On 2016-06-27 20:01, Greg Stark wrote:
On Wed, Jun 22, 2016 at 7:09 PM, Greg Stark <stark%mit.edu@localhost> wrote:I now have a working NetBSD network root install for 6.1.5 (because 7 has the broken GCC and that stops openssh from working). It's been building Perl 5.22 from the pkgsrc 6.1.5Q2 stable source for about a day now.My NetBSD 6.1.5 machine has been chugging away building pkgsrc packages I need for a while now. A couple questions... 1) Would it be useful if I uploaded the packages I've built in this process? In the binary archives there are only a fraction of the pkgsrc packages.
It could be useful, as it takes forever to build 'em. :-)That said, we should probably do it on a more current version than 6.1.5 for this.
2) Three times now I've seen spurious faults on illegal instructions. Twice was "awk" and once was some other binary. Occam's razor says this is just awk doing something like log(0) or some other unsupported libm call that's documented to cause an illegal instruction fault. But there are two reasons I'm skeptical. a) I was subsequently able to configure and build the same packages without a problem and dozens of other packages and b) The stack traces in the core dumps look bogus under gdb. They have only one or two frames which have implausible addresses.
Did you clean out the directory in between? Otherwise a retry will not do that file again, as you will end up with an empty .o file, which make considers as valid, and it will move on to the next file.
And with that said, the problem is usually not something like a log(0), but with problems with the native toolchain.
Johnny