app.py

import os
import threading
import time
from concurrent.futures import ThreadPoolExecutor, as_completed

import gradio as gr
import requests
import inspect
import pandas as pd

from agent import (
    DEFAULT_MODEL_ID,
    BasicAgent,
    anthropic_model_choices,
    default_anthropic_model,
)

# (Keep Constants as is)
# --- Constants ---
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
CONCURRENCY = max(1, int(os.getenv("AGENT_CONCURRENCY", "4")))
TOTAL_REQUESTS_PER_MINUTE = int(os.getenv("AGENT_REQUESTS_PER_MINUTE", "30"))
SUBMIT_ATTEMPTS = 4
SUBMIT_RETRY_DELAY = 15

# Answers from the last completed run, so a failed submission can be retried without re-running.
_last_run: dict[str, dict] = {}

_thread_local = threading.local()


def describe_backend() -> str:
    if os.getenv("ANTHROPIC_API_KEY"):
        return f"**Backend:** Anthropic (`ANTHROPIC_API_KEY` found) — default model `{default_anthropic_model()}`"
    return (
        "**Backend:** Hugging Face Inference (no `ANTHROPIC_API_KEY` set) — default model "
        f"`{os.getenv('HF_MODEL_ID') or DEFAULT_MODEL_ID}`"
    )


def get_worker_agent(model_id: str | None, requests_per_minute: int) -> BasicAgent:
    """One agent per worker thread: a CodeAgent's memory cannot be shared across runs in flight."""
    agent = getattr(_thread_local, "agent", None)
    if agent is None:
        agent = BasicAgent(model_id=model_id, requests_per_minute=requests_per_minute)
        _thread_local.agent = agent
    return agent


def answer_question(item: dict, model_id: str | None, requests_per_minute: int):
    """Returns (answer or None on failure, text to display)."""
    task_id = item.get("task_id")
    started = time.monotonic()
    try:
        agent = get_worker_agent(model_id, requests_per_minute)
        answer = agent(item["question"], task_id=task_id, file_name=item.get("file_name"))
    except Exception as e:
        print(f"Error running agent on task {task_id} after {time.monotonic() - started:.0f}s: {e}")
        return None, f"AGENT ERROR: {e}"
    print(f"Task {task_id} answered in {time.monotonic() - started:.0f}s")
    return answer, answer


def format_transcript(table) -> str:
    """Flatten the results table into one copy-pasteable block."""
    if table is None or len(table) == 0:
        return ""
    blocks = []
    for index, row in enumerate(pd.DataFrame(table).to_dict("records"), start=1):
        blocks.append(
            f"### {index}. {row.get('Task ID', '')}\n"
            f"Q: {row.get('Question', '')}\n"
            f"A: {row.get('Submitted Answer', '')}"
        )
    return "\n\n".join(blocks)


def run_and_submit_all( profile: gr.OAuthProfile | None, model_id: str, concurrency: int):
    """
    Fetches all questions, runs the BasicAgent on them, submits all answers,
    and displays the results.

    Questions are answered by a pool of worker threads, and progress is yielded
    as answers land so the browser keeps receiving data instead of waiting on
    one long-running request.
    """
    # --- Determine HF Space Runtime URL and Repo URL ---
    space_id = os.getenv("SPACE_ID") # Get the SPACE_ID for sending link to the code

    if profile:
        username= f"{profile.username}"
        print(f"User logged in: {username}")
    else:
        print("User not logged in.")
        yield "Please Login to Hugging Face with the button.", None
        return

    api_url = DEFAULT_API_URL
    questions_url = f"{api_url}/questions"
    submit_url = f"{api_url}/submit"

    # 1. Instantiate Agent ( modify this part to create your agent)
    model_id = (model_id or "").strip() or None
    workers = max(1, int(concurrency))
    worker_rpm = max(1, TOTAL_REQUESTS_PER_MINUTE // workers)
    try:
        # Fail fast on bad model configuration; workers build their own agents.
        resolved_model = BasicAgent(model_id=model_id).model_id
    except Exception as e:
        print(f"Error instantiating agent: {e}")
        yield f"Error initializing agent: {e}", None
        return
    # In the case of an app running as a hugging Face space, this link points toward your codebase ( usefull for others so please keep it public)
    agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
    print(agent_code)

    # 2. Fetch Questions
    print(f"Fetching questions from: {questions_url}")
    yield f"Model: {resolved_model}\nFetching questions...", None
    try:
        response = requests.get(questions_url, timeout=15)
        response.raise_for_status()
        questions_data = response.json()
        if not questions_data:
             print("Fetched questions list is empty.")
             yield "Fetched questions list is empty or invalid format.", None
             return
        print(f"Fetched {len(questions_data)} questions.")
    except requests.exceptions.RequestException as e:
        print(f"Error fetching questions: {e}")
        yield f"Error fetching questions: {e}", None
        return
    except requests.exceptions.JSONDecodeError as e:
         print(f"Error decoding JSON response from questions endpoint: {e}")
         print(f"Response text: {response.text[:500]}")
         yield f"Error decoding server response for questions: {e}", None
         return
    except Exception as e:
        print(f"An unexpected error occurred fetching questions: {e}")
        yield f"An unexpected error occurred fetching questions: {e}", None
        return

    # 3. Run your Agent
    answers_payload = []
    questions = [
        item for item in questions_data
        if item.get("task_id") and item.get("question") is not None
    ]
    skipped = len(questions_data) - len(questions)
    if skipped:
        print(f"Skipped {skipped} item(s) with missing task_id or question.")
    total = len(questions)
    rows: list[dict | None] = [None] * total
    started_all = time.monotonic()
    run_header = f"Model: {resolved_model} | {workers} parallel worker(s)"
    print(f"Running agent on {total} questions. {run_header}")
    yield f"{run_header}\nRunning agent on {total} questions...", None

    with ThreadPoolExecutor(max_workers=workers) as pool:
        futures = {
            pool.submit(answer_question, item, model_id, worker_rpm): index
            for index, item in enumerate(questions)
        }
        for done, future in enumerate(as_completed(futures), start=1):
            index = futures[future]
            item = questions[index]
            answer, display = future.result()
            if answer is not None:
                answers_payload.append({"task_id": item["task_id"], "submitted_answer": answer})
            rows[index] = {
                "Task ID": item["task_id"],
                "Question": item["question"],
                "Submitted Answer": display,
            }
            yield (
                f"{run_header}\n"
                f"Answered {done}/{total} questions "
                f"({time.monotonic() - started_all:.0f}s elapsed)...",
                pd.DataFrame([row for row in rows if row]),
            )

    results_log = [row for row in rows if row]
    print(f"All {total} questions finished in {time.monotonic() - started_all:.0f}s")

    if not answers_payload:
        print("Agent did not produce any answers to submit.")
        yield "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
        return

    # 4. Prepare Submission 
    submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
    _last_run[username] = {
        "submission_data": submission_data,
        "results_log": results_log,
        "model": resolved_model,
    }
    status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
    print(status_update)
    yield status_update, pd.DataFrame(results_log)

    # 5. Submit
    yield from submit_with_retry(submit_url, submission_data, results_log, resolved_model)


def submit_with_retry(submit_url: str, submission_data: dict, results_log: list, resolved_model: str):
    """POST the answers, retrying server-side failures, which the scoring API returns often."""
    print(f"Submitting {len(submission_data['answers'])} answers to: {submit_url}")
    results_df = pd.DataFrame(results_log)
    last_error = "Submission Failed: no attempt was made."
    for attempt in range(1, SUBMIT_ATTEMPTS + 1):
        try:
            response = requests.post(submit_url, json=submission_data, timeout=60)
            response.raise_for_status()
            result_data = response.json()
            final_status = (
                f"Submission Successful!\n"
                f"User: {result_data.get('username')}\n"
                f"Model: {resolved_model}\n"
                f"Overall Score: {result_data.get('score', 'N/A')}% "
                f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
                f"Message: {result_data.get('message', 'No message received.')}"
            )
            print("Submission successful.")
            yield final_status, results_df
            return
        except requests.exceptions.HTTPError as e:
            error_detail = f"Server responded with status {e.response.status_code}."
            try:
                error_json = e.response.json()
                error_detail += f" Detail: {error_json.get('detail', e.response.text)}"
            except requests.exceptions.JSONDecodeError:
                error_detail += f" Response: {e.response.text[:500]}"
            last_error = f"Submission Failed: {error_detail}"
            # 4xx means the payload is wrong, so retrying cannot help.
            retryable = e.response.status_code >= 500
        except requests.exceptions.Timeout:
            last_error = "Submission Failed: The request timed out."
            retryable = True
        except requests.exceptions.RequestException as e:
            last_error = f"Submission Failed: Network error - {e}"
            retryable = True
        except Exception as e:
            last_error = f"An unexpected error occurred during submission: {e}"
            retryable = False

