crates/flow-core/src/pipeline.rs at push-uses-login · andymccall/oxygenflow

◆ push-uses-login oxygenflow / crates / flow-core / src / pipeline.rs
49.0 KiB · Rust · last changed Report outputs: a directory with report.toml, uploaded file by file with a manifest 25 days ago source blame history raw
//! `.flow/pipeline.kdl`: the M0 subset of spec §3 (docs/m0.md §1), parsed
//! from KDL 2.0 and validated before anything runs. Every problem is
//! reported with its line, and all of them at once.

use serde::{Deserialize, Serialize};
use std::collections::{BTreeMap, BTreeSet};
use std::fmt;

/// Where a pipeline lives in the state being built.
pub const PIPELINE_PATH: &str = ".flow/pipeline.kdl";

#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct Pipeline {
    pub schema_version: u32,
    /// By name.
    pub sandboxes: BTreeMap<String, Sandbox>,
    /// By name.
    pub jobs: BTreeMap<String, Job>,
    /// When the pipeline's jobs run, unless a job says (oxygenflow#24);
    /// absent: on every push and from "Run pipeline".
    #[serde(default, skip_serializing_if = "Option::is_none")]
    pub on: Option<On>,
}

/// What starts a job (`on { push "main"; tag "v*"; manual; schedule "0 3 * * *" }`).
#[derive(Clone, Debug, Default, PartialEq, Eq, Serialize, Deserialize)]
pub struct On {
    /// Bookmark globs a push runs it for; `Some(vec![])`: any bookmark.
    #[serde(default, skip_serializing_if = "Option::is_none")]
    pub push: Option<Vec<String>>,
    /// Tag globs a pushed tag runs it for; `Some(vec![])`: any tag.
    #[serde(default, skip_serializing_if = "Option::is_none")]
    pub tag: Option<Vec<String>>,
    /// From "Run pipeline" (and `oh run start`).
    #[serde(default, skip_serializing_if = "std::ops::Not::not")]
    pub manual: bool,
    /// Five-field cron lines, on the hub's clock, for the default bookmark.
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    pub schedule: Vec<String>,
}

/// Why a run started, for [`On::matches`].
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub enum Event<'a> {
    /// A push moved this bookmark.
    Push(&'a str),
    /// This tag was pushed.
    Tag(&'a str),
    /// "Run pipeline", `oh run start`.
    Manual,
    /// A rerun of a scheduled run: any job with a schedule.
    Schedule,
    /// The hub's clock at a minute (UTC): the jobs with a line due then.
    ScheduleAt(When),
}

/// A minute of the hub's clock, UTC, as cron reads it.
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub struct When {
    pub minute: u32,
    pub hour: u32,
    pub day: u32,
    pub month: u32,
    /// 0 is Sunday.
    pub weekday: u32,
}

impl On {
    /// What a job without any `on` runs for: every push, and by hand.
    pub fn default_on() -> On {
        On { push: Some(vec![]), tag: None, manual: true, schedule: vec![] }
    }

    pub fn matches(&self, event: Event<'_>) -> bool {
        let any = |globs: &Option<Vec<String>>, name: &str| match globs {
            None => false,
            Some(g) if g.is_empty() => true,
            Some(g) => g.iter().any(|p| globset::Glob::new(p).is_ok_and(|p| p.compile_matcher().is_match(name))),
        };
        match event {
            Event::Push(b) => any(&self.push, b),
            Event::Tag(t) => any(&self.tag, t),
            Event::Manual => self.manual,
            Event::Schedule => !self.schedule.is_empty(),
            Event::ScheduleAt(w) => self
                .schedule
                .iter()
                .filter_map(|l| crate::cron::Cron::parse(l))
                .any(|c| c.matches(w.minute, w.hour, w.day, w.month, w.weekday)),
        }
    }
}

impl Pipeline {
    /// When `job` runs: its own `on`, else the pipeline's, else the default.
    pub fn on_of(&self, job: &Job) -> On {
        job.on.clone().or_else(|| self.on.clone()).unwrap_or_else(On::default_on)
    }
}

#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum Runtime {
    /// Tier 0 (spec §4.1); refused in M0.
    Wasm,
    /// Tier 1 (spec §4.2): namespaces, Landlock, seccomp, cgroups.
    Syscall,
    /// Tier 2 (spec §4.3), the default.
    Microvm,
}

impl Runtime {
    pub fn as_str(self) -> &'static str {
        match self {
            Runtime::Wasm => "wasm",
            Runtime::Syscall => "syscall",
            Runtime::Microvm => "microvm",
        }
    }
}

#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum Network {
    Isolated,
    Loopback,
    Egress,
}

impl Network {
    pub fn as_str(self) -> &'static str {
        match self {
            Network::Isolated => "isolated",
            Network::Loopback => "loopback",
            Network::Egress => "egress",
        }
    }
}

#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct Sandbox {
    pub name: String,
    pub runtime: Runtime,
    /// BLAKE3 of the rootfs image, hex.
    pub rootfs: String,
    pub memory_mib: u32,
    pub cpus: u32,
    pub network: Network,
    /// Environment every job in this sandbox starts from; a job's own `env`
    /// overrides it.
    #[serde(default, skip_serializing_if = "BTreeMap::is_empty")]
    pub env: BTreeMap<String, String>,
    /// Labels a runner must have to run jobs here (`runs-on`), sorted.
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    pub runs_on: Vec<String>,
    /// How long each job here may run, in seconds (`timeout "2h"`), when
    /// the job does not say; the hub caps it at the site's maximum.
    #[serde(default, skip_serializing_if = "Option::is_none")]
    pub timeout_s: Option<u64>,
    /// The architecture the jobs need (`arch "riscv64"`, spec §11.3);
    /// without it, the rootfs image's.
    #[serde(default, skip_serializing_if = "Option::is_none")]
    pub arch: Option<String>,
}

/// An architecture name as runners report it: `x86_64`, `aarch64`, `riscv64`.
pub fn arch_ok(a: &str) -> bool {
    (2..=16).contains(&a.len()) && a.bytes().all(|b| b.is_ascii_lowercase() || b.is_ascii_digit() || b == b'_')
}

