Timeline

The propagating form marked every segment ABOVE a landed one closed too: in a stack those check-ins descend from the landed tip, so propagation followed the DAG straight into work that was still open. The landing records stayed open and correct; only 'branch' and 'leaves' lied, which is exactly where you look for what still needs doing.

Fossil's own propagating close is safe because it closes a check-in with no descendants. A landed segment is a leaf by the same-branch rule while still having children on the branch above it. check-in: d97b39b272 user: dmestas tags: strata

Adopting an index inserted every entry into a StringHashMap before answering anything: 255,167 keys on a 302MB store, to serve the one or two lookups a tree listing makes. Measured at 5.5ms of a 24ms open by building the map twice and taking the delta.

v3 appends entry_count little-endian u32 rids ordered by digest, and lookup binary-searches it -- ~18 comparisons against a quarter-million insertions. Opening tcl drops 24.0ms to 20.0ms; with the hardware crc32 change before it, 38.5ms to 20.0ms. Output byte-identical.

The directory carries RIDS and the entry records stay in scan order. Sorting the entries themselves would renumber every rid, and ancestry, phantom snapshots and the sync walk all read that numbering.

lookup consults the map first, then the directory: entries appended after adoption -- a tail resume past the index boundary, and every write on a scan-path store -- are only in the map. Every held-check that was reading index.get directly now goes through lookup, because intern runs during a tail resume and a miss there mints a duplicate entry for a digest already held.

A v2 sidecar is refused and rewritten rather than adopted, as v1 was: adopting one would leave lookup with neither a map nor a directory and answer null for held artifacts. Pinned by a test that downgrades a v3 blob and re-checksums it.

The tests pin what equality cannot: that the map is EMPTY after adoption and every digest still resolves, that a digest which is not held returns null rather than a neighbour -- the failure a binary search has and a map does not -- and that digests written out of sort order still resolve, so a directory that merely mirrored insertion order would fail.

Closes #145 closed check-in: 0d16bc966d user: dmestas tags: index-v3-145

Every landed branch stayed an open leaf, which was survivable one landing at a time and is not with stacks: a six-deep stack left seven leaves, and 'leaves' is how you find work that still needs doing.

accept now writes fossil's propagating 'closed' tag on the branch it merged. reject does not -- a rejected branch is one the author may revise and propose again, and closing it would decide that for them.

Best-effort and last, like the retargets: a leaf that stayed open is untidy, never incorrect, and the diagnostic names the command that fixes it. check-in: c9b0147381 user: dmestas tags: strata

Checksumming was half the cost of opening a repository: 18.8ms of a 37ms `ls` on a 302MB store, across the two sidecars (26.3MB derived-state image at 13.5ms, 9.7MB log index at 5.3ms). Both measured by doubling the work and taking the delta, so the number is the checksum and not its neighbours.

ARMv8's crc32b/crc32x implement CRC-32/ISO-HDLC -- the same generator this file is already defined in terms of, in the same reflected convention -- so this is an instruction-count change and never a value change, which is what the module doc says every edit here must be. tcl opens in 24.5ms against 38.5ms, output byte-identical.

aarch64 only, deliberately. x86_64's SSE4.2 crc32 instruction is CRC-32C/Castagnoli, a DIFFERENT generator; taking it because the mnemonic matches would orphan every store already written. x86_64 keeps the table path, which stays load-bearing rather than becoming a fallback nobody runs.

The new test recomputes CRC-32 as a plain bit-at-a-time recurrence, independent of update and its tables, and holds both paths to it at every length 0..1024 -- all eight residues mod 8, including the lengths below the fold width. The two paths differ only in where the fold stops and the tail begins, so a divergence would appear at particular residues and nowhere else, which is the shape a spot check misses. Chunked updates are pinned too: the store checksums payloads in 4 KiB pieces, so a path correct only for single-shot input would corrupt exactly the largest artifacts.

Refs #143 closed check-in: b82aacc805 user: dmestas tags: crc-hardware-143

A landing may target another branch that has its own open landing; that pair is a stack, read from the targets rather than recorded anywhere. No new card, no dependency field -- the edge is already implied.

Stacks land bottom-up. accept refuses a landing whose target has its own open landing, and accepting one retargets whatever was stacked on it onto the branch it just landed on. That is the whole restack: a metadata write per child, because a segment's branch already descends from the one below it, so after that lands the merge sees only the segment's own change.

Adds 'landing retarget', a target reply resolved by the same latest-wins rule as a state transition. ls indents a stack by depth; show names what a segment sits on, what sits on it, and when its base was rejected.

Verified six deep on a shared file, which is the depth the design claim was previously only reasoned about. check-in: 3767f1eafc user: dmestas tags: strata

Adopting a scan index duplicated every entry hash: 255,167 allocations and copies on a 348MB store to answer, typically, one lookup. The hashes are fixed-length slices of a buffer that was being freed at the end of the open, so the store now holds that buffer for its lifetime and the entries point into it. Opening tcl drops 44.0ms -> 37.5ms; smaller stores scale down with the index.

openReadOnlyIndexed takes the blob UNCONDITIONALLY -- on success the store frees it in deinit, on every refusal it frees there. The first cut transferred only on success, both call sites looked right, and the second one double-freed inside commit's autosync. No type and no unit test caught it; only e2e did. A conditional transfer makes every call site restate the rule.

Ownership is a prefix, not a per-entry flag: adopted entries come first and a tail resume appends owned ones after them, so borrowed_entries is the split point deinit and the batch rollback both read.

The unit test pins that the hashes lie INSIDE the held blob, not merely that they compare equal -- equal bytes prove nothing about where they live, and a future dupe would restore the cost with every assertion still green.

This is one part of #143. The CRC that issue names as the largest fixed cost measures 5.2ms of the 44ms (1,860 MB/s over the 9.7MB index), so it is not where the time goes and its integrity check should stay.

Refs #143 closed check-in: 1234842809 user: dmestas tags: index-borrow-143

The current side becomes the working tree, so a divergent path compares what the operator actually has rather than what the check-in says is on disk -- which is the disagreement B-merge-14 existed to avoid. A clean tree scans to no divergences and every side reads from the store exactly as before.

