HardMediaGroup, CC BY-SA 3.0, via Wikimedia Commons

Pyrustic Hooking

Generic dual-paradigm hooking mechanism

This project is part of the Pyrustic Open Ecosystem.

Modules documentation

Overview
Examples
Functions and methods as targets
Anatomy of a hook
Tight coupling
- Override a target
- Wrap a target
Loose coupling
Miscellaneous
Installation

Overview

This project is a minimalist Python library that implements an intuitive, flexible, and generic dual-paradigm hooking mechanism.

In short, methods and functions, called targets, are decorated and assigned user-defined hooks. So when a target is called, the assigned hooks will be automatically executed upstream or downstream according to the specification provided by the programmer.

The programmer may wish to have tight or loose coupling between targets and hooks, depending on the requirement. Hence, for a nice developer experience, this library provides two interfaces each representing a paradigm (tight or loose coupling) to cover the needs.

From a hook, the programmer has access to the keyword arguments passed to the decorator, the arguments passed to the target, the target itself, and other useful information through a Context object.

Depending on whether the hook is executed upstream or downstream, the programmer can modify arguments to target, override the target itself with an arbitrary callable, break the execution of the chain of hooks, modify the return of the target, et cetera.

Why use this library

This library allows the programmer to wrap, augment, or override a function or method with either a tight or loose coupling. It is therefore the perfect solution to create a plugin mechanism for a project.

It can also be used for debugging, benchmarking, event-driven programming, implementing routing in a web development framework (Flask uses decorators to implement routing), et cetera.

The interface of this library is designed to be intuitive to use not only for crafting an API but also for consuming it.

Examples

Here are some examples of using this library in the tight and loose coupling paradigm.

Measure execution time

Here we will use the tight coupling paradigm to override the target function with a hook. From inside the hook, the target will be executed and we will measure the execution time.

The following example can be copy-pasted into a file (e.g. test.py) and run as is:

import time
from hooking import override


def timeit(context, *args, **kwargs):
    # execute and measure the target run time
    start_time = time.perf_counter()
    context.result = context.target(*args, **kwargs)
    total_time = time.perf_counter() - start_time
    # print elapsed time
    text = context.config.get("text")  # get 'text' from config data
    print(text.format(total=total_time))


# decorate 'heavy_computation' with 'override' (tight coupling)
# here, 'timeit' is the hook to execute when the target is called
# all following keyword arguments are part of the configuration data
@override(timeit, text="Done in {total:.3f} seconds !")
def heavy_computation(a, b):
    time.sleep(2)  # doing some heavy computation !
    return a*b


if __name__ == "__main__":
    # run 'heavy_computation'
    result = heavy_computation(6, 9)
    print("Result:", result)

$ python -m test
Done in 2.001 seconds !
Result: 54

Routing by a fictional web framework

This example is divided into two parts:

the server-side Python script;
and the internals of the web framework.

Server-side Python script

# Script for the server-side (API consumer side)
# web_script.py
from my_web_framework import Routing, start

# bind to 'home_view', three tags representing
# three paths. This view will be executed when
# the user will request the home page for example
@Routing.tag("/")
@Routing.tag("/home")
@Routing.tag("/index")
def home_view():
    return "Welcome !"


@Routing.tag("/about")
def about_view():
    return "About..."


if __name__ == "__main__":
    start()

Web framework internals

# Framework internals (API creator side)
# my_web_framework.py
import random
from hooking import H

# implementing custom routing mechanism by subclassing hooking.H
Routing = H.subclass("Routing")

def start():
    # entry point of the web framework
    # Get user request, then serve the appropriate page
    path = get_user_request()
    serve_page(path)

def get_user_request():
    # use randomness to simulate user page request
    paths = ("/", "/home", "/index", "/about")
    return random.choice(paths)

def serve_page(path):
    # get the list of functions and methods
    # tagged with the Routing.tag decorator
    cache = Routing.targets.get(path, list())
    for target_info in cache:
        view = target_info.target
        html = view()
        render_html(html)

def render_html(html):
    print(html)

Functions and methods as targets

As mentioned in the Overview section, functions and methods are the targets to which hooks are attached with a tight or loose coupling. A hook is a function that can be attached one or more times to one or more targets.

