It turned out to be a disk problem. It seems that trying to write to a particular spot on disk (i.e., where the fink build was trying to write) made the disk crash until a hard reboot. I would have thought that SMART would log the error and the extended Apple diagnostics would have picked it up, but apparently not. Solution: buy a new hard disk and copy over the data.
We haven't yet sent the drive in for warranty service. If you're interested in what they say, let me know privately, since the case is closed as far as this list is concerned. Regards, -Ken On Sun, Jul 12, 2009 at 9:15 PM, Kenneth Arnold<[email protected]> wrote: > I've run into what's probably an Apple bug (not sure if hardware or > software) that fink consistently hits; I'm reporting it here so that > anyone else who hits it can find it on Google, as an example of a > mostly successful troubleshooting involving (but not caused by) fink, > and to give any low-level Mac geeks who are watching a chance to > speculate on what's going on before I call Apple. (In short: this is > not a fink bug, and help is welcome but not expected.) > > fink install gcc44 gcc44-shlibs > ... wait ~20 minutes ... > "libtool: link: ar rc .libs/libstdc++.a"... and the machine dies. Can > still drag windows around, and the clock still updates, but nothing > else; "Device not configured" errors on any terminal when trying to > run any command that access disk, or when ssh-ing in. Can't even tell > it to shut down. > > This is 100% repeatable, in at least 5 trials (probably another 1 or 2 > crashes were due to this, but we weren't watching carefully). The last > two times I even used smcFanControl to crank up all the fans to max > speed. It sounds like a jet engine, stays at a reasonable temperature > (every sensor between 30 and 50 degrees C) and still crashes at > exactly the same place. Running the builtin hardware test immediately > after such a crash shows no errors -- basic or extended. > > The machine is a "MacPro3,1" 2x 2.8 GHz quad-core Xeon. It was running > with 4 GB of Apple RAM (one chip replaced because we caught errors > with it) plus 16 GB of iRam, in the matched configuration Apple > recommends. For the last trial, I took out all the aftermarket RAM, > and unloaded the VMWare kernel modules > (http://iamthekiller.net/2008/11/vmware-and-kernel-extentions/). Still > the exact same crash. > > It's not overheating or bad RAM; the machine does heavy numeric > computation normally without a problem (mostly, but the other glitch > we've seen is very rare and elusive). I doubt it's the disk (before > one trial I did rm -rf /sw/src/fink.build/*..., and fsck -fy is > totally clean). And when I ran "make" in the partially-built object > directory, it did a bunch of stuff (including compiling a bunch of > Java) and finished successfully. (Anyone have a theory, or something > else to test?) > > Thanks for thinking about this, and thanks for all the hard work on fink! > -Ken > > > PS - I can duplicate this near 100% on demand, even with fans on jet-engine, > by: > > cd > /sw/src/fink.build/gcc44-4.4.0-1000/darwin_objdir/i686-apple-darwin9/libstdc++-v3/ > make clean > make -j16 > > Once I had to do that twice. CPU temperature spiked 5 or 10 degrees, > but still sub-40C. The last few lines printed on the terminal were the > exact same as always: > ranlib: file: .libs/libstdc++.a(atexit_arm.o) has no symbols > ranlib: file: .libs/libstdc++.a(eh_arm.o) has no symbols > ranlib: file: .libs/libstdc++.a(tinfo2.o) has no symbols > ------------------------------------------------------------------------------ _______________________________________________ Fink-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/fink-users
