Merge pull request 'docs: clear PR #28 's docs-drift catalog entries' (#31 ) from chore/docs-catalog-2026-09-23 into main

docs: clear PR #28 's docs-drift catalog entries
Marks the token grant as verified now that #28's own merge ran release-tag and moved v1 to that commit (confirmed against the CI status API and the v1 tag on origin), and rewords the scenario-9 stranding framing to "a run that deferred and left no newer run behind it" rather than a concurrency-group cancellation ci.yaml no longer allows. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UkSjXXtU6JYcN2vPntWhfb
2026-09-23 03:02:39 +00:00 · 2026-09-22 21:53:32 -05:00 · 2026-09-22 23:12:59 +00:00 · 2026-09-22 16:14:41 -05:00 · 2026-09-22 15:01:05 -05:00 · 2026-09-22 14:01:28 -05:00
9 changed files with 620 additions and 68 deletions
@@ -101,3 +101,36 @@ jobs:
      # file.
      - name: Selftests
        run: bash scripts/selftest.sh
+
+  release-tag:
+    name: move v1 to main
+    # `needs: selftest` is what makes this "after the gate is green": a failed
+    # selftest skips this job, so v1 never advances onto a broken build. The
+    # `if:` restricts it to an actual push to main.
+    #
+    # No job-level `concurrency:` -- the lease in release-v1.sh already keeps
+    # v1 from moving backwards, and release-sweep.yaml picks up anything this
+    # job defers or misses.
+    needs: selftest
+    if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/main' }}
+    runs-on: ubuntu-latest
+    timeout-minutes: 2
+    # Requests write access from the run's built-in token -- whether that
+    # grant actually lets it push here is unobserved until the first merge
+    # (see README's Versioning section). Without this the checkout below
+    # still succeeds -- it's the push that would be rejected, which is a red
+    # job, not a silent no-op.
+    permissions:
+      contents: write
+    steps:
+      # Full history, so the ancestry checks in release-v1.sh can see how
+      # this commit relates to v1.
+      - uses: actions/checkout@v4
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}
+          fetch-depth: 0
+
+      # Releases this run's own commit only while it is still main's tip, and
+      # only forward -- see release-v1.sh.
+      - name: Move v1 to this commit if it is still main's tip
+        run: bash scripts/release-v1.sh merge "${{ github.sha }}"
@@ -0,0 +1,69 @@
+name: Release sweep
+
+# Keeps v1 from lagging main when ci.yaml's release-tag job defers or never
+# runs. Each tick either finds v1 already covering main's tip and exits, or
+# gates the tip exactly as ci.yaml's selftest job does and moves v1 to it. A
+# red run here means v1 is behind a main that fails its gate.
+#
+# Gitea registers schedules from the default branch only, so this fires once
+# it is on main.
+on:
+  schedule:
+    - cron: '*/15 * * * *'
+
+# A tick that arrives while another is still gating waits behind it rather
+# than gating the same tip twice.
+concurrency:
+  group: release-sweep
+  cancel-in-progress: false
+
+jobs:
+  sweep:
+    name: move v1 to main if it lags
+    runs-on: ubuntu-latest
+    timeout-minutes: 25
+    permissions:
+      contents: write
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}
+          fetch-depth: 0
+
+      # A scheduled run's github.sha is main as of the last push, not
+      # necessarily its tip, so the tip is read here instead.
+      - name: Check whether v1 lags main
+        id: check
+        run: bash scripts/release-v1.sh sweep-check
+
+      # Everything below runs only when v1 lags, and gates the tip itself,
+      # not the commit this run was created from.
+      - name: Check out main's tip
+        if: steps.check.outputs.needed == 'true'
+        run: git checkout -q --detach "${{ steps.check.outputs.tip }}"
+
+      - name: Install shellcheck
+        if: steps.check.outputs.needed == 'true'
+        uses: taiki-e/install-action@v2
+        with:
+          tool: shellcheck
+
+      # Same toolchains, same order, as ci.yaml's selftest job.
+      - name: Install Rust nightly
+        if: steps.check.outputs.needed == 'true'
+        uses: dtolnay/rust-toolchain@nightly
+      - name: Install Rust toolchain
+        if: steps.check.outputs.needed == 'true'
+        uses: dtolnay/rust-toolchain@stable
+
+      - name: shellcheck
+        if: steps.check.outputs.needed == 'true'
+        run: shellcheck -x --source-path=scripts scripts/*.sh
+
+      - name: Selftests
+        if: steps.check.outputs.needed == 'true'
+        run: bash scripts/selftest.sh
+
+      - name: Move v1 to the gated tip
+        if: steps.check.outputs.needed == 'true'
+        run: bash scripts/release-v1.sh push "${{ steps.check.outputs.tip }}" "${{ steps.check.outputs.v1 }}"
@@ -247,8 +247,8 @@ not collapsed into one:
  pass still deciding about one is never mistaken for a pass that died holding
  it. That settle window is a bound rather than a construction, and it is the
  only part of this that is. Until the rest of it was structural it was merely
-  policy — snapshots belong to protected refs, protected refs are never
-  eviction candidates — a property held by vigilance rather than by
+  policy — snapshots belong to protected refs, and a protected ref's snapshot
+  is never an eviction candidate — a property held by vigilance rather than by
  construction.
 - **Every other way the source can change mid-clone is detected, not
  prevented.** A `seed-fallback-dir` pointing at a directory something else
@@ -271,22 +271,45 @@ not collapsed into one:
 available to the clone that follows. Caches for branches that are DEAD are
 removed unconditionally; then, only if free space is under the requirement,
 live caches are evicted oldest-first; then, as a last resort, this run's own
-cache. Protected refs, the source this run is about to clone, and any cache
-held open by a running job are never candidates. Within the pressure pass,
-`target-*` directories are evicted before `snapshot-*` ones — the reverse of
-the obvious order, because a snapshot is hardlinked to everything cloned from
-it, so removing one frees almost no real bytes while costing every future PR
-its warm start.
