Graphical terminal improvements.

It makes using programs who like to refresh a ton much more tollerable
when I'm not in Emacs; I suspect others will like them too.

internal/app/terminal_events.go

@@ -5,15 +5,73 @@
 
 package app
 
-import "git.stormux.org/storm/cthulhu/internal/atspi"
+import (
+	"strings"
+	"unicode/utf8"
+
+	"git.stormux.org/storm/cthulhu/internal/atspi"
+	"git.stormux.org/storm/cthulhu/internal/input"
+)
 
 func should_ignore_terminal_scrollback(event atspi.Event) bool {
-	// VTE can remove old complete lines at the start of its accessible text
-	// while appending output at a different offset. These are not edits to
-	// the prompt. This event pattern occurs on both X11 and Wayland.
+	// VTE can remove old lines at the start of its accessible text while
+	// appending output at a different offset. A deletion entirely before
+	// the caret line is not an edit to the prompt. VTE wrapped rows have
+	// no trailing newline, so do not require one. This event pattern
+	// occurs on both X11 and Wayland.
 	return event.Type == atspi.TextDeletedEvent &&
 		event.Source.Role == "terminal" && event.Detail1 == 0 &&
 		event.Source.TextCaretKnown && event.Source.BrailleLineStart > 0 &&
-		event.TextChangeEndsLine && event.TextChangeLength > 0 &&
+		event.TextChangeLength > 0 &&
 		event.TextChangeLength <= event.Source.BrailleLineStart
 }
+
+// is_user_terminal_deletion identifies a deletion the user just caused at
+// the caret on its own line: a Backspace or Delete edit. The deleted text
+// is the user's confirmation that the edit took, so repaint coalescing
+// must not swallow it. Program repaints start above the caret line, and
+// deletions without a preceding delete key are program output churn.
+func is_user_terminal_deletion(event atspi.Event, intent input.Intent) bool {
+	return intent == input.DeleteIntent &&
+		event.Type == atspi.TextDeletedEvent && event.Source.Role == "terminal" &&
+		event.Source.TextCaretKnown &&
+		event.Detail1 >= event.Source.BrailleLineStart &&
+		event.Detail1 <= event.Source.BrailleCaret
+}
+
+// is_latest_wins_terminal_repaint identifies an in-place rewrite: an
+// insertion that replaced a just-deleted range. Status regions update
+// this way several times a second and only the newest state is worth
+// hearing. Editing feedback on the caret line after a typing or delete
+// key stays immediate; appended output has no preceding deletion and
+// never qualifies.
+func is_latest_wins_terminal_repaint(event atspi.Event, intent input.Intent, replacedDeletion bool) bool {
+	if !replacedDeletion || event.Type != atspi.TextInsertedEvent || event.Source.Role != "terminal" {
+		return false
+	}
+	if (intent == input.TypingIntent || intent == input.DeleteIntent) &&
+		event.Source.TextCaretKnown && event.Detail1 >= event.Source.BrailleLineStart {
+		return false
+	}
+	return true
+}
+
+// is_trivial_terminal_repaint identifies a held repaint whose payload is
+// only the few cells that changed in a ticking status (timer digits such
+// as "40"). Uttered alone it has no context; a fuller repaint of the same
+// region replaces it before the flush if one follows.
+func is_trivial_terminal_repaint(event atspi.Event) bool {
+	return event.Type == atspi.TextInsertedEvent && event.Source.Role == "terminal" &&
+		utf8.RuneCountInString(strings.TrimSpace(event.Payload)) <= 2
+}
+
+// terminalReplacementOverlap reports whether an insertion rewrites any of
+// the range a pending deletion removed. Offsets are not compared exactly:
+// a scroll shift at the start of the buffer between the deletion and its
+// replacement insertion moves the insertion's offset, and a repainted
+// region is routinely rewritten at a neighboring offset.
+func terminalReplacementOverlap(deletion, insertion atspi.Event) bool {
+	deletionEnd := deletion.Detail1 + max(deletion.TextChangeLength, 1)
+	insertionEnd := insertion.Detail1 + max(insertion.TextChangeLength, 1)
+	return insertion.Detail1 < deletionEnd && deletion.Detail1 < insertionEnd
+}