supercov_engine/

coverage_index.rs

//! Typed coverage columns stored in the immutable query-index container.
//!
//! This is not a serialized report. Records contain fixed-width values and
//! checked references into an interned UTF-8 string table. New query surfaces
//! add sections without forcing existing readers to parse unrelated data.

use std::collections::{BTreeMap, BTreeSet, HashMap};

use serde::Serialize;
use supercov_contracts::COVERAGE_MODEL_SCHEMA_VERSION;

use crate::{
    coverage_analysis::{
        CoverageCount, CoverageSummary, McdcVector, find_witnesses_for_conditions,
    },
    coverage_report::{
        CoverageModel, CoverageReport, CoverageView, TransportStats, coverage_summary_for_tests,
    },
    interned::Id,
    query_index::{QueryIndex, QueryIndexError, QueryIndexSection},
};

pub const SECTION_STRING_BYTES: u32 = 1;
pub const SECTION_STRINGS: u32 = 2;
pub const SECTION_STRING_RELATIONS: u32 = 3;
pub const SECTION_VIEW_SUMMARIES: u32 = 10;
pub const SECTION_FILE_GAPS: u32 = 11;
pub const SECTION_DECISION_GAPS: u32 = 12;
pub const SECTION_DIMENSIONS: u32 = 13;
pub const SECTION_PROJECTIONS: u32 = 14;
pub const SECTION_SCOPE_ENTRIES: u32 = 15;
pub const SECTION_CONFIDENCE: u32 = 16;
pub const SECTION_LINES: u32 = 17;
pub const SECTION_TEST_SUMMARIES: u32 = 18;
pub const SECTION_PHASE_SUMMARIES: u32 = 19;
pub const SECTION_ANCHORS: u32 = 20;
pub const SECTION_TEST_RETRIES: u32 = 21;
pub const SECTION_TEST_ATTEMPTS: u32 = 22;
pub const SECTION_TEST_LINES: u32 = 23;
pub const SECTION_TEST_HITS: u32 = 24;
pub const SECTION_TEST_DECISIONS: u32 = 25;
pub const SECTION_TEST_VECTORS: u32 = 26;
pub const SECTION_VECTOR_VALUES: u32 = 27;
pub const SECTION_HIT_METADATA: u32 = 28;
pub const SECTION_DECISION_METADATA: u32 = 29;
pub const SECTION_DECISION_DETAILS: u32 = 30;
pub const SECTION_DECISION_VECTOR_OBSERVATIONS: u32 = 31;
pub const SECTION_DECISION_CONDITIONS: u32 = 32;
pub const SECTION_LIMITATIONS: u32 = 33;
pub const SECTION_COVERAGE_MODEL: u32 = 34;

const STRING_RECORD_SIZE: usize = 16;
const SUMMARY_RECORD_SIZE: usize = 176;
const FILE_GAP_RECORD_SIZE: usize = 200;
const DECISION_GAP_RECORD_SIZE: usize = 96;
const DIMENSION_RECORD_SIZE: usize = 192;
const PROJECTION_RECORD_SIZE: usize = 544;
const SCOPE_ENTRY_RECORD_SIZE: usize = 96;
const CONFIDENCE_RECORD_SIZE: usize = 96;
const LINE_RECORD_SIZE: usize = 80;
const TEST_SUMMARY_RECORD_SIZE: usize = 64;
const PHASE_SUMMARY_RECORD_SIZE: usize = 64;
const ANCHOR_RECORD_SIZE: usize = 64;
const TEST_RETRY_RECORD_SIZE: usize = 16;
const TEST_ATTEMPT_RECORD_SIZE: usize = 24;
const TEST_LINE_RECORD_SIZE: usize = 24;
const TEST_HIT_RECORD_SIZE: usize = 16;
const TEST_DECISION_RECORD_SIZE: usize = 32;
const TEST_VECTOR_RECORD_SIZE: usize = 24;
const HIT_METADATA_RECORD_SIZE: usize = 64;
const DECISION_METADATA_RECORD_SIZE: usize = 64;
const DECISION_DETAIL_RECORD_SIZE: usize = 64;
const DECISION_VECTOR_OBSERVATION_RECORD_SIZE: usize = 64;
const DECISION_CONDITION_RECORD_SIZE: usize = 64;
const LIMITATION_RECORD_SIZE: usize = 64;
const COVERAGE_MODEL_RECORD_SIZE: usize = 48;
const NO_STRING: u32 = u32::MAX;

#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize)]
#[serde(rename_all = "lowercase")]
pub enum CoverageViewId {
    All = 0,
    Passed = 1,
    Failed = 2,
}

impl TryFrom<u8> for CoverageViewId {
    type Error = CoverageIndexError;

    fn try_from(value: u8) -> Result<Self, Self::Error> {
        match value {
            0 => Ok(Self::All),
            1 => Ok(Self::Passed),
            2 => Ok(Self::Failed),
            _ => Err(CoverageIndexError::InvalidRecord("coverage view")),
        }
    }
}

#[derive(Debug)]
pub enum CoverageIndexError {
    Container(QueryIndexError),
    InvalidRecord(&'static str),
    InvalidUtf8,
    SizeOverflow,
}

impl From<QueryIndexError> for CoverageIndexError {
    fn from(value: QueryIndexError) -> Self {
        Self::Container(value)
    }
}

impl std::fmt::Display for CoverageIndexError {
    fn fmt(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        match self {
            Self::Container(error) => write!(formatter, "{error}"),
            Self::InvalidRecord(reason) => write!(formatter, "invalid coverage index: {reason}"),
            Self::InvalidUtf8 => write!(formatter, "invalid UTF-8 in coverage index"),
            Self::SizeOverflow => write!(formatter, "coverage index exceeds format limits"),
        }
    }
}

impl std::error::Error for CoverageIndexError {}

fn put_u32(bytes: &mut [u8], offset: usize, value: u32) {
    bytes[offset..offset + 4].copy_from_slice(&value.to_le_bytes());
}

fn put_u64(bytes: &mut [u8], offset: usize, value: u64) {
    bytes[offset..offset + 8].copy_from_slice(&value.to_le_bytes());
}

/// A field of a record. Every record the reader takes is exactly its kind's
/// size -- [`CoverageIndex::new`] refuses an index whose sizes differ -- and
/// every field lies inside it, so a read cannot run past the record.
fn get_u32(record: &[u8], offset: usize) -> u32 {
    u32::from_le_bytes(
        record[offset..offset + 4]
            .try_into()
            .expect("a field lies inside its record"),
    )
}

fn get_u64(record: &[u8], offset: usize) -> u64 {
    u64::from_le_bytes(
        record[offset..offset + 8]
            .try_into()
            .expect("a field lies inside its record"),
    )
}

fn usize_u64(value: usize) -> Result<u64, CoverageIndexError> {
    u64::try_from(value).map_err(|_| CoverageIndexError::SizeOverflow)
}

fn usize_u32(value: usize) -> Result<u32, CoverageIndexError> {
    u32::try_from(value).map_err(|_| CoverageIndexError::SizeOverflow)
}

#[derive(Default)]
struct StringTable {
    ids: crate::interned::FastMap<String, u32>,
    strings: Vec<String>,
    // Keep each allocation alive: its address can then identify repeated Id
    // references without hashing the same long test name for every line.
    shared_ids: crate::interned::FastMap<usize, (Id, u32)>,
}

#[derive(Default)]
struct StringRelations {
    values: Vec<u32>,
    /// Where each distinct run already sits. Readers reach a run only through
    /// the (offset, count) pair the record stores, so two records naming the
    /// same run can share one copy of it and nothing downstream can tell.
    interned: crate::interned::FastMap<Vec<u32>, (u64, u64)>,
}

impl StringRelations {
    /// Append a run of interned strings, or point at an identical one already
    /// stored.
    ///
    /// The same run is named over and over: a filtered view repeats the whole
    /// relation of the view it filters, and one list of tests covers many
    /// lines of the file they exercise. Appending each time made this the
    /// largest section in the container -- 61 MB of a 210 MB index, 15.3
    /// million entries -- for a number of distinct runs far smaller.
    fn push(
        &mut self,
        values: impl IntoIterator<Item = impl AsRef<str>>,
        strings: &mut StringTable,
    ) -> Result<(u64, u64), CoverageIndexError> {
        let mut run = Vec::new();
        for value in values {
            run.push(strings.intern(value.as_ref())?);
        }
        self.push_run(run)
    }

    fn push_ids(
        &mut self,
        values: &[Id],
        strings: &mut StringTable,
    ) -> Result<(u64, u64), CoverageIndexError> {
        let mut run = Vec::with_capacity(values.len());
        for value in values {
            run.push(strings.intern_shared(value)?);
        }
        self.push_run(run)
    }

    fn push_run(&mut self, run: Vec<u32>) -> Result<(u64, u64), CoverageIndexError> {
        if let Some(found) = self.interned.get(&run) {
            return Ok(*found);
        }
        let offset = usize_u64(self.values.len())?;
        let count = usize_u64(run.len())?;
        self.values.extend_from_slice(&run);
        self.interned.insert(run, (offset, count));
        Ok((offset, count))
    }

    fn section(self) -> Result<QueryIndexSection, CoverageIndexError> {
        let mut bytes = Vec::with_capacity(self.values.len() * 4);
        for value in self.values {
            bytes.extend_from_slice(&value.to_le_bytes());
        }
        Ok(QueryIndexSection {
            kind: SECTION_STRING_RELATIONS,
            record_size: 4,
            count: usize_u64(bytes.len() / 4)?,
            bytes,
        })
    }
}

impl StringTable {
    fn intern_shared(&mut self, value: &Id) -> Result<u32, CoverageIndexError> {
        let allocation = value.as_str().as_ptr() as usize;
        if let Some((_, id)) = self.shared_ids.get(&allocation) {
            return Ok(*id);
        }
        // Different allocations of identical text still use the same string
        // table entry, including IDs built outside the report's interner.
        let id = self.intern(value.as_str())?;
        self.shared_ids.insert(allocation, (value.clone(), id));
        Ok(id)
    }

    fn intern(&mut self, value: &str) -> Result<u32, CoverageIndexError> {
        if let Some(id) = self.ids.get(value) {
            return Ok(*id);
        }
        let id = usize_u32(self.strings.len())?;
        self.ids.insert(value.into(), id);
        self.strings.push(value.into());
        Ok(id)
    }

    fn sections(self) -> Result<[QueryIndexSection; 2], CoverageIndexError> {
        let mut blob = Vec::new();
        let mut records = Vec::with_capacity(self.strings.len() * STRING_RECORD_SIZE);
        for string in self.strings {
            let offset = usize_u64(blob.len())?;
            let value = string.as_bytes();
            let length = usize_u32(value.len())?;
            blob.extend_from_slice(value);
            let mut record = [0_u8; STRING_RECORD_SIZE];
            put_u64(&mut record, 0, offset);
            put_u32(&mut record, 8, length);
            records.extend_from_slice(&record);
        }
        Ok([
            QueryIndexSection {
                kind: SECTION_STRING_BYTES,
                record_size: 0,
                count: usize_u64(blob.len())?,
                bytes: blob,
            },
            QueryIndexSection {
                kind: SECTION_STRINGS,
                record_size: STRING_RECORD_SIZE as u32,
                count: usize_u64(records.len() / STRING_RECORD_SIZE)?,
                bytes: records,
            },
        ])
    }
}

fn put_count(
    bytes: &mut [u8],
    offset: usize,
    count: &CoverageCount,
) -> Result<(), CoverageIndexError> {
    put_u64(bytes, offset, usize_u64(count.covered)?);
    put_u64(bytes, offset + 8, usize_u64(count.total)?);
    Ok(())
}

fn summary_record(
    id: CoverageViewId,
    view: &CoverageView,
    strings: &mut StringTable,
) -> Result<[u8; SUMMARY_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; SUMMARY_RECORD_SIZE];
    record[0] = id as u8;
    record[1] = u8::from(view.summary.coverage_complete);
    record[2] = match view.summary.completeness_blocked {
        None => 0,
        Some(false) => 1,
        Some(true) => 2,
    };
    put_u32(&mut record, 4, strings.intern(&view.generated_at)?);
    put_u32(&mut record, 8, strings.intern(&view.variant)?);
    let values = [
        view.summary.decisions,
        view.summary.executed_decisions,
        view.summary.covered_decisions,
        view.summary.conditions,
        view.summary.covered_conditions,
    ];
    for (index, value) in values.into_iter().enumerate() {
        put_u64(&mut record, 16 + index * 8, usize_u64(value)?);
    }
    for (index, count) in [
        &view.summary.lines,
        &view.summary.statements,
        &view.summary.functions,
        &view.summary.branches,
        &view.summary.decision_outcomes,
        &view.summary.condition_outcomes,
        &view.summary.value_selections,
    ]
    .into_iter()
    .enumerate()
    {
        put_count(&mut record, 56 + index * 16, count)?;
    }
    Ok(record)
}

#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct IndexedFileGap {
    #[serde(skip)]
    pub view: CoverageViewId,
    pub file: String,
    pub uncovered_lines: usize,
    pub uncovered_statements: usize,
    pub uncovered_functions: usize,
    pub missing_branches: usize,
    pub missing_mcdc_conditions: usize,
    pub measurement_limitations: usize,
    pub limitation_kinds: Vec<String>,
    pub covered_by_other_tests: IndexedGapDimensions,
    pub uncovered_everywhere: IndexedGapDimensions,
    /// What the file has to cover, and how much of it the selected tests do.
    /// The listing gave only what was missing, so no percentage could be
    /// worked out for a file, or summed for a directory.
    pub totals: IndexedGapDimensions,
    pub covered: IndexedGapDimensions,
    pub score: usize,
}

#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct IndexedGapDimensions {
    pub lines: usize,
    pub statements: usize,
    pub functions: usize,
    pub branches: usize,
    pub mcdc_conditions: usize,
}

#[derive(Debug, Clone, PartialEq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct IndexedCoverageSnapshot {
    pub all_summary: CoverageSummary,
    pub passed_summary: CoverageSummary,
    pub failed_summary: CoverageSummary,
    pub all_files: Vec<IndexedFileGap>,
    pub passed_files: Vec<IndexedFileGap>,
    pub failed_files: Vec<IndexedFileGap>,
}

#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct IndexedDecisionGap {
    #[serde(skip)]
    pub view: CoverageViewId,
    #[serde(skip)]
    pub file: String,
    pub id: String,
    pub line: usize,
    pub column: usize,
    pub kind: String,
    pub conditions: usize,
    pub missing_conditions: usize,
    pub source: String,
}

#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum CoverageDimension {
    Kind = 0,
    Runner = 1,
    /// Every test together: one unnamed entry, what the tests cover between
    /// them. The run's total less this is what ran outside any test.
    Tests = 2,
}

#[derive(Debug, Clone, PartialEq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct IndexedDimensionCoverage {
    #[serde(skip_serializing_if = "Option::is_none")]
    pub kind: Option<String>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub runner: Option<String>,
    pub tests: usize,
    pub setups: usize,
    /// How many of `tests` the summary below actually describes. Where this is
    /// zero and `tests` is not, the dimension ran tests whose coverage nothing
    /// can narrow, and its summary describes none of them.
    pub attributed: usize,
    pub summary: CoverageSummary,
}

#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct IndexedAttribution {
    pub browser_explicit: usize,
    pub browser_fallback: usize,
    pub server_explicit: usize,
    pub server_fallback: usize,
}

#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct IndexedOutcomeCounts {
    pub passed: usize,
    pub failed: usize,
    pub flaky: usize,
    pub skipped: usize,
    pub timed_out: usize,
    pub interrupted: usize,
    pub unknown: usize,
    pub unstarted: usize,
    /// node:test's todo tests, which the runner counts on their own: neither
    /// passed nor failed, whatever their body did.
    pub todo: usize,
}

#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct IndexedMeasurementKinds {
    #[serde(rename = "dynamic-code")]
    pub dynamic_code: usize,
    #[serde(rename = "semantic-safety")]
    pub semantic_safety: usize,
    #[serde(rename = "source-scope")]
    pub source_scope: usize,
}

#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct IndexedMeasurement {
    pub complete: bool,
    pub limitations: usize,
    pub evidence_corruptions: usize,
    pub blocking: usize,
    /// Limitations that declare a boundary of the denominator rather than
    /// blocking measurement inside it.
    pub declared: usize,
    pub files: usize,
    pub by_kind: IndexedMeasurementKinds,
}

#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct IndexedConfidenceLines {
    pub unexecuted: usize,
    pub executed: usize,
    pub action: usize,
    pub asserted: usize,
}

#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct IndexedSummaryConfidence {
    pub lines: IndexedConfidenceLines,
    pub assertion_covered_mcdc_conditions: usize,
}

#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct IndexedCoverageModel {
    pub schema_version: u32,
    pub variant: String,
    pub name: String,
    pub completeness_meaning: String,
    pub measured: Vec<String>,
    pub not_measured: Vec<String>,
}

#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct IndexedSourceScope {
    pub kind: String,
    pub language: String,
    pub model: String,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub mode: Option<String>,
    pub roots: Vec<String>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub unit: Option<String>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub measurement_complete: Option<bool>,
    pub included: usize,
    pub excluded: usize,
    pub ambiguous: usize,
}

#[derive(Debug, Clone, PartialEq)]
pub struct IndexedProjection {
    pub view: CoverageViewId,
    pub kind: Option<String>,
    pub runner: Option<String>,
    pub generated_at: String,
    pub summary: CoverageSummary,
    pub measurement: IndexedMeasurement,
    pub attribution: IndexedAttribution,
    pub transport: Option<TransportStats>,
    pub empty_evidence_tests: usize,
    pub first_empty_evidence_test: Option<String>,
    pub confidence: IndexedSummaryConfidence,
    pub files_with_gaps: usize,
    pub files_with_coverage_gaps: usize,
    pub tests: usize,
    pub setups: usize,
    pub test_outcomes: IndexedOutcomeCounts,
    pub source_scope: Option<IndexedSourceScope>,
}

#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct IndexedScopeEntry {
    pub file: String,
    pub status: String,
    pub reason: String,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub package_root: Option<String>,
    pub measurement_limitations: usize,
    pub limitation_kinds: Vec<String>,
}

#[derive(Debug, Clone, PartialEq)]
pub struct IndexedLine {
    pub file: String,
    pub line: usize,
    pub covered: bool,
    /// False when every obligation on the line was declined: the line stays
    /// addressable and carries its limitation, but it is neither covered nor
    /// uncovered.
    pub measured: bool,
    pub tests: Vec<String>,
    pub phases: Vec<String>,
    pub confidence: crate::coverage_report::CoverageConfidence,
}

#[derive(Debug, Clone, PartialEq)]
pub struct IndexedTestSummary {
    pub id: String,
    pub name: String,
    pub file: Option<String>,
    pub title: Option<String>,
    pub outcome: String,
    pub role: String,
    /// `exact`, or `run-wide` when this test ran and nothing can say what it
    /// reached. Read before any statement about this test's coverage: an
    /// empty hit set means "reached nothing" only under `exact`.
    pub attribution: String,
    pub provenance: crate::coverage_report::TestProvenance,
}

