[rfc] [patch] trivial patch to speed up bzr log

[Denys Duchier]

Robert Collins <robertc at robertcollins.net> writes:

> I'm concerned that this will double the memory footprint of bzr in large
> trees. (with -v on launchpad for instance, I'd be anticipating an 800Mb
> memory footprint with this change.)

Robert explained on IRC that during pull, many weaves are simultaneously live
and would all hold a cached copy of the last extracted revision.

> Caching an inventory as happens later in the thread is less of an issue
> for me - there is only one per branch :).

I understand, and for the case of "bzr log" that is indeed sufficient.  However,
wouldn't it be useful to have a generally safe way of caching stuff?  I believe
I have thought of such a scheme, provided that you are willing to let major
collections take care of clearing cached values.

The trick is to use an object with a self reference, that also holds the cached
value as an attribute.  Something like this:

import weakref

class Cached(object):
    def __init__(self, value):
        self.value = value
        self.self = self

we can then safely cache a value v as follows:

w = weakref.ref(Cached(v))

As long as no major collection occurs, w() returns the Cached object from which
we can retrieve the cached value v.  After a major collection, w() returns None.

What to you think?

Cheers,

--Denys