Removes the 2026-05-15 perf audit findings that have either shipped
(see CHANGELOG) or are tracked elsewhere, and replaces them with the
remaining palette-refinement notes: generic labels for focused
actions ("Close current agent") and a higher-level concern that the
palette has grown cluttered as features were added.
Added .mise.toml pinning zig = "0.15.2" (the minimum the vendored
Ghostty commit requires) and taught the Makefile to resolve zig
through mise when available, falling back to PATH. Contributors run
`mise install` once and `make deps` just works.
Re-ran the pipeline benchmarks after rebuilding libghostty-vt with
ReleaseFast (same hardware, AMD Ryzen 7 7800X3D):
Debug ReleaseFast speedup
Pipeline 8-colour @120fps 63 fps 2030 fps 32x
Pipeline truecolor @120fps 34 fps 931 fps 27x
Emulator-only truecolor 34 fps 2051 fps 60x
7-16x headroom over 120 fps for the heaviest workload (truecolor
full-screen redraws). Static library size 33 MiB -> 13 MiB.
TODO.md baseline numbers updated to reflect post-fix throughput;
the "Debug-mode lib" finding is folded into the result it produced
rather than left as an open item.
Added a full ASCII-video benchmark suite that hammers the renderer
with 30 KiB / 70 KiB full-screen frames at 30, 60, and 120 fps
targets — both renderer-only and full-pipeline (em.Write + renderer
+ stdout). Each stream benchmark reports µs/frame, fps_ceiling, and
percent of the per-frame budget consumed.
The pipeline benchmarks revealed we were missing 120 fps by a wide
margin (190%-350% of budget at 120fps, 60-90 fps ceiling). Isolating
em.Write confirmed libghostty-vt is the bottleneck — 16-29 ms per
truecolor frame, library file at 33 MiB.
Root cause: the Makefile invoked `zig build` with no
-Doptimize, and Zig's standardOptimizeOption defaults to Debug. So
the shipped libghostty-vt was unoptimised. Fixed by pinning
ReleaseFast in the Makefile (override via GHOSTTY_VT_OPTIMIZE for
debug builds of the upstream lib).
Existing checkouts need `make clean-deps && make deps` to pick up
the rebuild.
Live metrics (--profile):
- New metricsTracker instruments OnPTYOut, viewport renderer,
stdout writes, libghostty-vt Write/Title CGO calls, sidebar /
tabbar / status draws (with cache-hit accounting), snapshot
replays, and the chrome ticker (so we can see ticker fires that
did nothing).
- Writes metrics.jsonl (one snapshot per second) and metrics.json
+ summary.txt on exit, alongside the existing pprof files.
- All record* methods are nil-safe so disabled paths pay only a
cheap nil check; counters are atomic so the per-PTY-chunk hot
path stays lock-free.
Benchmark suite (go test -bench=.):
- Three workload fixtures — plain ASCII, SGR-styled lines, and a
ratatui-style cursor-shuffling burst — plus a containsOSC
microbenchmark. Reports ns/op, MB/s, allocs/op, B/op.
- Initial baseline numbers added to TODO under the perf-audit
section, alongside two new findings (renderer allocs ~1 per 4
bytes on styled chunks; styled throughput tops out near
90 MB/s) those benchmarks surfaced.
Codebase sweep for perf issues outside the per-PTY-chunk path that
recent CHANGELOG work already covered. Ten findings under a new
"Perf Audit (auto-generated)" section in TODO.md — anchored to
file:line, classified MEDIUM/LOW, with a sketched fix per entry.
None landed as code changes; review pending.
- Add --debug[=DIR] / --profile[=DIR] flags that write run artefacts
(patterm.log, events.jsonl, per-child raw PTY captures, CPU + heap
+ goroutine pprof) to a dir without polluting stdout/stderr.
- Strengthen vendor-TUI orientation in three places (MCP
initialize.instructions, the spawn_agent tool description, and
help('spawning')) to head off codex's habits of poking the Unix
socket via perl and shelling out to launch peers — both bypass
caller identity and produce orphaned top-level tabs.
