max / makenotwork

156.8 KB · 3742 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	/// Bounding each step, rather than the whole build, keeps one slow step from
99	/// being blamed on another and keeps a fan-out of slow-but-fine targets from
100	/// being 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 appending would
345	/// show 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 cannot run 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	/// Assert `host`'s build tree is at the release tag and return the commit,
663	/// which is what a recipe's `checkout` step logs. Fetch + checkout stream
664	/// into the current step's log; the sha comes from a separate `rev-parse` so
665	/// its stdout is only the sha.
666	///
667	/// The preflight has already put this tree at the tag before any recipe ran —
668	/// this is the same operation again, on purpose, so that a recipe's own
669	/// `checkout` step is a real step with a real log rather than a claim about
670	/// something that happened elsewhere. Re-running it is cheap and, because the
671	/// tree is Bento's own worktree, forcing: a build that has already dirtied it
672	/// must not be able to fail its own retry.
673	fn checkout_sha(self: &Arc<Self>, host: &str) -> Result<String> {
674	// Every command below runs ON `host`, so the path is that host's, not the
675	// daemon's. Windows is why: its worktree is under `C:/Users/me/Code/...`.
676	let repo = self.repo_for(host).to_string();
677	// A failing mirror is not a failing release: fetch is advisory, and only
678	// the checkout decides. Its output still streams into the step log, so an
679	// unreachable remote stays visible without being fatal.
680	let _ = self.run(host, &git_fetch_cmd(&repo))?;
681	let (code, err) = self.run(host, &git_worktree_pin_cmd(&repo, &self.tag))?;
682	if code != 0 {
683	let (probe, _) = self.run(host, &git_tag_exists_cmd(&repo, &self.tag))?;
684	anyhow::bail!(
685	"checkout of {} failed on `{host}`: {}",
686	self.tag,
687	worktree_failure_reason(&self.tag, probe == 0, &err)
688	);
689	}
690	let (code, tail) = self.run(host, &git_rev_parse_cmd(&repo))?;
691	anyhow::ensure!(code == 0, "rev-parse failed on `{host}`");
692	Ok(tail.trim().to_string())
693	}
694
695	/// Every artifact this run collected, `file name -> sha256`.
696	///
697	/// Already computed at `collect`, which is the only moment the bytes are
698	/// known to be the ones that landed.
699	pub fn artifact_hashes(&self) -> HashMap<String, String> {
700	self.artifact_hashes.lock().unwrap().clone()
701	}
702
703	/// Resolve `glob` on `host` to the single artifact it names. The `for` loop
704	/// lists each existing match on its own line (and prints nothing — rather
705	/// than a literal unexpanded pattern — when the glob matches no file), so
706	/// the count is unambiguous. `required` controls whether zero matches is an
707	/// error; more than one always is. See `resolve_artifact_match`.
708	fn resolve_artifact(
709	self: &Arc<Self>,
710	host: &str,
711	glob: &str,
712	required: bool,
713	) -> Result<String> {
714	ensure_glob_safe(glob)?;
715	// `[ -e ]` guards against a non-matching glob surviving as its literal
716	// self, and lists one path per line for the count.
717	let cmd = format!("for __f in {glob}; do [ -e \"$__f\" ] && printf '%s\\n' \"$__f\"; done");
718	let (code, tail) = self.run(host, &cmd)?;
719	anyhow::ensure!(
720	code == 0,
721	"resolving artifact glob `{glob}` on `{host}` exited {code}"
722	);
723	resolve_artifact_match(&tail, glob, required)
724	}
725	}
726
727	// ----- error bridging: anyhow -> Rhai runtime error -----
728
729	// Rhai host functions return `Result<_, Box<EvalAltResult>>` by convention, so
730	// this bridge must yield the boxed form to be usable with `.map_err(rhai_err)`.
731	#[allow(
732	clippy::unnecessary_box_returns,
733	reason = "rhai's error type is used boxed throughout its host-function API"
734	)]
735	fn rhai_err(e: impl std::fmt::Display) -> Box<EvalAltResult> {
736	Box::new(EvalAltResult::ErrorRuntime(
737	e.to_string().into(),
738	rhai::Position::NONE,
739	))
740	}
741
742	/// A crate's publish-relevant metadata, read from `cargo metadata`.
743	#[derive(Debug, Clone)]
744	pub struct CrateMeta {
745	pub name: String,
746	pub version: String,
747	pub repository: Option<String>,
748	pub description: Option<String>,
749	pub licensed: bool,
750	}
751
752	/// Parse the fields that matter for publishing out of `cargo metadata` JSON.
753	pub fn crate_meta_from_json(raw: &str) -> Result<CrateMeta> {
754	let v: serde_json::Value = serde_json::from_str(raw).context("parsing cargo metadata")?;
755	let p = v
756	.get("packages")
757	.and_then(\|p\| p.as_array())
758	.and_then(\|a\| a.first())
759	.context("cargo metadata reported no package")?;
760	let str_field = \|k: &str\| {
761	p.get(k)
762	.and_then(\|x\| x.as_str())
763	.filter(\|s\| !s.is_empty())
764	.map(str::to_string)
765	};
766	Ok(CrateMeta {
767	name: str_field("name").context("package has no name")?,
768	version: str_field("version").context("package has no version")?,
769	repository: str_field("repository"),
770	description: str_field("description"),
771	licensed: str_field("license").is_some() \|\| str_field("license_file").is_some(),
772	})
773	}
774
775	/// Everything wrong with a crate's metadata, as messages. Empty means publishable.
776	///
777	/// Checks only what crates.io records permanently. A published version cannot
778	/// be edited, only yanked, and yanking does not correct a wrong URL — so these
779	/// are the last moment any of it can be fixed.
780	pub fn crate_publish_problems(
781	meta: &CrateMeta,
782	repo_clonable: bool,
783	published: &[String],
784	credentials_present: bool,
785	) -> Vec<String> {
786	let mut out = Vec::new();
787	if !credentials_present {
788	out.push(
789	"no crates.io credentials on the publishing host: `cargo login` there first. \
790	Checked now rather than at the upload, so this fails in seconds instead of \
791	after a full build and verify."
792	.to_string(),
793	);
794	}
795	match &meta.repository {
796	None => out.push(
797	"no `repository` field: the crates.io page will show no source link, permanently"
798	.to_string(),
799	),
800	Some(url) if !repo_clonable => out.push(format!(
801	"`repository` is not publicly clonable: {url} \
802	(wrong URL, or the repo is private)"
803	)),
804	Some(_) => {}
805	}
806	if meta.description.is_none() {
807	out.push("no `description`: crates.io requires one".to_string());
808	}
809	if !meta.licensed {
810	out.push("no `license` or `license-file`".to_string());
811	}
812	if published.iter().any(\|v\| v == &meta.version) {
813	out.push(format!(
814	"version {} is already published; bump it",
815	meta.version
816	));
817	}
818	out
819	}
820
821	/// Read the app's version from its checkout on the daemon host. With
822	/// `version_path` set (topology `version_path`), read exactly that file — a
823	/// `.json` as a Tauri config, anything else as a `Cargo.toml`. Unset (the Tauri
824	/// default), try `src-tauri/tauri.conf.json` then the root `Cargo.toml`. Used by
825	/// the runner's default-version path.
826	pub fn version_from_repo(repo: &str, version_path: Option<&str>) -> Result<Version> {
827	let root = expand_tilde(repo);
828	if let Some(vp) = version_path {
829	let path = root.join(vp);
830	let raw = std::fs::read_to_string(&path)
831	.with_context(\|\| format!("reading version file {}", path.display()))?;
832	let ver = if std::path::Path::new(vp)
833	.extension()
834	.is_some_and(\|e\| e.eq_ignore_ascii_case("json"))
835	{
836	version_from_tauri_json(&raw)?
837	} else {
838	version_from_cargo_toml(&raw)?
839	};
840	return Version::parse(&ver).map_err(\|e\| anyhow::anyhow!(e));
841	}
842	let tauri_conf = root.join("src-tauri").join("tauri.conf.json");
843	if tauri_conf.exists() {
844	let raw = std::fs::read_to_string(&tauri_conf)
845	.with_context(\|\| format!("reading {}", tauri_conf.display()))?;
846	return Version::parse(&version_from_tauri_json(&raw)?).map_err(\|e\| anyhow::anyhow!(e));
847	}
848	let cargo_toml = root.join("Cargo.toml");
849	let raw = std::fs::read_to_string(&cargo_toml).with_context(\|\| {
850	format!(
851	"reading {} (no tauri.conf.json either)",
852	cargo_toml.display()
853	)
854	})?;
855	Version::parse(&version_from_cargo_toml(&raw)?).map_err(\|e\| anyhow::anyhow!(e))
856	}
857
858	/// Extract `version` from raw `tauri.conf.json` text.
859	fn version_from_tauri_json(raw: &str) -> Result<String> {
860	let v: serde_json::Value = serde_json::from_str(raw).context("parsing tauri.conf.json")?;
861	v.get("version")
862	.and_then(\|x\| x.as_str())
863	.map(str::to_owned)
864	.context("no `version` in tauri.conf.json")
865	}
866
867	/// Extract the version from raw `Cargo.toml` text — `[package].version` (a leaf
868	/// crate) or `[workspace.package].version` (a workspace that sets it).
869	fn version_from_cargo_toml(raw: &str) -> Result<String> {
870	let doc: toml::Value = toml::from_str(raw).context("parsing Cargo.toml")?;
871	doc.get("package")
872	.and_then(\|p\| p.get("version"))
873	.or_else(\|\| {
874	doc.get("workspace")
875	.and_then(\|w\| w.get("package"))
876	.and_then(\|p\| p.get("version"))
877	})
878	.and_then(\|v\| v.as_str())
879	.map(str::to_owned)
880	.context("no `[package].version` or `[workspace.package].version` in Cargo.toml")
881	}
882
883	/// Cross-check every version source in a repo and confirm they all agree with
884	/// the version being built, before a single host pulls or compiles.
885	///
886	/// `version_from_repo` reads exactly one file, so a `tauri.conf.json` at 0.5.0
887	/// and a root `Cargo.toml` still at 0.4.0 build happily and file artifacts under
888	/// whichever the runner happened to read. This reads every source present —
889	/// `version_path` (when set), `src-tauri/tauri.conf.json`, and the root
890	/// `Cargo.toml` — and fails loudly when any disagree, naming each file and its
891	/// version. A source that is absent is skipped (a library crate with only a
892	/// `Cargo.toml` has nothing to disagree with); the check never invents drift.
893	///
894	/// Scope: the JSON/TOML sources bentod itself reads. The iOS `gen/apple/project.yml`
895	/// path (rewritten by a build-time `sed`) is out of scope here — it is asserted at
896	/// its own build step — but the same drift class motivated this guard.
897	pub fn check_version_consistency(
898	repo: &str,
899	version_path: Option<&str>,
900	expected: &Version,
901	) -> Result<()> {
902	let root = expand_tilde(repo);
903	let mut sources: Vec<(String, String)> = Vec::new();
904	for rel in version_sources(version_path) {
905	let path = root.join(&rel);
906	// A source that is absent is skipped — a library crate with only a
907	// `Cargo.toml` has nothing to disagree with — but one the app NAMES
908	// must be readable, or the check would pass by failing to look.
909	match std::fs::read_to_string(&path) {
910	Ok(raw) => sources.push((rel, raw)),
911	Err(e) if version_path == Some(rel.as_str()) => {
912	return Err(e).with_context(\|\| format!("reading version file {}", path.display()));
913	}
914	Err(_) => {}
915	}
916	}
917	versions_agree(repo, &sources, version_path, expected)
918	}
919
920	/// The files a repo can state its version in, in the order they are read:
921	/// whatever the app names, then the two conventional ones.
922	///
923	/// The app's own `version_path` is never read twice, which is why this is a
924	/// function rather than a constant.
925	pub fn version_sources(version_path: Option<&str>) -> Vec<String> {
926	let mut rels: Vec<String> = version_path.into_iter().map(str::to_string).collect();
927	for conventional in ["src-tauri/tauri.conf.json", "Cargo.toml"] {
928	if version_path != Some(conventional) {
929	rels.push(conventional.to_string());
930	}
931	}
932	rels
933	}
934
935	/// The judgement half of [`check_version_consistency`], over sources somebody
936	/// else read.
937	///
938	/// Split out so the same rule can be applied to files read out of the release
939	/// TAG on a build host, which is where the question actually belongs: the tree
940	/// a release compiles is the tag's, so a `Cargo.toml` that disagrees with the
941	/// tag it is tagged in is the drift worth refusing. Reading the working copy
942	/// instead answered a question about a tree the release does not build.
943	///
944	/// `where_` is only for the error message — a path, or a tag and a host.
945	pub fn versions_agree(
946	where_: &str,
947	sources: &[(String, String)],
948	version_path: Option<&str>,
949	expected: &Version,
950	) -> Result<()> {
951	let mut found: Vec<(String, Version)> = Vec::new();
952	for (rel, raw) in sources {
953	// The app's own `version_path` can be either shape, so it is decided by
954	// extension; the two conventional sources are what they are.
955	let as_json = std::path::Path::new(rel)
956	.extension()
957	.is_some_and(\|e\| e.eq_ignore_ascii_case("json"));
958	let ver = if as_json {
959	version_from_tauri_json(raw)
960	} else {
961	// A Cargo.toml with neither `[package].version` nor
962	// `[workspace.package].version` (a pure virtual workspace) carries
963	// no version to check — skip it rather than fail. An app that NAMED
964	// this file is held to it.
965	match version_from_cargo_toml(raw) {
966	Ok(v) => Ok(v),
967	Err(e) if version_path == Some(rel.as_str()) => Err(e),
968	Err(_) => continue,
969	}
970	}?;
971	found.push((
972	rel.clone(),
973	Version::parse(&ver).map_err(\|e\| anyhow::anyhow!(e))?,
974	));
975	}
976
977	let disagree: Vec<&(String, Version)> = found.iter().filter(\|(_, v)\| v != expected).collect();
978	anyhow::ensure!(
979	disagree.is_empty(),
980	"version drift in {where_}: building {expected} but {}",
981	disagree
982	.iter()
983	.map(\|(src, v)\| format!("{src} says {v}"))
984	.collect::<Vec<_>>()
985	.join(", ")
986	);
987	Ok(())
988	}
989
990	/// Read one file as it exists in `tag`, without checking anything out.
991	///
992	/// `<rev>:./<path>` resolves the path relative to `-C`, so this is asked from
993	/// the app's own directory and needs no knowledge of where that sits inside the
994	/// repository. A non-zero exit means the file is not in the tag, which is the
995	/// same "absent, so nothing to disagree with" the local read treats it as.
996	pub fn git_show_file_cmd(dir: &str, tag: &str, rel: &str) -> String {
997	format!("git -C \"{dir}\" show \"{tag}:./{rel}\"")
998	}
999
1000	/// Every `X.Y.Z`-shaped version embedded in an artifact file name. Each maximal
1001	/// run of digits-and-dots contributes its first three numeric fields:
1002	/// `GoingsOn_0.5.0_aarch64.dmg` and `demo-9.9.9.bin` both yield one version (the
1003	/// trailing `.bin`/`.dmg` dot is tolerated), while `latest.json` yields `[]` and
1004	/// the `64` in `x86_64` is not three fields. Only the `major.minor.patch` core is
1005	/// taken; a prerelease/build suffix is separated by `-`/`+` and not needed here.
1006	fn versions_in_filename(name: &str) -> Vec<Version> {
1007	name.split(\|c: char\| !(c.is_ascii_digit() \|\| c == '.'))
1008	.filter_map(\|run\| {
1009	let f: Vec<&str> = run.split('.').filter(\|s\| !s.is_empty()).collect();
1010	if f.len() >= 3 && f[..3].iter().all(\|s\| s.chars().all(\|c\| c.is_ascii_digit())) {
1011	Version::parse(&format!("{}.{}.{}", f[0], f[1], f[2])).ok()
1012	} else {
1013	None
1014	}
1015	})
1016	.collect()
1017	}
1018
1019	/// Fail when a collected file's name embeds a version whose `major.minor.patch`
1020	/// is not the one being built. This is the guard against a stale checked-in
1021	/// artifact winning a glob: `ls -t <glob>` once let
1022	/// `AudioFiles-0.4.0-x86_64.AppImage` ship against 0.5.0. A file whose name
1023	/// carries no version (an updater `latest.json`, a `.sig`) is not asserted —
1024	/// there is nothing to compare. Compared on the core so a prerelease build's
1025	/// plain `X.Y.Z` in the filename still matches.
1026	fn assert_artifact_version(name: &str, expected: &Version) -> Result<()> {
1027	let versions = versions_in_filename(name);
1028	anyhow::ensure!(
1029	versions.is_empty() \|\| versions.iter().any(\|v\| v.core() == expected.core()),
1030	"collected artifact `{name}` carries version {} but the build is {expected}; \
1031	a stale artifact was left in the output dir — clean it so only {expected} remains",
1032	versions
1033	.iter()
1034	.map(ToString::to_string)
1035	.collect::<Vec<_>>()
1036	.join("/"),
1037	);
1038	Ok(())
1039	}
1040
1041	/// sha256 of a file, lowercase hex. Streams in 64 KiB chunks so a multi-GiB
1042	/// bundle never lands in memory whole.