Rollback needs no new machinery: captureMergeImages already reads the worktree, so an abandoned merge restores the uncommitted bytes.

Also adds the fault-injection coverage the carried update path was missing: a carried update that fails mid-apply must restore the WORKING bytes, not the check-in's. Verified by mutation -- breaking the baseline overlay fails both tests. closed check-in: 0627425be0 user: dmestas tags: merge-dirty-tree

Both opened `.full`, so each invocation re-derived the entire repository view before reading anything. The cost tracked the STORE, not the work: a one-line diff on a 302MB store took 0.33s, and cost the same whether the diff was 13 lines or 4126. diff read two file lists -- exactly what ls reads at 0.04s -- and artifact never touched the derived view at all, resolving through logStore alone.

diff inherited `.full` by omission: it called Repository.open directly instead of going through main.openRepo, which is where every other read verb chooses a mode. Nothing failed, so nothing caught it.

fossil 112MB 0.06s -> 0.01s tk 240MB 0.14s -> 0.03s sqlite 236MB 0.25s -> 0.03s tcl 302MB 0.33s -> 0.04s

Output byte-identical on all four.

readpath_test pins which verbs take which path, and pins the comment justifying each deliberate `.full` -- timeline reads the merge mark off adjacency, annotate walks ancestry. An unexplained `.full` is indistinguishable from this bug. grep is named there as an open question rather than pinned, since it also wires --no-cache, which has to move with the mode.

The OpenOptions comment claimed ls, cat, info, branch and tag were `.full`. All five have been fast_read since #98; reading it was how this stayed hidden.

Fixes #141 closed check-in: e845821a65 user: dmestas tags: readpath-141

The whole message went into the H card and the body both, so a multi-paragraph -m minted a multi-paragraph heading and carried the same text twice. An H card is one token (§4.7.7). The split is at the first newline, not the first blank line, so a two-line message cannot mint a two-line title either.

ls and show read only a title's first line: landings minted before this are immutable, so a reader has to cope with them, not merely stop making more. A message with no title line is refused before the branch lookups and the store scan.

Fixes #138 closed check-in: 3e565c1caa user: dmestas tags: landing-title-138

A dirty tree is now three-way merged (current check-in, working file, target) rather than rejected, so a branch tip that advanced under uncommitted work can be joined. Previously update, merge and stash all refused that state, leaving commit --allow-fork as the only exit -- the guard that keeps forks deliberate was manufacturing them. The fork guard itself is unchanged.

applyCheckinWith scans the tree once and calls back for the merge policy; handing carries in instead would scan every clean update twice, which cost a 20k-file no-op update 63%.

Paths that cannot merge either way (type change, edited here / deleted there) are named and refused before anything mutates.

Fixes #135 closed check-in: 7fcc9b5ee1 user: dmestas tags: update-dirty-tree-135

landing show answers the same question but must find the record first, which is a full-store walk. A consumer with its own index already knows the record and both tips; --probe lets it ask about the pair and pay only for the merge. Release build on a 20MB store: 0.01s against 0.90s.

Writes nothing, never autosyncs, and exits 0 whatever the answer -- 'has conflicts' is an answer, not a failure, so a caller reads the line rather than the status. Being a read it takes -R, and then --into is required because there is no checkout to supply the other side.

The probe moves out of landing and next to the merge engine: one implementation, three callers. A real merge now refuses -R by name instead of failing as a discovery miss that tells you to pass the flag you just passed. closed check-in: aff1d04310 user: dmestas tags: merge-probe

Lines differing only in whitespace compare equal, so a reindent or a CRLF/LF normalization produces no hunks and drops the file from --brief entirely. Comparison only: a line that does survive is emitted with its original bytes.

--brief reads content under -w, which it otherwise never does, and runs the same unified() the hunk path runs so the two cannot disagree about whether a file changed. A no-hunk result now claims 'mode changed' only on real byte equality -- with -w it also means a difference that was ignored. closed check-in: 23784ef673 user: dmestas tags: diff-ignore-ws

landing hardcoded N text/plain with no way to say otherwise, overriding the text/x-fossil-wiki default of artifact-format 4.7.12 on every record it wrote. It now takes --mimetype and, minting a forum artifact, follows forum's rule.

The wider bug was in what landing copied. Verified against fossil's source: wiki (wiki.c:2084) and forum (forum.c:1784) write N only when the mimetype is present, non-empty and NOT the default; technote (event.c:284) writes it whenever non-empty, with no suppression. zit wrote it unconditionally for all three, so an explicit --mimetype text/x-fossil-wiki minted different bytes, and a different artifact id, than fossil produces from the same input.

The draft states the READ default, not when the card is written. Only the source shows the suppression, and only the source shows that the three types disagree -- so tests pin the divergence rather than tidying it away. closed check-in: baa15c3362 user: dmestas tags: ncard-rule

Committed under a wrong description, corrected here. The message claimed this silenced an unused parameter and left two tests failing; it does neither. The content is the finished #126 fix -- appendRun honours flat, materializing the whole effective entry set rather than falling through to the S run -- and the suite is green.

The stub I described was a transient reversion in the shared checkout, made to prove the new regression test fails without the fix. It was restored before I committed. fossil commit auto-syncs, so my commit captured the restored tree while my message described the tree I had read minutes earlier.

Authored by the #126 work, not by this commit's message. check-in: 98ddd81b75 user: dmestas tags: trunk

The module doc claimed stores of this size read in milliseconds. They do not -- a 117MB mirror scanned for 62 seconds. The claim was carried over from a WRITE path taken once per ticket change onto a READ path taken per view, where neither the frequency nor the scale held. Removed rather than softened, with what the scan actually costs written down: still linear in store bytes, which the constant-factor fix did not change.

