Files
alkgit/tests/fetch_machines.rs
T
glm-5.3-flash b9c5aaee65 feat(fetch): accept include-tag, honor-by-omission — OQ-19 rider 3 (real-client fetch grammar tokens)
The rider-3 sessions ran real git 2.43.0 against the landed machine
over both substrate doors for the first time:

- tests/real_client.rs — the real-client sessions: the
  git-daemon-framed duplex door (POC-1's tcp bridge shape, the frame
  git-upload-pack <repo>\0host=…\0\0version=2\0) and the POC-3-shape
  smart-http stateless door; a real clone, a configured
  include-tag-sending fetch, and a transfer.advertiseSid=true client,
  per door, with packet-trace assertions. Rows skip quietly when the
  git binary is absent.
- Decision (flagged for the wave-5 gate): include-tag accepted and
  honored by omission — the token is un-gated in the real V2 client's
  send path (transport.c v2.43.0: args.include_tag = followtags;
  builtin/fetch.c sets it for every configured fetch) and V2's fetch=
  feature list has no slot to advertise or retract it, so the
  pre-decision decline demonstrably broke a real configured fetch
  (fatal: remote error: fetch: unexpected argument "include-tag");
  the pack is the want closure as generated, identical with and
  without the token.
- Decision (flagged for the wave-5 gate): session-id= stays declined —
  the honest retraction. git's V2 client sends the pre-line only when
  the server advertises the session-id capability (fetch-pack.c:
  advertise_sid && server_supports_v2("session-id"); captured live) —
  the decline is unreachable by a conforming client and accepting it
  would be advertising what we don't serve.
- src/fetch.rs — the accepted set grew by one (include_tag, the
  FetchArgs field + parse row); the session-id pre-line row pins the
  fail-closed decline; the advertisement is unchanged and honest.
- tests/fetch_machines.rs — the accepted-set row + the identical-pack
  honor-by-omission row.
- docs/research/git-protocol.md — the fetch-argument inventory
  corrected in place: grammar vs traffic (include-tag rides configured
  fetches only; the capability pre-lines are exactly agent +
  object-format).
- tasks/architecture/oq-19-wire-validations.md — the rider-3 section
  records both outcomes with bases.

Verification: cargo test (13 suites), -p alkgit,
-p alkgit-substrate --all-features, --all-features, both clippies
(-D warnings), fmt --check, cargo doc --no-deps, taskgraph validate
— all green; the real-client rows 3× repeat-stable.
2026-10-23 08:02:16 +00:00

1107 lines
36 KiB
Rust
Raw Permalink Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
//! The wave-2 fetch negotiation machine over the duplex session
//! substrate and the stateless response end, with the wave-2 backend
//! doubles (`core/w2-negotiation`): the ADR-014 ack loop, the done
//! round's packfile section, and the budgets — wire shapes asserted
//! byte-by-byte against negotiation-captures.md's captures.
//!
//! Rows are names + `Contract stamp:` versions, per
//! `tests/common` §The row/stamp discipline.
mod common;
use std::collections::BTreeSet;
use std::sync::{Arc, Mutex as StdMutex};
use std::time::Duration;
use common::{blob_fixture, commit_fixture, decode_pack, tree_fixture, MemObjects, MemPackGen};
use futures::io::AsyncWrite as FuturesAsyncWrite;
use gix_hash::ObjectId;
use tokio::io::{duplex, AsyncReadExt, AsyncWriteExt};
use tokio::sync::Semaphore;
use alkgit::{
parse_fetch_args, serve_fetch_round, FetchRounds, Limits, RepoRecord, Request, Service,
Session, StatelessRequest, SubstrateError, Visibility,
};
fn repo(id: &str) -> RepoRecord {
RepoRecord {
repo_id: id.to_string(),
visibility: Visibility::Private,
grants: std::collections::BTreeMap::new(),
}
}
fn oid(b: u8) -> ObjectId {
ObjectId::from([b; 20])
}
fn request(pre_lines: &[&str], args: &[&str]) -> Request {
Request {
command: "fetch".to_owned(),
pre_lines: pre_lines.iter().map(|l| l.to_string()).collect(),
args: args.iter().map(|l| l.to_string()).collect(),
}
}
fn want_oid(hex: &str) -> String {
format!("want {hex}")
}
fn have_oid(hex: &str) -> String {
format!("have {hex}")
}
const FLUSH: &[u8] = b"0000";
const DELIMITER: &[u8] = b"0001";
/// Encodes one pkt-line, client-side framing style.
fn pkt(payload: &[u8]) -> Vec<u8> {
let total = payload.len() + 4;
let mut out = Vec::with_capacity(total);
out.extend_from_slice(format!("{total:04x}").as_bytes());
out.extend_from_slice(payload);
out
}
/// Encodes a fetch round request's wire bytes.
