//! Coverage reports (oxygenhub#113): Coveralls built into the hub. An
//! OxygenFlow job that declares an artifact named `coverage` in LCOV
//! format (what `cargo llvm-cov --lcov`, Istanbul, coverage.py and
//! gcov/lcov write) has it parsed when it is uploaded. The totals and each
//! file's counts are kept per run; the per-line counts are read again from
//! the artifact's blob for the file view. Paths are taken relative to the
//! job's workspace (`/src`), and only files of the run's state are kept.

use crate::db::{now_ms, Pool};
use crate::Shared;
use std::collections::{BTreeMap, HashSet};
use std::sync::Mutex;

/// The artifact name a pipeline gives its report.
pub const ARTIFACT: &str = "coverage";
/// The largest report parsed.
pub const MAX_REPORT: usize = 64 << 20;

/// One file's coverage, from a report.
#[derive(Clone, Debug, Default, PartialEq, Eq)]
pub struct FileCoverage {
    /// Relative to the repository's root.
    pub path: String,
    /// Each instrumented line and how often it ran, by line number.
    pub lines: BTreeMap<u32, u64>,
    pub functions_found: u64,
    pub functions_hit: u64,
    pub branches_found: u64,
    pub branches_hit: u64,
    /// The file's own `LF`/`LH` totals, when it had one record: the tool's
    /// own count, which can name lines the `DA` lines leave out (llvm-cov
    /// does), so the hub agrees with what the tool printed.
    pub declared_lines: Option<(u64, u64)>,
}

impl FileCoverage {
    pub fn lines_found(&self) -> u64 {
        self.declared_lines.map(|d| d.0).unwrap_or(self.lines.len() as u64)
    }

    pub fn lines_hit(&self) -> u64 {
        self.declared_lines.map(|d| d.1).unwrap_or_else(|| self.lines.values().filter(|n| **n > 0).count() as u64)
    }
}

/// A path as a report names it, relative to the repository: the Flow
/// workspace's `/src/` and a leading `./` taken off. `None` for a path
/// outside the repository.
pub fn relative(path: &str) -> Option<String> {
    let p = path.trim();
    let p = p.strip_prefix("/src/").unwrap_or(p);
    let p = p.strip_prefix("./").unwrap_or(p);
    if p.is_empty() || p.starts_with('/') || p.contains('\\') || p.split('/').any(|s| s.is_empty() || s == "." || s == "..") {
        return None;
    }
    Some(p.to_string())
}

#[derive(Default)]
struct Acc {
    lines: BTreeMap<u32, u64>,
    functions: BTreeMap<String, u64>,
    branches: BTreeMap<(u32, String, String), u64>,
    declared_fn: (u64, u64),
    declared_br: (u64, u64),
    declared_lines: (Option<u64>, Option<u64>),
    records: u32,
}