The doc also named quarry throughout. zit provides the hooks; what is built on them is not zit's business, and naming one forge in the code reads as though it were. closed check-in: 78d56a2177 user: dmestas tags: landing-doc

The fold called artifact.parse on every artifact in the store and let it fail on the content blobs, which are most of a real store's bytes -- and a parse attempt pays a whole-bytes Z checksum before rejecting one. It now applies the reject-fast pre-screen frames --structural already uses, and reuses one plaintext buffer instead of allocating per artifact.

zit landing ls, release build: quarry 1.06s -> 0.15s, zit 2.19s -> 0.31s, libfossil 8.49s -> 0.91s. A forum-shaped store is unchanged, because there is nothing to reject there; it was never the slow case.

The screen only rejects. Every accept still takes the full parse, so nothing the fold would refuse can be admitted by being shaped like a card. closed check-in: 8776c0e5bb user: dmestas tags: landing-scan

A reviewer decides from the commits before they look at a diff, and a record whose discussion the display omits is one people hold somewhere else. Both were already computed: the fold walks every reply to derive state, and the ancestry query answers the rest -- so this prints what was read and thrown away.

Ordering comes from ancestryLinearized, not mtime: two check-ins made in the same second must not invert, and the verb owns no ancestry ordering of its own. A state change renders as an event rather than a remark, so the on-the-wire marker never reaches a reader. Both lists are bounded and say when they cut. closed check-in: cff154d675 user: dmestas tags: landing-show

zit landing show reports whether a branch would land cleanly, read from the store without writing anything. zit landing accept -R lands one with no checkout, merging in a working area it creates and removes, and refuses a conflicting landing with exit 5 and nothing written. check-in: d2bbc4f459 user: dmestas tags: trunk

accept -R merges in a working area it creates and removes, materializing only the files both sides changed -- sparse-cone 4.4 resolves the rest into the record without writing them. The check-in is minted by the same commit path a checkout uses, so its F, P and R cards match a dense merge exactly.

A landing that does not merge cleanly exits 5 with the store byte-unchanged and each conflicting path named. A forge never holds a half-finished merge and never commits conflict markers; the author resolves it from their own checkout.

runMergeIn/runCommitIn take the checkout as a parameter instead of reading the process's cwd, which is what makes an ephemeral tree addressable without mutating global state. closed check-in: 8951bc9d1e user: dmestas tags: landing-bare-store

show answers whether a branch would land cleanly with no working tree and without writing anything, so a forge can flag conflicts before anyone accepts. It calls merge3.classifyFile -- the same classifier the working-tree merge runs -- so the verdict agrees with the merge rather than being a second opinion. Content-level, because hash-level resolution refuses any file touched on both sides, which is every long-lived branch. check-in: dea1cc66d3 user: dmestas tags: landing-bare-store

A landing is a forum thread plus a raw tag on the check-in that lands, so a stock fossil clone reads all of it and no new artifact kind is minted. open/ls/show/comment/accept/reject; accept merges and commits first, then records the outcome, so a partial failure leaves an open landing over merged work rather than an accepted record over an unmerged branch. check-in: c9ff7a71ea user: dmestas tags: trunk

pack no longer re-encodes profitable deltas and answers immediately on an unchanged store -- which also means a no-op pack checks no integrity, so the reference now says verify is the pass that always reads everything. stat reports the user table; verify's untouched-row list includes user rows. caching.html gains the treestat and packed memos, and the log-scan index tier says what it actually carries now. check-in: 87b58a3943 user: dmestas tags: trunk

The client can request /user and latch its rows, and 'zit config pull user --user=NAME --password=' applies them -- merging by default, --overwrite to replace, fossil's own two modes. Credentials are required in practice because fossil serves the area to an administrator alone; without them the peer sends nothing and the verb says why instead of reporting empty success. zit's own serve does not emit /user: it has no capability model to gate it with. Smoke S12 drives a real fossil server for both halves (#125). check-in: 2170688c47 user: dmestas tags: trunk

A .user frame kind holds login/pw/cap/info/mtime rows, repo-local like shun and folded newest-mtime-wins; pack carries the folded table and the scan index (v2) carries it too. 'zit config import|merge FILE' applies a fossil configuration export -- whose cards are byte-identical to the sync protocol's -- so a repository's users cross with no SQLite reader and no server. 'zit users' prints them; the password hash needs --raw. Caps are opaque and verbatim (#124). check-in: 709c457d95 user: dmestas tags: trunk

Re-beating every existing delta paid best-of-N delta.create plus a zlib trial per artifact and never converged -- each run re-encoded ~100k frames again for ~16 bytes each (#122). A delta whose declared raw length clears pickBase's own 75% bar is copied verbatim; one that fails it (a peer's poor base) still re-encodes. A .zit-cache/packed generation key makes a repeat pack over an unchanged log a no-op. tk: 121.6s -> 76.7s first pack, 0.2s repeat. check-in: 27c4f30036 user: dmestas tags: trunk

The census pread 10 bytes per frame -- an O(frames) syscall floor, ~150k calls on tk (#121). A 64KB read-ahead window keeps the walk a measured physical pass at ~1/40th the syscalls, and the config/roll read adopts the scan index like every other cached verb. tk stat: 0.234s -> 0.026s, now ahead of fossil info. Output byte-identical. check-in: c17e0233ca user: dmestas tags: trunk

CheckinBaseline and the extras-family config read each paid a full frame-header scan per invocation (#120). A shared openLogAdoptingIndex adopts the .zit-cache sidecar (#112's pattern), refreshing on miss; extras' path is read-only per B-extras-16. tk single-file diff: 0.220s -> 0.020s. check-in: be281ab760 user: dmestas tags: trunk

pull/push/sync loaded every artifact into memory before the first wire byte -- 3.7GB resident on a 148k-artifact store, OOM beyond it (#119). The client now sits on LogIndexStore as serve does: names, held/phantom, shun mask, and hash-policy off the frame index, content materialized per-rid on demand, received artifacts in the appendix. No-op pull on tk: 3.7GB -> 88MB, 64s -> 13s. check-in: 46d49702f7 user: dmestas tags: trunk

Record (mtime, size, hash) per tracked file in .zit-cache/treestat after each scan; a matching stat answers without reading. Fossil's vfile heuristic (D. Richard Hipp), race window included. Advisory tier rules: malformed parses as empty, deletion costs a re-hash, never an answer. check-in: f34bbc730a user: dmestas tags: trunk

