Re: [PATCH] Modify align-c/align-c++ test case for AIX

[Pedro Alves <pedro@palves.net>]

Às 14:15 de 06/04/2023, Aditya Kamath1 escreveu:
> Hi Pedro, Ulrich, Tom and community,
> 
> I actually made a mistake while I was experimenting. I copied this line 
> in two places resulting in this hunk by mistake which was not required. 
> That is why I saw the error of initialising a non-static member. I 
> apologise for the same.

Ah, OK.  Makes sense now.  No worries.


> I am okay with the patch with change ID = 
> I874b717afde7b6fb1e45e526912b518a20a12716 pasted to me on the 29^th of 
> March. It is the same in AIX as well. Kindly commit the same.

Alright, merged.

> 
> 
> Only one more problem Pedro, Tom and Ulrich..
> Consider the code below..
> 
> #include <iostream>
> 
> using namespace std;
> 
> #define DEF(T,U) struct align_pair_ ## T ## _x_ ## U { T one; U two; }
> 
> DEF (char, double);
> 
> struct align_pair_char_x_double item_char_x_double;
> 
> int main (){
> 
>      int x = alignof (struct align_pair_char_x_double);
> 
>      printf ("%d \n", x);
> 
>      return 0;
> 
> }
> 
> The output of this in linux is 8. But in AIX is 4. Though the size is 8 
> for double, alignment is 4.
> 
> bash-5.1$ ~/gdb_tests/align
> 
> 4

That it's different between Linux and AIX does not by itself indicate
a problem.

> 
> So we have failures in all test cases involving double data type in AIX.

... because GDB's alignof operator returns 8?

It sounds like what the compiler says is the type's alignment differs
from GDB's built-in knowledge of the type's alignment.  I.e., GDB
alignof implementation returns the wrong alignment.  This is exactly the
sort of bug that the testcase is meant to catch.

Grepping around, it doesn't seem like AIX installs a gdbarch_type_align
hook, so the fix may to install one, and make it handle the double type
specially.

Pedro Alves