Merge lp://staging/~jelmer/bzr/hpss-_get-checkout-format into lp://staging/bzr

Does this actually make sense to ask across the network? Is this
something we could instead either get at the time the Branch is
opened, or just know from local data about the branch format?

On 11/29/2011 08:28 PM, Martin Pool wrote:
> Does this actually make sense to ask across the network? Is this
> something we could instead either get at the time the Branch is
> opened, or just know from local data about the branch format?
The checkout format can be different from that of the source branch in
some cases. We already have a BzrDir.cloning_metadir() HPSS method,
which behaves similarly, but is used for clone-oriented-operations (vs
checkout operations).

BzrBranch4 uses a metadir-based checkout format while it itself isn't
metadir-based, and SVN has a different checkout dir format than its
storage dir format. These probably aren't very relevant in practice, though.

Perhaps the HPSS call should be for BzrDir.checkout_metadir() rather
than Branch._get_checkout_format() though?

Cheers,

Jelmer

It's nice to see the counts go further down.

I have a couple of concerns about this patch and others like it
 * there's no real documentation of what the method does; because this
is an interface where old bzr code talks to newer code it seems useful
to be a bit more explicit
 * I do wonder if we need a separate roundtrip to get stuff that it
seems could reasonably be fetched when the object is opened.

--
Martin

Am 14/12/11 00:37, schrieb Martin Pool:
> It's nice to see the counts go further down.
>
> I have a couple of concerns about this patch and others like it
> * there's no real documentation of what the method does; because this
> is an interface where old bzr code talks to newer code it seems useful
> to be a bit more explicit
That makes sense, I'll have a look at adding more documentation.

> * I do wonder if we need a separate roundtrip to get stuff that it
> seems could reasonably be fetched when the object is opened.
There is a similar pattern for getting the appropriate metadir format
for branch cloning, where we call the "BzrDir.cloning_metadir" HPSS verb
to figure out what the target format should be.

I agree that we shouldconsider chaining some of the HPSS requests.
Another case that comes to mind is to get the stacked-on branch URL
automatically when opening a branch, rather than having to request that
location using Branch.get_stacked_on_url.

I'm worried about taking too big leaps though - and adding a new branch
open call is more a lot more complex than BzrDir.checkout_metadir. The
addition of BzrDir.checkout_metadir will already save a couple of
roundtrips per checkout, so I would prefer to take that win now. Once we
get down to less than 15 roundtrips for a checkout, it should be a lot
easier to see which calls we can combine to eliminate even more
roundtrips. If that point happens to be before 2.5.0, we can always
remove BzrDir.checkout_metadir - existing 2.5 clients that try it will
automatically fall back to VFS if it's not there.

Cheers,

Jelmer

works for me.