Both sides had made the repository-tier setting read adopt the scan index (#112); trunk's write-on-miss refresh is kept inside the shared helper the autosync read now also uses. check-in: 6f540c5d77 user: dmestas tags: trunk

A repository cloned from a Fossil server carries its versioned policy in .fossil-settings/, which zit ignored entirely -- extras, addremove, and clean answered differently from fossil on the same checkout. Every versionable setting now reads fossil's spelling first and zit's as the fallback, through the one resolver funnel; the ignore seam and the shadow note follow. Settings zit does not model stay unread. check-in: 296f2e5cc2 user: dmestas tags: trunk

Stage payloads once and rename them into place instead of writing them twice with an fsync each; keep the parent directory open across a path-sorted plan instead of re-walking it per file; read the repository setting through the persisted log index instead of a full frame scan; reuse the verb's open repository for the target's branch facts rather than deriving repo state a second (update: third) time.

tk (146k artifacts, 1,079 files): open 1.1s -> 0.48s; fossil is 0.34s. Rename-into-place also closes the checkout_sweep H2 finding: no crash cut can leave a hybrid file, so the pinned test now asserts that. closed check-in: 244523acfa user: dmestas tags: materialize-perf

open() resolves root_dir to the store's home, which for a detached worktree is another checkout's directory -- so currentCheckoutRid read THAT checkout's .zit-checkout, and leaves, branch, bare ls, and the update/merge plan base all took the other tree's branch as current while status told the truth. A new checkout_root field carries the invoking directory; checkout-state reads use it. Pinned by a two-checkout e2e asserting every verb stays in its own tree. check-in: 6f0ba0fd3d user: dmestas tags: trunk

repositorySetting opened the store with a bare openReadOnly -- a full frame-header scan to reach one config frame, 5.6s on a 1.17M-artifact store, and blind to the .zit-cache scan index because it had no directory in hand. It now takes Repository.logStore's own fast path: adopt the persisted index, resume over the tail, refresh on miss, with the cache directory resolved through the checkout's store pointer so a worktree never grows a stray sidecar. 5.6s to 0.12s. check-in: cf01ef34ba user: dmestas tags: trunk

serveAnswer resolved binary-glob and crlf-glob through a fresh LogStore.openReadOnly per setting per round -- a full frame-header scan each time, ~11s per round on a 1.17M-artifact store, turning a clone into half an hour of server CPU. The snapshot's per-build config read now carries the pair, like the submanifest facts beside it; freshness rides the same (size, mtime) key. Server CPU for the same 181-round clone: ~34 minutes to 15 seconds. check-in: 262ca49578 user: dmestas tags: trunk

An mtime divided by a day is already the day number Joffe's identity consumes (benjoffe.com/fast-day-of-week), so the UTC weekday falls out with no calendar decomposition. Flag-gated; unflagged output is byte-identical, pinned by test against a Sakamoto reference over 40,000 days.

log --long's record grammar is now a stated contract: consumers parse it positionally and no display option will change it. The flag skips info, which prints no date line. check-in: 4652c5cc31 user: dmestas tags: trunk

The record body now captures the U-card user and the changed-file set, diffed against the primary parent at derive time where the parsed manifest is already in hand (layout 5, checkpoint 4; one-time rebuild per store). 'zit crosslink' replays it: identity, comment, tags, and file changes as a versioned text stream, cache-served under the same key discipline as every read verb, byte-identical to --no-cache.

sqlite dumps 36,797 check-ins in 0.11s / 14.9MB against the 3.87GB structural stream. A parent the store does not hold reads as 'files unknown', never as an empty change set -- sqlite's corpus carries exactly one, a 2009 check-in with a genuinely absent parent.

Mixed binary versions thrash across the layout bump: each side rejects the other's cache and re-derives. Update every binary that shares a store. check-in: 2a68d2fc79 user: dmestas tags: trunk

Content blobs dominate a full stream (13.2GB on sqlite) and indexing consumers discard them. The gate is artifact.parseForImport on the reconstructed bytes -- never a cache -- behind an O(1) reject-fast screen that admits card shapes and signature wrappers; skipped frames are not hashed or emitted, and the cursor advances over them, so filtered and unfiltered polls compose against one cursor. sqlite 13.2GB -> 3.9GB, tcl -> 682MB. check-in: a2fb222e32 user: dmestas tags: trunk

release-manifest.json is the authoritative record of a release -- checksums, targets, qualification -- and nothing that produced it was in the repository. Its source_tree_sha256 had to be reverse-engineered from published output to cut 0.2.0: sha256 over 'sha256 path' lines sorted BY LINE, so by hash, which is the sha256sum | sort | sha256sum idiom.

The release whose manifest nobody could check is the one carrying false fields. 0.1.3 labelled a native build x86_64-linux-musl and claims 104 source files against a tarball holding 93.

The script re-verifies every archive against its own metadata fragment before describing it as qualified, and reproduces the published 0.2.0 manifest exactly. check-in: 520e388d6a user: dmestas tags: trunk

The 0.1.x line grew from 38 dispatched verbs to 61, gained the sub-manifest dialect, and grew a layered derived-state cache. 0.1.4's notes mention sub-manifests once and never say cache, so the notes here cover that machinery rather than only what changed since 0.1.4. check-in: 09e80bc9ef user: dmestas tags: trunk, version-0.2.0

An audit of all 57 verbs against their own --help found two gaps the existing pins could not see, because reference_test.zig checks that a verb HAS a section and #68 checks options-table flags against the usage block.

pack --drop-elided was in neither. The flag is real -- it is the sparse spec's narrow-then-compact deletion, destructive and irreversible -- and it appeared in no help output at all, so nothing compared it to anything. Now documented, including that unlike shun it leaves no tombstone, so a later widening re-fetches from a peer.

settings --global was accepted but missing from the reference.

