`750735d`

fix: approval bar focus and threading issues

**Issues Fixed:**

1. **Approval bar never got focus**
- Set can_focus=True in __init__ and show_approval
- Approval bar can now receive keyboard events (Y, n, e)

2. **Threading deadlock**
- Changed call_later to call_from_thread (worker-safe)
- Added 300s timeout to prevent infinite hangs
- Properly hide bar after confirmation

3. **No visibility into confirmation flow**
- Added extensive debug logging to trace:
- When bar is shown/focused
- When action methods are called
- When app handlers receive events
- When futures are resolved

**What This Fixes:**
- Tool calls now actually wait for user approval
- Pressing Y/n/e now works correctly
- No more 240s hangs with no output
- Approval bar properly shows, focuses, and responds

Authored by

espadonne 4 months ago

SHA: 750735df9283bf2907177149b8d1ddb9e2c4bd2b
Parents: a953bb4
Tree: 4c734a4

5 changed files

Status	File	+	-
A	`MODELS.md`	66	0
M	`README.md`	2	46
A	`build.sh`	4	0
M	`src/loader/ui/app.py`	45	4
M	`src/loader/ui/widgets/approval_bar.py`	33	1

MODELS.mdadded

 +# Recommended Ollama Models for Loader
++
 +## Currently Installed
++
 +| Model | Size |
 +|-------|------|
 +| qwen2.5-coder:14b | 9.0 GB |
 +| qwen2.5-coder:7b | 4.7 GB |
 +| qwen2.5:14b | 9.0 GB |
 +| deepseek-r1:14b | 9.0 GB |
 +| deepseek-coder-v2:16b | 8.9 GB |
 +| codellama:13b | 7.4 GB |
 +| gemma2:9b | 5.4 GB |
 +| mistral:7b | 4.4 GB |
 +| llama3.2:3b | 2.0 GB |
 +| phi3:mini | 2.2 GB |
++
 +## Models to Try Next
++
 +### Heavy Hitters (best quality, needs more VRAM)
++
 +| Model | Size | Why |
 +|-------|------|-----|
 +| `qwen2.5-coder:32b` | ~20GB | Best open coding model, rivals GPT-4 on benchmarks |
 +| `deepseek-r1:32b` | ~20GB | Larger reasoning model, even better multi-step logic |
 +| `codestral:22b` | ~13GB | Mistral's dedicated coding model, excellent tool use |
 +| `llama3.3:70b` | ~40GB | Meta's flagship, state-of-the-art instruction following |
++
 +### Mid-Size Sweet Spot
++
 +| Model | Size | Why |
 +|-------|------|-----|
 +| `starcoder2:15b` | ~9GB | BigCode's latest, trained on massive code corpus |
 +| `granite-code:20b` | ~12GB | IBM's code model, strong at enterprise patterns |
 +| `yi-coder:9b` | ~5.5GB | 01.AI's coding model, great at code completion |
 +| `phi4:14b` | ~8GB | Microsoft's latest, punches above its weight |
++
 +### Lightweight Speed Demons
++
 +| Model | Size | Why |
 +|-------|------|-----|
 +| `llama3.3:latest` | ~4.5GB | Latest Llama with improved instruction following |
 +| `qwen2.5-coder:3b` | ~2GB | Tiny but surprisingly capable for quick tasks |
 +| `deepseek-r1:7b` | ~4.7GB | Reasoning in a smaller package |
 +| `codegemma:7b` | ~5GB | Google's code-specific Gemma variant |
++
 +## Pull Commands
++
 +```bash
 +# Heavy hitters (if you have the VRAM)
 +ollama pull qwen2.5-coder:32b
 +ollama pull deepseek-r1:32b
 +ollama pull codestral:22b
++
 +# Mid-size (recommended next pulls)
 +ollama pull starcoder2:15b
 +ollama pull granite-code:20b
 +ollama pull yi-coder:9b
 +ollama pull phi4:14b
++
 +# Lightweight
 +ollama pull llama3.3
 +ollama pull qwen2.5-coder:3b
 +ollama pull deepseek-r1:7b
 +ollama pull codegemma:7b
 +```

README.mdmodified

 -# Loader
+-
 -Local agentic coding assistant. Runs on your hardware with local LLMs.
+-
 -## Install
+-
 -```bash
 -pip install -e .
 -```
+-
 -## Requirements
+-
 -- Python 3.11+
 -- Ollama running (`ollama serve`)
 -- A model pulled (`ollama pull llama3.1:8b`)
+-
 -## Usage
+-
 -```bash
 -# Interactive mode
 -loader
+-
 -# Single prompt
 -loader "Read main.py and explain it"
+-
 -# Skip confirmation prompts
 -loader -y "Create a hello.py file"
+-
 -# Use different model
 -loader -m qwen2.5:7b
 -```
+-
 -**In interactive mode:** type prompts, `clear` to reset, `exit` to quit.
+-
 -## Tools
+-
 -- `read` - read files
 -- `write` - write files
 -- `edit` - find/replace in files
 -- `glob` - find files by pattern
 -- `grep` - search file contents
 -- `bash` - run shell commands
+-
 -## License
+-
 -MIT
 +# FortranGoingOnForty
 +A tutorial on using Fortran for beginners.