+cache. A protected ref's *snapshot*, the source this run is about to clone,
+and any cache held open by a running job are never candidates in the pressure
+or self-clear passes. A protected ref's own *target* dir is an ordinary
+pressure-pass candidate, since it is a convenience cache the publisher's next
+run reseeds from the snapshot — but it is excluded from the liveness pass
+alone, because a branch's tip is trivially an ancestor of itself, and without
+that exclusion the merged-branch signal would read a publisher's own target
+dir as merged into itself and delete it every run, unconditionally. Within the
+pressure pass, `target-*` directories are evicted before `snapshot-*` ones:
+losing a target dir is cheap for exactly that reseeding reason, while
+evicting a snapshot forces every subsequent PR to start cold and, measured on
+the live volume (daniel/zemyna#1073), frees real disk rather than the
+near-nothing a shared-inode hardlink clone would suggest — `publish-snapshot.sh`
+unshares every executable after its `cp -al`, and executables are most of the
+tree by bytes.

-**A branch is dead in two ways, and the second is the one that reclaims
-anything here.** Gitea keeps a PR's branch after the merge unless the repo
-opts into delete-on-merge, so `git ls-remote` reports merged branches forever
-and "gone from origin" fires for almost nothing. The second signal is
+**A branch is dead in two ways, and neither signal makes the other
+redundant.** The first is that the branch is gone from origin. The second is
 ancestry: a branch still on origin whose tip is an ancestor of a protected
 branch's tip holds no commit that branch does not, so its cache will never be
-read again and goes in the same pass. On zemyna's volume that is the
-difference between reclaiming nothing and reclaiming a 40 GB directory per
-merged PR (gitdan-actions#20).
+read again and goes in the same pass.
+
+**Turn delete-on-merge on** (`default_delete_branch_after_merge`, per repo) —
+it is the setting this scheme is cheapest under, because a deleted branch is
+decidable from `ls-remote` alone, with no checkout, no objects and no walk.
+The ancestry signal is what covers the rest, and the rest is not a corner:
+
+- **Every branch merged before the setting was turned on.** They stay on
+  origin forever; nothing retroactively deletes them. zemyna carried 48 of
+  them at the time the setting was enabled, and ancestry is the only thing
+  that reclaims a cache dir belonging to any of them.
+- **Every merge the deletion declines or fails.** Gitea's delete is
+  best-effort and silent: it declines for a protected branch and for one
+  another open PR still uses, and an API merge that omits the flag — which
+  `tea pulls merge` does — simply never asks.
+- **Repos that have not enabled it**, which is the default.
+
+That is the difference between reclaiming nothing and reclaiming a 40 GB
+directory per merged PR on a full volume (issue 20).

 Ancestry is answered from the commits in the job's own checkout, so it is only
 answered where they are there to answer it — and "cannot tell" is never folded
@@ -460,7 +483,7 @@ directory; the line just doesn't say which.
 |---|---|---|
 | `cache-root` | `/cache` | mount point of the persistent volume inside the job container |
 | `cache-lineage` | *(empty)* | one directory level under `cache-root`, for a second job building the same ref for a different target or profile — see [Multiple jobs in one workflow](#multiple-jobs-in-one-workflow) |
-| `protected-branches` | `dev main` | refs that publish snapshots and are never evicted |
+| `protected-branches` | `dev main` | refs that publish snapshots, whose snapshots are never evicted (their target dirs are ordinary pressure-pass candidates) |
 | `min-free-percent` | `0` | an ADDITIONAL free-space floor, as a percentage of the volume. The gate is derived per run from what the seed is about to clone; this only ever raises it |
 | `restore-mtimes` | `true` | restore tracked-file mtimes from git history |
 | `prune` | `true` | run the eviction pass — before the seed, so what it frees is available to the clone |
@@ -611,13 +634,39 @@ entry, another permission — is a `v2`, not a `v1` move. Everything else moves
 `v1`: correctness fixes, new optional inputs, and anything internal to
 `scripts/`.

-**Moving the tag is a release step, and it is the operator's.** Merging to
-`main` ships nothing to anybody. `v1` is a lightweight tag and does not follow
-a branch, so until it is re-pointed every consumer keeps fetching the commit it
-already named, whatever `main` now says. The gap is deliberate: re-pointing
-`v1` changes what another repository's CI executes on its next run, so it is a
-decision taken once, knowingly, after the merge — never something a merge does
-by itself.
+**Moving the tag is automatic, gated on the same build that gates a PR.** Two
+jobs move it, both through `scripts/release-v1.sh`, both with the run's
+built-in `GITHUB_TOKEN`:
+
+- **`release-tag`** in `.gitea/workflows/ci.yaml` runs on every push to
+  `main`, `needs: selftest`, and moves `v1` to that run's own commit — but only
+  while that commit is still `main`'s tip. A run whose merge has already been
+  overtaken defers, as a successful no-op, rather than release a commit it
+  never gated.
+- **`release-sweep.yaml`** runs every 15 minutes. When `v1` already points at
+  `main`'s tip or a descendant of it, it exits after a checkout and one
+  comparison. Otherwise it runs the same shellcheck and selftests against the
+  tip and moves `v1` there only if they pass.
+
+Both jobs request `contents: write` on the run's built-in token, and that
+grant is now **verified**: PR #28's own merge ran `release-tag` successfully
+and moved `v1` to that merge's commit. The grant is still capped by the
+repo's and owner's maximum token permissions, and branch/tag protections on
+`v1` can't be read without admin access. A rejected push is a red job, not a
+silent no-op.
+
+So `v1` trails a green `main` by at most about one sweep interval plus one
+selftest run, and **a `main` that fails its gate shows up as a red sweep on every
+tick until it is fixed** — as does a push the token is not allowed to
+make. Both jobs push with `--force-with-lease` on the `v1` they read, so
+neither can move `v1` backwards over the other; a job that loses the lease to
+a newer `v1` finishes green.
