sunholo 0.74.7__py3-none-any.whl → 0.74.9__py3-none-any.whl

sunholo/tools/web_browser.py

@@ -4,6 +4,7 @@ import json
 from datetime import datetime
 import urllib.parse
 import time
+from io import BytesIO
 
 from ..logging import log
 
@@ -115,7 +116,14 @@ class BrowseWebWithImagePromptsBot:
     ```
     """
 #class BrowseWebWithImagePromptsBot:
-    def __init__(self, session_id, website_name, browser_type='chromium', headless=True, max_steps=10):
+    def __init__(
+            self,
+            website_name:str, 
+            session_id: str=None, 
+            browser_type:str='chromium', 
+            headless:bool=True, 
+            max_steps:int=10
+        ):
         try:
             from playwright.sync_api import sync_playwright
         except ImportError as err:
@@ -125,12 +133,19 @@ class BrowseWebWithImagePromptsBot:
         if not sync_playwright:
             raise ImportError("playright needed for BrowseWebWithImagePromptsBot class - install via `pip install sunholo[tools]`")
         
-        self.session_id = session_id or datetime.now().strftime("%Y%m%d%H%M%S")
+        log.info(f"Starting BrowseWebWithImagePromptsBot with {website_name=}, {session_id=}, {browser_type=}, {headless=}, {max_steps=}")
+        
+        # Assign session_id if it is None or 'None'
+        if not session_id or session_id == 'None':
+            self.session_id = datetime.now().strftime("%Y%m%d%H%M%S")
+        else:
+            self.session_id = session_id
+
         self.website_name = website_name
-        self.browser_type = browser_type
-        self.max_steps = max_steps
+        self.browser_type = browser_type or 'chromium'
+        self.max_steps = int(max_steps)
         self.steps = 0
-        self.screenshot_dir = f"browser_tool/{get_clean_website_name(website_name)}/{session_id}"
+        self.screenshot_dir = f"browser_tool/{get_clean_website_name(website_name)}/{self.session_id}"
         os.makedirs(self.screenshot_dir, exist_ok=True)
         self.cookie_file = os.path.join(self.screenshot_dir, "cookies.json")
         self.action_log_file = os.path.join(self.screenshot_dir, "action_log.json")
@@ -199,21 +214,81 @@ class BrowseWebWithImagePromptsBot:
             log.warning(f"navigate failed with {str(err)}")
             self.action_log.append(f"Tried to navigate to {url} but got an error")
 
+    def get_locator_via_roles_and_placeholder(self, selector: str):
+        interactive_roles = ["button", "link", "menuitem", "menuitemcheckbox", "menuitemradio", "tab", "option"]
+
+        for role in interactive_roles:
+            log.info(f'Trying role {role} for selector {selector}')
+            elements = self.page.get_by_role(role).get_by_placeholder(selector).locator("visible=true").all()
+            if elements:
+                log.info(f"Got {len(elements)} elements for selector {selector} with role {role}")
+                for element in elements:
+                    try:
+                        log.info(f"Trying {selector} with element.hover locator: {element}")
+                        try:
+                            element.hover(timeout=10000, trial=True)
+                            self.action_log.append(f"Successfully found element via selector: {selector}")
+
+                            return element
+                        
+                        except Exception as err:
+                            log.warning(f"Could not hover over element: {element} {str(err)} - trying next element")
+                    except Exception as e:
+                        log.error(f"Failed to get locator for selector '{selector}' with role {role}: {str(e)}")
+                    
+                    time.sleep(0.5)  # Wait for a bit before retrying
+            
+            log.info(f"No elements for '{selector}' within role '{role}'")
+
+        self.action_log.append(f"FAILED: Using page.get_by_role('role').get_by_placeholder('{selector}').locator('visible=true') could not find any valid element. Try something else.")
+        return None
+    
+    def get_locator_via_roles_and_text(self, selector: str):
+        interactive_roles = ["button", "link", "menuitem", "menuitemcheckbox", "menuitemradio", "tab", "option"]
+
+        for role in interactive_roles:
+            log.info(f'Trying role {role} for selector {selector}')
+            elements = self.page.get_by_role(role).get_by_text(selector).locator("visible=true").all()
+            if elements:
+                log.info(f"Got {len(elements)} elements for selector {selector} with role {role}")
+                for element in elements:
+                    try:
+                        log.info(f"Trying {selector} with element.hover locator: {element}")
+                        try:
+                            element.hover(timeout=10000, trial=True)
+                            self.action_log.append(f"Successfully found element via selector: {selector}")
+
+                            return element
+                        
+                        except Exception as err:
+                            log.warning(f"Could not hover over element: {element} {str(err)} - trying next element")
+                    except Exception as e:
+                        log.error(f"Failed to get locator for selector '{selector}' with role {role}: {str(e)}")
+                    
+                    time.sleep(0.5)  # Wait for a bit before retrying
+            
+            log.info(f"No elements for '{selector}' within role '{role}'")
+
+        self.action_log.append(f"FAILED: Using page.get_by_role('role').get_by_text('{selector}').locator('visible=true') could not find any valid element. Try something else.")
+        return None
+
     def get_locator(self, selector, by_text=True):
         if by_text:
-            elements = self.page.get_by_text(selector).all()
-            if elements:
-                return elements[0]
-            else:
-                log.warning(f"No elements found with text: {selector}")
-                return None
+            element = self.get_locator_via_roles_and_text(selector)
         else:
-            return self.page.locator(selector)
+            element = self.page.locator(selector)
+        
+        if element:
+            return element
+
+        log.error(f"Failed to get locator for selector {selector}")
+
+        return None
 
