On Sat, Jan 09, 2010 at 01:38:04PM +0000, Mindaugas Rasiukevicius wrote:
> David Young <dyo...@pobox.com> wrote:
> > Previously, the code distinguished garbage collection, locking
> > acquisition, and resource-reclamation from each other by using
> > more, and more meaningful, subroutine names. Also, in the common
> > case, it collected and reclaimed garbage immediately at the end of
> > config_detach(9) instead of at some arbitrary time in the future (for
> > example, in the next call to device_lookup()).
> >
> > Your change obfuscates the code and needlessly postpones a lot of
> > garbage collection. Please back it out.
>
> Your code had local a list of devices to G/C - it is redundant.
Now there is a global list, and there is still a local list of devices
to garbage collect. That is not more redundant?
> Having
> resource freeing/destroying tied with last unlock operation will make sure
> that it wont be done with locks held - that is a common way. Perhaps
> names "lock/unlock" confuse you? It can be "enter/exit" or similar.
The names are no longer descriptive. "Enter" and "exit" are vague.
> Can you explain "arbitrary time in the future" and "needlessly postpones"?
> It does not change the previous behaviour.
You changed the previous behavior in two ways:
config_detach: used to collect the garbage it just created, in the
common case where no other thread was in the device tree. Now it
does not.
device_lookup: collects garbage where previously it did not.
Back the changes out, and let's discuss how to achieve your purposes
with a smaller change.
Dave
--
David Young OJC Technologies
dyo...@ojctech.com Urbana, IL * (217) 278-3933