+
+This used to be a manual step, treated as a deliberate release decision taken
+once, knowingly, after the merge — in practice it was still forgotten
+(gitdan-actions#27): PR #25 merged to `main` and `v1` stayed on the previous
+release until someone asked whether it had moved. The manual form below is
+still the recovery path, for when neither job can push:

 ```bash
 git fetch origin
@@ -628,9 +677,8 @@ git ls-remote --tags origin v1     # must equal git rev-parse origin/main

 **Downstream** are emowheel, which pins `cargo-cache@v1` and
 `cargo-cache-publish@v1` across its CI workflow, and zemyna, migrating to the
-same pin. Both pick a move up on their next run with no change on their side,
-which is the whole point of the moving pointer and also the reason the move is
-not automatic.
+same pin. Both pick a move up automatically on their next run with no change
+on their side, which is the whole point of the moving pointer.

 ---

@@ -702,6 +750,7 @@ change here reaches all of them at once. That is what the gate is for.
 | `seed-target-dir-selftest.sh` | seed-source preference, lock-file stripping, two jobs racing on one cache key, **and one scenario per check a hardlink clone is validated against**: a source rotated wholesale, a subtree silently lost from the walk, a copy that reports failure over a tree both other checks read as whole, and a source identity that resolved at neither end — plus a staging tree that could not be privately owned being discarded rather than published, and the publisher's log showing it waited on the consumer's own reader-lock marker before reclaiming a rotated snapshot |
 | `publish-snapshot-selftest.sh` | the atomic swap, that a live consumer survives a republish, and the publisher's side of the rotation race: deferred reclamation under a live reader, and its sweep once the reader is gone |
 | `prune-cache-selftest.sh` | liveness in both its forms — a branch deleted from origin, and one still on it whose tip is already merged — plus protection, locking, eviction order, self-clear, **that a cache a job claims *inside* the check-to-unlink window survives it**, and that a requirement derived from the clone's mutable set evicts exactly enough and then fails rather than under-delivering. Against a real scratch `origin`, including a genuinely shallow clone of it and a `df` that answers from the fixture's own size, since a fixed one cannot show a pass stopping |
+| `release-v1-selftest.sh` | that `v1` reaches `main`'s tip only through a gate and never moves backwards: the sweep's no-op, tag and red-gate cases, the stranded-defer trace the sweep exists to recover, each lost-lease outcome — a newer `v1` skipped cleanly (with a control showing an unleased push steps it back), an older one retried, an unrelated one and a server rejection red — and a `v1` hand-placed on an unrelated commit before any push is ever attempted, also red. Against a real scratch `origin`; the other writer is sequenced between check and push, not raced |
 | `restore-mtimes-selftest.sh` | the merge hazard and the watermark that closes it, including the two-jobs-one-namespace case. Needs a real compiler. |

 Every suite runs the actual script, not a reimplementation of its logic, and
@@ -24,7 +24,9 @@ inputs:
    default: ''
  protected-branches:
    description: >-
-      Space-separated refs that publish snapshots and are never evicted.
+      Space-separated refs that publish snapshots. A protected ref's
+      snapshot is never evicted; its own target dir is an ordinary
+      pressure-pass candidate, reseeded from the snapshot on its next run.
      These are the branches PR caches layer over.
    required: false
    default: 'dev main'
@@ -11,8 +11,13 @@
 #      Waiting for pressure to notice means paying for dead caches until then.
 #   2. LIVE BRANCH SURVIVES despite being OLDER than the dead one — liveness,
 #      not age, is what decides pass 1.
-#   3. PROTECTED REFS NEVER EVICTED under forced disk pressure, even when
-#      their caches are the oldest on disk and would rank first for LRU.
+#   3. A PROTECTED REF'S SNAPSHOT NEVER EVICTED under forced disk pressure,
+#      even when it is the oldest on disk and would rank first for LRU — its
+#      own TARGET dir is an ordinary candidate and goes (gitdan-actions#24).
+#  3b. AND A PROTECTED REF'S TARGET DIR SURVIVES PASS 1 ANYWAY: its tip is
+#      trivially an ancestor of itself, so the merged-branch signal must never
+#      be allowed to evaluate it, or pass 1 — unconditional, not gated on
+#      pressure — would delete it every run.
 #   4. LOCKED CACHE PROTECTED even when dead, old, and under pressure — and
 #      the pass's closing summary agrees with the decline it just logged,
 #      rather than reporting that it found nothing.
@@ -28,7 +33,10 @@
 #      so evicting it frees almost nothing while costing every future PR its
 #      warm start.
 #   8. SELF-CLEAR REPORTS LOUDLY to the job summary, not just a log warning.
-#   9. OWN CACHE NEVER EVICTED by a sibling pass.
+#   9. OWN CACHE NEVER EVICTED by a sibling pass, genuinely under pressure —
+#      against a real, shrinking `df` (gitdan-actions#26). Checks CONTENTS,
+#      not just existence, so pass 3's self-clear can't mask a missed
+#      pass-2 guard.