-    def click(self, selector, by_text=True):
+    def click(self, selector):
         (x,y)=(0,0)
 
-        element = self.get_locator(selector, by_text=by_text)
+        element = self.get_locator_via_roles_and_text(selector)
         if element is None:
             self.action_log.append(f"Tried to click on text {selector} but it was not a valid location to click")
             return (x,y)
@@ -229,14 +304,14 @@ class BrowseWebWithImagePromptsBot:
         try:
             element.click()
             self.page.wait_for_load_state()
-            log.info(f"Clicked on element with selector {selector} at {x=},{y=}")
-            self.action_log.append(f"Clicked on element with selector {selector} at {x=},{y=}")
+            log.info(f"Clicked on {element=} with {selector=} at {x=},{y=}")
+            self.action_log.append(f"Clicked on {element=} with {selector=} at {x=},{y=}")
 
             return (x,y)
         
         except Exception as err:
             log.warning(f"click failed with {str(err)}")
-            self.action_log.append(f"Tried to click on element with selector {selector} at {x=},{y=} but got an error")  
+            self.action_log.append(f"Tried to click on {element=} with {selector=} at {x=},{y=} but got an error")  
 
             return (x,y)          
 
@@ -257,11 +332,11 @@ class BrowseWebWithImagePromptsBot:
             log.warning(f"Scrolled failed with {str(err)}")
             self.action_log.append(f"Tried to scroll {direction} by {amount} pixels but got an error")
 
-    def type_text(self, selector, text, by_text=True):
+    def type_text(self, selector, text):
         (x,y)=(0,0)
-        element = self.get_locator(selector, by_text=by_text)
+        element = self.get_locator_via_roles_and_placeholder(selector)
         if element is None:
-            self.action_log.append(f"Tried to type {text} via website text: {selector} but it was not a valid location to add text")
+            self.action_log.append(f"Tried to type {text} via website text: {selector} but it was not a valid element to add text")
             return (x,y)
 
         try:
@@ -275,50 +350,60 @@ class BrowseWebWithImagePromptsBot:
         try:
             element.fill(text)
             self.page.wait_for_load_state()
-            log.info(f"Typed text '{text}' into element with selector {selector} at {x=},{y=}")
-            self.action_log.append(f"Typed text '{text}' into element with selector {selector} at {x=},{y=}")
+            log.info(f"Typed text '{text}' into {element=} with {selector=} at {x=},{y=}")
+            self.action_log.append(f"Typed text '{text}' into {element=} with {selector=} at {x=},{y=}")
 
             return (x, y)
         
         except Exception as err:
             log.warning(f"Typed text failed with {str(err)}")
-            self.action_log.append(f"Tried to type text '{text}' into element with selector {selector} at {x=},{y=} but got an error")
+            self.action_log.append(f"Tried to type text '{text}' into {element=} with {selector=} at {x=},{y=} but got an error")
 
             return (x, y)
 
