smith

//! Staged nightly CI (`SMH-SPEC-SPEC0001`, testing acceptance).

use std::{ffi::OsString, fs, path::Path, process::ExitCode};

const SHARDS: usize = 8;
const LISTS: [&str; 4] = ["caught.txt", "missed.txt", "timeout.txt", "unviable.txt"];

/// Each Luci matrix child restores the same job's caches. Bootstrap clears
/// evidence; shard finding codes defer failure; report publishes before verdict.
pub(crate) fn run(extra: &[OsString]) -> ExitCode {
    let Some((phase, extra)) = crate::without_separator(extra).split_first() else {
        return crate::fail("nightly needs bootstrap, 0..7, report, or verdict\n");
    };
    let Some(phase) = phase.to_str() else {
        return crate::fail("nightly phase must be UTF-8\n");
    };
    execute(&crate::target_dir(), phase, extra, |step, args| step(args))
}

fn execute(
    target: &Path,
    phase: &str,
    extra: &[OsString],
    mut call: impl FnMut(crate::Step, &[OsString]) -> ExitCode,
) -> ExitCode {
    let stages = target.join("nightly");
    let summary = target.join("nightly-summary");
    let output = stages.join(phase);
    let result = (|| -> std::io::Result<ExitCode> {
        match phase {
            "bootstrap" => {
                if stages.exists() {
                    fs::remove_dir_all(&stages)?;
                }
                if summary.exists() {
                    fs::remove_dir_all(&summary)?;
                }
                for (step, args) in [
                    (crate::bootstrap as crate::Step, &[][..]),
                    (crate::cargo_nextest, extra),
                    (crate::ub, extra),
                ] {
                    let status = call(step, args);
                    if status != ExitCode::SUCCESS {
                        return Ok(status);
                    }
                }
                fs::create_dir_all(&output)?;
                fs::write(
                    output.join("index.md"),
                    "# Nightly bootstrap\n\nBaseline tests and Miri passed.\n",
                )?;
            }
            "report" => {
                aggregate(&stages, &summary)?;
                let status = call(report_stage, extra);
                if status != ExitCode::SUCCESS {
                    return Ok(status);
                }
                let report = target.join("report");
                for name in LISTS {
                    fs::copy(summary.join(name), report.join(name))?;
                }
                if output.exists() {
                    fs::remove_dir_all(&output)?;
                }
                fs::rename(report, &output)?;
            }
            "verdict" => {
                let report = stages.join("report");
                // No report means no complete run, even if stale shard lists exist.
                if !report.join("index.md").is_file()
                    || !LISTS.iter().all(|name| report.join(name).is_file())
                {
                    return Ok(crate::fail("nightly verdict requires the report stage\n"));
                }
                let Some([_, missed, timeout, _]) = crate::mutants_counts(&report) else {
                    return Ok(crate::fail("nightly verdict requires mutation evidence\n"));
                };
                fs::create_dir_all(&output)?;
                fs::write(
                    output.join("index.md"),
                    format!("# Nightly verdict\n\nMissed: {missed}; timeout: {timeout}.\n"),
                )?;
                if missed != 0 || timeout != 0 {
                    return Ok(crate::fail(
                        "nightly mutation findings; see the published report\n",
                    ));
                }
            }
            _ => {
                let shard = phase
                    .parse::<usize>()
                    .ok()
                    .filter(|&shard| shard < SHARDS && shard.to_string() == phase);
                let Some(shard) = shard else {
                    return Ok(crate::fail("unknown nightly phase\n"));
                };
                if output.exists() {
                    fs::remove_dir_all(&output)?;
                }
                let mut args = Vec::with_capacity(extra.len() + 5);
                args.push(OsString::from("--all"));
                args.extend_from_slice(extra);
                args.extend([
                    OsString::from("--shard"),
                    format!("{shard}/{SHARDS}").into(),
                    OsString::from("--output"),
                    output.into_os_string(),
                ]);
                let status = call(crate::cargo_mutants, &args);
                if ![ExitCode::SUCCESS, ExitCode::from(2), ExitCode::from(3)].contains(&status) {
                    return Ok(status);
                }
                let evidence = stages.join(phase).join("mutants.out");
                for name in LISTS.into_iter().chain(["outcomes.json"]) {
                    fs::metadata(evidence.join(name))?;
                }
            }
        }
        Ok(ExitCode::SUCCESS)
    })();
    result.unwrap_or_else(|error| crate::fail(&format!("nightly {phase}: {error}\n")))
}

fn report_stage(extra: &[OsString]) -> ExitCode {
    crate::report_with_mutants(extra, &crate::target_dir().join("nightly-summary"))
}