 #  10. SCOPED TO THE CACHE ROOT — a decoy outside it (standing in for another
 #      project's volume) is never touched.
 #  11. A LIVE READER MARKER PROTECTS A CACHE the same way a lock file does — a
@@ -46,10 +54,10 @@
 #      empty a tree its owner may still restore under a live cache name. Its
 #      fixture is an OLD directory renamed a moment ago — production's shape,
 #      and what lets it tell the two timestamps apart.
-#  15. A MERGED-BUT-UNDELETED BRANCH IS DEAD TOO. This forge keeps branches
-#      after merge, so `ls-remote` reports them forever and scenario 1's
-#      signal never fires for them — which is how three 40 GB caches sat on a
-#      full volume until somebody removed them by hand (gitdan-actions#20). A
+#  15. A MERGED-BUT-UNDELETED BRANCH IS DEAD TOO. A branch the forge did not
+#      delete at merge stays on `ls-remote` forever, so scenario 1's signal
+#      never fires for it — which is how three 40 GB caches sat on a full
+#      volume until somebody removed them by hand (gitdan-actions#20). A
 #      branch whose tip is an ancestor of a protected branch's tip is pruned
 #      like a deleted one; an unmerged branch beside it is not.
 #  16. AND "CANNOT TELL" IS STILL NOT DEATH, at both granularities: a branch
@@ -139,17 +147,84 @@ assert_kept "$root/target-$LIVE" "live branch survives despite an older marker t
 assert_log "no matching branch on origin" "eviction reason reported"

 echo
-echo "=== 3: protected refs never evicted under forced pressure ==="
+echo "=== 3: protected refs' SNAPSHOTS never evicted under forced pressure ==="
 reset_cache
 run_prune "1000000 1000"     # 0.1% free
-assert_kept "$root/target-$DEV" "dev's target dir survives disk pressure"
 assert_kept "$root/snapshot-$DEV" "dev's snapshot survives disk pressure"
-assert_kept "$root/target-$MAIN" "main's target dir survives disk pressure"
 assert_kept "$root/snapshot-$MAIN" "main's snapshot survives disk pressure"
+# Their TARGET dirs are ordinary candidates and go under the same pressure —
+# gitdan-actions#24: protecting them starved every second branch of room to
+# seed. Asserted here (gone, not kept) so this scenario still red-proves the
+# snapshot half if a future change reintroduces target protection.
+assert_gone "$root/target-$DEV" "dev's target dir is an ordinary pressure-pass candidate"
+assert_gone "$root/target-$MAIN" "main's target dir is an ordinary pressure-pass candidate"

 echo
-echo "=== 9: own cache never evicted by a sibling pass ==="
-assert_kept "$root/target-$OWN" "this run's own cache survives"
+echo "=== 3b: a protected ref's target dir is NOT pruned by pass 1's liveness ==="
+# The regression this fix could introduce and the selftest above cannot see:
+# a protected branch's tip is trivially an ancestor of itself, so once its
+# target dir stopped being excluded from pass 1 altogether, is_merged_dead
+# read it as "merged into itself" and pass 1 — unconditional, not gated on
+# pressure — deleted it on every single run. Plenty of free space, so only
+# pass 1 can be responsible for anything gone here.
+reset_cache
+run_prune "1000000 900000"   # 90% free: no pressure at all
+assert_kept "$root/target-$DEV" "dev's target dir survives pass 1 despite being its own ancestor"
+assert_kept "$root/target-$MAIN" "and so does main's"
+if grep -q "merged into" "$scratch/log"; then
+  fail "a protected ref's own target dir was evaluated by the merged-branch signal at all"
+fi
+ok "no protected ref's own target dir reaches the merged-branch check"
+
+echo
+echo "=== 9: own cache survives a sibling pass genuinely under pressure ==="
+# A real, shrinking `df` (the scenario-17 pattern), not CACHE_DF_OVERRIDE:
+# eviction has to actually free space for "pressure eases once enough is
+# freed" to mean anything. MIN_FREE_PCT=0 and a clone-headroom floor (not
+# the percentage floor) drive the requirement, so the requirement is an
+# exact, chosen KB rather than a percentage of a volume size this fixture
+# would otherwise have to reverse-engineer.
+rm -rf "$root"; mkdir -p "$root"
+blob_kb=4096
+mkdir -p "$root/target-$OWN"
+head -c $((blob_kb * 1024)) /dev/zero > "$root/target-$OWN/blob"
+touch -d '2020-01-01' "$root/target-$OWN/.cache-last-used"
+mkdir -p "$root/target-$LIVE"
+head -c $((blob_kb * 1024)) /dev/zero > "$root/target-$LIVE/blob"
+touch -d '2021-01-01' "$root/target-$LIVE/.cache-last-used"
+# The clone-headroom lookup's base-snapshot candidate — never read for its
+# content (CACHE_CLONE_HEADROOM_PERCENT=0 below), only for existing so the
+# floor alone becomes the requirement.
+mkdir -p "$root/snapshot-$DEV"
+
+real_du=$(command -v du)
+used9=$($real_du -sk "$root" | awk '{print $1}')
+cap9=$(( used9 + 2048 ))   # 2 MB to spare: under the requirement, over nothing else
+mkdir -p "$scratch/bin9"
+cat > "$scratch/bin9/df" <<DFEOF
+#!/usr/bin/env bash
+used=\$($real_du -sk "$root" | awk '{print \$1}')
+echo "Filesystem 1024-blocks Used Available Capacity Mounted-on"
+echo "fake $cap9 \$used \$(( $cap9 - used )) 50% $root"
+DFEOF
+chmod +x "$scratch/bin9/df"
+
+# Floor sits strictly between "0 evicted" (2048 KB free) and "1 evicted"
+# (2048 + blob_kb free) — satisfiable by evicting exactly one candidate.
+PATH="$scratch/bin9:$outer_path" \
+  CACHE_CLONE_HEADROOM_PERCENT=0 CACHE_CLONE_HEADROOM_FLOOR_KB=$(( 2048 + blob_kb / 2 )) \
+  GITHUB_STEP_SUMMARY="$scratch/summary" \
+  bash "$prune" "$root" "$root/target-$OWN" "dev main" 0 \
+    "$(cache_key unused-clone-probe)" "$DEV" "" \
+  > "$scratch/log" 2>&1 \
+  || { cat "$scratch/log"; fail "prune-cache.sh exited non-zero"; }
+assert_kept "$root/target-$OWN" "this run's own cache directory survives a genuinely pressured sibling pass"
+assert_kept "$root/target-$OWN/blob" "and its contents survive — not a recreated empty directory"
+assert_gone "$root/target-$LIVE" "the sibling is evicted instead, to make the same room"
+if grep -q 'clearing own' "$scratch/log"; then
+  fail "own cache was cleared by pass 3, not genuinely spared by pass 2 — this scenario proves nothing"
+fi
+ok "the requirement was met by pass 2 alone; pass 3 never ran"

 echo
 echo "=== 7: target dirs evicted before snapshots ==="
@@ -314,7 +389,7 @@ assert_log "may still be evicting it" "the deferral gives its actual reason"

 echo
 echo "=== 15: a merged-but-undeleted branch is dead too ==="
