Re: CVS commit: src/sys/arch

[Andrew Doran <ad%netbsd.org@localhost>]

On Sat, Dec 07, 2019 at 07:12:10AM +1100, matthew green wrote:
> > > > Why? I consider this totaly useless bloat, what was wrong with the boot.cfg
> > > > solution?
> > > 
> > > policy:  no default modules in the installation unless licenses
> > > issues force such, until module+kernel solution.
> > 
> > OK, but this is gone awry. The boot.cfg solution is great if anoyne
> > really needs this dmesg sugar, everyone else can go on using pcictl 
> > later instead.
> > 
> > So now the (reasonable) policy forces a (relatively harmless, as easily
> > overidable but still useless) change into bloat for everyone. Can we
> > just revert to the state before the initial boot.cfg change instead?
> 
> i'm not really a fan of this solution.
> 
> this code was removed from GENERIC back when ad was pushing
> hard on modules and reducing memory consumption using them,
> but we've got problems still to resolve about their safe
> usage.  (i really can't agree with your "harmless" line
> above.  modules are dangerous even with careful usage
> currently.)
> 
> so, this change is really about reverting a bad choice made
> 11 years ago, that still has not been properly resolved.
> we lost that functionality in netbsd 5.0 GENERIC with the
> idea it would be resolved "soon".  it's been a long, long
> "soon"...

Right.  And since then the kernel has doubled in size.  I don't like the
bloat, very frustrating, and I still think modules are the way to go.  But
whatever about the correct direction technically, people really seem to like
PCIVERBOSE so I'm throwing my hands up and saying that in retrospect, given
the way things have panned out, removing it was the wrong choice.

Andrew