turso_codebase_exploration.md

turso_codebase_exploration.md

You are GPT-5.3 Codex working inside the turso monorepo. Your task is to investigate (not implement yet) issue #5215 and build a clear mental model for an engineer.

Issue:

• tursodatabase/turso#5215
• Problem statement: In non-interactive CLI mode, query failures still return exit code 0; desired behavior is to return exit code 1 (SQLite-like behavior).

Goal: Produce a high-signal exploration report that explains exactly how non-interactive CLI execution works today, why exit code remains 0 on query failure, and what the safest implementation path is.

Scope:

• Focus on non-interactive CLI mode.
• Prioritize these likely files first:
  • cli/*
  • testing/runner/src/backends/cli.rs
• Explore any additional files needed to trace control flow end-to-end.

What to do:

1. Reproduce and baseline behavior

   • Run a failing multi-statement non-interactive query against tursodb and capture exit code.
   • Run equivalent sqlite3 command and capture exit code.
   • Note differences (e.g., whether execution stops after first error, output format, stderr/stdout split).

2. Build the execution flow map

   • Trace CLI startup path in non-interactive mode from argument parsing to query execution.
   • Identify where errors are produced, transformed, logged, swallowed, or returned.
   • Identify where process exit code is currently chosen (or defaults to success).

3. Build the error-propagation model

   • For each relevant layer/function, document:
     • input error type
     • wrapping/conversion behavior
     • return type/result handling
     • whether it influences exit code
   • Call out any “error observed but success returned” branches.

4. Test surface analysis

   • Review existing tests related to non-interactive behavior and exit codes.
   • Explain what is already covered vs missing.
   • Identify best place(s) to add tests (unit/integration/runner) with rationale.

5. Design options (no code changes yet)

   • Propose 2–3 implementation approaches to enforce exit code 1 on query failure.
   • For each approach: touched files, complexity, regression risk, compatibility with current behavior, and recommendation.
   • Recommend the smallest correct change consistent with SQLite behavior.

6. Edge cases checklist Analyze expected behavior for:

   • Single failing statement
   • Multi-statement input with middle failure
   • Parse/prepare error vs runtime error
   • Empty input / no-op input
   • Interactive mode (note only; not primary target)
   • Runner/fuzzer invocation paths

Output format (strict): A) “Current behavior” (short) B) “Control-flow map” (step-by-step with function names + file paths) C) “Error propagation table” (layer -> behavior -> impact on exit code) D) “Behavior matrix” (sqlite3 vs tursodb for key scenarios) E) “Gaps in tests” F) “Recommended implementation strategy” (with tradeoffs) G) “Open questions / assumptions” H) “Concrete implementation checklist” (ordered, small steps)

Quality bar:

• Be evidence-driven: cite exact file paths and function names for each claim.
• Keep it concise but complete enough that an engineer can start implementing immediately.
• Do not modify code in this run unless explicitly asked.