1043	fn sha256_file(path: &Path) -> Result<String> {
1044	let mut file =
1045	std::fs::File::open(path).with_context(\|\| format!("hashing {}", path.display()))?;
1046	let mut hasher = Sha256::new();
1047	std::io::copy(&mut file, &mut hasher)
1048	.with_context(\|\| format!("reading {} to hash", path.display()))?;
1049	Ok(hex_lower(&hasher.finalize()))
1050	}
1051
1052	/// Every regular file under `root`, as `(path relative to root, absolute path)`,
1053	/// sorted by the relative path.
1054	///
1055	/// This walk has to match `bundle::digest_dir` in sando, file for file. That
1056	/// function re-hashes an incoming bundle and refuses it when the bytes disagree
1057	/// with the manifest they arrived with, so a producer that walks differently
1058	/// produces a manifest the consumer will reject for an artifact nothing is wrong
1059	/// with. Three properties carry that agreement, and none is incidental:
1060	///
1061	/// - Recursive. A bundle may carry a directory (migrations, resources), and
1062	/// a top-level-only listing would omit its contents from the manifest while
1063	/// the verifier hashed them.
1064	/// - Symlinks are not followed, and not recorded. Following one would let
1065	/// content from outside the bundle into its identity; recording the link
1066	/// itself would name a file the verifier does not hash.
1067	/// - Relative paths, `/`-separated, sorted. Readdir order is not guaranteed,
1068	/// so an unsorted manifest would differ run to run on one machine, never mind
1069	/// between two.
1070	fn collected_files(root: &Path) -> std::io::Result<Vec<(String, PathBuf)>> {
1071	fn walk(dir: &Path, root: &Path, out: &mut Vec<(String, PathBuf)>) -> std::io::Result<()> {
1072	for entry in std::fs::read_dir(dir)? {
1073	let entry = entry?;
1074	let ft = entry.file_type()?;
1075	let path = entry.path();
1076	if ft.is_dir() {
1077	walk(&path, root, out)?;
1078	} else if ft.is_file() {
1079	let rel = path
1080	.strip_prefix(root)
1081	.unwrap_or(&path)
1082	.components()
1083	.map(\|c\| c.as_os_str().to_string_lossy())
1084	.collect::<Vec<_>>()
1085	.join("/");
1086	out.push((rel, path));
1087	}
1088	// Symlinks and other special files are intentionally ignored,
1089	// matching the verifier.
1090	}
1091	Ok(())
1092	}
1093	let mut out = Vec::new();
1094	walk(root, root, &mut out)?;
1095	out.sort_by(\|a, b\| a.0.cmp(&b.0));
1096	Ok(out)
1097	}
1098
1099	/// Lowercase-hex encode without pulling in a hex crate.
1100	fn hex_lower(bytes: &[u8]) -> String {
1101	use std::fmt::Write as _;
1102	let mut s = String::with_capacity(bytes.len() * 2);
1103	for b in bytes {
1104	let _ = write!(s, "{b:02x}");
1105	}
1106	s
1107	}
1108
1109	/// Refresh every remote's refs and tags, so the tag a release names is present
1110	/// locally however it was pushed. No branch/upstream assumptions — a bare
1111	/// `git pull --ff-only` needs a tracking branch the release path shouldn't
1112	/// depend on.
1113	///
1114	/// Best-effort on purpose. `fetch --all` exits non-zero if ANY remote fails, and
1115	/// the library repos carry three (`astra`, `mnw`, `srht`), so chaining this into
1116	/// the checkout with `&&` meant one unreachable mirror aborted the release and
1117	/// reported it as a missing tag. The checkout below is the step allowed to fail;
1118	/// this one only has to try. See [`git_worktree_pin_cmd`].
1119	///
1120	/// `repo` is interpolated UNQUOTED so a leading `~` is expanded by the remote
1121	/// host's shell (the checkout path is trusted topology config, not user input),
1122	/// matching how the recipes `cd` into it.
1123	pub fn git_fetch_cmd(repo: &str) -> String {
1124	format!("git -C {repo} fetch --all --tags --prune")
1125	}
1126
1127	/// Probe for the one failure a tracked `Cargo.lock` hits inside Bento's
1128	/// worktree. Exits 0 when the crate tracks a lock AND the checkout sits under a
1129	/// `.cargo/config.toml` declaring `[patch]`; 1 otherwise.
1130	///
1131	/// Both halves are needed and cargo reports neither. Under a `[patch]` block
1132	/// cargo re-resolves and rewrites the lock's `[[patch.unused]]` entries, so
1133	/// `cargo publish --dry-run` refuses the tree with "1 files in the working
1134	/// directory contain changes that were not yet committed into git: Cargo.lock"
1135	/// -- naming the lock and nothing about why it moved. pter 0.2.1 lost half an
1136	/// hour to that message on build 325.
1137	///
1138	/// `~/Code/.bento` is under `~/Code` deliberately, so that the patch block
1139	/// reaches the build (see [`crate::topology::Host::worktree_root`]). The patch
1140	/// block is therefore not the half to remove, which is why this is worth
1141	/// saying rather than leaving cargo to be cryptic about it.
1142	///
1143	/// `repo` is interpolated unquoted for the `~`, matching [`git_fetch_cmd`];
1144	/// `pwd -P` then hands the loop an absolute path to walk up from.
1145	pub fn tracked_lock_under_patch_cmd(repo: &str) -> String {
1146	format!(
1147	"git -C {repo} ls-files --error-unmatch Cargo.lock >/dev/null 2>&1 \|\| exit 1; \
1148	d=$(cd {repo} && pwd -P) \|\| exit 1; \
1149	while [ -n \"$d\" ] && [ \"$d\" != / ]; do \
1150	for c in \"$d/.cargo/config.toml\" \"$d/.cargo/config\"; do \
1151	[ -f \"$c\" ] && grep -q '^\\[patch' \"$c\" && exit 0; \
1152	done; d=$(dirname \"$d\"); done; exit 1"
1153	)
1154	}
1155
1156	/// What to say when [`tracked_lock_under_patch_cmd`] answers yes. Names both
1157	/// facts, because the error cargo would otherwise print names neither, and
1158	/// closes off the two wrong fixes that are both one flag away.
1159	pub fn tracked_lock_under_patch_problem(repo: &str) -> String {
1160	format!(
1161	"`Cargo.lock` is tracked and {repo} sits under a `.cargo/config.toml` \
1162	declaring `[patch]`. Cargo re-resolves there and rewrites the lock, so \
1163	`cargo publish --dry-run` will refuse the tree as dirty and name only \
1164	the lock. Untrack it: `git rm --cached Cargo.lock`. Every other library \
1165	in this tree already does. Not `--allow-dirty`, which publishes a lock \
1166	nobody reviewed, and not committing the rewritten lock, which resolves \
1167	differently on the next machine and fails there instead. The build \
1168	worktree is under `~/Code` on purpose so the patch block applies to it; \
1169	that is not the half to change."
1170	)
1171	}
1172
1173	/// Does `tag` resolve to a commit in this checkout? Run only when the checkout
1174	/// has already failed, to say WHY: an absent tag is an untagged or unpushed
1175	/// release, while a tag that resolves fine means the checkout was refused for a
1176	/// local reason (a dirty tree, most often) and the operator needs to hear that
1177	/// instead.
1178	pub fn git_tag_exists_cmd(repo: &str, tag: &str) -> String {
1179	format!("git -C {repo} rev-parse -q --verify \"refs/tags/{tag}^{{commit}}\"")
1180	}
1181
1182	/// Which repository `repo` belongs to, and where `repo` sits inside it, in one
1183	/// call: `--show-toplevel` then `--show-prefix`, one per line.
1184	///
1185	/// Both halves are needed to build in a worktree of a repo holding several
1186	/// products. The worktree is made of the repository (`~/Code/MNW`), and the
1187	/// recipe has to be pointed at the app inside it (`<worktree>/pom`).
1188	pub fn git_toplevel_and_prefix_cmd(repo: &str) -> String {
1189	format!("git -C {repo} rev-parse --show-toplevel --show-prefix")
1190	}
1191
1192	/// Read [`git_toplevel_and_prefix_cmd`]'s two lines.
1193	///
1194	/// The prefix is empty for a repo holding one product, where `repo` IS the
1195	/// repository root — and git prints an empty second line for it, so a missing
1196	/// line is a malformed answer rather than that case.
1197	pub fn parse_toplevel_and_prefix(out: &str) -> Option<(String, String)> {
1198	let mut lines = out.split('\n');
1199	let toplevel = lines.next()?.trim().to_string();
1200	let prefix = lines.next()?.trim().to_string();
1201	(!toplevel.is_empty()).then_some((toplevel, prefix))
1202	}
1203
1204	/// The repository's own directory name, which is what names its worktrees:
1205	/// `MNW` for `/home/max/Code/MNW`.
1206	///
1207	/// Splits on `/` only. Git reports `--show-toplevel` with forward slashes on
1208	/// every platform, Windows included, so this is the separator to read.
1209	pub fn repo_dir_name(toplevel: &str) -> &str {
1210	toplevel
1211	.trim_end_matches('/')
1212	.rsplit('/')
1213	.next()
1214	.unwrap_or(toplevel)
1215	}
1216
1217	/// Where the app being released sits inside its worktree: the worktree root for
1218	/// a repo holding one product, `<worktree>/pom` for one holding several.
1219	pub fn app_dir_in_worktree(worktree: &str, prefix: &str) -> String {
1220	let prefix = prefix.trim_matches('/');
1221	if prefix.is_empty() {
1222	worktree.to_string()
1223	} else {
1224	format!("{}/{prefix}", worktree.trim_end_matches('/'))
1225	}
1226	}
1227
1228	/// Does this worktree already exist? Run before deciding whether to create one.
1229	///
1230	/// `rev-parse --git-dir` rather than a shell test, because the one non-unix
1231	/// build host has no `test`: every command Bento renders for a host is a git
1232	/// command or something a recipe wrote.
1233	pub fn git_worktree_probe_cmd(worktree: &str) -> String {
1234	format!("git -C \"{worktree}\" rev-parse --git-dir")
1235	}
1236
1237	/// Forget worktrees whose directories are gone. Run before creating one: a
1238	/// directory somebody deleted by hand is still registered in the repository, and
1239	/// `worktree add` refuses the path as in use rather than rebuilding it.
1240	pub fn git_worktree_prune_cmd(toplevel: &str) -> String {
1241	format!("git -C \"{toplevel}\" worktree prune")
1242	}
1243
1244	/// Create this app's build worktree, detached at the release tag. Git creates
1245	/// the leading directories, so the worktree root needs no preparation.
1246	pub fn git_worktree_add_cmd(toplevel: &str, worktree: &str, tag: &str) -> String {
1247	format!("git -C \"{toplevel}\" worktree add --detach --force \"{worktree}\" \"{tag}\"")
1248	}
1249
1250	/// Put an existing build worktree at the release tag.
1251	///
1252	/// `--force` discards whatever the last release left in it — a rewritten
1253	/// `Cargo.lock`, most often — and that is safe here in a way it never was in the
1254	/// ordinary checkout: nothing but Bento writes in this tree, so there is no edit
1255	/// of anybody's to lose. Owning the tree is what buys the forcing.
1256	pub fn git_worktree_pin_cmd(worktree: &str, tag: &str) -> String {
1257	format!("git -C \"{worktree}\" checkout --detach --force \"{tag}\"")
1258	}
1259
1260	/// The operator-facing explanation for a worktree that could not be put at the
1261	/// tag.
1262	///
1263	/// An absent tag is an untagged or unpushed release and is the common case, so
1264	/// it is answered plainly. Anything else is git's own stderr, which says more
1265	/// about a path that is not a worktree, or a worktree another release holds,
1266	/// than a guess would.
1267	pub fn worktree_failure_reason(tag: &str, tag_exists: bool, stderr: &str) -> String {
1268	if !tag_exists {
1269	return format!("tag {tag} does not exist there (is it created and pushed?)");
1270	}
1271	let stderr = stderr.trim();
1272	if stderr.is_empty() {
1273	format!("tag {tag} exists, and git said nothing about why")
1274	} else {
1275	stderr.to_string()
1276	}
1277	}
1278
1279	/// The command a host runs to report the commit it has checked out, for the
1280	/// release preflight barrier.
1281	pub fn git_rev_parse_cmd(repo: &str) -> String {
1282	format!("git -C {repo} rev-parse HEAD")
1283	}
1284
1285	/// Expand a leading `~/` to `$HOME`. Paths in the topology are written with `~`.
1286	pub fn expand_tilde(p: &str) -> PathBuf {
1287	if let Some(rest) = p.strip_prefix("~/")
1288	&& let Ok(home) = std::env::var("HOME")
1289	{
1290	return Path::new(&home).join(rest);
1291	}
1292	PathBuf::from(p)
1293	}
1294
1295	/// Reject a glob that carries shell command metacharacters. Path and wildcard
1296	/// characters (`/ . * ? [ ] ~` etc.) are fine — the pattern reaches a login
1297	/// shell to be expanded — but a `;` or `$(...)` must not ride along and run.
1298	/// Not a privilege boundary (a recipe already runs arbitrary shell via `sh_ok`)
1299	/// but it keeps a malformed pattern from turning into a command. Shared by
1300	/// `collect` and `resolve_artifact`.
1301	fn ensure_glob_safe(glob: &str) -> Result<()> {
1302	anyhow::ensure!(
1303	!glob.chars().any(\|c\| matches!(
1304	c,
1305	';' \| '&' \| '\|' \| '$' \| '`' \| '\'' \| '"' \| '\\' \| ' ' \| '\n' \| '(' \| ')' \| '<' \| '>'
1306	)),
1307	"glob `{glob}` contains shell metacharacters"
1308	);
1309	Ok(())
1310	}
1311
1312	/// Decide the single artifact a glob resolves to from a newline-separated
1313	/// listing of the paths that matched it.
1314	///
1315	/// Demands exactly one match. Zero matches fail when `required` (return `""`
1316	/// when optional); more than one is always an error rather than an arbitrary
1317	/// newest-wins pick, because an ambiguous match means the build left stale
1318	/// artifacts behind and the wrong one could ship.
1319	fn resolve_artifact_match(listing: &str, glob: &str, required: bool) -> Result<String> {
1320	let matches: Vec<&str> = listing
1321	.lines()
1322	.map(str::trim)
1323	.filter(\|l\| !l.is_empty())
1324	.collect();
1325	match matches.as_slice() {
1326	[] if required => anyhow::bail!("no artifact matched glob `{glob}`"),
1327	[] => Ok(String::new()),
1328	[one] => Ok((*one).to_string()),
1329	many => anyhow::bail!(
1330	"glob `{glob}` is ambiguous: {} artifacts matched ({}). \
1331	The build left more than one behind; clean stale artifacts so exactly one remains.",
1332	many.len(),
1333	many.join(", ")
1334	),
1335	}
1336	}
1337
1338	/// Build a Rhai engine with the host API bound to `ctx`. Sandboxed: recipes
1339	/// touch the outside world only through these functions.
1340	pub fn build_engine(ctx: &Arc<RecipeCtx>) -> Engine {
1341	let mut engine = Engine::new();
1342	// Defensive caps — recipes are first-party but bound the blast radius.
1343	engine.set_max_operations(5_000_000);
1344	engine.set_max_call_levels(64);
1345	engine.set_max_string_size(0);
1346
1347	// --- step(name) ---
1348	{
1349	let ctx = ctx.clone();
1350	engine.register_fn(
1351	"step",
1352	move \|name: &str\| -> Result<(), Box<EvalAltResult>> {
1353	let step: Step = name.parse().map_err(rhai_err)?;
1354	ctx.begin_step(step).map_err(rhai_err)
1355	},
1356	);
1357	}
1358
1359	// --- sh(host, cmd) -> #{ code, stdout_tail } ---
1360	//
1361	// The branch-on-exit-code primitive: the recipe OWNS the outcome. A non-zero
1362	// exit is returned, not raised, and does NOT fail the step or bar publish —
1363	// use this only when the recipe inspects `code` and decides. For a command
1364	// that must succeed (build/sign/etc.), use `sh_ok`, which fails the step (and
1365	// therefore bars publish via the failed-step ledger) on a non-zero exit.
1366	{
1367	let ctx = ctx.clone();
1368	engine.register_fn(
1369	"sh",
1370	move \|host: &str, cmd: &str\| -> Result<Map, Box<EvalAltResult>> {
1371	let (code, tail) = ctx.run(host, cmd).map_err(rhai_err)?;
1372	let mut m = Map::new();
1373	m.insert("code".into(), (code as i64).into());
1374	m.insert("stdout_tail".into(), tail.into());
1375	Ok(m)
1376	},
1377	);
1378	}
1379
1380	// --- sh_ok(host, cmd): run + assert exit 0 (the must-succeed primitive) ---
1381	//
1382	// A non-zero exit fails the current step (added to the publish-barring
1383	// ledger) and aborts the recipe, so an artifact is never shipped after a
1384	// must-succeed command failed.
