fix(daemon): surface pane resize failures

commit: 25f341a386d81fabfdb29cce0b48fec2260b0b8a
author: brice · 3mo ago · 2026-07-21 05:02 UTC
parent: a2b1d86
signature: GPG VerifiedCD88F9E564E0102646FB6077A1A07C7C7C3E349D · arch-gpg Verified for the committer email

3 files · +66 -1

M internal/usecase/daemon/performance_benchmark_test.go +1 -1


@@ -132,7 +132,7 @@ func TestTransactionalResizeMetrics(t *testing.T) {
132	132		require.Equal(t, uint64(1), metrics.resizeCommits)
133	133		require.Equal(t, uint64(1), metrics.ptyFailures)
134	134		require.Equal(t, uint64(1), metrics.ptyRetries)
135		-	require.Equal(t, uint64(2), metrics.outputFrames, "commit plus successful full-reset retry")
	135	+	require.Equal(t, uint64(3), metrics.outputFrames, "commit, resize-failure notice toast, and successful full-reset retry")
136	136		require.Zero(t, metrics.frameGapEpochs, "every committed epoch emitted its required frame")
137	137		require.Equal(t, []domain.Size{{Cols: 50, Rows: 28}, {Cols: 50, Rows: 28}}, fixture.pty.requested())
138	138		require.Equal(t, domain.Size{Cols: 100, Rows: 30}, fixture.ac.size)


@@ -132,7 +132,7 @@ func TestTransactionalResizeMetrics(t *testing.T) {
132		require.Equal(t, uint64(1), metrics.resizeCommits)	132		require.Equal(t, uint64(1), metrics.resizeCommits)
133		require.Equal(t, uint64(1), metrics.ptyFailures)	133		require.Equal(t, uint64(1), metrics.ptyFailures)
134		require.Equal(t, uint64(1), metrics.ptyRetries)	134		require.Equal(t, uint64(1), metrics.ptyRetries)
135	-	require.Equal(t, uint64(2), metrics.outputFrames, "commit plus successful full-reset retry")	135	+	require.Equal(t, uint64(3), metrics.outputFrames, "commit, resize-failure notice toast, and successful full-reset retry")
136		require.Zero(t, metrics.frameGapEpochs, "every committed epoch emitted its required frame")	136		require.Zero(t, metrics.frameGapEpochs, "every committed epoch emitted its required frame")
137		require.Equal(t, []domain.Size{{Cols: 50, Rows: 28}, {Cols: 50, Rows: 28}}, fixture.pty.requested())	137		require.Equal(t, []domain.Size{{Cols: 50, Rows: 28}, {Cols: 50, Rows: 28}}, fixture.pty.requested())
138		require.Equal(t, domain.Size{Cols: 100, Rows: 30}, fixture.ac.size)	138		require.Equal(t, domain.Size{Cols: 100, Rows: 30}, fixture.ac.size)

M internal/usecase/daemon/transactional_resize.go +14 -0


@@ -71,6 +71,11 @@ func (d Daemon) prepareResize(sess session, size domain.Size) resizePlan {
71	71		// applyResize deliberately holds only a pane's resizeMu around its PTY call;
72	72		// no daemon, session, tab, or pane lock crosses the external boundary.
73	73		func (d Daemon) applyResize(plan resizePlan, current ...func() bool) bool {
	74	+	// Collect degradation across the loop and report once after it returns to
	75	+	// lock-free code: notify acquires sess.mu/d.mu and repaints, so it must never
	76	+	// run while a pane's resizeMu is held across the external PTY call below.
	77	+	failed := 0
	78	+	var lastErr error
74	79		for i := range plan.members {
75	80		if len(current) != 0 && !current[0]() {
76	81		return false

@@ -91,6 +96,8 @@ func (d Daemon) applyResize(plan resizePlan, current ...func() bool) bool {
91	96		m.ok = true
92	97		} else if err := pty.Resize(rectSize(m.rect)); err != nil {
93	98		d.log.Warn("pty resize failed", "err", err)
	99	+	failed++
	100	+	lastErr = err
94	101		d.replayResizePending(m.session, m.tab, m.pane, false, m.rect)
95	102		} else {
96	103		m.ok = true

@@ -99,6 +106,13 @@ func (d Daemon) applyResize(plan resizePlan, current ...func() bool) bool {
99	106		}
100	107		m.pane.resizeMu.Unlock()
101	108		}
	109	+	if failed > 0 {
	110	+	// Every resizeMu (and pane/session/daemon lock) is released here: each was
	111	+	// bracketed inside the loop. plan.members is non-empty because a failure
	112	+	// came from one of its members, so members[0] is always safe.
	113	+	d.notify(plan.members[0].session, domain.NoticeWarn, domain.NoticeResizeFailed,
	114	+	"pane resize failed; retrying in background", lastErr)
	115	+	}
102	116		return true
103	117		}
104	118


