Re: [OE-core] [PATCH] kernel-module-split: fix conf file generation when KERNEL_SPLIT_MODULES=0

[Trevor Woerner <twoerner@gmail.com>]

On Sun 2025-03-16 @ 01:16:47 AM, Dixit Parmar via lists.openembedded.org wrote:
> Can you show the conf files for the same kernel configuration
> with and without the kernel_split_modules enabled ? That way
> we know there's no change in existing behaviour.
> > I have confirmed that in my testing. Can you suggest how I can
> share that information here?

What data are you expecting to find, and where are you expecting to find
it, that prompted this fix? I assume that in one case there will be
information/files populated in the image's /etc/modprobe.d and in the other
case (the buggy case) that isn't happening? Or is there something else?

Can you provide complete instructions and/or your configuration demonstrating
the problem occurring before your fix, and the problem being solved by your
fix? Simply enabling KERNEL_SPLIT_MODULES by itself won't show anything unless
the user is trying to enable some module(s) too.

I found that I had to go read the bugzilla report to better understand the
issue and potentially how to show it occurring. Keep the reference to the
bugzilla, but ideally the commit message would have all the necessary
information to understand and reproduce the issue.

> We also should make a test for this in the OE selftests. We
> are adding conditional code paths, so they should be tested
> to ensure no regressions in either in the future.
> > Never done that before. May be I can do it given some direction
> as separate patch.
> I'm curious about the above line. It is unclear to me why we'd
> only have this postinst be relevant if none was previously set.
> > Reverted.
> Is there really a scenario where the directory won't exist ? Isn't
> this just running in our own install phase ? So all prerequisites
> and directories should be in place.
> > Ideally no, we kept it for safer side, I have added log warning.
> The walking and sorting seems quite heavy.  Isn't this called from do_split_packages indirectly ?
> Do we really need to walk and gather the information ? Is this mainly for the case of no-split
> on the kernel modules ? If that is the case, isn't there a way to short circuit the processing
> on the split-package case ?
> > Litterally I could not think of anything else here and not sure of there
> are any short-circuit options. I have limited knowledge in this. I am open to suggestions
> if this is not the best solution at the moment.

It is customary that quoted text have the start-of-line ">" symbols (the more
symbols, the further back in history the comment) and that replies not have
any start-of-line character, as I'm doing here. Your email program should have
an option to do this automatically.