max / makenotwork

143.0 KB · 3437 lines History Blame Raw

1	//! Rhai recipe engine + host-function API.
2	//!
3	//! A `(app, target)` resolves to a `.rhai` recipe composed from a shared step
4	//! vocabulary. The daemon embeds Rhai and registers the host functions recipes
5	//! call; the recipe is the orchestration, the host functions are the
6	//! privileged primitives (run a command, read a secret, collect artifacts,
7	//! publish). Recipes are otherwise sandboxed — no arbitrary FS/network except
8	//! through these functions — matching the Balanced Breakfast plugin model.
9	//!
10	//! Rhai is synchronous; the engine runs each recipe on a blocking thread
11	//! (`spawn_blocking`, see [`crate::runner`]) and host functions bridge to async
12	//! work via `Handle::block_on`. That is sound only off a runtime worker thread,
13	//! which `spawn_blocking` guarantees.
14
15	use crate::config::Config;
16	use crate::domain::{AppId, Status, Step, StepRunId, Target, Version};
17	use crate::events::{self, Event, EventTx};
18	use crate::ota::{OtaRegistry, PublishAuthority, Release};
19	use crate::state::ExecutorMap;
20	use crate::topology::{DeployTarget, Kind};
21	use anyhow::{Context as _, Result};
22	use ops_core::live_log::LiveLog;
23	use ops_exec::{Action, Executor, ObserveKind, Step as OpStep, SyncOpts};
24	use rhai::{Engine, EvalAltResult, Map};
25	use sha2::{Digest, Sha256};
26	use sqlx::SqlitePool;
27	use std::collections::HashMap;
28	use std::path::{Path, PathBuf};
29	use std::sync::atomic::{AtomicBool, Ordering};
30	use std::sync::{Arc, Mutex};
31	use tokio::runtime::Handle;
32	use tokio::sync::Mutex as AsyncMutex;
33
34	/// The capability label for a command, derived from the open recipe step. A
35	/// recipe's `sh("mbp", …)` under `step("sign")` becomes an `Action::Sign`, gated
36	/// by the mac host's `sign` grant — so recipes stay unchanged while every command
37	/// is capability-checked at its transport. `Verify` is read-only (an observe).
38	/// The capability a step's commands are gated on.
39	///
40	/// `Verify` depends on what is being released, which is the one place this is
41	/// not a property of the step alone. An app's verify is a Gatekeeper check on a
42	/// signed bundle, and the `gatekeeper` observe is granted implicitly to hosts
43	/// that can `sign` (`CapabilitySet::from_tokens`) precisely so that pairing
44	/// holds. A library's verify is a crate preflight: it runs `cargo` on the build
45	/// host and asks the registry a question. Gating that on Gatekeeper asks a Linux
46	/// host for a macOS code-signing capability it can never honestly hold, and the
47	/// only way to satisfy it would be to declare the capability falsely.
48	fn action_for(step: Step, kind: Kind) -> Action {
49	match step {
50	Step::Checkout \| Step::Prebuild \| Step::Build => Action::Build,
51	Step::Sign => Action::Sign,
52	Step::Notarize => Action::Notarize,
53	Step::Staple => Action::Staple,
54	Step::Package => Action::Package,
55	Step::Verify => match kind {
56	Kind::App => Action::Observe(ObserveKind::Custom("gatekeeper".into())),
57	// Running the build toolchain to inspect a crate or a service
58	// binary, which is what `build` means on a host. Neither has a
59	// bundle for Gatekeeper to have an opinion about.
60	Kind::Library \| Kind::Service => Action::Build,
61	},
62	// Publish/Collect/Handoff run on the daemon, not through a host
63	// executor; this label only applies if a recipe runs a bare `sh` while
64	// one is open. `handoff` is the daemon's own post-recipe motion and no
65	// recipe should open it at all — naming it here costs nothing and beats
66	// a wildcard that would silently absorb the next step somebody adds.
67	Step::Publish \| Step::Collect \| Step::Handoff => Action::Package,
68	// The one step that dispatches to a host OUTSIDE the build topology.
69	// Every command a recipe runs while `deploy` is open — the install, the
70	// restart, the health assertion — carries this action, so it reaches the
71	// service host only through the deploy grant and reaches a build host
72	// not at all (no build host is granted `deploy`).
73	Step::Deploy => Action::Deploy,
74	}
75	}
76
77	/// The currently-open step within a recipe run: its DB row id, which step it
78	/// is, and the live-log sink that `sh`/`log` stream into.
79	struct StepState {
80	run_id: StepRunId,
81	step: Step,
82	log: Arc<AsyncMutex<LiveLog>>,
83	/// Set when something in the step recorded a hard failure the recipe did
84	/// not abort on (e.g. `verify_gatekeeper` rejected the artifact but the
85	/// recipe ignored the bool). Forces the step's recorded status to `Failed`
86	/// and bars `publish` (the step-success ledger).
87	failed: bool,
88	/// Wall-clock deadline for this step. A command that runs past it fails the
89	/// step (and unwinds the recipe) rather than wedging under the old
90	/// whole-build guillotine, which a legitimate 5-target fan-out plus notary
91	/// queueing could trip — mismarking every target failed while the blocking
92	/// recipe bodies kept signing.
93	deadline: std::time::Instant,
94	}
95
96	/// Per-step wall-clock budget: a generous ceiling that catches a wedged command
97	/// (a hung ssh, a stuck notary poll) without killing legitimately slow work.
98	/// The old design bounded the whole build at 2h; this bounds each step so one
99	/// slow step can't be blamed on another and a fan-out of slow-but-fine targets
100	/// isn't guillotined. `Config::step_timeout_secs` overrides these per-kind
101	/// defaults for every step; see [`RecipeCtx::step_budget`].
102	fn default_step_budget(step: Step) -> std::time::Duration {
103	use std::time::Duration;
104	let mins = match step {
105	Step::Checkout => 10,
106	// clippy + full test suite, cold, on a workspace.
107	Step::Prebuild => 45,
108	// cargo tauri build, cold, universal bundles.
109	Step::Build => 90,
110	Step::Sign => 15,
111	// Apple's notary queue + this step's bounded retries.
112	Step::Notarize => 60,
113	Step::Staple => 10,
114	Step::Package => 30,
115	Step::Verify => 10,
116	// rsync of multi-GiB artifacts off the build host.
117	Step::Collect => 30,
118	Step::Publish => 20,
119	// A binary push, an install, a unit restart, and a health poll. Minutes
120	// of work; the ceiling is for a wedged transport, not slow work.
121	Step::Deploy => 15,
122	// Unused by any recipe — the daemon runs the handoff itself, outside a
123	// step's clock — and matched to `collect`, since it moves the same
124	// bytes the same way and would wedge for the same reasons.
125	Step::Handoff => 30,
126	};
127	Duration::from_secs(mins * 60)
128	}
129
130	/// Where a service's binary is staged on the host that will run it, before the
131	/// privileged installer moves it into place.
132	///
133	/// A fixed, unguessable-by-accident path rather than a recipe-chosen one,
134	/// because the installer refuses any source outside it. That refusal is the
135	/// only thing standing between the NOPASSWD sudo grant and `install`-as-root to
136	/// an arbitrary path, so both ends have to name the same constant. `/var/tmp`
137	/// rather than `/tmp` so a staged binary survives a `PrivateTmp` unit and a
138	/// systemd tmpfiles sweep between staging and install.
139	pub const DEPLOY_STAGING_ROOT: &str = "/var/tmp/bento-deploy";
140
141	/// Everything a recipe's host functions need, shared (Arc) into each closure.
142	pub struct RecipeCtx {
143	pub app: AppId,
144	pub version: Version,
145	pub target: Target,
146	/// Name of the host this target builds on (resolved from the topology by the
147	/// runner). Recipes read it via `build_host()` so one per-platform recipe can
148	/// dispatch to the right host across arches (linux x86_64 -> fw13, aarch64 ->
149	/// astra) without hard-coding a host name.
150	pub build_host: String,
151	/// The build host's SSH destination (topology `ssh`), as opposed to its
152	/// name. Deploy compares it against the service host's to tell "build and
153	/// run on the same box" from "two hosts that need a transfer" — a question
154	/// the host NAMES cannot answer, since a build host and a deploy
155	/// destination are declared in different files and need not agree on one.
156	pub build_host_ssh: String,
157	/// This release's git tag, rendered from the app's `tag_format`. Held here
158	/// rather than derived from the version because a repo holding several
159	/// products spells it per product (`pom-v0.4.1`), and the recipe, the
160	/// preflight barrier and the failure message all have to agree on it.
161	pub tag: String,
162	/// The app's default checkout path (topology `repo`, `~`-prefixed). Read it
163	/// through [`RecipeCtx::repo_for`] for anything that runs on a build host;
164	/// this field alone is the daemon-local answer.
165	pub repo: String,
166	/// Per-host checkout overrides (topology `repo_by_host`). Empty for every app
167	/// that has not declared one, which is all of them but the Windows-shipping
168	/// ones.
169	pub repo_by_host: HashMap<String, String>,
170	/// Cargo features this app's release builds enable (topology `features`).
171	/// Recipes read it via `feature_flags()`.
172	pub features: Vec<String>,
173	/// App or library. Decides which capability a `verify` step is gated on;
174	/// see [`action_for`].
175	pub kind: Kind,
176	pub target_run_id: i64,
177	/// Capability-scoped executor per build host. Recipe commands dispatch
178	/// through these — the transport (local / ssh / in-session agent) and the
179	/// capability gate are the executor's, not the engine's.
180	pub execs: Arc<ExecutorMap>,
181	/// Sync transport per build host, used only by `collect` to pull artifacts
182	/// back. Never the agent, even for an agent host — see `state::build_sync`.
183	/// Not an execution path.
184	pub syncs: Arc<ExecutorMap>,
185	/// Where this target installs, for a `kind = "service"` app. `None` for an
186	/// app or a library, which makes every deploy host function fail with that
187	/// as the reason rather than with a missing-host error.
188	///
189	/// The runner resolves it from the app manifest's `[[deploy]]` table and
190	/// registers its executor into `execs` under the destination's host string,
191	/// so `sh_ok(deploy_host(), ...)` reaches the service host through the same
192	/// capability gate as everything else.
193	pub deploy: Option<DeployTarget>,
194	pub pool: SqlitePool,
195	pub events: EventTx,
196	pub cfg: Arc<Config>,
197	pub ota: Arc<OtaRegistry>,
198	pub rt: Handle,
199	current: Mutex<Option<StepState>>,
200	/// Gatekeeper verdict recorded by `verify_gatekeeper`: `None` = never run,
201	/// `Some(false)` = ran and rejected, `Some(true)` = accepted. `publish`
202	/// requires `Some(true)` for a macOS/iOS artifact (the proof it is signed +
203	/// notarized).
204	gatekeeper_ok: Mutex<Option<bool>>,
205	/// Steps finalized as `Failed` during this run. A non-empty ledger bars
206	/// `publish` — an artifact is never shipped after a step failed, even if the
207	/// recipe ignored the failure and ran on.
208	failed_steps: Mutex<Vec<Step>>,
209	/// Set when a newer build supersedes this run. Checked at step boundaries
210	/// and before publish so a superseded recipe stops promptly rather than
211	/// running to completion (the blocking Rhai body can't be `abort()`ed).
212	cancel: Arc<AtomicBool>,
213	/// The all-targets-green publish gate (topology `require_all_targets`).
214	/// `Some(declared)` ⇒ `publish` refuses unless every one of `declared` has a
215	/// successful latest run for this `(app, version)`. `None` ⇒ gate off, each
216	/// target publishes independently.
217	all_green_required: Option<Vec<Target>>,
218	/// sha256 of each artifact hashed at `collect`, keyed by file name. `publish`
219	/// reads it to record `releases.artifact_hash` for the bytes it ships, so the
220	/// hash is the one computed when the artifact landed rather than a re-read
221	/// that could see a different file. Absent ⇒ `publish` hashes on demand.
222	artifact_hashes: Mutex<HashMap<String, String>>,
223	}
224
225	impl RecipeCtx {
226	/// Versions of `name` already on crates.io. A network failure yields an
227	/// empty list: preflight then cannot claim a version is a duplicate, and
228	/// `cargo publish` still refuses one, so the check degrades to advisory
229	/// rather than blocking a release on registry availability.
230	fn published_versions(name: &str) -> Vec<String> {
231	let url = format!("https://crates.io/api/v1/crates/{name}");
232	let Ok(out) = std::process::Command::new("curl")
233	.args([
234	"-sS",
235	"--max-time",
236	"15",
237	"-H",
238	"User-Agent: bento-preflight",
239	&url,
240	])
241	.output()
242	else {
243	return Vec::new();
244	};
245	let Ok(v) = serde_json::from_slice::<serde_json::Value>(&out.stdout) else {
246	return Vec::new();
247	};
248	v.get("versions")
249	.and_then(\|x\| x.as_array())
250	.map(\|a\| {
251	a.iter()
252	.filter_map(\|x\| x.get("num").and_then(\|n\| n.as_str()).map(str::to_string))
253	.collect()
254	})
255	.unwrap_or_default()
256	}
257
258	#[allow(clippy::too_many_arguments)]
259	pub fn new(
260	app: AppId,
261	version: Version,
262	target: Target,
263	build_host: String,
264	build_host_ssh: String,
265	tag: String,
266	repo: String,
267	features: Vec<String>,
268	kind: Kind,
269	target_run_id: i64,
270	execs: Arc<ExecutorMap>,
271	syncs: Arc<ExecutorMap>,
272	deploy: Option<DeployTarget>,
273	pool: SqlitePool,
274	events: EventTx,
275	cfg: Arc<Config>,
276	ota: Arc<OtaRegistry>,
277	rt: Handle,
278	cancel: Arc<AtomicBool>,
279	all_green_required: Option<Vec<Target>>,
280	) -> Self {
281	Self {
282	app,
283	version,
284	target,
285	build_host,
286	build_host_ssh,
287	tag,
288	repo,
289	repo_by_host: HashMap::new(),
290	features,
291	kind,
292	target_run_id,
293	execs,
294	syncs,
295	deploy,
296	pool,
297	events,
298	cfg,
299	ota,
300	rt,
301	current: Mutex::new(None),
302	gatekeeper_ok: Mutex::new(None),
303	failed_steps: Mutex::new(Vec::new()),
304	cancel,
305	all_green_required,
306	artifact_hashes: Mutex::new(HashMap::new()),
307	}
308	}
309
310	/// Declare the app's per-host checkout overrides (topology `repo_by_host`).
311	///
312	/// Separate from `new` because it is empty for every app that has not opted
313	/// in, and `new` already carries twenty arguments no test wants a
314	/// twenty-first of.
315	#[must_use]
316	pub fn with_repo_by_host(mut self, repo_by_host: HashMap<String, String>) -> Self {
317	self.repo_by_host = repo_by_host;
318	self
319	}
320
321	/// Where this app is checked out on `host` (topology `AppConfig::repo_for`).
322	///
323	/// Every git command a recipe or the engine runs on a build host goes
324	/// through this. The bare `repo` field is the daemon-local path.
325	pub fn repo_for(&self, host: &str) -> &str {
326	self.repo_by_host
327	.get(host)
328	.map_or(self.repo.as_str(), String::as_str)
329	}
330
331	/// Whether a newer build has superseded this run.
332	fn is_cancelled(&self) -> bool {
333	self.cancel.load(Ordering::SeqCst)
334	}
335
336	fn now() -> String {
337	chrono::Utc::now().to_rfc3339()
338	}
339
340	/// `<logs_root>/<app>/<version>/<target-with-slash-as-dash>/<step>.<run_id>.log`.
341	///
342	/// The run id is in the filename because the rest of the key is not unique:
343	/// re-running an app at a version it already built (a retry, or a rebuild of
344	/// an already-published release) reopens the same path, and the old file
345	/// appended to shows two runs' output with nothing marking the boundary. The
346	/// step ledger records a run id per step, so keying the file on it makes a
347	/// ledger row resolve to exactly one file and keeps the earlier run readable.
348	fn log_path(&self, step: Step, run_id: StepRunId) -> PathBuf {
349	self.log_dir()
350	.join(format!("{}.{}.log", step.as_str(), run_id.0))
351	}
352
353	/// `<logs_root>/<app>/<version>/<target-with-slash-as-dash>/`.
354	fn log_dir(&self) -> PathBuf {
355	let target_dir = self.target.to_string().replace('/', "-");
356	self.cfg
357	.logs_root
358	.join(self.app.as_str())
359	.join(self.version.to_string())
360	.join(target_dir)
361	}
362
363	/// Close the previous step (as `Ok`), open a new one: insert its DB row,
364	/// open a live log whose chunks broadcast `StepLogChunk`, emit `StepStart`.
365	fn begin_step(self: &Arc<Self>, step: Step) -> Result<()> {
366	anyhow::ensure!(
367	!self.is_cancelled(),
368	"build superseded by a newer request; aborting before `{}`",
369	step.as_str()
370	);
371	self.finish_step(Status::Ok)?;
372	let me = self.clone();
373	let started = Self::now();
374	let started_for_header = started.clone();
375	let run_id = self.rt.block_on(async move {
376	// The step row and the target's current_step pointer are one logical
377	// state — write them atomically so a failure can't leave a `running`
378	// step row while current_step still names the previous step.
379	let mut tx = me.pool.begin().await.context("begin step tx")?;
380	// log_ref names the run id, which only exists once the row does, so
381	// the path is written back inside the same transaction rather than
382	// guessed beforehand.
383	let id: i64 = sqlx::query_scalar(
384	"INSERT INTO step_runs (target_run_id, step, status, started_at)
385	VALUES (?, ?, 'running', ?) RETURNING id",
386	)
387	.bind(me.target_run_id)
388	.bind(step.as_str())
389	.bind(&started)
390	.fetch_one(&mut *tx)
391	.await
392	.context("insert step_run")?;
393	let log_ref = me
394	.log_path(step, StepRunId(id))
395	.to_string_lossy()
396	.into_owned();
397	sqlx::query("UPDATE step_runs SET log_ref = ? WHERE id = ?")
398	.bind(&log_ref)
399	.bind(id)
400	.execute(&mut *tx)
401	.await
402	.context("set step_run log_ref")?;
403	sqlx::query("UPDATE target_runs SET current_step = ? WHERE id = ?")
404	.bind(step.as_str())
405	.bind(me.target_run_id)
406	.execute(&mut *tx)
407	.await
408	.context("update current_step")?;
409	tx.commit().await.context("commit step tx")?;
410	anyhow::Ok(StepRunId(id))
411	})?;
412
413	// Live log: each chunk fans out as a StepLogChunk event keyed by run_id.
