test: add unified DOSBox-X milestone runner

.gitignore

@@ -7,6 +7,9 @@
 !.qemu/share/*.c
 .qemu/share/fec/
 
+# Reproducible DOSBox-X/Open Watcom development cache and ephemeral runs
+.dosboxx/
+
 .qemu/*.qcow2
 .qemu/*.img
 .qemu/*.iso

AGENTS.md

@@ -18,11 +18,16 @@ VM 자동화 **명령 목록과 플래그는 문서가 아니라 CLI가 규범**
 ```powershell
 uv run ferro-vm --help
 uv run ferro-vm <command> --help
+uv run ferro-test --help
 ```
 
 ## 검증 규칙
 
-- **실행 검증은 QEMU FreeDOS 내부에서만 한다.** DOSBox는 쓰지 않는다.
+- 개발 중 빠른 회귀 검사는 `uv run ferro-test run --through <milestone>`로
+  DOSBox-X에서 수행한다.
+  이 경로도 컴파일러 A를 DOS 내부 Open Watcom으로 매번 새로 빌드한다.
+- **마일스톤의 최종 공식 검증은 QEMU FreeDOS 내부에서만 한다.** DOSBox-X 결과는
+  개발용 smoke test이며 완료 게이트를 대체하지 않는다.
 - 컴파일러 A와 생성 C 모두 VM 안의 Open Watcom으로 컴파일한다.
   컴파일러 A와 bits16은 `WCL`, bits32 생성 C는 `WCL386`.
 - **호스트에서 컴파일하지 않는다.** WSL이나 호스트 C 컴파일러 결과는 정식 검증으로

pyproject.toml

@@ -5,11 +5,13 @@ description = "Ferro language compiler and QEMU development automation"
 requires-python = ">=3.12"
 dependencies = [
     "onnxruntime>=1.28.0",
+    "pytest>=9.0.0",
     "rapidocr>=3.9.2",
 ]
 
 [project.scripts]
 ferro-vm = "ferrolang_vm.cli:main"
+ferro-test = "ferrolang_vm.test_cli:main"
 
 [build-system]
 requires = ["hatchling"]

