fix: array sub-type option respects its value, not just its presence (#415)

[simonkundrik]

Fixes #415

An array option configured with a falsy sub-type flag still enabled that coercion. For example, { key: 'foo', number: false } coerced the array values as numbers:

const parser = require('yargs-parser')

parser('--foo dog cat', { array: { key: 'foo', number: false } })
// before: { _: [], foo: [ NaN, NaN ] }
// after:  { _: [], foo: [ 'dog', 'cat' ] }

parser('--foo dog cat', { array: { key: 'foo', number: undefined } })
// before: { _: [], foo: [ NaN, NaN ] }
// after:  { _: [], foo: [ 'dog', 'cat' ] }

Root cause

When registering array options, the sub-type (boolean / string / number) was selected by key existence rather than by value:

const assignment = Object.keys(opt).map(function (key) {
  const arrayFlagKeys = { boolean: 'bools', string: 'strings', number: 'numbers' }
  return arrayFlagKeys[key]          // <- fires whenever the key is present
}).filter(Boolean).pop()

So { number: false } (or undefined) still mapped to 'numbers' and turned on number coercion.

Fix

Only enable the sub-type coercion when its value is truthy:

return typeof opt === 'object' && (opt as Record<string, unknown>)[key]
  ? arrayFlagKeys[key]
  : undefined

{ number: true } / { string: true } / { boolean: true } are unchanged; only explicit falsy values now behave as "not set".

Tests

Added a regression test (should not coerce array values as numbers when 'number' is false (#415)) using non-numeric values, since numeric-looking strings are coerced by the global parse-numbers default regardless of this flag.

Verified locally: full suite 363 passing (1 new), gts lint clean.

[shadowspawn]

To be clear: is this Pull Request entirely AI/LLM generated?

There is already a Pull Request open (#536). Is this better than the existing PR?

[simonkundrik]

Thanks for flagging this, and sorry for the noise. I missed #533 when I opened this, it predates mine and fixes the same thing, so that one should take priority. I've closed this in its favor.

And to answer your question directly: yes, this was written with AI assistance. I did review and test the change myself, but I clearly should have checked the open PRs more carefully before opening. Appreciate the heads-up.