misc random

[timotheecour]

D20201208T190246

links

• misc oids · Issue #431 · timotheecour/Nim
• random

[timotheecour]

• we should port urandom/getentropy/arc4random + friends
  in particular, this would provide better seeding for std/random.randomize

see man urandom and man getentropy

/dev/urandom is a compatibility nod to Linux. On Linux, /dev/urandom will produce lower quality output if the entropy pool drains, while /dev/random will prefer to block and wait for additional entropy
     to be collected.  With Fortuna, this choice and distinction is not necessary, and the two devices behave identically. You may use either.

     The random device implements the Fortuna pseudo random number generator algorithm and maintains its entropy pool.  The kernel automatically seeds the algorithm with additional entropy during normal exe-
     cution.

etc

EDIT: would help properly fix nim-lang#8588

[timotheecour]

• port SecureRandom and PseudoRandom

see https://github.com/mongodb/mongo/blob/eb2b72cf9c0269f086223d499ac9be8a270d268c/src/mongo/platform/random.h

* More secure random numbers
 * Suitable for nonce/crypto
 * Slower than PseudoRandom, so only use when really need

[timotheecour]

• see also uuid implementation from https://github.com/mongodb/mongo/blob/eb2b72cf9c0269f086223d499ac9be8a270d268c/src/mongo/util/uuid.cpp

[ringabout]

• port SecureRandom and PseudoRandom

see https://github.com/mongodb/mongo/blob/eb2b72cf9c0269f086223d499ac9be8a270d268c/src/mongo/platform/random.h

* More secure random numbers
 * Suitable for nonce/crypto
 * Slower than PseudoRandom, so only use when really need

This is really useful.
https://docs.python.org/3/library/secrets.html
https://github.com/cheatfate/nimcrypto/blob/master/nimcrypto/sysrand.nim

[timotheecour]

• some bugs with rand(float) corner cases:

when true:
  #[
  D20201208T123017:here
  ]#
  import std/random
  import std/fenv
  proc main()=
    echo rand(-Inf .. Inf) # bug: nan
    echo rand(0.0 .. Inf) # ok: inf
    echo rand(0.0 .. float.maximumPositiveValue) # ok: 5.992393389465399e+307
    echo rand(-float.maximumPositiveValue .. float.maximumPositiveValue) # bug: inf
    echo rand(float.maximumPositiveValue) # ok: 7.976840732228883e+307
    echo rand(Inf) # ok: inf
  main()

[timotheecour]

• make random threadsafe if there is no meaningful overhead

This proc uses the default random number generator. Thus, it is not thread-safe.

(either using threadvar or using locks)

[timotheecour]

• rand at CT
  Cannot use rand at compile time · Issue #12028 · nim-lang/Nim (still broken despite issue closed)

[timotheecour]

• rand(T) should work for all ordinal types, eg: rand(char) should work
  (now that Ordinal don't contain enum with holes, refs Random: proc rand(x: HSlice) requires substraction nim-lang/Nim#7698 (comment))

[timotheecour]

• consider whether it would make sense to port alternative rand algorithm: wyrand
  refs: https://github.com/wangyi-fudan/wyhash

wyhash and wyrand are the ideal 64-bit hash function and PRNG respectively:
solid: wyhash passed SMHasher, wyrand passed BigCrush, practrand.
portable: 64-bit/32-bit system, big/little endian.
fastest: Efficient on 64-bit machines, especially for short keys.
simplest: In the sense of code size.

(suggested by @xflywind)

[ringabout]

• rand(var Rand, typedesc) overload
  rand(var Rand, typedesc) overload nim-lang/Nim#16425

[ringabout]

• rand(typedesc) / next doesn't work in JS backend
  rand(int8) => JS

[timotheecour]

• Add random.sampleBuffer nim-lang/Nim#16957 (comment)

• add sampleBuffer
• changing randomize() to use the fresh new std/sysrandom, then users can choose depending on how they call:

# CSPRNG
var r = initRand(opt = useSysrand)
sampleBuffer(r, buf, alphabet)

# PRNG
var r = initRand(opt = useTime)
sampleBuffer(r, buf, alphabet)

# CSPRNG
randomize() # could be changed to: randState = initRand(opt = useSysrand)
sampleBuffer(randState, buf, alphabet) # uses default RNG