#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct Job {
    pub name: String,
    pub sandbox: String,
    /// Glob patterns over repository paths (spec §6.2).
    pub scope: Vec<String>,
    /// Sorted.
    pub depends_on: Vec<String>,
    pub env: BTreeMap<String, String>,
    /// Executed as is: no shell, no expansion (spec §2.2 I2).
    pub exec: Vec<String>,
    /// Ephemeral scratch mounts, sorted by target (spec §3.1 `mount`).
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    pub mounts: Vec<Mount>,
    /// Files the job produces, sorted by name (spec §5.2 `output`).
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    pub outputs: Vec<Output>,
    /// Secrets the job mounts, sorted by name (spec §9 `secret`): the hub
    /// supplies each by name, the runner mounts it as a file.
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    pub secrets: Vec<Secret>,
    /// Identities the job asks the hub for (oxygenhub#103): an OIDC token
    /// for an audience, mounted as a file like a secret.
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    pub identities: Vec<Identity>,
    /// Labels a runner must have to take this job (`runs-on`), sorted;
    /// with its sandbox's, what the hub matches against a runner's.
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    pub runs_on: Vec<String>,
    /// How long the job may run, in seconds (`timeout "90m"`); its
    /// sandbox's when absent, else the site's default (oxygenflow#23).
    #[serde(default, skip_serializing_if = "Option::is_none")]
    pub timeout_s: Option<u64>,
    /// When it runs, over the pipeline's `on` (oxygenflow#24).
    #[serde(default, skip_serializing_if = "Option::is_none")]
    pub on: Option<On>,
}

impl Job {
    /// The job's environment on top of its sandbox's: what it runs with.
    pub fn in_sandbox(&self, sandbox: &Sandbox) -> Job {
        let mut job = self.clone();
        for (k, v) in &sandbox.env {
            job.env.entry(k.clone()).or_insert_with(|| v.clone());
        }
        job.runs_on.extend(sandbox.runs_on.iter().cloned());
        job.runs_on.sort();
        job.runs_on.dedup();
        job.timeout_s = job.timeout_s.or(sandbox.timeout_s);
        job
    }

    /// The job with variables that tell it what it builds (`FLOW_STATE`,
    /// `FLOW_BOOKMARK` …) added to its environment; a variable the pipeline
    /// sets itself keeps the pipeline's value.
    pub fn with_flow_env(&self, vars: &[(String, String)]) -> Job {
        let mut job = self.clone();
        for (k, v) in vars {
            job.env.entry(k.clone()).or_insert_with(|| v.clone());
        }
        job
    }
}

/// An ephemeral scratch mount: empty at the start of every job and gone
/// after it, so it cannot carry anything between jobs (spec §2.2 I4).
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct Mount {
    pub source: MountSource,
    /// An absolute path under `/src` or `/tmp`.
    pub target: String,
    pub size_mib: u32,
}

#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum MountSource {
    /// `ephemeral://tmpfs`: memory in the sandbox, counted against its memory-limit.
    Tmpfs,
    /// `ephemeral://disk`: a fresh sparse disk on the runner, for large build trees.
    Disk,
}

impl MountSource {
    pub fn as_str(self) -> &'static str {
        match self {
            MountSource::Tmpfs => "ephemeral://tmpfs",
            MountSource::Disk => "ephemeral://disk",
        }
    }
}

/// A file a job produces (spec §5.2): read from the job's disks when it
/// ends, kept in the repository as a blob, and named by its receipt.
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct Output {
    /// The artifact's name, unique in the job.
    pub name: String,
    /// A clean absolute path under `/src`, not on a tmpfs mount.
    pub path: String,
    /// `when "always"`: kept from a failing job too, such as a test report
    /// (oxygenflow#37). From a failing job it is never in the receipt.
    #[serde(default, skip_serializing_if = "std::ops::Not::not")]
    pub always: bool,
    /// An artifact is one file; a report is a directory with a
    /// `report.toml`, kept file by file with a manifest (oxygenhub#127).
    #[serde(default, skip_serializing_if = "OutputKind::is_artifact")]
    pub kind: OutputKind,
}

/// What an [`Output`] is.
#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum OutputKind {
    #[default]
    Artifact,
    Report,
}

impl OutputKind {
    pub fn is_artifact(&self) -> bool {
        *self == OutputKind::Artifact
    }

    pub fn word(&self) -> &'static str {
        match self {
            OutputKind::Artifact => "artifact",
            OutputKind::Report => "report",
        }
    }
}

/// A secret the job reads as a file (spec §9): never in the environment.
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct Secret {
    /// Which secret, as the hub names it; unique in the job.
    pub name: String,
    /// A clean absolute path under `/run/secrets/`, unique in the job;
    /// the file is 0400, the job's user's, on an in-memory mount.
    pub mount_path: String,
}

/// An OIDC token the hub mints for the job (oxygenhub#103), naming the
/// repository, bookmark, run and job, for `audience`; read as a file.
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct Identity {
    /// Who the token is for (`sts.amazonaws.com`, a Vault role's audience, …).
    pub audience: String,
    /// A clean absolute path under `/run/secrets/`, unique among the
    /// job's secrets and identities.
    pub mount_path: String,
}

/// A secret's mount path: under `/run/secrets/`, no `.` or `..`.
pub fn secret_path_ok(p: &str) -> bool {
    p.strip_prefix("/run/secrets/").is_some_and(|rest| !rest.is_empty() && rest.split('/').all(|c| !c.is_empty() && c != "." && c != ".."))
}

/// Artifacts a job may declare.
pub const MAX_OUTPUTS: usize = 16;

/// Whether `p` is a clean absolute path strictly under `/src`, with nothing
/// a debugfs command line would have to quote.
fn output_path_ok(p: &str) -> bool {
    p.strip_prefix("/src/").is_some_and(|rest| {
        !rest.is_empty()
            && rest.split('/').all(|c| !c.is_empty() && c != "." && c != "..")
            && p.bytes().all(|b| b.is_ascii_graphic() && b != b'"' && b != b'\\')
    })
}

