(j3.2006) definition of CFI_CDESC_T

[Robert Corbett]

On 04/24/12 01:28, Malcolm Cohen wrote:
> (cc'd to interop-tr which is probably a better place for discussion).
If the discussion is moved to interop-tr, please copy me on e-mail relevant to 
my comments.
>
> That does not look like "a type suitable for declaring a C descriptor" to me.
I agree that your interpretation is reasonable, but I do not think it is the 
only possible interpretation of the phrase.  I think the phrase needs to be 
clarified.  The definition I gave in my previous e-mail satisfies my 
understanding of the phrase.
> Surely, given
>   CFI_CDESC_T(2) x;
> then x must be a C descriptor, and therefore &x must be compatible with 
> CFI_cdesc_t*, and x.base_addr et al should be usable; neither of those things 
> appear to be true of this (that or my eyes have glazed over too much).
I am confused by your assertion that &x must be compatible with CFI_cdesc_t*.  I 
assume you mean that the type of &x and CFI_cdesc_t* becompatible types.  For 
that to be true, the type of x and CFI_cdesc_t must be compatible types [C99: 
6.7.5.1p2].  The TS effectively requires CFI_CDESC_T(2) to be a complete type 
and CFI_cdesc_t to be an incomplete type.  They will have different definitions, 
and therefore they will be different types.  Assuming the type specifiers for 
CFI_CDESC_T(2) and CFI_cdesc_t appear in the same translation unit, they will 
not be compatible types [C99: 6.2.7p1] regardless of how CFI_CDESC_T(2) is 
defined.  Glazed eyes are a natural reaction to reading this part of the C99 
standard.

I think the TS needs to clearly state if the types created by the macro 
CFI_CDESC_T are structure types, and if they are required to be structure types, 
if the members of those types are to be accessible.

Bob Corbett