Sam Steingold writes:
> > * Don Cohen:
> > I do a build, then make distrib, then copy the .tar.gz to another
> > machine where make fails with messages about undefined references to
> > dcngettext. What's to be done? It seems to be related to libintl.
That worked, in the sense that after I unpack the distrib the make
works without complaint. But running it gives a segfault.
does not get an error but adding -M /base/lispinit.mem does.
I could send an strace, I suppose.
gdb didn't show much, of course.
I suppose the next step would be --with-debug ? But of course that
seems to fix all such problems. So maybe gcc -g ?
I hope the following new data doesn't confuse things too much.
For clarification, the distrib that segfaults on another machine
was built from cvs as of ~10/3 with a few compiler fixes found in
the mean while. That was built on the machine I call h1, which
I characterize as
issue rh 7.0
The distrib segfaulted on one I call number4
issue rh 6.2
I now get a new cvs on number4, build it without incident,
it passes all the tests (whereas the one built on h1, as
I reported earlier, failed this one
Form: (PROGN (DEFCLASS T1 NIL ((FOO :ACCESSOR FOO :INITFORM :FOO)))
(LIST (FUNCALL (COMPILE NIL (LAMBDA NIL (TYPEP (MAKE-INSTANCE 'T1) 'T1))))
'(TYPEP (MAKE-INSTANCE 'T1) 'T1)))))
CORRECT: (T T)
CLISP : (T NIL)
), and when I distribute it the result works on h1.
I think the lesson is that builds on older software run on newer
software, but not vice versa. So always build on the oldest you can