Decouple connection from Blob

[dhermes]

NOTE: Has #822 as diffbase.

Towards #728

@tseaver I made a separate commit for each method with used Blob.connection but can combine them if you like.

I wanted to get rid of the Blob.connection property, but _PropertyMixin.reload() and _PropertyMixin.patch() use it as well as ObjectACL.reload() and ObjectACL.save().

[coveralls]

Coverage remained the same at 100.0% when pulling 5cf9023 on dhermes:decouple-connection-in-blob into 6283ab9 on GoogleCloudPlatform:master.

[tseaver]

9932f0d LGTM
e429f9a LGTM
8ebb280 see comments inline (setting connection=None in ctors to verify that override is used)
824ff20 same issue
5cf9023 same issue

[dhermes]

@tseaver I imagine this being a transient issue since at some point connection will not be bound to any of the types or to the stubs.

I'm not sure what would need to be checked either.

[tseaver]

@dhermes in the tests, we should pass None to the bucket ctor in order to ensure that the passed-in connection is being used (the tests as we have them would pass if the methods took the connection argument, but then fell back to using self.connection).

[dhermes]

The heisenbug that @jgeewax fixed in #834 raised it's head here too. Only the 2nd time I had seen (build failure was first). Will rebase.

---

@tseaver All the Blob methods I changed use the passed in connection only and don't fall back to self.connection.

They are moving towards no connection anywhere, so falling back is a no-no.

[coveralls]

Coverage remained the same at 100.0% when pulling ca34a4a on dhermes:decouple-connection-in-blob into 40d84e9 on GoogleCloudPlatform:master.

[tseaver]

That's why having the bucket initialized with an invalid connection is the Right Thing (TM): it ensures that fallback cannot cause the test to succeed.

[dhermes]

Gotcher. Current plan is to cherry-pick each of the 5 commits and then replace Bucket(connection) with Bucket(None) or similar. (About to spend an hour without internet on BART.)

LMK if this is a bad idea.

[tseaver]

SGTM

[dhermes]

@tseaver PTAL

[coveralls]

Coverage remained the same at 100.0% when pulling 60d43ac on dhermes:decouple-connection-in-blob into 40d84e9 on GoogleCloudPlatform:master.

[tseaver]

LGTM

[dhermes]

@tseaver Can we divvy up work and try to close out #728 together?

[tseaver]

How do you see the split?

[dhermes]

Can you handle the methods in acl.py and I'll try to cover the rest (Bucket and _PropertyMixin)?

Then we "converge" and remove the connection property from Blob (it gets used in acl.py).

[tseaver]

OK. I just inventoried those methods in #825.

[dhermes]

Awesome! Thanks