Nested named structs names are not detected. about ctypeslib

[trolldbois]

Ah, I think I understand, you mean you would like

• To not have a type declaration for an anonymous union, that is an un-named field (union_s_0)
• A type declaration for an anonymous union, that is named field should carry the name of the field (union_s_1 -> union_s_u)

is that what you were asking ?

[RobertHerreraEECS]

FWIW I'm using the following command:

clang2py -v --validate 1 -k mcdefstua --clang-args "-I/dir/includes -std=c11 " target_header.h

[RobertHerreraEECS]

Another note, after digging through the logs, the interesting thing is that I can see the actual name of the struct being detected as a field FIELD_DECL, where the expected name in this edited log snippet is bar.

`

2023-03-26 09:38:37,157:DEBUG:clangparser:Unregister foo_0_3
2023-03-26 09:38:37,157:DEBUG:clangparser:register: foo_0_3 
2023-03-26 09:38:37,157:DEBUG:cursorhandler:creating FIELD_DECL for FIELD_DECL/bar
2023-03-26 09:38:37,157:DEBUG:cursorhandler:FIELD_DECL: field offset is 0
2023-03-26 09:38:37,157:DEBUG:cursorhandler:FIELD_DECL: field is 80 bits
2023-03-26 09:38:37,157:DEBUG:cursorhandler:FIELD_DECL: we now look for the declaration name.kind CursorKind.STRUCT_DECL`

[RobertHerreraEECS]

The following is a simple test showcasing the behavior I'm seeing as well an accompanying c file with the desired use of the struct I would need in my python code.
example.zip

[trolldbois]

@RobertHerreraEECS can you confirm this is fixed in the main branch ?

[RobertHerreraEECS]

Thanks for getting this ticket first off!

This seems like a partial fix but definitely much better results. Unfortunately, it seems like the current patch and tests only account for the only 'fields' inheriting the TYPEDECL name and not propagating that into the class namespace as well.

For instance I would expect the following in test case

self.assertNotIn('union_s_0', self.namespace) self.assertIn('union_s_u', self.namespace)

Let me know if this doesn't fit within the scope of functionality ^. If it doesn't, feel free to close the ticket.

Thanks again.

[trolldbois]

@RobertHerreraEECS I'm sorry I do not understand the test case you are proposing.
Could you write python code you see failing / are not happy with, and the C headers ?

[RobertHerreraEECS]

Yes exactly! "A type declaration for an anonymous union, that is named field should carry the name of the field (union_s_1 -> union_s_u)".

Let me know what you think. I think its quite portable for most use cases. From a C standpoint if I see a structure that is both named and typedef usually the code base resorts to only using the typdef alias.

So my proposition would be if it the structure was unnamed (cursor.type.get_declaration().spelling == '' ) BUT cursor.spelling != '' (the typedef alias), then just use the typedef name rather using a numeric index.