/// Mounts a job may declare.
pub const MAX_MOUNTS: usize = 4;
/// The most secrets a job may mount.
pub const MAX_SECRETS: usize = 8;
/// The most identities a job may ask for.
pub const MAX_IDENTITIES: usize = 4;
/// Sizes a mount may have: 16 MiB to 64 GiB.
pub const MIN_MOUNT_MIB: u64 = 16;
pub const MAX_MOUNT_MIB: u64 = 64 * 1024;

/// Whether `t` is a clean absolute path strictly under `/src` or `/tmp`.
fn mount_target_ok(t: &str) -> bool {
    let under = t.strip_prefix("/src/").or_else(|| t.strip_prefix("/tmp/"));
    under.is_some_and(|rest| !rest.is_empty() && rest.split('/').all(|c| !c.is_empty() && c != "." && c != ".."))
}

/// One problem in a pipeline, at a 1-based line.
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct Problem {
    pub line: usize,
    pub message: String,
}

impl fmt::Display for Problem {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        write!(f, "line {}: {}", self.line, self.message)
    }
}

/// Every problem found, in line order.
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize, thiserror::Error)]
pub struct Problems(pub Vec<Problem>);

impl fmt::Display for Problems {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        let lines: Vec<String> = self.0.iter().map(|p| p.to_string()).collect();
        f.write_str(&lines.join("\n"))
    }
}

const MIN_MEMORY_MIB: u64 = 64;
const MAX_MEMORY_MIB: u64 = 16 * 1024;
const MAX_CPUS: i128 = 16;

/// Nodes of the full spec that M0 recognises and refuses.
const LATER_TOP: &[&str] = &["publish", "secret"];
const LATER_JOB: &[&str] = &["attest"];

struct Ctx<'a> {
    src: &'a str,
    problems: Vec<Problem>,
}

impl Ctx<'_> {
    fn line(&self, span: SourceSpan) -> usize {
        let at = span.offset().min(self.src.len());
        self.src[..at].bytes().filter(|b| *b == b'\n').count() + 1
    }

    fn err(&mut self, span: SourceSpan, message: impl Into<String>) {
        let line = self.line(span);
        self.problems.push(Problem { line, message: message.into() });
    }

    /// The node's arguments as strings; properties and other types are errors.
    fn strings(&mut self, node: &KdlNode) -> Vec<String> {
        let mut out = Vec::new();
        for e in node.entries() {
            if let Some(k) = e.name() {
                self.err(e.span(), format!("`{}` takes no property `{}`", node.name().value(), k.value()));
                continue;
            }
            match e.value() {
                KdlValue::String(s) => out.push(s.clone()),
                other => self.err(e.span(), format!("`{}` takes strings, not {}", node.name().value(), kind_of(other))),
            }
        }
        out
    }

    /// Exactly one string argument.
    fn one_string(&mut self, node: &KdlNode) -> Option<String> {
        let v = self.strings(node);
        if v.len() != 1 {
            self.err(node.span(), format!("`{}` takes one string", node.name().value()));
            return None;
        }
        v.into_iter().next()
    }

    fn one_integer(&mut self, node: &KdlNode) -> Option<i128> {
        match node.entries() {
            [e] if e.name().is_none() => match e.value() {
                KdlValue::Integer(n) => Some(*n),
                other => {
                    self.err(e.span(), format!("`{}` takes an integer, not {}", node.name().value(), kind_of(other)));
                    None
                }
            },
            _ => {
                self.err(node.span(), format!("`{}` takes one integer", node.name().value()));
                None
            }
        }
    }

    fn no_children(&mut self, node: &KdlNode) {
        if node.children().is_some() {
            self.err(node.span(), format!("`{}` takes no block", node.name().value()));
        }
    }

    fn name_ok(&mut self, span: SourceSpan, what: &str, name: &str) -> bool {
        let ok = !name.is_empty() && name.len() <= 64 && name.bytes().all(|b| b.is_ascii_alphanumeric() || b"._-".contains(&b));
        if !ok {
            self.err(span, format!("{what} name {name:?} must be 1 to 64 of A-Z a-z 0-9 . _ -"));
        }
        ok
    }
}

use kdl::{KdlDocument, KdlNode, KdlValue};
use miette::SourceSpan;

fn kind_of(v: &KdlValue) -> &'static str {
    match v {
        KdlValue::String(_) => "a string",
        KdlValue::Integer(_) => "an integer",
        KdlValue::Float(_) => "a number",
        KdlValue::Bool(_) => "a boolean",
        KdlValue::Null => "null",
    }
}

/// `"512MiB"`, `"4GiB"`, `"65536KiB"` to MiB.
fn parse_memory(s: &str) -> Option<u64> {
    let (num, unit) = s.split_at(s.find(|c: char| !c.is_ascii_digit())?);
    let n: u64 = num.parse().ok()?;
    let mib = match unit {
        "KiB" => n.is_multiple_of(1024).then_some(n / 1024)?,
        "MiB" => n,
        "GiB" => n.checked_mul(1024)?,
        _ => return None,
    };
    Some(mib)
}

/// The line of the first escape KDL 2.0 does not have but shells and C
/// do (`\0NN`, `\e`, `\xNN`), as someone colouring output writes them.
fn foreign_escape_line(src: &str) -> Option<usize> {
    src.lines()
        .position(|l| {
            let b = l.as_bytes();
            b.windows(2).enumerate().any(|(i, w)| {
                w[0] == b'\\'
                    && (i == 0 || b[i - 1] != b'\\')
                    && (w[1] == b'e' || w[1] == b'x' || (w[1] == b'0' && b.get(i + 2).is_some_and(u8::is_ascii_digit)))
            })
        })
        .map(|i| i + 1)
}

