Re: [PATCH] Make sparse-directories checkouts/updates work with old servers

Vlad Georgescu <vgeorgescu@gmail.com> writes:
> Hmm, I'm not sure it's something as simple as a bugfix. Suppose we
> checkout a greek tree at depth=empty, and then do a depth=infinity
> update. How is the (1.5+) server supposed to know it should send A/ and
> iota? All it sees is the top-level set-path with a depth of infinity,
> the same as a fully complete working copy might send.

Sorry it's taken me so long to come back to this thread; I drove to a
conference a thousand miles away, talked, and drove back :-).

It seems we have a problem here. The client needs to do two separate
things:

   1) Describe the working copy (including depth) to the server.
   2) Tell the server what depth the client *wants* to have.

Have we been carelessly confusing these two things?

The reporter set_path() calls are about (1), and the 'depth' parameter
to (for example) svn_ra_do_update2() is about (2).

(I know you know all this, I'm just stating everything very carefully
here to make sure my own understanding is correct!)

Now, (2) is a single depth, sent exactly once during the update. It
doesn't really matter if it's sent early or late in the request -- I
lean toward early, because in some future full-duplex universe, that
might allow the server to start sending back answers sooner.

I think this means that r24493, r24468, and r24453 were mistaken. We
should be transmitting a depth from client to server, and remembering
it on the server side as the *requested* depth, not confusing it with
the *described* (default) depth.

Does this analysis sound right to you?

> With old servers it gets more complicated. You'd probably have to call
> delete-path on each file or directory that's missing, except that a
> working copy doesn't actually have that information.

I don't think that's necessary, we're just going to have the client
ignore unwanted portions of the update (cf. Mike Pilato's recent
patch to the RA layers).

> And, of course, whatever solution we come up with must simultaneously
> work for both old and new servers, because we don't currently
> distinguish between them.

Oh, I think we're just going to have different solutions for the
different situations.

For old servers, the client sends a requested depth (which the server
ignores) and per-path depth information (which the server also
ignores), and then the client ignores extra information the server
sends back.

For new servers, the client doesn't need to ignore anything, because
the server doesn't send any extra information anyway. But if it did,
the client would just ignore it :-).

-Karl

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu May 3 23:04:03 2007