fix: local struct shadows included module

[FrankPortman]

fixes: #323

[FrankPortman]

@aisk not sure if you are the current maintainer but I see you have the most commits in the last half year so please take a look when you can 🙏

[FrankPortman]

Friendly bump - @cocolato tagging you as well since I see some recent activity from you.

Some initial direction to this PR or the underlying issue would be great!

[cocolato]

thriftpy2/thriftpy2/parser/parser.py

Lines 82 to 84 in 8e82f1c

	setattr(child, '__thrift_module_name__', child_module_name)
	setattr(thrift, child.__name__, child)
	_add_thrift_meta('includes', child)

thriftpy2/thriftpy2/parser/parser.py

Lines 244 to 248 in 8e82f1c

	def p_seen_struct(p):
	'''seen_struct : STRUCT IDENTIFIER '''
	val = _make_empty_struct(p[2])
	setattr(threadlocal.thrift_stack[-1], p[2], val)
	p[0] = val

The root cause appears to be that the p_seen_struct method overwrites the contents of the originally imported module via setattr. Although the current fix resolves this issue, I suggest using a prefix with setattr to better distinguish between the imported module and the struct. @aisk please take a look.

[FrankPortman]

thriftpy2/thriftpy2/parser/parser.py

Lines 82 to 84 in 8e82f1c

	setattr(child, '__thrift_module_name__', child_module_name)
	setattr(thrift, child.__name__, child)
	_add_thrift_meta('includes', child)

thriftpy2/thriftpy2/parser/parser.py

Lines 244 to 248 in 8e82f1c

	def p_seen_struct(p):
	'''seen_struct : STRUCT IDENTIFIER '''
	val = _make_empty_struct(p[2])
	setattr(threadlocal.thrift_stack[-1], p[2], val)
	p[0] = val

The root cause appears to be that the p_seen_struct method overwrites the contents of the originally imported module via setattr. Although the current fix resolves this issue, I suggest using a prefix with setattr to better distinguish between the imported module and the struct. @aisk please take a look.

Thank you for reviewing! You're correct that p_seen_struct overwrites the imported module - that's the root cause. However, using prefixes would be a breaking change since all existing code expects direct attribute access (e.g., thrift.MyStruct).

The current fix preserves backward compatibility by using the already-available __thrift_meta__['includes'] to correctly resolve qualified names when there's a collision. This follows the existing pattern where includes are already tracked separately in metadata.

If you'd prefer a different approach, I'm happy to adjust, but I believe this is the least disruptive fix for users.

Caveat - I am obviously very new to this project so it is possible I missed something in your recommendation. At the end of the day, any approach that fixes #323, either with this PR or a different PR, would be fine for me.

[aisk]

I'd like to mark this as pending further discussion. Since we don’t have a mechanism to put a PR on hold, I’ve set it to Request Changes instead.