1385	{
1386	let ctx = ctx.clone();
1387	engine.register_fn(
1388	"sh_ok",
1389	move \|host: &str, cmd: &str\| -> Result<(), Box<EvalAltResult>> {
1390	let (code, _) = ctx.run(host, cmd).map_err(rhai_err)?;
1391	if code != 0 {
1392	// Attribute the failure to the current step explicitly so the
1393	// ledger bars publish even if a future caller swallowed the error.
1394	ctx.fail_current_step();
1395	return Err(rhai_err(format!(
1396	"command on `{host}` exited {code}: {cmd}"
1397	)));
1398	}
1399	Ok(())
1400	},
1401	);
1402	}
1403
1404	// --- resolve_artifact(host, glob) -> path: the ONE artifact matching glob ---
1405	//
1406	// The artifact-selection primitive. Replaces `sh(host, "ls -t <glob> \| head
1407	// -1").stdout_tail.trim()` guarded on an empty string, which let a non-zero
1408	// `ls` pass quietly and a stale newest-by-mtime file win. This resolves the
1409	// glob on the host and demands exactly one match: zero matches or more than
1410	// one both throw (an ambiguous match means the build left stale artifacts,
1411	// and silently picking the newest is how the wrong bytes ship). Use
1412	// `resolve_artifact_opt` for an artifact that may legitimately be absent.
1413	{
1414	let ctx = ctx.clone();
1415	engine.register_fn(
1416	"resolve_artifact",
1417	move \|host: &str, glob: &str\| -> Result<String, Box<EvalAltResult>> {
1418	ctx.resolve_artifact(host, glob, true).map_err(rhai_err)
1419	},
1420	);
1421	}
1422
1423	// --- resolve_artifact_opt(host, glob) -> path \| "": zero-or-one match ---
1424	//
1425	// Same strict resolution as `resolve_artifact` but tolerates zero matches
1426	// (returns ""); more than one is still an error. For optional outputs like a
1427	// `.deb` or an updater bundle a recipe collects only when present.
1428	{
1429	let ctx = ctx.clone();
1430	engine.register_fn(
1431	"resolve_artifact_opt",
1432	move \|host: &str, glob: &str\| -> Result<String, Box<EvalAltResult>> {
1433	ctx.resolve_artifact(host, glob, false).map_err(rhai_err)
1434	},
1435	);
1436	}
1437
1438	// --- log(msg): operator-visible line into the current step's tail ---
1439	{
1440	let ctx = ctx.clone();
1441	engine.register_fn("log", move \|msg: &str\| -> Result<(), Box<EvalAltResult>> {
1442	let sink = ctx.ensure_step().map_err(rhai_err)?;
1443	let line = format!("[recipe] {msg}\n");
1444	ctx.rt.block_on(async {
1445	use ops_core::remote::LogSink;
1446	sink.lock().await.write_chunk(line.as_bytes()).await;
1447	});
1448	Ok(())
1449	});
1450	}
1451
1452	// --- version_of(app) -> string ---
1453	{
1454	let ctx = ctx.clone();
1455	engine.register_fn(
1456	"version_of",
1457	move \|app: &str\| -> Result<String, Box<EvalAltResult>> {
1458	// Only the current app is in scope; cross-app reads aren't needed.
1459	if app != ctx.app.as_str() {
1460	return Err(rhai_err(format!(
1461	"version_of: `{app}` is not the app being built"
1462	)));
1463	}
1464	Ok(ctx.version.to_string())
1465	},
1466	);
1467	}
1468
1469	// --- version() -> string: the version being built (no-arg form) ---
1470	{
1471	let ctx = ctx.clone();
1472	engine.register_fn("version", move \|\| -> String { ctx.version.to_string() });
1473	}
1474
1475	// --- build_host() -> string: the host this target builds on ---
1476	{
1477	let ctx = ctx.clone();
1478	engine.register_fn("build_host", move \|\| -> String { ctx.build_host.clone() });
1479	}
1480
1481	// --- repo() -> string: the app's checkout path on this target's build host
1482	// (`~`-prefixed on a unix host). Host-correct rather than one path per
1483	// app, so a recipe for a host whose checkout is elsewhere still calls
1484	// this instead of hard-coding the path — which is what kept the Windows
1485	// recipes off `checkout_sha`. ---
1486	{
1487	let ctx = ctx.clone();
1488	engine.register_fn("repo", move \|\| -> String {
1489	ctx.repo_for(&ctx.build_host).to_string()
1490	});
1491	}
1492
1493	// --- checkout_sha(host) -> sha: pin this host to the release tag and report
1494	// its commit. Replaces a recipe's `git pull --ff-only`, which builds
1495	// whatever `main` is at pull time; the daemon also runs the same pin as
1496	// a cross-host preflight barrier before any target builds. ---
1497	{
1498	let ctx = ctx.clone();
1499	engine.register_fn(
1500	"checkout_sha",
1501	move \|host: &str\| -> Result<String, Box<EvalAltResult>> {
1502	ctx.checkout_sha(host).map_err(rhai_err)
1503	},
1504	);
1505	}
1506
1507	// --- crate_preflight() -> string: verify this crate is safe to publish,
1508	// or abort the run. Everything it checks is immutable once published:
1509	// crates.io versions can be yanked but never edited, so a wrong
1510	// repository URL is permanent. pter 0.1.0 shipped with a dead one. ---
1511	{
1512	let ctx = ctx.clone();
1513	engine.register_fn(
1514	"crate_preflight",
1515	move \|\| -> Result<String, Box<EvalAltResult>> {
1516	// `repo`, not `repo_for(...)`: `cargo metadata` runs on the
1517	// daemon's own box, so this is the one checkout that is always
1518	// the local one. It is not a missed call site.
1519	let repo = expand_tilde(&ctx.repo);
1520
1521	let out = std::process::Command::new("cargo")
1522	.args(["metadata", "--no-deps", "--format-version", "1"])
1523	.current_dir(&repo)
1524	.output()
1525	.map_err(\|e\| format!("running cargo metadata in {}: {e}", repo.display()))?;
1526	if !out.status.success() {
1527	return Err(format!(
1528	"cargo metadata failed in {}: {}",
1529	repo.display(),
1530	String::from_utf8_lossy(&out.stderr).trim()
1531	)
1532	.into());
1533	}
1534	let meta = crate_meta_from_json(&String::from_utf8_lossy(&out.stdout))
1535	.map_err(\|e\| e.to_string())?;
1536
1537	// The real question is not whether a page renders but whether a
1538	// stranger with no credentials can fetch the source, so ask git.
1539	let clonable = meta.repository.as_ref().is_some_and(\|url\| {
1540	std::process::Command::new("git")
1541	.args(["ls-remote", url])
1542	.env("GIT_TERMINAL_PROMPT", "0")
1543	.output()
1544	.is_ok_and(\|o\| o.status.success())
1545	});
1546
1547	// Ask the publishing host whether cargo has credentials, rather
1548	// than moving the token anywhere. It stays in cargo's own 0600
1549	// store; a shell line carrying it would be visible in `ps`.
1550	// An exit code answers "are there credentials"; an Err answers
1551	// "the question could not be asked". Collapsing the second into
1552	// the first reported a capability denial as "no crates.io
1553	// credentials", which sent a real diagnosis three rounds the
1554	// wrong way. A check that cannot run is not a failed check.
1555	let creds =
1556	ctx.run(
1557	&ctx.build_host.clone(),
1558	"cargo login --help >/dev/null 2>&1 && \
1559	test -s \"${CARGO_HOME:-$HOME/.cargo}/credentials.toml\" \
1560	\|\| test -s \"${CARGO_HOME:-$HOME/.cargo}/credentials\"",
1561	)
1562	.map_err(\|e\| {
1563	format!(
1564	"could not check crates.io credentials on `{}`: {e}",
1565	ctx.build_host
1566	)
1567	})?
1568	.0 == 0;
1569
1570	// Asked of the build host rather than the daemon: the tree
1571	// that gets published is the worktree over there, and it is the
1572	// one whose `[patch]` ancestry decides this. An Err is "the
1573	// question could not be asked" and is not a finding -- same
1574	// rule as the credentials check above, for the same reason it
1575	// was written that way.
1576	let build_repo = ctx.repo_for(&ctx.build_host).to_string();
1577	let patched_lock =
1578	ctx.run(
1579	&ctx.build_host.clone(),
1580	&tracked_lock_under_patch_cmd(&build_repo),
1581	)
1582	.map_err(\|e\| {
1583	format!(
1584	"could not check for a tracked Cargo.lock on `{}`: {e}",
1585	ctx.build_host
1586	)
1587	})?
1588	.0 == 0;
1589
1590	let published = RecipeCtx::published_versions(&meta.name);
1591	let mut problems = crate_publish_problems(&meta, clonable, &published, creds);
1592	if patched_lock {
1593	problems.push(tracked_lock_under_patch_problem(&build_repo));
1594	}
1595	if !problems.is_empty() {
1596	return Err(format!(
1597	"{} {} is not safe to publish:\n - {}",
1598	meta.name,
1599	meta.version,
1600	problems.join("\n - ")
1601	)
1602	.into());
1603	}
1604	Ok(format!("{} {} passed preflight", meta.name, meta.version))
1605	},
1606	);
1607	}
1608
1609	// --- feature_flags() -> string: `--features a,b`, or "" when the app
1610	// declares none. Returns the whole flag rather than a bare list so an
1611	// app with no features cannot produce a dangling `--features`. ---
1612	{
1613	let ctx = ctx.clone();
1614	engine.register_fn("feature_flags", move \|\| -> String {
1615	if ctx.features.is_empty() {
1616	String::new()
1617	} else {
1618	format!("--features {}", ctx.features.join(","))
1619	}
1620	});
1621	}
1622
1623	// --- target() / platform() / arch(): the target axis, for one per-platform
1624	// recipe to branch on arch (bundle paths differ between x86_64/aarch64). ---
1625	{
1626	let ctx = ctx.clone();
1627	engine.register_fn("target", move \|\| -> String { ctx.target.to_string() });
1628	}
1629	{
1630	let ctx = ctx.clone();
1631	engine.register_fn("platform", move \|\| -> String {
1632	ctx.target.platform.as_str().to_string()
1633	});
1634	}
1635	{
1636	let ctx = ctx.clone();
1637	engine.register_fn("arch", move \|\| -> String {
1638	ctx.target.arch.as_str().to_string()
1639	});
1640	}
1641
1642	// --- secret(key) -> string (file under secrets_root; never logged) ---
1643	{
1644	let ctx = ctx.clone();
1645	engine.register_fn("secret", move \|key: &str\| -> Result<String, Box<EvalAltResult>> {
1646	// Guard against traversal out of secrets_root. Require every path
1647	// component to be `Normal` (rejects `..`, `.`, absolute roots and
1648	// drive prefixes) and forbid backslashes (a literal filename char on
1649	// Linux, but a separator elsewhere) — the per-component strength of
1650	// Sando's `safe()`. A multi-segment key like `app/token` is still
1651	// allowed; `foo..bar` (a legit filename) is no longer falsely blocked.
1652	let safe = !key.is_empty()
1653	&& !key.contains('\\')
1654	&& std::path::Path::new(key)
1655	.components()
1656	.all(\|c\| matches!(c, std::path::Component::Normal(_)));
1657	if !safe {
1658	return Err(rhai_err(
1659	"secret key must be a relative path under secrets_root (no `..`, `.`, absolute paths, or backslashes)",
1660	));
1661	}
1662	let path = ctx.cfg.secrets_root.join(key);
1663	std::fs::read_to_string(&path)
1664	.map(\|s\| s.trim_end().to_string())
1665	.map_err(\|e\| rhai_err(format!("secret `{key}`: {e}")))
1666	});
1667	}
1668
1669	// --- env(host, key) -> string ---
1670	{
1671	let ctx = ctx.clone();
1672	engine.register_fn(
1673	"env",
1674	move \|host: &str, key: &str\| -> Result<String, Box<EvalAltResult>> {
1675	// The key is interpolated into a `${...}` shell expansion, so it must
1676	// be a bare shell identifier — anything else (quotes, `}`, `$`, `;`)
1677	// could break out and run arbitrary commands on the host. Validate
1678	// before building the command; this is the one env read that can't
1679	// sh-quote its argument (a quoted var name doesn't expand).
1680	if key.is_empty()
1681	\|\| !key
1682	.chars()
1683	.next()
1684	.is_some_and(\|c\| c == '_' \|\| c.is_ascii_alphabetic())
1685	\|\| !key.chars().all(\|c\| c == '_' \|\| c.is_ascii_alphanumeric())
1686	{
1687	return Err(rhai_err(format!(
1688	"env name `{key}` must be a shell identifier ([A-Za-z_][A-Za-z0-9_]*)"
1689	)));
1690	}
1691	// Read via the shell so it works on remote hosts too.
1692	let (code, tail) = ctx
1693	.run(host, &format!("printf '%s' \"${{{key}}}\""))
1694	.map_err(rhai_err)?;
1695	if code != 0 {
1696	return Err(rhai_err(format!("env `{key}` on `{host}` failed")));
1697	}
1698	Ok(tail.trim().to_string())
1699	},
1700	);
1701	}
1702
1703	// --- collect(host, glob, app, version): pull artifacts to dist_root ---
1704	{
1705	let ctx = ctx.clone();
1706	engine.register_fn(
1707	"collect",
1708	move \|host: &str,
1709	glob: &str,
1710	app: &str,
1711	version: &str\|
1712	-> Result<(), Box<EvalAltResult>> {
1713	ctx.collect(host, glob, app, version).map_err(rhai_err)
1714	},
1715	);
1716	}
1717
1718	// --- publish(channel, app, target, version, artifact, meta) ---
1719	{
1720	let ctx = ctx.clone();
1721	engine.register_fn(
1722	"publish",
1723	move \|channel: &str,
1724	app: &str,
1725	target: &str,
1726	version: &str,
1727	artifact: &str,
1728	meta: Map\|
1729	-> Result<String, Box<EvalAltResult>> {
1730	ctx.publish(channel, app, target, version, artifact, &meta)
1731	.map_err(rhai_err)
1732	},
1733	);
1734	}
1735
1736	// --- deploy(binary) -> summary: install a service binary and restart its
1737	// unit. The terminal step for `kind = "service"`, the counterpart of
1738	// `publish` for something that is run rather than distributed.
1739	//
1740	// Takes only the binary's path on the build host: where it lands, on
1741	// which machine, and which unit restarts all come from the `[[deploy]]`
1742	// entry for the target already being built. A recipe cannot deploy the
1743	// aarch64 binary to the x86_64 box by naming the wrong host, because it
1744	// never names a host at all.
1745	{
1746	let ctx = ctx.clone();
1747	engine.register_fn(
1748	"deploy",
1749	move \|binary: &str\| -> Result<String, Box<EvalAltResult>> {
1750	ctx.deploy(binary).map_err(rhai_err)
1751	},
1752	);
1753	}
1754
1755	// --- deploy_host() -> string: the service host's ssh destination, so a
1756	// recipe can run its own assertions there (`sh_ok(deploy_host(), ...)`).
1757	// Commands run through it while the `deploy` step is open, so they are
1758	// gated on the deploy grant like the install itself. ---
1759	{
1760	let ctx = ctx.clone();
1761	engine.register_fn(
1762	"deploy_host",
1763	move \|\| -> Result<String, Box<EvalAltResult>> {
1764	ctx.deploy_target()
1765	.map(\|d\| d.host.clone())
1766	.map_err(rhai_err)
1767	},
1768	);
1769	}
1770
1771	// --- service_name() / install_path() / health_url(): the rest of the
1772	// `[[deploy]]` entry, so a recipe asserts against the configured values
1773	// rather than repeating them as literals that can drift. `health_url`
1774	// is "" when unset. ---
1775	{
1776	let ctx = ctx.clone();
1777	engine.register_fn(
1778	"service_name",
1779	move \|\| -> Result<String, Box<EvalAltResult>> {
1780	ctx.deploy_target()
1781	.map(\|d\| d.service.clone())
1782	.map_err(rhai_err)
1783	},
1784	);
1785	}
1786	{
1787	let ctx = ctx.clone();
1788	engine.register_fn(
1789	"install_path",
1790	move \|\| -> Result<String, Box<EvalAltResult>> {
1791	ctx.deploy_target()
1792	.map(\|d\| d.install_path.clone())
1793	.map_err(rhai_err)
1794	},
1795	);
1796	}
1797	{
1798	let ctx = ctx.clone();
1799	engine.register_fn(
1800	"health_url",
1801	move \|\| -> Result<String, Box<EvalAltResult>> {
1802	ctx.deploy_target()
1803	.map(\|d\| d.health_url.clone().unwrap_or_default())
1804	.map_err(rhai_err)
1805	},
1806	);
1807	}
1808
1809	// --- glibc_check(binary) -> string: assert the build host did not produce
1810	// a binary the service host's glibc is too old to exec. Aborts the run
1811	// if it did; returns "needs X, host has Y" for the log if it did not.
