//! Requirements traceability for DarkRoom.
//!
//! Extracts `TRACES:` tags from source, counts what `requirements.md` actually
//! defines, and reports coverage as the intersection of the two.
//!
//! # Why the arithmetic is written this way
//!
//! Adapted from the JellyTau tooling, including the bug it was repaired for.
//! That gate divided a traced count by *frozen literal* denominators; the
//! requirements file grew past them, and it reported **158% coverage**. A gate
//! reporting over 100% cannot fail its own threshold, so it silently stopped
//! being a gate at all.
//!
//! Two rules follow, and both are enforced by tests here:
//!
//! 1. **Denominators are parsed from `requirements.md` at run time.** Never
//!    hardcoded, never cached.
//! 2. **Coverage is `|traced ∩ defined| / |defined|`.** Using the raw traced
//!    count as the numerator is precisely what lets a ratio exceed 100%, since
//!    a tag naming a deleted requirement would count as covered. Such tags are
//!    reported as orphans instead.

use std::collections::{BTreeMap, BTreeSet};
use std::path::{Path, PathBuf};

use serde::Serialize;

/// Requirement ID prefixes that participate in coverage.
///
/// Test identifiers (UT, IT) are a separate taxonomy: they are *evidence* for
/// requirements, not requirements themselves. Counting them would inflate both
/// numerator and denominator, and flagging them as orphans would bury real
/// typos in noise.
pub const REQUIREMENT_TYPES: &[&str] = &["FR", "NFR", "R"];

/// Prefixes recognised in tags but deliberately excluded from coverage.
///
/// `D` are decisions, `S` spikes, `M` milestone items, `AC` acceptance
/// criteria, `UT`/`IT` tests. All are legitimate things to tag against, and
/// none is a requirement — counting them would inflate the denominator by 25
/// and make coverage look worse than it is, which is the same class of defect
/// as JellyTau's inflated ratio, just in the other direction.
pub const NON_REQUIREMENT_TYPES: &[&str] = &["UT", "IT", "AC", "M", "S", "D"];

/// One `TRACES:` tag found in source.
#[derive(Debug, Clone, Serialize, PartialEq, Eq)]
pub struct TraceEntry {
    pub file: String,
    pub line: usize,
    pub context: String,
    pub requirements: Vec<String>,
}

/// What `requirements.md` defines — the coverage denominators.
#[derive(Debug, Clone, Default)]
pub struct DefinedRequirements {
    pub ids: BTreeSet<String>,
    pub by_type: BTreeMap<String, usize>,
}

impl DefinedRequirements {
    pub fn total(&self) -> usize {
        self.ids.len()
    }
}

/// The coverage result.
#[derive(Debug, Clone, Serialize, PartialEq)]
pub struct Coverage {
    pub covered: usize,
    pub total: usize,
    pub percent: f64,
    /// Tagged in source but absent from `requirements.md` — a typo, or a
    /// requirement that was renumbered or deleted. Never counted as covered.
    pub orphaned: Vec<String>,
    /// Defined but never tagged anywhere.
    pub untraced: Vec<String>,
}

/// Parse the requirement IDs a markdown document *defines*.
///
/// A requirement is defined by a bolded heading-style declaration
/// (`**FR-CAT-1 — …**`) or as the leading cell of a table row (`| R1 | … |`).
/// Both forms appear in DarkRoom's requirements.md.
///
/// Deliberately *not* a bare scan for anything matching the ID shape: that
/// counts cross-references in prose and in "relates to" columns as
/// definitions, inflating the denominator. IDs are deduplicated because a
/// requirement may legitimately appear in both a definition and a summary
/// table.
pub fn parse_defined_requirements(markdown: &str) -> DefinedRequirements {
    let mut ids = BTreeSet::new();

    for line in markdown.lines() {
        let trimmed = line.trim_start();

