Speed up no-exclusive-tests and no-pending-tests

[szuend]

The isDescribe and isTestCase are doing a lot of work that is unnecessarily redone on each call expression.
This PR moves this work upfront, when the rule is created and re-uses the result for each call expression.

Please note that the choosen implementation is not a strong opinion. I tried to keep the code churn to a minimum and not export more astUtils internals. Feel free to suggest how you prefer this to be solved differently.

We noticed the performance while linting the DevTools frontend. Running with the following command line:

TIMING=1 third_party/node/node.py --output scripts/test/run_lint_check_js.js

Before the change:

Rule                                 | Time (ms) | Relative
:------------------------------------|----------:|--------:
mocha/no-exclusive-tests             |  3697.734 |    35.9%
mocha/no-pending-tests               |  2133.972 |    20.7%
@typescript-eslint/naming-convention |   761.571 |     7.4%
no-whitespace-before-property        |   373.331 |     3.6%
keyword-spacing                      |   359.593 |     3.5%
space-infix-ops                      |   266.812 |     2.6%
@typescript-eslint/func-call-spacing |   180.198 |     1.8%
space-in-parens                      |   175.388 |     1.7%
@typescript-eslint/comma-dangle      |   164.089 |     1.6%
no-trailing-spaces                   |   144.546 |     1.4%

After the change (the two eslint-plugin-mocha rules we use no longer show up in the TOP 10):

Rule                                 | Time (ms) | Relative
:------------------------------------|----------:|--------:
@typescript-eslint/naming-convention |   758.180 |    17.2%
no-whitespace-before-property        |   348.997 |     7.9%
keyword-spacing                      |   336.037 |     7.6%
space-infix-ops                      |   250.037 |     5.7%
space-in-parens                      |   215.183 |     4.9%
@typescript-eslint/func-call-spacing |   147.527 |     3.4%
@typescript-eslint/comma-dangle      |   144.443 |     3.3%
rulesdir/es_modules_import           |   127.713 |     2.9%
no-trailing-spaces                   |   122.335 |     2.8%
comma-style                          |   112.220 |     2.5%

[lo1tuma]

Nice! Thank you for that contribution. I’ve also already started work on a major refactoring of astUtils to also address this and a few other issues, but it is a big change and still not finished yet. So I really appreciate such a small change that improves the performance significantly.

In order to further decrease the performance I’ve introduce a benchmark test and set a performance budget. Can you check out this benchmark whether we can reduce the budged with this PR? See https://github.com/lo1tuma/eslint-plugin-mocha/blob/faa8bbf7dcd2a8f22bfe76a3a47b03bb678c2284/benchmarks/runtime.bench.js#L88.

[szuend]

Thanks for the really fast answer! The speed numbers that os.cpus() reports for my workstation are fluctuating quite heavily so I choose the new boundary rather conservatively.

I'd like to extend this approach for isDescribe and isTestCase to other rules that could benefit from them (but not other astUtils functions). Would you prefer a separate PR to keep commits small or should I modify this PR?

[lo1tuma]

Thanks. I would say let’s ship those changes. For further performance improvements we can create follow-up PRs.