beads/engdocs/TESTING.md
Bee 70b44da7cc
docs(ci): finish the retired-shard-job comment sweep after #7478 (wy-wqq7c1) (#7456)
#7478 deleted the frozen 15-shard proxied and 20-shard embedded manifest
blocks and refreshed most comments that described pr-risk.yml's retired
fork/push jobs as still running. Finish the #7382 adoption follow-ups on
what survived: the Close3/Close4 split is cited as #7173 (bd-ojckcg), and
the remaining present-tense references to the legacy jobs (bazel.yml's
embedded-lane rationale and cmd-dolt lane note, BUILD.bazel's socat note,
TESTING.md's socat paragraph) now say the jobs are retired (bd-g2w5rp).
bd-3lht (the hand-held 15-shard header round-trip) is moot: the block no
longer exists.

Comments and docs only; no shard assignment, workflow step or BUILD
setting changes.

Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
2026-10-15 06:50:08 +00:00

26 KiB

Testing Guide

TESTING.md is the single authority for test commands, test selection, and test design in this repository. Use the commands here for local work.

Building and Testing: Bazel Is the Gate

Bazel is beads' build and test system. CI gates every pull request on the bazel test lanes in .github/workflows/bazel.yml, and several checks exist only as Bazel targets: nogo (go test's vet checks plus the golangci-lint linters, tools/nogo), gofmt (//scripts/repochecks:fmt_test) and the repository guards under scripts/repochecks. A plain go test runs none of them, so it is an inner-loop convenience CI does not enforce; a change is ready when its Bazel lanes pass.

Install Bazelisk as bazel; it reads .bazelversion. The pre-commit hook (nogo on staged packages) and the pre-push hook (the test lane) both need it.

Where actions run is per machine, never committed:

  • Contributors read rbe-west's anonymous, read-only cache with --config=fork-cache: everything CI already executed is a cache hit, misses run on your machine, and nothing is uploaded. Make it the default with build --config=fork-cache in your gitignored .bazelrc.local, or pass it per command (make test BAZEL_FLAGS=--config=fork-cache).
  • Maintainers with an rbe-west client certificate execute remotely with --config=remote-exec; the executor endpoint and TLS lines go in the gitignored user.bazelrc or .bazelrc.local (see the .bazelrc comments).
  • Agent hosts whose ~/.bazelrc names the executor need neither flag.

Each lane, as bazel.yml runs it (add your --config=fork-cache or --config=remote-exec when it is not your default):

Lane (bazel.yml job) Command Notes
Test (bazel-test) make test, i.e. bazel test //... --config=ci PR Core's selection: race, -short, skips. Includes nogo, gofmt and the repository guards. The default gate for every Go change.
Lint, all platforms make ci-pr-lint nogo natively plus the windows/amd64 and darwin/arm64 passes. make lint-changed covers only your changed packages.
Pure-Go (bazel-pure) bazel build --config=pure //cmd/bd:bd //cmd/bd:bd_test cgo off. The job's cmd/bd test subset (PURE_CMD_BD_TESTS) and js/wasm step are in bazel.yml.
Release cross-compile (bazel-release-cross) ./scripts/ci/bazel-release-cross-compile.sh Every go_library/go_binary for each row of scripts/ci/release-targets.txt, cgo off, with nogo.
Integration (bazel-integration) bazel test //... --config=integration The integration-tagged build. Runs with the read-only cache too.
Dolt server (bazel-doltserver) bazel test //... --config=doltserver Starts its own dolt sql-server from the pinned binary; no docker.
cmd/bd Dolt server (bazel-cmd-dolt) bazel test //cmd/bd:bd_dolt_server_test --config=doltserver-cmd 32 shards of the integration-tagged cmd/bd suite (the slowest tests pinned to 16 of them).
Embedded Dolt (bazel-embedded) bazel test //... --config=embedded Remote execution only in CI; locally it is slow. Also runs the served http tiers with BEADS_HTTP_TEST_REQUIRED=1: //backend/http:http_served_test, //internal/httpclient:httpclient_served_test and the command corpus //backend/http/corpus:corpus_served_test, each sharded through tools/bazel/go_test_pinned_shard.sh.
Proxied server (bazel-proxied) bazel test //... --config=doltserver-proxied Remote execution only: 34 shards, each with a Dolt server.
Dolt server + proxied (bazel-dolt-race) bazel test //... --config=dolt-race Both rows above in one invocation; CI runs this instead of them in mode remote.
Server-Dolt storage (bazel-server-storage) bazel test //... --config=doltserver-integration Remote execution only.
Server storage + cmd/bd Dolt server (bazel-dolt-int) bazel test //... --config=dolt-int The two rows above in one invocation; CI runs this instead of them in mode remote. No job-level continue-on-error: it reports a server_storage_result output (plain-gated, like bazel-server-storage) and a cmd_dolt_result output (advisory unless BAZEL_CMD_DOLT_REQUIRED is true, like bazel-cmd-dolt) separately, via a per-tier classify step over the BEP, since its members have different gating categories.
Docs make check-docs //test/docsync:docsync_test and //scripts/repochecks:doc_freshness_test in the test lane's configuration, then scripts/check-doc-flags.sh against the pinned release.