414	let events = self.events.clone();
415	let cb_run_id = run_id;
416	let mut log = self.rt.block_on(LiveLog::open(
417	self.log_path(step, run_id),
418	Box::new(move \|seq, text\| {
419	events::emit(
420	&events,
421	Event::StepLogChunk {
422	run_id: cb_run_id,
423	seq,
424	text: text.to_string(),
425	},
426	);
427	}),
428	));
429
430	events::emit(
431	&self.events,
432	Event::StepStart {
433	run_id,
434	app: self.app.clone(),
435	version: self.version.clone(),
436	target: self.target,
437	step,
438	},
439	);
440
441	// A log that names its own run: reading one tells you which ledger row
442	// it belongs to without going back to the DB, and a file that somehow
443	// does get appended to still shows where the second run began. Written
444	// after `StepStart` so its chunk event cannot precede the step it
445	// belongs to.
446	let header = format!(
447	"=== bento {app} {version} {target} step={step} run_id={run_id} started={started_for_header} ===\n",
448	app = self.app.as_str(),
449	version = self.version,
450	target = self.target,
451	step = step.as_str(),
452	);
453	self.rt.block_on(async {
454	use ops_core::remote::LogSink as _;
455	log.write_chunk(header.as_bytes()).await;
456	});
457
458	*self.current.lock().unwrap() = Some(StepState {
459	run_id,
460	step,
461	log: Arc::new(AsyncMutex::new(log)),
462	failed: false,
463	deadline: std::time::Instant::now() + self.step_budget(step),
464	});
465	Ok(())
466	}
467
468	/// This build's budget for `step`: the `Config` override if set, else the
469	/// per-kind default.
470	fn step_budget(&self, step: Step) -> std::time::Duration {
471	self.cfg
472	.step_timeout_secs
473	.map_or_else(\|\| default_step_budget(step), std::time::Duration::from_secs)
474	}
475
476	/// The current step's wall-clock deadline (or a default if no step is open,
477	/// which only happens before the first `step()` — commands then run under an
478	/// implicit `Build` step opened by `ensure_step`).
479	fn step_deadline(&self) -> std::time::Instant {
480	self.current.lock().unwrap().as_ref().map_or_else(
481	\|\| std::time::Instant::now() + self.step_budget(Step::Build),
482	\|s\| s.deadline,
483	)
484	}
485
486	/// Drive `fut` on the runtime, but stop early on two conditions the recipe
487	/// bodies otherwise cannot observe (they run synchronously on a blocking
488	/// thread): the current step's deadline, and supersession by a newer build.
489	/// Either turns into an error that fails the step and unwinds the recipe, so
490	/// a wedged command no longer runs unbounded and a superseded build stops
491	/// mid-step instead of only at the next step boundary.
492	fn run_bounded<F, T>(&self, what: &str, fut: F) -> Result<T>
493	where
494	F: std::future::Future<Output = Result<T>>,
495	{
496	let deadline = self.step_deadline();
497	let cancel = self.cancel.clone();
498	self.rt.block_on(async move {
499	tokio::pin!(fut);
500	let watch = async {
501	// Poll the cooperative cancel flag; the finalizer and a
502	// superseding build both set it. Cheap next to a build step.
503	while !cancel.load(Ordering::SeqCst) {
504	tokio::time::sleep(std::time::Duration::from_millis(250)).await;
505	}
506	};
507	tokio::select! {
508	r = &mut fut => r,
509	() = tokio::time::sleep_until(deadline.into()) => {
510	Err(anyhow::anyhow!("`{what}` exceeded its per-step deadline"))
511	}
512	() = watch => {
513	Err(anyhow::anyhow!("build superseded by a newer request; aborting `{what}`"))
514	}
515	}
516	})
517	}
518
519	/// Flag the currently-open step as failed (no-op if none is open). Forces
520	/// its recorded status to `Failed` at `finish_step` and adds it to the
521	/// publish-barring ledger, even though the recipe kept running.
522	fn fail_current_step(&self) {
523	if let Some(st) = self.current.lock().unwrap().as_mut() {
524	st.failed = true;
525	}
526	}
527
528	/// Finalize the open step (if any): close its log, stamp the DB row, emit
529	/// `StepDone`. Idempotent when no step is open. A step flagged via
530	/// [`fail_current_step`] is recorded `Failed` regardless of the requested
531	/// status, and added to the ledger `publish` consults.
532	pub fn finish_step(self: &Arc<Self>, status: Status) -> Result<()> {
533	let st = self.current.lock().unwrap().take();
534	let Some(st) = st else { return Ok(()) };
535	let status = if st.failed { Status::Failed } else { status };
536	if status == Status::Failed {
537	self.failed_steps.lock().unwrap().push(st.step);
538	}
539	let me = self.clone();
540	self.rt.block_on(async move {
541	// Drop all log refs so the sink can be owned + flushed.
542	if let Ok(m) = Arc::try_unwrap(st.log) {
543	m.into_inner().close().await;
544	}
545	if let Err(e) = sqlx::query(
546	"UPDATE step_runs SET status = ?, finished_at = ? WHERE id = ?",
547	)
548	.bind(status.as_str())
549	.bind(Self::now())
550	.bind(st.run_id.0)
551	.execute(&me.pool)
552	.await
553	{
554	tracing::error!(step = st.step.as_str(), error = %e, "could not stamp step_run status");
555	}
556	});
557	events::emit(
558	&self.events,
559	Event::StepDone {
560	run_id: st.run_id,
561	app: self.app.clone(),
562	target: self.target,
563	step: st.step,
564	status,
565	},
566	);
567	Ok(())
568	}
569
570	/// Ensure a step is open; default to `Build` if a recipe runs a command
571	/// before declaring one.
572	fn ensure_step(self: &Arc<Self>) -> Result<Arc<AsyncMutex<LiveLog>>> {
573	if self.current.lock().unwrap().is_none() {
574	self.begin_step(Step::Build)?;
575	}
576	Ok(self.current.lock().unwrap().as_ref().unwrap().log.clone())
577	}
578
579	/// The step currently open, or `Build` as a default for failure
580	/// attribution before any step was declared.
581	pub fn current_step(&self) -> Step {
582	self.current
583	.lock()
584	.unwrap()
585	.as_ref()
586	.map_or(Step::Build, \|s\| s.step)
587	}
588
589	/// The capability-scoped executor for `name`, or an error if the host isn't
590	/// in the topology.
591	fn exec(&self, name: &str) -> Result<Arc<dyn ops_exec::Executor>> {
592	self.execs
593	.get(name)
594	.cloned()
595	.ok_or_else(\|\| anyhow::anyhow!("unknown build host `{name}` (not in topology)"))
596	}
597
598	/// The ssh string for `name` (for `collect`'s remote scp source).
599	/// The transport that moves artifacts off `name`. Distinct from
600	/// [`RecipeCtx::exec_for`]'s executor: an agent host signs over `AgentRpc`
601	/// but is collected from over ssh (`state::build_sync`).
602	fn host_sync(&self, name: &str) -> Result<Arc<dyn Executor>> {
603	self.syncs
604	.get(name)
605	.cloned()
606	.ok_or_else(\|\| anyhow::anyhow!("unknown build host `{name}` (not in topology)"))
607	}
608
609	/// Run `cmd` on `host` through its capability-scoped executor, streaming into
610	/// the current step's log. The command's [`Action`] is derived from the open
611	/// step (see [`action_for`]) and gated at the transport before dispatch — so a
612	/// `build` step on a host without the `build` grant is denied, and the macOS
613	/// sign steps ride the in-session `AgentRpc` transport automatically. Returns
614	/// exit code + a tail of stdout for the recipe to branch on.
615	fn run(self: &Arc<Self>, host: &str, cmd: &str) -> Result<(i32, String)> {
616	// The service host is addressed as a service host whatever step is open.
617	// Deriving the action from the step is right for a build host, where the
618	// step IS the work; on a service host it would ask for `build` during a
619	// `verify` and be denied for a reason unrelated to what was attempted.
620	let action = match &self.deploy {
621	Some(d) if d.host == host => Action::Deploy,
622	_ => action_for(self.current_step(), self.kind),
623	};
624	self.run_as(host, cmd, action)
625	}
626
627	/// `run`, with the [`Action`] stated rather than resolved. Used where the
628	/// caller already knows which plane it is on.
629	fn run_as(self: &Arc<Self>, host: &str, cmd: &str, action: Action) -> Result<(i32, String)> {
630	let sink = self.ensure_step()?;
631	let exec = self.exec(host)?;
632	let cur = self.current_step();
633	let step = OpStep::shell(action, cmd.to_string());
634	// Echo the command before running it. Without this a log says what
635	// happened but not what was asked, and a gate that prints nothing when
636	// it passes (`cargo fmt --all --check`) is indistinguishable from a gate
637	// that never ran.
638	let echo = format!("$ [{host}] {cmd}\n");
639	// Bounded by the step's deadline and interruptible on supersession, so a
640	// hung command fails its step instead of running unbounded, and a
641	// superseded build stops mid-step rather than only at the next boundary.
642	let label = format!("{cur} command on `{host}`");
643	let out = self.run_bounded(&label, async move {
644	use ops_core::remote::LogSink as _;
645	let mut guard = sink.lock().await;
646	guard.write_chunk(echo.as_bytes()).await;
647	exec.run_streaming(&step, &mut *guard).await
648	})?;
649	let code = out.status.code().unwrap_or(-1);
650	let stdout = String::from_utf8_lossy(&out.stdout);
651	let tail: String = stdout
652	.chars()
653	.rev()
654	.take(2000)
655	.collect::<Vec<_>>()
656	.into_iter()
657	.rev()
658	.collect();
659	Ok((code, tail))
660	}
661
662	/// Pin `host` to the release tag `v<version>` and return the commit it now
663	/// has checked out. Fetch + checkout stream into the current step's log;
664	/// the sha comes from a separate `rev-parse` so its stdout is only the sha.
665	fn checkout_sha(self: &Arc<Self>, host: &str) -> Result<String> {
666	// Every command below runs ON `host`, so the path is that host's, not the
667	// daemon's. Windows is why: its checkout is at `C:/Users/me/Code/...`.
668	let repo = self.repo_for(host).to_string();
669	// A failing mirror is not a failing release: fetch is advisory, and only
670	// the checkout decides. Its output still streams into the step log, so an
671	// unreachable remote stays visible without being fatal.
672	let _ = self.run(host, &git_fetch_cmd(&repo))?;
673	let (code, _) = self.run(host, &git_checkout_tag_cmd(&repo, &self.tag))?;
674	if code != 0 {
675	let (probe, _) = self.run(host, &git_tag_exists_cmd(&repo, &self.tag))?;
676	anyhow::bail!(
677	"checkout of {} failed on `{host}`: {}",
678	self.tag,
679	checkout_failure_reason(&self.tag, probe == 0)
680	);
681	}
682	let (code, tail) = self.run(host, &git_rev_parse_cmd(&repo))?;
683	anyhow::ensure!(code == 0, "rev-parse failed on `{host}`");
684	Ok(tail.trim().to_string())
685	}
686
687	/// Every artifact this run collected, `file name -> sha256`.
688	///
689	/// Already computed at `collect`, which is the only moment the bytes are
690	/// known to be the ones that landed.
691	pub fn artifact_hashes(&self) -> HashMap<String, String> {
692	self.artifact_hashes.lock().unwrap().clone()
693	}
694
695	/// Resolve `glob` on `host` to the single artifact it names. The `for` loop
696	/// lists each existing match on its own line (and prints nothing — rather
697	/// than a literal unexpanded pattern — when the glob matches no file), so
698	/// the count is unambiguous. `required` controls whether zero matches is an
699	/// error; more than one always is. See `resolve_artifact_match`.
700	fn resolve_artifact(
701	self: &Arc<Self>,
702	host: &str,
703	glob: &str,
704	required: bool,
705	) -> Result<String> {
706	ensure_glob_safe(glob)?;
707	// `[ -e ]` guards against a non-matching glob surviving as its literal
708	// self, and lists one path per line for the count.
709	let cmd = format!("for __f in {glob}; do [ -e \"$__f\" ] && printf '%s\\n' \"$__f\"; done");
710	let (code, tail) = self.run(host, &cmd)?;
711	anyhow::ensure!(
712	code == 0,
713	"resolving artifact glob `{glob}` on `{host}` exited {code}"
714	);
715	resolve_artifact_match(&tail, glob, required)
716	}
717	}
718
719	// ----- error bridging: anyhow -> Rhai runtime error -----
720
721	// Rhai host functions return `Result<_, Box<EvalAltResult>>` by convention, so
722	// this bridge must yield the boxed form to be usable with `.map_err(rhai_err)`.
723	#[allow(
724	clippy::unnecessary_box_returns,
725	reason = "rhai's error type is used boxed throughout its host-function API"
726	)]
727	fn rhai_err(e: impl std::fmt::Display) -> Box<EvalAltResult> {
728	Box::new(EvalAltResult::ErrorRuntime(
729	e.to_string().into(),
730	rhai::Position::NONE,
731	))
732	}
733
734	/// A crate's publish-relevant metadata, read from `cargo metadata`.
735	#[derive(Debug, Clone)]
736	pub struct CrateMeta {
737	pub name: String,
738	pub version: String,
739	pub repository: Option<String>,
740	pub description: Option<String>,
741	pub licensed: bool,
742	}
743
744	/// Parse the fields that matter for publishing out of `cargo metadata` JSON.
745	pub fn crate_meta_from_json(raw: &str) -> Result<CrateMeta> {
746	let v: serde_json::Value = serde_json::from_str(raw).context("parsing cargo metadata")?;
747	let p = v
748	.get("packages")
749	.and_then(\|p\| p.as_array())
750	.and_then(\|a\| a.first())
751	.context("cargo metadata reported no package")?;
752	let str_field = \|k: &str\| {
753	p.get(k)
754	.and_then(\|x\| x.as_str())
755	.filter(\|s\| !s.is_empty())
756	.map(str::to_string)
757	};
758	Ok(CrateMeta {
759	name: str_field("name").context("package has no name")?,
760	version: str_field("version").context("package has no version")?,
761	repository: str_field("repository"),
762	description: str_field("description"),
763	licensed: str_field("license").is_some() \|\| str_field("license_file").is_some(),
764	})
765	}
766
767	/// Everything wrong with a crate's metadata, as messages. Empty means publishable.
768	///
769	/// Checks only what crates.io records permanently. A published version cannot
770	/// be edited, only yanked, and yanking does not correct a wrong URL — so these
771	/// are the last moment any of it can be fixed.
772	pub fn crate_publish_problems(
773	meta: &CrateMeta,
774	repo_clonable: bool,
775	published: &[String],
776	credentials_present: bool,
777	) -> Vec<String> {
778	let mut out = Vec::new();
779	if !credentials_present {
780	out.push(
781	"no crates.io credentials on the publishing host: `cargo login` there first. \
782	Checked now rather than at the upload, so this fails in seconds instead of \
783	after a full build and verify."
784	.to_string(),
785	);
786	}
787	match &meta.repository {
788	None => out.push(
789	"no `repository` field: the crates.io page will show no source link, permanently"
790	.to_string(),
791	),
792	Some(url) if !repo_clonable => out.push(format!(
793	"`repository` is not publicly clonable: {url} \
794	(wrong URL, or the repo is private)"
795	)),
796	Some(_) => {}
797	}
798	if meta.description.is_none() {
799	out.push("no `description`: crates.io requires one".to_string());
800	}
801	if !meta.licensed {
802	out.push("no `license` or `license-file`".to_string());
803	}
804	if published.iter().any(\|v\| v == &meta.version) {
805	out.push(format!(
806	"version {} is already published; bump it",
807	meta.version
808	));
809	}
810	out
811	}
812
813	/// Read the app's version from its checkout on the daemon host. With
814	/// `version_path` set (topology `version_path`), read exactly that file — a
815	/// `.json` as a Tauri config, anything else as a `Cargo.toml`. Unset (the Tauri
816	/// default), try `src-tauri/tauri.conf.json` then the root `Cargo.toml`. Used by
817	/// the runner's default-version path.
818	pub fn version_from_repo(repo: &str, version_path: Option<&str>) -> Result<Version> {
819	let root = expand_tilde(repo);
820	if let Some(vp) = version_path {
821	let path = root.join(vp);
822	let raw = std::fs::read_to_string(&path)
823	.with_context(\|\| format!("reading version file {}", path.display()))?;
824	let ver = if std::path::Path::new(vp)
825	.extension()
826	.is_some_and(\|e\| e.eq_ignore_ascii_case("json"))
827	{
828	version_from_tauri_json(&raw)?
829	} else {
830	version_from_cargo_toml(&raw)?
831	};
832	return Version::parse(&ver).map_err(\|e\| anyhow::anyhow!(e));
833	}
834	let tauri_conf = root.join("src-tauri").join("tauri.conf.json");
835	if tauri_conf.exists() {
836	let raw = std::fs::read_to_string(&tauri_conf)
837	.with_context(\|\| format!("reading {}", tauri_conf.display()))?;
838	return Version::parse(&version_from_tauri_json(&raw)?).map_err(\|e\| anyhow::anyhow!(e));
839	}
840	let cargo_toml = root.join("Cargo.toml");
841	let raw = std::fs::read_to_string(&cargo_toml).with_context(\|\| {
842	format!(
843	"reading {} (no tauri.conf.json either)",
844	cargo_toml.display()
845	)
846	})?;
847	Version::parse(&version_from_cargo_toml(&raw)?).map_err(\|e\| anyhow::anyhow!(e))
848	}
849
850	/// Extract `version` from raw `tauri.conf.json` text.
851	fn version_from_tauri_json(raw: &str) -> Result<String> {
852	let v: serde_json::Value = serde_json::from_str(raw).context("parsing tauri.conf.json")?;
853	v.get("version")
854	.and_then(\|x\| x.as_str())
855	.map(str::to_owned)
856	.context("no `version` in tauri.conf.json")
857	}
858
859	/// Extract the version from raw `Cargo.toml` text — `[package].version` (a leaf
860	/// crate) or `[workspace.package].version` (a workspace that sets it).
861	fn version_from_cargo_toml(raw: &str) -> Result<String> {
862	let doc: toml::Value = toml::from_str(raw).context("parsing Cargo.toml")?;
863	doc.get("package")
864	.and_then(\|p\| p.get("version"))
865	.or_else(\|\| {
866	doc.get("workspace")
867	.and_then(\|w\| w.get("package"))
868	.and_then(\|p\| p.get("version"))
869	})
870	.and_then(\|v\| v.as_str())
871	.map(str::to_owned)
872	.context("no `[package].version` or `[workspace.package].version` in Cargo.toml")
873	}
874
875	/// Cross-check every version source in a repo and confirm they all agree with
876	/// the version being built, before a single host pulls or compiles.
