Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!tut.cis.ohio-state.edu!wsl.dec.com!bothner
From: bothner@wsl.dec.com
Newsgroups: gnu.g++.bug
Subject: (none)
Message-ID: <8908220842.AC00623@decwrl.dec.com>
Date: 22 Aug 89 08:42:16 GMT
Sender: daemon@tut.cis.ohio-state.edu
Distribution: gnu
Organization: GNUs Not Usenet
Lines: 42

Subject: cc1plus 1.36.0- crashes for sony_news
Date: Mon, 21 Aug 89 20:50:38 -0700
From: bothner@wsl.dec.com

First, the config.g++ for entries for the Sony News are wrong.
I tried the following, which *seemed* to work:
        news800|news)
                configuration_file=xm-m68k.h
                target_machine=tm-news.h
                cpu_type=m68k
                ;;
        news-gas | news-g)
                configuration_file=xm-m68k.h
                target_machine=tm-newsgas.h
                cpu_type=m68k
                ;;

I then found that cc1plus crashes on most files in libg++.
E.g. I run the preprocessor on stream.cc to get stream.cpp, and then:
/usr/local/lib/gcc-cc1plus stream.cpp -quiet -dumpbase stream.cc -fstrength-reduce -fsave-memoized -opt -version -o stream.s
GNU C++ version 1.36.0- (68k, Motorola syntax) compiled by GNU C version 1.35.96.
default target switches: -m68020 -mc68020 -m68881 -mbitfield
I get a Floating point exception in
layout_record (rec=(tree) 0x1c04cc) (stor-layout.c line 758)
754       if (var_size == 0)
755         TYPE_SIZE (rec)
756           = build_int (CEIL (CEIL (const_size, record_align) * record_align,
757                              size_unit));
758       else
The reason is that CEIL is trying to divide by record_align, which is 0.
On entry to layout_record, record_align was set to 8 by:
605       int record_align = MAX (BITS_PER_UNIT, TYPE_ALIGN (rec));
It was then set to 0 in line 621:
618       if (CLASSTYPE_N_BASECLASSES (rec))
619         {
620           const_size = TREE_INT_CST_LOW (CLASSTYPE_SIZE (rec));
621           record_align = CLASSTYPE_ALIGN (rec);
622         }
Btw, there seem to be no fields in this class, since the
loop starting at line 624 is skipped.
	--Per Bothner
bothner@wsl.dec.com   ...!decwrl!bothner