charisma-cli 0.1.5__tar.gz → 0.2.0__tar.gz

{charisma_cli-0.1.5 → charisma_cli-0.2.0}/.gitignore

@@ -132,3 +132,4 @@ ltm/snapshots/*
 backend/coverage.json
 */__pycache__/*
 allure-results/*
+allure-results-live/

{charisma_cli-0.1.5 → charisma_cli-0.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: charisma-cli
-Version: 0.1.5
+Version: 0.2.0
 Summary: CLI tool that watches allure-results and streams test results + attachments to Charisma.
 Author: Charisma Team
 License-Expression: MIT

{charisma_cli-0.1.5 → charisma_cli-0.2.0}/pyproject.toml

@@ -5,7 +5,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "charisma-cli"
-version = "0.1.5"
+version = "0.2.0"
 description = "CLI tool that watches allure-results and streams test results + attachments to Charisma."
 readme = "README.md"
 license = "MIT"

{charisma_cli-0.1.5 → charisma_cli-0.2.0}/src/charisma_cli/__init__.py

@@ -1,3 +1,3 @@
 """charisma-cli: Stream allure results to Charisma in real time."""
 
-__version__ = "0.1.2"
+__version__ = "0.1.6"

{charisma_cli-0.1.5 → charisma_cli-0.2.0}/src/charisma_cli/main.py

@@ -16,6 +16,15 @@ from charisma_cli.subprocess_mgr import SubprocessManager
 from charisma_cli.uploader import Uploader
 from charisma_cli.watcher import ResultsWatcher, classify_file
 
+# Post-subprocess settle window for the watch command. allure-pytest under heavy
+# xdist can finish flushing its result files to disk seconds after the test
+# subprocess exits; settle_scan keeps polling until the directory is quiet for
+# SETTLE_QUIET_PERIOD or SETTLE_MAX_WAIT elapses. The window is deliberately
+# wider than settle_scan's own defaults (1s/15s) because a real CI burst can
+# land well past 15s — a too-short window is the watch results_sent=0 bug.
+SETTLE_QUIET_PERIOD = 2.0
+SETTLE_MAX_WAIT = 60.0
+
 
 @click.group()
 @click.version_option(version=__version__, prog_name="charismactl")
@@ -185,13 +194,19 @@ def watch(
     mgr.spawn(command)
     exit_code = mgr.wait()
 
-    # After subprocess exits (crash or normal): flush debouncing files and scan
-    # for anything the watcher missed entirely. This prevents data loss when
-    # parallel test workers write results in bursts just before crash.
+    # Reconcile the results directory after the subprocess exits. allure-pytest
+    # writes its result files in a burst during test-process teardown; under
+    # heavy xdist parallelism that burst can finish flushing to disk seconds
+    # after the subprocess returns. flush_pending drains any files still in the
+    # 500ms debounce window; settle_scan then polls the directory until it is
+    # quiet, capturing the whole late burst before we drain. Without a wide
+    # enough window the queue is drained empty and nothing is sent (the watch
+    # results_sent=0 bug). The shared seen-set keeps every file exactly-once.
     watcher.flush_pending()
-    watcher.final_scan()
+    watcher.settle_scan(quiet_period=SETTLE_QUIET_PERIOD, max_wait=SETTLE_MAX_WAIT)
 
-    # Drain and close — stop watcher AFTER drain so all files are processed
+    # Drain and close — stop the watcher AFTER drain so the consumer's final
+    # queue flush still sees anything the observer enqueued during drain.
     uploader.drain(timeout=config.drain_timeout)
     watcher.stop()
     uploader.stop()

{charisma_cli-0.1.5 → charisma_cli-0.2.0}/src/charisma_cli/uploader.py

@@ -26,6 +26,11 @@ _BATCH_SIZE = 50
 _BATCH_TIMEOUT_SECONDS = 2.0
 _CONTAINER_ASSOCIATION_TIMEOUT = 60.0
 
+# Sentinel expectedTests for CLI launches: the total count is unknown at open
+# time (directory-watch streaming), so we send the backend maximum. This keeps
+# the launch in 'receiving' across all appends; close_launch() finalizes it.
+_UNKNOWN_EXPECTED_TESTS = 1_000_000
+
 
 def _epoch_ms_to_iso(epoch_ms: int) -> str:
     """Convert epoch milliseconds to ISO 8601 UTC string."""
@@ -173,9 +178,14 @@ class Uploader:
         """
         self.ensure_client()
 
+        # The CLI streams from a watched directory and does not know the total
+        # test count at open time. Send a high sentinel so the launch stays
+        # 'receiving' through every append; finalization is driven by
+        # close_launch() at the end of the run (or the stale-launch finalizer as
+        # a backstop), never by the append counter reaching expectedTests.
         payload: dict[str, Any] = {
             "projectAlias": self._config.project,
-            "expectedTests": 1000,
+            "expectedTests": _UNKNOWN_EXPECTED_TESTS,
         }
         if self._config.build_id:
             payload["buildId"] = self._config.build_id
@@ -306,16 +316,16 @@ class Uploader:
         if result.uuid:
             self._known_result_uuids.add(result.uuid)
 
-        # Build API payload
+        # Build API payload (snake_case wire format — shared TestResultInput contract)
         payload: dict[str, Any] = {
-            "testId": result.testId,
+            "test_id": result.testId,
             "outcome": result.outcome,
             "duration_ms": result.duration_ms,
         }
         if result.name:
             payload["name"] = result.name
         if result.full_name:
-            payload["fullName"] = result.full_name
+            payload["full_name"] = result.full_name
         if result.error_message:
             payload["error_message"] = result.error_message
         if result.stack_trace:
@@ -344,10 +354,11 @@ class Uploader:
         if result.ended_at is not None:
             payload["ended_at"] = _epoch_ms_to_iso(result.ended_at)
 
-        # Serialize Allure steps and links into labels dict (per design field mapping)
+        # Serialize Allure steps into the first-class `steps` array (StepInput
+        # wire shape). Previously packed as a JSON string in labels["steps"] —
+        # dropped in favor of the shared TestResultInput contract.
         steps = raw_data.get("steps")
         if steps:
-            # Normalize Allure step format to match the schema _parse_steps expects
             normalized_steps = []
             for step in steps:
                 status_details = step.get("statusDetails") or {}
@@ -359,17 +370,16 @@ class Uploader:
                     "duration_ms": (stop - start) if (start and stop) else None,
                     "error_message": status_details.get("message"),
                 })
-            labels_d = payload.get("labels", {})
-            labels_d["steps"] = json.dumps(normalized_steps)
-            payload["labels"] = labels_d
+            payload["steps"] = normalized_steps
 
+        # Links and description are label-like metadata not modeled by StepInput;
+        # keep them in labels (values are strings, matching labels: dict[str, str]).
         links = raw_data.get("links")
         if links:
             labels_d = payload.get("labels", {})
             labels_d["links"] = json.dumps(links)
             payload["labels"] = labels_d
 
-        # Add description to labels if present
         description = raw_data.get("description") or raw_data.get("descriptionHtml")
         if description:
             labels_d = payload.get("labels", {})

