feat(tui): host-side Mission Control TUI for hm run/dev/cloud watch#2
Closed
markovejnovic wants to merge 62 commits into
Closed
feat(tui): host-side Mission Control TUI for hm run/dev/cloud watch#2markovejnovic wants to merge 62 commits into
markovejnovic wants to merge 62 commits into
Conversation
Host-side ratatui-based TUI for hm run / dev / cloud build watch, designed to be the easiest and most screenshot-worthy way to run a Harmont deployment. Mission Control layout (animated DAG + gantt timeline + log pane + summary card), tachyonfx for subtle effects, NO_COLOR / non-TTY falls back to the existing human plugin. Single protocol-level addition: hm_build_event_emit host fn so the embedded cloud plugin can drive the host TUI from its WASM watch loop without lifting that loop out of the sandbox.
Maps the 2026-05-22-tui-mission-control-design spec onto bite-sized TDD tasks across 7 phases: foundation, event model + reducer, adapters (local/dev/cloud with the new hm_build_event_emit host fn), terminal/theme/fx, widgets (insta snapshot tests), main loop + overlays, command wiring (run/dev/cloud build watch), and demo artifacts + CI tape smoke.
Also enables uuid v4+v5 features in crates/hm to support uuid_from_deploy_id (Uuid::new_v5 for deterministic deploy IDs).
- Remove dead `use std::time::Instant` import and `_instant_unused_marker` fn - Add `#[allow(clippy::map_entry)]` on `apply` with comment explaining why entry() can't replace contains_key+insert in DeployLog arm - Rewrite `cycle_focus` using `isize/usize::try_from` to eliminate unsafe numeric casts - Add `#[must_use]` to `AppState::new()` and `focused_step_id()`
- scheduler.rs: replace `Lagged(_) => continue` with `=> {}` to
silence clippy::needless_continue in the extra_event_tx forwarder.
- tui/source/local.rs: wrap BuildEvent, TuiEvent, tokio::sync::mpsc,
and Lagged in backticks in the module doc to satisfy doc_markdown.
- tui/source/local.rs: add comment on chain_idx/label placeholder in
the StepStart arm explaining the reducer fills these in.
- commands/run/local.rs: annotate the None arg to orchestrator::run
clarifying TUI wiring is handled separately.
Add tui_event_tx field to OrchestratorState so both the local bus forwarder and the new cloud-watch host fn share the same TUI sink. Implement hm_build_event_emit as a raw-bytes host fn that deserialises a BuildEvent and try_sends it non-blocking into the TUI channel. Switch the Task 2.2 bus forwarder to read its sender from state_arc rather than the standalone extra_event_tx parameter.
Adds hm_build_event_emit extern + build_event_emit safe wrapper to the SDK host module. Cloud build watch now emits BuildStart/StepQueued/ StepStart/StepLog/StepEnd/BuildEnd (and ChainFailed on cancel) instead of writing raw bytes to stderr.
Full-screen summary card rendered after BuildEnd: HARMONT wordmark via tui-big-text Quadrant pixels, pass/cache/fail counts, total duration, cache hit %, and slowest step. Includes insta snapshot test.
TTY-detect in local.rs: when stdout is a terminal, format=human, and neither --no-tui nor NO_COLOR is set, spawn the TUI alongside the orchestrator via tui::source::local::spawn(). Thread --no-tui/--no-fx flags through RunContext instead of env vars (avoids unsafe set_var under unsafe_code=deny).
These were working docs used while building the TUI. Removing now that the implementation has landed so the repo doesn't carry in-progress planning artifacts.
Hardcoded 256-color indices ignored the terminal palette and rendered poorly on light backgrounds. Swap to ANSI 16 names so the terminal's own foreground/background mapping applies.
extism wraps a host-fn Err into a wasm trap whose Display string is
just "error while executing at wasm backtrace". The actual anyhow
chain from the host fn was lost. Format the trap with {:#} so the
chain shows, and tracing::error! the host-side cause inside the
docker pull / image_exists impls before extism eats it.
Scheduler always spawned the human formatter, which writes BuildEvent lines to stdout. With the TUI on the same stdout, those lines scrolled over the rendered widgets. Gate the output_subscriber spawn on state.tui_event_tx being None — the TUI is the sink.
Two-phase plan: (1) merge hm-plugin-output-{human,json} into the hm
binary as native formatters, keeping OutputFormatter SDK trait for
external plugins; (2) rewrite hand-rolled cell_mut rendering in the
TUI widgets using ratatui's Paragraph/List/Line/Span types.
