Skip to content

interpreter.md

Latest commit

 

History

History
38 lines (24 loc) · 1.81 KB

interpreter.md

File metadata and controls

38 lines (24 loc) · 1.81 KB

Interpreter

The interpreter executes RoboProgram (from RoboExecutable) as a plain C# loop over fake IL. It does not JIT to CLR IL or use reflection for user code.

Public entry points

  • RoboInterpreter: Run-to-completion API over a program, world, and TextWriter targets for stdout/stderr.
  • RoboInterpreterSession: Holds engine state for stepping / sessions (used when hosts need pause boundaries).

Shared execution logic lives in RoboInterpreterEngine (internal): initialize, then ExecuteNext for one instruction.

Contract (summary)

Matches v1-runtime-spec.md:

  • Deterministic: No threads, timers, or hidden randomness in v1.
  • Result-driven: Normal failures are RuntimeFault / ExecutionResult, not thrown exceptions for user-level mistakes (some defensive InvalidOperationException paths may still map to faults in the engine).

One step

Each ExecuteNext:

  1. Bounds-check the current instruction pointer.
  2. Fetch the current Instruction.
  3. Dispatch on RoboOpcode.
  4. Mutate evaluation stack, locals, heap arrays, world, or writers as defined for that opcode (il-instruction-set.md).
  5. Return null to continue, ExecutionResult.Completed when the call stack is drained, or ExecutionResult.Failed on fault.

Implementation vs v1 spec

The v1 spec describes one evaluation stack per call frame. The current engine uses a single shared stack with frame metadata; behavior remains stepable and teaching-friendly, but deep stack inspection UIs should be aware of the difference when showing “operand stack per frame.”

Related