Timeline

* fix(manifest): maintain the leaf table incrementally (RS-7.3-3)

The leaf table was recomputed from scratch at the end of every crosslink batch, so a check-in lost its leaf verdict whenever a later batch linked an untagged child to it. repo-state-01 Section 7.3, as corrected, and fossil keep that verdict: a check-in and its plink parents are rechecked only when a branch tag is written to it or propagation updates or deletes its branch row (src/tag.c:195, :100-102), and the checks run when the batch ends (src/leaf.c leaf_do_pending_checks).

The rule now lives in a new internal/leaf package: Check (leaf_check), Rebuild (the explicit full rebuild) and Pending (the deferred schedule). internal/tag schedules a recheck at exactly the branch-row writes; the crosslink sweep, receive linker and repair gate apply a batch's schedule at the gate, each phantom-fill cascade transaction and each commit or tag transaction apply their own before commit, and verify.Rebuild keeps the whole-table recompute. Reparenting schedules nothing, as in fossil, so the reparent case is split into a one-batch case and a per-artifact case.

Tracked in #426.

* fix(manifest): recheck leaves a cancel reaches; drop rechecks of a failed receive tx check-in: 628f932301 user: [email protected] tags: main, trunk

* test(conformance): re-pin the -01 drafts with the 2026-10-13 errata

Re-pin the repo-state and sync-protocol drafts to zit check-in 5a1b741b25d8, which applies five more -01 errata: repo-state §5.3, §7.3 and §8.3, sync-protocol §9.4/§9.5 and §10.1. zit now keeps each -01 text under the -00 file name; the copies keep their -01 names and conformance/rfcs/README.md records the zit path and check-in. The new errata are pinned, and the requirement tables are re-read against every changed hunk.

The §7.3 and §8.3 cases that were deviations now pass as written, and new cases cover a date tag's arrival order (§5.3), untagged leaves and a merge parent's recheck (§7.3).

The sync server answers each guard-passing ci-lock with pragma avoid-delta-manifests while forbid-delta-manifests is on, defaulting to on as the draft and fossil 2.29 do (fossil 2.28 defaults it off). The client leaves its transfer unchanged on that pragma. A file or cfile whose content does not match its hash now fails the client's round, in a pull and a clone: the reply rolls back, prefix included, and no further round starts.

The leaf table is still recomputed per crosslink batch, so a parent whose untagged child arrives in a later batch loses the verdict §7.3 keeps; that case is marked as a gap, tracked in #426.

* fix(sync): continue after a uvfile content error; renumber the avoid-delta-manifests row

A uvfile whose content does not match its HASH is a handler-reported content error: warn, commit the prefix and follow ordinary continuation (sync-protocol-01 10.1). The avoid-delta-manifests client row is SP-9.5-7 now that share-links took SP-9.5-5 and SP-9.5-6. Older unreleased changelog entries drop the pending-erratum wording. check-in: 600861484f user: [email protected] tags: main, trunk

* feat(sync): add share-links mode for pragma req-links and link

sync-protocol-01 Section 9.5 makes share-links mode OPTIONAL. With SyncOpts.ShareLinks a pulling session sends pragma req-links in its first request, as fossil's --share-links does, unless SyncOpts.ProjectCode names another project. It keeps each pragma link URL ARG MTIME as config link:URL = ARG, defossilized and otherwise verbatim, replacing an existing link only with a strictly newer MTIME. A wrong token count, a URL that is not HTTP(S) or an MTIME that is not a positive decimal drops the link.

Where fossil 2.28 differs the draft is followed: fossil fails fatally on a link URL its parser rejects and reads MTIME with strtoll base 0.

New sync.9.5 client cases run against fossil sync --share-links as well. SP-9.5-4 loses its scope note and is covered; SP-9.5-5 and SP-9.5-6 are new rows for which links a client keeps and when it sends req-links.

Tracked in #425.

* test(sync): cover links in later rounds and push+pull req-links; match project code case-insensitively check-in: 5473c06942 user: [email protected] tags: main, trunk

* fix(sync): answer a client clone_seqno with bad command; make Clone's phantom pull opt-in

Two deliberate #255 deviations from sync-protocol-01 move to conformance.

SP-5.2-6: the server now answers a client-sent clone_seqno with "error bad command" and processes the cards after it, as Section 5.2 requires for a reply-only card and as fossil's page_xfer does.

SP-8.2-5: Clone now sends nothing after clone_seqno 0 and its cleanup round, so by default it leaves the phantoms a clone produces in place, as fossil(1)'s clone does. The follow-up pull that fetched them is now opt-in through CloneOpts.PullPhantoms, whose doc comment says enabling it departs from Section 8.2.

Both cases leave the sync deviations map. Tracked in #424.