src/ferrolang_vm/dosboxx.py

@@ -0,0 +1,236 @@
+"""Reproducible DOSBox-X/Open Watcom development test backend."""
+from __future__ import annotations
+
+import hashlib
+import json
+import os
+import shutil
+import subprocess
+import tempfile
+import urllib.request
+import zipfile
+from dataclasses import dataclass
+from pathlib import Path
+
+from .daemon import ROOT
+from .suite import Case
+
+
+CACHE = ROOT / ".dosboxx"
+LOCK_PATH = ROOT / "tools" / "toolchains" / "dosboxx.lock.json"
+RUNS = CACHE / "runs"
+
+
+class DosboxError(RuntimeError):
+    pass
+
+
+def _lock() -> dict[str, object]:
+    return json.loads(LOCK_PATH.read_text(encoding="utf-8"))
+
+
+def _sha256(path: Path) -> str:
+    digest = hashlib.sha256()
+    with path.open("rb") as source:
+        while chunk := source.read(1024 * 1024):
+            digest.update(chunk)
+    return digest.hexdigest()
+
+
+def _download(name: str, spec: dict[str, object]) -> Path:
+    downloads = CACHE / "downloads"
+    downloads.mkdir(parents=True, exist_ok=True)
+    target = downloads / Path(str(spec["url"])).name
+    expected = str(spec["sha256"]).lower()
+    if target.is_file() and _sha256(target) == expected:
+        return target
+    target.unlink(missing_ok=True)
+    partial = target.with_suffix(target.suffix + ".part")
+    partial.unlink(missing_ok=True)
+    print(f"ferro-test: downloading {name} {spec['version']}...")
+    try:
+        with urllib.request.urlopen(str(spec["url"])) as response, partial.open("wb") as output:
+            shutil.copyfileobj(response, output, length=1024 * 1024)
+    except Exception:
+        partial.unlink(missing_ok=True)
+        raise
+    actual = _sha256(partial)
+    if actual != expected:
+        partial.unlink(missing_ok=True)
+        raise DosboxError(f"{name} SHA-256 mismatch: expected {expected}, got {actual}")
+    partial.replace(target)
+    return target
+
+
+def _safe_extract(archive: Path, destination: Path) -> None:
+    destination.parent.mkdir(parents=True, exist_ok=True)
+    temporary = Path(tempfile.mkdtemp(prefix=f".{destination.name}-", dir=destination.parent))
+    try:
+        with zipfile.ZipFile(archive) as bundle:
+            root = temporary.resolve()
+            for member in bundle.infolist():
+                target = (temporary / member.filename).resolve()
+                if target != root and root not in target.parents:
+                    raise DosboxError(f"unsafe archive member: {member.filename}")
+            bundle.extractall(temporary)
+        if destination.exists():
+            shutil.rmtree(destination)
+        temporary.replace(destination)
+    except Exception:
+        shutil.rmtree(temporary, ignore_errors=True)
+        raise
+
+
+def _required_paths(lock: dict[str, object]) -> tuple[Path, Path, list[Path]]:
+    dosbox_spec = lock["dosboxx"]
+    watcom_spec = lock["open_watcom"]
+    assert isinstance(dosbox_spec, dict) and isinstance(watcom_spec, dict)
+    dosbox = CACHE / "dosbox-x" / str(dosbox_spec["executable"])
+    watcom = CACHE / "watcom"
+    required = [watcom / str(item) for item in watcom_spec["required"]]
+    return dosbox, watcom, required
+
+
+def setup(*, accept_watcom_license: bool = False) -> tuple[Path, Path]:
+    if os.name != "nt":
+        raise DosboxError("the DOSBox-X development backend currently supports Windows only")
+    lock = _lock()
+    dosbox, watcom, watcom_required = _required_paths(lock)
+    marker = CACHE / "SETUP.OK"
+    lock_hash = _sha256(LOCK_PATH)
+    if (marker.is_file() and marker.read_text(encoding="ascii").strip() == lock_hash
+            and dosbox.is_file() and all(path.is_file() for path in watcom_required)):
+        return dosbox, watcom
+    if not accept_watcom_license:
+        raise DosboxError(
+            "Open Watcom is distributed under the Sybase Open Watcom Public License. "
+            "Review tools/toolchains/dosboxx.lock.json and rerun "
+            "`uv run ferro-test setup --accept-watcom-license`."
+        )
+    dosbox_spec = lock["dosboxx"]
+    watcom_spec = lock["open_watcom"]
+    assert isinstance(dosbox_spec, dict) and isinstance(watcom_spec, dict)
+    _safe_extract(_download("DOSBox-X", dosbox_spec), CACHE / "dosbox-x")
+    _safe_extract(_download("Open Watcom", watcom_spec), watcom)
+    dosbox, watcom, watcom_required = _required_paths(lock)
+    missing = [str(path.relative_to(CACHE)) for path in [dosbox, *watcom_required]
+               if not path.is_file()]
+    if missing:
+        raise DosboxError("toolchain archive is missing: " + ", ".join(missing))
+    marker.write_text(lock_hash + "\n", encoding="ascii")
