powercut 0.1.0

//! What came back, and how sure we are about it.
//!
//! The type that matters most here is [`Finding`], and the variant that
//! matters most is [`Finding::NotMeasured`]. A check that could not be
//! performed must never render as a pass — that is the whole reason this
//! module has four outcomes instead of two.

use std::fmt;

use crate::capture::{CapturedFile, Tier};
use crate::witness::Coverage;

/// What kind of thing went wrong, so a ratchet can tell *new* breakage from
/// breakage it already knows about.
///
/// ```
/// # use powercut::ViolationKind;
/// assert_eq!(ViolationKind::LostData.as_str(), "lost-data");
/// ```
#[non_exhaustive]
#[derive(Copy, Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub enum ViolationKind {
    /// Something the ack promised is simply absent.
    LostData,
    /// Something the ack promised came back with a different value.
    Divergent,
    /// State that existed **before** the workload ran is missing, so the
    /// captured copy is not a usable store and nothing else in the report is a
    /// durability measurement.
    Baseline,
}

impl ViolationKind {
    /// A short, stable, lowercase label. Baseline files contain these, so they
    /// are part of the public contract.
    ///
    /// ```
    /// # use powercut::ViolationKind;
    /// assert_eq!(ViolationKind::Baseline.as_str(), "baseline");
    /// ```
    #[must_use]
    pub const fn as_str(self) -> &'static str {
        match self {
            ViolationKind::LostData => "lost-data",
            ViolationKind::Divergent => "divergent",
            ViolationKind::Baseline => "baseline",
        }
    }

    /// Parse a label written by [`Self::as_str`].
    ///
    /// ```
    /// # use powercut::ViolationKind;
    /// assert_eq!(ViolationKind::parse("divergent"), Some(ViolationKind::Divergent));
    /// assert_eq!(ViolationKind::parse("nonsense"), None);
    /// ```
    #[must_use]
    pub fn parse(s: &str) -> Option<Self> {
        match s {
            "lost-data" => Some(ViolationKind::LostData),
            "divergent" => Some(ViolationKind::Divergent),
            "baseline" => Some(ViolationKind::Baseline),
            _ => None,
        }
    }
}

/// One way the acked promise failed to survive.
///
/// ```
/// # use powercut::{Violation, ViolationKind};
/// let v = Violation::lost("row id=7 the API acked is absent");
/// assert_eq!(v.kind, ViolationKind::LostData);
/// ```
#[derive(Clone, Debug, PartialEq, Eq)]
pub struct Violation {
    /// What kind of failure this is.
    pub kind: ViolationKind,
    /// The human-readable detail, written for whoever reads the failure.
    pub detail: String,
}

impl Violation {
    /// Something the ack promised is gone.
    ///
    /// ```
    /// # use powercut::Violation;
    /// assert!(Violation::lost("order 42 is absent").detail.contains("42"));
    /// ```
    #[must_use]
    pub fn lost(detail: impl Into<String>) -> Self {
        Self {
            kind: ViolationKind::LostData,
            detail: detail.into(),
        }
    }

    /// Something the ack promised came back different.
    ///
    /// ```
    /// # use powercut::{Violation, ViolationKind};
    /// assert_eq!(Violation::divergent("balance is 9, acked 10").kind, ViolationKind::Divergent);
    /// ```
    #[must_use]
    pub fn divergent(detail: impl Into<String>) -> Self {
        Self {
            kind: ViolationKind::Divergent,
            detail: detail.into(),
        }
    }

    /// Pre-workload state is missing, so this run measured nothing.
    ///
    /// ```
    /// # use powercut::{Violation, ViolationKind};
    /// let v = Violation::baseline("the schema itself is absent from the copy");
    /// assert_eq!(v.kind, ViolationKind::Baseline);
    /// ```
    #[must_use]
    pub fn baseline(detail: impl Into<String>) -> Self {
        Self {
            kind: ViolationKind::Baseline,
            detail: detail.into(),
        }
    }
}

impl fmt::Display for Violation {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        let prefix = match self.kind {
            ViolationKind::LostData => "LOST",
            ViolationKind::Divergent => "DIVERGENT",
            ViolationKind::Baseline => "HARNESS/BASELINE",
        };
        write!(f, "{prefix}: {}", self.detail)
    }
}