impl Pipeline {
    /// Parse and validate a pipeline document.
    pub fn parse(src: &str) -> Result<Pipeline, Problems> {
        let doc = match KdlDocument::parse_v2(src) {
            Ok(d) => d,
            Err(e) => {
                let mut ctx = Ctx { src, problems: Vec::new() };
                for d in &e.diagnostics {
                    let msg = d.message.clone().or_else(|| d.label.clone()).unwrap_or_else(|| "not valid KDL 2.0".into());
                    ctx.err(d.span, msg);
                }
                if ctx.problems.is_empty() {
                    ctx.problems.push(Problem { line: 1, message: "not valid KDL 2.0".into() });
                }
                if let Some(line) = foreign_escape_line(src) {
                    ctx.problems.push(Problem {
                        line,
                        message: "KDL strings have no \\033, \\e or \\x escapes: write ESC (for ANSI colours) as \\u{1b}".into(),
                    });
                }
                return Err(Problems(ctx.problems));
            }
        };
        let mut ctx = Ctx { src, problems: Vec::new() };
        let mut schema = None;
        let mut sandboxes = BTreeMap::new();
        let mut jobs: BTreeMap<String, (Job, SourceSpan)> = BTreeMap::new();
        let mut on = None;
        for node in doc.nodes() {
            match node.name().value() {
                "on" => {
                    if on.is_some() {
                        ctx.err(node.span(), "`on` is given twice");
                    }
                    on = Some(parse_on(&mut ctx, node));
                }
                "schema-version" => {
                    ctx.no_children(node);
                    match ctx.one_integer(node) {
                        Some(1) => schema = Some(1),
                        Some(n) => ctx.err(node.span(), format!("schema-version {n} is not supported; this runner reads 1")),
                        None => {}
                    }
                }
                "sandbox" => {
                    if let Some(sb) = parse_sandbox(&mut ctx, node) {
                        if sandboxes.contains_key(&sb.name) {
                            ctx.err(node.span(), format!("sandbox {:?} is declared twice", sb.name));
                        } else {
                            sandboxes.insert(sb.name.clone(), sb);
                        }
                    }
                }
                "job" => {
                    if let Some(job) = parse_job(&mut ctx, node) {
                        if jobs.contains_key(&job.name) {
                            ctx.err(node.span(), format!("job {:?} is declared twice", job.name));
                        } else {
                            jobs.insert(job.name.clone(), (job, node.span()));
                        }
                    }
                }
                n if LATER_TOP.contains(&n) => ctx.err(node.span(), format!("`{n}` is not in M0 yet")),
                n => ctx.err(node.span(), format!("unknown node `{n}`; expected schema-version, on, sandbox or job")),
            }
        }
        if schema.is_none() && !ctx.problems.iter().any(|p| p.message.contains("schema-version")) {
            ctx.problems.push(Problem { line: 1, message: "`schema-version 1` is required".into() });
        }
        if jobs.is_empty() && ctx.problems.is_empty() {
            ctx.problems.push(Problem { line: 1, message: "a pipeline needs at least one job".into() });
        }
        // References and cycles.
        for (job, span) in jobs.values() {
            if !job.sandbox.is_empty() && !sandboxes.contains_key(&job.sandbox) {
                ctx.err(*span, format!("job {:?} uses sandbox {:?}, which is not declared", job.name, job.sandbox));
            }
            for d in &job.depends_on {
                if !jobs.contains_key(d) {
                    ctx.err(*span, format!("job {:?} depends on {:?}, which is not declared", job.name, d));
                } else if d == &job.name {
                    ctx.err(*span, format!("job {:?} depends on itself", job.name));
                }
            }
        }
        let graph: BTreeMap<&str, Vec<&str>> =
            jobs.values().map(|(j, _)| (j.name.as_str(), j.depends_on.iter().map(String::as_str).collect())).collect();
        if let Some(cycle) = find_cycle(&graph) {
            let span = jobs[cycle[0]].1;
            ctx.err(span, format!("jobs depend on each other in a cycle: {}", cycle.join(" → ")));
        }
        if !ctx.problems.is_empty() {
            ctx.problems.sort_by_key(|p| p.line);
            return Err(Problems(ctx.problems));
        }
        Ok(Pipeline { schema_version: 1, sandboxes, jobs: jobs.into_iter().map(|(k, (j, _))| (k, j)).collect(), on })
    }

    /// Jobs in an order where each comes after what it depends on (ties by name).
    pub fn order(&self) -> Vec<&Job> {
        let mut done: BTreeSet<&str> = BTreeSet::new();
        let mut out = Vec::with_capacity(self.jobs.len());
        while out.len() < self.jobs.len() {
            let before = out.len();
            for j in self.jobs.values() {
                if !done.contains(j.name.as_str()) && j.depends_on.iter().all(|d| done.contains(d.as_str())) {
                    out.push(j);
                }
            }
            for j in &out[before..] {
                done.insert(j.name.as_str());
            }
            if out.len() == before {
                break; // validated pipelines have no cycles
            }
        }
        out
    }

    /// Each job's depth: 0 without dependencies, else one more than its deepest dependency.
    pub fn depths(&self) -> BTreeMap<&str, usize> {
        let mut depth: BTreeMap<&str, usize> = BTreeMap::new();
        for j in self.order() {
            let d = j.depends_on.iter().map(|x| depth.get(x.as_str()).copied().unwrap_or(0) + 1).max().unwrap_or(0);
            depth.insert(j.name.as_str(), d);
        }
        depth
    }
}

fn find_cycle<'a>(graph: &BTreeMap<&'a str, Vec<&'a str>>) -> Option<Vec<&'a str>> {
    #[derive(Clone, Copy, PartialEq)]
    enum Mark {
        Visiting,
        Done,
    }
    fn visit<'a>(
        n: &'a str,
        graph: &BTreeMap<&'a str, Vec<&'a str>>,
        marks: &mut BTreeMap<&'a str, Mark>,
        stack: &mut Vec<&'a str>,
    ) -> Option<Vec<&'a str>> {
        match marks.get(n) {
            Some(Mark::Done) => return None,
            Some(Mark::Visiting) => {
                let at = stack.iter().position(|s| *s == n).unwrap_or(0);
                let mut c = stack[at..].to_vec();
                c.push(n);
                return Some(c);
            }
            None => {}
        }
        marks.insert(n, Mark::Visiting);
        stack.push(n);
        for d in graph.get(n).into_iter().flatten() {
            if graph.contains_key(d) && *d != n {
                if let Some(c) = visit(d, graph, marks, stack) {
                    return Some(c);
                }
            }
        }
        stack.pop();
        marks.insert(n, Mark::Done);
        None
    }
    let mut marks = BTreeMap::new();
    for n in graph.keys() {
        let mut stack = Vec::new();
        if let Some(c) = visit(n, graph, &mut marks, &mut stack) {
            return Some(c);
        }
    }
    None
}