charisma_cli-0.2.0/src/charisma_cli/watcher.py

@@ -0,0 +1,348 @@
+"""Watchdog observer, stability debounce, and file classification for allure-results."""
+
+import logging
+import os
+import time
+from pathlib import Path
+from queue import PriorityQueue
+from threading import Lock, Timer
+
+from watchdog.events import FileSystemEvent, FileSystemEventHandler
+from watchdog.observers import Observer
+
+from charisma_cli.config import Config
+from charisma_cli.models import FileCategory, FileEvent
+
+logger = logging.getLogger(__name__)
+
+_TWO_MB = 2 * 1024 * 1024
+_STABILITY_SECONDS = 0.5
+
+
+class _SeenSet:
+    """Thread-safe set of absolute paths already enqueued.
+
+    Shared by every producer (live observer, flush_pending, final_scan,
+    settle_scan) so each file is enqueued exactly once. ``add`` returns True
+    only the first time a path is seen, letting callers gate the enqueue.
+    """
+
+    def __init__(self) -> None:
+        self._paths: set[str] = set()
+        self._lock = Lock()
+
+    def add(self, path: str) -> bool:
+        """Record a path. Returns True if newly added, False if already present."""
+        with self._lock:
+            if path in self._paths:
+                return False
+            self._paths.add(path)
+            return True
+
+    def __contains__(self, path: str) -> bool:
+        with self._lock:
+            return path in self._paths
+
+
+def classify_file(filename: str) -> FileCategory | None:
+    """Classify a filename into a FileCategory or None if excluded.
+
+    Args:
+        filename: The base filename (not full path).
+
+    Returns:
+        FileCategory.RESULT for *-result.json,
+        FileCategory.CONTAINER for *-container.json,
+        FileCategory.ATTACHMENT for all other valid files,
+        None for excluded files (dot-prefix or .tmp suffix).
+    """
+    if filename.startswith("."):
+        return None
+    if filename.endswith(".tmp"):
+        return None
+    if filename.endswith("-result.json"):
+        return FileCategory.RESULT
+    if filename.endswith("-container.json"):
+        return FileCategory.CONTAINER
+    return FileCategory.ATTACHMENT
+
+
+def should_skip(path: Path, config: Config) -> bool:
+    """Check whether a file should be skipped based on size.
+
+    Args:
+        path: Path to the file on disk.
+        config: Resolved CLI configuration.
+
+    Returns:
+        True if the file exceeds 2MB and skip_too_big is enabled.
+    """
+    if not config.skip_too_big:
+        return False
+    return path.stat().st_size > _TWO_MB
+
+
+def _enqueue_once(
+    queue: "PriorityQueue[FileEvent]",
+    config: Config,
+    enqueued: "_SeenSet",
+    path: Path,
+    category: FileCategory,
+) -> bool:
+    """Enqueue a file exactly once, guarding all filesystem access.
+
+    Single owner of the enqueue-decision pipeline shared by every producer
+    (the live observer's debounce handler, flush_pending, and the scans):
+    resolve the path, apply the size-skip guard, gate on the shared seen-set,
+    then put the event. Keeping this in one place ensures the exactly-once
+    invariant (skip BEFORE marking seen, one canonical resolved-path key)
+    stays consistent across all callers.
+
+    All filesystem calls (``resolve``, ``stat`` via ``should_skip``) are guarded:
+    the scans run on the main thread during teardown, so an unhandled OSError
+    here would abort launch-close and the summary. On any filesystem error the
+    file is skipped rather than raised.
+
+    Returns:
+        True if a FileEvent was enqueued, False if skipped (too big, already
+        seen, or unreadable).
+    """
+    try:
+        if should_skip(path, config):
+            return False
+        abs_path = str(path.resolve())
+    except (OSError, ValueError):
+        # File vanished, permission/symlink error, or unresolvable path —
+        # skip it rather than crash the caller (main-thread teardown).
+        return False
+
+    if not enqueued.add(abs_path):
+        return False
+    queue.put(FileEvent(path=path, category=category))
+    return True
+
+
+class _StabilityHandler(FileSystemEventHandler):
+    """Debounces filesystem events, enqueuing files after 500ms of stability."""
+
+    def __init__(self, queue: "PriorityQueue[FileEvent]", config: Config, enqueued: "_SeenSet") -> None:
+        super().__init__()
+        self._queue = queue
+        self._config = config
+        self._timers: dict[str, Timer] = {}
+        # Shared across the live observer, flush_pending, and the scans so a
+        # given file is enqueued exactly once no matter which producer sees it
+        # first. Without this, the end-of-run settle_scan re-enqueues files the
+        # live observer already delivered, doubling results_sent.
+        self._enqueued = enqueued
+
+    def on_created(self, event: FileSystemEvent) -> None:
+        """Handle file creation events."""
+        if not event.is_directory:
+            self._handle_event(event.src_path)
+
+    def on_modified(self, event: FileSystemEvent) -> None:
+        """Handle file modification events."""
+        if not event.is_directory:
+            self._handle_event(event.src_path)
+
+    def _handle_event(self, src_path: str) -> None:
+        """Reset the stability timer for a file path."""
+        path = Path(src_path)
+        filename = path.name
+
+        category = classify_file(filename)
+        if category is None:
+            return
+
+        # Cancel existing timer for this path
+        existing = self._timers.get(src_path)
+        if existing is not None:
+            existing.cancel()
+
+        # Set a new timer that enqueues after stability period
+        timer = Timer(_STABILITY_SECONDS, self._enqueue, args=(path, category))
+        timer.daemon = True
+        self._timers[src_path] = timer
+        timer.start()
+
+    def _enqueue(self, path: Path, category: FileCategory) -> None:
+        """Enqueue a file event after stability check passes."""
+        # Remove from timer dict
+        self._timers.pop(str(path), None)
+        _enqueue_once(self._queue, self._config, self._enqueued, path, category)
+
+    def cancel_all(self) -> None:
+        """Cancel all pending timers."""
+        for timer in self._timers.values():
+            timer.cancel()
+        self._timers.clear()
+
+    def flush_pending(self) -> None:
+        """Immediately enqueue all files currently in debounce windows.
+
+        Called after subprocess exit to ensure no result files are lost due
+        to pending debounce timers. Cancels all timers and enqueues their
