elm-safe-int

A safe 54-bit signed integer for use cases where normal Int isn't sufficient and 54-bit range will suffice.

import SafeInt

-- `2^40 / 10`
SafeInt.pow SafeInt.two (SafeInt.new 40)
    |> SafeInt.divBy (SafeInt.new 10)
    |> SafeInt.toInt
    --> Just 109951162777

-- `1 / 0` is undefined
SafeInt.div SafeInt.one SafeInt.zero
    |> SafeInt.toInt
    --> Nothing

Unchecked versions of functions are also provided for use cases where more speed is required at the cost of some lost safety and features.

import SafeInt.Unchecked as Unchecked


-- `2^40 / 10`
Unchecked.pow 2 40
    |> Unchecked.divBy 10
    --> 109951162777

Design Goals

Primary design goal is to fix many problems of normal Int with which:

• operations can crash
  • e.g. modBy 0 0
• operations can return e.g. NaN, -Infinity, +Infinity or 0.5
  • e.g. remainderBy 0 0, round(-1/0), round 0 ^ -1, round 2 ^ -1
• operations don't work for full 54-bit JavaScript safe integer range
  • e.g. 2^40 // 10
• operations are not consistent
  • e.g. modBy 0 0 crashes but remainderBy 0 0 returns NaN
• operations are not properly documented
  • e.g. many of the examples mentioned above
• behavior depends on compilation target
  • e.g. 2000000000 + 1000000000 will give different result when compiling Elm to JavaScript versus WebAssembly*

So in contrast to Int, with SafeInt:

• operations never crash
• operations either return an integer or undefined, nothing else
• operations work for full 54-bit JavaScript safe integer range
• operations are consistent
• operations are properly documented, including all cases which can result in undefined
• behavior is same on any compilation target which supports IEEE 754 binary64, including JavaScript and WebAssembly*

Secondary design goal is speed.

*) Elm compiler doesn't yet support compiling to WebAssembly.

Performance

I'm not interested in doing proper benchmarking, but in some quick benchmarks:

• SafeInt was up to 6 times slower than Int or Float
• Unchecked was up to 3 times slower than Int or Float
• SafeInt was up to 2.5 times slower than Unchecked

Benchmarking was done 2020-06-07 with Elm 0.19.1 and elm make --optimize and run in Firefox 68.8 in Debian.

Some example benchmarks are included in benchmarks directory of source.

Missing functionality

• Bitwise operations

  I don't think it makes sense to implement bitwise operations for SafeInt, but if use case is found I can reconsider this.

  As Float doesn't support bitwise operations directly these would need to be simulated somehow.