The new e2e pin holds every verb's real --help against its reference section both ways: a flag the verb accepts that the section never names, and a flag the section documents that --help never mentions. The second direction matters more than it looks -- an omission is discoverable from --help, a promise sends someone hunting for something that is not there.

It lives in the e2e lane because it needs the help TEXT, and the help constants are in main.zig, outside the library test module. Conventions flags (-R, --no-cache) count as covered wherever they are documented once; matching is whole-token so --cone cannot satisfy --cone-not.

Both directions were verified by breaking them deliberately. check-in: 03e606ea49 user: dmestas tags: trunk

merge3 already resolved every region except one -- the case where both sides changed it differently. That branch now takes a side instead of always writing an envelope; the clean cases above it are untouched, which is the whole point. A region only one side changed is still that side's under either strategy, so -X is not 'discard their work', it is 'stop asking me'. The mechanical cases are what it is for: a regenerated lockfile, a vendored tree taken whole from upstream.

Applies to content disagreements -- conflicting hunks, add-vs-add, and both-modified binaries. It deliberately does NOT resolve modify-vs-delete: that is a disagreement about whether the file should exist, not about its contents, and either reading silently destroys work. git declines it too.

-X ours is not --backout, and the help says so outright: --backout reverses NAME's change, -X ours keeps your side where the two collide and reverses nothing. They read alike and do different things.

Composes with precedent (#103) without either knowing about the other: precedent is a pass over the outcome set, so a strategy that leaves no conflict leaves it nothing to replay or record. Verified both ways -- -X records no precedent, and a plain merge still conflicts and records.

A bad or missing -X value is a usage error rather than a silent fall back to markers, which would look exactly like a strategy that ran and found nothing. check-in: 52b8c57f4c user: dmestas tags: trunk

'zit export-git --help' printed " \ import-fossil bring a fossil repository's history into zit" as its second line, backslash included. The verbatim help pins cover merge and stash only.

The new pin checks shape instead: no rendered help line may start with a backslash, which catches the paste for every verb. The reference gains merge --no-precedent, which was documented under the wrong verb. check-in: 5fe0d9eafe user: dmestas tags: trunk

The sidecar was absent from the control-path registry, so extras listed it and add accepted it: an ordinary add put the contents of stashed work into a check-in and synced it to every peer, recoverable only by shunning. The stash help already promised "never an artifact"; nothing enforced it.

Authoritative tier, matching what losing the file costs. check-in: 503637a369 user: dmestas tags: trunk

A merge records each conflict in the local .zit-precedent sidecar; commit attaches the resolution the operator settled on; a later merge whose conflicted bytes match replays it instead of writing markers. Keyed by path and conflicted bytes, exact because the diff3 envelope carries all three sides. Whole-file, so a file whose other conflict moved will not replay.

Replay is named per file, never silent. --no-precedent declines for one merge without forgetting. 'zit precedent list|show|forget', alias rerere. The sidecar is a reserved advisory name, so it can never become an artifact. check-in: a9a24a857c user: dmestas tags: trunk

The sidebar was display:none below 800px, which is most of the time an embedded doc is read -- a panel, a phone, a split window. A 60-section reference with no navigation at all is worst exactly where scrolling costs most, and the rule was invisible because it only fires at widths you do not develop at.

Below 800px the rail becomes a compact index above the content, links flowing as wrapped chips rather than one per line. It is collapsed by default there: expanded, it pushed the page title a full screen down, so a reader met a wall of verb names before learning what the page was. The disclosure is hidden entirely on the wide layout, where the rail is always open and the control would be noise.

Verified in a real browser at 1400, 780 and 500 pixels: 70 index entries reachable at every width, against 1 before. check-in: 65fe7b9ec5 user: dmestas tags: trunk

The command reference and the caching write-up move into the tree as embedded documentation, which quarry serves on fossil's own /doc/VERSION/PATH route, already implemented in its api/register.go.

Fossil's www/embeddeddoc.wiki gives the reason to prefer this over the wiki, and it is the reason the pin below can exist:

The documentation files are versioned together with the source code files so it is always clear what version of the documentation goes with a particular release.

A wiki page cannot be pinned by a test; nothing ties it to a checkout. A tracked file can. reference_test.zig checks both directions -- a dispatched verb with no section, and a section naming no verb -- with a reviewed-exception list for sections that document an absence on purpose, which is what the export-fossil section is. The reference had drifted for weeks: grep, purge, config and trust all shipped undocumented, and nothing noticed because noticing needed a human to compare two lists.

The caching page's ladder is corrected while embedding it. It reported 641.9s / 58.3s / 0.31s; those were Debug numbers. Measured in the mode that ships: 29.63s full derive, 0.34s record-body replay, 0.04s image. Having a cache is worth 87x and the image a further 8.5x -- both real, neither 188x. Its footnote about a separate environment reporting 0.38s 'unreconciled' is resolved: same machine, optimized build. info also moved to the image since it was written, leaving branch and tag on the record-body rung. check-in: 269e24b3c3 user: dmestas tags: trunk

The image carries the active tag set now -- name, tagtype, optional value, per rid -- so tagsOf, branches, tagNamesMatching and activeTagValue all answer from either backing. Stored resolved: the derive has applied §5.1's cancel rule and §5.3's precedence before serialize runs, so nothing here re-judges either. That is what fossil does, where tagxref holds one row per (rid, tag) with propagation flattened in by tag_propagate at write time.

It subsumes the §5.2 override columns from the previous change; those are gone rather than kept beside a general mechanism that already holds the same bytes.

Two readers still went straight to tagxref and had to be routed: activeBranchValue and hasActiveClosed. Both answered plausibly on an image rather than refusing -- the first as a null-deref in a postcondition under safety and, with that assert compiled out, a different branch listing; the second as a listing quietly missing every (closed) marker.

tagNamesMatching keeps a per-backing arm on purpose. The derive holds a (rid, tag) map with no per-rid index, so walking rids and asking each for its tags is O(rids x map) there; sharing one loop cost 42s where two cost 0.37s. check-in: ce9e7a1cd4 user: dmestas tags: trunk

Every rule was paid for by a wrong answer that looked convincing: an issue filed against this repository proposed an architectural change on the strength of a 58s figure that was a bare-zig-build Debug number; the ~25x pack figure this file carried came from a --release=fast that was quietly building ReleaseSafe; a single run reported a 20% regression for a change that was a 3.7x improvement; and an optimization that won by 50x was erroring out on the first artifact and printing nothing.

