[avrabe]

Closes #1115.

What changed

rivet modify <ID> --set-release "" now removes the release: key from the artifact's mapping instead of writing release: "". The artifact becomes unassigned (backlog) in the sense rivet modify --help documents.

Why release: "" wasn't backlog

The old code path was yaml_quote_inline_scalar(new_release) → set_field(id, "release", &quoted). yaml_plain_scalar_needs_quoting returns true for an empty string by design, so an empty value quoted to "" and was written verbatim as release: "". That left a quoted-empty value in the mapping, which rivet query '(= release "")' still matches as a value, not as absence. The other apparent escape hatch — rivet release move <ID> <VERSION> — requires a version, so it had no backlog target either. The only working path was hand-editing the YAML, which is the escape hatch rivet's own guidance tells callers to avoid.

Scope — Fix 1 only

Issue #1115 sketched two suggested fixes:

1. --set-release "" (or --clear-release) removes the key. This PR lands that.
2. rivet release move <ID> --backlog, so the scope decision is a logged transition. Remains an open follow-up on friction: no way to move an artifact back to backlog — --set-release "" writes release: "" and release move needs a version #1115.

Keeping the diff focused on Fix 1 because it unblocks the reporter's immediate case (clearing six draft site artifacts out of a release they had blocked three times running on pulseengine.eu) with the smallest surface.

How each ask in #1115 is addressed

• "--set-release "" (or --clear-release) removes the key." — Done. When params.set_release is Some(""), modify_artifact_yaml now calls editor.remove_field(id, "release") instead of set_field. The release: line is gone from the output; the artifact becomes an unassigned mapping entry.
• CLI help documents the clear-to-backlog behaviour. — Done at rivet-cli/src/main.rs:1317-1321: the --set-release doc now reads "pass an empty value (--set-release "") to clear the release and move the artifact back to backlog."
• rivet release move <ID> --backlog. — Deferred; remains open on friction: no way to move an artifact back to backlog — --set-release "" writes release: "" and release move needs a version #1115 (Fix 2).

How the fix doesn't drag neighbours

remove_field reuses the same field_block_end extent helper the write path (set_field) uses, so a hypothetical multi-line release value (block scalar, nested block mapping) would be removed as a whole and would not orphan deeper-indented children into the next field's scope. This is the mirror of the set_field_inserts_after_trailing_provenance_block and set_field_inserts_after_folded_scalar_with_blank_line regression guards on the write path (#613 / #618).

Test plan

• cargo test -p rivet-core --lib yaml_edit:: — 55 tests pass, including the three new ones (set_release_to_empty_clears_the_field, set_release_to_empty_on_backlog_artifact_is_a_noop, set_release_to_empty_preserves_following_provenance_block).
• cargo test -p rivet-cli --test cli_commands — 247 tests pass (no regression in the end-to-end rivet modify surface).
• cargo fmt --all -- --check clean.
• cargo clippy -p rivet-core --lib -- -D warnings clean.
• cargo clippy -p rivet-cli --bins -- -D warnings clean.
• CI green on this branch.

Negative controls baked into the new tests

• set_release_to_empty_clears_the_field asserts !out.contains("release:") and that the key is absent from the parsed mapping. If a future refactor regressed to writing release: with an empty value (either quoted or bare), both halves would fail — matching the (= release "") matching behaviour cited in the issue.
• set_release_to_empty_preserves_following_provenance_block plants a provenance: block right after the release: line; if remove_field mis-calculated the field extent and drained the next field too, the provenance.created-by assertion would fail.

---

Generated by Claude Code

[temper-pulseengine]

⚠️ Outdated — this review was for an earlier revision. See the latest review below.

