fcd2768c05
What: _write_fleet_restart_pending_marker reports whether a store took the restart and arm_commit_obligations raises (pre-move refusal) when neither did; the arming snapshot treats only FileNotFoundError as absence (any other read error refuses); disarm restores/unlinks a record only while its bytes are still exactly what this run's latest arm left there. Why: the tree could move with no restart owed; disarm replayed a stale snapshot over another install's newer host record (shared per OS user) and deleted a record it merely failed to read. Red (7847374318c update_cmd_commit.py + update_cmd_fleet.py): arm returned with both stores unwritable; disarm -> 'OLD' over 'OTHER-INSTALL'; read error -> record deleted. Green: arm raises OSError; newer record kept; unreadable record survives; plain arm/disarm still restores.
394 lines
18 KiB
Python
394 lines
18 KiB
Python
"""The single commit point of a source ``hermes update``: everything the tree move needs armed first.
|
|
|
|
Before git (or the ZIP swap) writes the first file of the new code, three things are durable:
|
|
|
|
* the interrupted-pull marker (``.git/hermes-update-pull``) naming the commit git moves to, for EVERY
|
|
tree move (branch switch, upstream fork pull, the pull itself) so a launch after any exit that left
|
|
a torn tree puts the old one back (``_early_recovery.restore_interrupted_pull``);
|
|
* the source-completion tail (``venv_sync.arm_completion``) and the host fleet-restart obligation,
|
|
so a kill after the move but before the completion child starts still leaves the tail owed to the
|
|
next launch (``venv_sync.prepare_launch``) and the restart owed to the next ``hermes update``.
|
|
|
|
Failures before the move are no-ops: ``disarm_commit_obligations`` puts both obligations back the way
|
|
this run found them once the tree is verified at its pre-update commit again.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import os
|
|
import subprocess
|
|
import sys
|
|
from contextlib import suppress
|
|
from pathlib import Path
|
|
from typing import Optional
|
|
|
|
from hermes_cli._early_recovery import (
|
|
RECOVERY_CLOSURE,
|
|
RECOVERY_CLOSURE_INIT,
|
|
RECOVERY_CLOSURE_MANIFEST,
|
|
blob_id,
|
|
interrupted_pull_marker,
|
|
is_object_id,
|
|
recovery_closure_dir,
|
|
recovery_closure_verified,
|
|
restore_interrupted_pull,
|
|
)
|
|
from hermes_cli.update_custody import run_git
|
|
|
|
# What this run found before it armed anything: {path: bytes or None}. None = nothing armed yet.
|
|
_armed_snapshot: Optional[dict[Path, Optional[bytes]]] = None
|
|
# What this run's latest arm left at each path: disarm hands back only records still exactly these.
|
|
_armed_bytes: dict[Path, Optional[bytes]] = {}
|
|
# (git_cmd, (HEAD, branch)) when this run reached its checkout phase, and the checkout it names.
|
|
_run_start: Optional[tuple[list, tuple[str, str]]] = None
|
|
_obligation_root: Optional[Path] = None
|
|
|
|
|
|
def _owns_live_checkout(root: Path) -> bool:
|
|
from hermes_cli.update_cmd import _m
|
|
|
|
return _m()._pytest_owns_live_checkout(root)
|
|
|
|
|
|
def _obligation_paths(root: Path) -> list[Path]:
|
|
from hermes_cli.update_cmd_fleet import _fleet_restart_pending_marker_path
|
|
from hermes_cli.update_host_obligation import host_obligation_path
|
|
from hermes_cli.venv_sync import completion_pending_path
|
|
|
|
return [completion_pending_path(root), _fleet_restart_pending_marker_path(), host_obligation_path()]
|
|
|
|
|
|
def _read_or_none(path: Path) -> Optional[bytes]:
|
|
"""The record's bytes, None when absent. Any other read error raises: custody is never guessed."""
|
|
try:
|
|
return path.read_bytes()
|
|
except FileNotFoundError:
|
|
return None
|
|
|
|
|
|
def arm_commit_obligations(root: Path, expected_sha: str) -> None:
|
|
"""Owe the completion tail and the fleet restart for ``expected_sha`` BEFORE the tree moves.
