Currently requires:
- Python 2.7
- CouchDB 1.0.1+ (untested on lower, 0.11 will probably work)
- http://pypi.python.org/pypi/CouchDB 0.8+ (untested on lower, 0.7 will probably work)
Source: http://github.com/wickedgrey/couchable
Package: http://pypi.python.org/pypi/couchable
API docs: http://packages.python.org/couchable
Blog: http://blog.nopinch.net/tag/couchable/
Example of Use
===============
import numpy
import couchdb
import couchable
couchable.registerAttachmentType(numpy.ndarray,
lambda obj: obj.dumps(),
lambda data: numpy.loads(data),
'application/octet-stream', gzip=True)
class BaseClass(object):
def __init__(self, shape, type_=numpy.uint16):
self.vol = numpy.zeros(shape, type_)
# ...
class ClassA(object):
def __init__(self, name):
self.name = name
# ...
couchable.registerDocType(ClassA,
lambda obj, cdb: couchable.newid(obj, lambda x: x.name),
lambda obj, cdb: None)
class SubClassB(BaseClass):
def __init__(self, container):
BaseClass.__init__(self, container.vol.shape, numpy.float32, container.pixelLength_mm)
self.container = container
# ...
couchable.registerDocType(SubClassB)
class Container(BaseClass):
def __init__(self):
BaseClass.__init__(self, (300, 400, 200), numpy.bool_)
# ...
couchable.registerDocType(Container)
def main(options, arguments):
cdb = couchable.CouchableDb(options.cdb_name)
view = couchdb.design.ViewDefinition('generic', 'byclass',
'''
function(doc) {
if ('couchable:' in doc) {
info = doc['couchable:'];
emit([info.module, info.class, doc._id], doc);
}
}''')
view.sync(cdb.db)
# ...
viewResult = view(cdb.db, include_docs=True, startkey=['example', 'Container'], endkey=['example', 'Container', {}])
container_list = cdb.load(viewResult.rows)
# ...
Examples of the JSON Structure of Stored Objects
================================================
All of the object dumps are taken from Futon.
>>> import couchable
>>> cdb=couchable.CouchableDb('example')
>>> class SimpleDoc(couchable.CouchableDoc):
... def __init__(self, **kwargs):
... for name, value in kwargs.items():
... setattr(self, name, value)
...
>>> a = SimpleDoc(name='AAA')
>>> cdb.store(a)
'main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7'
{
"_id": "main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7",
"_rev": "1-315ed02172dddb449a4ab38e54b8bb85",
"couchable:": {
"class": "SimpleDoc",
"module": "__main__"
},
"name": "AAA"
}
The key things to note are:
- The automatically generated id includes some hints about the type of the
object. It is only made if the object doesn't have an ID already.
- The object metadata is stored in "couchable:".
- Normal field names are stored at the top level of the object.
>>> a.int_ = 1234
>>> a.long_ = 1234567890L
>>> a.str_ = 'some str'
>>> cdb.store(a)
'main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7'
{
"_id": "main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7",
"_rev": "2-618b01362f076539f1abb481585b2f64",
"couchable:": {
"class": "SimpleDoc",
"module": "__main__"
},
"long_": 1234567890,
"name": "AAA",
"str_": "some str",
"int_": 1234
}
Numbers, strings and lists are stored in native JSON. Dicts with string keys
are as well, though non-string keys are disallowed by JSON, and so require
special handling (see below).
>>> del a.int_
>>> del a.long_
>>> del a.str_
>>> a.tuple_ = (1, 'two', 3.3)
>>> cdb.store(a)
'main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7'
{
"_id": "main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7",
"_rev": "3-a24d5422be27e8dca0f4b4375718bceb",
"name": "AAA",
"couchable:": {
"class": "SimpleDoc",
"module": "__main__"
},
"tuple_": {
"couchable:": {
"args": [
[
1,
"two",
3.3
]
],
"class": "tuple",
"module": "__builtin__",
"kwargs": {
}
}
}
}
Tuples do not have a native JSON representation, so they get treated much like
arbitrary objects do (note: since tuples don't have a __dict__, they need
special case support). Here, the "args" and "kwargs" will be passed into the
constructor of the type when it's time to load this object.
