Skip to main content

Overlays for Python

## Project description

------------
Introduction
------------

This package provides a form of layer management for Python.
It transparently substitutes an existing module with ones patch thereof.
This normalizes or standardizes the original modules across ones code base.

ApeMan intercepts ones import calls to substitute the desired module with a patched variant provided in an overlay.
An overlay\ [#gentoo]_ is simply a python package containing ones patches for other packages.
Additionally the overlays :file:__init__.py file must invoke ApeMan.

Overlays make ones patches available across multiple projects; consistently exposing the additional API features provided by them.
Similarly a set of patches may quickly be substituted for another by simply importing a different overlay.

This formalizes monkey patching where the Ape in question has an affection for books, dislikes their readers and discourages, quite aggressively one might say, the use of the m... word\ [#librarian]_.

.. rubric:: Footnotes

.. [#gentoo] The term overlay is taken from Portage the package manager for Gentoo Linux.
.. [#librarian] Someone out there was about to find out their worst nightmare was a maddened librarian. With a Badge.

.. Suppose...
.. ----------
..
.. One is writing a script that imports a :class:CLASS from a :mod:MODULE in ones Python installation,
.. ::
..
.. from MODULE import CLASS
..
.. print(CLASS())
..
.. and one desperately wished that the class had a certain feature, say a nicer string representation.
.. One implements the following
.. ::
.. from MODULE import CLASS
..
.. class CLASS(CLASS) :
.. def __str__(self):
.. return "Nicer {} representation ".format(str(self))
..
.. while in their original script they would now import the patch
.. ::
..
.. from .MODULE import CLASS
..
.. print(CLASS())
..
.. This proves so useful that one wishes to make their patched implementation of :class:CLASS available to all of their projects.
.. ApeMan allows one to package their patch into an overlay so that
.. ::
..
.. import OVERLAY
.. from MODULE import CLASS
..
.. print(CLASS())

-------
Problem
-------

Occasionally one wants to patch the functions and/or classes provided by some other package; when it lacks features or to normalize the provided |API|.
A naive implementation relying upon the following structure;
::

PROJECT/ # The root folder for ones project
PACKAGE/ # The root folder of ones package.
PATCH.py # The module containing ones patches.
... # The other modules in the package.
__main__.py # The main script importing and using the patched module.

would patch the features from the SOURCE module by importing and overloading it's FUNCTION\ s and CLASS\ es.
::

from SOURCE import *

_FUNCTION_ = FUNCTION
def FUNCTION(*args, **kvps):
...
_FUNCTION_(*args, **kvps)
...

class CLASS(CLASS):
...

Ones modules would then the PATCH in favour of the original package; pulling in the modifications.
::

from .PATCH import *
...

This works well for once off patches in standalone projects.

Now, should a particular patch be especially useful, one might wish to use it across multiple projects.
At this point, one might copy the patch across to the other project(s) creating copies; copies that diverge from one another over time as features are added.

Should the original patch grow over time may become necessary to duplicate more of the structure of the original package; creating more files and exacerbating the problem.
Eventually one ends up with multiple PATCH files, spread across various projects, whose contents deviate further from one another to increasingly varied degrees.

--------
Solution
--------

ApeMan offers, a hopefully better strategy, that consistently manages these patches.
It resolves this by placing ones patches into an overlay; a package dedicated to ones patches.
This may be done locally, within a sub-package, for one shot usage; or globally, within a separate package, for repeated usage by multiple packages.

.. . If the following represents ones package structure
.. . ::
.. .
.. . PACKAGE/ # The root folder of ones package.
.. . overlay/ # The overlay containing the patches.
.. . SOURCE # The target package one is wrapping or patching.
.. . __init__.py # The __init__.py script importing ApeMan.
.. . __main__.py # The packages main script, executed when invoked as a module.

.. The packages main file makes it's usual calls to import the SOURCE modules but by importing the overlay first ApeMan redirects later imports to use ones patched modules instead.
.. ::
..
.. import overlay
.. from SOURCE import *
..
.. ...

Overlay Structure
=================

Whether it is made available globally or locally ones structures their overlay(s) as follows::

OVERLAY/ # The root folder of the ApeMan overlay
_PACKAGE_.py # The module containing ones patches, renamed after the source module or package
... # Further patches provided by the overlay
__init__.py # The file invoking ApeMan; identifying it as an overlay

The overlays' :file:__init__.py file then imports and registers the :class:ApeMan instance;
::

from apeman import ApeMan;
apeman = ApeMan()

which intercepts later imports, substituting ones patches for the original modules.

Local Overlay(s)
================

Locally one may create an overlay at any level within their package by including a sibling package along side it's modules.
::

PROJECT/ # The root folder for ones project
PACKAGE/ # The root folder of ones package.
OVERLAY/ # The ApeMan overlay
... # The contents of the overlay
... # The other modules in the package.
__main__.py # The main script importing and using the patched module.

Other modules within ones package may then invoke the overlay via relative import.
::

import .OVERLAY
from SOURCE import *

...

Global Overlay(s)
=================

Globally, an overlay, is provided as a separate, standalone package.
::

PROJECT/ # The root folder for ones project
OVERLAY/ # The root folder of the ApeMan Overlay
... # The contents of the overlay
PACKAGE/ # The root folder of ones package.
... # The other modules in the package.
__main__.py # The main script importing and using the patched module.

In this case the modules in ones package must invoke the overlay using an absolute import.
::

import OVERLAY
from SOURCE import *

...

.. One must explicitly import the features they need as the OverlayImporter actually blocks further imports.

