non-deterministic closureinfo from same store #4840
Description
Describe the bug
Realizing the same closureinfo derivation twice leads to two different outputs.
That in and of itself is not that weird, because it can happen easily as long as the close includes any path that is not reproducible, what is weird however is it happening on the same host.
diff -ru /nix/store/v3gdspvjfp0ri4g8ahadcsrc26ra4lfx-closure-info/registration /nix/store/v3gdspvjfp0ri4g8ahadcsrc26ra4lfx-closure-info.check/registration
--- /nix/store/v3gdspvjfp0ri4g8ahadcsrc26ra4lfx-closure-info/registration 1970-01-01 01:00:01.000000000 +0100
+++ /nix/store/v3gdspvjfp0ri4g8ahadcsrc26ra4lfx-closure-info.check/registration 1970-01-01 01:00:01.000000000 +0100
@@ -665,8 +665,8 @@
/nix/store/57g080n6rmzzq93gm5m9yc87zvqlh4yi-glibc-2.32-46-bin
/nix/store/vaf92wc23m067q9aig6zjkcnjvrg768n-glibc-2.32-46
/nix/store/59ppqq4p0b8sgm2ka443lfv9jiv1p4s3-initrd-linux-5.10.37
-sha256:0h7393ar8cb7l7x8qk1anv9za7pw6cc6r15kr55m32bmwk4swbj0
-23039080
+sha256:0dcqf7i7piqh1098sabwrxx5c610fmfrnk6sbyi7j3nxv5gpnmnx
+23036720
9
/nix/store/14c4d94qg2qpriqqbi6jfk001h1j5514-keymap
@@ -1941,8 +1941,8 @@
/nix/store/vaf92wc23m067q9aig6zjkcnjvrg768n-glibc-2.32-46
/nix/store/xiwckad47jy5vfibqym7pmjjr30caw80-aws-c-common-0.5.5
/nix/store/dyc2fjjk7igl8vi6wnhn9y065zjgqcn9-kernel-modules
-sha256:0r9q39m85l1lfbbsp1jr8l2dr9szf5jxp75hp6aip3b0ihqa03rl
-6230104
+sha256:1hj3gfhkpdjs9ia62x8byfgcln61xd0hpyiy0599gqlhw566dc2p
+6230168
2
/nix/store/3pvvw40cddikklg56c6xzrjcy731djhj-linux-5.10.37
@@ -3120,8 +3120,8 @@
/nix/store/wq2b68y4hi98zzm8p1jk7ybf23ly56m0-coreutils-8.32
/nix/store/ylz1dd8ks4cfag4mwks5dqsfqs2lcyzq-glibc-locales-2.32-46
/nix/store/nrr6ddvxifp3bmlgrvmpsa2qp2kycir6-linux-5.10.37-modules-shrunk
-sha256:1s828j8p7hzqar707b7avkv59n099ia0pwnn7zg1wjlsr1zij1gi
-2095712
+sha256:0biw53s2mq31vz3zpcy6dc2r8ivzlxdrf1b2fw4zsyvkjbqfcqnw
+2091976
1
/nix/store/nrr6ddvxifp3bmlgrvmpsa2qp2kycir6-linux-5.10.37-modules-shrunk
diff -ru /nix/store/v3gdspvjfp0ri4g8ahadcsrc26ra4lfx-closure-info/total-nar-size /nix/store/v3gdspvjfp0ri4g8ahadcsrc26ra4lfx-closure-info.check/total-nar-size
--- /nix/store/v3gdspvjfp0ri4g8ahadcsrc26ra4lfx-closure-info/total-nar-size 1970-01-01 01:00:01.000000000 +0100
+++ /nix/store/v3gdspvjfp0ri4g8ahadcsrc26ra4lfx-closure-info.check/total-nar-size 1970-01-01 01:00:01.000000000 +0100
@@ -1 +1 @@
-1201968216
+1201962184The reason I care about this is because it leads to NixOS tests failing on my hydra, because of hash mismatches in nixos-install (because the nix.db in the test is generated with nix-store --load-db from that registration file). In this case:
copying path '/nix/store/dyc2fjjk7igl8vi6wnhn9y065zjgqcn9-kernel-modules' to 'local'...
hash mismatch importing path '/nix/store/dyc2fjjk7igl8vi6wnhn9y065zjgqcn9-kernel-modules';
wanted: sha256:0r9q39m85l1lfbbsp1jr8l2dr9szf5jxp75hp6aip3b0ihqa03rl
got: sha256:1hj3gfhkpdjs9ia62x8byfgcln61xd0hpyiy0599gqlhw566dc2p
Here, sha256:0r9q39m85l1lfbbsp1jr8l2dr9szf5jxp75hp6aip3b0ihqa03rl is the hash that this path has on cache.nixos.org and sha256:1hj3gfhkpdjs9ia62x8byfgcln61xd0hpyiy0599gqlhw566dc2p is what is in the nix store on that machine.
Steps To Reproduce
I'm not quire sure, to be honest. I guess…
- Have a store path in your local store and also be available on
cache.nixos.org, while both have different contents - Run
closureInfo(orexportReferencesGraph, same thing) on it - Observe that it sometimes returns the one and sometimes the other
Expected behavior
I expect exportReferencesGraph to return the same thing twice when run on the same host twice. Especially if it's about the narhash of a path that's locally available.
nix-env --version output
nix-env (Nix) 2.4pre20210503_6d2553a