Skip to content
← REPL Enhancements Testing Framework →

Debugging Tools for Dana

(This document is a placeholder. It will focus on the specific tooling aspects for debugging Dana programs, complementing the conceptual design in ../../02_dana_runtime_and_execution/debugging_profiling.md. This includes command-line debuggers, IDE integration specifics, and visualization tools.)

Key Tooling Aspects for Debugging:

  • Command-Line Debugger (Conceptual):
  • Features: Setting breakpoints, stepping (over, in, out), inspecting variables and SandboxContext scopes, evaluating expressions in current context, viewing call stacks.
  • Invocation: How to start a Dana script under the debugger.
  • Commands: Syntax for debugger commands (e.g., b <line|func>, n, s, c, p <expr>, bt, up, down).

  • Interaction with REPL: Can the debugger be invoked from or drop into a REPL state?

  • VS Code Debugger Integration (via Debug Adapter Protocol - DAP):

  • Implementing a Dana Debug Adapter.
  • Mapping DAP requests (e.g., setBreakpoints, next, stepIn, evaluate) to Dana debugger functionalities.
  • Displaying variables, call stack, and context within the VS Code UI.

  • Configuration (launch.json) for debugging Dana scripts.

  • Log Analysis Tools:

  • Utilities for filtering, searching, and visualizing structured logs generated by log() statements or system components.

  • Highlighting errors or specific log levels.

  • SandboxContext Visualization/Inspection Tools:

  • A tool (perhaps REPL magic command or standalone utility) to dump or interactively explore the SandboxContext at a given point (e.g., during a debugging session or post-mortem).

  • Visualizing scope hierarchy and variable states.

  • IPV Execution Tracing Tools:

  • Tools to visualize or inspect the execution flow through the INFER, PROCESS, and VALIDATE phases of IPV-enabled functions.
  • Displaying the IPVCallContext at each stage.

  • Logging inputs/outputs of LLM calls made during IPV.

  • Execution Path Visualization (Advanced):

  • Graphical representation of the execution flow of a Dana script, especially for complex orchestrations or agent decision trees.

  • Highlighting active paths, decision points, and resource interactions.

  • Post-Mortem Debugging Tools:

  • Saving SandboxContext state on unhandled exceptions.
  • Tools to load and inspect this saved state to understand the cause of a crash.

Considerations for Debugging LLM Interactions:

  • Debugging prompts sent to LLMs.
  • Inspecting raw LLM responses.
  • Tracing how LLM outputs are parsed and validated (especially in IPV validate_phase).
  • Handling non-determinism if the same LLM call can produce different results.

Self-reflection: Effective debugging tools are paramount for a productive developer experience. For Dana, this means not only traditional code debugging but also tools to understand the state and flow of context, resource interactions, and LLM calls.