/// Preserve each shard's JSON evidence; merge only the four text worklists
/// consumed by the existing report. Missing shards are infrastructure failure.
fn aggregate(stages: &Path, summary: &Path) -> std::io::Result<()> {
    fs::create_dir_all(summary)?;
    for name in LISTS {
        let mut combined = String::new();
        for shard in 0..SHARDS {
            let source = stages.join(shard.to_string()).join("mutants.out");
            let list = fs::read_to_string(source.join(name))?;
            combined.push_str(&list);
            if !list.is_empty() && !list.ends_with('\n') {
                combined.push('\n');
            }
        }
        fs::write(summary.join(name), combined)?;
    }
    for shard in 0..SHARDS {
        fs::copy(
            stages
                .join(shard.to_string())
                .join("mutants.out/outcomes.json"),
            summary.join(format!("shard-{shard}.json")),
        )?;
    }
    Ok(())
}

#[cfg(test)]
mod tests {
    use std::{ffi::OsString, fs, path::Path, process::ExitCode};

    fn fixture(name: &str) -> std::path::PathBuf {
        let root =
            std::env::temp_dir().join(format!("smith-nightly-{name}-{}", std::process::id()));
        let _ = fs::remove_dir_all(&root);
        fs::create_dir_all(&root).unwrap();
        root
    }

    fn evidence(root: &Path, shard: usize, missed: bool) {
        let dir = root
            .join("nightly")
            .join(shard.to_string())
            .join("mutants.out");
        fs::create_dir_all(&dir).unwrap();
        for name in ["caught", "missed", "timeout", "unviable"] {
            fs::write(
                dir.join(format!("{name}.txt")),
                if name == "missed" && missed {
                    "miss\n"
                } else {
                    ""
                },
            )
            .unwrap();
        }
        fs::write(dir.join("outcomes.json"), "{}").unwrap();
    }

    #[test]
    fn findings_keep_shards_running_but_fail_the_verdict() {
        let root = fixture("findings");
        for shard in 0..8 {
            let status = super::execute(&root, &shard.to_string(), &[], |step, args| {
                assert!(std::ptr::fn_addr_eq(
                    step,
                    crate::cargo_mutants as crate::Step
                ));
                assert_eq!(
                    args,
                    &[
                        OsString::from("--all"),
                        OsString::from("--shard"),
                        OsString::from(format!("{shard}/8")),
                        OsString::from("--output"),
                        root.join("nightly")
                            .join(shard.to_string())
                            .into_os_string(),
                    ]
                );
                evidence(&root, shard, shard == 0);
                if shard == 0 {
                    ExitCode::from(2)
                } else {
                    ExitCode::SUCCESS
                }
            });
            assert_eq!(status, ExitCode::SUCCESS);
        }
        let status = super::execute(&root, "report", &[], |_, _| {
            let out = root.join("report");
            fs::create_dir_all(&out).unwrap();
            fs::write(out.join("index.md"), "report").unwrap();
            ExitCode::SUCCESS
        });
        assert_eq!(status, ExitCode::SUCCESS);
        assert!(root.join("nightly/report/index.md").exists());
        assert_eq!(
            fs::read_to_string(root.join("nightly/report/missed.txt")).unwrap(),
            "miss\n"
        );
        assert_eq!(
            super::execute(&root, "verdict", &[], |_, _| panic!(
                "verdict runs no tools"
            )),
            ExitCode::FAILURE
        );
        fs::remove_dir_all(root).unwrap();
    }

    #[test]
    fn infrastructure_errors_are_not_findings() {
        for (name, status, write) in [
            ("infra", ExitCode::FAILURE, true),
            ("missing", ExitCode::SUCCESS, false),
            ("timeout", ExitCode::from(3), true),
        ] {
            let root = fixture(name);
            let result = super::execute(&root, "0", &[], |_, _| {
                if write {
                    evidence(&root, 0, false);
                }
                status
            });
            assert_eq!(
                result,
                if write && status == ExitCode::from(3) {
                    ExitCode::SUCCESS
                } else {
                    ExitCode::FAILURE
                }
            );
            fs::remove_dir_all(root).unwrap();
        }
    }