Static methods, class methods, or decorated methods or functions should work fine with this library as long as one comes as close as possible to the native definition of the function or method. Example:

from hooking import H, on_enter, on_leave

class MyClass:
    # Good !
    @staticmethod
    @H.tag  # innermost
    def do_something1(arg):
        pass

    # BAD !!!
    @H.tag  # outermost
    @classmethod
    def do_something2(cls, arg):
        pass

def my_hook(context, *arg, **kwargs):
    pass

# Good !
@ExoticDecorator
@on_enter(my_hook)  # innermost
def my_func():
    pass

# BAD !!!
@on_leave(my_hook)  # outermost
@ExoticDecorator
def my_func():
    pass

Anatomy of a hook

A hook is a callable that accepts an instance of hooking.Context and arguments passed to the target.

The hooking.Context instance exposes the following attributes:

cls: the hook class;
hid: the Hook ID (HID) as returned by the class methods H.wrap, H.on_enter, and H.on_leave;
tag: label string used to tag a function or method;
config: dictionary representing keyword arguments passed to the decorator;
spec: either the hooking.ENTER constant or the hooking.LEAVE constant;
target: the decorated function or method;
args: tuple representing the positional arguments passed to the target;
kwargs: dictionary representing the keyword arguments passed to the target;
result: depending on the context, this attribute may contain the value returned by the target;
shared: ordered dictionary to store shared data between hooks (from upstream to downstream).

The attributes listed above can be updated with the Context.update method that accepts keyword arguments.

from hooking import H

# defining my_hook
def my_hook(context, *args, **kwargs):
    if context.tag != "target":
        raise Exception("Wrong tag !")
    # reset arguments
    context.update(args=tuple(), kwargs=dict())

@H.tag("target")  # tagging my_func with "target"
def my_func(*args, **kwargs):
    pass

# binding my_hook to the tag "target"
H.on_enter("target", my_hook)

Modify arguments to target

From an upstream hook, we can change the arguments passed to a target:

# ...

def upstream_hook(context, *args, **kwargs):
    # positional arguments are represented with a tuple
    context.args = (val1, val2)
    # keywords arguments are represented with a dictionary
    context.kwargs = {"item1": val1, "item2": val2}
 
# ...

Override the target from a hook

The library exposes the hooking.override to override a target function:

from hooking import override

def myhook(context, *args, **kwargs):
    context.result = new_target_function(*args, **kwargs)

@override(myhook) 
def target():
    pass

but one can still override the target from an arbitrary upstream hook:

from hooking import on_enter

def upstream_hook(context, *args, **kwargs):
    # override target with a new target that
    # that accepts same type signature
    context.target = new_target_function
 
@on_enter(upstream_hook) 
def target():
    pass

Note that you can set None to context.target to prevent the library for automatically running the target between the execution of upstream and downstream hooks.

Modify the return of a target

From a downstream hook, we can change the return of a target:

# ...

def downstream_hook(context, *args, **kwargs):
    context.result = new_value
 
# ...

Tight coupling

In this paradigm, hooks are directly bound to target. The library exposes the following decorators to decorate targets:

Decorator	Description	Signature
`hooking.override`	Bind to a target a hook that will override it	`@override(hook, **config)`
`hooking.wrap`	Bind to a target two hooks that will be executed upstream and downstream	`@wrap(hook1, hook2, **config)`
`hooking.on_enter`	Bind to a target a hook that will be executed upstream, i.e, before the target	`@on_enter(hook, **config)`
`hooking.on_leave`	Bind to a target a hook that will be executed downstream, i.e, after the target	`@on_leave(hook, **config)`

Override a target

The library exposes the hooking.override decorator to bind to a target a hook that will override it:

from hooking import override

def myhook(context, *args, **kwargs):
    context.result = context.target(*args, **kwargs)
    # or
    my_new_target = lambda *args, **kwargs: print("New Target Here !")
    context.result = my_new_target(*args, **kwargs)

# override target with myhook
@override(hook3, foo=42, bar="Alex")  # foo and bar are config data
def target():
    pass

Note that with the hooking.override decorator, the programmer must execute the target or its replacement inside the hook and set the result to context.result.

