Cengal

Cengal is a versatile Python library designed for a wide range of applications. To enhance performance, certain modules within Cengal have been implemented using Cython and/or C.

API Design Approach

The Cengal library adheres to an API design approach used in frameworks such as Qt. For those familiar with the C++ language, I will draw comparisons between the approaches of Qt and the C++ Standard Template Library (STL). The API provided by the STL was designed to significantly reduce the burden on programmers who develop the STL. This decision was logical from the standpoint of marketing the STL among compiler creators. However, this led to the usability of the STL for the user not being great. This is evident in the fact that the STL provides the most minimal possible API, and any conveniences must be programmed anew by each programmer every time - constantly reinventing the wheel. In contrast, Qt uses the opposite approach to API construction: classes have many methods whose purposes are similar, but are aimed at different usage models. This simplifies the use of Qt for users, speeds up the writing of the final code, and avoids many errors that we usually make when we write our own 'bicycles' for the same actions each time (not because the we are not smart, but because we are humans and therefore prone to make mistakes from time to time).

Cengal compatibility and requirements

• Target platforms: Win32, Linux, OS X, Android, iOS, Emscripten
• Target architectures: x64, x86, ARM
• Target interpreters: CPython, PyPy
• Recommended Python versions: 3.8+

Installation

To get started with Cengal, you can easily install it along with all the mandatory dependencies by running pip install cengal. This command not only installs Cengal and its required dependencies but also takes care of fetching and installing prebuilt wheels tailored for the Cengal library. These wheels are compatible with both Windows and Linux systems and work seamlessly with both CPython and PyPy interpreters.

If you prefer to install Cengal without its dependencies, you can opt for the 'cengal-light' package, which serves as the backend for the 'cengal' package. Simply run pip install cengal-light to get the lightweight version of Cengal.

Documentation

Cengal Documentation

For example Cengal Coroutines Concepts & Usage

or partially:

Cengal Wiki

For example Wiki: Cengal Coroutines Concepts & Usage

Stand-Alone Packages for Specific Cengal Modules

To cater to varying needs and streamline the installation process, I've introduced stand-alone packages for select Cengal modules. These packages are designed to offer users the ability to install specific Cengal functionality without the burden of the library's full set of dependencies.

The core of this approach lies in our 'cengal-light' package, which houses both Python and compiled Cengal modules. The 'cengal' package itself serves as a lightweight shell, devoid of its own modules, but dependent on 'cengal-light[full]' for a complete Cengal library installation with all required dependencies.

For users seeking individual Cengal features or looking to minimize dependencies, our stand-alone packages provide a convenient solution. Each stand-alone package is dedicated to a specific Cengal module and relies on 'cengal-light' as its sole dependency.

Below, you'll find a list of these stand-alone packages, each corresponding to a distinct Cengal module:

• cengal_memory_barriers (package for cengal.hardware.memory.barriers module): Fast crossplatform memory barriers for Python.
• cengal_cpu_info (package for cengal.hardware.info.cpu module): Extended, cached CPU info with consistent output format.
• cengal_app_dir_path_finder (package for cengal.file_system.app_fs_structure.app_dir_path module): Offering a unified API for easy retrieval of OS-specific application directories, enhancing data management across Windows, Linux, and macOS.

Stay tuned for future additions to our collection of stand-alone packages!

Exclusive Features: No Alternatives Online

Build system (work in progress)

Automatic hackable build system for your package which supports Python modules made with different languages: Cython (including Python to Cython automatic compilation), C/C++, ObjectiveC, Go, Nim. Other languages support is in progress.

Compiles your code, gather binary artifacts and puts them into your wheel.

Examples

• Compilable Golang module
• Compilable Nim module
• Pure Python module auto-compiled with Cython

Concurrent Execution of blocking CPU-Bound and GUI Tasks on a Single Thread

Cengal offers a unique and powerful feature that allows you to execute a diverse set of tasks concurrently on a single thread, effectively managing CPU-bound and GUI-related operations without introducing the complexity of multithreading or multiprocessing. Notably, Cengal can convert blocking CPU-bound functions into proper asynchronous coroutines, preventing them from blocking the thread for extended periods.

Examples

In this example, an application concurrently (at the same time) executes all of the following components within a single thread:

• own blocking CPU-bound function
• third-party blocking CPU-bound function
• Tkinter application
• CustomTkinter application
• asyncio-based file reading task.

