ver Jan14th

setup method for Thunderbird composing tasks

desktop_env/controllers/setup.py

@@ -204,15 +204,15 @@ class SetupController:
         except requests.exceptions.RequestException as e:
             logger.error("An error occurred while trying to send the request: %s", e)
 
-    def _launch_setup(self, command: Union[str, List[str]]):
+    def _launch_setup(self, command: Union[str, List[str]], shell: bool = False):
         if not command:
             raise Exception("Empty command to launch.")
 
-        if isinstance(command, str) and len(command.split()) > 1:
+        if not shell and isinstance(command, str) and len(command.split()) > 1:
             logger.warning("Command should be a list of strings. Now it is a string. Will split it by space.")
             command = command.split()
 
-        payload = json.dumps({"command": command})
+        payload = json.dumps({"command": command, "shell": shell})
         headers = {"Content-Type": "application/json"}
 
         try:
@@ -239,7 +239,7 @@ class SetupController:
 
         while not terminates:
             try:
-                response = requests.post(self.http_server_setup_root + "/execute", headers=headers, data=payload)
+                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:

desktop_env/server/main.py

@@ -61,11 +61,12 @@ def execute_command():
 @app.route('/setup/launch', methods=["POST"])
 def launch_app():
     data = request.json
-    command: List[str] = data.get("command", [])
+    shell = data.get("shell", False)
+    command: List[str] = data.get("command", "" if shell else [])
 
     try:
-        subprocess.Popen(command)
-        return "{:} launched successfully".format(" ".join(command))
+        subprocess.Popen(command, shell=shell)
+        return "{:} launched successfully".format(command if shell else " ".join(command))
     except Exception as e:
         return jsonify({"status": "error", "message": str(e)}), 500
 

evaluation_examples/examples/thunderbird/d088f539-cab4-4f9a-ac92-9999fc3a656e.json

@@ -0,0 +1,59 @@
+{
+	"id": "d088f539-cab4-4f9a-ac92-9999fc3a656e",
+	"snapshot": "thunderbird",
+	"instruction": "Attach the my AWS bill to the email. The bill is stored at ~/aws-bill.pdf. Don't close it or send it. I haven't finish all the contents.",
+	"source": "https://support.mozilla.org/en-US/kb/how-use-attachments",
+	"config": [
+		{
+			"type": "download",
+			"parameters": {
+				"files": [
+					{
+						"url": "https://drive.usercontent.google.com/download?id=1JDUxe8Cs0D9La9ECoYbHJAIQ95DZfkDT&export=download&authuser=0&confirm=t&uuid=1e0f94d8-4fb1-4f9b-878a-516cf4a7ebc4&at=APZUnTXHBZKTuW_vQUzi-1XKuAYg:1705242554801",
+						"path": "/home/user/aws-bill.pdf"
+					},
+					{
+						"url": "https://drive.usercontent.google.com/download?id=1EHLRWzBCOsyERkSMUnTF2pnsR0n6ZvtR&export=download&authuser=0&confirm=t&uuid=de09bd5e-bef8-499a-b599-c642af190e10&at=APZUnTXqOsQkxl0zMSX6R1Sgp_v3:1704362491712",
+						"path": "/home/user/thunderbird-profile.tar.gz"
+					},
+					{
+						"url": "https://drive.usercontent.google.com/download?id=17LutYTkEXhM__sxvrdldar7p-eQba4Pm&export=download&authuser=0&confirm=t&uuid=34730e1c-0fac-401a-ab91-102a47f02224&at=APZUnTX4eFTnG9vnYlQEJX6N0jqP:1705227498314",
+						"path": "/home/user/.aws-bill-mail-body.html"
+					}
+				]
+			}
+		},
+		{
+			"type": "execute",
+			"parameters": {
+				"command": [
+					"tar",
+					"-xzv",
+					"--recursive-unlink",
+					"-f",
+					"/home/user/thunderbird-profile.tar.gz",
+					"-C",
+					"/home/user/"
+				]
+			}
+		},
+		{
+			"type": "launch",
+			"parameters": {
+				"command": "/usr/bin/thunderbird -compose \"from='Anonym Tester <anonym-x2024@outlook.com>',to=assistant@outlook.com,subject='New-month AWS Bill',body='$(cat /home/user/.aws-bill-mail-body.html)'\"",
+				"shell": true
+			}
+		}
+	],
+	"trajectory": "trajectories/d088f539-cab4-4f9a-ac92-9999fc3a656e",
+	"related_apps": [
+		"thunderbird"
+	],
+	"evaluator": {
+		"func": "run_sqlite3",
+		"result": {
+			"type": "rule",
+			"rules": {}
+		}
+	}
+}

main.py

@@ -44,9 +44,9 @@ def human_agent():
     Runs the Gym environment with human input.
     """
 
-    with open("evaluation_examples/examples/thunderbird/3d1682a7-0fb0-49ae-a4dc-a73afd2d06d5.json.nosetup", "r") as f:
+    with open("evaluation_examples/examples/thunderbird/d088f539-cab4-4f9a-ac92-9999fc3a656e.json", "r") as f:
         example = json.load(f)
-    example["snapshot"] = "Snapshot 20"
+    example["snapshot"] = "Snapshot 21"
 
     env = DesktopEnv( path_to_vm="../../../../大文件/镜像/Ubuntu-1218/Ubuntu/Ubuntu.vmx"
                     , action_space="computer_13"
@@ -87,8 +87,8 @@ def human_agent():
             logger.info("The episode is done.")
             break
 
-    result = env.evaluate()
-    logger.info("Result: %.2f", result)
+    #result = env.evaluate()
+    #logger.info("Result: %.2f", result)
 
     #input("PAUSING")