Draw a peer message a block at a time, and parse it off the drawing thread

A message from another agent was the one markdown in the app still rendered
whole: one parse and one display list for the entire thing. Every settled
reply has been cut into blocks since the transcript was made lazy, and
`warm` has been making those parses ahead on a background thread -- but it
filtered for assistant replies alone, so the longest message a transcript
holds was also the only one parsed on the thread that draws.

Measured on the emulator against a 43KB peer message, opening it: 177ms in
`markdown parsed while composing`, against none afterwards and 156 blocks
already ready. What is left is the card being a single list item, so all 156
blocks are still measured, placed and recorded at once -- 118ms of placement
in that same frame.

The `when` in `warm` is now the rule rather than a filter: every row that
draws markdown belongs in it.

Blocks are spaced by the transcript's own BLOCK_SPACING rather than the
renderer's internal padding, which moves a heading about 6px (2.3dp) closer
to the paragraph above it. The message's total height is unchanged, and it
now matches every reply in the transcript.

While here: FrameStats was remembered per session screen and DebugStats is a
global emptied only by the copy button, so the two halves of a render report
covered different stretches of time -- and `drawAccounting` divides one by
the other. A report copied after visiting two sessions claimed 36.8 seconds
of placement inside a 13.5 second window, and clamped "everything else" to
0.00ms (0%), which reads as a screen whose entire cost is this app's code.
One FrameStats for the app, so both halves mean "since this was last copied".

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

app/androidApp/src/main/kotlin/com/example/aiapp/FrameStats.kt

@@ -10,7 +10,6 @@ import android.view.FrameMetrics
 import android.view.Window
 import androidx.compose.runtime.Composable
 import androidx.compose.runtime.DisposableEffect
-import androidx.compose.runtime.remember
 import androidx.compose.ui.platform.LocalContext
 
 /**
@@ -26,8 +25,15 @@ import androidx.compose.ui.platform.LocalContext
  * The phases are the platform's own: [FrameMetrics] reports each frame's cost in nanoseconds,
  * broken down into the parts the UI thread is responsible for -- handling input, running
  * animations, measuring and laying out, recording the draw -- and the parts after it.
+ *
+ * One of these for the app, like [DebugStats], because the two are read as one report and
+ * [drawAccounting] divides one by the other. Held per screen it was emptied by leaving a session
+ * and the counters were not, so a report copied after visiting two sessions divided every session's
+ * work by the newest one's frame count -- and printed the result as a per-frame measurement. It
+ * said 36.8 seconds of placement inside a 13.5 second window, and left "everything else" clamped at
+ * 0.00ms (0%), which reads as a screen whose whole cost is this app's own code.
  */
-class FrameStats {
+object FrameStats {
     private val total = ArrayList<Long>()
     private val waited = ArrayList<Long>()
     private val input = ArrayList<Long>()
@@ -108,28 +114,28 @@ class FrameStats {
     }
 
     private fun percent(part: Int, whole: Int) = "%.1f%%".format(100.0 * part / whole)
-
-    private companion object {
-        /** Enough for a couple of minutes of scrolling; this is a diagnostic, not a log. */
-        const val CAP = 20_000
-    }
 }
 
+/** Enough for a couple of minutes of scrolling; this is a diagnostic, not a log. */
+private const val CAP = 20_000
+
 /**
- * Frame timings for as long as this screen is on it.
+ * Records into [FrameStats] for as long as this screen is on it.
+ *
+ * The listener is what comes and goes; what it writes into does not, so a report covers the same
+ * stretch of time as the counters beside it. See [FrameStats].
  *
  * The listener is handed its own thread because the platform calls it for every frame and the
  * documentation is explicit that doing that on the main thread taxes the very thing being measured.
  */
 @Composable
-fun rememberFrameStats(): FrameStats {
+fun recordFrames() {
-    val stats = remember { FrameStats() }
     val window = LocalContext.current.activity()?.window
     DisposableEffect(window) {
         if (window == null) return@DisposableEffect onDispose {}
         val thread = HandlerThread("frame-stats").apply { start() }
         val listener = Window.OnFrameMetricsAvailableListener { _, metrics, _ ->
-            stats.add(metrics)
+            FrameStats.add(metrics)
         }
         window.addOnFrameMetricsAvailableListener(listener, Handler(thread.looper))
         onDispose {
@@ -137,7 +143,6 @@ fun rememberFrameStats(): FrameStats {
             thread.quitSafely()
         }
     }
-    return stats
 }
 
