feat: improve Terminal emulator performance

kernel/src/Api/IoRedir.hpp

@@ -13,6 +13,8 @@
 
 namespace Montauk {
 
+    static constexpr uint32_t RedirOutputStreamCapacity = 64 * 1024;
+
     static int Sys_SpawnRedir(const char* path, const char* args) {
         char resolved[256];
         if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return -1;
@@ -28,7 +30,7 @@ namespace Montauk {
             return -1;
         }
 
-        Ipc::Stream* outStream = Ipc::CreateStream();
+        Ipc::Stream* outStream = Ipc::CreateStream(RedirOutputStreamCapacity);
         Ipc::Stream* inStream = Ipc::CreateStream();
         Ipc::Mailbox* keyMailbox = Ipc::CreateMailbox();
         if (outStream == nullptr || inStream == nullptr || keyMailbox == nullptr) {

programs/include/gui/terminal.hpp

@@ -20,11 +20,14 @@ struct TermCell {
 };
 
 static constexpr int TERM_MAX_SCROLLBACK = 500;
-static constexpr int TERM_MAX_READS_PER_POLL = 8;
+static constexpr int TERM_READ_BUF_SIZE = 8192;
+static constexpr int TERM_MAX_DRAIN_BYTES_PER_POLL = 256 * 1024;
+static constexpr int TERM_DRAIN_GRACE_YIELDS = 2;
 
 struct TerminalState {
     TermCell* cells;
     TermCell* alt_cells;         // alternate screen buffer
+    TermCell* render_cells;      // last visible cells painted into the window buffer
     int cols, rows;
     int cursor_x, cursor_y;
     int saved_cursor_x, saved_cursor_y;  // saved cursor for alternate screen
@@ -39,6 +42,15 @@ struct TerminalState {
     bool alt_screen_active;
     bool reverse_video;
     bool dirty;                  // true when content changed since last render
+    bool render_valid;
+    bool render_cursor_visible;
+    int render_cols, render_rows;
+    int render_term_cols, render_term_rows;
+    int render_base_row;
+    int render_pw, render_ph;
+    int render_cell_w, render_cell_h;
+    int render_cursor_x, render_cursor_y;
+    uint32_t* render_pixels;
 
     enum { STATE_NORMAL, STATE_ESC, STATE_CSI } parse_state;
     bool csi_private;            // true if '?' was seen after CSI
@@ -92,6 +104,24 @@ static inline TermCell* term_screen_row(TerminalState* t, int r) {
     return &t->cells[(t->scrollback_lines + r) * t->cols];
 }
 
+static inline bool term_color_equal(Color a, Color b) {
+    return a.r == b.r && a.g == b.g && a.b == b.b && a.a == b.a;
+}
+
+static inline bool term_cell_equal(const TermCell& a, const TermCell& b) {
+    return a.ch == b.ch && term_color_equal(a.fg, b.fg) && term_color_equal(a.bg, b.bg);
+}
+
+static inline void terminal_invalidate_render_cache(TerminalState* t) {
+    if (!t) return;
+    t->render_valid = false;
+    t->render_cursor_visible = false;
+}
+
+static inline Color terminal_erase_bg(TerminalState* t) {
+    return t->current_bg;
+}
+
 static inline void terminal_scroll_up(TerminalState* t) {
     if (!t->alt_screen_active && t->scrollback_lines < t->max_scrollback) {
         // Room for scrollback: top visible row becomes scrollback, no data movement
@@ -109,7 +139,7 @@ static inline void terminal_scroll_up(TerminalState* t) {
     // Clear the new bottom visible row
     TermCell* bottom = term_screen_row(t, t->rows - 1);
     for (int c = 0; c < t->cols; c++) {
-        bottom[c] = {' ', t->current_fg, colors::TERM_BG};
+        bottom[c] = {' ', t->current_fg, terminal_erase_bg(t)};
     }
 
     // Keep user's scrolled-back viewport stable
@@ -138,6 +168,21 @@ static inline void terminal_init_cells(TerminalState* t, int cols, int rows, int
     t->alt_screen_active = false;
     t->reverse_video = false;
     t->dirty = true;
+    t->render_cells = nullptr;
+    t->render_valid = false;
+    t->render_cursor_visible = false;
+    t->render_cols = 0;
+    t->render_rows = 0;
+    t->render_term_cols = 0;
+    t->render_term_rows = 0;
+    t->render_base_row = 0;
+    t->render_pw = 0;
+    t->render_ph = 0;
+    t->render_cell_w = 0;
+    t->render_cell_h = 0;
+    t->render_cursor_x = 0;
+    t->render_cursor_y = 0;
+    t->render_pixels = nullptr;
     t->parse_state = TerminalState::STATE_NORMAL;
     t->csi_private = false;
     t->csi_param_count = 0;
@@ -301,32 +346,33 @@ static inline void terminal_process_csi(TerminalState* t, char cmd) {
     }
     case 'J': {
         // Erase in display
+        Color erase_bg = terminal_erase_bg(t);
         if (p0 == 0) {
             // Clear from cursor to end
             TermCell* row = term_screen_row(t, t->cursor_y);
             for (int x = t->cursor_x; x < t->cols; x++)
-                row[x] = {' ', t->current_fg, colors::TERM_BG};
+                row[x] = {' ', t->current_fg, erase_bg};
             for (int r = t->cursor_y + 1; r < t->rows; r++) {
                 TermCell* rp = term_screen_row(t, r);
                 for (int c = 0; c < t->cols; c++)
-                    rp[c] = {' ', t->current_fg, colors::TERM_BG};
+                    rp[c] = {' ', t->current_fg, erase_bg};
             }
         } else if (p0 == 1) {
             // Clear from start to cursor
             for (int r = 0; r < t->cursor_y; r++) {
                 TermCell* rp = term_screen_row(t, r);
                 for (int c = 0; c < t->cols; c++)
-                    rp[c] = {' ', t->current_fg, colors::TERM_BG};
+                    rp[c] = {' ', t->current_fg, erase_bg};
             }
             TermCell* row = term_screen_row(t, t->cursor_y);
             for (int x = 0; x <= t->cursor_x; x++)
-                row[x] = {' ', t->current_fg, colors::TERM_BG};
+                row[x] = {' ', t->current_fg, erase_bg};
         } else if (p0 == 2) {
             // Clear entire screen
             for (int r = 0; r < t->rows; r++) {
                 TermCell* rp = term_screen_row(t, r);
                 for (int c = 0; c < t->cols; c++)
-                    rp[c] = {' ', t->current_fg, colors::TERM_BG};
+                    rp[c] = {' ', t->current_fg, erase_bg};
             }
             t->cursor_x = 0;
             t->cursor_y = 0;
@@ -339,9 +385,10 @@ static inline void terminal_process_csi(TerminalState* t, char cmd) {
         if (p0 == 0) { start = t->cursor_x; end = t->cols; }
         else if (p0 == 1) { start = 0; end = t->cursor_x + 1; }
         else if (p0 == 2) { start = 0; end = t->cols; }
+        Color erase_bg = terminal_erase_bg(t);
         TermCell* row = term_screen_row(t, t->cursor_y);
         for (int x = start; x < end; x++)
-            row[x] = {' ', t->current_fg, colors::TERM_BG};
+            row[x] = {' ', t->current_fg, erase_bg};
         break;
     }
     case 'm': {
@@ -520,122 +567,199 @@ static inline void terminal_feed(TerminalState* t, const char* data, int len) {
     }
 }
 
