Use OCP classes as much as possible in files_external

[RobinMcCorkell]

To progress towards #4774 , I've modified files_external so that it uses OCP as much as is possible with the current public interfaces.

Also makes the code slightly cleaner and easier to work with in unit tests.

cc @DeepDiver1975 @rullzer @PVince81 @icewind1991

[karlitschek]

👍

[rullzer]

Looks good! 👍
Lets see what Jenkins has to say :)

[RobinMcCorkell]

Should we merge for oC 8.1? @DeepDiver1975

[scrutinizer-notifier]

A new inspection was created.

[MorrisJobke]

I rebased this to invoke new CI run.

[ghost]

🚀 Test PASSed.🚀

[DeepDiver1975]

I'm going to revert this - please resubmit with externals storage tests not failing - THX

[RobinMcCorkell]

Eek, I wonder what happened. Investigating.

[DeepDiver1975]

Eek, I wonder what happened. Investigating.

no worries - shit happens

[RobinMcCorkell]

Right, it looks like (for some reason) the UserManager cannot find the user passed in to the files_external code, so it returns null, triggering the error when getHome() is called on it. The old OC_User::getHome() code checks for null, and returns what it hopes is a valid home path if the user cannot be found. IMHO this is extremely dangerous, as the user might not even exist on the system.

@PVince81 @icewind1991 @DeepDiver1975 What should we do? Emulate the old behaviour and guess? Or bomb out with some sort of UserNotFound exception?