fn block<'a>(ctx: &mut Ctx, node: &'a KdlNode) -> &'a [KdlNode] {
    match node.children() {
        Some(d) => d.nodes(),
        None => {
            ctx.err(node.span(), format!("`{}` needs a block {{ … }}", node.name().value()));
            &[]
        }
    }
}

fn parse_sandbox(ctx: &mut Ctx, node: &KdlNode) -> Option<Sandbox> {
    let name = ctx.one_string(node)?;
    let named = ctx.name_ok(node.span(), "sandbox", &name);
    let (mut runtime, mut rootfs, mut memory, mut cpus, mut network) =
        (Some(Runtime::Microvm), None, Some(1024u32), Some(1u32), Some(Network::Isolated));
    let mut seen = BTreeSet::new();
    let mut env = BTreeMap::new();
    let mut runs_on = Vec::new();
    let mut timeout_s = None;
    let mut arch: Option<String> = None;
    for c in block(ctx, node) {
        let key = c.name().value();
        if key != "runs-on" && !seen.insert(key.to_string()) {
            ctx.err(c.span(), format!("`{key}` is given twice"));
        }
        if key == "env" {
            parse_env(ctx, c, &mut env);
            continue;
        }
        if key == "runs-on" {
            ctx.no_children(c);
            parse_labels(ctx, c, &mut runs_on);
            continue;
        }
        ctx.no_children(c);
        match key {
            "runtime" => {
                runtime = match ctx.one_string(c).as_deref() {
                    Some("microvm") => Some(Runtime::Microvm),
                    Some("syscall") | Some("gvisor") => Some(Runtime::Syscall),
                    Some("wasm") => {
                        ctx.err(c.span(), "runtime \"wasm\" is not in M0 yet; use \"microvm\" or \"syscall\"");
                        None
                    }
                    Some(other) => {
                        ctx.err(c.span(), format!("runtime {other:?}: expected \"microvm\", \"syscall\" or \"wasm\""));
                        None
                    }
                    None => None,
                }
            }
            "rootfs" => {
                rootfs = match ctx.one_string(c) {
                    Some(r) => match r.strip_prefix("cas://blob/") {
                        Some(h) if h.len() == 64 && h.bytes().all(|b| b.is_ascii_hexdigit() && !b.is_ascii_uppercase()) => {
                            Some(h.to_string())
                        }
                        _ => {
                            ctx.err(c.span(), format!("rootfs {r:?}: expected \"cas://blob/\" and 64 lower-case hex digits"));
                            None
                        }
                    },
                    None => None,
                }
            }
            "arch" => {
                arch = match ctx.one_string(c) {
                    Some(a) if arch_ok(&a) => Some(a),
                    Some(a) => {
                        ctx.err(c.span(), format!("arch {a:?} is not an architecture name such as x86_64, aarch64 or riscv64"));
                        None
                    }
                    None => None,
                }
            }
            "memory-limit" => {
                memory = match ctx.one_string(c) {
                    Some(m) => match parse_memory(&m) {
                        Some(mib) if (MIN_MEMORY_MIB..=MAX_MEMORY_MIB).contains(&mib) => Some(mib as u32),
                        Some(_) => {
                            ctx.err(c.span(), format!("memory-limit {m:?} must be from 64MiB to 16GiB"));
                            None
                        }
                        None => {
                            ctx.err(c.span(), format!("memory-limit {m:?}: expected a whole number of KiB, MiB or GiB, as \"4GiB\""));
                            None
                        }
                    },
                    None => None,
                }
            }
            "cpu-limit" => {
                cpus = match ctx.one_integer(c) {
                    Some(n) if (1..=MAX_CPUS).contains(&n) => Some(n as u32),
                    Some(n) => {
                        ctx.err(c.span(), format!("cpu-limit {n} must be from 1 to {MAX_CPUS}"));
                        None
                    }
                    None => None,
                }
            }
            "network" => {
                network = match ctx.one_string(c).as_deref() {
                    Some("isolated") => Some(Network::Isolated),
                    // Loopback only: the guest's `lo` comes up, and it still has
                    // no device that reaches anything outside it.
                    Some("loopback") => Some(Network::Loopback),
                    Some("egress") => {
                        ctx.err(c.span(), "network \"egress\" is not in M0; M0 has \"isolated\" and \"loopback\"");
                        None
                    }
                    Some(other) => {
                        ctx.err(c.span(), format!("network {other:?}: expected \"isolated\", \"loopback\" or \"egress\""));
                        None
                    }
                    None => None,
                }
            }
            "timeout" => timeout_s = parse_timeout(ctx, c),
            other => ctx.err(c.span(), format!("unknown sandbox setting `{other}`")),
        }
    }
    if rootfs.is_none() && !seen.contains("rootfs") {
        ctx.err(node.span(), format!("sandbox {name:?} needs a rootfs"));
    }
    if !named {
        return None;
    }
    Some(Sandbox {
        name,
        runtime: runtime?,
        rootfs: rootfs?,
        memory_mib: memory?,
        cpus: cpus?,
        network: network?,
        env,
        runs_on,
        timeout_s,
        arch,
    })
}

