Sorry for slacking on this, this has been fixed upstream for a short while now, and just needs a version bump. The only wrinkle is there's now an "optional" dependency on gmp at build time, and at runtime if one wishes to use the sb-gmp contrib module.
On Sun, Sep 15, 2013 at 04:54:19PM +0200, Jérémie Courrèges-Anglas wrote: > > Hi folks, > > here's a diff to correct sbcl's assumptions about struct timeval on > OpenBSD i386. The second build just ended, quick testing shows no > regression. I've refreshed the existing patches while here. > > ok? > > Questions: > - is someone here already dealing with upstream? If no one steps up > I'll send them the patch soon (if they fix their DNS problems...) Fixed upstream already, and in a fairly portable and futureproof way that avoids those hardcoded struct definitions. > - I see subdirectories in the source code, named like "mips" or "alpha". > Has anyone analyzed the work that should be done to make this port > available on more architectures? It will be a bit of work, I think. The easiest port to start with would be hppa, I think. Possibly sparc if you can squeeze sbcl into the virtual address space there. The alpha port of is somewhat special, and I think would be more work to bring up on openbsd. The mips port is 32-bit, so it's out unless we get a 32-bit mips platform. > Index: Makefile > =================================================================== > RCS file: /cvs/ports/lang/sbcl/Makefile,v > retrieving revision 1.19 > diff -u -p -r1.19 Makefile > --- Makefile 6 Sep 2013 22:16:24 -0000 1.19 > +++ Makefile 15 Sep 2013 11:06:08 -0000 > @@ -2,12 +2,11 @@ > > # not yet ported to other arches > ONLY_FOR_ARCHS = amd64 i386 powerpc > -BROKEN-i386= build fails post-64-bit time_t > -# ^^ logs: http://rhaalovely.net/build-failures/i386/20130901/lang/sbcl.log > > COMMENT= compiler and runtime system for ANSI Common Lisp > > V = 1.1.8 > +REVISION= 0 > DISTNAME= sbcl-${V}-source > PKGNAME= sbcl-${V} > WRKDIST= ${WRKDIR}/sbcl-${V} > Index: patches/patch-contrib_asdf-module_mk > =================================================================== > RCS file: /cvs/ports/lang/sbcl/patches/patch-contrib_asdf-module_mk,v > retrieving revision 1.1 > diff -u -p -r1.1 patch-contrib_asdf-module_mk > --- patches/patch-contrib_asdf-module_mk 8 Jul 2011 11:42:09 -0000 > 1.1 > +++ patches/patch-contrib_asdf-module_mk 15 Sep 2013 12:11:41 -0000 > @@ -5,9 +5,9 @@ Fix 'all' target to allow building witho > Don't copy every single file when installing the contribs, only the > ones that are actually needed to load the system. > > ---- contrib/asdf-module.mk.orig Mon May 9 04:49:39 2011 > -+++ contrib/asdf-module.mk Fri Jul 8 13:39:44 2011 > -@@ -25,7 +25,8 @@ endif > +--- contrib/asdf-module.mk.orig Sun Jun 2 15:12:39 2013 > ++++ contrib/asdf-module.mk Sun Sep 15 14:11:33 2013 > +@@ -27,7 +27,8 @@ endif > > export CC SBCL EXTRA_CFLAGS EXTRA_LDFLAGS > > @@ -15,9 +15,9 @@ ones that are actually needed to load th > +all: $(EXTRA_ALL_TARGETS) $(SYSTEM).fasl > +$(SYSTEM).fasl: > $(MAKE) -C ../asdf > - $(SBCL) --eval '(defvar *system* "$(SYSTEM)")' --load ../asdf-stub.lisp > --eval '(quit)' > + $(SBCL) --eval '(defvar *system* "$(SYSTEM)")' --load ../asdf-stub.lisp > --eval '(exit)' > > -@@ -37,5 +38,4 @@ test: all > +@@ -40,5 +41,4 @@ test: all > # KLUDGE: There seems to be no portable way to tell tar to not to > # preserve owner, so chown after installing for the current user. > install: $(EXTRA_INSTALL_TARGETS) > Index: patches/patch-make-target-contrib_sh > =================================================================== > RCS