    #[test]
    fn report_requires_every_shard_without_reusing_old_evidence() {
        let root = fixture("incomplete");
        evidence(&root, 0, false);
        assert_eq!(
            super::execute(&root, "report", &[], |_, _| panic!(
                "incomplete shards must stop before coverage"
            )),
            ExitCode::FAILURE
        );
        assert_eq!(
            super::execute(&root, "bootstrap", &[], |_, _| ExitCode::SUCCESS),
            ExitCode::SUCCESS
        );
        assert!(!root.join("nightly/0").exists());
        assert_eq!(
            super::execute(&root, "verdict", &[], |_, _| panic!(
                "verdict runs no tools"
            )),
            ExitCode::FAILURE
        );
        fs::remove_dir_all(root).unwrap();
    }

    #[test]
    fn bootstrap_verifies_baseline_before_miri_and_stops_on_failure() {
        let root = fixture("baseline");
        let extra = [OsString::from("-p"), OsString::from("xtask")];
        let mut calls = 0;
        let status = super::execute(&root, "bootstrap", &extra, |step, args| {
            calls += 1;
            match calls {
                1 => {
                    assert!(std::ptr::fn_addr_eq(step, crate::bootstrap as crate::Step));
                    assert!(args.is_empty());
                    ExitCode::SUCCESS
                }
                2 => {
                    assert!(std::ptr::fn_addr_eq(
                        step,
                        crate::cargo_nextest as crate::Step
                    ));
                    assert_eq!(args, extra);
                    ExitCode::FAILURE
                }
                _ => panic!("Miri must not run after failing baseline"),
            }
        });
        assert_eq!(calls, 2);
        assert_eq!(status, ExitCode::FAILURE);
        fs::remove_dir_all(root).unwrap();
    }

    #[test]
    fn complete_run_passes_but_timeouts_fail_the_verdict() {
        let root = fixture("verdict");
        for shard in 0..8 {
            evidence(&root, shard, false);
        }
        let report = || {
            assert_eq!(
                super::execute(&root, "report", &[], |_, _| {
                    fs::create_dir_all(root.join("report")).unwrap();
                    fs::write(root.join("report/index.md"), "report").unwrap();
                    ExitCode::SUCCESS
                }),
                ExitCode::SUCCESS
            );
        };
        report();
        assert_eq!(
            super::execute(&root, "verdict", &[], |_, _| panic!("no tools")),
            ExitCode::SUCCESS
        );
        fs::write(root.join("nightly/7/mutants.out/timeout.txt"), "timeout").unwrap();
        report();
        assert_eq!(
            fs::read_to_string(root.join("nightly/report/timeout.txt")).unwrap(),
            "timeout\n"
        );
        assert_eq!(
            super::execute(&root, "verdict", &[], |_, _| panic!("no tools")),
            ExitCode::FAILURE
        );
        fs::remove_dir_all(root).unwrap();
    }

    #[test]
    fn native_exit_codes_distinguish_findings_from_failed_runs() {
        for code in [0, 1, 2, 3, 4, 5, 6, 70] {
            let expected = match code {
                0 => ExitCode::SUCCESS,
                2 => ExitCode::from(2),
                3 => ExitCode::from(3),
                _ => ExitCode::FAILURE,
            };
            assert_eq!(crate::mutants_exit(Some(code)), expected);
        }
        assert_eq!(crate::mutants_exit(None), ExitCode::FAILURE);
    }

    #[test]
    fn invalid_phases_do_not_run_tools() {
        let root = fixture("invalid");
        for phase in ["", "8", "-1", "00", "whatever"] {
            assert_eq!(
                super::execute(&root, phase, &[], |_, _| panic!(
                    "invalid phases must not run tools"
                )),
                ExitCode::FAILURE
            );
        }
        fs::remove_dir_all(root).unwrap();
    }

    #[test]
    fn kdl_stages_share_one_job_in_execution_order() {
        let config = include_str!("../../.ci/nightly.kdl");
        assert!(config.contains("matrix {\n    phase \"bootstrap\" \"0\" \"1\" \"2\" \"3\" \"4\" \"5\" \"6\" \"7\" \"report\" \"verdict\"\n  }"));
        assert!(config.contains("cargo run --locked -p xtask -- nightly ${phase}"));
        assert!(config.contains("from \"target/nightly/${phase}\""));
        assert!(!config.contains("|| echo"));
    }
}