Artifact 3609025f34164a7774565c51c54f4b7de169556f13fa572eb39d1363dade58ac:
- File
conformance/rfcs/README.md
— part of check-in
[05a53c393d]
at
2026-10-19 21:50:52
on branch feat/submanifest-reader
— fix(sync): gate sub-manifest trees on the send-submanifests pragma
draft-fossil-submanifests-00 §7.2 no longer gates S-bearing artifacts on the peer's announced version: fossil 2.29 announces 22900 and reads no tree. A peer that reads trees now says so with pragma send-submanifests, which go-libfossil sends in every request right after client-version and in every reply, clone replies included, right after server-version, whatever the sub-manifest settings. The version threshold and the 22900 reader marker are gone; client-version and server-version announce 22800, the fossil release whose protocol this library speaks.
A tree manifest or sub-manifest is refused to a peer that has not sent the pragma with "Peer does not accept sub-manifests.", never converted or omitted. A server decides from the whole request it answers, a client from the replies it has received in the exchange. avoid-submanifests moves from the ci-lock answer into the same leading run of every reply.
Enabling emission over a stored submanifest-k outside 1 to 12 without an explicit k is refused, naming the stored value (§8.1).
The conformance suite pins the draft with the 2026-10-19 errata applied: SM-7.2-1 and SM-6.3-3 are met rather than deviations, SM-7.2-2, SM-7.2-3, SM-6.3-1/2/7 and SM-8.1-3 follow the corrected text, and SM-7.2-4 covers the pragma's placement. The zit interop pushes a tree go-libfossil wrote to zit and checks both sides refuse trees when the pragma is stripped.
Tracked in #452. (user: [email protected], size: 7139) [annotate] [blame] [check-ins using]
# Pinned drafts The conformance suite reads these documents, never edits them. Each is a byte-for-byte copy of a file under `docs/rfcs/` in the zit repository, taken from the check-in below. The requirements tables in `../requirements/` pin each copy by its SHA-256, so a refreshed copy fails the coverage test until its table has been re-read against it. Since zit check-in `5a1b741b25d8bb65fe8b2e71a0e099b1ec13b418` (2026-10-13) the -01 files are gone from zit: each draft is `docs/rfcs/draft-fossil-X-00.txt`, whose content is the -01 text (the document name inside still reads `draft-fossil-X-01`) with the -01 errata applied. The copies here keep the -01 file names; the zit path is the second column. | File | zit path | zit check-in | zit artifact | SHA-256 | |---|---|---|---|---| | `draft-fossil-artifact-format-01.txt` | `draft-fossil-artifact-format-00.txt` | `3f86ff4261` | `b92b19b71f` | `3217c01c2d5a01a4…` | | `draft-fossil-repo-state-01.txt` | `draft-fossil-repo-state-00.txt` | `3f86ff4261` | `9980e22265` | `59dd17b9588497c6…` | | `draft-fossil-sync-protocol-01.txt` | `draft-fossil-sync-protocol-00.txt` | `3f86ff4261` | `bafc21adc5` | `c738d32f854495e9…` | | `draft-fossil-delta-format-00.txt` | `draft-fossil-delta-format-00.txt` | `3f86ff4261` | `d8025d6d75` | `6a28ee7e2e146f80…` | | `draft-fossil-submanifests-00.txt` | same | `0431958ff4` | `a71055781a` | `359344760a429d44…` | | `draft-fossil-sparse-cone-00.txt` | same | `86d4b67372` (branch `horizon-rfc`) | `cc2c259c19` | `22f7c2d847a31bc8…` | | `draft-fossil-horizon-00.txt` | same | `86d4b67372` (branch `horizon-rfc`) | `56bf5c0553` | `e1e92af6fa1c2acf…` | | `erratum/artifact-format-01-s4.4-z-card-last-by-ordering.md` | same | `3f86ff4261` | `dc53e482e1` | `3049c7ebc2aadea1…` | | `erratum/artifact-format-01-s4.6.4-julian-day-grammar.md` | same | `3f86ff4261` | `5a1adf7f9c` | `c50e6b5aef35b654…` | | `erratum/artifact-format-01-s4.7.3-delta-of-delta-delete-markers.md` | same | `3f86ff4261` | `fbc5c735d8` | `55195d37f9e74737…` | | `erratum/artifact-format-01-s4.7.8-j-card-empty-value.md` | same | `3f86ff4261` | `befe7488ef` | `d0880d6735b7890d…` | | `erratum/repo-state-01-s4.2-date-tag-event-time.md` | same | `3f86ff4261` | `b5e6383412` | `e55108ed39857f42…` | | `erratum/repo-state-01-s5.3-inherited-row-blocks-direct-tag.md` | same | `3f86ff4261` | `1ee5e540ce` | `2bfd0d2e76efe0bf…` | | `erratum/repo-state-01-s6.2-cancel-cards-from-parent.md` | same | `3f86ff4261` | `43f021a207` | `3d24910396abea9e…` | | `erratum/repo-state-01-s6.2-self-cancelling-sym-tag.md` | same | `3f86ff4261` | `821dd715d9` | `85d4f040bbede7ee…` | | `erratum/repo-state-01-s7.3-leaf-recheck-branch-tag.md` | same | `3f86ff4261` | `275509b8fc` | `25a27e819bae83cf…` | | `erratum/repo-state-01-s8.3-ckout-ci-plus.md` | same | `3f86ff4261` | `7e8ecce19e` | `a955cf371e420ebd…` | | `erratum/sync-protocol-01-s5.1-truncated-payload-receiver.md` | same | `3f86ff4261` | `8157e3faaf` | `0e08208def3084ac…` | | `erratum/sync-protocol-01-s5.3-errors-count.md` | same | `3f86ff4261` | `ca1539a1e9` | `752da0b6944f341a…` | | `erratum/sync-protocol-01-s6.6-file-card-after-refused-push.md` | same | `3f86ff4261` | `1c50fe8add` | `5857e99624ece02c…` | | `erratum/sync-protocol-01-s6.6-refusal-text-suffix.md` | same | `3f86ff4261` | `80923cf46f` | `19ff7626abc62a50…` | | `erratum/sync-protocol-01-s9.4-avoid-delta-manifests.md` | same | `3f86ff4261` | `e29beb0afc` | `caad0402360fdcca…` | | `erratum/sync-protocol-01-s10.1-wrong-hash-file-fails-the-round.md` | same | `3f86ff4261` | `e5dd8a4c9b` | `8cb2b8cbbdfe8089…` | The drafts carry every -01 erratum, taken from zit check-in `3f86ff4261484cc3b57aa6bbeb1cede7515d76ad` (2026-10-17): the six of check-in `100351e26ce269a29e0567dfd81010729be80f3d` (2026-10-12); the five of 2026-10-13 (repo-state §5.3, §7.3, §8.3; sync-protocol §9.4, §10.1), applied by check-in `5a1b741b25d8`, with the §9.4 erratum corrected at `00e1711e802e` to say the `forbid-delta-manifests` default is on since Fossil 2.29 and off in 2.28; and six more of 2026-10-14 and 2026-10-16: artifact-format §4.4/§6.2 (the Z card is last by ordering and the required-card rule; verification reads bytes by position), §4.7.3 (a named delta's delete markers are filtered), §4.7.8 (an absent and an empty `J` value decode alike), repo-state §5.3/§6.2 (an artifact's tag cards apply in card order, so a self-cancelling sym tag stays), and sync-protocol §6.6 (a client matches a refusal by its leading text). The sixth, repo-state §5.3 step 3 (a cancel's value stays on the stored row), is filed in zit as `erratum/repo-state-00-s5.3-cancel-value.md`, an erratum to -00 applied in the -01 text; like the other -00 errata it is not pinned here, and the -01 draft carries its text. The horizon draft merged into zit trunk on 2026-10-17 was backed out at `3f86ff4261`, so nothing of it is pinned. The errata are pinned for the Fossil 2.28 and 2.29 behavior they record, which the cases use as the reference where the drafts and fossil(1) disagree. The -00 errata are folded into the -01 drafts (zit check-in `44685f33b1`) and are no longer pinned. There is no delta-format -01, so the delta suite stays on -00; its file is unchanged at `3f86ff4261`. The sub-manifest draft (`draft-fossil-submanifests-00`) has no -01. It is taken from zit check-in `0431958ff45983c8676be02fd854baa8cd9bb0465b638526878ac6cfc56f939a` (2026-10-19), with the eight errata of 2026-10-19 applied (filed in zit as `erratum/submanifests-00-*.md`): §7.2 gates on the `send-submanifests` pragma, not the announced version (erratum check-in `6588d1bb3b`); §7.2 places `avoid-submanifests` in every reply's leading pragma run, and defines the sync session it binds (§7.2, §8.1); §6.3 adopts the settings once, at clone, through `reqconfig /config`, specifies the history replay, and names the newest main-branch leaf as the current check-in without a checkout; §8.1 refuses an enable over an out-of-range stored k; and §1.3 records that zit and go-libfossil implement it. Like the other -00 errata they are folded into the draft and not pinned on their own. Its Appendix A vectors are byte vectors: `../testdata/submanifest/` holds zit's copies of them, as authored for zit, taken at check-in `3f86ff4261` and unchanged at `0431958ff4` (`src/rfc_vectors/`), each with two leading `#` lines that keep it from reading as a live artifact on disk; the suite drops those lines. The sparse-cone and horizon drafts are not on zit's trunk yet. They are taken from check-in `86d4b6737219c3d403868015d6d396367e89c11c` (2026-10-18) on zit's `horizon-rfc` branch, and are re-pinned when zit lands them on trunk. The sparse-cone draft's §3.6 still says exclusion members are out of scope; that paragraph is stale (tracked in danmestas/zit#242), and the suite follows §3.3 and zit, which implement them. The same check-in adds the pragma names these two drafts define (`cone`, `cone-not`, and the `horizon` family) to the sync draft's §9 list of extension pragmas; the sync draft here is not re-pinned to it, so that addition is pending.