alkblobs/tests/git_interop.rs

//! The load-bearing interop test (ADR-002; POC #1 findings 4/5's
//! standard): byte-exact derivation against the `git hash-object` CLI
//! for both algorithms — validated against git, not just hardcoded
//! vectors — plus the external-oid round-trip into pool keys. The
//! git-object derivation family (ADR-018) rides the same standard:
//! the family's digest over git's loose-object form is byte-exact
//! git's oid of non-blob kinds, CLI-validated over real
//! `git hash-object -t` bodies, tree/commit/tag.
//!
//! Harness detail (POC finding 4, empirically learned): in current
//! git, `--object-format` is not a `hash-object` flag; the object
//! format is set on the repo (`git init --object-format=sha256`) and
//! the hash-object runs inside that repo. Repos are created fresh in
//! a per-call temp directory, with global/system config detached.

use alkblobs::{Algorithm, Key};
use std::fs;
use std::path::{Path, PathBuf};
use std::process::Command;
use std::time::{SystemTime, UNIX_EPOCH};

/// Family-explicit construction from an external oid (ADR-020 §Decision
/// 5 — the interop constructor died with its length-implied family):
/// the registry byte names the family, the oid bytes follow, and
/// `Key::from_bytes` validates the pair over the physical encoding —
/// the same decoder the backend boundary and the wire's ingress trust.
fn oid_key(oid_hex: &str, algorithm: Algorithm) -> Key {
    let oid = hex::decode(oid_hex).expect("git printed lowercase hex");
    assert_eq!(oid.len(), algorithm.digest_len(), "oid width matches");
    let mut tagged = vec![algorithm.byte()];
    tagged.extend_from_slice(&oid);
    Key::from_bytes(&tagged).expect("the tagged oid key decodes")
}

/// Content cases: empty, short text, binary-with-NULs, and a
/// multi-KiB case (75 KiB, deterministic).
fn content_cases() -> Vec<(&'static str, Vec<u8>)> {
    vec![
        ("empty", Vec::new()),
        ("short", b"hello".to_vec()),
        (
            "binary-nuls",
            vec![0u8, 1, 255, 0, 10, 13, 0, 7, 255, 255, 0, 0],
        ),
        (
            "multi-kib",
            (0..75 * 1024).map(|i| (i % 251) as u8).collect::<Vec<u8>>(),
        ),
    ]
}

fn scratch_dir(tag: &str) -> PathBuf {
    let nanos = SystemTime::now()
        .duration_since(UNIX_EPOCH)
        .expect("system clock before epoch")
        .as_nanos();
    let dir = std::env::temp_dir().join(format!(
        "alkblobs-git-interop-{}-{}-{}",
        tag,
        std::process::id(),
        nanos
    ));
    fs::create_dir_all(&dir).expect("temp dir creation");
    dir
}

fn detached_config(command: &mut Command) {
    let devnull = if cfg!(windows) { "NUL" } else { "/dev/null" };
    command
        .env("GIT_CONFIG_GLOBAL", devnull)
        .env("GIT_CONFIG_SYSTEM", devnull);
}

/// Initialize a fresh repo of the requested object format and return
/// the oid hex `git hash-object --stdin` prints for `content`. With a
/// `kind` the object type is set (`-t tree|commit|tag` — non-blob
/// hashes validate the body as that kind); `None` hashes as a blob.
fn git_hash_object(content: &[u8], sha256: bool) -> (PathBuf, String) {
    git_hash_object_kind(content, sha256, None)
}

fn git_hash_object_kind(content: &[u8], sha256: bool, kind: Option<&str>) -> (PathBuf, String) {
    let dir = scratch_dir("repo");
    let mut init = Command::new("git");
    detached_config(&mut init);
    init.arg("init").arg("-q").current_dir(&dir);
    if sha256 {
        init.arg("--object-format=sha256");
    }
    let out = init.output().expect("git init spawn");
    assert!(
        out.status.success(),
        "git init failed: {}",
        String::from_utf8_lossy(&out.stderr)
    );

    let mut hash = Command::new("git");
    detached_config(&mut hash);
    hash.arg("-C").arg(&dir).arg("hash-object").arg("--stdin");
    if let Some(kind) = kind {
        hash.arg("-t").arg(kind);
    }
    use std::io::Write as _;
    let mut child = hash
        .stdin(std::process::Stdio::piped())
        .stdout(std::process::Stdio::piped())
        .spawn()
        .expect("git hash-object spawn");
    child
        .stdin
        .as_mut()
        .expect("stdin piped")
        .write_all(content)
        .expect("stdin write");
    let out = child.wait_with_output().expect("git hash-object output");
    assert!(
        out.status.success(),
        "git hash-object failed: {}",
        String::from_utf8_lossy(&out.stderr)
    );
    let oid = String::from_utf8(out.stdout)
        .expect("git output utf8")
        .trim()
        .to_owned();
    (dir, oid)
}

