local-ci-cd-system/tests/python/test_commands_job.py

"""Tests for ``ci_orchestrator job`` (Phase A4)."""

from __future__ import annotations

from pathlib import Path
from typing import Any

import click
import pytest
from click.testing import CliRunner

import ci_orchestrator.commands.job as job_module
from ci_orchestrator.__main__ import cli
from ci_orchestrator.backends.errors import BackendOperationFailed
from ci_orchestrator.backends.protocol import VmHandle

# ──────────────────────────────────────────────────────────── fakes


class _FakeBackend:
    """Records every backend call and creates the VMX file on clone."""

    def __init__(
        self,
        *,
        clone_fail: bool = False,
        start_fail: bool = False,
        delete_fail: bool = False,
        ip: str | None = "10.99.0.42",
        running_after: int = 0,
        clone_guest_os: str = "ubuntu-64",
    ) -> None:
        self.clone_fail = clone_fail
        self.start_fail = start_fail
        self.delete_fail = delete_fail
        self.ip = ip
        self.running_after = running_after
        self.clone_guest_os = clone_guest_os
        self.calls: list[str] = []
        self._is_running_count = 0

    def clone_linked(
        self,
        template: str,
        snapshot: str,
        name: str,
        destination: str | None = None,
    ) -> VmHandle:
        self.calls.append("clone")
        if self.clone_fail:
            raise BackendOperationFailed("clone", 1, "boom")
        assert destination is not None
        dst = Path(destination)
        dst.parent.mkdir(parents=True, exist_ok=True)
        dst.write_text(
            f'config.version = "8"\nguestOS = "{self.clone_guest_os}"\n',
            encoding="utf-8",
        )
        return VmHandle(identifier=str(dst), name=name)

    def start(self, handle: VmHandle, headless: bool = True) -> None:
        self.calls.append("start")
        if self.start_fail:
            raise BackendOperationFailed("start", 1, "no")

    def stop(self, handle: VmHandle, hard: bool = False) -> None:
        self.calls.append(f"stop:{'hard' if hard else 'soft'}")

    def delete(self, handle: VmHandle) -> None:
        self.calls.append("delete")
        if self.delete_fail:
            raise BackendOperationFailed("deleteVM", 1, "locked")

    def is_running(self, handle: VmHandle) -> bool:
        self._is_running_count += 1
        return self._is_running_count > self.running_after

    def get_ip(self, handle: VmHandle, timeout: float = 0.0) -> str | None:
        return self.ip

    def list_snapshots(self, handle: VmHandle) -> list[str]:
        return []


def _patch_common(
    monkeypatch: pytest.MonkeyPatch,
    backend: _FakeBackend,
    *,
    transport_ready: bool = True,
    linux_build_raises: bool = False,
) -> dict[str, list[Any]]:
    """Patch all in-process collaborators of the ``job`` command."""
    calls: dict[str, list[Any]] = {
        "linux_build": [],
        "windows_build": [],
        "linux_collect": [],
        "windows_collect": [],
        "manifest": [],
    }

    monkeypatch.setattr(job_module, "load_backend", lambda _cfg: backend)

    class _ProbeOk:
        def __enter__(self) -> _ProbeOk:
            return self

        def __exit__(self, *_e: object) -> None:
            return None

        def is_ready(self) -> bool:
            return transport_ready

    class _Store:
        def get(self, _t: str) -> Any:
            from ci_orchestrator.credentials import Credential

            return Credential("ci", "pwd")

    monkeypatch.setattr(job_module, "WinRmTransport", lambda *a, **k: _ProbeOk())
    monkeypatch.setattr(job_module, "SshTransport", lambda *a, **k: _ProbeOk())
    monkeypatch.setattr(job_module, "KeyringCredentialStore", lambda: _Store())

    def _linux_build(**kwargs: Any) -> None:
        calls["linux_build"].append(kwargs)
        if linux_build_raises:
            raise RuntimeError("build exploded")

    def _windows_build(**kwargs: Any) -> None:
        calls["windows_build"].append(kwargs)

    def _linux_collect(**kwargs: Any) -> None:
        calls["linux_collect"].append(kwargs)
        # Simulate a transferred file so manifest writing has something to do.
        Path(kwargs["host_dir"], "out.txt").write_text("ok", encoding="utf-8")

    def _windows_collect(**kwargs: Any) -> None:
        calls["windows_collect"].append(kwargs)

    def _manifest(host_dir: Path, job_id: str, commit: str) -> int:
        calls["manifest"].append((str(host_dir), job_id, commit))
        return 1

    monkeypatch.setattr(job_module, "_linux_build", _linux_build)
    monkeypatch.setattr(job_module, "_windows_build", _windows_build)
    monkeypatch.setattr(job_module, "_linux_collect", _linux_collect)
    monkeypatch.setattr(job_module, "_windows_collect", _windows_collect)
    monkeypatch.setattr(job_module, "_write_manifest", _manifest)
    # Make poll loops instant.
    monkeypatch.setattr(job_module.time, "sleep", lambda _s: None)
    return calls