877	///
878	/// `version_from_repo` reads exactly one file, so a `tauri.conf.json` at 0.5.0
879	/// and a root `Cargo.toml` still at 0.4.0 build happily and file artifacts under
880	/// whichever the runner happened to read. This reads every source present —
881	/// `version_path` (when set), `src-tauri/tauri.conf.json`, and the root
882	/// `Cargo.toml` — and fails loudly when any disagree, naming each file and its
883	/// version. A source that is absent is skipped (a library crate with only a
884	/// `Cargo.toml` has nothing to disagree with); the check never invents drift.
885	///
886	/// Scope: the JSON/TOML sources bentod itself reads. The iOS `gen/apple/project.yml`
887	/// path (rewritten by a build-time `sed`) is out of scope here — it is asserted at
888	/// its own build step — but the same drift class motivated this guard.
889	pub fn check_version_consistency(
890	repo: &str,
891	version_path: Option<&str>,
892	expected: &Version,
893	) -> Result<()> {
894	let root = expand_tilde(repo);
895	// (human-readable source label, parsed version) for every source present.
896	let mut found: Vec<(String, Version)> = Vec::new();
897
898	let mut consider = \|rel: &str, raw: &str, as_json: bool\| -> Result<()> {
899	let ver = if as_json {
900	version_from_tauri_json(raw)
901	} else {
902	version_from_cargo_toml(raw)
903	}?;
904	let parsed = Version::parse(&ver).map_err(\|e\| anyhow::anyhow!(e))?;
905	found.push((rel.to_string(), parsed));
906	Ok(())
907	};
908
909	if let Some(vp) = version_path {
910	let path = root.join(vp);
911	let raw = std::fs::read_to_string(&path)
912	.with_context(\|\| format!("reading version file {}", path.display()))?;
913	let is_json = std::path::Path::new(vp)
914	.extension()
915	.is_some_and(\|e\| e.eq_ignore_ascii_case("json"));
916	consider(vp, &raw, is_json)?;
917	}
918	let tauri_conf = root.join("src-tauri").join("tauri.conf.json");
919	if version_path != Some("src-tauri/tauri.conf.json") && tauri_conf.exists() {
920	let raw = std::fs::read_to_string(&tauri_conf)
921	.with_context(\|\| format!("reading {}", tauri_conf.display()))?;
922	consider("src-tauri/tauri.conf.json", &raw, true)?;
923	}
924	let cargo_toml = root.join("Cargo.toml");
925	if version_path != Some("Cargo.toml") && cargo_toml.exists() {
926	// A Cargo.toml with neither `[package].version` nor
927	// `[workspace.package].version` (a pure virtual workspace) carries no
928	// version to check — skip it rather than fail.
929	if let Ok(raw) = std::fs::read_to_string(&cargo_toml)
930	&& version_from_cargo_toml(&raw).is_ok()
931	{
932	consider("Cargo.toml", &raw, false)?;
933	}
934	}
935
936	let disagree: Vec<&(String, Version)> = found.iter().filter(\|(_, v)\| v != expected).collect();
937	anyhow::ensure!(
938	disagree.is_empty(),
939	"version drift in {repo}: building {expected} but {}",
940	disagree
941	.iter()
942	.map(\|(src, v)\| format!("{src} says {v}"))
943	.collect::<Vec<_>>()
944	.join(", ")
945	);
946	Ok(())
947	}
948
949	/// Every `X.Y.Z`-shaped version embedded in an artifact file name. Each maximal
950	/// run of digits-and-dots contributes its first three numeric fields:
951	/// `GoingsOn_0.5.0_aarch64.dmg` and `demo-9.9.9.bin` both yield one version (the
952	/// trailing `.bin`/`.dmg` dot is tolerated), while `latest.json` yields `[]` and
953	/// the `64` in `x86_64` is not three fields. Only the `major.minor.patch` core is
954	/// taken; a prerelease/build suffix is separated by `-`/`+` and not needed here.
955	fn versions_in_filename(name: &str) -> Vec<Version> {
956	name.split(\|c: char\| !(c.is_ascii_digit() \|\| c == '.'))
957	.filter_map(\|run\| {
958	let f: Vec<&str> = run.split('.').filter(\|s\| !s.is_empty()).collect();
959	if f.len() >= 3 && f[..3].iter().all(\|s\| s.chars().all(\|c\| c.is_ascii_digit())) {
960	Version::parse(&format!("{}.{}.{}", f[0], f[1], f[2])).ok()
961	} else {
962	None
963	}
964	})
965	.collect()
966	}
967
968	/// Fail when a collected file's name embeds a version whose `major.minor.patch`
969	/// is not the one being built. This is the guard against a stale checked-in
970	/// artifact winning a glob: `ls -t <glob>` once let
971	/// `AudioFiles-0.4.0-x86_64.AppImage` ship against 0.5.0. A file whose name
972	/// carries no version (an updater `latest.json`, a `.sig`) is not asserted —
973	/// there is nothing to compare. Compared on the core so a prerelease build's
974	/// plain `X.Y.Z` in the filename still matches.
975	fn assert_artifact_version(name: &str, expected: &Version) -> Result<()> {
976	let versions = versions_in_filename(name);
977	anyhow::ensure!(
978	versions.is_empty() \|\| versions.iter().any(\|v\| v.core() == expected.core()),
979	"collected artifact `{name}` carries version {} but the build is {expected}; \
980	a stale artifact was left in the output dir — clean it so only {expected} remains",
981	versions
982	.iter()
983	.map(ToString::to_string)
984	.collect::<Vec<_>>()
985	.join("/"),
986	);
987	Ok(())
988	}
989
990	/// sha256 of a file, lowercase hex. Streams in 64 KiB chunks so a multi-GiB
991	/// bundle never lands in memory whole.
992	fn sha256_file(path: &Path) -> Result<String> {
993	let mut file =
994	std::fs::File::open(path).with_context(\|\| format!("hashing {}", path.display()))?;
995	let mut hasher = Sha256::new();
996	std::io::copy(&mut file, &mut hasher)
997	.with_context(\|\| format!("reading {} to hash", path.display()))?;
998	Ok(hex_lower(&hasher.finalize()))
999	}
1000
1001	/// Every regular file under `root`, as `(path relative to root, absolute path)`,
1002	/// sorted by the relative path.
1003	///
1004	/// This walk has to match `bundle::digest_dir` in sando, file for file. That
1005	/// function re-hashes an incoming bundle and refuses it when the bytes disagree
1006	/// with the manifest they arrived with, so a producer that walks differently
1007	/// produces a manifest the consumer will reject for an artifact nothing is wrong
1008	/// with. Three properties carry that agreement, and none is incidental:
1009	///
1010	/// - Recursive. A bundle may carry a directory (migrations, resources), and
1011	/// a top-level-only listing would omit its contents from the manifest while
1012	/// the verifier hashed them.
1013	/// - Symlinks are not followed, and not recorded. Following one would let
1014	/// content from outside the bundle into its identity; recording the link
1015	/// itself would name a file the verifier does not hash.
1016	/// - Relative paths, `/`-separated, sorted. Readdir order is not guaranteed,
1017	/// so an unsorted manifest would differ run to run on one machine, never mind
1018	/// between two.
1019	fn collected_files(root: &Path) -> std::io::Result<Vec<(String, PathBuf)>> {
1020	fn walk(dir: &Path, root: &Path, out: &mut Vec<(String, PathBuf)>) -> std::io::Result<()> {
1021	for entry in std::fs::read_dir(dir)? {
1022	let entry = entry?;
1023	let ft = entry.file_type()?;
1024	let path = entry.path();
1025	if ft.is_dir() {
1026	walk(&path, root, out)?;
1027	} else if ft.is_file() {
1028	let rel = path
1029	.strip_prefix(root)
1030	.unwrap_or(&path)
1031	.components()
1032	.map(\|c\| c.as_os_str().to_string_lossy())
1033	.collect::<Vec<_>>()
1034	.join("/");
1035	out.push((rel, path));
1036	}
1037	// Symlinks and other special files are intentionally ignored,
1038	// matching the verifier.
1039	}
1040	Ok(())
1041	}
1042	let mut out = Vec::new();
1043	walk(root, root, &mut out)?;
1044	out.sort_by(\|a, b\| a.0.cmp(&b.0));
1045	Ok(out)
1046	}
1047
1048	/// Lowercase-hex encode without pulling in a hex crate.
1049	fn hex_lower(bytes: &[u8]) -> String {
1050	use std::fmt::Write as _;
1051	let mut s = String::with_capacity(bytes.len() * 2);
1052	for b in bytes {
1053	let _ = write!(s, "{b:02x}");
1054	}
1055	s
1056	}
1057
1058	/// Refresh every remote's refs and tags, so the tag a release names is present
1059	/// locally however it was pushed. No branch/upstream assumptions — a bare
1060	/// `git pull --ff-only` needs a tracking branch the release path shouldn't
1061	/// depend on.
1062	///
1063	/// Best-effort on purpose. `fetch --all` exits non-zero if ANY remote fails, and
1064	/// the library repos carry three (`astra`, `mnw`, `srht`), so chaining this into
1065	/// the checkout with `&&` meant one unreachable mirror aborted the release and
1066	/// reported it as a missing tag. The checkout below is the step allowed to fail;
1067	/// this one only has to try. See [`git_checkout_tag_cmd`].
1068	///
1069	/// `repo` is interpolated UNQUOTED so a leading `~` is expanded by the remote
1070	/// host's shell (the checkout path is trusted topology config, not user input),
1071	/// matching how the recipes `cd` into it.
1072	pub fn git_fetch_cmd(repo: &str) -> String {
1073	format!("git -C {repo} fetch --all --tags --prune")
1074	}
1075
1076	/// Pin the host's checkout to the release tag. Runs after [`git_fetch_cmd`], and
1077	/// is the operation whose exit code decides whether the release proceeds.
1078	pub fn git_checkout_tag_cmd(repo: &str, tag: &str) -> String {
1079	format!("git -C {repo} checkout \"{tag}\"")
1080	}
1081
1082	/// Does `tag` resolve to a commit in this checkout? Run only when the checkout
1083	/// has already failed, to say WHY: an absent tag is an untagged or unpushed
1084	/// release, while a tag that resolves fine means the checkout was refused for a
1085	/// local reason (a dirty tree, most often) and the operator needs to hear that
1086	/// instead.
1087	pub fn git_tag_exists_cmd(repo: &str, tag: &str) -> String {
1088	format!("git -C {repo} rev-parse -q --verify \"refs/tags/{tag}^{{commit}}\"")
1089	}
1090
1091	/// The operator-facing explanation for a failed tag checkout, given whether the
1092	/// tag turned out to exist locally.
1093	pub fn checkout_failure_reason(tag: &str, tag_exists: bool) -> String {
1094	if tag_exists {
1095	format!(
1096	"tag {tag} exists but could not be checked out \
1097	(uncommitted changes in the checkout?)"
1098	)
1099	} else {
1100	format!("tag {tag} does not exist there (is it created and pushed?)")
1101	}
1102	}
1103
1104	/// Uncommitted changes to TRACKED files, one `XY path` line each, empty when the
1105	/// tree is clean.
1106	///
1107	/// `--untracked-files=no` on purpose: an untracked file is not built into the
1108	/// binary and a build host accumulates them (editor scratch, stray logs), so
1109	/// failing a release on one would be noise. A modified tracked file is the
1110	/// opposite — it is exactly what `cargo build` would pick up instead of the
1111	/// tagged content.
1112	///
1113	/// Scoped to `repo` with `-- .` rather than asking about the whole repository,
1114	/// which matters only for the repos holding more than one product. `repo` for
1115	/// pom is `~/Code/MNW/pom` inside the MNW monorepo, and an edit in `server/` is
1116	/// not something pom's build can compile. Refusing pom's release for it would be
1117	/// a gate that fires on unrelated work, which is how a gate gets bypassed. For a
1118	/// single-product repo `repo` is the root and this is the whole tree, unchanged.
1119	pub fn git_dirty_cmd(repo: &str) -> String {
1120	format!("git -C {repo} status --porcelain --untracked-files=no -- .")
1121	}
1122
1123	/// The branch a host's checkout is on, empty (and non-zero) on a detached HEAD.
1124	/// Read BEFORE the release pins the tag, so the checkout can be put back
1125	/// afterwards — see [`git_restore_branch_cmd`].
1126	pub fn git_current_branch_cmd(repo: &str) -> String {
1127	format!("git -C {repo} symbolic-ref -q --short HEAD")
1128	}
1129
1130	/// Put a checkout back on the branch it was on before the release pinned it to
1131	/// the tag.
1132	///
1133	/// The pin itself is correct and deliberate: a release must build the tagged
1134	/// commit, not the branch tip. What was missing is the other half. Leaving the
1135	/// tree detached is invisible — git does not warn, and commits made afterwards
1136	/// succeed normally while belonging to no branch. makeover shipped 2.3.0 from
1137	/// exactly that state on 2026-07-28: three commits, including the published one,
1138	/// existed only as a detached HEAD on one machine, on no branch and no remote.
1139	pub fn git_restore_branch_cmd(repo: &str, branch: &str) -> String {
1140	format!("git -C {repo} checkout \"{branch}\"")
1141	}
1142
1143	/// The command a host runs to report the commit it has checked out, for the
1144	/// release preflight barrier.
1145	pub fn git_rev_parse_cmd(repo: &str) -> String {
1146	format!("git -C {repo} rev-parse HEAD")
1147	}
1148
1149	/// Expand a leading `~/` to `$HOME`. Paths in the topology are written with `~`.
1150	pub fn expand_tilde(p: &str) -> PathBuf {
1151	if let Some(rest) = p.strip_prefix("~/")
1152	&& let Ok(home) = std::env::var("HOME")
1153	{
1154	return Path::new(&home).join(rest);
1155	}
1156	PathBuf::from(p)
1157	}
1158
1159	/// Reject a glob that carries shell command metacharacters. Path and wildcard
1160	/// characters (`/ . * ? [ ] ~` etc.) are fine — the pattern reaches a login
1161	/// shell to be expanded — but a `;` or `$(...)` must not ride along and run.
1162	/// Not a privilege boundary (a recipe already runs arbitrary shell via `sh_ok`)
1163	/// but it keeps a malformed pattern from turning into a command. Shared by
1164	/// `collect` and `resolve_artifact`.
1165	fn ensure_glob_safe(glob: &str) -> Result<()> {
1166	anyhow::ensure!(
1167	!glob.chars().any(\|c\| matches!(
1168	c,
1169	';' \| '&' \| '\|' \| '$' \| '`' \| '\'' \| '"' \| '\\' \| ' ' \| '\n' \| '(' \| ')' \| '<' \| '>'
1170	)),
1171	"glob `{glob}` contains shell metacharacters"
1172	);
1173	Ok(())
1174	}
1175
1176	/// Decide the single artifact a glob resolves to from a newline-separated
1177	/// listing of the paths that matched it.
1178	///
1179	/// The recipes used to select an artifact with `ls -t <glob> \| head -1` and
1180	/// guard only on an empty string, so a non-zero `ls` slipped past quietly and a
1181	/// stale newest-by-mtime file could win. This is the strict replacement: it
1182	/// demands exactly one match. Zero matches fail when `required` (return `""`
1183	/// when optional); more than one is always an error rather than an arbitrary
1184	/// newest-wins pick, because an ambiguous match means the build left stale
1185	/// artifacts behind and the wrong one could ship.
1186	fn resolve_artifact_match(listing: &str, glob: &str, required: bool) -> Result<String> {
1187	let matches: Vec<&str> = listing
1188	.lines()
1189	.map(str::trim)
1190	.filter(\|l\| !l.is_empty())
1191	.collect();
1192	match matches.as_slice() {
1193	[] if required => anyhow::bail!("no artifact matched glob `{glob}`"),
1194	[] => Ok(String::new()),
1195	[one] => Ok((*one).to_string()),
1196	many => anyhow::bail!(
1197	"glob `{glob}` is ambiguous: {} artifacts matched ({}). \
1198	The build left more than one behind; clean stale artifacts so exactly one remains.",
1199	many.len(),
1200	many.join(", ")
1201	),
1202	}
1203	}
1204
1205	/// Build a Rhai engine with the host API bound to `ctx`. Sandboxed: recipes
1206	/// touch the outside world only through these functions.
1207	pub fn build_engine(ctx: &Arc<RecipeCtx>) -> Engine {
1208	let mut engine = Engine::new();
1209	// Defensive caps — recipes are first-party but bound the blast radius.
1210	engine.set_max_operations(5_000_000);
1211	engine.set_max_call_levels(64);
1212	engine.set_max_string_size(0);
1213
1214	// --- step(name) ---
1215	{
1216	let ctx = ctx.clone();
1217	engine.register_fn(
1218	"step",
1219	move \|name: &str\| -> Result<(), Box<EvalAltResult>> {
1220	let step: Step = name.parse().map_err(rhai_err)?;
1221	ctx.begin_step(step).map_err(rhai_err)
1222	},
1223	);
1224	}
1225
1226	// --- sh(host, cmd) -> #{ code, stdout_tail } ---
1227	//
1228	// The branch-on-exit-code primitive: the recipe OWNS the outcome. A non-zero
1229	// exit is returned, not raised, and does NOT fail the step or bar publish —
1230	// use this only when the recipe inspects `code` and decides. For a command
1231	// that must succeed (build/sign/etc.), use `sh_ok`, which fails the step (and
1232	// therefore bars publish via the failed-step ledger) on a non-zero exit.
1233	{
1234	let ctx = ctx.clone();
1235	engine.register_fn(
1236	"sh",
1237	move \|host: &str, cmd: &str\| -> Result<Map, Box<EvalAltResult>> {
1238	let (code, tail) = ctx.run(host, cmd).map_err(rhai_err)?;
1239	let mut m = Map::new();
1240	m.insert("code".into(), (code as i64).into());
1241	m.insert("stdout_tail".into(), tail.into());
1242	Ok(m)
1243	},
1244	);
1245	}
1246
1247	// --- sh_ok(host, cmd): run + assert exit 0 (the must-succeed primitive) ---
1248	//
1249	// A non-zero exit fails the current step (added to the publish-barring
1250	// ledger) and aborts the recipe, so an artifact is never shipped after a
1251	// must-succeed command failed.
1252	{
1253	let ctx = ctx.clone();
1254	engine.register_fn(
1255	"sh_ok",
1256	move \|host: &str, cmd: &str\| -> Result<(), Box<EvalAltResult>> {
1257	let (code, _) = ctx.run(host, cmd).map_err(rhai_err)?;
1258	if code != 0 {
1259	// Attribute the failure to the current step explicitly so the
1260	// ledger bars publish even if a future caller swallowed the error.
