[I'm cross-posting to the nettle list and a debian bug, so please take
care when replying]
The issue is that there are some differences in generated nettle header
files depending on architecture and compiler version.
Francois Gouget <fgouget(a)free.fr> writes:
> Here is a proposed patch for the compiler version issue in nettle-stdint.h:
>
> $ cat debian/patches/multiarch.patch
> --- a/aclocal.m4
> +++ b/aclocal.m4
> @@ -857,7 +857,7 @@
> fi # shortcircut to system "stdint.h"
> # ------------------ PREPARE VARIABLES ------------------------------
> if test "$GCC" = "yes" ; then
> -ac_cv_stdint_message="using gnu compiler "`$CC --version | head -1`
> +ac_cv_stdint_message="using gnu compiler $CC"
> else
> ac_cv_stdint_message="using $CC"
> fi
If we go this path, maybe just drop the conditional and unconditionally
print "using $CC"? I'm not sure about the reason for displaying the
version, but I guess some older gcc versions did stdint and/or
inttypes.h differently.
If you have the time, it would be helpful to look at the latest version
of AX_CREATE_STDINT_H (from autoconf archive) and see if it does
anything differently. Nettle uses a pretty old version (which seems to
work fine).
> For GMP_NUMB_BITS here are some thoughts:
You must understand that nettle defines GMP_NUMB_BITS only if configured
with --enable-mini-gmp. This configuration is not intended to be
compatible with anything else, and should never be installed on a normal
debian system. It's intended for small embedded systems which needs to
verify digital signatures, but which consider the real libgmp too large.
> * Is it really necessary to install version.h?
> IMHO the right place for other packages to figure out library versions is through
> some scripting in the configure script rather than through a
> header.
This has been requested by users. Not everyone uses autoconf. And it's
very common practice, I think both gmp and gnutls have similar version
defines.
> * GMP_NUMB_BITS is already defined by libgmp-dev in gmp.h. More
> preceisely
Exactly, and when not configuring with --enable-mini-gmp, that's the
definitions which is used.
Getting into the details, Nettle's definition of GMP_NUMB_BITS
conceptually belongs in mini-gmp.h. However, mini-gmp is a standalone
package which doesn't use autoconf. And it has to be a preprocessor
*constant*, so definiting it like sizeof(unsigned long) * CHAR_BIT won't
cut it, since the preprocessor doesn't understand sizeof.
Therefore the definition, for the --enable-mini-gmp configuration which
needs it, should be in a nettle include file. Then the cleanest way
would be to omit the definition completely when not using
--enable-mini-gmp. I haven't done that, because it would make the
substitution logic harder, and because I was thinking that
#define NETTLE_USE_MINI_GMP 0
#if NETTLE_USE_MINI_GMP
# define GMP_NUMB_BITS 64
#endif
is harmless whatever value appears in the second define. If it helps, I
guess this could be changed into the architecture independent
#define NETTLE_USE_MINI_GMP 0
#if NETTLE_USE_MINI_GMP
# define GMP_NUMB_BITS the moon is made of green cheese
#endif
Opinions?
Regards,
/Niels
--
Niels Möller. PGP-encrypted email is preferred. Keyid C0B98E26.
Internet email is subject to wholesale government surveillance.
