Re: [PATCH v2 3/3] pack-objects: honor '.keep' files

Junio C Hamano wrote:

Brandon Casey [off-list ref] writes:

quoted

  <-a>
    -create a new pack containing all objects required by the repository
     including those accessible through alternates, but excluding objects
     in _local_ packs with .keep

I have a feeling that it is debatable if this "fattening to dissociate
from alternates" is what people want.

I'm not sure I understand you here.

Andreas has suggested previously that 'repack -a' should pack everything,
including objects in packs with .keep. Is that what you mean?

With my current understanding it seems that that would muddy the semantics
of repack. If -a does not honor packs with .keep, then would it be intuitive
to expect that adding -l (i.e. exclude alternate packed objects) _would_
honor .keep?

quoted

  <-a -l>
     -Restrict operation to only local objects. Only has any effect with -a|-A.
     -Like -a, but additionally exclude objects in packs accessible through
      alternates.

Presumably you meant "exclude objects accessible through alternates,
either in packs or in loose form"?  If so then I think it is a good thing
to have.

Would that be an enhancement to the current behavior? I don't think I saw
any mechanism to exclude packing remote loose objects.

The documentation for pack-objects --local says:

  --local
         This flag is similar to --incremental; instead of ignoring  all
         packed objects, it only ignores objects that are packed and not
         in the local object store (i.e. borrowed from an alternate).

It only mentions packed alternate objects.

I am not sure if listing the behaviour by combination of flags is a good
way to start thinking about this.  Wouldn't it be more productive to list
what kinds of repacking are needed, and then label them with combination
of flags?  Otherwise you would miss a potentially useful operation that
cannot be expressed with the current set of flags you have.

I agree. I made a list of the options because I was trying to understand what
effect each option had, then I turned it into an email.

I think the useful kinds are only these five:

 - scoop loose objects that exist in local repository into a new pack,
   without touching existing packs at all; exclude anything available in
   any existing pack or in alternate repository (either loose or packed);

 - pack everything that is needed by the local ref, except the ones that
   are borrowed from alternate repositories (either loose or packed), into
   a single new pack.  There are two variants of this: eject what is
   currently packed but unnecessary into loose format when existing local
   packs are replaced with the new pack, or lose them (i.e. -A).

 - fatten local repository by packing everything that is needed by the
   local ref into a single new pack, including things that are currently
   borrowed from alternates.  There are two variants of this: eject what
   is currently packed but unnecessary into loose format when existing
   local packs are replaced with the new pack, or lose them (i.e. -A).

You didn't say when local .keep should be honored, i.e. objects in local
packs with .keep should be excluded from repacking. always, never, only
with -l, new repack option?

-brandon