* fix(sync): quote a client clone_seqno as received in bad command; keep DST coverage of the opt-in pull

Tracked in #424. check-in: d3fb5397c8 user: [email protected] tags: main, trunk

sync-protocol-01 Section 9.5 makes share-links mode OPTIONAL. With SyncOpts.ShareLinks a pulling session sends pragma req-links in its first request, as fossil's --share-links does, unless SyncOpts.ProjectCode names another project. It keeps each pragma link URL ARG MTIME as config link:URL = ARG, defossilized and otherwise verbatim, replacing an existing link only with a strictly newer MTIME. A wrong token count, a URL that is not HTTP(S) or an MTIME that is not a positive decimal drops the link.

Where fossil 2.28 differs the draft is followed: fossil fails fatally on a link URL its parser rejects and reads MTIME with strtoll base 0.

New sync.9.5 client cases run against fossil sync --share-links as well. SP-9.5-4 loses its scope note and is covered; SP-9.5-5 and SP-9.5-6 are new rows for which links a client keeps and when it sends req-links.

Tracked in #425. check-in: a565d1c467 user: [email protected] tags: trunk, feat/425-share-links

A received delta found wrong when its basis arrives later in the pass fails sync-protocol-01 Section 7.7's commit check, which now rolls back everything the pass stored, as fossil's page_xfer and client_sync transactions do, instead of undoing the delta by hand and keeping the rest. The phantom-fill cascades and the server's closing leaf and tag repair run inside that transaction. Any other card error still commits the prefix before it (Section 10.1); on the client that prefix now includes the reply's buffered igot phantoms and visibility changes.

db.DB.WithBoundTx binds a DB to one write transaction so code written against *DB, nested WithTx included, joins it as savepoints instead of contending for the write lock from a second connection.

SP-7.7-4 is covered; the new SP-10.1-6 has sync.10.1.integrity-error-prefix. Tracked in #418. check-in: c5f5838169 user: [email protected] tags: main, trunk

Tracked in #412. check-in: 056f36e0bc user: [email protected] tags: main, trunk

* fix(sync): repair leaves and tags for check-ins the cascade links (#414)

A pulled check-in arrives as an announced phantom filled the next round, so the phantom-fill cascade links it. The session's closing sweep then found nothing left to link and skipped the leaf and tag-propagation repair, which gated on the sweep's own count: the check-in had an event row and no leaf row, and its parent stayed a leaf. The server handler had the same gap for a pushed artifact that fills a phantom.

AfterDephantomize and AfterDephantomizeBatch now return how many artifacts they linked. The sync client and the handler add that up over the session and close it with the new CrosslinkAfterCascades, which runs the repair once when the sweep or the cascades linked anything (repo-state-01 Section 7.3, RS-7.3-4), not once per file. The receive linker's cancellation paths gate on the same flag.

The repo-state harness gains a pull load: the first artifacts of a script are held and crosslinked locally, the rest pulled in process from a go-libfossil server holding them all. New cases repostate.7.3.pulled-child and repostate.7.6.pulled-retag, checked against fossil per artifact, each run as a .pull case too.

Tracked in #414.

* fix(sync): repair leaves of cascade links when a session fails (#414) check-in: 0035af8ac5 user: [email protected] tags: main, trunk

sync-protocol-01 Section 7.7 verifies a reconstructable non-clone file before storage, and SP-7.7-4 has every newly stored artifact whose bytes are available hashed before commit. The client and a clone stored a delta against a held basis as given, and the server checked only a file delta, so a delta that did not give HASH stayed stored under that name until something expanded it.

The check moves to storeReceived, the one choke point every receive path shares, so the client, the server (file and cfile) and a clone refuse a delta whose result is not HASH, or that does not apply. An unheld basis still leaves the delta stored dangling.

Tracked in #405. check-in: b3914d95b3 user: [email protected] tags: main, trunk

* fix(sync): stop the client reply loop at a gimme or igot with no artifact name

fossil's client reads gimme and igot only with blob_is_hname; any other line is an unknown command that ends the reply-card loop, as sync-protocol-01 Section 5.2 requires of a malformed defined card. A server case pins the existing bad-command answer to such a gimme.

Tracked in #397.

* fix(sync): stop the client reply loop at a file or cfile with no artifact name

Tracked in #397.

* test(conformance): split the case tables this round grew past 70 lines

Tracked in #397. Each half keeps its cases in order; only the table functions move. check-in: 636e873e6a user: [email protected] tags: main, trunk

sync-protocol-01 Section 7.5 has a pushed igot reclassify what the server holds: public igot makes it public, igot HASH 1 private, the latter from a requester without x too, where it still makes no phantom. This matches page_xfer's content_make_public and content_make_private.

Tracked in #391. check-in: 5004d66398 user: [email protected] tags: main, trunk

sync-protocol-01 Section 7.5 has a pushed igot reclassify what the server holds: public igot makes it public, igot HASH 1 private, the latter from a requester without x too, where it still makes no phantom. This matches page_xfer's content_make_public and content_make_private.