+        associated files directly.
+        """
+        # Snapshot and clear timers atomically
+        pending = dict(self._timers)
+        self._timers.clear()
+
+        for src_path, timer in pending.items():
+            timer.cancel()
+            path = Path(src_path)
+
+            category = classify_file(path.name)
+            if category is None:
+                continue
+
+            _enqueue_once(self._queue, self._config, self._enqueued, path, category)
+
+
+class ResultsWatcher:
+    """Watches the allure-results directory for new files using watchdog.
+
+    Creates the directory if it doesn't exist, observes file creation/modification,
+    debounces events for 500ms of stability, then enqueues FileEvent objects
+    into the provided PriorityQueue.
+    """
+
+    def __init__(self, results_dir: str, queue: "PriorityQueue[FileEvent]", config: Config) -> None:
+        self._results_dir = results_dir
+        self._queue = queue
+        self._config = config
+        # One shared seen-set gates every producer (live observer, flush_pending,
+        # final_scan, settle_scan) so each file is enqueued exactly once.
+        self._enqueued = _SeenSet()
+        self._handler = _StabilityHandler(queue, config, self._enqueued)
+        self._observer = Observer()
+
+    def start(self) -> None:
+        """Create the results directory and start observing for file events."""
+        os.makedirs(self._results_dir, exist_ok=True)
+        self._observer.schedule(self._handler, self._results_dir, recursive=False)
+        self._observer.start()
+
+    def stop(self) -> None:
+        """Stop the observer and cancel all pending stability timers."""
+        self._observer.stop()
+        self._observer.join()
+        self._handler.cancel_all()
+
+    def flush_pending(self) -> None:
+        """Flush all files currently in debounce windows into the queue.
+
+        Should be called BEFORE stop() when you want to capture all pending
+        files rather than discard them.
+        """
+        self._handler.flush_pending()
+
+    def final_scan(self) -> int:
+        """Scan the results directory and enqueue any files not already processed.
+
+        Performs a one-time sweep of all files in the results directory,
+        enqueuing any that pass classification and size checks. This catches
+        files that the filesystem watcher may have missed entirely (e.g., written
+        between watcher setup and observation start, or during high-throughput
+        bursts that overwhelm OS event buffers). Files already enqueued by any
+        producer are skipped via the shared seen-set.
+
+        Runs on the main thread during teardown, so all filesystem access is
+        guarded: a transient error listing or reading a file skips that file
+        rather than aborting the scan (and the launch-close that follows it).
+
+        Returns:
+            Number of new files enqueued by this scan.
+        """
+        results_path = Path(self._results_dir)
+        try:
+            entries = list(results_path.iterdir())
+        except (OSError, FileNotFoundError):
+            # Directory missing or unreadable — nothing to enqueue this pass.
+            return 0
+
+        enqueued = 0
+        for filepath in entries:
+            try:
+                if not filepath.is_file():
+                    continue
+            except OSError:
+                continue
+
+            category = classify_file(filepath.name)
+            if category is None:
+                continue
+
+            if _enqueue_once(self._queue, self._config, self._enqueued, filepath, category):
+                enqueued += 1
+
+        return enqueued
+
+    def settle_scan(
+        self,
+        quiet_period: float = 1.0,
+        max_wait: float = 15.0,
+        poll_interval: float = 0.25,
+    ) -> int:
+        """Repeatedly scan until the results directory stops producing new files.
+
+        Under high parallelism (e.g. pytest-xdist with many workers), allure
+        writes its result files in a burst during the test process's teardown.
+        A single ``final_scan()`` taken the instant the subprocess exits can run
+        before the OS has finished flushing that burst to disk, capturing zero
+        or a partial set — the ``watch results_sent=0`` bug. This method keeps
+        scanning until no new files have appeared for ``quiet_period`` seconds
+        (the directory has "settled"), or until ``max_wait`` seconds elapse.
+
+        Idempotent with ``final_scan()``: both share the watcher's internal
+        ``_enqueued`` set, so files already enqueued are never enqueued again.
+
+        Args:
+            quiet_period: Seconds with no new files before the directory is
+                considered settled.
+            max_wait: Absolute cap on total wait time, so a directory that keeps
+                changing can't block teardown forever.
+            poll_interval: Delay between scans.
+
+        Returns:
+            Total number of new files enqueued across all iterations.
+        """
+        total = 0
+        started_at = time.monotonic()
+        deadline = started_at + max_wait
+        last_new_at = started_at
+        settled = False
+
+        while time.monotonic() < deadline:
+            new_count = self.final_scan()
+            total += new_count
+            now = time.monotonic()
+            if new_count > 0:
+                last_new_at = now
+            elif now - last_new_at >= quiet_period:
+                # No new files for a full quiet period — directory has settled.
+                settled = True
+                break
+            time.sleep(poll_interval)
+
+        elapsed = time.monotonic() - started_at
+        if settled:
+            logger.info(
+                "Results directory settled after %.1fs (%d file(s) captured during settle)",
+                elapsed,
+                total,
+            )
+        else:
+            # Exited on max_wait, not quiescence: the directory was still
+            # producing files when we gave up. Surface this so an operator can
+            # see teardown burned the full window rather than settling — a
+            # signal that max_wait may be too short for this workload.
+            logger.warning(
+                "settle_scan hit max_wait (%.0fs) before the results directory "
+                "settled; captured %d file(s). Some late results may be unsent.",
+                max_wait,
+                total,
+            )
+
+        return total