/// Parse an LCOV report into each file's coverage, records for the same
/// file (from several test binaries) added together. Files whose paths
/// are outside the repository are counted in the second value.
pub fn parse_lcov(text: &str) -> Result<(Vec<FileCoverage>, usize), String> {
    let mut files: BTreeMap<String, Acc> = BTreeMap::new();
    let mut current: Option<Option<String>> = None;
    let mut outside = HashSet::new();
    let mut records = 0usize;
    // A new SF: record has begun and not yet been counted for its file.
    let mut started = false;
    let num = |s: &str| s.trim().parse::<u64>().unwrap_or(0);
    for line in text.lines() {
        let line = line.trim();
        if let Some(path) = line.strip_prefix("SF:") {
            records += 1;
            let rel = relative(path);
            if rel.is_none() {
                outside.insert(path.to_string());
            }
            current = Some(rel);
            started = true;
            continue;
        }
        if line == "end_of_record" {
            current = None;
            continue;
        }
        let Some(Some(path)) = &current else { continue };
        let acc = files.entry(path.clone()).or_default();
        if line == "TN:" || line.starts_with("TN:") {
            continue;
        }
        if std::mem::take(&mut started) {
            acc.records += 1;
        }
        if let Some(n) = line.strip_prefix("LF:") {
            acc.declared_lines.0 = Some(num(n));
        } else if let Some(n) = line.strip_prefix("LH:") {
            acc.declared_lines.1 = Some(num(n));
        } else if let Some(rest) = line.strip_prefix("DA:") {
            let mut parts = rest.split(',');
            if let (Some(l), Some(n)) = (parts.next(), parts.next()) {
                if let Ok(l) = l.trim().parse::<u32>() {
                    *acc.lines.entry(l).or_default() += num(n);
                }
            }
        } else if let Some(rest) = line.strip_prefix("FN:") {
            if let Some((_, name)) = rest.split_once(',') {
                acc.functions.entry(name.to_string()).or_default();
            }
        } else if let Some(rest) = line.strip_prefix("FNDA:") {
            if let Some((n, name)) = rest.split_once(',') {
                *acc.functions.entry(name.to_string()).or_default() += num(n);
            }
        } else if let Some(rest) = line.strip_prefix("BRDA:") {
            let parts: Vec<&str> = rest.split(',').collect();
            if let [l, block, branch, taken] = parts[..] {
                if let Ok(l) = l.trim().parse::<u32>() {
                    let taken = if taken.trim() == "-" { 0 } else { num(taken) };
                    *acc.branches.entry((l, block.to_string(), branch.to_string())).or_default() += taken;
                }
            }
        } else if let Some(n) = line.strip_prefix("FNF:") {
            acc.declared_fn.0 = acc.declared_fn.0.max(num(n));
        } else if let Some(n) = line.strip_prefix("FNH:") {
            acc.declared_fn.1 = acc.declared_fn.1.max(num(n));
        } else if let Some(n) = line.strip_prefix("BRF:") {
            acc.declared_br.0 = acc.declared_br.0.max(num(n));
        } else if let Some(n) = line.strip_prefix("BRH:") {
            acc.declared_br.1 = acc.declared_br.1.max(num(n));
        }
    }
    if records == 0 {
        return Err("this is not an LCOV report: it names no source file (no SF: line)".into());
    }
    let out = files
        .into_iter()
        .map(|(path, a)| {
            let (functions_found, functions_hit) = if a.functions.is_empty() {
                a.declared_fn
            } else {
                (a.functions.len() as u64, a.functions.values().filter(|n| **n > 0).count() as u64)
            };
            let (branches_found, branches_hit) = if a.branches.is_empty() {
                a.declared_br
            } else {
                (a.branches.len() as u64, a.branches.values().filter(|n| **n > 0).count() as u64)
            };
            let declared_lines = match (a.records, a.declared_lines) {
                (1, (Some(f), Some(h))) if h <= f && f >= a.lines.len() as u64 => Some((f, h)),
                _ => None,
            };
            FileCoverage { path, lines: a.lines, functions_found, functions_hit, branches_found, branches_hit, declared_lines }
        })
        .collect();
    Ok((out, outside.len()))
}

/// A percentage to one decimal place, or none when nothing was counted.
pub fn percent(hit: i64, found: i64) -> Option<f64> {
    (found > 0).then(|| (hit as f64 * 1000.0 / found as f64).round() / 10.0)
}

#[derive(Clone, Debug, sqlx::FromRow)]
pub struct Report {
    pub id: i64,
    pub repo_id: i64,
    pub run_id: i64,
    pub job_id: i64,
    pub job: String,
    pub run_number: i64,
    pub bookmark: String,
    pub state: String,
    pub heads: String,
    pub blob: String,
    pub format: String,
    pub lines_found: i64,
    pub lines_hit: i64,
    pub functions_found: i64,
    pub functions_hit: i64,
    pub branches_found: i64,
    pub branches_hit: i64,
    pub files: i64,
    pub dropped: i64,
    pub created_ms: i64,
}

impl Report {
    pub fn percent(&self) -> Option<f64> {
        percent(self.lines_hit, self.lines_found)
    }

    pub fn functions_percent(&self) -> Option<f64> {
        percent(self.functions_hit, self.functions_found)
    }

    pub fn branches_percent(&self) -> Option<f64> {
        percent(self.branches_hit, self.branches_found)
    }

    pub fn head_list(&self) -> Vec<&str> {
        self.heads.split(',').filter(|h| !h.is_empty()).collect()
    }
}