+    return dosbox, watcom
+
+
+def resolve_tools() -> tuple[Path, Path]:
+    lock = _lock()
+    dosbox, watcom, required = _required_paths(lock)
+    if not dosbox.is_file() or not all(path.is_file() for path in required):
+        raise DosboxError("toolchain is not installed; run `uv run ferro-test setup`")
+    return dosbox, watcom
+
+
+def _batch(cases: list[Case], *, show_dos: bool, trace_dos: bool) -> str:
+    lines = [
+        "@echo off", "if not exist RESULTS md RESULTS", "if not exist OUT md OUT",
+        "set WATCOM=W:", "set INCLUDE=W:\\H",
+        "set LIB=W:\\LIB286\\DOS;W:\\LIB286;W:\\LIB386\\DOS;W:\\LIB386",
+        "call BUILD.BAT", "if not exist BUILD.OK goto BUILDFAIL",
+        "echo PASS>RESULTS\\BUILD.RES",
+    ]
+    for index, case in enumerate(cases):
+        key = f"C{index:03d}"
+        command = case.command
+        if not trace_dos:
+            command += f" > RESULTS\\{key}.LOG"
+        lines.append(command)
+        if case.expect_success:
+            lines.extend([
+                f"if errorlevel 1 goto {key}F", f"echo PASS>RESULTS\\{key}.RES",
+                f"goto {key}D", f":{key}F", f"echo FAIL>RESULTS\\{key}.RES", f":{key}D",
+            ])
+        else:
+            lines.extend([
+                f"if errorlevel 1 goto {key}P", f"echo FAIL>RESULTS\\{key}.RES",
+                f"goto {key}D", f":{key}P", f"echo PASS>RESULTS\\{key}.RES", f":{key}D",
+            ])
+    lines.extend([
+        "goto FINISH", ":BUILDFAIL", "echo FAIL>RESULTS\\BUILD.RES", ":FINISH",
+        "echo DONE>RUN.OK", *(["pause"] if show_dos else []), "exit", "",
+    ])
+    return "\r\n".join(lines)
+
+
+@dataclass
+class SuiteRun:
+    root: Path
+    cases: list[Case]
+    keep: bool = False
+
+    @property
+    def fec(self) -> Path:
+        return self.root / "FEC"
+
+    def _key(self, case: Case) -> str:
+        return f"C{self.cases.index(case):03d}"
+
+    def result(self, case: Case | None = None) -> str:
+        name = "BUILD" if case is None else self._key(case)
+        path = self.fec / "RESULTS" / f"{name}.RES"
+        return path.read_text(encoding="ascii").strip() if path.is_file() else "MISSING"
+
+    def log(self, case: Case | None = None) -> str:
+        name = "BUILD" if case is None else self._key(case)
+        path = self.fec / "RESULTS" / f"{name}.LOG"
+        content = path.read_text(encoding="utf-8", errors="replace") if path.is_file() else ""
+        if content:
+            return content
+        errors = sorted(self.fec.glob("*.ERR"))
+        if errors:
+            return "\n".join(p.read_text(encoding="utf-8", errors="replace") for p in errors)
+        console = self.root / "CONSOLE.LOG"
+        return console.read_text(encoding="utf-8", errors="replace") if console.is_file() else ""
+
+    def cleanup(self) -> None:
+        if not self.keep:
+            shutil.rmtree(self.root, ignore_errors=True)
+
+
+def run_suite(cases: list[Case], *, keep: bool = False, show_dos: bool = False,
+              trace_dos: bool = False) -> SuiteRun:
+    dosbox, watcom = resolve_tools()
+    RUNS.mkdir(parents=True, exist_ok=True)
+    run_root = Path(tempfile.mkdtemp(prefix="suite-", dir=RUNS))
+    result = SuiteRun(run_root, cases, keep)
+    fec = result.fec
+    try:
+        shutil.copytree(ROOT / "fec" / "src", fec / "SRC")
+        shutil.copytree(ROOT / "fec" / "std", fec / "STD")
+        shutil.copytree(ROOT / "fec" / "tests", fec / "TESTS")
+        shutil.copy2(ROOT / "fec" / "build-dos.bat", fec / "BUILD.BAT")
+        console = run_root / "CONSOLE.LOG"
+        config = run_root / "DOSBOX.CON"
+        config.write_text(
+            f"[log]\nlogfile={console}\n[dosbox]\nlog console=quiet\n",
+            encoding="ascii",
+        )
+        (fec / "RUN.BAT").write_text(
+            _batch(cases, show_dos=show_dos, trace_dos=trace_dos),
+            encoding="ascii", newline="",
+        )
+        command = [str(dosbox)]
+        if not show_dos:
+            command.append("-silent")
+        command.extend([
+            "-fastlaunch", "-conf", str(config),
+            "-c", f'mount C "{run_root}"', "-c", f'mount W "{watcom}" -ro',
+            "-c", "C:", "-c", "cd \\FEC", "-c", "RUN.BAT",
+        ])
+        completed = subprocess.run(command, check=False, timeout=300)
+        if completed.returncode != 0:
+            raise DosboxError(f"DOSBox-X exited with status {completed.returncode}")
+        if not (fec / "RUN.OK").is_file():
+            raise DosboxError("DOSBox-X did not complete the test batch")
+        return result
+    except Exception:
+        result.keep = True
+        raise

