# A module to make JSON in Python easier

I recently wrote the livejson module as a way to make working with JSON in Python easier. The idea of the module is that you initialize a livejson.File, and then you can interact with your JSON file with the same interface as a Python dict, except your changes are reflected in the file in realtime.

I'd like to know how I can improve this.

• Is my design good, in terms of things like naming, OOP design, etc.?
• Are there ways I could improve performance without decreasing reliability? I implemented the context manager with a concept of "grouped writes" specifically so that repetitive operations could achieve the performance of the native objects.

# The code

Here's the code. I've tried to document it well with comments and docstrings, and also tried to follow PEP8 so I hope it's readable:

Imports:

"""A module implementing a pseudo-dict class which is bound to a JSON file.

As you change the contents of the dict, the JSON file will be updated in
real-time. Magic.
"""

import collections
import os
import json


The code begins with some helper functions and generic base classes:

# MISC HELPERS

def _initfile(path, data="dict"):
"""Initialize an empty JSON file."""
data = {} if data.lower() == "dict" else []
# The file will need to be created if it doesn't exist
if not os.path.exists(path):  # The file doesn't exist
# Raise exception if the directory that should contain the file doesn't
# exist
dirname = os.path.dirname(path)
if dirname and not os.path.exists(dirname):
raise IOError(
("Could not initialize empty JSON file in non-existant "
"directory '{}'").format(os.path.dirname(path))
)
# Write an empty file there
with open(path, "w") as f:
json.dump(data, f)
return True
elif len(open(path, "r").read()) == 0:  # The file is empty
with open(path, "w") as f:
json.dump(data, f)
else:  # The file exists and contains content
return False

class _ObjectBase(object):
"""Class inherited by most things.

Implements the lowest common denominator for all emulating classes.
"""
def __getitem__(self, key):
out = self.data[key]

# Nesting
if isinstance(out, (list, dict)):
# If it's the top level, we can use [] for the path
pathInData = self.pathInData if hasattr(self, "pathInData") else []
newPathInData = pathInData + [key]
# The top level, i.e. the File class, not a nested class. If we're
# already the top level, just use self.
toplevel = self.base if hasattr(self, "base") else self
nestClass = _NestedList if isinstance(out, list) else _NestedDict
return nestClass(toplevel, newPathInData)
# Not a list or a dict, don't worry about it
else:
return out

def __len__(self):
return len(self.data)

# Methods not-required by the ABC

def __str__(self):
return str(self.data)

def __repr__(self):
return repr(self.data)

# MISC
def _checkType(self, key):
"""Make sure the type of a key is appropriate."""
pass


One of the major challenges I faced was easily supporting nested JSON structures, with the easy syntax of my_file["a"]["b"] = "c". It was difficult to make these cases behave properly because __setitem__ is not called in this case. To overcome this, I set up simple "nesting" classes to replace Python lists and dicts in nested situations. These mainly just listen for __setitem__ and tell the top-level class to update the file's contents.

# NESTING CLASSES

class _NestedBase(_ObjectBase):
"""Inherited by _NestedDict and _NestedList, implements methods common
between them. Takes arguments 'fileobj' which specifies the parent File
object, and 'pathToThis' which specifies where in the JSON file this object
exists (as a list).
"""
def __init__(self, fileobj, pathToThis):
self.pathInData = pathToThis
self.base = fileobj

@property
def data(self):
d = self.base.data
# Navigate through the object to find where self.pathInData points
for i in self.pathInData:
d = d[i]
# And return the result
return d

def __setitem__(self, key, value):
self._checkType(key)
# Store the whole data
data = self.base.data
# Iterate through and find the right part of the data
d = data
for i in self.pathInData:
d = d[i]
# It is passed by reference, so modifying the found object modifies
# the whole thing
d[key] = value
# Update the whole file with the modification
self.base.set_data(data)

def __delitem__(self, key):
# See __setitem__ for details on how this works
data = self.base.data
d = data
for i in self.pathInData:
d = d[i]
del d[key]
self.base.set_data(data)

class _NestedDict(_NestedBase, collections.MutableMapping):
"""A pseudo-dict class to replace vanilla dicts inside a livejson.File.