file: /cvs/ports/lang/sbcl/patches/patch-make-target-contrib_sh,v > retrieving revision 1.2 > diff -u -p -r1.2 patch-make-target-contrib_sh > --- patches/patch-make-target-contrib_sh 11 Aug 2012 23:02:23 -0000 > 1.2 > +++ patches/patch-make-target-contrib_sh 15 Sep 2013 01:47:04 -0000 > @@ -4,9 +4,9 @@ Only run the contrib tests if $RUN_CONTR > allows the contribs to be build when USE_SYSTRACE=Yes, and the tests > to be run later in do-regress. > > ---- make-target-contrib.sh.orig Mon Dec 5 00:09:01 2011 > -+++ make-target-contrib.sh Sat Aug 11 16:55:34 2012 > -@@ -43,6 +43,7 @@ export SBCL SBCL_BUILDING_CONTRIB > +--- make-target-contrib.sh.orig Sun Jun 2 15:12:40 2013 > ++++ make-target-contrib.sh Sun Sep 15 03:26:47 2013 > +@@ -44,6 +44,7 @@ export SBCL SBCL_BUILDING_CONTRIB > # as SB-RT and SB-GROVEL, but FIXME: there's probably a better > # solution. -- CSR, 2003-05-30 > > @@ -14,7 +14,7 @@ to be run later in do-regress. > find contrib/ \( -name '*.fasl' -o \ > -name '*.FASL' -o \ > -name 'foo.c' -o \ > -@@ -57,6 +58,11 @@ find contrib/ \( -name '*.fasl' -o \ > +@@ -58,6 +59,11 @@ find contrib/ \( -name '*.fasl' -o \ > > find output -name 'building-contrib.*' -print | xargs rm -f > > @@ -26,7 +26,7 @@ to be run later in do-regress. > # Ignore all source registries. > CL_SOURCE_REGISTRY='(:source-registry :ignore-inherited-configuration)' > export CL_SOURCE_REGISTRY > -@@ -74,7 +80,7 @@ for i in $contribs_to_build; do > +@@ -75,7 +81,7 @@ for i in $contribs_to_build; do > # export INSTALL_DIR=$SBCL_HOME/`basename $i ` > test -f $i/test-passed && rm $i/test-passed > # hack to get exit codes right. > Index: patches/patch-src_code_unix_lisp > =================================================================== > RCS file: patches/patch-src_code_unix_lisp > diff -N patches/patch-src_code_unix_lisp > --- /dev/null 1 Jan 1970 00:00:00 -0000 > +++ patches/patch-src_code_unix_lisp 15 Sep 2013 12:18:20 -0000 > @@ -0,0 +1,40 @@ > +$OpenBSD$ > + > +Cope with 64-bit time_t. > + > +--- src/code/unix.lisp.orig Sun Jun 2 15:12:40 2013 > ++++ src/code/unix.lisp Sun Sep 15 14:17:59 2013 > +@@ -215,16 +215,19 @@ corresponds to NAME, or NIL if there is none." > + (tv-sec time-t) ; seconds > + (tv-usec suseconds-t))) ; and microseconds > + > +-;; The above definition doesn't work on 64-bit OpenBSD platforms. > +-;; Both tv_sec and tv_usec are declared as long instead of time_t, and > +-;; time_t is a typedef for int. > +-#!+(or openbsd netbsd) > ++;; The following definition was used on OpenBSD pre-5.5. > ++;; Both tv_sec and tv_usec were declared as long instead of time_t, and > ++;; time_t was a typedef for int. > ++;; Note: time_t is 64-bit on all architectures, starting from OpenBSD 5.5. > ++;; FIXME: it seems recent NetBSD releases use a 64-bit time_t too. > ++#!+netbsd > + (define-alien-type nil > + (struct timeval > + (tv-sec long) ; seconds > + (tv-usec long))) ; and microseconds > + > +-#!+win32 > ++;; OpenBSD 5.5 uses this struct layout. > ++#!+(or win32 openbsd) > + (define-alien-type nil > + (struct timeval > + (tv-sec time-t) ; seconds > +@@ -948,6 +951,8 @@ avoiding atexit(3) hooks, etc. Otherwise exit(2) is ca > + ;; Just as with struct timeval, 64-bit OpenBSD has problems with the > + ;; above definition. tv_sec is declared as time_t instead of long, > + ;; and time_t is a typedef for int. > ++;; Note: time_t is 64-bit on all architectures, starting from OpenBSD 5.5 > ++;; FIXME: it seems recent NetBSD releases use a 64-bit time_t too. > + #!+(or openbsd netbsd) > + (define-alien-type nil > + (struct timespec