src/ferrolang_vm/registry_m1_m3.py

@@ -0,0 +1,69 @@
+"""Explicit M1--M3 commands and expectations from TEST-DOS.BAT."""
+from __future__ import annotations
+
+from .suite import Case
+
+
+def _c(milestone: int, name: str, command: str, ok: bool = True) -> Case:
+    return Case(f"m{milestone}-{name}", milestone, command, ok)
+
+
+M1_M3_CASES: list[Case] = []
+for _name in ("basic", "literals", "keybuilt", "v012form"):
+    M1_M3_CASES.append(_c(1, f"{_name}-parse", f"FEC.EXE --dump-ast TESTS\\PASS\\{_name.upper()}.FE"))
+for _name in ("core", "fmt", "io", "list", "map", "mem", "str", "sys"):
+    M1_M3_CASES.append(_c(1, f"std-{_name}-parse", f"FEC.EXE --dump-ast STD\\{_name.upper()}.FE"))
+for _name in ("misssemi", "unclcomm", "logical"):
+    M1_M3_CASES.append(_c(1, f"{_name}-reject", f"FEC.EXE --dump-ast TESTS\\FAIL\\{_name.upper()}.FE", False))
+
+for _name in ("hello", "scopes"):
+    _upper = _name.upper()
+    M1_M3_CASES.extend([
+        _c(2, f"{_name}-emit", f"FEC.EXE --target=bits32 --emit-c TESTS\\M2\\{_upper}.FE -o TESTS\\M2\\{_upper}.C"),
+        _c(2, f"{_name}-build", f"WCL386 -q -za -bt=dos -fe=TESTS\\M2\\{_upper}.EXE TESTS\\M2\\{_upper}.C"),
+        _c(2, f"{_name}-run", f"TESTS\\M2\\{_upper}.EXE"),
+    ])
+M1_M3_CASES.extend([
+    _c(2, "castwhil-emit", "FEC.EXE --target=bits16 --emit-c TESTS\\M2\\CASTWHIL.FE -o TESTS\\M2\\CAST16.C"),
+    _c(2, "castwhil-build", "WCL -q -za -bt=dos -fe=TESTS\\M2\\CAST16.EXE TESTS\\M2\\CAST16.C"),
+    _c(2, "castwhil-run", "TESTS\\M2\\CAST16.EXE"),
+])
+_m2_outputs = {
+    "bad-cond": "BAD-CO", "bad-cast": "BAD-CA", "bad-asgn": "BAD-AS",
+    "bad-unk": "BAD-UN", "bad-ari": "BAD-AR", "bad-type": "BAD-TY",
+    "bad-ret": "BAD-RE", "bad-unit": "BAD-UI", "bad-void": "BAD-VO",
+}
+for _name, _output in _m2_outputs.items():
+    M1_M3_CASES.append(_c(2, f"{_name}-reject", "FEC.EXE --target=bits32 --emit-c "
+                            f"TESTS\\M2\\{_name.upper()}.FE -o TESTS\\M2\\{_output}.C", False))
+
+def _m3_runtime(name: str) -> list[Case]:
+    upper = name.upper()
+    return [
+        _c(3, f"{name}-emit", f"FEC.EXE --target=bits32 --emit-c TESTS\\M3\\{upper}.FE -o TESTS\\M3\\{upper}.C"),
+        _c(3, f"{name}-build", f"WCL386 -q -za -bt=dos -fe=TESTS\\M3\\{upper}.EXE TESTS\\M3\\{upper}.C"),
+        _c(3, f"{name}-run", f"TESTS\\M3\\{upper}.EXE"),
+    ]
+
+
+for _name in ("struct", "enum", "array", "mutable"):
+    M1_M3_CASES.extend(_m3_runtime(_name))
+for _name in ("bad-mlet", "bad-shwr"):
+    M1_M3_CASES.append(_c(3, f"{_name}-reject", "FEC.EXE --target=bits32 --emit-c "
+                            f"TESTS\\M3\\{_name.upper()}.FE -o TESTS\\M3\\{_name.upper()}.C", False))
+for _name in ("str", "for", "nested", "char", "arrayctx"):
+    M1_M3_CASES.extend(_m3_runtime(_name))
+for _name in ("bounds", "slcbound"):
+    _upper = _name.upper()
+    M1_M3_CASES.extend([
+        _c(3, f"{_name}-emit", f"FEC.EXE --target=bits32 --emit-c TESTS\\M3\\{_upper}.FE -o TESTS\\M3\\{_upper}.C"),
+        _c(3, f"{_name}-build", f"WCL386 -q -za -bt=dos -fe=TESTS\\M3\\{_upper}.EXE TESTS\\M3\\{_upper}.C"),
+        _c(3, f"{_name}-trap", f"TESTS\\M3\\{_upper}.EXE", False),
+    ])
+M1_M3_CASES.extend([
+    _c(3, "bounds-no-checks-emit", "FEC.EXE --target=bits32 --no-checks --emit-c TESTS\\M3\\BOUNDS.FE -o TESTS\\M3\\BOUNDS-N.C"),
+    _c(3, "bounds-no-checks-build", "WCL386 -q -za -bt=dos -fe=TESTS\\M3\\BOUNDS-N.EXE TESTS\\M3\\BOUNDS-N.C"),
+])
+for _name in ("badfld", "badmat", "badarr", "badcycle", "badstr", "badchar", "badfield", "badindex"):
+    M1_M3_CASES.append(_c(3, f"{_name}-reject", "FEC.EXE --target=bits32 --emit-c "
+                            f"TESTS\\M3\\{_name.upper()}.FE -o TESTS\\M3\\{_name.upper()}.C", False))