1812	//
1813	// WHICH RECIPES CALL THIS, AND WHY THE OTHERS MUST NOT. The rule is not
1814	// a style preference and it is not optional: this reads the recipe's
1815	// `[[deploy]]` entry to learn which machine runs the bytes, so a recipe
1816	// with no `[[deploy]]` cannot call it at all.
1817	//
1818	// - A service that installs ITSELF (`[[deploy]]` present: magicmirror,
1819	// wam, mnw-cli) SHOULD call it. Bento is both builder and installer
1820	// there, so it knows the service host, and nothing downstream will
1821	// check on its behalf.
1822	// - A service HANDED OFF to Sando (`[[deploy]]` absent: pom) MUST NOT,
1823	// and the absence is the Sando/Bento boundary rather than an omission.
1824	// Which machine runs the bytes is environment knowledge, which is
1825	// Sando's half. Sando covers it on the far side, more strongly: it runs
1826	// the node's own loader against the rsynced bytes before the symlink
1827	// swap (`sando_daemon::deploy::ldd_guard_script`), and since 0.2.12
1828	// also compares the bundle's glibc floor against the node's declared
1829	// `libc` before the rsync (`check_bundle_fits_node`).
1830	//
1831	// So a new service recipe takes its answer from whether it carries a
1832	// `[[deploy]]` table, not from whichever sibling recipe it was copied
1833	// from. Wiki `sando-bento-boundary`, `host-base-images`. ---
1834	{
1835	let ctx = ctx.clone();
1836	engine.register_fn(
1837	"glibc_check",
1838	move \|binary: &str\| -> Result<String, Box<EvalAltResult>> {
1839	let (needs, has) = ctx.glibc_check(binary).map_err(rhai_err)?;
1840	Ok(format!(
1841	"glibc: binary needs {needs}, service host has {has}"
1842	))
1843	},
1844	);
1845	}
1846
1847	// --- macOS signing helpers. They dispatch through the named host's
1848	// executor like any other step; when that host is the mac (transport =
1849	// "agent"), codesign/notarize/staple ride the in-session `AgentRpc`
1850	// transport — the only security session where the Developer ID key is
1851	// usable (design §7 "THE WALL"). Capability-gated by the host's `sign`
1852	// grant. ---
1853	register_macos_fns(&mut engine, ctx);
1854
1855	engine
1856	}
1857
1858	/// Highest `GLIBC_x.y` version referenced by a built binary, and the glibc a
1859	/// host actually has, both parsed from the text the commands print.
1860	///
1861	/// Native-per-architecture builds remove the cross-compile hazard `deploy.sh`
1862	/// was written against, but not this one: fw13 tracks a newer glibc than the
1863	/// Ubuntu 24.04 box in Hetzner, so a binary built here can reference a symbol
1864	/// version that box does not have and fail at exec — after the unit has already
1865	/// been restarted onto it. Comparing the two before the install is what makes
1866	/// that a failed step instead of a downed service.
1867	fn max_glibc_symbol(objdump_out: &str) -> Option<(u64, u64)> {
1868	objdump_out
1869	.split(\|c: char\| !(c.is_ascii_digit() \|\| c == '.' \|\| c == '_' \|\| c.is_ascii_alphabetic()))
1870	.filter_map(\|tok\| tok.strip_prefix("GLIBC_"))
1871	.filter_map(parse_glibc_version)
1872	.max()
1873	}
1874
1875	/// Parse `2.39` (or `2.39.1`, keeping major/minor) into a comparable pair.
1876	fn parse_glibc_version(s: &str) -> Option<(u64, u64)> {
1877	let mut parts = s.split('.');
1878	let major = parts.next()?.parse().ok()?;
1879	let minor = parts.next()?.parse().ok()?;
1880	Some((major, minor))
1881	}
1882
1883	/// The glibc version out of `ldd --version`'s first line, whose tail is the
1884	/// version however the distro decorates the rest (`ldd (Ubuntu GLIBC
1885	/// 2.39-0ubuntu8.8) 2.39`).
1886	fn glibc_from_ldd(ldd_out: &str) -> Option<(u64, u64)> {
1887	let first = ldd_out.lines().find(\|l\| !l.trim().is_empty())?;
1888	parse_glibc_version(first.split_whitespace().last()?)
1889	}
1890
1891	impl RecipeCtx {
1892	/// This target's install destination, or an error naming why there is none.
1893	fn deploy_target(&self) -> Result<&DeployTarget> {
1894	self.deploy.as_ref().ok_or_else(\|\| {
1895	anyhow::anyhow!(
1896	"no deploy destination for {} {}: the app is `kind = \"{}\"`, and only a \
1897	service declares [[deploy]] entries",
1898	self.app,
1899	self.target,
1900	match self.kind {
1901	Kind::App => "app",
1902	Kind::Library => "library",
1903	Kind::Service => "service",
1904	}
1905	)
1906	})
1907	}
1908
1909	/// Compare the built binary's glibc requirement against the service host's.
1910	/// Returns the two versions for the recipe to log.
1911	fn glibc_check(self: &Arc<Self>, binary: &str) -> Result<(String, String)> {
1912	let d = self.deploy_target()?.clone();
1913	// `objdump -T` on the build host; no symbols at all (a static binary)
1914	// means nothing to check, which is a pass rather than a failure.
1915	let (code, out) = self.run(
1916	&self.build_host.clone(),
1917	&format!(
1918	"objdump -T {binary} 2>/dev/null \| grep -o 'GLIBC_[0-9.]*' \| sort -uV \|\| true"
1919	),
1920	)?;
1921	anyhow::ensure!(code == 0, "reading glibc symbols from {binary} failed");
1922	let Some(needs) = max_glibc_symbol(&out) else {
1923	return Ok(("none".into(), "n/a".into()));
1924	};
1925	let (code, ldd) = self.run(&d.host, "ldd --version")?;
1926	anyhow::ensure!(
1927	code == 0,
1928	"could not read glibc version on service host `{}`",
1929	d.host
1930	);
1931	let has = glibc_from_ldd(&ldd).ok_or_else(\|\| {
1932	anyhow::anyhow!(
1933	"could not parse glibc version from `ldd --version` on `{}`",
1934	d.host
1935	)
1936	})?;
1937	anyhow::ensure!(
1938	needs <= has,
1939	"binary needs glibc {}.{} but `{}` has {}.{} — it would fail to exec after the \
1940	unit restarted onto it. Build on a host no newer than the service host.",
1941	needs.0,
1942	needs.1,
1943	d.host,
1944	has.0,
1945	has.1,
1946	);
1947	Ok((
1948	format!("{}.{}", needs.0, needs.1),
1949	format!("{}.{}", has.0, has.1),
1950	))
1951	}
1952
1953	/// Install `binary` (a path on the BUILD host) onto the service host and
1954	/// restart its unit, via the privileged installer the host holds a scoped
1955	/// sudo grant for.
1956	///
1957	/// Bento never runs the install itself. It stages the bytes and calls a
1958	/// root script whose arguments are re-checked on the far side — the same
1959	/// shape as Sando's `install-companion.sh`, and for the same reason: the
1960	/// sudoers grant is then ONE auditable script rather than a broad
1961	/// `install`+`systemctl` grant on a production box.
1962	///
1963	/// Only the binary moves. Config is deliberately untouched: pom's
1964	/// `pom-astra.toml` / `pom-hetzner.toml` differ per instance, and prod's
1965	/// carried a `[targets.mnw.tests]` block the repo did not have. A deploy
1966	/// that copies config over is how that block gets silently deleted.
1967	fn deploy(self: &Arc<Self>, binary: &str) -> Result<String> {
1968	anyhow::ensure!(
1969	!self.is_cancelled(),
1970	"build superseded by a newer request; refusing to deploy"
1971	);
1972	// A failed earlier step bars a deploy exactly as it bars a publish. An
1973	// artifact that failed its gates must not reach a production host just
1974	// because the recipe kept running.
1975	let failed = self.failed_steps.lock().unwrap().clone();
1976	anyhow::ensure!(
1977	failed.is_empty(),
1978	"refusing to deploy {} {}: {} failed earlier in this run",
1979	self.app,
1980	self.version,
1981	failed
1982	.iter()
1983	.map(ToString::to_string)
1984	.collect::<Vec<_>>()
1985	.join(", "),
1986	);
1987	let d = self.deploy_target()?.clone();
1988	ensure_glob_safe(binary)?;
1989
1990	// Stage under a fixed root the installer also insists on, so "what was
1991	// checked" and "what is installed" cannot drift apart.
1992	let staged = format!("{DEPLOY_STAGING_ROOT}/{}", self.app);
1993	let staged_bin = format!("{staged}/{}", self.app);
1994	let deploy_exec = self.exec(&d.host)?;
1995	anyhow::ensure!(
1996	deploy_exec.capabilities().permits(&Action::Deploy),
1997	"service host `{}` is not granted the `deploy` capability",
1998	d.host
1999	);
2000
2001	self.run_ok(&d.host, &format!("mkdir -p {staged}"))?;
2002	if self.build_host_ssh == d.host {
2003	// Same box: the binary is already there. Routing it through the
2004	// daemon would be two transfers to end up where it started. This is
2005	// pom's aarch64 leg — astra builds it and astra runs it.
2006	self.run_ok(&d.host, &format!("cp -f {binary} {staged_bin}"))?;
2007	} else {
2008	// Build host -> daemon -> service host. Two hops because an executor
2009	// reaches one host; a direct host-to-host transport would mean the
2010	// build host holding a credential for the production box.
2011	let tmp = tempfile::tempdir().context("staging dir for deploy")?;
2012	let local = tmp.path().join(self.app.as_str());
2013	self.pull_for_deploy(binary, &local)?;
2014	let (dest, opts) = (PathBuf::from(&staged), SyncOpts::default());
2015	let dir = tmp.path().to_path_buf();
2016	self.run_bounded(&format!("stage {} on `{}`", self.app, d.host), async move {
2017	deploy_exec.push_dir(&dir, &dest, &opts).await
2018	})
2019	.with_context(\|\| format!("staging {} onto `{}`", self.app, d.host))?;
2020	}
2021
2022	// The privileged half. Every argument is re-validated by the script,
2023	// which is the thing actually holding the sudo grant.
2024	self.run_ok(
2025	&d.host,
2026	&format!(
2027	"{} {staged_bin} {} {}",
2028	self.cfg.deploy_installer, d.install_path, d.service
2029	),
2030	)?;
2031	Ok(format!(
2032	"{} {} installed at {} on `{}`; {} restarted",
2033	self.app, self.version, d.install_path, d.host, d.service
2034	))
2035	}
2036
2037	/// Fetch one file off a host into a daemon-local path for re-pushing.
2038	///
2039	/// A local build host is read directly: `fw13` is the daemon's own box, so
2040	/// the file is already on this filesystem. Routing it through the
2041	/// artifact-pull gate instead would demand a `pull_root` covering every repo
2042	/// a service could be built in — today that is `~/Code/Apps`, and pom lives
2043	/// in `~/Code/MNW`. Widening it to `~/Code` would put `_private`, the
2044	/// secrets root, inside the collectable tree. This is pom's x86_64 leg.
2045	fn pull_for_deploy(self: &Arc<Self>, remote: &str, local: &Path) -> Result<()> {
2046	let host = self.build_host.clone();
2047	let remote_path = expand_tilde(remote);
2048	if self.build_host_ssh == "local" \|\| self.build_host_ssh.is_empty() {
2049	std::fs::copy(&remote_path, local).with_context(\|\| {
2050	format!("staging {} from the daemon host", remote_path.display())
2051	})?;
2052	return Ok(());
2053	}
2054	let sync = self.host_sync(&host)?;
2055	let (src, dst, opts) = (remote_path, local.to_path_buf(), SyncOpts::default());
2056	self.run_bounded(&format!("fetch {remote} from `{host}`"), async move {
2057	sync.pull_file(&src, &dst, &opts).await
2058	})
2059	.with_context(\|\| format!("fetching {remote} from `{host}` to deploy"))
2060	}
2061
2062	/// `run`, failing the step on a non-zero exit. The Rust-side twin of the
2063	/// recipe's `sh_ok`, for commands the deploy machinery issues itself.
2064	fn run_ok(self: &Arc<Self>, host: &str, cmd: &str) -> Result<String> {
2065	let (code, tail) = self.run(host, cmd)?;
2066	if code != 0 {
2067	self.fail_current_step();
2068	anyhow::bail!("command on `{host}` exited {code}: {cmd}\n{tail}");
2069	}
2070	Ok(tail)
2071	}
2072
2073	/// Where this run's collected files land locally.
2074	///
2075	/// Per target, not per version. Sharing one `dist_root/<app>/<version>/`
2076	/// across targets would have the hash loop below (which lists the directory)
2077	/// attribute a sibling's AppImage to the mac build's artifact record. It is
2078	/// also the layout the archive uses, and the two have to agree or the local
2079	/// copy and the deposited one are different shapes.
2080	fn collect_dest(&self, app: &str, version: &str) -> PathBuf {
2081	self.cfg
2082	.dist_root
2083	.join(app)
2084	.join(version)
2085	.join(crate::archive::target_slug(self.target))
2086	}
2087
2088	fn collect(self: &Arc<Self>, host: &str, glob: &str, app: &str, version: &str) -> Result<()> {
2089	let dest = self.collect_dest(app, version);
2090	let dest_s = dest.to_string_lossy().into_owned();
2091	// The glob reaches a remote login shell intact (that's what expands it),
2092	// so command metacharacters stay barred. Path/wildcard chars are fine.
2093	ensure_glob_safe(glob)?;
2094	std::fs::create_dir_all(&dest)
2095	.with_context(\|\| format!("creating collect dest {dest_s}"))?;
2096	// The SYNC transport, not the host's exec executor: artifacts move over
2097	// ssh/rsync even from an agent host, whose `/pull` is confined to a
2098	// narrow `pull_root` that deliberately excludes the repo checkout these
2099	// artifacts are built in (see `state::build_sync`). The daemon still
2100	// runs the transfer itself, as it always has.
2101	let sync = self.host_sync(host)?;
2102	let opts = SyncOpts::precompressed();
2103	// Bounded by the collect step's deadline (rsync of a multi-GiB artifact
2104	// can wedge on a stalled transport) and interruptible on supersession.
2105	let dest_pull = dest.clone();
2106	self.run_bounded(&format!("collect {glob} from `{host}`"), async move {
2107	sync.pull_glob(glob, &dest_pull, &opts).await
2108	})
2109	.with_context(\|\| format!("collect {glob} from `{host}`"))?;
2110	// Assert the version and hash every collected file. This is where a
2111	// stale artifact is caught: a file whose name embeds a different version
2112	// fails the collect (rather than silently winning a later glob), and the
2113	// sha256 recorded here is what `publish` writes into the release ledger
2114	// and what the artifact record's manifest is built from.
2115	//
2116	// Recursive, and keyed by path relative to the collect dir. That is not
2117	// a preference: Sando's intake re-hashes the bundle with its own walker,
2118	// which recurses and keys the same way, and refuses a bundle whose bytes
2119	// do not match the manifest it was handed. A top-level `read_dir` keyed
2120	// by file name agrees with that walker for a flat directory and diverges
2121	// the moment a bundle carries a subdirectory — the honest artifact would
2122	// be refused for a manifest that omitted everything nested. The two
2123	// walkers have to be the same walk. See `bundle::digest_dir` in sando.
2124	for (rel, path) in
2125	collected_files(&dest).with_context(\|\| format!("listing collect dest {dest_s}"))?
2126	{
2127	// The version check stays on the file NAME rather than the relative
2128	// path: it is looking for a stale `app_1.2.3.AppImage` beside the
2129	// one this release built, and a directory component is not that.
2130	let name = path
2131	.file_name()
2132	.map_or_else(\|\| rel.clone(), \|n\| n.to_string_lossy().into_owned());
2133	assert_artifact_version(&name, &self.version)?;
2134	let digest = sha256_file(&path)?;
2135	self.artifact_hashes.lock().unwrap().insert(rel, digest);
2136	}
2137	// Deposit at the archive path, so this target's bytes have one address
2138	// whichever host produced them. A no-op when no archive is configured.
2139	//
2140	// Inside `collect`, not after the recipe: a failure here fails the
2141	// collect step, before sign and publish, rather than putting a red mark
2142	// on a release that has already shipped. And it is a failure, not a
2143	// warning — a deposit that is quietly skipped leaves the archive path
2144	// wrong for exactly the release nobody was watching, which is the thing
2145	// having one address is for.
2146	let (cfg, app_id, version, target) = (
2147	self.cfg.clone(),
2148	self.app.clone(),
2149	self.version.clone(),
2150	self.target,
2151	);
2152	let dest_archive = dest.clone();
2153	self.run_bounded("deposit in the archive", async move {
2154	crate::archive::deposit(&cfg, &dest_archive, &app_id, &version, target).await
2155	})?;
2156	// Best-effort size accounting for the event.