Tracked in #391. leaf check-in: e245232360 user: [email protected] tags: trunk, fix/391-igot-reclassify

* fix(sync): end card processing when a pushed file fails its handler

sync-protocol-01 Section 5.2 has a file or cfile handler failure reset the reply, send the error and stop parsing, as fossil's xfer_accept_file path does. A wrong hash, a bad HASH or DELTASRC and a storage failure now break the card loop; a file delta with a held basis is verified before it is stored (Section 7.5).

Tracked in #390.

* fix(sync): model a bad file name in the clone-fatal probe

Tracked in #390. check-in: c36a69e366 user: [email protected] tags: main, trunk

page_xfer writes the clone batch, its catalog, clone_seqno and push card at the clone card and the req-clusters igots at the pragma, so any later reset discards them (sync-protocol-01 Section 10.1). They were sent after the card loop and survived a bad command or refused push. The clone snapshot cookie is now read ahead of the loop.

Tracked in #389. check-in: 65783dbd3e user: [email protected] tags: main, trunk

sync-protocol-01 Section 9.4 applies both pragmas with no Write test, as fossil's page_xfer does. The leaf sweep before each lock write bounds the lock table. nobody's lock row stores a null login; ci-lock-fail fossilizes LOGIN and names a login-less holder nobody. lock-timeout is read as db_get_int reads it.

Tracked in #392. check-in: e44a484403 user: [email protected] tags: main, trunk

* fix(manifest,hash): verify the R card at commit; reject an unrecognized hash-policy

A check-in reads its manifest back and requires the R card to equal the digest recomputed over the resolved tree, byte for byte (artifact-format-01 6.3.4, AF-6.3-5). A hash-policy setting outside the ordinals 0 through 4 is now an error instead of a fall back to the default (7.1, AF-7.1-2).

Tracked in #387.

* fix(manifest,repo): drive the commit R check from conformance; refuse a bad hash-policy on write check-in: c42a2b90c6 user: [email protected] tags: main, trunk

Check the remaining partial and uncovered sync rows, and fix the client private-gimme answer, the unversioned-file NAME encoding, the check-in lock table, the uv-hash name form and the uvgimme round limit they exposed. Tracked in #392. check-in: 161d6ab68f user: [email protected] tags: main, trunk

Tracked in #385.

The server verified the login card first and handled every control card before any data card. A file card ahead of its push card was stored, and cards ahead of a successful login got the login's capabilities. Cards are now handled in one in-order loop, as fossil's page_xfer reads them. A request holding an invalid clone sequence is first walked without side effects, which stands in for fossil's whole-request rollback.

A refused push without an earlier pull now resets and answers "not authorized to write"; a refused pull resets, answers "not authorized to read" and ends card processing (sync-protocol-01 Section 6.6).

An accepted push now asks for the server's phantoms after the card loop, at most 500, skipping what the client asked for, shunned names and private ones outside a private sync, as request_phantoms does. A pushed igot makes a phantom instead of drawing a gimme on the spot.

A clone reply card the client cannot read ends the reply-card loop instead of aborting: the session goes on, the second clone round is still made, and the clone then fails naming the card.

The conformance matchers that accepted "push denied" and "pull denied" are narrowed to the draft's text, and new cases put output ahead of a refusal and cards ahead of the push or login they depend on. check-in: 3beb6eeb15 user: [email protected] tags: main, trunk

Tracked in #385.

The server verified the login card first and handled every control card before any data card. A file card ahead of its push card was stored, and cards ahead of a successful login got the login's capabilities. Cards are now handled in one in-order loop, as fossil's page_xfer reads them. A request holding an invalid clone sequence is first walked without side effects, which stands in for fossil's whole-request rollback.

A refused push without an earlier pull now resets and answers "not authorized to write"; a refused pull resets, answers "not authorized to read" and ends card processing (sync-protocol-01 Section 6.6).

An accepted push now asks for the server's phantoms after the card loop, at most 500, skipping what the client asked for, shunned names and private ones outside a private sync, as request_phantoms does. A pushed igot makes a phantom instead of drawing a gimme on the spot.

A clone reply card the client cannot read ends the reply-card loop instead of aborting: the session goes on, the second clone round is still made, and the clone then fails naming the card.

The conformance matchers that accepted "push denied" and "pull denied" are narrowed to the draft's text, and new cases put output ahead of a refusal and cards ahead of the push or login they depend on. leaf check-in: 479df656d3 user: [email protected] tags: trunk, fix/sync-server-order

sync-protocol-01 Section 7.6 (SP-7.6-1): a sender replaces a uvfile whose HASH is a SHA3 name with the error "Fossil version 2.0 or later required." when the peer's recorded VERSION is below 20000. The client gated artifacts this way already but sent unversioned files to such a server regardless.

buildUVFileCards now sends the refusal in place of the card, with or without content, and drops the file from the send list, as fossil's send_unversioned_file does. A deletion tombstone names no hash and is still sent. check-in: 6bde4f3729 user: [email protected] tags: main, trunk

sync-protocol-01 Section 8.6: outside a private sync, a gimme for content held as private is answered with igot HASH 1 when the peer's date is 2020-04-13 or later, and with nothing for an earlier or unknown date. fossil's send_file decides it on remoteDate, which page_xfer reads from a client-version pragma carrying DATE and TIME.

