devcouncil 0.3.0 → 0.4.0

package/src/devcouncil/codeintel/sync/__init__.py

@@ -0,0 +1,19 @@
+"""Native filesystem synchronization for code-intelligence generations."""
+
+from devcouncil.codeintel.sync.coordinator import (
+    SyncCoordinator,
+    SyncState,
+    get_sync_coordinator,
+    stop_all_coordinators,
+)
+from devcouncil.codeintel.sync.incremental import sync_affected_paths
+from devcouncil.codeintel.sync.scope import IndexScope
+
+__all__ = [
+    "IndexScope",
+    "SyncCoordinator",
+    "SyncState",
+    "get_sync_coordinator",
+    "sync_affected_paths",
+    "stop_all_coordinators",
+]

package/src/devcouncil/codeintel/sync/coordinator.py

@@ -0,0 +1,430 @@
+"""Debounced native watcher with reconciliation and a single writer."""
+
+from __future__ import annotations
+
+import logging
+import subprocess
+import sys
+import threading
+import time
+from dataclasses import asdict, dataclass, field
+from pathlib import Path
+from typing import Any, Callable
+
+from devcouncil.codeintel.service import CodeIntelService, get_codeintel_service
+from devcouncil.codeintel.sync.lease import WriterLease
+from devcouncil.codeintel.sync.scope import IndexScope
+
+logger = logging.getLogger(__name__)
+
+
+def _fsevents_preflight(root: Path) -> bool:
+    """Probe FSEvents out of process because startup failure may abort Python.
+
+    Retries once: under heavy reconcile/build load the first probe can time out
+    even when FSEvents is healthy, which previously forced a permanent Kqueue
+    fallback for the process lifetime.
+    """
+    script = (
+        "import sys,time; "
+        "from watchdog.events import FileSystemEventHandler; "
+        "from watchdog.observers import Observer; "
+        "o=Observer(); o.schedule(FileSystemEventHandler(),sys.argv[1],recursive=True); "
+        "o.start(); time.sleep(0.35); "
+        "emitters=list(o.emitters); "
+        "ok=bool(emitters) and all(e.is_alive() for e in emitters); "
+        "o.stop(); o.join(2); raise SystemExit(0 if ok else 3)"
+    )
+    last_error = ""
+    for attempt in range(2):
+        try:
+            completed = subprocess.run(
+                [sys.executable, "-c", script, str(root)],
+                capture_output=True,
+                timeout=5.0,
+                check=False,
+            )
+        except subprocess.TimeoutExpired:
+            last_error = "timeout"
+            time.sleep(0.2)
+            continue
+        except OSError as exc:
+            last_error = str(exc)
+            break
+        if completed.returncode == 0:
+            return True
+        stderr = getattr(completed, "stderr", None) or b""
+        if isinstance(stderr, str):
+            stderr_text = stderr.strip()
+        else:
+            stderr_text = stderr.decode("utf-8", errors="replace").strip()
+        last_error = stderr_text or f"exit {completed.returncode}"
+        time.sleep(0.2)
+    if last_error:
+        logger.info("FSEvents preflight failed: %s", last_error)
+    return False
+
+
+@dataclass
+class SyncState:
+    state: str = "disabled"
+    backend: str = ""
+    backend_kind: str = ""
+    generation: int | None = None
+    pending: list[str] = field(default_factory=list)
+    last_sync_at: float | None = None
+    last_reconcile_at: float | None = None
+    degraded_reason: str = ""
+    last_error: str = ""
+    build_id: str = ""
+    build_state: str = "idle"
+    build_pid: int | None = None
+    build_phase: str = ""
+    build_completed: int = 0
+    build_total: int = 0
+    build_last_progress_at: float | None = None
+    generation_before: int | None = None
+    generation_after: int | None = None
+    compatibility_export: str = "unknown"
+
+    def as_dict(self) -> dict[str, object]:
+        return asdict(self)
+
+
+class SyncCoordinator:
+    def __init__(
+        self,
+        service: CodeIntelService,
+        *,
+        debounce_seconds: float = 0.75,
+        reconcile_seconds: float = 60.0,
+        sync_callback: Callable[[list[str]], object] | None = None,
+        allow_polling_fallback: bool = True,
+    ):
+        self.service = service
+        self.root = service.project_root
+        self.scope = IndexScope(self.root)
+        self.debounce_seconds = max(0.1, min(60.0, debounce_seconds))
+        self.reconcile_seconds = max(1.0, reconcile_seconds)
+        self.sync_callback = sync_callback or self._default_sync
+        self.allow_polling_fallback = allow_polling_fallback
+        self._state = SyncState(generation=service.store.current_generation())
+        self._pending: set[str] = set()