- Fix click-and-drag text selection from alt-screen TUIs. Host SGR
mouse reporting now follows the focused child's screen side
instead of being permanently armed; alt-screen TUIs that need
mouse re-enable it themselves and the toggle is forwarded.
- Move drawSidebar() off the per-PTY-chunk hot path. Long claude
session resume was paying a full sidebar rebuild for every
scrolled chunk; the chrome ticker now drains a dirty flag at 60 Hz.
- Gate the per-chunk Title() CGO poll on a containsOSC scan so
codex/ratatui's many SGR-only chunks no longer pay a CGO call each.
- Palette's per-child "Kill <name>" action is now labelled "Close <name>"
(action kind unchanged; still SIGTERM). Matches the existing "Close
agent: …" context entry and reads less violent for a graceful term.
- New "New Terminal" palette entry spawns a bare interactive $SHELL pane
via LaunchTerminal (kind=terminal). Replaces the default "shell"
process preset that was seeded on first run.
- Exited KindTerminal entries are now dropped from the session in
reapChild — terminals have no restart path, so leaving them behind as
greyed rows in the Processes sidebar was just clutter. processList
also filters defensively.
The stdin loop's scratchpad-input branch ran before the palette branch
and silently dropped every byte except a handful of app-level chords,
so palette typing and Esc never reached the palette while a pad was
focused. Skip the pad-input branch whenever st.palette != nil.
closePalette also called repaintFocused() on cancel / no-op action
paths, which paints the empty focused-child slot (focusedID == "" while
a pad is focused) and leaves the palette's top border drawn over the
pad. Route those branches through a restoreView helper that picks
repaintFocusedPad when a pad is focused.
Switching from a pad to a child via the palette now clears the pad
focus and wipes the viewport, matching focusProcess's pad-exit path.
Adds a harness scenario (palette_over_scratchpad) that opens a pad,
opens the palette, types a query, and verifies that Esc leaves the
pad correctly repainted with no palette chrome lingering.
Bundles the in-flight work into the first tagged release. See
CHANGELOG.md `[0.0.1] - 2026-05-14` for the full per-change list.
Highlights:
- Sidebar / chrome stability: clamp absolute cursor positioning and
printable bytes to the viewport so long-running TUIs (claude, codex)
can't spray into the right rail; bound tab bar's row clear to the
viewport width so the rail isn't wiped on every tab redraw; flag
scroll escapes (RI/IND/NEL/SU/SD/IL/DL) and clamp `CSI 0/1/2 J`/`K`
to viewport columns.
- Palette: "Spawn process…" form, macros (`sw `, `k `, `sp `), kill
entries mark the focused tab, dead agents drop out of the switch
list.
- Sidebar: split into Processes (session-wide) + Agent Tree
(per-active-agent) sections; relaunch indicator; Ctrl+W/S walks the
combined list, Ctrl+A/D steps tabs.
- MCP: protocol handshake (`initialize`, `tools/list`, `tools/call`,
`ping`), `mcp_injection.kind = cli_override / config_env` so codex
and opencode pick up the server with no file writes, `lifecycle`
help topic and tool-description cleanup-duty pointers.
- Lifecycle: orchestrator-spawned children cascade-killed when the
parent dies; orchestrator-injected prompts end with CR + delayed
Enter so claude submits cleanly.
Add a `lifecycle` help topic spelling out that the caller owns the
processes it spawns and should `close_process` when a sub-agent or
spawned child is no longer needed. The `spawn_agent` and `spawn_process`
descriptions advertised via `tools/list` now restate the same duty
inline (with a pointer to `help('lifecycle')`), so vendor TUIs see the
expectation at the moment they reach for the tool. The `spawning` topic
and `topics` index cross-reference the new content.
Bundles two already-staged improvements that fall in the same area:
- OnChildSpawned primes the snapshot-replay budget for new panes so
diff-based vendor TUIs come up clean without a manual Ctrl+W/Ctrl+S
refresh.