The handler now records the peer's date from client-version and sends the private igot in place of the skipped file. check-in: 13e5e9d229 user: [email protected] tags: main, trunk

Three server refusal paths now follow sync-protocol-01:

- Section 6.6 (erratum s6.6-file-card-after-refused-push): a file or cfile card without an accepted push resets the reply, sends "not authorized to write" and ends card processing, as page_xfer does. The pull's igot inventory still follows. - Section 6.4: a refused login ends card processing at its own position rather than ending the request. Cards before it keep their effect, the end-of-request inventory runs, and the reply carries the closing timestamp comment. - Section 5.3 (erratum s5.3-errors-count): errors N counts errors raised, including those a reset discarded, and skips the bad-command error, the private-content refusal and the SHA3 refusal. check-in: 5f4424320b user: [email protected] tags: main, trunk

* test: isolate fossil's global config in tests that run the fossil binary (#284)

Tracked in #284.

* test(testutil): document the file-level scope of the fossil home guard (#284)

* test(testutil): allow StartServer in the fossil home guard (#284) check-in: 8d3f18265e user: [email protected] tags: main, trunk

Start fossil server in its own process group through a shared testutil.StartServer helper and SIGKILL the whole group before Wait, so forked per-connection children cannot write into t.TempDir() during cleanup. Tracked in #352. check-in: bb864e6c58 user: [email protected] tags: main, trunk

fossil's trunk page_xfer sets g.perm.Read for a requester with Clone, and the section 7.3 erratum states it for uvgimme. The handler treated only o, i and Admin as Read, so a Clone-only requester got no unversioned files, no xtable-hash inventory and no read-gated configuration.

This is an access change. handler.canRead, CanRead or a Clone held before any login (as page_xfer derives Read ahead of its card loop), now owns every read gate: the unversioned catalog for uv-hash and for a clone, uvigot and uvgimme answers, xtable-hash, and reqconfig. auth.CanRead is unchanged, and pull still needs o. The reqconfig interop test compares a Clone-only requester with fossil 2.28 given Read too, since the 2.28 binary predates the trunk change.

Pull still needs o although fossil trunk's derived Read lets a Clone-only nobody pull; that difference is deliberate. check-in: a67fe49556 user: [email protected] tags: main, trunk

fossil's client moves its login card from the body into the x-f-l-c cookie once the server's server-version passes xfer_xflc_check (2.27, 2025-07-27 11:05 or later), and fossil's check_login, when a stored password is not 40 characters, tries the signature again against the shared secret that cleartext password derives (sections 6.4, 6.5). The client always logged in in the body, and the server checked only the stored value.

A request marks itself LoginCookieOK once the gate passes; Message.SplitLoginCookie moves its login card out of the body, and the HTTP transports (sync.HTTPTransport and the public NewHTTPTransport through the adapter) send it as the cookie, on every redirect too. The nonce is unchanged: it already covers what follows the login line, which in cookie form is the whole body. Transports that cannot set a cookie send the login in the body as before. The server side of the cookie landed with the server-version exchange.

HandleOpts gains LoginCookie, so an embedder serving HandleSyncWithOpts over its own HTTP handler can pass the cookie through; XferHandler and ServeHTTP read it themselves. check-in: 9d35c01fb3 user: [email protected] tags: main, trunk

fossil's xfer_accept_unversioned_file stores a received file under the HASH its card gave, takes only the mtime when the content is the same and newer (status 2), and stores an omitted payload as NULL with size 0; send_unversioned_file sends nothing for a name it does not hold; and the client seeds uv_tosend from rows whose hash is not NULL (sections 7.3, 8.5.1). Both sides stored a received file under a hash computed by local policy and dropped status 2 and omitted-content cards; the server answered an absent name with a tombstone; the client queued its tombstones.

uv.Receive now owns the acceptance rule, which the server and the client share instead of each re-deriving it from uv.Read, uv.Status and uv.Write. check-in: d33a37fffd user: [email protected] tags: main, trunk

fossil's send_compressed_file sends a cfile as the blob table stores it and pads a payload that does not end in LF; page_xfer sends a version 3 clone's reply as application/x-fossil-uncompressed, sends the unversioned catalog with a clone when uv-sync is on, and refuses a clone with a reset, the push card and "error not authorized to clone". The server recompressed each cfile, never padded, compressed the reply, sent no catalog, appended its refusal to the reply, and restarted a clone whose SEQNO was negative.

