Add a toggle for parsing noqa comments

[ragne]

Summary

The feature request arose when I tried to use ansible-lint as pre-commit hook and it took more than 10 seconds to complete.
Profiling shown that most of the time was spend in ruamel.yaml.load from append_skipped_rules. Which is perfectly logical, because CLoader AFAIK doesn't support comments.
When I added a toggle flag, the total time went down to ~1.5 seconds.

Thus it would be nice to have a speed boost when parsing #noqa isn't needed for some reason (e.g you have skip tag instead)

Issue Type

• Feature Idea

Additional Information

Example flamegraph:

[ssbarnea]

I am very well aware of speed issues, and your case is not even major. I know repositories where it takes more than a minute to run. Still, disabling some essential features does not seem like the right fix.

Instead I would focus on making it run in parallel much better. With 8 cores would would likely make it ~6-7x faster, kinda similar with your idea, but more important: without disabling features.

I would recommend using https://opendev.org/zuul/zuul-jobs/src/branch/master/roles as reference model for performance measurement as it contains lots of roles.

[ragne]

Unfortunately during my attempts to test with zuul repo I was repeatedly beaten by:

• too new version of ansible (of which you cannot attach line anymore)
• After merging your branch with fix I was facing a missing zuul_return module, which I don't know where to find

So, I used whatever handy repos I had. I'm sorry, I was unable to grasp where it would be really beneficial to introduce the parallelization (e.g on enumerating playbook files or go further and parallelize rule applying). I went with parallel run by file, which you can see in master...ragne:parallel

As for why I'm not bringing a PR, there's a couple of gotchas with my approach (and with parallelization in general). Either I did not figure it out, but joblib which I used refused to pickle any of rules (probably because of imports, didn't yet look), their "native" "loky" backend however chewed it fine and I was able to actually run 16 processes in parallel.
Now to the gotchas: I've noticed that decorating parse_yaml_linenumbers and extracting yaml loading from _append_skipped_rules would already give huge performance improvements. And synchronising LRUCache across different processes is a bit difficult and less effective.

I still have to measure impact of parallelizing I've done, but after applying lru (I'm sure you might find a couple other bottlenecks) it isn't so convincing anymore