make check runs the testing.Short policy, make ci-pr-lint and make test. After adding, removing or renaming Go files or changing imports or go.mod, run make bazel-sync (gazelle, the go_srcs filegroups and MODULE.bazel).

The Makefile keeps plain-Go twins under explicit -go names: make test-go (./scripts/test.sh), make check-go and make check-docs-go. They work offline and without Bazel, but they are not what CI enforces. GitHub Actions jobs that still run Go-native suites call the -go names, never make test (TestWorkflowsNameTheirTestEngine).

Native macOS C/C++ toolchain

Bazel uses pinned LLVM 22.1.8 Clang on macOS, with the installed Apple SDK and its native linker (xcrun --find ld). MODULE.bazel selects linker = "auto" only for Darwin arm64 and x86_64. The macOS 27 SDK contains arm64e.x1 TAPI targets that the bundled LLVM 22.1.8 ld64.lld cannot parse; pairing the SDK with Apple's linker avoids that incompatibility without changing the shared SDK, Xcode selection, or installed compiler. Darwin's SDK/linker are host dependencies, so this is not a hermetic macOS build.

Linux x86_64 still uses the SHA-pinned LLVM compiler and bundled LLD with the pinned Ubuntu 24.04 sysroot. Host C/C++ toolchain detection stays disabled on every platform; the Darwin override does not select local_config_cc. scripts/bazel_hermetic_cc_test.go guards both contracts. A focused local race build exercises the native cgo link through Go's race standard library:

bazel test //scripts:scripts_test --config=ci --test_filter='^TestBazelHermetic'

This link check is separate from full macOS gate portability: the Linux-only timeout test backends remain tracked in upstream issue #7362. Run the full make test gate on its supported Linux platform as well.

Choose the Smallest Useful Test

Test at the lowest seam that can fail for the user-visible reason. Add a higher tier only when it covers a distinct risk that the lower tier cannot show: integration wiring, a real persistence property, a process boundary, or an external contract.

This keeps feedback fast and failures readable. It does not mean every test is a unit test: use the real boundary when the defect could live there.

Need Run When
Docs-only validation git diff --check and make check-docs For prose-only changes; add any generated-doc or surface-specific link check the changed paths require. Do not run the full suite merely because a Markdown file changed.
Focused red/green loop bazel test //path/to/package:package_test --config=ci --test_filter='^TestExactName$', or the plain-Go ./scripts/test.sh -run '^TestExactName$' ./path/to/package/... While writing or fixing one behavior. The go test loop is fine here; the Bazel targets below are the gate.
Affected-package confidence bazel test //path/to/package/... --config=ci After the focused test passes; include directly affected neighbors when their contract changed.
Final gate make test Once after focused work on Go code is green: the whole test lane, mostly cache hits.
Another lane's risk that lane's command from the table above When the change touches what that lane covers (integration-tagged files, the Dolt server path, embedded Dolt, pure-Go builds).
Named CI wrapper make ci-pr-core or make ci-pr-lint Run the wrapper whose risk or surface is affected, or use it to reproduce that CI check. Do not run all three routinely for every edit.
Hook shims against real timeout implementations bazel test //tests/hook_timeout_backends:hook_timeout_backends_test After changing the hook generator in cmd/bd/hooks.go (then make githooks-regen) or anything under .githooks/. Runs the tracked managed sections against GNU coreutils, uutils, busybox and toybox timeout — the multicalls also installed as gtimeout alone — with and without Perl, under dash, bash and busybox ash (240 cases). About one deadline of wall time; needs no Go build. The PR-core lane runs it on every PR.