1261	ctx.fail_current_step();
1262	return Err(rhai_err(format!(
1263	"command on `{host}` exited {code}: {cmd}"
1264	)));
1265	}
1266	Ok(())
1267	},
1268	);
1269	}
1270
1271	// --- resolve_artifact(host, glob) -> path: the ONE artifact matching glob ---
1272	//
1273	// The artifact-selection primitive. Replaces `sh(host, "ls -t <glob> \| head
1274	// -1").stdout_tail.trim()` guarded on an empty string, which let a non-zero
1275	// `ls` pass quietly and a stale newest-by-mtime file win. This resolves the
1276	// glob on the host and demands exactly one match: zero matches or more than
1277	// one both throw (an ambiguous match means the build left stale artifacts,
1278	// and silently picking the newest is how the wrong bytes ship). Use
1279	// `resolve_artifact_opt` for an artifact that may legitimately be absent.
1280	{
1281	let ctx = ctx.clone();
1282	engine.register_fn(
1283	"resolve_artifact",
1284	move \|host: &str, glob: &str\| -> Result<String, Box<EvalAltResult>> {
1285	ctx.resolve_artifact(host, glob, true).map_err(rhai_err)
1286	},
1287	);
1288	}
1289
1290	// --- resolve_artifact_opt(host, glob) -> path \| "": zero-or-one match ---
1291	//
1292	// Same strict resolution as `resolve_artifact` but tolerates zero matches
1293	// (returns ""); more than one is still an error. For optional outputs like a
1294	// `.deb` or an updater bundle a recipe collects only when present.
1295	{
1296	let ctx = ctx.clone();
1297	engine.register_fn(
1298	"resolve_artifact_opt",
1299	move \|host: &str, glob: &str\| -> Result<String, Box<EvalAltResult>> {
1300	ctx.resolve_artifact(host, glob, false).map_err(rhai_err)
1301	},
1302	);
1303	}
1304
1305	// --- log(msg): operator-visible line into the current step's tail ---
1306	{
1307	let ctx = ctx.clone();
1308	engine.register_fn("log", move \|msg: &str\| -> Result<(), Box<EvalAltResult>> {
1309	let sink = ctx.ensure_step().map_err(rhai_err)?;
1310	let line = format!("[recipe] {msg}\n");
1311	ctx.rt.block_on(async {
1312	use ops_core::remote::LogSink;
1313	sink.lock().await.write_chunk(line.as_bytes()).await;
1314	});
1315	Ok(())
1316	});
1317	}
1318
1319	// --- version_of(app) -> string ---
1320	{
1321	let ctx = ctx.clone();
1322	engine.register_fn(
1323	"version_of",
1324	move \|app: &str\| -> Result<String, Box<EvalAltResult>> {
1325	// Only the current app is in scope; cross-app reads aren't needed.
1326	if app != ctx.app.as_str() {
1327	return Err(rhai_err(format!(
1328	"version_of: `{app}` is not the app being built"
1329	)));
1330	}
1331	Ok(ctx.version.to_string())
1332	},
1333	);
1334	}
1335
1336	// --- version() -> string: the version being built (no-arg form) ---
1337	{
1338	let ctx = ctx.clone();
1339	engine.register_fn("version", move \|\| -> String { ctx.version.to_string() });
1340	}
1341
1342	// --- build_host() -> string: the host this target builds on ---
1343	{
1344	let ctx = ctx.clone();
1345	engine.register_fn("build_host", move \|\| -> String { ctx.build_host.clone() });
1346	}
1347
1348	// --- repo() -> string: the app's checkout path on this target's build host
1349	// (`~`-prefixed on a unix host). Host-correct rather than one path per
1350	// app, so a recipe for a host whose checkout is elsewhere still calls
1351	// this instead of hard-coding the path — which is what kept the Windows
1352	// recipes off `checkout_sha`. ---
1353	{
1354	let ctx = ctx.clone();
1355	engine.register_fn("repo", move \|\| -> String {
1356	ctx.repo_for(&ctx.build_host).to_string()
1357	});
1358	}
1359
1360	// --- checkout_sha(host) -> sha: pin this host to the release tag and report
1361	// its commit. Replaces a recipe's `git pull --ff-only`, which builds
1362	// whatever `main` is at pull time; the daemon also runs the same pin as
1363	// a cross-host preflight barrier before any target builds. ---
1364	{
1365	let ctx = ctx.clone();
1366	engine.register_fn(
1367	"checkout_sha",
1368	move \|host: &str\| -> Result<String, Box<EvalAltResult>> {
1369	ctx.checkout_sha(host).map_err(rhai_err)
1370	},
1371	);
1372	}
1373
1374	// --- crate_preflight() -> string: verify this crate is safe to publish,
1375	// or abort the run. Everything it checks is immutable once published:
1376	// crates.io versions can be yanked but never edited, so a wrong
1377	// repository URL is permanent. pter 0.1.0 shipped with a dead one. ---
1378	{
1379	let ctx = ctx.clone();
1380	engine.register_fn(
1381	"crate_preflight",
1382	move \|\| -> Result<String, Box<EvalAltResult>> {
1383	// `repo`, not `repo_for(...)`: `cargo metadata` runs on the
1384	// daemon's own box, so this is the one checkout that is always
1385	// the local one. It is not a missed call site.
1386	let repo = expand_tilde(&ctx.repo);
1387
1388	let out = std::process::Command::new("cargo")
1389	.args(["metadata", "--no-deps", "--format-version", "1"])
1390	.current_dir(&repo)
1391	.output()
1392	.map_err(\|e\| format!("running cargo metadata in {}: {e}", repo.display()))?;
1393	if !out.status.success() {
1394	return Err(format!(
1395	"cargo metadata failed in {}: {}",
1396	repo.display(),
1397	String::from_utf8_lossy(&out.stderr).trim()
1398	)
1399	.into());
1400	}
1401	let meta = crate_meta_from_json(&String::from_utf8_lossy(&out.stdout))
1402	.map_err(\|e\| e.to_string())?;
1403
1404	// The real question is not whether a page renders but whether a
1405	// stranger with no credentials can fetch the source, so ask git.
1406	let clonable = meta.repository.as_ref().is_some_and(\|url\| {
1407	std::process::Command::new("git")
1408	.args(["ls-remote", url])
1409	.env("GIT_TERMINAL_PROMPT", "0")
1410	.output()
1411	.is_ok_and(\|o\| o.status.success())
1412	});
1413
1414	// Ask the publishing host whether cargo has credentials, rather
1415	// than moving the token anywhere. It stays in cargo's own 0600
1416	// store; a shell line carrying it would be visible in `ps`.
1417	// An exit code answers "are there credentials"; an Err answers
1418	// "the question could not be asked". Collapsing the second into
1419	// the first reported a capability denial as "no crates.io
1420	// credentials", which sent a real diagnosis three rounds the
1421	// wrong way. A check that cannot run is not a failed check.
1422	let creds =
1423	ctx.run(
1424	&ctx.build_host.clone(),
1425	"cargo login --help >/dev/null 2>&1 && \
1426	test -s \"${CARGO_HOME:-$HOME/.cargo}/credentials.toml\" \
1427	\|\| test -s \"${CARGO_HOME:-$HOME/.cargo}/credentials\"",
1428	)
1429	.map_err(\|e\| {
1430	format!(
1431	"could not check crates.io credentials on `{}`: {e}",
1432	ctx.build_host
1433	)
1434	})?
1435	.0 == 0;
1436
1437	let published = RecipeCtx::published_versions(&meta.name);
1438	let problems = crate_publish_problems(&meta, clonable, &published, creds);
1439	if !problems.is_empty() {
1440	return Err(format!(
1441	"{} {} is not safe to publish:\n - {}",
1442	meta.name,
1443	meta.version,
1444	problems.join("\n - ")
1445	)
1446	.into());
1447	}
1448	Ok(format!("{} {} passed preflight", meta.name, meta.version))
1449	},
1450	);
1451	}
1452
1453	// --- feature_flags() -> string: `--features a,b`, or "" when the app
1454	// declares none. Returns the whole flag rather than a bare list so an
1455	// app with no features cannot produce a dangling `--features`. ---
1456	{
1457	let ctx = ctx.clone();
1458	engine.register_fn("feature_flags", move \|\| -> String {
1459	if ctx.features.is_empty() {
1460	String::new()
1461	} else {
1462	format!("--features {}", ctx.features.join(","))
1463	}
1464	});
1465	}
1466
1467	// --- target() / platform() / arch(): the target axis, for one per-platform
1468	// recipe to branch on arch (bundle paths differ between x86_64/aarch64). ---
1469	{
1470	let ctx = ctx.clone();
1471	engine.register_fn("target", move \|\| -> String { ctx.target.to_string() });
1472	}
1473	{
1474	let ctx = ctx.clone();
1475	engine.register_fn("platform", move \|\| -> String {
1476	ctx.target.platform.as_str().to_string()
1477	});
1478	}
1479	{
1480	let ctx = ctx.clone();
1481	engine.register_fn("arch", move \|\| -> String {
1482	ctx.target.arch.as_str().to_string()
1483	});
1484	}
1485
1486	// --- secret(key) -> string (file under secrets_root; never logged) ---
1487	{
1488	let ctx = ctx.clone();
1489	engine.register_fn("secret", move \|key: &str\| -> Result<String, Box<EvalAltResult>> {
1490	// Guard against traversal out of secrets_root. Require every path
1491	// component to be `Normal` (rejects `..`, `.`, absolute roots and
1492	// drive prefixes) and forbid backslashes (a literal filename char on
1493	// Linux, but a separator elsewhere) — the per-component strength of
1494	// Sando's `safe()`. A multi-segment key like `app/token` is still
1495	// allowed; `foo..bar` (a legit filename) is no longer falsely blocked.
1496	let safe = !key.is_empty()
1497	&& !key.contains('\\')
1498	&& std::path::Path::new(key)
1499	.components()
1500	.all(\|c\| matches!(c, std::path::Component::Normal(_)));
1501	if !safe {
1502	return Err(rhai_err(
1503	"secret key must be a relative path under secrets_root (no `..`, `.`, absolute paths, or backslashes)",
1504	));
1505	}
1506	let path = ctx.cfg.secrets_root.join(key);
1507	std::fs::read_to_string(&path)
1508	.map(\|s\| s.trim_end().to_string())
1509	.map_err(\|e\| rhai_err(format!("secret `{key}`: {e}")))
1510	});
1511	}
1512
1513	// --- env(host, key) -> string ---
1514	{
1515	let ctx = ctx.clone();
1516	engine.register_fn(
1517	"env",
1518	move \|host: &str, key: &str\| -> Result<String, Box<EvalAltResult>> {
1519	// The key is interpolated into a `${...}` shell expansion, so it must
1520	// be a bare shell identifier — anything else (quotes, `}`, `$`, `;`)
1521	// could break out and run arbitrary commands on the host. Validate
1522	// before building the command; this is the one env read that can't
1523	// sh-quote its argument (a quoted var name doesn't expand).
1524	if key.is_empty()
1525	\|\| !key
1526	.chars()
1527	.next()
1528	.is_some_and(\|c\| c == '_' \|\| c.is_ascii_alphabetic())
1529	\|\| !key.chars().all(\|c\| c == '_' \|\| c.is_ascii_alphanumeric())
1530	{
1531	return Err(rhai_err(format!(
1532	"env name `{key}` must be a shell identifier ([A-Za-z_][A-Za-z0-9_]*)"
1533	)));
1534	}
1535	// Read via the shell so it works on remote hosts too.
1536	let (code, tail) = ctx
1537	.run(host, &format!("printf '%s' \"${{{key}}}\""))
1538	.map_err(rhai_err)?;
1539	if code != 0 {
1540	return Err(rhai_err(format!("env `{key}` on `{host}` failed")));
1541	}
1542	Ok(tail.trim().to_string())
1543	},
1544	);
1545	}
1546
1547	// --- collect(host, glob, app, version): pull artifacts to dist_root ---
1548	{
1549	let ctx = ctx.clone();
1550	engine.register_fn(
1551	"collect",
1552	move \|host: &str,
1553	glob: &str,
1554	app: &str,
1555	version: &str\|
1556	-> Result<(), Box<EvalAltResult>> {
1557	ctx.collect(host, glob, app, version).map_err(rhai_err)
1558	},
1559	);
1560	}
1561
1562	// --- publish(channel, app, target, version, artifact, meta) ---
1563	{
1564	let ctx = ctx.clone();
1565	engine.register_fn(
1566	"publish",
1567	move \|channel: &str,
1568	app: &str,
1569	target: &str,
1570	version: &str,
1571	artifact: &str,
1572	meta: Map\|
1573	-> Result<String, Box<EvalAltResult>> {
1574	ctx.publish(channel, app, target, version, artifact, &meta)
1575	.map_err(rhai_err)
1576	},
1577	);
1578	}
1579
1580	// --- deploy(binary) -> summary: install a service binary and restart its
1581	// unit. The terminal step for `kind = "service"`, the counterpart of
1582	// `publish` for something that is run rather than distributed.
1583	//
1584	// Takes only the binary's path on the build host: where it lands, on
1585	// which machine, and which unit restarts all come from the `[[deploy]]`
1586	// entry for the target already being built. A recipe cannot deploy the
1587	// aarch64 binary to the x86_64 box by naming the wrong host, because it
1588	// never names a host at all.
1589	{
1590	let ctx = ctx.clone();
1591	engine.register_fn(
1592	"deploy",
1593	move \|binary: &str\| -> Result<String, Box<EvalAltResult>> {
1594	ctx.deploy(binary).map_err(rhai_err)
1595	},
1596	);
1597	}
1598
1599	// --- deploy_host() -> string: the service host's ssh destination, so a
1600	// recipe can run its own assertions there (`sh_ok(deploy_host(), ...)`).
1601	// Commands run through it while the `deploy` step is open, so they are
1602	// gated on the deploy grant like the install itself. ---
1603	{
1604	let ctx = ctx.clone();
1605	engine.register_fn(
1606	"deploy_host",
1607	move \|\| -> Result<String, Box<EvalAltResult>> {
1608	ctx.deploy_target()
1609	.map(\|d\| d.host.clone())
1610	.map_err(rhai_err)
1611	},
1612	);
1613	}
1614
1615	// --- service_name() / install_path() / health_url(): the rest of the
1616	// `[[deploy]]` entry, so a recipe asserts against the configured values
1617	// rather than repeating them as literals that can drift. `health_url`
1618	// is "" when unset. ---
1619	{
1620	let ctx = ctx.clone();
1621	engine.register_fn(
1622	"service_name",
1623	move \|\| -> Result<String, Box<EvalAltResult>> {
1624	ctx.deploy_target()
1625	.map(\|d\| d.service.clone())
1626	.map_err(rhai_err)
1627	},
1628	);
1629	}
1630	{
1631	let ctx = ctx.clone();
1632	engine.register_fn(
1633	"install_path",
1634	move \|\| -> Result<String, Box<EvalAltResult>> {
1635	ctx.deploy_target()
1636	.map(\|d\| d.install_path.clone())
1637	.map_err(rhai_err)
1638	},
1639	);
1640	}
1641	{
1642	let ctx = ctx.clone();
1643	engine.register_fn(
1644	"health_url",
1645	move \|\| -> Result<String, Box<EvalAltResult>> {
1646	ctx.deploy_target()
1647	.map(\|d\| d.health_url.clone().unwrap_or_default())
1648	.map_err(rhai_err)
1649	},
1650	);
1651	}
1652
1653	// --- glibc_check(binary) -> string: assert the build host did not produce
1654	// a binary the service host's glibc is too old to exec. Aborts the run
1655	// if it did; returns "needs X, host has Y" for the log if it did not. ---
1656	{
1657	let ctx = ctx.clone();
1658	engine.register_fn(
1659	"glibc_check",
1660	move \|binary: &str\| -> Result<String, Box<EvalAltResult>> {
1661	let (needs, has) = ctx.glibc_check(binary).map_err(rhai_err)?;
1662	Ok(format!(
1663	"glibc: binary needs {needs}, service host has {has}"
1664	))
1665	},
1666	);
1667	}
1668
1669	// --- macOS signing helpers. They dispatch through the named host's
1670	// executor like any other step; when that host is the mac (transport =
1671	// "agent"), codesign/notarize/staple ride the in-session `AgentRpc`
1672	// transport — the only security session where the Developer ID key is
1673	// usable (design §7 "THE WALL"). Capability-gated by the host's `sign`
1674	// grant. ---
1675	register_macos_fns(&mut engine, ctx);
1676
1677	engine
1678	}
1679
1680	/// Highest `GLIBC_x.y` version referenced by a built binary, and the glibc a
1681	/// host actually has, both parsed from the text the commands print.
1682	///
1683	/// Native-per-architecture builds remove the cross-compile hazard `deploy.sh`
1684	/// was written against, but not this one: fw13 tracks a newer glibc than the
1685	/// Ubuntu 24.04 box in Hetzner, so a binary built here can reference a symbol
1686	/// version that box does not have and fail at exec — after the unit has already
1687	/// been restarted onto it. Comparing the two before the install is what makes
1688	/// that a failed step instead of a downed service.
1689	fn max_glibc_symbol(objdump_out: &str) -> Option<(u64, u64)> {
1690	objdump_out
1691	.split(\|c: char\| !(c.is_ascii_digit() \|\| c == '.' \|\| c == '_' \|\| c.is_ascii_alphabetic()))
1692	.filter_map(\|tok\| tok.strip_prefix("GLIBC_"))
1693	.filter_map(parse_glibc_version)
1694	.max()
1695	}
1696
1697	/// Parse `2.39` (or `2.39.1`, keeping major/minor) into a comparable pair.
1698	fn parse_glibc_version(s: &str) -> Option<(u64, u64)> {
1699	let mut parts = s.split('.');
1700	let major = parts.next()?.parse().ok()?;
1701	let minor = parts.next()?.parse().ok()?;
1702	Some((major, minor))
1703	}
1704
1705	/// The glibc version out of `ldd --version`'s first line, whose tail is the
1706	/// version however the distro decorates the rest (`ldd (Ubuntu GLIBC
1707	/// 2.39-0ubuntu8.8) 2.39`).
1708	fn glibc_from_ldd(ldd_out: &str) -> Option<(u64, u64)> {
1709	let first = ldd_out.lines().find(\|l\| !l.trim().is_empty())?;
1710	parse_glibc_version(first.split_whitespace().last()?)
1711	}
1712
1713	impl RecipeCtx {
1714	/// This target's install destination, or an error naming why there is none.
