perf(terminal): make screen clone match Zig, batch OSC payload copies

screen-clone was 2.6x slower than Zig. The main cost was in
PageList::clone_range: it created a throwaway page, recycled it into the
pool and Page::reinit then memset the entire page again, faulting in
every cell. Untouched pages now go to a clean_pool that create_node
reuses without re-zeroing. Page::init allocates cells with
bytemuck::allocation::try_zeroed_vec (Cell derives Zeroable), getting
lazily-zeroed memory from the allocator without unsafe code, and
clone_partial_row_from bulk-copies plain-cell runs in rows with managed
memory. screen-clone clone: 78 ms -> ~28 ms (Zig ~27.8 ms).

osc::Parser::next_slice appends payload bytes in bulk while behaving
exactly like per-byte next() (2048 limit, overflow), verified by a
property test; the osc-parser bench uses it (54.5 -> ~32 ms, Zig ~28).

Still slower: screen-clone render (~2.4x), dominated by per-row style
vectors in RenderState::update.

Test Plan:
- cargo test --workspace (no failures)
- ghostty-bench +screen-clone / +osc-parser with --profile production

Cargo.lock

@@ -1399,6 +1399,7 @@ dependencies = [
 name = "ghostty-terminal"
 version = "0.1.0"
 dependencies = [
+ "bytemuck",
  "crossbeam-channel",
  "flate2",
  "ghostty-config",

crates/ghostty-bench/src/benches.rs

@@ -231,9 +231,7 @@ impl Benchmark for OscParser {
                 Ok(_) => return Ok(()),
                 Err(_) => return Err(BenchmarkFailed),
             }
-            for &c in &osc_buf[..len] {
+            self.parser.next_slice(&osc_buf[..len]);
-                self.parser.next(c);
-            }
             black_box(self.parser.end(Some(0x07)));
             self.parser.reset();
         }

crates/ghostty-terminal/Cargo.toml

@@ -5,6 +5,7 @@ edition.workspace = true
 license.workspace = true
 
 [dependencies]
+bytemuck = { version = "1.25.2", features = ["derive", "extern_crate_alloc"] }
 crossbeam-channel = "0.5.17"
 flate2 = "1.1.10"
 ghostty-datastruct.workspace = true

crates/ghostty-terminal/src/page.rs

@@ -76,7 +76,7 @@ pub enum SemanticPrompt {
     PromptContinuation = 2,
 }
 /// Eight-byte terminal cell; all-zero bits describe an empty cell.
-#[derive(Debug, Default, Clone, Copy, PartialEq, Eq)]
+#[derive(Debug, Default, Clone, Copy, PartialEq, Eq, bytemuck::Zeroable)]
 #[repr(transparent)]
 pub struct Cell(u64);
 impl Cell {
@@ -744,12 +744,10 @@ impl Page {
             row.set_cells(usize::from(y) * usize::from(cap.cols));
             rows.push(row);
         }
-        let mut cells = Vec::new();
         let count = usize::from(cap.cols) * usize::from(cap.rows);
-        cells
+        // Zeroed allocation: large buffers come lazily zeroed from the OS.
-            .try_reserve_exact(count)
+        let cells: Vec<Cell> =
-            .map_err(|_| Error::OutOfMemory)?;
+            bytemuck::allocation::try_zeroed_vec(count).map_err(|()| Error::OutOfMemory)?;
-        cells.resize(count, Cell::default());
         let grapheme_alloc = ChunkAlloc::new(l.grapheme_alloc_layout, 16)?;
         let mut grapheme_data = Vec::new();
         let codepoints = grapheme_alloc.chunks * 4;
@@ -1486,15 +1484,18 @@ impl Page {
             // Row metadata/cell indices were validated above (`get_row`).
             let from = source_row.cells() + src_x;
             let to = original_row.cells() + dst_x;
-            for n in 0..len {
+            let src_cells = &src.cells[from..from + len];
-                let cell = src.cells[from + n];
+            let mut run = 0;
+            for (n, cell) in src_cells.iter().enumerate() {
                 if !cell.has_grapheme() && !cell.hyperlink() && !cell.has_styling() {
-                    // Plain cell: no managed memory to copy.
+                    continue;
-                    self.cells[to + n] = cell;
-                } else {
-                    self.clone_cell_from_cleared(src, from + n, dst_x + n, dst_y)?;
                 }
+                // Bulk-copy the preceding run of plain cells (no managed memory).
+                self.cells[to + run..to + n].copy_from_slice(&src_cells[run..n]);
+                self.clone_cell_from_cleared(src, from + n, dst_x + n, dst_y)?;
+                run = n + 1;
             }
+            self.cells[to + run..to + len].copy_from_slice(&src_cells[run..len]);
         } else {
             let from = source_row.cells() + src_x;
             let to = original_row.cells() + dst_x;