        // Form 1: a bolded definition, e.g. `**FR-CAT-1 — Scan.**`
        if let Some(rest) = trimmed.strip_prefix("**") {
            if let Some(id) = leading_id(rest) {
                ids.insert(id);
                continue;
            }
        }

        // Form 2: leading table cell, e.g. `| **R1** | … |` or `| R1 | … |`
        if let Some(rest) = trimmed.strip_prefix('|') {
            let cell = rest.trim().trim_start_matches("**");
            if let Some(id) = leading_id(cell) {
                ids.insert(id);
            }
        }
    }

    // Only requirement types enter the register. Decisions, spikes, milestone
    // items and test ids are all taggable, but none is a requirement, and
    // counting them would inflate the denominator.
    let ids: BTreeSet<String> = ids.into_iter().filter(|id| is_requirement(id)).collect();

    let mut by_type: BTreeMap<String, usize> = BTreeMap::new();
    for id in &ids {
        *by_type.entry(type_of(id).to_string()).or_insert(0) += 1;
    }

    DefinedRequirements { ids, by_type }
}

/// Extract a requirement ID anchored at the start of `s`.
///
/// Accepts `FR-CAT-1`, `NFR-P13`, `R1`, `FR-DEV-3a` — DarkRoom uses
/// alphanumeric segments and an optional trailing letter, not the fixed
/// three-digit form JellyTau assumed.
fn leading_id(s: &str) -> Option<String> {
    let bytes = s.as_bytes();
    if bytes.is_empty() || !bytes[0].is_ascii_uppercase() {
        return None;
    }

    let mut end = 0;
    let mut seen_digit = false;
    for (i, c) in s.char_indices() {
        match c {
            'A'..='Z' | '0'..='9' | '-' => {
                if c.is_ascii_digit() {
                    seen_digit = true;
                }
                end = i + c.len_utf8();
            }
            'a'..='z' if seen_digit => {
                // Trailing variant letter, e.g. FR-DEV-3a.
                end = i + c.len_utf8();
            }
            _ => break,
        }
    }

    if end == 0 || !seen_digit {
        return None;
    }

    let id = s[..end].trim_end_matches('-').to_string();
    // Must have a recognised prefix, or arbitrary capitalised words match.
    let ty = type_of(&id);
    if REQUIREMENT_TYPES.contains(&ty) || NON_REQUIREMENT_TYPES.contains(&ty) {
        Some(id)
    } else {
        None
    }
}

/// The type prefix of an ID: `FR-CAT-1` → `FR`, `R1` → `R`.
pub fn type_of(id: &str) -> &str {
    let end = id
        .find(|c: char| !c.is_ascii_uppercase())
        .unwrap_or(id.len());
    &id[..end]
}

/// Whether an ID participates in coverage.
pub fn is_requirement(id: &str) -> bool {
    REQUIREMENT_TYPES.contains(&type_of(id))
}

/// Extract every `TRACES:` tag from a source file's text.
pub fn extract_from_text(text: &str, path: &str) -> Vec<TraceEntry> {
    let lines: Vec<&str> = text.lines().collect();
    let mut out = Vec::new();

    for (idx, line) in lines.iter().enumerate() {
        let Some(pos) = line.find("TRACES:") else {
            continue;
        };
        let tail = &line[pos + "TRACES:".len()..];
        let ids = parse_ids(tail);
        if ids.is_empty() {
            continue;
        }
        out.push(TraceEntry {
            file: path.to_string(),
            line: idx + 1,
            context: find_context(&lines, idx),
            requirements: ids,
        });
    }

    out
}

/// Parse the ID list from a tag body: `FR-CAT-1, FR-CAT-2 | NFR-P1`.
///
/// The pipe groups types for readability; both separators are treated alike.
fn parse_ids(s: &str) -> Vec<String> {
    s.split(['|', ','])
        .filter_map(|part| leading_id(part.trim()))
        .collect()
}