fn fetch_request_bytes(pre_lines: &[&str], args: &[&str]) -> Vec<u8> {
let mut out = pkt(b"command=fetch\n");
for line in pre_lines {
out.extend_from_slice(&pkt(line.as_bytes()));
}
out.extend_from_slice(DELIMITER);
for arg in args {
out.extend_from_slice(&pkt(arg.as_bytes()));
}
out.extend_from_slice(FLUSH);
out
}
/// Reads pkt-lines from the client end until (and including) a flush.
async fn read_pkts(reader: &mut tokio::io::DuplexStream) -> Vec<Vec<u8>> {
let mut packets = Vec::new();
loop {
let mut header = [0u8; 4];
reader
.read_exact(&mut header)
.await
.expect("pkt-line header");
let len = usize::from_str_radix(std::str::from_utf8(&header).expect("hex header"), 16)
.expect("hex length");
if len < 4 {
packets.push(Vec::new());
break;
}
let mut payload = vec![0u8; len - 4];
reader
.read_exact(&mut payload)
.await
.expect("pkt-line payload");
packets.push(payload);
}
packets
}
/// A duplex session + its client end, over a buffer large enough to hold
/// a full packfile section (the machine streams before the rows read).
fn session(
limits: Limits,
) -> (
tokio::io::DuplexStream,
Session<tokio::io::DuplexStream>,
Limits,
) {
let (peer, stream) = duplex(1 << 20);
(
peer,
Session::new(stream, limits, Service::UploadPack),
limits,
)
}
/// Pkt-line payloads of a framed response body (the stateless rows).
fn parse_pkt_lines(body: &[u8]) -> Vec<Vec<u8>> {
let mut lines = Vec::new();
let mut cursor = body;
while !cursor.is_empty() {
let (header, rest) = cursor.split_at(4);
let len = usize::from_str_radix(std::str::from_utf8(header).expect("hex header"), 16)
.expect("hex length");
cursor = rest;
if len < 4 {
lines.push(Vec::new());
continue;
}
let (payload, rest) = cursor.split_at(len - 4);
cursor = rest;
lines.push(payload.to_vec());
}
lines
}
fn text(payload: &[u8]) -> String {
String::from_utf8_lossy(payload)
.trim_end_matches('\n')
.to_owned()
}
/// An async sink whose framed bytes the rows inspect after the fact —
/// the stateless response end's writer (the http door's surface shape).
#[derive(Clone, Default)]
struct VecSink(Arc<StdMutex<Vec<u8>>>);
impl VecSink {
fn body(&self) -> Vec<u8> {
self.0.lock().expect("sink is not poisoned").clone()
}
}
impl FuturesAsyncWrite for VecSink {
fn poll_write(
self: std::pin::Pin<&mut Self>,
_cx: &mut std::task::Context<'_>,
buf: &[u8],
) -> std::task::Poll<std::io::Result<usize>> {
self.0
.lock()
.expect("sink is not poisoned")
.extend_from_slice(buf);
std::task::Poll::Ready(Ok(buf.len()))
}
fn poll_flush(
self: std::pin::Pin<&mut Self>,
_cx: &mut std::task::Context<'_>,
) -> std::task::Poll<std::io::Result<()>> {
std::task::Poll::Ready(Ok(()))
}
fn poll_close(
self: std::pin::Pin<&mut Self>,
cx: &mut std::task::Context<'_>,
) -> std::task::Poll<std::io::Result<()>> {
self.poll_flush(cx)
}
}
/// Incompressible pseudo-random bytes (LCG) — forces multi-chunk bands.
fn noisy_bytes(len: usize) -> Vec<u8> {
let mut state = 0x12345678u64;
(0..len)
.map(|_| {
state = state
.wrapping_mul(6364136223846793005)
.wrapping_add(1442695040888963407);
(state >> 33) as u8
})
.collect()
}
/// A two-commit chain fixture: commit_b(child) → commit_a(root), each
/// with its own tree and blob. Returns the objects and the loose bytes
/// to seed.
struct Chain {
blob_a: ObjectId,
blob_b: ObjectId,
tree_a: ObjectId,
tree_b: ObjectId,
commit_a: ObjectId,
commit_b: ObjectId,
}
fn two_commit_chain(objects: &MemObjects, doubled: &RepoRecord) -> Chain {
let (blob_a, blob_a_bytes) = blob_fixture(b"alpha");
let (blob_b, blob_b_bytes) = blob_fixture(b"beta");
let (tree_a, tree_a_bytes) =
tree_fixture(&[(gix_object::tree::EntryKind::Blob, "a.txt", blob_a)]);
let (tree_b, tree_b_bytes) =
tree_fixture(&[(gix_object::tree::EntryKind::Blob, "b.txt", blob_b)]);
let (commit_a, commit_a_bytes) = commit_fixture(tree_a, &[], "alpha");
let (commit_b, commit_b_bytes) = commit_fixture(tree_b, &[commit_a], "beta");
for (id, bytes) in [
(blob_a, blob_a_bytes),
(blob_b, blob_b_bytes),
(tree_a, tree_a_bytes),
(tree_b, tree_b_bytes),
(commit_a, commit_a_bytes),
(commit_b, commit_b_bytes),
] {
objects.put(doubled, id, bytes);
}
Chain {
blob_a,
blob_b,
tree_a,
tree_b,
commit_a,
commit_b,
}
}
/// A no-`done` round is answered with the exact `acknowledgments`
/// grammar of negotiation-captures.md's captures: `acknowledgments`,
/// `ACK <oid>` per recognized have — the seam's output, bare 40-hex
/// oids, no suffixes — `NAK` absent when ACKs were sent, flush
/// terminator, never `ready`, never a packfile line.
///
/// Contract stamp: ADR-014 §1 (the round grammar + the ACK rule); ADR-014
/// §5 (the wire layer acks exactly `common_haves`' output, no
/// refinement); negotiation-captures.md §Round grammar (NAK placement,
/// FLUSH terminator, bare-oid ACK line format); ADR-014 §4 (`ready` is
/// never sent, `fetch=wait-for-done` honest).
#[tokio::test]
async fn no_done_round_answers_the_exact_acknowledgments_grammar() {
let doubled = repo("owner/doubled");
let objects = MemObjects::default();
let chain = two_commit_chain(&objects, &doubled);
let gen = Arc::new(MemPackGen {
objects: objects.clone(),
});
let (mut peer, mut session, limits) = session(Limits::default());
peer.write_all(&fetch_request_bytes(
&["agent=git/2.43.0", "object-format=sha1"],
&[
&want_oid(&chain.commit_b.to_string()),
&have_oid(&chain.commit_a.to_string()),
&have_oid(&oid(9).to_string()),
],
))
.await
.unwrap();
let raw = session.read_request().await.unwrap().expect("a request");
let args = parse_fetch_args(&raw).unwrap();
assert!(!args.done, "the no-done round parses without done");
let mut rounds = FetchRounds::default();
serve_fetch_round(
&mut session,
&raw,
&mut rounds,
gen.clone(),
&doubled,
&limits,
Arc::new(Semaphore::new(limits.max_concurrent_blocking_tasks)),
)
.await
.unwrap();
let packets = read_pkts(&mut peer).await;
let shapes: Vec<String> = packets.iter().map(|p| text(p)).collect();
assert_eq!(shapes[0], "acknowledgments");
assert_eq!(shapes[1], format!("ACK {}", chain.commit_a));
assert_eq!(packets.len(), 3, "the section ends at flush");
assert!(packets[2].is_empty(), "the flush terminator");
for forbidden in ["NAK", "ready"] {
assert!(
shapes.iter().all(|s| !s.contains(forbidden)),
"{forbidden} is absent when ACKs were sent (negotiation-captures.md)"
);
}
assert!(shapes.iter().all(|s| !s.contains("packfile")));
assert_eq!(rounds.rounds_used(), 1, "the round consumed the budget");
}
/// A round whose haves the seam recognizes none of — missing or
/// non-commit alike — is `acknowledgments` + `NAK` + flush: NAK only
/// when no ACKs in the round.
///
/// Contract stamp: ADR-014 §1 (NAK only when no ACKs); negotiation-captures.md
/// §Round grammar (the observed NAK-alone shape); ADR-014 §5 (the seam's
/// exists + is-commit predicate decides).
#[tokio::test]
async fn a_round_with_no_recognized_haves_yields_the_nak_only_shape() {
let doubled = repo("owner/doubled");
let objects = MemObjects::default();
let (blob, blob_bytes) = blob_fixture(b"leaf");
let (tree, tree_bytes) = tree_fixture(&[(gix_object::tree::EntryKind::Blob, "a.txt", blob)]);
let (commit, commit_bytes) = commit_fixture(tree, &[], "root");
for (id, bytes) in [
(blob, blob_bytes),
(tree, tree_bytes),
(commit, commit_bytes),
] {
objects.put(&doubled, id, bytes);
}
let gen = Arc::new(MemPackGen {
objects: objects.clone(),
});
let (mut peer, mut session, limits) = session(Limits::default());
peer.write_all(&fetch_request_bytes(
&[],
&[
&want_oid(&commit.to_string()),
&have_oid(&oid(9).to_string()),
&have_oid(&blob.to_string()),
],
))
.await
.unwrap();
let raw = session.read_request().await.unwrap().expect("a request");
let mut rounds = FetchRounds::default();
serve_fetch_round(
&mut session,
&raw,
&mut rounds,
gen.clone(),
&doubled,
&limits,
Arc::new(Semaphore::new(limits.max_concurrent_blocking_tasks)),
)
.await
.unwrap();
let packets = read_pkts(&mut peer).await;
let shapes: Vec<String> = packets.iter().map(|p| text(p)).collect();
assert_eq!(
shapes[..2],
["acknowledgments", "NAK"],
"the missing and the non-commit have are both unrecognized, NAK alone"
);
assert!(packets[2].is_empty(), "the flush terminator");
}
/// A want-less round is answered empty — `acknowledgments` + `NAK`
/// + flush, no pack: the degenerate no-work shape (an up-to-date
/// client's exit round), not an error.