crates/ghostty-terminal/src/page_list.rs

@@ -153,6 +153,8 @@ pub struct PageList {
     nodes: Vec<Option<Node>>,
     free_nodes: Vec<usize>,
     page_pool: Vec<Page>,
+    /// Never-written pages (all cells zero) that need no `reinit` memset.
+    clean_pool: Vec<Page>,
     /// Oldest page.
     pub first: Option<NodeId>,
     /// Last active page.
@@ -211,6 +213,7 @@ impl PageList {
             nodes: Vec::new(),
             free_nodes: Vec::new(),
             page_pool: Vec::new(),
+            clean_pool: Vec::new(),
             first: None,
             last: None,
             page_serial: 0,
@@ -257,7 +260,13 @@ impl PageList {
             return Err(Error::OutOfSpace);
         }
         self.nodes.try_reserve(1).map_err(|_| Error::OutOfMemory)?;
-        let mut data = if let Some(pos) = self.page_pool.iter().position(|p| p.capacity == cap) {
+        let mut data = if let Some(mut p) = self.clean_pool.pop().filter(|p| p.capacity == cap) {
+            p.size = page::Size {
+                cols: cap.cols,
+                rows: cap.rows,
+            };
+            p
+        } else if let Some(pos) = self.page_pool.iter().position(|p| p.capacity == cap) {
             let mut p = self.page_pool.swap_remove(pos);
             p.reinit()?;
             p
@@ -2737,7 +2746,8 @@ impl PageList {
         let mut result = Self::init(self.cols, self.rows, Some(self.explicit_max_size))?;
         while let Some(id) = result.first {
             let node = result.detach(id).ok_or(Error::InvalidId)?;
-            result.recycle(node);
+            // Freshly initialised and never written: reuse without re-zeroing.
+            result.clean_pool.push(node.data);
         }
         result.page_serial = 0;
         result.total_rows = 0;

crates/ghostty-vt/src/osc/mod.rs

@@ -496,6 +496,37 @@ impl Parser {
         self.next_state(c);
     }
 
+    /// Consume a run of characters; exactly equivalent to calling
+    /// [`Parser::next`] for each byte, but payload bytes are appended in bulk.
+    pub fn next_slice(&mut self, mut bytes: &[u8]) {
+        while let Some((&c, rest)) = bytes.split_first() {
+            if self.state == State::Invalid {
+                return;
+            }
+            match self.writer {
+                Writer::None => {
+                    self.next_state(c);
+                    bytes = rest;
+                }
+                Writer::Allocating => {
+                    self.buffer.extend_from_slice(bytes);
+                    return;
+                }
+                Writer::Fixed => {
+                    let room = Self::MAX_BUF.saturating_sub(self.buffer.len());
+                    if bytes.len() <= room {
+                        self.buffer.extend_from_slice(bytes);
+                    } else {
+                        // Overflow: keep what fits, discard the rest.
+                        self.buffer.extend_from_slice(&bytes[..room]);
+                        self.state = State::Invalid;
+                    }
+                    return;
+                }
+            }
+        }
+    }
+
     #[inline(never)]
     fn next_state(&mut self, c: u8) {
         #[allow(clippy::enum_glob_use)]
@@ -749,3 +780,31 @@ impl Parser {
         }
     }
 }
+
+#[cfg(test)]
+mod next_slice_tests {
+    use super::*;
+
+    #[test]
+    fn next_slice_matches_next() {
+        for alloc in [false, true] {
+            for prefix in ["0;", "52;c;", "4;", "7;", "8;;", "99;", "x"] {
+                for n in [0usize, 1, 5, 2040, 2046, 2047, 2048, 2049, 2050, 3000] {
+                    let mut input = prefix.as_bytes().to_vec();
+                    input.extend((0..n).map(|i| b'a' + (i % 26) as u8));
+                    let mut a = Parser::init(alloc);
+                    let mut b = Parser::init(alloc);
+                    for &c in &input {
+                        a.next(c);
+                    }
+                    let mid = input.len() / 2;
+                    b.next_slice(&input[..mid]);
+                    b.next_slice(&input[mid..]);
+                    assert_eq!(a.state, b.state);
+                    assert_eq!(a.buffer, b.buffer);
+                    assert_eq!(a.writer, b.writer);
+                }
+            }
+        }
+    }
+}

docs/rewrite/notes/perf-P1.md

@@ -76,3 +76,23 @@ utf8 0.85 s, sgr 1.6 s (no Zig counterpart; profile shows no pathology).
   buffer); `Parser::data()` also clones the buffer.
 - terminal-parser sgr (1.5x): moving the 64-byte `Option<Action>` array out of the
   parser; real stream path (`terminal-stream sgr`) is at parity.
+
+## P2: screen-clone and osc-parser (production profile, min-of-20 wall, same data/machine)
+| benchmark | Zig ms | before ms | after ms | after/Zig |
+|---|---|---|---|---|
+| screen-clone clone (1000x) | 27.1 | 78 | 28.2 | 1.04 |
+| screen-clone render (1000x) | 4.5 | 11.7 | 11.5 | 2.5 (unchanged) |
+| osc-parser (50 MB) | 27.9 | 54.5 | 31.3 | 1.12 |
+
+- `PageList::clone_range` built a throwaway page via `Self::init`, recycled it into the pool, then
+  `create_node` re-memset the whole page in `Page::reinit` (page-faulting every cell). Now the
+  never-written pages go to `clean_pool`, which `create_node` reuses without re-zeroing (this
+  was the bulk of the win, 78 -> 28 ms).
+- `Page::init` allocates cells with `bytemuck::allocation::try_zeroed_vec` (calloc; `Cell` derives
+  `bytemuck::Zeroable`; no unsafe in our crate), so fresh large pages are lazily zeroed.
+- `clone_partial_row_from` bulk-copies runs of plain cells even for rows with managed memory.
+- `osc::Parser::next_slice(&[u8])`: exact equivalent of per-byte `next` (incl. 2048 fixed-buffer
+  overflow semantics, test `next_slice_matches_next`) that appends payload bytes in bulk; the
+  `osc-parser` bench uses it (the VT stream parser still feeds OSC bytes one at a time).
+- Remaining: render (2.5x): `RenderState::update` per-row `Vec<Style>` resize (84 B/cell, every
+  row is styled in this benchmark) and cell copy; a lazy style vec did not help here, so reverted.