+static inline void terminal_fill_pixel_rect(uint32_t* pixels, int pw, int ph,
+                                            int x, int y, int w, int h,
+                                            uint32_t color) {
+    if (!pixels || pw <= 0 || ph <= 0 || w <= 0 || h <= 0) return;
+    int x0 = x < 0 ? 0 : x;
+    int y0 = y < 0 ? 0 : y;
+    int x1 = x + w > pw ? pw : x + w;
+    int y1 = y + h > ph ? ph : y + h;
+    for (int row = y0; row < y1; row++) {
+        uint32_t* dst = &pixels[row * pw + x0];
+        for (int col = x0; col < x1; col++)
+            *dst++ = color;
+    }
+}
+
+static inline void terminal_draw_bitmap_char(uint32_t* pixels, int pw, int ph,
+                                             int x, int y, unsigned char ch,
+                                             uint32_t color) {
+    const uint8_t* glyph = &font_data[ch * FONT_HEIGHT];
+    for (int fy = 0; fy < FONT_HEIGHT; fy++) {
+        int dy = y + fy;
+        if (dy >= ph) break;
+        if (dy < 0) continue;
+        uint8_t bits = glyph[fy];
+        for (int fx = 0; fx < FONT_WIDTH; fx++) {
+            if (bits & (0x80 >> fx)) {
+                int dx = x + fx;
+                if (dx >= pw) break;
+                if (dx >= 0) pixels[dy * pw + dx] = color;
+            }
+        }
+    }
+}
+
+static inline void terminal_draw_cell_pixels(const TermCell& cell,
+                                             uint32_t* pixels, int pw, int ph,
+                                             int px, int py, int cell_w, int cell_h,
+                                             bool fill_bg, bool use_ttf,
+                                             GlyphCache* gc) {
+    if (fill_bg) {
+        terminal_fill_pixel_rect(pixels, pw, ph, px, py, cell_w, cell_h,
+                                 cell.bg.to_pixel());
+    }
+
+    if (cell.ch <= 32 && cell.ch >= 0) return;
+
+    if (use_ttf) {
+        int baseline = py + gc->ascent;
+        fonts::mono->draw_char_to_buffer(pixels, pw, ph,
+            px, baseline, (unsigned char)cell.ch, cell.fg, gc);
+    } else {
+        terminal_draw_bitmap_char(pixels, pw, ph, px, py, (unsigned char)cell.ch,
+                                  cell.fg.to_pixel());
+    }
+}
+
+static inline void terminal_draw_cursor_pixels(const TermCell& cell,
+                                               uint32_t* pixels, int pw, int ph,
+                                               int px, int py, int cell_w, int cell_h,
+                                               bool use_ttf, GlyphCache* gc) {
+    terminal_fill_pixel_rect(pixels, pw, ph, px, py, cell_w, cell_h,
+                             colors::WHITE.to_pixel());
+
+    if (cell.ch <= 32 && cell.ch >= 0) return;
+
+    if (use_ttf) {
+        int baseline = py + gc->ascent;
+        fonts::mono->draw_char_to_buffer(pixels, pw, ph,
+            px, baseline, (unsigned char)cell.ch, colors::BLACK, gc);
+    } else {
+        terminal_draw_bitmap_char(pixels, pw, ph, px, py, (unsigned char)cell.ch,
+                                  colors::BLACK.to_pixel());
+    }
+}
+
+static inline bool terminal_prepare_render_cache(TerminalState* t,
+                                                 int visible_cols,
+                                                 int visible_rows) {
+    if (visible_cols <= 0 || visible_rows <= 0) return false;
+    if (t->render_cells && t->render_cols == visible_cols && t->render_rows == visible_rows)
+        return true;
+
+    if (t->render_cells) {
+        montauk::free(t->render_cells);
+        t->render_cells = nullptr;
+    }
+
+    int cells = visible_cols * visible_rows;
+    t->render_cells = (TermCell*)montauk::alloc(cells * sizeof(TermCell));
+    if (!t->render_cells) {
+        terminal_invalidate_render_cache(t);
+        t->render_cols = 0;
+        t->render_rows = 0;
+        return false;
+    }
+
+    t->render_cols = visible_cols;
+    t->render_rows = visible_rows;
+    terminal_invalidate_render_cache(t);
+    return true;
+}
+
+static inline TermCell terminal_visible_cell(TerminalState* t, int base_row, int x, int y) {
+    return t->cells[(base_row + y) * t->cols + x];
+}
+
 static inline void terminal_render(TerminalState* t, uint32_t* pixels, int pw, int ph) {
-    if (!t->dirty || !t->cells) return;
+    if (!t->dirty || !t->cells || !pixels || pw <= 0 || ph <= 0) return;
-    t->dirty = false;
 
     int cell_w = mono_cell_width();
     int cell_h = mono_cell_height();
+    if (cell_w <= 0 || cell_h <= 0) return;
+
     bool use_ttf = fonts::mono && fonts::mono->valid;
     GlyphCache* gc = use_ttf ? fonts::mono->get_cache(fonts::TERM_SIZE) : nullptr;
 
