bzr-svn 0.4.9 - still with memory leaks?

[Matt Nordhoff]

Paul Moore wrote:
> Interesting. It sounds like the lessons from all this are:
> 
> 1. Make the initial cloning by pulling in batches.
> 2. If the repository has case clashes anywhere in history, you need to
> use Unix for the initial clone.
> 
> Fixing the leaks in bzr-svn would solve the first (at the moment,
> either the Windows binaries don't include the python-svn leak fixes,
> contrary to what I understand, *or* there are other leaks which
> haven't been addressed yet, *or* memory use is expected to grow as the
> number of revisions being pulled, which could do with being
> documented).
> 
> A robust solution to case (in)sensitivity issues in Bazaar would
> resolve the second one [1]. The second is the worst issue, as there
> isn't a workaround.
> 
> Paul.
> 
> [1] By "robust", I mean something that ensures that pulling in batches
> through a case issue results in an identical repository to the one
> you'd get on a case sensitive system - this is a lot harder than just
> resolving clashes in the working directory files.

If you don't create a working tree, I doubt case-insensitive FS issues
would get in the way. "bzr init-repo --rich-root-pack --no-trees" when
creating the shared repo, or "bzr remove-tree" in the branch if it
already exists. (You can also set a preexisting repo to default to no
working trees by creating an empty .bzr/repository/no-working-trees file.)

It would be faster anyway. Updating the working tree after every pull
would just be a waste of CPU and I/O.
--