[PATCH] Build fixes and align binary names and installed binary lists between waf and autoconf

[Jelmer Vernooij]

On Wed, Nov 21, 2012 at 08:34:09PM +1100, Andrew Bartlett wrote:
> The rationale for each patch is in the comment, but overall this series
> of patches tries to fix up issues I noticed while doing some Debian
> packaging.  
> 
> In particular, it fixes up the last binary names that differ between the
> build systems, and will allow us to remove the 'binary mapping' system
> in make test eventually (once we are sure we don't need to rename
> again).  Removing the autoconf locktest and masktest is part of that.
> 
> This also removes the install flag from all our testing binaries, which
> should not be installed on the system or shipped by distributions.  I
> note in particular that Debian ignores this in autoconf, and picks the
> binaries out of the build tree.  This isn't practical with waf, and I
> hope we can agree not to package these (otherwise, debian can remove the
> install=False flag in a patch). 

> 
> Currently, we put strerror_r into libreplace even on systems with strerror_r.
> 
> Andrew Bartlett
> ---
>  lib/replace/replace.c | 2 +-
>  lib/replace/replace.h | 4 +---
>  2 files changed, 2 insertions(+), 4 deletions(-)
> 
> diff --git a/lib/replace/replace.c b/lib/replace/replace.c
> index 7ee5f4b..f37d69f 100644
> --- a/lib/replace/replace.c
> +++ b/lib/replace/replace.c
> @@ -785,7 +785,7 @@ char *rep_get_current_dir_name(void)
>  }
>  #endif
>  
> -#if !defined(HAVE_STRERROR_R) || !defined(STRERROR_R_PROTO_COMPATIBLE)
I think we should properly check the prototype and keep the
STRERROR_R_PROTO_COMPATIBLE define . The problem is that strerror_r
returns a char * on some platforms (notably, glibc) and an int on
others. Heimdal relies on the int version as far as I can tell.

See http://daniel.haxx.se/projects/portability/

Cheers,

Jelmer