@@ -71,6 +71,11 @@ func (d Daemon) prepareResize(sess session, size domain.Size) resizePlan {
71	// applyResize deliberately holds only a pane's resizeMu around its PTY call;	71		// applyResize deliberately holds only a pane's resizeMu around its PTY call;
72	// no daemon, session, tab, or pane lock crosses the external boundary.	72		// no daemon, session, tab, or pane lock crosses the external boundary.
73	func (d Daemon) applyResize(plan resizePlan, current ...func() bool) bool {	73		func (d Daemon) applyResize(plan resizePlan, current ...func() bool) bool {
		74	+	// Collect degradation across the loop and report once after it returns to
		75	+	// lock-free code: notify acquires sess.mu/d.mu and repaints, so it must never
		76	+	// run while a pane's resizeMu is held across the external PTY call below.
		77	+	failed := 0
		78	+	var lastErr error
74	for i := range plan.members {	79		for i := range plan.members {
75	if len(current) != 0 && !current[0]() {	80		if len(current) != 0 && !current[0]() {
76	return false	81		return false

@@ -91,6 +96,8 @@ func (d Daemon) applyResize(plan resizePlan, current ...func() bool) bool {
91	m.ok = true	96		m.ok = true
92	} else if err := pty.Resize(rectSize(m.rect)); err != nil {	97		} else if err := pty.Resize(rectSize(m.rect)); err != nil {
93	d.log.Warn("pty resize failed", "err", err)	98		d.log.Warn("pty resize failed", "err", err)
		99	+	failed++
		100	+	lastErr = err
94	d.replayResizePending(m.session, m.tab, m.pane, false, m.rect)	101		d.replayResizePending(m.session, m.tab, m.pane, false, m.rect)
95	} else {	102		} else {
96	m.ok = true	103		m.ok = true

@@ -99,6 +106,13 @@ func (d Daemon) applyResize(plan resizePlan, current ...func() bool) bool {
99	}	106		}
100	m.pane.resizeMu.Unlock()	107		m.pane.resizeMu.Unlock()
101	}	108		}
		109	+	if failed > 0 {
		110	+	// Every resizeMu (and pane/session/daemon lock) is released here: each was
		111	+	// bracketed inside the loop. plan.members is non-empty because a failure
		112	+	// came from one of its members, so members[0] is always safe.
		113	+	d.notify(plan.members[0].session, domain.NoticeWarn, domain.NoticeResizeFailed,
		114	+	"pane resize failed; retrying in background", lastErr)
		115	+	}
102	return true	116		return true
103	}	117		}
104		118

M internal/usecase/daemon/transactional_resize_test.go +51 -0


@@ -204,6 +204,57 @@ func TestTransactionalResizeApplyDoesNotHoldTabOrPaneLocks(t *testing.T) {
204	204		require.True(t, paneUnlocked, "PTY.Resize must run outside pane.mu")
205	205		}
206	206
	207	+	// P3 degradation: a failing PTY resize is surfaced as exactly one warn notice
	208	+	// per applyResize call, collected inside the resizeMu loop and reported once
	209	+	// after it returns to lock-free code. The count is the meaningful assertion:
	210	+	// notify acquires sess.mu/d.mu and must never run while any resizeMu is held.
	211	+	func TestTransactionalResizeApplyReportsFailureOncePerCall(t *testing.T) {
	212	+	countResizeFailed := func(d *Daemon) (int, domain.Notification) {
	213	+	var last domain.Notification
	214	+	count := 0
	215	+	for _, h := range d.notices.history() {
	216	+	if h.Code == domain.NoticeResizeFailed {
	217	+	count++
	218	+	last = h
	219	+	}
	220	+	}
	221	+	return count, last
	222	+	}
	223	+
	224	+	t.Run("one failed member yields one notice after return", func(t *testing.T) {
	225	+	failing := &transactionalResizePTY{errs: []error{errors.New("scripted resize failure")}}
	226	+	d, sess, _, _ := newManualSessionWithPTYs(t, failing)
	227	+	tb := sess.activeTab()
	228	+	p := tb.focusedPane()
	229	+	plan := resizePlan{members: []resizeMember{{session: sess, tab: tb, pane: p, rect: domain.Rect{Width: 100, Height: 20}, retry: true}}}
	230	+
	231	+	require.Empty(t, d.notices.history(), "no notice is recorded before applyResize")
	232	+	require.True(t, d.applyResize(&plan))
	233	+
	234	+	count, n := countResizeFailed(d)
	235	+	require.Equal(t, 1, count, "exactly one NoticeResizeFailed per applyResize call")
	236	+	require.Equal(t, domain.NoticeWarn, n.Severity)
	237	+	require.Equal(t, "pane resize failed; retrying in background", n.Message)
	238	+	require.False(t, plan.members[0].ok, "failed member is not marked ok")
	239	+	})
	240	+
	241	+	t.Run("two failed members still yield one notice per call", func(t *testing.T) {
	242	+	first := &transactionalResizePTY{errs: []error{errors.New("first fails")}}
	243	+	second := &transactionalResizePTY{errs: []error{errors.New("second fails")}}
	244	+	d, sess, _, _ := newManualSessionWithPTYs(t, first, second)
	245	+	tabs := sess.tabs
	246	+	plan := resizePlan{members: []resizeMember{
	247	+	{session: sess, tab: tabs[0], pane: tabs[0].focusedPane(), rect: domain.Rect{Width: 100, Height: 20}, retry: true},
	248	+	{session: sess, tab: tabs[1], pane: tabs[1].focusedPane(), rect: domain.Rect{Width: 100, Height: 20}, retry: true},
	249	+	}}
	250	+
	251	+	require.True(t, d.applyResize(&plan))
	252	+
	253	+	count, _ := countResizeFailed(d)
	254	+	require.Equal(t, 1, count, "one notice per applyResize call, not per failed member")
	255	+	})
	256	+	}
	257	+
207	258		// S3 acceptance: a failed member never publishes speculative parser/screen or
208	259		// rectangle state. Successful members and the client layout commit together;
209	260		// composition clips/pads the retained failed screen against that new layout.


