There are a few cases where the userspace application has to be built as the same bitsize (32-bit/64-bit) as the kernel. In a multilib configuration where the default multilib does not match the kernel's bitsize this is very difficult for a new user to do. For example: a powerpc64 kernel would usually have a 32-bit (powerpc) filesystem. Something like 'crash' needs to be built 64-bit in this case. So the user is forced to know they have to build "lib64-crash" and not just 'crash'. There should be a way the user can ask for 'crash', and instead they get 'lib64-crash' to build and be installed as necessary. This is important because image / packagegroup recipes may have things like 'crash' hardcoded into them, but that package won't actually work on the target system.
Jackie, Is this still a problem for the crash package? If so, it would be good to fix early in 2.6 unless you have a low risk change now for oe-core-2.5M3.
(In reply to comment #1) > Jackie, Is this still a problem for the crash package? Yes, it still be a problem. > If so, it would be good to fix early in 2.6 unless you have a low risk > change now for oe-core-2.5M3.
I think it could be fixed by adding crash to NON_MULTILIB_RECIPES which is introduced by commit 25f7c6c329038b443d36074fff45a30ba3712f7a.
Fixed by https://git.openembedded.org/meta-openembedded/commit/?id=d80fe97166347006fcfee2dae9bc80ec666174ae