build.shadded

 +#!/bin/bash
 +git checkout main
 +npm install
 +npm run build

src/loader/ui/app.pymodified

          self._pending_confirmation = loop.create_future()
          self._pending_command = details
 -        # Show the approval bar
 +        # Show the approval bar - must use call_from_thread since we're in worker
          approval_bar = self.query_one("#approval-bar", ApprovalBar)
 -        self.call_later(lambda: approval_bar.show_approval(tool_name, message, details))
 -        # Wait for user response
 +        def show_bar():
 +            try:
 +                approval_bar.show_approval(tool_name, message, details)
 +                # Debug log
 +                with open("/tmp/loader_debug.log", "a") as f:
 +                    f.write(f"[approval] Bar shown, waiting for user input\n")
 +            except Exception as e:
 +                with open("/tmp/loader_debug.log", "a") as f:
 +                    f.write(f"[approval] Error showing bar: {e}\n")
++
 +        self.call_from_thread(show_bar)
++
 +        # Wait for user response with timeout
          try:
 -            return await self._pending_confirmation
 +            result = await asyncio.wait_for(self._pending_confirmation, timeout=300.0)
 +            try:
 +                with open("/tmp/loader_debug.log", "a") as f:
 +                    f.write(f"[approval] Got result: {result}\n")
 +            except Exception:
 +                pass
 +            return result
 +        except asyncio.TimeoutError:
 +            try:
 +                with open("/tmp/loader_debug.log", "a") as f:
 +                    f.write(f"[approval] Timeout waiting for user\n")
 +            except Exception:
 +                pass
 +            return False
          finally:
              self._pending_confirmation = None
              self._pending_command = ""
 +            # Hide the bar
 +            self.call_from_thread(approval_bar.hide_approval)
      def on_approval_bar_approved(self, event: ApprovalBar.Approved) -> None:
          """Handle approval from the bar."""
 +        try:
 +            with open("/tmp/loader_debug.log", "a") as f:
 +                f.write(f"[approval] Approved handler called\n")
 +        except Exception:
 +            pass
          if self._pending_confirmation and not self._pending_confirmation.done():
              self._pending_confirmation.set_result(True)
      def on_approval_bar_rejected(self, event: ApprovalBar.Rejected) -> None:
          """Handle rejection from the bar."""
 +        try:
 +            with open("/tmp/loader_debug.log", "a") as f:
 +                f.write(f"[approval] Rejected handler called\n")
 +        except Exception:
 +            pass
          if self._pending_confirmation and not self._pending_confirmation.done():
              self._pending_confirmation.set_result(False)
          # Refocus input
      def on_approval_bar_edit_requested(self, event: ApprovalBar.EditRequested) -> None:
          """Handle edit request - put command in input for editing."""
 +        try:
 +            with open("/tmp/loader_debug.log", "a") as f:
 +                f.write(f"[approval] Edit handler called\n")
 +        except Exception:
 +            pass
          if self._pending_confirmation and not self._pending_confirmation.done():
              self._pending_confirmation.set_result(False)
          # Put the command in the input field for editing

src/loader/ui/widgets/approval_bar.pymodified

          self._tool_name: str = ""
          self._command_preview: str = ""
          self._full_command: str = ""
 +        # Make this widget focusable from the start
 +        self.can_focus = True
      def compose(self) -> ComposeResult:
          with Horizontal(id="approval-container"):
              preview = preview[:57] + "..."
          preview_label.update(preview)
 -        # Show the bar
 +        # Show the bar and focus it
          self.add_class("visible")
 +        self.can_focus = True  # Make sure it can receive focus
++
 +        # Debug logging
 +        try:
 +            with open("/tmp/loader_debug.log", "a") as f:
 +                f.write(f"[approval-bar] show_approval: tool={tool_name}, visible=True, focusing...\n")
 +        except Exception:
 +            pass
++
          self.focus()
 +        try:
 +            with open("/tmp/loader_debug.log", "a") as f:
 +                f.write(f"[approval-bar] focus() called, has_focus={self.has_focus}\n")
 +        except Exception:
 +            pass
++
      def hide_approval(self) -> None:
          """Hide the approval bar."""
          self.remove_class("visible")
      def action_approve(self) -> None:
          """Handle 'y' key - approve the action."""
 +        try:
 +            with open("/tmp/loader_debug.log", "a") as f:
 +                f.write(f"[approval-bar] action_approve called, posting Approved message\n")
 +        except Exception:
 +            pass
          self.post_message(self.Approved())
          self.hide_approval()
      def action_reject(self) -> None:
          """Handle 'n' or escape - reject the action."""
 +        try:
 +            with open("/tmp/loader_debug.log", "a") as f:
 +                f.write(f"[approval-bar] action_reject called, posting Rejected message\n")
 +        except Exception:
 +            pass
          self.post_message(self.Rejected())
          self.hide_approval()
      def action_edit(self) -> None:
          """Handle 'e' key - edit the command."""
 +        try:
 +            with open("/tmp/loader_debug.log", "a") as f:
 +                f.write(f"[approval-bar] action_edit called, posting EditRequested message\n")
 +        except Exception:
 +            pass
          self.post_message(self.EditRequested(self._full_command))
          self.hide_approval()