distribution/tests/store_content_addressed.sx

// Acceptance for P2.2 — the content-addressed artifact store.
//
// Drives a fresh store rooted under `.sx-tmp/` (never /tmp) and asserts
// the four Slice-3 invariants:
//   1. put → object lands at `objects/<sha256>` and its bytes round-trip;
//      the storage key equals std.hash, an independent `shasum -a 256`,
//      and the pinned SHA-256("abc") vector.
//   2. dedup — identical bytes are not stored twice and an existing
//      object is never rewritten.
//   3. atomicity — a staged-but-unpublished write is invisible at the
//      final path, and a publish that fails before/at the rename leaves
//      no object.
//   4. put_file — a file source produces the same key and bytes.
// Exits 0 only if every assertion holds (process.assert aborts otherwise).
#import "modules/std.sx";
fs      :: #import "modules/std/fs.sx";
hash    :: #import "modules/std/hash.sx";
process :: #import "modules/std/process.sx";
#import "../src/store/store.sx";

// SHA-256("abc"), the FIPS 180-4 one-block known-answer vector.
ABC_SHA256 :: "ba7816bf8f01cfea414140de5dae2223b00361a396177a9cb410ff61f20015ad";

// std.hash digest of `s` as a heap string key (independent of the store).
stdhash_key :: (s: string) -> string {
    d := hash.sha256_hex(s);
    view := string.{ ptr = @d[0], len = 64 };
    return substr(view, 0, 64);
}

// First 64 hex chars of `shasum -a 256` over `bytes`, via the shell.
// `bytes` must be shell-safe (the fixtures here are plain ASCII).
shasum_key :: (bytes: string) -> string {
    cmd := concat("printf '%s' ", concat(bytes, " | shasum -a 256"));
    r := process.run(cmd);
    process.assert(r != null, "shasum -a 256 must run");
    res := r!;
    process.assert(res.exit_code == 0, "shasum -a 256 must exit 0");
    return substr(res.stdout, 0, 64);
}

// Number of directory entries under `dir`, parsed from `ls -1 | wc -l`.
entry_count :: (dir: string) -> string {
    cmd := concat("ls -1 ", concat(dir, " | wc -l | tr -dc '0-9'"));
    r := process.run(cmd);
    process.assert(r != null, "ls/wc must run");
    res := r!;
    return res.stdout;
}

// Number of `put_file` staging temps (`incoming-*`) left under `dir`.
// 0 means every file-source put cleaned up its staging copy.
incoming_count :: (dir: string) -> string {
    cmd := concat("ls -1 ", concat(dir, " 2>/dev/null | grep -c '^incoming-' | tr -dc '0-9'"));
    r := process.run(cmd);
    process.assert(r != null, "ls/grep must run");
    res := r!;
    if res.stdout.len == 0 { return "0"; }
    return res.stdout;
}

main :: () -> i32 {
    root := ".sx-tmp/store-cas";
    process.run(concat("rm -rf ", root));   // fresh root, even after a crashed prior run

    st := Store.init(root);

    // ── 1. put + content addressing ─────────────────────────────────────
    fixture := "abc";
    key, e := st.put_bytes(fixture);
    process.assert(!e, "put_bytes(abc) must succeed");
    process.assert(key == ABC_SHA256, "key must equal pinned SHA-256(abc) vector");
    process.assert(key == stdhash_key(fixture), "store key must equal std.hash digest");
    process.assert(key == shasum_key(fixture), "store key must equal shasum -a 256");
    print("  store == std.hash == shasum == vector: {}\n", key);

    process.assert(st.has(key), "object must exist at objects/<sha256>");
    stored := fs.read_file(st.object_path(key));
    process.assert(stored != null, "stored object must be readable");
    process.assert(stored! == fixture, "stored bytes must equal the input");

