Headless integration tests drive the real TUI model over a seeded git repository and assert on the rendered output, with no terminal.
Running · Layout · Harness · Fixture · Inventory · Notes
go test ./library/tui/test/ # the quick tier
GITSOCIAL_TEST_FULL=1 go test ./library/tui/test/ # the full tier: adds TestSmoke, TestSequence, TestGolden/LayoutProperties
go test ./library/tui/test/ -run Golden -update # regenerate the golden files
go test ./library/tui/test/ -run TestGenerateFixture -generate # regenerate the fixture tarballs
scripts/test.sh -run Smoke ./library/tui/test/ # any of the above with streamed per-test progressThe tests create temporary directories and need no external services. Run times with a warm build cache:
| Run | Time |
|---|---|
| quick tier | 13 s |
| full tier | 78 s |
TestSmoke | 40 s |
TestGolden/LayoutProperties | 20 s |
library/tui/test/
├── harness.go # the headless model driver
├── fixture.go # repository setup and data seeding
├── assert.go # render assertions: ANSI stripping, pattern matching
├── main_test.go # the shared fixture, via TestMain
├── generate_test.go # fixture tarball generation, behind -generate
├── smoke_test.go # every key on every view
├── display_test.go # seeded content renders
├── golden_test.go # golden files and layout properties
├── navigation_test.go # view transitions
├── sequence_test.go # multi-step interactions
├── form_test.go # form submits
├── diff_test.go # diff view keys
├── cursor_test.go # list cursor stability
├── history_diff_test.go # history-diff footers
├── stack_test.go # stacked pull requests
├── proposal_test.go # cross-repo proposals
├── anchors_test.go # detail card anchors against its links
└── testdata/ # fixture-repo.tar.gz, fixture-fork.tar.gz, fixture.json, *.goldenHarness wraps tui.NewModel, sends tea.KeyMsg and tea.WindowSizeMsg directly, and reads the rendered output.
func New(t *testing.T, workdir, cacheDir string) *Harness // 120x40, Init() run, startup commands drained
func (h *Harness) SendKey(key string)
func (h *Harness) SendKeys(keys ...string)
func (h *Harness) Navigate(path string)
func (h *Harness) NavigateTo(loc tuicore.Location)
func (h *Harness) SetSize(w, h int)
func (h *Harness) DrainCmds()
func (h *Harness) Rendered() string
func (h *Harness) CurrentPath() string
func (h *Harness) CurrentContext() tuicore.Context
func (h *Harness) CurrentView() tuicore.View
func (h *Harness) BindingsForContext(ctx tuicore.Context) []tuicore.BindingEvery SendKey and Navigate drains the commands it produces. Key names: enter, esc, tab, shift+tab, the arrows, ctrl+c, ctrl+d, ctrl+u, space, backspace, home, end, pgup, pgdown; every other string is sent as runes.
Assertions: stripANSI, rendered(h), assertContains(t, output, substr), assertRendersItem(t, h, loc, want...), assertNotEmpty, assertMaxWidth(t, output, maxCols), assertFitsTerminal(t, h), assertFrameFits(t, frame, cols, lines).
Prefer assertRendersItem. It navigates, requires every named fragment of seeded content, and fails on an empty expectation. assertNotEmpty passes when a view shows only its frame (borders, title and footer), so use it only for a view with no seeded data.
TestMain extracts two repositories once per run, and the tests share them read-only:
testdata/fixture-repo.tar.gz: the workspace, with origin https://github.com/user/repotestdata/fixture-fork.tar.gz: a fork, with origin https://github.com/bob/repo, that has one cross-repo edit of the workspace issueEntity ids are in testdata/fixture.json.
func getFixture(t *testing.T) *Fixture // the shared fixture, for read-only tests
func SetupFixture(t *testing.T) *Fixture // a fresh copy, for tests that mutate the repositorySetupFixture closes the shared cache first and opens it again on cleanup; cache.Open is a no-op while a cache is open.