/// A small helper for writing checkers that put the baseline check first.
///
/// The ordering is the point. If the captured copy lost state that existed
/// *before* the workload, then every later assertion fails for a reason that
/// has nothing to do with durability — and a report that says "data loss" in
/// that situation is a false accusation.
///
/// ```
/// use powercut::{Checker, ViolationKind};
///
/// let mut c = Checker::new();
/// c.baseline("schema present", true, true);
/// c.survives("row 7", Some(42_i64), None);
/// let violations = c.finish();
///
/// assert_eq!(violations.len(), 1);
/// assert_eq!(violations[0].kind, ViolationKind::LostData);
/// ```
#[derive(Debug, Default)]
pub struct Checker {
    baseline: Vec<Violation>,
    rest: Vec<Violation>,
}

impl Checker {
    /// An empty checker.
    ///
    /// ```
    /// # use powercut::Checker;
    /// assert!(Checker::new().finish().is_empty());
    /// ```
    #[must_use]
    pub fn new() -> Self {
        Self::default()
    }

    /// Assert something that was true **before** the workload ran. A failure
    /// here means the harness, not the store, is the problem.
    ///
    /// ```
    /// # use powercut::{Checker, ViolationKind};
    /// let mut c = Checker::new();
    /// c.baseline("table count", 1_usize, 0_usize);
    /// assert_eq!(c.finish()[0].kind, ViolationKind::Baseline);
    /// ```
    pub fn baseline<T>(&mut self, what: &str, expected: T, actual: T)
    where
        T: PartialEq + fmt::Debug,
    {
        if expected != actual {
            self.baseline.push(Violation::baseline(format!(
                "{what}: the captured copy has {actual:?}, but this was already true \
                 BEFORE the workload ran (expected {expected:?}). The copy is not a usable \
                 store, so nothing else in this report is a durability measurement."
            )));
        }
    }

    /// Assert something the workload's ack promised.
    ///
    /// A missing value is reported as [`ViolationKind::LostData`], a different
    /// value as [`ViolationKind::Divergent`] — a distinction that matters,
    /// because losing a row and reading a stale row are different bugs.
    ///
    /// ```
    /// # use powercut::{Checker, ViolationKind};
    /// let mut c = Checker::new();
    /// c.survives("balance", 10_i64, 9_i64);
    /// assert_eq!(c.finish()[0].kind, ViolationKind::Divergent);
    /// ```
    pub fn survives<T>(&mut self, what: &str, expected: T, actual: T)
    where
        T: PartialEq + fmt::Debug + IsAbsent,
    {
        if expected == actual {
            return;
        }
        let detail = format!("{what}: acked {expected:?}, the durable set has {actual:?}");
        self.rest.push(if actual.is_absent() {
            Violation::lost(detail)
        } else {
            Violation::divergent(detail)
        });
    }

    /// Record a violation you built yourself.
    ///
    /// ```
    /// # use powercut::{Checker, Violation};
    /// let mut c = Checker::new();
    /// c.push(Violation::lost("the WAL replayed but the row is not in it"));
    /// assert_eq!(c.finish().len(), 1);
    /// ```
    pub fn push(&mut self, violation: Violation) {
        if violation.kind == ViolationKind::Baseline {
            self.baseline.push(violation);
        } else {
            self.rest.push(violation);
        }
    }

    /// Every violation, baseline failures first.
    ///
    /// ```
    /// # use powercut::{Checker, ViolationKind};
    /// let mut c = Checker::new();
    /// c.survives("row", 1_i64, 0_i64);
    /// c.baseline("schema", true, false);
    /// assert_eq!(c.finish()[0].kind, ViolationKind::Baseline);
    /// ```
    #[must_use]
    pub fn finish(mut self) -> Vec<Violation> {
        self.baseline.append(&mut self.rest);
        self.baseline
    }
}

/// Lets [`Checker::survives`] tell "absent" apart from "different".
///
/// The blanket rule is that only `None` counts as absent; implement it for
/// your own type if it has its own notion of emptiness.
///
/// ```
/// # use powercut::IsAbsent;
/// assert!(None::<i64>.is_absent());
/// assert!(!Some(0_i64).is_absent());
/// assert!(!7_i64.is_absent());
/// ```
pub trait IsAbsent {
    /// `true` when this value represents "nothing came back".
    fn is_absent(&self) -> bool;
}

impl<T> IsAbsent for Option<T> {
    fn is_absent(&self) -> bool {
        self.is_none()
    }
}

impl<T> IsAbsent for Vec<T> {
    fn is_absent(&self) -> bool {
        self.is_empty()
    }
}

macro_rules! not_absent {
    ($($t:ty),*) => {$(
        impl IsAbsent for $t {
            fn is_absent(&self) -> bool { false }
        }
    )*};
}
not_absent!(i8, i16, i32, i64, i128, isize, u8, u16, u32, u64, u128, usize, bool, String, &str);