    // ── 2. dedup: same bytes, one object, never rewritten ───────────────
    // Overwrite the object on disk; a deduped re-put must NOT touch it.
    process.assert(fs.write_file(st.object_path(key), "TAMPERED"), "tamper write must succeed");
    key2, e2 := st.put_bytes(fixture);
    process.assert(!e2, "second put_bytes must succeed");
    process.assert(key2 == key, "dedup: identical bytes yield the same key");
    after := fs.read_file(st.object_path(key));
    process.assert(after! == "TAMPERED", "dedup: existing object must not be rewritten");
    process.assert(entry_count(st.objects_dir()) == "1", "dedup: exactly one object stored");
    // Restore the real bytes so the store is left consistent.
    process.assert(fs.write_file(st.object_path(key), fixture), "restore write must succeed");
    print("  dedup: one object, copy skipped on re-put\n");

    // ── 3. atomicity: staged write is invisible until publish ───────────
    pending := "interrupted-upload-bytes";
    pkey := stdhash_key(pending);
    process.assert(!st.has(pkey), "fresh store: pending object must be absent");
    sp, se := st.stage_write(pkey, pending);
    process.assert(!se, "stage_write must succeed");
    process.assert(fs.exists(sp), "staged file must exist after stage_write");
    process.assert(!st.has(pkey), "atomicity: object must NOT exist before the rename");

    // A publish whose staging source is missing fails and creates nothing.
    missing := "1111111111111111111111111111111111111111111111111111111111111111";
    process.assert(!st.has(missing), "precondition: no object for the missing key");
    failed := false;
    st.publish(st.staging_path(missing), missing) catch { failed = true; };
    process.assert(failed, "publish of a missing staging file must fail");
    process.assert(!st.has(missing), "failed publish must leave no object");
    print("  atomicity: staged write invisible; failed publish leaves no object\n");

    // ── 4. put_file: single source read, key == digest of published object
    src := ".sx-tmp/store-cas-src.bin";
    file_bytes := "file-source-bytes-123";  // shell-safe: no spaces/newlines
    process.assert(fs.write_file(src, file_bytes), "fixture source file must be written");

    fkey, fe := st.put_file(src);
    process.assert(!fe, "put_file must succeed");
    process.assert(st.has(fkey), "put_file object must be published");

    // The returned key must be the SHA-256 of the bytes ACTUALLY published —
    // re-hash the stored object and confirm it equals the key (and equals
    // std.hash + shasum -a 256 of the original fixture).
    fstored := fs.read_file(st.object_path(fkey));
    process.assert(fstored != null, "published object must be readable");
    process.assert(fstored! == file_bytes, "put_file stored bytes must equal the file");
    process.assert(stdhash_key(fstored!) == fkey, "key must equal SHA-256 of the published object");
    process.assert(fkey == stdhash_key(file_bytes), "put_file key must equal std.hash of the file bytes");
    process.assert(fkey == shasum_key(file_bytes), "put_file key must equal shasum -a 256");
    process.assert(incoming_count(st.staging_dir()) == "0", "put_file must clean up its staging temp");
    objs_after_file := entry_count(st.objects_dir());
    print("  put_file: key {} == digest(published object)\n", fkey);

    // Cross-path dedup: put_bytes of identical content yields the SAME key
    // and adds no second object; the stored bytes are not rewritten.
    bkey, be := st.put_bytes(file_bytes);
    process.assert(!be, "cross-path put_bytes must succeed");
    process.assert(bkey == fkey, "put_file and put_bytes of identical content share a key");
    process.assert(entry_count(st.objects_dir()) == objs_after_file, "cross-path dedup adds no object");
    afterb := fs.read_file(st.object_path(fkey));
    process.assert(afterb! == file_bytes, "cross-path dedup must not rewrite the object");

    // A repeat put_file hits dedup and also drops its staging temp.
    fkey2, fe2 := st.put_file(src);
    process.assert(!fe2, "repeat put_file must succeed");
    process.assert(fkey2 == fkey, "repeat put_file dedup yields the same key");
    process.assert(entry_count(st.objects_dir()) == objs_after_file, "repeat put_file adds no object");
    process.assert(incoming_count(st.staging_dir()) == "0", "dedup put_file must clean up its staging temp");
    print("  put_file: cross-path dedup, one object, staging cleaned\n");

    // ── cleanup ─────────────────────────────────────────────────────────
    process.run(concat("rm -rf ", root));
    fs.delete_file(src);

    print("store_content_addressed: ALL CASES PASS\n");
    return 0;
}