szn

Bug 483 — Per-space writeBytes(" ") in the render loop

483. No run-length batching for blank cells

Severity: LOW (performance) Status: ✅ FIXED Files: src/server/render.zig:597-599, src/server/render.zig:709

Problem

if (cp == 0 or cp == ' ') {
    try self.writeBytes(" ");
    cur_cx += 1;
}

(render.zig:597-599) Each blank/sixel-anchor cell is one separate writeBytes(" ") append. Same pattern at render.zig:709 for status-bar padding. In the interactive loop writeBytes appends to the capture frame buffer, so 80 consecutive spaces per row is 80 small append calls where one appendSlice of a pre-sized 80-space string would do. On mostly-blank screens (the common case — prompts, sparse TUIs) blank cells are the dominant cell type, so this is the most frequent inner-loop append.

Proposed Fix

Accumulate a run-length of consecutive spaces and emit one writeBytes(spaces[0..run]) (a static space buffer sized to sx) instead of N one-byte appends. Same for the status-bar padding loop at line 709.

Resolution

renderContent (src/server/render.zig) batches a run of consecutive identical blank cells into one writeBytes(spaces[0..run]) using a 256-byte space scratch; runs are capped at the scratch size so wider rows spill into the next batch, padding cells stop a run, and cells whose last-known state was a sixel are excluded so the per-cell \x1b[X erase behaviour is unchanged. The status-bar padding loop now emits the remaining columns as space-run chunks of the same scratch buffer instead of one write per column.

Tests: a batched blank row is emitted as a single space run, and renderStatusBar pads a short line to the full width with the expected run.

Verified: zig build test -Doptimize=ReleaseFast — 1012/1012 tests pass.