-    // Fill background using row-copy: fill first row, then memcpy to the rest
-    uint32_t bg_px = colors::TERM_BG.to_pixel();
-    int row_bytes = pw * sizeof(uint32_t);
-    for (int i = 0; i < pw; i++) pixels[i] = bg_px;
-    for (int r = 1; r < ph; r++) {
-        montauk::memcpy(&pixels[r * pw], pixels, row_bytes);
-    }
-
-    // Determine which rows of the buffer to display
     int base_row = t->scrollback_lines - t->view_offset;
     if (base_row < 0) base_row = 0;
 
-    // Render each visible cell
     int visible_rows = ph / cell_h;
     int visible_cols = pw / cell_w;
     if (visible_rows > t->rows) visible_rows = t->rows;
     if (visible_cols > t->cols) visible_cols = t->cols;
+    if (visible_rows <= 0 || visible_cols <= 0) return;
+
+    bool have_cache = terminal_prepare_render_cache(t, visible_cols, visible_rows);
+    bool full_redraw = !have_cache || !t->render_valid ||
+        t->render_pixels != pixels ||
+        t->render_pw != pw || t->render_ph != ph ||
+        t->render_cell_w != cell_w || t->render_cell_h != cell_h ||
+        t->render_base_row != base_row ||
+        t->render_term_cols != t->cols || t->render_term_rows != t->rows;
+
+    uint32_t bg_px = colors::TERM_BG.to_pixel();
+    if (full_redraw) {
+        int row_bytes = pw * sizeof(uint32_t);
+        for (int i = 0; i < pw; i++) pixels[i] = bg_px;
+        for (int r = 1; r < ph; r++)
+            montauk::memcpy(&pixels[r * pw], pixels, row_bytes);
+    } else if (t->render_cursor_visible) {
+        int x = t->render_cursor_x;
+        int y = t->render_cursor_y;
+        if (x >= 0 && x < visible_cols && y >= 0 && y < visible_rows) {
+            TermCell cell = terminal_visible_cell(t, base_row, x, y);
+            terminal_draw_cell_pixels(cell, pixels, pw, ph,
+                                      x * cell_w, y * cell_h,
+                                      cell_w, cell_h, true, use_ttf, gc);
+            if (have_cache)
+                t->render_cells[y * visible_cols + x] = cell;
+        }
+    }
 
     for (int r = 0; r < visible_rows; r++) {
         int py = r * cell_h;
-        int src_row = base_row + r;
         for (int c = 0; c < visible_cols; c++) {
-            int idx = src_row * t->cols + c;
+            TermCell cell = terminal_visible_cell(t, base_row, c, r);
-            TermCell& cell = t->cells[idx];
+            int cache_idx = r * visible_cols + c;
+            bool changed = full_redraw || !have_cache ||
+                !term_cell_equal(t->render_cells[cache_idx], cell);
+            if (!changed) continue;
 
-            int px = c * cell_w;
+            bool fill_bg = !full_redraw || cell.bg.to_pixel() != bg_px;
-
+            terminal_draw_cell_pixels(cell, pixels, pw, ph,
-            // Only draw cell background if it differs from terminal bg
+                                      c * cell_w, py,
-            uint32_t cell_bg = cell.bg.to_pixel();
+                                      cell_w, cell_h, fill_bg, use_ttf, gc);
-            if (cell_bg != bg_px) {
+            if (have_cache)
-                for (int fy = 0; fy < cell_h && py + fy < ph; fy++) {
+                t->render_cells[cache_idx] = cell;
-                    uint32_t* row = &pixels[(py + fy) * pw + px];
-                    for (int fx = 0; fx < cell_w && px + fx < pw; fx++) {
-                        row[fx] = cell_bg;
-                    }
-                }
-            }
-
-            // Draw character glyph
-            if (cell.ch > 32 || cell.ch < 0) {
-                if (use_ttf) {
-                    int baseline = py + gc->ascent;
-                    fonts::mono->draw_char_to_buffer(pixels, pw, ph,
-                        px, baseline, (unsigned char)cell.ch, cell.fg, gc);
-                } else {
-                    uint32_t cell_fg = cell.fg.to_pixel();
-                    const uint8_t* glyph = &font_data[(unsigned char)cell.ch * FONT_HEIGHT];
-                    for (int fy = 0; fy < FONT_HEIGHT; fy++) {
-                        int dy = py + fy;
-                        if (dy >= ph) break;
-                        uint8_t bits = glyph[fy];
-                        for (int fx = 0; fx < FONT_WIDTH; fx++) {
-                            if (bits & (0x80 >> fx)) {
-                                int dx = px + fx;
-                                if (dx >= pw) break;
-                                pixels[dy * pw + dx] = cell_fg;
-                            }
-                        }
-                    }
-                }
-            }
         }
     }
 