This "watches" for changes made to its content, then tells
the base livejson.File instance to update itself so that the file always

This class is what allows for nested calls like this

>>> f = livejson.File("myfile.json")
>>> f["a"]["b"]["c"] = "d"

to update the file.
"""
def __iter__(self):
return iter(self.data)

def _checkType(self, key):
if not isinstance(key, str):
raise TypeError("JSON only supports strings for keys, not '{}'. {}"
.format(type(key).__name__, "Try using a list for"
" storing numeric keys" if
isinstance(key, int) else ""))

class _NestedList(_NestedBase, collections.MutableSequence):
"""A pseudo-list class to replace vanilla lists inside a livejson.File.

This "watches" for changes made to its content, then tells
the base livejson.File instance to update itself so that the file always

This class is what allows for nested calls involving lists like this:

>>> f = livejson.File("myfile.json")
>>> f["a"].append("foo")

to update the file.
"""
def insert(self, index, value):
# See _NestedBase.__setitem__ for details on how this works
data = self.base.data
d = data
for i in self.pathInData:
d = d[i]
d.insert(index, value)
self.base.set_data(data)


Here we have the main element of the module, the classes which envelop everything else:

# THE MAIN CLASSES

class _BaseFile(_ObjectBase):
"""Class inherited by DictFile and ListFile.

This implements all the required methods common between
collections.MutableMapping and collections.MutableSequence."""
def __init__(self, path, pretty=False, sort_keys=False):
self.path = path
self.path = path
self.pretty = pretty
self.sort_keys = sort_keys
self.indent = 2  # Default indentation level

_initfile(self.path,
"list" if isinstance(self, ListFile) else "dict")

def _data(self):
"""A simpler version of data to avoid infinite recursion in some cases.

Don't use this.
"""
if self.is_caching:
return self.cache
with open(self.path, "r") as f:

@property
def data(self):
"""Get a vanilla dict object to represent the file."""
# Update type in case it's changed
self._updateType()
# And return
return self._data()

def __setitem__(self, key, value):
self._checkType(key)
data = self.data
data[key] = value
self.set_data(data)

def __delitem__(self, key):
data = self.data
del data[key]
self.set_data(data)

def _updateType(self):
"""Make sure that the class behaves like the data structure that it
is, so that we don't get a ListFile trying to represent a dict."""
data = self._data()
# Change type if needed
if isinstance(data, dict) and isinstance(self, ListFile):
self.__class__ = DictFile
elif isinstance(data, list) and isinstance(self, DictFile):
self.__class__ = ListFile

# Bonus features!

def set_data(self, data):
"""Overwrite the file with new data. You probably shouldn't do
this yourself, it's easy to screw up your whole file with this."""
if self.is_caching:
self.cache = data
else:
fcontents = self.file_contents
with open(self.path, "w") as f:
try:
# Write the file. Keep user settings about indentation, etc
indent = self.indent if self.pretty else None
json.dump(data, f, sort_keys=self.sort_keys, indent=indent)
except Exception as e:
# Rollback to prevent data loss
f.seek(0)
f.truncate()
f.write(fcontents)
# And re-raise the exception
raise e
self._updateType()

def remove(self):
"""Delete the file from the disk completely."""
os.remove(self.path)

@property
def file_contents(self):
"""Get the raw file contents of the file."""
with open(self.path, "r") as f:

# Grouped writes

@property
def is_caching(self):
"""Returns a boolean value describing whether a grouped write is
underway.
"""
return hasattr(self, "cache")

def __enter__(self):
self.cache = self.data
return self  # This enables using "as"

def __exit__(self, *args):
# We have to write manually here because __setitem__ is set up to write
# to cache, not to file
with open(self.path, "w") as f:
json.dump(self.cache, f)
del self.cache

class DictFile(_BaseFile, collections.MutableMapping):
"""A class emulating Python's dict that will update a JSON file as it is
modified.
"""
def __iter__(self):
return iter(self.data)

def _checkType(self, key):
if not isinstance(key, str):
raise TypeError("JSON only supports strings for keys, not '{}'. {}"
.format(type(key).__name__, "Try using a list for"
" storing numeric keys" if
isinstance(key, int) else ""))

class ListFile(_BaseFile, collections.MutableSequence):
"""A class emulating a Python list that will update a JSON file as it is
modified. Use this class directly when creating a new file if you want the
base object to be an array.
"""
def insert(self, index, value):
data = self.data
data.insert(index, value)
self.set_data(data)

def clear(self):
# Under Python 3, this method is already in place. I've implemented it
# myself to maximize compatibility with Python 2. Note that the
# docstring here is stolen from Python 3.
"""L.clear() -> None -- remove all items from L."""
self.set_data([])


