From 5932d228012ca5ba7a02bd9e16a2fe8d3dcf7230 Mon Sep 17 00:00:00 2001
From: Miguel Jacq <mig@mig5.net>
Date: Mon, 3 Aug 2026 15:58:14 +1000
Subject: [PATCH] Make remote harvest zipapp stdlib-only

Lazy-load manifest, explain, and validation dependencies so the remote
harvest zipapp does not require jsonschema, PyYAML, Paramiko, or other
site-packages on the target host.

Run the remote zipapp with Python isolated mode and site-packages
disabled, while preserving existing CLI monkeypatch hooks.
---
 enroll/cli.py        | 24 +++++++++++++--
 enroll/remote.py     |  2 ++
 tests/test_remote.py | 71 ++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 94 insertions(+), 3 deletions(-)

diff --git a/enroll/cli.py b/enroll/cli.py
index a123fd1..b0147d3 100644
--- a/enroll/cli.py
+++ b/enroll/cli.py
@@ -18,20 +18,38 @@ from .diff import (
     post_webhook,
     send_email,
 )
-from .explain import explain_state
 from .harvest import harvest
 from .harvest_safety import ensure_safe_output_parent, write_text_output_file
-from .manifest import manifest
 from .remote import (
     remote_harvest,
     RemoteSudoPasswordRequired,
     RemoteSSHKeyPassphraseRequired,
 )
 from .sopsutil import SopsError, encrypt_file_binary
-from .validate import validate_harvest
 from .version import get_enroll_version
 
 
+def explain_state(*args, **kwargs):
+    """Load the explain implementation only when that command is used."""
+    from .explain import explain_state as _explain_state
+
+    return _explain_state(*args, **kwargs)
+
+
+def manifest(*args, **kwargs):
+    """Load manifest dependencies only when rendering a manifest."""
+    from .manifest import manifest as _manifest
+
+    return _manifest(*args, **kwargs)
+
+
+def validate_harvest(*args, **kwargs):
+    """Load jsonschema only when validation is requested."""
+    from .validate import validate_harvest as _validate_harvest
+
+    return _validate_harvest(*args, **kwargs)
+
+
 def _discover_config_path(argv: list[str]) -> Optional[Path]:
     """Return the config path to use, if any.
 
diff --git a/enroll/remote.py b/enroll/remote.py
index 83006d9..60eb7a3 100644
--- a/enroll/remote.py
+++ b/enroll/remote.py
@@ -1126,6 +1126,8 @@ def _remote_harvest(
             # Run remote harvest.
             argv: list[str] = [
                 remote_python,
+                "-I",
+                "-S",
                 rapp,
                 "harvest",
                 "--out",
diff --git a/tests/test_remote.py b/tests/test_remote.py
index 1e5ffcc..7e44316 100644
--- a/tests/test_remote.py
+++ b/tests/test_remote.py
@@ -313,6 +313,9 @@ def test_remote_harvest_happy_path(tmp_path: Path, monkeypatch):
     assert promote_argv.count("enroll.pyz") >= 2
     assert "/tmp/enroll-root-123/enroll.pyz" in harvest_cmd
     assert "/tmp/enroll-remote-123/enroll.pyz" not in harvest_cmd
+    harvest_argv = shlex.split(harvest_cmd)
+    pyz_i = harvest_argv.index("/tmp/enroll-root-123/enroll.pyz")
+    assert harvest_argv[pyz_i - 2 : pyz_i] == ["-I", "-S"]
     assert not any(_is_pyz_verify_cmd(c) for c, _pty in calls)
 
     # The trusted digest must be obtained while the archive is still private,
@@ -1251,6 +1254,74 @@ def test_remote_verify_pyz_sha256_rejects_nonzero_rc():
         )
 
 
+def test_build_enroll_pyz_harvest_starts_without_site_packages(tmp_path: Path):
+    """The remote payload must not depend on packages installed on the target.
+
+    ``-S`` disables site-packages, reproducing a target that has Python but no
+    jsonschema/PyYAML. ``-I`` also ignores PYTHON* environment variables and
+    the current directory. The old eager CLI imports failed here before
+    argparse could dispatch to ``harvest``.
+    """
+    import subprocess
+    import sys
+
+    import enroll.remote as r
+
+    pyz, _sha = r._build_enroll_pyz(tmp_path)
+    proc = subprocess.run(
+        [sys.executable, "-I", "-S", str(pyz), "harvest", "--help"],
+        check=False,
+        stdout=subprocess.PIPE,
+        stderr=subprocess.PIPE,
+        text=True,
+        timeout=15,
+    )
+
+    assert proc.returncode == 0, proc.stderr
+    assert "usage: enroll harvest" in proc.stdout
+    assert "ModuleNotFoundError" not in proc.stderr
+    assert "jsonschema" not in proc.stderr
+    assert "yaml" not in proc.stderr.lower()
+
+
+def test_build_enroll_pyz_runs_harvest_without_site_packages(tmp_path: Path):
+    """Exercise command dispatch and the actual harvest implementation under -S."""
+    import subprocess
+    import sys
+
+    import enroll.remote as r
+
+    build_dir = tmp_path / "build"
+    build_dir.mkdir()
+    pyz, _sha = r._build_enroll_pyz(build_dir)
+    out_dir = tmp_path / "bundle"
+    proc = subprocess.run(
+        [
+            sys.executable,
+            "-I",
+            "-S",
+            str(pyz),
+            "harvest",
+            "--out",
+            str(out_dir),
+            "--exclude-path",
+            "/**",
+            "--assume-safe-path",
+        ],
+        check=False,
+        stdout=subprocess.PIPE,
+        stderr=subprocess.PIPE,
+        text=True,
+        timeout=60,
+    )
+
+    assert proc.returncode == 0, proc.stderr
+    assert (out_dir / "state.json").is_file()
+    assert "ModuleNotFoundError" not in proc.stderr
+    assert "jsonschema" not in proc.stderr
+    assert "yaml" not in proc.stderr.lower()
+
+
 def test_build_enroll_pyz_excludes_tests_and_caches_and_returns_sha(tmp_path: Path):
     import zipfile