#[derive(Debug, Clone, PartialEq)]
pub struct IndexedPhaseSummary {
    pub id: String,
    pub kind: String,
    pub operation: String,
    pub source: Option<String>,
    pub test: String,
    pub status: Option<String>,
    pub caused_by_phase_id: Option<String>,
    pub lines: usize,
    pub decisions: usize,
}

#[derive(Debug, Clone, PartialEq)]
pub struct IndexedAnchor {
    pub kind: String,
    pub id: String,
    pub file: String,
    pub line: usize,
    pub column: usize,
    pub covered: bool,
    pub conditions: Option<usize>,
    pub covered_conditions: Option<usize>,
    pub tests: Vec<String>,
}

#[derive(Debug, Clone, PartialEq)]
pub struct IndexedTestDetail {
    pub summary: IndexedTestSummary,
    pub retries: Vec<usize>,
    pub attempts: Vec<crate::coverage_report::TestAttempt>,
    pub hits: Vec<String>,
    pub decisions: Vec<crate::coverage_report::TestDecisionResult>,
    pub lines: Vec<crate::coverage_report::SourceLine>,
}

#[derive(Debug, Clone, PartialEq, Eq)]
pub struct IndexedHitMetadata {
    pub id: String,
    pub obligation: String,
    pub branch_kind: Option<String>,
    pub file: String,
    pub line: usize,
    pub column: usize,
    pub label: Option<String>,
    pub alternative: Option<String>,
    pub parent_id: Option<String>,
    pub source: String,
    pub tests: Vec<String>,
}

#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
pub struct IndexedLimitation {
    pub id: String,
    pub kind: String,
    pub file: String,
    pub line: usize,
    pub column: usize,
    pub source: String,
    pub reason: String,
    /// Whether this blocks measurement of the denominator, as opposed to
    /// declaring a boundary of it.
    pub blocking: bool,
}

#[derive(Default)]
struct MutableFileGap {
    uncovered_lines: usize,
    uncovered_statements: usize,
    uncovered_functions: usize,
    missing_branches: usize,
    missing_mcdc_conditions: usize,
    measurement_limitations: usize,
    limitation_mask: u32,
    covered_by_other_tests: [usize; 5],
    uncovered_everywhere: [usize; 5],
    totals: [usize; 5],
}

fn limitation_kind(value: &serde_json::Value) -> Option<(&str, &str)> {
    Some((value.get("file")?.as_str()?, value.get("kind")?.as_str()?))
}

fn includes_selected<T: AsRef<str>>(
    tests: &[T],
    selected: Option<&Selected>,
    covered: bool,
) -> bool {
    selected.map_or(covered, |selected| {
        tests.iter().any(|test| selected.contains(test.as_ref()))
    })
}

fn classify(
    gap: &mut MutableFileGap,
    dimension: usize,
    selected: Option<&Selected>,
    covered_overall: bool,
) {
    if selected.is_some() && covered_overall {
        gap.covered_by_other_tests[dimension] += 1;
    } else {
        gap.uncovered_everywhere[dimension] += 1;
    }
}

/// A file's gap, its name copied only the first time the file is met.
fn gap_for<'m>(
    files: &'m mut BTreeMap<String, MutableFileGap>,
    file: &str,
) -> &'m mut MutableFileGap {
    if !files.contains_key(file) {
        files.insert(file.to_owned(), MutableFileGap::default());
    }
    files.get_mut(file).expect("gap was inserted")
}

fn file_gaps(
    view: &CoverageView,
    selected: Option<&Selected>,
) -> Result<Vec<(String, MutableFileGap)>, CoverageIndexError> {
    let mut files = BTreeMap::<String, MutableFileGap>::new();
    for line in &view.lines {
        let gap = gap_for(&mut files, &line.file);
        gap.totals[0] += usize::from(line.measured);
        if line.measured && !includes_selected(&line.tests, selected, line.covered) {
            gap.uncovered_lines += 1;
            classify(gap, 0, selected, line.covered);
        }
    }
    for point in &view.points {
        let gap = gap_for(&mut files, &point.meta.file);
        if point.measured {
            gap.totals[match point.meta.kind {
                crate::coverage_analysis::PointKind::Statement => 1,
                crate::coverage_analysis::PointKind::Function => 2,
            }] += 1;
        }
        if point.measured && !includes_selected(&point.tests, selected, point.covered) {
            match point.meta.kind {
                crate::coverage_analysis::PointKind::Statement => {
                    gap.uncovered_statements += 1;
                    classify(gap, 1, selected, point.covered);
                }
                crate::coverage_analysis::PointKind::Function => {
                    gap.uncovered_functions += 1;
                    classify(gap, 2, selected, point.covered);
                }
            }
        }
    }
    for branch in &view.branches {
        let gap = gap_for(&mut files, &branch.meta.file);
        gap.totals[3] += branch.alternatives.len();
        for alternative in &branch.alternatives {
            if !includes_selected(&alternative.tests, selected, alternative.covered) {
                gap.missing_branches += 1;
                classify(gap, 3, selected, alternative.covered);
            }
        }
    }
    for decision in &view.decisions {
        let gap = gap_for(&mut files, &decision.meta.file);
        gap.totals[4] += decision.meta.conditions.len();
        let selected_vectors = decision
            .vector_observations
            .iter()
            .filter(|observation| includes_selected(&observation.tests, selected, true))
            .map(|observation| observation.vector.clone())
            .collect::<Vec<_>>();
        let witnesses =
            find_witnesses_for_conditions(&selected_vectors, decision.meta.conditions.len())
                .map_err(|_| CoverageIndexError::InvalidRecord("MC/DC vector width"))?;
        for (index, witness) in witnesses.into_iter().enumerate() {
            if witness.is_none() {
                gap.missing_mcdc_conditions += 1;
                classify(gap, 4, selected, decision.conditions[index].covered);
            }
        }
    }
    for limitation in &view.limitations {
        let Some((file, kind)) = limitation_kind(limitation) else {
            continue;
        };
        let gap = files.entry(file.into()).or_default();
        gap.measurement_limitations += 1;
        gap.limitation_mask |= match kind {
            "dynamic-code" => 1,
            "semantic-safety" => 2,
            "source-scope" => 4,
            _ => 8,
        };
    }
    Ok(files.into_iter().collect())
}

fn file_gap_record(
    view_id: CoverageViewId,
    file: &str,
    gap: &MutableFileGap,
    kind: Option<&str>,
    runner: Option<&str>,
    strings: &mut StringTable,
) -> Result<[u8; FILE_GAP_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; FILE_GAP_RECORD_SIZE];
    record[0] = view_id as u8;
    put_u32(&mut record, 4, strings.intern(file)?);
    for (index, value) in [
        gap.uncovered_lines,
        gap.uncovered_statements,
        gap.uncovered_functions,
        gap.missing_branches,
        gap.missing_mcdc_conditions,
        gap.measurement_limitations,
    ]
    .into_iter()
    .enumerate()
    {
        put_u64(&mut record, 8 + index * 8, usize_u64(value)?);
    }
    put_u32(&mut record, 56, gap.limitation_mask);
    let score = gap.uncovered_lines
        + gap.uncovered_functions * 2
        + gap.missing_branches * 2
        + gap.missing_mcdc_conditions * 3
        + gap.measurement_limitations * 3;
    put_u64(&mut record, 64, usize_u64(score)?);
    put_u32(
        &mut record,
        72,
        kind.map_or(Ok(NO_STRING), |value| strings.intern(value))?,
    );
    put_u32(
        &mut record,
        76,
        runner.map_or(Ok(NO_STRING), |value| strings.intern(value))?,
    );
    for (index, value) in gap.covered_by_other_tests.into_iter().enumerate() {
        put_u64(&mut record, 80 + index * 8, usize_u64(value)?);
    }
    for (index, value) in gap.uncovered_everywhere.into_iter().enumerate() {
        put_u64(&mut record, 120 + index * 8, usize_u64(value)?);
    }
    for (index, value) in gap.totals.into_iter().enumerate() {
        put_u64(&mut record, 160 + index * 8, usize_u64(value)?);
    }
    Ok(record)
}

/// The tests a projection selects. Hashed: every line, point, branch and
/// vector of the view asks whether any of its tests is selected, once per
/// projection, and a tree of long test names answered each by comparing them.
type Selected = crate::interned::FastSet<String>;

fn projections(view: &CoverageView) -> Vec<(Option<String>, Option<String>, Selected)> {
    let kinds = view
        .tests
        .iter()
        .map(|test| test.provenance.kind.clone())
        .collect::<BTreeSet<_>>();
    let runners = view
        .tests
        .iter()
        .map(|test| test.provenance.runner.clone())
        .collect::<BTreeSet<_>>();
    let mut selectors = Vec::new();
    for kind in &kinds {
        selectors.push((Some(kind.clone()), None));
    }
    for runner in &runners {
        selectors.push((None, Some(runner.clone())));
    }
    for kind in &kinds {
        for runner in &runners {
            selectors.push((Some(kind.clone()), Some(runner.clone())));
        }
    }
    selectors
        .into_iter()
        .filter_map(|(kind, runner)| {
            let selected = view
                .tests
                .iter()
                .filter(|test| {
                    kind.as_ref()
                        .is_none_or(|value| test.provenance.kind == *value)
                        && runner
                            .as_ref()
                            .is_none_or(|value| test.provenance.runner == *value)
                })
                .map(|test| test.id.clone())
                .collect::<Selected>();
            (!selected.is_empty()).then_some((kind, runner, selected))
        })
        .collect()
}

fn decision_gap_record(
    view_id: CoverageViewId,
    decision: &crate::coverage_report::DecisionResult,
    selected: Option<&Selected>,
    kind: Option<&str>,
    runner: Option<&str>,
    strings: &mut StringTable,
) -> Result<[u8; DECISION_GAP_RECORD_SIZE], CoverageIndexError> {
    let vectors = decision
        .vector_observations
        .iter()
        .filter(|observation| includes_selected(&observation.tests, selected, true))
        .map(|observation| observation.vector.clone())
        .collect::<Vec<_>>();
    let witnesses = find_witnesses_for_conditions(&vectors, decision.meta.conditions.len())
        .map_err(|_| CoverageIndexError::InvalidRecord("MC/DC vector width"))?;
    let mut record = [0_u8; DECISION_GAP_RECORD_SIZE];
    record[0] = view_id as u8;
    put_u32(
        &mut record,
        4,
        kind.map_or(Ok(NO_STRING), |value| strings.intern(value))?,
    );
    put_u32(
        &mut record,
        8,
        runner.map_or(Ok(NO_STRING), |value| strings.intern(value))?,
    );
    put_u32(&mut record, 12, strings.intern(&decision.meta.id)?);
    put_u32(&mut record, 16, strings.intern(&decision.meta.file)?);
    put_u32(&mut record, 20, strings.intern(&decision.meta.kind)?);
    put_u32(
        &mut record,
        24,
        strings.intern(
            &decision
                .meta
                .source
                .split_whitespace()
                .collect::<Vec<_>>()
                .join(" "),
        )?,
    );
    put_u64(&mut record, 32, usize_u64(decision.meta.line)?);
    put_u64(&mut record, 40, usize_u64(decision.meta.column)?);
    put_u64(&mut record, 48, usize_u64(decision.meta.conditions.len())?);
    put_u64(
        &mut record,
        56,
        usize_u64(witnesses.iter().filter(|witness| witness.is_none()).count())?,
    );
    Ok(record)
}

fn put_summary_payload(
    record: &mut [u8],
    flags_offset: usize,
    base: usize,
    summary: &CoverageSummary,
) -> Result<(), CoverageIndexError> {
    record[flags_offset] = u8::from(summary.coverage_complete);
    record[flags_offset + 1] = match summary.completeness_blocked {
        None => 0,
        Some(false) => 1,
        Some(true) => 2,
    };
    for (index, value) in [
        summary.decisions,
        summary.executed_decisions,
        summary.covered_decisions,
        summary.conditions,
        summary.covered_conditions,
    ]
    .into_iter()
    .enumerate()
    {
        put_u64(record, base + index * 8, usize_u64(value)?);
    }
    for (index, count) in [
        &summary.lines,
        &summary.statements,
        &summary.functions,
        &summary.branches,
        &summary.decision_outcomes,
        &summary.condition_outcomes,
        &summary.value_selections,
    ]
    .into_iter()
    .enumerate()
    {
        put_count(record, base + 40 + index * 16, count)?;
    }
    Ok(())
}

fn dimension_record(
    view_id: CoverageViewId,
    dimension: CoverageDimension,
    value: &crate::coverage_report::DimensionCoverage,
    strings: &mut StringTable,
) -> Result<[u8; DIMENSION_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; DIMENSION_RECORD_SIZE];
    record[0] = view_id as u8;
    record[1] = dimension as u8;
    let name = match dimension {
        CoverageDimension::Kind => value.kind.as_deref(),
        CoverageDimension::Runner => value.runner.as_deref(),
        CoverageDimension::Tests => Some(""),
    }
    .ok_or(CoverageIndexError::InvalidRecord("dimension name"))?;
    put_u32(&mut record, 4, strings.intern(name)?);
    put_u64(&mut record, 8, usize_u64(value.tests)?);
    put_u64(&mut record, 16, usize_u64(value.setups)?);
    // Bytes 184..192 were reserved from the start and the reader refuses a
    // record that sets them, so this is an addition rather than a format
    // change: the record is the same 192 bytes it always was.
    put_u64(&mut record, 184, usize_u64(value.attributed)?);
    put_summary_payload(&mut record, 24, 32, &value.summary)?;
    Ok(record)
}

#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum ScopeKind {
    SourceDiscovery = 1,
    Compiler = 2,
}

struct ScopeProjection<'a> {
    kind: ScopeKind,
    language: &'a str,
    model: &'a str,
    mode: Option<&'a str>,
    roots: Vec<String>,
    unit: Option<&'a str>,
    measurement_complete: Option<bool>,
    entries: Option<&'a Vec<serde_json::Value>>,
}

fn scope_projection<'a>(
    scope: Option<&'a serde_json::Value>,
    coverage_language: &'a str,
    coverage_model: &'a str,
) -> Result<Option<ScopeProjection<'a>>, CoverageIndexError> {
    let Some(scope) = scope else {
        return Ok(None);
    };
    let object = scope
        .as_object()
        .ok_or(CoverageIndexError::InvalidRecord("coverage scope"))?;
    if let Some(mode) = object.get("mode") {
        let mode = mode
            .as_str()
            .ok_or(CoverageIndexError::InvalidRecord("source-scope mode"))?;
        let roots = object
            .get("roots")
            .and_then(serde_json::Value::as_array)
            .ok_or(CoverageIndexError::InvalidRecord("source-scope roots"))?
            .iter()
            .map(|root| {
                root.as_str()
                    .map(str::to_owned)
                    .ok_or(CoverageIndexError::InvalidRecord("source-scope root"))
            })
            .collect::<Result<Vec<_>, _>>()?;
        let entries = object
            .get("entries")
            .and_then(serde_json::Value::as_array)
            .ok_or(CoverageIndexError::InvalidRecord("source-scope entries"))?;
        return Ok(Some(ScopeProjection {
            kind: ScopeKind::SourceDiscovery,
            language: coverage_language,
            model: coverage_model,
            mode: Some(mode),
            roots,
            unit: None,
            measurement_complete: None,
            entries: Some(entries),
        }));
    }

    let language = object
        .get("language")
        .and_then(serde_json::Value::as_str)
        .ok_or(CoverageIndexError::InvalidRecord("compiler-scope language"))?;
    let model = object
        .get("model")
        .and_then(serde_json::Value::as_str)
        .ok_or(CoverageIndexError::InvalidRecord("compiler-scope model"))?;
    let unit = object
        .get("crate")
        .and_then(serde_json::Value::as_str)
        .ok_or(CoverageIndexError::InvalidRecord("compiler-scope unit"))?;
    let measurement_complete = object
        .get("measurementComplete")
        .and_then(serde_json::Value::as_bool)
        .ok_or(CoverageIndexError::InvalidRecord(
            "compiler-scope measurement completeness",
        ))?;
    Ok(Some(ScopeProjection {
        kind: ScopeKind::Compiler,
        language,
        model,
        mode: None,
        roots: Vec::new(),
        unit: Some(unit),
        measurement_complete: Some(measurement_complete),
        entries: None,
    }))
}

fn projection_record(
    view_id: CoverageViewId,
    view: &CoverageView,
    selected: Option<&Selected>,
    kind: Option<&str>,
    runner: Option<&str>,
    strings: &mut StringTable,
    relations: &mut StringRelations,
) -> Result<[u8; PROJECTION_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; PROJECTION_RECORD_SIZE];
    record[0] = view_id as u8;
    put_u32(
        &mut record,
        4,
        kind.map_or(Ok(NO_STRING), |value| strings.intern(value))?,
    );
    put_u32(
        &mut record,
        8,
        runner.map_or(Ok(NO_STRING), |value| strings.intern(value))?,
    );
    put_u32(&mut record, 12, strings.intern(&view.generated_at)?);

    let scope = scope_projection(view.scope.as_ref(), &view.model.language, &view.model.name)?;
    record[2] = u8::from(scope.is_some());
    record[3] = scope.as_ref().map_or(0, |scope| scope.kind as u8);
    put_u32(
        &mut record,
        16,
        scope
            .as_ref()
            .and_then(|scope| scope.mode)
            .map_or(Ok(NO_STRING), |value| strings.intern(value))?,
    );
    let roots = scope
        .as_ref()
        .map_or_else(Vec::new, |scope| scope.roots.clone());
    let (roots_offset, roots_count) = relations.push(roots, strings)?;
    put_u64(&mut record, 24, roots_offset);
    put_u32(
        &mut record,
        32,
        u32::try_from(roots_count).map_err(|_| CoverageIndexError::SizeOverflow)?,
    );

    let summary = selected.map_or_else(
        || Ok(view.summary.clone()),
        |ids| {
            coverage_summary_for_tests(view, &ids.iter().cloned().collect())
                .map_err(|_| CoverageIndexError::InvalidRecord("projection summary"))
        },
    )?;
    put_summary_payload(&mut record, 36, 40, &summary)?;

    let mut limitation_kinds = [0_usize; 3];
    let mut limitation_files = BTreeSet::new();
    for limitation in &view.limitations {
        if let Some((file, kind)) = limitation_kind(limitation) {
            limitation_files.insert(file);
            match kind {
                "dynamic-code" => limitation_kinds[0] += 1,
                "semantic-safety" => limitation_kinds[1] += 1,
                "source-scope" => limitation_kinds[2] += 1,
                _ => {}
            }
        }
    }
    let corrupt_records = view
        .transport
        .as_ref()
        .map_or(0, |value| value.corrupt_records);
    let corrupt_files = view
        .transport
        .as_ref()
        .map_or(0, |value| value.corrupt_files);
    // A limitation that declares a boundary of the denominator does not block
    // measurement inside it; corrupt evidence always does.
    let declared = view
        .limitations
        .iter()
        .filter(|limitation| !crate::coverage_report::blocking_limitation(limitation))
        .count();
    for (offset, value) in [
        (192, view.limitations.len()),
        (200, corrupt_records),
        (208, view.limitations.len() - declared + corrupt_records),
        (528, declared),
        (216, limitation_files.len() + corrupt_files),
        (224, limitation_kinds[0]),
        (232, limitation_kinds[1]),
        (240, limitation_kinds[2]),
    ] {
        put_u64(&mut record, offset, usize_u64(value)?);
    }

