beads/backend
Julian Knutsen d8d8bf587b
feat(cli): bd create --graph via one atomic BatchApplier request; --if-revision on bd reopen (#7209)
* feat(cli): bd create --graph via BatchApplier (spawner_id, thread_id)

Reapplies S11's graph-batch-apply half onto origin/main's post-#7203
state: `bd create --graph` now builds its plan through
issueops.BatchApplier instead of the removed buildDomainGraphPlan path
(create_proxied_server.go shrinks accordingly), spawner_id is stamped
only when a node names a spawner (RunBatchApplyStampsSpawnerIDOnlyWhenNamed
conformance case), and thread_id survives translation into
issueops.DepAddItem alongside the rest of GraphApplyEdge/GraphApplyNodeDep.

Entirely independent of #7203's --if-revision work: no file in this
commit overlaps #7203's or #7205's changed files.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

* feat(cli): extend #7203's --if-revision guard to bd reopen

#7203 (37acca96d) added --if-revision to update/close/assign/delete but
not reopen, even though issueops.ReopenRequest.ExpectedVersion already
existed upstream. Wire reopen.go and reopen_proxied_server.go through
the same parseIfRevisionFlag / requireSingleIfRevisionID /
reportIfRevisionFailure / ifRevisionFlagHelp helpers #7203 introduced,
so reopen gets identical one-id-only, exit-13-on-mismatch, gascity-
compatible-body behavior as the other four verbs, with no new helper
surface of its own.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

* test(cli): gc-matcher decode pin and proxied-server --if-revision coverage

#7203 never ran its mismatch body through gc's actual decode logic (only
through bd's own JSON assertions), and it only exercises --if-revision
through the embedded/direct route -- the proxied-server path
(*_proxied_server.go's uow-backed preflight/apply) had no coverage at
all on either side.

- if_revision_gc_matcher_test.go: a faithful port of gc's
  bdstore_conditional.go decode logic, run against a real built bd for
  all five guarded verbs (including reopen). Confirms against #7203's
  actual numeric-only body that no decimal-string twins are needed:
  encoding/json decodes a JSON number straight into an int64 struct
  field with no float64 intermediate, so there is no precision loss for
  gc's matcher to lose above 2^53 the way there would be decoding into
  an untyped interface{}.
- if_revision_proxied_integration_test.go: proxied-server-leg coverage
  for all five verbs (update/close/reopen/delete/assign), using a
  numeric-only proxiedPreconditionBody matching #7203's shape exactly
  (no _str fields, since the matcher test above shows none are needed).
  reopen's case is its only --if-revision coverage on the proxied route
  at all, since #7203 did not add the flag to reopen.

Both verified for real: TestGCConditionalMatcherDecode via
BEADS_TEST_EMBEDDED_DOLT=1, the five TestIfRevisionGuardProxied* tests
via BEADS_TEST_PROXIED_SERVER=1 BEADS_TEST_DOLT_SERVER=local against the
pinned dolt 2.2.0 binary (no Docker needed).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

* docs(changelog): document graph-batch-apply and if-revision-reopen deltas

Neither #7203 (--if-revision) nor #7205 (batch create/import perf)
touched CHANGELOG.md, so document this branch's two deltas on top of
them directly: bd create --graph via BatchApplier (spawner_id/thread_id,
field-survival test, real-store hierarchy/cycle scenarios), and
--if-revision's extension to bd reopen plus its gc-matcher decode pin
and new proxied-server coverage.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

* fix(cli): adapt reopen's reportIfRevisionFailure call to #7216's signature

#7216 (merged to main, landed in this rebase) added a fourth ifRevision
*int64 parameter to reportIfRevisionFailure for reassign-fence
disambiguation. Pass bd reopen's own declared --if-revision value at both
call sites (direct and proxied-server paths) to match.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

* build(bazel): auto-sync BUILD files

Applied from the bazel-sync-patch artifact of https://github.com/gastownhall/beads/actions/runs/37290168028
(`make bazel-sync`: gazelle, tools/bazel/go_srcs.py, bazel mod tidy).
Only BUILD.bazel, MODULE.bazel and MODULE.bazel.lock are ever pushed; see
scripts/bazel-autofix-push.sh.

* fix(cli): evaluate reopen's --if-revision guard before the already-open short-circuit

bd reopen's direct route checked issue.Status == StatusOpen before ever
calling ops.Reopen, so a stale --if-revision guard against an
already-open issue silently returned the "already open" no-op (exit 0)
instead of the precondition_failed refusal (exit 13) every other guarded
verb gives for a stale token. ops.Reopen itself already evaluates
ExpectedVersion first (ExecuteReopen -> CheckVersionInTx) on both the
dolt and embeddeddolt backends, and the proxied-server route already
called through unconditionally -- only the direct route's pre-check
bypassed it.

Fix: skip the already-open pre-check when --if-revision is set, letting
ops.Reopen evaluate the revision guard first. Without --if-revision,
today's behavior for already-open issues is unchanged.

