[Dev] Re: refcounted
vajda at osafoundation.org
Thu Oct 28 18:52:21 PDT 2004
> Also, I was curious to know why you prefer to leave non-pinning default when
> we have no plans to use it and Ted, Morgen, Lisa and I prefer the opposite?
As discussed with you several times before, the 'refcounted' mode, just like
the item stamping feature, is a, so far, non portable python-only feature.
And it's somewhat experimental still.
Until I can see a way to extend it to another language such as Java - ideas
exist but remain to be proven - I don't want it to be the default since it is
a pervasive programming style change. As long as this feature is not portable
elsewhere, I do not want to promote it since code written in refcounted style
is difficult to port to environments not supporting it - one has to
figure out all the places where pinning, the next best - or worse - solution,
is to take place.
If Chandler has no plans to ever use anything else, that's perfectly fine,
just don't, I have no objections to that. It is your responsible choice as to
how much portability you wish to build into the Chandler codebase. I don't
feel comfortable taking that responsibility by making it or promoting it as
Furthermore, I don't think that it not being the default is putting any undue
burden on you or that your-life-is-made-any-harder this way. In Chandler,
there is one place where you set this, and in the unit tests, one other.
There are a number of other configuration flags that need to be set when
opening or creating a new repository and, as performance is further focused
on, I expect the number of such flags to increase and these flags to become
As cool as this feature is, I do not endorse it as long as I care about
More information about the Dev