2157	events::emit(
2158	&self.events,
2159	Event::ArtifactCollected {
2160	app: self.app.clone(),
2161	target: self.target,
2162	path: dest_s,
2163	bytes: dir_size(&dest).unwrap_or(0),
2164	},
2165	);
2166	Ok(())
2167	}
2168
2169	/// The all-targets-green gate: err unless every declared target OTHER than
2170	/// the one publishing has a latest `target_runs` row of `ok` for this
2171	/// `(app, version)`. A sibling with no run, a running run, or a failed
2172	/// latest run all block the publish, naming what is not green.
2173	fn assert_siblings_green(self: &Arc<Self>, declared: &[Target]) -> Result<()> {
2174	let me = self.clone();
2175	let (app_s, ver_s) = (self.app.to_string(), self.version.to_string());
2176	let rows: Vec<(String, String)> = self.rt.block_on(async move {
2177	sqlx::query_as(
2178	"SELECT target, status FROM target_runs tr
2179	WHERE app = ?1 AND version = ?2
2180	AND id = (SELECT MAX(id) FROM target_runs
2181	WHERE app = ?1 AND version = ?2 AND target = tr.target)",
2182	)
2183	.bind(app_s)
2184	.bind(ver_s)
2185	.fetch_all(&me.pool)
2186	.await
2187	.unwrap_or_default()
2188	});
2189	let status_of = \|t: &Target\| -> Option<String> {
2190	let key = t.to_string();
2191	rows.iter()
2192	.find(\|(name, _)\| name == &key)
2193	.map(\|(_, s)\| s.clone())
2194	};
2195	let not_green: Vec<String> = declared
2196	.iter()
2197	.filter(\|t\| **t != self.target) // the publishing target is the last mile
2198	.filter(\|t\| status_of(t).as_deref() != Some("ok"))
2199	.map(\|t\| format!("{t} ({})", status_of(t).unwrap_or_else(\|\| "no run".into())))
2200	.collect();
2201	anyhow::ensure!(
2202	not_green.is_empty(),
2203	"all-targets-green gate: refusing to publish {} {} — not green: {}",
2204	self.app,
2205	self.version,
2206	not_green.join(", "),
2207	);
2208	Ok(())
2209	}
2210
2211	fn publish(
2212	self: &Arc<Self>,
2213	channel: &str,
2214	app: &str,
2215	target: &str,
2216	version: &str,
2217	artifact: &str,
2218	meta: &Map,
2219	) -> Result<String> {
2220	// Never let a superseded build ship. This is the last and most important
2221	// cooperative-cancel checkpoint: even if a long-running step finished
2222	// after supersession, the artifact must not reach the backend.
2223	anyhow::ensure!(
2224	!self.is_cancelled(),
2225	"build superseded by a newer request; refusing to publish"
2226	);
2227	// Opt-in all-targets-green gate: refuse a partial release. Every OTHER
2228	// declared target of this (app, version) must have a successful latest
2229	// run before this one ships, so macOS can't publish while windows is red
2230	// or still building. The publishing target itself is the last mile (it
2231	// reached publish, so its steps passed) and is not required to be green
2232	// in the ledger yet.
2233	if let Some(declared) = self.all_green_required.clone() {
2234	self.assert_siblings_green(&declared)?;
2235	}
2236	let backend = self
2237	.ota
2238	.get(channel)
2239	.ok_or_else(\|\| anyhow::anyhow!("unknown publish channel `{channel}`"))?;
2240	let target: Target = target.parse().map_err(\|e: String\| anyhow::anyhow!(e))?;
2241	let version = Version::parse(version).map_err(\|e\| anyhow::anyhow!(e))?;
2242	let app = AppId::new(app);
2243
2244	// The backend must actually handle this target (e.g. the desktop updater
2245	// disclaims iOS) — otherwise publish would push an artifact through a
2246	// backend that does not support it.
2247	anyhow::ensure!(
2248	backend.supports(target),
2249	"publish channel `{channel}` does not support target {target}",
2250	);
2251
2252	// Monotonicity: never publish a version that is not strictly newer than
2253	// the latest already published for this (app, target, channel). Without
2254	// this an older build could republish over a live newer release. The
2255	// `releases` column is TEXT, so compare by parsed semver precedence
2256	// (Version: Ord), not lexically.
2257	{
2258	let (app_s, target_s, chan_s) =
2259	(app.to_string(), target.to_string(), channel.to_string());
2260	let me = self.clone();
2261	let latest: Option<Version> = self.rt.block_on(async move {
2262	let rows: Vec<(String,)> = sqlx::query_as(
2263	"SELECT version FROM releases WHERE app = ? AND target = ? AND channel = ?",
2264	)
2265	.bind(app_s)
2266	.bind(target_s)
2267	.bind(chan_s)
2268	.fetch_all(&me.pool)
2269	.await
2270	.unwrap_or_default();
2271	rows.into_iter()
2272	.filter_map(\|(v,)\| Version::parse(&v).ok())
2273	.max()
2274	});
2275	if let Some(latest) = latest {
2276	anyhow::ensure!(
2277	version > latest,
2278	"refusing to publish {app} {version} to `{channel}` ({target}): \
2279	not newer than the last published {latest}",
2280	);
2281	}
2282	}
2283
2284	// Step-success ledger (the Bento analogue of Sando's gate fail-closed),
2285	// minted as an unforgeable PublishAuthority. `backend.publish` cannot be
2286	// called without one, so the unverified/post-failure ship path is sealed
2287	// at the type level rather than guarded by a separate runtime check.
2288	let authority = {
2289	let failed = self.failed_steps.lock().unwrap();
2290	let gatekeeper = *self.gatekeeper_ok.lock().unwrap();
2291	PublishAuthority::prove(target, failed.as_slice(), gatekeeper)?
2292	};
2293	let notes = meta
2294	.get("notes")
2295	.and_then(\|v\| v.clone().into_string().ok())
2296	.unwrap_or_default();
2297	// Resolve the artifact relative to the collected dist dir if not absolute.
2298	let artifact_path = {
2299	let p = PathBuf::from(artifact);
2300	if p.is_absolute() {
2301	p
2302	} else {
2303	self.collect_dest(app.as_str(), &version.to_string())
2304	.join(artifact)
2305	}
2306	};
2307	let rel = Release {
2308	app: &app,
2309	target,
2310	version: &version,
2311	notes,
2312	};
2313	let receipt = backend
2314	.publish(&rel, &artifact_path, &authority)
2315	.with_context(\|\| format!("publish to `{channel}`"))?;
2316	// Record for idempotency / monotonicity. This write is CHECKED, not
2317	// fire-and-forget: a swallowed failure here would silently re-arm the
2318	// monotonicity guard (which reads this same table), letting an older
2319	// version republish over a live release. Concurrent same-(app,target)
2320	// publishers can't race the read-then-insert because the latest-wins slot
2321	// (state::ActiveSlot) serializes them and a superseded run is cancelled
2322	// before it reaches publish.
2323	// The artifact's hash, recorded so the release ledger says exactly which
2324	// bytes shipped. Prefer the digest computed at `collect`; fall back to
2325	// hashing the file now (an absolute-path artifact never routed through
2326	// `collect`). A hash failure must not fail an already-published release,
2327	// so degrade to NULL rather than erroring.
2328	let artifact_hash: Option<String> = artifact_path
2329	.file_name()
2330	.and_then(\|n\| n.to_str())
2331	.and_then(\|n\| self.artifact_hashes.lock().unwrap().get(n).cloned())
2332	.or_else(\|\| sha256_file(&artifact_path).ok());
2333	let me = self.clone();
2334	let (app_s, target_s, ver_s, chan_s) = (
2335	app.to_string(),
2336	target.to_string(),
2337	version.to_string(),
2338	channel.to_string(),
2339	);
2340	self.rt
2341	.block_on(async move {
2342	sqlx::query(
2343	"INSERT OR IGNORE INTO releases (app, target, version, channel, artifact_hash, published_at)
2344	VALUES (?, ?, ?, ?, ?, ?)",
2345	)
2346	.bind(app_s)
2347	.bind(target_s)
2348	.bind(ver_s)
2349	.bind(chan_s)
2350	.bind(artifact_hash)
2351	.bind(Self::now())
2352	.execute(&me.pool)
2353	.await
2354	})
2355	.context("recording release in the idempotency ledger (artifact published but ledger write failed)")?;
2356	events::emit(
2357	&self.events,
2358	Event::PublishOk {
2359	app: self.app.clone(),
2360	target: self.target,
2361	channel: channel.to_string(),
2362	},
2363	);
2364	Ok(receipt)
2365	}
2366	}
2367
2368	fn dir_size(p: &Path) -> Option<i64> {
2369	let mut total = 0i64;
2370	for entry in std::fs::read_dir(p).ok()? {
2371	let entry = entry.ok()?;
2372	let md = entry.metadata().ok()?;
2373	if md.is_file() {
2374	total += md.len() as i64;
2375	} else if md.is_dir() {
2376	// Recurse so a bundle dir (a `.app`) reports its real size, not ~0.
2377	total += dir_size(&entry.path()).unwrap_or(0);
2378	}
2379	}
2380	Some(total)
2381	}
2382
2383	/// macOS signing/notarization host functions. Thin wrappers over the right
2384	/// shell incantations, dispatched through the named host's executor. On the mac
2385	/// host (`transport = "agent"`) they run via the in-session `ops-agent`, the only
2386	/// context where codesign can use the Developer ID key (design §7 "THE WALL"); a
2387	/// plain SSH session cannot. Each is gated by the host's `sign` capability.
2388	fn register_macos_fns(engine: &mut Engine, ctx: &Arc<RecipeCtx>) {
2389	{
2390	let ctx = ctx.clone();
2391	engine.register_fn(
2392	"verify_gatekeeper",
2393	move \|host: &str, path: &str\| -> Result<bool, Box<EvalAltResult>> {
2394	// spctl has no JSON mode, so assess on-host and decide there,
2395	// emitting an unambiguous sentinel as the final line. We match the
2396	// sentinel rather than substring-hunting `source=Notarized...` in a
2397	// 2000-char tail: truncation only drops the front, so the sentinel
2398	// is always present, and it can't be spoofed by spctl's own prose.
2399	// The full assess output is still streamed to the step log.
2400	let q = ops_core::remote::sh_quote(path);
2401	let cmd = format!(
2402	"out=$(spctl --assess -vv --type install {q} 2>&1); printf '%s\\n' \"$out\"; \
2403	printf '%s' \"$out\" \| grep -q 'source=Notarized Developer ID' \
2404	&& echo BENTO_GATEKEEPER_OK \|\| echo BENTO_GATEKEEPER_FAIL",
2405	);
2406	let (_, tail) = ctx.run(host, &cmd).map_err(rhai_err)?;
2407	let accepted = tail.contains("BENTO_GATEKEEPER_OK");
2408	// Record the verdict for the publish gate. A rejection also
2409	// fails the step, so the matrix shows red and `publish` is barred
2410	// even if the recipe ignores the returned bool.
2411	*ctx.gatekeeper_ok.lock().unwrap() = Some(accepted);
2412	if !accepted {
2413	ctx.fail_current_step();
2414	}
2415	Ok(accepted)
2416	},
2417	);
2418	}
2419	{
2420	let ctx = ctx.clone();
2421	engine.register_fn(
2422	"codesign",
2423	move \|host: &str, identity: &str, path: &str\| -> Result<(), Box<EvalAltResult>> {
2424	let cmd = format!(
2425	"codesign --force --options runtime --timestamp --sign {} {}",
2426	ops_core::remote::sh_quote(identity),
2427	ops_core::remote::sh_quote(path),
2428	);
2429	let (code, _) = ctx.run(host, &cmd).map_err(rhai_err)?;
2430	if code != 0 {
2431	return Err(rhai_err("codesign failed"));
2432	}
2433	Ok(())
2434	},
2435	);
2436	}
2437	{
2438	let ctx = ctx.clone();
2439	engine.register_fn(
2440	"staple",
2441	move \|host: &str, path: &str\| -> Result<(), Box<EvalAltResult>> {
2442	let (code, _) = ctx
2443	.run(
2444	host,
2445	&format!("xcrun stapler staple {}", ops_core::remote::sh_quote(path)),
2446	)
2447	.map_err(rhai_err)?;
2448	if code != 0 {
2449	return Err(rhai_err("stapler failed"));
2450	}
2451	Ok(())
2452	},
2453	);
2454	}
2455	{
2456	let ctx = ctx.clone();
2457	engine.register_fn(
2458	"notarize",
2459	move \|host: &str, path: &str\| -> Result<String, Box<EvalAltResult>> {
2460	ctx.notarize(host, path).map_err(rhai_err)
2461	},
2462	);
2463	}
2464	{
2465	let ctx = ctx.clone();
2466	engine.register_fn(
2467	"keychain_open",
2468	move \|host: &str, name: &str\| -> Result<(), Box<EvalAltResult>> {
2469	// The full build-keychain lifecycle lives in dist/build-keychain.sh
2470	// (design §7); this drives it by name so the recipe stays short.
2471	let (code, _) = ctx
2472	.run(
2473	host,
2474	&format!(
2475	". ~/.tauri/passwords.env && ./dist/build-keychain.sh open {}",
2476	ops_core::remote::sh_quote(name)
2477	),
2478	)
2479	.map_err(rhai_err)?;
2480	if code != 0 {
2481	return Err(rhai_err("keychain_open failed"));
2482	}
2483	Ok(())
2484	},
2485	);
2486	}
2487	{
2488	let ctx = ctx.clone();
2489	engine.register_fn(
2490	"keychain_close",
2491	move \|host: &str, name: &str\| -> Result<(), Box<EvalAltResult>> {
2492	let _ = ctx.run(
2493	host,
2494	&format!(
2495	"./dist/build-keychain.sh close {}",
2496	ops_core::remote::sh_quote(name)
2497	),
2498	);
2499	Ok(())
2500	},
2501	);
2502	}
2503	}
2504
2505	impl RecipeCtx {
2506	/// `xcrun notarytool submit --wait` with bounded retry (the one flaky,
2507	/// network-bound step). Emits `NotarizeRetry` per attempt.
2508	fn notarize(self: &Arc<Self>, host: &str, path: &str) -> Result<String> {
2509	const MAX_ATTEMPTS: u32 = 3;
2510	let backoff = self
2511	.cfg
2512	.notarize_backoff_secs
2513	.map_or(std::time::Duration::from_secs(15), \|s\| {
2514	std::time::Duration::from_secs(s)
2515	});
2516	let cmd = format!(
2517	". ~/.tauri/passwords.env && xcrun notarytool submit {} \
2518	--key \"$NOTARY_KEY\" --key-id \"$NOTARY_KEY_ID\" --issuer \"$NOTARY_ISSUER\" \
2519	--wait --output-format json",
2520	ops_core::remote::sh_quote(path),
2521	);
2522	let mut last = String::new();
2523	for attempt in 1..=MAX_ATTEMPTS {
2524	let (code, tail) = self.run(host, &cmd)?;
2525	if code == 0 && notary_accepted(&tail) {
2526	return Ok(tail);
2527	}
2528	last = tail;
2529	if attempt < MAX_ATTEMPTS {
2530	events::emit(
2531	&self.events,
2532	Event::NotarizeRetry {
2533	app: self.app.clone(),
2534	target: self.target,
2535	attempt,
2536	reason: format!("exit {code}"),
2537	},
2538	);
2539	self.rt.block_on(tokio::time::sleep(backoff));
2540	}
2541	}
2542	anyhow::bail!("notarization failed after {MAX_ATTEMPTS} attempts: {last}")
2543	}
2544	}
2545
2546	/// True iff `notarytool --output-format json` output reports `status: Accepted`.
2547	/// Isolates the JSON object (`{`..`}`) from any shell-sourcing noise and reads
2548	/// the typed `status` field, rather than substring-matching `"status":"Accepted"`
2549	/// in a possibly-truncated tail — which could match the literal inside an error
2550	/// message or miss it across a whitespace variant. Fails closed: any parse or
2551	/// field miss returns false.
2552	fn notary_accepted(output: &str) -> bool {
2553	let (Some(start), Some(end)) = (output.find('{'), output.rfind('}')) else {
2554	return false;
2555	};
2556	if start > end {
2557	return false;
2558	}
2559	serde_json::from_str::<serde_json::Value>(&output[start..=end])
2560	.ok()
2561	.and_then(\|v\| {
2562	v.get("status")
2563	.and_then(\|s\| s.as_str())
2564	.map(\|s\| s.eq_ignore_ascii_case("accepted"))
2565	})
2566	.unwrap_or(false)
2567	}
2568
2569	#[cfg(test)]
2570	mod tests {
2571	use super::*;
2572
2573	/// Build the shared cross-crate bundle fixture under `root`.
2574	///
2575	/// A binary at the top and two files in a subdirectory: the shape a service
2576	/// that ships its migrations has, and the case a flat walk gets wrong.
2577	pub(crate) fn write_bundle_fixture(root: &Path) {
2578	std::fs::create_dir_all(root.join("migrations")).unwrap();
2579	std::fs::write(root.join("pom"), b"binary-bytes").unwrap();
2580	std::fs::write(root.join("migrations/001_init.sql"), b"create table a;").unwrap();
2581	std::fs::write(root.join("migrations/002_next.sql"), b"alter table a;").unwrap();
2582	}
2583
2584	/// The manifest text the fixture must produce, in BOTH crates.