The extension APIs seed this data, taken from the protocol specs:
stack_test.go adds a dependent pull request to its own copyThe cache is filled with client.SyncWorkspace, workspace first and fork second. A cross-repo edit resolves only after its canonical version is in the cache. Commit timestamps are rewritten one second apart, with the last one fixtureCommitAge before the run, so each row shows the same relative time for the full run and the order is deterministic. Fixture generation sets HOME, XDG_CONFIG_HOME and GITSOCIAL_PERSONAL_REPO to a temporary directory.
seedMemoTiersPanic builds the memo tiers that are outside the workspace repository: the personal bare repository, the session tui-fixture, and the repository behind the inherits ref. Each has one memo, written through the memo API. TestMain sets MEMO_SESSION_ID to that session, and the repositories are under the isolated HOME. They stay after a fixture ends, so setup keeps an existing repository and only re-indexes it. Their commit dates are fixtureCommitAge before the run. The session picker reads the session age from the repository, not from the cache.
| File | Tests | Covers |
|---|---|---|
smoke_test.go | TestSmoke/AllKeysAllViews, GlobalShortcuts, UnregisteredKeysIgnored | each view with the keys of its context and one key that nothing binds, then each global shortcut once, on the first view that binds it (86 subtests), with no panic; full tier only |
display_test.go | TestDisplay/*: Timeline, Search, MyRepository, Board, IssuesList, Milestones, Sprints, PRList, ReleasesList, Notifications, Memos, ProjectMemos, MemoDetail, MemoHistory, MemoInherits, Forks, Settings, Site, Cache, Help | seeded content appears on each view |
golden_test.go | TestGolden/{timeline,board,issues,pr_list,releases,settings,help,memo_personal,memo_session,memo_session_items,memo_inherited,memo_inherits}_120x40 | ANSI-stripped renders against testdata/*.golden, each frame also 120 columns wide and 40 lines; the memo goldens have one memo per tier and the inherited source |
golden_test.go | TestGolden/LayoutProperties | each route fits the width and the height at 120x40, 80x24 and 200x60; full tier only |
golden_test.go | TestGoldenWidth_rejectsNavOneColumnTooWide | a navigation row one column too wide fails the check |
golden_test.go | TestGoldenTime_labelHoldsForALongRun | the fixture keeps one relative-time label for the full run |
navigation_test.go | TestNavigation/GlobalKeys (S, P, R, V, M), Back, SiteEditToggle, MultiLevelBack, Detail, Search, Help, Notifications | the global jump keys open their routes; esc goes back, and /, ? and @ open search, help and notifications |
sequence_test.go | TestSequence/*: AllExtensions, BrowseAndReturn, IssuesFlow, SettingsAndBack, QuickJumpOverridesHistory, the *OpensForm and *Navigates flows per item type, PostRetractShowsConfirm, SearchFlow, PRDiffNavigates, MultipleViewRenders, PushConfirmNamesRemote | multi-step flows; full tier only |
form_test.go | TestMilestoneFormCreates, TestSprintFormCreates | a submitted create form opens the detail view of the new item, on an isolated fixture |
diff_test.go | TestDiffTabMovesToNextFile | tab in a diff moves the cursor to the next file and leaves the panel focus alone, on an isolated fixture with a two-file commit |
cursor_test.go | TestTimelineCursorSurvivesFetch, TestTimelineCursorSurvivesBackNav | the timeline selection stays after a fetch and after a return from a detail view |
history_diff_test.go | TestHistoryDiffFooter, PostHistoryDiffRenders | every history-diff context registers its footer entries without duplicates, and the view renders |
stack_test.go | TestStackDisplay/BadgeOnPRList, TestStackBindings, TestStackNavigationBackend | the stack badge, the stack keys, and GetStack and getDependents behind them |
proposal_test.go | TestProposalDisplay/{IssueListMarker,IssueDetailBanner,HistoryRow,HistoryFooterOffersAcceptAndDecline}, TestProposalAccept, TestProposalDecline | the ✎ marker, the banner, the history row, and A and X that accept or decline a proposal, on an isolated fixture |
anchors_test.go | Test{Issue,PR,Milestone,Sprint,Release,Memo,Post}Detail_anchorsMatchLinks | the n-th anchor a detail card marks has the location of the n-th entry of its Links(), on each detail view with a body that holds a ref, a url and a mention; the issue carries an assignee, a label and a citing commit, the pull request a foreign head with a tip, a dependency, a reviewer, a label and a closed issue |
tea.BatchMsg. A command that would block is skipped by function name before execution: BlinkCmd, startFetch, tea.Tick (message timeouts, the auto-fetch heartbeat). These messages are dropped after execution: tea.QuitMsg, setWindowTitleMsg, execMsg (editor and process launches, counted in SkippedExecN), cursor.BlinkMsg. SetHeadless(true) skips the terminal-dependent commands in Init().tui.Model.Update returns either tui.Model or *tui.Model; toModel handles both.Activate calls, navigation dead ends and registration-order bugsassertFrameFits finds horizontal overflow on each route in TestGolden/LayoutProperties, at 120x40, 80x24 and 200x60, and assertFitsTerminal finds it on each TestDisplay view. Both measure printable cells with tuicore.AnsiWidth, so a border past the terminal edge fails on its own line, not as a golden diff.