On Thu, 2024-01-25 at 12:31 +0100, Jose E. Marchesi wrote:
[...]
> However, btf_dump_test_case_syntax.c explicitly tests the dumping of a C
> function like the above:
>
> * - `fn_ptr2_t`: function, taking anonymous struct as a first arg and pointer
> * to a function, that takes int and returns int, as a second arg; returning
> * a pointer to a const pointer to a char. Equivalent to:
> * typedef struct { int a; } s_t;
> * typedef int (*fn_t)(int);
> * typedef char * const * (*fn_ptr2_t)(s_t, fn_t);
>
> the function being:
>
> typedef char * const * (*fn_ptr2_t)(struct {
> int a;
> }, int (*)(int));
>
> which is not really equivalent to the above because one is an anonymous
> struct type, the other is named, and also the scope issue described
> above.
>
> That makes me wonder, since this is testing the C generation from BTF,
> what motivated this particular test? Is there some particular code in
> the kernel (or anywhere else) that uses anonymous struct types defined
> in parameter lists? If so, how are these functions used?
fwiw, I can't find any FUNC_PROTO in test kernel BTF that have
anonymous struct or pointer to anonymous struct as their parameter.
