Created attachment 4077 [details] diffoscope output, gawk This is a weird one, highly reproducible, maybe even 100%. I noticed this when building core-image-minimal (gawk_4.1.4-r0_i386.deb is built as part of the build). The .gnu_debuglink contains either: 1. gawk-4.1.4 or 2. gawk The .gnu_debuglink checksum is the same in both cases. 1. local.conf PACKAGE_CLASSES="package_deb" 2. local.conf: PACKAGE_CLASSES="package_deb" INHERIT+="rm_work" TMPDIR="/tmp/ramdisk/gawk" DEPLOY_DIR="/data/master-master/poky-contrib/gawk/tmp/deploy" Basically, when I use ramdisk as TMPDIR, I get different result. I could also reproduce this by building an individual package: $ bitbake gawk Build in two different folders, each containing local.conf as per above. In ramdisk I need to use "rm_work", else I run out of memory, but I don't think that is the culprit. I also witnessed other similar anomalies with respect to .gnu_debuglink: e2fsprogs-e2fsck e2fsprogs-mke2fs
Created attachment 4078 [details] diffoscope, e2fsprogs-e2fsck
Created attachment 4079 [details] diffoscope output e2fsprogs-mke2fs
I am not entirely convinced the culprit is different TMPDIR, the ramdisk IO is quite a bit faster than HD IO, so it could be a race/timing issue.
I have not seen this in last 6-8 weeks, so I am going to close this as RESOLVED/OBSOLETE. It may have something to do with gawk being upgraded to 4.2. Will re-open if encountered again.