@@ -204,6 +204,57 @@ func TestTransactionalResizeApplyDoesNotHoldTabOrPaneLocks(t *testing.T) {
204	require.True(t, paneUnlocked, "PTY.Resize must run outside pane.mu")	204		require.True(t, paneUnlocked, "PTY.Resize must run outside pane.mu")
205	}	205		}
206		206
		207	+	// P3 degradation: a failing PTY resize is surfaced as exactly one warn notice
		208	+	// per applyResize call, collected inside the resizeMu loop and reported once
		209	+	// after it returns to lock-free code. The count is the meaningful assertion:
		210	+	// notify acquires sess.mu/d.mu and must never run while any resizeMu is held.
		211	+	func TestTransactionalResizeApplyReportsFailureOncePerCall(t *testing.T) {
		212	+	countResizeFailed := func(d *Daemon) (int, domain.Notification) {
		213	+	var last domain.Notification
		214	+	count := 0
		215	+	for _, h := range d.notices.history() {
		216	+	if h.Code == domain.NoticeResizeFailed {
		217	+	count++
		218	+	last = h
		219	+	}
		220	+	}
		221	+	return count, last
		222	+	}
		223	+
		224	+	t.Run("one failed member yields one notice after return", func(t *testing.T) {
		225	+	failing := &transactionalResizePTY{errs: []error{errors.New("scripted resize failure")}}
		226	+	d, sess, _, _ := newManualSessionWithPTYs(t, failing)
		227	+	tb := sess.activeTab()
		228	+	p := tb.focusedPane()
		229	+	plan := resizePlan{members: []resizeMember{{session: sess, tab: tb, pane: p, rect: domain.Rect{Width: 100, Height: 20}, retry: true}}}
		230	+
		231	+	require.Empty(t, d.notices.history(), "no notice is recorded before applyResize")
		232	+	require.True(t, d.applyResize(&plan))
		233	+
		234	+	count, n := countResizeFailed(d)
		235	+	require.Equal(t, 1, count, "exactly one NoticeResizeFailed per applyResize call")
		236	+	require.Equal(t, domain.NoticeWarn, n.Severity)
		237	+	require.Equal(t, "pane resize failed; retrying in background", n.Message)
		238	+	require.False(t, plan.members[0].ok, "failed member is not marked ok")
		239	+	})
		240	+
		241	+	t.Run("two failed members still yield one notice per call", func(t *testing.T) {
		242	+	first := &transactionalResizePTY{errs: []error{errors.New("first fails")}}
		243	+	second := &transactionalResizePTY{errs: []error{errors.New("second fails")}}
		244	+	d, sess, _, _ := newManualSessionWithPTYs(t, first, second)
		245	+	tabs := sess.tabs
		246	+	plan := resizePlan{members: []resizeMember{
		247	+	{session: sess, tab: tabs[0], pane: tabs[0].focusedPane(), rect: domain.Rect{Width: 100, Height: 20}, retry: true},
		248	+	{session: sess, tab: tabs[1], pane: tabs[1].focusedPane(), rect: domain.Rect{Width: 100, Height: 20}, retry: true},
		249	+	}}
		250	+
		251	+	require.True(t, d.applyResize(&plan))
		252	+
		253	+	count, _ := countResizeFailed(d)
		254	+	require.Equal(t, 1, count, "one notice per applyResize call, not per failed member")
		255	+	})
		256	+	}
		257	+
207	// S3 acceptance: a failed member never publishes speculative parser/screen or	258		// S3 acceptance: a failed member never publishes speculative parser/screen or
208	// rectangle state. Successful members and the client layout commit together;	259		// rectangle state. Successful members and the client layout commit together;
209	// composition clips/pads the retained failed screen against that new layout.	260		// composition clips/pads the retained failed screen against that new layout.