Repository navigation
id(ssh): incremental (diff) ANSI renderer instead of full clear-repaint #562
Description
What and why
Render only changed cells/rows to the terminal. This is the terminal-side counterpart of F12.11 delta frames and the biggest win on xterm-class terminals. This is the terminal-side counterpart of F12.11 delta frames and the biggest win on xterm-class terminals.
Source
- RPT §14 Tier 2
- RPT §3 'no incremental diffing'
RPT = doc/2026-10-18T18-40-00Z_report_platform_world_state/2026-10-18T18-40-00Z_report_platform_world_state.md, REF = doc/2026-10-18T00-00-00Z_reference_platform_state/2026-10-18T00-00-00Z_reference_platform_state.md. Both are uncommitted reports in the platform-world-apps-migration worktree.
Current status
Not started. pkgs/id/src/world_presentation.rs:370 starts every render with \x1b[H\x1b[2J; no diff state is kept per session.
Acceptance criteria
- Each session keeps the previous frame and sends only changed cells or rows
- Resize and attach still send a full repaint
- A parity test shows applying the diffs reproduces the full render
- Byte savings are measured against the F11.9 baseline
Size
M
Prerequisites
- id(ssh): use the alternate screen for raw interactive SSH sessions #545 — D1-042 Alternate screen for the raw interactive SSH session
Activity
dezren39 commentedon Oct 24, 2026
id train triage (2026-10-24): too-big-needs-design — incremental (diff) ANSI renderer (D1-044), terminal-side counterpart of F12.11.
dezren39 commentedon Oct 26, 2026
Intake review
Fleet-monitor intake pass (2026-10-26). Body already carries a full
## Acceptance criteriasection, so this pass is label normalization plus a liveness check — no body edit.Verified live on main
e3467560(2026-10-26):pkgs/id/src/world_presentation.rs:370still starts every render with\x1b[H\x1b[2J(full clear-repaint), and the test fixture at line 524 confirms the same shape. No per-session diff state exists. Still live, exactly as filed.Related: prerequisite #545 (alternate screen for the raw SSH session) is still open and unenriched. Byte-savings measurement depends on the F11.9 baseline, #542. Server-side counterpart: #561 (redraw batching) — the two compose; neither blocks the other.
Risks / notes: diff-renderer bugs corrupt terminals subtly (stale cells on resize/attach are the classic failure); the parity test in the acceptance criteria is the load-bearing one and should land with the renderer, not after.
Suggested approach: keep the previous frame per session in the presentation layer, emit row/cell diffs, force full repaint on resize and attach, and gate the rollout on the parity test plus a byte-savings measurement against the #542 baseline.
2 remaining items
dezren39 commentedon Nov 1, 2026
agent-note:v1
event: done
task: 2157 (openchamber scheduled lane)
outcome: Triage complete; follow-up fix task 4837 queued. Issue remains open. Sandbox bookkeeping deferred to fleet reconciliation.
dezren39 commentedon Nov 1, 2026
agent-note:v1
event: picked_up
task: 4837 (repo-health-code-pilot, repo-worker-code)
state: Claimed small_fix; issue #562 is open and the targeted PR search found no active competing implementation. Revalidating scope and preparing a dedicated laptop worktree.
dezren39 commentedon Nov 1, 2026
agent-note:v1
event: handed_off
task: 4837 (repo-health-code-pilot, repo-worker-code)
state: Implementation pushed as PR #3773; per-session raw SSH screen diffs, resize/non-view invalidation, and commit-after-write cache semantics are included. Local targeted renderer tests are running; SSH channel regression test and CI are pending.
dezren39 commentedon Nov 1, 2026
agent-note:v1
event: done
task: 4837 (repo-health-code-pilot, repo-worker-code)
outcome: PR #3773 opened with per-session incremental ANSI rendering; renderer parity/byte-savings suite passed 15 tests and the raw SSH steady-state integration test passed. Issue remains open pending PR review/merge. Sandbox backlog and sweep-log bookkeeping deferred to fleet reconciliation.