[PATCH 15/18] Add __va_list to <sys/_types.h>

Yaakov Selkowitz yselkowitz@cygwin.com
Fri Apr 15 23:23:00 GMT 2016

On 2016-04-15 14:17, Corinna Vinschen wrote:
> On Apr 15 09:34, Yaakov Selkowitz wrote:
>> On 2016-04-15 06:22, Corinna Vinschen wrote:
>>> On Apr 15 12:52, Sebastian Huber wrote:
>>>> On 15/04/16 12:11, Corinna Vinschen wrote:
>>>>> On Apr 14 13:54, Sebastian Huber wrote:
>>>>>>> Add __va_list to <sys/_types.h> for BSD compatibility.  In FreeBSD this
>>>>>>> typedef is provided by the various architecture-specific
>>>>>>> <machine/_types.h> in a copy and paste manner.
>>>>> Is that supposed to replace __VALIST in stdio.h>
>>>> No, it is used to use, e.g. <syslog.h> from FreeBSD (see tail of file):
>>> I'm asking because FreeBSD uses __va_list throughout stdio.h rather than
>>> using this weird __VALIST.  I think we should do the same (not in this
>>> patch series, but again, something to keep in mind).
>>> I also just noticed that stdio.h always includes stddef.h to fetch the
>>> definition of va_list.  This seems wrong in terms of the feature test
>>> macros.  FreeBSD only defines va_list (as __va_list) if
>>> __POSIX_VISIBLE >= 200112 || __XSI_VISIBLE.
>> Since you said va_list, I'm guessing you meant <stdarg.h> here.
> Duh, right.
>> Looking at
>> each version of stdio.h in SUS though, va_list is marked XSI until SUSv4
>> where it is CX, so I think it should be __POSIX_VISIBLE >= 200809 ||
> Sounds good to me and matches glibc's stdio.h.  Do you want to create a
> patch?  As a sidenote, glibc's stdio.h only includes stdarg.h with
> non-gcc compilers.  Otherwise it typedef's va_list directly as
> _G_va_list aka __gnuc_va_list, just as our stdio.h does with __VALIST.

Yeah, I've started looking into this, and found a few other related 
tweaks to be made.


More information about the Newlib mailing list