///
/// Contract stamp: ADR-014 §3 (want-less rounds answered empty); negotiation-captures.md
/// §Client round behavior (the captured up-to-date client's round).
#[tokio::test]
async fn want_less_round_gets_the_empty_acknowledgments_shape() {
let doubled = repo("owner/doubled");
let objects = MemObjects::default();
let gen = Arc::new(MemPackGen {
objects: objects.clone(),
});
let (mut peer, mut session, limits) = session(Limits::default());
peer.write_all(&fetch_request_bytes(
&[],
&["thin-pack", "no-progress", &have_oid(&oid(9).to_string())],
))
.await
.unwrap();
let raw = session.read_request().await.unwrap().expect("a request");
let args = parse_fetch_args(&raw).unwrap();
assert!(args.wants.is_empty() && args.no_progress);
let mut rounds = FetchRounds::default();
serve_fetch_round(
&mut session,
&raw,
&mut rounds,
gen.clone(),
&doubled,
&limits,
Arc::new(Semaphore::new(limits.max_concurrent_blocking_tasks)),
)
.await
.unwrap();
let packets = read_pkts(&mut peer).await;
let shapes: Vec<String> = packets.iter().map(|p| text(p)).collect();
assert_eq!(shapes[0], "acknowledgments");
assert_eq!(shapes[1], "NAK");
assert_eq!(packets.len(), 3, "flush-terminated, no packfile section");
assert!(shapes.iter().all(|s| !s.contains("packfile")));
}
/// The done round generates via `GitPackGen` with the recognized-subset
/// boundary, streams the `packfile` section — header line, band-1
/// sideband chunks (the 65000 cap forces several), the section flush —
/// and no acknowledgments section appears. The reassembled band data
/// re-decodes as the V2 sha1 pack of closure(wants) − closure(haves).
///
/// Contract stamp: ADR-014 §2 (the done round, no acknowledgments
/// section); negotiation-captures.md (the done-path response shape —
/// `packfile` + sideband + flush, POC-1 re-confirmed); ADR-004 (the
/// generation pipeline, V2 sha1 packs); ADR-009 §sideband chunk size
/// (65000 fixed per protocol).
#[tokio::test]
async fn done_round_subtracts_the_recognized_boundary_and_streams_the_packfile_section() {
let doubled = repo("owner/doubled");
let objects = MemObjects::default();
let (blob_a, blob_a_bytes) = blob_fixture(b"alpha");
let (blob_b, blob_b_bytes) = blob_fixture(b"beta");
let (big, big_bytes) = blob_fixture(&noisy_bytes(150_000));
let (tree_a, tree_a_bytes) =
tree_fixture(&[(gix_object::tree::EntryKind::Blob, "a.txt", blob_a)]);
let (tree_b, tree_b_bytes) = tree_fixture(&[
(gix_object::tree::EntryKind::Blob, "b.txt", blob_b),
(gix_object::tree::EntryKind::Blob, "big.bin", big),
]);
let (commit_a, commit_a_bytes) = commit_fixture(tree_a, &[], "alpha");
let (commit_b, commit_b_bytes) = commit_fixture(tree_b, &[commit_a], "beta");
for (id, bytes) in [
(blob_a, blob_a_bytes),
(blob_b, blob_b_bytes),
(big, big_bytes),
(tree_a, tree_a_bytes),
(tree_b, tree_b_bytes),
(commit_a, commit_a_bytes),
(commit_b, commit_b_bytes),
] {
objects.put(&doubled, id, bytes);
}
let gen = Arc::new(MemPackGen {
objects: objects.clone(),
});
let (mut peer, mut session, limits) = session(Limits::default());
peer.write_all(&fetch_request_bytes(
&[],
&[
"thin-pack",
"ofs-delta",
"no-progress",
&want_oid(&commit_b.to_string()),
&have_oid(&commit_a.to_string()),
"done",
],
))
.await
.unwrap();
let raw = session.read_request().await.unwrap().expect("a request");
let mut rounds = FetchRounds::default();
serve_fetch_round(
&mut session,
&raw,
&mut rounds,
gen.clone(),
&doubled,
&limits,
Arc::new(Semaphore::new(limits.max_concurrent_blocking_tasks)),
)
.await
.unwrap();
let packets = read_pkts(&mut peer).await;
let shapes: Vec<String> = packets.iter().map(|p| text(p)).collect();
assert_eq!(
shapes[0], "packfile",
"the section opens with the header line"
);
// Text frames before the band data: only the header line — no
// acknowledgments section opens the done round (the band chunks are
// binary; the substring scan only covers framed text lines).
let text_lines: Vec<&String> = shapes[1..packets.len() - 1]
.iter()
.filter(|s| !s.is_empty() && s.chars().all(|c| c.is_ascii_graphic() || c == ' '))
.collect();
assert!(
text_lines.is_empty(),
"no text frames inside the band: {text_lines:?}"
);
// band-1 chunks, then the flush; the 150 KB blob forces >2 chunks.
let band_chunks: Vec<&[u8]> = packets[1..packets.len() - 1]
.iter()
.map(|p| {
assert_eq!(p[0], 1, "band 1");
&p[1..]
