Range tests on zero-length buffer+string are incompatible with bytewise, charwise & IDB

[vweevers]

They assume that a zero-length string or zero-length Buffer means "not defined". IMO that should be left up to an implementation.

abstract-leveldown/test/iterator-range-test.js

Lines 313 to 345 in fd64104

	if (testCommon.bufferKeys) {
	rangeTest('test iterator with gte as empty buffer', {
	gte: Buffer.alloc(0)
	}, data)
	rangeTest('test iterator with start as empty buffer - legacy', {
	start: Buffer.alloc(0)
	}, data)
	rangeTest('test iterator with lte as empty buffer', {
	lte: Buffer.alloc(0)
	}, data)
	rangeTest('test iterator with end as empty buffer - legacy', {
	end: Buffer.alloc(0)
	}, data)
	}
	rangeTest('test iterator with gte as empty string', {
	gte: ''
	}, data)
	rangeTest('test iterator with start as empty string - legacy', {
	start: ''
	}, data)
	rangeTest('test iterator with lte as empty string', {
	lte: ''
	}, data)
	rangeTest('test iterator with end as empty string - legacy', {
	end: ''
	}, data)

In bytewise, charwise and IndexedDB, an empty string is a significant type. In all three, the sort order of types is:

• null (not supported by IDB)
• false (not supported by IDB)
• true (not supported by IDB)
• number (numeric)
• date (numeric, epoch offset)
• binary (bitwise) (not supported by charwise and IDB First Edition)
• string (lexicographic)
• array (componentwise)
• undefined (not supported by IDB)

For example, charwise encodes an empty string as 'J'.

I suggest:

• Remove these tests from abstract-leveldown, document behavior (Remove range tests that assumed zero-length strings or Buffers meant "not defined" #319) (semver-patch)
• Copy the tests to leveldown, document the behavior (Add tests for zero-length range options leveldown#573) (semver-patch or RC, depending on when this lands)
• Add new tests to level-js on sort order of the above types, document the behavior (Fix zero-length range options level-js#156) (release as part of upgrade to abstract-leveldown@6)

[vweevers]

Note that level-js is currently passing the test suite of abstract-leveldown@5 because abstract-leveldown used to filter out range options that were a zero-length string or buffer.

[vweevers]

Correction: unlike bytewise, IDB sorts binary after string. For reference, here's a quick test of IDB type order:

Click to expand

function cmp (a, b) {
  // # https://www.w3.org/TR/IndexedDB-2/#key-construct
  // "Let ta be the type of a, let tb be the type of b."
  var ta = type(a)
  var tb = type(b)

  // "If ta is array and tb is binary, string, date or number, return 1."
  if (ta === 'array' && (tb === 'binary' || tb === 'string' || tb === 'date' || tb === 'number')) return 1

  // "If tb is array and ta is binary, string, date or number, return -1."
  if (tb === 'array' && (ta === 'binary' || ta === 'string' || ta === 'date' || ta === 'number')) return -1

  // "If ta is binary and tb is string, date or number, return 1."
  if (ta === 'binary' && (tb === 'string' || tb === 'date' || tb === 'number')) return 1

  // "If tb is binary and ta is string, date or number, return -1."
  if (tb === 'binary' && (ta === 'string' || ta === 'date' || ta === 'number')) return -1

  // "If ta is string and tb is date or number, return 1."
  if (ta === 'string' && (tb === 'date' || tb === 'number')) return 1

  // "If tb is string and ta is date or number, return -1."
  if (tb === 'string' && (ta === 'date' || ta === 'number')) return -1

  // "If ta is date and tb is number, return 1."
  if (ta === 'date' && (tb === 'number')) return 1

  // "If tb is date and ta is number, return -1."
  if (tb === 'date' && (ta === 'number')) return -1

  throw new Error('value comparison is not implemented')
}

function type (value) {
  const t = typeof value

  if (t === 'string' || t === 'number') return t
  if (Buffer.isBuffer(value)) return 'binary' // stored same as arraybuffer
  if (Array.isArray(value)) return 'array'
  if (Object.prototype.toString.call(value) === '[object Date]') return 'date'

  throw new Error('unsupported type')
}

// [ 3, <date>, '', <Buffer>, [] ]
console.log([
  new Date(),
  '',
  3,
  [],
  Buffer.alloc(0)
].sort(cmp))