A Hypothesis strategy for generating NetworkX graphs

## Project description

# Hypothesis-networkx

This module provides a Hypothesis strategy for generating networkx graphs.
This can be used to efficiently and thoroughly test your code.

## Installation

This module can be installed via `pip`:
```
pip install hypothesis-networkx
```

## User guide

The module exposes a single function: `graph_builder`. This function is a
hypothesis composite strategy for building graphs. You can use it as follows:

```python3
from hypothesis_networkx import graph_builder
from hypothesis import strategies as st
import networkx as nx

node_data = st.fixed_dictionaries({'name': st.text(),
'number': st.integers()})
edge_data = st.fixed_dictionaries({'weight': st.floats(allow_nan=False,
allow_infinity=False)})

builder = graph_builder(graph_type=nx.Graph,
node_keys=st.integers(),
node_data=node_data,
edge_data=edge_data,
min_nodes=2, max_nodes=10,
min_edges=1, max_edges=None,
self_loops=False,
connected=True)

graph = builder.example()
print(graph.nodes(data=True))
print(graph.edges(data=True))
```

Of course this builder is a valid hypothesis strategy, and using it to just
make examples is not super usefull. Instead, you can (and should) use it in

```python3
from hypothesis import given

@given(graph=builder)
def test_my_function(graph):
assert my_function(graph) == known_function(graph)

```

The meaning of the arguments given to `graph_builder` are pretty
self-explanatory, but they *must* be given as keyword arguments. Of particular
note are the following arguments:

- `graph_type`: This function (or class) will be called to create an empty
initial graph.
- `connected`: If True, the generated graph is garuanteed to be a single
connected component.
- `self_loops`: If False, there will be no self-loops in the generated graph.
Self-loops are edges between a node and itself.

## Known limitations

There are a few (minor) outstanding issues with this module:

- Graph generation may be slow for large graphs.
- The `min_edges` argument is not always respected when the produced graph
is too small.
- It currently works for Python 2.7, but this is considered deprecated and
may stop working without notice.

[Networkx](https://networkx.github.io/documentation/stable/index.html)