Check for unknown config options

[meejah]

The list of valid sections + config-items came from grep'ing the source for .get_config

[coveralls]

Coverage decreased (-0.008%) to 92.237% when pulling e2a34f0 on meejah:2809.unknown-configs.0 into 00ea41e on tahoe-lafs:master.

[warner]

I like it, but I think we need to make some changes before landing:

• Node is shared between Client and IntroducerNode, and the introducer uses a subset of the config options (really just the [node] section, I think), so we need separate validation rules for each. So I'm thinking that _validate_config() still lives on the Node class, but it accepts the what-is-legal dict as an argument. CONFIG_SECTIONS gets moved out to a top-level definition (not a class attribute) in client.py and introducer/server.py (with separate values in each), and the call to self._validate_config(CONFIG_SECTIONS) moves into Client.__init__ and IntroducerNode.__init__.
• tests! The approach in test_node.ClientNotListening should work pretty well: it makes a tempdir, writes a tahoe.cfg, then instantiates a Client and sees if it explodes or not. self.mktemp() is probably better for the tempdir than what I've currently got in there. I'd actually consider making _validate_config() a standalone function (instead of a method), so you could test it by making a config object from the file and then submitting it to the function (instead of instantiating a whole Client or IntroducerNode). My goal here is to run the tests quickly, and instantiating a whole Node means it must create a new Tub, with a new RSA key, and really all we care about is the config file. Another option is to make a FakeClient subclass of Client with a __init__ that strips out everything but calling the load-config/validate-config sections.

[meejah]

@warner does the latest commit address your code-location etc concerns? I still need to change the lists of valid config items for client vs introducer so not quite ready for merge

[meejah]

Okay so b4282b6 is my stab at the minimum list of things the introducer needs (based on searching the code).

[coveralls]

Coverage increased (+0.01%) to 92.257% when pulling b4282b6 on meejah:2809.unknown-configs.0 into 00ea41e on tahoe-lafs:master.

[coveralls]

Coverage increased (+0.01%) to 92.257% when pulling b4282b6 on meejah:2809.unknown-configs.0 into 00ea41e on tahoe-lafs:master.

[coveralls]

Coverage decreased (-0.007%) to 92.238% when pulling b4282b6 on meejah:2809.unknown-configs.0 into 00ea41e on tahoe-lafs:master.

[coveralls]

Coverage decreased (-0.007%) to 92.238% when pulling b4282b6 on meejah:2809.unknown-configs.0 into 00ea41e on tahoe-lafs:master.

[coveralls]

Coverage increased (+0.02%) to 92.262% when pulling 1d34de0 on meejah:2809.unknown-configs.0 into 00ea41e on tahoe-lafs:master.

[meejah]

I squashed everything, should be good-to-go incorporating our discussion.

[coveralls]

Coverage decreased (-0.01%) to 92.236% when pulling 2732c37 on meejah:2809.unknown-configs.0 into 00ea41e on tahoe-lafs:master.

[warner]

Looks good. I think I might add a test to feed the tahoe create-client default config through the validator (seems like a good sanity check), and also update the current tests to make sure the error message includes the right information about which key had a problem. But I'll merge this first.