Adds output::formatters::{human,json} modules to the hm crate. The
Human struct replicates the WASM plugin's render logic with instance-
owned step_key state (no global Mutex). Json is a stub for Task A2.
Check crate::output::formatters::builtin() first so --format human and --format json pass validation without being in the registry index. The available-list in the error path now includes both registry entries and the hard-coded built-in names (sorted, deduped).
Remove crates/hm-plugin-output-human and crates/hm-plugin-output-json from the workspace now that their logic lives in crates/hm/src/output/ formatters/ (A1–A5). Workspace Cargo.toml members list updated; no workspace.dependencies entries referenced these crates.
Replace hand-rolled buf.cell_mut().set_symbol() loops in LogPane with ratatui's Paragraph + Line + Span types. Snapshot test passes unchanged.
Replace hand-rolled buf.cell_mut/set_symbol loops in graph.rs with Paragraph + Line + Span, matching the pattern established in log.rs.
Previous run hung at 6h on cargo build --release (resource starvation on the runner). Debug profile + rust-cache should keep it under 20m. continue-on-error keeps PR CI green even if vhs flakes — the smoke test stays useful as an annotation but can't block merges.
Mirrors docker CLI's precedence: DOCKER_HOST > DOCKER_CONTEXT > ~/.docker/config.json currentContext > platform default. Fixes the Docker Desktop on Linux paper cut: Desktop ships a 'desktop-linux' context pointing at ~/.docker/desktop/docker.sock, but bollard's connect_with_local_defaults only looks at /var/run/docker.sock, so `hm run` would bail on a Desktop host without an explicit DOCKER_HOST export. We now read the context the same way docker does (sha256(name) -> contexts/meta/<hash>), so Desktop just works. Remote HTTPS contexts return a clear "not supported, set DOCKER_HOST" error rather than silently downgrading — bollard's ssl feature would pull rustls+ring transitively and isn't worth it for a niche case.
The two plugin_kv tests each set XDG_CONFIG_HOME (process-global) to their own tempdir. Run in parallel — which CI does — they clobber each other's env var between set/read, producing intermittent None reads. Local repro is hard because higher parallelism on dev boxes still happens to interleave reads between the two writes; CI's lower vCPU count widened the window. Merge the two tests into one serialised body. No new dev-deps, no test-thread-counts pinning.
Contributor
Author
|
slop |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
hm run,hm dev up, andhm cloud build watchwhen stdout is a TTY. Mission Control layout: animated DAG graph, gantt timeline, focusable log pane, summary card.NO_COLOR,--no-tui) falls through to the existing WASMhumanoutput plugin unchanged.hm_build_event_emit) so the embedded cloud plugin can drive the host TUI from its WASM watch loop. Wire types andHM_PLUGIN_API_VERSIONare unchanged.--no-fxandNO_COLORdisable animation.hm runandhm dev upplus a CI smoke workflow at.github/workflows/demo.ymlthat exercises the run tape on TUI-touching PRs.The two
docs/superpowers/{specs,plans}/*-tui-mission-control*.mdfiles are deleted here — they were in-progress planning docs that were committed to main earlier in the brainstorming session and shouldn't ship in the repo.Test plan
cargo build -p harmont-clicleancargo test -p harmont-cli --lib— 92 passedcargo build -p hm-plugin-cloud --target wasm32-wasip1clean (embedded plugin rebuild)hm run --no-tuiinexamples/rustfalls back to the streaming formatter (verified during implementation)hm runin a TTY enters the Mission Control TUI and shows the summary card on completionhm dev upin a TTY (with a deploy example) enters the TUIhm cloud build watchin a TTY enters the TUI (requires staging cloud + an in-flight build)demo-tape-smokeworkflow exits cleandocs/demo/run.gifregenerated (vhs not available in the dev env; will be produced by CI or locally before merge)Follow-ups (not in this PR)
docs/demo/run.gif(and PNG) so the README embed renders. CI will produce it on the first PR touchingcrates/hm/src/tui/.examples/dev-demo/with a minimal@hm.deployregistration sodocs/demo/dev.tapecan be smoke-tested in CI.tui::run(currently 169-linetokio::select!loop) into key/mouse/render helpers — flagged by the final reviewer as worth a follow-up refactor.future_not_sendcascades 7 false positives becausetachyonfx::Shader: !Send. Consider#[allow(clippy::future_not_send)]ontui::runonce the lint behavior is confirmed acceptable.