CFileCard's StoredBlob, set by the clone path from the blob table, is now what the encoder sends; xfer.Message.Uncompressed tells the HTTP server to send a reply as card text. The catalog goes to a requester with Read, as for uv-hash: fossil lets Clone imply Read there, which this server does not. A negative SEQNO, which section 8.1 exempts from its fatal, now ends the clone with clone_seqno 0, and the unit test that expected a restart expects that. check-in: 6c69fc1231 user: [email protected] tags: main, trunk

fossil's page_xfer refuses a pushed config card from a requester without Admin with a reset, "error not authorized to push configuration" and a break (section 9.3); drops a client's check-in locks on pragma ci-unlock (section 9.4); and answers pragma req-clusters with send_all_clusters, the cluster igots between "# sending-clusters" and "# end-of-clusters" (section 5.3). The server ignored the first two and recorded the third without answering it.

An Admin's config push is refused too, naming that this server does not apply pushed configuration, rather than dropped while the push appears to succeed. ci-unlock is honored for writers only, as ci-lock already is, and the cluster igots go to any requester, as in fossil: they name public artifacts and carry no content. check-in: 85c48f38eb user: [email protected] tags: main, trunk

fossil's page_xfer answers a client-version that carries DATE and TIME with pragma server-version, and both sides record the peer's release; send_file, send_compressed_file and send_unversioned_file then append "error Fossil version 2.0 or later required." in place of SHA3-named content for a peer older than 2.0, with a server treating a client that announced no release as one (sections 7.6, 9.1). The server sent no server-version and neither side kept or gated on the peer's release.

The server gates gimme, clone and uvgimme; the client gates its answer to a gimme. Because a server-version from a current release turns fossil's client to logging in through the x-f-l-c cookie (section 6.5), the HTTP server now reads that cookie: xfer.DecodeRequest puts its login card first with the nonce bound to the whole body, as page_xfer does. Without it a real fossil client's later rounds arrived as nobody. That moves sync.6.5.server-cookie-login, marked for #304, here. check-in: 6963b0a9dc user: [email protected] tags: main, trunk

fossil's client_sync continues a round when stored content or a new phantom leaves the phantom table non-empty, or when files are queued for the next request (section 3.7). The client instead continued on any received file and on any phantom until it had aged three rounds, kept a gimme it could not answer queued forever, and kept an igot's phantom only in the session's map. After a clone's first round it aborted on a server error, where the draft ends the reply-card loop and goes on.

converged now applies 3.7 (a) and (b). A gimme is queued only for content the client holds and may send. An igot's phantom is written to the phantom table, batched once per reply as content_new would write it, and the after-dephantomize cascades of a reply run as one batch (manifest.AfterDephantomizeBatch), so a pull no longer begins a crosslink transaction per artifact. A clone records a later round's server error, stops reading that reply, continues by 3.7, and fails at the end naming the error. Four unit tests that encoded the old continuation now expect fossil's. check-in: 62683d1d0f user: [email protected] tags: main, trunk

fossil's client_sync emits pragma send-private ahead of client-version in its first request and right after it in later ones (section 3.2); the client put it after its igot, file and gimme cards. And fossil's content_new makes the missing basis of a private card's delta a private phantom (section 7.4), where the client made it public.

buildRequest is split into header, content and extra card builders so the header can carry send-private in fossil's place; the cards it sends are otherwise unchanged. check-in: 9d3644134d user: [email protected] tags: main, trunk

* fix(sync): drop shunned artifacts and never ask for a shunned phantom (#302)

fossil's xfer_accept_file and xfer_accept_compressed_file drop a received artifact whose hash is on the shun list, and its request_phantoms and content_new leave shunned names out of what a client asks for (sections 3.3 and 7.7). The server, the sync client and clone stored a shunned artifact, and the client asked for shunned phantoms.

storeReceivedFile, which every receive path shares, now drops a shunned artifact; the server checks first so a dropped file is not reported as missing after store. The client's phantom scan and its igot handling skip shunned names, so a delta whose basis is shunned is kept as a delta and the basis is never requested.

* refactor(sync): split the client's reply handling into per-card helpers

processResponse had grown to 232 lines: one switch over every reply card followed by the convergence check. It is now a loop over handleReplyCard, which hands each card to a focused helper, and converged, which keeps the termination rules apart from card handling. Behavior is unchanged. check-in: aea957939f user: [email protected] tags: main, trunk

fossil's page_xfer accepts a uvfile on the unversioned-write capability (WrUnver, y) alone; `fossil uv sync` sends uvfile cards with no push card. The handler refused each with "rejected: no push card", so pushing unversioned files from fossil to a go-libfossil server stored nothing.

This is an access change: a requester with y and no Write may now store unversioned files, as in fossil. A refusal (no y, or a HASH that does not match the content) now resets the reply, answers the error and breaks out of the card loop, as xfer_accept_unversioned_file's caller does. Tests pin both sides of the gate, and an interop test pushes with fossil uv sync. The real-fossil test helper now points HOME and XDG_CONFIG_HOME, not only FOSSIL_HOME, at the test's directory. check-in: b45ccf4351 user: [email protected] tags: main, trunk

fossil's page_xfer compares a push or pull card's PROJECTCODE with the repository's project-code before any capability check and, on a mismatch, resets the reply, answers "error wrong project" and breaks out of the card loop (section 8.5). The handler accepted either card whatever its project code.