{charisma_cli-0.1.5 → charisma_cli-0.2.0}/tests/test_integration.py

@@ -2,6 +2,7 @@
 
 import json
 import sys
+import threading
 from pathlib import Path
 
 import httpx
@@ -228,10 +229,10 @@ class TestReportingFixesEndToEnd:
         # The skipped result was sent (not dropped)
         assert results_route.called
         sent_results = json.loads(results_route.calls[0].request.content)["results"]
-        outcomes = {r["testId"]: r["outcome"] for r in sent_results}
+        outcomes = {r["test_id"]: r["outcome"] for r in sent_results}
         assert outcomes.get("hist_skip") == "skipped"
         # duration defaulted to 0 for the missing stop
-        skip_payload = next(r for r in sent_results if r["testId"] == "hist_skip")
+        skip_payload = next(r for r in sent_results if r["test_id"] == "hist_skip")
         assert skip_payload["duration_ms"] == 0
 
         # environment.properties sent as variables on close
@@ -242,3 +243,132 @@ class TestReportingFixesEndToEnd:
             "aws_region": "us-west-2",
             "workers_number": "8",
         }
+
+
+class TestSettleScanCapturesLateTeardownBurst:
+    """Regression for the `watch results_sent=0` bug — isolates the true fix.
+
+    The one behavioral property that separates the old teardown from the new is
+    settle_scan's max_wait window. allure-pytest under heavy xdist can flush its
+    result files to disk SEVERAL SECONDS after the subprocess exits. The old
+    teardown used settle_scan()'s default max_wait=15.0; the new watch teardown
+    widens it to 60.0. A file that first lands on disk AFTER the window closes is
+    never enqueued and never sent — that is the production failure.
+
+    This test drives ResultsWatcher directly (not the full CLI) with scaled-down
+    timings so it is fast and deterministic: a result file is written by a
+    background timer at t=0.3s after settle_scan starts. With a SHORT window
+    (mimicking the old code) settle_scan gives up before the file lands and
+    enqueues nothing. With a LONGER window (mimicking the new code) it is still
+    polling when the file lands and enqueues it. No reliance on OS event buffers
+    or thread scheduling — only on-disk appearance time vs the window.
+    """
+
+    def _write_result_after(self, results_dir: Path, delay: float) -> threading.Timer:
+        """Schedule a valid Allure result file to appear after `delay` seconds."""
+
+        def _write() -> None:
+            (results_dir / "late-result.json").write_text(json.dumps({
+                "uuid": "late-1",
+                "historyId": "hist_late",
+                "fullName": "tests.test_late.test_case",
+                "name": "test_case",
+                "status": "passed",
+                "start": 1000,
+                "stop": 2000,
+                "labels": [],
+                "parameters": [],
+                "attachments": [],
+            }))
+
+        timer = threading.Timer(delay, _write)
+        timer.daemon = True
+        timer.start()
+        return timer
+
+    def test_short_window_misses_late_file_long_window_captures_it(
+        self, tmp_path: Path, default_config
+    ) -> None:
+        """settle_scan with a window shorter than the file's appearance misses it;
+        a window longer than the appearance captures it. This is exactly the
+        old (max_wait=15) vs new (max_wait=60) separation, scaled to sub-second."""
+        from queue import PriorityQueue
+
+        from charisma_cli.models import FileEvent
+        from charisma_cli.watcher import ResultsWatcher
+
+        # --- OLD behavior: window closes BEFORE the file lands (t=0.3s) ---
+        old_dir = tmp_path / "old"
+        old_dir.mkdir()
+        old_queue: PriorityQueue[FileEvent] = PriorityQueue()
+        old_watcher = ResultsWatcher(str(old_dir), old_queue, default_config)
+        t1 = self._write_result_after(old_dir, delay=0.3)
+        # Do not start the observer — isolate the scan behavior from OS events.
+        old_captured = old_watcher.settle_scan(quiet_period=0.1, max_wait=0.2)
+        t1.join()
+        assert old_captured == 0, "short window should give up before the late file lands"
+        assert old_queue.empty()
+
+        # --- NEW behavior: window stays open PAST the file landing (t=0.3s) ---
+        new_dir = tmp_path / "new"
+        new_dir.mkdir()
+        new_queue: PriorityQueue[FileEvent] = PriorityQueue()
+        new_watcher = ResultsWatcher(str(new_dir), new_queue, default_config)
+        t2 = self._write_result_after(new_dir, delay=0.3)
+        new_captured = new_watcher.settle_scan(quiet_period=0.3, max_wait=2.0)
+        t2.join()
+        assert new_captured == 1, "long window should still be polling when the late file lands"
+        assert new_queue.qsize() == 1
+
+
+class TestWatchStreamsPreexistingResult:
+    """Sanity E2E: watch streams a result already on disk through the launch API.
+
+    This does NOT isolate the old/new fix (any on-disk file is caught by both).
+    It guards the happy path: watch opens a launch, sends the result, closes,
+    and reports results_sent in the summary.
+    """
+
+    @respx.mock
+    def test_watch_sends_result_and_reports_summary(self, tmp_path: Path, monkeypatch) -> None:
+        results_dir = tmp_path / "allure-results"
+        results_dir.mkdir()
+        (results_dir / "pre-result.json").write_text(json.dumps({
+            "uuid": "pre-1",
+            "historyId": "hist_pre",
+            "fullName": "tests.test_pre.test_case",
+            "name": "test_case",
+            "status": "passed",
+            "start": 1000,
+            "stop": 2000,
+            "labels": [],
+            "parameters": [],
+            "attachments": [],
+        }))
+
+        respx.post("https://charisma.test/api/v1/launches").mock(
+            return_value=httpx.Response(201, json={"launchId": "launch-pre", "projectId": "proj-uuid"})
+        )
+        results_route = respx.post(
+            "https://charisma.test/api/v1/launches/launch-pre/results"
+        ).mock(return_value=httpx.Response(200, json={"accepted": 1}))
+        respx.post("https://charisma.test/api/v1/launches/launch-pre/close").mock(
+            return_value=httpx.Response(200, json={})
+        )
+
+        monkeypatch.setenv("CHARISMA_ENDPOINT", "https://charisma.test")
+        monkeypatch.setenv("CHARISMA_TOKEN", "pre-token")
+
+        runner = CliRunner()
+        result = runner.invoke(cli, [
+            "watch",
+            "--project", "pre-project",
+            "--results", str(results_dir),
+            "--", sys.executable, "-c", "import sys; sys.exit(0)",
+        ])
+
+        assert result.exit_code == 0
+        assert results_route.called
+        sent_results = json.loads(results_route.calls[0].request.content)["results"]
+        assert any(r["test_id"] == "hist_pre" for r in sent_results)
+        assert "results_sent=1" in result.output

{charisma_cli-0.1.5 → charisma_cli-0.2.0}/tests/test_watcher.py

@@ -520,8 +520,12 @@ class TestFinalScan:
         event = queue.get_nowait()
         assert event.path.name == "valid-result.json"
 
-    def test_skips_already_seen_files(self, tmp_path: Path) -> None:
-        """final_scan skips files in the already_seen set."""
+    def test_skips_files_already_enqueued(self, tmp_path: Path) -> None:
+        """final_scan skips files already enqueued by a prior producer.
+
+        Dedup is via the watcher's shared seen-set: a file the live observer
+        already claimed is not re-enqueued by a subsequent scan.
+        """
         results_dir = tmp_path / "allure-results"
         results_dir.mkdir()
 
@@ -534,8 +538,11 @@ class TestFinalScan:
         config = _make_config()
         watcher = ResultsWatcher(str(results_dir), queue, config)
 
-        already_seen = {str(file_a.resolve())}
-        count = watcher.final_scan(already_seen=already_seen)
+        # Simulate file_a already enqueued via the live-observer path.
+        watcher._handler._enqueue(file_a, FileCategory.RESULT)
+        queue.get_nowait()  # drain the observer's enqueue
+
+        count = watcher.final_scan()
 
         assert count == 1
         event = queue.get_nowait()
@@ -569,3 +576,249 @@ class TestFinalScan:
 