src/ferrolang_vm/registry_m4_m6.py

@@ -0,0 +1,49 @@
+"""Explicit M4--M5 cases from TEST-DOS.BAT and M6 fixture expectations.
+
+Only commands and their expected status live here; the ``.fe`` fixtures stay
+under ``fec/tests`` and are copied by the runner.
+"""
+from __future__ import annotations
+
+from .suite import Case
+
+
+def _c(milestone: int, name: str, command: str, ok: bool) -> Case:
+    return Case(f"m{milestone}-{name}", milestone, command, ok)
+
+
+M4_M6_CASES: list[Case] = [
+    _c(4, "format", "FEC.EXE --target=bits32 --emit-c TESTS\\M4\\FORMAT.FE -o TESTS\\M4\\FORMAT.C", True),
+    _c(4, "format-build", "WCL386 -q -za -wx -wcd=202 -bt=dos -fe=TESTS\\M4\\FORMAT.EXE TESTS\\M4\\FORMAT.C", True),
+    _c(4, "format-run", "TESTS\\M4\\FORMAT.EXE", True),
+    _c(4, "try-fpr", "FEC.EXE --target=bits32 --emit-c TESTS\\M4\\TRY-FPR.FE -o TESTS\\M4\\TRY-FPR.C", True),
+    _c(4, "try-fpr-build", "WCL386 -q -za -wx -wcd=202 -bt=dos -fe=TESTS\\M4\\TRY-FPR.EXE TESTS\\M4\\TRY-FPR.C", True),
+    _c(4, "try-fpr-run", "TESTS\\M4\\TRY-FPR.EXE", True),
+    _c(4, "prop", "FEC.EXE --target=bits32 --emit-c TESTS\\M4\\PROP.FE -o TESTS\\M4\\PROP.C", True),
+    _c(4, "prop-build", "WCL386 -q -za -wx -wcd=202 -bt=dos -fe=TESTS\\M4\\PROP.EXE TESTS\\M4\\PROPTEST.C", True),
+    _c(4, "prop-run", "TESTS\\M4\\PROP.EXE", True),
+]
+
+_m4_outputs = {"bad-type": "BAD-TYP", "bad-writ": "BAD-WRI"}
+for _name in ("bad-ari", "bad-verb", "bad-run", "bad-type", "bad-try", "bad-writ", "bad-bufw", "bad-many", "bad-open", "bad-cls"):
+    _output = _m4_outputs.get(_name, _name.upper())
+    M4_M6_CASES.append(_c(4, _name, "FEC.EXE --target=bits32 --emit-c "
+                           f"TESTS\\M4\\{_name.upper()}.FE -o TESTS\\M4\\{_output}.C", False))
+
+for _name in ("defer", "owned"):
+    M4_M6_CASES.append(_c(5, _name, "FEC.EXE --target=bits32 --emit-c "
+                           f"TESTS\\M5\\{_name.upper()}.FE -o TESTS\\M5\\{_name.upper()}.C", True))
+for _name in ("bad-move", "bad-dest", "bad-drop", "bad-dbl", "bad-cond", "bad-proj", "bad-clos", "bad-loop"):
+    _output = "BAD-DES" if _name == "bad-dest" else _name.upper()
+    M4_M6_CASES.append(_c(5, _name, "FEC.EXE --target=bits32 --emit-c "
+                           f"TESTS\\M5\\{_name.upper()}.FE -o TESTS\\M5\\{_output}.C", False))
+M4_M6_CASES += [
+    _c(5, "runtime", "FEC.EXE --target=bits32 --emit-c TESTS\\M5\\RUNTIME.FE -o TESTS\\M5\\RUNT-G.C", True),
+    _c(5, "runtime-build", "WCL386 -q -za -bt=dos -dmalloc=m5_malloc -dfree=m5_free -fe=TESTS\\M5\\RUNTIME.EXE TESTS\\M5\\RUNT-G.C TESTS\\M5\\RUNTIME.C", True),
+    _c(5, "runtime-run", "TESTS\\M5\\RUNTIME.EXE", True),
+]
+
+for _name in ("badarg", "badbinit", "badbrmov", "baddefer", "badfld", "badglob", "badgmut", "badinv", "badlocsl", "badloop", "badmove", "badmut", "badmut2", "badptr", "badret", "badrfld", "badridx", "badscop", "badself", "badshwr", "badslfld", "badtwo", "badup", "badweak"):
+    M4_M6_CASES.append(_c(6, _name, f"FEC.EXE --check TESTS\\M6\\{_name.upper()}.FE", False))
+for _name in ("okbranch", "okdefer", "okglobcp", "oklast", "okr8free", "okr8join", "okr8meth", "okr8stat", "okrebor", "okrtlast", "okshare", "okslreb", "okstatic", "oktemp", "oktrim", "okwcall"):
+    M4_M6_CASES.append(_c(6, _name, f"FEC.EXE --target=bits32 --emit-c -o OUT\\{_name.upper()}.C TESTS\\M6\\{_name.upper()}.FE", True))

