inspect-ai 0.3.70__py3-none-any.whl → 0.3.71__py3-none-any.whl

inspect_ai/tool/_tools/_web_browser/_resources/web_client.py

@@ -1,63 +1,213 @@
-"""A client to interract with the web server."""
-
+import argparse
 import sys
+from typing import Literal
 
-import grpc
-from dm_env_rpc.v1 import connection as dm_env_connection
-from dm_env_rpc.v1 import dm_env_adaptor, dm_env_rpc_pb2
+from constants import DEFAULT_SESSION_NAME, SERVER_PORT
+from rpc_client_helpers import RPCError, rpc_call
+from web_browser_rpc_types import (
+    ClickArgs,
+    CrawlerBaseArgs,
+    CrawlerResponse,
+    GoArgs,
+    NewSessionArgs,
+    NewSessionResponse,
+    ScrollArgs,
+    TypeOrSubmitArgs,
+)
 
-_DM_ENV_BASE_PORT = 9443
-_WORLD_NAME = "WebBrowser"
-_SESSION_FLAG = "--session_name="
+_SERVER_URL = f"http://localhost:{SERVER_PORT}/"
 
 
-def parse_args(cli_args: list[str]) -> tuple[str, str]:
-    if not cli_args:
-        return _WORLD_NAME, ""
+def main() -> None:
+    if len(sys.argv) > 1:
+        command, params = _parse_args()
+        _execute_command(command, params)
+    else:
+        _interactive_mode()
 
-    if cli_args[0].startswith(_SESSION_FLAG):
-        world_name = cli_args[0][len(_SESSION_FLAG) :]
 
-        if len(cli_args) == 1:
-            return world_name, ""
+def _execute_command(
+    command: str,
+    params: NewSessionArgs
+    | GoArgs
+    | ClickArgs
+    | TypeOrSubmitArgs
+    | ScrollArgs
+    | CrawlerBaseArgs,
+) -> None:
+    try:
+        if command == "new_session":
+            print(
+                rpc_call(
+                    _SERVER_URL, command, dict(params), NewSessionResponse
+                ).session_name
+            )
         else:
-            return world_name, " ".join(cli_args[1:])
+            response = rpc_call(
+                _SERVER_URL,
+                command,
+                dict(params),
+                CrawlerResponse,
+            )
+            for key, value in vars(response).items():
+                if value is not None:
+                    print(key, ": ", value)
 
-    else:
-        return _WORLD_NAME, " ".join(cli_args)
+    except RPCError as rpc_error:
+        _return_error(f"error: {rpc_error}")
 