1715	fn deploy_target(&self) -> Result<&DeployTarget> {
1716	self.deploy.as_ref().ok_or_else(\|\| {
1717	anyhow::anyhow!(
1718	"no deploy destination for {} {}: the app is `kind = \"{}\"`, and only a \
1719	service declares [[deploy]] entries",
1720	self.app,
1721	self.target,
1722	match self.kind {
1723	Kind::App => "app",
1724	Kind::Library => "library",
1725	Kind::Service => "service",
1726	}
1727	)
1728	})
1729	}
1730
1731	/// Compare the built binary's glibc requirement against the service host's.
1732	/// Returns the two versions for the recipe to log.
1733	fn glibc_check(self: &Arc<Self>, binary: &str) -> Result<(String, String)> {
1734	let d = self.deploy_target()?.clone();
1735	// `objdump -T` on the build host; no symbols at all (a static binary)
1736	// means nothing to check, which is a pass rather than a failure.
1737	let (code, out) = self.run(
1738	&self.build_host.clone(),
1739	&format!(
1740	"objdump -T {binary} 2>/dev/null \| grep -o 'GLIBC_[0-9.]*' \| sort -uV \|\| true"
1741	),
1742	)?;
1743	anyhow::ensure!(code == 0, "reading glibc symbols from {binary} failed");
1744	let Some(needs) = max_glibc_symbol(&out) else {
1745	return Ok(("none".into(), "n/a".into()));
1746	};
1747	let (code, ldd) = self.run(&d.host, "ldd --version")?;
1748	anyhow::ensure!(
1749	code == 0,
1750	"could not read glibc version on service host `{}`",
1751	d.host
1752	);
1753	let has = glibc_from_ldd(&ldd).ok_or_else(\|\| {
1754	anyhow::anyhow!(
1755	"could not parse glibc version from `ldd --version` on `{}`",
1756	d.host
1757	)
1758	})?;
1759	anyhow::ensure!(
1760	needs <= has,
1761	"binary needs glibc {}.{} but `{}` has {}.{} — it would fail to exec after the \
1762	unit restarted onto it. Build on a host no newer than the service host.",
1763	needs.0,
1764	needs.1,
1765	d.host,
1766	has.0,
1767	has.1,
1768	);
1769	Ok((
1770	format!("{}.{}", needs.0, needs.1),
1771	format!("{}.{}", has.0, has.1),
1772	))
1773	}
1774
1775	/// Install `binary` (a path on the BUILD host) onto the service host and
1776	/// restart its unit, via the privileged installer the host holds a scoped
1777	/// sudo grant for.
1778	///
1779	/// Bento never runs the install itself. It stages the bytes and calls a
1780	/// root script whose arguments are re-checked on the far side — the same
1781	/// shape as Sando's `install-companion.sh`, and for the same reason: the
1782	/// sudoers grant is then ONE auditable script rather than a broad
1783	/// `install`+`systemctl` grant on a production box.
1784	///
1785	/// Only the binary moves. Config is deliberately untouched: pom's
1786	/// `pom-astra.toml` / `pom-hetzner.toml` differ per instance, and prod's
1787	/// carried a `[targets.mnw.tests]` block the repo did not have. A deploy
1788	/// that copies config over is how that block gets silently deleted.
1789	fn deploy(self: &Arc<Self>, binary: &str) -> Result<String> {
1790	anyhow::ensure!(
1791	!self.is_cancelled(),
1792	"build superseded by a newer request; refusing to deploy"
1793	);
1794	// A failed earlier step bars a deploy exactly as it bars a publish. An
1795	// artifact that failed its gates must not reach a production host just
1796	// because the recipe kept running.
1797	let failed = self.failed_steps.lock().unwrap().clone();
1798	anyhow::ensure!(
1799	failed.is_empty(),
1800	"refusing to deploy {} {}: {} failed earlier in this run",
1801	self.app,
1802	self.version,
1803	failed
1804	.iter()
1805	.map(ToString::to_string)
1806	.collect::<Vec<_>>()
1807	.join(", "),
1808	);
1809	let d = self.deploy_target()?.clone();
1810	ensure_glob_safe(binary)?;
1811
1812	// Stage under a fixed root the installer also insists on, so "what was
1813	// checked" and "what is installed" cannot drift apart.
1814	let staged = format!("{DEPLOY_STAGING_ROOT}/{}", self.app);
1815	let staged_bin = format!("{staged}/{}", self.app);
1816	let deploy_exec = self.exec(&d.host)?;
1817	anyhow::ensure!(
1818	deploy_exec.capabilities().permits(&Action::Deploy),
1819	"service host `{}` is not granted the `deploy` capability",
1820	d.host
1821	);
1822
1823	self.run_ok(&d.host, &format!("mkdir -p {staged}"))?;
1824	if self.build_host_ssh == d.host {
1825	// Same box: the binary is already there. Routing it through the
1826	// daemon would be two transfers to end up where it started. This is
1827	// pom's aarch64 leg — astra builds it and astra runs it.
1828	self.run_ok(&d.host, &format!("cp -f {binary} {staged_bin}"))?;
1829	} else {
1830	// Build host -> daemon -> service host. Two hops because an executor
1831	// reaches one host; a direct host-to-host transport would mean the
1832	// build host holding a credential for the production box.
1833	let tmp = tempfile::tempdir().context("staging dir for deploy")?;
1834	let local = tmp.path().join(self.app.as_str());
1835	self.pull_for_deploy(binary, &local)?;
1836	let (dest, opts) = (PathBuf::from(&staged), SyncOpts::default());
1837	let dir = tmp.path().to_path_buf();
1838	self.run_bounded(&format!("stage {} on `{}`", self.app, d.host), async move {
1839	deploy_exec.push_dir(&dir, &dest, &opts).await
1840	})
1841	.with_context(\|\| format!("staging {} onto `{}`", self.app, d.host))?;
1842	}
1843
1844	// The privileged half. Every argument is re-validated by the script,
1845	// which is the thing actually holding the sudo grant.
1846	self.run_ok(
1847	&d.host,
1848	&format!(
1849	"{} {staged_bin} {} {}",
1850	self.cfg.deploy_installer, d.install_path, d.service
1851	),
1852	)?;
1853	Ok(format!(
1854	"{} {} installed at {} on `{}`; {} restarted",
1855	self.app, self.version, d.install_path, d.host, d.service
1856	))
1857	}
1858
1859	/// Fetch one file off a host into a daemon-local path for re-pushing.
1860	///
1861	/// A local build host is read directly: `fw13` is the daemon's own box, so
1862	/// the file is already on this filesystem. Routing it through the
1863	/// artifact-pull gate instead would demand a `pull_root` covering every repo
1864	/// a service could be built in — today that is `~/Code/Apps`, and pom lives
1865	/// in `~/Code/MNW`. Widening it to `~/Code` would put `_private`, the
1866	/// secrets root, inside the collectable tree. This is pom's x86_64 leg.
1867	fn pull_for_deploy(self: &Arc<Self>, remote: &str, local: &Path) -> Result<()> {
1868	let host = self.build_host.clone();
1869	let remote_path = expand_tilde(remote);
1870	if self.build_host_ssh == "local" \|\| self.build_host_ssh.is_empty() {
1871	std::fs::copy(&remote_path, local).with_context(\|\| {
1872	format!("staging {} from the daemon host", remote_path.display())
1873	})?;
1874	return Ok(());
1875	}
1876	let sync = self.host_sync(&host)?;
1877	let (src, dst, opts) = (remote_path, local.to_path_buf(), SyncOpts::default());
1878	self.run_bounded(&format!("fetch {remote} from `{host}`"), async move {
1879	sync.pull_file(&src, &dst, &opts).await
1880	})
1881	.with_context(\|\| format!("fetching {remote} from `{host}` to deploy"))
1882	}
1883
1884	/// `run`, failing the step on a non-zero exit. The Rust-side twin of the
1885	/// recipe's `sh_ok`, for commands the deploy machinery issues itself.
1886	fn run_ok(self: &Arc<Self>, host: &str, cmd: &str) -> Result<String> {
1887	let (code, tail) = self.run(host, cmd)?;
1888	if code != 0 {
1889	self.fail_current_step();
1890	anyhow::bail!("command on `{host}` exited {code}: {cmd}\n{tail}");
1891	}
1892	Ok(tail)
1893	}
1894
1895	/// Where this run's collected files land locally.
1896	///
1897	/// Per target, not per version. Every target used to share one
1898	/// `dist_root/<app>/<version>/`, so the hash loop below (which lists the
1899	/// directory) attributed a sibling's AppImage to the mac build's artifact
1900	/// record. It is also the layout the archive uses, and the two have to agree
1901	/// or the local copy and the deposited one are different shapes.
1902	fn collect_dest(&self, app: &str, version: &str) -> PathBuf {
1903	self.cfg
1904	.dist_root
1905	.join(app)
1906	.join(version)
1907	.join(crate::archive::target_slug(self.target))
1908	}
1909
1910	fn collect(self: &Arc<Self>, host: &str, glob: &str, app: &str, version: &str) -> Result<()> {
1911	let dest = self.collect_dest(app, version);
1912	let dest_s = dest.to_string_lossy().into_owned();
1913	// The glob reaches a remote login shell intact (that's what expands it),
1914	// so command metacharacters stay barred. Path/wildcard chars are fine.
1915	ensure_glob_safe(glob)?;
1916	std::fs::create_dir_all(&dest)
1917	.with_context(\|\| format!("creating collect dest {dest_s}"))?;
1918	// The SYNC transport, not the host's exec executor: artifacts move over
1919	// ssh/rsync even from an agent host, whose `/pull` is confined to a
1920	// narrow `pull_root` that deliberately excludes the repo checkout these
1921	// artifacts are built in (see `state::build_sync`). The daemon still
1922	// runs the transfer itself, as it always has.
1923	let sync = self.host_sync(host)?;
1924	let opts = SyncOpts::precompressed();
1925	// Bounded by the collect step's deadline (rsync of a multi-GiB artifact
1926	// can wedge on a stalled transport) and interruptible on supersession.
1927	let dest_pull = dest.clone();
1928	self.run_bounded(&format!("collect {glob} from `{host}`"), async move {
1929	sync.pull_glob(glob, &dest_pull, &opts).await
1930	})
1931	.with_context(\|\| format!("collect {glob} from `{host}`"))?;
1932	// Assert the version and hash every collected file. This is where a
1933	// stale artifact is caught: a file whose name embeds a different version
1934	// fails the collect (rather than silently winning a later glob), and the
1935	// sha256 recorded here is what `publish` writes into the release ledger
1936	// and what the artifact record's manifest is built from.
1937	//
1938	// Recursive, and keyed by path relative to the collect dir. That is not
1939	// a preference: Sando's intake re-hashes the bundle with its own walker,
1940	// which recurses and keys the same way, and refuses a bundle whose bytes
1941	// do not match the manifest it was handed. A top-level `read_dir` keyed
1942	// by file name agrees with that walker for a flat directory and diverges
1943	// the moment a bundle carries a subdirectory — the honest artifact would
1944	// be refused for a manifest that omitted everything nested. The two
1945	// walkers have to be the same walk. See `bundle::digest_dir` in sando.
1946	for (rel, path) in
1947	collected_files(&dest).with_context(\|\| format!("listing collect dest {dest_s}"))?
1948	{
1949	// The version check stays on the file NAME rather than the relative
1950	// path: it is looking for a stale `app_1.2.3.AppImage` beside the
1951	// one this release built, and a directory component is not that.
1952	let name = path
1953	.file_name()
1954	.map_or_else(\|\| rel.clone(), \|n\| n.to_string_lossy().into_owned());
1955	assert_artifact_version(&name, &self.version)?;
1956	let digest = sha256_file(&path)?;
1957	self.artifact_hashes.lock().unwrap().insert(rel, digest);
1958	}
1959	// Deposit at the archive path, so this target's bytes have one address
1960	// whichever host produced them. A no-op when no archive is configured.
1961	//
1962	// Inside `collect`, not after the recipe: a failure here fails the
1963	// collect step, before sign and publish, rather than putting a red mark
1964	// on a release that has already shipped. And it is a failure, not a
1965	// warning — a deposit that is quietly skipped leaves the archive path
1966	// wrong for exactly the release nobody was watching, which is the thing
1967	// having one address is for.
1968	let (cfg, app_id, version, target) = (
1969	self.cfg.clone(),
1970	self.app.clone(),
1971	self.version.clone(),
1972	self.target,
1973	);
1974	let dest_archive = dest.clone();
1975	self.run_bounded("deposit in the archive", async move {
1976	crate::archive::deposit(&cfg, &dest_archive, &app_id, &version, target).await
1977	})?;
1978	// Best-effort size accounting for the event.
1979	events::emit(
1980	&self.events,
1981	Event::ArtifactCollected {
1982	app: self.app.clone(),
1983	target: self.target,
1984	path: dest_s,
1985	bytes: dir_size(&dest).unwrap_or(0),
1986	},
1987	);
1988	Ok(())
1989	}
1990
1991	/// The all-targets-green gate: err unless every declared target OTHER than
1992	/// the one publishing has a latest `target_runs` row of `ok` for this
1993	/// `(app, version)`. A sibling with no run, a running run, or a failed
1994	/// latest run all block the publish, naming what is not green.
1995	fn assert_siblings_green(self: &Arc<Self>, declared: &[Target]) -> Result<()> {
1996	let me = self.clone();
1997	let (app_s, ver_s) = (self.app.to_string(), self.version.to_string());
1998	let rows: Vec<(String, String)> = self.rt.block_on(async move {
1999	sqlx::query_as(
2000	"SELECT target, status FROM target_runs tr
2001	WHERE app = ?1 AND version = ?2
2002	AND id = (SELECT MAX(id) FROM target_runs
2003	WHERE app = ?1 AND version = ?2 AND target = tr.target)",
2004	)
2005	.bind(app_s)
2006	.bind(ver_s)
2007	.fetch_all(&me.pool)
2008	.await
2009	.unwrap_or_default()
2010	});
2011	let status_of = \|t: &Target\| -> Option<String> {
2012	let key = t.to_string();
2013	rows.iter()
2014	.find(\|(name, _)\| name == &key)
2015	.map(\|(_, s)\| s.clone())
2016	};
2017	let not_green: Vec<String> = declared
2018	.iter()
2019	.filter(\|t\| **t != self.target) // the publishing target is the last mile
2020	.filter(\|t\| status_of(t).as_deref() != Some("ok"))
2021	.map(\|t\| format!("{t} ({})", status_of(t).unwrap_or_else(\|\| "no run".into())))
2022	.collect();
2023	anyhow::ensure!(
2024	not_green.is_empty(),
2025	"all-targets-green gate: refusing to publish {} {} — not green: {}",
2026	self.app,
2027	self.version,
2028	not_green.join(", "),
2029	);
2030	Ok(())
2031	}
2032
2033	fn publish(
2034	self: &Arc<Self>,
2035	channel: &str,
2036	app: &str,
2037	target: &str,
2038	version: &str,
2039	artifact: &str,
2040	meta: &Map,
2041	) -> Result<String> {
2042	// Never let a superseded build ship. This is the last and most important
2043	// cooperative-cancel checkpoint: even if a long-running step finished
2044	// after supersession, the artifact must not reach the backend.
2045	anyhow::ensure!(
2046	!self.is_cancelled(),
2047	"build superseded by a newer request; refusing to publish"
2048	);
2049	// Opt-in all-targets-green gate: refuse a partial release. Every OTHER
2050	// declared target of this (app, version) must have a successful latest
2051	// run before this one ships, so macOS can't publish while windows is red
2052	// or still building. The publishing target itself is the last mile (it
2053	// reached publish, so its steps passed) and is not required to be green
2054	// in the ledger yet.
2055	if let Some(declared) = self.all_green_required.clone() {
2056	self.assert_siblings_green(&declared)?;
2057	}
2058	let backend = self
2059	.ota
2060	.get(channel)
2061	.ok_or_else(\|\| anyhow::anyhow!("unknown publish channel `{channel}`"))?;
2062	let target: Target = target.parse().map_err(\|e: String\| anyhow::anyhow!(e))?;
2063	let version = Version::parse(version).map_err(\|e\| anyhow::anyhow!(e))?;
2064	let app = AppId::new(app);
2065
2066	// The backend must actually handle this target (e.g. the desktop updater
2067	// disclaims iOS) — otherwise publish would push an artifact through a
2068	// backend that does not support it.
2069	anyhow::ensure!(
2070	backend.supports(target),
2071	"publish channel `{channel}` does not support target {target}",
2072	);
2073
2074	// Monotonicity: never publish a version that is not strictly newer than
2075	// the latest already published for this (app, target, channel). Without
2076	// this an older build could republish over a live newer release. The
2077	// `releases` column is TEXT, so compare by parsed semver precedence
2078	// (Version: Ord), not lexically.
2079	{
2080	let (app_s, target_s, chan_s) =
2081	(app.to_string(), target.to_string(), channel.to_string());
2082	let me = self.clone();
2083	let latest: Option<Version> = self.rt.block_on(async move {
2084	let rows: Vec<(String,)> = sqlx::query_as(
2085	"SELECT version FROM releases WHERE app = ? AND target = ? AND channel = ?",
2086	)
2087	.bind(app_s)
2088	.bind(target_s)
2089	.bind(chan_s)
2090	.fetch_all(&me.pool)
2091	.await
2092	.unwrap_or_default();
2093	rows.into_iter()
2094	.filter_map(\|(v,)\| Version::parse(&v).ok())
2095	.max()
2096	});
2097	if let Some(latest) = latest {
2098	anyhow::ensure!(
2099	version > latest,
2100	"refusing to publish {app} {version} to `{channel}` ({target}): \
2101	not newer than the last published {latest}",
2102	);
2103	}
2104	}
2105
2106	// Step-success ledger (the Bento analogue of Sando's gate fail-closed),
2107	// minted as an unforgeable PublishAuthority. `backend.publish` cannot be
2108	// called without one, so the unverified/post-failure ship path is sealed
2109	// at the type level rather than guarded by a separate runtime check.
2110	let authority = {
2111	let failed = self.failed_steps.lock().unwrap();
2112	let gatekeeper = *self.gatekeeper_ok.lock().unwrap();
2113	PublishAuthority::prove(target, failed.as_slice(), gatekeeper)?
2114	};
2115	let notes = meta
2116	.get("notes")
2117	.and_then(\|v\| v.clone().into_string().ok())
2118	.unwrap_or_default();
2119	// Resolve the artifact relative to the collected dist dir if not absolute.
2120	let artifact_path = {
2121	let p = PathBuf::from(artifact);
2122	if p.is_absolute() {
2123	p
2124	} else {
2125	self.collect_dest(app.as_str(), &version.to_string())
2126	.join(artifact)
2127	}
2128	};
2129	let rel = Release {
2130	app: &app,
2131	target,
2132	version: &version,
2133	notes,
2134	};
2135	let receipt = backend
2136	.publish(&rel, &artifact_path, &authority)
2137	.with_context(\|\| format!("publish to `{channel}`"))?;
2138	// Record for idempotency / monotonicity. This write is CHECKED, not
2139	// fire-and-forget: a swallowed failure here would silently re-arm the
2140	// monotonicity guard (which reads this same table), letting an older
2141	// version republish over a live release. Concurrent same-(app,target)
2142	// publishers can't race the read-then-insert because the latest-wins slot
2143	// (state::ActiveSlot) serializes them and a superseded run is cancelled
2144	// before it reaches publish.
