Conversation

Notices

1. Embed this notice
   Luj (luj@chaos.social)'s status on Monday, 27-Jan-2025 23:03:32 JST Luj

   • Dr. Théo Zimmermann ?️‍?
   • Stefano Zacchiroli

   📢 I am very proud to announce the publication of my #research #paper "Does Functional Package Management Enable Reproducible Builds at Scale? Yes." at MSR'25, written with @zacchiro and @Zimm_i48 !

   -> If you're interested in bitwise reproducibility and NixOS, I wrote a #blog post summarizing the key takeaways for the article: https://luj.fr/blog/is-nixos-truly-reproducible.html

   -> I strongly advise curious readers to read our preprint: https://hal.science/hal-04913007

   🔁 Boosts would be much appreciated!

   • Embed this notice
     Stefano Zacchiroli (zacchiro@mastodon.xyz)'s status on Monday, 27-Jan-2025 23:04:23 JST Stefano Zacchiroli

     in reply to

     • Reproducible Builds
     • Dr. Théo Zimmermann ?️‍?

     @luj @Zimm_i48 Congrats Julien! Very important work for @reproducible_builds , @nix , and safety of free/#opensource software more broadly.

   • Embed this notice
     Luj (luj@chaos.social)'s status on Tuesday, 04-Feb-2025 02:21:33 JST Luj

     in reply to

     • Morten Linderud
     • Dr. Théo Zimmermann ?️‍?
     • Stefano Zacchiroli

     @Foxboron @zacchiro @Zimm_i48 very poor reproducibility rate (around 60%). Haskell even accounts for about 41% of all non-reproducible packages that we identified ! (About 47k non reproducible Haskell packages)

     Haelwenn /элвэн/ :triskell: likes this.
   • Embed this notice
     Luj (luj@chaos.social)'s status on Tuesday, 04-Feb-2025 02:21:34 JST Luj

     in reply to

     • Morten Linderud
     • Dr. Théo Zimmermann ?️‍?
     • Stefano Zacchiroli

     @Foxboron @zacchiro @Zimm_i48 On the question of the different ecosystems ingested by nixpkgs, we discuss in our RQ2 the disparities in reproducibility rates in function of ecosystem of provenance: we indeed remark that some large ecosystems (like Python for example) have excellent reproducibility rate and contribute positively in the overall nixpkgs reproducibility rate. On the other hand Haskell for example is also one of the largest ecosystem in nixpkgs and has 1/2

     Haelwenn /элвэн/ :triskell: repeated this.
   • Embed this notice
     Morten Linderud (foxboron@chaos.social)'s status on Tuesday, 04-Feb-2025 02:21:35 JST Morten Linderud

     in reply to

     • Dr. Théo Zimmermann ?️‍?
     • Stefano Zacchiroli

     @luj @zacchiro @Zimm_i48
     Another aspect of this problem is whether or not the 91% figure, which is not mentioned in the paper, is this bi-effect of the decade of work by the reproducible build effort?

     I'd be also interested if the 91% figure is inflated? nixpkgs consume entire package indexes where some ecosystems are more reproducible than others.

   • Embed this notice
     Morten Linderud (foxboron@chaos.social)'s status on Tuesday, 04-Feb-2025 02:21:38 JST Morten Linderud

     in reply to

     • Dr. Théo Zimmermann ?️‍?
     • Stefano Zacchiroli

     @luj @zacchiro @Zimm_i48

     And there are better citations for the improbable aspect of reliably reproducing nix packages from when the topic was originally discussed inn 2016.

     Only citing me is not giving me more fans!

   • Embed this notice
     Morten Linderud (foxboron@chaos.social)'s status on Tuesday, 04-Feb-2025 02:21:39 JST Morten Linderud

     in reply to

     • Dr. Théo Zimmermann ?️‍?
     • Stefano Zacchiroli

     @luj @zacchiro @Zimm_i48

     > Note that even projects that do monitor build reproducibility like Debian or Arch Linux only do it at a given point in time (when the package gets built) and not backward in time like we did in our experiment!

     Arch does have an archive going back to 2016~2018, and we are able to reliably reproduce packages from way back when. Fwiw.