Python hooks reloaded

[alexAubin]

The problem

This is a rewrite of #535 because there were too many changes ... :

Python hooks are required for the new diagnosis system modularity / extendability, and more generally if we want to implement some hooks in python for some reason. This could be the case for example of the famous app ssowatconf command which should be a classical regen-conf thing but can't be so far because it has too many python calls and json stuff to be implemented in bash.

Solution

Split the behavior of the hook_exec function depending on the nature of the hook. Python hooks are to be modules loaded by yunohost and expected to have a main function.

PR Status

Tested and working

How to test

Create some hook with a main function (e.g. just displaying hello world) and call yunohost hook list your_hook_type and yunohost diagnosis callback your_hook_type

Validation

• Principle agreement 0/2 :
• Quick review 0/1 :
• Simple test 0/1 :
• Deep review 0/1 :

[Josue-T]

Didn't tested, but look good to me.

[alexAubin]

Merging soonish

[Psycojoker]

Shouldn't we spawn a new python process that launch the python instead of calling it in the current python process? That would be better security and bug wise I think (but idk how hard would it be to do that)

[alexAubin]

Hm I kinda chose to do this because that's supposedly less ressource expensive, considering that we're likely to import code from yunohost's in each of those (e.g. 10~15ish hooks in the context of diagnosis), each likely to do stuff like importing requests or other libs that take hell of time to load on RPi for example

[alexAubin]

Proposing to merge soonish in 3.7 (not really needed or anything but at least that becomes stable code upon which the migration mechanism can be built and possibly other things)