src/ferrolang_vm/suite.py

@@ -0,0 +1,22 @@
+"""Shared types and selection for the explicit milestone registries."""
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True)
+class Case:
+    id: str
+    milestone: int
+    command: str
+    expect_success: bool
+
+
+def all_cases(*, through: int = 6, only: int | None = None) -> list[Case]:
+    from .registry_m1_m3 import M1_M3_CASES
+    from .registry_m4_m6 import M4_M6_CASES
+
+    cases = [*M1_M3_CASES, *M4_M6_CASES]
+    if only is not None:
+        return [case for case in cases if case.milestone == only]
+    return [case for case in cases if case.milestone <= through]

src/ferrolang_vm/test_cli.py

@@ -0,0 +1,72 @@
+"""Developer test entry point backed by a disposable DOSBox-X run."""
+from __future__ import annotations
+
+import argparse
+import os
+import sys
+from pathlib import Path
+
+from .dosboxx import DosboxError, setup
+
+
+MILESTONES = tuple(f"m{number}" for number in range(1, 7))
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(
+        prog="ferro-test",
+        description="Fast Ferro development tests in DOSBox-X/Open Watcom.",
+    )
+    commands = parser.add_subparsers(dest="command", required=True)
+    prepare = commands.add_parser("setup", help="download and verify pinned development tools")
+    prepare.add_argument("--accept-watcom-license", action="store_true",
+                         help="confirm acceptance of the Sybase Open Watcom Public License")
+    run = commands.add_parser("run", help="build once and run milestone pytest cases")
+    selection = run.add_mutually_exclusive_group()
+    selection.add_argument("--through", choices=MILESTONES, default="m6",
+                           help="run cumulatively through this milestone (default: m6)")
+    selection.add_argument("--only", choices=MILESTONES,
+                           help="run only this milestone's cases")
+    run.add_argument("-v", "--verbose", action="store_true", help="show every pytest case")
+    run.add_argument("--keep-failed", action="store_true",
+                     help="keep the disposable DOS filesystem after failures")
+    run.add_argument("--show-dos", action="store_true",
+                     help="show DOSBox-X and wait for a key before closing")
+    run.add_argument("--dos-log", action="store_true",
+                     help="print the captured DOS console after the run")
+    run.add_argument("--trace-dos", action="store_true",
+                     help="do not redirect case command output")
+    args = parser.parse_args()
+    try:
+        if args.command == "setup":
+            dosbox, watcom = setup(accept_watcom_license=args.accept_watcom_license)
+            print(f"DOSBox-X: {dosbox}")
+            print(f"Open Watcom: {watcom}")
+            return 0
+        if args.only:
+            os.environ["FERRO_TEST_ONLY"] = args.only
+        else:
+            os.environ["FERRO_TEST_THROUGH"] = args.through
+        for enabled, name in (
+            (args.keep_failed, "FERRO_TEST_KEEP_FAILED"),
+            (args.show_dos, "FERRO_TEST_SHOW_DOS"),
+            (args.trace_dos, "FERRO_TEST_TRACE_DOS"),
+            (args.dos_log, "FERRO_TEST_DOS_LOG"),
+        ):
+            if enabled:
+                os.environ[name] = "1"
+        import pytest
+        test_file = os.fspath(
+            Path(__file__).resolve().parents[2] / "tools" / "tests" / "test_milestones_dosboxx.py"
+        )
+        pytest_args = [test_file, "--tb=short", "-v" if args.verbose else "-q"]
+        if args.dos_log:
+            pytest_args.append("-s")
+        return int(pytest.main(pytest_args))
+    except (DosboxError, ValueError) as exc:
+        print(f"ferro-test: {exc}", file=sys.stderr)
+        return 2
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

tools/README.md

@@ -2,12 +2,37 @@
 
 ## Host support
 
-Automation currently supports **Windows 10/11 only**. It requires `uv`, QEMU
-with WHPX support, and `ffmpeg.exe` on `PATH`. The Python implementation uses
-portable APIs where possible, but other hosts are not supported yet.
+Automation currently supports **Windows 10/11 only**. The fast development loop
+requires only `uv`; it downloads pinned DOSBox-X and Open Watcom DOS releases.
+The final milestone gate additionally requires QEMU with WHPX support and
+`ffmpeg.exe` on `PATH`. Other hosts are not supported yet.
 