2145	// The artifact's hash, recorded so the release ledger says exactly which
2146	// bytes shipped. Prefer the digest computed at `collect`; fall back to
2147	// hashing the file now (an absolute-path artifact never routed through
2148	// `collect`). A hash failure must not fail an already-published release,
2149	// so degrade to NULL rather than erroring.
2150	let artifact_hash: Option<String> = artifact_path
2151	.file_name()
2152	.and_then(\|n\| n.to_str())
2153	.and_then(\|n\| self.artifact_hashes.lock().unwrap().get(n).cloned())
2154	.or_else(\|\| sha256_file(&artifact_path).ok());
2155	let me = self.clone();
2156	let (app_s, target_s, ver_s, chan_s) = (
2157	app.to_string(),
2158	target.to_string(),
2159	version.to_string(),
2160	channel.to_string(),
2161	);
2162	self.rt
2163	.block_on(async move {
2164	sqlx::query(
2165	"INSERT OR IGNORE INTO releases (app, target, version, channel, artifact_hash, published_at)
2166	VALUES (?, ?, ?, ?, ?, ?)",
2167	)
2168	.bind(app_s)
2169	.bind(target_s)
2170	.bind(ver_s)
2171	.bind(chan_s)
2172	.bind(artifact_hash)
2173	.bind(Self::now())
2174	.execute(&me.pool)
2175	.await
2176	})
2177	.context("recording release in the idempotency ledger (artifact published but ledger write failed)")?;
2178	events::emit(
2179	&self.events,
2180	Event::PublishOk {
2181	app: self.app.clone(),
2182	target: self.target,
2183	channel: channel.to_string(),
2184	},
2185	);
2186	Ok(receipt)
2187	}
2188	}
2189
2190	fn dir_size(p: &Path) -> Option<i64> {
2191	let mut total = 0i64;
2192	for entry in std::fs::read_dir(p).ok()? {
2193	let entry = entry.ok()?;
2194	let md = entry.metadata().ok()?;
2195	if md.is_file() {
2196	total += md.len() as i64;
2197	} else if md.is_dir() {
2198	// Recurse so a bundle dir (a `.app`) reports its real size, not ~0.
2199	total += dir_size(&entry.path()).unwrap_or(0);
2200	}
2201	}
2202	Some(total)
2203	}
2204
2205	/// macOS signing/notarization host functions. Thin wrappers over the right
2206	/// shell incantations, dispatched through the named host's executor. On the mac
2207	/// host (`transport = "agent"`) they run via the in-session `ops-agent`, the only
2208	/// context where codesign can use the Developer ID key (design §7 "THE WALL"); a
2209	/// plain SSH session cannot. Each is gated by the host's `sign` capability.
2210	fn register_macos_fns(engine: &mut Engine, ctx: &Arc<RecipeCtx>) {
2211	{
2212	let ctx = ctx.clone();
2213	engine.register_fn(
2214	"verify_gatekeeper",
2215	move \|host: &str, path: &str\| -> Result<bool, Box<EvalAltResult>> {
2216	// spctl has no JSON mode, so assess on-host and decide there,
2217	// emitting an unambiguous sentinel as the final line. We match the
2218	// sentinel rather than substring-hunting `source=Notarized...` in a
2219	// 2000-char tail: truncation only drops the front, so the sentinel
2220	// is always present, and it can't be spoofed by spctl's own prose.
2221	// The full assess output is still streamed to the step log.
2222	let q = ops_core::remote::sh_quote(path);
2223	let cmd = format!(
2224	"out=$(spctl --assess -vv --type install {q} 2>&1); printf '%s\\n' \"$out\"; \
2225	printf '%s' \"$out\" \| grep -q 'source=Notarized Developer ID' \
2226	&& echo BENTO_GATEKEEPER_OK \|\| echo BENTO_GATEKEEPER_FAIL",
2227	);
2228	let (_, tail) = ctx.run(host, &cmd).map_err(rhai_err)?;
2229	let accepted = tail.contains("BENTO_GATEKEEPER_OK");
2230	// Record the verdict for the publish gate. A rejection also
2231	// fails the step, so the matrix shows red and `publish` is barred
2232	// even if the recipe ignores the returned bool.
2233	*ctx.gatekeeper_ok.lock().unwrap() = Some(accepted);
2234	if !accepted {
2235	ctx.fail_current_step();
2236	}
2237	Ok(accepted)
2238	},
2239	);
2240	}
2241	{
2242	let ctx = ctx.clone();
2243	engine.register_fn(
2244	"codesign",
2245	move \|host: &str, identity: &str, path: &str\| -> Result<(), Box<EvalAltResult>> {
2246	let cmd = format!(
2247	"codesign --force --options runtime --timestamp --sign {} {}",
2248	ops_core::remote::sh_quote(identity),
2249	ops_core::remote::sh_quote(path),
2250	);
2251	let (code, _) = ctx.run(host, &cmd).map_err(rhai_err)?;
2252	if code != 0 {
2253	return Err(rhai_err("codesign failed"));
2254	}
2255	Ok(())
2256	},
2257	);
2258	}
2259	{
2260	let ctx = ctx.clone();
2261	engine.register_fn(
2262	"staple",
2263	move \|host: &str, path: &str\| -> Result<(), Box<EvalAltResult>> {
2264	let (code, _) = ctx
2265	.run(
2266	host,
2267	&format!("xcrun stapler staple {}", ops_core::remote::sh_quote(path)),
2268	)
2269	.map_err(rhai_err)?;
2270	if code != 0 {
2271	return Err(rhai_err("stapler failed"));
2272	}
2273	Ok(())
2274	},
2275	);
2276	}
2277	{
2278	let ctx = ctx.clone();
2279	engine.register_fn(
2280	"notarize",
2281	move \|host: &str, path: &str\| -> Result<String, Box<EvalAltResult>> {
2282	ctx.notarize(host, path).map_err(rhai_err)
2283	},
2284	);
2285	}
2286	{
2287	let ctx = ctx.clone();
2288	engine.register_fn(
2289	"keychain_open",
2290	move \|host: &str, name: &str\| -> Result<(), Box<EvalAltResult>> {
2291	// The full build-keychain lifecycle lives in dist/build-keychain.sh
2292	// (design §7); this drives it by name so the recipe stays short.
2293	let (code, _) = ctx
2294	.run(
2295	host,
2296	&format!(
2297	". ~/.tauri/passwords.env && ./dist/build-keychain.sh open {}",
2298	ops_core::remote::sh_quote(name)
2299	),
2300	)
2301	.map_err(rhai_err)?;
2302	if code != 0 {
2303	return Err(rhai_err("keychain_open failed"));
2304	}
2305	Ok(())
2306	},
2307	);
2308	}
2309	{
2310	let ctx = ctx.clone();
2311	engine.register_fn(
2312	"keychain_close",
2313	move \|host: &str, name: &str\| -> Result<(), Box<EvalAltResult>> {
2314	let _ = ctx.run(
2315	host,
2316	&format!(
2317	"./dist/build-keychain.sh close {}",
2318	ops_core::remote::sh_quote(name)
2319	),
2320	);
2321	Ok(())
2322	},
2323	);
2324	}
2325	}
2326
2327	impl RecipeCtx {
2328	/// `xcrun notarytool submit --wait` with bounded retry (the one flaky,
2329	/// network-bound step). Emits `NotarizeRetry` per attempt.
2330	fn notarize(self: &Arc<Self>, host: &str, path: &str) -> Result<String> {
2331	const MAX_ATTEMPTS: u32 = 3;
2332	let backoff = self
2333	.cfg
2334	.notarize_backoff_secs
2335	.map_or(std::time::Duration::from_secs(15), \|s\| {
2336	std::time::Duration::from_secs(s)
2337	});
2338	let cmd = format!(
2339	". ~/.tauri/passwords.env && xcrun notarytool submit {} \
2340	--key \"$NOTARY_KEY\" --key-id \"$NOTARY_KEY_ID\" --issuer \"$NOTARY_ISSUER\" \
2341	--wait --output-format json",
2342	ops_core::remote::sh_quote(path),
2343	);
2344	let mut last = String::new();
2345	for attempt in 1..=MAX_ATTEMPTS {
2346	let (code, tail) = self.run(host, &cmd)?;
2347	if code == 0 && notary_accepted(&tail) {
2348	return Ok(tail);
2349	}
2350	last = tail;
2351	if attempt < MAX_ATTEMPTS {
2352	events::emit(
2353	&self.events,
2354	Event::NotarizeRetry {
2355	app: self.app.clone(),
2356	target: self.target,
2357	attempt,
2358	reason: format!("exit {code}"),
2359	},
2360	);
2361	self.rt.block_on(tokio::time::sleep(backoff));
2362	}
2363	}
2364	anyhow::bail!("notarization failed after {MAX_ATTEMPTS} attempts: {last}")
2365	}
2366	}
2367
2368	/// True iff `notarytool --output-format json` output reports `status: Accepted`.
2369	/// Isolates the JSON object (`{`..`}`) from any shell-sourcing noise and reads
2370	/// the typed `status` field, rather than substring-matching `"status":"Accepted"`
2371	/// in a possibly-truncated tail — which could match the literal inside an error
2372	/// message or miss it across a whitespace variant. Fails closed: any parse or
2373	/// field miss returns false.
2374	fn notary_accepted(output: &str) -> bool {
2375	let (Some(start), Some(end)) = (output.find('{'), output.rfind('}')) else {
2376	return false;
2377	};
2378	if start > end {
2379	return false;
2380	}
2381	serde_json::from_str::<serde_json::Value>(&output[start..=end])
2382	.ok()
2383	.and_then(\|v\| {
2384	v.get("status")
2385	.and_then(\|s\| s.as_str())
2386	.map(\|s\| s.eq_ignore_ascii_case("accepted"))
2387	})
2388	.unwrap_or(false)
2389	}
2390
2391	#[cfg(test)]
2392	mod tests {
2393	use super::*;
2394
2395	/// Build the shared cross-crate bundle fixture under `root`.
2396	///
2397	/// A binary at the top and two files in a subdirectory — the shape a service
2398	/// that ships its migrations has, which is the case the flat walk used to
2399	/// get wrong.
2400	pub(crate) fn write_bundle_fixture(root: &Path) {
2401	std::fs::create_dir_all(root.join("migrations")).unwrap();
2402	std::fs::write(root.join("pom"), b"binary-bytes").unwrap();
2403	std::fs::write(root.join("migrations/001_init.sql"), b"create table a;").unwrap();
2404	std::fs::write(root.join("migrations/002_next.sql"), b"alter table a;").unwrap();
2405	}
2406
2407	/// The manifest text the fixture must produce, in BOTH crates.
2408	///
2409	/// Sando's `bundle::digest_dir` has the identical constant and the identical
2410	/// fixture. That is the whole point: bento writes this text into the artifact
2411	/// record, sando recomputes it from the bytes that arrive, and an artifact is
2412	/// refused when they differ. Two walks, one answer, pinned from both ends —
2413	/// if either crate's walk drifts, its own test fails and names the drift
2414	/// rather than a release failing intake for a bundle nothing is wrong with.
2415	pub(crate) const BUNDLE_FIXTURE_MANIFEST: &str = concat!(
2416	"e4c908e219c533fa7ad7ea1634398f9bf51637ba20717769ada545bab26d7368 migrations/001_init.sql\n",
2417	"b026fd51bae096b34672cefdb781b6585b13efb53bc301d50c305f422552a380 migrations/002_next.sql\n",
2418	"71227a7f160afca3fb3c39f448735886dda7bd366252580c2222fb87d4bb4d85 pom\n",
2419	);
2420
2421	/// The producer half of the contract above: what `collect` hashes, turned
2422	/// into a manifest, is exactly the text the verifier will recompute.
2423	///
2424	/// Nested files are included and addressed by relative path. Before this,
2425	/// `collect` listed only the top level, so `migrations/` contributed nothing
2426	/// to the manifest while sando's walker hashed both files in it — and the
2427	/// honest bundle was refused for a manifest that had omitted them.
2428	#[test]
2429	fn a_collected_bundle_manifests_exactly_as_the_verifier_will_read_it() {
2430	let dir = tempfile::tempdir().unwrap();
2431	write_bundle_fixture(dir.path());
2432
2433	let files = collected_files(dir.path()).unwrap();
2434	assert_eq!(
2435	files.iter().map(\|(r, _)\| r.as_str()).collect::<Vec<_>>(),
2436	vec!["migrations/001_init.sql", "migrations/002_next.sql", "pom"],
2437	"recursive, relative, sorted"
2438	);
2439
2440	let hashes: Vec<(String, String)> = files
2441	.into_iter()
2442	.map(\|(rel, path)\| (rel, sha256_file(&path).unwrap()))
2443	.collect();
2444	let manifest = ops_artifact::Manifest::new(hashes).unwrap();
2445	assert_eq!(manifest.to_text(), BUNDLE_FIXTURE_MANIFEST);
2446	}
2447
2448	/// A symlink is neither followed nor named. Following one would let bytes
2449	/// from outside the bundle into its identity; naming it would put a path in
2450	/// the manifest the verifier does not hash, which reads as a corrupt bundle.
2451	#[test]
2452	#[cfg(unix)]
2453	fn a_symlink_in_the_collect_dir_is_not_part_of_the_bundle() {
2454	let dir = tempfile::tempdir().unwrap();
2455	write_bundle_fixture(dir.path());
2456	let outside = dir.path().join("..").join("secret.env");
2457	std::fs::write(&outside, b"TOKEN=1").ok();
2458	std::os::unix::fs::symlink(&outside, dir.path().join("link.env")).unwrap();
2459
2460	let files = collected_files(dir.path()).unwrap();
2461	assert!(
2462	!files.iter().any(\|(rel, _)\| rel.contains("link.env")),
2463	"{files:?}"
2464	);
2465	}
2466
2467	/// Run 3 S1: once the cooperative cancel flag is set (a newer build
2468	/// superseded this run), a step boundary refuses to proceed — the blocking
2469	/// recipe stops at the next `step()` instead of running on and publishing.
2470	#[tokio::test]
2471	async fn begin_step_bails_when_cancelled() {
2472	let dir = tempfile::tempdir().unwrap();
2473	let cfg = Arc::new(Config::for_tests(dir.path()));
2474	let pool = crate::db::open(&cfg.db_path).await.unwrap();
2475	let cancel = Arc::new(AtomicBool::new(true));
2476	let ctx = Arc::new(RecipeCtx::new(
2477	AppId::new("demo"),
2478	Version::parse("0.1.0").unwrap(),
2479	"linux/x86_64".parse().unwrap(),
2480	"fw13".into(),
2481	"local".into(),
2482	"v0.1.0".into(),
2483	"/tmp".into(),
2484	vec![],
2485	Kind::App,
2486	1,
2487	Arc::new(std::collections::HashMap::new()),
2488	Arc::new(std::collections::HashMap::new()),
2489	None,
2490	pool,
2491	crate::events::channel(),
2492	cfg,
2493	Arc::new(OtaRegistry::standard("https://makenot.work")),
2494	tokio::runtime::Handle::current(),
2495	cancel.clone(),
2496	None,
2497	));
2498	// Cancelled: begin_step refuses before touching the DB (the ensure! is
2499	// ahead of any block_on, so this is safe to call from the async test).
2500	let err = ctx.begin_step(Step::Build).unwrap_err();
2501	assert!(err.to_string().contains("supersede"), "got: {err}");
2502	assert!(ctx.is_cancelled());
2503	}
2504
2505	/// `feature_flags()` returns a whole flag or nothing at all. An app with
2506	/// no declared features must not yield a bare `--features`, which would
2507	/// swallow the next word of the build command as its argument.
2508	#[tokio::test]
2509	async fn feature_flags_renders_whole_flag_or_empty() {
2510	async fn flags_for(features: Vec<String>) -> String {
2511	let dir = tempfile::tempdir().unwrap();
2512	let cfg = Arc::new(Config::for_tests(dir.path()));
2513	let pool = crate::db::open(&cfg.db_path).await.unwrap();
2514	let ctx = Arc::new(RecipeCtx::new(
2515	AppId::new("demo"),
2516	Version::parse("0.1.0").unwrap(),
2517	"linux/x86_64".parse().unwrap(),
2518	"fw13".into(),
2519	"local".into(),
2520	"v0.1.0".into(),
2521	"/tmp".into(),
2522	features,
2523	Kind::App,
2524	1,
2525	Arc::new(std::collections::HashMap::new()),
2526	Arc::new(std::collections::HashMap::new()),
2527	None,
2528	pool,
2529	crate::events::channel(),
2530	cfg,
2531	Arc::new(OtaRegistry::standard("https://makenot.work")),
2532	tokio::runtime::Handle::current(),
2533	Arc::new(AtomicBool::new(false)),
2534	None,
2535	));
2536	let engine = build_engine(&ctx);
2537	engine.eval::<String>("feature_flags()").unwrap()
2538	}
2539
2540	assert_eq!(flags_for(vec![]).await, "");
2541	assert_eq!(
2542	flags_for(vec!["supernote".into()]).await,
2543	"--features supernote"
2544	);
2545	assert_eq!(
2546	flags_for(vec!["supernote".into(), "extra".into()]).await,
2547	"--features supernote,extra"
2548	);
2549	}
2550
2551	/// `repo()` answers for the host this target builds on, not for the daemon.
2552	///
2553	/// This is what lets a Windows recipe call `repo()` and `checkout_sha(h)`
2554	/// instead of hard-coding `C:/Users/me/...` — and hard-coding it is what
2555	/// kept those recipes off the release-tag pin, since `checkout_sha` builds
2556	/// its git commands from the app's path and takes no override.