@pytest.fixture
def template_vmx(tmp_path: Path) -> Path:
    """Linux-flavoured VMX so auto-detect picks the SSH transport."""
    vmx = tmp_path / "templates" / "LinuxBuild2404.vmx"
    vmx.parent.mkdir(parents=True)
    vmx.write_text('guestOS = "ubuntu-64"\n', encoding="utf-8")
    return vmx


def _common_args(
    template: Path,
    *,
    clone_base: Path,
    artifact_base: Path,
    extra: list[str] | None = None,
) -> list[str]:
    return [
        "job",
        "--job-id",
        "ci-test-1",
        "--repo-url",
        "http://gitea.local/org/repo.git",
        "--branch",
        "main",
        "--template-path",
        str(template),
        "--clone-base-dir",
        str(clone_base),
        "--artifact-base-dir",
        str(artifact_base),
        "--ready-timeout",
        "30",
        "--poll-interval",
        "1",
        *(extra or []),
    ]


# ──────────────────────────────────────────────────────────── tests


def test_job_happy_path_linux(
    monkeypatch: pytest.MonkeyPatch, tmp_path: Path, template_vmx: Path
) -> None:
    backend = _FakeBackend(running_after=1)
    calls = _patch_common(monkeypatch, backend)

    base_clone = tmp_path / "build-vms"
    base_artifact = tmp_path / "artifacts"

    result = CliRunner().invoke(
        cli, _common_args(template_vmx, clone_base=base_clone, artifact_base=base_artifact)
    )
    assert result.exit_code == 0, result.output
    # Backend received the full lifecycle including delete in finally.
    assert backend.calls[0] == "clone"
    assert "start" in backend.calls
    assert backend.calls[-1] == "delete"
    # Linux build branch was taken (auto-detected from VMX guestOS).
    assert len(calls["linux_build"]) == 1
    assert calls["linux_build"][0]["build_command"] == ""
    # Artifacts were collected into the per-job directory and manifest written.
    assert (base_artifact / "ci-test-1" / "out.txt").is_file()
    assert calls["manifest"] and calls["manifest"][0][1] == "ci-test-1"
    # Clone directory removed.
    assert not any((base_clone).iterdir()) or all(not p.exists() for p in base_clone.iterdir())


def test_job_default_transport_in_guest_with_submodules(
    monkeypatch: pytest.MonkeyPatch, tmp_path: Path, template_vmx: Path
) -> None:
    backend = _FakeBackend(running_after=1)
    calls = _patch_common(monkeypatch, backend)
    result = CliRunner().invoke(
        cli,
        _common_args(
            template_vmx, clone_base=tmp_path / "v", artifact_base=tmp_path / "a"
        ),
    )
    assert result.exit_code == 0, result.output
    kw = calls["linux_build"][0]
    # Default: in-guest clone (clone_url set, no host source) + submodules.
    assert kw["host_source_dir"] is None
    assert kw["clone_url"] and "repo.git" in kw["clone_url"]
    assert kw["clone_submodules"] is True


def test_job_host_clone_transport(
    monkeypatch: pytest.MonkeyPatch, tmp_path: Path, template_vmx: Path
) -> None:
    backend = _FakeBackend(running_after=1)
    calls = _patch_common(monkeypatch, backend)
    root = tmp_path / "hostclone"
    src = root / "src"
    src.mkdir(parents=True)

    def _fake_host_clone(
        repo_url: str, branch: str, commit: str, submodules: bool
    ) -> tuple[Path, Path]:
        assert submodules is True
        return root, src

    monkeypatch.setattr(job_module, "_host_clone", _fake_host_clone)
    result = CliRunner().invoke(
        cli,
        _common_args(
            template_vmx,
            clone_base=tmp_path / "v",
            artifact_base=tmp_path / "a",
            extra=["--host-clone"],
        ),
    )
    assert result.exit_code == 0, result.output
    kw = calls["linux_build"][0]
    assert kw["host_source_dir"] == str(src)
    assert kw["clone_url"] is None
    assert kw["clone_submodules"] is True
    # Host-side temp clone removed after the build.
    assert not root.exists()


def test_job_no_submodules_flag(
    monkeypatch: pytest.MonkeyPatch, tmp_path: Path, template_vmx: Path
) -> None:
    backend = _FakeBackend(running_after=1)
    calls = _patch_common(monkeypatch, backend)
    result = CliRunner().invoke(
        cli,
        _common_args(
            template_vmx,
            clone_base=tmp_path / "v",
            artifact_base=tmp_path / "a",
            extra=["--no-submodules"],
        ),
    )
    assert result.exit_code == 0, result.output
    assert calls["linux_build"][0]["clone_submodules"] is False


