Re: code cleanup

On Wed, 28 Apr 2004, John Baldwin wrote:

> On Wednesday 28 April 2004 02:26 pm, Julian Elischer wrote:
> > On Wed, 28 Apr 2004, John Baldwin wrote:
> > > On Wednesday 28 April 2004 02:26 am, Alex Lyashkov wrote:
> > > > Hi All
> > > >
> > > > how i see many points at kernel work with allproc list direct, but
> > > > proc.h introduce macros FOREACH_PROC_IN_SYSTEM.
> > > > This patch clean this places.
> > >
> > > I'd actually rather see the FOREACH_PROC macro removed, I don't think
> > > hiding the fact that it's a TAILQ is all that useful.

I'd rather it had never been.

> > it makes it possible (well, easier) to do:
> >
> > FOREACH_PROC_IN_SYSTEM(p) {
> > 	FOREACH_KSEGROUP_IN_PROC(p, kg) {
> > 		FOREACH_THREAD_IN_GROUP(kg.td) {
> > 			something(td, kg);
> > 		}
> > 	}
> > }
> >
> > Which is a lot easier to read and understand
> > than the expanded version. You don't kave to remember the linkage
> > pointer's names and you can add debugging  to it
> > and check that the correct loks are held etc.
> > (the latter being a major reason I did it).

This macro seemed more reasonable when it was added because its scope
was limited and I thought it was temporary.

> Note that the allproc_lock protects the allproc list.  W/o the FOREACH_PROC
> macro, I can grep for 'allproc' in the source tree to find all users to
> verify locking, etc.  With the extra macro, I now have to do multiple greps.
> When you multiple the effect with several wrapper macros, it now becomes much
> more work to work on locking the lists of structures since you have to do
> multiple greps to find the places to look at.  I think remembering the
> linkages for lists is actually quite important to avoid using the same
> linkage for multiple lists incorrectly.

Macros are bad for debugging.  The above is a sort of high level aspect
of debugging.  One low level one is when need to look at the linkages
for the lists.

Bruce