Release Notes

Full Version History

February 26th 2013: version 3.1.0 released

This is a minor release with a couple of new features and some bug fixes.

New ‘pad’ token

This token can be used in reads and when packing/unpacking to indicate that you don’t care about the contents of these bits. Any padding bits will just be skipped over when reading/unpacking or zero-filled when packing.

>>> a, b = s.readlist('pad:5, uint:3, pad:1, uint:3')

Here only two items are returned in the list - the padding bits are ignored.

New clear and copy convenience methods

These methods have been introduced in Python 3.3 for lists and bytearrays, as more obvious ways of clearing and copying, and we mirror that change here.

t = s.copy() is equivalent to t = s[:], and s.clear() is equivalent to del s[:].

Other changes

  • Some bug fixes.

November 21st 2011: version 3.0.0 released

This is a major release which breaks backward compatibility in a few places.

Backwardly incompatible changes

Hex, oct and bin properties don’t have leading 0x, 0o and 0b

If you ask for the hex, octal or binary representations of a bitstring then they will no longer be prefixed with 0x, 0o or 0b. This was done as it was noticed that the first thing a lot of user code does after getting these representations was to cut off the first two characters before further processing.

>>> a = BitArray('0x123')
>>> a.hex, a.oct, a.bin
('123', '0443', '000100100011')

Previously this would have returned ('0x123', '0o0443', '0b000100100011')

This change might require some recoding, but it should all be simplifications.

ConstBitArray renamed to Bits

Previously Bits was an alias for ConstBitStream (for backward compatibility). This has now changed so that Bits and BitArray loosely correspond to the built-in types bytes and bytearray.

If you were using streaming/reading methods on a Bits object then you will have to change it to a ConstBitStream.

The ConstBitArray name is kept as an alias for Bits.

Stepping in slices has conventional meaning

The step parameter in __getitem__, __setitem__ and __delitem__ used to act as a multiplier for the start and stop parameters. No one seemed to use it though and so it has now reverted to the convential meaning for containers.

If you are using step then recoding is simple: s[a:b:c] becomes s[a*c:b*c].

Some examples of the new usage:

>>> s = BitArray('0x0000')
s[::4] = [1, 1, 1, 1]
>>> s.hex
>>> del s[8::2]
>>> s.hex

New features

New readto method

This method is a mix between a find and a read - it searches for a bitstring and then reads up to and including it. For example:

>>> s = ConstBitStream('0x47000102034704050647')
>>> s.readto('0x47', bytealigned=True)
>>> s.readto('0x47', bytealigned=True)
>>> s.readto('0x47', bytealigned=True)

pack function accepts an iterable as its format

Previously only a string was accepted as the format in the pack function. This was an oversight as it broke the symmetry between pack and unpack. Now you can use formats like this:

fmt = ['hex:8', 'bin:3']
a = pack(fmt, '47', '001')

June 18th 2011: version 2.2.0 released

This is a minor upgrade with a couple of new features.

New interleaved exponential-Golomb interpretations

New bit interpretations for interleaved exponential-Golomb (as used in the Dirac video codec) are supplied via uie and sie:

>>> s = BitArray(uie=41)
>>> s.uie
>>> s.bin

These are pretty similar to the non-interleaved versions - see the manual for more details. Credit goes to Paul Sargent for the patch.

New package-level bytealigned variable

A number of methods take a bytealigned parameter to indicate that they should only work on byte boundaries (e.g. find, replace, split). Previously this parameter defaulted to False. Instead it now defaults to bitstring.bytealigned, which itself defaults to False, but can be changed to modify the default behaviour of the methods. For example:

>>> a = BitArray('0x00 ff 0f ff')
>>> a.find('0x0f')
(4,)    # found first not on a byte boundary
>>> a.find('0x0f', bytealigned=True)
(16,)   # forced looking only on byte boundaries
>>> bitstring.bytealigned = True  # Change default behaviour
>>> a.find('0x0f')
>>> a.find('0x0f', bytealigned=False)

If you’re only working with bytes then this can help avoid some errors and save some typing!

Other changes

  • Fix for Python 3.2, correcting for a change to the binascii module.
  • Fix for bool initialisation from 0 or 1.
  • Efficiency improvements, including interning strategy.

February 23rd 2011: version 2.1.1 released

This is a release to fix a couple of bugs that were introduced in 2.1.0.

  • Bug fix: Reading using the ‘bytes’ token had been broken (Issue 102).
  • Fixed problem using some methods on ConstBitArray objects.
  • Better exception handling for tokens missing values.
  • Some performance improvements.

January 23rd 2011: version 2.1.0 released

New class hierarchy introduced with simpler classes

Previously there were just two classes, the immutable Bits which was the base class for the mutable BitString class. Both of these classes have the concept of a bit position, from which reads etc. take place so that the bitstring could be treated as if it were a file or stream.

Two simpler classes have now been added which are purely bit containers and don’t have a bit position. These are called ConstBitArray and BitArray. As you can guess the former is an immutable version of the latter.

The other classes have also been renamed to better reflect their capabilities. Instead of BitString you should use BitStream, and instead of Bits you can use ConstBitStream. The old names are kept as aliases for backward compatibility.

