Re: [Bug-gnulib] Re: [PATCH] ping on GNU/KFreeBSD patch sent a while ago

[Bruno Haible]

Robert Millan wrote:
> > approximate percentage of packages which test the $host values
> >    a) in order to know about some hardware/kernel issues,
>
> Very few. That I can remember right now, only binutils, gcc (gcc itself,
> not g++ etc) and some parts of libtool.
>
> >    b) in order to know about things tied to libc,
>
> Most of them. They tend to use triplets to define things like _GNU_SOURCE,
> dynamic linker stuff, Glibc-specific CFLAGS, etc.

Interesting. This means that in the quadruples

   CPU_TYPE-MANUFACTURER-KERNEL-OPERATING_SYSTEM

the kernel is not so important (because of (a)), and that instead it would
be more useful to have a simple test for glibc systems. Which neither the
previous naming scheme

   linux-gnulibc[15] [Linux libc5] -> no
   linux-gnu                       -> yes
   gnu               [Hurd]        -> yes

nor your advocated naming scheme

   netbsd*-gnu)                    -> no
   netbsd*)                        -> no
   knetbsd*-gnu)                   -> yes
   kfreebsd*-gnu)                  -> yes

gives.

IMO, we should either
  - move the AM_GLIBC macro (gnulib/m4/glibc.m4) into autoconf, so that
    more packages use it instead of making choices based on config.guess
    results when all they need is AC_GLIBC, or
  - ask RMS to include the libc type in a canonical quadruple/quintuple.

Bruno