|
|
|
|
Idempotent within a run (the first call snapshots what to restore on a no-op failure). An
|
|
unwritable install state, an unreadable record or a fleet restart neither store accepted
|
|
raises: nothing has moved yet, and moving without the obligation is exactly the
|
|
tail-never-runs state this exists to prevent.
|
|
"""
|
|
global _armed_snapshot
|
|
from hermes_cli.update_cmd_fleet import _write_fleet_restart_pending_marker
|
|
from hermes_cli.venv_sync import arm_completion
|
|
|
|
root = Path(root)
|
|
if _owns_live_checkout(root):
|
|
return
|
|
if _armed_snapshot is None:
|
|
_armed_snapshot = {path: _read_or_none(path) for path in _obligation_paths(root)}
|
|
try:
|
|
arm_completion(root)
|
|
owed = _write_fleet_restart_pending_marker(expected_sha=expected_sha or "")
|
|
finally: # even a half-done arm: disarm must still recognise what this run wrote
|
|
_armed_bytes.update({path: _read_or_none(path) for path in _armed_snapshot})
|
|
if not owed:
|
|
raise OSError("could not record the pending gateway-restart obligation")
|
|
|
|
|
|
def disarm_commit_obligations() -> None:
|
|
"""Restore both obligations to what this run found: the tree never left its pre-update commit.
|
|
|
|
Refused (obligations stay armed) unless HEAD is the commit this run STARTED from: a failed later
|
|
move (the upstream fork ff after a committed origin pull) is put back to a commit that is
|
|
already new code, and that tree still owes its tail.
|
|
"""
|
|
global _armed_snapshot
|
|
if _run_start is not None:
|
|
git_cmd, (start_head, _branch) = _run_start
|
|
root = Path(_obligation_root) if _obligation_root is not None else None
|
|
if root is None or not start_head or head_and_branch(git_cmd, root)[0] != start_head:
|
|
return
|
|
snapshot, _armed_snapshot = _armed_snapshot, None
|
|
armed = dict(_armed_bytes)
|
|
_armed_bytes.clear()
|
|
for path, data in (snapshot or {}).items():
|
|
try:
|
|
if path not in armed or _read_or_none(path) != armed[path]:
|
|
continue # rewritten since this run armed it (another install's update): theirs now
|
|
if data is None:
|
|
path.unlink(missing_ok=True)
|
|
else:
|
|
tmp = path.with_name(path.name + ".restore")
|
|
tmp.write_bytes(data)
|
|
os.replace(tmp, path)
|
|
except OSError:
|
|
pass # an owed tail/restart left armed is a retry, never a lost obligation
|
|
|
|
|
|
def commit_obligations_armed() -> bool:
|
|
return _armed_snapshot is not None
|
|
|
|
|
|
def arm_tree_move(git_cmd, root: Path, *, pre: str | None, target: str, stash: str | None,
|
|
rollback: str | None = None) -> Path:
|
|
"""Write the interrupted-pull marker for one git tree move (pre -> target).
|
|
|
|
``rollback`` (``branch``/``detach``): a syntax rollback that moves HEAD and the index back to
|
|
``pre`` before any file; a kill before that step finds HEAD still on ``target`` and the restore
|
|
redoes it first.
|
|
"""
|
|
marker = interrupted_pull_marker(root)
|
|
if pre:
|
|
with suppress(OSError, subprocess.SubprocessError): # no closure: the in-tree repair still runs
|
|
publish_recovery_closure(git_cmd, root, pre)
|
|
marker.write_text(f"pid={os.getpid()}\npre={pre or ''}\ntarget={target}\nstash={stash or ''}\n"
|
|
+ (f"rollback={rollback}\n" if rollback else "")
|
|
+ (f"git={git}\n" if (git := _absolute_git(git_cmd)) else ""), encoding="utf-8")
|
|
return marker
|
|
|
|
|
|
def publish_recovery_closure(git_cmd, root: Path, pre: str) -> Path:
|
|
"""Copy the launch repair's modules, as committed at ``pre``, beside the marker (outside the tree).
|
|
|
|
The bytes come from the commit the repair restores, never from the working tree: a rollback move
|
|
starts from a broken tree, and a second move in one run starts from code this process never ran.
|
|
A manifest names each file's blob id; the launcher runs the closure only when every file hashes
|
|
to it (``recovery_closure_verified``) and rebuilds it from git's objects otherwise. Files and
|
|
directory are fsynced before the atomic rename, so a power loss during the move that follows
|
|
finds the whole closure or none. An older run's closure (another ``pre``) is dropped.