>>> del a.tuple_
>>> a._implementationDetail = 'this is private'
>>> cdb.store(a)
'main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7'
{
"_id": "main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7",
"_rev": "5-5ac8698c7b9567642484e728e60b95b1",
"couchable:": {
"private": {
"_implementationDetail": "this is private"
},
"class": "SimpleDoc",
"module": "__main__"
},
"name": "AAA"
}
CouchDB reserves all top-level fields that start with an underscore. Since
python makes liberal use of underscores to denote "private" or implementation
detail fields, those end up inside of a second-level dictionary inside of the
"couchable:" dict. During loading, these will end up back inside the object
__dict__.
>>> del a._implementationDetail
>>> a.reserved = 'couchable: reserving the string "couchable:" since 2010'
>>> cdb.store(a)
'main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7'
{
"_id": "main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7",
"_rev": "6-a7ef4d2ba59a68128b076cf5d5d0aaee",
"couchable:": {
"class": "SimpleDoc",
"module": "__main__"
},
"reserved": "couchable:append:str:couchable: reserving the string \"couchable:\" since 2010",
"name": "AAA"
}
Any string that starts with "couchable:" is escaped, since couchable makes
heavy use of that prefix. See below.
>>> del a.reserved
>>> a.dict_ = {'foo':'FOO', 123:'bar', (45, 67):'baz'}
>>> cdb.store(a)
'main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7'
{
"_id": "main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7",
"_rev": "7-4706c617a5f8900956c76ecb6f2e2daa",
"couchable:": {
"keys": {
"couchable:key:tuple:(45, 67)": {
"couchable:": {
"args": [[45, 67]],
"class": "tuple",
"module": "__builtin__",
"kwargs": {
}
}
}
},
"class": "SimpleDoc",
"module": "__main__"
},
"dict_": {
"couchable:key:tuple:(45, 67)": "baz",
"foo": "FOO",
"couchable:repr:int:123": "bar"
},
"name": "AAA"
}
As we saw above, tuples are supported using similar structures to those used
by arbitrary objects. However, JSON only supports strings as dictionary keys.
We solve this by replacing the object-as-key with a string that acts as a
pointer to the actual object, which is stored inside doc['couchable:']['keys']
as the full object.
Similarly, ints cannot be JSON dictonary keys, but we can use the int repr to
fully represent the object, and so we do so in-place (note that this wouldn't
work for tuples because tuples can contain arbitrarily complex objects, not
just the ints that we have in the example).
>>> del a.dict_
>>> a.nested = {(1,1):{(2,2):{(3,3):'four'}}}
>>> cdb.store(a)
'main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7'
{
"_id": "main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7",
"_rev": "10-1fe944276e27bef87259857be9d2abd3",
"nested": {
"couchable:key:tuple:(1, 1)": {
"couchable:key:tuple:(2, 2)": {
"couchable:key:tuple:(3, 3)": "four"
}
}
},
"couchable:": {
"keys": {
"couchable:key:tuple:(1, 1)": {
"couchable:": {
"args": [[1, 1]],
"class": "tuple",
"module": "__builtin__",
"kwargs": {
}
}
},
"couchable:key:tuple:(2, 2)": {
"couchable:": {
"args": [[2, 2]],
"class": "tuple",
"module": "__builtin__",
"kwargs": {
}
}
},
"couchable:key:tuple:(3, 3)": {
"couchable:": {
"args": [[3, 3]],
"class": "tuple",
"module": "__builtin__",
"kwargs": {
}
}
}
},
"class": "SimpleDoc",
"module": "__main__"
},
"name": "AAA"
}
Here we have nested dictionaries with tuple keys, and dict values. All of the
tuples are replaced with string pointers, and fully saved in 'keys'. Note
that 'keys' is a *document* level construct, not an object level one. 'keys'
will never appear outside of doc['couchable:'] (at least not outside of naming
coincidences).
>>> del a.nested
>>> b=SimpleDoc(name='BBB')
>>> a.bbb = b
>>> cdb.store(a)
'main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7'
{
"_id": "main__.SimpleDoc:2a208810-467f-4feb-a5bb-98d0beb1e5e7",
"_rev": "11-5eccb48b99431431c9f91b5e84a4ed7b",
"bbb": "couchable:id:main__.SimpleDoc:544a1408-d3f3-41c4-a428-9f0d2d8e2372",
"couchable:": {
"class": "SimpleDoc",
"module": "__main__"
},
"name": "AAA"
}
{
"_id": "main__.SimpleDoc:544a1408-d3f3-41c4-a428-9f0d2d8e2372",
"_rev": "1-577fefe455995539e11c992b7b46e10a",
"couchable:": {
"class": "SimpleDoc",
"module": "__main__"
},
"name": "BBB"
}
Here, we show the behavior when storing multiple objects, each of which is
flagged as needing to be a full document. Similar to the other string
pointers encountered already, couchable:id points to an object stored in an
entirely different document.
Known Limitations
=================
In general, couchable doesn't play well with classes that override __new__ in
odd ways.
Couchable cannot (and will probably never) store instances of the following
types:
- Tuple subclasses that override __new__ that *don't* do so in a way that is
compatible with collections.namedtuple.
- Programmatically defined classes that are not importable (they're basically
impossible to reconstruct during loading).
This list may not be exhaustive; unknown limitations may exist.