The refusal goes through a new breakLoop, which the undecodable file card path now shares: the reply is reset to the one error, no later card is processed, and the post-loop work and timestamp still follow. Tests that synced repositories created with different project codes now create them with a shared one, and the DST simulator gives its leaves the upstream's. check-in: 821cf115ce user: [email protected] tags: main, trunk

The server refused application/x-fossil-debug with HTTP 415, where fossil dispatches it to sync like the other x-fossil types. The client read any 2xx as a sync reply and decoded a reply as compressed only when typed exactly application/x-fossil, so a reply with no Content-Type was read as plain cards and a proxy's HTML page was parsed as cards.

Replies now go through xfer.DecodeReply, which applies section 4's reply rule by case-insensitive prefix: no type is the compressed container, an x-fossil-debug or x-fossil-uncompressed prefix is card text, another x-fossil prefix is compressed, and anything else is refused. Only status 200 is a reply. check-in: 47dc6363c0 user: [email protected] tags: main, trunk

The server ignored reply-only cards a client should never send and the client read on past cards that end a reply.

On the server, a client-sent error or message card -- reply-only cards fossil's page_xfer has no branch for -- now answers "bad command" and resets the reply, not silence. A push card from a requester who may read but not write answers "message pull only - not authorized to push", as fossil does, rather than a refusal, so the pull proceeds. To reach that branch the client now sends its pull card before its push card, fossil's order: the server's read-vs-pull-only decision turns on having seen the pull card first.

On the client, an error card ends the reply-card loop (§10.1): the cards before it stand, those after it are not read. A "message pull only ..." (§10.2) stops the client pushing -- it keeps pulling but answers no further gimme and drops its push card one request later, matching fossil clearing SYNC_PUSH after it has already emitted the next request's header.

Behavior change worth a look: a first --private pull against a server that denies private content (no 'x') now receives no content that session, because the denial error ends the reply-card loop before the server's inventory, as fossil's client does. The private blob still never reaches an unauthorized client. TestSyncPrivateEndToEnd/no_x_capability is updated to this. check-in: c9436f82cc user: [email protected] tags: main, trunk

The card decoder refused a card by failing the entire message: the server answered HTTP 400 where §5.2 has it emit "bad command" and parse on, and the client dropped a whole reply including its valid prefix.

The decoder now yields a bad-command card for a line no rule accepts (an unknown operator, a bare pragma, a gimme with a surplus token, a config with a signed size) and a bad-file card for a file/cfile/uvfile line it cannot frame. The server resets its reply, emits the matching error (bad command, or the handler's "malformed file line" and the like) and either parses on (bad command) or stops (bad file), as fossil's page_xfer does; the client ends the reply-card loop, keeping the cards decoded before the bad one; a clone treats it as a non-retryable protocol error.

Tokens are split on ASCII whitespace only, within fossil's six-token window (blob_tokenize over aToken[6]): a no-break space no longer separates tokens, and a seventh token is ignored rather than rejected, so igot and private tolerate a surplus token. A full cfile's USIZE is carried through as metadata, not checked against the inflated size (the hash binds the content instead), and the byte after a config payload is consumed whatever it is, both matching fossil's xfer_accept_compressed_file and page_xfer.

This also resolves the bad-command cases tracked in #300 (sync.5.2.bad-command, .bad-command-resets, .unknown-prefix, sync.10.1.bad-command-keeps-mutations), whose marks are removed here. check-in: ddc4b5396d user: [email protected] tags: main, trunk

With no hash-policy row, NamingFor now follows fossil's hname_default_policy: sha3 if the repo holds a SHA3 artifact or no SHA1 one, auto otherwise (7.2). hash.PromoteFor, called by blob's single row writer and by sync's received-content store before an artifact enters, writes hash-policy=2 when an auto repo takes a SHA3 name (7.4), as content_put_ex does, so fossil opening the repo sees the promotion. A row still saying auto next to SHA3 artifacts, from before the promotion was written, keeps deriving sha3. check-in: cc76fd023f user: [email protected] tags: main, trunk

The delta format holds integers below 2^32 only, so a producer must not write a target length or copy offset at or past that. Create now returns ErrTooLarge for such inputs; Deltify then stores the content whole, as fossil does when a delta is not worthwhile, and the other callers propagate the error. Apply's header check becomes >= so it refuses exactly what Create will not write. The bound is tested on lengths, as a direct test needs a 4 GiB input. check-in: 38eddf43f1 user: [email protected] tags: main, trunk

* fix(sync): send content only to requesters allowed to read it

The sync server answered `gimme HASH` with the artifact whether or not a pull card had been accepted, and answered `pragma uv-hash`, `uvgimme` and `uvigot` with the unversioned catalog and files for any requester, anonymous ones included.

Match fossil's page_xfer:

- gimme is answered only when a pull card was accepted, or in a clone, which fossil also treats as pulling. - pragma uv-hash sends the catalog only to a requester with Read (o, i or Admin). uv-push-ok goes only to a requester with y; other readers get uv-pull-only, as fossil sends. - uvgimme and uvigot send nothing to a requester without Read, and uvigot adopts a client's mtime only for a requester with y. - Table sync follows the same rules: xgimme needs an accepted pull and pragma xtable-hash needs Read. - pragma ci-lock takes or reports a lock only for a requester with Write, so an anonymous requester cannot learn who holds one.

Pragma handling moves into handlePragma to keep handleControlCard short.

GHSA-mvx3-6793-8x4r

* docs(changelog): note that gating ci-lock on Write departs from fossil

GHSA-mvx3-6793-8x4r

* fix(sync): follow HTTP redirects as fossil does, never HTTPS to HTTP

Both HTTP transports used net/http's default redirect policy. It follows a redirect from HTTPS to plain HTTP, so a 307 or 308 resent the request, login card included, in cleartext. It also turns a 301 or 302 into a GET without the body and stops after 10 redirects, neither of which fossil does.

PostXfer now follows redirects the way fossil's http_exchange (src/http.c) and draft §4 describe. It replaces the client's redirect policy for these requests, then:

- refuses a redirect from HTTPS to HTTP, and to any scheme but http(s); - re-POSTs the same body and Content-Type for 301, 302, 307 and 308; - strips one trailing /xfer from the new location, as fossil's loop does (the draft says "repeatedly"; the code strips once); - fails on the twentieth redirect (MAX_REDIRECTS 20).

HTTPTransport and NewHTTPTransport both use it, including with a client passed through WithHTTPClient.

GHSA-3f8x-8fr5-mr98 check-in: 3a1014cf60 user: [email protected] tags: trunk

* fix(xfer): reject non-decimal card sizes and bound payload memory

A payload card's size was parsed with strconv.Atoi, so a peer sending `file HASH -1` reached readPayload's negative-size assertion and panicked the client or server decoding the message, and a signed size such as `+5` was accepted. Sizes now go through parseSize, which follows the sync draft's `decimal = 1*DIGIT` grammar and fossil's 32-bit limit and returns a decoding error for anything else. uvfile FLAGS use the same parser, so negative FLAGS are rejected as the draft requires. The assertion stays, now unreachable from peer input.

readPayload no longer allocates the declared size up front: its buffer grows with the bytes read, so a large size over a short message fails after allocating about what was sent.

A cfile payload may inflate only to its USIZE (a margin above it for a delta) and not past its 4-byte size prefix, and all cfile payloads in one message share a MaxCFileInflateBytes (1 GiB) budget, so a few megabytes of zlib cannot inflate to tens of gigabytes.

FuzzDecode exercises the decoder over both framings.

* fix(sync): refuse malformed xrow payloads and cap HTTP reply bodies

An xrow payload or xdelete PKData from a peer was decoded into a map and used as is. JSON null gave a nil map that reached coerceJSONNumbers' assertion, and a primary-key value of the wrong type or a nested value reached normalizeValue's, panicking the client or server. Both directions now decode through decodeRowJSON: one JSON object, nothing after it, and repo.ValidateRow's rules, which require scalar values and primary-key values that PKHash can normalize. A client refuses an xrow for a table it has not registered. Column types match case-insensitively, as table creation accepts them.

Both HTTP transports read replies through xfer.ReadBody, which stops at MaxBodyBytes (MaxDecompressedBytes plus framing headroom) and returns ErrBodyTooLarge, untagged for retry, instead of reading an endless body into memory.

FuzzTableSyncPayload drives the server handler and client over arbitrary payloads.

* fix(sync): check synced-table rows and identifiers before building SQL

repo.UpsertXRow wrote a row's map keys into its INSERT as column names, and the rows of an xrow payload or xdelete PKData come from the peer, so a key such as "val, mtime) VALUES(...) --" chose the statement's columns and values, and a key could select from other tables into a synced one. This held on the server for pushed rows and tombstones, and on the client for a server's rows.

A peer's row must now use exactly the table's declared columns: repo.ValidateRow refuses any other key, which also keeps a peer from setting _owner. UpsertXRow no longer trusts its caller: it reads the table's registered schema and refuses any key it does not declare, apart from _owner on an ownership-mode table. ListXRows, LookupXRowOwner and DeleteXRowByPKHash validate the table name and the column names they interpolate, so no synced-table identifier reaches SQL text unchecked.