    let phases = view
        .phases
        .iter()
        .filter(|phase| selected.is_none_or(|selected| selected.contains(phase.test.as_str())));
    let mut attribution = [0_usize; 4];
    for phase in phases {
        attribution[0] += phase.explicit_browser_events;
        attribution[1] += phase.inferred_browser_events;
        attribution[2] += phase.explicit_server_events;
        attribution[3] += phase.inferred_server_events;
    }
    for (index, value) in attribution.into_iter().enumerate() {
        put_u64(&mut record, 248 + index * 8, usize_u64(value)?);
    }

    let confidence_levels = ["unexecuted", "executed", "action", "asserted"];
    for (index, level) in confidence_levels.into_iter().enumerate() {
        put_u64(
            &mut record,
            280 + index * 8,
            usize_u64(
                view.lines
                    .iter()
                    .filter(|line| line.confidence.level == level)
                    .count(),
            )?,
        );
    }
    put_u64(
        &mut record,
        312,
        usize_u64(
            view.decisions
                .iter()
                .flat_map(|decision| &decision.conditions)
                .filter(|condition| condition.assertion_covered)
                .count(),
        )?,
    );

    let gaps = file_gaps(view, selected)?;
    put_u64(
        &mut record,
        320,
        usize_u64(
            gaps.iter()
                .filter(|(_, gap)| {
                    gap.uncovered_lines > 0
                        || gap.uncovered_statements > 0
                        || gap.uncovered_functions > 0
                        || gap.missing_branches > 0
                        || gap.missing_mcdc_conditions > 0
                        || gap.measurement_limitations > 0
                })
                .count(),
        )?,
    );
    put_u64(
        &mut record,
        328,
        usize_u64(
            gaps.iter()
                .filter(|(_, gap)| {
                    gap.uncovered_lines > 0
                        || gap.uncovered_statements > 0
                        || gap.uncovered_functions > 0
                        || gap.missing_branches > 0
                        || gap.missing_mcdc_conditions > 0
                })
                .count(),
        )?,
    );

    let selected_tests = view
        .tests
        .iter()
        .filter(|test| selected.is_none_or(|selected| selected.contains(&test.id)))
        .collect::<Vec<_>>();
    put_u64(
        &mut record,
        336,
        usize_u64(
            selected_tests
                .iter()
                .filter(|test| test.role == "test")
                .count(),
        )?,
    );
    put_u64(
        &mut record,
        344,
        usize_u64(
            selected_tests
                .iter()
                .filter(|test| test.role == "setup")
                .count(),
        )?,
    );
    for (index, outcome) in [
        "passed",
        "failed",
        "flaky",
        "skipped",
        "timedOut",
        "interrupted",
        "unknown",
    ]
    .into_iter()
    .enumerate()
    {
        put_u64(
            &mut record,
            352 + index * 8,
            usize_u64(
                selected_tests
                    .iter()
                    .filter(|test| test.role == "test" && test.outcome == outcome)
                    .count(),
            )?,
        );
    }
    put_u64(
        &mut record,
        520,
        usize_u64(
            selected_tests
                .iter()
                .filter(|test| test.role == "test" && test.outcome == "unstarted")
                .count(),
        )?,
    );

    if let Some(transport) = &view.transport {
        record[1] = 1;
        for (index, value) in [
            transport.processes,
            transport.child_launches,
            transport.remote_launches,
            transport.workspace_capabilities,
            transport.scoped_server_records,
            transport.background_server_records,
            transport.corrupt_records,
            transport.corrupt_files,
        ]
        .into_iter()
        .enumerate()
        {
            put_u64(&mut record, 408 + index * 8, usize_u64(value)?);
        }
        put_u64(&mut record, 536, usize_u64(transport.guest_processes)?);
    }

    let phase_tests = view
        .phases
        .iter()
        .map(|phase| phase.test.as_str())
        .collect::<BTreeSet<_>>();
    let empty_tests = selected_tests
        .iter()
        .filter(|test| {
            test.role == "test"
                && test.lines.is_empty()
                && test.hits.is_empty()
                && test.decisions.is_empty()
                && phase_tests.contains(test.id.as_str())
        })
        .collect::<Vec<_>>();
    put_u64(&mut record, 472, usize_u64(empty_tests.len())?);
    put_u32(
        &mut record,
        20,
        empty_tests
            .first()
            .map_or(Ok(NO_STRING), |test| strings.intern(&test.name))?,
    );

    let entries = scope.as_ref().and_then(|scope| scope.entries);
    for (index, status) in ["included", "excluded", "ambiguous"]
        .into_iter()
        .enumerate()
    {
        put_u64(
            &mut record,
            480 + index * 8,
            usize_u64(entries.map_or(0, |entries| {
                entries
                    .iter()
                    .filter(|entry| {
                        entry.get("status").and_then(serde_json::Value::as_str) == Some(status)
                    })
                    .count()
            }))?,
        );
    }
    put_u32(
        &mut record,
        504,
        scope
            .as_ref()
            .map_or(Ok(NO_STRING), |scope| strings.intern(scope.language))?,
    );
    put_u32(
        &mut record,
        508,
        scope
            .as_ref()
            .map_or(Ok(NO_STRING), |scope| strings.intern(scope.model))?,
    );
    put_u32(
        &mut record,
        512,
        scope
            .as_ref()
            .and_then(|scope| scope.unit)
            .map_or(Ok(NO_STRING), |value| strings.intern(value))?,
    );
    if let Some(measurement_complete) = scope.as_ref().and_then(|scope| scope.measurement_complete)
    {
        record[516] = 1;
        record[517] = u8::from(measurement_complete);
    }
    Ok(record)
}

fn scope_entry_records(
    view_id: CoverageViewId,
    view: &CoverageView,
    strings: &mut StringTable,
) -> Result<Vec<[u8; SCOPE_ENTRY_RECORD_SIZE]>, CoverageIndexError> {
    let Some(scope) = &view.scope else {
        return Ok(Vec::new());
    };
    if scope.get("mode").is_none() {
        return Ok(Vec::new());
    }
    let entries = scope
        .get("entries")
        .and_then(serde_json::Value::as_array)
        .ok_or(CoverageIndexError::InvalidRecord("source-scope entries"))?;
    let mut limitations = BTreeMap::<&str, (usize, u32)>::new();
    for limitation in &view.limitations {
        let Some((file, kind)) = limitation_kind(limitation) else {
            return Err(CoverageIndexError::InvalidRecord("coverage limitation"));
        };
        let value = limitations.entry(file).or_default();
        value.0 += 1;
        value.1 |= match kind {
            "dynamic-code" => 1,
            "semantic-safety" => 2,
            "source-scope" => 4,
            _ => {
                return Err(CoverageIndexError::InvalidRecord(
                    "coverage limitation kind",
                ));
            }
        };
    }
    entries
        .iter()
        .map(|entry| {
            let file = entry
                .get("file")
                .and_then(serde_json::Value::as_str)
                .ok_or(CoverageIndexError::InvalidRecord("source-scope file"))?;
            let status = entry
                .get("status")
                .and_then(serde_json::Value::as_str)
                .ok_or(CoverageIndexError::InvalidRecord("source-scope status"))?;
            let reason = entry
                .get("reason")
                .and_then(serde_json::Value::as_str)
                .ok_or(CoverageIndexError::InvalidRecord("source-scope reason"))?;
            let package_root = entry
                .get("packageRoot")
                .map(|value| {
                    value.as_str().ok_or(CoverageIndexError::InvalidRecord(
                        "source-scope package root",
                    ))
                })
                .transpose()?;
            let mut record = [0_u8; SCOPE_ENTRY_RECORD_SIZE];
            record[0] = view_id as u8;
            record[1] = match status {
                "included" => 0,
                "excluded" => 1,
                "ambiguous" => 2,
                _ => return Err(CoverageIndexError::InvalidRecord("source-scope status")),
            };
            put_u32(&mut record, 4, strings.intern(file)?);
            put_u32(&mut record, 8, strings.intern(reason)?);
            put_u32(
                &mut record,
                12,
                package_root.map_or(Ok(NO_STRING), |value| strings.intern(value))?,
            );
            let (count, mask) = limitations.get(file).copied().unwrap_or_default();
            put_u64(&mut record, 16, usize_u64(count)?);
            put_u32(&mut record, 24, mask);
            Ok(record)
        })
        .collect()
}

fn optional_string_id(
    value: Option<&str>,
    strings: &mut StringTable,
) -> Result<u32, CoverageIndexError> {
    value.map_or(Ok(NO_STRING), |value| strings.intern(value))
}

fn confidence_record(
    confidence: &crate::coverage_report::CoverageConfidence,
    strings: &mut StringTable,
    relations: &mut StringRelations,
) -> Result<[u8; CONFIDENCE_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; CONFIDENCE_RECORD_SIZE];
    record[0] = match confidence.level.as_str() {
        "unexecuted" => 0,
        "executed" => 1,
        "action" => 2,
        "asserted" => 3,
        _ => return Err(CoverageIndexError::InvalidRecord("confidence level")),
    };
    record[1] = u8::from(confidence.setup_only)
        | (u8::from(confidence.background_only) << 1)
        | (u8::from(confidence.asserted) << 2)
        | (u8::from(confidence.e2e) << 3);
    for (index, values) in [&confidence.tests, &confidence.asserted_tests]
        .into_iter()
        .enumerate()
    {
        let (offset, count) = relations.push_ids(values, strings)?;
        put_u64(&mut record, 8 + index * 16, offset);
        put_u64(&mut record, 16 + index * 16, count);
    }
    for (index, values) in [&confidence.runners, &confidence.kinds]
        .into_iter()
        .enumerate()
    {
        let (offset, count) = relations.push(values, strings)?;
        put_u64(&mut record, 40 + index * 16, offset);
        put_u64(&mut record, 48 + index * 16, count);
    }
    Ok(record)
}

fn line_record(
    view_id: CoverageViewId,
    line: &crate::coverage_report::LineResult,
    confidence_index: usize,
    strings: &mut StringTable,
    relations: &mut StringRelations,
) -> Result<[u8; LINE_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; LINE_RECORD_SIZE];
    record[0] = view_id as u8;
    record[1] = u8::from(line.covered);
    // Inverted: a line every frontend declined is the exception, and a zeroed
    // byte then still reads as measured.
    record[2] = u8::from(!line.measured);
    put_u32(&mut record, 4, strings.intern(&line.file)?);
    put_u64(&mut record, 8, usize_u64(line.line)?);
    let (tests_offset, tests_count) = relations.push_ids(&line.tests, strings)?;
    put_u64(&mut record, 16, tests_offset);
    put_u64(&mut record, 24, tests_count);
    let (phases_offset, phases_count) = relations.push_ids(&line.phases, strings)?;
    put_u64(&mut record, 32, phases_offset);
    put_u64(&mut record, 40, phases_count);
    put_u64(&mut record, 48, usize_u64(confidence_index)?);
    Ok(record)
}

fn test_summary_record(
    view_id: CoverageViewId,
    test: &crate::coverage_report::TestCoverageResult,
    strings: &mut StringTable,
) -> Result<[u8; TEST_SUMMARY_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; TEST_SUMMARY_RECORD_SIZE];
    record[0] = view_id as u8;
    record[1] = match test.role.as_str() {
        "test" => 0,
        "setup" => 1,
        "background" => 2,
        _ => return Err(CoverageIndexError::InvalidRecord("test role")),
    };
    // Byte 3 was reserved from the start and the reader refuses a record that
    // sets it, which is what makes this an addition rather than a format
    // change: the record is the same 64 bytes it always was.
    record[3] = match test.attribution.as_str() {
        crate::coverage_report::ATTRIBUTION_EXACT => 0,
        crate::coverage_report::ATTRIBUTION_RUN_WIDE => 1,
        crate::coverage_report::ATTRIBUTION_PARTIAL => 2,
        _ => return Err(CoverageIndexError::InvalidRecord("test attribution")),
    };
    record[2] = match test.outcome.as_str() {
        "passed" => 0,
        "failed" => 1,
        "flaky" => 2,
        "skipped" => 3,
        "timedOut" => 4,
        "interrupted" => 5,
        "unknown" => 6,
        "unstarted" => 7,
        "todo" => 8,
        _ => return Err(CoverageIndexError::InvalidRecord("test outcome")),
    };
    put_u32(&mut record, 4, strings.intern(&test.id)?);
    put_u32(&mut record, 8, strings.intern(&test.name)?);
    put_u32(
        &mut record,
        12,
        optional_string_id(test.file.as_deref(), strings)?,
    );
    put_u32(
        &mut record,
        16,
        optional_string_id(test.title.as_deref(), strings)?,
    );
    put_u32(&mut record, 20, strings.intern(&test.provenance.runner)?);
    put_u32(&mut record, 24, strings.intern(&test.provenance.kind)?);
    put_u32(
        &mut record,
        28,
        optional_string_id(test.provenance.project.as_deref(), strings)?,
    );
    put_u32(&mut record, 32, strings.intern(&test.provenance.source)?);
    Ok(record)
}

fn phase_summary_record(
    view_id: CoverageViewId,
    phase: &crate::coverage_report::PhaseResult,
    strings: &mut StringTable,
) -> Result<[u8; PHASE_SUMMARY_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; PHASE_SUMMARY_RECORD_SIZE];
    record[0] = view_id as u8;
    put_u32(&mut record, 4, strings.intern(&phase.phase.id)?);
    put_u32(&mut record, 8, strings.intern(&phase.phase.kind)?);
    put_u32(&mut record, 12, strings.intern(&phase.phase.operation)?);
    put_u32(
        &mut record,
        16,
        optional_string_id(phase.phase.source.as_deref(), strings)?,
    );
    put_u32(&mut record, 20, strings.intern(&phase.test)?);
    put_u32(
        &mut record,
        24,
        optional_string_id(phase.phase.status.as_deref(), strings)?,
    );
    put_u32(
        &mut record,
        28,
        optional_string_id(phase.phase.caused_by_phase_id.as_deref(), strings)?,
    );
    put_u64(&mut record, 32, usize_u64(phase.lines.len())?);
    put_u64(
        &mut record,
        40,
        usize_u64(
            phase
                .decisions
                .iter()
                .map(|decision| decision.vectors.len())
                .sum(),
        )?,
    );
    Ok(record)
}

struct AnchorInput<'a> {
    view_id: CoverageViewId,
    kind: u8,
    id: &'a str,
    file: &'a str,
    line: usize,
    column: usize,
    covered: bool,
    conditions: Option<(usize, usize)>,
    tests: &'a [Id],
}

fn anchor_record(
    input: AnchorInput<'_>,
    strings: &mut StringTable,
    relations: &mut StringRelations,
) -> Result<[u8; ANCHOR_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; ANCHOR_RECORD_SIZE];
    record[0] = input.view_id as u8;
    record[1] = input.kind;
    record[2] = u8::from(input.covered);
    put_u32(&mut record, 4, strings.intern(input.id)?);
    put_u32(&mut record, 8, strings.intern(input.file)?);
    put_u64(&mut record, 16, usize_u64(input.line)?);
    put_u64(&mut record, 24, usize_u64(input.column)?);
    if let Some((covered, total)) = input.conditions {
        put_u64(&mut record, 32, usize_u64(total)?);
        put_u64(&mut record, 40, usize_u64(covered)?);
    }
    let (tests_offset, tests_count) = relations.push_ids(input.tests, strings)?;
    put_u64(&mut record, 48, tests_offset);
    put_u64(&mut record, 56, tests_count);
    Ok(record)
}

fn test_retry_record(
    view_id: CoverageViewId,
    test_id: &str,
    retry: usize,
    strings: &mut StringTable,
) -> Result<[u8; TEST_RETRY_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; TEST_RETRY_RECORD_SIZE];
    record[0] = view_id as u8;
    put_u32(&mut record, 4, strings.intern(test_id)?);
    put_u64(&mut record, 8, usize_u64(retry)?);
    Ok(record)
}

fn test_attempt_record(
    view_id: CoverageViewId,
    test_id: &str,
    attempt: &crate::coverage_report::TestAttempt,
    strings: &mut StringTable,
) -> Result<[u8; TEST_ATTEMPT_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; TEST_ATTEMPT_RECORD_SIZE];
    record[0] = view_id as u8;
    put_u32(&mut record, 4, strings.intern(test_id)?);
    put_u64(&mut record, 8, usize_u64(attempt.retry)?);
    put_u32(&mut record, 16, strings.intern(&attempt.status)?);
    put_u32(
        &mut record,
        20,
        optional_string_id(attempt.expected_status.as_deref(), strings)?,
    );
    Ok(record)
}

fn test_line_record(
    view_id: CoverageViewId,
    test_id: &str,
    line: &crate::coverage_report::SourceLine,
    strings: &mut StringTable,
) -> Result<[u8; TEST_LINE_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; TEST_LINE_RECORD_SIZE];
    record[0] = view_id as u8;
    put_u32(&mut record, 4, strings.intern(test_id)?);
    put_u32(&mut record, 8, strings.intern(&line.file)?);
    put_u64(&mut record, 16, usize_u64(line.line)?);
    Ok(record)
}

fn test_hit_record(
    view_id: CoverageViewId,
    test_id: &str,
    hit: &str,
    strings: &mut StringTable,
) -> Result<[u8; TEST_HIT_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; TEST_HIT_RECORD_SIZE];
    record[0] = view_id as u8;
    put_u32(&mut record, 4, strings.intern(test_id)?);
    put_u32(&mut record, 8, strings.intern(hit)?);
    Ok(record)
}

fn test_vector_record(
    vector: &McdcVector,
    values: &mut Vec<u8>,
) -> Result<[u8; TEST_VECTOR_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; TEST_VECTOR_RECORD_SIZE];
    record[0] = u8::from(vector.outcome);
    put_u64(&mut record, 8, usize_u64(values.len())?);
    put_u64(&mut record, 16, usize_u64(vector.values.len())?);
    values.extend(vector.values.iter().map(|value| match value {
        None => 0,
        Some(false) => 1,
        Some(true) => 2,
    }));
    Ok(record)
}

fn test_decision_record(
    view_id: CoverageViewId,
    test_id: &str,
    decision_id: &str,
    vectors_offset: usize,
    vectors_count: usize,
    strings: &mut StringTable,
) -> Result<[u8; TEST_DECISION_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; TEST_DECISION_RECORD_SIZE];
    record[0] = view_id as u8;
    put_u32(&mut record, 4, strings.intern(test_id)?);
    put_u32(&mut record, 8, strings.intern(decision_id)?);
    put_u64(&mut record, 16, usize_u64(vectors_offset)?);
    put_u64(&mut record, 24, usize_u64(vectors_count)?);
    Ok(record)
}