2557	#[tokio::test]
2558	async fn repo_resolves_per_build_host() {
2559	async fn repo_on(build_host: &str) -> String {
2560	let dir = tempfile::tempdir().unwrap();
2561	let cfg = Arc::new(Config::for_tests(dir.path()));
2562	let pool = crate::db::open(&cfg.db_path).await.unwrap();
2563	let ctx = Arc::new(
2564	RecipeCtx::new(
2565	AppId::new("demo"),
2566	Version::parse("0.1.0").unwrap(),
2567	"linux/x86_64".parse().unwrap(),
2568	build_host.into(),
2569	"local".into(),
2570	"v0.1.0".into(),
2571	"~/Code/Apps/demo".into(),
2572	vec![],
2573	Kind::App,
2574	1,
2575	Arc::new(std::collections::HashMap::new()),
2576	Arc::new(std::collections::HashMap::new()),
2577	None,
2578	pool,
2579	crate::events::channel(),
2580	cfg,
2581	Arc::new(OtaRegistry::standard("https://makenot.work")),
2582	tokio::runtime::Handle::current(),
2583	Arc::new(AtomicBool::new(false)),
2584	None,
2585	)
2586	.with_repo_by_host(HashMap::from([(
2587	"windows-x86".to_string(),
2588	"C:/Users/me/Code/Apps/demo".to_string(),
2589	)])),
2590	);
2591	build_engine(&ctx).eval::<String>("repo()").unwrap()
2592	}
2593
2594	assert_eq!(repo_on("windows-x86").await, "C:/Users/me/Code/Apps/demo");
2595	assert_eq!(repo_on("fw13").await, "~/Code/Apps/demo");
2596	}
2597
2598	/// `secret(key)` reads a file under `secrets_root`, trims its trailing
2599	/// newline (the shape of a here-doc'd token file), and refuses any key that
2600	/// could escape the root. Covers the host-fn registered in `build_engine`.
2601	#[tokio::test]
2602	async fn secret_reads_under_root_and_blocks_traversal() {
2603	let dir = tempfile::tempdir().unwrap();
2604	let cfg = Config::for_tests(dir.path());
2605	// Seed a secret and one in a nested subdir; a trailing newline that the
2606	// read must strip.
2607	std::fs::create_dir_all(&cfg.secrets_root).unwrap();
2608	std::fs::write(cfg.secrets_root.join("token"), "s3cr3t\n").unwrap();
2609	std::fs::create_dir_all(cfg.secrets_root.join("app")).unwrap();
2610	std::fs::write(cfg.secrets_root.join("app").join("key"), "nested").unwrap();
2611	// Plant a file OUTSIDE the root that a traversal key would reach.
2612	std::fs::write(dir.path().join("outside"), "leak").unwrap();
2613
2614	let cfg = Arc::new(cfg);
2615	let pool = crate::db::open(&cfg.db_path).await.unwrap();
2616	let ctx = Arc::new(RecipeCtx::new(
2617	AppId::new("demo"),
2618	Version::parse("0.1.0").unwrap(),
2619	"linux/x86_64".parse().unwrap(),
2620	"fw13".into(),
2621	"local".into(),
2622	"v0.1.0".into(),
2623	"/tmp".into(),
2624	vec![],
2625	Kind::App,
2626	1,
2627	Arc::new(std::collections::HashMap::new()),
2628	Arc::new(std::collections::HashMap::new()),
2629	None,
2630	pool,
2631	crate::events::channel(),
2632	cfg,
2633	Arc::new(OtaRegistry::standard("https://makenot.work")),
2634	tokio::runtime::Handle::current(),
2635	Arc::new(AtomicBool::new(false)),
2636	None,
2637	));
2638	let engine = build_engine(&ctx);
2639
2640	// Happy path: read + trim.
2641	assert_eq!(
2642	engine.eval::<String>(r#"secret("token")"#).unwrap(),
2643	"s3cr3t"
2644	);
2645	// A multi-segment relative key is allowed.
2646	assert_eq!(
2647	engine.eval::<String>(r#"secret("app/key")"#).unwrap(),
2648	"nested"
2649	);
2650
2651	// Traversal, absolute paths, and empty keys are refused BEFORE any read,
2652	// so the file one `..` above the root is never disclosed.
2653	for bad in [
2654	r#"secret("../outside")"#,
2655	r#"secret("/etc/passwd")"#,
2656	r#"secret("")"#,
2657	] {
2658	let err = engine.eval::<String>(bad).unwrap_err().to_string();
2659	assert!(
2660	err.contains("relative path under secrets_root"),
2661	"`{bad}` should hit the traversal guard, got: {err}"
2662	);
2663	}
2664	// A missing key surfaces the filesystem error, not a panic, and does not
2665	// trip the traversal guard (it is a legitimate relative path).
2666	let err = engine
2667	.eval::<String>(r#"secret("nope")"#)
2668	.unwrap_err()
2669	.to_string();
2670	assert!(err.contains("secret `nope`"), "got: {err}");
2671	}
2672
2673	/// The two failures that actually shipped, as regression cases.
2674	#[test]
2675	fn preflight_catches_a_dead_repository_url() {
2676	// pter 0.1.0: repository pointed at a URL that does not exist. It
2677	// published clean and the link is now permanent for that version.
2678	let meta = CrateMeta {
2679	name: "pter".into(),
2680	version: "0.1.0".into(),
2681	repository: Some("https://github.com/maxjacobson/pter".into()),
2682	description: Some("d".into()),
2683	licensed: true,
2684	};
2685	let problems = crate_publish_problems(&meta, false, &[], true);
2686	assert_eq!(problems.len(), 1, "{problems:?}");
2687	assert!(
2688	problems[0].contains("not publicly clonable"),
2689	"{problems:?}"
2690	);
2691
2692	// Same metadata, reachable URL: nothing to report.
2693	assert!(crate_publish_problems(&meta, true, &[], true).is_empty());
2694	}
2695
2696	#[test]
2697	fn preflight_requires_the_fields_crates_io_bakes_in() {
2698	let bare = CrateMeta {
2699	name: "x".into(),
2700	version: "0.1.0".into(),
2701	repository: None,
2702	description: None,
2703	licensed: false,
2704	};
2705	let problems = crate_publish_problems(&bare, false, &[], true);
2706	assert_eq!(problems.len(), 3, "{problems:?}");
2707	assert!(problems.iter().any(\|p\| p.contains("repository")));
2708	assert!(problems.iter().any(\|p\| p.contains("description")));
2709	assert!(problems.iter().any(\|p\| p.contains("license")));
2710	}
2711
2712	// A library's verify is a crate preflight, not a Gatekeeper check on a
2713	// signed bundle. Gating it on `gatekeeper` asked a Linux host for a macOS
2714	// code-signing capability it can never hold, so the step was denied before
2715	// it ran a command; the denial then surfaced as "no crates.io credentials",
2716	// which is not what went wrong. The only way to satisfy the old gate was to
2717	// declare the capability falsely in the topology.
2718	#[test]
2719	fn a_library_verify_is_not_gated_on_gatekeeper() {
2720	assert_eq!(
2721	action_for(Step::Verify, Kind::Library),
2722	Action::Build,
2723	"a crate preflight runs the build toolchain; that is what it needs",
2724	);
2725	assert_eq!(
2726	action_for(Step::Verify, Kind::App),
2727	Action::Observe(ObserveKind::Custom("gatekeeper".into())),
2728	"an app's verify still proves the bundle is signed and notarized",
2729	);
2730	}
2731
2732	// The capability the default host grant actually carries. Without this the
2733	// fix above is only true by inspection.
2734	#[test]
2735	fn a_default_host_can_run_a_library_verify_and_not_an_app_one() {
2736	let caps =
2737	ops_exec::CapabilitySet::from_tokens(["build", "package"], ["build-log", "artifact"]);
2738	assert!(caps.permits(&action_for(Step::Verify, Kind::Library)));
2739	assert!(!caps.permits(&action_for(Step::Verify, Kind::App)));
2740	}
2741
2742	// Every other step is a property of the step alone; verify is the one that
2743	// depends on what is being released.
2744	#[test]
2745	fn no_other_step_changes_with_the_kind() {
2746	for step in [
2747	Step::Checkout,
2748	Step::Prebuild,
2749	Step::Build,
2750	Step::Sign,
2751	Step::Notarize,
2752	Step::Staple,
2753	Step::Package,
2754	Step::Publish,
2755	Step::Collect,
2756	] {
2757	assert_eq!(
2758	action_for(step, Kind::App),
2759	action_for(step, Kind::Library),
2760	"{step:?} should not depend on the kind",
2761	);
2762	}
2763	}
2764
2765	#[test]
2766	fn preflight_rejects_republishing_the_same_version() {
2767	let meta = CrateMeta {
2768	name: "makeover".into(),
2769	version: "0.10.0".into(),
2770	repository: Some("https://git.sr.ht/~maxmj/makeover".into()),
2771	description: Some("d".into()),
2772	licensed: true,
2773	};
2774	let problems =
2775	crate_publish_problems(&meta, true, &["0.9.0".into(), "0.10.0".into()], true);
2776	assert_eq!(problems.len(), 1, "{problems:?}");
2777	assert!(problems[0].contains("already published"), "{problems:?}");
2778
2779	// An unreleased version against the same history is fine.
2780	let mut next = meta.clone();
2781	next.version = "0.11.0".into();
2782	assert!(crate_publish_problems(&next, true, &["0.10.0".into()], true).is_empty());
2783	}
2784
2785	/// Missing credentials must surface at preflight, not at the upload. The
2786	/// publish step is the irreversible one and runs last, after a full build
2787	/// and verify; discovering there that cargo cannot authenticate wastes the
2788	/// whole run.
2789	#[test]
2790	fn preflight_reports_missing_credentials_up_front() {
2791	let meta = CrateMeta {
2792	name: "makeover".into(),
2793	version: "0.11.0".into(),
2794	repository: Some("https://git.sr.ht/~maxmj/makeover".into()),
2795	description: Some("d".into()),
2796	licensed: true,
2797	};
2798	// Metadata is perfect; only the token is absent.
2799	let problems = crate_publish_problems(&meta, true, &[], false);
2800	assert_eq!(problems.len(), 1, "{problems:?}");
2801	assert!(problems[0].contains("credentials"), "{problems:?}");
2802	assert!(
2803	problems[0].contains("cargo login"),
2804	"should say how to fix it"
2805	);
2806
2807	// Present: nothing to report.
2808	assert!(crate_publish_problems(&meta, true, &[], true).is_empty());
2809	}
2810
2811	#[test]
2812	fn crate_meta_reads_cargo_metadata_json() {
2813	let raw = r#"{"packages":[{"name":"makeover","version":"0.10.0",
2814	"repository":"https://git.sr.ht/~maxmj/makeover","description":"themes",
2815	"license":"MIT"}]}"#;
2816	let m = crate_meta_from_json(raw).unwrap();
2817	assert_eq!(m.name, "makeover");
2818	assert_eq!(m.version, "0.10.0");
2819	assert!(m.licensed);
2820	assert_eq!(
2821	m.repository.as_deref(),
2822	Some("https://git.sr.ht/~maxmj/makeover")
2823	);
2824
2825	// license_file alone also counts as licensed; empty strings do not
2826	// count as present.
2827	let lf = r#"{"packages":[{"name":"x","version":"0.1.0","license":"",
2828	"license_file":"LICENSE","description":""}]}"#;
2829	let m = crate_meta_from_json(lf).unwrap();
2830	assert!(m.licensed);
2831	assert!(m.description.is_none());
2832	}
2833
2834	/// Reads the ambient `HOME` rather than setting one. `set_var` is
2835	/// process-global and unsynchronized, so a test that overwrote HOME changed
2836	/// it for every other test in the binary — which is what silently disabled
2837	/// `topology::live_config_smoke` (it skips when `$HOME/.config/bento` is
2838	/// absent, and `/home/test` always is).
2839	#[test]
2840	fn expand_tilde_handles_home() {
2841	let home = PathBuf::from(std::env::var("HOME").expect("HOME is set"));
2842	assert_eq!(expand_tilde("~/Code/x"), home.join("Code/x"));
2843	assert_eq!(expand_tilde("/abs/path"), PathBuf::from("/abs/path"));
2844	}
2845
2846	// ---- artifact resolution (the M3 silent-`sh` fix) ----
2847
2848	#[test]
2849	fn resolve_artifact_match_wants_exactly_one() {
2850	// Exactly one match: the path, trimmed of the listing's line noise.
2851	assert_eq!(
2852	resolve_artifact_match(" /d/App.AppImage \n", "*.AppImage", true).unwrap(),
2853	"/d/App.AppImage"
2854	);
2855	}
2856
2857	#[test]
2858	fn resolve_artifact_match_zero_depends_on_required() {
2859	// Required + zero matches is the case the old empty-string guard caught;
2860	// keep failing it.
2861	let err = resolve_artifact_match("", "*.dmg", true).unwrap_err();
2862	assert!(err.to_string().contains("no artifact matched"), "{err}");
2863	// Optional + zero matches resolves to empty (recipe skips the collect).
2864	assert_eq!(
2865	resolve_artifact_match("\n \n", "*.deb", false).unwrap(),
2866	""
2867	);
2868	}
2869
2870	#[test]
2871	fn resolve_artifact_match_rejects_ambiguous() {
2872	// Two matches must throw rather than silently pick one — this is the
2873	// stale-newest-mtime hole the audit flagged. Applies even when optional.
2874	for required in [true, false] {
2875	let err =
2876	resolve_artifact_match("/d/old.deb\n/d/new.deb\n", "*.deb", required).unwrap_err();
2877	let msg = err.to_string();
2878	assert!(msg.contains("ambiguous"), "{msg}");
2879	assert!(msg.contains("old.deb") && msg.contains("new.deb"), "{msg}");
2880	}
2881	}
2882
2883	#[test]
2884	fn ensure_glob_safe_allows_paths_bars_commands() {
2885	// Path and wildcard characters pass.
2886	assert!(ensure_glob_safe("~/Code/app/dist/*.AppImage").is_ok());
2887	assert!(ensure_glob_safe("/t/App_1.2.3-x86_64.dmg").is_ok());
2888	// A command substitution or separator does not.
2889	for bad in ["*.dmg; rm -rf /", "$(evil)", "a\|b", "a b"] {
2890	assert!(ensure_glob_safe(bad).is_err(), "should reject {bad:?}");
2891	}
2892	}
2893
2894	// ---- version resolution ----
2895
2896	#[test]
2897	fn version_from_tauri_json_reads_version() {
2898	assert_eq!(
2899	version_from_tauri_json(r#"{"version":"0.4.2"}"#).unwrap(),
2900	"0.4.2"
2901	);
2902	assert!(version_from_tauri_json(r#"{"productName":"X"}"#).is_err());
2903	}
2904
2905	#[test]
2906	fn version_from_cargo_toml_prefers_package_then_workspace() {
2907	// A leaf crate's [package].version.
2908	assert_eq!(
2909	version_from_cargo_toml("[package]\nname = \"x\"\nversion = \"0.5.0\"\n").unwrap(),
2910	"0.5.0"
2911	);
2912	// A workspace that sets [workspace.package].version.
2913	assert_eq!(
2914	version_from_cargo_toml("[workspace.package]\nversion = \"1.2.3\"\n").unwrap(),
2915	"1.2.3"
2916	);
2917	// No version anywhere -> error, not a panic.
2918	assert!(version_from_cargo_toml("[workspace]\nmembers = []\n").is_err());
2919	}
2920
2921	#[test]
2922	fn version_from_repo_default_and_explicit_paths() {
2923	let tmp = tempfile::tempdir().unwrap();
2924	let root = tmp.path();
2925
2926	// Tauri app: default path reads src-tauri/tauri.conf.json.
2927	let tauri = root.join("tauri");
2928	std::fs::create_dir_all(tauri.join("src-tauri")).unwrap();
2929	std::fs::write(
2930	tauri.join("src-tauri/tauri.conf.json"),
2931	r#"{"version":"0.4.2"}"#,
2932	)
2933	.unwrap();
2934	assert_eq!(
2935	version_from_repo(tauri.to_str().unwrap(), None)
2936	.unwrap()
2937	.to_string(),
2938	"0.4.2"
2939	);
2940
2941	// Workspace egui app: no tauri.conf.json, explicit version_path at a member crate.
2942	let ws = root.join("ws");
2943	std::fs::create_dir_all(ws.join("crates/app")).unwrap();
2944	std::fs::write(
2945	ws.join("Cargo.toml"),
2946	"[workspace]\nmembers = [\"crates/app\"]\n",
2947	)
2948	.unwrap();
2949	std::fs::write(
2950	ws.join("crates/app/Cargo.toml"),
2951	"[package]\nname = \"app\"\nversion = \"0.5.0\"\n",
2952	)
2953	.unwrap();
2954	assert_eq!(
2955	version_from_repo(ws.to_str().unwrap(), Some("crates/app/Cargo.toml"))
2956	.unwrap()
2957	.to_string(),
2958	"0.5.0"
2959	);
2960	}
2961
2962	// ---- version-source cross-check (drift preflight) ----
2963
2964	fn ver(s: &str) -> Version {
2965	Version::parse(s).unwrap()
2966	}
2967
2968	#[test]
2969	fn version_consistency_passes_when_all_sources_agree() {
2970	let tmp = tempfile::tempdir().unwrap();
2971	let repo = tmp.path();
2972	std::fs::create_dir_all(repo.join("src-tauri")).unwrap();
2973	std::fs::write(
2974	repo.join("src-tauri/tauri.conf.json"),
2975	r#"{"version":"0.5.0"}"#,
2976	)
2977	.unwrap();
2978	std::fs::write(
2979	repo.join("Cargo.toml"),
2980	"[package]\nname = \"app\"\nversion = \"0.5.0\"\n",
2981	)
2982	.unwrap();
2983	check_version_consistency(repo.to_str().unwrap(), None, &ver("0.5.0")).unwrap();
2984	}
2985
2986	#[test]
2987	fn version_consistency_flags_tauri_vs_cargo_drift() {
2988	// The concrete finding: tauri.conf.json bumped to 0.5.0 but the root
2989	// Cargo.toml left at 0.4.0. version_from_repo (one file) would miss it.
2990	let tmp = tempfile::tempdir().unwrap();
2991	let repo = tmp.path();
2992	std::fs::create_dir_all(repo.join("src-tauri")).unwrap();
2993	std::fs::write(
2994	repo.join("src-tauri/tauri.conf.json"),
2995	r#"{"version":"0.5.0"}"#,
2996	)
2997	.unwrap();
2998	std::fs::write(
2999	repo.join("Cargo.toml"),
3000	"[package]\nname = \"app\"\nversion = \"0.4.0\"\n",
3001	)
3002	.unwrap();
3003	let err =
3004	check_version_consistency(repo.to_str().unwrap(), None, &ver("0.5.0")).unwrap_err();
3005	let msg = format!("{err:#}");
3006	assert!(msg.contains("Cargo.toml says 0.4.0"), "{msg}");
3007	}
3008
3009	#[test]
3010	fn version_consistency_flags_explicit_version_the_repo_does_not_reflect() {
3011	let tmp = tempfile::tempdir().unwrap();
3012	let repo = tmp.path();
3013	std::fs::create_dir_all(repo.join("src-tauri")).unwrap();
3014	std::fs::write(
3015	repo.join("src-tauri/tauri.conf.json"),
3016	r#"{"version":"0.5.0"}"#,
3017	)
3018	.unwrap();
3019	let err =
3020	check_version_consistency(repo.to_str().unwrap(), None, &ver("9.9.9")).unwrap_err();
3021	assert!(format!("{err:#}").contains("building 9.9.9"));
3022	}
3023
3024	#[test]
3025	fn version_consistency_single_source_never_invents_drift() {
3026	// A virtual-workspace root Cargo.toml (no version) alongside the member
3027	// crate the version_path points at: only one real source, so no drift.
