Kernel package and modules

[Werner Almesberger]

Andy Green wrote:
> If they're really "never used", let's stop making 'em.  Otherwise,
> someone is using them, and when they found they wanted them can be
> unpredictable.

Yes, that's what I meant: never used on a given system, but obviously
deemed potentially useful for someone. So the package paradigm would
make sense here.

> On the issue of modules out of sync with monolithic kernel, moving to
> modules in one package won't "substantially improve" loss of sync it
> will eliminate it.

Yes, but as long as a user stays within the boundaries of the package
management system, why would loss of sync even occur in the first
place ?

We have one example how sync can be lost from our end, namely because
a change in the kernel modularization wasn't coordinated with the
packaging. But now that everyone's aware of this, it should be
avoidable, no ?

The total space eaten by all the modules is enormous. Even on GTA02,
some 20% of total NAND storage would be taken up by them. So I think
just installing all the modules isn't a viable option in that setting.

> Similar as the effect of RREQUIRES we got now
> apparently, it will also pull in all the module updates each time.  So

Hmm, that sounds wrong. It should only update modules that have
actually been selected, not just the whole nightmare. In the latter
case, there would indeed be no point in having packaging with finer
granularity.

- Werner