The classes hierarchy is:

       /    \
      /      \
BitArray   ConstBitStream (formerly Bits)
      \      /
       \    /
      BitStream (formerly BitString)

Other changes

A lot of internal reorganisation has taken place since the previous version, most of which won’t be noticed by the end user. Some things you might see are:

  • New package structure. Previous versions have been a single file for the module and another for the unit tests. The module is now split into many more files so it can’t be used just by copying any more.
  • To run the unit tests there is now a script called in the test directory.
  • File based bitstring are now implemented in terms of an mmap. This should be just an implementation detail, but unfortunately for 32-bit versions of Python this creates a limit of 4GB on the files that can be used. The work around is either to get a 64-bit Python, or just stick with version 2.0.
  • The ConstBitArray and ConstBitStream classes no longer copy byte data when a slice or a read takes place, they just take a reference. This is mostly a very nice optimisation, but there are occassions where it could have an adverse effect. For example if a very large bitstring is created, a small slice taken and the original deleted. The byte data from the large bitstring would still be retained in memory.
  • Optimisations. Once again this version should be faster than the last. The module is still pure Python but some of the reorganisation was to make it more feasible to put some of the code into Cython or similar, so hopefully more speed will be on the way.

July 26th 2010: version 2.0.3 released

  1. Bug fix: Using peek and read for a single bit now returns a new bitstring as was intended, rather than the old behaviour of returning a bool.
  2. Removed HTML docs from source archive - better to use the online version.

July 25th 2010: version 2.0.2 released

This is a major release, with a number of backwardly incompatible changes. The main change is the removal of many methods, all of which have simple alternatives. Other changes are quite minor but may need some recoding.

There are a few new features, most of which have been made to help the stream-lining of the API. As always there are performance improvements and some API changes were made purely with future performance in mind.

The backwardly incompatible changes are:

Methods removed

About half of the class methods have been removed from the API. They all have simple alternatives, so what remains is more powerful and easier to remember. The removed methods are listed here on the left, with their equivalent replacements on the right:

s.advancebit()              ->   s.pos += 1
s.advancebits(bits)         ->   s.pos += bits
s.advancebyte()             ->   s.pos += 8
s.advancebytes(bytes)       ->   s.pos += 8*bytes
s.allunset([a, b])          ->   s.all(False, [a, b])
s.anyunset([a, b])          ->   s.any(False, [a, b])
s.delete(bits, pos)         ->   del s[pos:pos+bits]
s.peekbit()                 ->   s.peek(1)
s.peekbitlist(a, b)         ->   s.peeklist([a, b])
s.peekbits(bits)            ->   s.peek(bits)
s.peekbyte()                ->   s.peek(8)
s.peekbytelist(a, b)        ->   s.peeklist([8*a, 8*b])
s.peekbytes(bytes)          ->   s.peek(8*bytes)
s.readbit()                 ->
s.readbitlist(a, b)         ->   s.readlist([a, b])
s.readbits(bits)            ->
s.readbyte()                ->
s.readbytelist(a, b)        ->   s.readlist([8*a, 8*b])
s.readbytes(bytes)          ->*bytes)
s.retreatbit()              ->   s.pos -= 1
s.retreatbits(bits)         ->   s.pos -= bits
s.retreatbyte()             ->   s.pos -= 8
s.retreatbytes(bytes)       ->   s.pos -= 8*bytes
s.reversebytes(start, end)  ->   s.byteswap(0, start, end)                 ->   s.pos = pos
s.seekbyte(bytepos)         ->   s.bytepos = bytepos
s.slice(start, end, step)   ->   s[start:end:step]
s.tell()                    ->   s.pos
s.tellbyte()                ->   s.bytepos
s.truncateend(bits)         ->   del s[-bits:]
s.truncatestart(bits)       ->   del s[:bits]
s.unset([a, b])             ->   s.set(False, [a, b])

Many of these methods have been deprecated for the last few releases, but there are some new removals too. Any recoding needed should be quite straightforward, so while I apologise for the hassle, I had to take the opportunity to streamline and rationalise what was becoming a bit of an overblown API.

set / unset methods combined

The set/unset methods have been combined in a single method, which now takes a boolean as its first argument:

s.set([a, b])               ->   s.set(1, [a, b])
s.unset([a, b])             ->   s.set(0, [a, b])
s.allset([a, b])            ->   s.all(1, [a, b])
s.allunset([a, b])          ->   s.all(0, [a, b])
s.anyset([a, b])            ->   s.any(1, [a, b])
s.anyunset([a, b])          ->   s.any(0, [a, b])

all / any only accept iterables

The all and any methods (previously called allset, allunset, anyset and anyunset) no longer accept a single bit position. The recommended way of testing a single bit is just to index it, for example instead of:

>>> if s.all(True, i):

just use

>>> if s[i]:

If you really want to you can of course use an iterable with a single element, such as s.any(False, [i]), but it’s clearer just to write not s[i].

Exception raised on reading off end of bitstring

If a read or peek goes beyond the end of the bitstring then a ReadError will be raised. The previous behaviour was that the rest of the bitstring would be returned and no exception raised.

