OE recompile pending tonight

I posted about Xephyr X server missing:

https://bkhome.org/news/202203/xephyr-nested-server-missing.html

But was still unable to run Dunfell desktop in container, the 'pflask' utility failed. I chased this down to the 'libcap-ng' library not recognizing "CAP_SYS_MOUNT", which is a new Linux Capability that I patched into the Linux kernel.

I have posted about CAP_SYS_MOUNT before, for example here:

https://bkhome.org/news/201805/isolated-bootup-using-capsysmount.html

Anyway, it seems that in bumping 'linux-libc-headers' from 5.8 to 5.10, /usr/include/linux/capability.h lost the CAP_SYS_MOUNT patched entry. I have fixed that. I think that this is what has made libcap-ng go wrong.

Changing the subject, I wanted to compile 'gitg', a GUI for git management, but it wants 'glib-2.0' >= 2.68, whereas easy only has 2.62.6. So I have bumped glib-2.0 to 2.70.4.

I am a bit nervous bumping a core package like glib-2.0, as it could break something else.

Anyway, we shall see. I intend to do a complete OpenEmbedded recompile overnight.

Another thing that I fixed is an annoyance. Some OE developers have created recipes with hard-coded revision number. This is the "PR" variable. I set a global value for PR, and individual recipes should not override it.

Currently PR = "r8", and this appears in all package names. However, some binary packages have the hard-coded override, for example 'db-1_5.3.28-r1-nocona-64.tar.xz' and 'libsdl-1.2.15-r3-nocona-64.tar.xz'.

I wrote a little script to remove all of the revision-number overrides in recipes.

EDIT 2022-03-18:
Started the compile about 2pm yesterday, has finished at 9.15am today, that is over 19 hours.
This is on my Lenovo PC, with Intel 8th-gen i3 CPU, 32GB RAM, and compiling on an external (usb3) 1TB SATA SSD.   

Tags: easy