Interpreting Bitstrings¶
Bitstrings don’t know or care how they were created; they are just collections of bits. This means that you are quite free to interpret them in any way that makes sense.
Several Python properties are used to create interpretations for the bitstring. These properties call private functions which will calculate and return the appropriate interpretation. These don’t change the bitstring in any way and it remains just a collection of bits. If you use the property again then the calculation will be repeated.
Note that these properties can potentially be very expensive in terms of both computation and memory requirements. For example if you have initialised a bitstring from a 10 GiB file object and ask for its binary string representation then that string will be around 80 GiB in size!
If you’re in an interactive session then the pretty-print method pp can be useful as it will only convert the bitstring one chunk at a time for display.
Properties¶
Many of the more commonly used interpretations have single letter equivalents.
The hex, bin, oct, int, uint and float properties can be shortened to h, b, o, i, u and f respectively.
Properties can have bit lengths appended to them to make properties such as f64, u32 or floatle32.
When used as a getter these just add an extra check on the bitstring’s length - if the bitstring is not the stated length then an InterpretError is raised. When used as a setter they define the new length of the bitstring.
s = BitArray() # Empty bitstring
s.f32 = 101.5 # New length is 32 bits, representing a float
For the properties described below we will use these:
>>> a = BitArray('0x123')
>>> b = BitArray('0b111')
bin / hex / oct¶
The most fundamental interpretation is perhaps as a binary string (a ‘bitstring’). The bin property returns a string of the binary representation of the bitstring. All bitstrings can use this property and it is used to test equality between bitstrings.
>>> a.bin
'000100100011'
>>> b.b
'111'
Note that the initial zeros are significant; for bitstrings the zeros are just as important as the ones!
For whole-byte bitstrings the most natural interpretation is often as hexadecimal, with each byte represented by two hex digits.
If the bitstring does not have a length that is a multiple of four bits then an InterpretError exception will be raised. This is done in preference to truncating or padding the value, which could hide errors in user code.
>>> a.hex
'123'
>>> b.h
ValueError: Cannot convert to hex unambiguously - not multiple of 4 bits.
For an octal interpretation use the oct property.
If the bitstring does not have a length that is a multiple of three then an InterpretError exception will be raised.
>>> a.oct
'0443'
>>> b.o
'7'
>>> (b + '0b0').oct
ValueError: Cannot convert to octal unambiguously - not multiple of 3 bits.
Integer types¶
To interpret the bitstring as a binary (base-2) bit-wise big-endian unsigned integer (i.e. a non-negative integer) use the uint property.
>>> a.uint
283
>>> b.u
7
For byte-wise big-endian, little-endian and native-endian interpretations use uintbe, uintle and uintne respectively. These will raise a ValueError if the bitstring is not a whole number of bytes long.
>>> s = BitArray('0x000001')
>>> s.uint # bit-wise big-endian
1
>>> s.uintbe # byte-wise big-endian
1
>>> s.uintle # byte-wise little-endian
65536
>>> s.uintne # byte-wise native-endian (will be 1 on a big-endian platform!)
65536
For a two’s complement interpretation as a base-2 signed integer use the int property. If the first bit of the bitstring is zero then the int and uint interpretations will be equal, otherwise the int will represent a negative number.
>>> a.int
283
>>> b.i
-1
For byte-wise big, little and native endian signed integer interpretations use intbe, intle and intne respectively. These work in the same manner as their unsigned counterparts described above.
bytes¶
A common need is to retrieve the raw bytes from a bitstring for further processing or for writing to a file. For this use the bytes interpretation, which returns a bytes object.
If the length of the bitstring isn’t a multiple of eight then a ValueError will be raised. This is because there isn’t an unequivocal representation as bytes. You may prefer to use the method tobytes as this will be pad with between one and seven zero bits up to a byte boundary if necessary.
>>> open('somefile', 'wb').write(a.tobytes())
>>> open('anotherfile', 'wb').write(('0x0'+a).bytes)
>>> a1 = BitArray(filename='somefile')
>>> a1.hex
'1230'
>>> a2 = BitArray(filename='anotherfile')
>>> a2.hex
'0123'
Note that the tobytes method automatically padded with four zero bits at the end, whereas for the other example we explicitly padded at the start to byte align before using the bytes property.
Floating point types¶
For a floating point interpretation use the float property. This uses the IEEE 754 floating point representation and will only work if the bitstring is 16, 32 or 64 bits long.
Different endiannesses are provided via floatle and floatne.
Note that as floating point interpretations are only valid on whole-byte bitstrings there is no difference between the bit-wise big-endian float and the byte-wise big-endian floatbe.
Note also that standard floating point numbers in Python are stored in 64 bits, so use this size if you wish to avoid rounding errors.
Other floating point types¶
A range of floating point types that are mostly used in machine learning are also availabe.
They include bfloat16 which is a truncated float32, together with IEEE 8-bit formats and a range of OCP Microscaling 8-bit, 6-bit and 4-bit formats.
See Exotic Floating Point Formats for more information.
Exponential-Golomb types¶
Some variable length integer types are supported. The lengths of these types depends upon the data being read and they are mainly used in video codecs.
See Exponential-Golomb Codes for more information.