And finally, this is what a user will directly initialize (most of the time). File is a class so that I can have staticmethods on it, but it really behaves like a factory function since it immediately changes self.__class__. Initializing a File() will give you either a ListFile or a DictFile object.

class File(object):
"""The main interface of livejson. Emulates a list or a dict, updating a
JSON file in real-time as it is modified.

This will be automatically replaced with either a ListFile or as
DictFile based on the contents of your file (a DictFile is the default when
creating a new file).
"""

def __init__(self, path, pretty=False, sort_keys=True, indent=2):
# When creating a blank JSON file, it's better to make the top-level an
# Object ("dict" in Python), rather than an Array ("list" in python),
# because that's the case for most JSON files.
self.path = path
self.pretty = pretty
self.sort_keys = sort_keys
self.indent = indent

_initfile(self.path)

with open(self.path, "r") as f:
if isinstance(data, dict):
self.__class__ = DictFile
elif isinstance(data, list):
self.__class__ = ListFile

@staticmethod
def with_data(path, data):
"""Initialize a new file that starts out with some data. Pass data
as a list, dict, or JSON string.
"""
# De-jsonize data if necessary
if isinstance(data, str):

# Make sure this is really a new file
if os.path.exists(path):
raise ValueError("File exists, not overwriting data. Use "
"'set_data' if you really want to do this.")
else:
f = File(path)
f.set_data(data)
return f

# Aliases for backwards-compatibility
Database = File
ListDatabase = ListFile
DictDatabase = DictFile


# Usage

Here's example usage in case that's helpful to critique my API design:

Basic usage:

import livejson
f = livejson.File("test.json")
f["a"] = "b"
# That's it, the file has been written to!


Usage as a context manager:

import livejson
with livejson.File("test.json") as f:
f["a"] = "b"


If I could get some feedback on this that would be great.

From a first look pretty clean code, and you also presented it in a nice fashion, so off to a good start.

I'll go with your questions first:

• Naming looks mostly good, though I'd wager that at some point generic names with "Base" in it will make it harder to understand than necessary - unfortunately I don't have a good suggestion as to what other names would be more helpful.
• Any unspecific question about performance should get the answer "take a profiler and look" IMO. Really, there's more to worry about than raw numbers, including the fact that this is Python. For starts the regular json module also isn't the fastest. However correctness comes first, so
• not only is thread safety a concern, access from multiple processes is a concern even more so. The one pattern that should really be used here at least is to first create a temporary file with the to-be-written content, then atomically moving that to the destination. I'm not sure what's the canonical source for this, but this blog post is a nice start. The set_data method is particularly bad in that respect.

Okay, so the code is almost PEP8 compatible, but there's still a lot of camel case in there. IMO that's fine as long as it's consistent.

One note:

• Replacing self.__class__ ... that works? Even if it did that looks extremely fishy, c.f. this SO post for example. I share the sentiment that a different mechanism would be better - if I see ...File() as a constructor call I definitely don't expect a subclass of that to be the object in question. Not doing "clever" things will save people minutes to hours of debugging time, and this is definitely too clever. Just replace it with a regular function like with livejson.mapped("test.json") as f: or so and make it clear that you get a File subclass as a result would be enough for me.
• hmm, I read that SO post, but it doesn't seem like many of those points apply to my case. It works fine in my unit tests, which cover 100% of my code. Can you think of a scenario in my specific case when this might make debugging harder or confuse users? I feel like for my specific case, in which the class is replaced immediately, before the user can do anything with the class, it's roughly equivalent to a factory function. Could you elaborate on some downsides for my specific case? Otherwise, I agree completely with you, thanks a lot :D – Luke Taylor Jul 5 '16 at 20:17
• Yes, I get your point and yes it's roughly equivalent; the coverage and testing is really great btw., so then I'd maybe downgrade it and say "if you document this behaviour clearly" it should be fine. – ferada Jul 5 '16 at 20:28