lzy/sci-gui-agent-benchmark
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'origin/main'

Browse Source 
# Conflicts:
#	desktop_env/controllers/setup.py
#	desktop_env/evaluators/metrics/utils.py
This commit is contained in:
Timothyxxx
2024-01-12 17:30:15 +08:00
parent 5a93a32958 d6f694da1c
commit 186df65683
67 changed files with 4514 additions and 272 deletions
Download Patch File Download Diff File Expand all files Collapse all files

130
desktop_env/controllers/setup.py
View File

@@ -1,17 +1,18 @@
import json
import logging
import os.path
import time
import os.path
import traceback
import uuid
from typing import Any, Union
from typing import Dict, List
from typing import Any, Union, Optional
import requests
from playwright.sync_api import sync_playwright
from requests_toolbelt.multipart.encoder import MultipartEncoder
from desktop_env.evaluators.metrics.utils import compare_urls
import logging
logger = logging.getLogger("desktopenv.setup")
@@ -47,6 +48,8 @@ class SetupController:
assert hasattr(self, setup_function)
getattr(self, setup_function)(**parameters)
logger.info("SETUP: %s(%s)", setup_function, str(parameters))
# self._download_setup(config)
# self._change_wallpaper(config)
# self._tidy_desktop(config) todo: implement this
@@ -54,31 +57,31 @@ class SetupController:
# can add other setup steps
# ZDY_COMMENT: merged with launch
# def _command_setup(self, command: str):
# """
# Directly send a command into the virtual machine os for setting up.
# """
# payload = json.dumps({"command": command})
# headers = {
# 'Content-Type': 'application/json'
# }
# timeout = 5
# timout_whitelist = ["vlc"]
#
# try:
#
# response = requests.post(self.http_server + "/execute", headers=headers, data=payload, timeout=timeout)
# if response.status_code == 200:
# print("Command executed successfully:", response.text)
# else:
# print("Failed to execute command. Status code:", response.status_code)
# except requests.exceptions.Timeout as e:
# if command in timout_whitelist:
# print("Command executed successfully:", command)
# else:
# print("An error occurred while trying to execute the command:", e)
# except requests.exceptions.RequestException as e:
# print("An error occurred while trying to execute the command:", e)
#def _command_setup(self, command: str):
#"""
#Directly send a command into the virtual machine os for setting up.
#"""
#payload = json.dumps({"command": command})
#headers = {
#'Content-Type': 'application/json'
#}
#timeout = 5
#timout_whitelist = ["vlc"]
#
#try:
#
#response = requests.post(self.http_server + "/execute", headers=headers, data=payload, timeout=timeout)
#if response.status_code == 200:
#print("Command executed successfully:", response.text)
#else:
#print("Failed to execute command. Status code:", response.status_code)
#except requests.exceptions.Timeout as e:
#if command in timout_whitelist:
#print("Command executed successfully:", command)
#else:
#print("An error occurred while trying to execute the command:", e)
#except requests.exceptions.RequestException as e:
#print("An error occurred while trying to execute the command:", e)
def _download_setup(self, files: List[Dict[str, str]]):
"""
@@ -221,35 +224,60 @@ class SetupController:
except requests.exceptions.RequestException as e:
logger.error("An error occurred while trying to send the request: %s", e)
def _execute_setup(self, command: List[str], stdout: str = "", stderr: str = ""):
def _execute_setup( self, command: List[str]
, stdout: str = "", stderr: str = ""
, shell: bool = False, until: Optional[Dict[str, Any]] = None):
if not command:
raise Exception("Empty comman to launch.")
payload = json.dumps({"command": command})
until: Dict[str, Any] = until or {}
terminates: bool = False
nb_failings = 0
payload = json.dumps({"command": command, "shell": shell})
headers = {"Content-Type": "application/json"}
try:
response = requests.post(self.http_server + "/setup" + "/execute", headers=headers, data=payload)
if response.status_code == 200:
results: Dict[str, str] = response.json()
if stdout:
with open(os.path.join(self.cache_dir, stdout), "w") as f:
f.write(results["output"])
if stderr:
with open(os.path.join(self.cache_dir, stderr), "w") as f:
f.write(results["error"])
logger.info("Command executed successfully: %s -> %s"
, " ".join(command)
, response.text
)
else:
logger.error("Failed to launch application. Status code: %s", response.text)
except requests.exceptions.RequestException as e:
logger.error("An error occurred while trying to send the request: %s", e)
traceback.print_exc()
while not terminates:
try:
response = requests.post(self.http_server_setup_root + "/execute", headers=headers, data=payload)
if response.status_code == 200:
results: Dict[str, str] = response.json()
if stdout:
with open(os.path.join(self.cache_dir, stdout), "w") as f:
f.write(results["output"])
if stderr:
with open(os.path.join(self.cache_dir, stderr), "w") as f:
f.write(results["error"])
logger.info( "Command executed successfully: %s -> %s"
, " ".join(command)
, response.text
)
else:
logger.error("Failed to launch application. Status code: %s", response.text)
results = None
nb_failings += 1
except requests.exceptions.RequestException as e:
logger.error("An error occurred while trying to send the request: %s", e)
traceback.print_exc()
def _command_setup(self, command: List[str], stdout: str = "", stderr: str = ""):
self._execute_setup(command, stdout, stderr)
results = None
nb_failings += 1
if len(until)==0:
terminates = True
elif results is not None:
terminates = "returncode" in until and results["returncode"]==until["returncode"]\
or "stdout" in until and until["stdout"] in results["output"]\
or "stderr" in until and until["stderr"] in results["error"]
terminates = terminates or nb_failings>=5
if not terminates:
time.sleep(0.3)
def _command_setup(self, command: List[str], **kwargs):
self._execute_setup(command, **kwargs)
def _sleep_setup(self, seconds: float):
time.sleep(seconds)
def _act_setup(self, action_seq: List[Union[Dict[str, Any], str]]):
# TODO