* fix(sync): answer reqconfig only with fossil's transferable config

The sync server answered reqconfig NAME with any row of the config table, for any requester, including one with no capabilities, so project-code, server-code and every local setting could be read by anyone who could reach the sync endpoint.

handleReqConfig now follows fossil's page_xfer and configure.c. It answers only a /GROUP name, only for a requester with Read, matches the group by prefix in fossil's table order, and sends only the names fossil transfers in that group, in fossil's encoding. The user and subscriber tables also need Admin, which counts only for a verified login, and the concealed table needs RdAddr. A refused request gets no cards and no error.

The groups, their names and their capability rules live in one place, config_groups.go. A differential test answers the same requests with fossil's own server and with HandleSync on the same repository and requires identical config cards.

* test(sync): keep fossil http from leaving a backoffice process

* fix(sync): check the login nonce against the request body

The server verified a login card's signature over its nonce but never checked that the nonce was the SHA-1 of the request body after the login line, as fossil's check_tail_hash does (draft-fossil-sync-protocol-00 §6.2). A login card captured from one request therefore authenticated any other request body.

The decoder now records, for each login card, the SHA-1 of the decompressed bytes after the card's line, and VerifyLogin refuses a card whose nonce does not match it. A login card that was never decoded has no tail and is refused.

Login handling otherwise follows fossil's page_xfer and check_login: - a refused login ends the exchange with "error login failed" instead of continuing as nobody; - a second login card is refused with "multiple login cards"; - a successful login adds the user's capabilities to nobody's instead of replacing them; - nobody and anonymous log in without a password, and the reserved developer and reader logins and accounts with an empty password cannot be logged into.

Fixes GHSA-2g5p-q5f3-477p.

* fix(repo): seed nobody with fossil's default capabilities

A new repository seeded the nobody user with "cghijknorswy", so every request without a login card could push check-ins and unversioned files, write wiki pages and tickets, and held Setup. Seed the four built-in users `fossil new` creates, with fossil's capabilities: nobody "gjorz", anonymous "hz", developer "ei" and reader "kptw".

Setup and Admin now include check-in, read and clone, as they do in fossil, so the creating user can sync once given a password. A login to an account with no stored password is refused, as fossil's check_login refuses it; the creator starts with none, and the signature would otherwise be the SHA-1 of the nonce alone.

Tests that push without credentials grant nobody push explicitly. Existing repositories keep their stored capabilities.

* test(auth): reserved-login test updates the seeded developer and reader rows

New repositories now seed developer and reader as fossil does, so the test gives the existing rows a password instead of inserting them.

* docs(changelog): group the sync hardening fixes under Security

* test(sync): reqconfig tests sign their logins over the request body

The login nonce now has to cover the request body, so the reqconfig tests build their admin requests with signedRequest instead of a bare login card. check-in: 3a6126c3e9 user: [email protected] tags: trunk

The server answered a pull-only client's igot cards with gimme for artifacts it lacked. The client was not pushing, so it never sent them and kept going round after round until it hit the round cap. Fossil's page_xfer handles igot only when the client sent a push card, and does the same here: a pull-only igot just records what the client has.

The client also treated the server's gimme cards as outstanding work even when it was not pushing, so any server that did ask (an older libfossil, or another implementation) could hold a pull open. A session without push now ignores gimme cards.

Two handler tests relied on a pull card alone producing a gimme; they now send a push card. check-in: 1e3e2df2f2 user: [email protected] tags: trunk

* fix(sync): clone and sync exchanges follow the sync draft (#255)

* docs(changelog): clone and sync exchanges follow the sync draft (#255)

* fix(sync): fail unfinished clones; ask for held phantoms once (#255 review) check-in: e46a6c3f60 user: [email protected] tags: trunk

SeedConfig never wrote hash-policy, so naming fell back to deriving from the repo's own artifacts — and an empty repo has none, settling on SHA1 permanently. Seed sha3, which is what fossil init writes.

Clone drops the row the same way it drops project-code: a clone adopts its source's policy, and an absent row derives naming from the artifacts that arrive.

Moves the policy enum to db so the seed and hash.NamingFor read one definition, and updates fixtures that hardcoded SHA1 names to ask the repo how it names artifacts. check-in: 1c02c944ec user: [email protected] tags: trunk

Three tests started a real fossil server and killed it with Process.Kill() without waiting for the exit. Kill signals but does not reap, so the server could still be writing its -wal/-shm files while t.TempDir()'s cleanup walked the directory. A file appearing mid-walk fails RemoveAll with "directory not empty", which is how TestCloneFromRealFossilUncompressedReply failed on CI while passing locally -- it needs a loaded machine to lose the race.

internal/sync/integration_test.go already did this correctly, Kill followed by Wait, with a comment naming the same failure. This applies that established pattern at the three sites that lacked it.

Separately, the unit-test step ran with -timeout=120s, a per-package budget. CI runs the package binaries concurrently, so under contention internal/manifest and internal/sync exceeded it and panicked with "test timed out" despite finishing in 19s and 34s on an unloaded runner. Raised to 300s, matching the DST step, so the margin does not depend on how busy the runner is. check-in: 2256f5d7ce user: [email protected] tags: trunk