})
.collect();
assert!(
band_chunks.len() > 2,
"the 65000 cap forces several chunks: {}",
band_chunks.len()
);
assert!(
packets.last().expect("flush").is_empty(),
"the section flush"
);
let pack: Vec<u8> = band_chunks.concat();
let decoded = decode_pack(&pack);
assert_eq!(decoded.version, gix_pack::data::Version::V2);
let ids: BTreeSet<ObjectId> = decoded.objects.iter().map(|(id, _, _)| *id).collect();
assert_eq!(
ids,
BTreeSet::from([commit_b, tree_b, blob_b, big]),
"closure(commit_b) − closure(commit_a)"
);
assert_eq!(
decoded.object_count as usize,
ids.len(),
"the header count matches"
);
let _ = decoded.trailer;
}
/// The boundary set is the *recognized* subset: a have the store lacks
/// is never honored — the done round ships the full closure, and
/// nothing else changed vs the subtraction path.
///
/// Contract stamp: ADR-014 §2 as amended by review 001 A-4 (the
/// boundary set is request haves filtered through `common_haves` —
/// never an unverified have); backend.md §trait family bullet 4 (the
/// honest boundary).
#[tokio::test]
async fn an_unverified_have_is_never_subtracted() {
let doubled = repo("owner/doubled");
let objects = MemObjects::default();
let chain = two_commit_chain(&objects, &doubled);
let gen = Arc::new(MemPackGen {
objects: objects.clone(),
});
let (mut peer, mut session, limits) = session(Limits::default());
peer.write_all(&fetch_request_bytes(
&[],
&[
&want_oid(&chain.commit_b.to_string()),
&have_oid(&oid(9).to_string()),
"done",
],
))
.await
.unwrap();
let raw = session.read_request().await.unwrap().expect("a request");
let mut rounds = FetchRounds::default();
serve_fetch_round(
&mut session,
&raw,
&mut rounds,
gen.clone(),
&doubled,
&limits,
Arc::new(Semaphore::new(limits.max_concurrent_blocking_tasks)),
)
.await
.unwrap();
let packets = read_pkts(&mut peer).await;
let band_chunks: Vec<&[u8]> = packets[1..packets.len() - 1]
.iter()
.map(|p| &p[1..])
.collect();
let pack: Vec<u8> = band_chunks.concat();
let decoded = decode_pack(&pack);
let ids: BTreeSet<ObjectId> = decoded.objects.iter().map(|(id, _, _)| *id).collect();
assert_eq!(
ids,
BTreeSet::from([
chain.commit_a,
chain.blob_a,
chain.tree_a,
chain.commit_b,
chain.blob_b,
chain.tree_b,
]),
"the missing have subtracts nothing — the full closure ships"
);
}
/// A done round where the client already has everything (the want is
/// itself the recognized boundary) is a valid zero-object packfile —
/// header, no entries, verified trailer — a valid response, not an
/// error.
///
/// Contract stamp: ADR-014 §2 (an empty resulting pack is a valid
/// zero-object packfile response); ADR-004 (a valid pack for whatever
/// closure remains).
#[tokio::test]
async fn zero_object_done_round_is_a_valid_packfile_section() {
let doubled = repo("owner/doubled");
let objects = MemObjects::default();
let (blob, blob_bytes) = blob_fixture(b"solo");
let (tree, tree_bytes) = tree_fixture(&[(gix_object::tree::EntryKind::Blob, "a.txt", blob)]);
let (commit, commit_bytes) = commit_fixture(tree, &[], "solo");
for (id, bytes) in [
(blob, blob_bytes),
(tree, tree_bytes),
(commit, commit_bytes),
] {
objects.put(&doubled, id, bytes);
}
let gen = Arc::new(MemPackGen {
objects: objects.clone(),
});
let (mut peer, mut session, limits) = session(Limits::default());
peer.write_all(&fetch_request_bytes(
&[],
&[
&want_oid(&commit.to_string()),
&have_oid(&commit.to_string()),
"done",
],
))
.await
.unwrap();
let raw = session.read_request().await.unwrap().expect("a request");
let mut rounds = FetchRounds::default();
serve_fetch_round(
&mut session,
&raw,
&mut rounds,
gen.clone(),
&doubled,
&limits,
Arc::new(Semaphore::new(limits.max_concurrent_blocking_tasks)),
)
.await
.unwrap();
let packets = read_pkts(&mut peer).await;
assert_eq!(text(&packets[0]), "packfile");
let band_chunks: Vec<&[u8]> = packets[1..packets.len() - 1]
.iter()
.map(|p| &p[1..])
