Nixpkgs: x32 abi in future?

[vcunat]

It seems a good thing for the most common (desktop) use cases. https://sites.google.com/site/x32abi/ Probably report your interest/ideas here. Some mainstream distros are starting to support it (Gentoo, Debian), so pushing it to nixpkgs should not be so much work.

[shlevy]

Any benchmarks for common use cases?

[bluescreen303]

https://lwn.net/Articles/548838/

While the higher performance (compared to x86 and x86_64) in every (perhaps contrived) benchmark is nice, it's especially the far lower memory footprint that interests me.

RAM isn't cheap with most cloud/VPS providers.

[vcunat]

The only significant difference against x86_64 are 4-byte pointers (and "long" C type), which can increase speed due to saving cache space (but that IMO depends on particular case a lot). It's speculated that VMs like python and haskell would have larger benefits, as they maintain lots of pointers. I could find almost no benchmarks, though (e.g. a few days ago someone using it on debian wrote it is faster, but this is all quite vague).

[Profpatsch]

(triage) To we already have that?

[vcunat]

We don't, but x32 doesn't seem really worth the effort in this general-purpose nixpkgs case. Maybe for particular packages or use cases. It would take lots of resources to maintain (whole) nixpkgs for yet another "platform".

Thanks for pinging me; I recently remembered I wanted to close this, but I wasn't at a computer at that time and forgot afterwards.

[volth]

After @Ericson2314 efforts for cross-compilation, it might be much easier to add x32 as target platform.
At least for some packages.
I'd like to compare performance of pkgs.varnish and pkgs.pkgsx32Linux.varnish :)

Also, I have seen that some people install pkgs.pkgsi686Linux.firefox to save memory.
Having pkgs.pkgsx32Linux.firefox would be a nice alternative.

[vcunat]

32-bit firefox does save memory noticeably, which is why I commonly used it on my 4GiB-RAM systems (I have more RAM now); IIRC it/upstream didn't support x32abi when I last checked (perhaps most hopeful link).

[Ericson2314]

I do want to have hydra.nixos.org cross compile the whole world to some platform, say post 17.09. Doesn't really matter what the target platform is, so this could be it.

[volth]

@Ericson2314 BTW, is it currently possible to build an aarch64 system closure (for scaleway) on x86_64-linux? Last time I tried it was not possible without run some builds on aarch64

[Ericson2314]

@volth like a minimal NixOS system? I haven't tried. Also, currently cross compilation is fairly broken pending https://github.com/NixOS/nixpkgs/pull/26805.

[volth]

yes.
ah, sorry, I expected that #26805 is already merged (there were issues with propagatedBuildInputs and I mistakenly connected them with merge of #26805 and remembered as though the cross-compilation is already here)

[Ericson2314]

Heheh, I wish it was! But still getting the stdenvs to build, and then I need to rewrite the docs, and finally others need to review.

[volth]

It seems that with the latest progress in cross-compilation it would be cheap and trivial to add pkgsCross.x32