Hint: Speeding up mksquashfs when building for foreign architectures (e.g. on amd64 for arm64)

[Stefan Baur <newsgroups.mail2@stefanbaur.de>]

Hi List,

I noticed that when live-build runs mksquashfs to create the squashfs
file for a foreign architecture, it does so by calling the mksquashfs
inside the chroot - which means it gets executed by qemu, rather than
using the host-native mksquashfs and in addition to that, this means it
has access to only one CPU core.

I have worked around this with a hack.

Whenever I'm building for a foreign architecture, I am basically
checking when ./chroot/usr/bin/mksquashfs starts to exist.
As soon as it does, I move it to ./chroot/usr/bin/mksquashfs.real and
replace it with a shell script.

This shell script echoes $0 and $@ into a semaphore file
/tmp/filesystem.squashfs.temp (inside the chroot), then checks for the
existence of said file, and as long as that file exists, simply loops
with a 1 second interval. Once this semaphore file is gone, the script
deletes itself, and moves the real executable back into its place.

Now, on the host, I start a background task that checks if
$CHROOTDIR/tmp/filesystem.squashfs.temp exists. While it doesn't -
you've guessed it - it loops with a 1 second interval.

It then parses the contents of the file to call the mksquashfs on the
host, in its native architecture, with the corresponding parameters.

Once the mksquashfs job completes, it deletes the semaphore file.

As mksquashfs is now running natively, and able to make use of all
cores, the job completes significantly faster.

I wonder if this functionality could be built into live-build directly,
somehow, so it becomes less hackish?

Kind Regards,
Stefan Baur