struct HitMetadataInput<'a> {
    view_id: CoverageViewId,
    kind: u8,
    id: &'a str,
    file: &'a str,
    line: usize,
    column: usize,
    branch_kind: Option<&'a str>,
    label: Option<&'a str>,
    alternative: Option<&'a str>,
    source: &'a str,
    tests: &'a [Id],
}

fn hit_metadata_record(
    input: HitMetadataInput<'_>,
    strings: &mut StringTable,
    relations: &mut StringRelations,
) -> Result<[u8; HIT_METADATA_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; HIT_METADATA_RECORD_SIZE];
    record[0] = input.view_id as u8;
    record[1] = input.kind;
    put_u32(&mut record, 4, strings.intern(input.id)?);
    put_u32(&mut record, 8, strings.intern(input.file)?);
    put_u64(&mut record, 16, usize_u64(input.line)?);
    put_u64(&mut record, 24, usize_u64(input.column)?);
    put_u32(
        &mut record,
        32,
        optional_string_id(input.branch_kind, strings)?,
    );
    put_u32(&mut record, 36, optional_string_id(input.label, strings)?);
    put_u32(
        &mut record,
        40,
        optional_string_id(input.alternative, strings)?,
    );
    put_u32(&mut record, 44, strings.intern(input.source)?);
    let (tests_offset, tests_count) = relations.push_ids(input.tests, strings)?;
    put_u64(&mut record, 48, tests_offset);
    put_u64(&mut record, 56, tests_count);
    Ok(record)
}

fn limitation_record(
    view_id: CoverageViewId,
    limitation: &serde_json::Value,
    strings: &mut StringTable,
) -> Result<[u8; LIMITATION_RECORD_SIZE], CoverageIndexError> {
    let field = |name| {
        limitation
            .get(name)
            .and_then(serde_json::Value::as_str)
            .ok_or(CoverageIndexError::InvalidRecord("coverage limitation"))
    };
    let number = |name| {
        limitation
            .get(name)
            .and_then(serde_json::Value::as_u64)
            .ok_or(CoverageIndexError::InvalidRecord("coverage limitation"))
    };
    let mut record = [0_u8; LIMITATION_RECORD_SIZE];
    record[0] = view_id as u8;
    record[1] = u8::from(crate::coverage_report::blocking_limitation(limitation));
    put_u32(&mut record, 4, strings.intern(field("id")?)?);
    put_u32(&mut record, 8, strings.intern(field("kind")?)?);
    put_u32(&mut record, 12, strings.intern(field("file")?)?);
    put_u32(&mut record, 16, strings.intern(field("source")?)?);
    put_u32(&mut record, 20, strings.intern(field("reason")?)?);
    put_u64(&mut record, 24, number("line")?);
    put_u64(&mut record, 32, number("column")?);
    Ok(record)
}

fn decision_metadata_record(
    view_id: CoverageViewId,
    decision: &crate::coverage_report::DecisionResult,
    strings: &mut StringTable,
    relations: &mut StringRelations,
) -> Result<[u8; DECISION_METADATA_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; DECISION_METADATA_RECORD_SIZE];
    record[0] = view_id as u8;
    put_u32(&mut record, 4, strings.intern(&decision.meta.id)?);
    put_u32(&mut record, 8, strings.intern(&decision.meta.file)?);
    put_u32(&mut record, 12, strings.intern(&decision.meta.source)?);
    put_u32(&mut record, 16, strings.intern(&decision.meta.kind)?);
    put_u64(&mut record, 24, usize_u64(decision.meta.line)?);
    put_u64(&mut record, 32, usize_u64(decision.meta.column)?);
    let (conditions_offset, conditions_count) =
        relations.push(decision.meta.conditions.clone(), strings)?;
    put_u64(&mut record, 40, conditions_offset);
    put_u64(&mut record, 48, conditions_count);
    Ok(record)
}

fn decision_vector_observation_record(
    observation: &crate::coverage_report::VectorObservation,
    confidence_index: usize,
    vector_index: usize,
    strings: &mut StringTable,
    relations: &mut StringRelations,
) -> Result<[u8; DECISION_VECTOR_OBSERVATION_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; DECISION_VECTOR_OBSERVATION_RECORD_SIZE];
    put_u64(&mut record, 0, usize_u64(confidence_index)?);
    put_u64(&mut record, 8, usize_u64(vector_index)?);
    for (offset, values) in [
        (16, &observation.tests),
        (32, &observation.phases),
        (48, &observation.explicit_phases),
    ] {
        let (relation_offset, relation_count) = relations.push_ids(values, strings)?;
        put_u64(&mut record, offset, relation_offset);
        put_u64(&mut record, offset + 8, relation_count);
    }
    Ok(record)
}

fn decision_condition_record(
    condition: &crate::coverage_report::ConditionResult,
    witness_vectors: Option<(usize, usize)>,
    strings: &mut StringTable,
    relations: &mut StringRelations,
) -> Result<[u8; DECISION_CONDITION_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; DECISION_CONDITION_RECORD_SIZE];
    record[0] = u8::from(condition.covered)
        | (u8::from(condition.assertion_covered) << 1)
        | (u8::from(condition.witness.is_some()) << 2);
    put_u32(&mut record, 4, strings.intern(&condition.source)?);
    put_u64(&mut record, 8, usize_u64(condition.index)?);
    if let Some((first, second)) = witness_vectors {
        put_u64(&mut record, 16, usize_u64(first)?);
        put_u64(&mut record, 24, usize_u64(second)?);
    }
    let empty = [Vec::new(), Vec::new()];
    let witness_tests = condition.witness_tests.as_ref().unwrap_or(&empty);
    for (offset, values) in [(32, &witness_tests[0]), (48, &witness_tests[1])] {
        let (relation_offset, relation_count) = relations.push_ids(values, strings)?;
        put_u64(&mut record, offset, relation_offset);
        put_u64(&mut record, offset + 8, relation_count);
    }
    Ok(record)
}

struct DecisionDetailInput<'a> {
    view_id: CoverageViewId,
    decision: &'a crate::coverage_report::DecisionResult,
    confidence_index: usize,
    observations: (usize, usize),
    conditions: (usize, usize),
}

fn decision_detail_record(
    input: DecisionDetailInput<'_>,
    strings: &mut StringTable,
    relations: &mut StringRelations,
) -> Result<[u8; DECISION_DETAIL_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; DECISION_DETAIL_RECORD_SIZE];
    record[0] = input.view_id as u8;
    record[1] = u8::from(input.decision.executed) | (u8::from(input.decision.covered) << 1);
    put_u32(&mut record, 4, strings.intern(&input.decision.meta.id)?);
    put_u64(&mut record, 8, usize_u64(input.confidence_index)?);
    let (tests_offset, tests_count) = relations.push_ids(&input.decision.tests, strings)?;
    put_u64(&mut record, 16, tests_offset);
    put_u64(&mut record, 24, tests_count);
    put_u64(&mut record, 32, usize_u64(input.observations.0)?);
    put_u64(&mut record, 40, usize_u64(input.observations.1)?);
    put_u64(&mut record, 48, usize_u64(input.conditions.0)?);
    put_u64(&mut record, 56, usize_u64(input.conditions.1)?);
    Ok(record)
}

fn coverage_model_record(
    variant: &str,
    model: &CoverageModel,
    strings: &mut StringTable,
    relations: &mut StringRelations,
) -> Result<[u8; COVERAGE_MODEL_RECORD_SIZE], CoverageIndexError> {
    let mut record = [0_u8; COVERAGE_MODEL_RECORD_SIZE];
    put_u32(&mut record, 0, strings.intern(variant)?);
    put_u32(&mut record, 4, strings.intern(&model.name)?);
    put_u32(&mut record, 8, strings.intern(&model.completeness_meaning)?);
    let (measured_offset, measured_count) = relations.push(model.measured.clone(), strings)?;
    put_u64(&mut record, 16, measured_offset);
    put_u64(&mut record, 24, measured_count);
    let (not_measured_offset, not_measured_count) =
        relations.push(model.not_measured.clone(), strings)?;
    put_u64(&mut record, 32, not_measured_offset);
    put_u64(&mut record, 40, not_measured_count);
    Ok(record)
}

pub fn coverage_index_sections(
    report: &CoverageReport,
) -> Result<Vec<QueryIndexSection>, CoverageIndexError> {
    let views = [
        (CoverageViewId::All, &report.view),
        (CoverageViewId::Passed, &report.filters.passed),
        (CoverageViewId::Failed, &report.filters.failed),
    ];
    let mut strings = StringTable::default();
    let mut relations = StringRelations::default();
    let mut summaries = Vec::with_capacity(views.len() * SUMMARY_RECORD_SIZE);
    let mut gaps = Vec::new();
    let mut decision_gaps = Vec::new();
    let mut dimensions = Vec::new();
    let mut projection_records = Vec::new();
    let mut scope_entries = Vec::new();
    let mut confidence_records = Vec::new();
    let mut line_records = Vec::new();
    let mut test_summaries = Vec::new();
    let mut phase_summaries = Vec::new();
    let mut anchors = Vec::new();
    let mut test_retries = Vec::new();
    let mut test_attempts = Vec::new();
    let mut test_lines = Vec::new();
    let mut test_hits = Vec::new();
    let mut test_decisions = Vec::new();
    let mut test_vectors = Vec::new();
    let mut vector_values = Vec::new();
    let mut hit_metadata = Vec::new();
    let mut decision_metadata = Vec::new();
    let mut decision_details = Vec::new();
    let mut decision_vector_observations = Vec::new();
    let mut decision_conditions = Vec::new();
    let mut limitations = Vec::new();
    for (id, view) in views {
        summaries.extend_from_slice(&summary_record(id, view, &mut strings)?);
        projection_records.extend_from_slice(&projection_record(
            id,
            view,
            None,
            None,
            None,
            &mut strings,
            &mut relations,
        )?);
        for entry in scope_entry_records(id, view, &mut strings)? {
            scope_entries.extend_from_slice(&entry);
        }
        for limitation in &view.limitations {
            limitations.extend_from_slice(&limitation_record(id, limitation, &mut strings)?);
        }
        for line in &view.lines {
            let confidence_index = confidence_records.len() / CONFIDENCE_RECORD_SIZE;
            confidence_records.extend_from_slice(&confidence_record(
                &line.confidence,
                &mut strings,
                &mut relations,
            )?);
            line_records.extend_from_slice(&line_record(
                id,
                line,
                confidence_index,
                &mut strings,
                &mut relations,
            )?);
        }
        for test in &view.tests {
            test_summaries.extend_from_slice(&test_summary_record(id, test, &mut strings)?);
            for retry in &test.retries {
                test_retries.extend_from_slice(&test_retry_record(
                    id,
                    &test.id,
                    *retry,
                    &mut strings,
                )?);
            }
            for attempt in &test.attempts {
                test_attempts.extend_from_slice(&test_attempt_record(
                    id,
                    &test.id,
                    attempt,
                    &mut strings,
                )?);
            }
            for line in &test.lines {
                test_lines.extend_from_slice(&test_line_record(id, &test.id, line, &mut strings)?);
            }
            for hit in &test.hits {
                test_hits.extend_from_slice(&test_hit_record(id, &test.id, hit, &mut strings)?);
            }
            for decision in &test.decisions {
                let vectors_offset = test_vectors.len() / TEST_VECTOR_RECORD_SIZE;
                for vector in &decision.vectors {
                    test_vectors
                        .extend_from_slice(&test_vector_record(vector, &mut vector_values)?);
                }
                test_decisions.extend_from_slice(&test_decision_record(
                    id,
                    &test.id,
                    &decision.id,
                    vectors_offset,
                    decision.vectors.len(),
                    &mut strings,
                )?);
            }
        }
        for phase in &view.phases {
            phase_summaries.extend_from_slice(&phase_summary_record(id, phase, &mut strings)?);
        }
        for decision in &view.decisions {
            let confidence_index = confidence_records.len() / CONFIDENCE_RECORD_SIZE;
            confidence_records.extend_from_slice(&confidence_record(
                &decision.confidence,
                &mut strings,
                &mut relations,
            )?);
            let observations_offset =
                decision_vector_observations.len() / DECISION_VECTOR_OBSERVATION_RECORD_SIZE;
            for observation in &decision.vector_observations {
                let observation_confidence = confidence_records.len() / CONFIDENCE_RECORD_SIZE;
                confidence_records.extend_from_slice(&confidence_record(
                    &observation.confidence,
                    &mut strings,
                    &mut relations,
                )?);
                let vector_index = test_vectors.len() / TEST_VECTOR_RECORD_SIZE;
                test_vectors.extend_from_slice(&test_vector_record(
                    &observation.vector,
                    &mut vector_values,
                )?);
                decision_vector_observations.extend_from_slice(
                    &decision_vector_observation_record(
                        observation,
                        observation_confidence,
                        vector_index,
                        &mut strings,
                        &mut relations,
                    )?,
                );
            }
            let conditions_offset = decision_conditions.len() / DECISION_CONDITION_RECORD_SIZE;
            for condition in &decision.conditions {
                let witness_vectors = if let Some(witness) = &condition.witness {
                    let first = test_vectors.len() / TEST_VECTOR_RECORD_SIZE;
                    test_vectors
                        .extend_from_slice(&test_vector_record(&witness[0], &mut vector_values)?);
                    let second = test_vectors.len() / TEST_VECTOR_RECORD_SIZE;
                    test_vectors
                        .extend_from_slice(&test_vector_record(&witness[1], &mut vector_values)?);
                    Some((first, second))
                } else {
                    None
                };
                decision_conditions.extend_from_slice(&decision_condition_record(
                    condition,
                    witness_vectors,
                    &mut strings,
                    &mut relations,
                )?);
            }
            decision_details.extend_from_slice(&decision_detail_record(
                DecisionDetailInput {
                    view_id: id,
                    decision,
                    confidence_index,
                    observations: (observations_offset, decision.vector_observations.len()),
                    conditions: (conditions_offset, decision.conditions.len()),
                },
                &mut strings,
                &mut relations,
            )?);
            decision_metadata.extend_from_slice(&decision_metadata_record(
                id,
                decision,
                &mut strings,
                &mut relations,
            )?);
            anchors.extend_from_slice(&anchor_record(
                AnchorInput {
                    view_id: id,
                    kind: 0,
                    id: &decision.meta.id,
                    file: &decision.meta.file,
                    line: decision.meta.line,
                    column: decision.meta.column,
                    covered: decision.covered,
                    conditions: Some((
                        decision
                            .conditions
                            .iter()
                            .filter(|condition| condition.covered)
                            .count(),
                        decision.conditions.len(),
                    )),
                    tests: &decision.tests,
                },
                &mut strings,
                &mut relations,
            )?);
        }
        for branch in &view.branches {
            anchors.extend_from_slice(&anchor_record(
                AnchorInput {
                    view_id: id,
                    kind: 1,
                    id: &branch.meta.id,
                    file: &branch.meta.file,
                    line: branch.meta.line,
                    column: branch.meta.column,
                    covered: branch.covered,
                    conditions: None,
                    tests: &[],
                },
                &mut strings,
                &mut relations,
            )?);
            for alternative in &branch.alternatives {
                hit_metadata.extend_from_slice(&hit_metadata_record(
                    HitMetadataInput {
                        view_id: id,
                        kind: 2,
                        id: &alternative.id,
                        file: &branch.meta.file,
                        line: branch.meta.line,
                        column: branch.meta.column,
                        branch_kind: Some(&branch.meta.kind),
                        label: Some(&branch.meta.id),
                        alternative: Some(&alternative.label),
                        source: &branch.meta.source,
                        tests: &alternative.tests,
                    },
                    &mut strings,
                    &mut relations,
                )?);
            }
        }
        for point in &view.points {
            anchors.extend_from_slice(&anchor_record(
                AnchorInput {
                    view_id: id,
                    kind: match point.meta.kind {
                        crate::coverage_analysis::PointKind::Statement => 2,
                        crate::coverage_analysis::PointKind::Function => 3,
                    },
                    id: &point.meta.id,
                    file: &point.meta.file,
                    line: point.meta.line,
                    column: point.meta.column,
                    covered: point.covered,
                    conditions: None,
                    tests: &point.tests,
                },
                &mut strings,
                &mut relations,
            )?);
            hit_metadata.extend_from_slice(&hit_metadata_record(
                HitMetadataInput {
                    view_id: id,
                    kind: match point.meta.kind {
                        crate::coverage_analysis::PointKind::Statement => 0,
                        crate::coverage_analysis::PointKind::Function => 1,
                    },
                    id: &point.meta.id,
                    file: &point.meta.file,
                    line: point.meta.line,
                    column: point.meta.column,
                    branch_kind: None,
                    label: point.meta.label.as_deref(),
                    alternative: None,
                    source: &point.meta.source,
                    tests: &point.tests,
                },
                &mut strings,
                &mut relations,
            )?);
        }
        for value in &view.coverage_by_kind {
            dimensions.extend_from_slice(&dimension_record(
                id,
                CoverageDimension::Kind,
                value,
                &mut strings,
            )?);
        }
        for value in &view.coverage_by_runner {
            dimensions.extend_from_slice(&dimension_record(
                id,
                CoverageDimension::Runner,
                value,
                &mut strings,
            )?);
        }
        let role = |role: &str| view.tests.iter().filter(|test| test.role == role).count();
        dimensions.extend_from_slice(&dimension_record(
            id,
            CoverageDimension::Tests,
            &crate::coverage_report::DimensionCoverage {
                kind: None,
                runner: None,
                tests: role("test"),
                setups: role("setup"),
                attributed: view
                    .tests
                    .iter()
                    .filter(|test| {
                        test.role == "test"
                            && crate::coverage_report::coverage_is_its_own(&test.attribution)
                    })
                    .count(),
                summary: crate::coverage_report::summary_for_tests(view)
                    .map_err(|_| CoverageIndexError::InvalidRecord("coverage of the tests"))?,
            },
            &mut strings,
        )?);
        for decision in &view.decisions {
            decision_gaps.extend_from_slice(&decision_gap_record(
                id,
                decision,
                None,
                None,
                None,
                &mut strings,
            )?);
        }
        for (file, gap) in file_gaps(view, None)? {
            gaps.extend_from_slice(&file_gap_record(id, &file, &gap, None, None, &mut strings)?);
        }
        for (kind, runner, selected) in projections(view) {
            projection_records.extend_from_slice(&projection_record(
                id,
                view,
                Some(&selected),
                kind.as_deref(),
                runner.as_deref(),
                &mut strings,
                &mut relations,
            )?);
            for decision in &view.decisions {
                decision_gaps.extend_from_slice(&decision_gap_record(
                    id,
                    decision,
                    Some(&selected),
                    kind.as_deref(),
                    runner.as_deref(),
                    &mut strings,
                )?);
            }
            for (file, gap) in file_gaps(view, Some(&selected))? {
                gaps.extend_from_slice(&file_gap_record(
                    id,
                    &file,
                    &gap,
                    kind.as_deref(),
                    runner.as_deref(),
                    &mut strings,
                )?);
            }
        }
    }
    let model = coverage_model_record(
        &report.view.variant,
        &report.view.model,
        &mut strings,
        &mut relations,
    )?;
    let [blob, string_records] = strings.sections()?;
    let string_relations = relations.section()?;
    Ok(vec![
        blob,
        string_records,
        string_relations,
        QueryIndexSection {
            kind: SECTION_VIEW_SUMMARIES,
            record_size: SUMMARY_RECORD_SIZE as u32,
            count: usize_u64(summaries.len() / SUMMARY_RECORD_SIZE)?,
            bytes: summaries,
        },
        QueryIndexSection {
            kind: SECTION_FILE_GAPS,
            record_size: FILE_GAP_RECORD_SIZE as u32,
            count: usize_u64(gaps.len() / FILE_GAP_RECORD_SIZE)?,
            bytes: gaps,
        },
        QueryIndexSection {
            kind: SECTION_DECISION_GAPS,
            record_size: DECISION_GAP_RECORD_SIZE as u32,
            count: usize_u64(decision_gaps.len() / DECISION_GAP_RECORD_SIZE)?,
            bytes: decision_gaps,
        },
        QueryIndexSection {
            kind: SECTION_DIMENSIONS,
            record_size: DIMENSION_RECORD_SIZE as u32,
            count: usize_u64(dimensions.len() / DIMENSION_RECORD_SIZE)?,
            bytes: dimensions,
        },
        QueryIndexSection {
            kind: SECTION_PROJECTIONS,
            record_size: PROJECTION_RECORD_SIZE as u32,
            count: usize_u64(projection_records.len() / PROJECTION_RECORD_SIZE)?,
            bytes: projection_records,
        },
        QueryIndexSection {
            kind: SECTION_SCOPE_ENTRIES,
            record_size: SCOPE_ENTRY_RECORD_SIZE as u32,
            count: usize_u64(scope_entries.len() / SCOPE_ENTRY_RECORD_SIZE)?,
            bytes: scope_entries,
        },
        QueryIndexSection {
            kind: SECTION_CONFIDENCE,
            record_size: CONFIDENCE_RECORD_SIZE as u32,
            count: usize_u64(confidence_records.len() / CONFIDENCE_RECORD_SIZE)?,
            bytes: confidence_records,
        },
        QueryIndexSection {
            kind: SECTION_LINES,
            record_size: LINE_RECORD_SIZE as u32,
            count: usize_u64(line_records.len() / LINE_RECORD_SIZE)?,
            bytes: line_records,
        },
        QueryIndexSection {
            kind: SECTION_TEST_SUMMARIES,
            record_size: TEST_SUMMARY_RECORD_SIZE as u32,
            count: usize_u64(test_summaries.len() / TEST_SUMMARY_RECORD_SIZE)?,
            bytes: test_summaries,
        },
        QueryIndexSection {
            kind: SECTION_PHASE_SUMMARIES,
            record_size: PHASE_SUMMARY_RECORD_SIZE as u32,
            count: usize_u64(phase_summaries.len() / PHASE_SUMMARY_RECORD_SIZE)?,
            bytes: phase_summaries,
        },
        QueryIndexSection {
            kind: SECTION_ANCHORS,
            record_size: ANCHOR_RECORD_SIZE as u32,
            count: usize_u64(anchors.len() / ANCHOR_RECORD_SIZE)?,
            bytes: anchors,
        },
        QueryIndexSection {
            kind: SECTION_TEST_RETRIES,
            record_size: TEST_RETRY_RECORD_SIZE as u32,
            count: usize_u64(test_retries.len() / TEST_RETRY_RECORD_SIZE)?,
            bytes: test_retries,
        },
        QueryIndexSection {
            kind: SECTION_TEST_ATTEMPTS,
            record_size: TEST_ATTEMPT_RECORD_SIZE as u32,
            count: usize_u64(test_attempts.len() / TEST_ATTEMPT_RECORD_SIZE)?,
            bytes: test_attempts,
        },
        QueryIndexSection {
            kind: SECTION_TEST_LINES,
            record_size: TEST_LINE_RECORD_SIZE as u32,
            count: usize_u64(test_lines.len() / TEST_LINE_RECORD_SIZE)?,
            bytes: test_lines,
        },
        QueryIndexSection {
            kind: SECTION_TEST_HITS,
            record_size: TEST_HIT_RECORD_SIZE as u32,
            count: usize_u64(test_hits.len() / TEST_HIT_RECORD_SIZE)?,
            bytes: test_hits,
        },
        QueryIndexSection {
            kind: SECTION_TEST_DECISIONS,
            record_size: TEST_DECISION_RECORD_SIZE as u32,
            count: usize_u64(test_decisions.len() / TEST_DECISION_RECORD_SIZE)?,
            bytes: test_decisions,
        },
        QueryIndexSection {
            kind: SECTION_TEST_VECTORS,
            record_size: TEST_VECTOR_RECORD_SIZE as u32,
            count: usize_u64(test_vectors.len() / TEST_VECTOR_RECORD_SIZE)?,
            bytes: test_vectors,
        },
        QueryIndexSection {
            kind: SECTION_VECTOR_VALUES,
            record_size: 1,
            count: usize_u64(vector_values.len())?,
            bytes: vector_values,
        },
        QueryIndexSection {
            kind: SECTION_HIT_METADATA,
            record_size: HIT_METADATA_RECORD_SIZE as u32,
            count: usize_u64(hit_metadata.len() / HIT_METADATA_RECORD_SIZE)?,
            bytes: hit_metadata,
        },
        QueryIndexSection {
            kind: SECTION_DECISION_METADATA,
            record_size: DECISION_METADATA_RECORD_SIZE as u32,
            count: usize_u64(decision_metadata.len() / DECISION_METADATA_RECORD_SIZE)?,
            bytes: decision_metadata,
        },
        QueryIndexSection {
            kind: SECTION_DECISION_DETAILS,
            record_size: DECISION_DETAIL_RECORD_SIZE as u32,
            count: usize_u64(decision_details.len() / DECISION_DETAIL_RECORD_SIZE)?,
            bytes: decision_details,
        },
        QueryIndexSection {
            kind: SECTION_DECISION_VECTOR_OBSERVATIONS,
            record_size: DECISION_VECTOR_OBSERVATION_RECORD_SIZE as u32,
            count: usize_u64(
                decision_vector_observations.len() / DECISION_VECTOR_OBSERVATION_RECORD_SIZE,
            )?,
            bytes: decision_vector_observations,
        },
        QueryIndexSection {
            kind: SECTION_DECISION_CONDITIONS,
            record_size: DECISION_CONDITION_RECORD_SIZE as u32,
            count: usize_u64(decision_conditions.len() / DECISION_CONDITION_RECORD_SIZE)?,
            bytes: decision_conditions,
        },
        QueryIndexSection {
            kind: SECTION_LIMITATIONS,
            record_size: LIMITATION_RECORD_SIZE as u32,
            count: usize_u64(limitations.len() / LIMITATION_RECORD_SIZE)?,
            bytes: limitations,
        },
        QueryIndexSection {
            kind: SECTION_COVERAGE_MODEL,
            record_size: COVERAGE_MODEL_RECORD_SIZE as u32,
            count: 1,
            bytes: model.to_vec(),
        },
    ])
}

