py2neo-compat 1.0.1

Description

A (probably vain) attempt to create a cross-version compatibility layer for py2neo.

Purpose

Every major release of py2neo has radically (and gratuitously, it seems) changed the API: package/module organization, name of the database connection class, facilities for working with nodes & relationships – not to mention more complicated features like interacting with schema.

What is implemented here is a hodgepodge of monkey-patches and other hacks to make things work between v1.6 and v2.

Background

At work I inherited a codebase developed around py2neo v1.6, and data access was intricately tied with business logic. Tests were scarce and what tests existed were system tests that did not address errors, boundary conditions, etc., so updating every use of py2neo for a new version would be very risky. (Not to mention that I was just learning py2neo & Neo4j.)

My initial plan was to upgrade to py2neo v2 (current at the time) incrementally by monkey-patching & renaming to make v1.6 able to be used as if it were v2 and when the code had all been changed to use v2, upgrade to py2neo v2. Along the way I had the opportunity to develop a new, stand-alone application, so I started out creating a new dist for the application and developing some libraries that could be shared with legacy system, so I created a separate dist for the shared libraries. I did not want to develop the new application with the old v1.6, so my shared library would need to support both v1.6 and v2. I also started a separate project called gryaml, to load sample data from YAML files, which also needed to support multiple versions in order to be run as test fixtures.

Features

• Implementation-independent names, optionally monkey-patched into the py2neo namespace and classes upon call to py2neo_compat.monkey_patch_py2neo:

  • Graph

    • Graph.create_unique (only single path)

    • Graph.cypher.stream

    • Graph.cypher.execute

    • Graph.delete_all

    • Graph.find_one

    • Graph.legacy.delete_index

    • Graph.legacy.get_*index*

    • Graph.resource

    • Graph.uri

  • Node

    • Node.labels property (get-only)

    • Node.pull method (aliased to refresh in v1)

    • Node.push method (aliased to refresh in v1)

  • Relationship

    • Relationship.push method (aliased to refresh in v1)

    • Relationship.pull method (aliased to refresh in v1)

  • Record

  • node

  • rel

  • ServerError

  • ClientError

  • URI

  • Resource

  • py2neo.legacy.LegacyWriteBatch

  • py2neo.legacy.Index

  • py2neo.batch.WriteBatch

• Custom wrapper functions, which are not monkey-patched:

  • py2neo_compat

    • graph_metadata

    • create_node

    • py2neo_entity_to_dict aka to_dict: Dump the properties of a node or relationship as a dict.

  • schema:

    • schema_constraints - Yields tuples of schema constraints for all constraints (not just ones in use).

    • schema_indexes - Yields tuples of schema indexes for all indexes (not just ones in use).

    • drop_schema

    • drop_constraints

    • drop_indexes

    • create_schema

• Known limitations:

  • py2neo v1.6 will not create labels with an abstract Node.

  • There is no common way to modify labels after node creation; v1 uses special methods to manipulate the labels, whereas v2 uses a a subclass of set. The easy way would be to add the v1 methods to v2, but I am trying to avoid propagating the v1 API; the harder way would be to implement a set-like class for v1, but considerably more work. Another alternative is a compatibility function, which ends up being another API. I have not really needed to manipulate labels after creation.

• Free software: Apache Software License 2.0

Future

For future development, I am considering several options:

• Attempt to do just enough to support v3 and ignore v4.

• Ignore v3 and reimplement the features I need on v4.

• Ignore v3 and v4 altogether and build what I need on v2 and the Bolt-only Python driver.

Approaches other than monkey-patching:

• Instead of monkey-patching, subclass or wrap the appropriate classes and re-implement the differing bits. Clients would import directly from the py2neo_compat package, rather than importing, calling the monkey-patch function and then importing from py2neo.

• Adopt a function-based approach, where the receiver is an explicit parameter instead of having methods. This can get redundant, since the type of the primary receiver ends up being embedded in the name.

This would might work better for Nodes and Relationships and the basic Graph features than the schema, batch and legacy index support.

Credits

This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.

1.0.1 (2018-08-02)

• Restrict to only supported py2neo versions with environment-based override for testing with later versions.

1.0.0 (2018-08-01)

• First release on PyPI.