+        self._pending_since: float | None = None
+        self._condition = threading.Condition()
+        self._stop = threading.Event()
+        self._observer: Any = None
+        self._worker: threading.Thread | None = None
+        self._sync_lock = threading.Lock()
+        self._lease = WriterLease(self.root / ".devcouncil" / "codeintel" / "writer.lock")
+
+    def start(self) -> SyncState:
+        if self._worker is not None and self._worker.is_alive():
+            return self.status()
+        self._stop.clear()
+        self._start_observer()
+        self._worker = threading.Thread(target=self._run, name="devcouncil-codeintel-sync", daemon=True)
+        self._worker.start()
+        return self.status()
+
+    def stop(self, *, timeout: float = 5.0) -> None:
+        self._stop.set()
+        with self._condition:
+            self._condition.notify_all()
+        observer = self._observer
+        if observer is not None:
+            observer.stop()
+            observer.join(timeout=timeout)
+            self._observer = None
+        if self._worker is not None:
+            self._worker.join(timeout=timeout)
+            if self._worker.is_alive():
+                self._state.state = "stopping"
+                self._state.degraded_reason = "sync worker did not stop before timeout"
+                return
+            self._worker = None
+        if self._sync_lock.locked():
+            self._state.state = "stopping"
+            self._state.degraded_reason = "an in-flight sync is still stopping"
+            return
+        self._state.state = "disabled"
+
+    def status(self) -> SyncState:
+        from devcouncil.codeintel.build_control import read_build_status
+
+        build = read_build_status(self.root)
+        with self._condition:
+            if (
+                self._worker is not None
+                and not self._worker.is_alive()
+                and self._stop.is_set()
+            ):
+                self._worker = None
+                if self._state.state == "stopping":
+                    self._state.state = "disabled"
+            elif (
+                self._worker is None
+                and self._stop.is_set()
+                and not self._sync_lock.locked()
+                and self._state.state == "stopping"
+            ):
+                self._state.state = "disabled"
+            self._state.pending = sorted(self._pending)
+            self._state.generation = self.service.store.current_generation()
+            self._state.build_id = build.build_id
+            self._state.build_state = build.state
+            self._state.build_pid = build.pid
+            self._state.build_phase = build.phase
+            self._state.build_completed = build.completed
+            self._state.build_total = build.total
+            self._state.build_last_progress_at = build.last_progress_at
+            self._state.generation_before = build.generation_before
+            self._state.generation_after = build.generation_after
+            self._state.compatibility_export = build.compatibility_export
+            return SyncState(**asdict(self._state))
+
+    def mark_pending(self, path: str | Path) -> None:
+        try:
+            rel = self.scope.relative(path)
+        except ValueError:
+            return
+        # Deletions cannot pass an existence-based check, so extension + prefix
+        # filtering is enough for paths that were present in the committed index.
+        if not self.scope.includes(rel) and not self.service.store.has_indexed_path(rel):
+            return
+        with self._condition:
+            self._pending.add(rel)
+            self._pending_since = self._pending_since or time.monotonic()
+            self._state.state = "pending"
+            self._condition.notify_all()
+
+    def reconcile(self) -> list[str]:
+        indexed = self.service.store.file_metadata()
+        current: dict[str, tuple[int, int]] = {}
+        for rel in self.scope.files():
+            try:
+                stat = (self.root / rel).stat()
+            except OSError:
+                continue
+            current[rel] = (stat.st_size, stat.st_mtime_ns)
+        changed = {
+            rel
+            for rel, values in current.items()
+            if rel not in indexed or values != indexed[rel][:2]
+        }
+        changed.update(set(indexed) - set(current))
+        if changed:
+            # Every item came from IndexScope.files() or the persisted index, so
+            # repeating per-path git ignore probes through mark_pending() is both
+            # redundant and expensive.
