The only one we were uploading was `dmesg.log` and it no longer exists. Got lost in the rework of the image shrinking in !95. Avoids the `ERROR: No files to upload` at the end of every job.
It would appear that invent mangles files permissions so that we end up
with world writable files. Instead try to hold on to git. This may
ultimately cause trouble with docker but we'll have to try to know.
Relates to #27
- add ci builds
- add some scripting surrounding spinning up local containers
- use btrfs send-receive to populate the image for both speed, precision
and reliablity (won't miss ENSPC or something)
- no longer generate bootable images through mkosi, we generate two UKIs
anyway, mkosis extra work is entirely pointless and as it turns out
produces vastly larger UKIs for some reason. probably worth inspecting
at some point why that is
- the entire build has been moved from relying on arch packages to using
kde-builder. this effectively turns the way we build in development also
in the way we build for deployment, giving greater familiarity and
exercise of code paths
- move initrd from dracut to mkinitcpio as it is better integrating with
the arch base
- the initrd system is still driven by systemd rather than busybox (the
initcpio default)
- calamares tech still doesn't work entirely but got a bit of polish
- basic session blessing support (tracking whether an image is good or
not)
- boot failure tracking via the UKI filename (foo_1+3-1.efi etc)
- pam files imported from arch
- font files imported from neon
- aggressive debugging via systemd is on for the time being to aid with
development
- snapd support
- general AUR support (really cruddy because we are inside a bwrap that
runs with no-new-priviledges which means we can't sudo but arch tools
don't want to run as root - so everything is awkwardly detangled with
dependency installation happening manually in root scope, then the build
as a temporary user, then the installation in root scope again)
-