Wrap a target

The hooking.wrap decorator allows the programmer to wrap the target between two hooks that will be executed upstream and downstream. Two additional decorators hooking.on_enter and hooking.on_leave allow the programmer to bind either a upstream or a downstream hook to a target.

from hooking import wrap, on_enter, on_leave

def hook1(context, *args, **kwargs):
    pass

def hook2(context, *args, **kwargs):
    pass

# bind an upstream hook and a downstream hook to my_func1
@wrap(hook1, hook2, foo=42, bar="Alex")  # foo and bar are config data
def my_func1():
    pass

# bind an upstream hook to my_func2
@on_enter(hook1, foo=42, bar="Alex")
def my_func2():
    pass

# bind a downstream hook to my_func3
@on_leave(hook2, foo=42, bar="Alex")
def my_func3():
    pass

Loose coupling

In this paradigm, hooks aren't directly bound to target but to tags which are linked to targets. The library exposes the hooking.H class to support the loose coupling paradigm. In short, the hooking.H.tag decorator is used to tag targets, then class methods hooking.H.on_enter, hooking.H.on_leave, and hooking.H.wrap are used to bind hooks to tags.

Tagging mechanism

The hooking.H.tag class method allows you to tag a function or a method:

from hooking import H

@H.tag
def my_func(*args, **kwargs):
    pass

class MyClass:
    @H.tag
    def my_method(self, *args, **kwargs):
        pass

The hooking.H.tag decorator accepts a label string as argument. By default, when this argument isn't provided, the library uses the qualified name of the method or function as the label.

Here we provide the label argument:

from hooking import H

@H.tag("my_func")
def my_func(*args, **kwargs):
    pass

class MyClass:
    @H.tag("MyClass.my_method")
    def my_method(self, *args, **kwargs):
        pass

Bind hooks to tags

These are the hooking.H class methods to bind hooks to tags:

Class method	Description	Signature
`hooking.H.on_enter`	Bind to a tag a hook that will be executed downstream, i.e, before the target	`on_enter(tag, hook)`
`hooking.H.on_leave`	Bind to a tag a hook that will be executed downstream, i.e, after the target	`on_leave(tag, hook)`
`hooking.H.wrap`	Bind to a tag two hooks that will be executed upstream and downstream	`wrap(tag, hook1, hook2)`

from hooking import H

@H.tag("target")
def my_func(*args, **kwargs):
    pass

def my_hook1(context, *args, **kwargs):
    pass

def my_hook2(context, *args, **kwargs):
    pass

# bind my_hook1 to "target" and run it upstream
hook_id = H.on_enter("target", my_hook1)

# bind my_hook2 to "target" and run it downstream
hook_id = H.on_leave("target", my_hook2)

# bind my_hook1 and my_hook2 to "target"
hook_id1, hook_id2 = H.wrap("target", my_hook1, my_hook2)

Unbind hooks

Whenever a hook is bound to a tag, the Hook ID (HID) which could be used later to unbind the hook, is returned:

from hooking import H

def hook(context, *args, **kwargs):
    pass

# bind
hid = H.on_enter("tag", hook)

# unbind
H.unbind(hid)

Multiple hooks can be unbound in a single statement:

from hooking import H

def hook1(context, *args, **kwargs):
    pass

def hook2(context, *args, **kwargs):
    pass

# bind
hid1 = H.on_enter("tag", hook1)
hid2 = H.on_leave("tag", hook2)

# unbind multiple hooks manually
H.unbind(hid1, hid2)

Clear hooks bound to a specific tag

The clear class method of hooking.H unbinds all hooks bound to a specific tag:

from hooking import H

def hook1(context, *args, **kwargs):
    pass

def hook2(context, *args, **kwargs):
    pass

@H.tag
def target():
    pass

# bind
hid1 = H.on_enter("target", hook1)
hid2 = H.on_enter("target", hook2)

# unbind hook1 and hook2 from "target"
H.clear("target")

You can clear multiple tags in a single statement:

from hooking import H

def hook1(context, *args, **kwargs):
    pass

def hook2(context, *args, **kwargs):
    pass

@H.tag
def target1():
    pass

@H.tag
def target2():
    pass

# bind
hid1 = H.on_enter("target1", hook1)
hid2 = H.on_enter("target2", hook2)

# unbind hook1 and hook2 from "target1" and "target2"
H.clear("target1", "target2")

Chain break

This library exposes an Exception subclass to allow the programmer to break the execution of a chain of hooks:

from hooking import H, ChainBreak

@H.tag("target")
def my_func(*args, **kwargs):
    pass
    
def hook1(context, *args, **kwargs):
    pass

def hook2(context, *args, **kwargs):
    raise ChainBreak

def hook3(context, *args, **kwargs):
    pass


# bind hook1, hook2 and hook3 to 'target'
for hook in (hook1, hook2, hook3):
    H.on_enter("target", hook)

# call the target
my_func()

# since the target was called,
# the chain of hooks (hook1, hook2, hook3)
# must be executed.

# hook2 having used ChainBreak,
# the chain of execution will be broken
# and hook3 will be ignored

Freeze the hooking class

We could freeze the hooking class and thus prevent the execution of all hooks:

from hooking import H

@H.tag
def my_func(*args, **kwargs):
    pass

H.freeze()

# from now, no hook will be executed anymore

Unfreeze the hooking class

To unfreeze the hooking class, use the H.unfreeze class method:

from hooking import H

H.unfreeze()

# from now, hooks will be executed when needed

Exposed data

The library exposes data through a class method, class variables, and data transfer objects (namedtuples).

Get the list of upstream and downstream hooks

Upstream and downstream hooks bound to a specific tag can be retrieved with the get_hooks class method.

from hooking import H

# returns a 2-tuple of upstream hooks and
# downstream hooks
upstream_hooks, downstream_hooks = H.get_hooks("tag")

# iterate through upstream_hooks which is
# a list of instances of hooking.HookInfo
for hook_info in upstream_hooks:
    print(hook_info)

Read-only class variables

The hooking.H class exposes the following class variables:

Class variable	Description
`targets`	Ordered dictionary. Keys are tags and values are lists of instances of `hooking.TargetInfo`. Example: {"tag1": [TargetInfo(), TargetInfo(), ...], ...}
`hooks`	Ordered dictionary. Keys are tags and values are lists of instances of `hooking.HookInfo`. Example: {"tag1": [HookInfo(), HookInfo(), ...], ...}
`tags`	The set of registered tags
`frozen`	Boolean to tell whether the hooking mechanism is frozen or not

Note: it is not recommended to modify the contents of these class variables directly. Use the appropriate class methods for this purpose.

Both hooking.TargetInfo and hoooking.HookInfo are namedtuples that will be explored in the next subsection.

Data transfer object

Here are the fields from the hooking.TargetInfo namedtuple:

cls: the hooking class;
tag: the string label that represents the tag;
target: the target method or function;
config: dictionary containing the configuration data passed to the decorator.

Here are the fields from the hooking.HookInfo namedtuple:

cls: the hooking class;
hid: the hook identifier;
hook: the callable representing the hook;
tag: the string label that represents the tag;
spec: either hooking.ENTER or hooking.LEAVE.

Reset the hooking class

You may need to reset the hooking class, i.e., reinitialize the contents of the following class variables: hooking.H.hooks, hooking.H.tags, and hooking.H.frozen. In this case, you just have to call the hooking.H.reset class method.

Note: targets won't be removed.

Subclassing the hooking class

This library is flexible enough to allow the programmer to create their own subclass of hooking.H like this:

from hooking import H

MyCustomHookingClass = H.subclass("MyCustomHookingClass")

Subclassing hooking.H allows the programmer to apply the separation of concerns. For example, a web framework creator might create a Routing subclass to implement a routing mechanism, and also create an Extension subclass to implement a plugin mechanism. Each subclass would have its own set of tags, hooks, and targets.

Note: class variables are automatically reset when subclassing hooking.H.

Miscellaneous

Multithreading

Whenever threads are introduced into a program, the state shared between threads becomes vulnerable to corruption. To avoid this situation, this library uses threading.Lock as a synchronization tool.

Installation

Hooking is cross-platform and should work on Python 3.5 or newer.

First time

$ pip install hooking

Upgrade

$ pip install hooking --upgrade --upgrade-strategy eager

Show package information

$ pip show hooking

pyrustic/hooking