pub struct CoverageIndex<'a> {
    index: &'a QueryIndex,
}

impl<'a> CoverageIndex<'a> {
    pub fn new(index: &'a QueryIndex) -> Result<Self, CoverageIndexError> {
        for (kind, size) in [
            (SECTION_STRINGS, STRING_RECORD_SIZE),
            (SECTION_VIEW_SUMMARIES, SUMMARY_RECORD_SIZE),
            (SECTION_FILE_GAPS, FILE_GAP_RECORD_SIZE),
            (SECTION_DECISION_GAPS, DECISION_GAP_RECORD_SIZE),
            (SECTION_DIMENSIONS, DIMENSION_RECORD_SIZE),
            (SECTION_PROJECTIONS, PROJECTION_RECORD_SIZE),
            (SECTION_SCOPE_ENTRIES, SCOPE_ENTRY_RECORD_SIZE),
            (SECTION_CONFIDENCE, CONFIDENCE_RECORD_SIZE),
            (SECTION_LINES, LINE_RECORD_SIZE),
            (SECTION_TEST_SUMMARIES, TEST_SUMMARY_RECORD_SIZE),
            (SECTION_PHASE_SUMMARIES, PHASE_SUMMARY_RECORD_SIZE),
            (SECTION_ANCHORS, ANCHOR_RECORD_SIZE),
            (SECTION_TEST_RETRIES, TEST_RETRY_RECORD_SIZE),
            (SECTION_TEST_ATTEMPTS, TEST_ATTEMPT_RECORD_SIZE),
            (SECTION_TEST_LINES, TEST_LINE_RECORD_SIZE),
            (SECTION_TEST_HITS, TEST_HIT_RECORD_SIZE),
            (SECTION_TEST_DECISIONS, TEST_DECISION_RECORD_SIZE),
            (SECTION_TEST_VECTORS, TEST_VECTOR_RECORD_SIZE),
            (SECTION_VECTOR_VALUES, 1),
            (SECTION_HIT_METADATA, HIT_METADATA_RECORD_SIZE),
            (SECTION_DECISION_METADATA, DECISION_METADATA_RECORD_SIZE),
            (SECTION_DECISION_DETAILS, DECISION_DETAIL_RECORD_SIZE),
            (
                SECTION_DECISION_VECTOR_OBSERVATIONS,
                DECISION_VECTOR_OBSERVATION_RECORD_SIZE,
            ),
            (SECTION_DECISION_CONDITIONS, DECISION_CONDITION_RECORD_SIZE),
            (SECTION_LIMITATIONS, LIMITATION_RECORD_SIZE),
            (SECTION_COVERAGE_MODEL, COVERAGE_MODEL_RECORD_SIZE),
        ] {
            if index.descriptor(kind)?.record_size as usize != size {
                return Err(CoverageIndexError::InvalidRecord("record size"));
            }
        }
        index.descriptor(SECTION_STRING_BYTES)?;
        if index.descriptor(SECTION_STRING_RELATIONS)?.record_size != 4 {
            return Err(CoverageIndexError::InvalidRecord(
                "string-relation record size",
            ));
        }
        Ok(Self { index })
    }

    pub fn model(&self) -> Result<IndexedCoverageModel, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_COVERAGE_MODEL)?;
        if descriptor.count != 1 {
            return Err(CoverageIndexError::InvalidRecord(
                "coverage model record count",
            ));
        }
        let record = self.index.record(SECTION_COVERAGE_MODEL, 0)?;
        if record[12..16].iter().any(|byte| *byte != 0) {
            return Err(CoverageIndexError::InvalidRecord(
                "coverage model reserved bytes",
            ));
        }
        Ok(IndexedCoverageModel {
            schema_version: COVERAGE_MODEL_SCHEMA_VERSION,
            variant: self.string(get_u32(record, 0))?,
            name: self.string(get_u32(record, 4))?,
            completeness_meaning: self.string(get_u32(record, 8))?,
            measured: self.relation_strings(get_u64(record, 16), get_u64(record, 24))?,
            not_measured: self.relation_strings(get_u64(record, 32), get_u64(record, 40))?,
        })
    }

    fn string(&self, id: u32) -> Result<String, CoverageIndexError> {
        let record = self.index.record(SECTION_STRINGS, u64::from(id))?;
        if record[12..].iter().any(|byte| *byte != 0) {
            return Err(CoverageIndexError::InvalidRecord("string reserved bytes"));
        }
        let offset = get_u64(record, 0);
        let length = u64::from(get_u32(record, 8));
        let value = self.index.bytes(SECTION_STRING_BYTES, offset, length)?;
        std::str::from_utf8(value)
            .map(str::to_owned)
            .map_err(|_| CoverageIndexError::InvalidUtf8)
    }

    pub fn summary(&self, view: CoverageViewId) -> Result<CoverageSummary, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_VIEW_SUMMARIES)?;
        if descriptor.count != 3 {
            return Err(CoverageIndexError::InvalidRecord("summary view count"));
        }
        let mut found = None;
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_VIEW_SUMMARIES, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            if record[3] != 0 || record[12..16].iter().any(|byte| *byte != 0) {
                return Err(CoverageIndexError::InvalidRecord("summary reserved bytes"));
            }
            self.string(get_u32(record, 4))?;
            self.string(get_u32(record, 8))?;
            let count = |offset: usize| -> Result<CoverageCount, CoverageIndexError> {
                let covered = usize::try_from(get_u64(record, offset))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?;
                let total = usize::try_from(get_u64(record, offset + 8))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?;
                if covered > total {
                    return Err(CoverageIndexError::InvalidRecord("covered exceeds total"));
                }
                Ok(CoverageCount {
                    covered,
                    total,
                    percentage: percentage(covered, total),
                })
            };
            let value = |offset: usize| -> Result<usize, CoverageIndexError> {
                usize::try_from(get_u64(record, offset))
                    .map_err(|_| CoverageIndexError::SizeOverflow)
            };
            let conditions = value(40)?;
            let covered_conditions = value(48)?;
            if covered_conditions > conditions {
                return Err(CoverageIndexError::InvalidRecord(
                    "covered conditions exceed total",
                ));
            }
            let decisions = value(16)?;
            let executed_decisions = value(24)?;
            let covered_decisions = value(32)?;
            if covered_decisions > executed_decisions || executed_decisions > decisions {
                return Err(CoverageIndexError::InvalidRecord("decision count ordering"));
            }
            let summary = CoverageSummary {
                unmeasured_obligations: None,
                exact_fraction_pct: None,
                decisions,
                executed_decisions,
                covered_decisions,
                conditions,
                covered_conditions,
                condition_coverage_pct: percentage(covered_conditions, conditions),
                lines: count(56)?,
                statements: count(72)?,
                functions: count(88)?,
                branches: count(104)?,
                decision_outcomes: count(120)?,
                condition_outcomes: count(136)?,
                value_selections: count(152)?,
                coverage_complete: bool_field(record[1])?,
                completeness_blocked: match record[2] {
                    0 => None,
                    1 => Some(false),
                    2 => Some(true),
                    _ => return Err(CoverageIndexError::InvalidRecord("optional boolean")),
                },
            };
            if found.replace(summary).is_some() {
                return Err(CoverageIndexError::InvalidRecord("duplicate coverage view"));
            }
        }
        found.ok_or(CoverageIndexError::InvalidRecord("missing coverage view"))
    }

    pub fn file_gaps(
        &self,
        view: CoverageViewId,
        kind: Option<&str>,
        runner: Option<&str>,
    ) -> Result<Vec<IndexedFileGap>, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_FILE_GAPS)?;
        let mut gaps = Vec::new();
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_FILE_GAPS, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            if record[1..4].iter().any(|byte| *byte != 0)
                || record[60..64].iter().any(|byte| *byte != 0)
            {
                return Err(CoverageIndexError::InvalidRecord("file-gap reserved bytes"));
            }
            let number = |offset: usize| -> Result<usize, CoverageIndexError> {
                usize::try_from(get_u64(record, offset))
                    .map_err(|_| CoverageIndexError::SizeOverflow)
            };
            let mask = get_u32(record, 56);
            if mask & !15 != 0 {
                return Err(CoverageIndexError::InvalidRecord("limitation mask"));
            }
            let measurement_limitations = number(48)?;
            if (measurement_limitations == 0) != (mask == 0) {
                return Err(CoverageIndexError::InvalidRecord(
                    "limitation count and kinds disagree",
                ));
            }
            let uncovered_lines = number(8)?;
            let uncovered_statements = number(16)?;
            let uncovered_functions = number(24)?;
            let missing_branches = number(32)?;
            let missing_mcdc_conditions = number(40)?;
            let score = number(64)?;
            let expected_score = uncovered_lines
                + uncovered_functions * 2
                + missing_branches * 2
                + missing_mcdc_conditions * 3
                + measurement_limitations * 3;
            if score != expected_score {
                return Err(CoverageIndexError::InvalidRecord("file-gap score"));
            }
            let record_kind = self.optional_string(get_u32(record, 72))?;
            let record_runner = self.optional_string(get_u32(record, 76))?;
            if record_kind.as_deref() != kind || record_runner.as_deref() != runner {
                continue;
            }
            let mut limitation_kinds = Vec::new();
            for (bit, kind) in [
                (1, "dynamic-code"),
                (2, "semantic-safety"),
                (4, "source-scope"),
                (8, "unknown"),
            ] {
                if mask & bit != 0 {
                    limitation_kinds.push(kind.into());
                }
            }
            let totals = IndexedGapDimensions {
                lines: number(160)?,
                statements: number(168)?,
                functions: number(176)?,
                branches: number(184)?,
                mcdc_conditions: number(192)?,
            };
            let covered = |total: usize, missing: usize| {
                total
                    .checked_sub(missing)
                    .ok_or(CoverageIndexError::InvalidRecord("file-gap totals"))
            };
            gaps.push(IndexedFileGap {
                view,
                file: self.string(get_u32(record, 4))?,
                uncovered_lines,
                uncovered_statements,
                uncovered_functions,
                missing_branches,
                missing_mcdc_conditions,
                measurement_limitations,
                limitation_kinds,
                covered_by_other_tests: IndexedGapDimensions {
                    lines: number(80)?,
                    statements: number(88)?,
                    functions: number(96)?,
                    branches: number(104)?,
                    mcdc_conditions: number(112)?,
                },
                uncovered_everywhere: IndexedGapDimensions {
                    lines: number(120)?,
                    statements: number(128)?,
                    functions: number(136)?,
                    branches: number(144)?,
                    mcdc_conditions: number(152)?,
                },
                totals: totals.clone(),
                covered: IndexedGapDimensions {
                    lines: covered(totals.lines, uncovered_lines)?,
                    statements: covered(totals.statements, uncovered_statements)?,
                    functions: covered(totals.functions, uncovered_functions)?,
                    branches: covered(totals.branches, missing_branches)?,
                    mcdc_conditions: covered(totals.mcdc_conditions, missing_mcdc_conditions)?,
                },
                score,
            });
        }
        gaps.sort_by(|left, right| {
            right
                .score
                .cmp(&left.score)
                .then_with(|| left.file.cmp(&right.file))
        });
        Ok(gaps)
    }

    fn optional_string(&self, id: u32) -> Result<Option<String>, CoverageIndexError> {
        if id == NO_STRING {
            Ok(None)
        } else {
            self.string(id).map(Some)
        }
    }

    fn relation_ids(&self, offset: u64, count: u64) -> Result<Vec<Id>, CoverageIndexError> {
        Ok(self
            .relation_strings(offset, count)?
            .into_iter()
            .map(Id::from)
            .collect())
    }

