How do I serialize a Python dictionary into a string, and then back to a dictionary?

It depends on what you're wanting to use it for. If you're just trying to save it, you should use pickle (or, if you’re using CPython 2.x, cPickle, which is faster).

>>> import pickle
>>> pickle.dumps({'foo': 'bar'})
b'\x80\x03}q\x00X\x03\x00\x00\x00fooq\x01X\x03\x00\x00\x00barq\x02s.'
>>> pickle.loads(_)
{'foo': 'bar'}

If you want it to be readable, you could use json:

>>> import json
>>> json.dumps({'foo': 'bar'})
'{"foo": "bar"}'
>>> json.loads(_)
{'foo': 'bar'}

json is, however, very limited in what it will support, while pickle can be used for arbitrary objects (if it doesn't work automatically, the class can define __getstate__ to specify precisely how it should be pickled).

>>> pickle.dumps(object())
b'\x80\x03cbuiltins\nobject\nq\x00)\x81q\x01.'
>>> json.dumps(object())
Traceback (most recent call last):
  ...
TypeError: <object object at 0x7fa0348230c0> is not JSON serializable

Use Python's json module, or simplejson if you don't have python 2.6 or higher.

Pickle is great but I think it's worth mentioning literal_eval from the ast module for an even lighter weight solution if you're only serializing basic python types. It's basically a "safe" version of the notorious eval function that only allows evaluation of basic python types as opposed to any valid python code.

Example:

>>> d = {}
>>> d[0] = range(10)
>>> d['1'] = {}
>>> d['1'][0] = range(10)
>>> d['1'][1] = 'hello'
>>> data_string = str(d)
>>> print data_string
{0: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9], '1': {0: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9], 1: 'hello'}}

>>> from ast import literal_eval
>>> d == literal_eval(data_string)
True

One benefit is that the serialized data is just python code, so it's very human friendly. Compare it to what you would get with pickle.dumps:

>>> import pickle
>>> print pickle.dumps(d)
(dp0
I0
(lp1
I0
aI1
aI2
aI3
aI4
aI5
aI6
aI7
aI8
aI9
asS'1'
p2
(dp3
I0
(lp4
I0
aI1
aI2
aI3
aI4
aI5
aI6
aI7
aI8
aI9
asI1
S'hello'
p5
ss.

The downside is that as soon as the the data includes a type that is not supported by literal_ast you'll have to transition to something else like pickling.