         assert count == 0
         assert queue.empty()
+
+
+# ---------------------------------------------------------------------------
+# settle_scan() tests — the fix for the `watch results_sent=0` bug under
+# high pytest-xdist parallelism (burst of result files written at teardown).
+# ---------------------------------------------------------------------------
+
+
+class TestSettleScan:
+    """settle_scan rescans until the results directory stops changing.
+
+    A single final_scan() taken the instant the test subprocess exits can run
+    before the OS finishes flushing allure's end-of-session burst, capturing
+    nothing (the results_sent=0 bug). settle_scan keeps scanning until the
+    directory is quiet, so the whole burst is captured regardless of how many
+    files or workers produced it.
+    """
+
+    def test_captures_files_present_before_scan(self, tmp_path: Path) -> None:
+        """Files already on disk are enqueued by the first settle_scan pass."""
+        results_dir = tmp_path / "allure-results"
+        results_dir.mkdir()
+        for i in range(5):
+            (results_dir / f"r{i}-result.json").write_text("{}")
+
+        queue: PriorityQueue = PriorityQueue()
+        watcher = ResultsWatcher(str(results_dir), queue, _make_config())
+
+        enqueued = watcher.settle_scan(quiet_period=0.3, max_wait=3.0, poll_interval=0.1)
+
+        assert enqueued == 5
+        assert queue.qsize() == 5
+
+    def test_captures_single_result_flushed_after_first_scan(self, tmp_path: Path) -> None:
+        """Single-test case: one result file flushed AFTER the initial scan.
+
+        Reproduces the observed 1-test / 1-worker `results_sent=0` failure —
+        allure-pytest flushes its one result file during process teardown, which
+        surfaces to the filesystem just after the subprocess exits. A single
+        final_scan taken at that instant captures 0; settle_scan waits and
+        captures the 1 late file. This is the N=1 boundary of the burst race.
+        """
+        import threading
+
+        results_dir = tmp_path / "allure-results"
+        results_dir.mkdir()
+
+        queue: PriorityQueue = PriorityQueue()
+        watcher = ResultsWatcher(str(results_dir), queue, _make_config())
+
+        def write_single_late() -> None:
+            time.sleep(0.3)
+            (results_dir / "only-result.json").write_text("{}")
+
+        writer = threading.Thread(target=write_single_late)
+        writer.start()
+
+        # Directory is empty when the first scan runs (the old bug: captures 0).
+        first = watcher.final_scan()
+        assert first == 0
+
+        settled = watcher.settle_scan(quiet_period=0.5, max_wait=10.0, poll_interval=0.1)
+        writer.join()
+
+        assert first + settled == 1
+        assert queue.qsize() == 1
+
+    def test_captures_burst_written_after_first_scan(self, tmp_path: Path) -> None:
+        """The core fix: files that appear AFTER the initial scan are still captured.
+
+        Reproduces the teardown-burst race — a single final_scan would have
+        captured only the one baseline file; settle_scan waits for the rest.
+        """
+        import threading
+
+        results_dir = tmp_path / "allure-results"
+        results_dir.mkdir()
+        (results_dir / "baseline-result.json").write_text("{}")
+
+        queue: PriorityQueue = PriorityQueue()
+        watcher = ResultsWatcher(str(results_dir), queue, _make_config())
+
+        def write_burst() -> None:
+            time.sleep(0.2)
+            for i in range(30):
+                (results_dir / f"burst{i}-result.json").write_text("{}")
+                time.sleep(0.02)
+
+        writer = threading.Thread(target=write_burst)
+        writer.start()
+
+        # A single scan would see only the baseline file.
+        first = watcher.final_scan()
+        assert first == 1
+
+        # settle_scan reconciles the burst that lands afterward.
+        settled = watcher.settle_scan(quiet_period=0.5, max_wait=10.0, poll_interval=0.1)
+        writer.join()
+
+        assert first + settled == 31
+        assert queue.qsize() == 31
+
+    def test_no_duplicate_enqueue_across_scans(self, tmp_path: Path) -> None:
+        """Repeated scans never enqueue the same file twice (shared seen-set)."""
+        results_dir = tmp_path / "allure-results"
+        results_dir.mkdir()
+        for i in range(4):
+            (results_dir / f"r{i}-result.json").write_text("{}")
+
+        queue: PriorityQueue = PriorityQueue()
+        watcher = ResultsWatcher(str(results_dir), queue, _make_config())
+
+        first = watcher.settle_scan(quiet_period=0.3, max_wait=3.0, poll_interval=0.1)
+        second = watcher.settle_scan(quiet_period=0.3, max_wait=3.0, poll_interval=0.1)
+
+        assert first == 4
+        assert second == 0  # nothing new — no double-enqueue
+        assert queue.qsize() == 4
+
+    def test_observer_enqueue_and_settle_scan_dedup(self, tmp_path: Path) -> None:
+        """Files the live observer already enqueued are not re-enqueued by settle_scan.
+
