Re: [PATCH 2/6] Only define NEEDS_SOCKET if libsocket is usable

From: Junio C Hamano <gitster@pobox.com>
Date: Aug 17, 2008, 09:16 UTC
Message-ID: <7v63q080io.fsf@gitster.siamese.dyndns.org>
In-Reply-To: <2EA0EACE-D8FE-476A-BAE2-7E12EE961C46@web.de>

Andreas Färber <andreas.faerber@web.de> writes:

Show 19 quoted lines

> diff --git a/configure.ac b/configure.ac
> index 7c2856e..75ec83a 100644
> --- a/configure.ac
> +++ b/configure.ac
> @@ -223,11 +223,11 @@ AC_LINK_IFELSE(ZLIBTEST_SRC,
>  LIBS="$old_LIBS"
>  AC_SUBST(NO_DEFLATE_BOUND)
>  #
> -# Define NEEDS_SOCKET if linking with libc is not enough (SunOS,
> +# Define NEEDS_SOCKET if linking with libc is required (SunOS,
>  # Patrick Mauritz).
> -AC_CHECK_LIB([c], [socket],
> -[NEEDS_SOCKET=],
> -[NEEDS_SOCKET=YesPlease])
> +AC_CHECK_LIB([socket], [socket],
> +[NEEDS_SOCKET=YesPlease],
> +[NEEDS_SOCKET=])
>  AC_SUBST(NEEDS_SOCKET)
>  test -n "$NEEDS_SOCKET" && LIBS="$LIBS -lsocket"

Doesn't this force linkage with -lsocket even if -lc is enough to use socket(2) calls?

In other words, "checking libc is not enough" is only half correct. The right thing to do is "check libc and if it is sufficient be happy, but otherwise do not automatically assume -lsocket is Ok." Something like:

AC_CHECK_LIB([c], [socket],
[NEEDS_SOCKET=],
[AC_CHECK_LIB([socket], [socket],
        	[NEEDS_SOCKET=YesPlease],
                [NEEDS_SOCKET=])])

Other patches seemed Ok from my cursory look; I do not know the people whose Ack's were on your patch submission, though...

Thanks.