2585	///
2586	/// Sando's `bundle::digest_dir` has the identical constant and the identical
2587	/// fixture. That is the whole point: bento writes this text into the artifact
2588	/// record, sando recomputes it from the bytes that arrive, and an artifact is
2589	/// refused when they differ. Two walks, one answer, pinned from both ends —
2590	/// if either crate's walk drifts, its own test fails and names the drift
2591	/// rather than a release failing intake for a bundle nothing is wrong with.
2592	pub(crate) const BUNDLE_FIXTURE_MANIFEST: &str = concat!(
2593	"e4c908e219c533fa7ad7ea1634398f9bf51637ba20717769ada545bab26d7368 migrations/001_init.sql\n",
2594	"b026fd51bae096b34672cefdb781b6585b13efb53bc301d50c305f422552a380 migrations/002_next.sql\n",
2595	"71227a7f160afca3fb3c39f448735886dda7bd366252580c2222fb87d4bb4d85 pom\n",
2596	);
2597
2598	/// The producer half of the contract above: what `collect` hashes, turned
2599	/// into a manifest, is exactly the text the verifier will recompute.
2600	///
2601	/// Nested files are included and addressed by relative path. Before this,
2602	/// `collect` listed only the top level, so `migrations/` contributed nothing
2603	/// to the manifest while sando's walker hashed both files in it — and the
2604	/// honest bundle was refused for a manifest that had omitted them.
2605	#[test]
2606	fn a_collected_bundle_manifests_exactly_as_the_verifier_will_read_it() {
2607	let dir = tempfile::tempdir().unwrap();
2608	write_bundle_fixture(dir.path());
2609
2610	let files = collected_files(dir.path()).unwrap();
2611	assert_eq!(
2612	files.iter().map(\|(r, _)\| r.as_str()).collect::<Vec<_>>(),
2613	vec!["migrations/001_init.sql", "migrations/002_next.sql", "pom"],
2614	"recursive, relative, sorted"
2615	);
2616
2617	let hashes: Vec<(String, String)> = files
2618	.into_iter()
2619	.map(\|(rel, path)\| (rel, sha256_file(&path).unwrap()))
2620	.collect();
2621	let manifest = ops_artifact::Manifest::new(hashes).unwrap();
2622	assert_eq!(manifest.to_text(), BUNDLE_FIXTURE_MANIFEST);
2623	}
2624
2625	/// A symlink is neither followed nor named. Following one would let bytes
2626	/// from outside the bundle into its identity; naming it would put a path in
2627	/// the manifest the verifier does not hash, which reads as a corrupt bundle.
2628	#[test]
2629	#[cfg(unix)]
2630	fn a_symlink_in_the_collect_dir_is_not_part_of_the_bundle() {
2631	let dir = tempfile::tempdir().unwrap();
2632	write_bundle_fixture(dir.path());
2633	let outside = dir.path().join("..").join("secret.env");
2634	std::fs::write(&outside, b"TOKEN=1").ok();
2635	std::os::unix::fs::symlink(&outside, dir.path().join("link.env")).unwrap();
2636
2637	let files = collected_files(dir.path()).unwrap();
2638	assert!(
2639	!files.iter().any(\|(rel, _)\| rel.contains("link.env")),
2640	"{files:?}"
2641	);
2642	}
2643
2644	/// Run 3 S1: once the cooperative cancel flag is set (a newer build
2645	/// superseded this run), a step boundary refuses to proceed — the blocking
2646	/// recipe stops at the next `step()` instead of running on and publishing.
2647	#[tokio::test]
2648	async fn begin_step_bails_when_cancelled() {
2649	let dir = tempfile::tempdir().unwrap();
2650	let cfg = Arc::new(Config::for_tests(dir.path()));
2651	let pool = crate::db::open(&cfg.db_path).await.unwrap();
2652	let cancel = Arc::new(AtomicBool::new(true));
2653	let ctx = Arc::new(RecipeCtx::new(
2654	AppId::new("demo"),
2655	Version::parse("0.1.0").unwrap(),
2656	"linux/x86_64".parse().unwrap(),
2657	"fw13".into(),
2658	"local".into(),
2659	"v0.1.0".into(),
2660	"/tmp".into(),
2661	vec![],
2662	Kind::App,
2663	1,
2664	Arc::new(std::collections::HashMap::new()),
2665	Arc::new(std::collections::HashMap::new()),
2666	None,
2667	pool,
2668	crate::events::channel(),
2669	cfg,
2670	Arc::new(OtaRegistry::standard("https://makenot.work")),
2671	tokio::runtime::Handle::current(),
2672	cancel.clone(),
2673	None,
2674	));
2675	// Cancelled: begin_step refuses before touching the DB (the ensure! is
2676	// ahead of any block_on, so this is safe to call from the async test).
2677	let err = ctx.begin_step(Step::Build).unwrap_err();
2678	assert!(err.to_string().contains("supersede"), "got: {err}");
2679	assert!(ctx.is_cancelled());
2680	}
2681
2682	/// `feature_flags()` returns a whole flag or nothing at all. An app with
2683	/// no declared features must not yield a bare `--features`, which would
2684	/// swallow the next word of the build command as its argument.
2685	#[tokio::test]
2686	async fn feature_flags_renders_whole_flag_or_empty() {
2687	async fn flags_for(features: Vec<String>) -> String {
2688	let dir = tempfile::tempdir().unwrap();
2689	let cfg = Arc::new(Config::for_tests(dir.path()));
2690	let pool = crate::db::open(&cfg.db_path).await.unwrap();
2691	let ctx = Arc::new(RecipeCtx::new(
2692	AppId::new("demo"),
2693	Version::parse("0.1.0").unwrap(),
2694	"linux/x86_64".parse().unwrap(),
2695	"fw13".into(),
2696	"local".into(),
2697	"v0.1.0".into(),
2698	"/tmp".into(),
2699	features,
2700	Kind::App,
2701	1,
2702	Arc::new(std::collections::HashMap::new()),
2703	Arc::new(std::collections::HashMap::new()),
2704	None,
2705	pool,
2706	crate::events::channel(),
2707	cfg,
2708	Arc::new(OtaRegistry::standard("https://makenot.work")),
2709	tokio::runtime::Handle::current(),
2710	Arc::new(AtomicBool::new(false)),
2711	None,
2712	));
2713	let engine = build_engine(&ctx);
2714	engine.eval::<String>("feature_flags()").unwrap()
2715	}
2716
2717	assert_eq!(flags_for(vec![]).await, "");
2718	assert_eq!(
2719	flags_for(vec!["supernote".into()]).await,
2720	"--features supernote"
2721	);
2722	assert_eq!(
2723	flags_for(vec!["supernote".into(), "extra".into()]).await,
2724	"--features supernote,extra"
2725	);
2726	}
2727
2728	/// `repo()` answers for the host this target builds on, not for the daemon.
2729	///
2730	/// This is what lets a Windows recipe call `repo()` and `checkout_sha(h)`
2731	/// instead of hard-coding `C:/Users/me/...` — and hard-coding it is what
2732	/// kept those recipes off the release-tag pin, since `checkout_sha` builds
2733	/// its git commands from the app's path and takes no override.
2734	#[tokio::test]
2735	async fn repo_resolves_per_build_host() {
2736	async fn repo_on(build_host: &str) -> String {
2737	let dir = tempfile::tempdir().unwrap();
2738	let cfg = Arc::new(Config::for_tests(dir.path()));
2739	let pool = crate::db::open(&cfg.db_path).await.unwrap();
2740	let ctx = Arc::new(
2741	RecipeCtx::new(
2742	AppId::new("demo"),
2743	Version::parse("0.1.0").unwrap(),
2744	"linux/x86_64".parse().unwrap(),
2745	build_host.into(),
2746	"local".into(),
2747	"v0.1.0".into(),
2748	"~/Code/Apps/demo".into(),
2749	vec![],
2750	Kind::App,
2751	1,
2752	Arc::new(std::collections::HashMap::new()),
2753	Arc::new(std::collections::HashMap::new()),
2754	None,
2755	pool,
2756	crate::events::channel(),
2757	cfg,
2758	Arc::new(OtaRegistry::standard("https://makenot.work")),
2759	tokio::runtime::Handle::current(),
2760	Arc::new(AtomicBool::new(false)),
2761	None,
2762	)
2763	.with_repo_by_host(HashMap::from([(
2764	"windows-x86".to_string(),
2765	"C:/Users/me/Code/Apps/demo".to_string(),
2766	)])),
2767	);
2768	build_engine(&ctx).eval::<String>("repo()").unwrap()
2769	}
2770
2771	assert_eq!(repo_on("windows-x86").await, "C:/Users/me/Code/Apps/demo");
2772	assert_eq!(repo_on("fw13").await, "~/Code/Apps/demo");
2773	}
2774
2775	/// `secret(key)` reads a file under `secrets_root`, trims its trailing
2776	/// newline (the shape of a here-doc'd token file), and refuses any key that
2777	/// could escape the root. Covers the host-fn registered in `build_engine`.
2778	#[tokio::test]
2779	async fn secret_reads_under_root_and_blocks_traversal() {
2780	let dir = tempfile::tempdir().unwrap();
2781	let cfg = Config::for_tests(dir.path());
2782	// Seed a secret and one in a nested subdir; a trailing newline that the
2783	// read must strip.
2784	std::fs::create_dir_all(&cfg.secrets_root).unwrap();
2785	std::fs::write(cfg.secrets_root.join("token"), "s3cr3t\n").unwrap();
2786	std::fs::create_dir_all(cfg.secrets_root.join("app")).unwrap();
2787	std::fs::write(cfg.secrets_root.join("app").join("key"), "nested").unwrap();
2788	// Plant a file OUTSIDE the root that a traversal key would reach.
2789	std::fs::write(dir.path().join("outside"), "leak").unwrap();
2790
2791	let cfg = Arc::new(cfg);
2792	let pool = crate::db::open(&cfg.db_path).await.unwrap();
2793	let ctx = Arc::new(RecipeCtx::new(
2794	AppId::new("demo"),
2795	Version::parse("0.1.0").unwrap(),
2796	"linux/x86_64".parse().unwrap(),
2797	"fw13".into(),
2798	"local".into(),
2799	"v0.1.0".into(),
2800	"/tmp".into(),
2801	vec![],
2802	Kind::App,
2803	1,
2804	Arc::new(std::collections::HashMap::new()),
2805	Arc::new(std::collections::HashMap::new()),
2806	None,
2807	pool,
2808	crate::events::channel(),
2809	cfg,
2810	Arc::new(OtaRegistry::standard("https://makenot.work")),
2811	tokio::runtime::Handle::current(),
2812	Arc::new(AtomicBool::new(false)),
2813	None,
2814	));
2815	let engine = build_engine(&ctx);
2816
2817	// Happy path: read + trim.
2818	assert_eq!(
2819	engine.eval::<String>(r#"secret("token")"#).unwrap(),
2820	"s3cr3t"
2821	);
2822	// A multi-segment relative key is allowed.
2823	assert_eq!(
2824	engine.eval::<String>(r#"secret("app/key")"#).unwrap(),
2825	"nested"
2826	);
2827
2828	// Traversal, absolute paths, and empty keys are refused BEFORE any read,
2829	// so the file one `..` above the root is never disclosed.
2830	for bad in [
2831	r#"secret("../outside")"#,
2832	r#"secret("/etc/passwd")"#,
2833	r#"secret("")"#,
2834	] {
2835	let err = engine.eval::<String>(bad).unwrap_err().to_string();
2836	assert!(
2837	err.contains("relative path under secrets_root"),
2838	"`{bad}` should hit the traversal guard, got: {err}"
2839	);
2840	}
2841	// A missing key surfaces the filesystem error, not a panic, and does not
2842	// trip the traversal guard (it is a legitimate relative path).
2843	let err = engine
2844	.eval::<String>(r#"secret("nope")"#)
2845	.unwrap_err()
2846	.to_string();
2847	assert!(err.contains("secret `nope`"), "got: {err}");
2848	}
2849
2850	/// The two failures that actually shipped, as regression cases.
2851	#[test]
2852	fn preflight_catches_a_dead_repository_url() {
2853	// pter 0.1.0: repository pointed at a URL that does not exist. It
2854	// published clean and the link is now permanent for that version.
2855	let meta = CrateMeta {
2856	name: "pter".into(),
2857	version: "0.1.0".into(),
2858	repository: Some("https://github.com/maxjacobson/pter".into()),
2859	description: Some("d".into()),
2860	licensed: true,
2861	};
2862	let problems = crate_publish_problems(&meta, false, &[], true);
2863	assert_eq!(problems.len(), 1, "{problems:?}");
2864	assert!(
2865	problems[0].contains("not publicly clonable"),
2866	"{problems:?}"
2867	);
2868
2869	// Same metadata, reachable URL: nothing to report.
2870	assert!(crate_publish_problems(&meta, true, &[], true).is_empty());
2871	}
2872
2873	#[test]
2874	fn preflight_requires_the_fields_crates_io_bakes_in() {
2875	let bare = CrateMeta {
2876	name: "x".into(),
2877	version: "0.1.0".into(),
2878	repository: None,
2879	description: None,
2880	licensed: false,
2881	};
2882	let problems = crate_publish_problems(&bare, false, &[], true);
2883	assert_eq!(problems.len(), 3, "{problems:?}");
2884	assert!(problems.iter().any(\|p\| p.contains("repository")));
2885	assert!(problems.iter().any(\|p\| p.contains("description")));
2886	assert!(problems.iter().any(\|p\| p.contains("license")));
2887	}
2888
2889	// A library's verify is a crate preflight, not a Gatekeeper check on a
2890	// signed bundle. Gating it on `gatekeeper` asked a Linux host for a macOS
2891	// code-signing capability it can never hold, so the step was denied before
2892	// it ran a command; the denial then surfaced as "no crates.io credentials",
2893	// which is not what went wrong. The only way to satisfy the old gate was to
2894	// declare the capability falsely in the topology.
2895	#[test]
2896	fn a_library_verify_is_not_gated_on_gatekeeper() {
2897	assert_eq!(
2898	action_for(Step::Verify, Kind::Library),
2899	Action::Build,
2900	"a crate preflight runs the build toolchain; that is what it needs",
2901	);
2902	assert_eq!(
2903	action_for(Step::Verify, Kind::App),
2904	Action::Observe(ObserveKind::Custom("gatekeeper".into())),
2905	"an app's verify still proves the bundle is signed and notarized",
2906	);
2907	}
2908
2909	// The capability the default host grant actually carries. Without this the
2910	// fix above is only true by inspection.
2911	#[test]
2912	fn a_default_host_can_run_a_library_verify_and_not_an_app_one() {
2913	let caps =
2914	ops_exec::CapabilitySet::from_tokens(["build", "package"], ["build-log", "artifact"]);
2915	assert!(caps.permits(&action_for(Step::Verify, Kind::Library)));
2916	assert!(!caps.permits(&action_for(Step::Verify, Kind::App)));
2917	}
2918
2919	// Every other step is a property of the step alone; verify is the one that
2920	// depends on what is being released.
2921	#[test]
2922	fn no_other_step_changes_with_the_kind() {
2923	for step in [
2924	Step::Checkout,
2925	Step::Prebuild,
2926	Step::Build,
2927	Step::Sign,
2928	Step::Notarize,
2929	Step::Staple,
2930	Step::Package,
2931	Step::Publish,
2932	Step::Collect,
2933	] {
2934	assert_eq!(
2935	action_for(step, Kind::App),
2936	action_for(step, Kind::Library),
2937	"{step:?} should not depend on the kind",
2938	);
2939	}
2940	}
2941
2942	#[test]
2943	fn preflight_rejects_republishing_the_same_version() {
2944	let meta = CrateMeta {
2945	name: "makeover".into(),
2946	version: "0.10.0".into(),
2947	repository: Some("https://git.sr.ht/~maxmj/makeover".into()),
2948	description: Some("d".into()),
2949	licensed: true,
2950	};
2951	let problems =
2952	crate_publish_problems(&meta, true, &["0.9.0".into(), "0.10.0".into()], true);
2953	assert_eq!(problems.len(), 1, "{problems:?}");
2954	assert!(problems[0].contains("already published"), "{problems:?}");
2955
2956	// An unreleased version against the same history is fine.
2957	let mut next = meta.clone();
2958	next.version = "0.11.0".into();
2959	assert!(crate_publish_problems(&next, true, &["0.10.0".into()], true).is_empty());
2960	}
2961
2962	/// Missing credentials must surface at preflight, not at the upload. The
2963	/// publish step is the irreversible one and runs last, after a full build
2964	/// and verify; discovering there that cargo cannot authenticate wastes the
2965	/// whole run.
2966	#[test]
2967	fn preflight_reports_missing_credentials_up_front() {
2968	let meta = CrateMeta {
2969	name: "makeover".into(),
2970	version: "0.11.0".into(),
2971	repository: Some("https://git.sr.ht/~maxmj/makeover".into()),
2972	description: Some("d".into()),
2973	licensed: true,
2974	};
2975	// Metadata is perfect; only the token is absent.