/// An `on` block: `push` and `tag` with globs (none: any), `manual`,
/// `schedule` with five-field cron lines.
fn parse_on(ctx: &mut Ctx, node: &KdlNode) -> On {
    let mut on = On::default();
    if !node.entries().is_empty() {
        ctx.err(node.span(), "`on` takes a block: on { push \"main\"; tag \"v*\"; manual }");
    }
    let globs = |ctx: &mut Ctx, c: &KdlNode| -> Vec<String> {
        let g = ctx.strings(c);
        for p in &g {
            if let Err(e) = globset::Glob::new(p) {
                ctx.err(c.span(), format!("{p:?}: {e}"));
            }
        }
        g
    };
    for c in block(ctx, node) {
        ctx.no_children(c);
        match c.name().value() {
            "push" => on.push.get_or_insert_with(Vec::new).extend(globs(ctx, c)),
            "tag" => on.tag.get_or_insert_with(Vec::new).extend(globs(ctx, c)),
            "manual" => on.manual = true,
            "schedule" => {
                for line in ctx.strings(c) {
                    if crate::cron::Cron::parse(&line).is_some() {
                        on.schedule.push(line);
                    } else {
                        ctx.err(
                            c.span(),
                            format!("schedule {line:?}: expected five cron fields (minute hour day month weekday), as \"0 3 * * *\""),
                        );
                    }
                }
            }
            other => ctx.err(c.span(), format!("unknown `on` event `{other}`; expected push, tag, manual or schedule")),
        }
    }
    if on.push.is_none() && on.tag.is_none() && !on.manual && on.schedule.is_empty() {
        ctx.err(node.span(), "`on` names no event, so nothing would run it");
    }
    on
}

/// The longest a pipeline may ask a job to run.
pub const MAX_TIMEOUT_S: u64 = 24 * 3600;

/// A duration: whole hours, minutes and seconds in that order, each
/// optional, as "90m", "2h", "1h30m" or "45s".
pub fn parse_duration(s: &str) -> Option<u64> {
    let (mut total, mut num, mut last) = (0u64, String::new(), 0u8);
    for ch in s.chars() {
        if ch.is_ascii_digit() {
            num.push(ch);
            continue;
        }
        let (unit, mult) = match ch {
            'h' => (1, 3600),
            'm' => (2, 60),
            's' => (3, 1),
            _ => return None,
        };
        // Each unit once, largest first, each with its number.
        if num.is_empty() || unit <= last {
            return None;
        }
        last = unit;
        total = total.checked_add(num.parse::<u64>().ok()?.checked_mul(mult)?)?;
        num.clear();
    }
    (num.is_empty() && total > 0).then_some(total)
}

/// A `timeout` line: one duration, from a minute to a day.
fn parse_timeout(ctx: &mut Ctx, c: &KdlNode) -> Option<u64> {
    let t = ctx.one_string(c)?;
    match parse_duration(&t) {
        Some(secs) if (60..=MAX_TIMEOUT_S).contains(&secs) => Some(secs),
        Some(_) => {
            ctx.err(c.span(), format!("timeout {t:?} must be from 1m to 24h"));
            None
        }
        None => {
            ctx.err(c.span(), format!("timeout {t:?}: expected hours, minutes and seconds, as \"90m\" or \"1h30m\""));
            None
        }
    }
}

/// The labels of a `runs-on` line: lower-case letters, digits, `-`, `_`
/// and `.`, at most 64 bytes each; into `into`, sorted, once each.
fn parse_labels(ctx: &mut Ctx, c: &KdlNode, into: &mut Vec<String>) {
    let labels = ctx.strings(c);
    if labels.is_empty() {
        ctx.err(c.span(), "`runs-on` needs one or more labels, as runs-on \"riscv64\"");
    }
    for l in labels {
        let ok = !l.is_empty()
            && l.len() <= 64
            && l.bytes().all(|b| b.is_ascii_lowercase() || b.is_ascii_digit() || matches!(b, b'-' | b'_' | b'.'));
        if ok {
            into.push(l);
        } else {
            ctx.err(c.span(), format!("runs-on {l:?}: a label is lower-case letters, digits, -, _ and ., at most 64"));
        }
    }
    into.sort();
    into.dedup();
}

/// An `env` block of `NAME "value"` lines, into `env`.
fn parse_env(ctx: &mut Ctx, c: &KdlNode, env: &mut BTreeMap<String, String>) {
    if !c.entries().is_empty() {
        ctx.err(c.span(), "`env` takes a block of NAME \"value\" lines");
    }
    for e in block(ctx, c) {
        let k = e.name().value().to_string();
        let valid =
            !k.is_empty() && !k.starts_with(|c: char| c.is_ascii_digit()) && k.bytes().all(|b| b.is_ascii_alphanumeric() || b == b'_');
        if !valid {
            ctx.err(e.span(), format!("environment name {k:?} must be letters, digits and _"));
            continue;
        }
        ctx.no_children(e);
        if let Some(v) = ctx.one_string(e) {
            if env.insert(k.clone(), v).is_some() {
                ctx.err(e.span(), format!("environment {k} is set twice"));
            }
        }
    }
}

/// `mount { source "ephemeral://disk"; target "/src/target"; size "8GiB" }`.
fn parse_mount(ctx: &mut Ctx, node: &KdlNode) -> Option<Mount> {
    let (mut source, mut target, mut size) = (None, None, None);
    let mut seen = BTreeSet::new();
    for c in block(ctx, node) {
        let key = c.name().value();
        if !seen.insert(key.to_string()) {
            ctx.err(c.span(), format!("`{key}` is given twice"));
            continue;
        }
        ctx.no_children(c);
        match key {
            "source" => {
                source = match ctx.one_string(c).as_deref() {
                    Some("ephemeral://tmpfs") => Some(MountSource::Tmpfs),
                    Some("ephemeral://disk") => Some(MountSource::Disk),
                    Some(other) => {
                        ctx.err(
                            c.span(),
                            format!("mount source {other:?} is not in M0 yet; M0 mounts \"ephemeral://tmpfs\" or \"ephemeral://disk\""),
                        );
                        None
                    }
                    None => None,
                }
            }
            "target" => {
                target = match ctx.one_string(c) {
                    Some(t) if mount_target_ok(&t) => Some(t),
                    Some(t) => {
                        ctx.err(
                            c.span(),
                            format!("mount target {t:?} must be an absolute path under /src or /tmp, without . or .. or a trailing /"),
                        );
                        None
                    }
                    None => None,
                }
            }
            "size" => {
                size = match ctx.one_string(c) {
                    Some(m) => match parse_memory(&m) {
                        Some(mib) if (MIN_MOUNT_MIB..=MAX_MOUNT_MIB).contains(&mib) => Some(mib as u32),
                        Some(_) => {
                            ctx.err(c.span(), format!("mount size {m:?} must be from 16MiB to 64GiB"));
                            None
                        }
                        None => {
                            ctx.err(c.span(), format!("mount size {m:?}: expected a whole number of KiB, MiB or GiB, as \"8GiB\""));
                            None
                        }
                    },
                    None => None,
                }
            }
            "readonly" | "include" => ctx.err(c.span(), format!("mount `{key}` is not in M0 yet")),
            other => ctx.err(c.span(), format!("unknown mount setting `{other}`")),
        }
    }
    for (name, missing) in [("source", source.is_none()), ("target", target.is_none()), ("size", size.is_none())] {
        if missing && !seen.contains(name) {
            ctx.err(node.span(), format!("a mount needs a {name}"));
        }
    }
    Some(Mount { source: source?, target: target?, size_mib: size? })
}

