On Tue, 2021-04-13 at 00:45 -0500, Gustavo A. R. Silva wrote:
> Hi Martin,
>
> On 4/12/21 23:52, Martin K. Petersen wrote:
>
> > Silencing analyzer warnings shouldn't be done at the expense of
> > human
> > readers. If it is imperative to switch to flex_array_size() to
> > quiesce
> > checker warnings, please add a comment in the code explaining that
> > the
> > size evaluates to nseg_new-1 sge_ieee1212 structs.
>
> Done:
>
> https://lore.kernel.org/lkml/20210413054032.GA276102@embeddedor/
I think the reason everyone gets confused is that they think the first
argument should do something. If flex_array_size had been defined
#define flex_array_size(p, count) \
array_size(count, \
sizeof(*(p)) + __must_be_array(p))
Then we could have used
flex_array_size(sge, nseg_new - 1)
or
flex_array_size(rio->sge, nseg_new - 1)
and everyone would have understood either expression. This would also
have been useful, as the first example demonstrates, when we have a
pointer rather than a flexible member ... although that means the macro
likely needs a new name.
However, perhaps just do
array_size(nseg_new - 1, sizeof(*sge));
And lose the comment?
James