2976	let problems = crate_publish_problems(&meta, true, &[], false);
2977	assert_eq!(problems.len(), 1, "{problems:?}");
2978	assert!(problems[0].contains("credentials"), "{problems:?}");
2979	assert!(
2980	problems[0].contains("cargo login"),
2981	"should say how to fix it"
2982	);
2983
2984	// Present: nothing to report.
2985	assert!(crate_publish_problems(&meta, true, &[], true).is_empty());
2986	}
2987
2988	#[test]
2989	fn crate_meta_reads_cargo_metadata_json() {
2990	let raw = r#"{"packages":[{"name":"makeover","version":"0.10.0",
2991	"repository":"https://git.sr.ht/~maxmj/makeover","description":"themes",
2992	"license":"MIT"}]}"#;
2993	let m = crate_meta_from_json(raw).unwrap();
2994	assert_eq!(m.name, "makeover");
2995	assert_eq!(m.version, "0.10.0");
2996	assert!(m.licensed);
2997	assert_eq!(
2998	m.repository.as_deref(),
2999	Some("https://git.sr.ht/~maxmj/makeover")
3000	);
3001
3002	// license_file alone also counts as licensed; empty strings do not
3003	// count as present.
3004	let lf = r#"{"packages":[{"name":"x","version":"0.1.0","license":"",
3005	"license_file":"LICENSE","description":""}]}"#;
3006	let m = crate_meta_from_json(lf).unwrap();
3007	assert!(m.licensed);
3008	assert!(m.description.is_none());
3009	}
3010
3011	/// Reads the ambient `HOME` rather than setting one. `set_var` is
3012	/// process-global and unsynchronized, so a test that overwrote HOME changed
3013	/// it for every other test in the binary — which is what silently disabled
3014	/// `topology::live_config_smoke` (it skips when `$HOME/.config/bento` is
3015	/// absent, and `/home/test` always is).
3016	#[test]
3017	fn expand_tilde_handles_home() {
3018	let home = PathBuf::from(std::env::var("HOME").expect("HOME is set"));
3019	assert_eq!(expand_tilde("~/Code/x"), home.join("Code/x"));
3020	assert_eq!(expand_tilde("/abs/path"), PathBuf::from("/abs/path"));
3021	}
3022
3023	// ---- artifact resolution (the M3 silent-`sh` fix) ----
3024
3025	#[test]
3026	fn resolve_artifact_match_wants_exactly_one() {
3027	// Exactly one match: the path, trimmed of the listing's line noise.
3028	assert_eq!(
3029	resolve_artifact_match(" /d/App.AppImage \n", "*.AppImage", true).unwrap(),
3030	"/d/App.AppImage"
3031	);
3032	}
3033
3034	#[test]
3035	fn resolve_artifact_match_zero_depends_on_required() {
3036	// Required + zero matches is the case the old empty-string guard caught;
3037	// keep failing it.
3038	let err = resolve_artifact_match("", "*.dmg", true).unwrap_err();
3039	assert!(err.to_string().contains("no artifact matched"), "{err}");
3040	// Optional + zero matches resolves to empty (recipe skips the collect).
3041	assert_eq!(
3042	resolve_artifact_match("\n \n", "*.deb", false).unwrap(),
3043	""
3044	);
3045	}
3046
3047	#[test]
3048	fn resolve_artifact_match_rejects_ambiguous() {
3049	// Two matches must throw rather than silently pick one — this is the
3050	// stale-newest-mtime hole the audit flagged. Applies even when optional.
3051	for required in [true, false] {
3052	let err =
3053	resolve_artifact_match("/d/old.deb\n/d/new.deb\n", "*.deb", required).unwrap_err();
3054	let msg = err.to_string();
3055	assert!(msg.contains("ambiguous"), "{msg}");
3056	assert!(msg.contains("old.deb") && msg.contains("new.deb"), "{msg}");
3057	}
3058	}
3059
3060	#[test]
3061	fn ensure_glob_safe_allows_paths_bars_commands() {
3062	// Path and wildcard characters pass.
3063	assert!(ensure_glob_safe("~/Code/app/dist/*.AppImage").is_ok());
3064	assert!(ensure_glob_safe("/t/App_1.2.3-x86_64.dmg").is_ok());
3065	// A command substitution or separator does not.
3066	for bad in ["*.dmg; rm -rf /", "$(evil)", "a\|b", "a b"] {
3067	assert!(ensure_glob_safe(bad).is_err(), "should reject {bad:?}");
3068	}
3069	}
3070
3071	// ---- version resolution ----
3072
3073	#[test]
3074	fn version_from_tauri_json_reads_version() {
3075	assert_eq!(
3076	version_from_tauri_json(r#"{"version":"0.4.2"}"#).unwrap(),
3077	"0.4.2"
3078	);
3079	assert!(version_from_tauri_json(r#"{"productName":"X"}"#).is_err());
3080	}
3081
3082	#[test]
3083	fn version_from_cargo_toml_prefers_package_then_workspace() {
3084	// A leaf crate's [package].version.
3085	assert_eq!(
3086	version_from_cargo_toml("[package]\nname = \"x\"\nversion = \"0.5.0\"\n").unwrap(),
3087	"0.5.0"
3088	);
3089	// A workspace that sets [workspace.package].version.
3090	assert_eq!(
3091	version_from_cargo_toml("[workspace.package]\nversion = \"1.2.3\"\n").unwrap(),
3092	"1.2.3"
3093	);
3094	// No version anywhere -> error, not a panic.
3095	assert!(version_from_cargo_toml("[workspace]\nmembers = []\n").is_err());
3096	}
3097
3098	#[test]
3099	fn version_from_repo_default_and_explicit_paths() {
3100	let tmp = tempfile::tempdir().unwrap();
3101	let root = tmp.path();
3102
3103	// Tauri app: default path reads src-tauri/tauri.conf.json.
3104	let tauri = root.join("tauri");
3105	std::fs::create_dir_all(tauri.join("src-tauri")).unwrap();
3106	std::fs::write(
3107	tauri.join("src-tauri/tauri.conf.json"),
3108	r#"{"version":"0.4.2"}"#,
3109	)
3110	.unwrap();
3111	assert_eq!(
3112	version_from_repo(tauri.to_str().unwrap(), None)
3113	.unwrap()
3114	.to_string(),
3115	"0.4.2"
3116	);
3117
3118	// Workspace egui app: no tauri.conf.json, explicit version_path at a member crate.
3119	let ws = root.join("ws");
3120	std::fs::create_dir_all(ws.join("crates/app")).unwrap();
3121	std::fs::write(
3122	ws.join("Cargo.toml"),
3123	"[workspace]\nmembers = [\"crates/app\"]\n",
3124	)
3125	.unwrap();
3126	std::fs::write(
3127	ws.join("crates/app/Cargo.toml"),
3128	"[package]\nname = \"app\"\nversion = \"0.5.0\"\n",
3129	)
3130	.unwrap();
3131	assert_eq!(
3132	version_from_repo(ws.to_str().unwrap(), Some("crates/app/Cargo.toml"))
3133	.unwrap()
3134	.to_string(),
3135	"0.5.0"
3136	);
3137	}
3138
3139	// ---- version-source cross-check (drift preflight) ----
3140
3141	fn ver(s: &str) -> Version {
3142	Version::parse(s).unwrap()
3143	}
3144
3145	#[test]
3146	fn version_consistency_passes_when_all_sources_agree() {
3147	let tmp = tempfile::tempdir().unwrap();
3148	let repo = tmp.path();
3149	std::fs::create_dir_all(repo.join("src-tauri")).unwrap();
3150	std::fs::write(
3151	repo.join("src-tauri/tauri.conf.json"),
3152	r#"{"version":"0.5.0"}"#,
3153	)
3154	.unwrap();
3155	std::fs::write(
3156	repo.join("Cargo.toml"),
3157	"[package]\nname = \"app\"\nversion = \"0.5.0\"\n",
3158	)
3159	.unwrap();
3160	check_version_consistency(repo.to_str().unwrap(), None, &ver("0.5.0")).unwrap();
3161	}
3162
3163	#[test]
3164	fn version_consistency_flags_tauri_vs_cargo_drift() {
3165	// The concrete finding: tauri.conf.json bumped to 0.5.0 but the root
3166	// Cargo.toml left at 0.4.0. version_from_repo (one file) would miss it.
3167	let tmp = tempfile::tempdir().unwrap();
3168	let repo = tmp.path();
3169	std::fs::create_dir_all(repo.join("src-tauri")).unwrap();
3170	std::fs::write(
3171	repo.join("src-tauri/tauri.conf.json"),
3172	r#"{"version":"0.5.0"}"#,
3173	)
3174	.unwrap();
3175	std::fs::write(
3176	repo.join("Cargo.toml"),
3177	"[package]\nname = \"app\"\nversion = \"0.4.0\"\n",
3178	)
3179	.unwrap();
3180	let err =
3181	check_version_consistency(repo.to_str().unwrap(), None, &ver("0.5.0")).unwrap_err();
3182	let msg = format!("{err:#}");
3183	assert!(msg.contains("Cargo.toml says 0.4.0"), "{msg}");
3184	}
3185
3186	#[test]
3187	fn version_consistency_flags_explicit_version_the_repo_does_not_reflect() {
3188	let tmp = tempfile::tempdir().unwrap();
3189	let repo = tmp.path();
3190	std::fs::create_dir_all(repo.join("src-tauri")).unwrap();
3191	std::fs::write(
3192	repo.join("src-tauri/tauri.conf.json"),
3193	r#"{"version":"0.5.0"}"#,
3194	)
3195	.unwrap();
3196	let err =
3197	check_version_consistency(repo.to_str().unwrap(), None, &ver("9.9.9")).unwrap_err();
3198	assert!(format!("{err:#}").contains("building 9.9.9"));
3199	}
3200
3201	#[test]
3202	fn version_consistency_single_source_never_invents_drift() {
3203	// A virtual-workspace root Cargo.toml (no version) alongside the member
3204	// crate the version_path points at: only one real source, so no drift.
3205	let tmp = tempfile::tempdir().unwrap();
3206	let repo = tmp.path();
3207	std::fs::create_dir_all(repo.join("crates/app")).unwrap();
3208	std::fs::write(
3209	repo.join("Cargo.toml"),
3210	"[workspace]\nmembers = [\"crates/app\"]\n",
3211	)
3212	.unwrap();
3213	std::fs::write(
3214	repo.join("crates/app/Cargo.toml"),
3215	"[package]\nname = \"app\"\nversion = \"0.5.0\"\n",
3216	)
3217	.unwrap();
3218	check_version_consistency(
3219	repo.to_str().unwrap(),
3220	Some("crates/app/Cargo.toml"),
3221	&ver("0.5.0"),
3222	)
3223	.unwrap();
3224	}
3225
3226	// ---- artifact filename version assertion + hashing ----
3227
3228	#[test]
3229	fn versions_in_filename_extracts_only_real_semvers() {
3230	assert_eq!(
3231	versions_in_filename("GoingsOn_0.5.0_aarch64.dmg"),
3232	vec![ver("0.5.0")]
3233	);
3234	assert_eq!(
3235	versions_in_filename("AudioFiles-0.4.0-x86_64.AppImage"),
3236	vec![ver("0.4.0")]
3237	);
3238	// No three-part token ⇒ nothing (an updater manifest, a bare signature).
3239	assert!(versions_in_filename("latest.json").is_empty());
3240	assert!(versions_in_filename("app.sig").is_empty());
3241	}
3242
3243	#[test]
3244	fn assert_artifact_version_rejects_a_stale_artifact() {
3245	// The 0.4.0 file sitting in the output dir against a 0.5.0 build.
3246	let err =
3247	assert_artifact_version("AudioFiles-0.4.0-x86_64.AppImage", &ver("0.5.0")).unwrap_err();
3248	assert!(format!("{err:#}").contains("stale artifact"), "{err:#}");
3249	// The matching version passes, and a versionless file is not asserted.
3250	assert_artifact_version("GoingsOn_0.5.0_aarch64.dmg", &ver("0.5.0")).unwrap();
3251	assert_artifact_version("latest.json", &ver("0.5.0")).unwrap();
3252	}
3253
3254	/// The comparison that decides whether a binary can exec on the box that is
3255	/// about to be restarted onto it. Both sides are parsed out of text a tool
3256	/// printed, so both parsers are worth pinning: fw13 tracks a newer glibc
3257	/// than the Ubuntu 24.04 host in Hetzner, and getting this backwards means a
3258	/// dead unit rather than a failed step.
3259	#[test]
3260	fn glibc_versions_parse_from_what_the_tools_actually_print() {
3261	// `objdump -T \| grep -o 'GLIBC_[0-9.]*'` output: highest wins, and the
3262	// comparison is numeric (2.9 must not beat 2.34 lexically).
3263	let objdump = "GLIBC_2.2.5\nGLIBC_2.34\nGLIBC_2.9\nGLIBC_2.17\n";
3264	assert_eq!(max_glibc_symbol(objdump), Some((2, 34)));
3265	// A static binary references none: nothing to check.
3266	assert_eq!(max_glibc_symbol(""), None);
3267
3268	// `ldd --version` first line, however the distro decorates it.
3269	assert_eq!(
3270	glibc_from_ldd("ldd (Ubuntu GLIBC 2.39-0ubuntu8.8) 2.39\nCopyright...\n"),
3271	Some((2, 39))
3272	);
3273	assert_eq!(
3274	glibc_from_ldd("ldd (GNU libc) 2.41\nCopyright (C) 2025\n"),
3275	Some((2, 41))
3276	);
3277	assert_eq!(glibc_from_ldd(""), None);
3278	}
3279
3280	/// A binary needing MORE than the host has is the failure this check exists
3281	/// for; equal and less are both fine (glibc symbol versioning is backward
3282	/// compatible, so an older requirement runs on a newer host).
3283	#[test]
3284	fn glibc_requirement_is_satisfied_by_equal_or_newer_only() {
3285	let needs = max_glibc_symbol("GLIBC_2.41").unwrap();
3286	assert!(needs > glibc_from_ldd("ldd (Ubuntu GLIBC 2.39) 2.39").unwrap());
3287	assert!(needs <= glibc_from_ldd("ldd (GNU libc) 2.41").unwrap());
3288	assert!(needs <= glibc_from_ldd("ldd (GNU libc) 2.42").unwrap());
3289	assert!(needs <= glibc_from_ldd("ldd (GNU libc) 3.0").unwrap());
3290	}
3291
3292	/// Every deploy host function fails with the app's KIND as the reason when
3293	/// there is no destination, rather than with a missing-host error from
3294	/// somewhere deeper. A recipe calling `deploy()` on a library is a recipe
3295	/// written against the wrong kind, and the message should say so.
3296	#[tokio::test]
3297	async fn deploy_host_fns_explain_a_missing_destination_by_kind() {
3298	let dir = tempfile::tempdir().unwrap();
3299	let cfg = Arc::new(Config::for_tests(dir.path()));
3300	let pool = crate::db::open(&cfg.db_path).await.unwrap();
3301	let ctx = Arc::new(RecipeCtx::new(
3302	AppId::new("demo"),
3303	Version::parse("0.1.0").unwrap(),
3304	"linux/x86_64".parse().unwrap(),
3305	"fw13".into(),
3306	"local".into(),
3307	"v0.1.0".into(),
3308	"/tmp".into(),
3309	vec![],
3310	Kind::Library,
3311	1,
3312	Arc::new(std::collections::HashMap::new()),
3313	Arc::new(std::collections::HashMap::new()),
3314	None,
3315	pool,
3316	crate::events::channel(),
3317	cfg,
3318	Arc::new(OtaRegistry::standard("https://makenot.work")),
3319	tokio::runtime::Handle::current(),
3320	Arc::new(AtomicBool::new(false)),
3321	None,
3322	));
3323	let engine = build_engine(&ctx);
3324	for call in [
3325	"deploy_host()",
3326	"service_name()",
3327	"install_path()",
3328	"health_url()",
3329	r#"deploy("/tmp/x")"#,
3330	] {
3331	let err = engine.eval::<String>(call).unwrap_err().to_string();
3332	assert!(
3333	err.contains("library") && err.contains("no deploy destination"),
3334	"`{call}` must fail on the kind, got: {err}"
3335	);
3336	}
3337	}
3338
3339	/// A service host is addressed on the DEPLOY plane whatever step is open.
3340	///
3341	/// The subtle one. Actions are normally derived from the step, which is
3342	/// right for a build host — the step is what that host is being asked to do.
3343	/// A service host is granted `deploy`/`restart` and must never be granted
3344	/// `build`, so the same rule would have `glibc_check` ask it for `build`
3345	/// during a `verify` step and get denied for a reason unrelated to what was
3346	/// attempted. `verify` is the step that check belongs in, so without this
3347	/// routing the glibc gate cannot run at all.