fn drop_dir(dir: &Path) {
    let _ = fs::remove_dir_all(dir);
}

/// The load-bearing test: `Digest`'s derivation matches the git CLI
/// byte-exact, both algorithms, small and multi-KiB content.
#[test]
fn derivation_is_byte_exact_against_git_hash_object() {
    for (tag, content) in content_cases() {
        let (dir, oid) = git_hash_object(&content, false);
        drop_dir(&dir);
        assert_eq!(oid.len(), 40, "git printed a sha1 oid");
        assert_eq!(
            oid_key(&oid, Algorithm::GitBlobSha1)
                .digest()
                .bytes()
                .to_vec(),
            Algorithm::GitBlobSha1
                .derive(&content, content.len() as u64)
                .bytes()
                .to_vec(),
            "sha1 {tag}: derived digest != git hash-object",
        );

        let (dir, oid) = git_hash_object(&content, true);
        drop_dir(&dir);
        assert_eq!(oid.len(), 64, "git printed a sha256 oid");
        assert_eq!(
            oid_key(&oid, Algorithm::GitBlobSha256)
                .digest()
                .bytes()
                .to_vec(),
            Algorithm::GitBlobSha256
                .derive(&content, content.len() as u64)
                .bytes()
                .to_vec(),
            "sha256 {tag}: derived digest != git hash-object",
        );
    }
}

/// The external-oid round-trip (POC finding 4 corollary, key level):
/// an oid produced by external git (`Key::from_bytes` over the
/// physical encoding → `as_bytes` → `from_bytes`) becomes exactly the
/// pool key the store core would derive — no mapping layer.
#[test]
fn external_oid_round_trips_into_pool_keys() {
    for sha256 in [false, true] {
        let content = b"external producer blob \0 bytes";
        let (dir, oid) = git_hash_object(content, sha256);
        drop_dir(&dir);

        let algorithm = if sha256 {
            Algorithm::GitBlobSha256
        } else {
            Algorithm::GitBlobSha1
        };
        let external: Key = oid_key(&oid, algorithm);
        let derived: Key = Algorithm::GitBlobSha256
            .derive(content, content.len() as u64)
            .into();
        if sha256 {
            assert_eq!(external, derived, "sha256: external oid == derived key");
        } else {
            assert_eq!(
                external,
                Key::from(Algorithm::GitBlobSha1.derive(content, content.len() as u64)),
                "sha1: external oid == derived key"
            );
        }
        let decoded = Key::from_bytes(&external.as_bytes()).expect("physical encoding decodes");
        assert_eq!(decoded, external);
        assert_eq!(decoded.as_bytes(), external.as_bytes());
    }
}

/// The facade put seam against git (store-api.md §Invariants 1's
/// source-of-truth leg): both put arms (known-length, unknown-length)
/// key entries git-identically on both tiers under the named blob
/// derivation, and the get path round-trips git's content
/// byte-exactly. mem-gated: the wave-2 constructible shape
/// (`dual_tier_mem_local`).
#[cfg(feature = "mem")]
#[tokio::test]
async fn put_seam_keys_match_git_hash_object_on_both_tiers_and_paths() {
    use std::io::Cursor;

    use alkblobs::Store;

    let root = scratch_dir("store-root");
    let store = Store::dual_tier_mem_local(root.join("root"), 32)
        .await
        .expect("dual-tier store constructs");
    let small = b"small git-interop blob";
    let big = (0..200).map(|i| (i % 253) as u8).collect::<Vec<u8>>();

    let (repo, oid) = git_hash_object(small, true);
    drop_dir(&repo);
    let expected_small: Key = oid_key(&oid, Algorithm::GitBlobSha256);
    let (repo, oid) = git_hash_object(&big, true);
    drop_dir(&repo);
    let expected_big: Key = oid_key(&oid, Algorithm::GitBlobSha256);

    let (key, pin) = store
        .put_as(
            Algorithm::GitBlobSha256,
            Some(small.len() as u64),
            Cursor::new(&small[..]),
        )
        .await
        .expect("known-length put");
    assert_eq!(key, expected_small, "kv put keys git-identical");
    drop(pin);
    let (key, pin) = store
        .put_as(Algorithm::GitBlobSha256, None, Cursor::new(&small[..]))
        .await
        .expect("unknown-length put");
    assert_eq!(key, expected_small, "buffered form keys git-identical");
    let (len, blob) = store
        .get(&expected_small.as_bytes())
        .await
        .expect("get succeeds")
        .expect("entry present");
    assert_eq!(blob.read_at(0, len).await.unwrap(), &small[..]);
    drop(pin);