BitStringError renamed to Error

The base class for errors in the bitstring module is now just Error, so it will likely appears in your code as bitstring.Error instead of the rather repetitive bitstring.BitStringError.

Single bit slices and reads return a bool

A single index slice (such as s[5]) will now return a bool (i.e. True or False) rather than a single bit bitstring. This is partly to reflect the style of the bytearray type, which returns an integer for single items, but mostly to avoid common errors like:

>>> if s[0]:
...     do_something()

While the intent of this code snippet is quite clear (i.e. do_something if the first bit of s is set) under the old rules s[0] would be true as long as s wasn’t empty. That’s because any one-bit bitstring was true as it was a non-empty container. Under the new rule s[0] is True if s starts with a 1 bit and False if s starts with a 0 bit.

The change does not affect reads and peeks, so s.peek(1) will still return a single bit bitstring, which leads on to the next item...

Empty bitstrings or bitstrings with only zero bits are considered False

Previously a bitstring was False if it had no elements, otherwise it was True. This is standard behaviour for containers, but wasn’t very useful for a container of just 0s and 1s. The new behaviour means that the bitstring is False if it has no 1 bits. This means that code like this:

>>> if s.peek(1):
...     do_something()

should work as you’d expect. It also means that Bits(1000), Bits(0x00) and Bits('uint:12=0') are all also False. If you need to check for the emptiness of a bitstring then instead check the len property:

if s                ->   if s.len
if not s            ->   if not s.len

Length and offset disallowed for some initialisers

Previously you could create bitstring using expressions like:

>>> s = Bits(hex='0xabcde', offset=4, length=13)

This has now been disallowed, and the offset and length parameters may only be used when initialising with bytes or a file. To replace the old behaviour you could instead use

>>> s = Bits(hex='0xabcde')[4:17]

Renamed format parameter fmt

Methods with a format parameter have had it renamed to fmt, to prevent hiding the built-in format. Affects methods unpack, read, peek, readlist, peeklist and byteswap and the pack function.

Iterables instead of * format accepted for some methods

This means that for the affected methods (unpack, readlist and peeklist) you will need to use an iterable to specify multiple items. This is easier to show than to describe, so instead of

>>> a, b, c, d = s.readlist('uint:12', 'hex:4', 'bin:7')

you would instead write

>>> a, b, c, d = s.readlist(['uint:12', 'hex:4', 'bin:7'])

Note that you could still use the single string 'uint:12, hex:4, bin:7' if you preferred.

Bool auto-initialisation removed

You can no longer use True and False to initialise single bit bitstrings. The reasoning behind this is that as bool is a subclass of int, it really is bad practice to have Bits(False) be different to Bits(0) and to have Bits(True) different to Bits(1).

If you have used bool auto-initialisation then you will have to be careful to replace it as the bools will now be interpreted as ints, so Bits(False) will be empty (a bitstring of length 0), and Bits(True) will be a single zero bit (a bitstring of length 1). Sorry for the confusion, but I think this will prevent bigger problems in the future.

There are a few alternatives for creating a single bit bitstring. My favourite is to use a list with a single item:

Bits(False)            ->   Bits([0])
Bits(True)             ->   Bits([1])

New creation from file strategy

Previously if you created a bitstring from a file, either by auto-initialising with a file object or using the filename parameter, the file would not be read into memory unless you tried to modify it, at which point the whole file would be read.

The new behaviour depends on whether you create a Bits or a BitString from the file. If you create a Bits (which is immutable) then the file will never be read into memory. This allows very large files to be opened for examination even if they could never fit in memory.

If however you create a BitString, the whole of the referenced file will be read to store in memory. If the file is very big this could take a long time, or fail, but the idea is that in saying you want the mutable BitString you are implicitly saying that you want to make changes and so (for now) we need to load it into memory.

The new strategy is a bit more predictable in terms of performance than the old. The main point to remember is that if you want to open a file and don’t plan to alter the bitstring then use the Bits class rather than BitString.

Just to be clear, in neither case will the contents of the file ever be changed - if you want to output the modified BitString then use the tofile method, for example.

find and rfind return a tuple instead of a bool

If a find is unsuccessful then an empty tuple is returned (which is False in a boolean sense) otherwise a single item tuple with the bit position is returned (which is True in a boolean sense). You shouldn’t need to recode unless you explicitly compared the result of a find to True or False, for example this snippet doesn’t need to be altered:

>>> if s.find('0x23'):
...     print(s.bitpos)

but you could now instead use

>>> found = s.find('0x23')
>>> if found:
...     print(found[0])

The reason for returning the bit position in a tuple is so that finding at position zero can still be True - it’s the tuple (0,) - whereas not found can be False - the empty tuple ().

The new features in this release are:

New count method

This method just counts the number of 1 or 0 bits in the bitstring.

