alkdev/alkgit

0 Fork 0

Files

T

glm-5.3-flash e76f91f6d7 docs(architecture): resolve OQ-04 — receive-pack state machine (ADR-013)

- ADR-013: V0-framed push machine grounded in real git 2.43.0 captures
  (file://, git://, smart-http mock, raw stdio into real receive-pack):
  V0-shaped ref advertisement (caps on first ref line, capabilities^{}
  sentinel only for empty repos), served capability set, shallow requests
  rejected for v1, thin packs accepted with server-odb bases (no
  capability involved; push.thin default), ingestion bound to
  Bundle::write_to_directory_eagerly + gix-fsck + one gix-ref transaction
  per push (.keep-guarded), unpack-first CAS timing with observed
  upstream order, band-1 pkt-line-framed status report, http framing
  (probe/Content-Length/chunked), v1 update policy (CAS only; deletes
  and force-push allowed)
- docs/research/push-captures.md: the normative push wire record
- transport.md/backend.md/doors.md: receive-pack sections rewritten to
  the decided shapes; backend.md ingestion composition bound; stale
  OQ-04 references resolved
- ADR-003 amended: V2-only governs fetch; push is V0-framed by upstream
  design (fixes the V2-only contradiction found in review)
- ADR-009 amended: haves default reconciled with the client's stateless
  ceiling (16384); blocking-pipeline budget covers generation+ingestion
- OQ-04 resolved; tracker task closed; CAS-fail-fast optimization
  tracked (tasks/architecture/oq-13-cas-failfast.md)
- research index: poc findings + capture docs listed

Verification: cargo test / clippy -D warnings / fmt --check / doc pass

2026-09-25 04:05:07 +00:00

3.9 KiB

Raw Blame History

ADR-003: Protocol V2-first with honest capability advertisement

Status

Accepted

Context

The git smart protocol has three versions in the wild: V0 (the original), V1 (capability line prefixed), and V2 (command-based, default since git 2.18+). POC-1 and POC-2 validated the full V2 fetch path against real git 2.43 end-to-end (advertisement, ls-refs, fetch with sideband packs) over both duplex and http substrates. V0/V1 stateless-http is a different framing (per-round POSTs, stateless ack re-derivation).

The research inventory (git-protocol.md §"Open items") left the V0/V1 question open: support V0/V1 on http? On ssh? Which first?

Decision drivers:

Our deployment anchor (vision §"Primary deployment target") is modern git clients over http and ssh; every git client from the last six years speaks V2.
Honest capability advertisement (vision principle 5) is easiest to keep honest with fewer served shapes.
V0/V1 stateless-http needs multi-round negotiation state re-derivation — meaningful complexity for a client population that is effectively extinct.

Decision

Serve protocol V2 first. alkgit v1 serves V2 on both doors:

Advertisement is once-per-session (duplex) / per-request-set (http), exactly as observed in POC-1 (re-advertising between commands hangs real git).
Advertised capabilities are exactly what we serve: ls-refs=unborn, fetch=wait-for-done (the full-closure pack on done policy validated in POC-1/2; the multi-round ack loop, ADR-014, needs no capability change — the acknowledgments section is grammar, not capability), object-format=sha1 (+ sha256 under the feature flag, if and when tested — OQ-05). Nothing unimplemented is advertised (shallow, filter, packfile-uris, object-info, server-option are declined by omission; POC-1 confirmed real git accepts this). On the push side, git-upload-archive is not served at all — ssh exec requests for it get a fixed refusal (doors.md; ADR-008's never-execute rule).
HTTP requests protocol V2 only; ssh requests protocol V2 only (see below).

V0/V1 policy: decided now as V2-only for v1 — fetch only. Both doors (http and ssh) speak V2 for fetch; clients that cannot speak V2 get a clear pkt-line error. Reversal is wire-visible (the advertised version set is a wire format), so treat this as effectively one-way once published — the decision is still made now, and revisiting it needs a new ADR plus a deprecation window for clients. The motivation stands: a concrete consumer needing V0/V1 (e.g. very old CI images) has not been identified. Receive-pack is exempt: the push path is V0-framed by upstream design on every client generation (no version negotiation exists on push at all — ADR-013 §1), so the V2-only rule never touches git push. The V0/V1 decline applies to the fetch path only.

Sequencing note: the v1 fetch policy that ships first is full-closure-on-done (POC-validated). Multi-round V2 negotiation (haves/acks without done) is in v1's scope but lands after the done-path works end-to-end; its ack loop is now decided (ADR-014) and needs no capability-advertisement change.

Consequences

Smallest honest surface; POC-validated shapes only.
Old-client support is declined explicitly (error, not silence).
The POC's observed done-path shape (no acknowledgments section, no ready, response ends at flush) is our normative wire behavior, not the docs' grammar — git-protocol.md §V2 records this.
If V0/V1 is ever added, the state machines in transport must be re-shaped around a version-dispatch at session start (single point, by design — see ADR-005).

References

docs/research/git-protocol.md §"Protocol surface inventory", §"Negotiation policy"
docs/research/poc-1-findings.md (advertisement-once, capability declination), POC-3 (http V2 framing)
ADR-005 (substrate types), ADR-007 (ACL before advertisement)
transport.md, doors.md

3.9 KiB Raw Blame History