/// The nearest preceding declaration, for the report's context column.
fn find_context(lines: &[&str], from: usize) -> String {
    const MARKERS: &[&str] = &[
        "pub fn ",
        "fn ",
        "pub struct ",
        "struct ",
        "pub enum ",
        "enum ",
        "impl ",
        "pub trait ",
        "trait ",
        "#[test]",
        "component ",
        "export ",
    ];

    // Look forward first — a doc comment precedes what it documents.
    for line in lines.iter().skip(from + 1).take(6) {
        if MARKERS.iter().any(|m| line.contains(m)) {
            return trim_body(line);
        }
    }
    // Then backward, for tags placed inside a body.
    for i in (from.saturating_sub(6)..from).rev() {
        if MARKERS.iter().any(|m| lines[i].contains(m)) {
            return lines[i].trim().trim_end_matches('{').trim().to_string();
        }
    }
    "—".to_string()
}

/// Strip a trailing body opener so the context reads as a signature.
///
/// Handles both `fn f() {` and `fn f() {}` — the latter is why this is a
/// helper rather than a single `trim_end_matches('{')`.
fn trim_body(line: &str) -> String {
    line.trim()
        .trim_end_matches("{}")
        .trim_end()
        .trim_end_matches('{')
        .trim_end()
        .to_string()
}

/// Compute coverage as the intersection of traced and defined IDs.
///
/// This signature is the fix for the 158% bug: `defined` is required, so there
/// is nowhere for a frozen denominator to hide.
/// TRACES: NFR-OPS-1
pub fn compute_coverage(traced: &BTreeSet<String>, defined: &DefinedRequirements) -> Coverage {
    let traced_reqs: BTreeSet<&String> = traced.iter().filter(|id| is_requirement(id)).collect();

    let covered: Vec<&String> = traced_reqs
        .iter()
        .filter(|id| defined.ids.contains(**id))
        .copied()
        .collect();

    let orphaned: Vec<String> = traced_reqs
        .iter()
        .filter(|id| !defined.ids.contains(**id))
        .map(|s| s.to_string())
        .collect();

    let untraced: Vec<String> = defined
        .ids
        .iter()
        .filter(|id| !traced.contains(*id))
        .cloned()
        .collect();

    let total = defined.total();
    Coverage {
        covered: covered.len(),
        total,
        percent: if total == 0 {
            0.0
        } else {
            (covered.len() as f64 / total as f64) * 100.0
        },
        orphaned,
        untraced,
    }
}

/// Source file extensions scanned for tags.
///
/// `.yaml` is here because a develop operation is now declared rather than
/// written: `core/dr-pipeline/ops/<id>.yaml` is the whole node, and the Rust
/// implementing it is generated into `OUT_DIR`, which is not scanned and could
/// not be linked to from the report if it were. Without this a node would have
/// nowhere to record the requirement it satisfies.
pub const SOURCE_SUFFIXES: &[&str] = &[".rs", ".slint", ".wgsl", ".yaml"];

/// Directories never scanned.
const EXCLUDED: &[&str] = &["target", "target-android", ".git", "node_modules", "temp"];

/// Walk `roots` under `base`, returning every source file.
pub fn collect_sources(base: &Path, roots: &[&str]) -> Vec<PathBuf> {
    let mut out = Vec::new();
    for root in roots {
        let dir = base.join(root);
        if dir.exists() {
            walk(&dir, &mut out);
        }
    }
    out.sort();
    out
}

fn walk(dir: &Path, out: &mut Vec<PathBuf>) {
    let Ok(entries) = std::fs::read_dir(dir) else {
        return;
    };
    for entry in entries.flatten() {
        let path = entry.path();
        let name = entry.file_name();
        let name = name.to_string_lossy();

        if EXCLUDED.iter().any(|e| *e == name) {
            continue;
        }
        if path.is_dir() {
            walk(&path, out);
        } else if SOURCE_SUFFIXES.iter().any(|s| name.ends_with(s)) {
            out.push(path);
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    // ---- the 158% regression ------------------------------------------