    fn relation_strings(&self, offset: u64, count: u64) -> Result<Vec<String>, CoverageIndexError> {
        let end = offset
            .checked_add(count)
            .ok_or(CoverageIndexError::SizeOverflow)?;
        let descriptor = self.index.descriptor(SECTION_STRING_RELATIONS)?;
        if end > descriptor.count {
            return Err(CoverageIndexError::InvalidRecord("string relation range"));
        }
        (offset..end)
            .map(|index| {
                let record = self.index.record(SECTION_STRING_RELATIONS, index)?;
                self.string(get_u32(record, 0))
            })
            .collect()
    }

    pub fn projection(
        &self,
        view: CoverageViewId,
        kind: Option<&str>,
        runner: Option<&str>,
    ) -> Result<IndexedProjection, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_PROJECTIONS)?;
        let mut found = None;
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_PROJECTIONS, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            if record[38..40].iter().any(|byte| *byte != 0)
                || record[518..520].iter().any(|byte| *byte != 0)
            {
                return Err(CoverageIndexError::InvalidRecord(
                    "projection reserved bytes",
                ));
            }
            let record_kind = self.optional_string(get_u32(record, 4))?;
            let record_runner = self.optional_string(get_u32(record, 8))?;
            if record_kind.as_deref() != kind || record_runner.as_deref() != runner {
                continue;
            }
            if found.is_some() {
                return Err(CoverageIndexError::InvalidRecord(
                    "duplicate coverage projection",
                ));
            }
            let number = |offset: usize| -> Result<usize, CoverageIndexError> {
                usize::try_from(get_u64(record, offset))
                    .map_err(|_| CoverageIndexError::SizeOverflow)
            };
            let limitations = number(192)?;
            let evidence_corruptions = number(200)?;
            let blocking = number(208)?;
            let declared = number(528)?;
            if blocking + declared != limitations + evidence_corruptions {
                return Err(CoverageIndexError::InvalidRecord(
                    "measurement blocking count",
                ));
            }
            let mut transport_values = (0..8)
                .map(|index| number(408 + index * 8))
                .collect::<Result<Vec<_>, _>>()?;
            transport_values.push(number(536)?);
            let transport = if bool_field(record[1])? {
                Some(TransportStats {
                    processes: transport_values[0],
                    child_launches: transport_values[1],
                    remote_launches: transport_values[2],
                    workspace_capabilities: transport_values[3],
                    guest_processes: transport_values[8],
                    scoped_server_records: transport_values[4],
                    background_server_records: transport_values[5],
                    corrupt_records: transport_values[6],
                    corrupt_files: transport_values[7],
                })
            } else {
                if transport_values.iter().any(|value| *value != 0) {
                    return Err(CoverageIndexError::InvalidRecord("transport presence flag"));
                }
                None
            };
            let has_scope = bool_field(record[2])?;
            let scope_kind = match record[3] {
                0 => None,
                1 => Some((ScopeKind::SourceDiscovery, "source-discovery")),
                2 => Some((ScopeKind::Compiler, "compiler")),
                _ => return Err(CoverageIndexError::InvalidRecord("coverage scope kind")),
            };
            let scope_mode = self.optional_string(get_u32(record, 16))?;
            let scope_language = self.optional_string(get_u32(record, 504))?;
            let scope_model = self.optional_string(get_u32(record, 508))?;
            let scope_unit = self.optional_string(get_u32(record, 512))?;
            let has_measurement_complete = bool_field(record[516])?;
            let measurement_complete = bool_field(record[517])?;
            if !has_measurement_complete && measurement_complete {
                return Err(CoverageIndexError::InvalidRecord(
                    "scope measurement completeness flag",
                ));
            }
            if has_scope
                != (scope_kind.is_some() && scope_language.is_some() && scope_model.is_some())
            {
                return Err(CoverageIndexError::InvalidRecord("scope presence flag"));
            }
            let source_scope = if let Some((kind, kind_name)) = scope_kind {
                match kind {
                    ScopeKind::SourceDiscovery => {
                        if scope_mode.is_none() || scope_unit.is_some() || has_measurement_complete
                        {
                            return Err(CoverageIndexError::InvalidRecord(
                                "source-discovery scope shape",
                            ));
                        }
                    }
                    ScopeKind::Compiler => {
                        if scope_mode.is_some()
                            || scope_unit.is_none()
                            || !has_measurement_complete
                            || get_u32(record, 32) != 0
                            || number(480)? != 0
                            || number(488)? != 0
                            || number(496)? != 0
                        {
                            return Err(CoverageIndexError::InvalidRecord("compiler scope shape"));
                        }
                    }
                }
                Some(IndexedSourceScope {
                    kind: kind_name.into(),
                    language: scope_language.expect("validated scope language"),
                    model: scope_model.expect("validated scope model"),
                    mode: scope_mode,
                    roots: self
                        .relation_strings(get_u64(record, 24), u64::from(get_u32(record, 32)))?,
                    unit: scope_unit,
                    measurement_complete: has_measurement_complete.then_some(measurement_complete),
                    included: number(480)?,
                    excluded: number(488)?,
                    ambiguous: number(496)?,
                })
            } else {
                if get_u32(record, 32) != 0
                    || scope_mode.is_some()
                    || scope_language.is_some()
                    || scope_model.is_some()
                    || scope_unit.is_some()
                    || has_measurement_complete
                    || number(480)? != 0
                    || number(488)? != 0
                    || number(496)? != 0
                {
                    return Err(CoverageIndexError::InvalidRecord("absent scope data"));
                }
                None
            };
            let empty_evidence_tests = number(472)?;
            let first_empty_evidence_test = self.optional_string(get_u32(record, 20))?;
            if (empty_evidence_tests == 0) != first_empty_evidence_test.is_none() {
                return Err(CoverageIndexError::InvalidRecord(
                    "empty-evidence diagnostic identity",
                ));
            }
            found = Some(IndexedProjection {
                view,
                kind: record_kind,
                runner: record_runner,
                generated_at: self.string(get_u32(record, 12))?,
                summary: decode_summary(record, 36, 40)?,
                measurement: IndexedMeasurement {
                    complete: blocking == 0,
                    limitations,
                    evidence_corruptions,
                    blocking,
                    declared,
                    files: number(216)?,
                    by_kind: IndexedMeasurementKinds {
                        dynamic_code: number(224)?,
                        semantic_safety: number(232)?,
                        source_scope: number(240)?,
                    },
                },
                attribution: IndexedAttribution {
                    browser_explicit: number(248)?,
                    browser_fallback: number(256)?,
                    server_explicit: number(264)?,
                    server_fallback: number(272)?,
                },
                transport,
                empty_evidence_tests,
                first_empty_evidence_test,
                confidence: IndexedSummaryConfidence {
                    lines: IndexedConfidenceLines {
                        unexecuted: number(280)?,
                        executed: number(288)?,
                        action: number(296)?,
                        asserted: number(304)?,
                    },
                    assertion_covered_mcdc_conditions: number(312)?,
                },
                files_with_gaps: number(320)?,
                files_with_coverage_gaps: number(328)?,
                tests: number(336)?,
                setups: number(344)?,
                test_outcomes: {
                    let mut outcomes = IndexedOutcomeCounts {
                        passed: number(352)?,
                        failed: number(360)?,
                        flaky: number(368)?,
                        skipped: number(376)?,
                        timed_out: number(384)?,
                        interrupted: number(392)?,
                        unknown: number(400)?,
                        unstarted: number(520)?,
                        todo: 0,
                    };
                    // The projection record has no slot left, so todo is the
                    // tests no other outcome counts; zero in older indexes.
                    outcomes.todo = number(336)?.saturating_sub(
                        outcomes.passed
                            + outcomes.failed
                            + outcomes.flaky
                            + outcomes.skipped
                            + outcomes.timed_out
                            + outcomes.interrupted
                            + outcomes.unknown
                            + outcomes.unstarted,
                    );
                    outcomes
                },
                source_scope,
            });
        }
        found.ok_or(CoverageIndexError::InvalidRecord(
            "missing coverage projection",
        ))
    }

    pub fn decision_gaps(
        &self,
        view: CoverageViewId,
        kind: Option<&str>,
        runner: Option<&str>,
        file: &str,
    ) -> Result<Vec<IndexedDecisionGap>, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_DECISION_GAPS)?;
        let mut decisions = Vec::new();
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_DECISION_GAPS, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            if record[1..4].iter().any(|byte| *byte != 0)
                || record[28..32].iter().any(|byte| *byte != 0)
                || record[64..].iter().any(|byte| *byte != 0)
            {
                return Err(CoverageIndexError::InvalidRecord(
                    "decision-gap reserved bytes",
                ));
            }
            let record_kind = self.optional_string(get_u32(record, 4))?;
            let record_runner = self.optional_string(get_u32(record, 8))?;
            if record_kind.as_deref() != kind || record_runner.as_deref() != runner {
                continue;
            }
            let record_file = self.string(get_u32(record, 16))?;
            if record_file != file {
                continue;
            }
            let number = |offset: usize| -> Result<usize, CoverageIndexError> {
                usize::try_from(get_u64(record, offset))
                    .map_err(|_| CoverageIndexError::SizeOverflow)
            };
            let conditions = number(48)?;
            let missing_conditions = number(56)?;
            if conditions == 0 || missing_conditions > conditions {
                return Err(CoverageIndexError::InvalidRecord(
                    "decision condition counts",
                ));
            }
            decisions.push(IndexedDecisionGap {
                view,
                file: record_file,
                id: self.string(get_u32(record, 12))?,
                line: number(32)?,
                column: number(40)?,
                kind: self.string(get_u32(record, 20))?,
                conditions,
                missing_conditions,
                source: self.string(get_u32(record, 24))?,
            });
        }
        Ok(decisions)
    }

    pub fn dimensions(
        &self,
        view: CoverageViewId,
        dimension: CoverageDimension,
    ) -> Result<Vec<IndexedDimensionCoverage>, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_DIMENSIONS)?;
        let mut values = Vec::new();
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_DIMENSIONS, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            let record_dimension = match record[1] {
                0 => CoverageDimension::Kind,
                1 => CoverageDimension::Runner,
                2 => CoverageDimension::Tests,
                _ => return Err(CoverageIndexError::InvalidRecord("dimension type")),
            };
            if record_dimension != dimension {
                continue;
            }
            if record[26..32].iter().any(|byte| *byte != 0) {
                return Err(CoverageIndexError::InvalidRecord(
                    "dimension reserved bytes",
                ));
            }
            let name = self.string(get_u32(record, 4))?;
            values.push(IndexedDimensionCoverage {
                kind: (dimension == CoverageDimension::Kind).then(|| name.clone()),
                runner: (dimension == CoverageDimension::Runner).then_some(name),
                tests: usize::try_from(get_u64(record, 8))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?,
                setups: usize::try_from(get_u64(record, 16))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?,
                attributed: usize::try_from(get_u64(record, 184))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?,
                summary: decode_summary(record, 24, 32)?,
            });
        }
        Ok(values)
    }

    pub fn scope_entries(
        &self,
        view: CoverageViewId,
    ) -> Result<Vec<IndexedScopeEntry>, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_SCOPE_ENTRIES)?;
        let mut entries = Vec::new();
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_SCOPE_ENTRIES, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            if record[2..4].iter().any(|byte| *byte != 0)
                || record[28..].iter().any(|byte| *byte != 0)
            {
                return Err(CoverageIndexError::InvalidRecord(
                    "source-scope reserved bytes",
                ));
            }
            let status = match record[1] {
                0 => "included",
                1 => "excluded",
                2 => "ambiguous",
                _ => return Err(CoverageIndexError::InvalidRecord("source-scope status")),
            };
            let measurement_limitations = usize::try_from(get_u64(record, 16))
                .map_err(|_| CoverageIndexError::SizeOverflow)?;
            let mask = get_u32(record, 24);
            if mask & !7 != 0 || (measurement_limitations == 0) != (mask == 0) {
                return Err(CoverageIndexError::InvalidRecord(
                    "source-scope limitation annotation",
                ));
            }
            let mut limitation_kinds = Vec::new();
            for (bit, kind) in [
                (1, "dynamic-code"),
                (2, "semantic-safety"),
                (4, "source-scope"),
            ] {
                if mask & bit != 0 {
                    limitation_kinds.push(kind.into());
                }
            }
            entries.push(IndexedScopeEntry {
                file: self.string(get_u32(record, 4))?,
                status: status.into(),
                reason: self.string(get_u32(record, 8))?,
                package_root: self.optional_string(get_u32(record, 12))?,
                measurement_limitations,
                limitation_kinds,
            });
        }
        Ok(entries)
    }

    fn confidence(
        &self,
        index: u64,
    ) -> Result<crate::coverage_report::CoverageConfidence, CoverageIndexError> {
        let record = self.index.record(SECTION_CONFIDENCE, index)?;
        if record[2..8].iter().any(|byte| *byte != 0)
            || record[72..].iter().any(|byte| *byte != 0)
            || record[1] & !15 != 0
        {
            return Err(CoverageIndexError::InvalidRecord("confidence record"));
        }
        let values = (0..4)
            .map(|index| {
                self.relation_strings(
                    get_u64(record, 8 + index * 16),
                    get_u64(record, 16 + index * 16),
                )
            })
            .collect::<Result<Vec<_>, _>>()?;
        Ok(crate::coverage_report::CoverageConfidence {
            level: match record[0] {
                0 => "unexecuted",
                1 => "executed",
                2 => "action",
                3 => "asserted",
                _ => return Err(CoverageIndexError::InvalidRecord("confidence level")),
            }
            .into(),
            setup_only: record[1] & 1 != 0,
            background_only: record[1] & 2 != 0,
            asserted: record[1] & 4 != 0,
            e2e: record[1] & 8 != 0,
            tests: values[0].iter().map(Id::from).collect(),
            asserted_tests: values[1].iter().map(Id::from).collect(),
            runners: values[2].clone(),
            kinds: values[3].clone(),
        })
    }

    pub fn line(
        &self,
        view: CoverageViewId,
        file: &str,
        line: usize,
    ) -> Result<Option<IndexedLine>, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_LINES)?;
        let mut found = None;
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_LINES, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            if record[3] != 0 || record[56..].iter().any(|byte| *byte != 0) {
                return Err(CoverageIndexError::InvalidRecord("line record"));
            }
            let record_file = self.string(get_u32(record, 4))?;
            let record_line = usize::try_from(get_u64(record, 8))
                .map_err(|_| CoverageIndexError::SizeOverflow)?;
            if record_file != file || record_line != line {
                continue;
            }
            if found.is_some() {
                return Err(CoverageIndexError::InvalidRecord("duplicate line"));
            }
            found = Some(IndexedLine {
                file: record_file,
                line: record_line,
                covered: bool_field(record[1])?,
                measured: !bool_field(record[2])?,
                tests: self.relation_strings(get_u64(record, 16), get_u64(record, 24))?,
                phases: self.relation_strings(get_u64(record, 32), get_u64(record, 40))?,
                confidence: self.confidence(get_u64(record, 48))?,
            });
        }
        Ok(found)
    }

    pub fn lines(&self, view: CoverageViewId) -> Result<Vec<IndexedLine>, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_LINES)?;
        let mut lines = Vec::new();
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_LINES, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            if record[3] != 0 || record[56..].iter().any(|byte| *byte != 0) {
                return Err(CoverageIndexError::InvalidRecord("line record"));
            }
            lines.push(IndexedLine {
                file: self.string(get_u32(record, 4))?,
                line: usize::try_from(get_u64(record, 8))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?,
                covered: bool_field(record[1])?,
                measured: !bool_field(record[2])?,
                tests: self.relation_strings(get_u64(record, 16), get_u64(record, 24))?,
                phases: self.relation_strings(get_u64(record, 32), get_u64(record, 40))?,
                confidence: self.confidence(get_u64(record, 48))?,
            });
        }
        Ok(lines)
    }

    pub fn test_summaries(
        &self,
        view: CoverageViewId,
    ) -> Result<Vec<IndexedTestSummary>, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_TEST_SUMMARIES)?;
        let mut tests = Vec::new();
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_TEST_SUMMARIES, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            if record[36..].iter().any(|byte| *byte != 0) {
                return Err(CoverageIndexError::InvalidRecord("test summary record"));
            }
            tests.push(IndexedTestSummary {
                id: self.string(get_u32(record, 4))?,
                name: self.string(get_u32(record, 8))?,
                file: self.optional_string(get_u32(record, 12))?,
                title: self.optional_string(get_u32(record, 16))?,
                role: match record[1] {
                    0 => "test",
                    1 => "setup",
                    2 => "background",
                    _ => return Err(CoverageIndexError::InvalidRecord("test role")),
                }
                .into(),
                attribution: match record[3] {
                    0 => crate::coverage_report::ATTRIBUTION_EXACT,
                    1 => crate::coverage_report::ATTRIBUTION_RUN_WIDE,
                    2 => crate::coverage_report::ATTRIBUTION_PARTIAL,
                    _ => return Err(CoverageIndexError::InvalidRecord("test attribution")),
                }
                .into(),
                outcome: match record[2] {
                    0 => "passed",
                    1 => "failed",
                    2 => "flaky",
                    3 => "skipped",
                    4 => "timedOut",
                    5 => "interrupted",
                    6 => "unknown",
                    7 => "unstarted",
                    8 => "todo",
                    _ => return Err(CoverageIndexError::InvalidRecord("test outcome")),
                }
                .into(),
                provenance: crate::coverage_report::TestProvenance {
                    runner: self.string(get_u32(record, 20))?,
                    kind: self.string(get_u32(record, 24))?,
                    project: self.optional_string(get_u32(record, 28))?,
                    source: self.string(get_u32(record, 32))?,
                },
            });
        }
        Ok(tests)
    }

    fn test_vector(&self, index: u64) -> Result<McdcVector, CoverageIndexError> {
        let record = self.index.record(SECTION_TEST_VECTORS, index)?;
        if record[1..8].iter().any(|byte| *byte != 0) {
            return Err(CoverageIndexError::InvalidRecord("test vector record"));
        }
        let offset = get_u64(record, 8);
        let count = get_u64(record, 16);
        let descriptor = self.index.descriptor(SECTION_VECTOR_VALUES)?;
        let end = offset
            .checked_add(count)
            .ok_or(CoverageIndexError::InvalidRecord("vector value range"))?;
        if end > descriptor.count {
            return Err(CoverageIndexError::InvalidRecord("vector value range"));
        }
        let mut values = Vec::with_capacity(
            usize::try_from(count).map_err(|_| CoverageIndexError::SizeOverflow)?,
        );
        for index in offset..end {
            values.push(match self.index.record(SECTION_VECTOR_VALUES, index)?[0] {
                0 => None,
                1 => Some(false),
                2 => Some(true),
                _ => return Err(CoverageIndexError::InvalidRecord("vector value")),
            });
        }
        Ok(McdcVector {
            values,
            outcome: bool_field(record[0])?,
        })
    }