|
|
"""
|
|
import shutil
|
|
|
|
if not is_object_id(pre):
|
|
raise OSError(f"not a commit id: {pre!r}")
|
|
dest = recovery_closure_dir(Path(root), pre)
|
|
if recovery_closure_verified(dest, pre):
|
|
return dest
|
|
git_cmd = [*([git_cmd] if isinstance(git_cmd, str) else git_cmd), "--no-replace-objects"]
|
|
listed = run_git(git_cmd, ["ls-tree", "-z", "--full-tree", pre, "--", *RECOVERY_CLOSURE], cwd=str(root),
|
|
capture_output=True, stdin=subprocess.DEVNULL, timeout=120)
|
|
ids = {}
|
|
for entry in (listed.stdout or b"").split(b"\0") if listed.returncode == 0 else ():
|
|
meta, _, rel = entry.decode("utf-8", "replace").partition("\t")
|
|
if meta.split()[1:2] == ["blob"]:
|
|
ids[rel] = meta.split()[2]
|
|
blobs = {RECOVERY_CLOSURE_INIT: b""}
|
|
for rel in RECOVERY_CLOSURE:
|
|
shown = run_git(git_cmd, ["cat-file", "blob", ids.get(rel, "")], cwd=str(root), capture_output=True,
|
|
stdin=subprocess.DEVNULL, timeout=120) if rel in ids else None
|
|
if shown is None or shown.returncode != 0 or blob_id(shown.stdout or b"", pre) != ids[rel]:
|
|
raise OSError(f"{rel} is not in {pre[:10]}")
|
|
blobs[rel] = shown.stdout
|
|
manifest = "".join(f"{blob_id(data, pre)} {rel}\n" for rel, data in blobs.items()).encode("utf-8")
|
|
staging = dest.with_name(f".{pre}.{os.getpid()}.staging")
|
|
shutil.rmtree(staging, ignore_errors=True)
|
|
try:
|
|
for rel, data in {**blobs, RECOVERY_CLOSURE_MANIFEST: manifest}.items():
|
|
(staging / rel).parent.mkdir(parents=True, exist_ok=True)
|
|
with open(staging / rel, "wb") as handle:
|
|
handle.write(data)
|
|
handle.flush()
|
|
os.fsync(handle.fileno())
|
|
for directory in (staging / "hermes_cli", staging):
|
|
_fsync_dir(directory)
|
|
shutil.rmtree(dest, ignore_errors=True)
|
|
os.replace(staging, dest)
|
|
_fsync_dir(dest.parent)
|
|
finally:
|
|
shutil.rmtree(staging, ignore_errors=True)
|
|
for stale in dest.parent.iterdir():
|
|
if stale != dest:
|
|
shutil.rmtree(stale, ignore_errors=True)
|
|
return dest
|
|
|
|
|
|
def _fsync_dir(directory: Path) -> None:
|
|
"""POSIX: make a directory's entries durable. Windows opens no directory handle; NTFS journals
|
|
the rename itself."""
|
|
if os.name == "nt":
|
|
return
|
|
fd = os.open(directory, os.O_RDONLY)
|
|
try:
|
|
os.fsync(fd)
|
|
finally:
|
|
os.close(fd)
|
|
|
|
|
|
def _absolute_git(git_cmd) -> str:
|
|
"""The git this run moves the tree with, as an absolute path for the next launch's repair.
|
|
|
|
A Windows install's git is often PM's store copy, on PATH only inside the updater
|
|
(``expose_pm_git``); the repair cannot look it up through ``pm`` when the killed move tore a
|
|
module ``pm`` imports (``hermes_constants``)."""
|
|
import shutil
|
|
|
|
head = (list(git_cmd) or ["git"])[0] if not isinstance(git_cmd, str) else git_cmd
|
|
found = shutil.which(head)
|
|
return os.path.abspath(found) if found else ""
|
|
|
|
|
|
def files_added_by(git_cmd, root: Path, pre: str, target: str | None) -> list[str]:
|
|
"""Paths ``target`` adds over ``pre``. A rollback's mixed reset un-tracks them and ``reset --hard``
|
|
never touches untracked files, so ``drop_added_files`` removes them by name afterwards."""