-    // Draw cursor (only when viewing live position)
+    bool cursor_now_visible = t->view_offset == 0 && t->cursor_visible &&
-    if (t->view_offset == 0 && t->cursor_visible &&
+        t->cursor_x >= 0 && t->cursor_y >= 0 &&
-        t->cursor_x < visible_cols && t->cursor_y < visible_rows) {
+        t->cursor_x < visible_cols && t->cursor_y < visible_rows;
-        int cx = t->cursor_x * cell_w;
+    if (cursor_now_visible) {
-        int cy = t->cursor_y * cell_h;
+        TermCell cell = terminal_visible_cell(t, base_row, t->cursor_x, t->cursor_y);
-        uint32_t cursor_px = colors::WHITE.to_pixel();
+        terminal_draw_cursor_pixels(cell, pixels, pw, ph,
-        for (int fy = 0; fy < cell_h; fy++) {
+                                    t->cursor_x * cell_w, t->cursor_y * cell_h,
-            int dy = cy + fy;
+                                    cell_w, cell_h, use_ttf, gc);
-            if (dy >= ph) break;
-            for (int fx = 0; fx < cell_w; fx++) {
-                int dx = cx + fx;
-                if (dx >= pw) break;
-                pixels[dy * pw + dx] = cursor_px;
-            }
-        }
-        // Draw character on top of cursor in black
-        if (t->cursor_y < t->rows && t->cursor_x < t->cols) {
-            TermCell* row = term_screen_row(t, t->cursor_y);
-            char ch = row[t->cursor_x].ch;
-            if (ch > 32 || ch < 0) {
-                if (use_ttf) {
-                    int baseline = cy + gc->ascent;
-                    fonts::mono->draw_char_to_buffer(pixels, pw, ph,
-                        cx, baseline, (unsigned char)ch, colors::BLACK, gc);
-                } else {
-                    const uint8_t* glyph = &font_data[(unsigned char)ch * FONT_HEIGHT];
-                    uint32_t black_px = colors::BLACK.to_pixel();
-                    for (int fy = 0; fy < FONT_HEIGHT; fy++) {
-                        int dy = cy + fy;
-                        if (dy >= ph) break;
-                        uint8_t bits = glyph[fy];
-                        for (int fx = 0; fx < FONT_WIDTH; fx++) {
-                            if (bits & (0x80 >> fx)) {
-                                int dx = cx + fx;
-                                if (dx >= pw) break;
-                                pixels[dy * pw + dx] = black_px;
-                            }
-                        }
-                    }
-                }
-            }
-        }
     }
+
+    t->render_valid = have_cache;
+    t->render_cursor_visible = cursor_now_visible;
+    t->render_cursor_x = t->cursor_x;
+    t->render_cursor_y = t->cursor_y;
+    t->render_term_cols = t->cols;
+    t->render_term_rows = t->rows;
+    t->render_base_row = base_row;
+    t->render_pw = pw;
+    t->render_ph = ph;
+    t->render_cell_w = cell_w;
+    t->render_cell_h = cell_h;
+    t->render_pixels = pixels;
+    t->dirty = false;
 }
 
 static inline void terminal_resize(TerminalState* t, int new_cols, int new_rows) {
@@ -683,6 +807,10 @@ static inline void terminal_resize(TerminalState* t, int new_cols, int new_rows)
 
     if (t->cells) montauk::free(t->cells);
     if (t->alt_cells) montauk::free(t->alt_cells);
+    if (t->render_cells) {
+        montauk::free(t->render_cells);
+        t->render_cells = nullptr;
+    }
 
     t->cells = new_cells;
     t->alt_cells = new_alt;
@@ -691,6 +819,7 @@ static inline void terminal_resize(TerminalState* t, int new_cols, int new_rows)
     t->scrollback_lines = new_scrollback;
     t->cursor_x = new_cursor_x;
     t->cursor_y = new_cursor_y;
+    terminal_invalidate_render_cache(t);
 
     // Clamp view offset
     if (t->view_offset > t->scrollback_lines)
@@ -716,15 +845,28 @@ static inline void terminal_handle_key(TerminalState* t, const Montauk::KeyEvent
 // Returns false if the child process has exited
 static inline bool terminal_poll(TerminalState* t) {
     if (t->child_pid <= 0) return false;
-    char buf[4096];
+    char buf[TERM_READ_BUF_SIZE];
-    // Drain a bounded amount per frame so window close/input events stay responsive.
+    int drained = 0;
-    for (int reads = 0; reads < TERM_MAX_READS_PER_POLL; reads++) {
+    int grace_yields = 0;
+
+    // Drain bursty TUI output as one frame. When a child was blocked on a full
+    // stream, a short yield lets it finish the redraw before we present.
+    for (;;) {
         int n = montauk::childio_read(t->child_pid, buf, sizeof(buf));
         if (n > 0) {
             terminal_feed(t, buf, n);
+            drained += n;
+            grace_yields = 0;
+            if (drained >= TERM_MAX_DRAIN_BYTES_PER_POLL)
+                break;
         } else {
             // n == -1 means child process is gone; n == 0 means no data yet
             if (n < 0) { t->child_pid = 0; return false; }
+            if (drained > 0 && grace_yields < TERM_DRAIN_GRACE_YIELDS) {
+                grace_yields++;
+                montauk::yield();
+                continue;
+            }
             break;
         }
     }

programs/src/klog/main.cpp

@@ -148,6 +148,7 @@ static void klog_handle_mouse(const Montauk::WinEvent& ev) {
 static void klog_cleanup() {
     if (g_klog.term.cells) montauk::free(g_klog.term.cells);
     if (g_klog.term.alt_cells) montauk::free(g_klog.term.alt_cells);
+    if (g_klog.term.render_cells) montauk::free(g_klog.term.render_cells);
     if (g_klog.klog_buf) montauk::mfree(g_klog.klog_buf);
     montauk::memset(&g_klog, 0, sizeof(g_klog));
 }

programs/src/terminal/main.cpp

@@ -93,6 +93,7 @@ static void term_free_tab(TerminalState* ts) {
     if (ts->child_pid > 0) term_kill_process_tree(ts->child_pid);
     if (ts->cells) montauk::free(ts->cells);
     if (ts->alt_cells) montauk::free(ts->alt_cells);
+    if (ts->render_cells) montauk::free(ts->render_cells);
     montauk::mfree(ts);
 }
 
@@ -133,6 +134,7 @@ static void term_close_tab(int idx) {
     if (g_tabs.active_tab >= g_tabs.tab_count)
         g_tabs.active_tab = g_tabs.tab_count - 1;
 