>>> s = Bits('0x31fff4')
>>> s.count(1)

read and peek methods accept integers

The read, readlist, peek and peeklist methods now accept integers as parameters to mean “read this many bits and return a bitstring”. This has allowed a number of methods to be removed from this release, so for example instead of:

>>> a, b, c = s.readbits(5, 6, 7)
>>> if s.peekbit():
...     do_something()

you should write:

>>> a, b, c = s.readlist([5, 6, 7])
>>> if s.peek(1):
...     do_something()

byteswap used to reverse all bytes

The byteswap method now allows a format specifier of 0 (the default) to signify that all of the whole bytes should be reversed. This means that calling just byteswap() is almost equivalent to the now removed bytereverse() method (a small difference is that byteswap won’t raise an exception if the bitstring isn’t a whole number of bytes long).

Auto initialise with bytearray or (for Python 3 only) bytes

So rather than writing:

>>> a = Bits(bytes=some_bytearray)

you can just write

>>> a = Bits(some_bytearray)

This also works for the bytes type, but only if you’re using Python 3. For Python 2 it’s not possible to distinguish between a bytes object and a str. For this reason this method should be used with some caution as it will make you code behave differently with the different major Python versions.

>>> b = Bits(b'abcd\x23\x00') # Only Python 3!

set, invert, all and any default to whole bitstring

This means that you can for example write:

>>> a = BitString(100)       # 100 zero bits
>>> a.set(1)                 # set all bits to 1
>>> a.all(1)                 # are all bits set to 1?
>>> a.any(0)                 # are any set to 0?
>>> a.invert()               # invert every bit

New exception types

As well as renaming BitStringError to just Error there are also new exceptions which use Error as a base class.

These can be caught in preference to Error if you need finer control. The new exceptions sometimes also derive from built-in exceptions:

  1. ByteAlignError(Error) - whole byte position or length needed.
  2. ReadError(Error, IndexError) - reading or peeking off the end of the bitstring.
  3. CreationError(Error, ValueError) - inappropriate argument during bitstring creation.
  4. InterpretError(Error, ValueError) - inappropriate interpretation of binary data.

March 18th 2010: version 1.3.0 for Python 2.6 and 3.x released

New features

byteswap method for changing endianness

Changes the endianness in-place according to a format string or integer(s) giving the byte pattern. See the manual for details.

>>> s = BitString('0x00112233445566')
>>> s.byteswap(2)
>>> s
>>> s.byteswap('h')
>>> s
>>> s.byteswap([2, 5])
>>> s

Multiplicative factors in bitstring creation and reading

For example:

>>> s = Bits('100*0x123')

Token grouping using parenthesis

For example:

>>> s = Bits('3*(uint:6=3, 0b1)')

Negative slice indices allowed

The start and end parameters of many methods may now be negative, with the same meaning as for negative slice indices. Affects all methods with these parameters.

Sequence ABCs used

The Bits class now derives from collections.Sequence, while the BitString class derives from collections.MutableSequence.

Keywords allowed in readlist, peeklist and unpack

Keywords for token lengths are now permitted when reading. So for example, you can write

>>> s = bitstring.pack('4*(uint:n)', 2, 3, 4, 5, n=7)
>>> s.unpack('4*(uint:n)', n=7)
[2, 3, 4, 5]

start and end parameters added to rol and ror

join function accepts other iterables

Also its parameter has changed from ‘bitstringlist’ to ‘sequence’. This is technically a backward incompatibility in the unlikely event that you are referring to the parameter by name.

__init__ method accepts keywords

Rather than a long list of initialisers the __init__ methods now use a **kwargs dictionary for all initialisers except ‘auto’. This should have no effect, except that this is a small backward incompatibility if you use positional arguments when initialising with anything other than auto (which would be rather unusual).

More optimisations

A number of methods have been speeded up.

Bug fixed in replace method

(it could fail if start != 0).

January 19th 2010: version 1.2.0 for Python 2.6 and 3.x released

New ‘Bits’ class

Introducing a brand new class, Bits, representing an immutable sequence of bits.

The Bits class is the base class for the mutable BitString. The differences between Bits and BitStrings are:

  • Bits are immutable, so once they have been created their value cannot change. This of course means that mutating methods (append, replace, del etc.) are not available for Bits.
  • Bits are hashable, so they can be used in sets and as keys in dictionaries.
  • Bits are potentially more efficient than BitStrings, both in terms of computation and memory. The current implementation is only marginally more efficient though - this should improve in future versions.

You can switch from Bits to a BitString or vice versa by constructing a new object from the old.

>>> s = Bits('0xabcd')
>>> t = BitString(s)
>>> t.append('0xe')
>>> u = Bits(t)

The relationship between Bits and BitString is supposed to loosely mirror that between bytes and bytearray in Python 3.

Deprecation messages turned on

A number of methods have been flagged for removal in version 2. Deprecation warnings will now be given, which include an alternative way to do the same thing. All of the deprecated methods have simpler equivalent alternatives.

>>> t = s.slice(0, 2)
__main__:1: DeprecationWarning: Call to deprecated function slice.
Instead of 's.slice(a, b, c)' use 's[a:b:c]'.

The deprecated methods are: advancebit, advancebits, advancebyte, advancebytes, retreatbit, retreatbits, retreatbyte, retreatbytes, tell, seek, slice, delete, tellbyte, seekbyte, truncatestart and truncateend.