 
-def main() -> None:
-    world_name, command = parse_args(sys.argv[1:])
-
-    # Keep connection open even when empty data pings are received. This is
-    # required to avoid a "too many pings" error.
-    options = [
-        ("grpc.keepalive_permit_without_calls", 0),
-        ("grpc.keepalive_timeout_ms", 20000),
-        ("grpc.http2.max_pings_without_data", 0),
-        ("grpc.http2.max_pings_without_data", 0),
-        ("grpc.http2.max_ping_strikes", 0),
-        ("grpc.http2.min_recv_ping_interval_without_data_ms", 0),
-    ]
-    channel = grpc.secure_channel(
-        f"localhost:{_DM_ENV_BASE_PORT}",
-        grpc.local_channel_credentials(),
-        options=options,
+def _interactive_mode() -> None:
+    print(
+        "Welcome to the Playwright Crawler interactive mode!\n"
+        "commands:\n"
+        "  web_go <URL> - goes to the specified url.\n"
+        "  web_click <ELEMENT_ID> - clicks on a given element.\n"
+        "  web_scroll <up/down> - scrolls up or down one page.\n"
+        "  web_forward - navigates forward a page.\n"
+        "  web_back - navigates back a page.\n"
+        "  web_refresh - reloads current page (F5).\n"
+        "  web_type <ELEMENT_ID> <TEXT> - types the specified text into the input with the specified id.\n"
+        "  web_type_submit <ELEMENT_ID> <TEXT> - types the specified text into the input with the specified id and presses ENTER to submit the form."
+    )
+
+    session_created = False
+    while True:
+        try:
+            user_input = input("Enter command: ").strip()
+            if user_input.lower() in {"exit", "quit"}:
+                break
+            args = user_input.split()
+            sys.argv = ["cli"] + args
+            command, params = _parse_args()
+            print(f"command: {command}, params: {params}")
+            if not session_created:
+                _execute_command("new_session", NewSessionArgs(headful=True))
+                session_created = True
+            _execute_command(command, params)
+        except Exception as e:  # pylint: disable=broad-exception-caught
+            print(f"Error: {e}")
+
+
+def _return_error(error: str) -> None:
+    print(error, file=sys.stderr)
+    sys.exit(1)
+
+
+def _create_main_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(prog="web_client")
+    parser.add_argument(
+        "--session_name",
+        type=str,
+        required=False,
+        default=DEFAULT_SESSION_NAME,
+        help="Session name",
     )
-    connection = dm_env_connection.Connection(channel)
-
-    specs = connection.send(
-        dm_env_rpc_pb2.JoinWorldRequest(world_name=world_name)
-    ).specs
-
-    with dm_env_adaptor.DmEnvAdaptor(
-        connection=connection, specs=specs, nested_tensors=False
-    ) as env:
-        time_step = env.step({"command": command})
-        for key, value in time_step.observation.items():
-            print(key, ": ", value)
-        env.close()
+    return parser
+
+
+def _create_command_parser() -> argparse.ArgumentParser:
+    result = argparse.ArgumentParser(prog="web_client")
+
+    subparsers = result.add_subparsers(dest="command", required=True)
+
+    go_parser = subparsers.add_parser("web_go")
+    go_parser.add_argument("url", type=str, help="URL to navigate to")
+
+    click_parser = subparsers.add_parser("web_click")
+    click_parser.add_argument("element_id", type=str, help="ID of the element to click")
+
+    scroll_parser = subparsers.add_parser("web_scroll")
+    scroll_parser.add_argument(
+        "direction",
+        type=str,
+        choices=["up", "down"],
+        help="Direction to scroll (up or down)",
+    )
+    subparsers.add_parser("web_forward")
+    subparsers.add_parser("web_back")
+    subparsers.add_parser("web_refresh")
+
+    type_parser = subparsers.add_parser("web_type")
+    type_parser.add_argument(
+        "element_id", type=str, help="ID of the element to type into"
+    )
+    type_parser.add_argument("text", type=str, help="The text to type")
+
+    submit_parser = subparsers.add_parser("web_type_submit")
+    submit_parser.add_argument(
+        "element_id",
+        type=str,
+        help="ID of the element to type into and submit",
+    )
+    submit_parser.add_argument("text", type=str, help="The text to type")
+
+    # Add common argument to all subparsers
+    for name, subparser in subparsers.choices.items():
+        if name != "new_session":
+            subparser.add_argument(
+                "--session_name",
+                type=str,
+                nargs="?",
+                required=False,
+                default=DEFAULT_SESSION_NAME,
+                help="Session name",
+            )
+
+    return result
+
+
+main_parser = _create_main_parser()
+command_parser = _create_command_parser()
+
+
+def _parse_args() -> (
+    tuple[Literal["web_go"], GoArgs]
+    | tuple[Literal["web_click"], ClickArgs]
+    | tuple[Literal["web_type", "web_type_submit"], TypeOrSubmitArgs]
+    | tuple[Literal["web_scroll"], ScrollArgs]
+    | tuple[Literal["web_forward", "web_back", "web_refresh"], CrawlerBaseArgs]
+):
+    # web_client.py supports a very non-standard command line. It has a required named
+    # parameter, --session_name, before the command.
+    # Unfortunately, because we can't break backwards compatibility, we're stuck
+    # with that. To properly parse it, we'll be forced to have a separate parser
+    # for --session_name and merge the results with the normal command parser.
+
+    main_args, remaining_args = main_parser.parse_known_args()
+    session_name = main_args.session_name or DEFAULT_SESSION_NAME
+
+    command_args = command_parser.parse_args(remaining_args)
+    command_args_dict = vars(command_args)
+
+    match command_args.command:
+        case "web_go":
+            return command_args_dict["command"], GoArgs(
+                url=command_args_dict["url"],
+                session_name=session_name,
+            )
+        case "web_click":
+            return command_args_dict["command"], ClickArgs(
+                element_id=command_args_dict["element_id"],
+                session_name=session_name,
+            )
+        case "web_type" | "web_type_submit":
+            return command_args_dict["command"], TypeOrSubmitArgs(
+                element_id=command_args_dict["element_id"],
+                text=command_args_dict["text"],
+                session_name=session_name,
+            )
+        case "web_scroll":
+            return command_args_dict["command"], ScrollArgs(
+                direction=command_args_dict["direction"],
+                session_name=session_name,
+            )
+        case "web_forward" | "web_back" | "web_refresh":
+            return command_args_dict["command"], CrawlerBaseArgs(
+                session_name=session_name,
+            )
+        case _:
+            raise ValueError("Unexpected command")
 