-# This forge keeps a PR's branch after the merge, so `ls-remote` reports it
+# A branch the forge did not delete at merge, which `ls-remote` then reports
 # forever. Built the way that happens: a branch merged into dev with a merge
 # commit, still pushed, beside one branched at the same point and NOT merged.
 git="git -c user.email=t@t -c user.name=t -c commit.gpgsign=false"
@@ -28,8 +28,9 @@
 #      removed UNCONDITIONALLY, not gated on free space. A directory for a
 #      branch nothing will build again is pure loss; waiting for disk pressure
 #      to notice means paying for it until then. Two signals make a branch
-#      dead, and the second exists because the first alone is inert on a forge
-#      that keeps branches after merge (gitdan-actions#20):
+#      dead, and the second exists because the first alone is inert wherever
+#      a merged branch stays on origin — the default, and still the outcome
+#      whenever delete-on-merge declines or is not asked (gitdan-actions#20):
 #
 #        DELETED — the branch is no longer on origin at all.
 #        MERGED  — the branch is still on origin, but its tip is an ancestor
@@ -66,21 +67,32 @@
 # physics, not an arbitrary GB number.
 #
 # EVICTION ORDER, and why it is the reverse of the obvious one: within the
-# pressure pass, `target-*` directories are evicted BEFORE `snapshot-*` ones.
-# A snapshot is a hardlink clone of a live target dir and of every consumer
-# cloned from it, so removing it frees almost no real bytes — its inodes stay
-# alive through those other links — while costing every future PR its warm
-# start. Evicting snapshots first would be nearly pure loss. Target
-# directories are where a branch's own divergent artifacts actually live, so
-# they are what freeing space means.
+# pressure pass, `target-*` directories are evicted BEFORE `snapshot-*` ones
+# (a publisher's own target dir included — see PROTECTED below). A publisher
+# branch's target dir is a convenience cache that its own next run reseeds
+# from the snapshot, so losing it is cheap; evicting the snapshot instead
+# forces every subsequent PR to start cold. Measured on the live volume
+# (daniel/zemyna#1073), that cold start is not the near-free move it looks
+# like either: a snapshot shares almost nothing with the target dirs cloned
+# from it, because publish-snapshot.sh unshares every executable after its
+# `cp -al` (cargo and the linker rewrite binaries in place, so a shared
+# original would corrupt under them), and executables are the great majority
+# of the tree by bytes. So a snapshot eviction is a real, large disk cost as
+# well as a cold-start one — reserved for last because both costs are larger
+# than a target dir's.
 #
 # Two exclusions every pass respects:
 #
-#   PROTECTED — the publisher branches' target and snapshot directories, and
-#   this run's own target dir, are never candidates in any pass. Evicting a
-#   publisher's snapshot doesn't free real disk (every open PR's clone keeps
-#   the data alive) but does force every subsequent PR to start cold, which is
-#   the entire benefit this scheme exists to deliver.
+#   PROTECTED — a publisher branch's SNAPSHOT directory, and this run's own
+#   target dir, are never candidates in the pressure or self-clear passes. A
+#   publisher branch's own TARGET dir is not protected there: it is an
+#   ordinary pressure-pass candidate, evicted oldest-first like any other,
+#   because nothing downstream depends on it surviving — the publisher's own
+#   next run reseeds it from the snapshot. It IS excluded from pass 1 alone
+#   (is_protected_from_liveness): a branch's tip is trivially an ancestor of
+#   itself, so without this exclusion the merged-branch signal would read a
+#   publisher's own target dir as "merged into itself" and pass 1 —
+#   unconditional, not gated on pressure — would delete it every run.
 #
 #   LOCKED — a directory carrying a .ci-lock-* marker younger than
 #   STALE_LOCK_SECONDS is held open by a running job, or named by a live
@@ -179,32 +191,52 @@ else
 fi

 declare -A protected_ns=()
+# A protected ref's own target dir is excluded from pass 1 ONLY (see
+# is_protected_from_liveness below), never from the pressure pass. It must
+# stay out of pass 1 for a reason that has nothing to do with disk: a
+# protected ref's tip is trivially an ancestor of itself, so without this
+# is_merged_dead would read dev's own target dir as "merged into dev" and
+# pass 1 — unconditional, not gated on pressure — would delete it on every
+# single run.
+declare -A protected_target_ns=()
 for ref in $PROTECTED_REFS; do
  suffix=$(cache_key "$ref")
-  protected_ns["target-${suffix}"]=1
  protected_ns["snapshot-${suffix}"]=1
+  protected_target_ns["target-${suffix}"]=1
 done

-# Prints why <dir> is off limits to every pass, or nothing when it is a
-# candidate. The reason is not decoration: it is what the failure report at
-# the bottom lists against each directory it kept while running out of space.
+# Prints why <dir> is off limits to the pressure/self-clear passes, or
+# nothing when it is a candidate there. The reason is not decoration: it is
+# what the failure report at the bottom lists against each directory it kept
+# while running out of space.
 #
-# SEED_SRC is the third exclusion and the one this script did not used to need.
-# The prune ran after the seed, so the source had already been cloned and the
-# reader marker over it was gone; running BEFORE the seed puts the directory
-# this run is about to read squarely in the candidate set, and pass 1 would
-# take it the moment its branch merged.
+# SEED_SRC is the third exclusion and the one this script did not used to
+# need. The prune ran after the seed, so the source had already been cloned
+# and the reader marker over it was gone; running BEFORE the seed puts the
+# directory this run is about to read squarely in the candidate set, and
+# pass 1 would take it the moment its branch merged.
 protected_reason() {
  local dir="$1" name
  name=$(basename "$dir")
  [ "$dir" = "$OWN_DIR" ] && { printf 'this run own cache'; return 0; }
  [ -n "$SEED_SRC" ] && [ "$dir" = "$SEED_SRC" ] && { printf 'the source this run is about to clone'; return 0; }
-  [ -n "${protected_ns[$name]:-}" ] && { printf 'a protected branch cache'; return 0; }
+  [ -n "${protected_ns[$name]:-}" ] && { printf 'a protected branch snapshot'; return 0; }
  return 1
 }

 is_protected() { protected_reason "$1" >/dev/null; }

+# Pass 1 (liveness) only: also excludes a protected ref's own target dir, for
+# the self-ancestor reason above protected_target_ns documents. The pressure
+# pass does not call this — is_protected is what it uses, via protected_reason
+# directly.
