Re: [PATCH] hlist: drop the node parameter from iterators

[Sasha Levin <sasha.levin@oracle.com>]

On 01/30/2013 06:47 PM, Andrew Morton wrote:
> On Tue, 29 Jan 2013 20:08:30 -0500
> Sasha Levin <sasha.levin@oracle.com> wrote:
> 
>> Also, ping :)
>>
>> ...
>>
>>>  216 files changed, 1031 insertions(+), 1526 deletions(-)
> 
> Whimper.
> 
> I don't really see a sane way of avoiding a single huge atomic smash
> here.  Normally we'd use a multistep process:
> 
> 1: Create a new and differently named macro, say "sasha_is_a_pita()".
> 
> 2: Convert hlist_for_each_entry() to sasha_is_a_pita() in as many
>    sites as we can.
> 
> 3: Once we think all sites are converted, delete the now-unused
>    hlist_for_each_entry() definition.
> 
> Problem is, there just isn't any other identifier we can use here apart
> from hlist_for_each_entry().
> 
> I suppose we could add additional steps:
> 
> 4: Add hlist_for_each_entry(), which is identical to sasha_is_a_pita().
> 
> 5: Convert sasha_is_a_pita() to hlist_for_each_entry() in as many
>    sites as we can.
> 
> 6: Once we think all sites are converted, delete the now-unused
>    sasha_is_a_pita() definition.
> 
> But geeze.
> 
> 
> The alternative is to do the huge atomic smash immediately after the
> 3.9-rc1 release, when the amount of pending out-of-tree code is at a
> minimum.

I would accept a "it's not worth the effort" if you think it's not worth
the effort. I don't have a secret diabolic plan to make Linus and yourself
do a bunch of conflict merges because I like to see you suffer.

Regarding the multistep process, the downside there is that instead of
dealing with conflicts once you'll be dealing with them twice: once when
you move to the new macro, and once when you move back. If you think that's
preferable we can do that.

If not, should I send it over to you on -rc1?


Thanks,
Sasha