mollom_validate_analysis() currently triggers the fallback mode when it encounters a mollom.checkContent response with a 'spam' value of 0 (unknown).
This 'spam' response value is undocumented and should normally not occur. A recently reported bug revealed that mollom.checkContent returns a 'spam' value of 0 when only checking for profanity; i.e., passing 'checks=profanity'. While Mollom's backend is going to be fixed in parallel, the client should not bail out this dramatically when encountering a 'spam' value of 0.
This is already covered by tests, but the checkContent responses are built differently in testing mode.
The question of how the client is expected to behave remains though. Treat it like 'ham'? Just let it through without performing any special adjustment/action?
| Comment | File | Size | Author |
|---|---|---|---|
| #1 | mollom.spam-unknown.1.patch | 871 bytes | sun |
Comments
Comment #1
sunAttached patch fixes the issue.
Comment #2
sunComment #3
dries commentedI'm a bit torn about this, thought about it, but finally decided to commit this to 'master' and '6.x-1.x'.
Comment #4
dries commentedLooking at this some more -- should we not report this as WATCHDOG_ERROR?
Comment #5
sunI was equally unsure whether to log it as notice (committed) or as an error.
Went with the notice, because it's not really an error in the client - it's a problem on the backend. In light of that, I was even tempted to not log it at all, but kept the logging in the hope that it might help to resolve future customer support requests more easily.