It sits beside the build rather than in the wiki because each rule is only true of THIS build script -- bare zig build means Debug because of the line above it, and every --release resolves to the preference for the same reason. A page elsewhere would rot the day one of those changes and nothing would fail. check-in: 4e355fbfd1 user: dmestas tags: trunk

The default listing prints an ID per check-in and reads nothing else, yet it reconstructed and re-parsed every reachable manifest to find P cards. On a repository whose manifests inline the whole file set -- sqlite's are 2,219 F cards and 186KB each -- that is 6GB of text decompressed and parsed to answer a question that reads none of it.

Fossil does not do this. manifest_crosslink parses a manifest once and writes the edge to a persistent table:

CREATE TABLE plink(pid, cid, isprim, mtime, baseid, UNIQUE(pid,cid)); CREATE INDEX plink_i2 ON plink(cid,pid);

Every later traversal is an indexed read of it -- timeline.c's ancestor walk is a recursive CTE over plink and touches no manifest. The tell is fossil's parsed-manifest cache: MX_MANIFEST_CACHE is 6, and its comment says it exists for rebuild. Nothing on the read path needed more.

zit's derive holds the same relation, with the is_primary flag plink's isprim carries, and since #98 the zero-parse image carries it too, so this walk reads no artifact content at all. --long still needs D/U/C, so it stays on the reading walk.

log tip, min of repeated runs, warm, ReleaseFast:

libfossil 1.28s -> 0.02s libspatialite 4.81s -> 0.00s fossil 20.40s -> 0.11s tcl 40.05s -> 0.65s sqlite 62.90s -> 0.51s

Output byte-identical across all seven corpora.

WHAT MOVED: the reading walk re-hashed and re-validated every reachable artifact, so log doubled as an integrity check. It no longer does -- which is fossil's split, where validation happens at crosslink and test-integrity re-checks. zit verify is that verb, and the wrong-full-ID test now asserts verify catches what log stopped catching, so the coverage cannot go missing unnoticed. check-in: cc996c1af2 user: dmestas tags: trunk

info was the last of the three verbs held on the record-body rung, and it was held by one thing: amend --comment stores its override as a tag, and tagxref is empty on an image backing, so the fast path printed the manifest's original comment. On tcl that cost 58s to print one check-in.

The image carries the three display overrides now -- comment, user, date -- RESOLVED. The derive has already applied the cancel rule and mtime precedence by the time they are written, so the image stores the answer and never re-judges it, the same trade branch_id already makes with the §6.1 fallback. That is also what fossil does: tagxref holds one row per (rid, tag) with propagation flattened in by tag_propagate at write time, so its readers do group-bys rather than precedence.

activeTagValue now asserts derived backing instead of answering null there. It returned null for a caller that had a tag, which is the silent shape of wrong this whole issue is about; displayOverride is the image-safe door for the subset the image actually carries. check-in: 8c3abc36c7 user: dmestas tags: trunk

On tcl's tip zit ls drops from 58.3s to 0.31s: the derived state it needed was already in the cache, in the zero-parse image, and the verb asked for the record-body replay anyway.

Adjacency arrived in the image last change; what remained was every read that went around the accessors. events, its count, and store.count are all empty on an image backing, so a bound taken from them read zero and a branch value read from tagxref fell back to main_branch for every rid -- the section 8.4 walk then never left the branch and answered start:/root: with a check-in from somewhere else, exit 0 and no diagnostic. Those reads route through hasEvent, mtimeOf, hashOf, ridCount and eventsIter now, which name one enumeration order for both backings rather than inheriting insertion order from one of them.

info stays on the full derive: amend --comment stores its override as a tag, and the image carries no tag tables, so on the fast path info printed the manifest's original comment. An existing test caught that the moment the verb moved; branch and tag stay behind for the same reason. check-in: 2772dd3942 user: dmestas tags: trunk

walkLog reconstructed every manifest with log.content, which walks the delta chain from scratch. Check-in manifests are routinely stored as deltas against their neighbours, so the same base chain was re-resolved once per check-in -- work that grows with the history behind a node rather than with the node. Per-check-in cost climbed with repository size (0.5ms on libfossil against 2.2ms on sqlite) instead of staying flat.

Every other read path in this file already took #84's PlaintextCache; this one did not. Output is byte-identical -- contentCached is the same bytes, only less work.

log tip, min of repeated runs, warm, ReleaseFast:

pikchr 0.14s -> 0.04s wapp 0.05s -> 0.01s libfossil 1.26s -> 0.34s libspatialite 4.69s -> 2.28s fossil 19.36s -> 7.24s sqlite 59.70s -> 23.81s tcl 38.29s -> 5.99s

The image held mtimes, hashes, branch ids and the symbol table but not the parent/child relation, which is why every verb needing ancestry re-derived the whole log on open. Six CSR columns carry it now, both directions with the primary flag -- primary-parent selection and merge-edge skipping both read it, so an image without it answers wrongly rather than refusing.

The two backings hold the same relation in different shapes, so edgesOf hides which one a caller got, and the equivalence test compares them in order with a guard that fails when the corpus has no edges to compare.

Behaviour-neutral: no assert relaxed, no verb moved. Freeing those needs an events equivalent as well -- resolve reads self.events directly, and that table is empty on an image backing. check-in: 82de6f83d0 user: dmestas tags: trunk

Measured on wapp warm, modes interleaved and repeated: pack is ~8.5s fast, ~9.5s safe, ~10.7s small, and the read path is 0.31/0.33/0.36 with leaves at 0.01s in all three. So fast buys ~12% on writes and nothing measurable on reads.