|
|
if not target or target == pre:
|
|
return []
|
|
cp = run_git(git_cmd, ["diff", "--name-only", "-z", "--no-renames", "--diff-filter=A", pre, target],
|
|
cwd=str(root), capture_output=True, text=True, encoding="utf-8", errors="replace",
|
|
stdin=subprocess.DEVNULL, timeout=120)
|
|
return [p for p in cp.stdout.split("\0") if p] if cp.returncode == 0 else []
|
|
|
|
|
|
def drop_added_files(root: Path, added: list[str]) -> bool:
|
|
"""Every one is the move's own file: git refuses to move a checkout over an untracked file.
|
|
|
|
False when one of them is still there (the caller keeps the marker; the restore retries)."""
|
|
root = Path(root)
|
|
gone = True
|
|
for rel in added:
|
|
try:
|
|
(root / rel).unlink(missing_ok=True)
|
|
except OSError:
|
|
gone = False
|
|
for parent in sorted({p for rel in added for p in Path(rel).parents if str(p) != "."},
|
|
key=lambda p: len(p.parts), reverse=True):
|
|
try:
|
|
(root / parent).rmdir() # only when empty: anything else inside keeps it
|
|
except OSError:
|
|
pass
|
|
return gone and not any(os.path.lexists(root / rel) for rel in added)
|
|
|
|
|
|
def tree_whole_at(git_cmd, root: Path, sha: str) -> bool:
|
|
"""HEAD is ``sha``, no ``index.lock`` is left and no tracked file differs from it."""
|
|
def run(*args: str) -> subprocess.CompletedProcess:
|
|
return run_git(git_cmd, [*args], cwd=str(root), capture_output=True, text=True, encoding="utf-8",
|
|
errors="replace", stdin=subprocess.DEVNULL, timeout=120)
|
|
|
|
head = run("rev-parse", "-q", "--verify", "HEAD")
|
|
if head.returncode != 0 or head.stdout.strip() != sha or (interrupted_pull_marker(Path(root)).parent / "index.lock").exists():
|
|
return False
|
|
status = run("status", "--porcelain", "-z", "--untracked-files=no")
|
|
return status.returncode == 0 and not status.stdout.strip("\0")
|
|
|
|
|
|
def settle_failed_tree_move(root: Path) -> bool:
|
|
"""Git exited without finishing a tree move: put back what it wrote, keep the marker if we can't.
|
|
|
|
True when the tree is verified whole again (at the pre-move commit, or at the target when git got
|
|
there anyway); False leaves the marker for the next launch's ``restore_interrupted_pull``.
|
|
"""
|
|
restore_interrupted_pull(Path(root), after_failure=True)
|
|
return not interrupted_pull_marker(Path(root)).is_file()
|
|
|
|
|
|
def requires_other_python(pyproject: bytes | str | None) -> bool:
|
|
"""True when a target's ``requires-python`` excludes the running interpreter.
|
|
|
|
Its startup modules may then use a newer Python's syntax that this interpreter's ``compile()``
|
|
cannot judge (PM provisions the required Python with the new code). Unknown -> False: keep
|
|
checking, so a real syntax error is never waved through on a guess.
|
|
"""
|
|
if not pyproject:
|
|
return False
|
|
import platform
|
|
import re
|
|
import tomllib
|
|
|
|
try:
|
|
text = pyproject.decode("utf-8") if isinstance(pyproject, bytes) else pyproject
|
|
spec = tomllib.loads(text)["project"]["requires-python"]
|
|
try:
|
|
from packaging.specifiers import SpecifierSet
|
|
except ImportError: # the lower bound is the part a Python bump moves
|
|
floor = re.search(r">=\s*(\d+)\.(\d+)", spec)
|
|
return bool(floor) and sys.version_info[:2] < (int(floor[1]), int(floor[2]))
|
|
return not SpecifierSet(spec).contains(platform.python_version(), prereleases=True)
|
|
except (KeyError, TypeError, ValueError): # TOMLDecodeError / InvalidSpecifier are ValueErrors
|
|
return False
|
|
|
|
|
|
def target_syntax_error(git_cmd, root: Path, target_ref: str, relpaths) -> tuple[str, str] | None:
|
|
"""``(path, error)`` for the first startup-critical file that does not compile at ``target_ref``.