 
 if __name__ == "__main__":

inspect_ai/tool/_tools/_web_browser/_resources/web_client_new_session.py

@@ -1,33 +1,34 @@
-"""Simple script to run and test the RPC server."""
+import argparse
+import sys
 
-import grpc
-from dm_env_rpc.v1 import connection as dm_env_connection
-from dm_env_rpc.v1 import dm_env_rpc_pb2
+from constants import SERVER_PORT
+from rpc_client_helpers import RPCError, rpc_call
+from web_browser_rpc_types import NewSessionArgs, NewSessionResponse
 
-_DM_ENV_BASE_PORT = 9443
 
-
-def main():
-    # Keep connection open even when empty data pings are received. This is
-    # required to avoid a "too many pings" error.
-    options = [
-        ("grpc.keepalive_permit_without_calls", 0),
-        ("grpc.keepalive_timeout_ms", 20000),
-        ("grpc.http2.max_pings_without_data", 0),
-        ("grpc.http2.max_ping_strikes", 0),
-        ("grpc.http2.min_recv_ping_interval_without_data_ms", 0),
-    ]
-    # Creating a world with the headless browser.
-    channel = grpc.secure_channel(
-        f"localhost:{_DM_ENV_BASE_PORT}",
-        grpc.local_channel_credentials(),
-        options=options,
+def main() -> None:
+    parser = argparse.ArgumentParser(prog="web_client_new_session")
+    parser.add_argument(
+        "--headful", action="store_true", help="Run in headful mode for testing"
     )
-    connection = dm_env_connection.Connection(channel)
-    world_name = connection.send(dm_env_rpc_pb2.CreateWorldRequest()).world_name
-    print(world_name)
+    args_class = parser.parse_args()
+    args_dict = vars(args_class)
+    # TODO: Frick. this does no validation
+    params_typed_dict = NewSessionArgs(headful=args_dict["headful"])
+    params = dict(params_typed_dict)
 
-    connection.close()
+    try:
+        print(
+            rpc_call(
+                f"http://localhost:{SERVER_PORT}/",
+                "new_session",
+                params,
+                NewSessionResponse,
+            ).session_name
+        )
+    except RPCError as rpc_error:
+        print(rpc_error, file=sys.stderr)
+        sys.exit(1)
 
 
 if __name__ == "__main__":

inspect_ai/tool/_tools/_web_browser/_resources/web_server.py

@@ -1,52 +1,191 @@
-"""Simple script to run and test the RPC server."""
+import threading
+from typing import Awaitable, Callable, Unpack
 
-from concurrent import futures
+from aiohttp.web import Application, Request, Response, run_app
+from jsonrpcserver import Result, Success, async_dispatch, method
 