    pub fn test_details(
        &self,
        view: CoverageViewId,
    ) -> Result<Vec<IndexedTestDetail>, CoverageIndexError> {
        let summaries = self.test_summaries(view)?;
        let positions = summaries
            .iter()
            .enumerate()
            .map(|(index, test)| (test.id.clone(), index))
            .collect::<HashMap<_, _>>();
        if positions.len() != summaries.len() {
            return Err(CoverageIndexError::InvalidRecord("duplicate test summary"));
        }
        let mut details = summaries
            .into_iter()
            .map(|summary| IndexedTestDetail {
                summary,
                retries: Vec::new(),
                attempts: Vec::new(),
                hits: Vec::new(),
                decisions: Vec::new(),
                lines: Vec::new(),
            })
            .collect::<Vec<_>>();
        let position = |record: &[u8]| -> Result<Option<usize>, CoverageIndexError> {
            if CoverageViewId::try_from(record[0])? != view {
                return Ok(None);
            }
            let id = self.string(get_u32(record, 4))?;
            positions
                .get(&id)
                .copied()
                .map(Some)
                .ok_or(CoverageIndexError::InvalidRecord("unknown test relation"))
        };
        let descriptor = self.index.descriptor(SECTION_TEST_RETRIES)?;
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_TEST_RETRIES, index)?;
            if record[1..4].iter().any(|byte| *byte != 0) {
                return Err(CoverageIndexError::InvalidRecord("test retry record"));
            }
            if let Some(position) = position(record)? {
                details[position].retries.push(
                    usize::try_from(get_u64(record, 8))
                        .map_err(|_| CoverageIndexError::SizeOverflow)?,
                );
            }
        }
        let descriptor = self.index.descriptor(SECTION_TEST_ATTEMPTS)?;
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_TEST_ATTEMPTS, index)?;
            if record[1..4].iter().any(|byte| *byte != 0) {
                return Err(CoverageIndexError::InvalidRecord("test attempt record"));
            }
            if let Some(position) = position(record)? {
                details[position]
                    .attempts
                    .push(crate::coverage_report::TestAttempt {
                        retry: usize::try_from(get_u64(record, 8))
                            .map_err(|_| CoverageIndexError::SizeOverflow)?,
                        status: self.string(get_u32(record, 16))?,
                        expected_status: self.optional_string(get_u32(record, 20))?,
                    });
            }
        }
        let descriptor = self.index.descriptor(SECTION_TEST_LINES)?;
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_TEST_LINES, index)?;
            if record[1..4].iter().any(|byte| *byte != 0)
                || record[12..16].iter().any(|byte| *byte != 0)
            {
                return Err(CoverageIndexError::InvalidRecord("test line record"));
            }
            if let Some(position) = position(record)? {
                details[position]
                    .lines
                    .push(crate::coverage_report::SourceLine {
                        file: self.string(get_u32(record, 8))?.into(),
                        line: usize::try_from(get_u64(record, 16))
                            .map_err(|_| CoverageIndexError::SizeOverflow)?,
                    });
            }
        }
        let descriptor = self.index.descriptor(SECTION_TEST_HITS)?;
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_TEST_HITS, index)?;
            if record[1..4].iter().any(|byte| *byte != 0)
                || record[12..].iter().any(|byte| *byte != 0)
            {
                return Err(CoverageIndexError::InvalidRecord("test hit record"));
            }
            if let Some(position) = position(record)? {
                details[position]
                    .hits
                    .push(self.string(get_u32(record, 8))?);
            }
        }
        let descriptor = self.index.descriptor(SECTION_TEST_DECISIONS)?;
        let vectors = self.index.descriptor(SECTION_TEST_VECTORS)?.count;
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_TEST_DECISIONS, index)?;
            if record[1..4].iter().any(|byte| *byte != 0)
                || record[12..16].iter().any(|byte| *byte != 0)
            {
                return Err(CoverageIndexError::InvalidRecord("test decision record"));
            }
            if let Some(position) = position(record)? {
                let offset = get_u64(record, 16);
                let count = get_u64(record, 24);
                let end = offset
                    .checked_add(count)
                    .ok_or(CoverageIndexError::InvalidRecord("test vector range"))?;
                if end > vectors {
                    return Err(CoverageIndexError::InvalidRecord("test vector range"));
                }
                let mut observed = Vec::with_capacity(
                    usize::try_from(count).map_err(|_| CoverageIndexError::SizeOverflow)?,
                );
                for vector in offset..end {
                    observed.push(self.test_vector(vector)?);
                }
                details[position]
                    .decisions
                    .push(crate::coverage_report::TestDecisionResult {
                        id: self.string(get_u32(record, 8))?,
                        vectors: observed,
                    });
            }
        }
        Ok(details)
    }

    pub fn hit_metadata(
        &self,
        view: CoverageViewId,
    ) -> Result<Vec<IndexedHitMetadata>, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_HIT_METADATA)?;
        let mut metadata = Vec::new();
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_HIT_METADATA, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            if record[2..4].iter().any(|byte| *byte != 0)
                || record[12..16].iter().any(|byte| *byte != 0)
            {
                return Err(CoverageIndexError::InvalidRecord("hit metadata record"));
            }
            let obligation = match record[1] {
                0 => "statement",
                1 => "function",
                2 => "branch",
                _ => return Err(CoverageIndexError::InvalidRecord("hit obligation")),
            };
            let metadata_label = self.optional_string(get_u32(record, 36))?;
            metadata.push(IndexedHitMetadata {
                id: self.string(get_u32(record, 4))?,
                obligation: obligation.into(),
                file: self.string(get_u32(record, 8))?,
                line: usize::try_from(get_u64(record, 16))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?,
                column: usize::try_from(get_u64(record, 24))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?,
                branch_kind: self.optional_string(get_u32(record, 32))?,
                label: (obligation != "branch")
                    .then_some(metadata_label.clone())
                    .flatten(),
                alternative: self.optional_string(get_u32(record, 40))?,
                parent_id: (obligation == "branch").then_some(metadata_label).flatten(),
                source: self.string(get_u32(record, 44))?,
                tests: self.relation_strings(get_u64(record, 48), get_u64(record, 56))?,
            });
        }
        Ok(metadata)
    }

    pub fn limitations(
        &self,
        view: CoverageViewId,
    ) -> Result<Vec<IndexedLimitation>, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_LIMITATIONS)?;
        let mut limitations = Vec::new();
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_LIMITATIONS, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            if record[2..4].iter().any(|byte| *byte != 0)
                || record[40..].iter().any(|byte| *byte != 0)
            {
                return Err(CoverageIndexError::InvalidRecord("limitation record"));
            }
            limitations.push(IndexedLimitation {
                id: self.string(get_u32(record, 4))?,
                kind: self.string(get_u32(record, 8))?,
                file: self.string(get_u32(record, 12))?,
                source: self.string(get_u32(record, 16))?,
                reason: self.string(get_u32(record, 20))?,
                line: usize::try_from(get_u64(record, 24))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?,
                column: usize::try_from(get_u64(record, 32))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?,
                blocking: bool_field(record[1])?,
            });
        }
        Ok(limitations)
    }

    pub fn decision_metadata(
        &self,
        view: CoverageViewId,
    ) -> Result<Vec<crate::coverage_report::DecisionMeta>, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_DECISION_METADATA)?;
        let mut metadata = Vec::new();
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_DECISION_METADATA, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            if record[1..4].iter().any(|byte| *byte != 0)
                || record[20..24].iter().any(|byte| *byte != 0)
                || record[56..].iter().any(|byte| *byte != 0)
            {
                return Err(CoverageIndexError::InvalidRecord(
                    "decision metadata record",
                ));
            }
            metadata.push(crate::coverage_report::DecisionMeta {
                id: self.string(get_u32(record, 4))?,
                file: self.string(get_u32(record, 8))?,
                source: self.string(get_u32(record, 12))?,
                kind: self.string(get_u32(record, 16))?,
                line: usize::try_from(get_u64(record, 24))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?,
                column: usize::try_from(get_u64(record, 32))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?,
                conditions: self.relation_strings(get_u64(record, 40), get_u64(record, 48))?,
            });
        }
        Ok(metadata)
    }

    fn decision_vector_observation(
        &self,
        index: u64,
    ) -> Result<crate::coverage_report::VectorObservation, CoverageIndexError> {
        let record = self
            .index
            .record(SECTION_DECISION_VECTOR_OBSERVATIONS, index)?;
        Ok(crate::coverage_report::VectorObservation {
            confidence: self.confidence(get_u64(record, 0))?,
            vector: self.test_vector(get_u64(record, 8))?,
            tests: self.relation_ids(get_u64(record, 16), get_u64(record, 24))?,
            phases: self.relation_ids(get_u64(record, 32), get_u64(record, 40))?,
            explicit_phases: self.relation_ids(get_u64(record, 48), get_u64(record, 56))?,
        })
    }

    fn decision_condition(
        &self,
        index: u64,
    ) -> Result<crate::coverage_report::ConditionResult, CoverageIndexError> {
        let record = self.index.record(SECTION_DECISION_CONDITIONS, index)?;
        if record[0] & !7 != 0 || record[1..4].iter().any(|byte| *byte != 0) {
            return Err(CoverageIndexError::InvalidRecord(
                "decision condition record",
            ));
        }
        let has_witness = record[0] & 4 != 0;
        let witness = if has_witness {
            Some([
                self.test_vector(get_u64(record, 16))?,
                self.test_vector(get_u64(record, 24))?,
            ])
        } else {
            None
        };
        let first_tests = self.relation_strings(get_u64(record, 32), get_u64(record, 40))?;
        let second_tests = self.relation_strings(get_u64(record, 48), get_u64(record, 56))?;
        if !has_witness && (!first_tests.is_empty() || !second_tests.is_empty()) {
            return Err(CoverageIndexError::InvalidRecord(
                "condition witness tests without witness",
            ));
        }
        Ok(crate::coverage_report::ConditionResult {
            index: usize::try_from(get_u64(record, 8))
                .map_err(|_| CoverageIndexError::SizeOverflow)?,
            source: self.string(get_u32(record, 4))?,
            covered: record[0] & 1 != 0,
            assertion_covered: record[0] & 2 != 0,
            witness,
            witness_tests: has_witness.then_some([
                first_tests.into_iter().map(Id::from).collect(),
                second_tests.into_iter().map(Id::from).collect(),
            ]),
        })
    }

    pub fn decision_details(
        &self,
        view: CoverageViewId,
    ) -> Result<Vec<crate::coverage_report::DecisionResult>, CoverageIndexError> {
        let metadata = self
            .decision_metadata(view)?
            .into_iter()
            .map(|meta| (meta.id.clone(), meta))
            .collect::<HashMap<_, _>>();
        let descriptor = self.index.descriptor(SECTION_DECISION_DETAILS)?;
        let observation_count = self
            .index
            .descriptor(SECTION_DECISION_VECTOR_OBSERVATIONS)?
            .count;
        let condition_count = self.index.descriptor(SECTION_DECISION_CONDITIONS)?.count;
        let mut decisions = Vec::new();
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_DECISION_DETAILS, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            if record[1] & !3 != 0 || record[2..4].iter().any(|byte| *byte != 0) {
                return Err(CoverageIndexError::InvalidRecord("decision detail record"));
            }
            let executed = record[1] & 1 != 0;
            let covered = record[1] & 2 != 0;
            if covered && !executed {
                return Err(CoverageIndexError::InvalidRecord(
                    "covered unexecuted decision",
                ));
            }
            let range = |offset: usize,
                         available: u64,
                         label: &'static str|
             -> Result<std::ops::Range<u64>, CoverageIndexError> {
                let start = get_u64(record, offset);
                let count = get_u64(record, offset + 8);
                let end = start
                    .checked_add(count)
                    .ok_or(CoverageIndexError::InvalidRecord(label))?;
                if end > available {
                    return Err(CoverageIndexError::InvalidRecord(label));
                }
                Ok(start..end)
            };
            let observations = range(32, observation_count, "decision observation range")?
                .map(|index| self.decision_vector_observation(index))
                .collect::<Result<Vec<_>, _>>()?;
            let conditions = range(48, condition_count, "decision condition range")?
                .map(|index| self.decision_condition(index))
                .collect::<Result<Vec<_>, _>>()?;
            let id = self.string(get_u32(record, 4))?;
            let meta = metadata
                .get(&id)
                .cloned()
                .ok_or(CoverageIndexError::InvalidRecord(
                    "missing decision metadata",
                ))?;
            if conditions.len() != meta.conditions.len()
                || conditions.iter().enumerate().any(|(index, condition)| {
                    condition.index != index || condition.source != meta.conditions[index]
                })
            {
                return Err(CoverageIndexError::InvalidRecord(
                    "decision condition denominator",
                ));
            }
            decisions.push(crate::coverage_report::DecisionResult {
                meta,
                executed,
                covered,
                vectors: observations
                    .iter()
                    .map(|observation| observation.vector.clone())
                    .collect(),
                vector_observations: observations,
                conditions,
                tests: self.relation_ids(get_u64(record, 16), get_u64(record, 24))?,
                confidence: self.confidence(get_u64(record, 8))?,
            });
        }
        Ok(decisions)
    }

    pub fn phase_summaries(
        &self,
        view: CoverageViewId,
    ) -> Result<Vec<IndexedPhaseSummary>, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_PHASE_SUMMARIES)?;
        let mut phases = Vec::new();
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_PHASE_SUMMARIES, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            if record[1..4].iter().any(|byte| *byte != 0)
                || record[48..].iter().any(|byte| *byte != 0)
            {
                return Err(CoverageIndexError::InvalidRecord("phase summary record"));
            }
            phases.push(IndexedPhaseSummary {
                id: self.string(get_u32(record, 4))?,
                kind: self.string(get_u32(record, 8))?,
                operation: self.string(get_u32(record, 12))?,
                source: self.optional_string(get_u32(record, 16))?,
                test: self.string(get_u32(record, 20))?,
                status: self.optional_string(get_u32(record, 24))?,
                caused_by_phase_id: self.optional_string(get_u32(record, 28))?,
                lines: usize::try_from(get_u64(record, 32))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?,
                decisions: usize::try_from(get_u64(record, 40))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?,
            });
        }
        Ok(phases)
    }

    pub fn anchors(
        &self,
        view: CoverageViewId,
        file: &str,
        line: usize,
    ) -> Result<Vec<IndexedAnchor>, CoverageIndexError> {
        let descriptor = self.index.descriptor(SECTION_ANCHORS)?;
        let mut anchors = Vec::new();
        for index in 0..descriptor.count {
            let record = self.index.record(SECTION_ANCHORS, index)?;
            if CoverageViewId::try_from(record[0])? != view {
                continue;
            }
            if record[3] != 0 || record[12..16].iter().any(|byte| *byte != 0) {
                return Err(CoverageIndexError::InvalidRecord("anchor record"));
            }
            let record_file = self.string(get_u32(record, 8))?;
            let record_line = usize::try_from(get_u64(record, 16))
                .map_err(|_| CoverageIndexError::SizeOverflow)?;
            if record_file != file || record_line != line {
                continue;
            }
            let total = usize::try_from(get_u64(record, 32))
                .map_err(|_| CoverageIndexError::SizeOverflow)?;
            let covered_conditions = usize::try_from(get_u64(record, 40))
                .map_err(|_| CoverageIndexError::SizeOverflow)?;
            let (kind, conditions, covered_conditions) = match record[1] {
                0 => {
                    if total == 0 || covered_conditions > total {
                        return Err(CoverageIndexError::InvalidRecord(
                            "decision anchor conditions",
                        ));
                    }
                    ("decision", Some(total), Some(covered_conditions))
                }
                1 => ("branch", None, None),
                2 => ("statement", None, None),
                3 => ("function", None, None),
                _ => return Err(CoverageIndexError::InvalidRecord("anchor kind")),
            };
            if kind != "decision" && (total != 0 || covered_conditions.is_some()) {
                return Err(CoverageIndexError::InvalidRecord("anchor conditions"));
            }
            anchors.push(IndexedAnchor {
                kind: kind.into(),
                id: self.string(get_u32(record, 4))?,
                file: record_file,
                line: record_line,
                column: usize::try_from(get_u64(record, 24))
                    .map_err(|_| CoverageIndexError::SizeOverflow)?,
                covered: bool_field(record[2])?,
                conditions,
                covered_conditions,
                tests: self.relation_strings(get_u64(record, 48), get_u64(record, 56))?,
            });
        }
        anchors.sort_by_key(|anchor| anchor.column);
        Ok(anchors)
    }

    pub fn snapshot(&self) -> Result<IndexedCoverageSnapshot, CoverageIndexError> {
        Ok(IndexedCoverageSnapshot {
            all_summary: self.summary(CoverageViewId::All)?,
            passed_summary: self.summary(CoverageViewId::Passed)?,
            failed_summary: self.summary(CoverageViewId::Failed)?,
            all_files: self.file_gaps(CoverageViewId::All, None, None)?,
            passed_files: self.file_gaps(CoverageViewId::Passed, None, None)?,
            failed_files: self.file_gaps(CoverageViewId::Failed, None, None)?,
        })
    }
}

fn bool_field(value: u8) -> Result<bool, CoverageIndexError> {
    match value {
        0 => Ok(false),
        1 => Ok(true),
        _ => Err(CoverageIndexError::InvalidRecord("boolean")),
    }
}

fn decode_summary(
    record: &[u8],
    flags_offset: usize,
    base: usize,
) -> Result<CoverageSummary, CoverageIndexError> {
    let number = |offset: usize| -> Result<usize, CoverageIndexError> {
        usize::try_from(get_u64(record, offset)).map_err(|_| CoverageIndexError::SizeOverflow)
    };
    let count = |offset: usize| -> Result<CoverageCount, CoverageIndexError> {
        let covered = number(offset)?;
        let total = number(offset + 8)?;
        if covered > total {
            return Err(CoverageIndexError::InvalidRecord("covered exceeds total"));
        }
        Ok(CoverageCount {
            covered,
            total,
            percentage: percentage(covered, total),
        })
    };
    let decisions = number(base)?;
    let executed_decisions = number(base + 8)?;
    let covered_decisions = number(base + 16)?;
    let conditions = number(base + 24)?;
    let covered_conditions = number(base + 32)?;
    if covered_decisions > executed_decisions
        || executed_decisions > decisions
        || covered_conditions > conditions
    {
        return Err(CoverageIndexError::InvalidRecord("summary count ordering"));
    }
    Ok(CoverageSummary {
        unmeasured_obligations: None,
        exact_fraction_pct: None,
        decisions,
        executed_decisions,
        covered_decisions,
        conditions,
        covered_conditions,
        condition_coverage_pct: percentage(covered_conditions, conditions),
        lines: count(base + 40)?,
        statements: count(base + 56)?,
        functions: count(base + 72)?,
        branches: count(base + 88)?,
        decision_outcomes: count(base + 104)?,
        condition_outcomes: count(base + 120)?,
        value_selections: count(base + 136)?,
        coverage_complete: bool_field(record[flags_offset])?,
        completeness_blocked: match record[flags_offset + 1] {
            0 => None,
            1 => Some(false),
            2 => Some(true),
            _ => return Err(CoverageIndexError::InvalidRecord("optional boolean")),
        },
    })
}

fn percentage(covered: usize, total: usize) -> f64 {
    if total == 0 {
        100.0
    } else {
        ((covered as f64 / total as f64) * 10_000.0).round() / 100.0
    }
}