3348	#[tokio::test]
3349	async fn a_service_host_is_addressed_on_the_deploy_plane_in_any_step() {
3350	let dir = tempfile::tempdir().unwrap();
3351	let cfg = Arc::new(Config::for_tests(dir.path()));
3352	let pool = crate::db::open(&cfg.db_path).await.unwrap();
3353	sqlx::query(
3354	"INSERT INTO builds (id, app, version, status, created_at) \
3355	VALUES (1, 'demo', '0.1.0', 'running', '2026-07-30T00:00:00Z')",
3356	)
3357	.execute(&pool)
3358	.await
3359	.unwrap();
3360	sqlx::query(
3361	"INSERT INTO target_runs (id, build_id, app, version, target, status, started_at) \
3362	VALUES (1, 1, 'demo', '0.1.0', 'linux/x86_64', 'running', '2026-07-30T00:00:00Z')",
3363	)
3364	.execute(&pool)
3365	.await
3366	.unwrap();
3367
3368	let deploy = crate::topology::DeployTarget {
3369	target: "linux/x86_64".parse().unwrap(),
3370	host: "local".into(),
3371	port: None,
3372	install_path: "/usr/local/bin/demo".into(),
3373	service: "demo.service".into(),
3374	health_url: None,
3375	};
3376	let mut execs: crate::state::ExecutorMap = std::collections::HashMap::new();
3377	execs.insert("local".into(), crate::state::build_deploy_executor(&deploy));
3378	// The service host's grant is exactly deploy + restart. If this ever
3379	// widens to include `build`, the test below stops proving anything.
3380	assert!(!execs["local"].capabilities().permits(&Action::Build));
3381	assert!(execs["local"].capabilities().permits(&Action::Deploy));
3382
3383	let ctx = Arc::new(RecipeCtx::new(
3384	AppId::new("demo"),
3385	Version::parse("0.1.0").unwrap(),
3386	"linux/x86_64".parse().unwrap(),
3387	"fw13".into(),
3388	"local".into(),
3389	"v0.1.0".into(),
3390	"/tmp".into(),
3391	vec![],
3392	Kind::Service,
3393	1,
3394	Arc::new(execs),
3395	Arc::new(std::collections::HashMap::new()),
3396	Some(deploy),
3397	pool,
3398	crate::events::channel(),
3399	cfg,
3400	Arc::new(OtaRegistry::standard("https://makenot.work")),
3401	tokio::runtime::Handle::current(),
3402	Arc::new(AtomicBool::new(false)),
3403	None,
3404	));
3405
3406	let ctx_blocking = ctx.clone();
3407	tokio::task::spawn_blocking(move \|\| {
3408	// `verify` on a service derives Action::Build — which the service
3409	// host does not grant. The command must still run.
3410	ctx_blocking.begin_step(Step::Verify).unwrap();
3411	assert_eq!(
3412	action_for(Step::Verify, Kind::Service),
3413	Action::Build,
3414	"the step's own action is the one that would be denied",
3415	);
3416	let (code, out) = ctx_blocking
3417	.run("local", "echo reached-the-service-host")
3418	.expect("a service host must be reachable during a verify step");
3419	assert_eq!(code, 0, "{out}");
3420	assert!(out.contains("reached-the-service-host"), "{out}");
3421	})
3422	.await
3423	.unwrap();
3424	}
3425
3426	/// A step that finalized `Failed` bars the deploy, exactly as it bars a
3427	/// publish. Without this, a recipe that inspects `sh(...).code` and carries
3428	/// on regardless still lands a binary on a production host — the precise
3429	/// hazard a pipeline exists to remove. The check is the ledger, not the
3430	/// control flow, so it holds whether or not the recipe noticed.
3431	#[tokio::test]
3432	async fn a_failed_step_bars_the_deploy() {
3433	let dir = tempfile::tempdir().unwrap();
3434	let cfg = Arc::new(Config::for_tests(dir.path()));
3435	let pool = crate::db::open(&cfg.db_path).await.unwrap();
3436	let deploy = crate::topology::DeployTarget {
3437	target: "linux/x86_64".parse().unwrap(),
3438	host: "local".into(),
3439	port: None,
3440	install_path: "/usr/local/bin/demo".into(),
3441	service: "demo.service".into(),
3442	health_url: None,
3443	};
3444	// A real build + target run, so the step rows this test finalizes have
3445	// the parents the schema requires.
3446	sqlx::query(
3447	"INSERT INTO builds (id, app, version, status, created_at) \
3448	VALUES (1, 'demo', '0.1.0', 'running', '2026-07-30T00:00:00Z')",
3449	)
3450	.execute(&pool)
3451	.await
3452	.unwrap();
3453	sqlx::query(
3454	"INSERT INTO target_runs (id, build_id, app, version, target, status, started_at) \
3455	VALUES (1, 1, 'demo', '0.1.0', 'linux/x86_64', 'running', '2026-07-30T00:00:00Z')",
3456	)
3457	.execute(&pool)
3458	.await
3459	.unwrap();
3460
3461	let mut execs: crate::state::ExecutorMap = std::collections::HashMap::new();
3462	execs.insert("local".into(), crate::state::build_deploy_executor(&deploy));
3463	let ctx = Arc::new(RecipeCtx::new(
3464	AppId::new("demo"),
3465	Version::parse("0.1.0").unwrap(),
3466	"linux/x86_64".parse().unwrap(),
3467	"fw13".into(),
3468	"local".into(),
3469	"v0.1.0".into(),
3470	"/tmp".into(),
3471	vec![],
3472	Kind::Service,
3473	1,
3474	Arc::new(execs),
3475	Arc::new(std::collections::HashMap::new()),
3476	Some(deploy),
3477	pool,
3478	crate::events::channel(),
3479	cfg,
3480	Arc::new(OtaRegistry::standard("https://makenot.work")),
3481	tokio::runtime::Handle::current(),
3482	Arc::new(AtomicBool::new(false)),
3483	None,
3484	));
3485
3486	// A gate ran, failed, and the recipe did not abort — the swallowed
3487	// failure. Finalizing it is what puts it in the ledger.
3488	let ctx_blocking = ctx.clone();
3489	tokio::task::spawn_blocking(move \|\| {
3490	ctx_blocking.begin_step(Step::Prebuild).unwrap();
3491	ctx_blocking.fail_current_step();
3492	ctx_blocking.finish_step(Status::Ok).unwrap();
3493
3494	let err = ctx_blocking.deploy("/tmp/demo").unwrap_err().to_string();
3495	assert!(
3496	err.contains("refusing to deploy") && err.contains("prebuild"),
3497	"must refuse and name the failed step, got: {err}"
3498	);
3499	})
3500	.await
3501	.unwrap();
3502	}
3503
3504	#[test]
3505	fn every_step_has_a_nonzero_default_budget() {
3506	// A zero/missing budget would deadline-fail a step instantly. Cover the
3507	// whole matrix so a new Step variant can't silently get a 0 budget.
3508	for step in Step::ALL {
3509	assert!(
3510	default_step_budget(step) >= std::time::Duration::from_mins(1),
3511	"{step} budget must be a sane ceiling",
3512	);
3513	}
3514	}
3515
3516	#[test]
3517	fn sha256_file_is_lowercase_hex_of_contents() {
3518	let tmp = tempfile::tempdir().unwrap();
3519	let f = tmp.path().join("a.bin");
3520	std::fs::write(&f, b"abc").unwrap();
3521	// Known SHA-256 of "abc".
3522	assert_eq!(
3523	sha256_file(&f).unwrap(),
3524	"ba7816bf8f01cfea414140de5dae2223b00361a396177a9cb410ff61f20015ad"
3525	);
3526	}
3527
3528	// ---- publish step-success gate ----
3529
3530	fn target(s: &str) -> Target {
3531	s.parse().unwrap()
3532	}
3533
3534	#[test]
3535	fn publish_gate_blocks_macos_without_verification() {
3536	// Never verified -> blocked, with a message pointing at verify_gatekeeper.
3537	let err = PublishAuthority::prove(target("macos/aarch64"), &[], None).unwrap_err();
3538	assert!(format!("{err:#}").contains("never verified"), "{err:#}");
3539	}
3540
3541	#[test]
3542	fn publish_gate_blocks_macos_when_gatekeeper_rejected() {
3543	let err = PublishAuthority::prove(target("macos/aarch64"), &[], Some(false)).unwrap_err();
3544	assert!(
3545	format!("{err:#}").contains("Gatekeeper rejected"),
3546	"{err:#}"
3547	);
3548	}
3549
3550	#[test]
3551	fn publish_gate_allows_macos_when_gatekeeper_accepted() {
3552	PublishAuthority::prove(target("macos/aarch64"), &[], Some(true)).unwrap();
3553	// iOS is gated the same way.
3554	PublishAuthority::prove(target("ios/universal"), &[], Some(true)).unwrap();
3555	assert!(PublishAuthority::prove(target("ios/universal"), &[], None).is_err());
3556	}
3557
3558	#[test]
3559	fn publish_gate_does_not_require_gatekeeper_for_non_apple_targets() {
3560	// Linux/Windows aren't notarized; no gatekeeper proof needed.
3561	PublishAuthority::prove(target("linux/x86_64"), &[], None).unwrap();
3562	PublishAuthority::prove(target("windows/x86_64"), &[], None).unwrap();
3563	}
3564
3565	#[test]
3566	fn publish_gate_blocks_when_any_prior_step_failed() {
3567	// A failed step bars publish on every target, even a verified macOS one.
3568	let err =
3569	PublishAuthority::prove(target("linux/x86_64"), &[Step::Build], None).unwrap_err();
3570	assert!(
3571	format!("{err:#}").contains("prior step(s) failed"),
3572	"{err:#}"
3573	);
3574	assert!(
3575	format!("{err:#}").contains("build"),
3576	"names the failed step: {err:#}"
3577	);
3578
3579	let err = PublishAuthority::prove(target("macos/aarch64"), &[Step::Sign], Some(true))
3580	.unwrap_err();
3581	assert!(
3582	format!("{err:#}").contains("prior step(s) failed"),
3583	"{err:#}"
3584	);
3585	}
3586
3587	#[test]
3588	fn notary_accepted_parses_status_field() {
3589	assert!(notary_accepted(
3590	r#"{"id":"abc","status":"Accepted","message":"ok"}"#
3591	));
3592	// Embedded in shell-sourcing noise: the object is isolated and parsed.
3593	assert!(notary_accepted(
3594	"sourcing env...\n{\n \"status\": \"Accepted\"\n}\nbye"
3595	));
3596	// Whitespace variant that a tight substring `"status":"Accepted"` misses.
3597	assert!(notary_accepted(r#"{ "status" : "Accepted" }"#));
3598	}
3599
3600	#[test]
3601	fn notary_accepted_rejects_non_accepted_and_garbage() {
3602	assert!(!notary_accepted(r#"{"status":"Invalid"}"#));
3603	assert!(!notary_accepted(r#"{"status":"In Progress"}"#));
3604	assert!(!notary_accepted("no json here"));
3605	assert!(!notary_accepted("")); // empty / truncated -> fail closed
3606	// A truncated tail whose opening brace was cut off cannot parse -> closed.
3607	assert!(!notary_accepted(r#""status":"Accepted"}"#));
3608	// The literal appearing inside an error string must NOT pass as success.
3609	assert!(!notary_accepted(
3610	r#"{"status":"Invalid","message":"expected status:Accepted"}"#
3611	));
3612	}
3613
3614	/// Both shapes of repo: one holding several products, and one holding a
3615	/// single crate, where git prints an empty prefix line.
3616	#[test]
3617	fn toplevel_and_prefix_read_both_shapes_of_repo() {
3618	let (top, prefix) =
3619	parse_toplevel_and_prefix("/home/max/Code/MNW\npom/\n").expect("two lines");
3620	assert_eq!(top, "/home/max/Code/MNW");
3621	assert_eq!(prefix, "pom/");
3622	// A repo holding one product: git prints an empty second line.
3623	let (top, prefix) =
3624	parse_toplevel_and_prefix("/home/max/Code/Libraries/pter\n\n").expect("two lines");
3625	assert_eq!(top, "/home/max/Code/Libraries/pter");
3626	assert_eq!(prefix, "");
3627	assert!(
3628	parse_toplevel_and_prefix("").is_none(),
3629	"no answer is not an answer"
3630	);
3631	}
3632
3633	#[test]
3634	fn repo_dir_name_is_the_last_segment_on_every_platform() {
3635	assert_eq!(repo_dir_name("/home/max/Code/MNW"), "MNW");
3636	assert_eq!(repo_dir_name("/home/max/Code/MNW/"), "MNW");
3637	// Git reports forward slashes on Windows too.
3638	assert_eq!(repo_dir_name("C:/Users/me/Code/Apps/goingson"), "goingson");
3639	}
3640
3641	/// The app's directory inside its worktree, for both repo shapes.
3642	#[test]
3643	fn app_dir_in_worktree_follows_the_prefix() {
3644	assert_eq!(
3645	app_dir_in_worktree("/home/max/Code/.bento/MNW/pom", "pom/"),
3646	"/home/max/Code/.bento/MNW/pom/pom"
3647	);
3648	assert_eq!(
3649	app_dir_in_worktree("/home/max/Code/.bento/pter/pter", ""),
3650	"/home/max/Code/.bento/pter/pter"
3651	);
3652	}
3653
3654	/// A missing tag is the common failure and gets a plain answer; anything
3655	/// else is git's own stderr, which says more than a guess.
3656	#[test]
3657	fn worktree_failure_reason_names_the_tag_or_repeats_git() {
3658	let missing = worktree_failure_reason("pom-v0.4.5", false, "irrelevant");
3659	assert!(missing.contains("does not exist"), "{missing}");
3660	let held = worktree_failure_reason(
3661	"pom-v0.4.5",
3662	true,
3663	"fatal: '/home/max/Code/.bento/MNW/pom' already exists",
3664	);
3665	assert!(held.contains("already exists"), "{held}");
3666	let silent = worktree_failure_reason("pom-v0.4.5", true, " ");
3667	assert!(silent.contains("said nothing"), "{silent}");
3668	}
3669
3670	/// Run the probe for real rather than asserting on its text: it is shell,
3671	/// and the thing worth knowing is whether `sh` agrees, not whether the
3672	/// string looks right.
3673	fn probe(repo: &std::path::Path) -> bool {
3674	std::process::Command::new("sh")
3675	.arg("-c")
3676	.arg(tracked_lock_under_patch_cmd(&repo.display().to_string()))
3677	.status()
3678	.unwrap()
3679	.success()
3680	}
3681
3682	/// A crate under a `[patch]` root, with and without its lock tracked.
3683	///
3684	/// pter 0.2.1 is the case: the tracked half was true, the patch half was
3685	/// true, and cargo reported only "Cargo.lock" (build 325). Untracking the
3686	/// lock in `a9969a9` is what fixed it, and this asserts the probe agrees
3687	/// with that fix in both directions.
3688	#[test]
3689	fn a_tracked_lock_is_only_a_problem_under_a_patch_block() {
3690	let root = tempfile::tempdir().unwrap();
3691	let repo = root.path().join("crate");
3692	std::fs::create_dir_all(&repo).unwrap();
3693	let git = \|args: &[&str]\| {
3694	std::process::Command::new("git")
3695	.args(args)
3696	.current_dir(&repo)
3697	.env("GIT_AUTHOR_NAME", "t")
3698	.env("GIT_AUTHOR_EMAIL", "t@t")
3699	.env("GIT_COMMITTER_NAME", "t")
3700	.env("GIT_COMMITTER_EMAIL", "t@t")
3701	.output()
3702	.unwrap()
3703	};
3704	git(&["init", "-q", "."]);
3705	std::fs::write(repo.join("Cargo.lock"), "# lock\n").unwrap();
3706
3707	// Lock present but untracked, no patch anywhere: nothing to say.
3708	assert!(!probe(&repo));
3709
3710	// Tracked, still no patch root. Every library that commits a lock and
3711	// builds outside `~/Code` lives here, and it publishes fine.
3712	git(&["add", "Cargo.lock"]);
3713	git(&["commit", "-qm", "lock"]);
3714	assert!(!probe(&repo));
3715
3716	// The ancestor declares `[patch]`. Both halves now hold.
3717	std::fs::create_dir_all(root.path().join(".cargo")).unwrap();
3718	std::fs::write(
3719	root.path().join(".cargo/config.toml"),
3720	"[patch.\"https://makenot.work/git/max/docengine.git\"]\ndocengine = { path = \"x\" }\n",
3721	)
3722	.unwrap();
3723	assert!(probe(&repo));
3724
3725	// Untracking the lock is the fix, and the probe has to agree that it is.
3726	git(&["rm", "-q", "--cached", "Cargo.lock"]);
3727	assert!(!probe(&repo));
3728	}
3729
3730	/// The message exists because cargo's names neither fact and both wrong
3731	/// fixes are one flag away. Assert it still says all four things.
3732	#[test]
3733	fn the_tracked_lock_message_names_both_facts_and_refuses_both_wrong_fixes() {
3734	let msg = tracked_lock_under_patch_problem("~/Code/.bento/pter/pter");
3735	assert!(msg.contains("Cargo.lock"), "{msg}");
3736	assert!(msg.contains("[patch]"), "{msg}");
3737	assert!(msg.contains("git rm --cached"), "{msg}");
3738	assert!(msg.contains("--allow-dirty"), "{msg}");
3739	assert!(msg.contains("~/Code/.bento/pter/pter"), "{msg}");
3740	}
3741	}
3742