/// The four-valued outcome. Two of these are not "pass" and not "fail".
///
/// ```
/// # use powercut::Verdict;
/// assert!(!Verdict::NotMeasured.is_pass());
/// assert!(Verdict::NotMeasured.severity() < Verdict::Lost.severity());
/// ```
#[non_exhaustive]
#[derive(Copy, Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub enum Verdict {
    /// Everything the ack promised came back from the durable set.
    Survived,
    /// The check could not be performed soundly. **Not a pass.**
    NotMeasured,
    /// The durable set was missing or diverged from what was acked.
    Lost,
    /// The harness itself is broken, so the red proves nothing about the store.
    HarnessBroken,
}

impl Verdict {
    /// A short, stable label. Baseline files contain these.
    ///
    /// ```
    /// # use powercut::Verdict;
    /// assert_eq!(Verdict::HarnessBroken.as_str(), "harness-broken");
    /// ```
    #[must_use]
    pub const fn as_str(self) -> &'static str {
        match self {
            Verdict::Survived => "survived",
            Verdict::NotMeasured => "not-measured",
            Verdict::Lost => "lost",
            Verdict::HarnessBroken => "harness-broken",
        }
    }

    /// Parse a label written by [`Self::as_str`].
    ///
    /// ```
    /// # use powercut::Verdict;
    /// assert_eq!(Verdict::parse("lost"), Some(Verdict::Lost));
    /// assert_eq!(Verdict::parse("green"), None);
    /// ```
    #[must_use]
    pub fn parse(s: &str) -> Option<Self> {
        match s {
            "survived" => Some(Verdict::Survived),
            "not-measured" => Some(Verdict::NotMeasured),
            "lost" => Some(Verdict::Lost),
            "harness-broken" => Some(Verdict::HarnessBroken),
            _ => None,
        }
    }

    /// Only [`Verdict::Survived`] is a pass. In particular
    /// [`Verdict::NotMeasured`] is not.
    ///
    /// ```
    /// # use powercut::Verdict;
    /// assert!(Verdict::Survived.is_pass());
    /// assert!(!Verdict::NotMeasured.is_pass());
    /// ```
    #[must_use]
    pub fn is_pass(self) -> bool {
        matches!(self, Verdict::Survived)
    }

    /// Ranking used by the CI ratchet: lower is better.
    ///
    /// ```
    /// # use powercut::Verdict;
    /// assert!(Verdict::Survived.severity() < Verdict::HarnessBroken.severity());
    /// ```
    #[must_use]
    pub const fn severity(self) -> u8 {
        match self {
            Verdict::Survived => 0,
            Verdict::NotMeasured => 1,
            Verdict::Lost => 2,
            Verdict::HarnessBroken => 3,
        }
    }
}

impl fmt::Display for Verdict {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.write_str(self.as_str())
    }
}

/// The outcome with its evidence attached.
///
/// ```
/// # use powercut::{Finding, Verdict};
/// let f = Finding::NotMeasured { why: "the engine syncs where we cannot see".into() };
/// assert_eq!(f.verdict(), Verdict::NotMeasured);
/// ```
#[non_exhaustive]
#[derive(Clone, Debug)]
pub enum Finding {
    /// The durable set honoured the ack.
    Survived,
    /// The durable set did not honour the ack, and coverage was complete
    /// enough for that to mean something.
    Lost {
        /// What was missing or different.
        violations: Vec<Violation>,
    },
    /// **Never a silent pass, never a false accusation.** The check ran but
    /// its result is not sound — most often because the witness admitted it
    /// cannot see all of the store's sync activity, so an unobserved sync
    /// might have saved the data.
    NotMeasured {
        /// Why this could not be measured, in one sentence.
        why: String,
    },
    /// The harness is broken: the captured copy was not a usable store, or the
    /// teeth control — the run that *must* pass — did not.
    HarnessBroken {
        /// What is wrong with the harness.
        why: String,
        /// The violations that proved it.
        violations: Vec<Violation>,
    },
}

