> Donn:
>
> > potentially several headers
>
> Typically, this is #include <sys/types.h>
> followed by #include <foo.h> for the function foo().
>
> It is very desirable to state that the POSIX mandated
> headers can be included in any order. Once one says so,
> the #include <foo.h> cannot count on <sys/types.h>
> being included already, so must include it itself.
>
> So, as I see it, the goal is not `one include per function',
> but the goal is `independence of inclusion order', and
> a useful side effect is that one usually will have only
> one include per function. I do not think this useful
> side effect should be stated as a goal or requirement.
>
> Andries
>
>
Order independence and one include file do not necessarily
yield the same results.
The question is whether including <sys/stat.h>, for example,
would allow all of the side effects of including <sys/types.h>
or not so that anything defined in <sys/types.h> could be visible.
I don't think that requiring order independence or requiring
a single include for any interface, does not allow the side
effects of including headers that the header you specify includes.
David Korn
research!dgk
yyy@xxxxxxxxxxxxxxxx
|