.. Note that an overlay package is meant to reside alongside its sibling module to afford the most flexibility.
.. Whether or not this is possible at every level within a package depends upon how python enforces scoping.

-------
Example
-------

Consider patching the :class:Decimal class from the :mod:decimal module.

Monkey Patching
===============

The following structure is the simplest to implement.
::

PACKAGE/ # The root folder of ones package.
_decimal_.py # The module containing ones patches to the decimal module.
__main__.py # The packages main script, executed when invoked as a module.

Within :file:_decimal_.py import everything from the :mod:decimal module then subclass and monkey patch the Decimal class; modifying it's string representation.
::

from decimal import *

class Decimal(Decimal):
def __str__(self) :
return super().__str__().split("'")[1]

Then within the :file:__main__.py file one would import and use the patch as follows::

import ._decimal_ as decimal
from decimal import Decimal

print(Decimal(42))

This should output 42 instead of Decimal('42') when we invoke the package using :code:python -m PACKAGE.

Ape Patching
============

Using ApeMan we would move the _decimal_.py file into a sub-folder called overlay, with the resulting structure;
::

PACKAGE/ # The root folder of ones package.
overlay/ # The overlay containing the patches.
_decimal_.py # The module containing ones patches to the decimal module.
__init__.py # The __init__.py script importing ApeMan.
__main__.py # The packages main script, executed when invoked as a module.

accordingly the :file:__init__.py file should contain ::

from apeman import ApeMan
apeman = ApeMan

The main file is then adapted to reflect the following.
::

import .overlay
from decimal import Decimal

print(Decimal(42))

Without ...
===========

One might argue that a cleaner structure still, is as follows
::

PACKAGE/ # The root folder of ones package.
decimal.py # The module containing ones patches to the decimal module.
__main__.py # The packages main script, executed when invoked as a module.

but this results in a whole series of clashes and the following error
::

AttributeError: 'module' object has no attribute 'Decimal'

.. Other related errors include :
.. SystemError: Parent module '' not loaded, cannot perform relative import

.. Essentially the :file:decimal.py module gets installed within the decimal name space preventing the import of the original library.

Essentially the :file:PACKAGE/decimal.py file gets loaded as the :mod:decimal module and is assigned under :attr:sys.modules reserving the decimal key; preventing the subsequent import of the actual :mod:decimal module.

.. note ::

This method actually works if one tells python it's executing a module using the -m switch, :code:python -m PACKAGE, but only I found this out after creating the package.

-------------
Compatability
-------------

The machinery underlying :meth:import has undergone some radical changes over the lastest releases of Python (Particularly versions 3.3-3.5 and next in 3.7).
In light of this ApeMan aims to support a minimal set of features; namely explicit and implicit overlays providing patches whose structure matches their intended source packages.
Any functionality offered beyond these base features is considered sugar e.g. repeated imports, stacked overlays, restructured patches and substructured patches;

.. tested by the :mod:*Explcit :mod:*Implcit and :mod:*Init tests (See :ref:Testing).

.. table :: Set of features supported by the Python import system in the different Python implementations
:align: center
:widths: auto

====================== === === === ===
Python 2.7 3.3 3.4 3.7
====================== === === === ===
explicit packages X X X X
implicit packages X X X
lazy loading X
C implementation X X
Python implementation X X
====================== === === === ===

It should be noted that Apeman has only been developed and tested in Python 2.7, 3.4 and 3.6.
The Python 3.4 implementation was last tested when the author still had it installed, before switching to 3.6.
The author also maintains a flaky build of Python 3.5 but this is not a good testing envvironment as a result ApeMan in 3.5 is flaky.
Generally speaking if you're using ApeMan in anything other then Python 2.7 or 3.6 you're on your own.

.. table :: Set of features supported ApeMan under the different Python implementations
:align: center
:widths: 4,1,1,1,1

======================= === === === ===
Python 2.7 3.4 3.5 3.6
======================= === === === ===
explicit packages X X ? X
implicit packages N/A X ? X
repeated imports ? ? ? X
Substructured overlays
Restructured overlays
Stacking overlays
======================= === === === ===

.. There are tworules for success ...
.. 1) Never reveal everything that you know

-a command line option
-b path command line option

Having said that the 2.7 implementation ought to work in Python 2.7-3.3.
Python 3.4 saw a big overhaul from 3.3 but one did develop ApeMan against this version and, unless one is grossly mistaken, the implementation should still work.
Python 3.5 included a few leftovers that were forgotten for Python 3.4.
The 3.4 implementation ought to work in 3.5 but the current 3.5 implementation diverges from the one for 3.4 and appears broken upon the authors machine.
The Python 3.6 implementation is presently the most tested and stable while Python 3.7 has not been attempted just yet.

------------------
Live and Let Die !
------------------

This is largely inspired by Portage, the package manager for Gentoo Linux and the tutorial by David Beasley.
However it is only possible through the contributions of Brett Cannon, who ported the Python import machinery from C/C++ to Python.

In general a big thank you is also made to the developers of Python and all the other third party packages that come withit.

.. .. tikz:: Title
.. :libs: calc
..
.. \draw (0,0) circle (3em) circle (4em) circle (5em);

---------
Licencing
---------

This software is licenced under a GPL v3 licence.
One requests that anyone hosting a fork of this code inform the author accordingly so that any useful modifications one has made may periodically be merged into the code base.

## Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

### Source Distribution

ApeMan-0.1.1.tar.gz (897.2 kB view hashes)

Uploaded source