 ## Getting started
 
+```powershell
+uv run ferro-test setup --accept-watcom-license
+uv run ferro-test run --through m6 -v
+```
+
+`setup` reads `tools/toolchains/dosboxx.lock.json`, downloads the exact official
+archives, verifies their SHA-256 hashes, and extracts them under ignored
+`.dosboxx/`. Review the Open Watcom license referenced by the lock file before
+accepting it. Archives and installed tools are deliberately not committed.
+
+Each `run` creates a disposable DOS drive, copies the current compiler, standard
+library, and fixtures, then builds `FEC.EXE` once inside DOS with Open Watcom.
+All selected milestone commands execute sequentially in that same DOSBox-X
+instance, while pytest reports every emit, Watcom build, runtime, and rejection
+check separately. Thus stale QEMU binaries cannot make the test pass.
+
+`--through m6` runs cumulatively from M1; `--only m6` selects one milestone.
+Use `--keep-failed` to preserve a failed drive under `.dosboxx/runs/`,
+`--dos-log` to print the captured DOS console, `--trace-dos` to disable command
+output redirection, and `--show-dos` to keep the GUI open until a key is pressed.
+
+This is the quick development smoke test. Run the QEMU/FreeDOS workflow below
+for the authoritative milestone completion gate.
+
 ```powershell
 uv run ferro-vm start
 uv run ferro-vm status
@@ -18,6 +43,7 @@ The command list lives in the CLI itself, not in this file:
 ```powershell
 uv run ferro-vm --help
 uv run ferro-vm <command> --help
+uv run ferro-test --help
 ```
 
 Working rules, verification gates, and DOS build traps are in `AGENTS.md`.

tools/tests/test_milestones_dosboxx.py

@@ -0,0 +1,54 @@
+from __future__ import annotations
+
+import os
+
+import pytest
+
+from ferrolang_vm.dosboxx import SuiteRun, run_suite
+from ferrolang_vm.suite import Case, all_cases
+
+
+def _number(name: str) -> int:
+    if not name.startswith("m") or not name[1:].isdigit():
+        raise ValueError(f"invalid milestone: {name}")
+    value = int(name[1:])
+    if value not in range(1, 7):
+        raise ValueError(f"unsupported milestone: {name}")
+    return value
+
+
+ONLY = os.environ.get("FERRO_TEST_ONLY")
+CASES = all_cases(
+    through=_number(os.environ.get("FERRO_TEST_THROUGH", "m6")),
+    only=_number(ONLY) if ONLY else None,
+)
+
+
+@pytest.fixture(scope="session")
+def suite_run() -> SuiteRun:
+    run = run_suite(
+        CASES,
+        keep=os.environ.get("FERRO_TEST_KEEP_FAILED") == "1",
+        show_dos=os.environ.get("FERRO_TEST_SHOW_DOS") == "1",
+        trace_dos=os.environ.get("FERRO_TEST_TRACE_DOS") == "1",
+    )
+    yield run
+    if os.environ.get("FERRO_TEST_DOS_LOG") == "1":
+        console = run.root / "CONSOLE.LOG"
+        if console.is_file():
+            print(console.read_text(encoding="utf-8", errors="replace"))
+    run.cleanup()
+
+
+def test_compiler_build(suite_run: SuiteRun) -> None:
+    assert suite_run.result() == "PASS", suite_run.log()
+
+
+@pytest.mark.parametrize("case", CASES, ids=lambda case: case.id)
+def test_milestone_case(case: Case, suite_run: SuiteRun) -> None:
+    if suite_run.result() != "PASS":
+        pytest.skip("compiler build failed")
+    assert suite_run.result(case) == "PASS", (
+        f"DOS command: {case.command}\nExpected success: {case.expect_success}\n"
+        f"{suite_run.log(case)}"
+    )

tools/toolchains/dosboxx.lock.json

@@ -0,0 +1,23 @@
+{
+  "schema": 1,
+  "dosboxx": {
+    "version": "2026.08.02",
+    "url": "https://github.com/joncampbell123/dosbox-x/releases/download/dosbox-x-v2026.08.02/dosbox-x-vsbuild-win64-2026.08.02-portable.zip",
+    "sha256": "ca28208f5fee25a74caf3a02cc0189c7f7943a42ce37c02848f5fc03450a96cf",
+    "executable": "bin/x64/Release/dosbox-x.exe"
+  },
+  "open_watcom": {
+    "version": "2026-08-01-Build",
+    "url": "https://github.com/open-watcom/open-watcom-v2/releases/download/2026-08-01-Build/open-watcom-2_0-c-dos.exe",
+    "sha256": "80db4ab340f382e59bf3d396280576ec837964c2ef00e8ddd3b2b3724ab63edf",
