Fixed touch(), with time=0 (with tests).

[niconoe]

The touch() method was throwing an error when time=0. The error itself wasn't showed properly because of a byte/string issue in a self.debuglog() call.

This PR fix both issues, and provides regression tests.

[timgraham]

I'm not sure if the time.sleep() technique is ideal. Requiring the test suite runtime to increase by multiple seconds when testing timeout doesn't seem like a good path to head down. It might be better to use mocking or to look at other cache libraries for inspiration.

Is it possible to add tests for the debuglog?

[niconoe]

About the time.sleep() technique: I agree, but don't really found a simple solution so far, so I thought it was good enough for now. It seems pylibmc use this technique too, pymemcache don't even properly test touch(), and I couldn't find a well-maintained Python-based memcached mock object.

The debuglog() call is implicitly tested on line 87, when trying to touch a nonexistent key (without the fix, TypeError is raised on Python 3). Do you want me to add a comment about this in tests?

[timgraham]

About verifying debuglog -- I think the tests should capture stderr output and check it rather than the test output looking like:

...MemCached: while expecting 'DELETED', got unexpected response 'NOT_FOUND'
MemCached: while expecting 'DELETED', got unexpected response 'NOT_FOUND'
....MemCached: while expecting 'STORED', got unexpected response 'SERVER_ERROR object too large for cache'
......MemCached: MemCache: inet:127.0.0.1:11211: test.  Marking dead.
.....MemCached: MemCache: inet:memcached:11211: connection closed in readline().  Marking dead.
MemCached: MemCache: inet:memcached:11211 (dead until 1481895267): connection closed in readline().  Marking dead.
.

It's a separate issues though.

[timgraham]

Is this supposed to be a regression test? As far as I can tell, it's passing before the fix is applied (returning True). Why assertNotEqual rather than assertEqual?

[niconoe]

I added this test because that case (delete with time=0) wasn't covered by the test suite before, and there was a risk my touch() fix would change this behaviour. So I considered it as a safety net, with the added benefit of increasing test coverage.

assertNotEqual(self.mc.set("my_key", "my_val"), 0) to be consistent with the docstrings in memcache.py. It is stated:

• if delete(time=0), subsequent set() should succeed immediately.
• a successful set() returns nonzero.

[timgraham]

Doesn't seem feasible. Again, I think we should use mocking to verify the values sent to the server without having to wait around to ensure the server handles those values correctly. I'll give it a try if you're stuck.

[niconoe]

I thought using mocking was larger than this issue and could be solved in another PR, but I can give it a try here!

[timgraham]

Updated in #137.