impl Finding {
    /// The matching [`Verdict`].
    ///
    /// ```
    /// # use powercut::{Finding, Verdict};
    /// assert_eq!(Finding::Survived.verdict(), Verdict::Survived);
    /// ```
    #[must_use]
    pub fn verdict(&self) -> Verdict {
        match self {
            Finding::Survived => Verdict::Survived,
            Finding::Lost { .. } => Verdict::Lost,
            Finding::NotMeasured { .. } => Verdict::NotMeasured,
            Finding::HarnessBroken { .. } => Verdict::HarnessBroken,
        }
    }

    /// The violations behind this finding, if any.
    ///
    /// ```
    /// # use powercut::Finding;
    /// assert!(Finding::Survived.violations().is_empty());
    /// ```
    #[must_use]
    pub fn violations(&self) -> &[Violation] {
        match self {
            Finding::Lost { violations } | Finding::HarnessBroken { violations, .. } => violations,
            Finding::Survived | Finding::NotMeasured { .. } => &[],
        }
    }
}

/// The result of one case: one store, one tier, one checker.
///
/// A `Report` is written to be **self-adjudicating**. Every failure prints the
/// full manifest, every warrant, the notes the capture made, and — when the
/// tier is [`Tier::PowerCut`] — whether the data was present in the tier-1
/// byte copy taken at the same instant. That last line is usually the answer:
/// *present at tier 1, absent at tier 2* means the write reached the
/// filesystem and no further, which is a missing fsync, not a lost write.
///
/// ```
/// use powercut::{Checker, ScratchDir, Scenario, SyncJournal, Tier, Verdict};
///
/// let store = ScratchDir::new("doc-report").unwrap();
/// let journal = SyncJournal::complete();
///
/// let report = Scenario::new(store.path(), journal.clone())
///     .label("no-fsync-at-all")
///     .tier(Tier::PowerCut)
///     .ack(|| std::fs::write(store.path().join("app.log"), b"acked").unwrap())
///     .unwrap()
///     .check(|crash, ()| {
///         let mut c = Checker::new();
///         let survived = std::fs::read(crash.dir().join("app.log")).ok();
///         c.survives("app.log", Some(b"acked".to_vec()), survived);
///         c.finish()
///     });
///
/// assert_eq!(report.verdict(), Verdict::Lost);
/// assert!(report.reached_filesystem() == Some(true));
/// ```
#[derive(Debug)]
pub struct Report {
    pub(crate) label: String,
    pub(crate) tier: Tier,
    pub(crate) finding: Finding,
    pub(crate) coverage: Coverage,
    pub(crate) manifest: Vec<CapturedFile>,
    pub(crate) manifest_text: String,
    pub(crate) notes: Vec<String>,
    pub(crate) reference_violations: Option<Vec<Violation>>,
    pub(crate) control_violations: Option<Vec<Violation>>,
    pub(crate) used_assumption: bool,
}

impl Report {
    /// The case name. Baseline files are keyed on it, so keep it stable.
    #[must_use]
    pub fn label(&self) -> &str {
        &self.label
    }

    /// Which tier was measured.
    #[must_use]
    pub fn tier(&self) -> Tier {
        self.tier
    }

    /// The outcome with its evidence.
    #[must_use]
    pub fn finding(&self) -> &Finding {
        &self.finding
    }

    /// The outcome as a plain, matchable value.
    #[must_use]
    pub fn verdict(&self) -> Verdict {
        self.finding.verdict()
    }

    /// What the witness admitted it could see.
    #[must_use]
    pub fn coverage(&self) -> &Coverage {
        &self.coverage
    }

    /// The surviving files and their warrants.
    #[must_use]
    pub fn manifest(&self) -> &[CapturedFile] {
        &self.manifest
    }

    /// For a [`Tier::PowerCut`] run: did the same check pass against the
    /// tier-1 byte copy taken at the same instant?
    ///
    /// `Some(true)` with a `Lost` verdict is the classic missing-fsync
    /// signature. `Some(false)` means the data was not even on the filesystem,
    /// which is a different bug — or a wrong store path. `None` means the run
    /// was already tier 1.
    #[must_use]
    pub fn reached_filesystem(&self) -> Option<bool> {
        self.reference_violations.as_ref().map(|v| v.is_empty())
    }