Tests:
- if_revision_reopen_embedded_test.go (new): direct route, all four
  combinations (already-open/closed x stale/matching revision). Verified
  live that already_open_mismatch_refuses fails against the pre-fix
  ordering and passes against the fix.
- if_revision_proxied_integration_test.go: added the same already-open
  coverage to TestIfRevisionGuardProxiedReopen to confirm the
  proxied-server route already behaved correctly and keeps doing so.

Also documents in CHANGELOG.md that a `bd create --graph` plan over
issueops.MaxApplyBatchItems (1000) is refused outright with
GraphApplyTooLargeError (non-blocking review item).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

* fix(cli): put guard tests in CI lanes; fence unguarded reopen errors

Maintainer fixup for #7209, review loop iteration 1. It replays the prior
adoption's stranded fixup (f71ac4f66) on top of the contributor's own fix
for the direct-route reopen guard (266cacfca1). The contributor's fence
and tests are kept, and the fixup's overlapping reopen fence and
open-issue pins are dropped.

CI lanes: seven env-gated tests matched no lane's discovery pattern, so
no CI job ran them, including the reopen fix's own pin. The proxied tests
are now TestProxiedServerIfRevisionGuard{Update,Close,Reopen,Delete,
Assign}. The gc matcher test moves to
if_revision_gc_matcher_embedded_test.go as
TestEmbeddedGCConditionalMatcherDecode, and the direct reopen test is
TestEmbeddedIfRevisionReopenGuardDirect. BUILD.bazel and both shard
manifests are regenerated, and each renamed test is selected by exactly
one shard of every lane.

bd reopen:
- reportIfRevisionFailure now runs only when --if-revision is set, on
  both routes, as update, assign and delete already do. Since #7256 the
  classifier maps storage.ErrNotFound to precondition_failed, so an
  unguarded reopen that lost a race with a delete reported a guard
  refusal (exit 13) for a command that carried no guard.
- The gc matcher test gains a reopen_open row: a stale guard on an open
  issue gets the same refusal body. The matching-guard no-op subtest now
  also checks the "already open" line and that the revision did not move.

bd create --graph:
- A node naming both a plan-local parent and a parent_id is parented under
  the plan key again, as both legs did before the BatchApplier translation
  and as the dry-run preview and the local cycle check assume
  (TestBuildGraphApplyBatchRequestParentKeyWinsOverParentID).
- Removed dead code: graphApplyParentDepPairs, graphApplyDepPairKey and
  GraphApplyUnsupportedFieldError, with the helper-only test.
- A path from a node's parent to the node through a waits-for hop is now
  stored, as bd dep add and bd batch apply already store it, because the
  shared end gate walks the scheduling edges only. That shape can stall
  ready work once the spawner has an open child. The acceptance is
  documented in the CHANGELOG and the buildGraphApplyBatchRequest comment
  and pinned by TestExecuteGraphApplyAcceptsParentToChildPathThroughWaitsFor.
  Refusing the shape belongs in the shared role (types.IsSchedulingEdge),
  for every front door at once; that is a follow-up.
- The deleted fake-store rejection scenarios are ported to real-store
  tests: reverse parent-to-child blocking edges, a reverse parent-child
  edge cycle, and the end gate alone with SkipPerEdgeCycleCheck set.
- The graphApplyEdgeBatchMetadata comment names the real
  issueops.StampWaitsForSpawnerID.

CHANGELOG: the spawner stamp is described as a waits-for edge's
metadata.spawner_id, and the reopen entry says a guarded reopen of an
open issue is judged against the guard on both routes.

Agent-Signature: claude-code-claude-opus-5-5-max on behalf of Gas City Adopt-PR
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* test(cli): name #7292 in the waits-for acceptance marker

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: Eddie the Engineer <adopt-pr@gascity.com>
2026-10-07 07:53:35 +00:00
..
conformance feat(cli): bd create --graph via one atomic BatchApplier request; --if-revision on bd reopen (#7209) 2026-10-07 07:53:35 +00:00
backend.go feat(backends): registry Remote + OpenWith options; beads.OpenBestAvailableWith for embedders (#7188) 2026-10-04 09:34:36 -07:00
BUILD.bazel ci(bazel): run the scripts and source-walking tests remotely; retire their go test legs (#7320) 2026-10-06 22:20:16 +00:00
completeness_test.go test(bazel): runfiles helpers + declared inputs for scan-guard tests (#6828) 2026-09-27 08:24:08 -07:00
errors.go fix(dolt): preserve transaction outcome boundaries (#5341) 2026-08-06 04:34:46 -07:00
errors_test.go fix(dolt): preserve transaction outcome boundaries (#5341) 2026-08-06 04:34:46 -07:00
types.go perf(storage): SearchIssueSummaries narrow-projection foundation (be-nu4.3) (#3458) 2026-10-01 01:19:24 -07:00