Incremental text: parley cannot, the app already does it, and the 9.5ms is the fixture

Iris asked to look into incremental text rendering, hoping parley
supported it. It does not, by design: a `Layout` re-linebreaks and
re-aligns freely but "if the text content or the styles applied to that
content change then a new `Layout` must be created", its LRU cache holds
harfrust's per-font shaper data rather than shaped runs, and its own
`PlainEditor` rebuilds the whole layout from the whole buffer on every
keystroke.

The app already does what incremental layout would buy: `RowBlocks::
apply_delta` keeps one `TextEdit` per markdown block and re-shapes only
the one a delta landed in. Re-splitting the markdown to find it is 18µs
at 18,000 characters; comparing the blocks is 470ns.

What is left is one `TextBuffer::shape` of that block, linear in its
length at ~0.23ms per 1,000 characters here -- and the bench fixture's
streamed message is 14,888 characters in a *single* block, a run-on
paragraph with no blank line in it, so every delta reshapes all of it.
That is 3.5ms of the measured 3.86ms frame.

Real replies are not that: across 7,706 top-level blocks from 3,675 real
assistant messages on this machine (lengths only, no content copied
anywhere), p50 147 characters, p90 449, p99 836, largest 1,580, nothing
above 4,000; code fences p50 126, largest 589. At those sizes a reshape
is 48µs to 372µs here, roughly 0.12-0.93ms on the phone -- inside a
120Hz budget with no incremental anything.

So the recommendation is not to build it, and to give the fixture's
streamed message the paragraph structure a real reply has instead. Three
runs added to `frame_profile.rs` so none of this is re-derived: what
reshaping a growing message costs (including at the sizes real replies
reach), where a delta's cost is, and what the fixture actually streams.

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

AGENTS.md

@@ -405,12 +405,19 @@ Each exists because something was invisible without it.
   transcript -- every later pass over the same rows is p99 0.26ms. A
   **streamed event** is, and it is not where it looks: folding the event
   is 0.35ms and applying the diff to the widget tree is 0.41ms, while the
-  *frame* is 3.86ms here and 9.5ms on Iris's phone. 401 streamed events
+  *frame* is 3.86ms here and 9.5ms on Iris's phone. (The fold was the
-  move the item count from 652 to 654, so almost every one is a delta
-  into the same row -- the cost is re-laying out and re-shaping one
-  growing message on every delta, not the fold. (The fold was the
   hypothesis, from `foldEvent`'s Compose lesson under "Things that have
-  bitten"; measuring it is what ruled it out.)
+  bitten"; measuring it is what ruled it out.) That frame is one
+  `TextBuffer::shape` of the block a delta landed in, and **the fixture's
+  is 14,888 characters in a single block** -- against a largest-ever
+  1,580 across 7,706 blocks of real replies. So the stream phase's number
+  is a property of the fixture, not of streaming; docs/RUST.md's
+  "Incremental text" has the measurements and why parley cannot help.
+
+  The last two runs (`where_a_streamed_deltas_cost_is`,
+  `what_the_fixture_streams`) exist to keep that answerable: what a delta
+  costs to re-split and re-compare, and what the fixture actually
+  streams.
 - **The emulator is a GLES rig, deliberately** (Iris, 2026-09-08;
   docs/RUST.md). Its guest has no hardware Vulkan -- only SwiftShader
   in software -- while its GLES *is* the host's real GPU through virgl at