-import dm_env_servicer
-import grpc
-import web_environment
-from dm_env_rpc.v1 import connection as dm_env_connection
-from dm_env_rpc.v1 import dm_env_rpc_pb2, dm_env_rpc_pb2_grpc
+from constants import DEFAULT_SESSION_NAME, SERVER_PORT
+from playwright_browser import PlaywrightBrowser
+from playwright_crawler import PlaywrightCrawler
+from scale_factor import get_screen_scale_factor
+from web_browser_rpc_types import (
+    ClickArgs,
+    CrawlerBaseArgs,
+    CrawlerResponse,
+    GoArgs,
+    NewSessionArgs,
+    NewSessionResponse,
+    ScrollArgs,
+    TypeOrSubmitArgs,
+)
 
-_DM_ENV_BASE_PORT = 9443
 
+class Sessions:
+    def __init__(self) -> None:
+        self._lock = threading.Lock()
+        self._browser: PlaywrightBrowser | None = None
+        self._sessions: dict[str, PlaywrightCrawler] = {}
 
-def main():
-    # Keep connection open even when empty data pings are received. This is
-    # required to avoid a "too many pings" error.
-    options = [
-        ("grpc.keepalive_permit_without_calls", 0),
-        ("grpc.keepalive_timeout_ms", 20000),
-        ("grpc.http2.max_pings_without_data", 0),
-        ("grpc.http2.max_ping_strikes", 0),
-        ("grpc.http2.min_recv_ping_interval_without_data_ms", 0),
-    ]
-    grpc_server = grpc.server(
-        # We must have a single worker thread since the web environment is not
-        # thread safe.
-        futures.ThreadPoolExecutor(max_workers=1),
-        options=options,
-    )
-    env_service = dm_env_servicer.EnvironmentService(web_environment.WebEnvironment)
-    dm_env_rpc_pb2_grpc.add_EnvironmentServicer_to_server(env_service, grpc_server)
+    async def new_session(self, headful: bool) -> str:
+        with self._lock:
+            if not self._browser:
+                self._browser = await PlaywrightBrowser.create(headless=not headful)
+            current_count = len(self._sessions)
+            name = (
+                DEFAULT_SESSION_NAME
+                if current_count == 0
+                else f"{DEFAULT_SESSION_NAME}_{current_count}"
+            )
+            crawler = await PlaywrightCrawler.create(
+                await self._browser.get_new_context(),
+                device_scale_factor=get_screen_scale_factor() if headful else 1,
+            )
+            self._sessions[name] = crawler
+            return name
 
-    grpc_server.add_secure_port(
-        f"localhost:{_DM_ENV_BASE_PORT}", grpc.local_server_credentials()
-    )
+    async def get_crawler_for_session(self, name: str) -> PlaywrightCrawler:
+        if not self._sessions:
+            await self.new_session(False)
+        return self._sessions[name]
 
-    grpc_server.start()
 
-    # Creating a world with the headless browser after the server started.
-    channel = grpc.secure_channel(
-        f"localhost:{_DM_ENV_BASE_PORT}",
-        grpc.local_channel_credentials(),
-        options=options,
+sessions = Sessions()
+
+
+@method
+async def new_session(**kwargs: Unpack[NewSessionArgs]) -> NewSessionResponse:
+    return Success(
+        NewSessionResponse(
+            session_name=await sessions.new_session(kwargs.get("headful", False))
+        ).model_dump()
     )
-    connection = dm_env_connection.Connection(channel)
-    connection.send(dm_env_rpc_pb2.CreateWorldRequest())
-    connection.close()
 
