Dependency Algorithm

Here is my take on an algorithm in Python 3.7 that resolves dependencies. The best way to illustrate how this works is with an example...

Example

Let's say that we have the following dictionary where the keys are items, and the values are the dependencies of those items. Each dependency must itself be an item, and items with no dependencies have an empty list [] as they dependency:

my_items = {
    'A': ['B', 'C', 'D'],  # -- A is dependent on B, C, D,
    'B': [],  # -- B is dependent on nothing, etc.
    'C': ['D'],
    'D': ['B', 'E'],
    'E': ['F'],
    'F': [],
    'Z': ['A', 'B', 'C', 'D']
}

Note that we've only provided a partial dictionary of items to their dependencies...for example, notice how C is dependent on D, which is dependent on B (no dependencies) and E, which is dependent on F...therefore, C is dependent on D, B, E, and F.

We can use the Dependencies class to get the complete list of dependencies for an item, like so:

from dependency_algorithm import Dependencies

# Creating a Dependencies object
dependencies = Dependencies(my_items)
dependencies.complete_dependencies("C")

>>> ['D', 'B', 'E', 'F']

More importantly, we can return the items in an order such that the dependencies resolve:

dependencies.resolve_dependencies()

>>> ['B', 'F', 'E', 'D', 'C', 'A', 'Z']

In many cases, there are multiple correct ordering of our items such that each item's dependencies resolve. If we're interested in all possible correct orderings, the Dependencies class can permutate over all possible orderings, and identify the correct ones (albeit at a high computational cost), like so:

dependencies.all_possible_resolution_orders(verbose=True)

>>> Number of permutations: 5040
>>> Number of correct orderings: 3
>>> Number of incorrect orderings: 5037
>>> [('B', 'F', 'E', 'D', 'C', 'A', 'Z'),
>>>  ('F', 'B', 'E', 'D', 'C', 'A', 'Z'),
>>>  ('F', 'E', 'B', 'D', 'C', 'A', 'Z')]

That's pretty much it! The Dependencies class also performs two checks, one for any dependencies that are "missing" (i.e., they are not keys in the input dictionary of items and dependencies), and another for cirular dependencies (i.e., A is dependent on B which is dependent on A which is...and so on...).

Installation

Requires Python 3.7 or greater.

pip install dependency_algorithm

Running the unit tests with pytest

git clone https://github.com/jakesherman/dependency_algorithm.git
cd dependency_algorithm
pip install -e .
python -m pytest

Future work

• New version of Dependencies._enhanced_list_dependencies that uses iteration instead of recursion
• Improved version of Dependencies.all_possible_resolution_orders that uses a more efficient algorithm than looping through permutations, ex. a recursive algorithm