Allow garbage-collecting a closure by lisanna-dettwyler · Pull Request #15236 · NixOS/nix

lisanna-dettwyler · 2026-02-14T06:13:19Z

Add option --skip-alive to nix store delete

nix store delete --recursive --skip-alive is morally equivalent to

for validPath in $(nix path-info --recursive foo); do
  nix store delete "$validPath" || true
done

~~Also adds utility function nix dev-env-path to facilitate use with devShells.~~ will be split off into separate PR

Resolves #7239

Motivation

This is the most requested unresolved feature in Nix. (https://github.com/NixOS/nix/issues?q=is%3Aissue%20state%3Aopen%20sort%3Areactions-%2B1-desc)

Context

#7239

Replaces #8417 because the author appears to have stopped working on it years ago.

This is a separate command to avoid the pitfall of nix store gc $(getSomePaths) resolving to nix store gc, which would unintentionally GC the entire store.

Add 👍 to pull requests you find important.

The Nix maintainer team uses a GitHub project board to schedule and track reviews.

xokdvium

Makes sense to me. Nice that we can hack it in without changing the actual worker protocol too much.

Ericson2314 · 2026-03-23T20:02:26Z

@roberth I am trying to remember the CLI guidelines. Should this be nix store delete --recursive? What other flavors of deleting things should we have?

lisanna-dettwyler · 2026-03-25T13:34:50Z

@roberth I am trying to remember the CLI guidelines. Should this be nix store delete --recursive? What other flavors of deleting things should we have?

Maybe nix store delete --recursive --ignore-alive? Or is that too similar sounding to --ignore-liveness, which has the opposite effect?

roberth · 2026-03-30T17:42:48Z

+        // This violates the convention that an empty `pathsToDelete` corresponds
+        // to the whole store, but deleting the whole store doesn't make sense,
+        // and `nix-store --delete` is a valid command that deletes nothing, so
+        // we need to keep it as-it-is.
+        return;


So basically we maintain the invariant that an added installable increases the number of candidates to try delete?
I agree it's the more expected and safer behavior.

roberth · 2026-03-30T17:45:22Z

+    input0=$(realpath "$TEST_ROOT/gc-root")
+    input1=$(nix build -f dependencies.nix input1_drv --no-link --print-out-paths)
+    input2=$(nix build -f dependencies.nix input2_drv --no-link --print-out-paths)
+    top=$(nix build -f dependencies.nix --no-link --print-out-paths)


Related

Make nix store delete ./result work #9475

Its implementation could drop the --no-link here.

No action required.

roberth · 2026-03-30T17:50:17Z


+        if (!options.pathsToDelete.empty() && options.action == GCAction::gcDeleteDead
+            && !conn.protoVersion.features.contains(WorkerProto::featureGcClosure)) {
+            throw Error("Garbage collecting a closure requested but it is not supported by the negotiated protocol");


The way I understand it, this is near-unreachable and only a nix-as-client logic error would trigger this?

Yeah, I don't think this is reachable currently and would only happen due to a future introduced logic error

Ericson2314 · 2026-04-10T19:37:03Z

        auto store = openStore();
        auto & gcStore = require<GcStore>(*store);
        options.action = dryRun ? GCOptions::gcReturnDead : GCOptions::gcDeleteDead;
+        options.pathsToDelete = GCOptions::WholeStore{};


Nice, I like how it forces us to be explicit!

This allows garbage collecting a closure. `nix store delete --recursive --skip-alive` is morally equivalent to ```bash for validPath in $(nix path-info --recursive foo); do nix store delete "$validPath" || true done ``` Resolves NixOS#7239 Co-authored-by: Théophane Hufschmitt <theophane.hufschmitt@tweag.io> Co-authored-by: Alexander Bantyev <balsoft@balsoft.ru> Signed-off-by: Lisanna Dettwyler <lisanna.dettwyler@gmail.com>

Ericson2314

All fixed now!

github-actions Bot added documentation new-cli Relating to the "nix" command with-tests Issues related to testing. PRs with tests have some priority store Issues and pull requests concerning the Nix store labels Feb 14, 2026

lisanna-dettwyler force-pushed the gc-closure-2 branch 2 times, most recently from 86012d3 to 2f34a72 Compare February 14, 2026 06:15

lisanna-dettwyler changed the title ~~Allow garbage-collecting whithin a closure~~ Allow garbage-collecting within a closure Feb 14, 2026

lisanna-dettwyler force-pushed the gc-closure-2 branch 3 times, most recently from 8f3db57 to 34bc281 Compare February 14, 2026 06:55

lisanna-dettwyler marked this pull request as ready for review February 14, 2026 07:39

lisanna-dettwyler requested review from Ericson2314 and edolstra as code owners February 14, 2026 07:39

lisanna-dettwyler force-pushed the gc-closure-2 branch 2 times, most recently from 6082386 to 38782d1 Compare February 16, 2026 18:48

lisanna-dettwyler force-pushed the gc-closure-2 branch 4 times, most recently from e9bb1d4 to 1b5a1b2 Compare March 18, 2026 20:19

xokdvium reviewed Mar 18, 2026

View reviewed changes

Comment thread tests/functional/gc-closure.sh Outdated

Comment thread src/libstore/gc.cc Outdated

Comment thread tests/functional/gc-closure.sh Outdated

lisanna-dettwyler force-pushed the gc-closure-2 branch from 1b5a1b2 to e93193b Compare March 18, 2026 20:42

xokdvium approved these changes Mar 18, 2026

View reviewed changes

lisanna-dettwyler force-pushed the gc-closure-2 branch 3 times, most recently from 02002bb to 395130a Compare March 30, 2026 17:33

lisanna-dettwyler changed the title ~~Allow garbage-collecting within a closure~~ Allow garbage-collecting a closure Mar 30, 2026

roberth reviewed Mar 30, 2026

View reviewed changes

roberth approved these changes Apr 8, 2026

View reviewed changes

lisanna-dettwyler force-pushed the gc-closure-2 branch from efeb271 to b93025d Compare April 9, 2026 22:19

Eveeifyeve approved these changes Apr 10, 2026

View reviewed changes