+is_protected_from_liveness() {
+  local dir="$1" name
+  is_protected "$dir" && return 0
+  name=$(basename "$dir")
+  [ -n "${protected_target_ns[$name]:-}" ]
+}
+
 # is_locked <dir> [name]
 #
 # `name` is the directory's own name for reporting and for the reader-marker
@@ -434,8 +466,11 @@ fi
 #
 # True when the branch this directory belongs to is still on origin but every
 # commit it holds is already on a protected branch — a merged PR whose branch
-# the forge did not delete, which is the case zemyna hits on every merge and
-# which the deleted-branch signal above can never see.
+# the forge did not delete, which the deleted-branch signal above can never
+# see. Enabling delete-on-merge narrows this to the branches merged before it
+# was enabled, the ones its deletion declines (protected, or used by another
+# open PR), and the merges that never ask (an API merge without the flag);
+# see README's eviction section.
 #
 # Memoised per tip because target-<key> and snapshot-<key> share one branch,
 # and because the "cannot tell" warning belongs to the branch rather than to
@@ -476,7 +511,7 @@ if [ "$LIVENESS_AVAILABLE" = "1" ]; then
  for dir in "$ROOT"/target-* "$ROOT"/snapshot-*; do
    [ -d "$dir" ] || continue
    name=$(basename "$dir")
-    is_protected "$dir" && continue
+    is_protected_from_liveness "$dir" && continue
    if [ -z "${live_ns[$name]:-}" ]; then
      why="no matching branch on origin"
      why_summary="branch no longer exists on origin"
@@ -0,0 +1,180 @@
+#!/usr/bin/env bash
+# Regression test for release-v1.sh against a scratch bare origin: v1 reaches
+# main's tip once it has been gated, never lands on an ungated commit, and
+# never moves backwards when two writers race (gitdan-actions#27).
+#
+# run_sweep mirrors release-sweep.yaml's step order -- check, gate only when
+# needed, push the gated tip leased on the v1 the check read -- with the gate
+# stood in for by a command, so a failing gate is a failing sweep.
+set -euo pipefail
+script_dir=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
+release="$script_dir/release-v1.sh"
+
+scratch=$(mktemp -d)
+trap 'rm -rf "$scratch"' EXIT
+pass_count=0
+fail() { echo "ASSERTION FAILED: $*" >&2; exit 1; }
+ok() { pass_count=$((pass_count + 1)); echo "PASS: $*"; }
+
+export GIT_AUTHOR_NAME=t GIT_AUTHOR_EMAIL=t@t GIT_COMMITTER_NAME=t GIT_COMMITTER_EMAIL=t@t
+unset GITHUB_OUTPUT
+
+# A fresh origin with main at one commit and v1 on it; dev pushes to main,
+# ci and ci2 are the runners' clones.
+fresh() {
+  rm -rf "$scratch/w"; mkdir -p "$scratch/w"
+  git init -q --bare "$scratch/w/origin.git"
+  git clone -q "$scratch/w/origin.git" "$scratch/w/dev" 2>/dev/null
+  commit_to_main >/dev/null
+  git -C "$scratch/w/dev" push -q origin HEAD:refs/tags/v1
+  git clone -q "$scratch/w/origin.git" "$scratch/w/ci"
+  git clone -q "$scratch/w/origin.git" "$scratch/w/ci2"
+}
+commit_to_main() {
+  git -C "$scratch/w/dev" commit -q --allow-empty -m "c$RANDOM"
+  git -C "$scratch/w/dev" push -q origin HEAD:refs/heads/main
+  git -C "$scratch/w/dev" rev-parse HEAD
+}
+origin_v1() { git -C "$scratch/w/origin.git" rev-parse -q --verify 'refs/tags/v1^{commit}' || true; }
+origin_tip() { git -C "$scratch/w/origin.git" rev-parse refs/heads/main; }
+in_ci() { (cd "$scratch/w/${CLONE:-ci}" && bash "$release" "$@"); }
+field() { sed -n "s/^$1=//p"; }
+
+run_sweep() {
+  local gate="$1" out tip v1
+  out=$(in_ci sweep-check)
+  [ "$(field needed <<<"$out")" = true ] || return 0
+  tip=$(field tip <<<"$out"); v1=$(field v1 <<<"$out")
+  "$gate" || return 1
+  in_ci push "$tip" "$v1"
+}
+
+echo "=== 1. sweep with v1 at the tip is a no-op ==="
+fresh
+before=$(origin_v1)
+out=$(in_ci sweep-check)
+[ "$(field needed <<<"$out")" = false ] || fail "a current v1 was reported as lagging"
+run_sweep false || fail "a current v1 ran the gate"
+[ "$(origin_v1)" = "$before" ] || fail "a no-op sweep moved v1"
+ok "v1 == tip: needed=false, gate not run, v1 unchanged"
+
+echo
+echo "=== 2. sweep with v1 ahead of the tip is a no-op ==="
+fresh
+ahead=$(git -C "$scratch/w/dev" commit-tree -p HEAD -m ahead 'HEAD^{tree}')
+git -C "$scratch/w/dev" push -q -f origin "$ahead:refs/tags/v1"
+out=$(in_ci sweep-check)
+[ "$(field needed <<<"$out")" = false ] || fail "a v1 descending from the tip was reported as lagging"
+ok "v1 descends from tip: needed=false"
+
+echo
+echo "=== 3. sweep with v1 behind and a passing gate tags the tip ==="
+fresh
+tip=$(commit_to_main)
+run_sweep true || fail "a passing sweep failed"
+[ "$(origin_v1)" = "$tip" ] || fail "v1 is $(origin_v1), not the gated tip $tip"
+ok "v1 behind, gate green: v1 -> tip"
+
+echo
+echo "=== 4. sweep with v1 behind and a failing gate goes red and tags nothing ==="
+fresh
+before=$(origin_v1)
+commit_to_main >/dev/null
+if run_sweep false; then fail "a sweep over a failing gate succeeded"; fi
