Add structured error location reporting to execution errors (D128)

[isaacbrodsky]

Summary

This change adds structured error location information (where) to all execution error responses, allowing the UI to highlight the exact line of user code that failed. Both the subprocess and in-process execution paths now report errors uniformly with file, line number, function name, and source code snippet.

Key Changes

• Error response shape: Added where field to error objects in all execution paths. where is either null (for harness/infrastructure errors) or a dict with {file, line, func, source} pointing to the deepest user-code frame.

• Shared error diagnostics (_binding.py):

  • trim_harness_frames(): Strips leading runner/bootstrap frames from tracebacks so user code is prominent
  • user_location(): Extracts the deepest frame in the user's script as a structured location dict, handling both runtime exceptions and SyntaxErrors

• Subprocess execution (_child.py):

  • Uses trim_harness_frames() and user_location() to build clean tracebacks and error locations
  • Formats exceptions with format_exception_only() for harness errors (no stack) and format_exception() for user errors (with trimmed stack)

• In-process execution (executor.py):

  • Mirrors subprocess behavior for consistency
  • Applies same frame trimming and location extraction to in-process helper errors

• Fused engine (engine.py):

  • Added _user_location() to parse where from already-formatted traceback strings (text-based counterpart to live exception handling)
  • Extracts location from both runtime frames and SyntaxError display blocks

• UI updates (runtime.js, template.html):

  • Display where information prominently above the full traceback
  • Shows file, line, function name, and source code snippet for quick error identification

• Tests (test_executor.py, test_engine.py):

  • Comprehensive test coverage for error location extraction across different error types
  • Validates that harness errors have where=null and user errors point to the correct frame

Implementation Details

• Both execution paths (subprocess and in-process) use identical logic from _binding.py to ensure consistent error reporting
• Harness-raised errors (missing params, missing main, unserializable return) have where=null and no stack trace
• User-code errors point at the deepest frame in the user's file, even when the error originated in a library
• SyntaxErrors are handled specially since they don't produce runtime frames—location comes from the exception object itself
• Frozen importlib bootstrap frames are filtered out to reduce noise in tracebacks

https://claude.ai/code/session_01Ad7XvN7x4887vZoe4VP24h

---

Note

Low Risk
Additive error payload and UI-only display changes; existing consumers that ignore where keep working, with broad test coverage on both execution paths.

Overview
Python run failures now return a structured where (file, line, func, source) on the deepest frame in the user script, with traceback trimmed so it no longer leads with _child.py / executor harness frames. Harness-only failures (bad params, missing main, timeouts, missing file) keep where: null and often a message-only traceback.

Shared helpers trim_harness_frames and user_location in _binding.py drive both the subprocess worker and in-process executor; the fused engine adds _user_location to parse the same shape from cleaned traceback text. Specs, architecture notes, and authoring docs are updated for the wire shape (PY-14 / RH-3).

runtime.js and the API inspector template headline the failing line when where is present, collapse the full traceback behind Show traceback (sr-only so select-all still copies it), and add Copy error. New tests/test_executor.py and engine tests cover deepest-frame semantics, library errors blaming the caller, syntax errors, and harness cases.

^{Reviewed by Cursor Bugbot for commit 31bc606. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Coverage report

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
fused_render
_binding.py					36-41, 54-70, 75
_child.py					23-34, 65-77
engine.py
executor.py					146-155
server.py					3737
Project Total

This report was generated by python-coverage-comment-action

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 214be36. Configure here.}