Re: [buildd] r-base
On Wed, Feb 21, 2007 at 02:04:46PM +0100, Michael Schmitz wrote:
> > Hi,
> > I tried to build r-base on aahz, it got a bit further than kullervo.
> > kullervo failed with an ICE:
> >
> > gcc -I. -I../../src/include -I../../src/include -DHAVE_CONFIG_H -fpic
> > -std=gnu99 -O3 -pipe -c vfonts.c -o vfonts.o
> > gfortran -fpic -O3 -pipe -c xxxpr.f -o xxxpr.o
> > <built-in>:0: internal compiler error: Segmentation fault
> >
> > ahhz built that fine, but a little later the whole machine froze:
> >
> > building package 'base'
> > mkdir -p -- ../../../library/base/demo
> > mkdir -p -- ../../../library/base/po
> > mkdir -p -- ../../../library/base/man
> >
> > I guess the command that made the machine freeze is not shown here.
> > This is from dmesg on aahz, lots of NETDEV, some hda and lots of eps0
> > messages after that:
> >
> > NETDEV WATCHDOG: eth0: transmit timed out
> > NETDEV WATCHDOG: eth0: transmit timed out
> > hda: status timeout: status=0xff { Busy }
> > ide: failed opcode was: unknown
> > hda: drive not ready for command
> > ide0(?): unexpected interrupt, status=0xff, count=1
> > esp0: Gross error sreg=7f
> > esp0: No current cmd during gross error, resetting bus
> > esp0: Resetting scsi bus
> > esp0: Gross error sreg=7f
> > esp0: No current cmd during gross error, resetting bus
> > esp0: Resetting scsi bus
> >
> > The system is install on hda, the SCSI disk is not used for building, maybe
> > I should disconnect it? Can R be responsible for SCSI or IDE errors???
(I missed this at first)
> Unlikely - is that the debug=mem and AmigaOS dmesg output?
Agreed.
> R might be heavy on swap, thus triggering disk problems, but the kernel
> should recover from that.
God no. Look at Quantlib -- THAT is heavy on swap. Takes me around 90
minutes on a 2 gb dual-athlon 1.8 GHz. C++ and lots, I mean lots, of
templates.
R is just ANSI C and Fortran. It may be that the fortran compiler is
killing you again. In the R Core team, Brian Ripley (who is actually
rather knowledgeable) still recommends sticking with f77. I overruled
that for Debian to be 'all gcc 4.*'. Try rolling it back. Also try
mucking down my debian/rules to use fewer configure option eg forget
Atlas / Refblas for a test build -- Christian knows how to do this as
he kindly maintained backports at CRAN for a while.
That said, I really don't know why you all are doing this become no
one will use it on m68k. But if you must, this is the route I'd take
for debugging.
Dirk
--
Hell, there are no rules here - we're trying to accomplish something.
-- Thomas A. Edison
Reply to: