Windows support (#276)

* Supporting for Windows

* Supporting for Windows

* Supporting for Windows

* Upload coverage report

* Upload coverage report

* up

* No need for error details

* Revert "No need for error details"

This reverts commit a6b9f6e.

* Restore original __all__

* Use platform.system(), remove Python version check

* Make black happy

* Black & typing adjustments

* _WindowsAdapter class

* Remove test markers

* Decorator to adapt tests for Windows

* Move _WindowsAdapter to _utils module

* Adapt all tests for Windows

* Update readme about Windows

* Placeholder changelog entry for upcoming release

* Rename coverage report CI step

* Add pull request id to changelog

* CI: add CODECOV_TOKEN to env (Windows)

* Run twisted test on Windows too

* Readme adjustments

* Remove unused check for Deferred

* asyncio reactor is not a requirement on Windows

---------

Co-authored-by: sanzenwin <[email protected]>

Author: elacuesta

.github/workflows/tests.yml

@@ -13,6 +13,8 @@ jobs:
         include:
           - os: macos-latest
             python-version: "3.12"
+          - os: windows-latest
+            python-version: "3.12"
 
     steps:
     - uses: actions/checkout@v4
@@ -48,3 +50,12 @@ jobs:
         curl -Os https://uploader.codecov.io/latest/macos/codecov
         chmod +x codecov
         ./codecov
+
+    - name: Upload coverage report (Windows)
+      if: runner.os == 'Windows'
+      env:
+        CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
+      run: |
+        $ProgressPreference = 'SilentlyContinue'
+        Invoke-WebRequest -Uri https://uploader.codecov.io/latest/windows/codecov.exe -Outfile codecov.exe
+        .\codecov.exe

README.md

@@ -56,10 +56,13 @@ See the [changelog](docs/changelog.md) document.
 
 ## Activation
 
+### Download handler
+
 Replace the default `http` and/or `https` Download Handlers through
 [`DOWNLOAD_HANDLERS`](https://docs.scrapy.org/en/latest/topics/settings.html):
 
 ```python
+# settings.py
 DOWNLOAD_HANDLERS = {
     "http": "scrapy_playwright.handler.ScrapyPlaywrightDownloadHandler",
     "https": "scrapy_playwright.handler.ScrapyPlaywrightDownloadHandler",
@@ -70,12 +73,19 @@ Note that the `ScrapyPlaywrightDownloadHandler` class inherits from the default
 `http/https` handler. Unless explicitly marked (see [Basic usage](#basic-usage)),
 requests will be processed by the regular Scrapy download handler.
 
-Also, be sure to [install the `asyncio`-based Twisted reactor](https://docs.scrapy.org/en/latest/topics/asyncio.html#installing-the-asyncio-reactor):
+
+### Twisted reactor
+
+When running on GNU/Linux or macOS you'll need to
+[install the `asyncio`-based Twisted reactor](https://docs.scrapy.org/en/latest/topics/asyncio.html#installing-the-asyncio-reactor):
 
 ```python
+# settings.py
 TWISTED_REACTOR = "twisted.internet.asyncioreactor.AsyncioSelectorReactor"
 ```
 
+This is not a requirement on Windows (see [Windows support](#windows-support))
+
 
 ## Basic usage
 
@@ -112,6 +122,20 @@ does not match the running Browser. If you prefer the `User-Agent` sent by
 default by the specific browser you're using, set the Scrapy user agent to `None`.
 
 
+## Windows support
+
+Windows support is possible by running Playwright in a `ProactorEventLoop` in a separate thread.
+This is necessary because it's not possible to run Playwright in the same
+asyncio event loop as the Scrapy crawler:
+* Playwright runs the driver in a subprocess. Source:
+  [Playwright repository](https://github.com/microsoft/playwright-python/blob/v1.44.0/playwright/_impl/_transport.py#L120-L130).
+* "On Windows, the default event loop `ProactorEventLoop` supports subprocesses,
+  whereas `SelectorEventLoop` does not". Source:
+  [Python docs](https://docs.python.org/3/library/asyncio-platforms.html#asyncio-windows-subprocess).
+* Twisted's `asyncio` reactor requires the `SelectorEventLoop`. Source:
+  [Twisted repository](https://github.com/twisted/twisted/blob/twisted-24.3.0/src/twisted/internet/asyncioreactor.py#L31)
+
+
 ## Supported [settings](https://docs.scrapy.org/en/latest/topics/settings.html)
 
 ### `PLAYWRIGHT_BROWSER_TYPE`
@@ -851,6 +875,12 @@ Refer to the
 [upstream docs](https://docs.scrapy.org/en/latest/topics/extensions.html#module-scrapy.extensions.memusage)
 for more information about supported settings.
 
+### Windows support
+
+Just like the [upstream Scrapy extension](https://docs.scrapy.org/en/latest/topics/extensions.html#module-scrapy.extensions.memusage), this custom memory extension does not work
+on Windows. This is because the stdlib [`resource`](https://docs.python.org/3/library/resource.html)
+module is not available.
+
 
 ## Examples
 
@@ -912,23 +942,6 @@ See the [examples](examples) directory for more.
 
 ## Known issues
 
-### Lack of native support for Windows
-
-This package does not work natively on Windows. This is because:
-
-* Playwright runs the driver in a subprocess. Source:
-[Playwright repository](https://github.com/microsoft/playwright-python/blob/v1.28.0/playwright/_impl/_transport.py#L120-L129).
-* "On Windows, the default event loop `ProactorEventLoop` supports subprocesses,
-whereas `SelectorEventLoop` does not". Source:
-[Python docs](https://docs.python.org/3/library/asyncio-platforms.html#asyncio-windows-subprocess).
-* Twisted's `asyncio` reactor requires the `SelectorEventLoop`. Source:
-[Twisted repository](https://github.com/twisted/twisted/blob/twisted-22.4.0/src/twisted/internet/asyncioreactor.py#L31).
-
-Some users have reported having success
-[running under WSL](https://github.com/scrapy-plugins/scrapy-playwright/issues/7#issuecomment-817394494).
-See also [#78](https://github.com/scrapy-plugins/scrapy-playwright/issues/78)
-for information about working in headful mode under WSL.
-
 ### No per-request proxy support
 Specifying a proxy via the `proxy` Request meta key is not supported.
 Refer to the [Proxy support](#proxy-support) section for more information.

docs/changelog.md

@@ -1,5 +1,10 @@
 # scrapy-playwright changelog
 
+### [v0.0.36](https://github.com/scrapy-plugins/scrapy-playwright/releases/tag/v0.0.36) (2024-MM-DD)
+
+* Windows support (#276)
+
+
 ### [v0.0.35](https://github.com/scrapy-plugins/scrapy-playwright/releases/tag/v0.0.35) (2024-06-01)
 
 * Update exception message check

scrapy_playwright/_utils.py

@@ -1,10 +1,15 @@
+import asyncio
+import concurrent
 import logging
+import platform
+import threading
 from typing import Awaitable, Iterator, Optional, Tuple, Union
 
+import scrapy
 from playwright.async_api import Error, Page, Request, Response
-from scrapy import Spider
 from scrapy.http.headers import Headers
 from scrapy.utils.python import to_unicode
+from twisted.internet.defer import Deferred
 from w3lib.encoding import html_body_declared_encoding, http_content_type_encoding
 
 
@@ -53,7 +58,7 @@ def _is_safe_close_error(error: Error) -> bool:
 
 async def _get_page_content(
     page: Page,
-    spider: Spider,
+    spider: scrapy.Spider,
     context_name: str,
     scrapy_request_url: str,
     scrapy_request_method: str,
@@ -89,3 +94,38 @@ async def _get_header_value(
         return await resource.header_value(header_name)
     except Exception:
         return None
+
+
+if platform.system() == "Windows":
+
+    class _WindowsAdapter:
+        """Utility class to redirect coroutines to an asyncio event loop running
+        in a different thread. This allows to use a ProactorEventLoop, which is
+        supported by Playwright on Windows.
+        """
+
+        loop = None
+        thread = None
+
+        @classmethod
+        def get_event_loop(cls) -> asyncio.AbstractEventLoop:
+            if cls.thread is None:
+                if cls.loop is None:
+                    policy = asyncio.WindowsProactorEventLoopPolicy()  # type: ignore
+                    cls.loop = policy.new_event_loop()
+                    asyncio.set_event_loop(cls.loop)
+                if not cls.loop.is_running():
+                    cls.thread = threading.Thread(target=cls.loop.run_forever, daemon=True)
+                    cls.thread.start()
+                    logger.info("Started loop on separate thread: %s", cls.loop)
+            return cls.loop
+
+        @classmethod
+        async def get_result(cls, coro) -> concurrent.futures.Future:
+            return asyncio.run_coroutine_threadsafe(coro=coro, loop=cls.get_event_loop()).result()
+
+    def _deferred_from_coro(coro) -> Deferred:
+        return scrapy.utils.defer.deferred_from_coro(_WindowsAdapter.get_result(coro))
+
+else:
+    _deferred_from_coro = scrapy.utils.defer.deferred_from_coro

scrapy_playwright/handler.py

@@ -1,5 +1,6 @@
 import asyncio
 import logging
+import platform
 from contextlib import suppress
 from dataclasses import dataclass
 from ipaddress import ip_address
@@ -25,7 +26,6 @@
 from scrapy.http.headers import Headers
 from scrapy.responsetypes import responsetypes
 from scrapy.settings import Settings
-from scrapy.utils.defer import deferred_from_coro
 from scrapy.utils.misc import load_object
 from scrapy.utils.reactor import verify_installed_reactor
 from twisted.internet.defer import Deferred, inlineCallbacks
@@ -38,6 +38,7 @@
     _get_page_content,
     _is_safe_close_error,
     _maybe_await,
+    _deferred_from_coro,
 )
 
 
@@ -101,7 +102,8 @@ class ScrapyPlaywrightDownloadHandler(HTTPDownloadHandler):
 
     def __init__(self, crawler: Crawler) -> None:
         super().__init__(settings=crawler.settings, crawler=crawler)
-        verify_installed_reactor("twisted.internet.asyncioreactor.AsyncioSelectorReactor")
+        if platform.system() != "Windows":
+            verify_installed_reactor("twisted.internet.asyncioreactor.AsyncioSelectorReactor")
         crawler.signals.connect(self._engine_started, signals.engine_started)
         self.stats = crawler.stats
 
@@ -134,7 +136,7 @@ def from_crawler(cls: Type[PlaywrightHandler], crawler: Crawler) -> PlaywrightHa
 
     def _engine_started(self) -> Deferred:
         """Launch the browser. Use the engine_started signal as it supports returning deferreds."""
-        return deferred_from_coro(self._launch())
+        return _deferred_from_coro(self._launch())
 
     async def _launch(self) -> None:
         """Launch Playwright manager and configured startup context(s)."""
@@ -290,7 +292,7 @@ def _set_max_concurrent_context_count(self):
     def close(self) -> Deferred:
         logger.info("Closing download handler")
         yield super().close()
-        yield deferred_from_coro(self._close())
+        yield _deferred_from_coro(self._close())
 
     async def _close(self) -> None:
         await asyncio.gather(*[ctx.context.close() for ctx in self.context_wrappers.values()])
@@ -305,7 +307,7 @@ async def _close(self) -> None:
 
     def download_request(self, request: Request, spider: Spider) -> Deferred:
         if request.meta.get("playwright"):
-            return deferred_from_coro(self._download_request(request, spider))
+            return _deferred_from_coro(self._download_request(request, spider))
         return super().download_request(request, spider)
 
     async def _download_request(self, request: Request, spider: Spider) -> Response:

tests/__init__.py

@@ -1,10 +1,38 @@
+import inspect
+import logging
+import platform
 from contextlib import asynccontextmanager
+from functools import wraps
 
 from scrapy import Request
 from scrapy.http.response.html import HtmlResponse
 from scrapy.utils.test import get_crawler
 
 
+logger = logging.getLogger("scrapy-playwright-tests")
+
+
+if platform.system() == "Windows":
+    from scrapy_playwright._utils import _WindowsAdapter