+    terminal_invalidate_render_cache(g_tabs.tabs[g_tabs.active_tab]);
     g_tabs.tabs[g_tabs.active_tab]->dirty = true;
     term_request_redraw();
 }
@@ -192,6 +194,7 @@ static bool term_render() {
     }
 
     if (g_tabs.active_tab != g_tabs.prev_active_tab) {
+        terminal_invalidate_render_cache(g_tabs.tabs[g_tabs.active_tab]);
         g_tabs.tabs[g_tabs.active_tab]->dirty = true;
         g_tabs.prev_active_tab = g_tabs.active_tab;
     }

template/sysroot/include/gui/terminal.hpp

@@ -20,10 +20,14 @@ struct TermCell {
 };
 
 static constexpr int TERM_MAX_SCROLLBACK = 500;
+static constexpr int TERM_READ_BUF_SIZE = 8192;
+static constexpr int TERM_MAX_DRAIN_BYTES_PER_POLL = 256 * 1024;
+static constexpr int TERM_DRAIN_GRACE_YIELDS = 2;
 
 struct TerminalState {
     TermCell* cells;
     TermCell* alt_cells;         // alternate screen buffer
+    TermCell* render_cells;      // last visible cells painted into the window buffer
     int cols, rows;
     int cursor_x, cursor_y;
     int saved_cursor_x, saved_cursor_y;  // saved cursor for alternate screen
@@ -38,6 +42,15 @@ struct TerminalState {
     bool alt_screen_active;
     bool reverse_video;
     bool dirty;                  // true when content changed since last render
+    bool render_valid;
+    bool render_cursor_visible;
+    int render_cols, render_rows;
+    int render_term_cols, render_term_rows;
+    int render_base_row;
+    int render_pw, render_ph;
+    int render_cell_w, render_cell_h;
+    int render_cursor_x, render_cursor_y;
+    uint32_t* render_pixels;
 
     enum { STATE_NORMAL, STATE_ESC, STATE_CSI } parse_state;
     bool csi_private;            // true if '?' was seen after CSI
@@ -69,11 +82,46 @@ static inline Color term_ansi_color(int idx) {
     }
 }
 
+// ANSI 256-color palette (0-15 = standard, 16-231 = RGB cube, 232-255 = grayscale)
+static inline Color term_ansi_256_color(int idx) {
+    if (idx < 0) return colors::TERM_FG;
+    if (idx <= 15) return term_ansi_color(idx);
+    if (idx <= 231) {
+        // 6x6x6 RGB cube: idx 16 = (0,0,0), idx 231 = (5,5,5)
+        int r = ((idx - 16) / 36) * 51;
+        int g = (((idx - 16) % 36) / 6) * 51;
+        int b = ((idx - 16) % 6) * 51;
+        return Color::from_rgb(r, g, b);
+    }
+    // Grayscale: idx 232 = #080808, idx 255 = #eeeeee
+    int gray = 8 + (idx - 232) * 10;
+    if (gray > 238) gray = 238;
+    return Color::from_rgb(gray, gray, gray);
+}
+
 // Helper: pointer to start of screen-relative row r in the cells buffer
 static inline TermCell* term_screen_row(TerminalState* t, int r) {
     return &t->cells[(t->scrollback_lines + r) * t->cols];
 }
 
+static inline bool term_color_equal(Color a, Color b) {
+    return a.r == b.r && a.g == b.g && a.b == b.b && a.a == b.a;
+}
+
+static inline bool term_cell_equal(const TermCell& a, const TermCell& b) {
+    return a.ch == b.ch && term_color_equal(a.fg, b.fg) && term_color_equal(a.bg, b.bg);
+}
+
+static inline void terminal_invalidate_render_cache(TerminalState* t) {
+    if (!t) return;
+    t->render_valid = false;
+    t->render_cursor_visible = false;
+}
+
+static inline Color terminal_erase_bg(TerminalState* t) {
+    return t->current_bg;
+}
+
 static inline void terminal_scroll_up(TerminalState* t) {
     if (!t->alt_screen_active && t->scrollback_lines < t->max_scrollback) {
         // Room for scrollback: top visible row becomes scrollback, no data movement
@@ -91,7 +139,7 @@ static inline void terminal_scroll_up(TerminalState* t) {
     // Clear the new bottom visible row
     TermCell* bottom = term_screen_row(t, t->rows - 1);
     for (int c = 0; c < t->cols; c++) {
-        bottom[c] = {' ', t->current_fg, colors::TERM_BG};
+        bottom[c] = {' ', t->current_fg, terminal_erase_bg(t)};
     }
 
