Re: Ensuring posix-ness

2011-09-08 Thread Reuben Thomas
On 8 September 2011 22:14, Eric Blake wrote: > > Done: > > http://sourceware.org/bugzilla/show_bug.cgi?id=13171 Thanks, and I'm glad I didn't try to do it myself, since on reading the bug report I seemed to have misunderstood XSI shading. -- http://rrt.sc3d.org

Re: Ensuring posix-ness

2011-09-08 Thread Eric Blake
On 09/08/2011 01:47 PM, Reuben Thomas wrote: On 8 September 2011 11:04, Bruno Haible wrote: Would you care to raise a bug report against glibc? Likewise about encrypt() in and setkey() in. I'm not sure to whom this request to raise a bug report was addressed, but given the need for preci

Re: Ensuring posix-ness

2011-09-08 Thread Bruce Korb
On 09/08/11 00:11, Eric Blake wrote: On 09/07/2011 10:33 PM, Bruno Haible wrote: The crypt example is particularly annoying, as without _XOPEN_SOURCE defined it's not defined, even though POSIX_VERSION is>= 200112L, and it's in that version of POSIX (of course, that's a glibc bug Setting _POSI

Re: Ensuring posix-ness

2011-09-08 Thread Reuben Thomas
On 8 September 2011 08:11, Eric Blake wrote: > > We could add a gnulib crypt module that exposes the declaration by default > (basically, a stub module that depends on extensions, since the use of the > extensions module is also sufficient to convince glibc to declare crypt). If someone could add

Re: Ensuring posix-ness

2011-09-08 Thread Reuben Thomas
On 8 September 2011 11:04, Bruno Haible wrote: > >> Would you care to raise a bug report against glibc? > > Likewise about encrypt() in and setkey() in . I'm not sure to whom this request to raise a bug report was addressed, but given the need for precision I'd much rather it was one of you than

Re: Ensuring posix-ness

2011-09-08 Thread Bruno Haible
Eric Blake wrote: > > But it is marked as [XSI]. I don't know whether that implies that glibc is > > right in hiding the declaration, or it should enable it? > > Glibc bug. XSI shading means that a non-XSI implementation need not > have crypt by default, but glibc is an XSI implementation, so req

Re: Ensuring posix-ness

2011-09-08 Thread Bruno Haible
Hi Reuben, > > Certainly the fact that crypt() is not defined by default could be mentioned > > in doc/posix-functions/crypt.texi. > > Patch follows: > > diff --git a/ChangeLog b/ChangeLog > index 79d9777..9a64036 100644 > --- a/ChangeLog > +++ b/ChangeLog > @@ -1,3 +1,7 @@ > +2011-09-07 Reuben

Re: Ensuring posix-ness

2011-09-08 Thread Eric Blake
On 09/07/2011 10:33 PM, Bruno Haible wrote: The crypt example is particularly annoying, as without _XOPEN_SOURCE defined it's not defined, even though POSIX_VERSION is>= 200112L, and it's in that version of POSIX (of course, that's a glibc bug Setting _POSIX_C_SOURCE to 200809L _should_ automat

Re: Ensuring posix-ness

2011-09-07 Thread Reuben Thomas
On 7 September 2011 22:33, Bruno Haible wrote: > > For crypt() to be declared in , you need nothing. But neither crypt(3) nor crypt(3posix) documents accessing crypt via #include , which suggests that's a non-standard way to get the function. > Note that must be the first header file to be incl

Re: Ensuring posix-ness

2011-09-07 Thread Bruno Haible
Hi Reuben, > I find that some APIs on GNU systems need > extra treatment (i.e. GNU is not POSIX-1.2001 by default). For > example, for crypt(3) I need to define _XOPEN_SOURCE For crypt() to be declared in , you need nothing. But for it to be declared in , you need _XOPEN_SOURCE or _GNU_SOURCE. _G

Ensuring posix-ness

2011-09-07 Thread Reuben Thomas
Trying to build some code (Lua bindings for POSIX APIs) with a gnulib-ized build system, I find that some APIs on GNU systems need extra treatment (i.e. GNU is not POSIX-1.2001 by default). For example, for crypt(3) I need to define _XOPEN_SOURCE, and for some realtime functions such as clock_getre