Re: Re: one-header-per-function

[yyyyyyyyyyyyyyy@xxxxxx]

David Korn:

> 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.

Hmm. Difficult to parse sentence. Are you saying

  I do think that requiring order independence or requiring
  a single include for any interface, does allow the side
  effects of including headers that the header you specify includes.

?


The function wctob() is described with includes
<stdio.h> and <wchar.h>.
The function timer_create() is described with includes
<time.h> and <signal.h>.
Etc - sometimes several headers are required to describe
the types of the parameters of a function.

I do not think we want to change that.


For the function stat() it suffices to include <sys/stat.h>.
So, this single include file defines a struct stat.
But then it must define the elements of such a struct,
and in particular also for example the type dev_t.
Maybe we must state that it is implementation dependent
whether after inclusion of <sys/stat.h> also the type
pthread_mutex_t is defined.

Andries