     // Keep user's scrolled-back viewport stable
@@ -120,6 +168,21 @@ static inline void terminal_init_cells(TerminalState* t, int cols, int rows, int
     t->alt_screen_active = false;
     t->reverse_video = false;
     t->dirty = true;
+    t->render_cells = nullptr;
+    t->render_valid = false;
+    t->render_cursor_visible = false;
+    t->render_cols = 0;
+    t->render_rows = 0;
+    t->render_term_cols = 0;
+    t->render_term_rows = 0;
+    t->render_base_row = 0;
+    t->render_pw = 0;
+    t->render_ph = 0;
+    t->render_cell_w = 0;
+    t->render_cell_h = 0;
+    t->render_cursor_x = 0;
+    t->render_cursor_y = 0;
+    t->render_pixels = nullptr;
     t->parse_state = TerminalState::STATE_NORMAL;
     t->csi_private = false;
     t->csi_param_count = 0;
@@ -283,32 +346,33 @@ static inline void terminal_process_csi(TerminalState* t, char cmd) {
     }
     case 'J': {
         // Erase in display
+        Color erase_bg = terminal_erase_bg(t);
         if (p0 == 0) {
             // Clear from cursor to end
             TermCell* row = term_screen_row(t, t->cursor_y);
             for (int x = t->cursor_x; x < t->cols; x++)
-                row[x] = {' ', t->current_fg, colors::TERM_BG};
+                row[x] = {' ', t->current_fg, erase_bg};
             for (int r = t->cursor_y + 1; r < t->rows; r++) {
                 TermCell* rp = term_screen_row(t, r);
                 for (int c = 0; c < t->cols; c++)
-                    rp[c] = {' ', t->current_fg, colors::TERM_BG};
+                    rp[c] = {' ', t->current_fg, erase_bg};
             }
         } else if (p0 == 1) {
             // Clear from start to cursor
             for (int r = 0; r < t->cursor_y; r++) {
                 TermCell* rp = term_screen_row(t, r);
                 for (int c = 0; c < t->cols; c++)
-                    rp[c] = {' ', t->current_fg, colors::TERM_BG};
+                    rp[c] = {' ', t->current_fg, erase_bg};
             }
             TermCell* row = term_screen_row(t, t->cursor_y);
             for (int x = 0; x <= t->cursor_x; x++)
-                row[x] = {' ', t->current_fg, colors::TERM_BG};
+                row[x] = {' ', t->current_fg, erase_bg};
         } else if (p0 == 2) {
             // Clear entire screen
             for (int r = 0; r < t->rows; r++) {
                 TermCell* rp = term_screen_row(t, r);
                 for (int c = 0; c < t->cols; c++)
-                    rp[c] = {' ', t->current_fg, colors::TERM_BG};
+                    rp[c] = {' ', t->current_fg, erase_bg};
             }
             t->cursor_x = 0;
             t->cursor_y = 0;
@@ -321,15 +385,44 @@ static inline void terminal_process_csi(TerminalState* t, char cmd) {
         if (p0 == 0) { start = t->cursor_x; end = t->cols; }
         else if (p0 == 1) { start = 0; end = t->cursor_x + 1; }
         else if (p0 == 2) { start = 0; end = t->cols; }
+        Color erase_bg = terminal_erase_bg(t);
         TermCell* row = term_screen_row(t, t->cursor_y);
         for (int x = start; x < end; x++)
-            row[x] = {' ', t->current_fg, colors::TERM_BG};
+            row[x] = {' ', t->current_fg, erase_bg};
         break;
     }
     case 'm': {
         // SGR - Set Graphics Rendition
         for (int i = 0; i < t->csi_param_count; i++) {
             int code = t->csi_params[i];
+
+            // Handle extended color sequences: 38;5;N or 48;5;N or 38;2;R;G;B or 48;2;R;G;B
+            if ((code == 38 || code == 48) && i + 2 < t->csi_param_count) {
+                if (t->csi_params[i + 1] == 5) {
+                    // 256-color: 38;5;N or 48;5;N
+                    int color_idx = t->csi_params[i + 2];
+                    if (code == 38) {
+                        t->current_fg = term_ansi_256_color(color_idx);
+                    } else {
+                        t->current_bg = term_ansi_256_color(color_idx);
+                    }
+                    i += 2;
+                    continue;
+                } else if (t->csi_params[i + 1] == 2 && i + 4 < t->csi_param_count) {
+                    // Truecolor: 38;2;R;G;B or 48;2;R;G;B
+                    int r = t->csi_params[i + 2];
+                    int g = t->csi_params[i + 3];
+                    int b = t->csi_params[i + 4];
+                    if (code == 38) {
+                        t->current_fg = Color::from_rgb(r, g, b);
+                    } else {
+                        t->current_bg = Color::from_rgb(r, g, b);
+                    }
+                    i += 4;
+                    continue;
+                }
+            }
+
             if (code == 0) {
                 t->current_fg = colors::TERM_FG;
                 t->current_bg = colors::TERM_BG;
@@ -474,122 +567,199 @@ static inline void terminal_feed(TerminalState* t, const char* data, int len) {
     }
 }
 
+static inline void terminal_fill_pixel_rect(uint32_t* pixels, int pw, int ph,
+                                            int x, int y, int w, int h,
+                                            uint32_t color) {
+    if (!pixels || pw <= 0 || ph <= 0 || w <= 0 || h <= 0) return;
+    int x0 = x < 0 ? 0 : x;
+    int y0 = y < 0 ? 0 : y;
+    int x1 = x + w > pw ? pw : x + w;
+    int y1 = y + h > ph ? ph : y + h;
+    for (int row = y0; row < y1; row++) {
+        uint32_t* dst = &pixels[row * pw + x0];
+        for (int col = x0; col < x1; col++)
+            *dst++ = color;
+    }
+}
+
+static inline void terminal_draw_bitmap_char(uint32_t* pixels, int pw, int ph,
+                                             int x, int y, unsigned char ch,
+                                             uint32_t color) {
+    const uint8_t* glyph = &font_data[ch * FONT_HEIGHT];
+    for (int fy = 0; fy < FONT_HEIGHT; fy++) {
+        int dy = y + fy;
+        if (dy >= ph) break;
+        if (dy < 0) continue;
+        uint8_t bits = glyph[fy];
+        for (int fx = 0; fx < FONT_WIDTH; fx++) {
+            if (bits & (0x80 >> fx)) {
+                int dx = x + fx;
+                if (dx >= pw) break;
+                if (dx >= 0) pixels[dy * pw + dx] = color;
+            }
+        }
+    }
+}
+
+static inline void terminal_draw_cell_pixels(const TermCell& cell,
+                                             uint32_t* pixels, int pw, int ph,
+                                             int px, int py, int cell_w, int cell_h,
+                                             bool fill_bg, bool use_ttf,
+                                             GlyphCache* gc) {
+    if (fill_bg) {
+        terminal_fill_pixel_rect(pixels, pw, ph, px, py, cell_w, cell_h,
+                                 cell.bg.to_pixel());
+    }
+
+    if (cell.ch <= 32 && cell.ch >= 0) return;
+
+    if (use_ttf) {
+        int baseline = py + gc->ascent;
+        fonts::mono->draw_char_to_buffer(pixels, pw, ph,
+            px, baseline, (unsigned char)cell.ch, cell.fg, gc);
+    } else {
+        terminal_draw_bitmap_char(pixels, pw, ph, px, py, (unsigned char)cell.ch,
+                                  cell.fg.to_pixel());
+    }
+}
+
+static inline void terminal_draw_cursor_pixels(const TermCell& cell,
+                                               uint32_t* pixels, int pw, int ph,
+                                               int px, int py, int cell_w, int cell_h,
+                                               bool use_ttf, GlyphCache* gc) {
+    terminal_fill_pixel_rect(pixels, pw, ph, px, py, cell_w, cell_h,
+                             colors::WHITE.to_pixel());
+
+    if (cell.ch <= 32 && cell.ch >= 0) return;
+
+    if (use_ttf) {
+        int baseline = py + gc->ascent;
+        fonts::mono->draw_char_to_buffer(pixels, pw, ph,
+            px, baseline, (unsigned char)cell.ch, colors::BLACK, gc);
+    } else {
+        terminal_draw_bitmap_char(pixels, pw, ph, px, py, (unsigned char)cell.ch,
+                                  colors::BLACK.to_pixel());
+    }
+}
+
+static inline bool terminal_prepare_render_cache(TerminalState* t,
+                                                 int visible_cols,
+                                                 int visible_rows) {
+    if (visible_cols <= 0 || visible_rows <= 0) return false;
+    if (t->render_cells && t->render_cols == visible_cols && t->render_rows == visible_rows)
+        return true;
+
+    if (t->render_cells) {
+        montauk::free(t->render_cells);
+        t->render_cells = nullptr;
+    }
+
+    int cells = visible_cols * visible_rows;
+    t->render_cells = (TermCell*)montauk::alloc(cells * sizeof(TermCell));
+    if (!t->render_cells) {
+        terminal_invalidate_render_cache(t);
+        t->render_cols = 0;
+        t->render_rows = 0;
+        return false;
+    }
+
+    t->render_cols = visible_cols;
+    t->render_rows = visible_rows;
+    terminal_invalidate_render_cache(t);
+    return true;
+}
+
+static inline TermCell terminal_visible_cell(TerminalState* t, int base_row, int x, int y) {
+    return t->cells[(base_row + y) * t->cols + x];
+}
+
 static inline void terminal_render(TerminalState* t, uint32_t* pixels, int pw, int ph) {
-    if (!t->dirty || !t->cells) return;
+    if (!t->dirty || !t->cells || !pixels || pw <= 0 || ph <= 0) return;
-    t->dirty = false;
 
