cvs commit: ports/lang/gcc34 Makefile
David O'Brien
obrien at FreeBSD.org
Sat Mar 20 11:53:13 PST 2004
On Fri, Mar 19, 2004 at 02:13:50AM +0100, Gerald Pfeifer wrote:
> On Wed, 17 Mar 2004, David O'Brien wrote:
> >>> Have you tested this following the instructions at
> >>> http://www.freebsd.org/doc/en_US.ISO8859-1/books/porters-handbook/porting-testing.html
> >> I *DID* test on i386 (5.2-CURRENT, full 'make world' just before testing
> >> the port).
> > I just re-tested on FreeBSD/i386:
> >
> > kieu(r):/usr/ports/lang/gcc34> /usr/bin/time -h make package
> > ...
> > Creating bzip'd tar ball in '/usr/ports/lang/gcc34/gcc-3.4.0_20040310.tbz'
> > 49m57.11s real 16m21.78s user 7m13.72s sys
> > kieu(r):/usr/ports/lang/gcc34> ll gcc-3.4.0_20040310.tbz
> > -rw-r--r-- 1 root ncvs 13.0M Mar 17 09:48 gcc-3.4.0_20040310.tbz
>
> Well, this is just one part of the testing required for port updates.
>
> If you follow the instructions in the Porters' Handbook, you'll note
> that `make deinstall` got broken, as I just verified on beast
>
> rmdir: /j/gerald/lib/gcc/alpha-portbld-freebsd5.2: Directory not empty
> rmdir: /j/gerald/lib/gcc: Directory not empty
Actually this is a bug in your pkg-plist. I just tested this with your
latest commits (Makefile rev 1.166, pkg-plist rev 1.55). This patch
(which I just committed) makes the existing breakage obvious:
@dirrm libexec/gcc/%%GNU_HOST%%/%%GCC_VER%%
+ at dirrm libexec/gcc/%%GNU_HOST%%
+ at dirrm libexec/gcc
%%LIBJAVA%%share/java/libgcj-%%GCC_VER%%.jar
I suggest we use the mechanism I used for lib/gcc/. That would have
prevented this pkg-plist bug.
If you look at ports/Mk/bsd.port.mk, you'll see that CONFIGURE_TARGET is
defined as ${MACHINE_ARCH}-portbld-freebsd${OSREL}, which is effectively
the same as the definition I removed and you re-added.
Can we agree to this patch now?
.if ${ARCH} == "amd64"
CONFIGURE_TARGET= x86_64-portbld-freebsd${OSREL}
-.else
-CONFIGURE_TARGET= ${ARCH}-portbld-freebsd${OSREL}
.endif
--
-- David
More information about the cvs-all
mailing list