#[derive(Clone, Debug, sqlx::FromRow)]
pub struct FileRow {
    pub report_id: i64,
    pub path: String,
    pub lines_found: i64,
    pub lines_hit: i64,
    pub functions_found: i64,
    pub functions_hit: i64,
    pub branches_found: i64,
    pub branches_hit: i64,
}

impl FileRow {
    pub fn percent(&self) -> Option<f64> {
        percent(self.lines_hit, self.lines_found)
    }

    pub fn missed(&self) -> i64 {
        self.lines_found - self.lines_hit
    }
}

/// Repositories with a report, so their pages show the Coverage tab:
/// keyed by the hub's repository root and the id, unique even between
/// hubs in one process.
static HAS: Mutex<Option<HashSet<String>>> = Mutex::new(None);

fn key(app: &Shared, repo_id: i64) -> String {
    format!("{}#{repo_id}", app.repos.root().display())
}

pub fn has_reports(app: &Shared, repo_id: i64) -> bool {
    HAS.lock().unwrap_or_else(|p| p.into_inner()).as_ref().is_some_and(|s| s.contains(&key(app, repo_id)))
}

fn mark(app: &Shared, repo_id: i64) {
    HAS.lock().unwrap_or_else(|p| p.into_inner()).get_or_insert_with(HashSet::new).insert(key(app, repo_id));
}

/// At start: which repositories have reports.
pub async fn load(app: &Shared) -> sqlx::Result<()> {
    let ids: Vec<(i64,)> = sqlx::query_as("SELECT DISTINCT repo_id FROM coverage_reports").fetch_all(&app.db).await?;
    for (id,) in ids {
        mark(app, id);
    }
    Ok(())
}

/// What a run's report is kept with.
pub struct Origin<'a> {
    pub repo_id: i64,
    pub run_id: i64,
    pub run_number: i64,
    pub job_id: i64,
    pub job: &'a str,
    pub bookmark: &'a str,
    pub state: &'a str,
    pub heads: &'a [String],
    pub blob: &'a str,
}

/// Keep a parsed report: the files outside `in_state` are dropped and
/// counted. Replaces an earlier report of the same job.
pub async fn keep(app: &Shared, o: &Origin<'_>, files: Vec<FileCoverage>, outside: usize, in_state: &HashSet<String>) -> sqlx::Result<i64> {
    let (kept, missing): (Vec<FileCoverage>, Vec<FileCoverage>) = files.into_iter().partition(|f| in_state.contains(&f.path));
    let sum = |f: &dyn Fn(&FileCoverage) -> u64| kept.iter().map(f).sum::<u64>() as i64;
    let mut tx = crate::db::begin_write(&app.db).await?;
    sqlx::query("DELETE FROM coverage_reports WHERE job_id = ?").bind(o.job_id).execute(&mut *tx).await?;
    let id: i64 = sqlx::query_scalar(
        "INSERT INTO coverage_reports (repo_id, run_id, job_id, job, run_number, bookmark, state, heads, blob, format, \
         lines_found, lines_hit, functions_found, functions_hit, branches_found, branches_hit, files, dropped, created_ms) \
         VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, 'lcov', ?, ?, ?, ?, ?, ?, ?, ?, ?) RETURNING id",
    )
    .bind(o.repo_id)
    .bind(o.run_id)
    .bind(o.job_id)
    .bind(o.job)
    .bind(o.run_number)
    .bind(o.bookmark)
    .bind(o.state)
    .bind(o.heads.join(","))
    .bind(o.blob)
    .bind(sum(&|f| f.lines_found()))
    .bind(sum(&|f| f.lines_hit()))
    .bind(sum(&|f| f.functions_found))
    .bind(sum(&|f| f.functions_hit))
    .bind(sum(&|f| f.branches_found))
    .bind(sum(&|f| f.branches_hit))
    .bind(kept.len() as i64)
    .bind((missing.len() + outside) as i64)
    .bind(now_ms())
    .fetch_one(&mut *tx)
    .await?;
    for f in &kept {
        sqlx::query(
            "INSERT INTO coverage_files (report_id, path, lines_found, lines_hit, functions_found, functions_hit, branches_found, branches_hit) \
             VALUES (?, ?, ?, ?, ?, ?, ?, ?)",
        )
        .bind(id)
        .bind(&f.path)
        .bind(f.lines_found() as i64)
        .bind(f.lines_hit() as i64)
        .bind(f.functions_found as i64)
        .bind(f.functions_hit as i64)
        .bind(f.branches_found as i64)
        .bind(f.branches_hit as i64)
        .execute(&mut *tx)
        .await?;
    }
    tx.commit().await?;
    mark(app, o.repo_id);
    Ok(id)
}

pub async fn report(pool: &Pool, repo_id: i64, id: i64) -> sqlx::Result<Option<Report>> {
    sqlx::query_as("SELECT * FROM coverage_reports WHERE id = ? AND repo_id = ?").bind(id).bind(repo_id).fetch_optional(pool).await
}