     int cell_w = mono_cell_width();
     int cell_h = mono_cell_height();
+    if (cell_w <= 0 || cell_h <= 0) return;
+
     bool use_ttf = fonts::mono && fonts::mono->valid;
     GlyphCache* gc = use_ttf ? fonts::mono->get_cache(fonts::TERM_SIZE) : nullptr;
 
-    // Fill background using row-copy: fill first row, then memcpy to the rest
-    uint32_t bg_px = colors::TERM_BG.to_pixel();
-    int row_bytes = pw * sizeof(uint32_t);
-    for (int i = 0; i < pw; i++) pixels[i] = bg_px;
-    for (int r = 1; r < ph; r++) {
-        montauk::memcpy(&pixels[r * pw], pixels, row_bytes);
-    }
-
-    // Determine which rows of the buffer to display
     int base_row = t->scrollback_lines - t->view_offset;
     if (base_row < 0) base_row = 0;
 
-    // Render each visible cell
     int visible_rows = ph / cell_h;
     int visible_cols = pw / cell_w;
     if (visible_rows > t->rows) visible_rows = t->rows;
     if (visible_cols > t->cols) visible_cols = t->cols;
+    if (visible_rows <= 0 || visible_cols <= 0) return;
+
+    bool have_cache = terminal_prepare_render_cache(t, visible_cols, visible_rows);
+    bool full_redraw = !have_cache || !t->render_valid ||
+        t->render_pixels != pixels ||
+        t->render_pw != pw || t->render_ph != ph ||
+        t->render_cell_w != cell_w || t->render_cell_h != cell_h ||
+        t->render_base_row != base_row ||
+        t->render_term_cols != t->cols || t->render_term_rows != t->rows;
+
+    uint32_t bg_px = colors::TERM_BG.to_pixel();
+    if (full_redraw) {
+        int row_bytes = pw * sizeof(uint32_t);
+        for (int i = 0; i < pw; i++) pixels[i] = bg_px;
+        for (int r = 1; r < ph; r++)
+            montauk::memcpy(&pixels[r * pw], pixels, row_bytes);
+    } else if (t->render_cursor_visible) {
+        int x = t->render_cursor_x;
+        int y = t->render_cursor_y;
+        if (x >= 0 && x < visible_cols && y >= 0 && y < visible_rows) {
+            TermCell cell = terminal_visible_cell(t, base_row, x, y);
+            terminal_draw_cell_pixels(cell, pixels, pw, ph,
+                                      x * cell_w, y * cell_h,
+                                      cell_w, cell_h, true, use_ttf, gc);
+            if (have_cache)
+                t->render_cells[y * visible_cols + x] = cell;
+        }
+    }
 
     for (int r = 0; r < visible_rows; r++) {
         int py = r * cell_h;
-        int src_row = base_row + r;
         for (int c = 0; c < visible_cols; c++) {
-            int idx = src_row * t->cols + c;
+            TermCell cell = terminal_visible_cell(t, base_row, c, r);
-            TermCell& cell = t->cells[idx];
+            int cache_idx = r * visible_cols + c;
+            bool changed = full_redraw || !have_cache ||
+                !term_cell_equal(t->render_cells[cache_idx], cell);
+            if (!changed) continue;
 
-            int px = c * cell_w;
+            bool fill_bg = !full_redraw || cell.bg.to_pixel() != bg_px;
-
+            terminal_draw_cell_pixels(cell, pixels, pw, ph,
-            // Only draw cell background if it differs from terminal bg
+                                      c * cell_w, py,
-            uint32_t cell_bg = cell.bg.to_pixel();
+                                      cell_w, cell_h, fill_bg, use_ttf, gc);
-            if (cell_bg != bg_px) {
+            if (have_cache)
-                for (int fy = 0; fy < cell_h && py + fy < ph; fy++) {
+                t->render_cells[cache_idx] = cell;
-                    uint32_t* row = &pixels[(py + fy) * pw + px];
-                    for (int fx = 0; fx < cell_w && px + fx < pw; fx++) {
-                        row[fx] = cell_bg;
-                    }
-                }
-            }
-
-            // Draw character glyph
-            if (cell.ch > 32 || cell.ch < 0) {
-                if (use_ttf) {
-                    int baseline = py + gc->ascent;
-                    fonts::mono->draw_char_to_buffer(pixels, pw, ph,
-                        px, baseline, (unsigned char)cell.ch, cell.fg, gc);
-                } else {
-                    uint32_t cell_fg = cell.fg.to_pixel();
-                    const uint8_t* glyph = &font_data[(unsigned char)cell.ch * FONT_HEIGHT];
-                    for (int fy = 0; fy < FONT_HEIGHT; fy++) {
-                        int dy = py + fy;
-                        if (dy >= ph) break;
-                        uint8_t bits = glyph[fy];
-                        for (int fx = 0; fx < FONT_WIDTH; fx++) {
-                            if (bits & (0x80 >> fx)) {
-                                int dx = px + fx;
-                                if (dx >= pw) break;
-                                pixels[dy * pw + dx] = cell_fg;
-                            }
-                        }
-                    }
-                }
-            }
         }
     }
 
