Bugs with rich-root-packs

[Martin Pool]

On Thu, Mar 20, 2008 at 9:46 AM, John Arbash Meinel
<john at arbash-meinel.com> wrote:
>  Alexander Belchenko wrote:
>  | It seems like today (at least in bzr 1.2 and 1.3rc1 problem exists)
>  | upgrade to rich-root/rich-root-packs format is completely broken.
>  |
>  |     https://bugs.launchpad.net/bugs/177874
>  |     https://bugs.launchpad.net/bzr/+bug/202884
>  |     https://bugs.launchpad.net/bzr/+bug/203607
>  |
>  | I'm inclined to mark these bugs as critical. What core guys think about it?

In general, if you think something is a critical bug, I would rather
that you mark it as such and then we downgrade it later if necessary.

As far as I can see they are duplicates of 177874 from 2007-12.

>  By hacking in a bit of path awareness into the KnitGraphIndex code (I'm
>  including the diff), I can see that it is indeed TREE_ROOT causing problems, so
>  it seems like it is bug 177874.
>
>  Specifically, the error is:
>  bzr: ERROR: Revision {('mbp at sourcefrog.net-20050711054655-47ea25e1f75a8813',)}
>  not present in
>  "<KnitGraphIndex(CombinedGraphIndex(<bzrlib.index.InMemoryGraphIndex object at
>  0x021D2B30>) file_id:TREE_ROOT)>".
>
>  So I'm not sure if it should be creating the nodes at that point, since it is
>  the InMemoryGraphIndex which is failing.
>
>  I don't know that these are release blocking bugs, as I would guess this isn't a
>  regression. It *is* something that needs to be fixed before we make rich roots a
>  default. (Which was discussed at the sprint, at least a --pack-1.3 which is
>  rich-root-pack with a couple small updates.)

It has been hit by several people, it leaves them pretty solidly stuck
when they do hit it, and it seems like it may be hit by many people
using bzr+svn.  On the other hand the point of time based releases is
to get them out on time with what has been landed, and this is not new
to this releases.

John suggests <https://bugs.launchpad.net/bzr/+bug/177874/comments/2>
that we should special case

 if root_id == ROOT_ID or root_id.startswith('tree_root'):
   # Ignore if there is a missing versioned file

So overall I think we should go ahead with 1.3 today but prioritise
this for 1.4, or even do a 1.3.1 next week.

-- 
Martin <http://launchpad.net/~mbp/>