def test_job_skip_artifact_branch(
    monkeypatch: pytest.MonkeyPatch, tmp_path: Path, template_vmx: Path
) -> None:
    backend = _FakeBackend(running_after=0)
    calls = _patch_common(monkeypatch, backend)
    result = CliRunner().invoke(
        cli,
        _common_args(
            template_vmx,
            clone_base=tmp_path / "vms",
            artifact_base=tmp_path / "art",
            extra=["--skip-artifact"],
        ),
    )
    assert result.exit_code == 0, result.output
    assert calls["linux_collect"] == []
    assert calls["windows_collect"] == []
    assert calls["manifest"] == []
    assert backend.calls[-1] == "delete"


def test_job_cleanup_on_build_failure(
    monkeypatch: pytest.MonkeyPatch, tmp_path: Path, template_vmx: Path
) -> None:
    """If the build raises, the VM must still be destroyed (try/finally)."""
    backend = _FakeBackend(running_after=0)
    _patch_common(monkeypatch, backend, linux_build_raises=True)
    result = CliRunner().invoke(
        cli,
        _common_args(template_vmx, clone_base=tmp_path / "vms", artifact_base=tmp_path / "art"),
    )
    assert result.exit_code != 0
    assert "build exploded" in result.output
    assert "delete" in backend.calls
    # Stop attempt happened before delete.
    assert any(c.startswith("stop:") for c in backend.calls)


def test_job_cleanup_on_clone_failure(
    monkeypatch: pytest.MonkeyPatch, tmp_path: Path, template_vmx: Path
) -> None:
    backend = _FakeBackend(clone_fail=True)
    _patch_common(monkeypatch, backend)
    result = CliRunner().invoke(
        cli,
        _common_args(template_vmx, clone_base=tmp_path / "vms", artifact_base=tmp_path / "art"),
    )
    assert result.exit_code != 0
    assert "clone failed" in result.output
    # Backend.start/delete must not have been called: handle is None.
    assert backend.calls == ["clone"]


def test_job_rejects_missing_template(
    monkeypatch: pytest.MonkeyPatch, tmp_path: Path
) -> None:
    backend = _FakeBackend()
    _patch_common(monkeypatch, backend)
    result = CliRunner().invoke(
        cli,
        _common_args(
            tmp_path / "ghost.vmx",
            clone_base=tmp_path / "vms",
            artifact_base=tmp_path / "art",
        ),
    )
    assert result.exit_code != 0
    assert "Template VMX not found" in result.output
    assert backend.calls == []


def test_job_windows_branch_with_overrides(
    monkeypatch: pytest.MonkeyPatch, tmp_path: Path
) -> None:
    """Windows-flavoured VMX selects the WinRM build path; CPU/RAM override applied."""
    vmx = tmp_path / "templates" / "WinBuild2025.vmx"
    vmx.parent.mkdir(parents=True)
    vmx.write_text('guestOS = "windows10srv-64"\n', encoding="utf-8")

    backend = _FakeBackend(running_after=0, clone_guest_os="windows10srv-64")
    calls = _patch_common(monkeypatch, backend)
    extra = [
        "--guest-cpu",
        "4",
        "--guest-memory-mb",
        "8192",
        "--extra-env-json",
        '{"BUILD_TAG":"abc"}',
    ]
    result = CliRunner().invoke(
        cli,
        _common_args(
            vmx,
            clone_base=tmp_path / "vms",
            artifact_base=tmp_path / "art",
            extra=extra,
        ),
    )
    assert result.exit_code == 0, result.output
    assert calls["windows_build"], "Windows build branch not taken"
    assert calls["windows_build"][0]["extra_env"] == {"BUILD_TAG": "abc"}
    # CPU/RAM override patched the clone VMX.
    clone_vmx = next((tmp_path / "vms").rglob("*.vmx"), None)
    # After successful job the clone dir is deleted, so the file is gone.
    # Check that the Windows collect helper got the collect dir instead.
    assert calls["windows_collect"][0]["guest_path"] == "C:\\CI\\output"
    assert clone_vmx is None  # cleaned up
    assert backend.calls[-1] == "delete"


def test_job_invalid_extra_env_json(
    monkeypatch: pytest.MonkeyPatch, tmp_path: Path, template_vmx: Path
) -> None:
    backend = _FakeBackend()
    _patch_common(monkeypatch, backend)
    result = CliRunner().invoke(
        cli,
        _common_args(
            template_vmx,
            clone_base=tmp_path / "vms",
            artifact_base=tmp_path / "art",
            extra=["--extra-env-json", '{"bad-key":"x"}'],
        ),
    )
    assert result.exit_code != 0
    assert "not a valid env var name" in result.output
    # Failure is parameter-time; clone never started.
    assert backend.calls == []