-    // Draw cursor (only when viewing live position)
+    bool cursor_now_visible = t->view_offset == 0 && t->cursor_visible &&
-    if (t->view_offset == 0 && t->cursor_visible &&
+        t->cursor_x >= 0 && t->cursor_y >= 0 &&
-        t->cursor_x < visible_cols && t->cursor_y < visible_rows) {
+        t->cursor_x < visible_cols && t->cursor_y < visible_rows;
-        int cx = t->cursor_x * cell_w;
+    if (cursor_now_visible) {
-        int cy = t->cursor_y * cell_h;
+        TermCell cell = terminal_visible_cell(t, base_row, t->cursor_x, t->cursor_y);
-        uint32_t cursor_px = colors::WHITE.to_pixel();
+        terminal_draw_cursor_pixels(cell, pixels, pw, ph,
-        for (int fy = 0; fy < cell_h; fy++) {
+                                    t->cursor_x * cell_w, t->cursor_y * cell_h,
-            int dy = cy + fy;
+                                    cell_w, cell_h, use_ttf, gc);
-            if (dy >= ph) break;
-            for (int fx = 0; fx < cell_w; fx++) {
-                int dx = cx + fx;
-                if (dx >= pw) break;
-                pixels[dy * pw + dx] = cursor_px;
-            }
-        }
-        // Draw character on top of cursor in black
-        if (t->cursor_y < t->rows && t->cursor_x < t->cols) {
-            TermCell* row = term_screen_row(t, t->cursor_y);
-            char ch = row[t->cursor_x].ch;
-            if (ch > 32 || ch < 0) {
-                if (use_ttf) {
-                    int baseline = cy + gc->ascent;
-                    fonts::mono->draw_char_to_buffer(pixels, pw, ph,
-                        cx, baseline, (unsigned char)ch, colors::BLACK, gc);
-                } else {
-                    const uint8_t* glyph = &font_data[(unsigned char)ch * FONT_HEIGHT];
-                    uint32_t black_px = colors::BLACK.to_pixel();
-                    for (int fy = 0; fy < FONT_HEIGHT; fy++) {
-                        int dy = cy + fy;
-                        if (dy >= ph) break;
-                        uint8_t bits = glyph[fy];
-                        for (int fx = 0; fx < FONT_WIDTH; fx++) {
-                            if (bits & (0x80 >> fx)) {
-                                int dx = cx + fx;
-                                if (dx >= pw) break;
-                                pixels[dy * pw + dx] = black_px;
-                            }
-                        }
-                    }
-                }
-            }
-        }
     }
+
+    t->render_valid = have_cache;
+    t->render_cursor_visible = cursor_now_visible;
+    t->render_cursor_x = t->cursor_x;
+    t->render_cursor_y = t->cursor_y;
+    t->render_term_cols = t->cols;
+    t->render_term_rows = t->rows;
+    t->render_base_row = base_row;
+    t->render_pw = pw;
+    t->render_ph = ph;
+    t->render_cell_w = cell_w;
+    t->render_cell_h = cell_h;
+    t->render_pixels = pixels;
+    t->dirty = false;
 }
 
 static inline void terminal_resize(TerminalState* t, int new_cols, int new_rows) {
@@ -637,6 +807,10 @@ static inline void terminal_resize(TerminalState* t, int new_cols, int new_rows)
 
     if (t->cells) montauk::free(t->cells);
     if (t->alt_cells) montauk::free(t->alt_cells);
+    if (t->render_cells) {
+        montauk::free(t->render_cells);
+        t->render_cells = nullptr;
+    }
 
     t->cells = new_cells;
     t->alt_cells = new_alt;
@@ -645,6 +819,7 @@ static inline void terminal_resize(TerminalState* t, int new_cols, int new_rows)
     t->scrollback_lines = new_scrollback;
     t->cursor_x = new_cursor_x;
     t->cursor_y = new_cursor_y;
+    terminal_invalidate_render_cache(t);
 
     // Clamp view offset
     if (t->view_offset > t->scrollback_lines)
@@ -670,15 +845,28 @@ static inline void terminal_handle_key(TerminalState* t, const Montauk::KeyEvent
 // Returns false if the child process has exited
 static inline bool terminal_poll(TerminalState* t) {
     if (t->child_pid <= 0) return false;
-    char buf[4096];
+    char buf[TERM_READ_BUF_SIZE];
-    // Drain all available data so large output renders in one frame
+    int drained = 0;
+    int grace_yields = 0;
+
+    // Drain bursty TUI output as one frame. When a child was blocked on a full
+    // stream, a short yield lets it finish the redraw before we present.
     for (;;) {
         int n = montauk::childio_read(t->child_pid, buf, sizeof(buf));
         if (n > 0) {
             terminal_feed(t, buf, n);
+            drained += n;
+            grace_yields = 0;
+            if (drained >= TERM_MAX_DRAIN_BYTES_PER_POLL)
+                break;
         } else {
             // n == -1 means child process is gone; n == 0 means no data yet
             if (n < 0) { t->child_pid = 0; return false; }
+            if (drained > 0 && grace_yields < TERM_DRAIN_GRACE_YIELDS) {
+                grace_yields++;
+                montauk::yield();
+                continue;
+            }
             break;
         }
     }