hte-cli 0.2.8__py3-none-any.whl → 0.2.16__py3-none-any.whl

hte_cli/cli.py

@@ -311,41 +311,20 @@ def session_join(ctx, session_id: str, force_setup: bool):
                     continue
 
                 # Need to pull - show progress
-                last_status = ["connecting..."]  # Use list for closure mutability
+                last_status = ["connecting..."]
                 with console.status(f"[yellow]↓[/yellow] {short_name} [dim]connecting...[/dim]") as status:
                     def show_progress(image: str, line: str):
-                        # Parse docker pull output for layer progress
-                        # Lines look like: "abc123: Downloading [====>   ] 10MB/50MB"
-                        # Or: "abc123: Extracting [====>   ] 10MB/50MB"
-                        # Or: "abc123: Pull complete", "Digest: sha256:...", "Status: ..."
-                        display_text = None
+                        # Show docker output directly - includes MB progress from PTY
+                        # Lines look like: "abc123: Downloading  360.9MB/4.075GB"
                         if ": " in line:
                             parts = line.split(": ", 1)
                             if len(parts) == 2:
-                                layer_id = parts[0][-8:]  # Last 8 chars of layer ID
-                                layer_status = parts[1]  # Don't truncate - keep full progress
-                                # Show progress bars for Downloading/Extracting with MB info
-                                if "Downloading" in layer_status or "Extracting" in layer_status:
-                                    # Keep progress: "[====>   ] 10.5MB/50MB"
-                                    display_text = f"{layer_id}: {layer_status[:50]}"
-                                elif "Pull complete" in layer_status:
-                                    display_text = f"{layer_id}: done"
-                                elif "Download complete" in layer_status:
-                                    display_text = f"{layer_id}: download done"
-                                elif "Already exists" in layer_status:
-                                    display_text = f"{layer_id}: cached"
-                                elif "Waiting" in layer_status:
-                                    display_text = f"{layer_id}: waiting"
-                                elif "Verifying" in layer_status:
-                                    display_text = f"{layer_id}: verifying"
-                                else:
-                                    display_text = line[:55]
-                        elif line.strip():
-                            display_text = line[:55]
-
-                        if display_text and display_text != last_status[0]:
-                            last_status[0] = display_text
-                            status.update(f"[yellow]↓[/yellow] {short_name} [dim]{display_text}[/dim]")
+                                layer_id = parts[0][-8:]
+                                layer_status = parts[1][:45]
+                                display = f"{layer_id}: {layer_status}"
+                                if display != last_status[0]:
+                                    last_status[0] = display
+                                    status.update(f"[yellow]↓[/yellow] {short_name} [dim]{display}[/dim]")
 
                     success = pull_image_with_progress(img, on_progress=show_progress)
 
@@ -375,6 +354,15 @@ def session_join(ctx, session_id: str, force_setup: bool):
         console.print("[dim]Skipping setup (use --force-setup to re-run)[/dim]")
         console.print()
 
+    # Check if session was cancelled during setup
+    try:
+        updated_session = api.join_session(session_id)
+        if updated_session.get("status") == "cancelled":
+            console.print("[yellow]Session was cancelled. Exiting.[/yellow]")
+            sys.exit(0)
+    except APIError:
+        pass  # Continue if we can't check - server might be temporarily unavailable
+
     # Step 4: Show instructions
     if session_info.get("instructions"):
         console.print(Panel(session_info["instructions"], title="Task Instructions"))
@@ -383,18 +371,19 @@ def session_join(ctx, session_id: str, force_setup: bool):
     # Step 3: Run the task using TaskRunner
     step_num = "3" if (not is_reconnect or force_setup) and images else "2" if (not is_reconnect or force_setup) else "1"
     console.print(f"[bold]Step {step_num}:[/bold] Starting task environment...")
+    console.print("[dim]Launching Docker containers...[/dim]")
+    console.print()
 
     events.docker_started()
 
