[SOT] Avoid bind wrong instance on patched method

[SigureMo]

PR Category

Execute Infrastructure

PR Types

Bug fixes

Description

同 #73038，paddle.distributed.fleet 会有同样的问题，这导致在调用分布式 API 的情况下会报错，本 PR 避免 bind 时候 bind 错 instance

Paddle/python/paddle/distributed/fleet/__init__.py

Lines 55 to 114 in b0cdc64

	fleet = Fleet()
	_final_strategy = fleet._final_strategy
	_get_applied_meta_list = fleet._get_applied_meta_list
	_get_applied_graph_list = fleet._get_applied_graph_list
	init = fleet.init
	is_first_worker = fleet.is_first_worker
	worker_index = fleet.worker_index
	worker_num = fleet.worker_num
	node_num = fleet.node_num
	rank = fleet.worker_index
	nranks = fleet.worker_num
	world_size = fleet.worker_num
	# device id in current trainer
	local_device_ids = fleet.local_device_ids
	# device ids in world
	world_device_ids = fleet.world_device_ids
	# rank in node
	local_rank = fleet.local_rank
	rank_in_node = local_rank
	is_worker = fleet.is_worker
	is_coordinator = fleet.is_coordinator
	init_coordinator = fleet.init_coordinator
	make_fl_strategy = fleet.make_fl_strategy
	get_fl_client = fleet.get_fl_client
	worker_endpoints = fleet.worker_endpoints
	server_num = fleet.server_num
	server_index = fleet.server_index
	server_endpoints = fleet.server_endpoints
	is_server = fleet.is_server
	util = UtilBase()
	barrier_worker = fleet.barrier_worker
	all_reduce = fleet.all_reduce
	init_worker = fleet.init_worker
	init_server = fleet.init_server
	run_server = fleet.run_server
	stop_worker = fleet.stop_worker
	distributed_optimizer = distributed_optimizer
	save_inference_model = fleet.save_inference_model
	save_persistables = fleet.save_persistables
	save_cache_model = fleet.save_cache_model
	check_save_pre_patch_done = fleet.check_save_pre_patch_done
	save_one_table = fleet.save_one_table
	save_dense_params = fleet.save_dense_params
	load_model = fleet.load_model
	load_inference_model = fleet.load_inference_model
	load_one_table = fleet.load_one_table
	set_date = fleet.set_date
	print_table_stat = fleet.print_table_stat
	minimize = fleet.minimize
	distributed_model = distributed_model
	shrink = fleet.shrink
	get_hybrid_communicate_group = fleet.get_hybrid_communicate_group
	distributed_scaler = distributed_scaler
	set_log_level = log_util.set_log_level
	get_log_level_code = log_util.get_log_level_code
	get_log_level_name = log_util.get_log_level_name
	check_memory_usage = log_util.check_memory_usage
	save_cache_table = fleet.save_cache_table
	collective_perf = fleet.collective_perf
	from .. import auto_parallel as auto # noqa: F401

cpython 在 LOAD_METHOD 时如果不是 obj.__class__.<name>.__get__(obj, obj.__class__) 这种形式的情况下，是不会走第一个 meth_found 分支的

对于当前 case，这是因为 obj.__dict__ 有 <name>，即 patched method，这种情况会被认为 meth_found=0

https://github.com/python/cpython/blob/1173f8068b9a8c1168780532c466c391be10bf67/Objects/object.c#L1174-L1181

另外 super().fn 之前也走了错误的路径，但是能跑，tp_getattro 实际是 super_getattro，所以在 Py_TYPE(obj)->tp_getattro != PyObject_GenericGetAttr 就会走到 meth_found=0

https://github.com/python/cpython/blob/1173f8068b9a8c1168780532c466c391be10bf67/Objects/object.c#L1147-L1151

[paddle-bot]

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

[Copilot]

Pull request overview

This PR fixes a bug where patched methods (methods attached to instances rather than defined on their class) were incorrectly binding the wrong instance when accessed through the SOT (Symbolic Opcode Translator) system. The fix adds a check to determine if an attribute exists on the class type before binding an instance, preventing incorrect bindings for dynamically-attached methods.

Key changes:

• Added logic to detect patched methods by checking if attributes exist on the class type
• When attributes don't exist on the class, the instance is set to None, allowing the correct instance to be extracted from the method's __self__
• Applied the fix in both attr_proxy_getter and load_method to ensure consistency

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
test/sot/test_patched_method.py	Adds test case validating that patched methods (methods bound to a different instance) are correctly handled without breaking the graph
python/paddle/jit/sot/opcode_translator/executor/variables/base.py	Adds check for whether attribute exists on class type; if not, sets instance to None to avoid binding wrong instance
python/paddle/jit/sot/opcode_translator/executor/opcode_executor.py	Adds the same hasattr check when loading methods to ensure consistent handling of patched methods

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[DrRyanHuang]

最小复现demo真简洁喵 👍

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (develop@ad925a3). Learn more about missing BASE report.

Additional details and impacted files

@@             Coverage Diff             @@
##             develop    #77227   +/-   ##
===========================================
  Coverage           ?   100.00%           
===========================================
  Files              ?         2           
  Lines              ?         4           
  Branches           ?         0           
===========================================
  Hits               ?         4           
  Misses             ?         0           
  Partials           ?         0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.