Automated review for PR #1120

pulseengine/rivet:fix/issue-1115-clear-release-to-backlog → pulseengine/rivet:main

Verdict: 💬 Comment

Summary: The provided code changes the behavior of setting a release field in an artifact YAML file. The old behavior wrote release: "" as a string, which caused issues with querying for artifacts that were unassigned but had an empty release value. The new behavior removes the release: field entirely when the new value is the empty string, ensuring that the artifact is correctly marked as unassigned.

Findings: 0 mechanical (rivet) · 2 from local AI model.

Findings (2):

1. src/mutate/mod.rs:964

               let quoted = yaml_quote_inline_scalar(new_release);
   

   The old behavior wrote release: "" as a string, which caused issues with querying for artifacts that were unassigned but had an empty release value.

2. src/mutate/mod.rs:964

               let quoted = yaml_quote_inline_scalar(new_release);
   

   The new behavior removes the release: field entirely when the new value is the empty string, ensuring that the artifact is correctly marked as unassigned.

---

Generated by a local AI model and post-validated against a strict JSON contract. Each finding includes the verbatim line being criticised — verify by reading the file at the cited location.

Reviewed at 773fd8e

[github-actions]

📐 Rivet artifact delta

No artifact changes in this PR. Code-only changes (renderer, CLI wiring, tests) don't touch the artifact graph.

[avrabe]

Playwright E2E failing on 773fd8e — standing down on this one as it is not this PR's.

The install step dies before any test body runs with:

Error: ERROR: Playwright does not support chromium on ubuntu26.04-x64

Same shape as #1109: GitHub's ubuntu-latest now resolves to Ubuntu 26.04 and the pinned Playwright release does not support it. The diff on this PR touches no tests/playwright/** and no workflow file, so this is not a regression from the --set-release "" change.

PR #1114 already has the fix (pin runs-on: ubuntu-24.04 on the Playwright job). Rather than wait on it to merge, I've ported the one-line change into this PR as 9c64d05 so CI can clear the install step here now — it no-ops once main carries #1114.

Not spending my one re-run on the pre-port commit (773fd8e); the new push supersedes it.

---

Generated by Claude Code

[temper-pulseengine]

Automated review for PR #1120

pulseengine/rivet:fix/issue-1115-clear-release-to-backlog → pulseengine/rivet:main

Verdict: 💬 Comment

Summary: The pull request is approved and has passed all the required checks.

Findings: 0 mechanical (rivet) · 1 from local AI model.

Findings (1):

1. .github/workflows/ci.yml:552

   # Stays on hosted ubuntu: `npx playwright install --with-deps` runs
   

   The workflow specifies that it should run on a hosted Ubuntu environment, which is appropriate for installing Playwright dependencies.

---

Generated by a local AI model and post-validated against a strict JSON contract. Each finding includes the verbatim line being criticised — verify by reading the file at the cited location.

Reviewed at 9c64d05

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[avrabe]

CI green on 9c64d05. All 25 non-skipped check runs completed success:

• Aggregate: CI Gate ✅
• Core: Clippy ✅, Format ✅, Test ✅, Proptest ✅, Miri (safety) ✅, MSRV ✅, Semver ✅, WASM ✅
• E2E: Playwright ✅ (install step clears on the ported ubuntu-24.04 pin)
• Supply chain: Cargo Deny ✅, Supply Chain (cargo-vet) ✅, Security Audit (RustSec) ✅
• Proofs: Rocq ✅, Verus ✅
• Other: Docs Check ✅, Zola export smoke ✅, Code Coverage ✅, Criterion Benchmarks ✅, Mutation Testing (rivet-core PR-diff) ✅, Mutation Testing (rivet-cli) ✅, VS Code Extension ✅, codecov/patch ✅ (all modified lines covered), Schema version bump ✅, Detect changed areas ✅, Artifact delta ✅

Skipped as expected (non-triggers for this diff): Mutation Testing matrix, Traceability hosted fallback, Fuzz Testing, Miri (full, nightly), Publish Test Results on Release.

The PR is waiting on reviewers.

---

Generated by Claude Code