app-rust/tests/frame_profile.rs

@@ -188,3 +188,162 @@ fn what_a_streamed_event_costs() {
     summarise("apply", &apply);
     summarise("frame", &frame);
 }
+
+/// What re-shaping a *growing* message costs, isolated from everything
+/// else a frame does -- the measurement that decides whether an
+/// incremental-text design would pay for itself (Iris, 2026-09-09:
+/// "we should definitely investigate incremental text rendering").
+///
+/// Grows one text buffer a delta at a time, the way a streamed reply
+/// grows one row, and reports what `TextBuffer::shape` costs at each
+/// length. Linear per-delta cost means the total over a reply is
+/// quadratic in its length, which is the thing an incremental shaper
+/// would remove.
+#[test]
+#[ignore]
+fn what_reshaping_a_growing_message_costs() {
+    use iris::prelude::*;
+
+    let mut h = Harness::new(phone_size(), PHONE_SCALE);
+    // A reply-sized paragraph built a delta at a time. The deltas are
+    // words rather than characters because that is what a model streams.
+    const DELTA: &str = "the quick brown fox jumps over the lazy dog ";
+    let attrs = TextAttrs::default();
+    let width = Some(phone_size().x);
+
+    let mut buffer = iris::prelude::TextBuffer::new_empty();
+    let mut text = String::new();
+    let mut per_delta = Vec::new();
+    let mut total = Duration::ZERO;
+    for n in 1..=400 {
+        text.push_str(DELTA);
+        buffer.set_text(text.clone());
+        let at = Instant::now();
+        buffer.shape(&mut h.rsc.ui.text, &attrs, width, PHONE_SCALE);
+        let took = at.elapsed();
+        per_delta.push(took);
+        total += took;
+        if n % 100 == 0 {
+            println!(
+                "  after {n:>3} deltas ({:>6} chars): this reshape {:?}",
+                text.len(),
+                took
+            );
+        }
+    }
+    println!(
+        "  400 deltas: {:?} of shaping in total, {:?} per delta at the end",
+        total,
+        per_delta.last().unwrap()
+    );
+    summarise("reshape", &per_delta);
+
+    // The same measurement at the sizes real replies actually reach.
+    // Measured 2026-09-09 over 7,706 top-level blocks from 3,675 real
+    // assistant messages on this machine: p50 147 chars, p90 449, p99
+    // 836, largest 1,580, and *nothing* above 4,000. The bench fixture's
+    // streamed message is one 14,888-character block, which is 9x the
+    // largest real one -- so the sizes below are what a live reshape
+    // actually costs and the run above is what the benchmark measures.
+    println!("  at the sizes real replies reach:");
+    for chars in [147usize, 449, 836, 1580] {
+        let mut sample = String::new();
+        while sample.len() < chars {
+            sample.push_str(DELTA);
+        }
+        sample.truncate(chars);
+        let mut buffer = iris::prelude::TextBuffer::new(sample);
+        let at = Instant::now();
+        buffer.shape(&mut h.rsc.ui.text, &attrs, width, PHONE_SCALE);
+        println!("    {chars:>5} chars: {:?}", at.elapsed());
+    }
+}
+
+/// Where a streamed delta's cost actually is, given that `RowBlocks::
+/// apply_delta` already re-shapes only the block the delta landed in.
+/// Three candidates, all of which scale with the *whole* message rather
+/// than the delta: re-parsing the markdown to find the blocks, comparing
+/// them against the ones already drawn, and re-shaping the last block.
+#[test]
+#[ignore]
+fn where_a_streamed_deltas_cost_is() {
+    use ai_app::client::markdown_blocks::{common_prefix, split_blocks};
+
+    // A reply with real block structure -- paragraphs separated by blank
+    // lines, the way a model writes -- so the last block is one paragraph
+    // rather than the whole message.
+    const SENTENCE: &str = "The quick brown fox jumps over the lazy dog. ";
+    let mut src = String::new();
+    let mut blocks = Vec::new();
+
+    let mut split = Vec::new();
+    let mut compare = Vec::new();
+    for n in 1..=400 {
+        src.push_str(SENTENCE);
+        // A paragraph break every eight deltas, so the trailing block
+        // stays a normal size and only the message grows.
+        if n % 8 == 0 {
+            src.push_str("\n\n");
+        }
+
+        let at = Instant::now();
+        let new_blocks = split_blocks(&src);
+        split.push(at.elapsed());
+
+        let at = Instant::now();
+        let _ = common_prefix(&blocks, &new_blocks);
+        compare.push(at.elapsed());
+
+        blocks = new_blocks;
+        if n % 100 == 0 {
+            println!(
+                "  after {n:>3} deltas ({:>6} chars, {} blocks): split {:?} compare {:?}",
+                src.len(),
+                blocks.len(),
+                split[n - 1],
+                compare[n - 1],
+            );
+        }
+    }
+    summarise("split_blocks", &split);
+    summarise("common_prefix", &compare);
+    let total: Duration = split.iter().chain(compare.iter()).sum();
+    println!("  400 deltas: {total:?} in block-splitting and comparison alone");
+}
+
+/// What the bench fixture's streamed tail actually is, since the cost of
+/// a delta depends entirely on how big the block it lands in gets.
+#[test]
+#[ignore]
+fn what_the_fixture_streams() {
+    use ai_app::client::markdown_blocks::split_blocks;
+
+    let mut h = Harness::new(phone_size(), PHONE_SCALE);
+    let opened = ai_app::ui::fixture::open(&mut h.rsc, &mut h.state).expect("the fixture folds");
+    let mut items = opened.items;
+    let before = items.len();
+    for event in &opened.stream_tail {
+        items = ai_app::client::transcript_fold::fold_event(&items, event);
+    }
+    println!("{} items -> {}", before, items.len());
+    // The last few items are where the stream landed. Only the message
+    // variants matter -- those are what a delta appends to.
+    use ai_app::client::transcript_fold::TranscriptItem;
+    for item in items.iter().rev().take(4) {
+        let (kind, text) = match item {
+            TranscriptItem::AssistantMsg { text, .. } => ("AssistantMsg", text.clone()),
+            TranscriptItem::UserMsg { text, .. } => ("UserMsg", text.clone()),
+            TranscriptItem::ToolRun { input, output, .. } => {
+                ("ToolRun", format!("{input}{output}"))
+            }
+            _ => ("other", String::new()),
+        };
+        let blocks = split_blocks(&text);
+        println!(
+            "  tail {kind}: {} chars in {} blocks; longest block {} chars",
+            text.len(),
+            blocks.len(),
+            blocks.iter().map(|b| b.source.len()).max().unwrap_or(0),
+        );
+    }
+}