+            with self._condition:
+                self._pending.update(changed)
+                self._pending_since = self._pending_since or time.monotonic()
+                self._state.state = "pending"
+                self._condition.notify_all()
+        self._state.last_reconcile_at = time.time()
+        return sorted(changed)
+
+    def sync_now(self, paths: list[str] | None = None) -> bool:
+        from devcouncil.codeintel.build_control import (
+            GraphBuildBusy,
+            _lease_timeouts,
+            graph_build_session,
+        )
+
+        with self._sync_lock:
+            if paths:
+                for path in paths:
+                    self.mark_pending(path)
+            with self._condition:
+                batch = sorted(self._pending)
+            if not batch:
+                return True
+            try:
+                # Short bounded wait: another watcher/MCP writer often holds the
+                # lease briefly. Pending stays queued on timeout so the debounce
+                # loop retries instead of clearing work as healthy.
+                sync_timeout = _lease_timeouts(self.root)[1]
+                with graph_build_session(
+                    self.root, lease=self._lease, timeout=sync_timeout
+                ):
+                    self._state.state = "syncing"
+                    self.sync_callback(batch)
+                with self._condition:
+                    self._pending.difference_update(batch)
+                    self._pending_since = None if not self._pending else time.monotonic()
+                if self._stop.is_set():
+                    self._state.state = "stopping"
+                else:
+                    # Drop transient lease-busy reasons so a later successful sync
+                    # does not stay permanently ``degraded`` / ``read_only``.
+                    if "writer lease" in (self._state.degraded_reason or "").lower():
+                        self._state.degraded_reason = ""
+                    self._state.state = (
+                        "healthy" if not self._state.degraded_reason else "degraded"
+                    )
+                self._state.last_sync_at = time.time()
+                self._state.last_error = ""
+                self._state.generation = self.service.store.current_generation()
+                return True
+            except GraphBuildBusy as exc:
+                # Keep pending queued; debounce/reconcile will retry after backoff.
+                self._state.state = "pending" if batch or self._pending else "read_only"
+                self._state.degraded_reason = str(exc)
+                return False
+            except Exception as exc:
+                logger.exception("code-intelligence sync failed")
+                self._state.state = "degraded"
+                self._state.last_error = f"{type(exc).__name__}: {exc}"
+                return False
+
+    def wait_until_fresh(self, *, timeout: float = 2.0) -> bool:
+        deadline = time.monotonic() + max(0.0, timeout)
+        while time.monotonic() < deadline:
+            if not self.status().pending:
+                return True
+            self.sync_now()
+            if not self.status().pending:
+                return True
+            time.sleep(0.05)
+        return not self.status().pending
+
+    def _run(self) -> None:
+        # Initial reconciliation is intentionally asynchronous.  MCP clients must
+        # be able to complete initialize/list_tools immediately while the index is
+        # brought current in the background.