.collect();
let pack: Vec<u8> = band_chunks.concat();
let decoded = decode_pack(&pack);
assert_eq!(decoded.version, gix_pack::data::Version::V2);
assert_eq!(decoded.object_count, 0, "the zero-object pack ships");
assert!(decoded.objects.is_empty());
}
/// A `generate` `ObjectMissing` abort maps to the session's protocol
/// error (the entries' error-band mapping — this machine's shape) with
/// **nothing emitted**: no `packfile` header, no partial band chunks —
/// never a broken pack downstream.
///
/// Contract stamp: ADR-004 (missing objects abort the fetch — sideband
/// error band, never a broken pack); ADR-018 §7 (`ObjectMissing`); ADR-005
/// §4 (protocol errors band-map at the entries; the stateless side maps
/// at the door).
#[tokio::test]
async fn object_missing_done_round_aborts_without_emitting_a_broken_pack() {
let doubled = repo("owner/doubled");
let objects = MemObjects::default();
let gen = Arc::new(MemPackGen {
objects: objects.clone(),
});
let (mut peer, mut session, limits) = session(Limits::default());
peer.write_all(&fetch_request_bytes(
&[],
&[&want_oid(&oid(9).to_string()), "done"],
))
.await
.unwrap();
let raw = session.read_request().await.unwrap().expect("a request");
let mut rounds = FetchRounds::default();
let error = serve_fetch_round(
&mut session,
&raw,
&mut rounds,
gen.clone(),
&doubled,
&limits,
Arc::new(Semaphore::new(limits.max_concurrent_blocking_tasks)),
)
.await
.unwrap_err();
match error {
SubstrateError::Protocol(message) => {
assert!(message.contains("does not exist"), "{message}");
assert!(message.contains(&oid(9).to_string()), "{message}");
}
other => panic!("the abort is a session error: {other:?}"),
}
drop(session);
let mut buf = Vec::new();
peer.read_to_end(&mut buf).await.unwrap();
assert_eq!(buf, Vec::<u8>::new(), "no pack byte, no header was emitted");
}
/// The budgets fail closed (ADR-009, ADR-014 §6, no new budget kinds):
/// the second fetch round of a capped session errors with nothing
/// emitted, and a round carrying more haves than the budget errors
/// before any backend call.
///
/// Contract stamp: ADR-009 (rounds/haves budgets, breach ends the
/// session); ADR-014 §6 (the loop consumes the rounds cap).
#[tokio::test]
async fn rounds_and_haves_budgets_fail_closed() {
let doubled = repo("owner/doubled");
let objects = MemObjects::default();
let gen = Arc::new(MemPackGen {
objects: objects.clone(),
});
// (a) rounds: a session capped at one round serves round 1 and
// fails round 2 before anything is written.
let limits = Limits {
max_negotiation_rounds: 1,
..Limits::default()
};
let (mut peer, mut session, limits) = session(limits);
peer.write_all(&fetch_request_bytes(&[], &[&want_oid(&oid(1).to_string())]))
.await
.unwrap();
let raw = session.read_request().await.unwrap().expect("a request");
let mut rounds = FetchRounds::default();
serve_fetch_round(
&mut session,
&raw,
&mut rounds,
gen.clone(),
&doubled,
&limits,
Arc::new(Semaphore::new(limits.max_concurrent_blocking_tasks)),
)
.await
.unwrap();
read_pkts(&mut peer).await;
session.reset_request();
peer.write_all(&fetch_request_bytes(&[], &[&want_oid(&oid(1).to_string())]))
.await
.unwrap();
let raw = session.read_request().await.unwrap().expect("a request");
let error = serve_fetch_round(
&mut session,
&raw,
&mut rounds,
gen.clone(),
&doubled,
&limits,
Arc::new(Semaphore::new(limits.max_concurrent_blocking_tasks)),
)
.await
.unwrap_err();
match error {
SubstrateError::Protocol(message) => {
assert!(message.contains("rounds budget"), "{message}");
}
other => panic!("a rounds breach is a session error: {other:?}"),
}
drop(session);
let mut extra = Vec::new();
peer.read_to_end(&mut extra).await.unwrap();
assert_eq!(extra, Vec::<u8>::new(), "the breaching round wrote nothing");
// (b) haves: a round over the haves budget errors before any
// backend call or wire byte.
