Re: [hackers] Re: [sbase] Orphan patches and unsent patches

From: NRK <nrk_AT_disroot.org>
Date: Sun, 3 Mar 2024 00:50:28 +0000

On Sun, Mar 03, 2024 at 12:58:20AM +0100, Elie Le Vaillant wrote:
> I'm using the web interface to the mailing list to check what has been
> sent, and these patches were not sent.

The web archive is not reliable and often drops mails.

> + * Copied off OpenBSD (original is arc4random_uniform)
> + */
> +uint32_t
> +random_uniform(uint32_t upper_bound)

I'd recommend using the bitmask approach described in here:
https://www.pcg-random.org/posts/bounded-rands.html#bitmask-with-rejection-unbiased-apples-method

And if you want a portable clz without depending on __builtin_clz() then see:
https://graphics.stanford.edu/~seander/bithacks.html#RoundUpPowerOf2

> + r = random(); /* arc4random() is better, but we don't always have it */

I'm not sure why you think arc4random() is better. I doubt you need any
cryptograhic guarantees here. Personally I'd just go with a simple PCG
construction. It takes about ~6 lines of code [0], is fast and has good
statistical properties. Would also avoid spending time locking/unlocking
mutex inside libc for no reason.

[0]: https://github.com/imneme/pcg-c-basic/blob/bc39cd76ac3d541e618606bcc6e1e5ba5e5e6aa3/pcg_basic.c#L60-L67
or: https://codeberg.org/NRK/slashtmp/src/branch/master/prng/pcg_32.c

- NRK
Received on Sun Mar 03 2024 - 01:50:28 CET