-    def take_screenshot(self, final=False, full_page=False, mark_action=None):
+    def take_screenshot(self, full_page=False, mark_action=None):
+
+        from PIL import Image
+
         timestamp = datetime.now().strftime("%Y%m%d%H%M%S")
         parsed_url = urllib.parse.urlparse(self.page.url)
 
         url_path = parsed_url.path
         if url_path == "/":
             url_path = "index.html"
-        if final:
-            screenshot_path = os.path.join(self.screenshot_dir, f"final/{timestamp}_{url_path}.png")
         else:
-            screenshot_path = os.path.join(self.screenshot_dir, f"{timestamp}_{url_path}.png")
-        self.page.screenshot(path=screenshot_path, full_page=full_page)
+            url_path = url_path.replace("/","_")
+        
+        if get_clean_website_name(url_path) != self.website_name:
+            url_path = f"{get_clean_website_name(url_path)}_{url_path}"
+            
+        screenshot_path = os.path.join(self.screenshot_dir, f"{timestamp}_{url_path}.png")
+        screenshot_bytes = self.page.screenshot(full_page=full_page, scale='css')
         
         if mark_action:
-            self.mark_screenshot(screenshot_path, mark_action)
+            image = self.mark_screenshot(screenshot_bytes, mark_action)
+        else:
+            image = Image.open(BytesIO(screenshot_bytes))
+        
+        image.save(screenshot_path, format='png')
 
         log.info(f"Screenshot {self.page.url} taken and saved to {screenshot_path}")
         #self.action_log.append(f"Screenshot {self.page.url} taken and saved to {screenshot_path}")
         self.session_screenshots.append(screenshot_path)
 
-        return screenshot_path
+        return screenshot_bytes
 
-    def mark_screenshot(self, screenshot_path, mark_action):
+    def mark_screenshot(self, screenshot_bytes, mark_action):
         """
         Marks the screenshot with the specified action.
 
         Parameters:
-            screenshot_path (str): The path to the screenshot.
+            screenshot_bytes (bytes): The bytes of the screenshot.
             mark_action (dict): Action details for marking the screenshot.
         """
         from PIL import Image, ImageDraw
-
-        image = Image.open(screenshot_path)
+        time.sleep(1) # maybe pass in bytes to avoid waiting for file to be written
+        image = Image.open(BytesIO(screenshot_bytes))
         draw = ImageDraw.Draw(image)
         
         if mark_action['type'] == 'click':
@@ -329,7 +414,7 @@ class BrowseWebWithImagePromptsBot:
             x, y = mark_action['position']
             draw.rectangle((x-5, y-5, x+5, y+5), outline='blue', width=3)
         
-        image.save(screenshot_path)
+        return image
 
     def get_latest_screenshot_path(self):
         screenshots = sorted(
@@ -364,6 +449,10 @@ class BrowseWebWithImagePromptsBot:
         else:
             log.warning(f'Unknown response: {response=} {type(response)}')
             output = None
+        
+        if not output:
+            log.error(f'Got no output from response: {response=}')
+            return None
 
         if 'status' not in output:
             log.error(f'Response did not contain status')
@@ -377,9 +466,9 @@ class BrowseWebWithImagePromptsBot:
 
         return output        
 
-    def send_screenshot_to_llm(self, screenshot_path, last_message):
-        with open(screenshot_path, "rb") as image_file:
-            encoded_image = base64.b64encode(image_file.read()).decode('utf-8')
+    def send_screenshot_to_llm(self, screenshot_bytes, last_message):
+
+        encoded_image = base64.b64encode(screenshot_bytes).decode('utf-8')
         
         prompt_vars = self.create_prompt_vars(last_message)
         response = self.send_prompt_to_llm(prompt_vars, encoded_image)  # Sending prompt and image separately
@@ -396,6 +485,8 @@ This method should be implemented by subclasses: `def send_prompt_to_llm(self, p
 """)
 
