lineapy.system_tracing package

Submodules

lineapy.system_tracing.exec_and_record_function_calls module

lineapy.system_tracing.exec_and_record_function_calls.exec_and_record_function_calls(code: code, globals_: Dict[str, object]) → TraceFunc[source]

Execute the code while recording all the function calls which originate from the code object.

lineapy.system_tracing.function_call module

class lineapy.system_tracing.function_call.FunctionCall(fn: ~typing.Callable, args: ~typing.List[~typing.Any] = <factory>, kwargs: ~typing.Dict[str, ~typing.Any] = <factory>, res: ~typing.Optional[~typing.Any] = None)[source]

A record of a function call that happened in the tracer.

lineapy.system_tracing.function_calls_to_side_effects module

lineapy.system_tracing.function_calls_to_side_effects.function_calls_to_side_effects(function_inspector: FunctionInspector, function_calls: Iterable[FunctionCall], input_nodes: Mapping[LineaID, object], output_globals: Mapping[str, object]) → Iterable[Union[MutatedNode, ViewOfNodes, AccessedGlobals, ImplicitDependencyNode]][source]

Translates a list of function calls to a list of side effects, by mapping objects to nodes.

Parameters

  • function_inspector – The function inspector to use to lookup what side effects each function call has.

  • function_calls – The function calls that were recorded.

  • input_nodes – Mapping of node ID to value for all the nodes that were passed in to this execution.

  • output_globals – Mapping of global identifier to the value of all globals that were set during this execution.

Module contents

This module adds support for using sys.settrace to understand what happens during a subset of code execution that’s passed in. In the context of how it’s currently used, it’s limited to the “blackbox” execs—l_exec_statement It can be used used and tested independently.

At a high level, users could:

  1. Use the exec_and_record_function_calls.py as an entry point and uses sys.settrace to trace every bytecode execution and _op_stack.py to look at the bytecode stack during tracing. It translates different bytecode instructions into the corresponding Python function calls.

  2. Use function_calls_to_side_effects.py to translate the sequence of calls that were recorded into the side effects produced on nodes (mapping Python changes to graph changes).