+        The live-observer path (_StabilityHandler._enqueue) and settle_scan share
+        the watcher's seen-set. Without this, settle_scan would double-count files
+        the observer already delivered — the results_sent doubling regression.
+        """
+        results_dir = tmp_path / "allure-results"
+        results_dir.mkdir()
+        files = []
+        for i in range(10):
+            f = results_dir / f"r{i}-result.json"
+            f.write_text("{}")
+            files.append(f)
+
+        queue: PriorityQueue = PriorityQueue()
+        watcher = ResultsWatcher(str(results_dir), queue, _make_config())
+
+        # Simulate the live observer having already enqueued the first 4 files
+        # by driving the handler's enqueue path directly.
+        for f in files[:4]:
+            watcher._handler._enqueue(f, FileCategory.RESULT)
+        assert queue.qsize() == 4
+
+        # settle_scan should add only the remaining 6, not re-add the first 4.
+        added = watcher.settle_scan(quiet_period=0.3, max_wait=3.0, poll_interval=0.1)
+
+        assert added == 6
+        assert queue.qsize() == 10  # 10 unique, zero duplicates
+
+        seen: set[str] = set()
+        while not queue.empty():
+            ev: FileEvent = queue.get_nowait()
+            key = str(ev.path.resolve())
+            assert key not in seen, "duplicate file enqueued"
+            seen.add(key)
+        assert len(seen) == 10
+
+    def test_settles_quickly_when_directory_empty(self, tmp_path: Path, caplog) -> None:
+        """An empty directory settles after one quiet period and enqueues nothing.
+
+        Also asserts the settle path logs at INFO (not WARNING) — settling
+        cleanly is the normal case and must not raise a max_wait warning.
+        """
+        import logging
+
+        results_dir = tmp_path / "allure-results"
+        results_dir.mkdir()
+
+        queue: PriorityQueue = PriorityQueue()
+        watcher = ResultsWatcher(str(results_dir), queue, _make_config())
+
+        start = time.monotonic()
+        with caplog.at_level(logging.INFO, logger="charisma_cli.watcher"):
+            enqueued = watcher.settle_scan(quiet_period=0.3, max_wait=5.0, poll_interval=0.1)
+        elapsed = time.monotonic() - start
+
+        assert enqueued == 0
+        assert queue.empty()
+        assert elapsed < 4.0  # returned on the quiet-period, not the max_wait cap
+        # Clean settle → INFO log, never a max_wait WARNING.
+        assert any(r.levelno == logging.INFO for r in caplog.records)
+        assert not any(r.levelno == logging.WARNING for r in caplog.records)
+
+    def test_respects_max_wait_when_directory_never_settles(self, tmp_path: Path, caplog) -> None:
+        """max_wait caps total time even if files keep arriving continuously.
+
+        This documents the one bounded caveat: a directory that never goes quiet
+        stops at max_wait rather than blocking teardown forever. Also asserts a
+        WARNING is emitted so the up-to-max_wait teardown is observable rather
+        than silent.
+        """
+        import logging
+        import threading
+
+        results_dir = tmp_path / "allure-results"
+        results_dir.mkdir()
+
+        queue: PriorityQueue = PriorityQueue()
+        watcher = ResultsWatcher(str(results_dir), queue, _make_config())
+
+        stop = threading.Event()
+
+        def write_forever() -> None:
+            i = 0
+            while not stop.is_set():
+                (results_dir / f"c{i}-result.json").write_text("{}")
+                i += 1
+                time.sleep(0.05)
+
+        writer = threading.Thread(target=write_forever)
+        writer.start()
+        try:
+            start = time.monotonic()
+            with caplog.at_level(logging.WARNING, logger="charisma_cli.watcher"):
+                watcher.settle_scan(quiet_period=0.5, max_wait=1.5, poll_interval=0.1)
+            elapsed = time.monotonic() - start
+        finally:
+            stop.set()
+            writer.join()
+
+        # Should stop at ~max_wait, not run indefinitely.
+        assert 1.5 <= elapsed < 4.0
+        # Hitting max_wait must surface a WARNING (observability, not silent).
+        assert any(
+            r.levelno == logging.WARNING and "max_wait" in r.getMessage()
+            for r in caplog.records
+        )
+
+    def test_scales_to_large_burst(self, tmp_path: Path) -> None:
+        """A large number of files (simulating many workers/tests) is fully captured.
+
+        Correctness is independent of file count — settle_scan keeps scanning
+        until quiet, so 500 files enqueue exactly once each.
+        """
+        results_dir = tmp_path / "allure-results"
+        results_dir.mkdir()
+        for i in range(500):
+            (results_dir / f"r{i}-result.json").write_text("{}")
+
+        queue: PriorityQueue = PriorityQueue()
+        watcher = ResultsWatcher(str(results_dir), queue, _make_config())
+
+        enqueued = watcher.settle_scan(quiet_period=0.3, max_wait=10.0, poll_interval=0.1)
+
+        assert enqueued == 500
+        assert queue.qsize() == 500

