missing configure checks with the waf build

[Andrew Bartlett]

On Thu, 2011-11-24 at 10:07 +0100, Andreas Schneider wrote:
> On Thursday 24 November 2011 08:44:39 Andrew Tridgell wrote:
> > Hi Andreas,
> 
> Hi Andrew,
> 
> > > * waf needs to support all configure checks which the Samba3 autoconf
> > > 
> > >   provides. There are still options and defines missing.
> > 
> > could you provide a list and say why they are needed?
> 
> Build Samba 4 with waf and Samba 3 with autoconf. Then diff the config.h 
> files.

Andreas,

Thanks for mentioning this.  I've started looking into the issue, and
found it is more challenging than you might expect, as the config.h
files are not directly comparable.

First, the krb5 defines need to be ignored, as these are testing the
system krb5 compared with the internal heimdal.  Once this is done, then
we need to discount the init macros and a number of other defines that
are simply handled in other ways in waf.

Once these are removed, the list of tests that run on my Fedora system,
and which give a #define on autoconf but not on waf is:

+# define __EXTENSIONS__ 1
+#define _FILE_OFFSET_BITS 64
+#define HAVE_BUILTIN_HAVE_ISBLANK 1
+#define HAVE_DECL_RL_EVENT_HOOK 0
+#define HAVE_LINUX_XFS_QUOTAS 1
+#define HAVE_MREMAP 1
a1719384 lib/replace/libreplace.m4         (Simo Sorce        2011-08-14
18:05:27 -0400 103) 

+#define HAVE_QUOTACTL_LINUX 1
8779f74e (Stefan Metzmacher     2005-09-12 10:23:14 +0000 4927) 

+#define HAVE_RES_NINIT 1
4f6b1e8c (Volker Lendecke       2011-04-25 10:55:58 +0200 1426)

+#define HAVE_SYS_QUOTAS 1

+#define _LARGEFILE64_SOURCE 1
+# define _POSIX_PTHREAD_SEMANTICS 1
+#define WITH_QUOTAS 1

(I've noted the commit where the autoconf test was added, in a few cases
where I have investigated). 

There is a larger number of tests which do not pass on my autoconf
build, but which do not appear in the waf config.h (as /* #undef */).
These are still a concern, and a bit more of a challenge to investigate,
but I'll keep looking into this.

The other difference I noticed is that the waf build is more aggressive
in locating AIO functions.

Thanks for raising this, as it gives me a concrete list of tasks that
remains, as well as hope, given that so few of the defines turned out
differently. 

Andrew Bartlett

-- 
Andrew Bartlett                                http://samba.org/~abartlet/
Authentication Developer, Samba Team           http://samba.org