- TODO drops the three items now actioned (prompt-injection preface,
agent cleanup duty, opencode→claude view corruption) and keeps the
unicode `<?>` entry with the investigation notes.
This batches the in-flight [Unreleased] block from CHANGELOG.md into a
single commit. Highlights:
- Real MCP protocol layer (initialize / tools/list / tools/call) so
vendor MCP clients can complete the handshake against the per-PID
socket. Legacy direct-dispatch preserved for the harness.
- New mcp_injection kinds — cli_override for codex, config_env for
opencode — joining the existing env-var and config_file paths so
patterm can slot into more agents without touching their real
config or auth.
- Ctrl+A/D and Ctrl+W/S focus navigation across tabs and intra-tab
process lists, recognised in legacy / kitty CSI u / xterm
modifyOtherKeys encodings.
- Palette macros (sw / k / sp ) and reordering so open sessions
surface above spawn-new entries.
- Two-row tab bar, sidebar/tabbar/status chrome cache, viewport-wipe
on agent spawn, CR-terminated orchestrator injections, and split-
Enter PTY writes so paste-detecting TUIs see Enter as a key event.
Also fixes the bug logged in TODO: claude's Ctrl+O tool-call expansion
emits CSI 0 J, which the viewport renderer was forwarding verbatim —
wiping the sidebar to the right of the cursor and leaving the chrome
cache convinced nothing had changed. CSI 0 J and CSI 1 J are now
translated into per-row ECH sequences clamped to the viewport, same
as CSI 2 J and CSI K already were.
Agent guides (CLAUDE.md / AGENTS.md) now spell out the
TODO->CHANGELOG workflow so completed items land in the changelog
rather than as ticked entries left behind in TODO.
Module renamed github.com/harrybrwn/patterm → github.com/hjbdev/patterm
across imports.
Chrome:
- Palette redrawn with rounded box-drawing borders, accent left-bar
for the selected item, dim hints, and a separator-aware footer.
- Tab bar grew from 1 row to 3: labels with breathing room, a dim
argv subtitle truncated to each tab's width, and an accent thick
underline for the focused tab with a faint divider extending across
the rest of the host width. Layout, viewport-renderer, and screen-
renderer tests updated for the new mainTop.
- Sidebar reuses the same palette: accent section headers, `▎`
selection marker, `●`/`○` status glyphs, dim previews.
- Shared SGR constants moved into internal/app/style.go.
Palette input:
- Adjacent duplicate arrow events (legacy `\x1b[B` + kitty
`\x1b[57353u` for one keypress, or two of the same form) are now
collapsed via peekArrowEvent + chunk-level dedupe in processStdin.
- On open, push `\x1b[>0u` onto the host's kitty keyboard stack so
palette input is in plain legacy mode regardless of what the child
pushed (codex/ratatui pushes its own flags which had been leaking
to the host). Popped on close.
Tab-switch repaint (repaintFocused):
- Use the emulator's SerializeVT bytes (with SGR / cursor / DECSTBM
/ tabstops) instead of plain text, fed through the per-focused
viewport renderer so the shifter translates row positions.
- Prelude resets host SGR / DECOM / DECSTBM (pinned to viewport) /
cursor visibility before the replay, so leftover modes from the
previously-focused child don't distort the new snapshot.
- Re-emit the saved cursor as a child-space CUP after the
serialized bytes so the host cursor lands at the emulator's
actual position (overriding DECSTBM's home side-effect and the
tabstop-setup CHA sequences) AND the renderer's vr.row/vr.col
get re-synced via trackCSI.
- cursorShifter now carries childRows and rewrites empty
`\x1b[r` to `\x1b[<mainTop>;<mainBottom>r` (host coords) — the
default (1,1) shifted to (4,4) was producing a one-row scrolling
region that scroll-exploded the replay.
- After the snapshot lands, nudge the focused child with a one-row
PTY winsize toggle so the kernel emits SIGWINCH and ratatui-style
TUIs throw away their diff state and emit a fresh frame.
Codex still renders incorrectly after a focus switch; see TODO.md
"Switch-back render divergence" for the deep investigation handoff.
