nar seralization hash mismatch

[domenkozar]

For /nix/store/00mzv9k8gjw0pgzla4ysraq5kdsjbkb2-libffi-3.4.2, there's a mismatch in hashes comparing to what nix-store --dump and nar streaming would generate.

This hints there's bug somewhere.

[domenkozar]

@sorki any ideas?

[domenkozar]

To reproduce:

$ nix-store -r /nix/store/00mzv9k8gjw0pgzla4ysraq5kdsjbkb2-libffi-3.4.2
$ nix-store --dump /nix/store/00mzv9k8gjw0pgzla4ysraq5kdsjbkb2-libffi-3.4.2 > store
$ nix-shell -p "haskellPackages.ghcWithPackages (pkgs: [pkgs.hnix-store-core])"
$ ghci
ghci> import System.Nix.Nar 
ghci> import System.IO
ghci> handle <- openFile "hnix" WriteMode
ghci> buildNarIO narEffectsIO "/nix/store/00mzv9k8gjw0pgzla4ysraq5kdsjbkb2-libffi-3.4.2" handle
$ diff -y <(xxd store) <(xxd hnix)

[domenkozar]

Most likely we need to sort the files/directories list or something trivial like that.

[s-and-witch]

Symlinks to the directories seems to be the reason of this issue. MRE:

$ cat default.nix
let
  pkgs = import <nixpkgs> {};
  in
derivation {
  system = "x86_64-linux";
  name = "test";
  builder = "${pkgs.bash}/bin/bash";
  args = [ ./builder.sh ];
  coreutils = pkgs.coreutils;
}
$ cat ./builder.sh
export PATH="$coreutils/bin"

mkdir $out
mkdir $out/1
echo 11 > $out/1/1

ln -s $out/1 $out/link1

$ nix-build  
this derivation will be built:
  /nix/store/67rnrqr2dwz9nqdv0awqrnvpv1gwjz2b-test.drv
building '/nix/store/67rnrqr2dwz9nqdv0awqrnvpv1gwjz2b-test.drv'...
/nix/store/bxgc56np7s0h35xl2dwmkcd65hxpxxqf-test
$ nix-store --dump /nix/store/bxgc56np7s0h35xl2dwmkcd65hxpxxqf-test > store
$ cabal repl hnix-store-core     
ghci> import System.Nix.Nar 
ghci> import System.IO
ghci>  withFile "hnix" WriteMode \handle -> buildNarIO narEffectsIO "/nix/store/bxgc56np7s0h35xl2dwmkcd65hxpxxqf-test" handle
$  diff -y <(xxd hnix-store-core/hnix) <(xxd store)    

[domenkozar]

@sorki any chance you can make a release?

[sorki]

Yep! Thanks for the fixes!

[ghostbuster91]

Most likely we need to sort the files/directories list or something trivial like that.

I have just spent whole day trying to debug an issue in my own implementation of mkStorePath. Then I found this comment and it turned out to be it!!

Thank you for posting it :100: