fix(datagrid): carry identity columns through so a new row pre-fills DEFAULT #2085

Workflow file for this run

.github/workflows/macos-tests.yml at 625a802

	name: macOS Tests

	on:
	# No paths filter here on purpose. A required status check has to report on every pull
	# request, and a workflow skipped by a paths filter reports nothing at all, which leaves
	# the check pending forever and blocks docs-only work. The filtering moved into the
	# changes job below, so the gate job always runs and always reports.
	pull_request:
	push:
	branches: [main]
	paths:
	- "TablePro/**"
	- "Plugins/**"
	- "Packages/**"
	- "LocalPackages/**"
	- "TableProTests/**"
	- "TableProUITests/**"
	- "Native/**"
	- "project.yml"
	- "TableProMobile/project.yml"
	- "Configs/**"
	- "Libs/**"
	- "scripts/**"
	- "TablePro.xcodeproj/project.xcworkspace/xcshareddata/swiftpm/Package.resolved"
	- ".github/actions/**"
	- ".github/macos-test-quarantine.txt"
	- ".github/macos-ui-test-quarantine.txt"
	- ".github/workflows/macos-tests.yml"
	workflow_dispatch:
	workflow_call:

	# One run per PR/branch at a time. On a pull request a new push cancels the run in flight, because
	# only the newest commit's verdict is wanted and the older one is about to be replaced anyway.
	#
	# A push to main is different: its run is the only verdict that commit will ever get, and cancelling
	# it leaves main untested. That is not hypothetical. Of the last 30 main runs of this workflow, 13
	# were cancelled by the next merge, and the merges are frequent enough that whole days passed with
	# no conclusive result on main at all.
	#
	# Queueing them does not pile up. With cancel-in-progress false GitHub still replaces the pending
	# run in the group: "any existing pending job or workflow in the same concurrency group will be
	# canceled and the new queued job or workflow will take its place." So a burst of merges holds at
	# most one running plus one queued, and the one that ends up queued is the newest.
	concurrency:
	group: macos-tests-${{ github.ref }}
	cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}

	# A downgrade or an exact match is always allowed for a called workflow; only asking for more than
	# the caller holds fails at run creation. Declaring read here therefore cannot break build.yml's
	# workflow_call, and it stops every job in this file from inheriting a write token if the repository
	# default is ever flipped.
	permissions:
	contents: read

	env:
	XCODE_VERSION: "26.4.1"
	XCODE_PROJECT: TablePro.xcodeproj
	XCODE_SCHEME: TablePro
	TEST_DESTINATION: "platform=macOS"
	# Inside the workspace rather than the default location, because the build job tars it and the
	# test jobs unpack it. Every path inside the generated .xctestrun is written relative to
	# __TESTROOT__, so the products relocate to any directory.
	DERIVED_DATA: DerivedData

	jobs:
	# Reproduces the paths filter this workflow used to carry on `on: pull_request`. Keep this
	# list and the `push:` paths above identical, so a pull request and a push to main run the
	# same suites. `Native/` is on both because the Dameng steps below build and test the Rust
	# bridge that lives there, and a bridge-only change would otherwise compile nowhere.
	changes:
	name: Detect relevant changes
	runs-on: ubuntu-latest
	timeout-minutes: 5
	outputs:
	run: ${{ steps.decide.outputs.run }}
	steps:
	- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
	with:
	fetch-depth: 0

	- name: Decide whether the macOS suites need to run
	id: decide
	env:
	EVENT_NAME: ${{ github.event_name }}
	BASE_SHA: ${{ github.event.pull_request.base.sha }}
	REF: ${{ github.ref }}
	# The reasoning lives in scripts/ci/detect-changed-paths.sh. It was inline here and in
	# ios-tests.yml, with a comment asking a human to keep the two copies in step and nothing
	# enforcing it.
	run: \|
	set -euo pipefail
	RUN=$(scripts/ci/detect-changed-paths.sh \
	--event "$EVENT_NAME" \
	--base "$BASE_SHA" \
	--ref "$REF" \
	--skip-release-commit \
	'TablePro/' 'Plugins/' 'Packages/' 'LocalPackages/' \
	'TableProTests/' 'TableProUITests/' 'Native/' 'Configs/' 'Libs/' 'scripts/' \
	'\.github/actions/' '\.github/macos-(ui-)?test-quarantine\.txt$' \
	'TablePro\.xcodeproj/project\.xcworkspace/xcshareddata/swiftpm/Package\.resolved$' \
	'TableProMobile/project\.yml$' 'project\.yml$' \
	'\.github/workflows/macos-tests\.yml$')
	echo "run=$RUN" >> "$GITHUB_OUTPUT"

	packages:
	name: Package Tests
	needs: changes
	if: needs.changes.outputs.run == 'true'
	runs-on: macos-26
	timeout-minutes: 20
	steps:
	- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1

	- name: Select Xcode
	uses: maxim-lobanov/setup-xcode@ed7a3b1fda3918c0306d1b724322adc0b8cc0a90 # v1.7.0
	with:
	xcode-version: ${{ env.XCODE_VERSION }}

	- name: Run TableProCore package tests
	run: swift test --package-path Packages/TableProCore

	# CodeEditSourceEditor is deliberately not run here. `swift test` cannot build it on this
	# runner: its CodeEditSymbols dependency fails with "type 'Bundle' has no member 'module'",
	# after a SwiftPM cache warning about a missing maintenance.lock. It builds fine inside the
	# Xcode project, so the app target still compiles it; only the standalone package test run
	# is broken. Its own suites also fail on main (HighlighterTests, TagEditingTests, plus one
	# that aborts the runner), so this needs fixing at the package level before it can gate.
	- name: Run CodeEditTextView package tests
	run: swift test --package-path LocalPackages/CodeEditTextView

	# Compiles the app, both test bundles and all 31 plugins once, then hands the products to the
	# test jobs. That build used to happen inside the same job that ran the tests, which meant the
	# unit step spent 12 of its 15.3 minutes compiling and the UI step could not start until the
	# unit step had finished. All 12,662 unit cases execute in 206 seconds, so the build was almost
	# all of the cost and none of it was parallel.
	build:
	name: Build for testing
	needs: changes
	if: needs.changes.outputs.run == 'true'
	runs-on: macos-26
	timeout-minutes: 45
	steps:
	- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1

	- name: Select Xcode
	uses: maxim-lobanov/setup-xcode@ed7a3b1fda3918c0306d1b724322adc0b8cc0a90 # v1.7.0
	with:
	xcode-version: ${{ env.XCODE_VERSION }}

	- name: Install xcbeautify
	run: brew list xcbeautify &>/dev/null \|\| brew install xcbeautify

	- name: Cache static libraries
	uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
	with:
	path: Libs
	key: ${{ runner.os }}-libs-${{ hashFiles('Libs/checksums.sha256', 'Plugins/MSSQLDriverPlugin/CFreeTDS/include/sybdb.h') }}

	- name: Download static libraries
	env:
	GH_TOKEN: ${{ github.token }}
	run: scripts/download-libs.sh

	# Compiles a C probe against Libs/libbson and parses every filter document the MongoDB
	# query builder can emit. It guarded that invariant and ran nowhere; it needs clang and the
	# vendored libraries, so this is the first job where it can run at all.
	- name: Check the MongoDB filter shapes against the query builder
	run: scripts/check-mongodb-filter-shapes.sh

	# Cargo and rustup need access outside Xcode's user-script sandbox. Build the
	# native bridge explicitly before compiling the driver target.
	- name: Build Dameng native bridge
	run: scripts/build-dameng.sh arm64

	# Only the bridge's own FFI tests live here. The DM8 protocol crates are a pinned
	# revision of TableProApp/rust-dameng and carry their own suite in that repository.
	# Run from the bridge directory so rustup reads its rust-toolchain.toml: rustup resolves
	# that file from the working directory, not from --manifest-path, and testing on the
	# runner's default Rust would not be the toolchain the shipped staticlib is built with.
	- name: Test Dameng native bridge
	working-directory: Native/DamengBridge
	run: cargo test --locked

	# Secrets.xcconfig is gitignored. Tests do not need analytics keys, and the project only
	# needs the variable to resolve, so an empty value is enough and no secret is read here.
	- name: Create Secrets.xcconfig
	run: echo "ANALYTICS_HMAC_SECRET = " > Configs/Secrets.xcconfig

	- name: Setup XcodeGen
	uses: ./.github/actions/setup-xcodegen

	- name: Generate Xcode project
	run: scripts/generate-project.sh

	# Package.resolved pins every revision and is tracked, which is why it can key the cache.
	# build.yml has cached these checkouts for a while; this workflow paid 1.5 minutes a run
	# re-resolving them.
	- name: Cache Swift package checkouts
	uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
	with:
	path: ~/.spm-cache
	key: ${{ runner.os }}-spm-${{ hashFiles('TablePro.xcodeproj/project.xcworkspace/xcshareddata/swiftpm/Package.resolved') }}
	restore-keys: ${{ runner.os }}-spm-

	- name: Resolve Swift package dependencies
	run: \|
	xcodebuild -resolvePackageDependencies \
	-project "$XCODE_PROJECT" \
	-scheme "$XCODE_SCHEME" \
	-clonedSourcePackagesDirPath ~/.spm-cache \
	-skipPackagePluginValidation

	# The TablePro scheme only builds the 14 bundled plugins, so a change confined to one
	# of the 17 registry-only plugins compiles nowhere else in CI. AllPlugins is the only
	# target that covers them, and build-plugin.yml does not run until a release tag.
	- name: Compile every plugin
	run: \|
	set -o pipefail
	xcodebuild build \
	-project "$XCODE_PROJECT" \
	-scheme AllPlugins \
	-destination "$TEST_DESTINATION" \
	-derivedDataPath "$DERIVED_DATA" \
	-clonedSourcePackagesDirPath ~/.spm-cache \
	-skipPackagePluginValidation \
	CODE_SIGNING_ALLOWED=NO \
	\| xcbeautify --renderer github-actions

	# Builds and signs used to be the whole check: nothing ever tried to load one. "Bundle
	# failed to load executable" has shipped twice, and dlopen is exactly the dyld path that
	# fails in that case.
	- name: Verify every plugin loads
	run: scripts/ci/verify-plugin-loads.sh "$DERIVED_DATA/Build/Products/Debug"

	- name: Build the app and both test bundles
	run: \|
	set -o pipefail
	xcodebuild build-for-testing \
	-project "$XCODE_PROJECT" \
	-scheme "$XCODE_SCHEME" \
	-destination "$TEST_DESTINATION" \
	-derivedDataPath "$DERIVED_DATA" \
	-clonedSourcePackagesDirPath ~/.spm-cache \
	-skipPackagePluginValidation \
	CODE_SIGNING_ALLOWED=NO \
	\| xcbeautify --renderer github-actions

	# DamengDriverTests is the one plugin-owned unit bundle: its suites import CDameng, so
	# they cannot move into TableProTests the way pure-logic plugin tests do. Without this
	# step nothing runs them and they rot silently.
	- name: Build the Dameng driver tests
	run: \|
	set -o pipefail
	xcodebuild build-for-testing \
	-project "$XCODE_PROJECT" \
	-scheme DamengDriverTests \
	-destination "$TEST_DESTINATION" \
	-derivedDataPath "$DERIVED_DATA" \
	-clonedSourcePackagesDirPath ~/.spm-cache \
	-skipPackagePluginValidation \
	CODE_SIGNING_ALLOWED=NO \
	\| xcbeautify --renderer github-actions

	# Tarred rather than uploaded as a directory. actions/upload-artifact does not preserve
	# the executable bit or symlinks, and both matter to an .app bundle: the app would arrive
	# unable to run. The archive measures 144 MB against 559 MB on disk.
	- name: Pack the built products
	run: tar -czf products.tar.gz -C "$DERIVED_DATA/Build" Products

	- name: Upload the built products
	uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
	with:
	name: test-products
	path: products.tar.gz
	retention-days: 1
	compression-level: 0

	unit:
	name: Unit tests
	needs: [changes, build]
	if: needs.changes.outputs.run == 'true'
	runs-on: macos-26
	timeout-minutes: 30
	steps:
	- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1

	- name: Select Xcode
	uses: maxim-lobanov/setup-xcode@ed7a3b1fda3918c0306d1b724322adc0b8cc0a90 # v1.7.0
	with:
	xcode-version: ${{ env.XCODE_VERSION }}

	- name: Install xcbeautify
	run: brew list xcbeautify &>/dev/null \|\| brew install xcbeautify

	- uses: ./.github/actions/unpack-test-products

	# Quarantined cases are listed one per line in .github/macos-test-quarantine.txt. Burn that
	# list down. The script enumerates the target first and refuses to run if any entry would
	# skip nothing, because xcodebuild accepts -skip-testing for a case that does not exist and
	# says nothing: that left one line inert for 479 commits, and it is also what happens to a
	# Swift Testing case written without its parentheses.
	- name: Run unit tests
	run: \|
	set -o pipefail
	# To a file on its own line, not through a process substitution. `< <(...)` discards the
	# producer's exit status, so the script refusing an entry that would skip nothing printed
	# its diagnostic, exited 1, and the step carried on to run the whole suite unskipped and
	# report success. The refusal the comment above promises never once failed a job.
	SKIP_FILE="${RUNNER_TEMP:-.}/quarantine-skip-args.txt"
	scripts/ci/quarantine-args.sh .github/macos-test-quarantine.txt TableProTests "$XCTESTRUN" "$TEST_DESTINATION" > "$SKIP_FILE"
	# A read loop, not mapfile: the macOS runner's /bin/bash is 3.2, which has neither
	# mapfile nor readarray, and a `run:` block gets that bash.
	SKIP_ARGS=()
	while IFS= read -r arg; do
	# A blank line would become a literal empty argument, and xcodebuild reads that as a
	# build action and refuses the whole run. An empty array expands to nothing and is
	# fine; one empty string is not.
	[ -n "$arg" ] \|\| continue
	SKIP_ARGS+=("$arg")
	done < "$SKIP_FILE"
	xcodebuild test-without-building \
	-xctestrun "$XCTESTRUN" \
	-destination "$TEST_DESTINATION" \
	-only-testing:TableProTests \
	${SKIP_ARGS[@]+"${SKIP_ARGS[@]}"} \
	-parallel-testing-enabled NO \
	-resultBundlePath TestResults.xcresult \
	\| xcbeautify --renderer github-actions

	- name: Run Dameng driver tests
	run: \|
	set -o pipefail
	xcodebuild test-without-building \
	-xctestrun "$DAMENG_XCTESTRUN" \
	-destination "$TEST_DESTINATION" \
	-parallel-testing-enabled NO \
	\| xcbeautify --renderer github-actions

	# Always, including on failure: a red run is when the durations matter most.
	- name: Summarise test durations
	if: ${{ !cancelled() }}
	run: scripts/ci/summarize-xcresult.sh TestResults.xcresult "Unit tests"

	- name: Upload test results
	if: ${{ !cancelled() }}
	uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
	with:
	name: macos-test-results
	path: TestResults.xcresult
	retention-days: 7

	# UI tests drive the real app, so they need the runner's GUI session and cannot run beside
	# the unit run. Every case launches the app against a throwaway storage sandbox that
	# UITestCase hands it, so nothing here touches a real store.
	#
	# Sharded across runners rather than parallelised on one. Three things in the suite are shared
	# per machine and cannot be made per-worker: the single `com.TablePro.uitest` defaults domain,
	# the preferences sweep in UITestCase's class setUp, and the menu bar itself, which belongs to
	# whichever app is frontmost. A separate runner per shard has its own of all three.
	ui:
	name: UI tests ${{ matrix.shard }}/${{ matrix.of }}
	needs: [changes, build]
	if: needs.changes.outputs.run == 'true'
	runs-on: macos-26
	timeout-minutes: 40
	strategy:
	fail-fast: false
	matrix:
	shard: [0, 1, 2]
	of: [3]
	steps:
	- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1

	- name: Select Xcode
	uses: maxim-lobanov/setup-xcode@ed7a3b1fda3918c0306d1b724322adc0b8cc0a90 # v1.7.0
	with:
	xcode-version: ${{ env.XCODE_VERSION }}

	- name: Install xcbeautify
	run: brew list xcbeautify &>/dev/null \|\| brew install xcbeautify

	- uses: ./.github/actions/unpack-test-products

	# One retry is allowed. A full local suite flaked once in three runs on a single
	# waitForExistence, which without a retry would fail roughly a third of runs and train
	# people to re-run until green. A case that fails twice is a real failure.
	- name: Run UI tests
	env:
	SHARD: ${{ matrix.shard }}
	SHARD_COUNT: ${{ matrix.of }}
	run: \|
	set -o pipefail
	# To a file on its own line, so a failure of the lister fails the step. A process
	# substitution discards its exit status, which reads here as "the shard selected nothing".
	ONLY_FILE="${RUNNER_TEMP:-.}/shard-only-args.txt"
	scripts/ci/list-tests.sh \
	--xctestrun "$XCTESTRUN" \
	--target TableProUITests \
	--quarantine .github/macos-ui-test-quarantine.txt \
	--shard "$SHARD/$SHARD_COUNT" > "$ONLY_FILE"
	# A read loop, not mapfile: the macOS runner's /bin/bash is 3.2, which has neither.
	ONLY_ARGS=()
	while IFS= read -r arg; do
	# The count check below cannot see an empty argument: one blank line makes the count 1
	# and xcodebuild then reads that empty argument as a build action.
	[ -n "$arg" ] \|\| continue
	ONLY_ARGS+=("$arg")
	done < "$ONLY_FILE"
	[ "${#ONLY_ARGS[@]}" -gt 0 ] \|\| { echo "::error::shard $SHARD selected no cases"; exit 1; }
	echo "shard $SHARD of $SHARD_COUNT runs ${#ONLY_ARGS[@]} cases"
	xcodebuild test-without-building \
	-xctestrun "$XCTESTRUN" \
	-destination "$TEST_DESTINATION" \
	"${ONLY_ARGS[@]}" \
	-parallel-testing-enabled NO \
	-test-iterations 2 \
	-retry-tests-on-failure \
	-resultBundlePath UITestResults.xcresult \
	\| xcbeautify --renderer github-actions

	- name: Summarise UI test durations
	if: ${{ !cancelled() }}
	run: scripts/ci/summarize-xcresult.sh UITestResults.xcresult "UI tests ${{ matrix.shard }}/${{ matrix.of }}"

	- name: Upload UI test results
	if: ${{ !cancelled() }}
	uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
	with:
	name: macos-ui-test-results-${{ matrix.shard }}
	path: UITestResults.xcresult
	retention-days: 7

	# The one check to mark required on main. It reports on every pull request, so a docs-only
	# change passes on skipped suites instead of hanging, and a red suite cannot be merged past.
	gate:
	name: macOS Tests Gate
	if: always()
	needs: [changes, packages, build, unit, ui]
	runs-on: ubuntu-latest
	timeout-minutes: 5
	steps:
	- name: Require every suite to have passed or been skipped
	env:
	RESULTS: ${{ join(needs.*.result, ' ') }}
	run: \|
	set -euo pipefail
	echo "suite results: $RESULTS"
	for result in $RESULTS; do
	case "$result" in
	success\|skipped) ;;
	*)
	echo "A macOS suite reported '$result'." >&2
	exit 1
	;;
	esac
	done

Navigation Menu

Uh oh!

Uh oh!

fix(datagrid): carry identity columns through so a new row pre-fills DEFAULT #2085

Workflow file

fix(datagrid): carry identity columns through so a new row pre-fills DEFAULT #2085

Uh oh!

Workflow file for this run