-    grpc_server.wait_for_termination()
+
+@method
+async def web_go(**kwargs: Unpack[GoArgs]) -> Result:
+    async def handler(crawler: PlaywrightCrawler):
+        await (await crawler.current_page).go_to_url(kwargs["url"])
+
+    return await _execute_crawler_command(kwargs["session_name"], handler)
+
+
+@method
+async def web_click(**kwargs: Unpack[ClickArgs]) -> Result:
+    async def handler(crawler: PlaywrightCrawler):
+        await (await crawler.current_page).click(kwargs["element_id"])
+
+    return await _execute_crawler_command(kwargs["session_name"], handler)
+
+
+@method
+async def web_scroll(**kwargs: Unpack[ScrollArgs]) -> Result:
+    async def handler(crawler: PlaywrightCrawler):
+        await (await crawler.current_page).scroll(kwargs["direction"])
+
+    return await _execute_crawler_command(kwargs["session_name"], handler)
+
+
+@method
+async def web_forward(**kwargs: Unpack[CrawlerBaseArgs]) -> Result:
+    async def handler(crawler: PlaywrightCrawler):
+        await (await crawler.current_page).forward()
+
+    return await _execute_crawler_command(kwargs["session_name"], handler)
+
+
+@method
+async def web_back(**kwargs: Unpack[CrawlerBaseArgs]) -> Result:
+    async def handler(crawler: PlaywrightCrawler):
+        await (await crawler.current_page).back()
+
+    return await _execute_crawler_command(kwargs["session_name"], handler)
+
+
+@method
+async def web_refresh(**kwargs: Unpack[CrawlerBaseArgs]) -> Result:
+    async def handler(crawler: PlaywrightCrawler):
+        await (await crawler.current_page).refresh()
+
+    return await _execute_crawler_command(kwargs["session_name"], handler)
+
+
+@method
+async def web_type(**kwargs: Unpack[TypeOrSubmitArgs]) -> Result:
+    async def handler(crawler: PlaywrightCrawler):
+        await (await crawler.current_page).type(
+            kwargs["element_id"], _str_from_str_or_list(kwargs["text"])
+        )
+
+    return await _execute_crawler_command(kwargs["session_name"], handler)
+
+
+@method
+async def web_type_submit(**kwargs: Unpack[TypeOrSubmitArgs]) -> Result:
+    async def handler(crawler: PlaywrightCrawler):
+        await (await crawler.current_page).clear(kwargs["element_id"])
+        await (await crawler.current_page).type(
+            kwargs["element_id"], _str_from_str_or_list(kwargs["text"]) + "\n"
+        )
+
+    return await _execute_crawler_command(kwargs["session_name"], handler)
+
+
+async def _execute_crawler_command(
+    session_name: str, handler: Callable[[PlaywrightCrawler], Awaitable[None]]
+) -> Result:
+    if not sessions:
+        await new_session()
+    try:
+        crawler = await sessions.get_crawler_for_session(session_name)
+        await handler(crawler)
+        await (await crawler.current_page).update()
+
+        # If there's a cookies message click to sort it out.
+        await _auto_click_cookies(crawler)
+
+        return Success(
+            CrawlerResponse(
+                web_url=(await crawler.current_page).url.split("?")[0],
+                main_content=(await crawler.current_page).render_main_content(),
+                web_at=(await crawler.current_page).render_at(),
+                error=None,
+            ).model_dump()
+        )
+    except Exception as e:  # pylint: disable=broad-exception-caught
+        return Success(
+            CrawlerResponse(
+                web_url=(await crawler.current_page).url.split("?")[0],
+                web_at="encountered error",
+                error=str(e),
+            ).model_dump()
+        )
+
+
+def _str_from_str_or_list(str_or_list: str | list[str]) -> str:
+    return str_or_list if isinstance(str_or_list, str) else " ".join(str_or_list)
+
+
+async def _auto_click_cookies(crawler: PlaywrightCrawler):
+    """Autoclick any cookies popup."""
+    try:
+        accept_node = (await crawler.current_page).lookup_node("<Accept all>")
+    except LookupError:
+        return
+    await (await crawler.current_page).click(accept_node.node_id)
+    await (await crawler.current_page).update()
+
+
+def main():
+    async def handle_request(request: Request) -> Response:
+        return Response(
+            text=await async_dispatch(await request.text()),
+            content_type="application/json",
+        )
+
+    app = Application()
+    app.router.add_post("/", handle_request)
+
+    run_app(app, port=SERVER_PORT)
 
 
 if __name__ == "__main__":