Initialise from bool

Booleans have been added to the list of types that can ‘auto’ initialise a bitstring.

>>> zerobit = BitString(False)
>>> onebit = BitString(True)

Improved efficiency

More methods have been speeded up, in particular some deletions and insertions.

Bug fixes

A rare problem with truncating the start of bitstrings was fixed.

A possible problem outputting the final byte in tofile() was fixed.

December 22nd 2009: version 1.1.3 for Python 2.6 and 3.x released

This version hopefully fixes an installation problem for platforms with case-sensitive file systems. There are no new features or other bug fixes.

December 18th 2009: version 1.1.2 for Python 2.6 and 3.x released

This is a minor update with (almost) no new features.

Improved efficiency

The speed of many typical operations has been increased, some substantially.

Initialise from integer

A BitString of ‘0’ bits can be created using just an integer to give the length in bits. So instead of

>>> s = BitString(length=100)

you can write just

>>> s = BitString(100)

This matches the behaviour of bytearrays and (in Python 3) bytes.

  • A defect related to using the set / unset functions on !BitStrings initialised from a file has been fixed.

November 24th 2009: version 1.1.0 for Python 2.6 and 3.x released

Note that this version will not work for Python 2.4 or 2.5. There may be an update for these Python versions some time next year, but it’s not a priorty quite yet. Also note that only one version is now provided, which works for Python 2.6 and 3.x (done with the minimum of hackery!)

New features

Improved efficiency

A fair number of functions have improved efficiency, some quite dramatically.

New bit setting and checking functions

Although these functions don’t do anything that couldn’t be done before, they do make some common use cases much more efficient. If you need to set or check single bits then these are the functions you need.

  • set / unset : Set bit(s) to 1 or 0 respectively.

  • allset / allunset : Check if all bits are 1 or all 0.

  • anyset / anyunset : Check if any bits are 1 or any 0.

    >>> s = BitString(length=1000)
    >>> s.set((10, 100, 44, 12, 1))
    >>> s.allunset((2, 22, 222))
    >>> s.anyset(range(7, 77))

New rotate functions

ror / rol : Rotate bits to the right or left respectively.

>>> s = BitString('0b100000000')
>>> s.ror(2)
>>> s.bin
>>> s.rol(5)
>>> s.bin

Floating point interpretations

New float initialisations and interpretations are available. These only work for BitStrings of length 32 or 64 bits.

>>> s = BitString(float=0.2, length=64)
>>> s.float
>>> t = bitstring.pack('<3f', -0.4, 1e34, 17.0)
>>> t.hex

‘bytes’ token reintroduced

This token returns a bytes object (equivalent to a str in Python 2.6).

>>> s = BitString('0x010203')
>>> s.unpack('bytes:2, bytes:1')
['\x01\x02', '\x03']

‘uint’ is now the default token type

So for example these are equivalent:

a, b = s.readlist('uint:12, uint:12')
a, b = s.readlist('12, 12')

October 10th 2009: version 1.0.1 for Python 3.x released

This is a straight port of version 1.0.0 to Python 3.

For changes since the last Python 3 release read all the way down in this document to version 0.4.3.

This version will also work for Python 2.6, but there’s no advantage to using it over the 1.0.0 release. It won’t work for anything before 2.6.

October 9th 2009: version 1.0.0 for Python 2.x released

Version 1 is here!

This is the first release not to carry the ‘beta’ tag. It contains a couple of minor new features but is principally a release to fix the API. If you’ve been using an older version then you almost certainly will have to recode a bit. If you’re not ready to do that then you may wish to delay updating.

So the bad news is that there are lots of small changes to the API. The good news is that all the changes are pretty trivial, the new API is cleaner and more ‘Pythonic’, and that by making it version 1.0 I’m promising not to tweak it again for some time.

API Changes

New read / peek functions for returning multiple items

The functions read, readbits, readbytes, peek, peekbits and peekbytes now only ever return a single item, never a list.

The new functions readlist, readbitlist, readbytelist, peeklist, peekbitlist and peekbytelist can be used to read multiple items and will always return a list.

So a line like:

>>> a, b ='uint:12, hex:32')


>>> a, b = s.readlist('uint:12, hex:32')

Renaming / removing functions

Functions have been renamed as follows:

``seekbit`` -> ``seek``

``tellbit`` -> ``tell``

``reversebits`` -> ``reverse``

``deletebits`` -> ``delete``

``tostring`` -> ``tobytes``

and a couple have been removed altogether:

  • deletebytes - use delete instead.
  • empty - use not s rather than s.empty().

Renaming parameters

The parameters ‘startbit’ and ‘endbit’ have been renamed ‘start’ and ‘end’. This affects the methods slice, find, findall, rfind, reverse, cut and split.

The parameter ‘bitpos’ has been renamed to ‘pos’. The affects the methods seek, tell, insert, overwrite and delete.

Mutating methods return None rather than self

This means that you can’t chain functions together so

>>> s.append('0x00').prepend('0xff')
>>> t = s.reverse()

Needs to be rewritten

>>> s.append('0x00')
>>> s.prepend('0xff')
>>> s.reverse()
>>> t = s

Affects truncatestart, truncateend, insert, overwrite, delete, append, prepend, reverse and reversebytes.

