Turn on object mode for the readable side

[mourner]

Fixes #12 (better late than never!). Also has a nice side effect of ~8% performance improvement.

I assume this would warrant a major version bump just to be on the safe side?

[mourner]

@cc @tyrasd @myndzi thoughts?

[myndzi]

Seems to do the thing. I still have a difference in behavior with my tests regarding empty tokens and empty delimiters primarily; for example, with the input "abcde" and a delimiter of "", I emit ["a", "b", "c", "d", "e"], while this package seems to emit "abcde". This could be seen as a behavioral choice and not a bug, but diverges from String.prototype.split. Equally, my code emits empty values in the case of an input such as "abba" with a delimiter of "b": ["a", "", "a"], while this code emits ["a", "a"]. Same thing. Neither really affects this PR, but they do block the PR I was originally preparing when filing #12 (which was to integrate my test suite into this package).

[myndzi]

(If you're interested, the WIP branch is here: https://github.com/myndzi/binary-split/tree/integrate-new-tests -- I didn't bother to convert the test runner yet since the tests simply fail)

[mourner]

@myndzi good points! I think we should follow the same logic for edge cases as split and split2 packages — let's check how those behave.

[mourner]

@myndzi results for split2 are mixed:

• '.a..b.' / '.' -> ['', 'a', '', 'b']
• 'abcd' / '' -> ['abcd']