    let (key, pin) = store
        .put_as(
            Algorithm::GitBlobSha256,
            Some(big.len() as u64),
            Cursor::new(&big[..]),
        )
        .await
        .expect("known-length large put");
    assert_eq!(key, expected_big, "staged put keys git-identical");
    drop(pin);
    let (key, pin) = store
        .put_as(Algorithm::GitBlobSha256, None, Cursor::new(&big[..]))
        .await
        .expect("unknown-length large put");
    assert_eq!(
        key, expected_big,
        "staged-bytes derivation keys git-identical"
    );
    let (len, blob) = store
        .get(&expected_big.as_bytes())
        .await
        .expect("get succeeds")
        .expect("entry present");
    assert_eq!(blob.read_at(0, len).await.unwrap(), big);
    drop(pin);

    let _ = fs::remove_dir_all(root);
}

/// Tagged-key coexistence (POC finding 5): git oids of both formats
/// address the same flat key universe from one external producer.
#[test]
fn both_oid_formats_coexist_in_one_pool() {
    let content = b"same content, two formats";
    let (dir1, oid1) = git_hash_object(content, false);
    drop_dir(&dir1);
    let (dir2, oid2) = git_hash_object(content, true);
    drop_dir(&dir2);

    let mut pool = std::collections::BTreeSet::new();
    pool.insert(oid_key(&oid1, Algorithm::GitBlobSha1));
    pool.insert(oid_key(&oid2, Algorithm::GitBlobSha256));
    assert_eq!(pool.len(), 2, "both algorithms coexist in one flat pool");
}

/// The git-object derivation family (ADR-018): the digest over git's
/// loose-object form (`"<type> <len>\0" + content`) is byte-exact
/// git's oid of the object, for non-blob kinds. The reference oids
/// come from the CLI itself (`hash-object -t <kind> --stdin` over a
/// crafted body — the CLI prepends its own header, so the bodies are
/// bare contents and the CLI computes the true oid; a tree/commit/tag
/// body's shape keeps the CLI's kind validation honest).
#[test]
fn git_object_derivation_is_byte_exact_against_git_hash_object() {
    let tree_body: Vec<u8> = {
        // "100644 f\0" + sha1 'hello' bytes
        let mut t = "100644 f\0".as_bytes().to_vec();
        t.extend_from_slice(
            &hex::decode("b6fc4c620b67d95f953a5c1c1230aaab5db5a1b0").expect("blob oid decodes"),
        );
        t
    };
    let commit_body = format!(
        "tree {}\nauthor t <t@t> 1792210192 +0000\ncommitter t <t@t> 1792210192 +0000\n\nx\n",
        hex::encode(
            Algorithm::GitObjectSha1
                .derive(&tree_body, tree_body.len() as u64)
                .bytes()
        )
    )
    .into_bytes();
    let tag_body = format!(
        "object {}\ntype commit\ntag v1\ntagger t <t@t> 1792210192 +0000\n\nx\n",
        hex::encode(
            Algorithm::GitObjectSha1
                .derive(&commit_body, commit_body.len() as u64)
                .bytes()
        )
    )
    .into_bytes();

    for (kind, body) in [
        ("tree", tree_body),
        ("commit", commit_body),
        ("tag", tag_body),
    ] {
        let (dir, oid) = git_hash_object_kind(&body, false, Some(kind));
        drop_dir(&dir);
        assert_eq!(oid.len(), 40);

        let loose_form = {
            let mut v = format!("{} {}\0", kind, body.len()).into_bytes();
            v.extend_from_slice(&body);
            v
        };
        let derived = Algorithm::GitObjectSha1.derive(&loose_form, loose_form.len() as u64);
        assert_eq!(
            hex::encode(derived.bytes()),
            oid,
            "{kind}: git-object derivation != git's oid (body len {})",
            body.len()
        );

        let key = Key::from_bytes(&Key::from(derived).as_bytes()).expect("git-object key decodes");
        assert_eq!(key.algorithm(), Algorithm::GitObjectSha1);
    }

    // The sha-256 sibling: same loose form, 64-hex oid from the CLI.
    let (dir, oid) = git_hash_object(b"x", true);
    drop_dir(&dir);
    let loose = b"blob 1\0x";
    let derived = Algorithm::GitObjectSha256.derive(loose, loose.len() as u64);
    assert_eq!(
        hex::encode(derived.bytes()),
        oid,
        "sha256 git-object derivation"
    );
}