        print(last_error)
        if not retryable or attempt == SUBMIT_ATTEMPTS:
            break
        delay = SUBMIT_RETRY_DELAY * attempt
        yield (
            f"{last_error}\n\nThe scoring server is flaky; retrying in {delay}s "
            f"(attempt {attempt + 1}/{SUBMIT_ATTEMPTS}). Your answers are kept below.",
            results_df,
        )
        time.sleep(delay)

    yield (
        f"{last_error}\n\nYour answers are preserved below - use 'Resubmit last answers' "
        "to try again without re-running the agent.",
        results_df,
    )


def resubmit_last_answers(profile: gr.OAuthProfile | None):
    if not profile:
        yield "Please Login to Hugging Face with the button.", None
        return
    cached = _last_run.get(profile.username)
    if not cached:
        yield "No previous run found in this session. Run the evaluation first.", None
        return
    yield from submit_with_retry(
        f"{DEFAULT_API_URL}/submit",
        cached["submission_data"],
        cached["results_log"],
        cached["model"],
    )


# --- Build Gradio Interface using Blocks ---
with gr.Blocks() as demo:
    gr.Markdown("# Basic Agent Evaluation Runner")
    gr.Markdown(
        """
        **Instructions:**

        1.  Please clone this space, then modify the code to define your agent's logic, the tools, the necessary packages, etc ...
        2.  Log in to your Hugging Face account using the button below. This uses your HF username for submission.
        3.  Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see the score.

        ---
        **Disclaimers:**
        Once clicking on the "submit button, it can take quite some time ( this is the time for the agent to go through all the questions).
        This space provides a basic setup and is intentionally sub-optimal to encourage you to develop your own, more robust solution. For instance for the delay process of the submit button, a solution could be to cache the answers and submit in a seperate action or even to answer the questions in async.
        """
    )

    gr.Markdown(describe_backend())

    gr.LoginButton()

    model_dropdown = gr.Dropdown(
        choices=list(anthropic_model_choices()),
        value=default_anthropic_model(),
        allow_custom_value=True,
        label="Model",
        info="Latest Anthropic Sonnet / Opus / Haiku (requires ANTHROPIC_API_KEY). Any other model id can be typed in.",
    )

    concurrency_slider = gr.Slider(
        minimum=1,
        maximum=10,
        step=1,
        value=CONCURRENCY,
        label="Parallel questions",
        info="Questions answered at the same time. Lower this if the provider returns rate limit errors.",
    )

    run_button = gr.Button("Run Evaluation & Submit All Answers")
    resubmit_button = gr.Button("Resubmit last answers")

    status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
    # Removed max_rows=10 from DataFrame constructor
    results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)

    transcript_output = gr.Textbox(
        label="Questions & answers (use the copy button to keep them for later)",
        lines=10,
        max_lines=20,
        interactive=False,
        show_copy_button=True,
    )

    results_table.change(
        fn=format_transcript,
        inputs=results_table,
        outputs=transcript_output,
    )

    run_button.click(
        fn=run_and_submit_all,
        inputs=[model_dropdown, concurrency_slider],
        outputs=[status_output, results_table]
    )

    resubmit_button.click(
        fn=resubmit_last_answers,
        outputs=[status_output, results_table]
    )

demo.queue(default_concurrency_limit=1)

if __name__ == "__main__":
    print("\n" + "-"*30 + " App Starting " + "-"*30)
    # Check for SPACE_HOST and SPACE_ID at startup for information
    space_host_startup = os.getenv("SPACE_HOST")
    space_id_startup = os.getenv("SPACE_ID") # Get SPACE_ID at startup

    if space_host_startup:
        print(f"✅ SPACE_HOST found: {space_host_startup}")
        print(f"   Runtime URL should be: https://{space_host_startup}.hf.space")
    else:
        print("ℹ️  SPACE_HOST environment variable not found (running locally?).")

    if space_id_startup: # Print repo URLs if SPACE_ID is found
        print(f"✅ SPACE_ID found: {space_id_startup}")
        print(f"   Repo URL: https://huggingface.co/spaces/{space_id_startup}")
        print(f"   Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
    else:
        print("ℹ️  SPACE_ID environment variable not found (running locally?). Repo URL cannot be determined.")

    print("-"*(60 + len(" App Starting ")) + "\n")

    print("Launching Gradio Interface for Basic Agent Evaluation...")
    demo.launch(debug=True, share=False)