/// `output { artifact "name" { path "/src/target/release/tool" } }`.
/// `secret "deploy-token" { mount-path "/run/secrets/token" }` (spec §9.1).
/// The hub is the vault: the secret of that name set for the repository,
/// its organisation or the site. `vault-ref` is not in this slice.
fn parse_secret(ctx: &mut Ctx, node: &KdlNode, job: &mut Job) {
    let Some(name) = ctx.one_string(node) else {
        ctx.err(node.span(), "`secret` takes a name and a block with mount-path");
        return;
    };
    if !ctx.name_ok(node.span(), "secret", &name) {
        return;
    }
    let mut path = None;
    for c in block(ctx, node) {
        ctx.no_children(c);
        match c.name().value() {
            "mount-path" => match ctx.one_string(c) {
                Some(p) if secret_path_ok(&p) => path = Some(p),
                Some(p) => {
                    ctx.err(c.span(), format!("secret mount-path {p:?} must be an absolute path under /run/secrets/, without . or .."))
                }
                None => {}
            },
            "vault-ref" => ctx.err(c.span(), "`vault-ref` is not in M0; the hub supplies the secret of that name"),
            other => ctx.err(c.span(), format!("unknown secret setting `{other}`")),
        }
    }
    let Some(mount_path) = path else {
        ctx.err(node.span(), format!("secret {name:?} needs a mount-path"));
        return;
    };
    if job.secrets.iter().any(|s| s.name == name) {
        ctx.err(node.span(), format!("secret {name:?} is given twice"));
    } else if job.secrets.iter().any(|s| s.mount_path == mount_path) || job.identities.iter().any(|i| i.mount_path == mount_path) {
        ctx.err(node.span(), format!("two files mount at {mount_path:?}"));
    } else if job.secrets.len() == MAX_SECRETS {
        ctx.err(node.span(), format!("a job has at most {MAX_SECRETS} secrets"));
    } else {
        job.secrets.push(Secret { name, mount_path });
    }
}

/// `identity "sts.amazonaws.com" { mount-path "/run/secrets/aws-token" }`:
/// the hub mints an OIDC token for that audience and the runner mounts it
/// as a file (oxygenhub#103).
fn parse_identity(ctx: &mut Ctx, node: &KdlNode, job: &mut Job) {
    let Some(audience) = ctx.one_string(node) else {
        ctx.err(node.span(), "`identity` takes an audience and a block with mount-path");
        return;
    };
    let aud_ok = !audience.is_empty() && audience.len() <= 255 && !audience.chars().any(|c| c.is_whitespace() || c.is_control());
    if !aud_ok {
        ctx.err(node.span(), format!("identity audience {audience:?} must be 1 to 255 characters without spaces"));
        return;
    }
    let mut path = None;
    for c in block(ctx, node) {
        ctx.no_children(c);
        match c.name().value() {
            "mount-path" => match ctx.one_string(c) {
                Some(p) if secret_path_ok(&p) => path = Some(p),
                Some(p) => {
                    ctx.err(c.span(), format!("identity mount-path {p:?} must be an absolute path under /run/secrets/, without . or .."))
                }
                None => {}
            },
            other => ctx.err(c.span(), format!("unknown identity setting `{other}`")),
        }
    }
    let Some(mount_path) = path else {
        ctx.err(node.span(), format!("identity {audience:?} needs a mount-path"));
        return;
    };
    if job.identities.iter().any(|i| i.audience == audience) {
        ctx.err(node.span(), format!("identity {audience:?} is given twice"));
    } else if job.identities.iter().any(|i| i.mount_path == mount_path) || job.secrets.iter().any(|s| s.mount_path == mount_path) {
        ctx.err(node.span(), format!("two files mount at {mount_path:?}"));
    } else if job.identities.len() == MAX_IDENTITIES {
        ctx.err(node.span(), format!("a job has at most {MAX_IDENTITIES} identities"));
    } else {
        job.identities.push(Identity { audience, mount_path });
    }
}

fn parse_output(ctx: &mut Ctx, node: &KdlNode, job: &mut Job) {
    if !node.entries().is_empty() {
        ctx.err(node.span(), "`output` takes a block of artifacts and reports");
    }
    for a in block(ctx, node) {
        let kind = match a.name().value() {
            "artifact" => OutputKind::Artifact,
            "report" => OutputKind::Report,
            other => {
                ctx.err(
                    a.span(),
                    format!("unknown output `{other}`; outputs are `artifact \"name\" {{ path \"/src/file\" }}` and `report \"name\" {{ path \"/src/dir\" }}`"),
                );
                continue;
            }
        };
        let word = kind.word();
        let Some(name) = ctx.one_string(a) else { continue };
        if !ctx.name_ok(a.span(), word, &name) {
            continue;
        }
        let mut path = None;
        let mut always = false;
        for c in block(ctx, a) {
            ctx.no_children(c);
            match c.name().value() {
                "path" => match ctx.one_string(c) {
                    Some(p) if output_path_ok(&p) => path = Some(p),
                    Some(p) => ctx
                        .err(c.span(), format!("{word} path {p:?} must be an absolute path under /src, without . or .., quotes or spaces")),
                    None => {}
                },
                "format" if kind == OutputKind::Artifact => {
                    if ctx.one_string(c).as_deref() != Some("blob") {
                        ctx.err(c.span(), "artifact format is \"blob\"");
                    }
                }
                "when" => match ctx.one_string(c).as_deref() {
                    Some("passed") => always = false,
                    Some("always") => always = true,
                    Some(_) => ctx.err(c.span(), format!("{word} `when` is \"passed\" (the default) or \"always\"")),
                    None => {}
                },
                other => ctx.err(c.span(), format!("unknown {word} setting `{other}`")),
            }
        }
        let Some(path) = path else {
            ctx.err(a.span(), format!("{word} {name:?} needs a path"));
            continue;
        };
        if job.outputs.iter().any(|o| o.name == name) {
            ctx.err(a.span(), format!("output {name:?} is given twice (artifacts and reports share names)"));
        } else if job.outputs.len() == MAX_OUTPUTS {
            ctx.err(a.span(), format!("a job has at most {MAX_OUTPUTS} artifacts and reports"));
        } else {
            job.outputs.push(Output { name, path, always, kind });
        }
    }
}

