Hi Vagrant!

Tangent: I sense some undeserved mysticism surrounding squashfs. 
It is not designed to be loop-mounted, any more than ext2 was.  It 
does not enjoy it.  People should stop doing it.

But they won't, because many distributions still insist that the 
same installer image must be both a bootable CD/DVD *and* boot 
when dd'd to a USB drive, on every PC ever made.

That ‘isohybrid’ dream justifies doing unmentionable things to an 
iso9660 file system (and only an iso9660 file system), so they 
must put the real squashfs on top of that and loop-mount it and 
ignore the screams I guess and--

...sorry; I got carried away.

Vagrant Cascadian 写道：
> Well, the suggestion to use squashfs does bear merit;

It's not a *bad* suggestion, just a bit obvious.

> it would require
> having some type of writeable filesystem on top, such as using 
> overlay
> fs to mount the installer rootfs with squashfs for the readonly 
> bits,
> and tmpfs for the writeable bits.

We've always done this.

> As a bonus, using a tmpfs overlay would solve the issue brought 
> up
> recently by someone who tried using the same installer image 
> multiple
> times, and /gnu/store and /var/guix got out of sync due to the 
> cow-store
> only writing to the newly installed system, so that the second 
> install
> failed.

...so no, it definitely wouldn't, but I think it's valuable to 
understand why you thought so!

Could you elaborate?

> Another angle might be to use a compressable but writeable 
> filesystem
> (btrfs?).

A persistently mutable . . . installation medium . . . ?

Very lost, slightly frightened,

T G-R