#[cfg(test)]
mod tests {
    use std::{
        fs,
        path::PathBuf,
        sync::atomic::{AtomicU64, Ordering},
        time::{SystemTime, UNIX_EPOCH},
    };

    use crate::{
        coverage_analysis::{McdcVector, PointKind},
        coverage_report::{
            CoverageManifest, CoverageReportRequest, DecisionMeta, ExitCodeInput, PointMeta,
            RawTestResult, RuntimeSnapshot, TestProvenance, analyze_coverage_results,
        },
        query_index::{QueryIndexIdentity, write_query_index},
    };

    use super::*;

    static ROOT_SEQUENCE: AtomicU64 = AtomicU64::new(0);

    fn root() -> PathBuf {
        let nonce = SystemTime::now()
            .duration_since(UNIX_EPOCH)
            .unwrap()
            .as_nanos();
        let root = std::env::temp_dir().join(format!(
            "supercov-coverage-index-{}-{nonce}-{}",
            std::process::id(),
            ROOT_SEQUENCE.fetch_add(1, Ordering::Relaxed),
        ));
        fs::create_dir_all(&root).unwrap();
        root
    }

    fn identity() -> QueryIndexIdentity {
        QueryIndexIdentity {
            evidence_sha256: [1; 32],
            evidence_bytes: 100,
            analysis_sha256: [2; 32],
            producer_sha256: [3; 32],
            archive_schema_version: 2,
        }
    }

    fn report() -> CoverageReport {
        let decision = DecisionMeta {
            id: "d".into(),
            file: "src/a.js".into(),
            line: 1,
            column: 1,
            source: "a && b".into(),
            conditions: vec!["a".into(), "b".into()],
            kind: "if".into(),
        };
        analyze_coverage_results(&CoverageReportRequest {
            run_id: "run".into(),
            manifest: CoverageManifest {
                unmeasured: Vec::new(),
                decisions: vec![decision.clone()],
                points: vec![PointMeta {
                    id: "point".into(),
                    kind: PointKind::Statement,
                    file: "src/a.js".into(),
                    line: 2,
                    column: 3,
                    source: "work();".into(),
                    label: None,
                }],
                branches: Vec::new(),
                limitations: vec![serde_json::json!({
                    "id": "dynamic",
                    "kind": "dynamic-code",
                    "file": "src/a.js",
                    "line": 3,
                    "column": 1,
                    "source": "eval(code)",
                    "reason": "dynamic source"
                })],
                scope: None,
                assertion_sites: Vec::new(),
            },
            raw_results: vec![RawTestResult {
                test_id: Some("test".into()),
                scope: None,
                test: "test".into(),
                test_file: Some("tests/a.js".into()),
                title: None,
                retry: Some(0),
                status: Some("passed".into()),
                expected_status: None,
                flaky: false,
                provenance: TestProvenance {
                    runner: "node:test".into(),
                    kind: "unit".into(),
                    project: None,
                    source: "runner-default".into(),
                },
                role: "test".into(),
                attribution: crate::coverage_report::ATTRIBUTION_EXACT.into(),
                phases: Vec::new(),
                runtime: vec![RuntimeSnapshot {
                    decisions: vec![crate::coverage_report::DecisionSnapshot {
                        meta: decision,
                        vectors: vec![McdcVector {
                            values: vec![Some(false), None],
                            outcome: false,
                        }],
                    }],
                    hits: vec!["point".into()],
                    events: Vec::new(),
                    logicals: Vec::new(),
                    phase_id: None,
                }],
                browser: Vec::new(),
                server: Vec::new(),
            }],
            generated_at: "time".into(),
            coverage_model: None,
            integrity: None,
            test_exit_code: ExitCodeInput::Present(Some(0)),
        })
        .unwrap()
    }

    fn report_attributed(attribution: &str) -> CoverageReport {
        let mut report = report();
        for view in [
            &mut report.view,
            &mut report.filters.passed,
            &mut report.filters.failed,
        ] {
            for test in &mut view.tests {
                test.attribution = attribution.to_owned();
            }
        }
        report
    }

    #[test]
    fn declined_lines_and_points_are_not_reported_as_coverage_gaps() {
        let mut report = report();
        for line in &mut report.view.lines {
            line.measured = false;
            line.covered = false;
            line.tests.clear();
        }
        for point in &mut report.view.points {
            point.measured = false;
            point.covered = false;
            point.tests.clear();
        }
        let selected = Selected::default();
        for filter in [None, Some(&selected)] {
            let gaps = file_gaps(&report.view, filter).unwrap();
            let (_, gap) = &gaps[0];
            assert_eq!(gap.uncovered_lines, 0);
            assert_eq!(gap.uncovered_statements, 0);
            assert_eq!(gap.uncovered_functions, 0);
            assert_eq!(gap.uncovered_everywhere[0..3], [0, 0, 0]);
            assert_eq!(gap.measurement_limitations, 1);
        }
        // Actual missed obligations must still appear.
        report.view.lines[0].measured = true;
        report.view.points[0].measured = true;
        let gaps = file_gaps(&report.view, None).unwrap();
        assert_eq!(gaps[0].1.uncovered_lines, 1);
        assert_eq!(gaps[0].1.uncovered_statements, 1);
    }

    #[test]
    fn a_test_keeps_the_attribution_it_was_written_with() {
        // Attribution rides in a byte the reader used to refuse outright, so a
        // value that does not survive the trip is not a wrong number on a
        // report -- it is a run that will not open. Only `exact` was ever
        // written by a test, which left the other two arms of both the encoder
        // and the decoder standing on the claim that they matched.
        let root = root();
        for attribution in [
            crate::coverage_report::ATTRIBUTION_EXACT,
            crate::coverage_report::ATTRIBUTION_RUN_WIDE,
            crate::coverage_report::ATTRIBUTION_PARTIAL,
        ] {
            let report = report_attributed(attribution);
            let path = root.join(format!("query-index-{attribution}.v1.bin"));
            write_query_index(
                &coverage_index_sections(&report).unwrap(),
                &identity(),
                &path,
            )
            .unwrap();
            let container = QueryIndex::open(&path, &identity()).unwrap();
            let index = CoverageIndex::new(&container).unwrap();
            let tests = index.test_summaries(CoverageViewId::All).unwrap();
            assert_eq!(tests.len(), 1);
            assert_eq!(
                tests[0].attribution, attribution,
                "written as `{attribution}` and read back as `{}`",
                tests[0].attribution
            );
        }
    }

    #[test]
    fn an_attribution_the_byte_has_no_room_for_is_refused_at_the_write() {
        // Refusing here is what keeps the reserved byte honest: a value nobody
        // taught the encoder must not be written as some neighbouring one and
        // read back as a measurement.
        let error = coverage_index_sections(&report_attributed("probably-exact")).unwrap_err();
        assert!(
            matches!(error, CoverageIndexError::InvalidRecord("test attribution")),
            "{error:?}"
        );
    }

    // The same run is named over and over -- a filtered view repeats the
    // relation of the view it filters, and one list of tests covers many lines
    // of the file they exercise. Storing each naming separately made this the
    // largest section in the container: 61 MB of a 210 MB index.
    #[test]
    fn a_relation_named_twice_is_stored_once() {
        let mut strings = StringTable::default();
        let mut relations = StringRelations::default();
        let run = || ["alpha".to_string(), "beta".to_string(), "gamma".to_string()];

        let first = relations.push(run(), &mut strings).unwrap();
        let second = relations.push(run(), &mut strings).unwrap();
        assert_eq!(first, second, "the second naming points at the first run");
        assert_eq!(relations.values.len(), 3, "and nothing was appended for it");

        // A different run still gets its own storage, after the one already
        // there -- readers reach a run only by (offset, count), so the
        // existing pair must keep meaning what it meant.
        let other = relations
            .push(["alpha".to_string(), "delta".to_string()], &mut strings)
            .unwrap();
        assert_eq!(other, (3, 2));
        assert_eq!(relations.values.len(), 5);

        // A prefix of a stored run is a different run, and is stored as one
        // rather than aliasing into the middle of another.
        let prefix = relations.push(["alpha".to_string()], &mut strings).unwrap();
        assert_eq!(prefix, (5, 1));

        // What each pair resolves to is what was pushed.
        let read = |(offset, count): (u64, u64)| {
            relations.values[offset as usize..(offset + count) as usize].to_vec()
        };
        assert_eq!(read(first), read(second));
        assert_eq!(read(first).len(), 3);
        assert_ne!(read(first), read(other));
    }

    #[test]
    fn shared_id_relations_match_text_relations_across_allocations() {
        let mut shared_strings = StringTable::default();
        let mut text_strings = StringTable::default();
        let mut shared_relations = StringRelations::default();
        let mut text_relations = StringRelations::default();
        let original: Vec<Id> = ["", "test::a[1]", "test::a[10]", "unicodé ✓"]
            .into_iter()
            .map(Id::from)
            .collect();
        let copies = original.clone();
        let separate: Vec<Id> = original.iter().map(|id| Id::from(id.as_str())).collect();
        for values in [&original[..], &copies, &separate, &separate[1..], &[]] {
            assert_eq!(
                shared_relations
                    .push_ids(values, &mut shared_strings)
                    .unwrap(),
                text_relations.push(values, &mut text_strings).unwrap(),
            );
        }
        // The table owns references to cached allocations even after callers
        // drop theirs, so a later ID cannot reuse an earlier cache address.
        drop(original);
        drop(copies);
        drop(separate);
        for value in ["new", "test::a[1]", ""] {
            let id = Id::from(value);
            assert_eq!(
                shared_strings.intern_shared(&id).unwrap(),
                text_strings.intern(value).unwrap(),
            );
        }
        assert_eq!(shared_relations.values, text_relations.values);
        assert_eq!(shared_strings.strings, text_strings.strings);
    }

    #[test]
    fn typed_columns_round_trip_all_outcome_views_without_json() {
        let report = report();
        let root = root();
        let path = root.join("query-index.v1.bin");
        write_query_index(
            &coverage_index_sections(&report).unwrap(),
            &identity(),
            &path,
        )
        .unwrap();
        let container = QueryIndex::open(&path, &identity()).unwrap();
        let index = CoverageIndex::new(&container).unwrap();
        assert_eq!(
            index.model().unwrap(),
            IndexedCoverageModel {
                schema_version: COVERAGE_MODEL_SCHEMA_VERSION,
                variant: report.view.variant.clone(),
                name: report.view.model.name.clone(),
                completeness_meaning: report.view.model.completeness_meaning.clone(),
                measured: report.view.model.measured.clone(),
                not_measured: report.view.model.not_measured.clone(),
            }
        );
        for (id, view) in [
            (CoverageViewId::All, &report.view),
            (CoverageViewId::Passed, &report.filters.passed),
            (CoverageViewId::Failed, &report.filters.failed),
        ] {
            assert_eq!(index.summary(id).unwrap(), view.summary);
        }
        let gaps = index.file_gaps(CoverageViewId::All, None, None).unwrap();
        assert_eq!(gaps.len(), 1);
        assert_eq!(gaps[0].file, "src/a.js");
        assert_eq!(gaps[0].missing_mcdc_conditions, 2);
        let projection = index.projection(CoverageViewId::All, None, None).unwrap();
        assert_eq!(projection.summary, report.view.summary);
        assert_eq!(projection.tests, 1);
        assert_eq!(projection.setups, 0);
        assert_eq!(projection.test_outcomes.passed, 1);
        assert!(projection.source_scope.is_none());
        let line = index
            .line(CoverageViewId::All, "src/a.js", 2)
            .unwrap()
            .unwrap();
        assert!(line.covered);
        assert_eq!(line.tests, ["test"]);
        assert_eq!(line.confidence.level, "executed");
        let tests = index.test_summaries(CoverageViewId::All).unwrap();
        assert_eq!(tests.len(), 1);
        assert_eq!(tests[0].provenance.runner, "node:test");
        // The decision, and its two outcomes as a branch.
        let anchors = index.anchors(CoverageViewId::All, "src/a.js", 1).unwrap();
        assert_eq!(anchors.len(), 2);
        let decision = anchors
            .iter()
            .filter(|anchor| anchor.kind == "decision")
            .collect::<Vec<_>>();
        assert_eq!(decision.len(), 1);
        assert_eq!(decision[0].conditions, Some(2));
        assert_eq!(decision[0].tests, ["test"]);
        assert!(anchors.iter().any(|anchor| anchor.kind == "branch"));
        let point = index.anchors(CoverageViewId::All, "src/a.js", 2).unwrap();
        assert_eq!(point.len(), 1);
        assert_eq!(point[0].kind, "statement");
        assert_eq!(point[0].tests, ["test"]);
        let details = index.test_details(CoverageViewId::All).unwrap();
        assert_eq!(details.len(), 1);
        assert_eq!(details[0].retries, [0]);
        assert_eq!(details[0].attempts.len(), 1);
        assert_eq!(details[0].hits, ["point"]);
        assert_eq!(details[0].lines.len(), 1);
        assert_eq!(details[0].lines[0].line, 2);
        assert_eq!(details[0].decisions.len(), 1);
        assert_eq!(details[0].decisions[0].vectors.len(), 1);
        assert_eq!(
            details[0].decisions[0].vectors[0].values,
            [Some(false), None]
        );
        let hits = index.hit_metadata(CoverageViewId::All).unwrap();
        // The point, and the decision's two outcomes.
        assert_eq!(hits.len(), 3);
        let hits = hits
            .into_iter()
            .filter(|hit| hit.obligation != "branch")
            .collect::<Vec<_>>();
        assert_eq!(hits.len(), 1);
        assert_eq!(hits[0].id, "point");
        assert_eq!(hits[0].source, "work();");
        assert_eq!(hits[0].tests, ["test"]);
        let decisions = index.decision_metadata(CoverageViewId::All).unwrap();
        assert_eq!(decisions.len(), 1);
        assert_eq!(decisions[0].conditions, ["a", "b"]);
        assert_eq!(
            index.decision_details(CoverageViewId::All).unwrap(),
            report.view.decisions
        );
        let limitations = index.limitations(CoverageViewId::All).unwrap();
        assert_eq!(limitations.len(), 1);
        assert_eq!(limitations[0].kind, "dynamic-code");
        assert_eq!(limitations[0].line, 3);
        fs::remove_dir_all(root).unwrap();
    }

    #[test]
    fn index_preserves_catalogued_unstarted_tests_without_attempts() {
        let report = analyze_coverage_results(&CoverageReportRequest {
            run_id: "run".into(),
            manifest: CoverageManifest {
                unmeasured: Vec::new(),
                decisions: Vec::new(),
                points: Vec::new(),
                branches: Vec::new(),
                limitations: Vec::new(),
                scope: None,
                assertion_sites: Vec::new(),
            },
            raw_results: vec![RawTestResult {
                test_id: Some("unstarted".into()),
                scope: None,
                test: "unstarted".into(),
                test_file: Some("tests/a.rs".into()),
                title: None,
                retry: None,
                status: Some("unstarted".into()),
                expected_status: Some("passed".into()),
                flaky: false,
                provenance: TestProvenance {
                    runner: "rust-nextest".into(),
                    kind: "unit".into(),
                    project: None,
                    source: "selected-but-not-started".into(),
                },
                role: "test".into(),
                attribution: crate::coverage_report::ATTRIBUTION_EXACT.into(),
                phases: Vec::new(),
                runtime: Vec::new(),
                browser: Vec::new(),
                server: Vec::new(),
            }],
            generated_at: "time".into(),
            coverage_model: None,
            integrity: None,
            test_exit_code: ExitCodeInput::Present(Some(100)),
        })
        .unwrap();
        let root = root();
        let path = root.join("query-index.v1.bin");
        write_query_index(
            &coverage_index_sections(&report).unwrap(),
            &identity(),
            &path,
        )
        .unwrap();
        let container = QueryIndex::open(&path, &identity()).unwrap();
        let index = CoverageIndex::new(&container).unwrap();
        let summaries = index.test_summaries(CoverageViewId::All).unwrap();
        assert_eq!(summaries.len(), 1);
        assert_eq!(summaries[0].outcome, "unstarted");
        let details = index.test_details(CoverageViewId::All).unwrap();
        assert!(details[0].retries.is_empty());
        assert!(details[0].attempts.is_empty());
        let projection = index.projection(CoverageViewId::All, None, None).unwrap();
        assert_eq!(projection.tests, 1);
        assert_eq!(projection.test_outcomes.unstarted, 1);
        fs::remove_dir_all(root).unwrap();
    }

    #[test]
    fn source_discovery_scope_preserves_python_language() {
        let mut report = report();
        let scope = serde_json::json!({
            "version": 1,
            "mode": "automatic",
            "roots": ["."],
            "entries": [{
                "file": "src/a.py",
                "status": "included",
                "reason": "python source",
                "packageRoot": "."
            }]
        });
        for view in [
            &mut report.view,
            &mut report.filters.passed,
            &mut report.filters.failed,
        ] {
            view.scope = Some(scope.clone());
            view.model.language = "python".into();
            view.model.name = "python-probes-v1".into();
        }
        let root = root();
        let path = root.join("query-index.python.bin");
        write_query_index(
            &coverage_index_sections(&report).unwrap(),
            &identity(),
            &path,
        )
        .unwrap();
        let container = QueryIndex::open(&path, &identity()).unwrap();
        let index = CoverageIndex::new(&container).unwrap();
        let projection = index.projection(CoverageViewId::All, None, None).unwrap();
        let source_scope = projection.source_scope.unwrap();
        assert_eq!(source_scope.language, "python");
        assert_eq!(source_scope.model, "python-probes-v1");
        assert_eq!(source_scope.kind, "source-discovery");
        assert_eq!(source_scope.included, 1);
        assert_eq!(index.scope_entries(CoverageViewId::All).unwrap().len(), 1);
        fs::remove_dir_all(root).unwrap();
    }

    #[test]
    fn compiler_owned_scope_round_trips_without_javascript_scope_fields() {
        let mut report = report();
        let scope = serde_json::json!({
            "language": "rust",
            "model": "rust-source-v1",
            "crate": "fixture",
            "measurementComplete": false
        });
        report.view.scope = Some(scope.clone());
        report.filters.passed.scope = Some(scope.clone());
        report.filters.failed.scope = Some(scope);
        let root = root();
        let path = root.join("query-index.v2.bin");
        write_query_index(
            &coverage_index_sections(&report).unwrap(),
            &identity(),
            &path,
        )
        .unwrap();
        let container = QueryIndex::open(&path, &identity()).unwrap();
        let index = CoverageIndex::new(&container).unwrap();
        let projection = index.projection(CoverageViewId::All, None, None).unwrap();
        assert_eq!(
            projection.source_scope,
            Some(IndexedSourceScope {
                kind: "compiler".into(),
                language: "rust".into(),
                model: "rust-source-v1".into(),
                mode: None,
                roots: Vec::new(),
                unit: Some("fixture".into()),
                measurement_complete: Some(false),
                included: 0,
                excluded: 0,
                ambiguous: 0,
            })
        );
        assert!(index.scope_entries(CoverageViewId::All).unwrap().is_empty());
        fs::remove_dir_all(root).unwrap();
    }
}