+        try:
+            self.reconcile()
+        except Exception as exc:  # noqa: BLE001
+            logger.exception("initial code-intelligence reconciliation failed")
+            self._state.state = "degraded"
+            self._state.last_error = f"{type(exc).__name__}: {exc}"
+        next_reconcile = time.monotonic() + self.reconcile_seconds
+        while not self._stop.is_set():
+            with self._condition:
+                self._condition.wait(timeout=min(self.debounce_seconds, 1.0))
+                pending_since = self._pending_since
+            now = time.monotonic()
+            if pending_since is not None and now - pending_since >= self.debounce_seconds:
+                self.sync_now()
+            if now >= next_reconcile:
+                self.reconcile()
+                next_reconcile = now + self.reconcile_seconds
+
+    def _start_observer(self) -> None:
+        try:
+            from watchdog.events import FileSystemEventHandler
+            from watchdog.observers import Observer
+            from watchdog.observers.polling import PollingObserver
+
+            coordinator = self
+
+            class Handler(FileSystemEventHandler):
+                def on_any_event(self, event):  # noqa: ANN001
+                    if getattr(event, "is_directory", False):
+                        return
+                    source = getattr(event, "src_path", "")
+                    destination = getattr(event, "dest_path", "")
+                    if source:
+                        coordinator.mark_pending(source)
+                    if destination:
+                        coordinator.mark_pending(destination)
+
+            candidates: list[tuple[str, Callable[[], Any]]] = [("native", Observer)]
+            try:
+                from watchdog.observers.kqueue import KqueueObserver
+
+                if KqueueObserver is not Observer:
+                    candidates.append(("native-fallback", KqueueObserver))
+            except (ImportError, AttributeError):
+                pass
+
+            if self.allow_polling_fallback:
+                candidates.append((
+                    "polling",
+                    lambda: PollingObserver(timeout=max(1.0, self.debounce_seconds)),
+                ))
+
+            observer = None
+            failures: list[str] = []
+            backend = ""
+            selected_kind = ""
+            for kind, factory in candidates:
+                candidate = factory()
+                try:
+                    if candidate.__class__.__name__ == "FSEventsObserver" and not _fsevents_preflight(self.root):
+                        raise RuntimeError("FSEvents preflight failed in isolated process")
+                    candidate.schedule(Handler(), str(self.root), recursive=True)
+                    candidate.start()
+                    emitters = list(getattr(candidate, "emitters", []))
+                    if emitters:
+                        # FSEvents can report startup failure only from its
+                        # emitter thread, after Observer.start() has returned.
+                        time.sleep(0.1)
+                        if not all(emitter.is_alive() for emitter in emitters):
+                            raise RuntimeError("observer emitter terminated during startup")
+                    observer = candidate
+                    backend = candidate.__class__.__name__
+                    selected_kind = kind
+                    break
+                except Exception as exc:
+                    failures.append(f"{candidate.__class__.__name__}: {type(exc).__name__}: {exc}")
+                    try:
+                        candidate.stop()
+                        candidate.join(timeout=1.0)
+                    except Exception:
+                        logger.debug("failed to stop rejected observer", exc_info=True)
+            if observer is None:
+                raise RuntimeError("; ".join(failures) or "no observer backend available")
+            if failures or selected_kind == "polling":
+                self._state.degraded_reason = "; ".join(failures) or "polling fallback selected"
+                logger.warning(
+                    "filesystem observer degraded to %s: %s",
+                    backend,
+                    self._state.degraded_reason,
+                )
+            self._observer = observer
+            self._state.backend = backend
+            self._state.backend_kind = "polling" if selected_kind == "polling" else "native"
+            self._state.state = "degraded" if self._state.degraded_reason else "healthy"
+        except Exception as exc:
+            self._observer = None
+            self._state.state = "degraded"
+            self._state.degraded_reason = f"watcher unavailable: {type(exc).__name__}: {exc}"
+
+    def _default_sync(self, paths: list[str]) -> None:
+        # Keep graph watch and ``dev map --watch`` on the same path: incremental
+        # graph sync plus a full repo-map rebuild (subsystems/dependents/guides).
+        from devcouncil.indexing.graph.build import refresh_map_for_paths
+
+        refresh_map_for_paths(self.service.project_root, paths, liveness=True)
+
+
+_COORDINATORS: dict[Path, SyncCoordinator] = {}
+_COORDINATORS_LOCK = threading.Lock()
+
+
+def get_sync_coordinator(root: Path, **kwargs: object) -> SyncCoordinator:
+    service = get_codeintel_service(root)
+    with _COORDINATORS_LOCK:
+        coordinator = _COORDINATORS.get(service.project_root)
+        if coordinator is None:
+            coordinator = SyncCoordinator(service, **kwargs)  # type: ignore[arg-type]
+            _COORDINATORS[service.project_root] = coordinator
+        else:
+            for name, requested in kwargs.items():
+                current = getattr(coordinator, name)
+                if current != requested:
+                    raise ValueError(
+                        f"sync coordinator for {service.project_root} already uses "
+                        f"a different {name}"
+                    )
+        return coordinator
+
+
+def stop_all_coordinators() -> None:
+    with _COORDINATORS_LOCK:
+        coordinators = list(_COORDINATORS.values())
+        _COORDINATORS.clear()
+    for coordinator in coordinators:
+        coordinator.stop()