On Tue, Apr 12, 2011 at 8:00 AM, C. Michael Pilato <cmpilato_at_collab.net> wrote:
>> You are looking at changelists as a way to learn how to move operations into
>> wc_db properly, but just like that temp table for notifications I don't see
>> this as the way to go forward.
>>
>> I really don't see why users want to add thousands of nodes to changelists
>> while we still don't support changelists on directories. And if it is just a
>> handful of nodes the old code worked fine.
>
> This was one of the wrestling matches that I had with myself when I started
> looking at this very bit of code that Hyrum has changed. As I *understood*
> it, we had an internal goal of losing the svn_wc__node_walk_children().
> "It's slow." But in some cases -- namely this one -- it just seemed like
> doing so would require adding obnoxious or otherwise unpleasant code.
>
> Changelist operations are, I would suspect, pretty rare, so if folks don't
> like the approach Hyrum has taken, I would suggest that he just revert the
> whole of his effort in this space, delete notes/wc_node_walkers.txt, add a
> note to the svn_wc__node_walk_children() docstring encouraging developers to
> consider using a more batch-based approach if possible when considering
> additional uses of the function, and then move on. If we're going to spin
> our wheels somewhere, let's not do it on our arguably lesser-used features,
> please.
The point of this entire exercise was not to make setting changelists
faster (though that is a nifty side effect), but rather to get more
insight into how we are going to deal with them when doing this type
of thing for use cases that do matter, such as recursive propset.
We've got changelists all over the code, and since the database can do
changelist filtering for us, I presumed that learning how to use that
capability would be a fruitful use of time. I guess the insight is:
"we can't do it using current methods."
I'll revert this work sometime today.
-Hyrum
Received on 2011-04-12 15:14:34 CEST