Hi Jeremy
On 2026/03/12 19:44, Jeremy Bícha wrote:
What are the downsides to providing a Debian Live image for Lomiri?
IMHO it would be ok to at least enable it for weekly builds for now to
see how it goes.
This is the first I've heard of a desire to discontinue some existing
Debian Live images, except for the January discussion here about not
wanting to build live images for oldstable any more.
I'm a strong proponent of radically reducing images, so that there is
more time/resources/space to add more images.
But, I have very different ideas for getting there.
There's a patch somewhere for live-boot or live-config (I can't find it
right now) that lets you specify a path within a squashfs for the
rootfs. This sounds great, because we have lots of duplicated content
across the ISOs and it might be possible to combine some ISOs... so,
let's experiment!
I extracted the squashfs filesystems for gnome, kde, lxde, lxqt, mate,
standard, xfce, and cinnamon.
How big is this directory?
46G extracted
Yikes. Ok lets see how much mksquashfs can shave off that with a huge
dictionary and xz compression:
4.4G extracted.squashfs
Woah. That's less than 10% of the total uncompressed data!
So, what if we could shave off enough unique parts to get it (along with
the kernel, initramfs, package pool) to fit on 4GB?
It's my next step to check, but what happens if dropping some big
packages like libreoffice (and its many translation files) could make up
for that last bit of space that we need? That means we could have one
generic live image where you could try or install pretty much all the
major desktop environments from.
Also, a single image has the benefit of better total seeding via
bittorrent, which is also something I want to push, because btfs works
great and you could also have a netinst installation image where you
boot a small stub from either USB, local network or over https and then
you could extract the squashfs over bittorrent, but I digress.
Now, reducing the live images to a single image doesn't really reduce
testing times so much. But, building this one big squashfs happens a lot
faster than building the 9 images individually. On my E5-2698 CPU,
building this squashfs took 36m58.627s. casulana, the Debian machine on
which these images are built, has two E5-2699 CPUs, so it should even
take just under half that time on there. On release day, we spend quite
some time waiting for these images, and if we can have an image
available much faster (and also a lot less downloading to do) then
perhaps we can make better use of people's time on release day.
Ah yes, the new images. As I've alluded to before, I'd really like to
take a stab at a netinst/netboot live image, but also I think we need
more architectures. There's an increasing amount of nice arm64 hardware
out there, and it looks like things like the new Apple Neo will sell
quite well, it would be a shame if we didn't have an image for it once
the hardware is properly supported. Also, more and more arm64 hardware
have nvme storage where installation works more like a traditional
installation than just writing an image to mmc. And also, it would be
nice to have a riscv64 image too, even if it's just for experimental builds.
Also, it's 2am here, so sorry if any of this sounds a bit incoherent, I
couldn't sleep so thought I would come and write this mail I've been
meaning to write all weekend :)
-Jonathan
