Add a `clean` command that deletes obsolete packs and uploads

Making this a separate command rather than an option to pack doesn't make any difference. Let's just add the option as in bug 304320

On Fri, 2009-05-08 at 03:56 +0000, Tim Penhey wrote:
>
> I checked my other repository that contained the branch that I was
> branching from, and I noticed 101M of obsolete packs, and 60M in the
> upload directory. Now I know that nothing is going on on that machine
> either uploading to it nor pushing from it. However I don't know when
> is a "safe" time to remove code from these directories, but I do know
> that they are not giving a favourable size comparison against other
> DVCSs.

Just to note - stuff in uploads should be deleted; with the smart server
we should be quite good at deleting that these days. It would be
reasonable for 'check' to report on noise in uploads.

-Rob

I don't think this is a duplicate.

Robert has indicated that there are times when the obsolete packs shouldn't be removed as part of pack.

Having a `bzr pack --delete-obsolete-packs` that doesn't actually pack seems wrong to me.

Having a command that explicitly removes old "stuff" seems more intuitive. It is more than just obsolete_packs, but the upload directory too.

Other tools have a fsck command; 'bzr check' is our equivalent. Perhaps
a flag to check to repair things it can [and another to not spend days
getting to cleanup upload and obsolete-packs].

-Rob

Just mentioning that since we have 'clean-tree', 'clean-repository' or 'clean-repo' would seem to be an appropriate name.

I think other systems also tend to have something like 'garbage-collect' ('gc'). Which unreferenced data in 'obsolete_packs' and 'upload' would fit.

Also, last I checked deleting an upload file will appropriately kill the running process when it tries to install the file. It *might* first create unreferenced indexes in the 'indices' directory, so we probably also want to prune unreferenced items from there. You also can have .pack files in packs/ which are unreachable (because they are not listed in pack-names). This is usually pretty rare, but it would also be something along the same lines for what to collect.

We could alternatively introduce a garbage-collect command, that would start at just doing this, and then grow the functionality to remove revisions/inventories/texts that are not referenced. (Note that those sorts of things start to become dangerous in the presence of things like stacked-branches, where we can no longer tell what revisions are *actually* referenced.)