    #[test]
    fn coverage_never_exceeds_one_hundred_percent() {
        // The JellyTau failure, reproduced: more tags than the register
        // defines. Every extra tag must land in `orphaned`, never inflate
        // the numerator.
        let defined = parse_defined_requirements("**FR-CAT-1 — Scan.**\n");
        let traced: BTreeSet<String> = ["FR-CAT-1", "FR-CAT-2", "FR-CAT-3", "FR-CAT-4"]
            .iter()
            .map(|s| s.to_string())
            .collect();

        let cov = compute_coverage(&traced, &defined);
        assert_eq!(cov.covered, 1);
        assert_eq!(cov.total, 1);
        assert_eq!(cov.percent, 100.0);
        assert!(cov.percent <= 100.0, "coverage must never exceed 100%");
        assert_eq!(cov.orphaned, vec!["FR-CAT-2", "FR-CAT-3", "FR-CAT-4"]);
    }

    #[test]
    fn orphan_tags_are_reported_not_counted() {
        let defined = parse_defined_requirements("**FR-CAT-1 — Scan.**\n");
        let traced: BTreeSet<String> = ["FR-CAT-1", "FR-TYPO-9"]
            .iter()
            .map(|s| s.to_string())
            .collect();

        let cov = compute_coverage(&traced, &defined);
        assert_eq!(cov.covered, 1);
        assert_eq!(cov.orphaned, vec!["FR-TYPO-9"]);
    }

    #[test]
    fn empty_register_is_zero_not_a_division_by_zero() {
        let defined = parse_defined_requirements("");
        let traced = BTreeSet::new();
        let cov = compute_coverage(&traced, &defined);
        assert_eq!(cov.percent, 0.0);
        assert_eq!(cov.total, 0);
    }

    // ---- denominator parsing ------------------------------------------

    #[test]
    fn definitions_come_from_declarations_not_cross_references() {
        // Only the first line defines FR-CAT-1. The prose reference and the
        // "relates to" cell must not each count as another definition.
        let md = "\
**FR-CAT-1 — Scan.** The app shall scan roots.

This is discussed in FR-CAT-1 and also FR-CAT-1 again.

| Spike | Answers | Relates to |
|---|---|---|
| S1 | whether it works | FR-CAT-1 |
";
        let defined = parse_defined_requirements(md);
        // FR-CAT-1 once, despite three mentions. S1 is a spike, not a
        // requirement, so it does not enter the register at all.
        assert_eq!(defined.total(), 1);
        assert!(defined.ids.contains("FR-CAT-1"));
    }

    #[test]
    fn decisions_and_spikes_are_not_requirements() {
        // D and S are taggable but are not requirements; counting them would
        // inflate the denominator and understate real coverage.
        let md = "\
**FR-CAT-1 — Scan.**
| D1 | Language | Rust |
| S1 | Zero-copy spike | proves ARCH 6.1 |
";
        let defined = parse_defined_requirements(md);
        assert_eq!(defined.total(), 1, "only FR-CAT-1 is a requirement");
        assert!(defined.ids.contains("FR-CAT-1"));
        assert!(!defined.ids.contains("D1"));
        assert!(!defined.ids.contains("S1"));
    }

    #[test]
    fn tagging_a_decision_neither_covers_nor_orphans() {
        let defined = parse_defined_requirements("**FR-CAT-1 — Scan.**\n");
        let traced: BTreeSet<String> = ["FR-CAT-1", "D1"].iter().map(|s| s.to_string()).collect();
        let cov = compute_coverage(&traced, &defined);
        assert_eq!(cov.covered, 1);
        assert!(cov.orphaned.is_empty(), "D1 is a valid tag, not an orphan");
    }

    #[test]
    fn table_row_ids_are_definitions() {
        let md = "| **R1** | Cross-platform | Same core on both |\n\
                  | R2 | Efficient display | 60fps |\n";
        let defined = parse_defined_requirements(md);
        assert!(defined.ids.contains("R1"));
        assert!(defined.ids.contains("R2"));
    }