The '599ms fast against 15.2s safe, ~25x' this file carried reproduces in neither the ratio nor the absolutes; the delta encoder's search was reworked since it was taken. Left uncorrected it would have justified almost any correctness cost, which is the expensive kind of stale number to leave in a comment that decides the shipped build. check-in: 6181a57e0c user: dmestas tags: trunk

zit ships ReleaseFast, so --release names it. This reverses the ReleaseSafe pin; pack runs ~25x faster and read verbs are indistinguishable between the modes (ls over tcl's 300MB store is 0.31s fast against 0.33s safe, leaves 0.01s in both).

The cost is real: over a thousand std.debug.assert calls compile to nothing, so release_assert_test.zig stops being belt-and-braces and becomes the thing keeping an assert from standing in for validation of untrusted input. Its rationale is rewritten around that.

Stating a preference makes it the ONLY reachable release mode, which the old comment got wrong: it documented --release=fast as an escape hatch from the ReleaseSafe pin, and that flag produced a ReleaseSafe binary anyway -- safety strings present to prove it. Anything benchmarked through it before now measured the checked build. The ci pin moves from 'keeps assertions alive' to 'the mode is stated, not left to whoever packages', which is the part that survives the reversal. check-in: da904fbe1c user: dmestas tags: trunk

smoke is the only lane that runs a real fossil against zit, so runners on different fossils make the conformance claim mean a different thing on each -- a regression against the newer can pass on the older, and newer-only behaviour reddens the older for something that is not a defect. Both are silent, because the version is printed hundreds of log lines before the failure it explains. fossil sqlite3 quoting its output on 2.26 but not 2.28 cost three round trips that way. The printed line stays; the release notes still need it. check-in: ab3a1983cb user: dmestas tags: trunk

-S reports the check-ins where the occurrence count of PATTERN changed, -G those where a changed line mentions it. A string added once and deleted later is two lines, where plain grep reports every version that held it in between.

The issue expected this to need a diff per revision. It does not: a path whose blob hash matches its parent's cannot have moved either answer, so the walk compares F-card hashes and reads only the blobs that changed. Deletions are picked up from the parent side, which a walk over the child's files alone would miss.

Merges compare against the primary parent, as git log -S does; counting against every parent reports one appearance twice. check-in: 28c48bb3f7 user: dmestas tags: trunk

The runner on 2.26 returns the check-in uuid wrapped in single quotes where 2.28 returns it bare, so the name handed to fossil tarball could not resolve -- and the failure named the hash, which reads like a missing check-in rather than a quoting difference. The harness strips to hex and then insists on a 40- or 64-character hash, so a future output change is refused by name here instead of failing two commands later. The CI step no longer runs its own copy of the query; one place for this to go wrong is enough. check-in: f37a155d45 user: dmestas tags: trunk

One runner carries fossil 2.26 where the others carry 2.28, and 2.26 refuses options before the positionals -- fossil documents 'tarball VERSION OUTPUTFILE [OPTIONS]' and 2.28 merely tolerated the other order.

The reason that took a full CI round trip to find is the second change: a failure said 'see $WORK/x.log' on a machine nobody can log into. It now prints the log, so fossil's own sentence reaches the build output. check-in: e492130985 user: dmestas tags: trunk

The discovery message tells the operator to name the repository with -R. For a -R failure that advises supplying the flag already supplied, and it reads the same whether the path was mistyped or simply holds no repository -- so a stale path in a script looked like a bug in the verb. Reported against tarball; every -R verb had it. One reporter, routed through all nine resolveRoot call sites; the no-flag discovery message is unchanged. check-in: 10a1236a7a user: dmestas tags: trunk

The runners have no unzip, so the harness exited 2 for a missing tool and the step failed on its own coverage guard instead. Adding a third tool to the battery for one step is the worse trade: python3 is already required here, and ZipFile.extractall drops modes and symlinks -- the exact defect this differential exists to catch -- so the external attributes are applied explicitly rather than trusted to an extractor. check-in: 2420f6fb88 user: dmestas tags: trunk

archive-fossil-diff.sh writes the per-format comparison to stderr, but the step piped only stdout into diff.log, so the 'kinds compared' line the coverage guard greps for was never in the file -- it counted 0 on every platform and the step could not pass.

pipefail with it: the pipeline was taking tee's status, so the differential's own failure, including its exit 2 for a missing tool, was swallowed and only the guard could ever fail this step. check-in: 98243aad6a user: dmestas tags: trunk

Removes a check-in and the blobs only it introduced, into a local graveyard that purge undo restores and purge obliterate empties.

The drop set is (seed + reachable(seed)) minus reachable(everything retained), so a blob a surviving check-in still names is never freed -- dropping it would create a phantom rather than reclaim anything. Only leaves go, judged over the whole named set: purging a parent alongside its child is accepted, purging it alone is refused.

Reuses shun's rewriteAndSwap and records no tombstone: a purge is local, so content that already synced comes back from a peer. The help says so and points at shun. check-in: 8536f59fba user: dmestas tags: trunk

A blob's plaintext length is already in every frame kind -- a zlib container states it outright, a delta carries it in its own header -- so --size reports one without resolving a chain. The binary verdict cannot be had that way (a NUL may be the last byte), so --long still reads every blob and keeps its exact columns.

The larger cost was neither: resolveHexPrefix scanned the whole store per file to resolve a full-width name the digest index answers outright, which slowed plain ls too. On tcl's tip --long drops 167s to 93s and --size costs what plain ls costs. check-in: 3b7e8c7569 user: dmestas tags: trunk

Unlike S10, which its own comment notes survives deleting the server's admission check, this one genuinely exercises the decision: the violating file is ordinary new content fossil pushes because it is new.

Records the limitation in the header: fossil exits 0 after printing the refusal, so the assertions are on the store and on fossil's output rather than its exit status. zit push exits 1. check-in: abc41cae82 user: dmestas tags: trunk

#81's predicates bind a checkout; nothing stopped a push from a clone that never adopted them, or from stock fossil. Fossil cannot close this at all -- its hooks are before-commit (client-side) and after-receive.

The policy is the SERVER's, resolved per round from its own settings, never from the artifacts arriving: a push carrying its own policy would otherwise arrive bearing permission to land. Only inspecting rules cross the wire; before-commit has no server-side counterpart.