let limits = Limits {
max_haves_per_round: 2,
..Limits::default()
};
let want = want_oid(&oid(1).to_string());
let haves: Vec<String> = (2..6u8).map(|b| have_oid(&oid(b).to_string())).collect();
let mut args = vec![want];
args.extend(haves);
let arg_refs: Vec<&str> = args.iter().map(String::as_str).collect();
let raw = request(&[], &arg_refs);
let mut rounds = FetchRounds::default();
let sink = VecSink::default();
let mut post = StatelessRequest::new(
futures::io::empty(),
sink.clone(),
limits,
Service::UploadPack,
);
let error = serve_fetch_round(
post.response(),
&raw,
&mut rounds,
gen.clone(),
&doubled,
&limits,
Arc::new(Semaphore::new(limits.max_concurrent_blocking_tasks)),
)
.await
.unwrap_err();
match error {
SubstrateError::Protocol(message) => {
assert!(message.contains("haves budget"), "{message}");
}
other => panic!("a haves breach is a session error: {other:?}"),
}
assert_eq!(
sink.body(),
Vec::<u8>::new(),
"the breaching round wrote nothing"
);
}
/// A saturated pipeline-concurrency semaphore delays the done round's
/// `generate` call — the permit wraps the call (ADR-009's enforcement
/// point); releasing a permit lets the round complete.
///
/// Contract stamp: ADR-009 (the blocking-pool budget, the permit
/// acquired around each `GitPackGen::generate` call); backend.md
/// §concurrency model (the wire layer enforces; exhaustion delays —
/// the 503-mapped class).
#[tokio::test]
async fn a_saturated_concurrency_permit_delays_the_done_round_generate() {
let doubled = repo("owner/doubled");
let objects = MemObjects::default();
let (blob, blob_bytes) = blob_fixture(b"paid");
let (tree, tree_bytes) = tree_fixture(&[(gix_object::tree::EntryKind::Blob, "a.txt", blob)]);
let (commit, commit_bytes) = commit_fixture(tree, &[], "paid");
for (id, bytes) in [
(blob, blob_bytes),
(tree, tree_bytes),
(commit, commit_bytes),
] {
objects.put(&doubled, id, bytes);
}
let gen = Arc::new(MemPackGen {
objects: objects.clone(),
});
let limits = Limits::default();
let pipeline = Arc::new(Semaphore::new(1));
let held = pipeline.clone().acquire_owned().await.unwrap();
let raw = request(
&[],
&[
&want_oid(&commit.to_string()),
&have_oid(&oid(9).to_string()),
"done",
],
);
let sink = VecSink::default();
let task = {
let gen = gen.clone();
let doubled = doubled.clone();
let pipeline = pipeline.clone();
let raw = raw.clone();
let task_sink = sink.clone();
tokio::spawn(async move {
let mut post =
StatelessRequest::new(futures::io::empty(), task_sink, limits, Service::UploadPack);
let mut rounds = FetchRounds::default();
serve_fetch_round(
post.response(),
&raw,
&mut rounds,
gen,
&doubled,
&limits,
pipeline,
)
.await
.expect("the done round completes once a permit frees");
})
};
// The machine is parked on the permit: nothing finished, nothing emitted.
tokio::time::sleep(Duration::from_millis(200)).await;
assert!(!task.is_finished(), "the round must wait for the permit");
assert_eq!(
sink.body(),
Vec::<u8>::new(),
"nothing emitted while parked"
);
drop(held);
task.await.unwrap();
assert!(
!sink.body().is_empty(),
"releasing the permit lets it stream"
);
let lines = parse_pkt_lines(&sink.body());
assert_eq!(text(&lines[0]), "packfile");
let band_chunks: Vec<&[u8]> = lines[1..lines.len() - 1].iter().map(|p| &p[1..]).collect();
let decoded = decode_pack(&band_chunks.concat());
assert_eq!(
decoded.object_count, 3,
"the delayed generate ran to the pack"
);
}
/// The machines run unchanged through the stateless response end (the
/// http door's surface): the acknowledgments shape and the packfile
/// section with identical wire text, each response one-shot
/// flush-terminated.
///
/// Contract stamp: ADR-005 §3 (two session substrates feeding one state
/// machine core); ADR-005 §2 (the stateless response rules — one-shot
/// flush, no `0002`); negotiation-captures.md §Statelessness (each POST
/// is one round, no cross-request state).
#[tokio::test]
async fn the_machine_runs_unchanged_through_the_stateless_response_end() {
let doubled = repo("owner/doubled");
let objects = MemObjects::default();
let chain = two_commit_chain(&objects, &doubled);
let gen = Arc::new(MemPackGen {
objects: objects.clone(),
});
let limits = Limits::default();
let pipeline = Arc::new(Semaphore::new(limits.max_concurrent_blocking_tasks));
// The no-done round: acknowledgments + ACK + flush.