Do not replace the focused loop with repeated full-suite runs. Run the final make test once the affected tests are green. For docs-only changes, use the docs, link, and diff checks instead.

Git Hooks

make install points core.hooksPath at .githooks. The pre-commit hook formats staged Go files and runs nogo on their packages (make lint-changed LINT_CHANGED_SCOPE=staged). The pre-push hook runs scripts/pre-push-suite.sh for branch pushes whose commits change Go or Bazel inputs: bazel test //... --config=ci, with --config=remote-exec when some rc file names an executor and --config=fork-cache otherwise. BD_PREPUSH_SUITE=rbe|cache|go picks the mode; go (and the fallback when bazel is not installed) runs make test-go under a banner saying it is not the suite CI gates on.

Plain go test (Inner Loop)

./scripts/test.sh is the plain-Go runner behind make test-go. It sources .buildflags, creates an isolated test environment, applies .test-skip, and defaults to a 25m per-package Go-test timeout. The timeout is a hang backstop, not a target runtime. Override it only when diagnosing a legitimate slow path:

TEST_TIMEOUT=30m ./scripts/test.sh ./cmd/bd/...
TEST_VERBOSE=1 ./scripts/test.sh ./cmd/bd/...
TEST_RUN='^TestExactName$' ./scripts/test.sh ./cmd/bd/...

# Equivalent command-line options.
./scripts/test.sh -v -run '^TestExactName$' ./cmd/bd/...
./scripts/test.sh -timeout 30m ./cmd/bd/...

Use the opt-in ICU regex path only when the change requires it:

make test-icu-path

It is maintainer-only and not part of normal validation. make test-full-cgo and ./scripts/test-cgo.sh remain deprecated compatibility aliases.

Use a named specialized target only when its risk is in scope:

make test-regression
make test-upgrade
make test-cross-version
make test-migration

For a failing GitHub Actions check, run the command of its bazel.yml lane from the table above; for the jobs that are not Bazel lanes, follow the workflow and its Makefile target.

Test Environment and Readiness

The runner isolates HOME, Git configuration, and Dolt state. By default its test environment adds dolt to BEADS_TEST_SKIP; set BEADS_TEST_ENV_RUN_DOLT=1 only when deliberately exercising the Dolt path and its prerequisites are available. Do not make ordinary tests depend on a developer's database, daemon, global Git configuration, or filesystem state.

To skip an optional service explicitly, use the existing skip mechanism:

BEADS_TEST_SKIP=dolt ./scripts/test.sh ./...

Tests that need a Dolt SQL server get one from internal/testutil (EnsureDoltContainerForTestMain, RequireDoltContainer, StartIsolatedDoltContainer[Handle], NewContainerProvider). Two backends sit behind that API, selected by BEADS_TEST_DOLT_SERVER:

  • container: the dolthub/dolt-sql-server image through testcontainers (needs docker and the pulled image). The default under plain go test.
  • local: a dolt sql-server started by the test process from the pinned dolt CLI (BEADS_TEST_DOLT_BINARY, else dolt on PATH; it must be the image's version). No docker. Used only when explicitly selected, under go test and bazel test alike (a Bazel target's env, or --test_env=BEADS_TEST_DOLT_SERVER=local); unset means container, which in a Bazel action without docker keeps the usual skip.

BEADS_TEST_REQUIRE_DOLT_CONTAINER=1 turns an unavailable backend into a failure (per test and in every TestMain) instead of a skip; lanes that exist to run the Dolt suites set it. BEADS_TEST_REQUIRE_SOCAT=1 does the same for the proxied subtests that bridge an external endpoint with socat (external-unix, the outage/reconnect matrix); //cmd/bd:bd_proxied_test sets it; the legacy GitHub proxied jobs retired in #7353 had no socat and left it unset.