Properties renamed

The ‘data’ property has been renamed to ‘bytes’. Also if the BitString is not a whole number of bytes then a ValueError exception will be raised when using ‘bytes’ as a ‘getter’.

Properties ‘len’ and ‘pos’ have been added to replace ‘length’ and ‘bitpos’, although the longer names have not been removed so you can continue to use them if you prefer.

Other changes

  • The unpack method now always returns a list, never a single item.
  • BitStrings are now ‘unhashable’, so calling hash on one or making a set will fail.
  • The colon separating the token name from its length is now mandatory. So for example BitString('uint12=100') becomes BitString('uint:12=100').
  • Removed support for the ‘bytes’ token in format strings. Instead of'bytes:4') use'bits:32').

New features

Added endswith and startswith functions

These do much as you’d expect; they return True or False depending on whether the BitString starts or ends with the parameter.

>>> BitString('0xef342').startswith('0b11101')

September 11th 2009: version 0.5.2 for Python 2.x released

Finally some tools for dealing with endianness!

New interpretations are now available for whole-byte BitStrings that treat them as big, little, or native-endian

>>> big = BitString(intbe=1, length=16) # or BitString('intbe:16=1') if you prefer.
>>> little = BitString(intle=1, length=16)
>>> print big.hex, little.hex
0x0001 0x0100
>>> print big.intbe, little.intle
1 1

‘Struct’-like compact format codes

To save some typing when using pack, unpack, read and peek, compact format codes based on those used in the struct and array modules have been added. These must start with a character indicating the endianness (>, < or @ for big, little and native-endian), followed by characters giving the format:

b   1-byte signed int
B   1-byte unsigned int
h   2-byte signed int
H   2-byte unsigned int
l   4-byte signed int
L   4-byte unsigned int
q   8-byte signed int
Q   8-byte unsigned int

For example:

>>> s = bitstring.pack('<4h', 0, 1, 2, 3)

creates a BitString with four little-endian 2-byte integers. While

>>> x, y, z ='>hhl')

reads them back as two big-endian two-byte integers and one four-byte big endian integer.

Of course you can combine this new format with the old ones however you like:

>>> s.unpack('<h, intle:24, uint:5, bin')
[0, 131073, 0, '0b0000000001100000000']

August 26th 2009: version 0.5.1 for Python 2.x released

This update introduces pack and unpack functions for creating and dissembling BitStrings.

New pack() and unpack() functions

The pack function provides a flexible new method for creating BitStrings. Tokens for BitString ‘literals’ can be used in the same way as in the constructor.

>>> from bitstring import BitString, pack
>>> a = pack('0b11, 0xff, 0o77, int:5=-1, se=33')

You can also leave placeholders in the format, which will be filled in by the values provided.

>>> b = pack('uint:10, hex:4', 33, 'f')

Finally you can use a dictionary or keywords.

>>> c = pack('bin=a, hex=b, bin=a', a='010', b='ef')

The unpack method is similar to the read method except that it always unpacks from the start of the BitString.

>>> x, y = b.unpack('uint:10, hex')

If a token is given without a length (as above) then it will expand to fill the remaining bits in the BitString. This also now works with read and peek.

New tostring() and tofile() methods

The tostring method just returns the data as a string, with up to seven zero bits appended to byte align. The tofile method does the same except writes to a file object.

>>> f = open('myfile', 'wb')
>>> BitString('0x1234ff').tofile(f)

Other changes

The use of = is now mandatory in ‘auto’ initialisers. Tokens like uint12 100 will no longer work. Also the use of a : before the length is encouraged, but not yet mandated. So the previous example should be written as uint:12=100.

The ‘auto’ initialiser will now take a file object.

>>> f = open('myfile', 'rb')
>>> s = BitString(f)

July 19th 2009: version 0.5.0 for Python 2.x released

This update breaks backward compatibility in a couple of areas. The only one you probably need to be concerned about is the change to the default for bytealigned in find, replace, split, etc.

See the user manual for more details on each of these items.

Expanded abilities of ‘auto’ initialiser

More types can be initialised through the ‘auto’ initialiser. For example instead of

>>> a = BitString(uint=44, length=16)

you can write

>>> a = BitString('uint16=44')

Also, different comma-separated tokens will be joined together, e.g.

>>> b = BitString('0xff') + 'int8=-5'

can be written

>>> b = BitString('0xff, int8=-5')

New formatted read and peek methods

These takes a format string similar to that used in the auto initialiser. If only one token is provided then a single value is returned, otherwise a list of values is returned.

>>> start_code, width, height ='hex32, uint12, uint12')

is equivalent to

>>> start_code = s.readbits(32).hex
>>> width = s.readbits(12).uint
>>> height = s.readbits(12).uint

The tokens are:

int n   : n bits as an unsigned integer.
uint n  : n bits as a signed integer.
hex n   : n bits as a hexadecimal string.
oct n   : n bits as an octal string.
bin n   : n bits as a binary string.
ue      : next bits as an unsigned exp-Golomb.
se      : next bits as a signed exp-Golomb.
bits n  : n bits as a new BitString.
bytes n : n bytes as a new BitString.