     def close(self):
+        log.info(f"Session {self.session_id} finished")
+        self.take_screenshot()
         self.save_cookies()
         self.browser.close()
         self.playwright.stop()
@@ -437,14 +528,14 @@ This method should be implemented by subclasses: `def send_prompt_to_llm(self, p
                 log.warning(f"Reached the maximum number of steps: {self.max_steps}")
                 return
         time.sleep(2) 
-        screenshot_path = self.take_screenshot(mark_action=mark_action)
+        screenshot_bytes = self.take_screenshot(mark_action=mark_action)
         next_browser_instructions = self.send_screenshot_to_llm(
-                screenshot_path, 
+                screenshot_bytes, 
                 last_message=last_message)
             
         return next_browser_instructions
 
-    def create_gif_from_pngs(self, frame_duration=500):
+    def create_gif_from_pngs(self, frame_duration=300):
         """
         Creates a GIF from a folder of PNG images.
 
@@ -501,9 +592,17 @@ This method should be implemented by subclasses: `def send_prompt_to_llm(self, p
                             log.error('Browser status: "in-progress" but no new_instructions')
                         last_message = next_instructions['message']
                         self.action_log.append(last_message)
-                        next_instructions = self.execute_instructions(
-                            next_instructions['new_instructions'], 
-                            last_message=last_message)
+                        try:
+                            next_instructions = self.execute_instructions(
+                                next_instructions['new_instructions'], 
+                                last_message=last_message)
+                        except Exception as err:
+                            log.error(f'session aborted due to: {str(err)}')
+                            next_instructions = {
+                                'status': 'error',
+                                'message': f'session errored with: {str(err)}'
+                            }
+                            break
                     else:
                         log.info(f'Session finished due to status={next_instructions["status"]}')
                         in_session=False
@@ -513,9 +612,6 @@ This method should be implemented by subclasses: `def send_prompt_to_llm(self, p
                     in_session=False
                     break
             
-            log.info("Session finished")
-            self.take_screenshot()
-            
             self.close()
 
             answer = None

sunholo/utils/parsers.py

@@ -186,6 +186,6 @@ def escape_braces(text):
     text = re.sub(r'(?<!})}(?!})', '}}', text)  # Replace '}' with '}}' if not already double braced
     return text
 
-def get_clean_website_name(url):
+def get_clean_website_name(url: str):
     parsed_url = urllib.parse.urlparse(url)
     return parsed_url.netloc

{sunholo-0.74.7.dist-info → sunholo-0.74.9.dist-info}/METADATA

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: sunholo
-Version: 0.74.7
+Version: 0.74.9
 Summary: Large Language Model DevOps - a package to help deploy LLMs to the Cloud.
 Home-page: https://github.com/sunholo-data/sunholo-py
-Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.74.7.tar.gz
+Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.74.9.tar.gz
 Author: Holosun ApS
 Author-email: multivac@sunholo.com
 License: Apache License, Version 2.0