    /// Everything worth reading, as text. This is what the assertions print.
    #[must_use]
    pub fn to_text(&self) -> String {
        let mut out = String::new();
        out.push_str(&format!(
            "powercut: {} [tier {}] => {}\n",
            self.label,
            self.tier.as_str(),
            self.verdict()
        ));
        out.push_str(&format!("  tier models {}\n", self.tier.models()));
        match &self.coverage {
            Coverage::Complete => {
                out.push_str("  witness coverage: COMPLETE\n");
            }
            Coverage::Partial { unobserved } => {
                out.push_str(&format!("  witness coverage: PARTIAL — {unobserved}\n"));
            }
        }
        if self.used_assumption {
            out.push_str(
                "  !! this run used an ASSUMED-durable file. Assumptions only ever ENLARGE \
                 the durable set, so a pass here is weaker than a measured one.\n",
            );
        }
        // Only worth saying when something went wrong: on a pass the tier-1
        // copy always holds the data too, so the line would be pure noise.
        match self
            .reached_filesystem()
            .filter(|_| !self.verdict().is_pass())
        {
            Some(true) => out.push_str(
                "  asymmetry: the acked data IS present in the tier-1 byte copy taken at the \
                 same instant. It reached the filesystem and no further — the missing thing \
                 is an fsync, not a write.\n",
            ),
            Some(false) => out.push_str(
                "  asymmetry: the acked data is ALSO absent from the tier-1 byte copy, which \
                 keeps every byte that reached the filesystem. So the write never reached the \
                 filesystem at all — it is still in the engine's user-space buffers — or the \
                 store paths / checker are wrong. A teeth control tells those apart.\n",
            ),
            None => {}
        }
        if let Some(control) = &self.control_violations {
            if control.is_empty() {
                out.push_str(
                    "  teeth control: PASSED — with an explicit flush the same copy and the \
                     same checker do find the data, so the red above is about durability.\n",
                );
            } else {
                out.push_str("  teeth control: FAILED — see below.\n");
            }
        } else if self.verdict() == Verdict::Lost {
            out.push_str(
                "  teeth control: NOT RUN. Without one, a red cannot be distinguished from a \
                 broken harness. Add `.with_teeth_control(...)`.\n",
            );
        }
        out.push_str(&self.manifest_text);
        for note in &self.notes {
            out.push_str(&format!("  note: {note}\n"));
        }
        let violations = self.finding.violations();
        if !violations.is_empty() {
            out.push_str("  findings:\n");
            for v in violations {
                out.push_str(&format!("    {v}\n"));
            }
        }
        if let Finding::NotMeasured { why } = &self.finding {
            out.push_str(&format!("  NOT MEASURED: {why}\n"));
        }
        out
    }

    /// One markdown table row: `| label | tier | verdict | detail |`.
    #[must_use]
    pub fn to_markdown_row(&self) -> String {
        let detail = match &self.finding {
            Finding::Survived => "—".to_string(),
            Finding::NotMeasured { why } => escape(why),
            Finding::Lost { violations } | Finding::HarnessBroken { violations, .. } => violations
                .iter()
                .map(|v| escape(&v.to_string()))
                .collect::<Vec<_>>()
                .join("; "),
        };
        format!(
            "| `{}` | {} | **{}** | {} |",
            self.label,
            self.tier.as_str(),
            self.verdict(),
            detail
        )
    }

    /// Panic unless the verdict is [`Verdict::Survived`], printing the whole
    /// self-adjudicating report.
    ///
    /// # Panics
    ///
    /// On any verdict other than `Survived` — including `NotMeasured`, which
    /// is the point: a check that could not run is not a pass.
    pub fn assert_survived(&self) {
        assert!(
            self.verdict().is_pass(),
            "\n{}\n{}",
            self.to_text(),
            self.control_detail()
        );
    }

    fn control_detail(&self) -> String {
        match &self.control_violations {
            Some(v) if !v.is_empty() => {
                let mut s = String::from(
                    "  the teeth control FAILED. With everything explicitly flushed the \
                     checker still could not find the data, so this run measured the harness, \
                     not the store:\n",
                );
                for violation in v {
                    s.push_str(&format!("    {violation}\n"));
                }
                s
            }
            _ => String::new(),
        }
    }
}

pub(crate) fn escape(s: &str) -> String {
    s.replace('|', "\\|").replace('\n', " ")
}