/// The newest reports on a bookmark, newest first.
pub async fn on_bookmark(pool: &Pool, repo_id: i64, bookmark: &str, limit: i64) -> sqlx::Result<Vec<Report>> {
    sqlx::query_as("SELECT * FROM coverage_reports WHERE repo_id = ? AND bookmark = ? ORDER BY created_ms DESC, id DESC LIMIT ?")
        .bind(repo_id)
        .bind(bookmark)
        .bind(limit)
        .fetch_all(pool)
        .await
}

pub async fn for_run(pool: &Pool, run_id: i64) -> sqlx::Result<Vec<Report>> {
    sqlx::query_as("SELECT * FROM coverage_reports WHERE run_id = ? ORDER BY job").bind(run_id).fetch_all(pool).await
}

pub async fn files(pool: &Pool, report_id: i64) -> sqlx::Result<Vec<FileRow>> {
    sqlx::query_as("SELECT * FROM coverage_files WHERE report_id = ? ORDER BY path").bind(report_id).fetch_all(pool).await
}

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

    const LCOV: &str = "TN:\nSF:/src/crates/a/src/lib.rs\nFN:3,alpha\nFN:9,beta\nFNDA:4,alpha\nFNDA:0,beta\nFNF:2\nFNH:1\n\
DA:3,4\nDA:4,4\nDA:9,0\nDA:10,0\nBRDA:4,0,0,3\nBRDA:4,0,1,-\nBRF:2\nBRH:1\nLF:4\nLH:2\nend_of_record\n\
SF:/src/crates/a/src/lib.rs\nDA:9,1\nend_of_record\n\
SF:./README.md\nDA:1,1\nend_of_record\n\
SF:/home/runner/.cargo/registry/src/dep/lib.rs\nDA:1,1\nend_of_record\n";

    #[test]
    fn lcov_reports_are_read_and_merged() {
        let (files, outside) = parse_lcov(LCOV).unwrap();
        assert_eq!(outside, 1, "the dependency outside the workspace");
        assert_eq!(files.len(), 2);
        let lib = &files[1];
        assert_eq!(lib.path, "crates/a/src/lib.rs");
        assert_eq!(lib.lines, [(3, 4), (4, 4), (9, 1), (10, 0)].into_iter().collect());
        assert_eq!((lib.lines_found(), lib.lines_hit()), (4, 3), "the second record ran line 9");
        assert_eq!((lib.functions_found, lib.functions_hit), (2, 1));
        assert_eq!((lib.branches_found, lib.branches_hit), (2, 1));
        assert_eq!(files[0].path, "README.md");
        assert_eq!(lib.declared_lines, None, "two records: counted from the lines");
        // One record whose totals name lines its DA lines leave out, as llvm-cov writes.
        let (one, _) = parse_lcov("SF:/src/a.rs\nDA:1,1\nDA:2,0\nLF:5\nLH:3\nend_of_record\n").unwrap();
        assert_eq!((one[0].lines_found(), one[0].lines_hit()), (5, 3));
        assert!(parse_lcov("not a report\n").unwrap_err().contains("not an LCOV report"));
    }

    #[test]
    fn paths_are_taken_relative_to_the_workspace() {
        assert_eq!(relative("/src/a/b.rs").as_deref(), Some("a/b.rs"));
        assert_eq!(relative("./a/b.rs").as_deref(), Some("a/b.rs"));
        assert_eq!(relative("a/b.rs").as_deref(), Some("a/b.rs"));
        for bad in ["/etc/passwd", "a/../b", "", "/src/", "a//b", "C:\\x\\y.rs"] {
            assert_eq!(relative(bad), None, "{bad}");
        }
        assert_eq!(percent(1, 3), Some(33.3));
        assert_eq!(percent(0, 0), None);
    }
}