YouTube Showcase

Source code

• rich_example.py
• third_party_cpu_bound.py

Tutorial

• Decorator which converts blocking code to concurrent code

True Interprocess Shared Memory (Proof of Concept)

Cengal introduces a novel approach to interprocess shared memory, currently at the proof of concept stage. With this feature, you can seamlessly share data between your Python processes (currently limited to 2 processes) and work with them just as you would in a single process. The underlying mechanism optimizes cross-process communication by employing efficient memory barriers instead of resource-intensive system calls.

Supported data types (current stage):

• shared Numpy arrays
• list: Unlike multiprocessing.shared_memory.ShareableList, Cengal's shared lists are both mutable and resizable between different processes. They support various container types (lists, tuples, dicts) as items and implement all standard list methods. Plus, they offer superior performance compared to multiprocessing.shared_memory.ShareableList.
• dict: Currently immutable.
• tuple
• str
• bytes
• bytearray
• bool
• float: Cengal's shared float values support Addition Assignment (shared_list[20] += 999.3) and all other native methods and operators, unlike values in multiprocessing.shared_memory.ShareableList.
• int: Currently limited to int64. Similar to shared float values, Cengal's shared integers support Addition Assignment (shared_list[15] += 999) and all other native methods and operators.
• None

Examples

General example:

• shared_memory_example.py

Messages transmit:

• shmem_sender.py
• shmem_receiver.py

Shared Numpy Array:

• numpy_array_shmem_main.py
• numpy_array_shmem_worker.py

and smaller:

from multiprocessing import Process
from cengal.hardware.memory.shared_memory import *


shared_memory_name = 'test_shared_mem'
shared_memory_size = 200 * 1024 * 1024
switches = 1000
changes_per_switch = 2000


def work(manager, shared_data)
    index = 0
    while index < switches:
        with wait_my_turn(manager):
            # emulatin our working process
            for i in range(changes_per_switch):
                shared_data[1] += 1

def second_process():
    consumer: SharedMemory = SharedMemory('test_shmem', False)
    consumer.wait_for_messages()
    with wait_my_turn(consumer):
        shared_data = consumer.take_message()
    
    work(consumer, shared_data)


creator: SharedMemory = SharedMemory(shared_memory_name, True, shared_memory_size)
p = Process(target=second_process)
p.start()
creator.wait_consumer_ready()
with wait_my_turn(creator):
    data = [
        'hello',
        0,
        (8, 2.0, False),
        {
            b'world': -6,
            5: 4
        }
    ]
    shared_data = creator.put_message(data)

work(creator, shared_data)
p.join()

Performance Benchmark results

In the realm of performance, Cengal's shared list container, although not yet fully optimized, is already outpacing the performance of multiprocessing.shared_memory.ShareableList. What sets it apart is its comprehensive support for native methods and operators, including Addition Assignment (shared_list[15] += 999), a feature unavailable in multiprocessing.shared_memory.ShareableList.

Cengal's shared list container demonstrates remarkable speed, boasting the ability to handle over 30,000,000 reads/writes per second for an int64 value (shared_list[2] = 1234 / val = shared_list[7]), or more than 1,450,000 addition assignments per second (shared_list[15] += 999). These performance figures underscore the efficiency and versatility of Cengal's interprocess shared memory solution, even in its current state.

Benchmark Results

Roadmap

• Continuosly moving more logic to Cython
• Implement mutable dict and set using an appropricate C hashmap library or C++ code (depending what will be faster in our case)
• Increase number of interacting processes from 2 to variable value
• Implement garbage collector for shared data in addition to manual free() call
• Implement an appropriate Service for cengal.parallel_execution.coroutines - for comfortable shared memory usage inside an async code (including asyncio)
• Improve memory allocation algorithm in an attempt of making it faster

Async LMDB database API

An example of usage (unit test of the module):

• test__db.py

Async logging into LMDB database

Developer can observe their logs in runtime using cengal.parallel_execution.coroutines.coro_tools.loop_administration.admin_tk module (made with Async Tkinter GUI):

• admin_tk.py

An example of usage of the admin_tk:

• admin_test.py

Alternatively, developer can load logs in off-line mode using Log Viewer application (made with async Tkinter GUI):

• log_viewer.py

Async Tkinter and Customtkinter

• tkinter_0.py
• customtkinter_0.py

Async wxPython

• async_wxpython_example.py

Async QT (PySide, PySide2, PySide6, PyQt4, PyQt5, PyQt6)

• pyside6__coro_slot_example_0.py

Async PyTermGUI

• hello_world_app_autoexit.py
• hello_world.py

Transparent background for your desktop applications (TBA)

• Target OS: Windows 11, Windows 10, Windows 8, Windows 7, Windows Vista.
• Target frameworks: PySide, PyQt, Kivy, PyWebView

,

Tkinter True Borderless apps for Windows platform (TBA)

• Target OS: Windows 11, Windows 10, Windows 8, Windows 7, Windows Vista.
• Target frameworks: CustomTkinter, Tkinter, ttkbootstrap, ...