charisma_cli-0.1.5/src/charisma_cli/watcher.py

@@ -1,223 +0,0 @@
-"""Watchdog observer, stability debounce, and file classification for allure-results."""
-
-import os
-from pathlib import Path
-from queue import PriorityQueue
-from threading import Timer
-
-from watchdog.events import FileSystemEvent, FileSystemEventHandler
-from watchdog.observers import Observer
-
-from charisma_cli.config import Config
-from charisma_cli.models import FileCategory, FileEvent
-
-_TWO_MB = 2 * 1024 * 1024
-_STABILITY_SECONDS = 0.5
-
-
-def classify_file(filename: str) -> FileCategory | None:
-    """Classify a filename into a FileCategory or None if excluded.
-
-    Args:
-        filename: The base filename (not full path).
-
-    Returns:
-        FileCategory.RESULT for *-result.json,
-        FileCategory.CONTAINER for *-container.json,
-        FileCategory.ATTACHMENT for all other valid files,
-        None for excluded files (dot-prefix or .tmp suffix).
-    """
-    if filename.startswith("."):
-        return None
-    if filename.endswith(".tmp"):
-        return None
-    if filename.endswith("-result.json"):
-        return FileCategory.RESULT
-    if filename.endswith("-container.json"):
-        return FileCategory.CONTAINER
-    return FileCategory.ATTACHMENT
-
-
-def should_skip(path: Path, config: Config) -> bool:
-    """Check whether a file should be skipped based on size.
-
-    Args:
-        path: Path to the file on disk.
-        config: Resolved CLI configuration.
-
-    Returns:
-        True if the file exceeds 2MB and skip_too_big is enabled.
-    """
-    if not config.skip_too_big:
-        return False
-    return path.stat().st_size > _TWO_MB
-
-
-class _StabilityHandler(FileSystemEventHandler):
-    """Debounces filesystem events, enqueuing files after 500ms of stability."""
-
-    def __init__(self, queue: PriorityQueue, config: Config) -> None:
-        super().__init__()
-        self._queue = queue
-        self._config = config
-        self._timers: dict[str, Timer] = {}
-
-    def on_created(self, event: FileSystemEvent) -> None:
-        """Handle file creation events."""
-        if not event.is_directory:
-            self._handle_event(event.src_path)
-
-    def on_modified(self, event: FileSystemEvent) -> None:
-        """Handle file modification events."""
-        if not event.is_directory:
-            self._handle_event(event.src_path)
-
-    def _handle_event(self, src_path: str) -> None:
-        """Reset the stability timer for a file path."""
-        path = Path(src_path)
-        filename = path.name
-
-        category = classify_file(filename)
-        if category is None:
-            return
-
-        # Cancel existing timer for this path
-        existing = self._timers.get(src_path)
-        if existing is not None:
-            existing.cancel()
-
-        # Set a new timer that enqueues after stability period
-        timer = Timer(_STABILITY_SECONDS, self._enqueue, args=(path, category))
-        timer.daemon = True
-        self._timers[src_path] = timer
-        timer.start()
-
-    def _enqueue(self, path: Path, category: FileCategory) -> None:
-        """Enqueue a file event after stability check passes."""
-        # Remove from timer dict
-        self._timers.pop(str(path), None)
-
-        # Check file still exists and size at enqueue time
-        try:
-            if should_skip(path, self._config):
-                return
-        except (OSError, FileNotFoundError):
-            return  # File was deleted between debounce and enqueue
-
-        self._queue.put(FileEvent(path=path, category=category))
-
-    def cancel_all(self) -> None:
-        """Cancel all pending timers."""
-        for timer in self._timers.values():
-            timer.cancel()
-        self._timers.clear()
-
-    def flush_pending(self) -> None:
-        """Immediately enqueue all files currently in debounce windows.
-
-        Called after subprocess exit to ensure no result files are lost due
-        to pending debounce timers. Cancels all timers and enqueues their
-        associated files directly.
-        """
-        # Snapshot and clear timers atomically
-        pending = dict(self._timers)
-        self._timers.clear()
-
-        for src_path, timer in pending.items():
-            timer.cancel()
-            path = Path(src_path)
-            filename = path.name
-
-            category = classify_file(filename)
-            if category is None:
-                continue
-
-            try:
-                if should_skip(path, self._config):
-                    continue
-            except (OSError, FileNotFoundError):
-                continue
-
-            self._queue.put(FileEvent(path=path, category=category))
-
-
-class ResultsWatcher:
-    """Watches the allure-results directory for new files using watchdog.
-
-    Creates the directory if it doesn't exist, observes file creation/modification,
-    debounces events for 500ms of stability, then enqueues FileEvent objects
-    into the provided PriorityQueue.
-    """
-
-    def __init__(self, results_dir: str, queue: PriorityQueue, config: Config) -> None:
-        self._results_dir = results_dir
-        self._queue = queue
-        self._config = config
-        self._handler = _StabilityHandler(queue, config)
-        self._observer = Observer()
-
-    def start(self) -> None:
-        """Create the results directory and start observing for file events."""
-        os.makedirs(self._results_dir, exist_ok=True)
-        self._observer.schedule(self._handler, self._results_dir, recursive=False)
-        self._observer.start()
-
-    def stop(self) -> None:
-        """Stop the observer and cancel all pending stability timers."""
-        self._observer.stop()
-        self._observer.join()
-        self._handler.cancel_all()
-
-    def flush_pending(self) -> None:
-        """Flush all files currently in debounce windows into the queue.
-
-        Should be called BEFORE stop() when you want to capture all pending
-        files rather than discard them.
-        """
-        self._handler.flush_pending()
-
-    def final_scan(self, already_seen: set[str] | None = None) -> int:
-        """Scan the results directory and enqueue any files not already processed.
-
-        Performs a one-time sweep of all files in the results directory,
-        enqueuing any that pass classification and size checks. This catches
-        files that the filesystem watcher may have missed entirely (e.g., written
-        between watcher setup and observation start, or during high-throughput
-        bursts that overwhelm OS event buffers).
-
-        Args:
-            already_seen: Optional set of absolute path strings that were
-                already enqueued. Files in this set are skipped.
-
-        Returns:
-            Number of new files enqueued by this scan.
-        """
-        results_path = Path(self._results_dir)
-        if not results_path.exists():
-            return 0
-
-        enqueued = 0
-        seen = already_seen or set()
-
-        for filepath in results_path.iterdir():
-            if not filepath.is_file():
-                continue
-
-            abs_path = str(filepath.resolve())
-            if abs_path in seen:
-                continue
-
-            category = classify_file(filepath.name)
-            if category is None:
-                continue
-
-            try:
-                if should_skip(filepath, self._config):
-                    continue
-            except (OSError, FileNotFoundError):
-                continue
-
-            self._queue.put(FileEvent(path=filepath, category=category))
-            enqueued += 1
-
-        return enqueued