Re: [PATCH] Don't build libnsl for new ABIs

[Florian Weimer <fweimer at redhat dot com>]

* Andreas Schwab:

> diff --git a/Makeconfig b/Makeconfig
> index a9e50e5b60..fbcf69e7c2 100644
> --- a/Makeconfig
> +++ b/Makeconfig

I don't feel competent to review this changes, but I trust your
expertise.

> diff --git a/scripts/haveversions.awk b/scripts/haveversions.awk
> new file mode 100644
> index 0000000000..aecfcc7eef
> --- /dev/null
> +++ b/scripts/haveversions.awk
> @@ -0,0 +1,11 @@
> +# This Script read the contents of Versions.all and outputs a definition
> +# of variable have-VERSION for each symbol version VERSION which is
> +# defined.

“This script reads”

Could you add something how this is to be used?  I see you used ifndef,
which isn't what we'd use in C sources, but then make doesn't have an
equivalent of -Wundef.  Maybe something like this?

# The have-VERSION variables can be used to check that a port supports a
# particular symbol version in makefiles.  A test for a compatibility
# symbol which was superseded with a GLIBC_2.15 version could be tested
# like this:
#
# ifdef HAVE-GLIBC_2.14
# tests += tst-spawn4-compat
# endif # HAVE-GLIBC_2.14
#
# (NB: GLIBC_2.14 is the symbol version that immediately precedes
# GLIBC_2.15.)

> +
> +NF == 1 && $1 != "}" {
> +  haveversion[$1] = 1
> +}
> +END {
> +  for (i in haveversion)
> +    printf "have-%s = 1\n", i

Should this be:

+    printf "have-%s = yes\n", i

?  I think we use "no" in other places.

Thanks,
Florian