bytesparse 1.0.0

Objectives

This library aims to provide utilities to work with a virtual memory, which consists of a virtual addressing space where sparse chunks of data can be stored.

In order to be easy to use, its interface should be close to that of a bytearray, which is the closest pythonic way to store dynamic data. The main downside of a bytearray is that it requires a contiguous data allocation starting from address 0. This is not good when sparse data have to be stored, such as when emulating the addressing space of a generic microcontroller.

The main idea is to provide a bytearray-like class with the possibility to internally hold the sparse blocks of data. A block is ideally a tuple (start, data) where start is the start address and data is the container of data items (e.g. bytearray). The length of the block is len(data). Those blocks are usually not overlapping nor contiguous, and sorted by start address.

Python implementation

This library provides a pure Python implementation, for maximum compatibility.

Its implementation should be correct and robust, whilst trying to be as fast as common sense suggests. This means that the code should be reasonably optimized for general use, while still providing features that are less likely to be used, yet compatible with the existing Python API (e.g. bytearray or dict).

The Python implementation can also leverage the capabilities of its powerful int type, so that a virtually infinite addressing space can be used, even with negative addresses!

Data chunks are stored as common mutable bytearray objects, whose size is limited by the Python engine (i.e. that of size_t).

The bytesparse package provides the following virtual memory types:

• bytesparse.Memory, a generic virtual memory with infinite address range.

• bytesparse.bytesparse, a subclass behaving more like bytearray.

All the implementations inherit the behavior of collections.abc.MutableSequence and collections.abc.MutableMapping. Please refer to the collections.abc reference manual for more information about the interface API methods and capabilities.

Cython implementation

The library also provides an experimental Cython implementation. It tries to mimic the same algorithms of the Python implementation, while leveraging the speedup of compiled C code.

Please refer to the cbytesparse Python package for more details.

Examples

Here’s a quick usage example of bytesparse objects:

>>> from bytesparse import Memory
>>> from bytesparse import bytesparse

>>> m = bytesparse(b'Hello, World!')  # creates from bytes
>>> len(m)  # total length
13
>>> str(m)  # string representation, with bounds and data blocks
"<[[0, b'Hello, World!']]>"
>>> bytes(m)  # exports as bytes
b'Hello, World!'
>>> m.to_bytes()  # exports the whole range as bytes
b'Hello, World!'

>>> m.extend(b'!!')  # more emphasis!!!
>>> bytes(m)
b'Hello, World!!!'

>>> i = m.index(b',')  # gets the address of the comma
>>> m[:i] = b'Ciao'  # replaces 'Hello' with 'Ciao'
>>> bytes(m)
b'Ciao, World!!!'

>>> i = m.index(b',')  # gets the address of the comma
>>> m.insert(i, b'ne')  # inserts 'ne' to make 'Ciaone' ("big ciao")
>>> bytes(m)
b'Ciaone, World!!!'

>>> i = m.index(b',')  # gets the address of the comma
>>> m[(i - 2):i] = b' ciao'  # makes 'Ciaone' --> 'Ciao ciao'
>>> bytes(m)
b'Ciao ciao, World!!!'

>>> m.pop()  # less emphasis --> 33 == ord('!')
33
>>> bytes(m)
b'Ciao ciao, World!!'

>>> del m[m.index(b'l')]  # makes 'World' --> 'Word'
>>> bytes(m)
b'Ciao ciao, Word!!'

>>> m.popitem()  # less emphasis --> pops 33 (== '!') at address 16
(16, 33)
>>> bytes(m)
b'Ciao ciao, Word!'

>>> m.remove(b' ciao')  # self-explanatory
>>> bytes(m)
b'Ciao, Word!'

>>> i = m.index(b',')  # gets the address of the comma
>>> m.clear(start=i, endex=(i + 2))  # makes empty space between the words
>>> m.to_blocks()  # exports as data block list
[[0, b'Ciao'], [6, b'Word!']]
>>> m.contiguous  # multiple data blocks (emptiness inbetween)
False
>>> m.content_parts  # two data blocks
2
>>> m.content_size  # excluding emptiness
9
>>> len(m)  # including emptiness
11

>>> m.flood(pattern=b'.')  # replaces emptiness with dots
>>> bytes(m)
b'Ciao..Word!'
>>> m[-2]  # 100 == ord('d')
100

>>> m.peek(-2)  # 100 == ord('d')
100
>>> m.poke(-2, b'k')  # makes 'Word' --> 'Work'
>>> bytes(m)
b'Ciao..Work!'

