A corollary to the below: loggy will stick around for a couple/three
weeks. Go ahead and clean it, change it, or treat it with disrespect.
As you wish.
I'm currently investigating how to get rid of all "entries-hash"
handling in our code. That will move us to reading/writing single
entry_t structures. Some performance gains, I'm hoping. But also a
good chunk of conceptual simplicity since *writing* a full hash of
entries is implemented a bit wonky in our system right (delete
everything, write it all back), and that cannot stick around as we
move to a one-directory solution.
Cheers,
-g
On Mon, Sep 21, 2009 at 03:40, Greg Stein <gstein_at_gmail.com> wrote:
> Hey all... figured that I would drop my current thoughts on wc-ng...
>
> A few days ago, I got my workqueue stuff checked in, and then I've
> added a couple operations to it. In that process, I've discovered that
> the bulk of the work done by those is manipulating the pristine text
> bases, and the property files. Yet, in our future, complete wc-ng
> environment *none* of those operations would be necessary. In other
> words, the vast majority of that work is directly caused by *pre*
> wc-ng systems.
>
> I think it is going to be inefficient to continue on the path of "move
> loggy to workqueue" in a brute force fashion, like I've just done. If
> we could get to the new pristine system, and get the props stored into
> the database, then these workqueue operations would be *much* simpler.
>
> So I believe that my next step is to get the pristine files and the
> props advanced to our next generation. There are a lot of assumptions
> in our codebase that need to be solved to continue forward. Much of
> the code is not ready or flexible enough for the wc-ng approach. There
> are a lot of bits that assume various files, and they manipulate those
> files, and shove them around, and do whatever. Lots of trouble.
>
> From my investigation this past evening, we have a giant knot. We have
> pristines, properties, per-dir, loggy, and workqueue all tightening
> into a ball. The simple edge cases are getting fixed, and leaving us
> with an ever-nastier ball of twine to unravel. This email is kind of
> an update to say that I'll be teasing at that knot.. trying to find
> the right piece to unwind it all. It's a bit hidden now.
>
> Two action items:
>
> 1) continue the shift to wc-ng by removing adm_access and entry_t
> 2) find me on IRC; I'd like to bounce un-knotting ideas around
>
>
> Obviously, any questions, comments, ideas, suggestions, and patches
> would be very helpful. Email discussion is great, and IRC is faster;
> either works.
>
> Cheers,
> -g
>
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=2400139
Received on 2009-09-25 10:58:54 CEST