Under Bazel, bazel test //... --config=doltserver runs the Dolt-backed domain, uow, tracker, doctor/fix, protocol and testutil suites on the local backend (the dolt-server targets); they need no docker and execute remotely with --config=remote-exec. No Bazel target uses the container backend; run go test with docker and the pulled image to exercise it. PR Risk's heavier server tiers have configs of their own, run by bazel.yml only with remote execution, each in a job of its own (bazel-proxied, which in mode remote is folded with bazel-doltserver into bazel-dolt-race, and bazel-server-storage, which in mode remote is folded with bazel-cmd-dolt into bazel-dolt-int): --config=doltserver-proxied is the proxied-server cmd/bd tier ("Test (Proxied Dolt Cmd N/15)", //cmd/bd:bd_proxied_test), and --config=doltserver-integration the server-Dolt storage tier ("Test (Server Dolt Conformance)", "Test (Server Dolt Full Suite N/16)", //internal/storage/dolt:dolt_server_*_test), which builds with the integration tag like --config=integration. Each shard runs its CI job's shard script, so for --config=doltserver-integration Bazel shard k runs the tests of job k+1 (both split the manifest's 16-shard block the same way). --config=doltserver-proxied's bd_proxied_test instead runs the manifest's own 34-shard block, bin-packed by measured duration (the manifest's legacy 15-shard, bd-init-cost-proxy block, once read by PR Risk's retired fork/push jobs, has been deleted; those jobs no longer exist).

--config=doltserver-cmd (//cmd/bd:bd_dolt_server_test, bazel.yml's advisory bazel-cmd-dolt job) runs the whole integration-tagged cmd/bd suite on the local backend: the Dolt-gated cmd/bd tests (TestCLI_*, the init and store-backed suites) that every other lane skips with BEADS_TEST_SKIP=dolt or leaves out of its manifest. It shares --config=integration's build and runs it in 32 shards: 16 run the slowest tests cmd/bd/dolt_server_pinned_shards.txt pins to them, and 16 are the Go binary's own round-robin shards over every other top-level test (tools/bazel/go_test_pinned_shard.sh; //tests/regression:regression_test splits the same way). After a change that moves test durations, regenerate the manifest from a run of the lane with tools/bazel/pin_shards.py (see its usage); a stale manifest only costs balance, never coverage. It runs where the integration lane runs (remote, or with the read-only cache). pr.yml's gate requires it once BAZEL_CMD_DOLT_REQUIRED is "true"; pr.yml then also passes bazel.yml cmd-dolt-required: true, and the PR's bazel-integration lane runs //... -//cmd/bd:bd_test, so each cmd/bd integration-build test runs once (push, nightly and bazel-farm runs keep bd_test in the integration lane). Locally: bazel test //cmd/bd:bd_dolt_server_test --config=doltserver-cmd.

An ambient BEADS_DOLT_SERVER_PORT or BEADS_DOLT_PORT is never honored by the suites that call testutil.EnsureDoltContainerForTestMain. When a test container is started, that container's port overwrites both variables; when one cannot be started -- for any reason, including BEADS_TEST_SKIP=dolt -- both are cleared, so no environment-named server can be resolved. Point a test run at a specific Dolt server by starting a container for it, not by exporting a port.

The one sanctioned way to hand these suites a server you started yourself is ./scripts/test.sh with BEADS_TEST_SHARED_SERVER=1: it starts one dolt sql-server, exports its port as BEADS_DOLT_PORT, and marks it by exporting BEADS_TEST_SHARED_DOLT_SERVER set to that same port number. It starts nothing if either port variable is still set when it gets there. The runner's test environment clears both first, so that only happens when that isolation is skipped (for example BEADS_TEST_ENV_DISABLE=1), and the script then says so on stderr. The marker -- set only by that script, only for a port it allocated -- is what the helper treats as container-equivalent, and only for a variable holding exactly the port it names: that variable survives the clearing above, and any other port variable is still cleared. A container still wins where one can be started; the shared server is the Docker-less path.

Clearing the environment variables closes the channel the ambient port travelled on; it does not make a port unresolvable in general. Resolution continues into the file chain (.beads/dolt-server.port, config.yaml, metadata.json), and internal/storage/dolt's production-port detection is narrower than that resolution -- see be-rl6tm, which tracks the remaining gap.

Tests that need a temporary repository or store should use t.TempDir() and t.Cleanup(). Temporary repositories must set a repository-local hooks path; do not inherit the developer's global hooks configuration.

In cmd/bd, fresh-workspace command fixtures should call isolateBeadsDirForTest(t) before setup or dispatch. It clears inherited BEADS_DIR and restores that variable exactly at cleanup, even after raw command-dispatch mutations. The TestMain reset only isolates startup. These fixtures must not use t.Parallel(). Tests intentionally selecting a workspace should use t.Setenv("BEADS_DIR", ...); initConfigForTest and ensureCleanGlobalState preserve that selection.

For manual CLI experiments, run both initialization and subsequent commands from a disposable working directory:

beads_manual_dir="$(mktemp -d)"
(
  set -e
  cd "$beads_manual_dir"
  bd init --quiet --prefix test --skip-hooks --skip-agents
  bd create "Test issue" -p 1
)
rm -rf -- "$beads_manual_dir"

BEADS_DB selects a database for database-opening commands, but it does not by itself redirect bd init workspace setup. Never run a manual bd init from a production workspace merely because BEADS_DB points elsewhere.

Tmpfs hosts: the cmd/bd test suite creates an isolated $HOME and several test binaries under $TMPDIR. They are normally cleaned by the test process, but a SIGKILLed or OOMed run can leave orphans behind. On hosts where /tmp is tmpfs (e.g. Fedora Atomic / Bluefin), run make clean-test-tmp between test runs if du -sh /tmp/beads-* /tmp/bd-* shows accumulation. See bd-3q2u.

testing.Short() is for genuine runtime, stress, or large-fixture skips. It is not a substitute for declaring an integration, end-to-end, API, Docker, or external-dependency boundary. Keep new uses within the repository policy:

make check-testing-short

Dolt Container Tests (podman-rootless)

Anything that does not carry BEADS_TEST_SKIP=dolt — including BEADS_TEST_ENV_RUN_DOLT=1 and a bare go test — reaches a real dolt sql-server: by default through testcontainers-go where a container runtime and the pinned image are present (otherwise those suites self-skip), and from the local dolt CLI in the suites that use testutil.RequireDoltBinary. Lanes that set BEADS_TEST_REQUIRE_DOLT_CONTAINER=1 fail instead of skipping. Two limitations of the containerized path are worth recognizing before reading a failure as a product bug. Neither failure mode manifests in production or on GitHub Actions: CI exercises this same containerized path green on every risk-tier PR, and .github/workflows/pr-risk.yml pulls the pinned image precisely to defeat the self-skip. A hang there is a real bug, not this section's subject.

Migration 0032 hangs over the wire protocol. Migration 0032 (drop_schema_migrations_applied_at) hangs indefinitely when applied through the containerized sql-server. How it surfaces depends on the caller's context. A go test ./internal/storage/uow/... -count=1 with no BEADS_TEST_SKIP passes context.Background(), so it sits in initSchema until the package timeout rather than failing — that is the shape you will normally see, with no deadline to cut it short. A caller that does supply one gets the hang reported by the server-mode store open (newServerMode in internal/storage/dolt/store.go) instead:

failed to initialize schema: context deadline exceeded

The cause is the podman-rootless container port-forwarding path, not migration 0032's SQL: the identical statement completes normally through the embedded/CLI engine, and against a bare-host-process dolt sql-server matching the deployed shape. Evidence chain in be-j3szz. Use BEADS_TEST_SKIP=dolt unless the container path is what you are testing.

Disabling Ryuk removes the container safety net. This repository sets no TESTCONTAINERS_RYUK_DISABLED, so CI runs with Ryuk — testcontainers-go's orphan-reaper sidecar — enabled. The podman-rootless flow generally requires turning it off by hand (TESTCONTAINERS_RYUK_DISABLED=true), because under rootless podman Ryuk frequently cannot start: it wants the runtime socket. See be-w3n2m. With Ryuk off, nothing reaps a container whose test process exited without running its cleanup — os.Exit reached before a deferred TerminateDoltContainer, for instance. be-5kkk6 records the cost: 101 leaked containers, and an exhausted swap. When running with Ryuk disabled, keep teardown on the normal return path using the testMainInner pattern (beads_test.go), and check for strays afterwards. Carry the rootless socket this flow runs on — a bare docker ps talks to the CLI's default endpoint and reports a false all-clear — and read the tag from the pin rather than copying it, so the command cannot drift when the pin moves. An empty read would be the same false all-clear (ancestor= matches nothing), so the command refuses to run without a tag:

dolt_image=$(sed -n 's/.*DoltDockerImage = "\(.*\)".*/\1/p' \
  "$(git rev-parse --show-toplevel)/internal/testutil/testdoltcommon.go")
DOCKER_HOST=unix:///run/user/$(id -u)/podman/podman.sock \
  docker ps -a --filter "ancestor=${dolt_image:?could not read the Dolt image pin}"

Test Design

Seams, Scenarios, and Doubles

Write one scenario at the smallest seam that demonstrates the behavior. Cover the boundary or failure mode that changes the user result; use table-driven subtests when examples share setup. Do not repeat the same scenario through a helper, every caller, and the CLI merely because all are available.

A recording double or fake should be narrow: model only the calls, inputs, outputs, and failures the test needs. It should not recreate a storage engine, process manager, or another subsystem just to make a unit test look realistic.

A behavioral fake is different. If it stands in for a contract shared by multiple production implementations, give it the same semantic-conformance suite as those implementations. That shared suite defines observable behavior; it prevents the fake from teaching callers a contract production code does not honor.

Semantic conformance asks whether an implementation produces the promised results, errors, and state transitions for the same operation. Persistence conformance asks whether the real persistence boundary preserves its required durability, transaction, migration, and recovery properties. They answer different questions. Do not claim backend parity unless a stated contract and its conformance suite establish it.

Tier Admission

An integration test belongs above the unit seam only when it covers a distinct boundary that a narrow double cannot prove, such as configuration wiring, a real filesystem or Git interaction, a subprocess protocol, or persistence behavior.

An end-to-end test is admitted only when all of these are true:

  1. The failure would be user-visible.
  2. A real process, setup, or wiring boundary owns a distinct failure risk that no lower seam can prove.
  3. Lower-tier tests cover the underlying behavior where practical, leaving the end-to-end test focused on that boundary.
  4. No existing end-to-end test already covers the same boundary risk.

Lower-tier coverage of the same user journey does not disqualify the end-to-end test; duplicate coverage of the same boundary risk does. State that risk in the test name or nearby documentation.

Avoid Incidental Complexity

Avoid these patterns unless they are the behavior under test:

  • Duplicate scenarios at several layers.
  • Global-state reset choreography. Prefer per-test state and cleanup.
  • Subprocesses, listeners, sleeps, or real-store setup for a unit-level claim.
  • Assertions on private implementation shape when observable behavior is the contract.
  • Performance claims without a repeatable measurement and stated workload.

Sleeps, listeners, and real stores are appropriate when the test is specifically about timing, lifecycle, protocol, or persistence. Keep the setup scoped and make that reason apparent.

Failures, Skips, and Review

.test-skip is a local, temporary exception list. If an unrelated failure is already listed, report it rather than silently broadening the skip. Before adding a new skip, record the issue it tracks and remove the skip when the underlying failure is fixed.

Before opening a PR:

  1. For docs-only changes, run the applicable docs, link, freshness, and diff checks; do not run the full suite by default.
  2. For Go code, keep the focused and affected-package tests green, then run one final make test (the Bazel test lane), plus the lane of any other tier the change touches.
  3. Run only the named CI wrapper, specialized target, or risk gate required by the changed surface, or the one needed to reproduce a CI result.

For historical CI inventory and maintainer planning context, consult CI_TEST_SURFACE_AUDIT.md and CI_CLEANUP_PLAN.md. For current commands, use the workflow files and Makefile; these context documents are not a second testing guide or a live CI inventory.