gerson@parc.xerox.com (Dan Gerson) (10/09/90)
Given the notes that Keith Gorlen kindly posted here a few days ago, I was able
to make NIHCL compile using Sun's version of the AT&T C++ 2.0 translator. However,
I did run across a few problems which I thought I would post in case it might
help others doing the same. (This is on a Sun-4 using SunOS 4.0.3)
1) NIHCL attempts to install the error facility into libC.a. However, under Sun's
C++, libC.a is rarely used since normally the shared libraries (libC.so.* etc) are
used instead. The symptom is that when you try to compile anything using NIHCL,
such as the nihcltest or examples, ld complains that it can't find various
error symbols.
To fix this, you either have to tell ld to do a static link instead of a
dynamic link (the default), or else you have to put the error facility elsewhere.
I did the latter, which meant either playing around with test/MAKEFILE,
vectest/MAKEFILE, and ex/MAKEFILE to find the library (since LFLAGS are not
passed down from Makefile), or else installing the error facility in both
libnihcl.a and libnihclmi.a. Since I don't expect to be using the error
facility outside of NIHCL, I choose to put it in the nihcl libraries. This
makes me happier than modifying libC.a.
(I suppose you could try to add the error facility to the shared libraries,
but that is a pain for various reasons.)
I would suggest changing the various makefile scripts to honor CLIBDIR and
LIB_ID when linking the tests and examples.
2) Related to this, NIHCL by default puts the error include files in
/usr/include/CC. If you change this, as you should if you put the
error facility someplace other than in libC, you have to not only update
ERRINCDIR, but also add the include file to the "I" list. Which is fine,
except that you have to add the second file with a -I prefix and modify
Makefile and test/MAKEFILE to do I="${I}" instead of I=${I}.
3) If you change any of the other error facility locations (which we do
in general since in our system's configuration /usr/local... is local to
the workstation, and things such as the NIHCL libraries are placed on
our file servers on paths such as /import/CC/local/include etc.), even if you
leave the libC installation alone, you need to modify the Makefile to do:
errorfacility:
cd ${ERRFAC}; \
${MAKE} -f MAKEFILE ${MFLAGS} ${COMMONFLAGS} \
ERRGENDIR=${ERRGENDIR} ERRTABDIR=${ERRTABDIR} ERRINCDIR=${ERRINCDIR}
cd ${ERRFAC}; \
${MAKE} -f MAKEFILE ${MFLAGS} install ${COMMONFLAGS} \
ERRGENDIR=${ERRGENDIR} ERRTABDIR=${ERRTABDIR} ERRINCDIR=${ERRINCDIR} \
CLIBDIR=${CLIBDIR} LIB_ID=${LIB_ID}
Otherwise, it was not propagating the ERRGENDIR etc bindings down in the
first make. Probably some of these bindings aren't needed, I didn't
bother to think much here.
Then it should compile and test fine. If you don't mind modifying libC,
you don't need to do all of the above, EXCEPT that you will then have to at
least modify the makefiles to specify a ld flag of -Bstatic, and do this
for all programs using NIHCL.
--
Dan Gerson gerson@parc.xerox.com
Xerox Palo Alto Research Center 415-494-4745
3333 Coyote Hill Road
Palo Alto, CA 94304
USA