Refactor facts to improve performance:

[raphink]

• Add an apt_has_updates boolean fact
• Make other facts depend on it

These facts call the apt-check command 3 times for every run. This command usually takes 2 seconds or more to return, thus taking 6 seconds or more just for these 3 facts.

This PR proposes a refactoring which runs the command only once when no packages need to be updated, and twice if packages need updating (to get the package names).

I am well aware that it puts global variables outside of setcode blocks, but it adds a great performance gain.

Note: I refactored the spec tests because they were stubbing stuff blindly so most of them were passing by pure luck.

[daenney]

I'm not adverse to the globals, a 4s speedup is a good enough reason. I'll try this out for a sec, see if I can replicate the performance issues.

[daenney]

If I just run time /usr/lib/update-notifier/apt-check -p 2>&1 the worst speed I get is 0.6s. Where did you get your 2s+ from?

[raphink]

@daenney On Ubuntu laptops here (even using an SSD), this is what I get:

# time /usr/lib/update-notifier/apt-check -p 2>&1

real    0m2.008s
user    0m1.934s
sys     0m0.062s

[daenney]

Uhg okay, that's a problem

[raphink]

I'm probably not the only one in this situation, so I'd rather fix this in the facts than debugging apt-check.

[daenney]

Oh totally. Besides, by the time apt-check updates have made it to every platform and release involved we'll be twenty years older.

[daenney]

You know, from what I can see this looks good. It works and the tests are better. Plus I can really appreciate not calling out to the same command a bunch of times but 'caching' that response.