let raw = request(
&["object-format=sha1"],
&[
&want_oid(&chain.commit_b.to_string()),
&have_oid(&chain.commit_a.to_string()),
],
);
let sink = VecSink::default();
let mut post = StatelessRequest::new(
futures::io::empty(),
sink.clone(),
limits,
Service::UploadPack,
);
let mut rounds = FetchRounds::default();
serve_fetch_round(
post.response(),
&raw,
&mut rounds,
gen.clone(),
&doubled,
&limits,
pipeline.clone(),
)
.await
.unwrap();
assert!(post.response().is_flushed(), "the response ends at flush");
let shapes: Vec<String> = parse_pkt_lines(&sink.body())
.iter()
.map(|p| text(p))
.collect();
assert_eq!(shapes[0], "acknowledgments");
assert_eq!(shapes[1], format!("ACK {}", chain.commit_a));
// The done round: packfile + band-1 chunks + flush.
let raw = request(
&[],
&[
&want_oid(&chain.commit_b.to_string()),
&have_oid(&chain.commit_a.to_string()),
"done",
],
);
let sink = VecSink::default();
let mut post = StatelessRequest::new(
futures::io::empty(),
sink.clone(),
limits,
Service::UploadPack,
);
let mut rounds = FetchRounds::default();
serve_fetch_round(
post.response(),
&raw,
&mut rounds,
gen.clone(),
&doubled,
&limits,
pipeline.clone(),
)
.await
.unwrap();
assert!(post.response().is_flushed(), "the response ends at flush");
let lines = parse_pkt_lines(&sink.body());
assert_eq!(text(&lines[0]), "packfile");
let band_chunks: Vec<&[u8]> = lines[1..lines.len() - 1].iter().map(|p| &p[1..]).collect();
let pack: Vec<u8> = band_chunks.concat();
let decoded = decode_pack(&pack);
let ids: BTreeSet<ObjectId> = decoded.objects.iter().map(|(id, _, _)| *id).collect();
assert_eq!(
ids,
BTreeSet::from([chain.commit_b, chain.blob_b, chain.tree_b]),
"closure(commit_b) − closure(commit_a) over the stateless surface"
);
let _ = decoded.trailer;
}
/// `include-tag` is an accepted argument honored by omission (the OQ-19
/// rider-3 decision): a done round carrying it serves the identical wire
/// response — the same `packfile` section, the same closure, the same
/// pack bytes — as the identical round without it. Tag carrying is never
/// synthesized; the pack is the want closure as generated.
///
/// Contract stamp: OQ-19 rider 3 (accept-and-honor-by-omission — the
/// `thin-pack` shape, tasks/architecture/oq-19-wire-validations.md §3);
/// ADR-004 (the generation pipeline over the want closure — no tag
/// synthesis); transport.c v2.43.0 (`args.include_tag =
/// data->options.followtags`, un-gated in V2's send path —
/// `core/w5-oq19-real-client-tokens`'s capture basis);
/// tests/real_client.rs (the end-to-end validation).
#[tokio::test]
async fn include_tag_is_honored_by_omission_serving_the_identical_pack() {
let doubled = repo("owner/doubled");
let objects = MemObjects::default();
let chain = two_commit_chain(&objects, &doubled);
let gen = Arc::new(MemPackGen {
objects: objects.clone(),
});
let (mut peer, mut server, limits) = session(Limits::default());
peer.write_all(&fetch_request_bytes(
&[],
&[
&want_oid(&chain.commit_b.to_string()),
&have_oid(&chain.commit_a.to_string()),
"done",
],
))
.await
.unwrap();
let raw = server.read_request().await.unwrap().expect("a request");
let mut rounds = FetchRounds::default();
serve_fetch_round(
&mut server,
&raw,
&mut rounds,
gen.clone(),
&doubled,
&limits,
Arc::new(Semaphore::new(limits.max_concurrent_blocking_tasks)),
)
.await
.unwrap();
let without = read_pkts(&mut peer).await;
let (mut peer2, mut session2, limits2) = session(Limits::default());
peer2
.write_all(&fetch_request_bytes(
&[],
&[
"include-tag",
&want_oid(&chain.commit_b.to_string()),
&have_oid(&chain.commit_a.to_string()),
"done",
],
))
.await
.unwrap();
let raw = session2.read_request().await.unwrap().expect("a request");
assert!(
parse_fetch_args(&raw).unwrap().include_tag,
"the accepted set parsed the token"
);
let mut rounds = FetchRounds::default();
serve_fetch_round(
&mut session2,
&raw,
&mut rounds,
gen.clone(),
&doubled,
&limits2,
Arc::new(Semaphore::new(limits2.max_concurrent_blocking_tasks)),
)
.await
.unwrap();
let with = read_pkts(&mut peer2).await;
assert_eq!(
without, with,
"the identical wire response with and without include-tag"
);
let band_chunks: Vec<&[u8]> = with[1..with.len() - 1].iter().map(|p| &p[1..]).collect();
let decoded = decode_pack(&band_chunks.concat());
let ids: BTreeSet<ObjectId> = decoded.objects.iter().map(|(id, _, _)| *id).collect();
assert_eq!(
ids,
BTreeSet::from([chain.commit_b, chain.blob_b, chain.tree_b]),
"the pack is the want closure as generated"
);
let _ = decoded.trailer;
}