Celery's rdb is not compatible with pdbpp > 0.10

[Hugo-C]

Hello,

The Celery project has a remote debbuger called rdb. It was working fine with pdbpp up to 0.10.3. 0.11.5 (the next available release on pypi), breaks with the following exception:

Traceback (most recent call last):
File "/path/to/src/replicate_rdb_init_error/main.py", line 8, in
main()
~~~~^^
File "/path/to/src/replicate_rdb_init_error/main.py", line 4, in main
rdb.set_trace()
~~~~~~~~~~~~~^^
File "/path/to/lib/python3.13/site-packages/celery/contrib/rdb.py", line 187, in set_trace
return debugger().set_trace(frame)
~~~~~~~~^^
File "/path/to/lib/python3.13/site-packages/celery/contrib/rdb.py", line 179, in debugger
rdb = _current[0] = Rdb()
~~~^^
File "/path/to/lib/python3.13/site-packages/pdbpp.py", line 279, in call
obj.init(*args, **kwargs)
~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
TypeError: Rdb.init() got an unexpected keyword argument 'start_filename'

A minimal reproducible example is available here: https://github.com/Hugo-C/replicate-rdb-init-error
It boils down to:

from celery.contrib import rdb

def main():
    rdb.set_trace()
    print('hello')

Is it something you consider like a bug on your side ? Or would you rather see it fixed in Celery ?

[bretello]

Thanks for the detailed report. It looks like the culprit is here:

pdbpp/src/pdbpp.py

Lines 283 to 294 in 5694940

	obj = cls.__new__(cls)
	if called_for_set_trace:
	kwargs.setdefault("start_filename", called_for_set_trace.f_code.co_filename)
	kwargs.setdefault("start_lineno", called_for_set_trace.f_lineno)
	if "set_global_pdb" in kwargs:
	set_global_pdb = kwargs.pop("set_global_pdb", use_global_pdb)
	if set_global_pdb:
	obj._force_use_as_global_pdb = True
	else:
	set_global_pdb = use_global_pdb
	obj.__init__(*args, **kwargs)

the PdbMeta class attempts to reuse existing Pdb (pdbpp.Pdb) instances when possible.

When subclassing Pdb (class Rdb(Pdb)) with pdbpp installed and enabled, this breaks calling set_trace.

I have a tentative fix for this here: #61

This basically breaks reuse of existing instances for Pdb subclasses and causes some tests to fail ( testing/test_pdb.py::test_config_gets_start_filename), so I need a bit more time to figure out if there's a better way to do this.