Improve the OOP ABI (page 7)

On Tuesday, 23 January 2024 at 21:37:42 UTC, Walter Bright wrote: > On 1/23/2024 3:49 AM, Andrea Fontana wrote: >> So probably you like xxhash128 >> >> https://xxhash.com/ > > Thanks! I do, it has a 64 bit version, does anyone know what the collision rate is? For 64 bits, your best best is CRC64. However, the risk of collision starts to be a bit too high to rely exclusively on this.

January 24

Re: Improve the OOP ABI

Posted by Siarhei Siamashka
in reply to deadalnix

Permalink

Siarhei Siamashka

Posted in reply to deadalnix

Permalink

On Wednesday, 24 January 2024 at 01:18:18 UTC, deadalnix wrote:

On Tuesday, 23 January 2024 at 22:15:11 UTC, Siarhei Siamashka wrote:

On Tuesday, 23 January 2024 at 21:37:42 UTC, Walter Bright wrote:

On 1/23/2024 3:49 AM, Andrea Fontana wrote:

So probably you like xxhash128

https://xxhash.com/

Thanks! I do, it has a 64 bit version, does anyone know what the collision rate is?

If you really need a 64-bit version, then you can simply take the first 64 bits of MD5. Assuming no defects in a 64-bit hash implementation, the collision probability is 1.0 / 2^^64 for a pair of hashed strings. And it becomes a https://en.wikipedia.org/wiki/Birthday_problem if the number of hashed strings is larger. There's no magic algorithmic sauce that can make a small hash collision resistant. A quote from the Wikipedia article:

While it is not possible to do better than this, it is certainly possible - and in fact way easier than you'd expect - to do worse.

Go ahead and show me truncated output of any hash, designed to be cryptographically secure, behaving badly. I know that xxhash, ahash, fnv or any other non-secure hashes are prone to have collision issues, as they are primarily optimized for speed while being borderline acceptable from the collision standpoint.

On Wednesday, 24 January 2024 at 01:18:18 UTC, deadalnix wrote:

This thread has gone in the typical bullshit D goes into, arguing about the intricacies of some details that do not matter at all for the value delivered in the end.

Why try to demotivate people? this could be a good entry point for a new contributor, it's not "bullshit", what's bullshit is sitting here doing nothing but complain

Submit PRs, encourage people to fix things, don't be negative, reading you feels like there is no point even participating, not everyone is a compiler developper, not everyone is able to "improve your OOP ABI", everyone is however able to benchmark and change the hashing function for a faster one

If you want "Improve the OOP ABI", it better make the compiler remain fast, that detail, is not futile

Forums