Problems building egcs-1.1.2 on NCR MP-RAS 3.02

["Steve Huckle" <steve at yossarian dot demon dot co dot uk>]

Hiya,

Apologies to all those who subscribe to gnu.gcc.help or
comp.sys.ncr....since you've already seen all this, but below are details
regarding problems I'm having building egcs-1.1.2 on an NCR 4300 running NCR
MP-RAS 3.02.  Apparently this might be the best place to post my SOS's (thnx
Tim)....indeed it makes perfect sense to me ;-)

If anyone can help me out b4 I go mad then I would v.much appreciate it.

Thnx

Steve H

============================================================================
=
Hello,

.....the plot thickens %-<

I have just built (with absolutely no problems) gcc-2.7.2 on the same
machine I have been having lots of problems trying to put together gcc-2.8.1
& egcs-1.1.2 (I incorrectly said in the original posting that this was an
NCR5100 (micro channel based), when in fact it's a 4300 (PCI based)).

....egcs-1.1.2 still reliably core dumps during the build whilst running
cc1obj.  Tracing cc1obj whilst it runs I believe its trying to run something
at an illegal address.

So my question is now be what's changed from gcc-2.7.2 to egcs-1.1.2 that's
stopping me completing the build?

....however, here's where the plot thickens.  I also have an old machine
around (an NCR 3400) that's running NCR MP-RAS 3.0 (I am currently trying a
build on a machine running 3.02) .  I tried the egcs-1.12 build there...&
guess what...it sailed through!  ...argghhh.

...I am currently looking for help on 3 fronts then...this newsgroup, the
comp.sys.ncr newsgroup, and NCR's Global Support Centre.....but my problems
are all a bit circular....is it NCR's problem or FSF's?...or neither in
which case I'm stuffed!

.....banging my head against a large wall....but then it wouldn't be fun if
everything worked would it! 8^

Steve H

============================================================================
=================================

Hiya,

Following the advice offered below, I desisted trying to build gcc2.8.1 and
started
on an egcs-1.1.2 build for an NCR5100 running NCR MP-RAS 3.02.

However, I'm still hitting problems.  After running CC=gcc CXX=g++
MAKE=gmake
../configure --host=i586-ncr-sysv4.3.03, and then gmake bootstrap (as
suggested by the distribution), The build fails here:

../cc1obj -print-objc-runtime-info tmp-runtime >>objc/runtime-info.h
Segmentation Fault - core dumped
gmake[2]: *** [objc/runtime-info.h] Error 139
gmake[2]: Leaving directory `/tooldir/src/egcs-1.1.2/gcc'
gmake[1]: *** [bootstrap] Error 2
gmake[1]: Leaving directory `/tooldir/src/egcs-1.1.2/gcc'
gmake: *** [bootstrap] Error 2

I traced the cc1obj program, here's the output.  It looks like cc1obj is
trying to write somewhere it shouldn't:

19551:  execve("./cc1obj", 0x08047BFC, 0x08047C0C)  argc = 3
19551:   argv: ./cc1obj -print-objc-runtime-info tmp-runtime
19551:  xstat(2, "/usr/lib/libc.so.1", 0x08047ABC)      = 0
19551:  open("/dev/zero", O_RDONLY, 0)                  = 3
19551:  mmap(0x00000000, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE, 3, 0) =
0xBFF9
9000
19551:  close(3)                                        = 0
19551:      Incurred fault #6, FLTBOUNDS  %pc = 0xBFFB3744
19551:        siginfo: SIGSEGV SEGV_MAPERR addr=0x08858828
19551:      Received signal #11, SIGSEGV [default]
19551:        siginfo: SIGSEGV SEGV_MAPERR addr=0x08858828
19551:          *** process killed ***

....any ideas?

Original mail===============================================================
>I am having big problems building gcc-2.8.1 on an NCR MP-RAS 3.02 UNIX box.
>I am using gcc-2.7.2 and GNU make 3.77 to do the task.  I am used to
>building GNU utilities....but have never come across such an a**e ache as
it
>is proving to be on NCR kit.

 If you mean to use your GCC for serious work, the buggy one-year-old
2.8.1 is the wrong selection. The current egcs-1.1.2 (= the current
official GCC) is much better to start with.  Please look at:
ftp://egcs.cygnus.com

>Anyways, 'snuff drivel.  I have configured the Makefile, by typing:
>
>CC=gcc CXX=g++ MAKE=gmake
>./configure --host=i586-ncr-sysv4.3.03 --prefix=/tooldir/gcc/gcc-2.8.1

 I built egcs-1.1.2 for 'i486-sysv4.2' target (UnixWare 1.1) a couple
of weeks ago, and had no problems... But your system is newer and
comparable to some newer UnixWare version, 2.1 ?  UW 7 ?  So, browsing
the 'comp.unix.sco.programming' etc. SCO newsgroups and searching any
comments about building GCC for UnixWare will surely help...  You
probably find the recommendation to use egcs in every comment...

 I don't remember how things went with the gcc-2.8.1 one year ago..

>Then I run gmake LANGUAGES=c.  This all appears fine, except for the
>following warnings/errors that get output:
>
>include/signal.h:75: warning: `struct sigstack' declared inside parameter
>list
>include/signal.h:75: warning: its scope is only this definition or
>declaration,
>include/signal.h:75: warning: which is probably not what you want.

 I didn't find any reference to 'struct sigstack' in the UW 1.1 and
MIPS-SVR4.2MP headers I have, so the struct definition is in some
header which hasn't been '#include'd into your
'/usr/ccs/include/signal.h' , or somewhere else, which GCC then copied
to the '<build>/include'.

  Anyway you can look (with 'grep sigstack *.h' ) where it is defined
and fix your copy of 'include/signal.h' to include the header
somewhere in the beginning of 'include/signal.h'.

>./enquire -f > tmp-float.h
>Abort - core dumped

  If the 'enquire' doesn't work when you compile it with the
stage1-compiler, the compiler which was compiled using your old
gcc-2.7.2 , it is quite vain to continue...

>./genattr ../../src/gcc-2.8.1/config/i386/i386.md > tmp-attr.h
>Abort - core dumped

 There is now some basic problem, when all executables crash when
tried with the stage1 GCC...

>fixproto: populating `include'
>fix-header: fixing rpc/auth.h
>fix-header: fixing X11/Xfuncs.h
>fix-header: fixing X11/extensions/xtraplibp.h

 If you 'fix it, if it ain't broken', you may broke it...

 Using a 'fixincludes=Makefile.in' in the 'configure' entry for
'i[34567]86-*-sysv4*'  will disable the system header fixing...
I would recommend using this as the first step...  If there is no
'fixincludes=something' -row, the generic 'fixproto' will probably be
used...

  At least you should think that you are wiser than the stupid machine
and check what on earth it did to the 'fixed headers'...  The
'fixproto' is a generic script, and may do some really insane things
to the headers, if it doesn't understand anything...

 Just do a 'diff -cr include /usr/ccs/include | more' (or something)
to see the differences...

 There is something more suitable for the SVR4 header-fixing,
'fixinc.svr4', or something...

 Ok, my advices are to use egcs-1.1.2 instead, disable or change the
header fixing script,  check the 'sigstack' problem (if it still
exists...) and make sure that the stage1-compiler works (the enquire
is quite good a test for this...), before going to the stages2 and
3...

 You can try to get the gcc-2.8.1 stage1 to work as an learning
experience, but later the egcs-1.1.2 should be built for production
use....

 Cheers, Kai
================================================================