docs/RUST.md

@@ -528,10 +528,54 @@ every delta.
 
 `fold_event`'s `items.to_vec()` per event was the hypothesis -- it is the
 exact shape of the Compose lesson in AGENTS.md's "Things that have
-bitten" -- and measuring it is what ruled it out. **Not yet designed**:
+bitten" -- and measuring it is what ruled it out.
-making a row's text append incrementally rather than reshape touches how
+
-`TranscriptRow` holds its shaped text, which is load-bearing enough to
+### Incremental text: parley cannot, and it turns out not to matter (2026-09-09)
-raise before building.
+
+Iris asked to investigate incremental text rendering and hoped parley
+supported it. **It does not, by design.** The crate's own docs: a
+`Layout` "supports re-linebreaking and re-aligning many times... but if
+the text content or the styles applied to that content change then a new
+`Layout` must be created". Its `LruCache` caches harfrust's per-font
+shaper data, instance and plan -- not shaped runs -- and its own
+`PlainEditor::update_layout` rebuilds the whole layout from the whole
+buffer on **every keystroke**. So there is nothing to adopt, and adding
+it would be upstream work in parley.
+
+**And the app already does the thing incremental layout would buy.**
+`RowBlocks::apply_delta` keeps one `TextEdit` per top-level markdown
+block and re-shapes only the block a delta landed in; re-splitting the
+markdown to find that block is 18µs at 18,000 characters and comparing
+the blocks is 470ns. Neither is the cost.
+
+**The 9.5ms is a bench-fixture artifact.** Measured with
+`frame_profile.rs`:
+
+- Re-shaping a block is linear in its length -- ~0.23ms per 1,000
+  characters on this desktop, so a message grown to 17,600 characters
+  costs 4.1ms on its *last* delta and 842ms of shaping over the whole
+  reply.
+- The fixture's streamed message is **14,888 characters in one block** --
+  a synthetic run-on paragraph with no blank line in it, so every delta
+  reshapes all of it. That is the whole of the frame: 3.5ms of the
+  measured 3.86ms.
+- Real replies are not like that. Over **7,706 top-level blocks from
+  3,675 real assistant messages** on this machine (block lengths only;
+  no content left the machine): p50 147 characters, p90 449, p99 836,
+  largest 1,580, and **nothing above 4,000**. Code fences are smaller
+  still -- 170 of them, p50 126, largest 589.
+- At those sizes a live reshape is 48µs (p50), 208µs (p99) and 372µs
+  (the largest block ever seen), or roughly 0.12-0.93ms on the phone.
+  Comfortably inside a 120Hz budget, with no incremental anything.
+
+**So: do not build incremental text.** What is worth doing instead is
+giving the fixture's streamed message the paragraph structure a real
+reply has, so the stream phase measures something that happens. Both
+apps read the same fixture, so the Compose/iris comparison stays sound,
+but numbers from before and after the change are not comparable to each
+other. Whether to keep a pathological block as a *labelled* stress case
+alongside it is Iris's call -- the danger of the current one is only that
+its number reads as "streaming costs 9.5ms" when nothing does.
 
 ### The Android release profile is `opt-level = 3`, not `"s"` (2026-09-09)