     runner = TaskRunner()
     eval_log_bytes = None
     try:
-        with console.status("[dim]Launching Docker containers (this may take a minute)...[/dim]"):
-            result = runner.run_from_assignment(
-                assignment=assignment,
-                compose_yaml=compose_yaml,
-                files_zip=files_zip,
-            )
+        result = runner.run_from_assignment(
+            assignment=assignment,
+            compose_yaml=compose_yaml,
+            files_zip=files_zip,
+        )
         # Read eval log before cleanup
         if result.eval_log_path and result.eval_log_path.exists():
             eval_log_bytes = result.eval_log_path.read_bytes()
@@ -439,7 +428,7 @@ def session_join(ctx, session_id: str, force_setup: bool):
                     eval_log_bytes=eval_log_bytes,
                     score=result.score,
                     score_binarized=result.score_binarized,
-                    agent_id=result.agent_id,
+                    agent_id=None,  # TODO: extract from task files if needed
                 )
             except APIError as e:
                 console.print(f"[red]Failed to upload result: {e}[/red]")

hte_cli/image_utils.py

@@ -1,6 +1,10 @@
 """Docker image utilities for pre-pulling compose images."""
 
 import logging
+import os
+import pty
+import re
+import select
 import subprocess
 from collections.abc import Callable
 
@@ -61,32 +65,77 @@ def pull_image_with_progress(
     on_complete: Callable[[str, bool], None] | None = None,
 ) -> bool:
     """
-    Pull a Docker image with progress callbacks.
+    Pull a Docker image with progress callbacks using PTY for real progress output.
 
     Args:
         image: Image name to pull
-        on_progress: Callback(image, status_line) called for each line of output
+        on_progress: Callback(image, status_line) called for each progress update
         on_complete: Callback(image, success) called when pull completes
 
     Returns:
         True if pull succeeded, False otherwise
     """
     try:
+        # Use PTY to get real progress output from docker
+        master_fd, slave_fd = pty.openpty()
+
         process = subprocess.Popen(
             ["docker", "pull", image],
-            stdout=subprocess.PIPE,
-            stderr=subprocess.STDOUT,
-            text=True,
-            bufsize=1,
+            stdout=slave_fd,
+            stderr=slave_fd,
+            stdin=slave_fd,
+            close_fds=True,
         )
 
-        # Stream output line by line
-        for line in iter(process.stdout.readline, ""):
-            line = line.strip()
-            if line and on_progress:
-                on_progress(image, line)
+        os.close(slave_fd)  # Close slave in parent
+
+        # Read output from master with timeout
+        output_buffer = ""
+        # Regex to parse docker progress: "abc123: Downloading [===>  ] 10.5MB/50MB"
+        progress_pattern = re.compile(
+            r"([a-f0-9]+):\s*(Downloading|Extracting|Verifying Checksum|Download complete|Pull complete|Already exists|Waiting)(?:\s+\[.*?\]\s+)?(\d+\.?\d*\s*[kMG]?B)?(?:/(\d+\.?\d*\s*[kMG]?B))?"
+        )
 
-        process.wait()
+        while True:
+            # Check if process is done
+            ret = process.poll()
+            if ret is not None:
+                # Read any remaining output
+                try:
+                    while True:
+                        ready, _, _ = select.select([master_fd], [], [], 0.1)
+                        if not ready:
+                            break
+                        chunk = os.read(master_fd, 4096)
+                        if not chunk:
+                            break
+                except OSError:
+                    pass
+                break
+
+            # Read available output
+            try:
+                ready, _, _ = select.select([master_fd], [], [], 0.1)
+                if ready:
+                    chunk = os.read(master_fd, 4096)
+                    if chunk:
+                        output_buffer += chunk.decode("utf-8", errors="replace")
+
+                        # Parse and report progress
+                        # Docker uses carriage returns to update lines in place
+                        lines = output_buffer.replace("\r", "\n").split("\n")
+                        output_buffer = lines[-1]  # Keep incomplete line
+
+                        for line in lines[:-1]:
+                            line = line.strip()
+                            # Strip ANSI escape codes
+                            line = re.sub(r"\x1b\[[0-9;]*[a-zA-Z]", "", line)
+                            if line and on_progress:
+                                on_progress(image, line)
+            except OSError:
+                break
+
+        os.close(master_fd)
         success = process.returncode == 0
 