# ──────────────────────────────────────────────── helper-level tests


def test_apply_vmx_overrides_replaces_existing_lines(tmp_path: Path) -> None:
    vmx = tmp_path / "x.vmx"
    vmx.write_text(
        'config.version = "8"\nnumvcpus = "1"\nmemsize = "2048"\n',
        encoding="utf-8",
    )
    job_module._apply_vmx_overrides(vmx, cpu=8, memory_mb=16384)
    text = vmx.read_text(encoding="utf-8")
    assert 'numvcpus = "8"' in text
    assert 'memsize = "16384"' in text


def test_apply_vmx_overrides_appends_when_missing(tmp_path: Path) -> None:
    vmx = tmp_path / "x.vmx"
    vmx.write_text('config.version = "8"\n', encoding="utf-8")
    job_module._apply_vmx_overrides(vmx, cpu=2, memory_mb=4096)
    text = vmx.read_text(encoding="utf-8")
    assert 'numvcpus = "2"' in text
    assert 'memsize = "4096"' in text


def test_read_guest_os_detects_linux(tmp_path: Path) -> None:
    vmx = tmp_path / "x.vmx"
    vmx.write_text('guestOS = "ubuntu-64"\n', encoding="utf-8")
    assert job_module._read_guest_os_from_vmx(vmx) == "linux"


def test_read_guest_os_defaults_to_windows(tmp_path: Path) -> None:
    vmx = tmp_path / "x.vmx"
    vmx.write_text("# nothing useful\n", encoding="utf-8")
    assert job_module._read_guest_os_from_vmx(vmx) == "windows"


def test_parse_extra_env_json_empty_inputs() -> None:
    assert job_module._parse_extra_env_json("") == {}
    assert job_module._parse_extra_env_json("{}") == {}


def test_parse_extra_env_json_rejects_non_object() -> None:
    with pytest.raises(click.exceptions.UsageError):
        job_module._parse_extra_env_json("[]")


def test_read_guest_os_handles_oserror(tmp_path: Path) -> None:
    """Unreadable VMX (e.g. a directory) returns the safe ``windows`` default."""
    p = tmp_path / "is_a_dir.vmx"
    p.mkdir()
    assert job_module._read_guest_os_from_vmx(p) == "windows"


def test_apply_vmx_overrides_noop_when_both_zero(tmp_path: Path) -> None:
    vmx = tmp_path / "x.vmx"
    vmx.write_text('numvcpus = "1"\n', encoding="utf-8")
    job_module._apply_vmx_overrides(vmx, cpu=0, memory_mb=0)
    assert vmx.read_text(encoding="utf-8") == 'numvcpus = "1"\n'


def test_parse_extra_env_json_invalid_json_raises() -> None:
    with pytest.raises(click.exceptions.UsageError):
        job_module._parse_extra_env_json("{not json")


def test_parse_extra_env_json_coerces_value_types() -> None:
    out = job_module._parse_extra_env_json('{"A":null,"B":42,"C":"x"}')
    assert out == {"A": "", "B": "42", "C": "x"}


def test_parse_extra_env_json_rejects_non_string_key() -> None:
    # JSON keys are always strings, so coerce via raw call to guarantee branch.
    with pytest.raises(click.exceptions.UsageError):
        job_module._parse_extra_env_json('{"1bad":"x"}')


def test_job_wait_running_swallows_backend_error(
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    from ci_orchestrator.backends.errors import BackendError
    from ci_orchestrator.backends.protocol import VmHandle

    class _Boom:
        def is_running(self, _h: VmHandle) -> bool:
            raise BackendError("nope")

    monkeypatch.setattr(job_module.time, "sleep", lambda _s: None)
    seq = iter([0.0, 1.0])
    monkeypatch.setattr(job_module, "_now", lambda: next(seq))
    assert (
        job_module._wait_running(_Boom(), VmHandle("x"), deadline=0.5, poll=0.001)  # type: ignore[arg-type]
        is False
    )


def test_job_wait_for_ip_returns_none_on_backend_error(
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    from ci_orchestrator.backends.errors import BackendError
    from ci_orchestrator.backends.protocol import VmHandle

    class _Boom:
        def get_ip(self, _h: VmHandle, timeout: float = 0.0) -> str | None:
            raise BackendError("ip nope")

    monkeypatch.setattr(job_module.time, "sleep", lambda _s: None)
    seq = iter([0.0, 1.0])
    monkeypatch.setattr(job_module, "_now", lambda: next(seq))
    assert (
        job_module._wait_for_ip(_Boom(), VmHandle("x"), deadline=0.5, poll=0.001)  # type: ignore[arg-type]
        is None
    )