>>> m.crop(start=m.index(b'W'))  # keeps 'Work!'
>>> m.to_blocks()
[[6, b'Work!']]
>>> m.span  # address range of the whole memory
(6, 11)
>>> m.start, m.endex  # same as above
(6, 11)

>>> m.bound_span = (2, 10)  # sets memory address bounds
>>> str(m)
"<2, [[6, b'Work']], 10>"
>>> m.to_blocks()
[[6, b'Work']]

>>> m.shift(-6)  # shifts to the left; NOTE: address bounds will cut 2 bytes!
>>> m.to_blocks()
[[2, b'rk']]
>>> str(m)
"<2, [[2, b'rk']], 10>"

>>> a = bytesparse(b'Ma')
>>> a.write(0, m)  # writes [2, b'rk'] --> 'Mark'
>>> a.to_blocks()
[[0, b'Mark']]

>>> b = Memory.from_bytes(b'ing', offset=4)
>>> b.to_blocks()
[[4, b'ing']]

>>> a.write(0, b)  # writes [4, b'ing'] --> 'Marking'
>>> a.to_blocks()
[[0, b'Marking']]

>>> a.reserve(4, 2)  # inserts 2 empty bytes after 'Mark'
>>> a.to_blocks()
[[0, b'Mark'], [6, b'ing']]

>>> a.write(4, b'et')  # --> 'Marketing'
>>> a.to_blocks()
[[0, b'Marketing']]

>>> a.fill(1, -1, b'*')  # fills asterisks between the first and last letters
>>> a.to_blocks()
[[0, b'M*******g']]

>>> v = a.view(1, -1)  # creates a memory view spanning the asterisks
>>> v[::2] = b'1234'  # replaces even asterisks with numbers
>>> a.to_blocks()
[[0, b'M1*2*3*4g']]
>>> a.count(b'*')  # counts all the asterisks
3
>>> v.release()  # release memory view

>>> c = a.copy()  # creates a (deep) copy
>>> c == a
True
>>> c is a
False

>>> del a[a.index(b'*')::2]  # deletes every other byte from the first asterisk
>>> a.to_blocks()
[[0, b'M1234']]

>>> a.shift(3)  # moves away from the trivial 0 index
>>> a.to_blocks()
[[3, b'M1234']]
>>> list(a.keys())
[3, 4, 5, 6, 7]
>>> list(a.values())
[77, 49, 50, 51, 52]
>>> list(a.items())
[(3, 77), (4, 49), (5, 50), (6, 51), (7, 52)]

>>> c.to_blocks()  # reminder
[[0, b'M1*2*3*4g']]
>>> c[2::2] = None  # clears (empties) every other byte from the first asterisk
>>> c.to_blocks()
[[0, b'M1'], [3, b'2'], [5, b'3'], [7, b'4']]
>>> list(c.intervals())  # lists all the block ranges
[(0, 2), (3, 4), (5, 6), (7, 8)]
>>> list(c.gaps())  # lists all the empty ranges
[(None, 0), (2, 3), (4, 5), (6, 7), (8, None)]

>>> c.flood(pattern=b'xy')  # fills empty spaces
>>> c.to_blocks()
[[0, b'M1x2x3x4']]

>>> t = c.cut(c.index(b'1'), c.index(b'3'))  # cuts an inner slice
>>> t.to_blocks()
[[1, b'1x2x']]
>>> c.to_blocks()
[[0, b'M'], [5, b'3x4']]
>>> t.bound_span  # address bounds of the slice (automatically activated)
(1, 5)

>>> k = bytesparse.from_blocks([[4, b'ABC'], [9, b'xy']], start=2, endex=15)  # bounded
>>> str(k)  # shows summary
"<2, [[4, b'ABC'], [9, b'xy']], 15>"
>>> k.bound_span  # defined at creation
(2, 15)
>>> k.span  # superseded by bounds
(2, 15)
>>> k.content_span  # actual content span (min/max addresses)
(4, 11)
>>> len(k)  # superseded by bounds
13
>>> k.content_size  # actual content size
5

>>> k.flood(pattern=b'.')  # floods between span
>>> k.to_blocks()
[[2, b'..ABC..xy....']]

Background

This library started as a spin-off of hexrec.blocks.Memory. That was based on a simple Python implementation using immutable objects (i.e. tuple and bytes). While good enough to handle common hexadecimal files, it was totally unsuited for dynamic/interactive environments, such as emulators, IDEs, data editors, and so on. Instead, bytesparse should be more flexible and faster, hopefully suitable for generic usage.

While developing the Python implementation, why not also jump on the Cython bandwagon, which permits even faster algorithms? Moreover, Cython itself is an interesting intermediate language, which brings to the speed of C, whilst being close enough to Python for the like.

Too bad, one great downside is that debugging Cython-compiled code is quite an hassle – that is why I debugged it in a crude way I cannot even mention, and the reason why there must be dozens of bugs hidden around there, despite the test suite :-) Moreover, the Cython implementation is still experimental, with some features yet to be polished (e.g. reference counting).

Documentation

For the full documentation, please refer to:

https://bytesparse.readthedocs.io/

Installation

From PyPI (might not be the latest version found on github):

$ pip install bytesparse

From the source code root directory:

$ pip install .

Development

To run the all the tests:

$ pip install tox
$ tox