 /** The activity behind a composable's context, which is what owns the window. */

app/androidApp/src/main/kotlin/com/example/aiapp/PeerMessage.kt

@@ -27,6 +27,10 @@ import androidx.compose.ui.unit.dp
  * Drawn as its own kind rather than as the reader's own bubble. They did not say this, and a
  * transcript that puts it in their voice is making a claim about who asked for the work that
  * follows -- which is exactly the question a peer message is usually the answer to.
+ *
+ * Opened, it is drawn a block at a time ([BlockedMarkdown]) for the reason every reply already is:
+ * these are the longest messages a transcript holds, and one of them as a single render is one
+ * parse and one display list proportional to the whole of it. See [markdownBlocks].
  */
 @Composable
 fun PeerMessageRow(
@@ -52,7 +56,7 @@ fun PeerMessageRow(
                     )
                 }
             }
-            if (expanded) MarkdownText(item.text, replies, Modifier.padding(top = 6.dp))
+            if (expanded) BlockedMarkdown(item.text, replies, Modifier.padding(top = 6.dp))
         }
     }
 }

app/androidApp/src/main/kotlin/com/example/aiapp/SessionScreen.kt

@@ -1049,7 +1049,7 @@ fun SessionScreen(settings: ServerSettings, summary: SessionSummary, onBack: ()
     // One poll for this machine's limits, read by the two things that show them: the bar under
     // the header, and the colour of the button that opens the dialog.
     val usage = rememberSessionUsage(settings, summary.setup)
-    val frames = rememberFrameStats()
+    recordFrames()
     var usageOpen by remember { mutableStateOf(false) }
     var settingsOpen by remember { mutableStateOf(false) }
 
@@ -1150,9 +1150,9 @@ fun SessionScreen(settings: ServerSettings, summary: SessionSummary, onBack: ()
                                             "  ${expandedTools.size} tool calls and" +
                                                 " ${expandedGroups.size} groups open",
                                         ),
-                                    frames = frames.lines(context.refreshHz()),
+                                    frames = FrameStats.lines(context.refreshHz()),
                                     accounting =
-                                        frames.drawPhase().let { (nanos, count) ->
+                                        FrameStats.drawPhase().let { (nanos, count) ->
                                             drawAccounting(nanos, count)
                                         },
                                     crash = lastCrash(context),
@@ -1170,7 +1170,7 @@ fun SessionScreen(settings: ServerSettings, summary: SessionSummary, onBack: ()
                             // Emptied by the copy, so pressing it twice measures two separate
                             // stretches
                             // of scrolling rather than one and then the same one again.
-                            frames.reset()
+                            FrameStats.reset()
                             DebugStats.reset()
                             Toast.makeText(context, "Copied render report", Toast.LENGTH_SHORT)
                                 .show()

app/androidApp/src/main/kotlin/com/example/aiapp/TranscriptItems.kt

@@ -466,32 +466,46 @@ private fun updateTool(
 private val parsingThreads = Dispatchers.Default.limitedParallelism(2)
 
 /**
- * Parses the replies among [rows], off whatever thread is drawing.
+ * Parses the markdown among [rows], off whatever thread is drawing.
  *
  * Called where a page of transcript is folded rather than where a row is composed, which is the
  * whole point: the work happens seconds before the reader reaches the rows it was done for. See
  * [ParsedReplies].
  *
  * What is warmed mirrors what the rows draw, unit by unit -- prose split into its blocks, a memory
- * note whole -- because a string warmed under a key no row ever looks up is a miss that nothing
+ * note whole, a peer message split the same way prose is -- because a string warmed under a key no
- * reports; see [transcriptUnits], which is the flatten this has to agree with. It reads the same
+ * row ever looks up is a miss that nothing reports; see [transcriptUnits], which is the flatten
- * [ParsedReplies.partsOf] and [ParsedReplies.blocksOf] caches the flatten does, so a message is
+ * this has to agree with. It reads the same [ParsedReplies.partsOf] and [ParsedReplies.blocksOf]
- * scanned once however many pages hand it back through here, while the whole loaded transcript
+ * caches the flatten does, so a message is scanned once however many pages hand it back through
- * crosses this on every page.
+ * here, while the whole loaded transcript crosses this on every page.
+ *
+ * Every kind of row that draws markdown belongs in the `when` below. That is the rule the peer
+ * message was missing: this used to filter for assistant replies alone, so the one row type nobody
+ * had thought about paid its whole parse in the frame it appeared in, with no counter saying which
+ * row it was.
  */
 suspend fun warm(replies: ParsedReplies, rows: List<TranscriptItem>) {
     withContext(parsingThreads) {
-        val texts =
+        val texts = rows.flatMap { row ->
-            rows
+            when (row) {
-                .filterIsInstance<TranscriptItem.AssistantMsg>()
+                is TranscriptItem.AssistantMsg ->
-                .flatMap { replies.partsOf(it.text) }
+                    replies.partsOf(row.text).flatMap { part ->
-                .flatMap { part ->
+                        when (part) {
-                    when (part) {
+                            is MessagePart.Prose -> replies.blocksOf(part.text)
-                        is MessagePart.Prose -> replies.blocksOf(part.text)
+                            // Drawn as one MarkdownText, so its whole text is the key
-                        // Drawn as one MarkdownText, so its whole text is the key looked up.
+                            // looked up.
-                        is MessagePart.Remembered -> listOf(part.text)
+                            is MessagePart.Remembered -> listOf(part.text)
+                        }
                     }
-                }
+                // A message from another agent is markdown too, and it is the longest thing
+                // in a transcript often enough that leaving it out was the whole of why one
+                // cost a fifth of a second to open: it was the only markdown in the app
+                // parsed on the thread that draws. Its blocks, not its text, because
+                // [PeerMessageRow] draws it a block at a time.
+                is TranscriptItem.PeerNote -> replies.blocksOf(row.text)
+                else -> emptyList()
+            }
+        }
         if (texts.isNotEmpty()) replies.warm(texts)
     }
 }