{sunholo-0.74.7.dist-info → sunholo-0.74.9.dist-info}/RECORD

@@ -99,7 +99,7 @@ sunholo/streaming/streaming.py,sha256=9z6pXINEopuL_Z1RnmgXAoZJum9dzyuOxqYtEYnjf8
 sunholo/summarise/__init__.py,sha256=MZk3dblUMODcPb1crq4v-Z508NrFIpkSWNf9FIO8BcU,38
 sunholo/summarise/summarise.py,sha256=C3HhjepTjUhUC8FLk4jMQIBvq1BcORniwuTFHjPVhVo,3784
 sunholo/tools/__init__.py,sha256=5NuYpwwTX81qGUWvgwfItoSLXteNnp7KjgD7IPZUFjI,53
-sunholo/tools/web_browser.py,sha256=SGNSObQ6bxgFGrI6FoKI31F1EuSQ1XonJxauStYWtTg,21780
+sunholo/tools/web_browser.py,sha256=wms9vHndrbD2NaAvNpYG291YiGP8wVHJ6iljsMVZc5w,26292
 sunholo/utils/__init__.py,sha256=Hv02T5L2zYWvCso5hzzwm8FQogwBq0OgtUbN_7Quzqc,89
 sunholo/utils/api_key.py,sha256=Ct4bIAQZxzPEw14hP586LpVxBAVi_W9Serpy0BK-7KI,244
 sunholo/utils/big_context.py,sha256=gJIP7_ZL-YSLhOMq8jmFTMqH1wq8eB1NK7oKPeZAq2s,5578
@@ -108,7 +108,7 @@ sunholo/utils/config_class.py,sha256=4fm2Bwn_zFhVJBiUnMBzfCA5LKhTcBMU3mzhf5seXrw
 sunholo/utils/config_schema.py,sha256=Wv-ncitzljOhgbDaq9qnFqH5LCuxNv59dTGDWgd1qdk,4189
 sunholo/utils/gcp.py,sha256=uueODEpA-P6O15-t0hmcGC9dONLO_hLfzSsSoQnkUss,4854
 sunholo/utils/gcp_project.py,sha256=0ozs6tzI4qEvEeXb8MxLnCdEVoWKxlM6OH05htj7_tc,1325
-sunholo/utils/parsers.py,sha256=aCIT08VjVbu8E3BAxepIiqFQa8zwu4bTgEstU_qjyg8,5414
+sunholo/utils/parsers.py,sha256=akLSZLdvHf5T9OKVj5C3bo4g3Y8puATd8rxnbB4tWDs,5419
 sunholo/utils/timedelta.py,sha256=BbLabEx7_rbErj_YbNM0MBcaFN76DC4PTe4zD2ucezg,493
 sunholo/utils/user_ids.py,sha256=SQd5_H7FE7vcTZp9AQuQDWBXd4FEEd7TeVMQe1H4Ny8,292
 sunholo/utils/version.py,sha256=P1QAJQdZfT2cMqdTSmXmcxrD2PssMPEGM-WI6083Fck,237
@@ -117,9 +117,9 @@ sunholo/vertex/extensions_class.py,sha256=4PsUM9dSYrIPpq9bZ3K2rL9MRb_rlqAgnMsW0o
 sunholo/vertex/init.py,sha256=-w7b9GKsyJnAJpYHYz6_zBUtmeJeLXlEkgOfwoe4DEI,2715
 sunholo/vertex/memory_tools.py,sha256=pomHrDKqvY8MZxfUqoEwhdlpCvSGP6KmFJMVKOimXjs,6842
 sunholo/vertex/safety.py,sha256=S9PgQT1O_BQAkcqauWncRJaydiP8Q_Jzmu9gxYfy1VA,2482
-sunholo-0.74.7.dist-info/LICENSE.txt,sha256=SdE3QjnD3GEmqqg9EX3TM9f7WmtOzqS1KJve8rhbYmU,11345
-sunholo-0.74.7.dist-info/METADATA,sha256=7UcUFn-TT6y-4aTHe3JyuBpoK1iiBVbFsusIAXys94U,7010
-sunholo-0.74.7.dist-info/WHEEL,sha256=y4mX-SOX4fYIkonsAGA5N0Oy-8_gI4FXw5HNI1xqvWg,91
-sunholo-0.74.7.dist-info/entry_points.txt,sha256=bZuN5AIHingMPt4Ro1b_T-FnQvZ3teBes-3OyO0asl4,49
-sunholo-0.74.7.dist-info/top_level.txt,sha256=wt5tadn5--5JrZsjJz2LceoUvcrIvxjHJe-RxuudxAk,8
-sunholo-0.74.7.dist-info/RECORD,,
+sunholo-0.74.9.dist-info/LICENSE.txt,sha256=SdE3QjnD3GEmqqg9EX3TM9f7WmtOzqS1KJve8rhbYmU,11345
+sunholo-0.74.9.dist-info/METADATA,sha256=tf6DzIGnmFeb47QUwocgunQWJD2WEmYI3Kv73pcCNy4,7010
+sunholo-0.74.9.dist-info/WHEEL,sha256=y4mX-SOX4fYIkonsAGA5N0Oy-8_gI4FXw5HNI1xqvWg,91
+sunholo-0.74.9.dist-info/entry_points.txt,sha256=bZuN5AIHingMPt4Ro1b_T-FnQvZ3teBes-3OyO0asl4,49
+sunholo-0.74.9.dist-info/top_level.txt,sha256=wt5tadn5--5JrZsjJz2LceoUvcrIvxjHJe-RxuudxAk,8
+sunholo-0.74.9.dist-info/RECORD,,