Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!GSBACD.UCHICAGO.EDU!phd_ivo
From: phd_ivo@GSBACD.UCHICAGO.EDU
Newsgroups: gnu.g++.bug
Subject: I've sent this to comp.sys.next. Maybe it helps your or the gcc people.
Message-ID: <8905240246.AA02559@life.ai.mit.edu>
Date: 24 May 89 02:36:00 GMT
Sender: daemon@tut.cis.ohio-state.edu
Distribution: gnu
Organization: GNUs Not Usenet
Lines: 65


I  spent the better  part of yesterday  trying to  install  g++.   The
installation documentation is not accurate, and  thus a real  problem.
For example, among the directories g++ expects to exist, is  gcc-test/
.  This directory exists  neither on my NeXT,  nor  is it mentioned in
the  current  gnu c 1.35.0  distribution.   Moreover, bison.simple was
hardly  ever mentioned, but some  of the  construction files  create a
link to a bison.simple in the gnu cc (I think) directory; anyway, it's
not  there on Next  but elsewhere (for non-NeXT   computer: it doesn't
come in the gcc tar files, either!).

I have never tried to  install either  gnu c nor g++.  One  thing that
astonished me is how many warning messages I got when I compiled gnu c
1.35.0 with the NeXT version of gnu c. Many of these seem benign,  but
as a user, one never knows. The  following are some  observations that
may make it easier for someone who knows what he is doing (I don't) to
go about installing g++ on NeXT.

----------------------------
The instructions mean that one should do a
	make -f Makefile maketest
Specifying  this  line  would  make   it much cleaner   and easier  to
understand than 'do a make of the "maketest" target in the makefile'.

Among the more disconcerting compile-time errors were a  definition in
cplus-lex.c
	extern double atof ();
which in NeXT's /usr/include/stdlib.h was moved to be translated into
	#define atof(nptr) strtod(nptr,(char**)NULL)
Obviously, this causes syntax error.

A similar problem appears in expr.c (due to---I  think---gnu  c 1.35.1
expr.h file).  What happens is that CONST0_RTX() is a  macro including
a conditional expression that looks like a typecast abort
	(rtx) abort()
Well, on NeXT, abort() is void and causes the cast to be a syntax error.

In final.c, the problems were getting more serious. A  tmp file  co(.s
at 4026) signalled an unknown  pseudo-op. I have no  idea  how to  fix
this.

In  genrecog, function fatal, I  decided to check some of  the warning
messages about improper casts. The function uses an  integer passed as
the string control argument to fprintf. I realize that  the purpose of
this exercise  is not to  have to program  in C  (but in G++), but  it
seems that this  sort of warning can be  easily cleaned up, and  would
save some wondering.   (I think genrecog  may  come   from gcc,  so no
criticism of G++ intended).

My final hour came when cc of intry-output.c died with signal 11.

Some  other problems were redefinitions of   __builtin_alloca, mallocs
and  NULL. I assume  NeXT has  decided  on its own   versions of these
routines.

-----------------------

In  any  case, installation  of  g++   on  NeXT   is not   a   trivial
streightforward exercise.  It  seems that there are many  other people
who would like to have a version of C++  running, so a  porting effort
by someone   "in the  know"  would  be   very  worthwhile  enterprise.
Definitely,  please  let me  know  if  you (the  unknown  porter) have
succeeded.

/ivo	phd_ivo@gsbacd.uchicago.edu