Re: [9front] [PATCH] libc: replace lrand's algorithm

[qwx@sciops.net]

On Fri Mar 29 17:02:32 +0100 2024, eolien55@disroot.org wrote:
> > lrand is used by rand(2) itself and is used throughout the system
> > and elsewhere, and who knows what impact changing it might have.
> 
> I looked through every usage of lrand-related functions in the codebase,
> and it seems that all programs use it simply as "a uniform pseudo-random
> number generator", without other expectations on the output (things
> that would vary from algorithm to algorithm). However the manpage should
> be changed to reflect the algorithm change.

I agree, it doesn't hurt to at least mention this in the manpage.


> However, through looking at the full source tree, it appears there
> are at least 3 lrand duplicates in the source tree, each with its
> own algorithm: one for the kernel, in /sys/src/9/port/random.c, which
> uses xoroshiro128+, which is quite good; one for normal userspace in
> /sys/src/libc/lrand.c, which uses the aforementioned ALFG, which is
> quite suboptimal; and one for ape in /sys/src/lib/v/rand.c, which
> seemingly points to this[1] article. The algorithm is not in common
> use, so maybe it should be changed to one whose properties are
> better-known?
> 
> Given that kernel's lrand is xoroshiro128+, it would make sense to
> either change kernel's lrand to PCG too, or change the proposed
> algorithm to xoroshiro128+.

I like the idea of simplifying the tree and deleting some code,
but keep in mind that the xoroshiro128+ variant is what is
exposed by /dev/random via truerand(2).  It's of little use to
touch the APE code at all at this point.  Most people are
uninsterested in a change here because of the dichotomy between
rand and truerand, the negligeable performance impact, etc.,
and beyond this, it gets into "this is like, my opinion man"
territory.  I wouldn't mind replacing libc's lrand with a PCG
variant also exposing a 64bit version, which we don't have, if
it indeed also improves the statistical properties of the prng.
But that's like, my opinion man.


> > Applications which need guarantees about distribution or
> > performance are not supposed to use rand(2) to begin with.
> 
> Well in that case, some code in /sys/src/cmd needs to be changed
> to not use rand, like spin and venti.

I agree, though again, I don't know what the impact of such
a change would be.

> Cheers,
> Elie Le Vaillant


Cheers,
qwx