A push sends artifacts by hash and only a manifest names paths, so the walk is manifest-first -- the policy gates what becomes REACHABLE, not what bytes transit. Decided before ingest, so a refusal stores nothing, and the whole round is refused rather than half of it.

Verified against fossil 2.28: it prints the refusal text verbatim and lands nothing. Note it exits 0 while doing so, where zit push exits 1. closed check-in: e091f2ac80 user: dmestas tags: push-policy

zit had no content search at all -- not git's pickaxe, not fossil grep's file-scoped walk. This is the latter: each named file's stored versions newest-first, every version whose content contains PATTERN reported with line numbers. A run of check-ins that never touched the file shares one blob and reports once.

PATTERN is literal; zit carries no regex engine and git log -S is literal for the same reason. Working-tree search stays ripgrep's job -- what only zit can do is read content that is not on disk.

A version the store cannot produce is counted and reported on stderr, not skipped: a search that quietly examined less looks like one that found nothing. Exit 0 found, 1 not found. check-in: a0ab591180 user: dmestas tags: trunk

Selection, top-level directory, reproducibility, and the refusal to write a partial archive are shared with tarball whole; only the container differs. Zip's header carries the size and CRC, so a member is compressed before its header is written, and deflate is used only when it makes the member smaller. Modes live in the central directory's external attributes, which is where a reader looks and where an executable silently becomes 0644 if you get it wrong -- so the fossil differential compares both formats now. check-in: c454de8df9 user: dmestas tags: trunk

Comparing zit to itself cannot catch a wrong permission bit -- writer and reader agree in the same direction. Fossil is an independent witness that already gets it right, so the harness unpacks both archives and compares member set, modes, and contents. It found one: -R was dropped at dispatch, which every test missed by running inside a checkout. Fossil versions symlinks where zit add still refuses them, so an imported corpus is also what finally exercises the l-perm path. check-in: 96a6d46d19 user: dmestas tags: trunk

The effective-set walk already resolves sub-manifests and delta chains in canonical order, so this is that walk with an archive writer. Members carry the F card's permission, so an executable stays executable. Two runs of one version are byte-identical: history supplies the order and the timestamps, and no uid, gid, or build host is recorded. A file the store does not hold is refused by name -- a partial archive looks exactly like success. check-in: 580d0691c7 user: dmestas tags: trunk

Plain tag listing now reflects only singleton symbolic release names, while raw listing preserves complete repository tag state.

Refs ticket a679b1197b61127930f7e2a6478145c561dd54cc. check-in: 24356ebb90 user: dmestas tags: trunk

Plain tag list now shows only singleton symbolic names globally and on a check-in, excluding branch machinery, raw properties, and propagating tags. Raw listing preserves every active tag, stored prefix, and value.

Add a tag-type-filtered repository-state query and end-to-end coverage for mixed singleton, propagating, raw, and branch tags.

Refs ticket a679b1197b61127930f7e2a6478145c561dd54cc. closed check-in: 41852f0a9e user: dmestas tags: fix-tag-list-release-filter

ls --long adds size, the F card's permission token, and text/binary. The permission is the fact no other verb can supply -- it lives in the manifest, and the blob it names does not know it is executable -- so a listing could not mark executables and mis-rendered symlinks. --under bounds the listing to one directory, which is also what makes --long's per-file content read affordable.

log --long adds each check-in's timestamp, user and comment. log CHECKIN PATH already selected the check-ins touching PATH; emitting only hashes meant a file-history page cost a subprocess per row.

Both are opt-in; the default outputs are byte-identical. A blob the store cannot produce reports 'absent' rather than a zero-byte text file. closed check-in: f01eb5b6a0 user: dmestas tags: quarry-support

Longest match across both polarities decides coverage, so the verdict alone cannot say why a file is missing. The member prints bare so it pastes into zit cone remove. Exits 0 when every path is covered, 1 when any is elided. closed check-in: 3cb62134a3 user: dmestas tags: sparse-cone

Both are opt-in: an unset glob is no opinion, not refuse-everything, so a repository that never set one commits exactly as before. Set binary-glob to *.png and PNGs are the binaries the project expects, so a stray .o is refused. Neither runs code, so unlike before-commit they bind a fresh clone with no consent step -- that asymmetry is why they exist.

The check gates what a commit INTRODUCES: an all-files commit re-captures every managed path, so classifying unchanged content would make declaring a glob block every later commit until the whole tree already conformed. Comparing against the parent's blob grandfathers inherited files, per content rather than per path, so editing one re-checks it.

--no-checks waives it, the same flag that waives before-commit. check-in: 2711adbb03 user: dmestas tags: trunk

project-code and hash-policy join stat's fields, and stat joins the -R family it was refused for a plumbing reason rather than a policy one. A consumer needing the project code had to find a free port, spawn zit serve, poll for readiness, POST a clone container and scrape the push card -- a subprocess and a listening socket per repository -- because the value was reachable only by speaking the sync protocol to a server you started yourself.

The no-checkout diagnostic said 'no .zit found' in directories where .zit was sitting right there; discovery looks for .zit-checkout, so it now says that and names -R as the way in. closed check-in: cd748a57c2 user: dmestas tags: project-code

keep-glob protects files clean must never delete while extras still lists them; manifest becomes a project setting driving build stamps. Both resolve across the tier chain, so a tracked .zit-settings file clones with the code.

before-commit splits declaration from consent (#82): the project ships the check as a tracked file, zit trust records the SHA-256 this checkout agreed to run. Cloning executes nothing, and an edit revokes consent by itself because the record pins content, not a flag. Non-zero exit aborts before anything is minted.

zit config pull shun is fossil's explicit request side; carriage stays clone-time otherwise so no peer can suppress content here on its own say-so.

Storage gains one keyed-string block: unknown keys are skipped by length, so the next string setting needs a key rather than another shape tag. Existing frames are unchanged.

clean-glob is deliberately absent: it means never-prompt and no-undo, which zit clean already is. check-in: 998dbf9d43 user: dmestas tags: trunk