Cengal Coroutines and Asyncio Administration and Monitoring Page

Observe loop performance, services state and coroutines list with details. Use an async interactive console in order to interact with your application from inside.

YouTube Showcase

Examples

admin_test.py

Modules with unique functionality

• "parallel_execution"
  • "coroutines" - asynchronous loop with almost preemptive multitasking within the single thread. Brings an async approach to an unmodified Tkinter, Qt, Kivy, etc. Unlike asyncio/trio/curio, it uses microkernel (services-based) approach which makes it highly- and easily-expandable. Can be executed both independently (asyncio/uvloop loop will be injected within the Cengal-coroutine when needed) and within already executed asyncio/uvloop loop. Can be used from the PyScript for the Web app creation.
    • "coro_standard_services" - set of standard services. You can replace standard service by yours for your app or for third-party module without code changes: by registering your own alias.
      • "loop_yield" - automatically kinda yield from your loops from time to time (priority based). Can be used to make a proper coroutine (which will not hangs on its endless loops) even from the long-running CPU-hungry third-party function (by function's bytecode modification made in runtime).
      • "tkinter" - make your Tkninter app async easily. Run any number of asynchronous Tkinter apps in single thread.
      • "db" - async wrapper around LMDB which provides an appropriate async API
      • "asyncio_loop" - use asyncio-based code directly from your async Cengal-coroutine neither Trio nor Curio able to to do this
      • "wait_coro" - 'put_atomic' request is an analogue of Trio's Nurseries for list of one or more coroutines; 'put_fastest' - returns when at least N of coroutines from the list were done successfully; etc.
      • "read_write_locker" - sync primitive usefull for DB creation (was made for a TagDB)
      • "remote_nodes" - in progress - connect to any opened listener/port of the node (TCP/UDP/Unix_Socket - doesn't matter), and identify your receiver by name (defined once - during the connection creation process). Uses improved version of the asyncio.streams as a backend in order to have a back pressure and an improved performance (see "efficient_streams" module description below).
    • "coro_tools" - tools
      • "await_coro" - await Cengal-coroutine or await for a call to the Cengal-service from your asyncio code
      • "low_latency" - use standard json module from your coroutines without hangs on huge Json-data (which usually hung even fast json implementation like orjson)
    • "integrations" -
      • "Qt" - wrapper around an unmodified Qt (supports: PySide, PySide2, PySide6, PyQt4, PyQt5, PyQt6). Adds asynchronous behavior to Slots. Doesn't require total reimplementation of your Qt app unlike other suggestions and competitors.
      • "customtkinter" - wrapper around an unmodified customtkinter. Implements an additional call, Customtkinter async apps needs to be executed for a proper work
      • "nicegui" - wrapper around an unmodified NiceGUI. Execute nicegui instance from within your code (administrative page for example). Build your pages in an asynchronous way in order to improve your server latency (NiceGUI makes it in a sync way).
      • "uvicorn" - wrapper around an unmodified uvicorn. Run uvicorn as a usual asyncio coroutine.
      • "uvloop" - an easy-install for a uvloop (if awailable).
      • "PyTermGUI" - wrapper around an unmodified PyTermGUI. Adds asynchronous behavior. No competitors currently.
  • "asyncio" - tools for an asyncio
    • "efficient_streams" - more efficient remake of an asyncion.streams. Better awailable traffic limits utilisation. Less kerner-calls number. Back pressure. Unlike asyncio, UDP version is planned but is not ready yet.
• "code_flow_control" -
  • "python_bytecode_manipulator" - modify your or third-party Python function's code in runtime easily
  • "chained_flow" - easy to use monad. Execute your your code if all/none/some of steps were completed withot an exceptions. Use all/none/some resutls of your steps at the final part of monad execution.
  • "multiinterface_essence" - Make your model and add different interfaces to it easily. Can be used for example in games: create "chair", "ball", "person" models and add to them your library of general interfaces like "touch", "push", "sit", "shot", "burn", "wet", etc.
• "hardware" - hardware related
  • "memory" - RAM related
    • "barriers" - fast full memory barriers for both x86/x64 and ARM (Windows, Linux, OS X, iOS, Android).
• "time_management" -
  • "high_precision_sync_sleep" - provides an ability to put your thread into legetimate sleep for at least 10x smaller time period than time.sleep() from the Python's Standard Library able to do on same Operating System: uses nanosleep() on Linux and periodic SwitchToThread() on Windows.
  • "cpu_clock_cycles" - Returnes value of RDTSCP on x86/x86_64 or CNTVCT_EL0 on ARM. Fast implementation: 6-12 times faster than all other competitors on Github. Note: CPU Time Stamp Counter (TSC) is not depends on actual current CPU frequency in modern CPUs (starting from around year 2007) so can be safely used as a high precision clock (see time_management.cpu_clock module). Windows, Linux and other Operating Systems are using it internaly.
  • "cpu_clock" - like perf_counter() but 25% faster. Supports both x86/x86_64 and ARM. cpu_clock is slightly faster than cpu_clock_cycles because double (float in Python terms) transfered from C-code to Python code more efficiently than 64-bit int (which needs an addition internal logic inside the Python itself for conversion). Highest-precision possible since it is CPU Time Stamp Counter based which is not depends on actual current CPU frequency in modern CPUs (starting from around year 2007) so can be safely used as a high precision clock (and Windows, Linux and other Operating Systems are using it internaly in this way). Benchmark: cpu_clock_test.py