         if on_complete:
@@ -94,7 +143,7 @@ def pull_image_with_progress(
 
         return success
 
-    except (subprocess.TimeoutExpired, FileNotFoundError, OSError) as e:
+    except (FileNotFoundError, OSError) as e:
         logger.error(f"Failed to pull {image}: {e}")
         if on_complete:
             on_complete(image, False)

{hte_cli-0.2.8.dist-info → hte_cli-0.2.16.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hte-cli
-Version: 0.2.8
+Version: 0.2.16
 Summary: Human Time-to-Completion Evaluation CLI
 Project-URL: Homepage, https://github.com/sean-peters-au/lyptus-mono
 Author: Lyptus Research

{hte_cli-0.2.8.dist-info → hte_cli-0.2.16.dist-info}/RECORD

@@ -1,15 +1,15 @@
 hte_cli/__init__.py,sha256=fDGXp-r8bIoLtlQnn5xJ_CpwMhonvk9bGjZQsjA2mDI,914
 hte_cli/__main__.py,sha256=63n0gNGfskidWDU0aAIF2N8lylVCLYKVIkrN9QiORoo,107
 hte_cli/api_client.py,sha256=m42kfFZS72Nu_VuDwxRsLNy4ziCcvgk7KNWBh9gwqy0,9257
-hte_cli/cli.py,sha256=W9R_jHBLhLho2GyroKzCCg6EhBluCrFJdZ9zCaKFGuo,42745
+hte_cli/cli.py,sha256=XpV7x2HcHlVg-ynr-Ih8MxbHc74PT9mQzCICavtpV_0,41623
 hte_cli/config.py,sha256=42Xv__YMSeRLs2zhGukJkIXFKtnBtYCHnONfViGyt2g,3387
 hte_cli/errors.py,sha256=1J5PpxcUKBu6XjigMMCPOq4Zc12tnv8LhAsiaVFWLQM,2762
 hte_cli/events.py,sha256=Zn-mroqaLHNzdT4DFf8st1Qclglshihdc09dBfCN070,5522
-hte_cli/image_utils.py,sha256=454yoZEI1duNYrZC8UjhfZzDRP4Nxdrf2TvnZ_54G1k,4439
+hte_cli/image_utils.py,sha256=TLwJdswUQrSD2bQcAXW03R8j8WG2pbHzd12TWcE7zy4,6418
 hte_cli/runner.py,sha256=DhC8FMjHwfLR193iP4thLDRZrNssYA9KH1WYKU2JKeg,13535
 hte_cli/scorers.py,sha256=sFoPJePRt-K191-Ga4cVmrldruJclYXTOLkU_C9nCDI,6025
 hte_cli/version_check.py,sha256=WVZyGy2XfAghQYdd2N9-0Qfg-7pgp9gt4761-PnmacI,1708
-hte_cli-0.2.8.dist-info/METADATA,sha256=XT1WLIfvC2PRQtcFB19qTpBdMApP_UKabzal-Fcn8Cw,3767
-hte_cli-0.2.8.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-hte_cli-0.2.8.dist-info/entry_points.txt,sha256=XbyEEi1H14DFAt0Kdl22e_IRVEGzimSzYSh5HlhKlFA,41
-hte_cli-0.2.8.dist-info/RECORD,,
+hte_cli-0.2.16.dist-info/METADATA,sha256=pNGPVDfi_RV-IWuC3pGzrqJCUzXgA2rf9-jXI2tNk3s,3768
+hte_cli-0.2.16.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+hte_cli-0.2.16.dist-info/entry_points.txt,sha256=XbyEEi1H14DFAt0Kdl22e_IRVEGzimSzYSh5HlhKlFA,41
+hte_cli-0.2.16.dist-info/RECORD,,