+    "required": [
+      "binw/wcl.exe",
+      "binw/wcl386.exe",
+      "binp/wlink.exe",
+      "h/stdio.h",
+      "lib286/dos/clibl.lib",
+      "lib386/dos/clib3r.lib",
+      "license.txt"
+    ]
+  }
+}

uv.lock

@@ -175,12 +175,14 @@ version = "0.1.0"
 source = { editable = "." }
 dependencies = [
     { name = "onnxruntime" },
+    { name = "pytest" },
     { name = "rapidocr" },
 ]
 
 [package.metadata]
 requires-dist = [
     { name = "onnxruntime", specifier = ">=1.28.0" },
+    { name = "pytest", specifier = ">=9.0.0" },
     { name = "rapidocr", specifier = ">=3.9.2" },
 ]
 
@@ -201,6 +203,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/1e/5e/d4e9f1a599fb8e573b7b87160658329fbf28d19eac2718f51fc3def3aa5a/idna-3.18-py3-none-any.whl", hash = "sha256:7f952cbe720b688055e3f87de14f5c3e5fdaa8bc3928985c4077ca689de849a2", size = 65455 },
 ]
 
+[[package]]
+name = "iniconfig"
+version = "2.3.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/72/34/14ca021ce8e5dfedc35312d08ba8bf51fdd999c576889fc2c24cb97f4f10/iniconfig-2.3.0.tar.gz", hash = "sha256:c76315c77db068650d49c5b56314774a7804df16fee4402c1f19d6d15d8c4730", size = 20503 }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/cb/b1/3846dd7f199d53cb17f49cba7e651e9ce294d8497c8c150530ed11865bb8/iniconfig-2.3.0-py3-none-any.whl", hash = "sha256:f631c04d2c48c52b84d0d0549c99ff3859c98df65b3101406327ecc7d53fbf12", size = 7484 },
+]
+
 [[package]]
 name = "numpy"
 version = "2.5.2"
@@ -418,6 +429,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/3d/68/1f3066acedf37673694a7141381d8f811ae97f30d34413d236abe7d489f1/pillow-12.3.0-cp315-cp315t-win_arm64.whl", hash = "sha256:06ff022112bc9cbf83b60f8e028d94ad87b60621706487e65f673de61610ab59", size = 2567491 },
 ]
 
+[[package]]
+name = "pluggy"
+version = "1.6.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/f9/e2/3e91f31a7d2b083fe6ef3fa267035b518369d9511ffab804f839851d2779/pluggy-1.6.0.tar.gz", hash = "sha256:7dcc130b76258d33b90f61b658791dede3486c3e6bfb003ee5c9bfb396dd22f3", size = 69412 }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/54/20/4d324d65cc6d9205fabedc306948156824eb9f0ee1633355a8f7ec5c66bf/pluggy-1.6.0-py3-none-any.whl", hash = "sha256:e920276dd6813095e9377c0bc5566d94c932c33b27a3e3945d8389c374dd4746", size = 20538 },
+]
+
 [[package]]
 name = "protobuf"
 version = "7.35.1"
@@ -463,6 +483,31 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/8f/dc/53df8b6931d47080b4fe4ee8450d42e660ee1c5c1556c7ab73359182b769/pyclipper-1.4.0-cp314-cp314t-win_amd64.whl", hash = "sha256:29dae3e0296dff8502eeb7639fcfee794b0eec8590ba3563aee28db269da6b04", size = 117608 },
 ]
 
+[[package]]
+name = "pygments"
+version = "2.20.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/c3/b2/bc9c9196916376152d655522fdcebac55e66de6603a76a02bca1b6414f6c/pygments-2.20.0.tar.gz", hash = "sha256:6757cd03768053ff99f3039c1a36d6c0aa0b263438fcab17520b30a303a82b5f", size = 4955991 }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/f4/7e/a72dd26f3b0f4f2bf1dd8923c85f7ceb43172af56d63c7383eb62b332364/pygments-2.20.0-py3-none-any.whl", hash = "sha256:81a9e26dd42fd28a23a2d169d86d7ac03b46e2f8b59ed4698fb4785f946d0176", size = 1231151 },
+]
+
+[[package]]
+name = "pytest"
+version = "9.1.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "colorama", marker = "sys_platform == 'win32'" },
+    { name = "iniconfig" },
+    { name = "packaging" },
+    { name = "pluggy" },
+    { name = "pygments" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/e4/47/b9efed96c114afcfa3c9d3fe98a76a1d14c74a9e266d397cf6eb64be5e01/pytest-9.1.1.tar.gz", hash = "sha256:1088fbde8f2b49d95a549a195707afa7a76a3ce9bcadc26b6d71f0ffda5fe313", size = 1636369 }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/24/25/1de2678b631f5a49215c6c96fff41ba892b0a34df68d6d80292b1b48aa7f/pytest-9.1.1-py3-none-any.whl", hash = "sha256:37a86b45efb9a47a61a36449063e8e18d0cab3161329fc099eb21783169c4f0c", size = 386536 },
+]
+
 [[package]]
 name = "pyyaml"
 version = "6.0.3"