|
|
|
|
Read from the object store, never written to the tree: this runs BEFORE HEAD moves, so a broken
|
|
release is refused with the install untouched (the post-pull rollback stays as the backstop).
|
|
Skipped for a target that requires a Python this interpreter is not (``requires_other_python``).
|
|
"""
|
|
pyproject = run_git(git_cmd, ["show", f"{target_ref}:pyproject.toml"], cwd=str(root),
|
|
capture_output=True, stdin=subprocess.DEVNULL, timeout=120)
|
|
if pyproject.returncode == 0 and requires_other_python(pyproject.stdout):
|
|
return None
|
|
for rel in relpaths:
|
|
shown = run_git(git_cmd, ["show", f"{target_ref}:{rel}"], cwd=str(root), capture_output=True,
|
|
stdin=subprocess.DEVNULL, timeout=120)
|
|
if shown.returncode != 0:
|
|
continue # absent at the target (or unreadable): the post-pull guard has the last word
|
|
try:
|
|
compile(shown.stdout, rel, "exec", dont_inherit=True)
|
|
except (SyntaxError, ValueError) as exc:
|
|
return rel, f"{type(exc).__name__}: {exc}"
|
|
return None
|
|
|
|
|
|
def head_and_branch(git_cmd, root: Path) -> tuple[str, str]:
|
|
def out(*args: str) -> str:
|
|
cp = run_git(git_cmd, [*args], cwd=str(root), capture_output=True, text=True, encoding="utf-8", errors="replace",
|
|
stdin=subprocess.DEVNULL, timeout=60)
|
|
return cp.stdout.strip() if cp.returncode == 0 else ""
|
|
|
|
return out("rev-parse", "HEAD"), out("rev-parse", "--abbrev-ref", "HEAD")
|
|
|
|
|
|
def checkout_untouched(git_cmd, root: Path, start: tuple[str, str] | None) -> bool:
|
|
"""True when the checkout is exactly where this run found it: same HEAD, same branch, no
|
|
autostash taken, no tree move armed. Only then may a failed git update fall back to ZIP."""
|
|
from hermes_cli.update_cmd_stash import _unrestored_autostash_notice
|
|
|
|
if start is None or not start[0]:
|
|
return False
|
|
if _unrestored_autostash_notice() is not None or interrupted_pull_marker(root).is_file():
|
|
return False
|
|
return head_and_branch(git_cmd, root) == start
|
|
|
|
|
|
def record_run_start(git_cmd, root: Path) -> None:
|
|
global _run_start, _obligation_root
|
|
_run_start = (list(git_cmd), head_and_branch(git_cmd, root))
|
|
_obligation_root = Path(root)
|
|
|
|
|
|
def run_checkout_untouched(root: Path) -> bool:
|
|
"""``checkout_untouched`` for this run; True before the checkout phase (nothing can have moved)."""
|
|
if _run_start is None:
|
|
return True
|
|
return checkout_untouched(_run_start[0], Path(root), _run_start[1])
|
|
|
|
|
|
def preflight_refusal(git_cmd, root: Path, target_ref: str, critical_files) -> str | None:
|
|
"""Why this update must not start, checked BEFORE the first tree move; ``None`` to proceed.
|
|
|
|
* a venv owned by another OS user (the completion child used to refuse only after the swap);
|
|
* a target whose startup-critical modules do not compile (rollback used to be the only guard).
|
|
"""
|
|
root = Path(root)
|
|
if not _owns_live_checkout(root):
|
|
try:
|
|
from hermes_cli.venv_sync import refuse_foreign_owned_venv
|
|
|
|
refuse_foreign_owned_venv(root)
|
|
except ImportError:
|
|
pass
|
|
except Exception as exc: # pm's refusal carries its own remediation text
|
|
return f"✗ {exc}"
|
|
broken = target_syntax_error(git_cmd, root, target_ref, critical_files)
|
|
if broken is not None:
|
|
path, error = broken
|
|
return (f"✗ The update target has a syntax error in a critical file:\n {path}\n "
|
|
+ "\n ".join(error.splitlines()[:6]))
|
|
return None
|
|
|
|
|
|
def reset_for_tests() -> None:
|
|
global _armed_snapshot, _run_start, _obligation_root
|
|
_armed_snapshot = None
|
|
_armed_bytes.clear()
|
|
_run_start = None
|
|
_obligation_root = None
|