fn parse_job(ctx: &mut Ctx, node: &KdlNode) -> Option<Job> {
    let name = ctx.one_string(node)?;
    let named = ctx.name_ok(node.span(), "job", &name);
    let mut job = Job {
        name: name.clone(),
        sandbox: String::new(),
        scope: vec![],
        depends_on: vec![],
        env: BTreeMap::new(),
        exec: vec![],
        mounts: vec![],
        secrets: vec![],
        identities: vec![],
        outputs: vec![],
        runs_on: vec![],
        timeout_s: None,
        on: None,
    };
    let mut seen = BTreeSet::new();
    let mut outputs_at = None;
    for c in block(ctx, node) {
        let key = c.name().value();
        if key != "scope"
            && key != "depends-on"
            && key != "mount"
            && key != "secret"
            && key != "identity"
            && key != "runs-on"
            && !seen.insert(key.to_string())
        {
            ctx.err(c.span(), format!("`{key}` is given twice"));
        }
        match key {
            "sandbox" => {
                ctx.no_children(c);
                job.sandbox = ctx.one_string(c).unwrap_or_default();
            }
            "scope" => {
                ctx.no_children(c);
                for g in ctx.strings(c) {
                    match globset::Glob::new(&g) {
                        Ok(_) => job.scope.push(g),
                        Err(e) => ctx.err(c.span(), format!("scope {g:?}: {e}")),
                    }
                }
            }
            "depends-on" => {
                ctx.no_children(c);
                job.depends_on.extend(ctx.strings(c));
            }
            "runs-on" => {
                ctx.no_children(c);
                parse_labels(ctx, c, &mut job.runs_on);
            }
            "timeout" => {
                ctx.no_children(c);
                job.timeout_s = parse_timeout(ctx, c);
            }
            "on" => job.on = Some(parse_on(ctx, c)),
            "env" => parse_env(ctx, c, &mut job.env),
            "exec" => {
                ctx.no_children(c);
                job.exec = ctx.strings(c);
                if job.exec.is_empty() {
                    ctx.err(c.span(), "`exec` needs a program and its arguments");
                } else if job.exec.iter().any(|a| a.contains('\0')) {
                    ctx.err(c.span(), "`exec` arguments cannot contain NUL");
                }
            }
            "mount" => {
                if !c.entries().is_empty() {
                    ctx.err(c.span(), "`mount` takes a block of source, target and size");
                }
                if let Some(m) = parse_mount(ctx, c) {
                    if job.mounts.len() == MAX_MOUNTS {
                        ctx.err(c.span(), format!("a job has at most {MAX_MOUNTS} mounts"));
                    } else if let Some(other) = job.mounts.iter().find(|o| {
                        o.target == m.target
                            || m.target.starts_with(&format!("{}/", o.target))
                            || o.target.starts_with(&format!("{}/", m.target))
                    }) {
                        ctx.err(c.span(), format!("mount {:?} overlaps mount {:?}", m.target, other.target));
                    } else {
                        job.mounts.push(m);
                    }
                }
            }
            "output" => {
                outputs_at = Some(c.span());
                parse_output(ctx, c, &mut job);
            }
            "secret" => parse_secret(ctx, c, &mut job),
            "identity" => parse_identity(ctx, c, &mut job),
            n if LATER_JOB.contains(&n) => ctx.err(c.span(), format!("`{n}` is not in M0 yet")),
            other => ctx.err(c.span(), format!("unknown job setting `{other}`")),
        }
    }
    job.mounts.sort_by(|a, b| a.target.cmp(&b.target));
    job.outputs.sort_by(|a, b| a.name.cmp(&b.name));
    job.secrets.sort_by(|a, b| a.name.cmp(&b.name));
    job.identities.sort_by(|a, b| a.audience.cmp(&b.audience));
    // A tmpfs is gone when the job ends; nothing on it can be kept.
    for o in &job.outputs {
        if let Some(m) = job.mounts.iter().find(|m| m.source == MountSource::Tmpfs && o.path.starts_with(&format!("{}/", m.target))) {
            ctx.err(
                outputs_at.unwrap_or(node.span()),
                format!(
                    "artifact {:?} is on the tmpfs mount {:?}, which is gone when the job ends; use ephemeral://disk",
                    o.name, m.target
                ),
            );
        }
    }
    if job.sandbox.is_empty() && !seen.contains("sandbox") {
        ctx.err(node.span(), format!("job {name:?} needs a sandbox"));
    }
    if job.exec.is_empty() && !seen.contains("exec") {
        ctx.err(node.span(), format!("job {name:?} needs an exec"));
    }
    job.depends_on.sort();
    job.depends_on.dedup();
    // Kept even when incomplete, so its references are checked too; the
    // problems above already make the pipeline invalid.
    named.then_some(job)
}
`/`	Search
`g` `c`	Code
`g` `h`	History
`g` `b`	Bookmarks
`g` `t`	Trust
`y`	Permalink to this state (files and trees)
`?`	This sheet
Keyboard
