Created attachment 448720 [details] Build output I get this really odd error when I was doing an `emerge -auDNv world`. It looks like an invalid tar file, plus gzip fails with a backtrace. I've removed the offending file from /usr/portage/disfiles to force a re-download and also tried another emerge --sync. It doesn't valid on the hash verification, so did an invalid archive get hashed and committed/mirrored? I can't find any other bugs on this, which seems like it should be wide reaching.
Created attachment 448722 [details] emerge info
The tar file is fine. The problem lies in the CFLAGS with which your gzip was built ("-O3 -march=nocona"). As proof, the following command should be enough to reproduce the issue on your machine: gzip -d < /usr/portage/distfiles/libosinfo-0.3.1.tar.gz > /dev/null It seems building gzip with "-O3 -march=nocona" is incompatible with GCC's Stack Smashing Protector. Feel free to report this issue upstream. As a workaround, change "-march=nocona" to something more appropriate for your machine or replace "-O3" with "-O2"; then rebuild gzip.
I've been copying this old image for a while and have never adjusted my CFLAGS. :) I switched to -march=native, recompiled gzip and libsosinfo was able to ungzip and install. I find it interesting this didn't affect any other .gz files. Who should this be reported to upstream? Gzip?
It is indeed peculiar, though I cannot reproduce this with gzip 1.8 on a different OS using GCC 6.2.1 and the same compilation flags, so perhaps this is a bug in older GCC versions or something Gentoo-specific. If you really want to investigate further, you can emerge an unstable version of GCC on your box (e.g. 5.4.0), rebuild gzip and then see if the bug is still present. If not, then it is probably a compiler bug already fixed upstream. If the error persists, well... keep trying :)
Does this affect any newer versions (4.9.4, 5.4, 6.x)?
(In reply to Andreas K. Hüttel from comment #5) > Does this affect any newer versions (4.9.4, 5.4, 6.x)? No reply.