Release v0.79.0

[max-sixty]

Cuts 0.79.0 from 75baa2cd3. Minor bump: cargo semver-checks reports five breaking library lints, so patch is disallowed pre-1.0.

Two commits — the changelog and version bump, plus one fix the release's data-loss review turned up.

The cache fix

has-added-changes and merge-add-probe are keyed on the branch and target tips with no TTL and no invalidation, so an entry answers for whatever code wrote it until its key recurs — which on a finished branch is never.

#4146 moved every diff wt parses onto plumbing so display config can no longer change what it reads. The porcelain reads it replaced wrote those two kinds, and under diff.relative or submodule.<name>.ignore they could record a branch's changes as absent. Nothing clears the cache on upgrade, so a poisoned entry outlives the fix: wt remove reads the old answer and deletes the branch without -D.

The exposure is narrow — it needs one of those two settings, and for diff.relative also a run from a subdirectory the branch's changes sit outside. The commits stay in the object store; it's the ref that goes. But nothing about the upgrade clears it, so for an affected repository the fix in #4146 does nothing until the cache is discarded. Reproduced against the fixed binary with a hand-written pre-upgrade entry:

✓ Removed branch feature (no added changes on main, ⊂)
branch still present? DELETED

CACHE_EPOCH in src/cache.rs versions what the cached values mean. ensure_epoch runs from cache_dir, the one funnel every read and write passes through, and discards a tree stamped with a different epoch. The lock spans the removal, not just the bookkeeping, because wt list fans the SHA-keyed reads over a thread pool. The sweep attempts every kind and returns the first error, so one undeletable directory doesn't strand the rest in read_dir order. Both wt config state clear and wt config state cache clear drop the stamp with the entries it describes, so a cleared tree re-stamps on the next command rather than vouching for whatever lands there next.

UNVERSIONED_KINDS exempts summary, ci-status and pr-number, so an upgrade doesn't charge a model call per branch and a forge round-trip for a diff fix. ci-status and pr-number hold the forge's own answer, so no change to how wt computes things can make them wrong. summary is exempt on weaker grounds, stated as such in the docstring: its key hashes the diff, but the prompt template and llm::prepare_diff's filtering sit downstream of that hash, so rewording the prompt does leave a stale summary — accepted because the value is display-only, and closing it means hashing the rendered prompt rather than bumping the epoch.

Known residual, deliberately deferred: the stamp records the last worktrunk to check a tree, not what wrote each entry, so a 0.78.0 binary run afterwards writes porcelain answers this one trusts. Closing that means putting the epoch in each entry's path, which invalidates by construction and costs the same in a later release. It is named in the CACHE_EPOCH comment.

TestRepo::assemble stamps the tree it builds, since a test that seeds .git/wt/cache/ by hand means entries this version produced. test_remove_branch_only_unmerged_ignores_cache_from_an_older_epoch covers the upgrade, a table-driven unit test covers all three stamp states and the exemption, and both clear handlers are pinned against dropping their clear_epoch call. Each assertion was checked under mutation rather than left to pass by construction.

Checks

cargo run -- hook pre-merge --yes green (4925 tests). This PR runs the full cross-platform matrix itself — full-tests on macOS, Linux, Windows and git 2.43.0, plus feature-powerset, nix-flake and the release targets — and nightly.yaml re-runs on the merged tip before the tag, since the green run on the cut-from tip predates the cache fix. Changelog verified over three passes against the diffs, the pre-change files, the rendered docs, and a forced cargo-semver-checks run.

This was written by Claude Code on behalf of max-sixty

🤖 Generated with Claude Code

[worktrunk-bot]

.git/wt/cache/.epoch is a new file Worktrunk creates, and it isn't in the FAQ's What files does Worktrunk create? inventory. Section 4 enumerates every .git/wt/ path down to .git/wt/cache/picker-preview/*.json and closes with "No files outside the six sections above", and CLAUDE.md's Data Safety section names that inventory as what new code changing this surface gets reviewed against. The row is also where to say the stamp deliberately survives wt config state clear — a leading-dot name has no extension(), so clear_json_files skips it — since the same section tells users that command "removes all repository data". Happy to push the row if you want it.

Three more inline: the failed-removal arm, what the whole-tree wipe costs for two of the kinds, and one changelog line that lost its meaning on the way out of the commit message.

[worktrunk-bot]

The epoch is the right shape for this — one funnel, one constant, and the residual named where the next bumper reads it. Two gaps, both about where the stamp's behavior did and didn't get wired through; inline.

[worktrunk-bot]

One new gap, on the funnel itself: cache_dir is now a mutation, and wt config state get's inspection reads go through it — so the command documented as pure inspection performs the wipe and then reports the tree it just emptied. Inline.

[worktrunk-bot]

The UNVERSIONED_KINDS exemption settles the summary/ci-status overcharge at the root rather than in the changelog wording, and clear_epoch in handle_state_clear_all closes the stronger of the two clears. Two things on the new code, both inline.

I ran the two new tests under mutation. The integration test holds up: drop the .epoch removal so the poisoned entry is trusted, and it fails with ✓ Removed branch feature-unmerged (no added changes on main, ⊂). The unit test's new exemption assertion does not — it stays green with the UNVERSIONED_KINDS check removed from discard_governed_kinds entirely.

The state get thread from the last pass narrows but doesn't close: summaries and CI status now survive an inspection read, while sha_cache::count_all still reports the counts the command left behind and # Reading vs resolving still says the aggregate get never persists.

[worktrunk-bot]

The per-kind sweep and the clear_epoch drop in handle_state_clear_all close what the last pass raised. Two gaps on the new code, both inline: the clear_epoch calls have nothing pinning them, and the exemption docstring states a rule its own third entry doesn't meet.

[worktrunk-bot]

The threads from the last four passes all close: the per-kind sweep carries on past a failure, the exemption test seeds off cache_root so the mutation fails, both clear handlers drop the stamp and both are pinned, and the UNVERSIONED_KINDS rule covers pr-number. Two new things, both inline — a changelog gap that will otherwise surface after merge, and one clause of that rule that summary doesn't meet.

On the data-loss gate for the release: discard_governed_kinds is a new fs::remove_dir_all that takes any non-exempt directory it finds, but its root is .git/wt/cache/ and src/cache.rs holds the only join("cache") in the tree — staging/promote and staging/relocate sit one level up at .git/wt/, out of reach. Everything it can delete is regenerable, so the hold doesn't apply.

[worktrunk-bot]

#4180 landed on main after the cut-from tip 75baa2cd3, so it is already an ancestor of this release commit and ships in 0.79.0 — but nothing here covers it. This is the release skill's step 12 case: the merge squashes onto whatever main tip exists at merge time, so the drift ships whether or not the changelog names it. Folding it in now costs a line; catching it at step 12 costs a follow-up squash PR and another pass over the coverage list.

Whether a Further-reading removal clears the bar for an entry is your call — the section it would join already covers doc-site sweeps of comparable weight. If it does:

Suggested change

	- **The homepage and README drop "Further reading"**: the section listed five external links; the one about Worktrunk itself — @DevOpsToolbox's video — moves into Next steps, beside the other things to do after installing. ([#4180](https://github.com/max-sixty/worktrunk/pull/4180))