3028	let tmp = tempfile::tempdir().unwrap();
3029	let repo = tmp.path();
3030	std::fs::create_dir_all(repo.join("crates/app")).unwrap();
3031	std::fs::write(
3032	repo.join("Cargo.toml"),
3033	"[workspace]\nmembers = [\"crates/app\"]\n",
3034	)
3035	.unwrap();
3036	std::fs::write(
3037	repo.join("crates/app/Cargo.toml"),
3038	"[package]\nname = \"app\"\nversion = \"0.5.0\"\n",
3039	)
3040	.unwrap();
3041	check_version_consistency(
3042	repo.to_str().unwrap(),
3043	Some("crates/app/Cargo.toml"),
3044	&ver("0.5.0"),
3045	)
3046	.unwrap();
3047	}
3048
3049	// ---- artifact filename version assertion + hashing ----
3050
3051	#[test]
3052	fn versions_in_filename_extracts_only_real_semvers() {
3053	assert_eq!(
3054	versions_in_filename("GoingsOn_0.5.0_aarch64.dmg"),
3055	vec![ver("0.5.0")]
3056	);
3057	assert_eq!(
3058	versions_in_filename("AudioFiles-0.4.0-x86_64.AppImage"),
3059	vec![ver("0.4.0")]
3060	);
3061	// No three-part token ⇒ nothing (an updater manifest, a bare signature).
3062	assert!(versions_in_filename("latest.json").is_empty());
3063	assert!(versions_in_filename("app.sig").is_empty());
3064	}
3065
3066	#[test]
3067	fn assert_artifact_version_rejects_a_stale_artifact() {
3068	// The 0.4.0 file sitting in the output dir against a 0.5.0 build.
3069	let err =
3070	assert_artifact_version("AudioFiles-0.4.0-x86_64.AppImage", &ver("0.5.0")).unwrap_err();
3071	assert!(format!("{err:#}").contains("stale artifact"), "{err:#}");
3072	// The matching version passes, and a versionless file is not asserted.
3073	assert_artifact_version("GoingsOn_0.5.0_aarch64.dmg", &ver("0.5.0")).unwrap();
3074	assert_artifact_version("latest.json", &ver("0.5.0")).unwrap();
3075	}
3076
3077	/// The comparison that decides whether a binary can exec on the box that is
3078	/// about to be restarted onto it. Both sides are parsed out of text a tool
3079	/// printed, so both parsers are worth pinning: fw13 tracks a newer glibc
3080	/// than the Ubuntu 24.04 host in Hetzner, and getting this backwards means a
3081	/// dead unit rather than a failed step.
3082	#[test]
3083	fn glibc_versions_parse_from_what_the_tools_actually_print() {
3084	// `objdump -T \| grep -o 'GLIBC_[0-9.]*'` output: highest wins, and the
3085	// comparison is numeric (2.9 must not beat 2.34 lexically).
3086	let objdump = "GLIBC_2.2.5\nGLIBC_2.34\nGLIBC_2.9\nGLIBC_2.17\n";
3087	assert_eq!(max_glibc_symbol(objdump), Some((2, 34)));
3088	// A static binary references none: nothing to check.
3089	assert_eq!(max_glibc_symbol(""), None);
3090
3091	// `ldd --version` first line, however the distro decorates it.
3092	assert_eq!(
3093	glibc_from_ldd("ldd (Ubuntu GLIBC 2.39-0ubuntu8.8) 2.39\nCopyright...\n"),
3094	Some((2, 39))
3095	);
3096	assert_eq!(
3097	glibc_from_ldd("ldd (GNU libc) 2.41\nCopyright (C) 2025\n"),
3098	Some((2, 41))
3099	);
3100	assert_eq!(glibc_from_ldd(""), None);
3101	}
3102
3103	/// A binary needing MORE than the host has is the failure this check exists
3104	/// for; equal and less are both fine (glibc symbol versioning is backward
3105	/// compatible, so an older requirement runs on a newer host).
3106	#[test]
3107	fn glibc_requirement_is_satisfied_by_equal_or_newer_only() {
3108	let needs = max_glibc_symbol("GLIBC_2.41").unwrap();
3109	assert!(needs > glibc_from_ldd("ldd (Ubuntu GLIBC 2.39) 2.39").unwrap());
3110	assert!(needs <= glibc_from_ldd("ldd (GNU libc) 2.41").unwrap());
3111	assert!(needs <= glibc_from_ldd("ldd (GNU libc) 2.42").unwrap());
3112	assert!(needs <= glibc_from_ldd("ldd (GNU libc) 3.0").unwrap());
3113	}
3114
3115	/// Every deploy host function fails with the app's KIND as the reason when
3116	/// there is no destination, rather than with a missing-host error from
3117	/// somewhere deeper. A recipe calling `deploy()` on a library is a recipe
3118	/// written against the wrong kind, and the message should say so.
3119	#[tokio::test]
3120	async fn deploy_host_fns_explain_a_missing_destination_by_kind() {
3121	let dir = tempfile::tempdir().unwrap();
3122	let cfg = Arc::new(Config::for_tests(dir.path()));
3123	let pool = crate::db::open(&cfg.db_path).await.unwrap();
3124	let ctx = Arc::new(RecipeCtx::new(
3125	AppId::new("demo"),
3126	Version::parse("0.1.0").unwrap(),
3127	"linux/x86_64".parse().unwrap(),
3128	"fw13".into(),
3129	"local".into(),
3130	"v0.1.0".into(),
3131	"/tmp".into(),
3132	vec![],
3133	Kind::Library,
3134	1,
3135	Arc::new(std::collections::HashMap::new()),
3136	Arc::new(std::collections::HashMap::new()),
3137	None,
3138	pool,
3139	crate::events::channel(),
3140	cfg,
3141	Arc::new(OtaRegistry::standard("https://makenot.work")),
3142	tokio::runtime::Handle::current(),
3143	Arc::new(AtomicBool::new(false)),
3144	None,
3145	));
3146	let engine = build_engine(&ctx);
3147	for call in [
3148	"deploy_host()",
3149	"service_name()",
3150	"install_path()",
3151	"health_url()",
3152	r#"deploy("/tmp/x")"#,
3153	] {
3154	let err = engine.eval::<String>(call).unwrap_err().to_string();
3155	assert!(
3156	err.contains("library") && err.contains("no deploy destination"),
3157	"`{call}` must fail on the kind, got: {err}"
3158	);
3159	}
3160	}
3161
3162	/// A service host is addressed on the DEPLOY plane whatever step is open.
3163	///
3164	/// The subtle one. Actions are normally derived from the step, which is
3165	/// right for a build host — the step is what that host is being asked to do.
3166	/// A service host is granted `deploy`/`restart` and must never be granted
3167	/// `build`, so the same rule would have `glibc_check` ask it for `build`
3168	/// during a `verify` step and get denied for a reason unrelated to what was
3169	/// attempted. `verify` is the step that check belongs in, so without this
3170	/// routing the glibc gate cannot run at all.
3171	#[tokio::test]
3172	async fn a_service_host_is_addressed_on_the_deploy_plane_in_any_step() {
3173	let dir = tempfile::tempdir().unwrap();
3174	let cfg = Arc::new(Config::for_tests(dir.path()));
3175	let pool = crate::db::open(&cfg.db_path).await.unwrap();
3176	sqlx::query(
3177	"INSERT INTO builds (id, app, version, status, created_at) \
3178	VALUES (1, 'demo', '0.1.0', 'running', '2026-07-30T00:00:00Z')",
3179	)
3180	.execute(&pool)
3181	.await
3182	.unwrap();
3183	sqlx::query(
3184	"INSERT INTO target_runs (id, build_id, app, version, target, status, started_at) \
3185	VALUES (1, 1, 'demo', '0.1.0', 'linux/x86_64', 'running', '2026-07-30T00:00:00Z')",
3186	)
3187	.execute(&pool)
3188	.await
3189	.unwrap();
3190
3191	let deploy = crate::topology::DeployTarget {
3192	target: "linux/x86_64".parse().unwrap(),
3193	host: "local".into(),
3194	port: None,
3195	install_path: "/usr/local/bin/demo".into(),
3196	service: "demo.service".into(),
3197	health_url: None,
3198	};
3199	let mut execs: crate::state::ExecutorMap = std::collections::HashMap::new();
3200	execs.insert("local".into(), crate::state::build_deploy_executor(&deploy));
3201	// The service host's grant is exactly deploy + restart. If this ever
3202	// widens to include `build`, the test below stops proving anything.
3203	assert!(!execs["local"].capabilities().permits(&Action::Build));
3204	assert!(execs["local"].capabilities().permits(&Action::Deploy));
3205
3206	let ctx = Arc::new(RecipeCtx::new(
3207	AppId::new("demo"),
3208	Version::parse("0.1.0").unwrap(),
3209	"linux/x86_64".parse().unwrap(),
3210	"fw13".into(),
3211	"local".into(),
3212	"v0.1.0".into(),
3213	"/tmp".into(),
3214	vec![],
3215	Kind::Service,
3216	1,
3217	Arc::new(execs),
3218	Arc::new(std::collections::HashMap::new()),
3219	Some(deploy),
3220	pool,
3221	crate::events::channel(),
3222	cfg,
3223	Arc::new(OtaRegistry::standard("https://makenot.work")),
3224	tokio::runtime::Handle::current(),
3225	Arc::new(AtomicBool::new(false)),
3226	None,
3227	));
3228
3229	let ctx_blocking = ctx.clone();
3230	tokio::task::spawn_blocking(move \|\| {
3231	// `verify` on a service derives Action::Build — which the service
3232	// host does not grant. The command must still run.
3233	ctx_blocking.begin_step(Step::Verify).unwrap();
3234	assert_eq!(
3235	action_for(Step::Verify, Kind::Service),
3236	Action::Build,
3237	"the step's own action is the one that would be denied",
3238	);
3239	let (code, out) = ctx_blocking
3240	.run("local", "echo reached-the-service-host")
3241	.expect("a service host must be reachable during a verify step");
3242	assert_eq!(code, 0, "{out}");
3243	assert!(out.contains("reached-the-service-host"), "{out}");
3244	})
3245	.await
3246	.unwrap();
3247	}
3248
3249	/// A step that finalized `Failed` bars the deploy, exactly as it bars a
3250	/// publish. Without this, a recipe that inspects `sh(...).code` and carries
3251	/// on regardless still lands a binary on a production host — the precise
3252	/// hazard a pipeline exists to remove. The check is the ledger, not the
3253	/// control flow, so it holds whether or not the recipe noticed.
3254	#[tokio::test]
3255	async fn a_failed_step_bars_the_deploy() {
3256	let dir = tempfile::tempdir().unwrap();
3257	let cfg = Arc::new(Config::for_tests(dir.path()));
3258	let pool = crate::db::open(&cfg.db_path).await.unwrap();
3259	let deploy = crate::topology::DeployTarget {
3260	target: "linux/x86_64".parse().unwrap(),
3261	host: "local".into(),
3262	port: None,
3263	install_path: "/usr/local/bin/demo".into(),
3264	service: "demo.service".into(),
3265	health_url: None,
3266	};
3267	// A real build + target run, so the step rows this test finalizes have
3268	// the parents the schema requires.
3269	sqlx::query(
3270	"INSERT INTO builds (id, app, version, status, created_at) \
3271	VALUES (1, 'demo', '0.1.0', 'running', '2026-07-30T00:00:00Z')",
3272	)
3273	.execute(&pool)
3274	.await
3275	.unwrap();
3276	sqlx::query(
3277	"INSERT INTO target_runs (id, build_id, app, version, target, status, started_at) \
3278	VALUES (1, 1, 'demo', '0.1.0', 'linux/x86_64', 'running', '2026-07-30T00:00:00Z')",
3279	)
3280	.execute(&pool)
3281	.await
3282	.unwrap();
3283
3284	let mut execs: crate::state::ExecutorMap = std::collections::HashMap::new();
3285	execs.insert("local".into(), crate::state::build_deploy_executor(&deploy));
3286	let ctx = Arc::new(RecipeCtx::new(
3287	AppId::new("demo"),
3288	Version::parse("0.1.0").unwrap(),
3289	"linux/x86_64".parse().unwrap(),
3290	"fw13".into(),
3291	"local".into(),
3292	"v0.1.0".into(),
3293	"/tmp".into(),
3294	vec![],
3295	Kind::Service,
3296	1,
3297	Arc::new(execs),
3298	Arc::new(std::collections::HashMap::new()),
3299	Some(deploy),
3300	pool,
3301	crate::events::channel(),
3302	cfg,
3303	Arc::new(OtaRegistry::standard("https://makenot.work")),
3304	tokio::runtime::Handle::current(),
3305	Arc::new(AtomicBool::new(false)),
3306	None,
3307	));
3308
3309	// A gate ran, failed, and the recipe did not abort — the swallowed
3310	// failure. Finalizing it is what puts it in the ledger.
3311	let ctx_blocking = ctx.clone();
3312	tokio::task::spawn_blocking(move \|\| {
3313	ctx_blocking.begin_step(Step::Prebuild).unwrap();
3314	ctx_blocking.fail_current_step();
3315	ctx_blocking.finish_step(Status::Ok).unwrap();
3316
3317	let err = ctx_blocking.deploy("/tmp/demo").unwrap_err().to_string();
3318	assert!(
3319	err.contains("refusing to deploy") && err.contains("prebuild"),
3320	"must refuse and name the failed step, got: {err}"
3321	);
3322	})
3323	.await
3324	.unwrap();
3325	}
3326
3327	#[test]
3328	fn every_step_has_a_nonzero_default_budget() {
3329	// A zero/missing budget would deadline-fail a step instantly. Cover the
3330	// whole matrix so a new Step variant can't silently get a 0 budget.
3331	for step in Step::ALL {
3332	assert!(
3333	default_step_budget(step) >= std::time::Duration::from_mins(1),
3334	"{step} budget must be a sane ceiling",
3335	);
3336	}
3337	}
3338
3339	#[test]
3340	fn sha256_file_is_lowercase_hex_of_contents() {
3341	let tmp = tempfile::tempdir().unwrap();
3342	let f = tmp.path().join("a.bin");
3343	std::fs::write(&f, b"abc").unwrap();
3344	// Known SHA-256 of "abc".
3345	assert_eq!(
3346	sha256_file(&f).unwrap(),
3347	"ba7816bf8f01cfea414140de5dae2223b00361a396177a9cb410ff61f20015ad"
3348	);
3349	}
3350
3351	// ---- publish step-success gate ----
3352
3353	fn target(s: &str) -> Target {
3354	s.parse().unwrap()
3355	}
3356
3357	#[test]
3358	fn publish_gate_blocks_macos_without_verification() {
3359	// Never verified -> blocked, with a message pointing at verify_gatekeeper.
3360	let err = PublishAuthority::prove(target("macos/aarch64"), &[], None).unwrap_err();
3361	assert!(format!("{err:#}").contains("never verified"), "{err:#}");
3362	}
3363
3364	#[test]
3365	fn publish_gate_blocks_macos_when_gatekeeper_rejected() {
3366	let err = PublishAuthority::prove(target("macos/aarch64"), &[], Some(false)).unwrap_err();
3367	assert!(
3368	format!("{err:#}").contains("Gatekeeper rejected"),
3369	"{err:#}"
3370	);
3371	}
3372
3373	#[test]
3374	fn publish_gate_allows_macos_when_gatekeeper_accepted() {
3375	PublishAuthority::prove(target("macos/aarch64"), &[], Some(true)).unwrap();
3376	// iOS is gated the same way.
3377	PublishAuthority::prove(target("ios/universal"), &[], Some(true)).unwrap();
3378	assert!(PublishAuthority::prove(target("ios/universal"), &[], None).is_err());
3379	}
3380
3381	#[test]
3382	fn publish_gate_does_not_require_gatekeeper_for_non_apple_targets() {
3383	// Linux/Windows aren't notarized; no gatekeeper proof needed.
3384	PublishAuthority::prove(target("linux/x86_64"), &[], None).unwrap();
3385	PublishAuthority::prove(target("windows/x86_64"), &[], None).unwrap();
3386	}
3387
3388	#[test]
3389	fn publish_gate_blocks_when_any_prior_step_failed() {
3390	// A failed step bars publish on every target, even a verified macOS one.
3391	let err =
3392	PublishAuthority::prove(target("linux/x86_64"), &[Step::Build], None).unwrap_err();
3393	assert!(
3394	format!("{err:#}").contains("prior step(s) failed"),
3395	"{err:#}"
3396	);
3397	assert!(
3398	format!("{err:#}").contains("build"),
3399	"names the failed step: {err:#}"
3400	);
3401
3402	let err = PublishAuthority::prove(target("macos/aarch64"), &[Step::Sign], Some(true))
3403	.unwrap_err();
3404	assert!(
3405	format!("{err:#}").contains("prior step(s) failed"),
3406	"{err:#}"
3407	);
3408	}
3409
3410	#[test]
3411	fn notary_accepted_parses_status_field() {
3412	assert!(notary_accepted(
3413	r#"{"id":"abc","status":"Accepted","message":"ok"}"#
3414	));
3415	// Embedded in shell-sourcing noise: the object is isolated and parsed.
3416	assert!(notary_accepted(
3417	"sourcing env...\n{\n \"status\": \"Accepted\"\n}\nbye"
3418	));
3419	// Whitespace variant that a tight substring `"status":"Accepted"` misses.
3420	assert!(notary_accepted(r#"{ "status" : "Accepted" }"#));
3421	}
3422
3423	#[test]
3424	fn notary_accepted_rejects_non_accepted_and_garbage() {
3425	assert!(!notary_accepted(r#"{"status":"Invalid"}"#));
3426	assert!(!notary_accepted(r#"{"status":"In Progress"}"#));
3427	assert!(!notary_accepted("no json here"));
3428	assert!(!notary_accepted("")); // empty / truncated -> fail closed
3429	// A truncated tail whose opening brace was cut off cannot parse -> closed.
3430	assert!(!notary_accepted(r#""status":"Accepted"}"#));
3431	// The literal appearing inside an error string must NOT pass as success.
3432	assert!(!notary_accepted(
3433	r#"{"status":"Invalid","message":"expected status:Accepted"}"#
3434	));
3435	}
3436	}
3437