See the user manual for more details.

hex and oct methods removed

The special methods for hex and oct have been removed. Please use the hex and oct properties instead.

>>> hex(s)


>>> s.hex

join made a method

The join function must now be called on a BitString object, which will be used to join the list together. You may need to recode slightly:

>>> s = bitstring.join('0x34', '0b1001', '0b1')


>>> s = BitString().join('0x34', '0b1001', '0b1')

More than one value allowed in readbits, readbytes, peekbits and peekbytes

If you specify more than one bit or byte length then a list of BitStrings will be returned.

>>> a, b, c = s.readbits(10, 5, 5)

is equivalent to

>>> a = readbits(10)
>>> b = readbits(5)
>>> c = readbits(5)

bytealigned defaults to False, and is at the end of the parameter list

Functions that have a bytealigned paramater have changed so that it now defaults to False rather than True. Also its position in the parameter list has changed to be at the end. You may need to recode slightly (sorry!)

readue and readse methods have been removed

Instead you should use the new read function with a ‘ue’ or ‘se’ token:

>>> i = s.readue()


>>> i ='ue')

This is more flexible as you can read multiple items in one go, plus you can now also use the peek method with ue and se.

Minor bugs fixed

See the issue tracker for more details.

June 15th 2009: version 0.4.3 for Python 2.x released

This is a minor update. This release is the first to bundle the bitstring manual. This is a PDF and you can find it in the docs directory.

New ‘cut’ method

This method returns a generator for constant sized chunks of a BitString.

>>> for byte in s.cut(8):
...     do_something_with(byte)

You can also specify a startbit and endbit, as well as a count, which limits the number of items generated:

>>> first100TSPackets = list(s.cut(188*8, count=100))

‘slice’ method now equivalent to __getitem__

This means that a step can also be given to the slice method so that the following are now the same thing, and it’s just a personal preference which to use:

>>> s1 = s[a:b:c]
>>> s2 = s.slice(a, b, c)

findall gets a ‘count’ parameter

So now

>>> list(a.findall(s, count=n))

is equivalent to

>>> list(a.findall(s))[:n]

except that it won’t need to generate the whole list and so is much more efficient.

Changes to ‘split’

The split method now has a ‘count’ parameter rather than ‘maxsplit’. This makes the interface closer to that for cut, replace and findall. The final item generated is now no longer the whole of the rest of the BitString.

  • A couple of minor bugs were fixed. See the issue tracker for details.

May 25th 2009: version 0.4.2 for Python 2.x released

This is a minor update, and almost doesn’t break compatibility with version 0.4.0, but with the slight exception of findall() returning a generator, detailed below.

Stepping in slices

The use of the step parameter (also known as the stride) in slices has been added. Its use is a little non-standard as it effectively gives a multiplicative factor to apply to the start and stop parameters, rather than skipping over bits.

For example this makes it much more convenient if you want to give slices in terms of bytes instead of bits. Instead of writing s[a*8:b*8] you can use s[a:b:8].

When using a step the BitString is effectively truncated to a multiple of the step, so s[::8] is equal to s if s is an integer number of bytes, otherwise it is truncated by up to 7 bits. So the final seven complete 16-bit words could be written as s[-7::16].

Negative slices are also allowed, and should do what you’d expect. So for example s[::-1] returns a bit-reversed copy of s (which is similar to s.reversebits(), which does the same operation on s in-place). As another example, to get the first 10 bytes in reverse byte order you could use s_bytereversed = s[0:10:-8].

Removed restrictions on offset

You can now specify an offset of greater than 7 bits when creating a BitString, and the use of offset is also now permitted when using the filename initialiser. This is useful when you want to create a BitString from the middle of a file without having to read the file into memory.

>>> f = BitString(filename='reallybigfile', offset=8000000, length=32)

Integers can be assigned to slices

You can now assign an integer to a slice of a BitString. If the integer doesn’t fit in the size of slice given then a ValueError exception is raised. So this is now allowed and works as expected:

>>> s[8:16] = 106

and is equivalent to

>>> s[8:16] = BitString(uint=106, length=8)

Less exceptions raised

Some changes have been made to slicing so that less exceptions are raised, bringing the interface closer to that for lists. So for example trying to delete past the end of the BitString will now just delete to the end, rather than raising a ValueError.

Initialisation from lists and tuples

A new option for the auto initialiser is to pass it a list or tuple. The items in the list or tuple are evaluated as booleans and the bits in the BitString are set to 1 for True items and 0 for False items. This can be used anywhere the auto initialiser can currently be used. For example:

>>> a = BitString([True, 7, False, 0, ()])     # 0b11000
>>> b = a + ['Yes', '']                        # Adds '0b10'
>>> (True, True, False) in a


  • reversebits now has optional startbit and endbit parameters.
  • As an optimisation findall will return a generator, rather than a list. If you still want the whole list then of course you can just call list() on the generator.
  • Improved efficiency of rfind.
  • A couple of minor bugs were fixed. See the issue tracker for details.