+[ "$(origin_v1)" = "$before" ] || fail "a failing gate still moved v1"
+ok "v1 behind, gate red: sweep red, v1 unchanged"
+
+echo
+echo "=== 5. a lost lease to a newer writer is a clean skip, never a step back ==="
+fresh
+t1=$(commit_to_main)
+out=$(in_ci sweep-check); v1_read=$(field v1 <<<"$out")
+t2=$(commit_to_main)
+CLONE=ci2 in_ci merge "$t2" >/dev/null
+[ "$(origin_v1)" = "$t2" ] || fail "the merge job did not release its own tip"
+in_ci push "$t1" "$v1_read" || fail "a lease lost to a newer v1 went red"
+[ "$(origin_v1)" = "$t2" ] || fail "v1 went backwards from $t2 to $(origin_v1)"
+ok "older writer lost the lease: exit 0, v1 stays at the newer $t2"
+git -C "$scratch/w/ci" push -q -f origin "$t1:refs/tags/v1"
+[ "$(origin_v1)" = "$t1" ] || fail "control: an unleased push did not step v1 back"
+ok "control: the same push without the lease steps v1 back to $t1"
+
+echo
+echo "=== 6. a lease lost to an older writer retries and lands the newer commit ==="
+fresh
+t1=$(commit_to_main)
+t2=$(commit_to_main)
+out=$(in_ci sweep-check); v1_read=$(field v1 <<<"$out")
+git -C "$scratch/w/dev" push -q -f origin "$t1:refs/tags/v1"
+in_ci push "$t2" "$v1_read" || fail "a lease lost to an older v1 went red"
+[ "$(origin_v1)" = "$t2" ] || fail "v1 is $(origin_v1), not $t2"
+ok "v1 moved to an ancestor under us: retried, v1 -> $t2"
+
+echo
+echo "=== 7. a lease lost to an unrelated commit goes red ==="
+fresh
+tip=$(commit_to_main)
+out=$(in_ci sweep-check); v1_read=$(field v1 <<<"$out")
+stray=$(git -C "$scratch/w/dev" commit-tree -m stray 'HEAD^{tree}')
+git -C "$scratch/w/dev" push -q -f origin "$stray:refs/tags/v1"
+if in_ci push "$tip" "$v1_read" 2>/dev/null; then fail "a v1 moved sideways was accepted"; fi
+[ "$(origin_v1)" = "$stray" ] || fail "the stray v1 was overwritten"
+ok "v1 moved to a commit neither ahead nor behind: red, v1 untouched"
+
+echo
+echo "=== 8. a push rejected for another reason goes red ==="
+fresh
+tip=$(commit_to_main)
+mkdir -p "$scratch/w/origin.git/hooks"
+printf '#!/bin/sh\nexit 1\n' > "$scratch/w/origin.git/hooks/pre-receive"
+chmod +x "$scratch/w/origin.git/hooks/pre-receive"
+before=$(origin_v1)
+if in_ci merge "$tip" 2>"$scratch/err"; then fail "a rejected push reported success"; fi
+[ "$(origin_v1)" = "$before" ] || fail "v1 moved despite the rejection"
+grep -q 'not a lost lease' "$scratch/err" || fail "the rejection was not diagnosed as one: $(cat "$scratch/err")"
+ok "server rejection with v1 unmoved: red, not a lost lease"
+
+echo
+echo "=== 9. the merge job defers on a moved tip; the next sweep catches up ==="
+# The stranded trace: C1's job runs after C2 merged and defers — a run that
+# deferred and left no newer run behind it — and merges stop.
+fresh
+before=$(origin_v1)
+c1=$(commit_to_main)
+c2=$(commit_to_main)
+in_ci merge "$c1" >/dev/null || fail "the deferring merge job went red"
+[ "$(origin_v1)" = "$before" ] || fail "the merge job released a commit that was not the tip"
+run_sweep true || fail "the catch-up sweep failed"
+[ "$(origin_v1)" = "$c2" ] || fail "v1 is $(origin_v1), not the tip $c2"
+ok "C1 deferred, C2 never ran: the sweep moved v1 to $c2"
+
+echo
+echo "=== 10. the merge job releases its own tip, and creates a missing v1 ==="
+fresh
+tip=$(commit_to_main)
+in_ci merge "$tip" >/dev/null
+[ "$(origin_v1)" = "$tip" ] || fail "the merge job did not release the tip"
+git -C "$scratch/w/dev" push -q origin :refs/tags/v1
+tip=$(commit_to_main)
+in_ci merge "$tip" >/dev/null
+[ "$(origin_v1)" = "$tip" ] || fail "the merge job did not create an absent v1"
+[ "$(origin_tip)" = "$tip" ] || fail "main moved"
+ok "tip == gated sha: released, including onto an absent v1"
+
+echo
+echo "=== 11. a v1 hand-placed on an unrelated commit is never silently overwritten ==="
+# Unlike #7, nothing races here -- v1 already sits on the stray commit before
+# the very first push attempt, so force-with-lease sees exactly the value it
+# expects and would otherwise succeed outright.
+fresh
+stray=$(git -C "$scratch/w/dev" commit-tree -m stray 'HEAD^{tree}')
+git -C "$scratch/w/dev" push -q -f origin "$stray:refs/tags/v1"
+tip=$(commit_to_main)
+if in_ci merge "$tip" 2>"$scratch/err"; then fail "an unrelated hand-placed v1 was overwritten"; fi
+[ "$(origin_v1)" = "$stray" ] || fail "v1 moved off the hand-placed $stray"
+grep -q "$stray" "$scratch/err" || fail "the error did not name the stray v1: $(cat "$scratch/err")"
+grep -q "$tip" "$scratch/err" || fail "the error did not name the gated sha: $(cat "$scratch/err")"
+ok "hand-placed v1, unrelated to tip: red on the first push, v1 untouched"
+
+echo
+echo "release-v1-selftest: all $pass_count assertions passed"
@@ -0,0 +1,109 @@
+#!/usr/bin/env bash
+# Moves the floating `v1` tag forward to a gated commit on `main`, and never
+# backwards. Run from a clone whose `origin` is this repository.
