Cross-architecture: Why does wrapped structure has _pack_ set to True about ctypeslib

[fpf3]

Seconding this. I have found a similar issue. Furthermore, if I am attempting to create an unpacked structure, it does not set _pack_ = False, as it should.

It seems that that clang2py has knowledge of how they are configured in the header file, and ignores it. You get something like:

class myStruct(Structure):
    _pack_ = True # source:False
    _fields_ = ...

I can only assume what "source:" refers to is the .h file.

[trolldbois]

Short version: It's the result of the cross-architecture ability of ctypeslib.

My recollection is that pack and alignment are different attributes returned by clang.
In ctypeslib, pack will always be True to ensure that python ctypes fields are actually in the correct index.
If the alignment provided by tells clang says a field is not packed next the the previous one, a Padding field will be inserted in-between.

Given that ctypeslib is aware of multi-architecture issues, it's a requirement.
You just have to trust that the code & alignment, issued for the platform chosen by clang is correct.

You cannot ask a python 2.7 ctypes lib on a x86 arch to know how to pack a structure for a arm 128 arch.

E.g: the python source file can be used on any other system with possibly a different architecture for the python binary, and still work for the targeted arch. You can read x64 memory bytes from x86, and vice-versa.

[jpvolkmann]

My understanding is that pack expects an integer number, not a boolean expression (according https://docs.python.org/2/library/ctypes.html#structure-union-alignment-and-byte-order ).
Because Python casts "True" to the integer 1 this works for ctypes library. Setting it to False will lead to a integer of 0 which is not valid