April 23rd 2009: Python 3 only version 0.4.1 released

This version is just a port of version 0.4.0 to Python 3. All the unit tests pass, but beyond that only limited ad hoc testing has been done and so it should be considered an experimental release. That said, the unit test coverage is very good - I’m just not sure if anyone even wants a Python 3 version!

April 11th 2009: version 0.4.0 released

New methods

Added rfind, findall and replace. These do pretty much what you’d expect - see the docstrings or the wiki for more information.

More special methods

Some missing methods were added: __repr__, __contains__, __rand__, __ror__, __rxor__ and __delitem__.


A couple of small bugs were fixed (see the issue tracker).

There are some small backward incompatibilities relative to version 0.3.2:

Combined find and findbytealigned

findbytealigned has been removed, and becomes part of find. The default start position has changed on both find and split to be the start of the BitString. You may need to recode:

>>> s1.find(bs)
>>> s2.findbytealigned(bs)
>>> s2.split(bs)


>>> s1.find(bs, bytealigned=False, startbit=s1.bitpos)
>>> s2.find(bs, startbit=s1.bitpos)  # bytealigned defaults to True
>>> s2.split(bs, startbit=s2.bitpos)

Reading off end of BitString no longer raises exception

Previously a read or peek function that encountered the end of the BitString would raise a ValueError. It will now instead return the remainder of the BitString, which could be an empty BitString. This is closer to the file object interface.

Removed visibility of offset

The offset property was previously read-only, and has now been removed from public view altogether. As it is used internally for efficiency reasons you shouldn’t really have needed to use it. If you do then use the _offset parameter instead (with caution).

March 11th 2009: version 0.3.2 released

Better performance

A number of methods (especially find and findbytealigned) have been sped up considerably.

Bit-wise operations

Added support for bit-wise AND (&), OR (|) and XOR (^). For example:

>>> a = BitString('0b00111')
>>> print a & '0b10101'


Added seekbit and seekbyte methods. These complement the ‘advance’ and ‘retreat’ functions, although you can still just use bitpos and bytepos properties directly.

>>> a.seekbit(100)                   # Equivalent to a.bitpos = 100

Allowed comparisons between BitString objects and strings. For example this will now work:

>>> a = BitString('0b00001111')
>>> a == '0x0f'

February 26th 2009: version 0.3.1 released

This version only adds features and fixes bugs relative to 0.3.0, and doesn’t break backwards compatibility.

Octal interpretation and initialisation

The oct property now joins bin and hex. Just prefix octal numbers with ‘0o’:

>>> a = BitString('0o755')
>>> print a.bin

Simpler copying

Rather than using b = copy.copy(a) to create a copy of a BitString, now you can just use b = BitString(a).

More special methods

Lots of new special methods added, for example bit-shifting via << and >>, equality testing via == and !=, bit inversion (~) and concatenation using *.

Also __setitem__ is now supported so BitString objects can be modified using standard index notation.

Proper installer

Finally got round to writing the distutils script. To install just python install.

February 15th 2009: version 0.3.0 released

Simpler initialisation from binary and hexadecimal

The first argument in the BitString constructor is now called ‘auto’ and will attempt to interpret the type of a string. Prefix binary numbers with ‘0b’ and hexadecimals with ‘0x’:

>>> a = BitString('0b0')         # single zero bit
>>> b = BitString('0xffff')      # two bytes

Previously the first argument was ‘data’, so if you relied on this then you will need to recode:

>>> a = BitString('\x00\x00\x01\xb3')   # Don't do this any more!


>>> a = BitString(data='\x00\x00\x01\xb3')

or just

>>> a = BitString('0x000001b3')

This new notation can also be used in functions that take a BitString as an argument. For example:

>>> a = BitString('0x0011') + '0xff'
>>> a.insert('0b001', 6)
>>> a.find('0b1111')

BitString made more mutable

The methods append, deletebits, insert, overwrite, truncatestart and truncateend now modify the BitString that they act upon. This allows for cleaner and more efficient code, but you may need to rewrite slightly if you depended upon the old behaviour:

>>> a = BitString(hex='0xffff')
>>> a = a.append(BitString(hex='0x00'))
>>> b = a.deletebits(10, 10)


>>> a = BitString('0xffff')
>>> a.append('0x00')
>>> b = copy.copy(a)
>>> b.deletebits(10, 10)

Thanks to Frank Aune for suggestions in this and other areas.

Changes to printing

The binary interpretation of a BitString is now prepended with ‘0b’. This is in keeping with the Python 2.6 (and 3.0) bin function. The prefix is optional when initialising using bin=.

Also, if you just print a BitString with no interpretation it will pick something appropriate - hex if it is an integer number of bytes, otherwise binary. If the BitString representation is very long it will be truncated by ‘...’ so it is only an approximate interpretation.

>>> a = BitString('0b0011111')
>>> print a
>>> a += '0b0'
>>> print a

More convenience functions

Some missing methods such as advancebit and deletebytes have been added. Also a number of ‘peek’ methods make an appearance as have prepend and reversebits. See the Tutorial for more details.

January 13th 2009: version 0.2.0 released

Some fairly minor updates, not really deserving of a whole version point update.

December 29th 2008: version 0.1.0 released

First release!