Some Other modules

• "parallel_execution"
  • "coroutines" -
    • "coro_tools" - tools
      • "wait_coro" - decorate your coroutine in order to be able to execute it from the plain sunc code as a sync function
      • "run_in_loop" - serves the same purpose as an asyncio.run() call
      • "prepare_loop" - creates and returns loop. You may use it later
  • "asyncio" - tools for an asyncio
    • "run_loop" - similar to asyncio.run() but ends only when all background tasks will be finished (main coro can be finished long before this moment).
    • "timed_yield" - simple (dum-dum but faster) version of the "loop_yield" (see above) but made directly for an asyncio.
• "bulk_pip_actions" - install lists of required modules. Lists can be different for a different operating systems
• "code_inspection" -
  • "auto_line_tracer" - smart and easy to use line logger (current func name, file, lines numbers, surrounding code)
  • "line_tracer" - - easy to use line logger (current func name, file, line number)
  • "line_profiling" - confinient work with a line_profiler if awailable
• "data_containers" - usefull data containers like stack, fast fifo, etc. Some of them are highly optimized for speed
• "data_manipulation" -
  • "conversion" -
    • "bit_cast_like" - similar to std::bit_cast from C++
    • "reinterpret_cast" - similar to reinterpret_cast from C++. You have a third-party object and you want to change its type (and behavior) in runtime.
  • "serialization" - automatically choose a fastest appropriate serializer for your type and structure of data (json, simplejson, ujson, ojson, msgpack, cbor, cbor2, marshal, pickle, cloudpickle, ...)
  • "tree_traversal" - both recrsive and nonrecursive tree traversal algorithms
• "ctypes_tools" - ctypes code and structures used by Cengal.
  • "tools" - ctypes tools usefull for your code
• "file_system" - normalized relative path, etc.
  • "app_fs_structure" - unified list of the default app directories (data, cache, temp, etc.) recommended by OS (Linux, Windows, Mac OS X) in a runtime for a given application name or a service name. Results are cached. Cache size can be modified in runtime.
• "hardware" - hardware related
  • "info" - hardware info
    • "cpu" - normalized results from cpuinfo extended with an info from psutil.
• "introspection" -
  • "inspect" - find out function parameters, entity owners list (method -> subclass -> class -> module), entitie's own properties (excluding parent's properties), etc.
  • "third_party" -
    • "ctypes" - provice an instance of ctypes Structure and take a dict with all internals of this structure. Good for inspecting/logging/printing values of a given structure with all values of all its substructures.
• "io" -
  • "used_ports" - database of known TCP/UDP ports. Updates from an appropriate Wikipedia page once per Cengal release but you can update if for your self anytime if you want to.
  • "serve_free_ports" - provide ports range with an interested port types set and receive number of the first open appropriate port on your machine within given port range.
  • "named_connections_manager" - base for the "remote_nodes" (see above) and similar entities
  • "net_io" - an experimental networking library with an expandable architecture. Has implemented modules for epoll and select.
• "math" -
  • "algebra" -
    • "fast_algorithms" - Fast inverse square root (the one from Quake III) implemented efficiently
  • "geometry" -
    • "ellipse" - ellipse related. Also several algorithms for precisely or efficiently compute an ellipse perimeter
    • "point" - numpy (if awailable) or python implementation of points (1D, 2D, 3D, nD)
    • "vector" - numpy (if awailable) or python algotithms on vectors (1D, 2D, 3D, nD). Implements CoordinateVectorNd, VectorNd, DirectedGraphNd
• "modules_management" - reload module, import drop-in-replacement module if an original is not awailable
• "statistics" -
  • "normal_distribution" - compute the normal distribution of your data. Booth count or use a formula. 99, 95, 68; standard_deviation: diff_mean, sqrt(variance), max_deviation, min_deviation.
• "text_processing" - text parsing, patching, detect BOM and encoding
• "time_management" -
  • "timer" - timer for any synchronous code
  • "sleep_tools" - sleep for a production code. Using usual sleep you may get not wat you want if you are not really into your target OS internals (Windows/Linux)
  • "repeat_for_a_time" - measures code/function executions per second. But it smart and eficiently repeats target code/function not N times but up to a T seconds. Results to a high precision measurements for even smallest and fastest pieces of code.
  • "relative_time" - time related module for a business purposes (calendars, payments, etc.)
• "unittest" -
  • "patcher" - set of context manager for monkey patching builtins or other entities
• "user_interface" -
  • "gui" -
    • "nt" -
      • "blur_behind" - Turn on Aero Glass backgrownd in Winndows 7, 10, 11 using documented or undocumented API (which one is awailable)
      • "dpi_awareness" - Turn on DPI awareness
• "web_tools" -
  • "detect_browsers_host_device_type" -
    • "by_http_user_agent" - detects is it mobile or tablet device by analizing its http user_agent string