+
+    def allow_windows(test_method):
+        """Wrap tests with the _WindowsAdapter class on Windows."""
+        if not inspect.iscoroutinefunction(test_method):
+            raise RuntimeError(f"{test_method} must be an async def method")
+
+        @wraps(test_method)
+        async def wrapped(self, *args, **kwargs):
+            logger.debug("Calling _WindowsAdapter.get_result for %r", self)
+            await _WindowsAdapter.get_result(test_method(self, *args, **kwargs))
+
+        return wrapped
+
+else:
+
+    def allow_windows(test_method):
+        return test_method
+
+
 @asynccontextmanager
 async def make_handler(settings_dict: dict):
     """Convenience function to obtain an initialized handler and close it gracefully"""

tests/conftest.py

@@ -1,3 +1,20 @@
+import platform
+
+import pytest
+
+
+@pytest.hookimpl(tryfirst=True)
+def pytest_configure(config):
+    # https://twistedmatrix.com/trac/ticket/9766
+    # https://github.com/pytest-dev/pytest-twisted/issues/80
+
+    if config.getoption("reactor", "default") == "asyncio" and platform.system() == "Windows":
+        import asyncio
+
+        selector_policy = asyncio.WindowsSelectorEventLoopPolicy()
+        asyncio.set_event_loop_policy(selector_policy)
+
+
 def pytest_sessionstart(session):  # pylint: disable=unused-argument
     """
     Called after the Session object has been created and before performing

tests/tests_asyncio/test_browser_contexts.py

@@ -10,11 +10,12 @@
 from scrapy import Spider, Request
 from scrapy_playwright.page import PageMethod
 
-from tests import make_handler
+from tests import allow_windows, make_handler
 from tests.mockserver import StaticMockServer
 
 
 class MixinTestCaseMultipleContexts:
+    @allow_windows
     async def test_context_kwargs(self):
         settings_dict = {
             "PLAYWRIGHT_BROWSER_TYPE": self.browser_type,
@@ -37,6 +38,7 @@ async def test_context_kwargs(self):
                 with pytest.raises(PlaywrightTimeoutError):
                     await handler._download_request(req, Spider("foo"))
 
+    @allow_windows
     async def test_contexts_max_pages(self):
         settings = {
             "PLAYWRIGHT_BROWSER_TYPE": self.browser_type,
@@ -71,6 +73,7 @@ async def test_contexts_max_pages(self):
 
             assert handler.stats.get_value("playwright/page_count/max_concurrent") == 4
 
+    @allow_windows
     async def test_max_contexts(self):
         def cb_close_context(task):
             response = task.result()
@@ -105,6 +108,7 @@ def cb_close_context(task):
 
             assert handler.stats.get_value("playwright/context_count/max_concurrent") == 4
 
+    @allow_windows
     async def test_contexts_startup(self):
         settings = {
             "PLAYWRIGHT_BROWSER_TYPE": self.browser_type,
@@ -143,6 +147,7 @@ async def test_contexts_startup(self):
             assert cookie["value"] == "bar"
             assert cookie["domain"] == "example.org"
 
+    @allow_windows
     async def test_persistent_context(self):
         temp_dir = f"{tempfile.gettempdir()}/{uuid4()}"
         settings = {
@@ -161,6 +166,7 @@ async def test_persistent_context(self):
             assert handler.context_wrappers["persistent"].persistent
             assert not hasattr(handler, "browser")
 
+    @allow_windows
     async def test_mixed_persistent_contexts(self):
         temp_dir = f"{tempfile.gettempdir()}/{uuid4()}"
         settings = {
@@ -183,6 +189,7 @@ async def test_mixed_persistent_contexts(self):
             assert not handler.context_wrappers["non-persistent"].persistent
             assert isinstance(handler.browser, Browser)
 
+    @allow_windows
     async def test_contexts_dynamic(self):
         async with make_handler({"PLAYWRIGHT_BROWSER_TYPE": self.browser_type}) as handler:
             assert len(handler.context_wrappers) == 0

tests/tests_asyncio/test_extensions.py

@@ -1,3 +1,4 @@
+import platform
 from asyncio.subprocess import Process as AsyncioProcess
 from unittest import IsolatedAsyncioTestCase
 from unittest.mock import MagicMock, patch
@@ -34,6 +35,10 @@ class MockMemoryInfo:
     rss = 999
 
 
+@pytest.mark.skipif(
+    platform.system() == "Windows",
+    reason="resource stdlib module is not available on Windows",
+)
 @patch("scrapy.extensions.memusage.MailSender")
 class TestMemoryUsageExtension(IsolatedAsyncioTestCase):
     async def test_process_availability(self, _MailSender):