FormatError


One of the biggest additions of rawutil is references. With rawutil, it is possible to use a value previously read as a repeat count for another element, and to insert custom values in a structure at run-time.

There are 3 types of references.

External references

An external reference is a reference to a value given at run-time — namely through the refdata argument of all rawutil functions In the format string, those are denoted by a #n element, with the index in refdata as n. For example, in the structure "#0B #1s", #0 will be replaced by the element 0 of refdata, and #1 by the element 1.

Example :

>>> unpack("#0B #1s", b"\x01\x02\x03foobar", refdata=(3, 6))
[1, 2, 3, b'foobar']

In the case above, it is equivalent to have "3B 6s" as the structure — but when you have to use several times the same structures with different repeat counts, it is possible to pre-compile the structure in a Struct object with external references, and then use the same object every time with different value, and without re-parsing the structure each time.

Absolute references

Absolute references allow to use a value previously read as a repeat count for another element further in the structure. Those are denoted with /n, with the index of the referenced element in the structure as n. For example, in the structure "I /0s", the integer is used to tell the length of the string, and the reference allows to read the string with that length.

Example :

>>> unpack("3B /0s /1s /2s", b"\x04\x03\x04spamhameggs")
[4, 3, 4, b'spam', b'ham', b'eggs']

Relative references

Relative references are similar to absolute references, except that they are relative to their location in the structure. They are denoted with /pN, where N is the number of elements to go back in the structure to find the referenced element. For example, /p1 gives the immediately previous element, /p2 the one before, and so on.

Example :

>>> unpack("B /p1s 2B /p2s /p2s", b"\x04spam\x03\x04hameggs")
[4, b'spam', 3, 4, b'ham', b'eggs']

This is especially useful in cases where there are a variable amount of elements before the referenced element, when the absolute references are unpractical — or when the structure is very long and absolute references become less practical.


The other big addition in rawutil is the substructures elements. Those can be used to isolate values in their own group instead of diluted in the global scope, or to easily read several times a same group of structure elements.

Note that a substructure always count as a single element towards references, and that references are local to their group : a /0 reference inside of a substructure will point to the first element of that substructure.

Alignments are also local to their substructure, thus will always align relative to the beginning of the substructure.


A group is simply a group of values isolated in their own list. Those are defined between parentheses (…). The values in a group are then extracted in a sub-list, and must be in a sub-list when packed.

Example :

>>> unpack("<I (3B) I", b"\xff\xff\xff\xff\x01\x02\x03\xff\xff\xff\xff")
[4294967295, [1, 2, 3], 4294967295]
>>> pack("<I (3B) I", 0xFFFFFFFF, (1, 2, 3), 0xFFFFFFFF)

When a repeat count is set to a group (as a number or as a reference, both are always valid), it will extract the group several times, but in the same sub-list, as a contrary to iterators that are described below.

Example :

>>> unpack("B 3(n)", b"\x0afoo\x00bar\x00foo2\x00")
[10, [b'foo', b'bar', b'foo2']]
>>> unpack("B /0(n)", b"\x03foo\x00bar\x00foo2\x00")
[3, [b'foo', b'bar', b'foo2']]


An iterator will extract its substructure as many times as it is told by its repeat count, in separate sub-lists. It is defined between square brackets […]

Example :

>>> unpack("B /0[B /0s]", b"\x03\x03foo\x03bar\x06foobar")
[3, [[3, b'foo'], [3, b'bar'], [6, b'foobar']]]
>>> pack("B /0[B /0s]", 2, ((3, b"foo"), (3, b"bar")))

Unbound iterators

While [] iterators are more or less equivalent to a for i in range(count), those are equivalent to a while. This kind of iterator is defined between curly brackets {…}, and extracts its substructure into a list of lists just like [], except that it extracts until there are no more data left to read. Thus you must not give it any repeat count (doing so will throw a FormatError), and it must always be the last element of its structure (it also raises an exception otherwise). The data to read must be an exact multiple of that substructure, otherwise it will throw an OperationError when attempting to unpack it.

Example :

>>> unpack("4s {Bn}", b"TEST\x00\foo\x00\x01bar\x00\x02foobar\x00")
[b'TEST', [[0, b'\x0coo'], [1, b'bar'], [2, b'foobar']]]
>>> pack("4s {Hn4a}", b"TEST", ((1, b"foo"), (1295, b"bar")))

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for rawutil, version 2.7.1
Filename, size File type Python version Upload date Hashes
Filename, size rawutil-2.7.1-py3-none-any.whl (15.6 kB) File type Wheel Python version py3 Upload date Hashes View
Filename, size rawutil-2.7.1.tar.gz (18.7 kB) File type Source Python version None Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page