Preparing to use futures in storage.

[dhermes]

Wraps setting/getting of object _properties in custom methods. This will allow centralized detection of a future in a response and will also allow replacing with the value on access if it is ready.

Towards #775

---

@tseaver LMK what you think of this. The next step (I have yet to write it) is to make Batch._do_request return a future object. These objects will be owned and kept in order by the Batch and eventually populated. (This may be an issue because JSONConnection.api_request will try to parse the response if we use expect_json=True. We simultaneously want to parse the non-batched JSON responses and just ignore the futures)

[coveralls]

Coverage remained the same at 100.0% when pulling 35ac718 on dhermes:prepare-for-storage-futures into 41b5b72 on GoogleCloudPlatform:master.

[coveralls]

Changes Unknown when pulling b0af13c on dhermes:prepare-for-storage-futures into * on GoogleCloudPlatform:master*.

[tseaver]

The whole self._properties -> self._get_properties() dance seems to me to call for having the "future" bits hold a _properties which is a special object, whose __getitem__/__setitem__/get would raise the appropriate error. Converting the future would then be done by replacing the _properties attribute with a real dict.

[dhermes]

1. I figured you'd have good ideas on this, so I pushed it out without thinking too hard about it.
2. I don't particularly like _(get|set)_properties
3. I think your suggestion is a bit too high-powered for what is happening. I prefer simplicity if possible.
4. Writing this and Using futures in batched requests #812 I very much feel like this stuff is too intertwined (i.e. modules / classes know / assume too much about each other)

[coveralls]

Coverage remained the same at 100.0% when pulling fafb25c on dhermes:prepare-for-storage-futures into f08e71b on GoogleCloudPlatform:master.

[tseaver]

I imagined the following:

class UnrealizedFuture(ValueError):
    pass

class _UnrealizedFutureDict(object):

    def get(self, key, default=None):
        raise UnrealizedFuture()

    def __getitem__(self, key):
        raise UnrealizedFuture()

    def __setitem__(self, key, value):
        raise UnrealizedFuture()

unrealized = _UnrealizedFutureDict()

Where every "future" Blob / Bucket would have the unrealized singleton as its _properties. That doesn't seem "high-powered" to me, compared to the ugliness of the _get_properties() bit scattered everywhere.

[dhermes]

That's much simpler than I imagined! Let me give it a spin.

[coveralls]

Coverage remained the same at 100.0% when pulling 50da3a3 on dhermes:prepare-for-storage-futures into 028a15d on GoogleCloudPlatform:master.

[dhermes]

@tseaver I got rid of _get_properties in light of your suggestion but kept around _set_properties since it also resets _changes for us. Kept the old version of this PR around in https://github.com/dhermes/gcloud-python/tree/prepare-for-storage-futures-OLD for now, but will delete soon-ish.

I will give this a spin in #812 with your suggestion. It totally assuages my concerns about modules knowing too much about each other.

[tseaver]

LGTM