On Tue, Mar 24, 2009 at 8:32 AM, Joseph S. Myers
<jos...@codesourcery.com> wrote:
> On Tue, 24 Mar 2009, H.J. Lu wrote:
>
>> > I see nothing about passing by value different from structure assignment,
>> > which ignores the flexible array member (see 6.7.2.1 paragraph 22 (in
>> > N1256) for an example stating this). Although argument passing and return
>> > aren't strictly assignments, they generally act in the same way, so such a
>> > structure passed by value or used as a function return value loses the
>> > flexible array members in the process. I suppose an optional warning for
>> > this might be useful.
>> >
>>
>> How should be they passed on x86-64? psABI isn't clear and gcc isn't
>> consistent
>> with char contents[0] vs char contents[].
>
> The [0] case is outside the scope of ISO C and so outside the scope of any
> ABI that limits itself to ISO C.
Well, gcc has to follow an ABI to pass the [0] case.
> For the [] case, it's a good question that could affect not just the
> x86-64 ABI but various other processor ABIs that look at the elements of
> structures (not just their sizes) to determine how they are passed - it
> would be useful to add target-independent tests to the compat testsuite so
> compatibility of GCC with itself and GCC with other compilers can be
> tested in this regard. They'd cover various combinations of types for the
> first element or first few elements, and the final [] array - both for
> argument passing and for function return.
>
On x86-64, gcc 4.3 passes/returns
struct line {
int length;
char contents[];
};
in memory and passes/returns
struct line {
int length;
char contents[0];
};
in register. I couldn't find anything in psABI which mandates such behaviors.
--
H.J.
