Artifact 11bd03015f17114489fd756d404c47acc2f67c067f45e0ba0e71f9cbfdcad150:
- File docs/RELEASE-NOTES-0.14.7.md — part of check-in [b6450a99d7] at 2026-11-15 11:27:53 on branch trunk — zit 0.14.7 (user: dmestas, size: 3816) [annotate] [blame] [check-ins using]
zit 0.14.7
zit 0.14.7 fixes three faults in how crosslink and diff read tree
check-ins, the ones whose file list is held in sub-manifests.
This release is from trunk. Stores and the wire are unchanged from 0.14.6. The record cache beside each store changes format and is upgraded in place on first open; nothing is rebuilt. See Compatibility.
A late sub-manifest is noticed
A tree check-in can arrive before one of its sub-manifests: a push split
in two, or a mirror that converges over several pulls. crosslink then
reports files unknown for it, which is right. But it went on reporting
that after the sub-manifest arrived, until crosslink --no-cache rebuilt
the cache. The cached record did not say what it was waiting for, so
nothing knew to compute it again.
A record that is unknown because an artifact is absent now names that
artifact, and is computed again on the next open once the store holds it.
The file-history index, which log PATH and ls --long read, kept the
same stale entry and is corrected with it.
crosslink and diff agree on a malformed tree
crosslink accepted a tree check-in whose sub-manifests overlap or are out
of name order and printed a change set for it, while diff --checkin
refused the same check-in. diff had a gap of its own: where a
sub-manifest shared with the parent ran past a new one beside it, both
commands accepted the tree.
Both now read a tree through one checked walk. crosslink reports such a
check-in as malformed. Its output for well-formed trees is unchanged.
One case is still accepted by both: a check-in that shares, with a malformed parent, a region broken inside a sub-manifest and not where two of them meet.
Missing and malformed are told apart
A caller could not tell a check-in that will read correctly after the next sync from one that never will.
crosslinkprints one of three lines, each beginning with the words it printed before:files unknown missing HASH: an artifact the walk needs is absent. It is computed again once HASH arrives.files unknown malformed: the tree breaks the walk's rules. It is not tried again.files unknown: any other cause.
diff --checkinsays "sub-manifest HASH is missing; a sync may bring it", or "check-in HASH has a malformed tree:" and the reason. HASH names whichever side's tree failed, which may be the parent's. Both exit 1, as before.
Compatibility
Stores, check-ins, the sync protocol and the daemon protocol version are unchanged.
The record cache in .zit-cache and the checkpoint file change format. An
older one is upgraded in place the first time 0.14.7 opens the store: every
known record is kept, and only records marked unknown are computed again,
which also corrects any that had gone stale. On a store of 1,300 tree
check-ins that first open took 9 ms, against about a second to rebuild the
cache in full.
Do not let an older zit open a store after 0.14.7 has: it does not read the new cache and rebuilds its own over it.
Source layout
Not user-facing. The command sources are now one file per verb under
src/cmd/, the end-to-end tests one file per verb under src/e2e/, and
the help text is in src/help.zig. The store adapter, receive-side
persistence and repository configuration moved from the command layer to
src/core/.
Qualification
The release manifest records native qualification against the exact tagged source archive on macOS ARM64, Linux x86-64 and Linux ARM64, using the pinned Zig 0.17.0 toolchain. Publication is gated on unit and end-to-end tests, Git differentials, deterministic simulation, seeded fuzzing, and Fossil interoperability checks on every distributed target.
For the basis of the conformance claim, see the 0.1.0 release notes.