QUERY · ISSUE

sys.settrace issue (bug?)

openby wolfc01opened 2021-02-03updated 2024-09-13

needs-info

Hi,

when I have the following code

def test():
	for a in range(10):
		if a == 7:
			print("it worked")	
			print("again")	
			print("or not.")	
			
test()

then sys.settrace fires at the line containing print("or not.") even if the condition a==7 is False.

for the debugger I am writing, see https://github.com/wolfc01/micropython_debugger/blob/main/debugger.py

Probably this has to do with entering or leaving some local scope?

When I run the very same debugger and example in python 3, above effect does not occur.

Best regards
Carl.

CANDIDATE · PULL REQUEST

Profiling: sys.settrace implementation.

closedby malinahopened 2018-10-24updated 2019-08-30

This is a working concept of profiling API related to issue #3864.

Please review this pull-request and comment especially on the overall integration into the Micropython.

So far the main concern of the profiler is runtime instruction tracing used to create detailed code coverage reports. To achieve this the analogue of sys.settrace was implemented and puts Micropython closer to feature a debugger #3009.

Building and Running

So far only the Unix port is concerned.

Enable profiling in ports/unix/mpconfigport.h like so:

#define MICROPY_PY_SYS_SETTRACE (1)

MICROPY_PERSISTENT_CODE_SAVE is required by the MICROPY_PY_SYS_TRACE.

Python API

`sys.settrace(tracefunc)`

As defined in original Python3 documentation sys.settrace(tracefunc). The Frame object and Code objects are only partially implemented.

Assessment

S Sonnet · high

Both items concern sys.settrace, but they address different aspects. PR #4265 is the initial feature implementation of sys.settrace in MicroPython — it adds the API and infrastructure (Frame/Code objects, bytecode tracing hooks, vm.c integration). Issue #6836 is a bug report about incorrect line number reporting in an existing sys.settrace implementation: the trace callback fires on unreachable lines when a conditional block is skipped, diverging from CPython behavior. The candidate PR implements the feature but does not address the line-number accuracy problem described in the query. It provides useful context (the implementation choices in vm.c and profiling.c may be the root cause of the bug) but leaves the query's problem open.

Suggested action

Link PR #4265 in issue #6836 as the originating implementation for context. The bug likely resides in how vm.c emits line trace events — investigate the line-number tracking logic introduced in that PR (py/vm.c, py/profiling.c) to determine whether line events are emitted for bytecode at the start of a block regardless of whether control flow actually reaches those instructions.