Size of the Cengal library

At the moment of 19 Mar 2024:

Around 200 modules

-------------------------------------------------------------------------------
Language                     files          blank        comment           code
-------------------------------------------------------------------------------
Python                         751          23696          30083          77396
Cython                          10            727            472           1892
C                                2             39             26            163
C/C++ Header                     2             14             26             37
Go                               3             19             37             88
Nim                              2             14              6             36
-------------------------------------------------------------------------------
SUM:                           770          24509          30650          79612
-------------------------------------------------------------------------------

Counted with cloc util.

Examples

• Can be found in examples dir
• Each module has a development folder. Examples are usually placed there
• Some of old examples can be fined inside the tests dir.

Cengal.coroutines examples

• General idea, greenlet main Cengal.coro
• General idea, async main Cengal.coro
• Transparent interconnection between Cengal.coroutines and asyncio

Text processing example

Ensures and updates copyright (with dates) in each Cengal's source file

Build

pip install git+https://github.com/FI-Mihej/Cengal.git

Installation process requires compilation. So ensure that:

• GCC/Clang is installed in your Linux/WSL (sudo apt-get --yes install build-essential for Ubuntu. And ./before_install_on_wsl.sh for Ubuntu under WSL for UI like Tkinter or Qt if you are using some kind of XServer on your host Windows)
• At least Visual Studio Community - Build Tools are installed on your Windows and you are installing Cengal from within its Developer Command Prompt for an appropriate target CPU architecture (x64 Native Tools Command Prompt for VS 2022 for example). Make sure that you have compatible version of Visual Studio for your target CPython interpreter (see python -VV command output. For example Python 3.9.11 (tags/v3.9.11:2de452f, Mar 16 2022, 14:33:45) [MSC v.1929 64 bit (AMD64)]: this python interpreter requires Visual Studio 2019 version 16.11.2+ according to 1929 word search in Wikipedia page)

Projects using Cengal

• flet_async - wrapper which makes Flet async and brings booth Cengal.coroutines and asyncio to Flet (Flutter based UI)
• justpy_containers - wrapper around JustPy in order to bring more security and more production-needed features to JustPy (VueJS based UI)
• Bensbach - decompiler from Unreal Engine 3 bytecode to a Lisp-like script and compiler back to Unreal Engine 3 bytecode. Made for a game modding purposes
• Realistic-Damage-Model-mod-for-Long-War - Mod for both the original XCOM:EW and the mod Long War. Was made with a Bensbach, which was made with Cengal
• SmartCATaloguer.com - TagDB based catalog of images (tags), music albums (genre tags) and apps (categories)

License

Copyright © 2012-2024 ButenkoMS. All rights reserved.

Licensed under the Apache License, Version 2.0.