+#
+#   release-v1.sh merge <gated-sha>     merge-triggered job: release <gated-sha>
+#                                       if it is still main's tip
+#   release-v1.sh sweep-check           scheduled sweep: report whether v1 lags
+#                                       main (tip=, v1=, needed= to
+#                                       $GITHUB_OUTPUT, or stdout without one)
+#   release-v1.sh push <gated-sha> <v1-as-read>
+#                                       scheduled sweep, after gating the tip
+#
+# Every push is leased on the v1 value the caller reasoned about. A lost lease
+# means another writer moved v1 first: that is a clean skip once v1 is at or
+# ahead of <gated-sha>, a retry against the new value while v1 is still behind
+# it, and a failure otherwise.
+set -euo pipefail
+
+MAX_ATTEMPTS=3
+
+fetch_main() {
+  git fetch -q origin +refs/heads/main:refs/remotes/origin/main
+  git rev-parse refs/remotes/origin/main
+}
+
+# Prints origin's v1 commit, or nothing when origin has no v1.
+fetch_v1() {
+  if [ -z "$(git ls-remote origin refs/tags/v1)" ]; then
+    git update-ref -d refs/release-v1/seen 2>/dev/null || true
+    return 0
+  fi
+  git fetch -q origin +refs/tags/v1:refs/release-v1/seen
+  git rev-parse 'refs/release-v1/seen^{commit}'
+}
+
+# True when v1 already covers <sha>: at it, or a descendant of it.
+covers() {
+  local sha="$1" v1="$2"
+  [ -n "$v1" ] && git merge-base --is-ancestor "$sha" "$v1"
+}
+
+push_leased() {
+  local sha="$1" expect="$2" now attempt
+
+  # force-with-lease only compares the ref's current value, not ancestry, so
+  # an unrelated v1 -- neither behind <sha> nor covering it -- would
+  # otherwise be silently overwritten on the very first push.
+  if [ -n "$expect" ] && ! covers "$sha" "$expect" && ! git merge-base --is-ancestor "$expect" "$sha"; then
+    echo "ERROR: v1 ($expect) is neither an ancestor of $sha nor at/ahead of it -- refusing to overwrite an unrelated v1" >&2
+    return 1
+  fi
+
+  for ((attempt = 1; attempt <= MAX_ATTEMPTS; attempt++)); do
+    if git push -q --force-with-lease="refs/tags/v1:$expect" origin "$sha:refs/tags/v1"; then
+      echo "v1 moved ${expect:-<absent>} -> $sha"
+      return 0
+    fi
+    now=$(fetch_v1)
+    if [ "$now" = "$expect" ]; then
+      echo "ERROR: push of v1 -> $sha rejected while v1 was still ${expect:-<absent>} -- not a lost lease" >&2
+      return 1
+    fi
+    if covers "$sha" "$now"; then
+      echo "lost the lease: another writer moved v1 to $now, at or ahead of $sha -- nothing to do"
+      return 0
+    fi
+    if [ -n "$now" ] && ! git merge-base --is-ancestor "$now" "$sha"; then
+      echo "ERROR: v1 moved to $now, which is neither behind nor ahead of $sha" >&2
+      return 1
+    fi
+    echo "lost the lease: v1 moved to ${now:-<absent>}, still behind $sha -- retrying"
+    expect="$now"
+  done
+  echo "ERROR: lost the lease on v1 $MAX_ATTEMPTS times running" >&2
+  return 1
+}
+
+cmd="${1:?usage: release-v1.sh merge <sha> | sweep-check | push <sha> <v1-as-read>}"
+shift
+case "$cmd" in
+  merge)
+    SHA="${1:?usage: release-v1.sh merge <gated-sha>}"
+    TIP=$(fetch_main)
+    if [ "$TIP" != "$SHA" ]; then
+      echo "main's tip ($TIP) is past this run's gated commit ($SHA) -- deferring; the sweep releases the tip"
+      exit 0
+    fi
+    V1=$(fetch_v1)
+    if covers "$SHA" "$V1"; then
+      echo "v1 ($V1) already at or ahead of $SHA -- nothing to do"
+      exit 0
+    fi
+    push_leased "$SHA" "$V1"
+    ;;
+  sweep-check)
+    TIP=$(fetch_main)
+    V1=$(fetch_v1)
+    if covers "$TIP" "$V1"; then NEEDED=false; else NEEDED=true; fi
+    echo "main=$TIP v1=${V1:-<absent>} release-needed=$NEEDED"
+    printf 'tip=%s\nv1=%s\nneeded=%s\n' "$TIP" "$V1" "$NEEDED" >> "${GITHUB_OUTPUT:-/dev/stdout}"
+    ;;
+  push)
+    push_leased "${1:?usage: release-v1.sh push <gated-sha> <v1-as-read>}" "${2-}"
+    ;;
+  *)
+    echo "release-v1.sh: unknown command '$cmd'" >&2
+    exit 2
+    ;;
+esac
@@ -13,7 +13,7 @@ script_dir=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
 FAST=0
 [ "${1:-}" = "--fast" ] && FAST=1

-FIXTURE_TESTS=(cache-root-selftest.sh seed-target-dir-selftest.sh publish-snapshot-selftest.sh prune-cache-selftest.sh)
+FIXTURE_TESTS=(cache-root-selftest.sh seed-target-dir-selftest.sh publish-snapshot-selftest.sh prune-cache-selftest.sh release-v1-selftest.sh)
 CARGO_TESTS=(hardlink-clone-selftest.sh restore-mtimes-selftest.sh)

 TESTS=("${FIXTURE_TESTS[@]}")