optimization in flatten function

[marijaselakovic]

Similarly as in https://github.com/jashkenas/underscore/pull/2496, I found that flatten function can be optimized by swapping the order of checks in logical expression in line 500 (the second check is false most of the time). I provide the screenshot with the results for unit tests which performance got improved (on V8 4.4 engine). Inputs are the same as in unit tests in the repository and performance is tested with Benchmark.js library.

The code for tests can be download from: https://s3.eu-central-1.amazonaws.com/underscorebenchmark/benchmarks2.zip

To run the tests, just call

node run.js

I am maybe wrong, but can isArrayLike(value) check be omitted here?

[coveralls]

Coverage remained the same at 96.863% when pulling f126f19 on marijaselakovic:master into 5500925 on jashkenas:master.

[megawac]

Yes, isArrayLike can be removed here

[jridgewell]

Yes, isArrayLike can be removed here

Not quite. This opt will only benefit isArray since it's faster than isArrayLike + isArray on non-array objects. But this'll hurt the isArguments check, since that uses the slower toString comparison.

Instead, we should guard only the isArguments check:

if (_.isArray(value) || (isArrayLike(value) && _.isArguments(value)) {
  //...
}

[jgonggrijp]

isArrayLike could be removed, as it is only there to improve performance. The current check is already optimal, however:

	arrays	arguments	other
current	isArrayLike + isArray	isArrayLike + isArray + isArguments	isArrayLike
marijaselakovic	isArray + isArrayLike	isArray + isArguments + isArrayLike	isArray + isArguments + isArrayLike
megawac	isArray	isArray + isArguments	isArray + isArguments
jridgewell	isArray	isArray + isArrayLike + `isArguments	isArray + isArrayLike

The toString-based checks isArray and isArguments are more expensive than the heuristic isArrayLike. See the comments in the resolved part of the second review in #2840. Avoiding isArrayLike for arrays is only a small optimization while having isArray for other values is a big pessimization.