watchpr: detect merge/close in poll loop (was hanging after baseline)

watchpr only reported meaningful changes as transitions from the poll
baseline. A PR already merged or closed when the watch started was
captured as the baseline and never produced a transition, so the loop
polled the dead PR forever (process alive, never exiting) -- the
single-PR --interval case observed in production.

Add a terminal-state check applied to the baseline snapshot: a PR that
is already merged or closed the moment watchpr starts is reported and
exits immediately, since it can never change again. Extract the
baseline+poll loop into agent.Watch behind a StateFetcher interface so
the loop, its open->merged/close detection, and its poll-error
resilience are unit-testable with a fake client.

internal/agent/watch_test.go

@@ -1,6 +1,10 @@
 package agent
 
-import "testing"
+import (
+	"errors"
+	"testing"
+	"time"
+)
 
 func base() PRState {
 	return PRState{
@@ -124,6 +128,151 @@ func TestMeaningfulChangeStaysFailed(t *testing.T) {
 	}
 }
 
+// fakeFetcher returns a scripted sequence of (state, error) results per call,
+// so tests can drive Watch across baseline and successive polls.
+type fakeFetcher struct {
+	states []PRState
+	errs   []error
+	calls  int
+}
+
+func (f *fakeFetcher) FetchState(ref PRRef, agentLogin string) (PRState, error) {
+	i := f.calls
+	if i >= len(f.states) {
+		i = len(f.states) - 1
+	}
+	f.calls++
+	var err error
+	if f.calls-1 < len(f.errs) {
+		err = f.errs[f.calls-1]
+	}
+	return f.states[i], err
+}
+
+func TestTerminalState(t *testing.T) {
+	open := base()
+	open.State = "open"
+	if term, _ := terminalState(open); term {
+		t.Errorf("open PR should not be terminal")
+	}
+
+	merged := base()
+	merged.State = "closed"
+	merged.Merged = true
+	if term, reason := terminalState(merged); !term || reason != "PR merged" {
+		t.Errorf("merged PR: got (%v, %q), want (true, %q)", term, reason, "PR merged")
+	}
+
+	closed := base()
+	closed.State = "closed"
+	if term, reason := terminalState(closed); !term || reason != "PR closed without merging" {
+		t.Errorf("closed PR: got (%v, %q), want (true, %q)", term, reason, "PR closed without merging")
+	}
+}
+
+// The production hang: a PR that is already merged when watchpr starts must be
+// reported at baseline and exit, without ever consuming a tick. Before the fix,
+// Watch only reported transitions, so a terminal baseline was polled forever.
+func TestWatchExitsWhenAlreadyMergedAtBaseline(t *testing.T) {
+	merged := base()
+	merged.State = "closed"
+	merged.Merged = true
+	f := &fakeFetcher{states: []PRState{merged}}
+
+	ticks := make(chan time.Time) // never fires; a hang would block here
+	res, err := Watch(f, []PRRef{merged.Ref}, "unkin-agent", ticks, nil, nil)
+	if err != nil {
+		t.Fatalf("Watch: %v", err)
+	}
+	if res.Reason != "PR merged" {
+		t.Errorf("reason = %q, want %q", res.Reason, "PR merged")
+	}
+	if f.calls != 1 {
+		t.Errorf("fetch calls = %d, want 1 (baseline only)", f.calls)
+	}
+}
+
+// A PR already closed-without-merge at baseline must also exit immediately.
+func TestWatchExitsWhenAlreadyClosedAtBaseline(t *testing.T) {
+	closed := base()
+	closed.State = "closed"
+	f := &fakeFetcher{states: []PRState{closed}}
+
+	ticks := make(chan time.Time)
+	res, err := Watch(f, []PRRef{closed.Ref}, "unkin-agent", ticks, nil, nil)
+	if err != nil {
+		t.Fatalf("Watch: %v", err)
+	}
+	if res.Reason != "PR closed without merging" {
+		t.Errorf("reason = %q, want %q", res.Reason, "PR closed without merging")
+	}
+}
+
+// An open→merged transition observed during polling must be detected and end
+// the watch.
+func TestWatchDetectsMergeAfterBaseline(t *testing.T) {
+	open := base()
+	merged := base()
+	merged.State = "closed"
+	merged.Merged = true
+	f := &fakeFetcher{states: []PRState{open, merged}} // baseline open, then merged
+
+	baselineFired := false
+	ticks := make(chan time.Time, 1)
+	ticks <- time.Now()
+	res, err := Watch(f, []PRRef{open.Ref}, "unkin-agent",
+		ticks, func() { baselineFired = true }, nil)
+	if err != nil {
+		t.Fatalf("Watch: %v", err)
+	}
+	if !baselineFired {
+		t.Errorf("onBaseline should fire for an open baseline")
+	}
+	if res.Reason != "PR merged" {
+		t.Errorf("reason = %q, want %q", res.Reason, "PR merged")
+	}
+}
+
+// A transient poll error must be reported and the loop must keep polling; a
+// merge on the following tick still ends the watch.
+func TestWatchContinuesPastPollError(t *testing.T) {
+	open := base()
+	merged := base()
+	merged.State = "closed"
+	merged.Merged = true
+	// baseline ok, first poll errors, second poll sees the merge.
+	f := &fakeFetcher{
+		states: []PRState{open, open, merged},
+		errs:   []error{nil, errors.New("HTTP 502"), nil},
+	}
+
+	var gotErr error
+	ticks := make(chan time.Time, 2)
+	ticks <- time.Now()
+	ticks <- time.Now()
+	res, err := Watch(f, []PRRef{open.Ref}, "unkin-agent", ticks, nil,
+		func(_ PRRef, e error) { gotErr = e })
+	if err != nil {
+		t.Fatalf("Watch: %v", err)
+	}
+	if gotErr == nil {
+		t.Errorf("onError should have received the transient poll error")
+	}
+	if res.Reason != "PR merged" {
+		t.Errorf("reason = %q, want %q (loop must survive the error)", res.Reason, "PR merged")
+	}
+}
+
+// A baseline fetch error aborts the watch (nothing to establish a baseline
+// from), unlike a mid-loop poll error.
+func TestWatchBaselineErrorAborts(t *testing.T) {
+	f := &fakeFetcher{states: []PRState{base()}, errs: []error{errors.New("HTTP 500")}}
+	ticks := make(chan time.Time)
+	if _, err := Watch(f, []PRRef{base().Ref}, "unkin-agent", ticks, nil, nil); err == nil {
+		t.Fatal("Watch should return the baseline fetch error")
+	}
+}
+
 func TestCountNonAgentComments(t *testing.T) {
 	comments := []Comment{
 		{User: User{Login: "unkin-agent"}},