    #[test]
    fn darkroom_id_shapes_parse() {
        // Not JellyTau's fixed three-digit form.
        assert_eq!(leading_id("FR-CAT-1 — Scan").as_deref(), Some("FR-CAT-1"));
        assert_eq!(
            leading_id("NFR-P13 — Next image").as_deref(),
            Some("NFR-P13")
        );
        assert_eq!(
            leading_id("FR-DEV-3a — Descriptors").as_deref(),
            Some("FR-DEV-3a")
        );
        assert_eq!(leading_id("R1 | Cross-platform").as_deref(), Some("R1"));
    }

    #[test]
    fn prose_is_not_an_id() {
        assert_eq!(leading_id("The app shall scan"), None);
        assert_eq!(leading_id("GPU results never"), None);
        // A recognised prefix with no digits is not an ID either.
        assert_eq!(leading_id("FR without a number"), None);
    }

    // ---- tag extraction -----------------------------------------------

    #[test]
    fn extracts_tags_with_both_separators() {
        let src = "\
/// TRACES: FR-CAT-1, FR-CAT-2 | NFR-P1
pub fn scan() {}
";
        let traces = extract_from_text(src, "x.rs");
        assert_eq!(traces.len(), 1);
        assert_eq!(
            traces[0].requirements,
            vec!["FR-CAT-1", "FR-CAT-2", "NFR-P1"]
        );
        assert_eq!(traces[0].line, 1);
        assert_eq!(traces[0].context, "pub fn scan()");
    }

    #[test]
    fn context_looks_forward_then_backward() {
        // Doc comments precede their item.
        let fwd = extract_from_text("// TRACES: R1\npub struct Catalog;", "x.rs");
        assert_eq!(fwd[0].context, "pub struct Catalog;");

        // A tag inside a body refers to the enclosing item.
        let back = extract_from_text("pub fn render() {\n    // TRACES: R1\n}", "x.rs");
        assert_eq!(back[0].context, "pub fn render()");
    }

    #[test]
    fn a_tag_with_no_ids_is_ignored() {
        let traces = extract_from_text("// TRACES: see the design doc\n", "x.rs");
        assert!(traces.is_empty());
    }

    #[test]
    fn test_ids_are_extracted_but_not_counted_as_requirements() {
        let traces = extract_from_text("// TRACES: FR-CAT-1 | UT-001\nfn f(){}", "x.rs");
        assert_eq!(traces[0].requirements, vec!["FR-CAT-1", "UT-001"]);

        // UT is a separate taxonomy: evidence, not a requirement.
        assert!(is_requirement("FR-CAT-1"));
        assert!(!is_requirement("UT-001"));

        // So it neither covers nor orphans.
        let defined = parse_defined_requirements("**FR-CAT-1 — Scan.**\n");
        let traced: BTreeSet<String> = ["FR-CAT-1", "UT-001"]
            .iter()
            .map(|s| s.to_string())
            .collect();
        let cov = compute_coverage(&traced, &defined);
        assert_eq!(cov.covered, 1);
        assert!(
            cov.orphaned.is_empty(),
            "UT must not be reported as orphaned"
        );
    }

    #[test]
    fn type_prefixes_split_correctly() {
        assert_eq!(type_of("FR-CAT-1"), "FR");
        assert_eq!(type_of("NFR-P13"), "NFR");
        assert_eq!(type_of("R1"), "R");
        assert_eq!(type_of("UT-001"), "UT");
    }

    #[test]
    fn untraced_requirements_are_listed() {
        let defined = parse_defined_requirements("**FR-A-1 — One.**\n**FR-B-2 — Two.**\n");
        let traced: BTreeSet<String> = ["FR-A-1"].iter().map(|s| s.to_string()).collect();
        let cov = compute_coverage(&traced, &defined);
        assert_eq!(cov.untraced, vec!["FR-B-2"]);
    }
}