/// The family's facade put/get round trip (store level): a non-blob
/// object put through the one put seam under the family's algorithm
/// lands under its git oid key, verifies on get, dedups, and — the
/// negative evidence for DQ-1's original problem statement — is not
/// reachable through the blob-family tags (no cross-family collision:
/// the same oid digest under a different tag addresses nothing).
#[cfg(feature = "mem")]
#[tokio::test]
async fn put_as_family_member_round_trips_under_the_oid_key() {
    use std::io::Cursor;

    use alkblobs::Store;

    let store = Store::mem_only();
    let tree_body: Vec<u8> = {
        let mut t = "100644 f\0".as_bytes().to_vec();
        t.extend_from_slice(
            &hex::decode("b6fc4c620b67d95f953a5c1c1230aaab5db5a1b0").expect("blob oid decodes"),
        );
        t
    };
    let loose_form = {
        let mut v = format!("tree {}\0", tree_body.len()).into_bytes();
        v.extend_from_slice(&tree_body);
        v
    };

    let (key, pin) = store
        .put_as(
            Algorithm::GitObjectSha1,
            Some(loose_form.len() as u64),
            Cursor::new(loose_form.clone()),
        )
        .await
        .expect("git-object put");
    assert_eq!(
        key.algorithm(),
        Algorithm::GitObjectSha1,
        "the key lands under the git-object family"
    );
    drop(pin);

    // Dedup by construction: the re-put resolves to the same key.
    let (key2, pin2) = store
        .put_as(
            Algorithm::GitObjectSha1,
            Some(loose_form.len() as u64),
            Cursor::new(&loose_form[..]),
        )
        .await
        .expect("re-put");
    assert_eq!(key2, key);
    drop(pin2);

    // The get verifies byte-exactly (verification-on-get against the
    // preimage-identity derivation) and round-trips.
    let (len, blob) = store
        .get(&key.as_bytes())
        .await
        .expect("get ok")
        .expect("entry present");
    let value = blob.read_at(0, len).await.expect("read");
    assert_eq!(value, loose_form);

    // The stat semantic (ADR-018 §Decision 3): the preimage length —
    // the loose form, git's header included.
    let meta = store
        .stat(&key.as_bytes())
        .await
        .expect("stat ok")
        .expect("stat sees the entry");
    assert_eq!(
        meta.len,
        loose_form.len() as u64,
        "stat.len is the stored-preimage length"
    );
    assert_eq!(meta.algorithm, Algorithm::GitObjectSha1);

    // The negative evidence: the same oid addressed through the
    // blob-family tag finds nothing (the tree oid is not a blob key) —
    // family-explicit construction names the tag; this one names the
    // wrong family on purpose (Key::from_bytes over [blob tag] + oid).
    let blob_tagged = {
        let mut tagged = vec![Algorithm::GitBlobSha1.byte()];
        tagged.extend_from_slice(key.digest().bytes());
        tagged
    };
    let blob_tagged_key = Key::from_bytes(&blob_tagged).expect("blob-tag key decodes");
    assert!(
        store
            .get(&blob_tagged_key.as_bytes())
            .await
            .expect("blob-tag get ok")
            .is_none(),
        "the tree oid is not reachable through the blob tags"
    );
}

/// The seam's blob arm is the old canonical put's behavior (ADR-020
/// §Decision 5 — no dedicated method remains to guard): a blob
/// algorithm through `put_as` lands under the canonical blob
/// derivation, git-identically — the known-answer vector is
/// `git hash-object`'s.
#[cfg(feature = "mem")]
#[tokio::test]
async fn put_as_blob_algorithm_lands_under_the_canonical_derivation() {
    use std::io::Cursor;

    use alkblobs::Store;

    let store = Store::mem_only();
    let content = b"hello";
    let expected = [
        (
            "8aec4e4876f854f688d0ebfc8f37598f38e5fd6903cccc850ca36591175aeb60",
            Algorithm::GitBlobSha256,
        ),
        (
            "b6fc4c620b67d95f953a5c1c1230aaab5db5a1b0",
            Algorithm::GitBlobSha1,
        ),
    ];
    for (vector, algorithm) in expected {
        let (key, pin) = store
            .put_as(algorithm, Some(5), Cursor::new(content))
            .await
            .expect("blob algorithms are legal at the seam");
        assert_eq!(key.algorithm(), algorithm);
        assert_eq!(
            hex::encode(key.digest().bytes()),
            vector,
            "{algorithm:?}: the seam's blob arm is the canonical derivation"
        );
        drop(pin);
    }
}