While addressing the problem in my last message I took a closer look
at the cc.darwin scripts in src/cmd/INIT. Here's some suggestions for
improvements:
* With ast-open-2014-09-29 you upped MACOSX_DEPLOYMENT_TARGET from
10.3 to 10.7 in cc.darwin and cc.darwin.i386. For cc.darwin in
particular this doesn't make sense since the last version of Mac OS
X that ran on Power PC was 10.5. For cc.darwin I changed this back
to 10.3. This seems most appropriate since, if I'm reading the case
statement properly, the first case appears to be for Mac OS 10.2 and
before. Since the first Intel version of Mac OS X was 10.4 I changed
MACOSX_DEPLOYMENT_TARGET to 10.4 in cc.darwin.i386. In
cc.darwin.i386-64 10.7 is a reasonable value for
MACOSX_DEPLOYMENT_TARGET.
* Again, if I'm reading the case statement properly, the first case
appears to be for Mac OS 10.2 and before. So this first case simply
doesn't apply to cc.darwin.i386*. Therefore, I reduced the entire
case statement to just the second case in cc.darwin.i386 and
cc.darwin.i386-64.
* And now for perhaps the most contentious change, I removed the
-Wl,-flat_namespace option in all the cc.darwin scripts. Those who
are used to flat namespaces from other dialects of Unix think that
that's the way to go on Darwin. But, Apple's philosophy is that
two-level namespaces are better. As a result, they've geared the
dynamic loader to work better with two-level namespaces (at least
with Mac OS 10.3 and above as I seem to recall). That's why
two-level namespaces are the default when building dynamic libraries
on Darwin. In support of the change I have to make the following report:
After I tested all my fixes to ast-open-2014-09-29 I rebuilt it
with debugging turned off. This, of course, reapplied dynamic
libraries to the build. Then I tried "package test" again. In
http://lists.research.att.com/pipermail/ast-users/2014q4/004686.html
I reported a problem with the dynamic loader not being able to
find the symbol, _nv_lastdict, while attempting to run "regress
.../kshlib/dss/dss.tst". But, in this latest test run that error
went away. I don't have time to prove it, but I firmly believe
that the removal of the -Wl,-flat_namespace option in cc.darwin
inadvertently fixed this problem. None of the other changes I
made were to address this problem.
I've attached cc.darwin.diff, cc.darwin.i386 and cc.darwin.i386-64.diff
so that the changes can be applied. Note that the changes to
cc.darwin.i386 and cc.darwin.i386-64 have not been tested since I don't
have an Intel Mac.
Terrence Doyle
--- src/cmd/INIT/cc.darwin.orig 2013-11-07 11:04:31.000000000 -0600
+++ src/cmd/INIT/cc.darwin 2014-12-22 19:41:18.000000000 -0600
@@ -50,13 +50,13 @@
dll) # what a compatibility mess -- surely they can get the apis to play nice
tmp=/tmp/cc.${USER:-$LOGNAME}.$$.err
trap "rm -f $tmp" EXIT
- case `MACOSX_DEPLOYMENT_TARGET=10.7 ld -undefined dynamic_lookup 2>&1`
in
+ case `MACOSX_DEPLOYMENT_TARGET=10.3 ld -undefined dynamic_lookup 2>&1`
in
*undefined*dynamic_lookup*)
ld -m -flat_namespace -undefined suppress -dylib -dynamic \
-ldylib1.o "$@" -lcc_dynamic -framework System >$tmp
2>&1
status=$?
;;
- *) MACOSX_DEPLOYMENT_TARGET=10.7 $CC -Wl,-flat_namespace
-dynamiclib -undefined dynamic_lookup "$@" >$tmp 2>&1
+ *) MACOSX_DEPLOYMENT_TARGET=10.3 $CC -dynamiclib -undefined
dynamic_lookup "$@" >$tmp 2>&1
status=$?
;;
esac
--- src/cmd/INIT/cc.darwin.i386.orig 2013-11-07 11:04:46.000000000 -0600
+++ src/cmd/INIT/cc.darwin.i386 2014-12-22 19:31:05.000000000 -0600
@@ -50,16 +50,8 @@
dll) # what a compatibility mess -- surely they can get the apis to play nice
tmp=/tmp/cc.${USER:-$LOGNAME}.$$.err
trap "rm -f $tmp" EXIT
- case `MACOSX_DEPLOYMENT_TARGET=10.7 ld -undefined dynamic_lookup 2>&1`
in
- *undefined*dynamic_lookup*)
- ld -m -flat_namespace -undefined suppress -dylib -dynamic \
- -ldylib1.o "$@" -lcc_dynamic -framework System >$tmp
2>&1
- status=$?
- ;;
- *) MACOSX_DEPLOYMENT_TARGET=10.7 $CC -Wl,-flat_namespace
-dynamiclib -undefined dynamic_lookup "$@" >$tmp 2>&1
- status=$?
- ;;
- esac
+ MACOSX_DEPLOYMENT_TARGET=10.4 $CC -dynamiclib -undefined dynamic_lookup
"$@" >$tmp 2>&1
+ status=$?
egrep -v ' (warning .*multiple definitions|definition) of ' $tmp >&2
exit $status
;;
--- src/cmd/INIT/cc.darwin.i386-64.orig 2013-11-07 11:11:02.000000000 -0600
+++ src/cmd/INIT/cc.darwin.i386-64 2014-12-22 19:38:04.000000000 -0600
@@ -50,16 +50,8 @@
dll) # what a compatibility mess -- surely they can get the apis to play nice
tmp=/tmp/cc.${USER:-$LOGNAME}.$$.err
trap "rm -f $tmp" EXIT
- case `MACOSX_DEPLOYMENT_TARGET=10.7 ld -undefined dynamic_lookup 2>&1`
in
- *undefined*dynamic_lookup*)
- ld -m -flat_namespace -undefined suppress -dylib -dynamic \
- -ldylib1.o "$@" -lcc_dynamic -framework System >$tmp
2>&1
- status=$?
- ;;
- *) MACOSX_DEPLOYMENT_TARGET=10.7 $CC -Wl,-flat_namespace
-dynamiclib -undefined dynamic_lookup "$@" >$tmp 2>&1
- status=$?
- ;;
- esac
+ MACOSX_DEPLOYMENT_TARGET=10.7 $CC -dynamiclib -undefined dynamic_lookup
"$@" >$tmp 2>&1
+ status=$?
egrep -v ' (warning .*multiple definitions|definition) of ' $tmp >&2
exit $status
;;
_______________________________________________
ast-users mailing list
ast-users@lists.research.att.com
http://lists.research.att.com/mailman/listinfo/ast-users
